Environ. Sci. Technol. 2005, 39, 5363-5371
Biochemical Reaction Network Modeling: Predicting Metabolism of Organic Chemical Mixtures ARTHUR N. MAYENO, RAYMOND S. H. YANG, AND BRAD REISFELD* Quantitative and Computational Toxicology Group, Department of Environmental and Radiological Health Sciences, Colorado State University, Foothills Campus, Fort Collins, Colorado 80523-1690
All organisms are exposed to multiple xenobiotics, through food, environmental contamination, and drugs. These xenobiotics often undergo biotransformation, a complex process that plays a critical role in xenobiotic elimination or bioactivation to toxic metabolites. Here we describe the results of a new computer-based simulation tool that predicts metabolites from exposure to multiple chemicals and interconnects their metabolic pathways, using four common drinking water pollutants (trichloroethylene, perchloroethylene, methylchloroform, and chloroform) as a test case. The simulation tool interconnected the metabolic pathways for these compounds, predicted reactive intermediates, such as epoxides and acid chlorides, and uncovered points in the metabolic pathways where typical endogenous compounds, such as glutathione or carbon dioxide, are consumed or generated. Moreover, novel metabolites, not previously reported, were predicted via this methodology. Metabolite prediction is based on a reactionmechanism-based methodology, which applies fundamental organic and enzyme chemistry. The tool can be used to (a) complement experimental studies of chemical mixtures, (b) aid in risk assessment, and (c) help understand the effects of complex chemical mixtures. Our results indicate that this tool is useful for predictive xenobiotic metabolomics, providing new and important insights into metabolites and the interrelationship between diverse chemicals that hitherto may have remained unnoticed.
Introduction All living organisms are continuously exposed to a myriad of xenobiotics, ranging from environmental pollutants and pesticides to household chemicals and drugs. Biotransformation is a biological process whereby lipophilic xenobiotics are usually detoxified through conversion to more soluble forms that facilitate excretion (1). However, this process can also bioactivate a xenobiotic, changing it to a more harmful form (1). Bioactivation can be significant in human health, as exemplified by the activation of aflatoxin B1 (2) or vinyl chloride (3) to carcinogenic epoxide metabolites or acetaminophen to the hepatotoxic quinoneimine (1) by cytochrome P450 (CYP) enzymes (4, 5). Thus, biotransformation is important in toxicology, environmental sciences, ecotoxicology, and pharmaceutical research, affecting toxicity, * Corresponding author phone: (970) 491-1019; fax: (970) 4918304; e-mail:
[email protected]. 10.1021/es0479991 CCC: $30.25 Published on Web 06/11/2005
2005 American Chemical Society
bioavailability, and efficacy and converting prodrugs into active drugs. Biotransformation of a single xenobiotic can be complex, occurring in several tissues, involving numerous enzymes, and yielding several metabolites (1). For multiple-chemical exposure, the biotransformation pathways and interactions are even more complicated: different chemicals may produce the same metabolites, creating interconnected metabolic pathways; exposure to mixtures may lead to metabolic inhibition, induction of metabolizing enzymes, or other effects, not observed through single-chemical exposure (6). Moreover, exposure to multiple chemicals occurs under various scenarios (e.g., simultaneous or sequential, acute or chronic). In light of the complexity of the biotransformation of xenobiotic mixtures in biological systems, the creation and application of computer-assisted tools to predict and simulate metabolism of xenobiotic mixtures is indispensable in understanding the possible health risk effects and the underlying mode of action. Chlorinated hydrocarbons are among the most common environmental pollutants (7). This class of chemicals includes the chlorinated solvents trichloroethylene (Cl2CdCHCl, TCE), perchloroethylene (Cl2CdCCl2, Perc), methylchloroform (CH3CCl3, MC), and chloroform (CHCl3), all of which have been used extensively in a variety of industrial applications (8). Due to their wide use, volatility, and resistance to degradation, these volatile organic chemicals (VOCs) are widely distributed in the environment as pollutants and are common contaminants at many chemical waste sites (8). As reported by the ATSDR (Agency for Toxic Substances and Disease Registry), CHCl3 and TCE are listed among the top 20 priority chemicals, while Perc and MC are also high on the 2003 CERCLA (Comprehensive Environmental Response, Compensation, and Liability Act) Priority List of Hazardous Substances (8). TCE is the chemical most commonly detected at Superfund sites and one of the most common groundwater contaminants (7). Furthermore, these chlorinated chemicals are likely to coexist in groundwater and/or drinking water as a result of environmental pollution (9, 10). As a result, it is likely that the general population is exposed to low levels of these chemicals. Occupational exposure among those who work at facilities that use these solvents is likely much higher. Each of these four VOCs can exert harmful health effects, and TCE, Perc, and CHCl3 have been shown to be carcinogenic in animal studies (11-13). Reactive metabolites of these compounds, such as epoxides and acid chlorides (Figure 1), are believed to be responsible, in large part, for the toxicity and carcinogenicity of these chemicals (11-13). Thus, the ability to accurately predict these metabolites, including their chemical and metabolic interactions, is crucial in assessing the human health risk associated with a chemical or chemical mixture. The complexity and significance of multiple chemical interactions can be illustrated using the metabolism of TCE in the presence of the other VOCs. TCE is metabolized via two principal pathways (11), CYP2E1 and glutathione Stransferase (GST) (Figure 1a), with CYP2E1 ∼3000-fold more efficient than GST (14, 15). However, as CYP2E1 is a “highaffinity” but “low-capacity” enzyme for TCE (11), exposure to high levels of multiple VOCs overwhelms the metabolic capacity of CYP2E1, and TCE metabolism is shunted toward the alternative GST pathway. The GST pathway leads to a reactive thioketene metabolite in the kidney, and renal toxicity, including renal cancer, is possible. Similar pathway complexities are present for the metabolism of MC, Perc, and CHCl3 (Figure 1b-d). These examples demonstrate that, VOL. 39, NO. 14, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
5363
FIGURE 1. Metabolic pathways of (a) trichloroethylene (11), (b) methylchloroform (8, 60), (c) perchloroethylene (12), and (d) chloroform (8, 61). when multiple metabolites from a single chemical are possible or, more relevantly, when multiple metabolites from a mixture of chemicals are possible, the degree of bioactivation or detoxification depends on which metabolic pathways predominate; specifically, the types of metabolites that are produced as well as their rates of formation and degradation are critical factors in determining chemical toxicity. Although in vivo and in vitro experimentation has been employed successfully to study metabolism and toxicology, these approaches have intrinsic limitations (e.g., detection limits toward short-lived or trace metabolites, variability among individuals and species, and, in the case of mixtures, high cost and resource requirements due to the large number of chemical combinations and conditions). Thus, computerassisted modeling tools that can predict and describe the 5364
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 14, 2005
quantitative, spatial, and temporal relationships of exogenous and endogenous chemicals and their metabolites are extremely useful. These tools will be the foundation for “predictive metabolomics”, which can provide scientific insight into metabolic processes and may aid in efficiently designing targeted experiments. For metabolite prediction, a variety of software tools are now available (16-22). These packages can be classified into two major categories (16), “expert systems” (application of fundamental biochemical transformation rules via an expert system or heuristics) and “database systems” (extraction of relevant information from a large database of experimentally verified biotransformations), or a combination of these approaches. Each of these packages has its strengths and limitations, with a common limitation the inability to handle mixtures or incorporate systems biology, which integrates
FIGURE 2. Information flow through BioTRaNS. computer technology with experimental biology to examine biological systems (23). To fill this important niche, we have been developing and applying a systems biology approach called “biochemical reaction network modeling”. Biochemical reaction network modeling is an outgrowth of biochemistry and reaction network modeling that was historically used in chemical and petroleum engineering (24, 25) and more recently in a limited context for biological applications (26, 27). Biochemical reaction network modeling is an in silico approach that can predict the chemical structures of the metabolites of one chemical or a mixture of chemicals, generate the metabolic pathways, and interconnect these pathways through metabolites in common. Moreover, once validated with experimental data, this methodology can predict and simulate in a quantitative and time-dependent manner the formation and disappearance of all metabolites, including potentially toxic reactive intermediates, such as epoxides. Thus, it can aid in predicting metabolism and toxicity, and in understanding the modes of action of individual chemicals as well as chemical mixtures. As our first test case, our biochemical reaction network model focuses on the biotransformation of a mixture of TCE, Perc, MC, and CHCl3. The empirically based metabolic pathways for each of the four test chemicals, individually, is shown in Figure 1. Although extensively studied as individual chemicals, few studies have examined mixtures of these chemicals (15, 28).
Methodology Structure of the Biochemical Reaction Network Modeling Framework. In silico prediction of metabolites and generation of the biotransformation pathways were accomplished using a biochemical reaction network simulation tool called BioTRaNS (biochemical tool for reaction network simulation), which we have been developing and refining in house. The conceptual flow of information through BioTRaNS is shown in Figure 2. A “chemical mixture” (concentration of a single chemical or multiple chemicals) and “reaction environment” (types and amounts of enzymes and other “agents”) are input by the user. The chemical mixture is converted to a set of “virtual molecules” (vMols), each with its own specifications (i.e., a canonical SMILES and list of computed geometric, energetic, and physicochemical properties). SMILES (simplified molecular input line entry specification) is a general-purpose chemical nomenclature and data-exchange format used in computational chemistry (29). In general, properties are retrieved from associated databases when available, or are computed as needed. The reaction environment is translated into a set of “virtual enzymes” (vEnzs), representing the actual enzymes, and
“virtual agents” (vAgnts), representing nonenzymatic reactions. Each vEnz is endowed with (i) an appropriate binding calculator and (ii) “transforms” governing the biotransformations that the vEnz can mediate. The binding calculator computes the feasibility of a particular vMol binding to a vEnz, on the basis of quantitative structure-activity relationships (or decision tree) derived from properties of known substrates for each enzyme. Transformations are stored as a list of SMIRKS-based representations (30) (see Table 1 for examples). SMIRKS (SMILES reaction specification) is a superset of the SMILES language that describes chemical transformations. For any transformation, more than one SMIRKS can be written, depending on the specificity desired. A single SMIRKS (or transform) can perform a general reaction, such as converting any primary alcohol to an aldehyde. Moreover, our methodology is unique in implementing transforms that describe steps of a reaction mechanism, including those for enzyme-mediated reactions (31). The use of mechanistic steps is key to automatically and accurately predicting metabolites (vide infra for an example). After the vMol is created, the binding calculator for each vEnz assesses binding feasibility; if feasible, the vMol becomes an eligible reactant. All eligible reactants then undergo appropriate virtual biotransformations, creating specific chemical reactions and converting the vMol into one or more metabolites, which also are checked for reaction feasibility. The information on the metabolites and the associated transform/agent interconnecting each substrate-product pair is converted to a “virtual reaction” (vRxn). Thus, each vRxn comprises the vMols and vEnzs/vAgnts from which it was derived, as well as appropriate kinetic properties. The ability to delineate the metabolites and visualize where pathways intersect is key in gaining new insight into interrelationships among chemicals and their metabolic pathways. With this concept in mind, the aggregate information in the vRxns is used to construct a reaction network which captures these interrelationships. This network is generated in the form of a mathematical graph (32), wherein the metabolites constitute the vertexes or nodes, and the reactions form the edges. Well-described in discrete mathematics, graphs can be efficiently analyzed using algorithms of computer science (33); graph theory allows efficient determination of characteristics such as common reactants between various pathways, the shortest path to a given metabolite, and common (or all) routes to given metabolites. These network graphs can be depicted at different levels of detail, from highly detailed, showing steps of a reaction mechanism, to overview, showing only key metabolites. Simultaneous to the graph creation, kinetic properties in each vRxn are used to create the appropriate reaction rate equations (ordinary differential equations, ODEs). These properties include (i) rate constants (e.g., the Michaelis constant, Km, and maximum velocity, Vmax, for enzymecatalyzed reactions and k for nonenzymatic reactions), (ii) inhibitor constants, Ki, and (iii) modes of inhibition or “allosterism” (34-36). The total set of rate equations and specified initial conditions forms an initial value problem that is solved by a stiff ODE solver for the concentrations of all species as a function of time. The essential output of BioTRaNS is a list of metabolites, their interconnections (through biotransformations), and their time-course concentrations. This output can be expressed to the user in many forms, including graphs of the metabolic pathways, enumerations of all (or certain specified types of) metabolites and/or reactions, and plots of concentration versus time for species of interest. Virtual Enzymes and Other Virtual Agents. The constituent transforms for each virtual enzyme were compiled by carefully culling the literature for data on enzymes known to act on the VOCs and VOC metabolites. Substrate feasibility VOL. 39, NO. 14, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
5365
TABLE 1. Examples of Transformsa
a
For any transformation, more than one SMIRKS can be written, depending on the specificity desired.
is based on properties of chemicals known to be substrates for each enzyme. Biochemical Reaction Network Simulation Tool Platform and Supporting Software Packages. BioTRaNS is being developed and executed on the Linux platform, with Python (37) as the principal language for coding the framework. Daylight programmer’s toolkits (30) were used to manipulate SMILES and carry out SMILES transformations. Open Babel (38), MOPAC (39) (public domain version), and MySQL (40) were used for translations between chemical information formats, for quantum chemical calculations, and for data storage and retrieval, respectively. CORINA (41) was used to predict three-dimensional molecular configurations. Whenever possible, programming tools, code, and components falling under one of the Open Source approved licenses were chosen (42), which allow users to read, redistribute, and modify the source code for the software. Metabolism Predictions. For simulation of biotransformation, a single compartment model was used in this work, primarily for clarity, and the actual tissues where metabolism occurs have not been specified or modeled. For example, although oxidative metabolism for the chemicals described herein occurs primarily in the liver, other metabolic steps, such as those involving β-lyase activities, are known to occur in the kidneys (11, 12, 43). In all of the simulations presented herein, both enzymatic (vEnzs) and non-enzyme-catalyzed (vAgnts) transforms were employed to generate metabolite pathways; however, only limited sets of vEnzs and vAgnts, and constituent transforms, were invoked, as described in the text. Though some of the metabolites predicted by our simulation are likely to be formed in small amounts or only under certain conditions, these have been included in the figures to illustrate the predictive ability of BioTRaNS.
Results and Discussion The BioTRaNS-generated biotransformation pathway of TCE is shown in Figure 3a, in the form of a graph, using selected vEnzs (e.g., CYP2E1, alcohol dehydrogenase, aldehyde dehydrogenase, and aldehyde oxidase) and vAgnts (e.g., “NIHshift,” acid chloride hydrolysis, and dehydrohalogenation of gem-halohydrins). The constituent transforms for each vEnz and vAgnt were based on the literature for each step. Only certain vEnzs and constituent transforms were invoked to keep the graph simple (i.e., at a “high” or “overview” level). On the basis of these agents, all of the major metabolites, such as choral (Cl3CCHO) and its hydrate form (Cl3CCH5366
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 14, 2005
(OH)2), trichloroethanol (Cl3CCH2OH), trichloroethanol glucuronide (Cl3CCH2OC6H9O6), and trichloroacetic acid (Cl3CCO2H, TCA), were predicted, as well as minor metabolites, such as dichloroacetic acid (Cl2CHCO2H) and oxalic acid (HO2CCO2H). Metabolites generated via the GST pathway, such as S-(1,2-dichlorovinyl)glutathione and N-acetyl-S-(1,2dichlorovinyl)-L-cysteine, were also predicted. Reactive species, such as TCE oxide and a thioketene (SdCdCHCl), were generated as well. Because only a limited set of agents and constituent transforms were used, Figure 3a shows a simplified depiction of all of the possible metabolites. For instance, more than one isomer of the glutathione adduct with TCE is known to be formed in vivo (14), and in aqueous solution, TCE oxide decomposes to carbon monoxide, formic acid, dichloroacetic acid, and glyoxylic acid (44). The vAgnts that describe all of the TCE oxide degradation pathways and the constituent transforms of the vEnz representing GST that describe the formation of other glutathione adducts with TCE were not invoked. Moreover, in Figure 3, TCE oxide is depicted as directly converting to chloral, although experimental evidence indicates that this epoxide does not rearrange to chloral in aqueous solution (44, 45); this transformation was implemented as a simplification, and more accurate pathway predictions, based on the empirically based, mechanistic pathways, are presented in Figure 4. For each vEnz, substratefeasibility conditions facilitated more accurate prediction of metabolites and were based on the molecular weight cutoff, log P, and other computationally derived parameters and user-stipulated. Without conditions, all possible metabolites based on the transforms (and thus pathways) will be generated, which can, in fact, be advantageous in expanding the current understanding of possible metabolites. The BioTRaNS-generated “high-level” biotransformation reaction network of the four VOCs (TCE, Perc, MC, and CHCl3) as a mixture is shown in Figure 3b. Again, only a limited number of agents were used to keep the graph simple. As predicted and rendered by the computer, all four pathways are interconnected, with large portions of the metabolic pathways overlapping. TCA is the intermediate common to all pathways. The BioTRaNS-generated, combined pathways are in good agreement with the data found in the literature for the metabolism of each of the individual VOCs (cf. Figure 1), and when these pathways are manually integrated together. The connection between the first three VOCs (TCE, Perc, and MC) and CHCl3 was predicted from the transform
VOL. 39, NO. 14, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9
FIGURE 3. BioTRaNS-generated biotransformation pathways for (a) TCE and (b) a mixture of TCE, Perc, MC, and CHCl3. For clarity, only selected agents (vEnzs and vAgnts) were used to generate these graphs. For (b), reactive metabolites are highlighted as follows: epoxides (brown, box, dashed); acid chlorides (orange, box, solid); thioketene (turquoise, box, solid); starting chemicals (blue, ellipse, solid).
5367
5368 9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 14, 2005
FIGURE 4. BioTRaNS-generated mechanism-based pathways for (a) the oxidation of TCE by CYP and (b) more comprehensive biotransformation pathways of TCE, Perc, MC, and CHCl3. Chemicals are highlighted as follows: starting chemicals (blue, ellipse, solid); epoxides (brown, box, dashed); acid chlorides (orange, box, solid); radicals (red, octagon, solid); Fe-O-substrate complexes (magenta, ellipse, dashed); carbocations (salmon, ellipse, dashed).
describing the decarboxylation of trihaloacetic acids. The conversion of TCA to CHCl3 is well documented in the chemical literature (46), although the authors are not aware of any studies demonstrating the occurrence of this reaction within mammalian systems. As the reaction is slow in vitro at acidic or neutral pH (46), it is presumably slow under physiological conditions; however, even the generation of minute amounts of a chemical may have important consequences, especially for processes such as carcinogenesis. In compartments within a cell that are more alkaline, decarboxylation would likely proceed faster. Biochemical reaction network modeling allows scientists to “see” connections between pathways that hitherto may not have been considered. Moreover, any species of interest, such as epoxides, as well as unstable intermediates, can be easily highlighted within the graph. In Figure 3b, epoxides, acid chlorides, and thioketenes are highlighted with various enclosing shapes and colors. Thus, with BioTRaNS, an investigator can examine the nature of species of interest and, in the context of health risks, easily locate highly reactive species. An innovative feature of BioTRaNS is the use of enzymeand reaction-mechanism-based transformations. Figure 4a shows the BioTRaNS-generated stepwise oxidation of TCE by CYP2E1. The figure was generated by BioTRaNS, on the basis of transforms representing each step of the proposed reaction mechanism (45). The first step involves formation of an intermediate between the high-valent iron-oxo complex, (FeO)3+, of the CYP heme (47) and the alkene, forming a carbocationic intermediate (FeIII-O-C-C+), which has been proposed to explain the 1,2-shifts (of H and Cl) leading to the observed products (3). Subsequent steps were performed by a transform representing ring closure to the epoxide and by transforms representing 1,2-shifts of H or Cl to give the rearranged product. Although a radical mechanism has also been hypothesized for the CYP-mediated oxidation of TCE, 1,2-shifts of hydrogen are not energetically favored due to the resulting three-electron transition state (31, 48). 1,2-Shifts of hydrogen to a carbocation, however, are energetically favorable. Nevertheless, transforms representing a radical mechanism, as well as those involving different oxidizing species (e.g., hydroperoxo-iron species (49, 50)), can also be prescribed and depicted. Thus, when the literature is unclear about a mechanism or if a reaction proceeds via different mechanisms under different conditions, alternative reaction mechanisms can be applied. The ability to predict products on the basis of reaction mechanisms should yield more accurate results, as species that are energetically unfavorable can be eliminated by setting an appropriate condition (e.g., transition-state energy threshold). Interestingly, with the inclusion of transforms describing hydrogen abstraction followed by “oxygen rebound” (47) (a proposed mechanism for hydroxylation of alkanes), BioTRaNS predicts the formation of 2,2,3-trichloro-3-hydroxyoxirane and other downstream intermediates not described in TCE metabolism literature. Further experimental research, beyond the scope of this current paper, is required to determine if these BioTRaNS-predicted, reactive intermediates are actually formed, but this example illustrates the potential predictive power of BioTRaNS. A more complete depiction of the interconnected biotransformation pathways of all four VOCs is presented in Figure 4b. The complexity of the pathways is clearly illustrated, despite using a noncomprehensive list of agents. [The graph can be greatly simplified by converting the output from detailed to overview (Figure 3b)]. Another useful feature is the ability to see how and where endogenous compounds (e.g., glutathione, oxalic acid, and carbon dioxide) are consumed or released. Glutathione is an important cellular antioxidant, and its depletion is known to adversely affect cell function (51, 52). The formation of other byproducts,
such as HCl, can also be depicted. Though reaction networks may contain thousands of reactions and chemicals, an investigator can easily interrogate the underlying graph data structure for information of interest. The use of highlighting provides immediate visual feedback on species of interest, even in highly complex networks. BioTRaNS continues to undergo development. The current version of BioTRaNS can calculate quantitative timecourse information on the intermediates, using user-input kinetic data. In future versions, we envision estimating reaction rates, regioselectivity, and stereoselectivity (thus metabolite distribution), by coupling each SMIRKS-generated reaction step with data, such as energies of transition states, calculated using computational chemistry (ab initio or semiempirical) approaches. Through such kinetic analyses, it should be possible to predict the major and minor pathways under different biological conditions (e.g., various CYP isozyme levels). Recent advances, such as the reporting of the crystal structures of human CYP2C8 (53), CYP2C9 (54), and CYP3A4 (55), will facilitate these efforts. In addition, BioTRaNS’s ability to predict metabolite profiles may aid in the design of “targeted” experiments. For instance, BioTRaNS may generate different product profiles when different reaction mechanisms are implemented. Thus, focused experiments can be devised to test each mechanism, especially if an unusual or unexpected metabolite is predicted and whose presence may have been overlooked if not targeted. For example, 1,2-shift products are predicted by transforms describing a cationic mechanism but not by transforms describing radical reactions. At present, our database of vEnzs represents only a fraction of phase I and phase II enzymes, and is being expanded. Although this paper focused on small, chlorinated VOCs, this methodology can be expanded to other classes of environmental contaminants, including polychlorinated biphenyls (PCBs) and polycyclic aromatic hydrocarbons. When linked to physiologically based pharmacokinetic (56) (PBPK) models (which can model biological processes such as diffusion, transport, and enzyme induction), xenobiotic metabolism can be predicted and quantitatively simulated for an entire organism, a characteristic of systems biology. Cumulative risk assessment has gained widespread interest in toxicology and human health risk assessment following the enactment of the Food Quality Protection Act of 1996 (57-59). Under Congressional mandate, the U.S. Environmental Protection Agency must develop methodologies to assess the risk to health resulting from exposure to multiple chemicals from a variety of sources and routes of exposure. The significance of this area is clear when one considers that individuals are typically exposed to numerous chemicals simultaneously and sequentially, and exposure to multiple chemicals can exert health effects different from exposure to an individual chemical. The linkage of PBPK and biochemical reaction network modeling would provide a powerful tool for the accurate prediction of pharmacokinetics and biotransformation of chemical mixtures in cumulative risk assessment. For instance, if the production of a certain reactive intermediate, such as a specific epoxide, is believed to mediate toxicity, then BioTRaNS can be used to screen for all chemicals that generate this epoxide. Here, we have demonstrated some applications of BioTRaNS, using four VOCs, by predicting novel metabolites, interconnecting metabolic pathways through metabolites in common, and identifying points in the metabolic pathways where endogenous chemicals can be consumed or generated. This work represents the early stages of development of this simulation tool and just the beginning of its potential applications. As with any in silico tool, experimental validation of model results is essential. These studies are currently under way. VOL. 39, NO. 14, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
5369
Acknowledgments The work was supported by NIEHS Grants ES11146-02, ES09655, and T32 ES07321 and CDC/NIOSH Grant OH0755601. We thank J. Lyle for technical assistance.
Supporting Information Available Each of the BioTRaNS-generated metabolic pathways (Figures 3 and 4). This material is available free of charge via the Internet at http://pubs.acs.org.
Literature Cited (1) Parkinson, A. In Casarett and Doull’s Toxicology: The Basic Science of Poisons, 6th ed.; Klaassen, C. D., Ed.; McGraw-Hill: New York, 2001; pp 133-224. (2) Guengerich, F. P.; Johnson, W. W.; Shimada, T.; Ueng, Y. F.; Yamazaki, H.; Langouet, S. Activation and detoxication of aflatoxin B-1. Mutat. Res.sFundam. Mol. Mech. Mutagen. 1998, 402, 121-128. (3) Guengerich, F. P. Cytochrome P450 oxidations in the generation of reactive electrophiles: epoxidation and related reactions. Arch. Biochem. Biophys. 2003, 409, 59-71. (4) Ortiz de Montellano, P. R., Ed.; Cytochrome P450: Structure, Mechanism, and Biochemistry, 2nd ed.; Plenum Press: New York, 1995. (5) Lewis, D. F. V. Guide to Cytochromes P450: Structure and Function; Taylor & Francis: London, 2001. (6) Andersen, M. E.; Dennison, J. E. Mechanistic approaches for mixture risk assessmentssPresent capabilities with simple mixtures and future directions. Environ. Toxicol. Pharmacol. 2004, 16, 1-11. (7) Abelson, P. H. Volatile Contaminants of Drinking-Water. Science 1990, 247, 141-141. (8) Agency for Toxic Substances and Disease Registry. 2003 CERCLA Priority List Hazardous Substances. http://www.atsdr.cdc.gov/ clist.html. (9) Squillace, P. J.; Moran, M. J.; Lapham, W. W.; Price, C. V.; Clawges, R. M.; Zogorski, J. S. Volatile organic compounds in untreated ambient groundwater of the United States, 1985-1995. Environ. Sci. Technol. 1999, 33, 4176-4187. (10) Williams, P.; Benton, L.; Warmerdam, J.; Sheehan, P. Comparative risk analysis of six volatile organic compounds in California drinking water. Environ. Sci. Technol. 2002, 36, 4721-4728. (11) Lash, L. H.; Fisher, J. W.; Lipscomb, J. C.; Parker, J. C. Metabolism of trichloroethylene. Environ. Health Perspect. 2000, 108 (Suppl. 2), 177-200. (12) Lash, L. H.; Parker, J. C. Hepatic and renal toxicities associated with perchloroethylene. Pharmacol. Rev. 2001, 53, 177-208. (13) Fabrizi, L.; Taylor, G. W.; Canas, B.; Boobis, A. R.; Edwards, R. J. Adduction of the chloroform metabolite phosgene to lysine residues of human histone H2B. Chem. Res. Toxicol. 2003, 16, 266-275. (14) Bloemen, L. J.; Monster, A. C.; Kezic, S.; Commandeur, J. N.; Veulemans, H.; Vermeulen, N. P.; Wilmer, J. W. Study on the cytochrome P-450- and glutathione-dependent biotransformation of trichloroethylene in humans. Int. Arch. Occup. Environ. Health 2001, 74, 102-108. (15) Dobrev, I. D.; Andersen, M. E.; Yang, R. S. H. In silico toxicology: simulating interaction thresholds for human exposure to mixtures of trichloroethylene, tetrachloroethylene, and 1,1,1trichloroethane. Environ. Health Perspect. 2002, 110, 10311039. (16) Langowski, J.; Long, A. Computer systems for the prediction of xenobiotic metabolism. Adv. Drug Delivery Rev. 2002, 54, 407415. (17) AKos Consulting and Solutions GmbH. PASS Metabolite. http:// akosgmbh.de/pass/pass_metabolite.htm, 2003. (18) Borodina, Y.; Sadym, A.; Filimonov, D.; Blinova, V.; Dmitriev, A.; Poroikov, V. Predicting biotransformation potential from molecular structure. J. Chem. Inf. Comput. Sci. 2003, 43, 16361646. (19) Mekenyan, O. G.; Dimitrov, S. D.; Pavlov, T. S.; Veith, G. D. A systematic approach to simulating metabolism in computational toxicology. I. The TIMES heuristic modelling framework. Curr. Pharm. Des. 2004, 10, 1273-1293. (20) GeneGo. MetaDrug. www.genego.com, 2004. (21) Csizmadia, F. JChem: Java applets and modules supporting chemical database handling from web browsers. J. Chem. Inf. Comput. Sci. 2000, 40, 323-324. 5370
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 14, 2005
(22) Bruker Daltonics. Metabolitetools. http://www.brukerdaltonik.de/metabolite_toolbox.html, 2003. (23) Kitano, H. Systems biology: a brief overview. Science 2002, 295, 1662-1664. (24) Watson, B. A.; Klein, M. T.; Harding, R. H. Catalytic cracking of alkylbenzenes: Modeling the reaction pathways and mechanisms. Appl. Catal., A 1997, 160, 13-39. (25) Kumar, A.; Campbell, D. M.; Klein, M. T. Computer-assisted kinetic modeling: Interfacing structure and reaction network model builders. Abstr. Pap.sAm. Chem. Soc. 1997, 214, 82PETR, Part 2. (26) Liao, K. H.; Dobrev, I. D.; Dennison, J. E.; Andersen, M. E.; Reisfeld, B.; Reardon, K. F.; Campain, J. A.; Wei, W.; Klein, M. T.; Quann, R. J.; Yang, R. S. H. Application of biologically based computer modeling to simple or complex mixtures. Environ. Health Perspect. 2002, 110 (Suppl. 6), 957-963. (27) Klein, M. T.; Hou, G.; Quann, R. J.; Wei, W.; Liao, K. H.; Yang, R. S. H.; Campain, J. A.; Mazurek, M. A.; Broadbelt, L. J. BioMOL: a computer-assisted biological modeling tool for complex chemical mixtures and biological processes at the molecular level. Environ. Health Perspect. 2002, 110 (Suppl. 6), 1025-1029. (28) Dobrev, I. D.; Andersen, M. E.; Yang, R. S. H. Assessing interaction thresholds for trichloroethylene in combination with tetrachloroethylene and 1,1,1-trichloroethane using gas uptake studies and PBPK modeling. Arch. Toxicol. 2001, 75, 134-144. (29) Weininger, D. SMILES, a Chemical Language and InformationSystem. 1. Introduction to Methodology and Encoding Rules. J. Chem. Inf. Comput. Sci. 1988, 28, 31-36. (30) Daylight Chemical Information Systems Inc. Daylight Toolkits, version 4.82. www.daylight.com. (31) Silverman, R. B. Organic Chemistry of Enzyme-Catalyzed Reactions; revised ed.; Academic Press: London, 2002. (32) Balaban, A. T. Chemical Applications of Graph Theory; Academic Press: New York, 1976. (33) Gross, J. L.; Yellen, J. Graph Theory and Its Applications; CRC Press: Boca Raton, FL, 1999. (34) Hlavica, P.; Lewis, D. F. V. Allosteric phenomena in cytochrome P450-catalyzed monooxygenations. Eur. J. Biochem. 2001, 268, 4817-4832. (35) Atkins, W. M.; Wang, R. W.; Lu, A. Y. H. Allosteric behavior in cytochrome P450-dependent in vitro drug-drug interactions: A prospective based on conformational dynamics. Chem. Res. Toxicol. 2001, 14, 338-347. (36) Yoon, M. Y.; Campbell, A. P.; Atkins, W. M. “Allosterism” in the elementary steps of the cytochrome P450 reaction cycle. Drug Metab. Rev. 2004, 36, 219-230. (37) Python Software Foundation. Python, version 2.3. http:// www.python.org. (38) Stahl, M. Open Babel, version 1.100.1. http://openbabel. sourceforge.net/. (39) Stewart, J. J. P. MOPAC Reference Manual and Release Notes, 6th ed.; Frank Seller Research Laboratory: U.S. Air Force Academy, CO, 1990. (40) MySQL AB. MySQL, version 3.23. www.mysql.com. (41) Schwab, C. CORINA, version 2.63. http://www2.chemie. uni-erlangen.de/software/corina/index.html. (42) Open Source Initiative. Open SourcesThe Approved Licenses. http://www.opensource.org/licenses/, 2003. (43) Goeptar, A. R.; Commandeur, J. N.; van Ommen, B.; van Bladeren, P. J.; Vermeulen, N. P. Metabolism and kinetics of trichloroethylene in relation to toxicity and carcinogenicity. Relevance of the mercapturic acid pathway. Chem. Res. Toxicol. 1995, 8, 3-21. (44) Cai, H. L.; Guengerich, F. P. Mechanism of aqueous decomposition of trichloroethylene oxide. J. Am. Chem. Soc. 1999, 121, 11656-11663. (45) Miller, R. E.; Guengerich, F. P. Oxidation of trichloroethylene by liver microsomal cytochrome P-450: evidence for chlorine migration in a transition state not involving trichloroethylene oxide. Biochemistry 1982, 21, 1090-1097. (46) Urbansky, E. T. The fate of the haloacetates in drinking waters Chemical kinetics in aqueous solution. Chem. Rev. 2001, 101, 3233-3243. (47) Guengerich, F. P. Common and uncommon cytochrome P450 reactions related to metabolism and chemical toxicity. Chem. Res. Toxicol. 2001, 14, 611-650. (48) Brooks, M. A.; Scott, L. T. 1,2-shifts of hydrogen atoms in aryl radicals. J. Am. Chem. Soc. 1999, 121, 5444-5449. (49) Coon, M. J. Multiple oxidants and multiple mechanisms in cytochrome P450 catalysis. Biochem. Biophys. Res. Commun. 2003, 312, 163-168.
(50) Newcomb, M.; Hollenberg, P. F.; Coon, M. J. Multiple mechanisms and multiple oxidants in P450-catalyzed hydroxylations. Arch. Biochem. Biophys. 2003, 409, 72-79. (51) Townsend, D. M.; Tew, K. D.; Tapiero, H. The importance of glutathione in human disease. Biomed. Pharmacother. 2003, 57, 145-155. (52) Dringen, R.; Hirrlinger, J. Glutathione pathways in the brain. Biol. Chem. 2003, 384, 505-516. (53) Schoch, G. A.; Yano, J. K.; Wester, M. R.; Griffin, K. J.; Stout, C. D.; Johnson, E. F. Structure of human microsomal cytochrome P4502C8sEvidence for a peripheral fatty acid binding site. J. Biol. Chem. 2004, 279, 9497-9503. (54) Williams, P. A.; Cosme, J.; Ward, A.; Angove, H. C.; Matak Vinkoviaˆc, D.; Jhoti, H. Crystal structure of human cytochrome P450 2C9 with bound warfarin. Nature 2003, 424, 464-468. (55) Williams, P. A.; Cosme, J.; Vinkovic, D. M.; Ward, A.; Angove, H. C.; Day, P. J.; Vonrhein, C.; Tickle, I. J.; Jhoti, H. Crystal structures of human cytochrome P450 3A4 bound to metyrapone and progesterone. Science 2004, 305, 683-686. (56) Krishnan, K.; Andersen, M. E. In Principles and Methods of Toxicology, 3rd ed.; Hayes, A. W., Ed.; Raven Press: New York, 1994; pp 149-188.
(57) Sielken, R. L. Risk metrics and cumulative risk assessment methodology for the FQPA. Regul. Toxicol. Pharmacol. 2000, 31, 300-307. (58) Fenner-Crisp, P. A. FQPA science issues: Common mechanism of toxicity and cumulative risk assessment. Regul. Toxicol. Pharmacol. 2000, 31, 308-310. (59) Wilkinson, C. F.; Christoph, G. R.; Julien, E.; Kelley, J. M.; Kronenberg, J.; McCarthy, J.; Reiss, R. Assessing the risks of exposures to multiple chemicals with a common mechanism of toxicity: How to cumulate? Regul. Toxicol. Pharmacol. 2000, 31, 30-43. (60) Nolan, R. J.; Freshour, N. L.; Rick, D. L.; McCarty, L. P.; Saunders: J. H. Kinetics and Metabolism of Inhaled Methyl Chloroform (1,1,1-Trichloroethane) in Male-Volunteers. Fundam. Appl. Toxicol. 1984, 4, 654-662. (61) Gemma, S.; Vittozzi, L.; Testai, E. Metabolism of chloroform in the human liver and identification of the competent P450s. Drug Metab. Dispos. 2003, 31, 266-274.
Received for review December 18, 2004. Revised manuscript received April 18, 2005. Accepted May 3, 2005. ES0479991
VOL. 39, NO. 14, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
5371
