molecular circuits, biological switches, and nonlinear dose

Environmental Health Perspectives • VOLUME 110 | SUPPLEMENT 6 | DECEMBER 2002 971

Risk assessment approaches increasingly usetwo complementary concepts—mode ofaction and tissue dose—to organize availabletoxicological and epidemiological studies, todecide on the shape of the dose–responsecurve (linear, nonlinear, or threshold), andto conduct low-dose, interspecies, doseroute, and interindividual extrapolations(Andersen and Dennison 2001). Mode ofaction provides a basis to decide qualita-tively on the shape of the dose–responsecurve (U.S. EPA 1999). By knowing thenature of tissue dose, i.e., whether it is par-ent chemical, metabolite, receptor-boundtoxicant, etc., physiologically based phar-macokinetic (PBPK) models can be used tocalculate tissue dose metrics for variousdose routes, species, and dose levels to sup-port extrapolations (Clewell and Andersen1985). These two organizing principles,mode of action and tissue dosimetry, havethe potential to encourage application of awide array of mechanistic data and biologi-cally based modeling in chemical riskassessment. However, for nonlinear andthreshold responses, acceptable exposurelimits are still primarily derived by a proce-dure based on objective evaluation ofdose–response curves, followed by applica-tion of multiple uncertainty factors thatadjust for interspecies differences inresponse, for interindividual response dif-ferences in humans, for adequacy of avail-able data, and to adjust to lifetime exposureperiods. Many of these uncertainty factorsare used because of ignorance about the

shape of dose–response curves at low inci-dence levels for these responses.

A variety of toxicants interfere with cel-lular signaling by endogenous endocrine sys-tem hormones and biological receptors(Kavlock and Ankley 1996). Our under-standing of these signaling pathways, at leastqualitatively, has increased markedly inrecent years through new techniques in mol-ecular biology, including studies usingknockout and transgenic animals and thedevelopment of high throughput methodsin genomics, transcriptomics, and pro-teomics. Some major signaling motifs thatare of interest for toxic responses includeactions of nuclear receptors, such as mem-bers of the steroid hormone family receptors(Landers and Spelsberg 1992), and G-pro-tein–coupled cell-surface receptors, such asthose proteins that recognize peptide-stimu-lating hormones secreted by the anteriorhypothalamus (Clement et al. 2001). A pos-sible impediment to the quantitative appli-cation of these data in risk assessment is thesheer volume of the information being col-lected at different levels of biological detail,i.e., the molecular, cellular, organ system,organism, and population levels. How willwe order and make sense of all this informa-tion to provide a more complete under-standing of dose–response curves forendogenous signaling components and forexogenous compounds that interfere withthese signaling motifs?

Noble recently emphasized the com-plementary roles of observation and

computational modeling in studying thephysiological function of biological systems.He writes:

The amount of biological data generated overthe past decade by new technologies has com-pletely overwhelmed our ability to understand it. . . Indeed, it is hard to see how [the] unravelingof complex physiological processes can occurwithout the iterative interaction between experi-ment and simulation . . . In a few years’ time weshall all wonder how we ever managed to dowithout [computational models]. . . . (Noble2002).

These same comments apply to thegeneration of information on mode ofaction in chemical risk assessment. Studieson mode of action are essentially qualita-tive in nature and must be organized byquantitative computational models tomake predict ions of the shape of thedose–response curves and to suggestimportant new experimentation. In riskassessment these computational modelsare referred to as biological ly baseddose–response (BBDR) models and pro-vide the substrate for simulations that linkmode of action research with predictedphysiological consequences of exposures(Setzer et al. 2001). Here we discuss riskassessment approaches for nonlinear toxi-cological processes that take into accountdose–response behaviors of native signal-ing molecules required for normal func-tion, perturbations of these systems by thepresence of toxicants, and BBDR model-ing of the underlying molecular circuitryassociated with normal and impaired func-tion of these signaling motifs.

This article is part of the monograph Application ofTechnology to Chemical Mixture Research.

Address correspondence to M.E. Andersen, CIIT-Centers for Health Research, PO Box 12137,6 Davis Dr., Research Triangle Park, NC 27709USA. Telephone: (919) 558-1205. Fax: (919) 558-1404. E-mail: [email protected]

American Chemistry Council Project RSK0005provided support for our work on switching inhepatocytes. National Institute of EnvironmentalHealth Sciences (NIEHS) Superfund BasicResearch Program project grant P42 ES05949,Agency for Toxic Substances and Disease Registrycooperative agreement U61/ATU 881475, andNIEHS Quantitative Toxicology training grantT32 ES07321 provided support for other aspects ofthe project.

Received 13 April 2002; accepted 2 October 2002.

Chemical Mixtures

Signaling motifs (nuclear transcriptional receptors, kinase/phosphatase cascades, G-coupled pro-tein receptors, etc.) have composite dose–response behaviors in relation to concentrations of pro-tein receptors and endogenous signaling molecules. “Molecular circuits” include the biologicalcomponents and their interactions that comprise the workings of these signaling motifs. Many ofthese molecular circuits have nonlinear dose–response behaviors for endogenous ligands and forexogenous toxicants, acting as switches with “all-or-none” responses over a narrow range of con-centration. In turn, these biological switches regulate large-scale cellular processes, e.g., commit-ment to cell division, cell differentiation, and phenotypic alterations. Biologically baseddose–response (BBDR) models accounting for these biological switches would improve riskassessment for many nonlinear processes in toxicology. These BBDR models must account fornormal control of the signaling motifs and for perturbations by toxic compounds. We describeseveral of these biological switches, current tools available for constructing BBDR models ofthese processes, and the potential value of these models in risk assessment. Key words: biologicalswitches, dose–response relationships, endocrine-active compounds, estrogen, molecularcircuitry, pharmacodynamic models, risk assessment, TCDD. Environ Health Perspect 110(suppl6):971–978 (2002).http://ehpnet1.niehs.nih.gov/docs/2002/suppl-6/971-978andersen/abstract.html

Molecular Circuits, Biological Switches, and Nonlinear Dose–ResponseRelationships

Melvin E. Andersen, Raymond S.H. Yang, C. Tenley French, Laura S. Chubb, and James E. Dennison

Quantitative and Computational Toxicology Group, Center for Environmental Toxicology and Technology, Foothills Campus, ColoradoState University, Fort Collins, Colorado, USA.

972 VOLUME 110 | SUPPLEMENT 6 | DECEMBER 2002 • Environmental Health Perspectives

Perturbations of SignalingPathways

Molecular CircuitryGene and protein arrays, simultaneouslyassaying the expression of hundreds or thou-sands of genes and proteins, provide infor-mation on expression of suites of genes andbatteries of protein products that are coordi-nately regulated and the manner in whichtoxicants alter their expression. However,the data from these arrays must be inte-grated to provide an understanding of bio-logical function rather than serving simplyas a catalog of change. Lander andWeinberg have expressed their opinionregarding the overall goal of genomics:

The long-term goal is to use this information toreconstruct the complex molecular circuitry thatoperates within the cell to map out the networkof interacting proteins that determines theunderlying logic of various cellular biologicalfunctions including cell proliferation, responsesto physiologic stresses, and acquisition andmaintenance of tissue-specific differentiationfunctions. A longer term goal, whose feasibilityremains unclear, is to create mathematical mod-els of these biological circuits and thereby pre-dict these various types of cell biologicalbehavior. (Lander and Weinberg 2000)

The underlying concept here is thatbiological functions require the successfuloperation of specific circuits that coordinateinformation flow and govern cellular behav-ior under a variety of physiological condi-tions. Analogous to electric circuits,molecular circuits consist of components(proteins, RNAs, signaling molecules, etc.,and cellular targets) that organize flows ofcellular information, although there isarguably more variability in the biologicalthan in the electrical system. Toxicants,especially those that interfere with hor-mones and signaling motifs, can interferewith normal functioning of these molecularcircuits, leading to altered function andultimately to toxicity.

Biological Switches

Molecular circuits are controlled by energyprovided in the form of receptors and lig-ands that activate signaling motifs, leading todownstream biological consequences. Mostresponses to these signaling compounds arethemselves nonlinear. The process by whichsome endogenous ligands or toxicant com-pounds cause nonlinear responses is referredto here as “switching.” Switches activate fun-damental changes in functional behaviors ofcells in an all-or-none fashion. The morenonlinear the response is, the more switch-like it is. Switches are another functionalcomponent of circuits. Signaling motifs withlinear dose–response behavior would beexpected to show graded responses to

changes in signal concentration. However,very few examples of graded transcriptionalresponse have been reported for eukaryoticgene expression (Louis and Becskei 2002).The carefully timed development of organ-isms from a single fertilized cell requirescoordination of a series of cellular switchesto complete the transformation from a singlefertilized cell to a mature organism(Davidson et al. 2002).

Several nonlinear switching processeshave been examined quantitatively.Progesterone induces maturation of Xenopusoocytes. The dose response for maturationof individual oocytes was described with aHill equation for the activation of mitogen-activated protein kinase (MAPK) (Ferrelland Machleder 1998):

[1]

where C is concentration of progesteroneand K d

n is the concentration at half-maximalresponse. In the Hill equation the exponentn (Hill coefficient) determines the steepnessof the dose–response curve. In simpler mol-ecular systems such as dimerization of estro-gen receptors, an n value of 2.0 for afunctional response such as gene expressionmay simply indicate involvement of abimolecular process. With oxygen bindingto hemoglobin, n values of 3–4 indicate thatbinding of the first oxygen molecule to thehemoglobin tetramer leads to structuralchanges, facilitating binding to the remain-ing three sites (allosteric binding). WithXenopus oocyte maturation, Hill coefficientsfor individual oocytes were reported to bebetween 20 and 30 (Ferrell and Machleder1998). These high values indicate an all-or-none switch for maturation in any individ-ual oocyte. For populations of oocytes theprogesterone concentrations causing matu-ration were distributed fairly broadly.Evaluation of populations of oocytes didnot show the all-or-none responses noted inindividual cells. In the oocyte, progesteronereceptor and MAPK cascade-signalingmotifs combine to control a switch that ini-tiates maturation circuitry.

Receptor upregulation may also producenonlinear biological responses. Certain bac-teria change phenotypic characteristicsunder conditions that encourage bacterialgrowth. As bacterial concentrations increase,secondary metabolites are released into thesurrounding media. As the concentration ofthis signaling molecule increases, themetabolite, in concert with a cytosolicreceptor, initiates transcription of a set ofgenes leading to new phenotypic character-istics in the bacteria. Photobacterium fischreii

regulates genes that control phosphores-cence via this mechanism (Fuqua et al.1994). The suites of genes controlled by thesecondary metabolite–receptor interactioninclude the receptor protein itself and anenzyme that converts the secondary metabo-lite to a higher-affinity ligand. Thus, thesemetabolites indirectly serve as surrogates forthe concentration of bacteria. Because of therelationship to bacterial number, theseresponses are called “quorum sensing.” Themetabolite accumulation signals the bacteriathat they are present in sufficient quantityto change phenotypic characteristics. Manybacteria, including film formation in somespecies (Costerton et al. 1995; Davies et al.1998), use quorum-sensing motifs torespond to environmental stimuli. Efforts tomodel these biological processes are alsounder way (Koerber et al. 2002).

Estrogen receptor upregulation controlsvitellogenesis in some fish, leading to non-linear dose–response relationships and stabledifferentiation of hepatocytes treated withhigh doses of estrogenic compounds(Shapiro et al. 1989). The possible role ofreceptor upregulation in these nonlinearresponses was investigated by simulationwith a pharmacodynamic gene inductionmodel. Several generic gene induction mod-els were developed and exercised (Andersenand Barton 1999). These models recapitu-lated nonlinear behaviors with very highHill coefficients. More recently a transcrip-tional switch controlling methylation ofnucleosomes and transcriptional cofactorswas described (Xu et al. 2001) that was asso-ciated with coactivator-associated argininemethyltransferase-1 (CARM-1). This mole-cule acts as a coactivator for nuclear hor-mone signaling via histone methylation andas a co-repressor of cyclic-AMP–associatedsignaling pathways. This switch involveslimiting concentrations of the CARM-1coactivator and acts via histone modificationto increase access to specific genes and pro-moter regions. Activation of transcriptionby the nuclear receptors, in concert withCARM-1, is expected to activate groups ofgenes and to silence others.

The proper function of all these signalingpathways requires correct concentrations of avariety of endogenous proteins and signalingmolecules. To a very large extent, doseresponse, a concept used frequently in toxicol-ogy in regard to adverse responses, now has anequally important position in normal biology.The ideas expressed in this regard are propergene dosages in cells, leading to proper con-centrations of receptors and ligands for nor-mal function. The proper functioning ofmolecular circuits and maintenance of healthyconditions in the organism requires appropri-ate doses of various signaling components and

Re

Re,sponse

sponse=

×

+MAX

n

dn n

C

K C

Chemical Mixtures • Andersen et al.


their presence at appropriate times for activa-tion of biological switches.

Endocrine-Active CompoundsMany endocrine-active compounds (EACs)interact with the normal cell circuitry tomimic or antagonize the actions and functionsof normal signaling systems. Excess of EACsor deficiencies of natural hormones alter hor-monal system function, leading to impairedhealth. Impaired health covers a wide range ofresponses, including loss of viability, impairedperformance, altered reproductive success, anddelayed maturation. The actions of many sig-naling elements—cell-surface receptors,cytosolic transcriptional factors, kinase/phos-phatase cascades—are now more completelyunderstood than just a few years ago. Studiesof the toxic responses to EACs need to beginwith examination of normal function of thesesignaling motifs and how the normal functionbecomes perturbed by exogenous compounds(Figure 1). The first requirement in a BBDRmodel, then, is to develop an adequate repre-sentation for the dose–response control of nor-mal function. Secondarily, the focus is on theperturbation of normal function by exogenouscompounds. This reorientation to a perturba-tion approach to normal biology rather thanan emphasis on final pathology providesnew avenues and strategies to evaluatedose–response relationships in biology and intoxicology. These EACs serve as examplesthrough the remainder of this article.

BBDR Models for MolecularCircuits and Switches

Concerted Cellular Responses

Most dose–response assessment models intoxicology assume smooth, continuous

changes in response to dose. These modelsdescribe many chemical processes by statisti-cal methods with average behaviors of mole-cules, as the numbers of particles involvedin most reactions and interactions are verylarge. The real world of cells demonstrates amore complex variety of interacting cir-cuitry. At the cellular level, behaviors aremore likely to be nonlinear and stochastic.A cell either divides or it does not divide. Inmoving from one phenotypic state toanother, all the components have to changein concert to achieve a smooth pleiotropicalteration in cell characteristics. A challengein formulating the mathematical models ofcellular functions is the requirement tograsp the manner in which continuouschanges of chemical variables (i.e., ligandand receptor concentrations) lead to sto-chastic responses such as apoptosis, prolifer-ation, differentiation, or activation of globalcellular circuitry by exposure to chemicals.

Stochastic, nonlinear models of cellular-level responses may provide the basis fordeveloping tools that will simulate nonlineardose–response behaviors toward toxic expo-sures. Some stochastic models assess cancerrisks based on rates of cell division, celldeath, and cell mutation. The Moolgavkar-Venzon-Knudson (MVK) model (Figure 2)represents a stochastic model of carcinogen-esis (Moolgavkar and Knudson 1981).These cancer models have to be initially setto describe tumor incidence in the controlanimals. These background rates are thenaltered, i.e., perturbed by the actions of tox-icant compounds. In developing BBDRmodels it is necessary to evaluate the effectof dose on intrinsic biological parameters ofthe model. The effects can be describedempirically, as has usually been done, ormechanistically. For the cancer models thestochastic aspect involves some probability

of division, death, or mutation that occursrandomly. Mechanistically, the requirementis to understand (model) the relationship ofthese probabilities with dose and todescribe the manner in which dose changesthe probability of division, death, or muta-tion during a time interval. The relation-ships between dose and cell proliferation orbetween dose and cell apoptosis are unlikelyto be simple continuous functions. Thecontrol of biological circuitry and the tran-sition between different states of the cellu-lar circuitry in response to exogenoussignaling molecules should determine thedose–response manifestations for prolifera-tion, apoptosis, and mutation in many ofthese cancer models.

Receptor-Mediated Control of Gene ProductsMany EACs directly or indirectly interferewith gene expression. In discussing molec-ular circuits the changes are generally coor-dinate alterations in groups of genes thatlead to altered biological characteristics ofthe affected cells. It is often possible tomeasure responses of single genes withgreat precision using modern techniquessuch as polymerase chain reaction amplifi-cation of gene transcripts. Molecular mark-ers, such as induction of cytochromeP4501A1 (CYP1A1) message by 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD)(Vanden Heuvel et al. 1994), allows obser-vation of the dose–response curves in lowerdose ranges than possible when examiningovert adverse responses of the organism.However, these measurements lead toquestions about the linkage between theseprecursor effects and clearly adverseresponses of the organism. For instance,should the observation of a 1% increase inCYP1A1 mRNA after 2,3,7,8-TCDDtreatment be considered adverse? Thisfocus on a single gene may not be the cor-rect one for assessing toxicity. Dioxin andsimilar receptor-mediated EACs do notsimply control expression of a single genein the intact liver. They alter concentra-tions of a battery of gene products toinduce a concerted, pleiotropic response inhepatocytes (Bock 1993).

An Example with Tumor PromotionDioxin is a liver carcinogen in rats and atumor promoter (Kociba et al. 1978; Pitotet al. 1987). Many liver tumor promotersact by transiently increasing proliferation ofhepatocytes with longer-term adaptation tothe exposures. The adaptation, with pheno-barbital, involves elaboration of transform-ing growth factor-β, a specific growthfactor that constrains hepatocyte prolifera-tion (Jirtle et al. 1991). Cells resistant to

Chemical Mixtures • Biological switches and risk assessment

Exposure

Tissue dose

Biochemicalinteractions

Biologicalconsequences

PBPK models

BBDR models

Homeostasis,adaptation,and repair

Normalstate

Alteredstate

Altered physiologyPathologyDisease/death

�

�

Figure 1. Dose–response models for perturbationsof signaling motifs focus on normal biology includ-ing dose–response behavior for endogenous sig-naling molecules and cognate receptors. Theactions of EACs would appear as perturbations onthe normal, nonlinear control of molecular circuitryand the switching modules moving between andamong various circuits.

α1

µ1

β1 β2

µ2

α2

N I M

Figure 2. The MVK model for cancer, including celldivision, cell death, and probabilities of mutationduring replication. Although the model itself is sto-chastic, the biological processes represented bybirth and death may themselves represent toxicantactions on nonlinear signaling motifs associatedwith perturbation of cellular circuits by the pres-ence of toxicants. N represents the population ofnormal cells, I the initiated cells, and M themutated cells. The parameter α represents the cellbirth rate, β the cell death rate, and µ the transfor-mation rate, where subscript 1 refers to the normalpopulation and 2 refers to the initiated population.

cytoinhibition are presumed to derive agrowth advantage and grow out to preneo-plastic foci under the selection pressurefrom the promoter. The dose–responserelationship for carcinogenesis requirescharacterization of the dose of promoterrequired to increase the proliferation ofhepatocytes.

Phenobarbital, in common with a largenumber of liver tumor promoters, has areceptor-mediated mode of action. Thesepromoters interact with protein receptorsthat serve as transcriptional modulators toalter expression of batteries of genes in thehepatocytes. Phenobarbital interacts via theconstitutive androstane receptor (Waxman1999). Other liver tumor promoters act viathe aryl hydrocarbon receptor (AhR)(Wilson and Safe 1998), the peroxisomeproliferation–activating receptor (Weghorstet al. 1994), or the pregnane-X receptor(Waxman 1999). All these receptors, inconcert with the toxic compounds, act toincrease expression of batteries of genes,leading to several alterations in expression ofmany individual gene products, includingspecific genes. The effects on cell-level char-acteristics are likely to be associated withthese pleiotropic responses.

Among these promoters, 2,3,7,8-TCDDhas received a great deal of attention in thepast decade as the U.S. EnvironmentalProtection Agency has reevaluated the risksof exposure to this environmental contami-nant (U.S. EPA 2000). PBPK and proteininduction models describe dioxin kinetics inthe body, including binding to the AhR

with activation of specific gene products(Kohn et al. 1993; Leung et al. 1990).These models have also described the induc-tion of specific genes through interactions ofthe AhR–TCDD complex with DNA-response elements for the AhR and variouspartnering molecules (Andersen et al.1997b. Like most mathematical models ofbiological systems, the BBDR modelspresently available for dioxin represent a sig-nificant simplification of the individualmolecular processes. Simplifications are nec-essary to attain a computationally tractablemodel and may be useful to gain insightsabout dose–response behavior (Suk andYang 2002). Bailey has emphasized thevalue and necessity of using simplificationsin modeling complex, biological systems(Bailey 2001). However, it is important thatthe simplifications retain the important bio-logical aspects of the responses.

The regulation of gene expression bydioxin and dioxinlike polychlorinatedbiphenyls (PCBs) and the AhR has beeninvestigated in a variety of systems, includ-ing many cell constructs with the CYP1A1promoter upstream of particular markergenes (Garrison et al. 2000; Jeon and Esser2000). Such constructs allow evaluation ofthe components necessary to control geneexpression by the AhR in a system with anavailable promoter. However, in the intactanimal, the silencing or activation ofgenomic structures that are not present inthe cell constructs may alter gene expres-sion. We are studying AhR-mediated induc-tion of CYP1A1 in primary hepatocytes.

Induction does not follow coherentdose–response relationships expected for aHill relationship with a low n value. Thisbehavior is true both in vivo (Chubb et al.2002) and in vitro in the primary hepato-cytes (French et al. 2002), as shown inFigures 3 and 4, respectively.

With AhR agonists, the response of cellsin the liver and of cells in vitro does notappear to follow a continuous response pat-tern where a 50% induction in the total liveris reflected by a 50% induction in all hepa-tocytes. The induction of individual cellsappears to occur almost in an all-or-nonefashion. Cells are either induced or remainin a basal state (Andersen et al. 1997a;Tritscher et al. 1992). This response, a con-certed response of a cell to the receptor–lig-and complex, is not yet well understood.The molecular circuitry that causes thisswitchlike behavior leads to a qualitativealteration in response over a narrow range ofdose, moving the cell from one state to anew one. In the liver different acinar regionshave varying sensitivity for induction. At lowdoses centrilobular cells are induced. As doseincreases, more of the cells in the liverbecome induced and the region of inducedcells progresses toward the periportal area ofthe liver acinus (Figure 3). Andersen et al.modeled regional enzyme induction using asemiempirical induction model (Andersen et al. 1997a) that could represent the differen-tial induction throughout the liver (Figure 5).This regional induction model coupled aPBPK model for the disposition of dioxin inthe body, a geometric representation of the



1.0 µg/kg PCB 126

0.1 µg/kg PCB 126

10 µg/kg PCB 126

0.1 µg/kg PCB 126 + 10 µg/kg PCB 153 1.0 µg/kg PCB 126 + 100 µg/kg PCB 153

10 µg/kg PCB 126 + 1,000 µg/kg PCB 153 10 µg/kg PCB 126 + 10,000 µg/kg PCB 153

Corn oil control

Figure 3. Induction of CYP1A1 and 1A2 by PCB 126 (3,4,5-3´,4´-pentachlorobiphenyl) in rat liver and the influence of a second promoter. (A) The pattern of immuno-histochemical staining for CYP1A1 protein is consistent with a switch that induces a change from a normal phenotype to an AhR-responding phenotype over anarrow range of dose. In the context of the topic of this article, this AhR-agonist PCB has altered the molecular circuitry of these hepatocytes, leading to activa-tion of a cellular switch. (B) In the presence of high daily doses of PCB 153, a phenobarbital-like enzyme inducer that increases concentrations of a differentcytochrome (CYP2B1/2), PCB 126 no longer induces cells in the centrilobular region of the liver. The presence of high PCB 153 (10,000 µg/kg) has turned off theAhR switch (Chubb et al. 2002).

liver acinus, and a more empiricalal descrip-tion of enzyme induction. Enzyme induc-tion in each zone of the liver acini wasmodeled with a Hill-type equation, withvariable affinities for AhR–ligand–DNAinteractions in each acinar region. Successfulmodeling of induction in a five-compart-ment liver acinus required that bindingaffinities differ by a factor of three betweenadjacent acinar regions, with high Hill coef-ficients for induction (i.e., 4–5) in eachregion of the acinus. The next step will be toprovide a more mechanistic description ofthe molecular circuit and switching thatcomprises this all-or-none behavior.

These switching responses of hepato-cytes appear to represent a reversible differ-entiation to a new stable state (a newphenotype) of the hepatocyte. This differen-tiation includes the concerted induction of abattery of genes. Interestingly, the currentinduction models (Andersen et al. 1997b;Kohn et al. 1993), if extended to describemultiple genes with independent promoters,would give rise to competitive, not coopera-tive, interactions. Biological mechanismsthat might explain nonlinear, concertedresponses of gene batteries include receptorautoregulation, genomic level switches, suchas that noted for histone methylation, orkinase cascades. We are now developing anexperimental system to study responses ofisolated hepatocytes to dioxinlike com-pounds (French et al. 2002). This experi-mental system is intended to permitevaluation of the mechanistic characteristics

of the hepatocyte switch, including the roleof kinase cascades or histone modification,in these processes. The accumulation ofmore mechanistic data on induction is nec-essary to provide sufficient biological detailto predict low-dose behavior.

Discussion

Modeling Tools for DescribingBiological Switches

Chaos and attractors. Our continuing eval-uation of induction responses of hepatocyteshas led us to a set of new concepts for futureexposure dose–response assessments. Theyinclude biological switching, molecular cir-cuits, and multiple stable states of the cells,in addition to our old concerns regardingthe relationship of molecular-level responsesand the ultimate expression of toxicity.How will we model these responses to pre-dict responses over a wide range of dosebased on biological characteristics of cellularswitches? Chaos and complexity theoristshave discussed concepts of stable attractorsin complex systems. In the context of mole-cular biology, an attractor is the proteomicstate of the cell (including the antecedentgenomic state) that is stable because of itsability to maintain homeostasis within arange of conditions. The attractor conceptimplies that there are a finite number of sta-ble states that exist rather than a smoothtransition between infinite numbers of cellphenotypes. In particular, these conceptssuggest that mammalian cells may exist in a

suite of differentiated forms that representstable attractors for the overall behavior ofthe genetic content of the cell (Kaufman1995). Shapiro and colleagues (1989) andSimon et al. (1988) have pursued limitedmodeling of stable states for hormoneresponsiveness of cells for estrogen-respon-sive actions. The basal or induced states inhepatocytes caused by tumor promotersmay represent two stable attractors. Theincreasing concentration of the receptor–lig-and complex may alter the concentration ofa limited set of initial gene products thatmove the circuitry from that for one stableattractor to a second stable attractor. Overtime, the overall content and behaviorchange, consistent with the new stable state.The new state determines the pathologicalor physiological consequences of inductionfor the cell, whereas the dose response of theprocess is more likely determined by someof the early interactions of the ligand andthe receptor molecules in the most sensitivepopulation of cells. Tyson et al. (2001) haveused similar cellular paradigms to describethe cell cycle.

Early (transient) and late (persistent)responses to signaling molecules. Switches arelikely to be organized by positive feedbackcircuits to drive transitions from one state toanother. In normal maintenance of cell func-tion in a given state, homeostatic responsesare more generally associated with negative



DMSO treatment PCB 126 0.01 µM

PCB 126 1 µM PCB 126 100 µM

Figure 4. CYP1A1 staining in vitro. Immunohistochemical staining for CYP1A1 in rat primary hepatocytestreated with various concentrations of PCB 126 for 24 hr. The staining occurs in increasing numbers ofcells with increasing dose rather than increasing concentrations of protein in each cell proportional todose (French et al. 2002).

Figure 5. The predicted staining pattern for dioxininduction of CYP1A1 in rat liver at various doses ofdioxin. A PBPK model was linked to a nonlinear,semiempirical model of gene induction to examinethe degree of nonlinearity indicated by the regionalinduction data with various daily doses of dioxin,as observed by Tritscher et al. (1992). The regionalHill coefficient for protein induction required toprovide demarcation between adjacent regions ofthe liver, 4–5, indicated a switch controlling differ-ent phenotypic behaviors of the hepatocytes.Reproduced from Andersen et al. (1997a) with per-mission from Academic Press.

feedback, as with the feedback processes forendocrine target-organ function and releaseof stimulating hormones from the pituitary.One concept involved in steroid hormonefunction (Landers and Spelsberg 1992) andin memory storage was the involvement ofearly and late responses organized by tran-scriptional receptor or nervous system activa-tion of cells (Kandel 2001). Here, earlyresponses are more transient; however, ifthese signals persist or are of sufficient mag-nitude, they initiate more permanent alter-ations of genetic expression of gene batteriesand alterations of cell characteristics.

Some possible experimental and compu-tational models. Among a wider range ofpossibilities, it is apparent that nonlinearswitching modules exist for receptorautoregulation (Shapiro, et al. 1989),kinase/phosphatase cascades (Ferrell andMachleder 1998), and Ca2+-mediatednerve-cell signaling related to long-termpotentiation (Bhalla and Iyengar 1999).Although the nonlinear characteristics ofthese switches are evident, none have beenexamined in sufficient detail to provide aquantitative understanding of the molecularbasis of the switch and its influence on thedose–response curve at low incidence levels.Bhalla and co-workers have modeled variouscascade interactions that may be involved inlong-term potentiation in neurons andmaintaining memory. Their work includedbiologically realistic kinetic models of theseprocesses that capture the emergence ofaltered cellular characteristics arising fromparticular pulse trains at the cell surface(Bhalla and Iyengar 1999). Their modelstructures and other efforts to create virtualcells should aid in providing the biologicaldetail for realistic BBDR models for varioussignaling motifs.

Glycoprotein-stimulating hormonessuchas thyroid-stimulating hormone, follicle-stimulating hormone (FSH), and luteinizinghormone are released by the pituitary andhave end-organ effects on endocrine tissues.These hormones cause cellular responses bybinding to cell-surface G-protein–coupledreceptors. Binding leads to activation ofadenylate cyclase (AC), with production ofcyclic adenosine monophosphate (cAMP).Phthalate esters interfere with FSH-mediatedsignaling in Sertoli cells (Heindel and Chapin1989), although the exact sites of interactionremain uncertain. The responses of variousendocrine tissues to these hormones dependon cAMP and on a variety of other signalingmolecules in the cell (Richards 2001), includ-ing inducible kinases and guanosine triphos-phatases. The efforts to unravel thesesignaling pathways should lead to a represen-tation of the key functional elementsinvolved in G-protein–coupled signaling in

these cells and an improved understanding ofthe role of these cascades in toxicity, disease,and health.

Generic tools. New methods for model-ing the control of gene batteries in normalsystems may use Boolean networks(Kaufman 1995) or apply neural networkmodels (Vohradsky 2001) for expression ofmultiple gene families. Quantum computa-tional or predictive structural activity rela-tionship approaches, such as the reactionnetwork modeling approach being devel-oped in our laboratory (Liao et al. 2002),should facilitate the simulation of molecularcircuits and cellular switches. The contri-butions from groups developing more quan-titative tools to assess physiologicalcoordination of multiple cellular activities,e.g., the Physiome Project (Physiome 2002),may significantly expand the mathematicaldose–response modeling approaches used bytoxicologists and risk assessors for integrat-ing the exposure–response–dose paradigminto a perturbation paradigm assessing toxicpotential of compounds. Programming toolsfor modeling neuronal function (Genesis2002; Wilson et al. 1989) or the entire cell(Tomita et al. 1999) are also available forevaluating integrated signaling motifs.Although there are many candidate tools fordeveloping models of genetic circuitry, theimmediate future in this area will requiretrial and error to discern the mathematic/

simulation tools that will allow the mostrapid progress. A brief synopsis of some ofthe major signaling motifs and several avail-able tools for analysis of biological circuits islisted in Table 1.

Systems theory. Increasingly, integrativebiological research relies on systems theoryfor connecting biological experiments withcomputational descriptions of cellular struc-ture and dynamics. Systems theory, i.e., thestudy of systems that are conceptualizedusing networks to define information flows,uses engineering concepts such as robust-ness, fragility, and failure cascades (Cseteand Doyle 2002) and serves as an organiza-tional scaffold to support complex systemmodeling. Current interest in systems biol-ogy is partly an outgrowth of the need toconceptualize, hypothesize, modularize,design experiments, communicate ideas,and share results between research institu-tions. It is also partly a result of the need tointegrate new computational methods, toconstruct and interface complex modelsfrom different sources and disciplines, andto integrate diverse information fromgenomics, proteomics, and metabonomicsinto coherent models readily shared amongdifferent research groups. Bifurcation theory(Tyson et al. 2001) may serve as a computa-tional aid to identify attractors, even thoughthe parameter space for the individual com-ponents of the biological system becomes



Table 1. Some signaling motifs and possible analysis tool.

Motif or tool Mechanism Reference

Empirical transcriptional Enzyme induction, receptor binding Kohn et al. 1993activation models and transcriptional upregulatory Andersen et al. 1997a,b

mechanismsSteroid hormones Vitellogenesis Shapiro et al. 1989

Tata et al. 1993Oocyte maturation Ferrell and Machleder 1998Early/late responses Landers and Spelsberg 1992

Schuchard et al. 1993Peptide hormones FSH-and G-protein–coupled surface Clement et al. 2001

receptorsCell cycle Positive feedback in protein regulation Tyson et al. 2001

through phosphorylation and relatedcontrol mechanisms

Cell cycle map Kohn 1999DNA methylation Transcriptional regulation through Xu et al. 2001

histone modificationCellular signaling Alliance for cellular signaling AFCS 2002Stochastic processes Two-stage tumorogenesis Moolgavkar and Knudson 1981

Two-stage growth model for Conolly and Kimbell 1994precancerous lesions and tumors

Neuronal function Genesis Genesis 2002Systems biology Systems Biology Markup Language Systems Biology 2002Modeling tools Berkeley Madonna Macey and Oster 2002

Advanced Continuous Simulation ACSL 2002Language (ACSL)

MatLab MathWorks 2002Scope SCoP 2002BioSpice (under development) Biospice 2002

very large. This approach examines theeffects of parameter variation on solutionsof the nonlinear circuit model.

Rapid progress applying systems biologyis apparent in a number of research fields,including immune function (Germain2001), the cardiac system (Noble 2002),developmental biology (Davidson et al.2002), and prokaryotic systems (Kitano2002; Weng et al. 1999). Eventually,BBDR models for alterations in signalingmotifs and nonlinear toxicological responsesmay be linked to organ system descriptionsof physiology to predict both early responses(i.e., activation or deactivation of biologicalswitches regulating signaling motifs) andadverse responses (i.e., diminished physio-logical function or diminished adaptabilityto stress). Such models would fulfill thegoals proposed by Noble of having tools tosimulate expected results, aid in experimen-tal design, and predict biological/toxicologi-cal consequences throughout the full rangeof exposure situations with environmentaltoxicants (Noble 2002).

REFERENCES

ACSL. 2002. Advanced Continuous Simulation Language.Huntsville, AL:AEgis Technologies Group, Inc. Available:http://www.aegistg.com/ [accessed 20 Sept 2002].

AFCS. 2002. Alliance for Cellular Signaling. Dallas, TX:Alliancefor Cellular Signaling. Available: http://www.afcs.org[accessed 20 Sept 2002].

Andersen ME, Barton HA. 1999. Biological regulation of recep-tor-hormone complex concentrations in relation to dose-response assessments for endocrine-active compounds.Toxicol Sci 48:38–50.

Andersen ME, Birnbaum LS, Barton HA, Eklund CR. 1997a.Regional hepatic CYP1A1 and CYP1A2 induction with2,3,7,8-tetrachlorodibenzo-p-dioxin evaluated with a multi-compartment geometric model of hepatic zonation.Toxicol Appl Pharmacol 144:145–155.

Andersen ME, Dennison JE. 2001. Mode of action and tissuedosimetry in current and future risk assessments. SciTotal Environ 274:3–14.

Andersen ME, Eklund CR, Mills JJ, Barton HA, Birnbaum LS.1997b. A multicompartment geometric model of the liver inrelation to regional induction of cytochrome P450s.Toxicol Appl Pharmacol 144:135–144.

Bailey JE. 2001. Complex biology with no parameters. NatureBiotechnol 19:503–504.

Bhalla US, Iyengar R. 1999. Emergent properties of networks ofbiological signaling pathways. Science 283:381–387.

Biospice. 2002. Biospice. Berkeley, CA:University of California.Available: http://genomics.lbl.gov/index.html [accessed 20Sept 2002].

Bock KW. 1993. Aryl hydrocarbon or dioxin receptor: biologicand toxic responses. Rev Physiol Biochem Pharmacol125:1–42.

Chubb LS, Benjamin SA, Painter JT, Dean CE, Andersen ME.2002. Regional induction of CYP1A1 protein during treat-ment of rats with mixtures of PCB126 and PCB153[Abstract]. Toxicologist 66:93.

Clement F, Monniaux D, Stark J, Hardy K, Thalabard JC,Franks S, et al. 2001. Mathematical model of FSH-inducedcAMP production in ovarian follicles. Am J PhysiolEndocrinol Metab 281:E35–E53.

Clewell HJ, 3rd, Andersen ME. 1985. Risk assessment extrapo-lations and physiological modeling. Toxicol Ind Health1:111–131.

Conolly RB, Kimbell JS. 1994. Computer simulation of cellgrowth governed by stochastic processes: application toclonal growth cancer models. Toxicol Appl Pharmacol124:284–295.

Costerton JW, Lewandowski Z, Caldwell DE, Korber DR,Lappin-Scott HM. 1995. Microbial biofilms. Annu RevMicrobiol 49:711–745.

Csete ME, Doyle JC. 2002. Reverse engineering of biologicalcomplexity. Science 295:1664–1669.

Davidson EH, Rast JP, Oliveri P, Ransick A, Calestani C, YuhCH, et al. 2002. A Genomic regulatory network for develop-ment. Science 295:1669–1678.

Davies DG, Parsek MR, Pearson JP, Iglewski BH, CostertonJW, Greenberg EP. 1998. The involvement of cell-to-cellsignals in the development of a bacterial biofilm. Science280:295–298.

Ferrell JE Jr, Machleder EM. 1998. The biochemical basis of anall-or-none cell fate switch in Xenopus oocytes. Science280:895–898.

French CT, Chubb LS, Bill ings R, Andersen ME. 2002.Developing an in vitro model system to evaluate switch-like behaviors of hepatocytes in response to variousenzyme inducers [Abstract]. Toxicologist 66:89.

Fuqua WC, Winans SC, Greenberg EP. 1994. Quorum sensing inbacteria: the LuxR-LuxI family of cell density- responsivetranscriptional regulators. J Bacteriol 176:269–275.

Garrison PM, Rogers JM, Brackney WR, Denison MS. 2000.Effects of histone deacetylase inhibitors on the Ah recep-tor gene promoter. Arch Biochem Biophys 374:161–171.

Genesis. 2002. General Neural Simulation System. Pasadena,CA:California Institute of Technology. Available:http://www.genesis-sim.org/GENESIS/ [accessed 20 Sept2002].

Germain RN. 2001. The art of the probable: system control inthe adaptive immune system. Science 293:240–245.

Heindel JJ, Chapin RE. 1989. Inhibition of FSH-stimulatedcAMP accumulation by mono(2-ethylhexyl) phthalate inprimary rat Sertoli cell cultures. Toxicol Appl Pharmacol97:377–385.

Jeon MS, Esser C. 2000. The murine IL-2 promoter containsdistal regulatory elements responsive to the Ah receptor,a member of the evolutionarily conserved bHLH-PAS tran-scription factor family. J Immunol 165:6975–6983.

Jirtle RL, Meyer SA, Brockenbrough JS. 1991. Liver tumor pro-moter phenobarbital: a biphasic modulator of hepatocyteproliferation. Prog Clin Biol Res 369:209–216.

Kandel ER. 2001. The molecular biology of memory storage: adialogue between genes and synapses. Science294:1030–1038.

Kaufman S. 1995. At Home in the Universe. New York:OxfordUniversity Press.

Kavlock RJ, Ankley GT. 1996. A perspective on the risk assess-ment process for endocrine-disruptive effects on wildlifeand human health. Risk Anal 16:731–739.

Kitano H. 2002. Systems biology: a brief overview. Science295:1662–1664.

Kociba RJ, Keyes DG, Beyer JE, Carreon RM, Wade CE,Dittenber DA, et al. 1978. Results of a two-year chronictoxicity and oncogenicity study of 2,3,7,8-tetra-chlorodibenzo-p-dioxin in rats. Toxicol Appl Pharmacol46:279–303.

Koerber AJ, King JR, Ward JP, Williams P, Croft JM, SockettRE. 2002. A mathematical model of partial-thickness burn-wound infection by Pseudomonas aeruginosa: quorumsensing and the build-up to invasion. Bull Math Biol64:239–259.

Kohn KW. 1999. Molecular interaction map of the mammaliancell cycle control and DNA repair systems. Mol Biol Cell10:2703–2734.

Kohn MC, Lucier GW, Clark GC, Sewall C, Tritscher AM, PortierCJ. 1993. A mechanistic model of effects of dioxin on geneexpression in the rat liver. Toxicol Appl Pharmacol120:138–154.

Lander ES, Weinberg RA. 2000. Genomics: journey to the cen-ter of biology. Science 287:1777–1782.

Landers JP, Spelsberg TC. 1992. New concepts in steroid hor-mone action: transcription factors, proto-oncogenes, andthe cascade model for steroid regulation of gene expres-sion. Crit Rev Eukaryot Gene Expr 2:19–63.

Leung HW, Paustenbach DJ, Murray FJ, Andersen ME. 1990. Aphysiological pharmacokinetic description of the tissuedistribution and enzyme-inducing properties of 2,3,7,8-tetrachlorodibenzo-p-dioxin in the rat. Toxicol ApplPharmacol 103:399–410.

Liao KH, Dobrev ID, Dennison JE Jr, Andersen ME, Reisfeld B,Reardon KF, et al. 2002. Application of biologically basedcomputer modeling to simple or complex mixtures.Environ Health Perspect 110(suppl 6):957–963.

Louis M, Becskei A. 2002. Binary and Graded Responses inGene Networks. Washington, DC:Science’s SignalTransduction Signal Network. Available: www.stke.org/cgi/content/full/sigtrans;2002/143/pe33 [accessed 20 Sept2002].

Macey RI, Oster GF. 2002. Berkeley Madonna. Berkeley,CA:University of California. Available: http://www.berke-leymadonna.com/ [accessed 20 Sept 2002].

MathWorks. 2002. MatLab. Natick, MA:The Mathworks, Inc.Available: http://www.mathworks.com/ [accessed 20 Sept2002].

Moolgavkar SH, Knudson AG Jr. 1981. Mutation and cancer: amodel for human carcinogenesis. J Natl Cancer Inst66:1037–1052.

Noble D. 2002. Modeling the heart—from genes to cells to thewhole organ. Science 295:1678–1682.

Physiome. 2002. The Physiome Project. Seattle, WA:Universityof Washington. Available: www.physiome.org [accessed20 Sept 2002].

Pitot HC, Goldsworthy TL, Moran S, Kennan W, Glauert HP,Maronpot RR, et al. 1987. A method to quantitate the rela-tive initiating and promoting potencies of hepatocarcino-genic agents in their dose-response relationships toaltered hepatic foci. Carcinogenesis 8:1491–1499.

Richards JS. 2001. New signaling pathways for hormones andcyclic adenosine 3´,5´-monophosphate action inendocrine cells. Mol Endocrinol 15:209–218.

Schuchard M, Landers JP, Sandhu NP, Spelsberg TC. 1993.Steroid hormone regulation of nuclear proto-oncogenes.Endocr Rev 14:659–669.

SCoP. 2002. Simulation Control Program. Redlands,CA:Simulation Resources, Inc. Available: http://www.sim-resinc.com [accessed 20 Sept 2002].

Setzer RW, Lau C, Mole ML, Copeland MF, Rogers JM, KavlockRJ. 2001. Toward a biologically based dose-responsemodel for developmental toxicity of 5-fluorouracil in therat: a mathematical construct. Toxicol Sci 59:49–58.

Shapiro DJ, Barton MC, McKearin DM, Chang TC, Lew D,Blume J, et al. 1989. Estrogen regulation of gene tran-scription and mRNA stability. Recent Prog Horm Res45:29–58.

Simon Z, Vlad A, Michalas GI. 1988. Positive gene control andsteroid hormone receptor protein interactions.Mathematical model. Studia Biophysica 125:137–142.

Suk WA, Olden, K, Yang RSH. 2002. Chemical mixturesresearch: future perspectives. Environ Health Perspect110(suppl 6):891–892.

Systems Biology Workbench Development Group. 2002.Systems Biology Markup Language. Pasadena,CA:Caltech Erato Kitano Systems Biology WorkbenchDevelopment Group. Available: www.cds.caltech.edu/erato/ [accessed 20 Sept 2002].

Tata J, Baker B, Machuca I, Rabelo E, Yamauchi K. 1993.Autoinduction of nuclear receptor genes and its signifi-cance. J Steroid Biochem Mol Biol 46:105–119.

Tomita M, Hashimoto K, Takahashi K, Shimizu TS, Matsuzaki Y,Miyoshi F, et al. 1999. E-CELL: software environment forwhole-cell simulation. Bioinformatics 15:72–84.

Tritscher AM, Goldstein JA, Portier CJ, McCoy Z, Clark GC,Lucier GW. 1992. Dose-response relationships for chronicexposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin in a rattumor promotion model: quantification and immunolocal-ization of CYP1A1 and CYP1A2 in the liver. Cancer Res52:3436–3442.

Tyson JJ, Chen K, Novak B. 2001. Network dynamics and cellphysiology. Nat Rev Mol Cell Biol 2:908–916.

U.S. EPA. 1999. Guidelines for Carcinogen Risk Assessment.NCEA-F-0644. Washington, DC:U.S. EnvironmentalProtection Agency.

———. 2000. Integrated Summary and Risk Characterizationfor 2,3,7,8-Tetrachlorodibenzo-p-Dioxin (TCDD) andRelated Compounds. Part II: Health Assessment for2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD) and RelatedCompounds. Chapter 8. Dose-response Modeling for2,3,7,8-TCDD. SAB Review Draft. Report NCEA-I-0835.Washington, DC:U.S. Environmental Protection Agency.

Vanden Heuvel JP, Clark GC, Kohn MC, Tritscher AM,Greenlee WF, Lucier GW, et al. 1994. Dioxin-responsivegenes: examination of dose-response relationships usingquantitative reverse transcriptase-polymerase chainreaction. Cancer Res 54:62–68.

Vohradsky J. 2001. Neural network model of gene expression.FASEB J 15:846–854.

Waxman DJ. 1999. P450 gene induction by structurally diverse



xenochemicals: central role of nuclear receptors CAR,PXR, and PPAR. Arch Biochem Biophys 369:11–23.

Weghorst CM, Devor DE, Henneman JR, Ward JM. 1994.Promotion of hepatocellular foci and adenomas by di(2-ethylhexyl) phthalate and phenobarbital in C3H/HeNCrmice following exposure to N-nitrosodiethylamine at 15days of age. Exp Toxicol Pathol 45:423–431.

Weng G, Bhalla US, Iyengar R. 1999. Complexity in biologicalsignaling systems. Science 284:92–96.

Wilson CL, Safe S. 1998. Mechanisms of ligand-induced arylhydrocarbon receptor-mediated biochemical and toxicresponses. Toxicol Pathol 26:657–671.

Wilson MA, Bhalla US, Uhley JD, Bower JM. 1989. GENESIS: asystem for simulating neural networks. In: Advances in

Neural Information Processing Systems (Touretzky D, ed).San Mateo, CA:Morgan Kaufmann, 485–492.

Xu W, Chen H, Du K, Asahara H, Tini M, Emerson BM, et al.2001. A transcriptional switch mediated by cofactormethylation. Science 294:2507–2511.



molecular circuits, biological switches, and nonlinear dose

Documents