Update on EPA's ToxCast Program: Providing High ... - ACS Publications

Apr 21, 2012 - support tools based on ToxCast in vitro screening results to help prioritize .... ambitious goals of the program for broad chemical-tox...
0 downloads 0 Views 4MB Size
Download PDF
Perspective pubs.acs.org/crt

Update on EPA’s ToxCast Program: Providing High Throughput Decision Support Tools for Chemical Risk Management Robert Kavlock,*,† Kelly Chandler,†,‡ Keith Houck,† Sid Hunter,‡ Richard Judson,† Nicole Kleinstreuer,† Thomas Knudsen,† Matt Martin,† Stephanie Padilla,‡ David Reif,† Ann Richard,† Daniel Rotroff,† Nisha Sipes,† and David Dix† †

National Center for Computational Toxicology and ‡National Health and Environmental Effects Research Laboratory, Office of Research and Development, U. S. Environmental Protection Agency, Research Triangle Park, North Carolina 27711, United States ABSTRACT: The field of toxicology is on the cusp of a major transformation in how the safety and hazard of chemicals are evaluated for potential effects on human health and the environment. Brought on by the recognition of the limitations of the current paradigm in terms of cost, time, and throughput, combined with the ever increasing power of modern biological tools to probe mechanisms of chemical−biological interactions at finer and finer resolutions, 21st century toxicology is rapidly taking shape. A key element of the new approach is a focus on the molecular and cellular pathways that are the targets of chemical interactions. By understanding toxicity in this manner, we begin to learn how chemicals cause toxicity, as opposed to merely what diseases or health effects they might cause. This deeper understanding leads to increasing confidence in identifying which populations might be at risk, significant susceptibility factors, and key influences on the shape of the dose−response curve. The U. S. Environmental Protection Agency (EPA) initiated the ToxCast, or “toxicity forecaster”, program 5 years ago to gain understanding of the strengths and limitations of the new approach by starting to test relatively large numbers (hundreds) of chemicals against an equally large number of biological assays. Using computational approaches, the EPA is building decision support tools based on ToxCast in vitro screening results to help prioritize chemicals for further investigation, as well as developing predictive models for a number of health outcomes. This perspective provides a summary of the initial, proof of concept, Phase I of ToxCast that has laid the groundwork for the next phases and future directions of the program.

■

CONTENTS

1. Introduction 2. ToxCast Chemical Library 3. ToxCast Bioassay Library 3.1. ACEA (ACEA Biosciences, Inc., San Diego, CA) 3.2. Apredica (Formerly Cellumen, Now a Subsidiary of Cyprotex, Watertown, MA) 3.3. Attagene (Attagene Inc., Morrisville, NC) 3.4. BioMAP (BioSeek Inc., Now a Subsidiary of Asterand, South San Francisco, CA) 3.5. BioReliance (BioReliance, Rockville, MD) 3.6. CellzDirect (CellzDirect Is a Subsidiary of Life Technologies Corp., Durham, NC) 3.7. Embryonic Stem Cells (National Health and Environmental Effects Research Laboratory, US EPA, Research Triangle Park, NC) 3.8. GreenScreen HC (Gentronix, Manchester, UK) 3.9. NCGC (National Institutes of Health (NIH) Chemical Genomics Center, Gaithersburg, MD)

3.10. NovaScreen (NovaScreen Biosciences Corporation Is a Subsidiary of Caliper Life Sciences Inc., Hanover, MD) 3.11. Odyssey Thera (Odyssey Thera, San Ramon, CA) 3.12. Vala (Vala Sciences, San Diego, CA) 3.13. Zebrafish Development (National Health and Environmental Effects Research Laboratory, U. S. EPA, Research Triangle Park, NC) 3.14. Coverage of Biological Space 4. Public Availability of Data 5. Data Analysis and Development of Predictive Signatures 5.1. ToxPi 5.2. Prediction of Rat Liver Cancer 5.3. Prediction of Reproductive Toxicity 5.4. Prediction of Developmental Toxicity 5.5. Vascular Mechanisms of Developmental Toxicity 6. Potential for High Throughput Risk Assessment

1288 1288 1289 1291 1291 1291 1291 1292 1292

1292 1292

1292

This article not subject to U.S. Copyright. Published 2012 by the American Chemical Society

1292 1293 1293

1293 1293 1293 1294 1294 1295 1296 1297 1297 1299

Received: March 2, 2012 Published: April 21, 2012 1287

dx.doi.org/10.1021/tx3000939 | Chem. Res. Toxicol. 2012, 25, 1287−1302

Chemical Research in Toxicology 7. Future Directions 8. Summary Author Information Acknowledgments Abbreviations References

Perspective

larger suite of in vitro assays, and is described in the Future Directions section of this document. Here, we provide an integrated overview of Phase I of the ToxCast program and its major products approximately 5 years after its initiation. Our objective is to summarize recent advances in the evaluation of HTS assays for predictive toxicology and lessons learned from the first-generation prediction models built from the Phase I HTS data. All data sets associated with Phase I have been made publicly available to the scientific community for independent analysis. As such, this perspective serves as a reflection on Phase I and as a preamble to Phase II in bringing the vision of toxicology in the 21st century closer to fruition.

1299 1300 1300 1300 1300 1301

1. INTRODUCTION The U. S. Environmental Protection Agency (EPA) ToxCast research program was launched in 2007 with the goal of evaluating the use of high-throughput bioassays to probe key biological events and pathways that are potential targets for chemicals whose interactions could cause diseases such as cancer, reproductive toxicity, or birth defects.1 As noted by the National Research Council (NRC) in its vision for a new paradigm in toxicity testing, the traditional approach to toxicology uses expensive and time-consuming animal-based testing approaches that cannot keep pace with the large numbers of chemicals in commerce.2 In addition, since the traditional approaches do not provide mechanistic information on how the chemicals exert toxicity, there remain large uncertainties in extrapolating data across dose, species, and life stages. ToxCast is designed to address many of the issues raised in the NRC report and echoed in the more recent report from the Council of Canadian Academies.3 It is a multiyear, multimillion dollar effort to comprehensively apply hundreds of in vitro tests against hundreds to thousands of chemicals, many of which have known toxicological phenotypes derived from traditional guideline studies for cancer, reproductive impairment and developmental disorders.4−6 The goal is to acquire sufficient information on a range of chemicals so that “bioactivity profiles” or “in vitro signatures” can be discerned that predict patterns of toxic effects or phenotypes observed in traditional animal toxicity testing. Although similar proposals have been made relative to the utility of high information-content data sets coming from omics technologies, particularly transcriptomics, where the high degree of parallelism can inform toxicant-responsive pathways, the ToxCast approach offers greater depth and detail about the molecular targets and cellular consequences of exposure across a range of concentrations. One factor that differentiates the ToxCast program from previous such efforts is the testing of a much larger number of compounds than any of the omics-centered projects (∼1000 chemicals included in Phases I and II) and that these chemicals were chosen to be of direct relevance to the environmental toxicity community. The resulting predictive bioactivity signatures are being developed based upon physicochemical properties, biochemical activities from high-throughput screening (HTS) and cell-based phenotypic assays, gene expression analyses of cells in vitro, and physiological responses in nonmammalian model organisms. A central tenet of the program is a commitment to transparency and public release of all data (see http://www.epa.gov/actor for access). ToxCast is being implemented in a phased approach. The initial Phase I, conducted between 2007 and 2010, was a proof of concept involving several hundred chemicals with wellcharacterized toxicity profiles. The results of that phase have been published in the peer reviewed literature and are the subject of this perspective. ToxCast Phase II is currently underway and involves a larger number of chemicals against a

2. TOXCAST CHEMICAL LIBRARY One way in which ToxCast represents a major departure from traditional toxicology testing is by virtue of the number of chemicals being simultaneously analyzed using a large number of HTS technologies. Hence, the ToxCast chemical library is a central, foundational element of the ToxCast effort (Table 1). Furthermore, the extent to which the chemical library provides sufficient coverage of chemical structure feature and property space in relation to in vivo toxicity end points both enables and limits the ability to derive useful associations from HTS in vitro data. The original ToxCast Phase I chemical library (referred to here as PhaseI_v1) consisted of 309 unique compounds, the large majority being pesticide active ingredients for which rich in vivo guideline study data were available (Table 1). The remainder of PhaseI_v1 included a small number of industrial chemicals of high interest to EPA programs (e.g., bisphenol A and perfluorooctanoic acid) and 10 known active metabolites relating to 12 pesticides, also included as part of the inventory. The full testing library of 320 compounds included 11 compounds that were source replicates (e.g., the same compound procured from two different vendors) or technical replicates (e.g., same source compound plated at three different locations). Additional chemical selection criteria included the following physicochemical filters and practical considerations: log of the octanol/water partition coefficient (LogP) in the range of −1 to 6 (97.5% met this criterion); molecular weight 250−1000 (90% met this criterion); commercial availability with purity >90% (98% met this criterion); and solubility in dimethyl sulfoxide (DMSO), determined after procurement and prior to HTS plating. Despite a majority representation of pesticide active ingredients, the PhaseI_v1 chemical library spans a relatively wide range of property values and diverse structure feature space, representing over 40 chemical functional classes (e.g., pyrazole, sulfonamide, organochlorine, pyrethroid, carbamate, organophosphate, etc.) and 24 known pesticidal mode-of-action (MoA) classes (e.g., phenylurea herbicides, organophosphate insecticides, dinitroaniline herbicides, etc.). A summary of compound classes included in the Phase I library is provided in Table 2. Chemical identifiers (e.g., names, CAS Registry Numbers, and chemical structures) for the PhaseI_v1 library were curated, quality reviewed, and structure-annotated within EPA’s DSSTox (Distributed Structure Searchable Toxicity) project.7 Subsequent analytical quality control (QC) was performed to confirm parent identity, stability, and purity for a major portion of the chemical library. The analytical methods employed (LC- and GC-MS) were deemed unsuitable for approximately 10% of the library (e.g., metal-containing and low molecular weight compounds), and evidence of significant sample decomposition in DMSO over time was found for a 1288

dx.doi.org/10.1021/tx3000939 | Chem. Res. Toxicol. 2012, 25, 1287−1302

Chemical Research in Toxicology

Perspective

Table 1. Characteristics of Chemicals Included in HTS Assaysa compd library

no. of unique chemicals

ToxCast Phase I

309

ToxCast Phase II

776

e1k

1860

Tox21

8193

a

compd characteristics

prevalence of toxicity data

prevalence of human exposure

pesticidal actives commercial chemicals commercial chemicals pesticide actives failed pharmaceuticals commercial chemicals reference chemicals pesticidal actives commercial chemicals all marketed pharmaceuticals

+++ ++ +/− ++ ++ +/− ++ ++ +/− +/?

++ +/− +/− + +++ +/− +/− + +/− ++

assays (no. of platforms) 664 (10) ∼700 (13)

206 (4) ∼50−100 (1)

See http://www.epa.gov/ncct/toxcast/chemicals.html for complete listing of chemicals.

Table 2. List of Top 40 Representative Pesticidal Active Compound Classes Contained in the ToxCast Phase I Inventorya

a

chemical class

count

chemical class

count

organophosphorus amide organothiophosphate urea conazole sulfonylurea carbamate anilide heterocyclic_organothiophosphate phenoxy pyrethroid triazole triazinylsulfonylurea pyridine pyrimidinylsulfonylurea phenoxypropionic aryloxyphenoxypropionic dicarboximide dinitroaniline aliphatic_organothiophosphate

35 24 24 22 18 16 15 14 12 12 12 12 11 10 9 8 7 7 7 6

aromatic_acid dithiocarbamate imidazole organophosphate phthalate pyrazole thiocarbamate imidazolinone nicotinoid phenyl_organothiophosphate aromatic chloroacetanilide chlorotriazine organochlorine oxazole oxime_carbamate phenylurea strobilurin sulfonamide thiazole other (3 cmpds or fewer in class)

6 6 6 6 6 6 6 5 5 5 4 4 4 4 4 4 4 4 4 4 114

Compounds may be assigned to multiple classes.

class of sulfurons that likely undergo acid hydrolysis (14 total). A DSSTox tabular listing and Structure Data Format (SDF) file for the full ToxCast chemical library, with PhaseI_v1 compounds indicated and sulfurons flagged (and eliminated from subsequent testing), is available at http://www.epa.gov/ NCCT/dsstox/sdf_toxcst.html. The chemical property space distribution of the ToxCast PhaseI_v1 inventory in relation to a representative portion of chemicals from the DSSTox Master inventory, which spans a large diversity of environmental and commercial chemicals (see http://www.epa.gov/NCCT/dsstox/DataFiles.html), is shown in Figure 1. Considering the comparatively small size of the Phase I inventory (309 compounds), the plot illustrates a reasonable coverage of property space, supporting the initial objectives of the ToxCast program. However, this plot also indicates regions that are not well covered by the Phase I inventory (e.g., consisting of drugs, antimicrobials, food additives, etc.), underscoring the need for expanded coverage of chemical space in future ToxCast testing phases if the ambitious goals of the program for broad chemical-toxicity mechanism and end point coverage are to be fully achieved.

These considerations and the experiences gained in compiling the PhaseI_v1 chemical inventory were factored into the selection of the larger set of ToxCast Phase II compounds, for which testing is currently underway.

3. TOXCAST BIOASSAY LIBRARY The selection of an appropriate suite of in vitro assays capable of providing broad coverage of potential toxicity targets is a formidable challenge. Whereas advances in HTS technologies have permitted the development of screening assays against a great diversity of molecular and cellular targets, defining those targets and validating their link to specific toxicity end points remains a major obstacle to defining a toxicity testing paradigm.8 With this viewpoint in mind, the ToxCast suite of biological assays was developed to yield an extensive sampling of biological space, i.e., molecular targets chosen from all major protein superfamilies, complemented by cellular assays covering key signaling pathways and phenotypic end points. Known toxicity targets (e.g., the hERG potassium ion channel, steroid hormone receptors, cholinesterases and ion channels, functional mitochondrial assays, etc.) were included to the extent 1289

dx.doi.org/10.1021/tx3000939 | Chem. Res. Toxicol. 2012, 25, 1287−1302

Chemical Research in Toxicology

Perspective

to be perturbed by xenobiotic small molecules. The assays were all available in high-throughput format and amenable to testing tens of thousands of chemicals. (Note that the term “highthroughput” refers here to assays capable of being run in 96well plate format, or greater, in automated systems, over a relatively short period of time, days to weeks.) The initial testing of the Phase I library encompassed 10 assay platforms and covered over 650 features (see Table 3). Analysis of the results using statistical methods to associate in vitro results with in vivo end points provided some success in building predictive models for end points that included reproductive function, developmental toxicity and vascular development (discussed below). As time progressed, assays developed by intramural researchers in EPA were also added to the suite, and additional assay platforms were added in 2011 using the RFP process. Collectively, these new assays provided additional coverage of genotoxicity, developmental biology, cell signaling pathways and xenobiotic metabolism capabilities. It is expected that the collection of assays used in an in vitro toxicity testing approach will be constantly evolving to keep pace with new technologies and advances in the understanding of toxicity pathways. All platforms were run in concentration response, with between 4 and 15 concentrations depending upon assay complexity, capacity, and cost. All data were received by EPA and processed through work flows that included normalization, curve fits using Hill equations, visual plots of the concentration−response relationships, and, finally, calculation of the concentration causing half maximal response (AC50) or, in some platforms, the Lowest Effect Concentration (LEC).9 A ‘hit’ is defined as to when an AC50 (or LEC) is recorded for a chemical-assay pair. The range of AC50s for any particular assay platform reflected the nature and cellular complexity of assays that spanned cell-free biochemical assays to complex cell culture systems. The most sensitive platforms, on average, were the mouse embryonic stem cell assay and zebrafish embryo

Figure 1. Comparison of the ToxCast Phase I inventory in relation to a representative portion of the DSSTox published inventory (5400 compounds) spanning industrial chemicals and drugs, according to 3 computed properties, LogP (octanol/water partition coefficient), TPSA (log of the total polar surface area), and complexity (log of complexity computed based on paths, branching, and atoms); properties were computed using the Adrianna software (donated by Molecular Networks GmbH, Erlangen, Germany).

possible. Access to these assays was achieved primarily by contracts generated in response to public Requests For Proposals (RFP) in 2006 that described the goals of the project in Statements of Work. The respondents to the RFPs were, by and large, companies supporting the drug discovery community. Thus, assays tended to be directed at targets and pathways generally shown to be critical components of pathological processes and, in general, that may be expected Table 3. HTS Components of ToxCast Phase Ia technology platform source

assay features

description

ACEA

real time cell electronic sensing (RT-CES) of growth of A549 cells

7

Apredica (formerly Cellumen) Attagene Biorelianceb Bioseek CellzDirect

cellular high content screening (HCS) evaluating cellular markers such as stress pathways, mitochondrial involvement, cell cycle, cell loss, mitotic arrest, and the cytoskeleton in HepG2 cells

19 (at 3 times)

multiplexed transcription factor profiling in HepG2 cells gene mutation and DNA and chromosomal damage assays ELISA based readouts of interactions of cocultures of primary human cells qNPA on select genes relevant to xenobiotic metabolism in primary human hepatocytes

81 TBD 174 16 (at 3 times) 1 4

Gentronix mouse embryonic stem cellsc NCGC NovaScreen Odyssey-Therab Vala Sciencesb zebrafishc

GreenScreen genetic toxicity assay using GADD45a GFP in TK6 cells mouse embryonic stem cell cytotoxicity and differentiation qHTS profiling of nuclear receptor function in agonist and antagonist mode by reporter genes using a variety of cell types biochemical profiling, largely using human proteins, of receptor binding, enzyme assays, GPCRs, and ion channels protein complementation assays for a wide variety of intracellular signaling networks. high content multiparameter assays providing quantitative digital imaging of cultured cells as well as information on a variety of cellular proteins/structures/function zebrafish embryonic development assay

ref no individual publication available; results included in ref 12 no individual publication available; results included in ref 12 16 18 17 19 24 22

19

26

273

28

TBD TBD

29 30 and 31

1

35

a

See http://actor.epa.gov/actor/faces/ToxCastDB/DataCollectionList.jsp for a full listing of the assays for each platform. bPlatform added to ToxCast suite following the completion of Phase I. cAssay provided by EPA research laboratory. 1290

dx.doi.org/10.1021/tx3000939 | Chem. Res. Toxicol. 2012, 25, 1287−1302

Chemical Research in Toxicology

Perspective

of differentiation with time in culture. End points included features such as overt cytotoxicity in the form of cell loss, cell cycle status, mitochondrial mass and function, apoptosis, steatosis, microtubule stability, DNA damage, oxidative stress, and stress kinase activation. The most frequently observed effect in either cell type was cell loss, with HepG2 cells significantly more sensitive than hepatocytes, an effect that is likely due to the active cell cycling of this line. Cell loss is determined using the Hoescht 33342 fluorescent nuclear dye that stains the nuclei of living or fixed cells. Cells detached from the plate, lysed, or having undergone dramatic changes in nuclear morphology usually associated with cell death will not be recognized as intact cells and, thus, will be scored positive for cell loss. An inhibition of proliferation relative to control treatment would also generate a positive result for cell loss. Another consistent observation was that mitochondrial membrane potential was affected to a much higher degree in HepG2 cells than in primary hepatocytes, perhaps due to rapid detoxification of the chemicals in the metabolically competent hepatocytes. Effects on mitochondrial properties were often coupled to oxidative stress and/or stress kinase activation and subsequent cell death. 3.3. Attagene (Attagene Inc., Morrisville, NC). This cellular biosensor system (Factorial) enables rapid, highcontent screening (HCS) assessment of a compound’s impact on gene regulatory networks. The factorial biosensors combined libraries of cis- and trans-regulated transcription factor reporter constructs with a highly homogeneous method of detection, enabling simultaneous evaluation of multiplexed transcription factor activities. We demonstrated the applicability of this technology by quantitatively evaluating the effects of 309 environmental chemicals on 25 nuclear receptors (TRANS) and 48 transcription factor response elements (CIS). There were 10 corresponding receptor family targets between the CIS and TRANS assays. For example, the TRANS estrogen receptor alpha and the CIS estrogen response element assay results were highly correlated. In addition to coherent activity among related targets across the CIS and TRANS platforms, nuclear redox factor (Nrf2) activity and other markers of oxidative stress were highly correlated to the overall promiscuity of a chemical and have subsequently aided in the interpretation of positive results in many of the nuclear receptor targets.16 3.4. BioMAP (BioSeek Inc., Now a Subsidiary of Asterand, South San Francisco, CA). This assay system models complex human disease and tissue biology with cocultures of primary human cells under various stimulatory conditions. 17 Cell types consisted of endothelial cells, fibroblasts, bronchial epithelial cells, smooth muscle cells, keratinocytes, and peripheral blood mononuclear cells in 8 different culture conditions. Cells were stimulated with a variety of important biological effectors such as TNF-α, LPS, IL-1β, IFN-γ, EGF, and TGF-β, and increased or decreased levels of downstream response proteins were determined by the enzyme-linked immunosorbent assay (ELISA). Measures of overt cytotoxicity and proliferation are also used. The effects of each of the ToxCast chemicals on a total of 87 features were determined. In addition to using the results of individual features in the larger ToxCast profiling data set, the BioMAP results for individual chemicals were compared for similarity by pairwise correlation analysis, resulting in functionally related clusters. Comparing these clusters to reference BioMAP profiles of compounds with known mechanisms of action led

development assay, which had the lowest values (see corresponding sections below). In the following sections, each Phase I assay technology is briefly described, together with some general findings and performance characteristics. 3.1. ACEA (ACEA Biosciences, Inc., San Diego, CA). This assay platform uses a technique called Real-Time Cell Electronic Sensing (RT-CES), which is a noninvasive, impedance-based technique to continuously measure the state of cells exposed to a chemical.10,11 Cells are seeded in microtiter plates that are integrated with microelectronic sensor arrays. Application of a low voltage (less than 20 mV) leads to the generation of a microampere current that is differentially modulated by the number of cells in the well, the morphology of the cells, and the strength of cell attachment. The application of cell sensor impedance technology for cell-based assays provides several fundamental advantages. First and foremost, cell sensor impedance allows for the noninvasive monitoring of cells over the entire history of the experiment, including cell attachment and spreading, cell morphology, proliferation and any treatment or manipulations.11 Because data are collected continuously, both short-term and long-term cellular responses to treatments can be monitored in the same well. In addition, the unique nature of the impedance readout, combined with real-time data acquisition and display for each treatment (chemical and concentration), leads to a unique response profile. The impedance measured by the RT-CES system is sensitive to three cellular phenotypes: cell number (and, hence, the real-time growth rate), cell morphology, and cell−cell adhesion. Therefore, chemicals that affect any of these properties will result in specific changes in the time-dependent impedance curves, from which kinetic parameters can be extracted. For analysis of the ToxCast chemicals, we developed a simple biologically based concentration−response (BBCR) model that links the time-dependent assay data to cell growth and cell morphology kinetics as a function of concentration and time.12 This model accounts for the effects that chemicals exert on both growth dynamics and cell morphology. Chemicals were classified based on the different types of cell growth curves that they exhibited. In addition to this qualitative information (e.g., cell growth trajectories), quantitative information for the different growth curves was determined (e.g., AC50s or LECs). 3.2. Apredica (Formerly Cellumen, Now a Subsidiary of Cyprotex, Watertown, MA). Chemical-induced liver toxicity is a major concern for safety assessment since the liver serves as the major detoxification organ in the body and is the site of active metabolism of xenobiotics. Effects of the ToxCast compound library on two liver models of hepatotoxicity, the HepG2 human hepatoma cell line and primary rat hepatocytes, were evaluated using an automated high-content imaging platform. HepG2 cells, while lacking many fully differentiated hepatocyte functions, are a well-studied cell model and undergo continuous proliferation in culture, an important feature that was reported as the most sensitive end point correlating with hepatotoxicity in previous studies.13,14 Freshly isolated primary rat hepatocytes, in contrast, have physiological levels of Phase I and II metabolic enzymes but lack proliferative capacity.15 Following chemical exposure for 1, 24, and 72 h (HepG2) or 1, 24, and 48 h (primary rat hepatocytes), multiplexed end points in both culture systems were imaged by automated fluorescent microscopy and analyzed by imaging algorithms. These exposure times were selected to account for acute as well as more chronic effects, with primary hepatocytes limited to 48 h exposure due to a loss 1291

dx.doi.org/10.1021/tx3000939 | Chem. Res. Toxicol. 2012, 25, 1287−1302

Chemical Research in Toxicology

Perspective

transcription factor binding assays and impaired ES cell cardiomyocyte differentiation, whereas chemical subsets that increased cytotoxicity alone did not have this association. Multiple developmentally critical genes were strongly associated with impaired ES cell differentiation. In addition, key players in reactive oxygen species (ROS) signaling pathways were strongly associated with decreased cardiomyocyte differentiation. A multivariate model built from these data revealed that up-regulation of ABCG2 was a strong predictor of decreased ES cell differentiation.22 3.8. GreenScreen HC (Gentronix, Manchester, UK). This assay has provided high-throughput detection of a marker of genotoxicity using the TK6, human lymphoblastoid cell line. This cell line was engineered to contain a green fluorescent protein (GFP) reporter gene regulated by the human GADD45a (growth arrest and DNA damage) gene promoter.23 This promoter contains p53 response elements and is responsive to all classes of genotoxic damage. The ToxCast Phase I library was tested in this assay at concentrations up to 200 μM with concurrent cytotoxicity determined because this can confound interpretation of results.24 GreenScreen HC results were compared to mutagenicity (Ames) and animal tumorigenicity data from ToxRefDB (http://actor.epa.gov/ toxrefdb). Overall, the assay showed low sensitivity but high specificity (88%) for animal tumorigens. Highest concordance was 74−78% for compounds producing tumors in multiple sites in rodents. The low sensitivity may be explained by the lack of metabolic activation (a caveat for many of the in vitro platforms) and weaker upstream p53-activating signals for nongenotoxic carcinogens. 3.9. NCGC (National Institutes of Health (NIH) Chemical Genomics Center, Gaithersburg, MD). The NCGC is a member of the broader Tox21 consortium (see Future Directions) and provided quantitative HTS assays against 10 human nuclear receptors (AR, ERα, FXR, GR, LXRβ, PPARγ, PPARδ, RXRβ, TRβ, and VDR) in both agonist and antagonist modes.25 The assays used GeneBLAzerbla cell lines that constitutively coexpress a fusion protein comprising the ligand-binding domains of related human nuclear receptors coupled to the DNA-binding domain of the yeast transcription factor, GAL4. When activated, these fusion proteins stimulate β-lactamase reporter gene expression. The assay was conducted in 1536 well format using 14 or 15 concentrations, ranging from 0.5 nM to 92 μM. Data were filtered for autofluorescence and/or cytotoxicity in the analysis.26 3.10. NovaScreen (NovaScreen Biosciences Corporation Is a Subsidiary of Caliper Life Sciences Inc., Hanover, MD). These cell-free biochemical and ligand binding-assays were selected from a commercial panel for preclinical drug development based on published evidence linking assay targets to pathways of toxicity, cell signaling, and xenobiotic metabolism. The biochemical HTS assay portfolio included a total of 292 assay features: 77 G-protein coupled receptor (GPCR) binding assays; 32 CYP-450-related enzyme activities; enzymatic assays for 72 kinases, 22 phosphatases, 15 proteases, 6 histone deactylases (HDACs), 3 cholinesterases, and 14 other enzyme activities; 18 nuclear receptor binding assays; 20 ion channel and ligand-gated ion channel activities; and 9 transporter proteins, 2 mitochondrial pore proteins, and 2 other receptor types.6 Due to cost considerations, a singleconcentration screen was run for each chemical-assay combination followed by a concentration−response for all active and selected inactive calls. Quality control checks showed

to the association of several of the clusters with potential toxicity mechanisms, including mitochondrial dysfunction, endoplasmic reticulum stress, NFκB inhibition, cAMP elevation, and microtubule destabilization.17 3.5. BioReliance (BioReliance, Rockville, MD). This assay platform was added to the ToxCast testing suite after completion of Phase I. Standard in vitro genotoxicity testing in support of various regulatory requirements (e.g., FDA, EPA, OECD, EU, JMHW, and JMAFF (see Abbreviations)) are conducted in compliance with Good Laboratory Practice (GLP) standards and are cost-prohibitive in high-throughput mode. Hence, efforts have been made to develop HTS versions of the assays that can be used to prioritize chemicals for the standard genotoxicity batteries. The GreenScreen HC assay described below is one example of an HTS genotoxicity assay and is available through BioReliance. In addition, the Ames II assay for mutagenicity is a 384-well-based, liquid culture version of the Ames assay that has throughput compatible with testing large numbers of chemicals. The in vitro micronuclease assay, designed to detect damage to chromosomes or the spindle apparatus of cells, has also been adapted to a higher throughput format using detection by flow cytometry.18 Although the ToxCast Phase I library contained few genotoxicants, as the majority of the chemicals were food-use pesticide active ingredients, a battery of high-throughput genotoxicity assays will be valuable for prioritizing the more diverse sets of chemicals in Phase II and beyond. 3.6. CellzDirect (CellzDirect Is a Subsidiary of Life Technologies Corp., Durham, NC). The metabolic role of the liver makes it an important toxicological target for xenobiotics. Primary human hepatocytes afford an in vitro model system retaining representative metabolic capacity, cell signaling, and transporter functionality. The effect of ToxCast Phase I compounds on the expression of 14 genes regulated by various nuclear receptor (NR) signaling pathways was evaluated using the quantitative nuclease protection assay (qNPA) by CellzDirect in primary hepatocytes from two human donors at 6, 24, and 48 h time points. Gene targets were selected as sentinels for five key hepatic NR signaling pathways: the aryl hydrocarbon receptor (AhR), constitutive androstane receptor (CAR), pregnane X receptor (PXR), farnesoid X receptor (FXR), and peroxisome proliferater-activated receptor-alpha (PPARα) signaling pathways. Concentration−response curves were fit to the data to determine potency and efficacy for each chemical’s response. Results showed that various chemicals in the ToxCast chemical library were capable of significantly altering the expression of certain gene targets in this culture system, indicative of chemical impact on NR signaling. Overall, out of 12,978 chemical−gene−time combinations, there were a total of 2,313 significant gene expression responses.19 3.7. Embryonic Stem Cells (National Health and Environmental Effects Research Laboratory, US EPA, Research Triangle Park, NC). Embryonic stem (ES) cells have been utilized in developmental toxicity screening for more than a decade.20 In the context of toxicity testing, ES cells may provide insight into key developmental signaling pathways and/ or biological networks that are perturbed upon exposure. This assay was developed in an EPA research lab as a sensitive, highthroughput method for quantitatively measuring cytotoxicity and cardiomyocyte differentiation simultaneously in a mouse ES cell (MESC) model system and applied to the ToxCast Phase I library.21,22 Interestingly, statistical analysis revealed significant associations for a subset of chemicals that perturbed 1292

dx.doi.org/10.1021/tx3000939 | Chem. Res. Toxicol. 2012, 25, 1287−1302

Chemical Research in Toxicology

Perspective

developmental signaling pathways and their regulation are conserved between fish and mammals, the zebrafish provides a model for studying mammalian disease as well as for molecular dissection of developmental pathways. The goals were to test the ToxCast Phase I 309 chemicals for toxicity toward the zebrafish embryo using a developmental assay with general, phenotypic end points (lethality, nonhatching, and malformations). The embryos were exposed throughout development to a wide concentration range (i.e., 5 orders of magnitude; highest dose = 80 μM) of each chemical. The results revealed that the majority (62%) of the chemicals were toxic to the developing embryo at concentrations at or below 80 μM.35 This toxicity, in terms of both incidence and potency, was correlated with chemical class as well as the relative hydrophobicity of the chemical. Furthermore, the conditions of the assay were such that the inter- and intraplate variability was low, and the data are consistent with previously published studies of selected chemicals tested in zebrafish embryos.33 3.14. Coverage of Biological Space. An important question to ask about the ToxCast approach is whether the included assays provide sufficient coverage of biological space to enable development of models to predict toxicity. In particular, the approach can be contrasted with whole-genome omics approaches, which were not employed in ToxCast except in pilot mode for a few chemicals. The present approach has been driven by the desire to employ a variety of molecular and cellular properties (as opposed to gene expression changes only), a variety of cell systems, and to be able to run all assays in concentration−response. These requirements have largely excluded whole-genome approaches for cost reasons, although with the advent of new whole genome technologies, this limitation is likely to disappear. In analogy to genomics approaches, however, gene coverage can be used to estimate the current coverage of biological space by the ToxCast assay set. The full suite of ToxCast assays probe a total 327 genes (mostly human) that cover a wide variety of molecular pathways. In particular, ToxCast has at least one assay mapping to a gene in 293 out of 592 pathways from KEGG, 520 out of 1117 pathways from REACTOME, and 313 out of 1050 pathways from PathwayCommons.36−38

excellent assay reproducibility and accuracy. For example, hit concordance was 74.7%, and overall AC50 concordance was 99.0% for replicate chemicals. A typical drug discovery effort may screen a chemical library for active compounds at lower chemical concentrations (e.g., 1 μM) to avoid false positives, whereas the ToxCast strategy used high concentration testing to minimize false negatives. The Phase I true positive rate (e.g., chemicals active in single-concentration versus concentration− response series) was 64.4% versus 2.5% false negative rate for the biochemical HTS. As such, the preponderance of false positives over false negatives is consistent with a conservative strategy that emphasizes sensitivity over specificity.27 In contrast, a comparison of assay orthologues across human and rat (or in some cases bovine) species showed weaker concordance for diverse endocrine, neurological, and xenobiotic enzyme targets.28 3.11. Odyssey Thera (Odyssey Thera, San Ramon, CA). This assay platform was added after the completion of ToxCast Phase I. It uses a proprietary assay technology called “PCA” (Protein-Fragment Complementation Assay). A PCA consists of a gene encoding a reporter protein that is dissected into two fragments. A test protein of interest (“A”) is fused in-frame to one of the reporter fragments, and the other test protein (“B”) is fused to the other reporter fragment. When “A” and “B” interact, the reporter protein is assembled, producing a detectable signal. Specific biochemical nodes within important human cell signaling pathways are monitored by appropriately choosing proteins “A” and “B”. Selection of the nodes was based on Odyssey Thera’s experience screening known toxicants, failed drugs, and other pharmacological probes. Use of automated, high-content fluorescent imaging makes this approach amenable to a high-throughput format with an added advantage of observing the subcellular location of the interactions being measured. The strategy, therefore, allows detection and quantification of toxicant-induced changes in the presence and localization of specific protein complexes in living human cells.29 3.12. Vala (Vala Sciences, San Diego, CA). Another new addition to the Phase I testing set, this technology offers a range of high-content multiparameter assays, providing quantitative digital imaging of cultured cells and tissues, as well as information on a variety of cellular proteins/structures/ functions. Assays utilizing embryonic stem cells examine chemical effects on differentiation pathways, neurological function, and insulin signaling, whereas the C. elegans small model organism is used to study germ-line proliferation and the Notch/Delta pathway. Lipolysis and adipogenesis assays have the potential to help identify endocrine disrupting compounds and obesogens, respectively. Assays that measure lipid droplet formation pertaining to liver steatosis can assist in identifying chemical toxicity that may lead to fatty liver and associated pathologies. Cultures of human primary cells are used to examine chemical perturbation of gap junction proteins that are critical to angiogenic processes in embryonic development and tumor progression.30−32 3.13. Zebrafish Development (National Health and Environmental Effects Research Laboratory, U. S. EPA, Research Triangle Park, NC). The zebrafish (Danio rerio) is an alternative vertebrate species that has become popular in embryology, pharmacology, and biomedical research. It is particularly amenable to large-scale screening of chemical libraries because the embryos mature rapidly (6 days) and can be maintained in microtiter plates.33,34 Because many key

4. PUBLIC AVAILABILITY OF DATA A major goal of the ToxCast program is to make all of the generated data publicly accessible in multiple formats. This helps leverage the EPA investment in data generation by allowing other groups to analyze and build independent models from the data. In addition, public release of all data used for model building within EPA (e.g., for setting priorities for further testing) supports the objectives of total transparency. Data are being made available in through three primary public venues: the ToxCast Web site (http://www.epa.gov/ncct/ toxcast); the EPA ACToR (Aggregated Computational Toxicology Resource) database (http://actor.epa.gov); and other public data portals, such as PubChem (http://pubchem. ncbi.nlm.nih.gov/).9,39 The ToxCast Web site is the primary source for information on the program, links to publications, and downloads of primary data in flat-file formats (text and MS Excel, primarily). This is also the best site for accessing computation-ready ToxCast data sets. All of the data are also loaded into the ACToR system in both the main ACToR database and in the specialized ToxCastDB database.4,5,9 ACToR is a data warehouse containing hazard, exposure, use, and other types of data from over 1,000 sources on several 1293

dx.doi.org/10.1021/tx3000939 | Chem. Res. Toxicol. 2012, 25, 1287−1302

Chemical Research in Toxicology

Perspective

Table 4. Descriptive Statistics on the Detection of Chemical Activity by the Various ToxCast Assay Platforms Used in Phase Ia platform

no. of chemicals

no. of assays

hitsb per assay (min)

hits per assay (max) [%]

hits per assay (med) [%]

AC50 (min)

AC50 (max)

AC50 (med)

ACEA Apredica Attagene BioSeek CellzDirect Gentronix MESC NCGC Novascreen Zebrafish

309 309 309 309 309 309 309 309 309 309

7 57 73 174 42 1 4 19 292 1

0 0 0 0 0 0 0 0 0 0

6 [86] 31 [54] 30 [37] 72 [41] 24 [50] 1 [100] 4 [50] 4 [21] 76 [26] 1 [100]

1 [14] 5 [9] 4 [5] 15 [9] 6 [12] 0 [0] 0 [0] 0 [0] 3 [1] 0 (0)

0.047 0.039 0.0087 1.5 0.0054 12 0.0012 0.19 0.0056