HIGH-THROUGHPUTSCREENING METHODSIN TOXICITY TESTING

HIGH-THROUGHPUTSCREENING METHODSIN TOXICITY TESTING

Edited by

PABLO STEINBERGInstitute for Food Toxicology and Analytical ChemistryUniversity of Veterinary Medicine HannoverHannover, Germany

Copyright C© 2013 by John Wiley & Sons, Inc. All rights reserved.

Published by John Wiley & Sons, Inc., Hoboken, New Jersey.Published simultaneously in Canada.

No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form orby any means, electronic, mechanical, photocopying, recording, scanning, or otherwise, except aspermitted under Section 107 or 108 of the 1976 United States Copyright Act, without either the priorwritten permission of the Publisher, or authorization through payment of the appropriate per-copy fee tothe Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, (978) 750-8400,fax (978) 750-4470, or on the web at www.copyright.com. Requests to the Publisher for permissionshould be addressed to the Permissions Department, John Wiley & Sons, Inc., 111 River Street, Hoboken,NJ 07030, (201) 748-6011, fax (201) 748-6008, or online at http://www.wiley.com/go/permission.

Limit of Liability/Disclaimer of Warranty: While the publisher and author have used their best efforts inpreparing this book, they make no representations or warranties with respect to the accuracy orcompleteness of the contents of this book and specifically disclaim any implied warranties ofmerchantability or fitness for a particular purpose. No warranty may be created or extended by salesrepresentatives or written sales materials. The advice and strategies contained herein may not be suitablefor your situation. You should consult with a professional where appropriate. Neither the publisher norauthor shall be liable for any loss of profit or any other commercial damages, including but not limited tospecial, incidental, consequential, or other damages.

For general information on our other products and services or for technical support, please contact ourCustomer Care Department within the United States at (800) 762-2974, outside the United Statesat (317) 572-3993 or fax (317) 572-4002.

Wiley also publishes its books in a variety of electronic formats. Some content that appears in print maynot be available in electronic formats. For more information about Wiley products, visit our web site atwww.wiley.com.

Library of Congress Cataloging-in-Publication Data:

High-throughput screening methods in toxicity testing / edited by Pablo Steinberg.pages cm

Includes index.ISBN 978-1-118-06563-1 (hardback)

1. High throughput screening (Drug development) 2. Toxicity testing.I. Steinberg, Pablo, editor if compilation.

RS419.5.H542 2013615.1′9–dc23

2012035755

Printed in the United States of America

ISBN: 9781118065631

10 9 8 7 6 5 4 3 2 1

CONTENTS

PREFACE ix

CONTRIBUTORS xi

PART I GENERAL ASPECTS

1 ToxCast: Predicting Toxicity Potential Through High-ThroughputBioactivity Profiling 3Keith A. Houck, Ann M. Richard, Richard S. Judson, Matthew T. Martin,David M. Reif, and Imran Shah

2 High-Throughput Toxicity Testing in Drug Development: Aims,Strategies, and Novel Trends 33Willem G.E.J. Schoonen, Walter M.A. Westerink, Femke M. van de Water, andHorbach G. Jean

3 Incorporating Human Dosimetry and Exposure Informationwith High-Throughput Screening Data in ChemicalToxicity Assessment 77Barbara A. Wetmore and Russell S. Thomas

4 The Use of Human Embryonic Stem Cells in High-ThroughputToxicity Assays 97Xin Huang, Dan-yan Zhu, and Yi-jia Lou

v

vi CONTENTS

PART II HIGH-THROUGHPUT ASSAYS TO ASSESS DIFFERENTCYTOTOXICITY ENDPOINTS

5 High-Throughput Screening Assays for the Assessment ofCytotoxicity 109Andrew L. Niles, Richard A. Moravec, Tracy J. Worzella, Nathan J. Evans,and Terry L. Riss

6 High-Throughput Flow Cytometry Analysis of Apoptosis 129Francesca de Giorgi and Francois Ichas

7 High Content Imaging-Based Screening for Cellular ToxicityPathways 143Bram Herpers and Bob van de Water

8 The Keratinosens Assay: A High-Throughput Screening Assay toAssess Chemical Skin Sensitization 159Andreas Natsch

9 High-Throughput Screening Assays to Assess ChemicalPhototoxicity 177Satomi Onoue, Yoshiki Seto, and Shizuo Yamada

PART III HIGH-THROUGHPUT ASSAYS TO ASSESS DNA DAMAGEAND CARCINOGENESIS

10 Ames IITM and Ames Liquid Format MutagenicityScreening Assays 193Kamala Pant

11 High-Throughput Bacterial Mutagenicity Testing: VitotoxTM Assay 213Luc Verschaeve

12 Genotoxicity and Carcinogenicity: Regulatory and NovelTest Methods 233Walter M.A. Westerink, Joe C.R. Stevenson, G. Jean Horbach, Femke M. van deWater, Beppy van de Waart, and Willem G.E.J. Schoonen

13 High-Throughput Genotoxicity Testing: The Greenscreen Assay 271Jorg Blumel and Nadine Krause

14 High-Throughput Assays to Quantify the Formation ofDNA Strand Breaks 285Marıa Moreno-Villanueva and Alexander Burkle

CONTENTS vii

15 High-Throughput Versions of the Comet Assay 295Irene Witte and Andre Stang

16 Automated Soft Agar Colony Formation Assay for theHigh-Throughput Screening of Malignant CellTransformation 309Pablo Steinberg

17 High-Throughput Quantification of Morphologically TransformedFoci in Bhas 42 Cells (v-Ha-ras Transfected BALB/c 3T3) UsingSpectrophotometry 317Kiyoshi Sasaki, Ayako Sakai, and Noriho Tanaka

PART IV HIGH-THROUGHPUT ASSAYS TO ASSESSREPRODUCTIVE TOXICITY, CARDIOTOXICITY,AND HAEMATOTOXICITY

18 ReProGlo: A New Stem-Cell-Based High-Throughput Assay toPredict the Embryotoxic Potential of Chemicals 343Frederik Uibel and Michael Schwarz

19 Embryonic Stem Cell Test (EST): Molecular Endpoints TowardHigh-Throughput Analysis of Chemical Embryotoxic Potential 357Peter T. Theunissen, Esther de Jong, Joshua F. Robinson,and Aldert H. Piersma

20 Zebrafish Development: High-Throughput Test Systems to AssessDevelopmental Toxicity 371Stephanie Padilla

21 Single Cell Imaging Cytometry-Based High-Throughput Analysisof Drug-Induced Cardiotoxicity 385Min Jung Kim and Joon Myong Song

22 High-Throughput Screening Assays to Evaluate the CardiotoxicPotential of Drugs 403Carl-Fredrik Mandenius and Thomas Meyer

23 High-Throughput Screening Assays to Evaluate the HematotoxicPotential of Drugs 421Caroline Haglund, Rolf Larsson, and Martin Hoglund

viii CONTENTS

PART V HIGH-THROUGHPUT ASSAYS TO ASSESS DRUGMETABOLISM AND RECEPTOR-RELATED TOXICITY

24 High-Throughput Enzyme Biocolloid Systems for DrugMetabolism and Genotoxicity Profiling Using LC–MS/MS 433James F. Rusling and John Schenkman

25 Higher-Throughput Screening Methods to Identify CytochromeP450 Inhibitors and Inducers: Current Applications and Practice 453David M. Stresser and George Zhang

26 High-Throughput Yeast-Based Assays to Study Receptor-MediatedToxicity 479Johanna Rajasarkka and Marko Virta

27 Evaluating the Peroxisomal Phenotype in High ContentToxicity Profiling 501Jonathan Z. Sexton and Kevin P. Williams

28 A Panel of Quantitative Calux R© Reporter Gene Assays forReliable High-Throughput Toxicity Screening of Chemicals andComplex Mixtures 519Bart van der Burg, Sander van der Linden, Hai-yen Man, Roos Winter, LydiaJonker, Barbara van Vugt-Lussenburg, and Abraham Brouwer

29 DR-Calux R©: A High-Throughput Screening Assay for theDetection of Dioxin and Dioxin-Like Compounds in Food and Feed 533Barbara van Vugt-Lussenburg, Harrie T. Besselink, Bart van der Burg, andAbraham Brouwer

INDEX 547

PREFACE

Conventional approaches to toxicity testing of chemicals and drugs are often decadesold, costly, do not allow high-throughput testing, and are of questionable value whenwanting to estimate human risk. The publication of the document entitled “ToxicityTesting in the 21st Century: A Vision and Strategy” by the US National ResearchCouncil and the implementation of the European legislation on the Registration, Eval-uation, Authorisation and Restriction of Chemicals (REACH) have led to a paradigmshift regarding the strategy to be pursued when evaluating the toxic potential of chem-icals and drugs. Namely, toxicity evaluation should be preponderantly performed byusing high-throughput in vitro methods and toxicity testing methods in animals shouldplay, if at all, a minimal role.

The book gives an overview on a variety of high-throughput screening methodsbeing used in toxicity testing nowadays and should be of help to all scientists workingin the field of toxicity evaluation and risk assessment of chemicals and drugs inchemico-pharmaceutical as well as biotechnology companies, contract laboratories,academia as well as regulatory agencies. The book chapters are written in such a waythat they lend support to those wanting to establish these methods in their laboratoriesas well as those having to evaluate the data generated. Each chapter describes theprinciple of the method and includes detailed information on data generation, dataanalysis, and the use/application(s) in risk assessment. Moreover, the chapters notonly list the advantages of the high-throughput method over the “conventional”methods used up to now in safety evaluation of chemicals and drugs but also pointout limitations and pitfalls.

The book is divided into five parts. Part I includes the strategies pursued nowa-days to predict the toxicity potential of chemicals and drugs through high-throughputbioactivity profiling, the incorporation of human dosimetry and exposure data into

ix

x PREFACE

high-throughput in vitro toxicity screening, and the use of human embryonic stemcells in high-throughput toxicity assays. Part II presents a variety of high-throughputassays to assess different cytotoxicity endpoints; Part III describes high-throughput assays to assess DNA damage and carcinogenesis; Part IV includes high-throughput assays to assess reproductive toxicity, cardiotoxicity, and hematotoxicity;and Part V presents high-throughput assays to assess drug metabolism and receptor-related toxicity. By including all these above-mentioned aspects, the book should beof great value to toxicologists, pharmacologists, analytical chemists, and pharmaceu-tical scientists working in academic institutions, industry, and regulatory agenciesthat are involved in safety evaluation and risk assessment of chemicals and drugsand an excellent complement to the current literature on toxicology in general andsafety evaluation/risk assessment in particular. Because of the test systems and thetoxicity endpoints described, this book could also be extremely interesting for all sci-entists working in the fields of biochemistry, cell biology, molecular biology, systemsbiology, and computational toxicology.

I hereby would like to thank all authors for their excellent contributions. Onlybecause of them it was possible to conceive a book including such a broad spectrumof toxicity testing methods. The development of high-throughput methods to screenthe toxic potential of drugs and chemicals is a rapidly evolving field. If the one orthe other method was missed, then this omission was not on purpose and an incentiveto actualize this version of the book in the future.

Pablo Steinberg

CONTRIBUTORS

Harrie T. Besselink, BioDetection Systems BV, Amsterdam, The Netherlands

Jorg Blumel, MedImmune, Gaithersburg, MD, USA

Abraham Brouwer, BioDetection Systems BV, Amsterdam, The Netherlands,and Department of Animal Ecology, VU University Amsterdam, Amsterdam,The Netherlands

Bart van der Burg, BioDetection Systems BV, Amsterdam, The Netherlands

Alexander Burkle, Molecular Toxicology Group, Department of Biology, Univer-sity of Konstanz, Konstanz, Germany

Nathan J. Evans, Research Department, Promega Corporation, Madison, WI, USA

Francesca de Giorgi, FluoFarma, Pessac, France

Caroline Haglund, Division of Clinical Pharmacology, Department of Medical Sci-ences, Uppsala University, Uppsala, Sweden

Bram Herpers, Division of Toxicology, The Leiden Amsterdam Center for DrugResearch, Leiden University, Leiden, The Netherlands

Martin Hoglund, Division of Hematology, Department of Medical Sciences, Upp-sala University, Uppsala, Sweden

G. Jean Horbach, Department of Toxicology and Drug Disposition, Merck Sharp& Dohme, Oss, The Netherlands

xi

xii CONTRIBUTORS

Keith A. Houck, National Center for Computational Toxicology, Office of Researchand Development, US Environmental Protection Agency, Research Triangle Park,NC, USA

Xin Huang, Cardiovascular Institute, Clinical Research Center, 2nd Affiliated Hos-pital at School of Medicine, Zhejiang University, Hangzhou, People’s Republic ofChina

Francois Ichas, FluoFarma, Pessac, France

Esther de Jong, Laboratory for Health Protection Research, National Institute forPublic Health and the Environment (RIVM), Bilthoven, The Netherlands, and Insti-tute for Risk Assessment Sciences, Faculty of Veterinary Medicine, Utrecht Univer-sity, Utrecht, The Netherlands

Lydia Jonker, BioDetection Systems BV, Amsterdam, The Netherlands

Richard S. Judson, National Center for Computational Toxicology, Office ofResearch and Development, US Environmental Protection Agency, Research Tri-angle Park, NC, USA

Min Jung Kim, College of Pharmacy, Seoul National University, Seoul, South Korea

Nadine Krause, Nonclinical Safety, Merz Pharmaceuticals GmbH, Frankfurt(Main), Germany

Rolf Larsson, Division of Clinical Pharmacology, Department of Medical Sciences,Uppsala University, Uppsala, Sweden

Sander van der Linden, BioDetection Systems BV, Amsterdam, The Netherlands

Yi-jia Lou, Institute of Pharmacology, Toxicology and Biochemical Pharmaceutics,Zhejiang University, Hangzhou, People’s Republic of China

Hai-yen Man, BioDetection Systems BV, Amsterdam, The Netherlands

Carl-Fredrik Mandenius, Division of Biotechnology, Department of Physics,Chemistry and Biology, Linkoping University, Linkoping, Sweden

Matthew T. Martin, National Center for Computational Toxicology, Office ofResearch and Development, US Environmental Protection Agency, Research Tri-angle Park, NC, USA

Thomas Meyer, Multi Channel Systems GmbH, Reutlingen, Germany

Richard A. Moravec, Research Department, Promega Corporation, Madison, WI,USA

Marıa Moreno-Villanueva, Molecular Toxicology Group, Department of Biology,University of Konstanz, Konstanz, Germany

Andreas Natsch, Givaudan Schweiz AG, Duebendorf, Switzerland

Andrew L. Niles, Research Department, Promega Corporation, Madison, WI, USA

CONTRIBUTORS xiii

Satomi Onoue, Department of Pharmacokinetics and Pharmacodynamics, Schoolof Pharmaceutical Sciences, University of Shizuoka, Shizuoka, Japan

Stephanie Padilla, Integrated Systems Toxicology Division, US Environmental Pro-tection Agency, Research Triangle Park, NC, USA

Kamala Pant, BioReliance Corporation, Rockville, MD, USA

Aldert H. Piersma, Laboratory for Health Protection Research, National Institutefor Public Health and the Environment (RIVM), Bilthoven, The Netherlands, andInstitute for Risk Assessment Sciences, Faculty of Veterinary Medicine, UtrechtUniversity, Utrecht, The Netherlands

Johanna Rajasarkka, Department of Food and Environmental Sciences, Universityof Helsinki, Helsinki, Finland

David M. Reif, National Center for Computational Toxicology, Office of Researchand Development, US Environmental Protection Agency, Research Triangle Park,NC, USA

Ann M. Richard, National Center for Computational Toxicology, Office of Researchand Development, US Environmental Protection Agency, Research Triangle Park,NC, USA

Terry L. Riss, Research Department, Promega Corporation, Madison, WI, USA

Joshua F. Robinson, Laboratory for Health Protection Research, NationalInstitute for Public Health and the Environment (RIVM), Bilthoven, TheNetherlands, and Department of Toxicogenomics, Maastricht University, Maastricht,The Netherlands

James F. Rusling, Department of Chemistry, University of Connecticut, Storrs, CT,USA, and Department of Cell Biology, University of Connecticut Health Center,Farmington, CT, USA

Ayako Sakai, Laboratory of Cell Carcinogenesis, Division of Alternative ToxicologyTest, Hatano Research Institute, Food and Drug Safety Center, Hadano, Kanagawa,Japan

Kiyoshi Sasaki, Laboratory of Cell Carcinogenesis, Division of Alternative Tox-icology Test, Hatano Research Institute, Food and Drug Safety Center, Hadano,Kanagawa, Japan

John Schenkman, Department of Cell Biology, University of Connecticut HealthCenter, Farmington, CT, USA

Willem G.E.J. Schoonen, Department of Toxicology and Drug Disposition, MerckSharp & Dohme, Oss, The Netherlands

Michael Schwarz, Department of Toxicology, Institute of Experimental and ClinicalPharmacology and Toxicology, University of Tubingen, Tubingen, Germany

xiv CONTRIBUTORS

Yoshiki Seto, Department of Pharmacokinetics and Pharmacodynamics, School ofPharmaceutical Sciences, University of Shizuoka, Shizuoka, Japan

Jonathan Z. Sexton, Biomanufacturing Research Institute and Technology Enter-prise, North Carolina Central University, Durham, NC, USA

Imran Shah, National Center for Computational Toxicology, Office of Researchand Development, US Environmental Protection Agency, Research Triangle Park,NC, USA

Joon Myong Song, College of Pharmacy, Seoul National University, Seoul, SouthKorea

Andre Stang, Institut fur Biologie und Umweltwissenschaften, Carl von OssietzkyUniversitat Oldenburg, Oldenburg, Germany

Pablo Steinberg, Institute for Food Toxicology and Analytical Chemistry, Univer-sity of Veterinary Medicine Hannover, Hannover, Germany

Joe C.R. Stevenson, Department of Toxicology and Drug Disposition, Merck Sharp& Dohme, The Netherlands

David M. Stresser, Corning GentestSM Contract Research Services, Corning LifeSciences - Discovery Labware, Woburn, MA, USA

Noriho Tanaka, Laboratory of Cell Carcinogenesis, Division of Alternative Tox-icology Test, Hatano Research Institute, Food and Drug Safety Center, Hadano,Kanagawa, Japan

Peter T. Theunissen, Laboratory for Health Protection Research, National Institutefor Public Health and the Environment (RIVM), Bilthoven, The Netherlands, andDepartment of Toxicogenomics, Maastricht University, Maastricht, The Netherlands

Russell S. Thomas, The Hamner Institutes for Health Sciences, Research TrianglePark, NC, USA

Frederik Uibel, Department of Toxicology, Institute of Experimental and ClinicalPharmacology and Toxicology, University of Tubingen, Tubingen, Germany

Luc Verschaeve, Scientific Institute of Public Health, Operational Direction PublicHealth & Surveillance, Laboratory of Toxicology, Brussels, Belgium

Marko Virta, Department of Food and Environmental Sciences, University ofHelsinki, Helsinki, Finland

Barbara van Vugt-Lussenburg, BioDetection Systems BV, Amsterdam, TheNetherlands

Beppy van de Waart, Department of In Vitro & Environmental Toxicology, WILResearch, DD ’s-Hertogenbosch, The Netherlands

Bob van de Water, Division of Toxicology, The Leiden Amsterdam Center for DrugResearch, Leiden University, The Netherlands

CONTRIBUTORS xv

Femke M. van de Water, Department of Toxicology and Drug Disposition, MerckSharp & Dohme, Oss, The Netherlands

Walter M.A. Westerink, Department of In Vitro and Environmental Toxicology,WIL Research, DD’s-Hertogenbosch, The Netherlands

Barbara A. Wetmore, The Hamner Institutes for Health Sciences, Research Trian-gle Park, NC, USA

Kevin P. Williams, Biomanufacturing Research Institute and Technology Enter-prise, North Carolina Central University, Durham, NC, USA

Roos Winter, BioDetection Systems BV, Amsterdam, The Netherlands

Irene Witte, Institut fur Biologie und Umweltwissenschaften, Carl von OssietzkyUniversitat Oldenburg, Oldenburg, Germany

Tracy J. Worzella, Research Department, Promega Corporation, Madison, WI, USA

Shizuo Yamada, Department of Pharmacokinetics and Pharmacodynamics, Schoolof Pharmaceutical Sciences, University of Shizuoka, Shizuoka, Japan

George Zhang, Corning GentestSM Contract Research Services, Corning LifeSciences – Discovery Labware, Woburn, MA, USA

Dan-yan Zhu, Institute of Pharmacology, Toxicology and Biochemical Pharmaceu-tics, Zhejiang University, Hangzhou, People’s Republic of China

PART I

GENERAL ASPECTS

1ToxCast: PREDICTING TOXICITYPOTENTIAL THROUGHHIGH-THROUGHPUT BIOACTIVITYPROFILING

Keith A. Houck, Ann M. Richard, Richard S. Judson,Matthew T. Martin, David M. Reif, and Imran Shah

1.1 INTRODUCTION

Chemical safety assessment has long relied on exposing a few species of laboratoryanimals to high doses of chemicals and observing adverse effects. These results areextrapolated to humans by applying safety factors (uncertainty factors) to account forspecies differences, susceptible sub-populations, establishing no observed adverseeffect levels (NOAEL) from the lowest observed adverse effect levels, and data gapsyielding theoretically safe exposure limits. This approach is often criticized for lackof relevance to human health effects due to the many demonstrated differences inphysiology, metabolism, and toxicological effects between humans and rodents orother laboratory animals [1]. Such criticism exists mainly due to the lack of knowl-edge of specific mechanisms of toxicity and whether these are relevant to humans.Toxicological modes of action (MOA) have been elucidated for only a limited numberof chemicals; even fewer chemicals have had their specific molecular mechanismsof action determined. Having such detailed knowledge would facilitate higher con-fidence in species extrapolation and setting of exposure limits. However, tens ofthousands of chemicals currently in commerce and with some potential for humanexposure lack even traditional toxicity testing and much less elucidated modes ormechanisms of toxicity [2]. Understanding mechanisms of toxicity usually results

High-Throughput Screening Methods in Toxicity Testing, First Edition. Edited by Pablo Steinberg.© 2013 John Wiley & Sons, Inc. Published 2013 by John Wiley & Sons, Inc.

3

4 ToxCast

from decades-long research dedicated to single chemicals of interest, a model unsuit-able for such vast numbers of chemicals. Even with dedicated research, such effortsare not guaranteed to succeed; the extended focus on understanding the mechanism oftoxicity of 2,3,7,8-tetrachlorodibenzodioxin (TCDD) is an example [3]. Traditionalanimal testing, in addition to the criticisms discussed above, is not appropriate or fea-sible for the large numbers of untested chemicals due to the high costs and numberof animals required [1].

One major effort to address this dilemma by providing a high-capacity alternativeis underway, facilitated by integration of the fields of computational toxicology andhigh-throughput in vitro testing [4, 5]. The ultimate goals of this approach are themeans to screen and prioritize thousands of chemicals, predict the potential for humanhealth effects, and derive safe exposure levels for the myriad of chemicals to which weare exposed. This approach relies on a shift in toxicology research away from “black-box” testing on whole animals and toward an understanding of the direct interactionsof chemicals with a broad spectrum of potential toxicity targets comprising specificmolecular entities and cellular phenotypes. This bioactivity profiling of chemicalsgenerated through the use of high-throughput approaches produces characteristicchemical response profiles, or signatures, which may describe the potential for toxicityof that chemical [6].

Computational analysis and modeling of the results are required to provide insightinto complex datasets and support the development of predictive toxicity algorithmsthat ultimately may serve as the foundation of an in vitro toxicity testing approachreplacing most or all animal testings. The groundwork required for a computationaltoxicology approach is the generation of datasets comprising the quantitative effectsof chemicals on biological targets. Two types of data are required. The first are the testresults from in vitro and/or in silico assays that can be run in high-throughput modeand provide bioactivity profiles for hundreds to thousands of chemicals. The secondis a dataset that details the effects of these chemicals on whole organisms, ideally thespecies of interest. These data are used to anchor and build predictive models thatcan then be applied to chemicals that lack in vivo testing. Generation of the in vitrodataset has become feasible and widely available as high-throughput in vitro screeningtechnology, developed in support of the drug discovery community. The selection anduse of these assays for computational toxicology will be discussed further in Section4. Obtaining the latter dataset of in vivo effects necessary to build the computationalmodels presents unique challenges. Although thousands of chemicals have beentested using in vivo approaches, only a limited amount of this information has beenreadily available. Much of it lies in formats not readily conducive to computationalanalysis, for example, paper records, in the data stores of private corporations, orprotected by confidentiality clauses [7], and generation of extensive new in vivo datato support the approach is cost prohibitive. The access and collation of these datainto a relational database useful for computational toxicology will be discussed inSection 5.

Beyond the technical aspects of generating the data, assembling the collection ofrequired datasets to support computational approaches is a challenging task in itself.Robust, efficient, and accurate knowledge discovery from large datasets require a

INTRODUCTION 5

robust data infrastructure. There are a number of critical steps in the process begin-ning with designing an underlying architecture to manage the data. Appropriate datamust be selected and preprocessed into common formats usable to computer programs(e.g., standardized field names for the types of attributes being measured, standardizedchemical names and links to other data sources). The use of standardized ontologiescan be particularly useful in the sharing of information across organizations [8].Because of the complexities of achieving this on a large scale, these approaches areperhaps best conducted by large organizations with access to computational scientistsin addition to experts in chemistry, toxicology, statistics, and high-throughput screen-ing (HTS). Examples of integration of these diverse areas of expertise include the U.S.Environmental Protection Agency’s (EPA) ToxCast program [4] and the Tox21 col-laboration between the EPA, the National Toxicology Program, the National Institutesof Health Center for Translational Therapeutics—NCTT (formally the NIH ChemicalGenomics Center [NCGC]), and the U.S. Food and Drug Administration [9, 10]. Inaddition, a number of large pharmaceutical companies have internal programs in thisarea relying on their own, extensive in-house expertise [11, 12].

As described, the ultimate goal is to use high-throughput in vitro assays to rapidlyand inexpensively profile the bioactivity of chemicals of unknown toxicity and makepredictions about their potential for causing various adverse endpoints [4]. Achievinga robust, predictive toxicology testing program is a long-range goal that will need toproceed through a number of systematic stages including proof-of-concept, extensionof chemical and bioassay diversity, refinement, and ultimately, supplementation orreplacement of existing methods. The initial stage involves multiple steps including(1) selecting an appropriate chemical test set for which in vivo data are available; (2)selecting high-throughput biological assays for screening the chemicals; (3) generat-ing the screening data on the chemicals; (4) collating the in vivo anchoring data for thechemicals; and (5) building up predictive models. Such models can then be validatedthrough testing of additional chemicals with known toxicity endpoints to determinethe robustness of the models. It is likely that the development of the test systems,as well as the computational models, will be an iterative process. New biologicalassays and statistical approaches are evaluated for potential inclusion in the program,whereas assays and models not producing useful results are dropped.

The success of this stage of the process would be models judged useful forprioritizing chemicals for the potential to cause specific toxicity endpoints. Thisprioritization will be valuable in the short term by allowing focused and limitedin vivo use of testing resources on chemicals most likely to be of concern. Theresults of targeted testing of designated chemicals for specific endpoints shouldensure a reduced use of test animals as only limited endpoints would need to beevaluated. This targeted testing will also provide an additional validation method forthe testing program, that is, do the adverse endpoints predicted by the models occurto a significant extent in the tested chemicals? Ultimately, refinement of the testingand modeling approaches should allow high-confidence prediction of the likelihoodfor toxicity, thereby avoiding animal testing altogether for many chemicals. Theremainder of this chapter will focus more specifically on providing background onthe steps undertaken in developing the initial stages of the ToxCast testing program at

6 ToxCast

EPA, as well as examples of applications of the program in prioritizing environmentalchemicals for multiple toxicity endpoints.

1.2 CHEMICAL LANDSCAPE

A major driver of the development and use of HTS methods in toxicology is thescope of the chemical problem, that is, tens of thousands of chemicals to whichindividuals are potentially exposed, the majority of which have never been tested inany significant way [2] . What chemicals are of interest and the kind of data that islikely to be available depends on the use of the chemical, which in turn is relatedto the regulations to which the chemicals are subjected. To understand the world ofchemicals that are of concern for potential toxicity and candidates for testing, it isuseful to discuss a set of chemical inventories, some of which are overlapping.

1.2.1 Pesticide Active Ingredients

These are typically the active compounds in pesticide formulations, which aredesigned to be toxic against select types of organisms. A related category of com-pounds falling under this general label are antimicrobials, which are also designed tobe toxic to certain organisms, in this case-targeting fungi or bacteria. These groups ofchemicals are further divided into food-use and nonfood-use actives for the purposeof regulation. EPA sets tolerance levels for pesticides that may be used in specificfoods, for particular reasons, and at particular exposure levels. Thus, EPA regulatesthe maximum amount of pesticide residue permitted to remain on a food approvedfor pesticide application. FDA, in contrast, has the authority to monitor and enforcelevels of food-use pesticides and ensure that they comply with EPA regulations.FDA has additional authority regarding the use of antimicrobials in food packaging[13]. Food-use pesticide actives have the highest data requirements and, for these,a company will typically generate data from 2-year chronic/cancer bioassays in ratsand mice, developmental toxicity studies in rats and rabbits, multigenerational repro-ductive toxicity studies in rats, and other specialized in vivo studies [14]. These aresimilar to the complete set of preclinical studies that are required for human phar-maceuticals. Because of this large data requirement, these chemicals are ideal foruse in building up toxicity prediction models, since one will have near-complete invitro and in vivo datasets. It is not surprising that pesticide actives have some of thesame features and chemical properties as pharmaceutical products, given that theyare often designed to interact with a specific molecular target.

1.2.2 Pesticidal Inerts

These are all of the ingredients in a pesticide product or formulation other than theactive ingredients. Although they are labeled as “inert”, there is no requirement thatthey be nontoxic. These can range from solvents (e.g., benzene) to animal attractants,such as peanut butter or rancid milk. As with the actives, inerts are classified as

CHEMICAL LANDSCAPE 7

food-use and nonfood-use. Regulatory data requirements are, in general, limited,thus resulting in the availability of little in vivo data [15].

1.2.3 Industrial Chemicals

This is an extremely broad class of chemicals including solvents, detergents, plas-tic monomers and polymers, fuels, synthesis intermediates, and dyes. As such, theyare typically not designed to be bioactive, although many do have bioactivity, some-times through interaction with enzymes and receptors, or by chemically reacting withbiomolecules or via physical interactions (e.g., by disrupting cell membranes). Manyof these compounds are manufactured in very large quantities, posing greater potentialrisks. Such chemicals typically have less stringent regulatory oversight and toxicitytesting requirements but are subject to reporting rules under the Toxics SubstancesControl Act (TSCA). Under TSCA, different reporting requirements and regulatoryscrutiny are applied depending on production volume levels (MPV—medium produc-tion volume chemicals, >25 K tons/year; HPV—high-production volume chemicals,>1 M tons/year). On average, these industrial compounds have lower molecularweight than pesticidal actives or pharmaceuticals, and include many more volatileand semivolatile compounds.

1.2.4 Pharmaceuticals

These are the active ingredients in drugs and, hence, are designed to have specificbioactivity. It is well known that many drugs have toxic side effects, often throughunexpected off-target interactions, and that this is a major economic concern forthe pharmaceutical industry driving up the costs of drug development. In addition,there is increasing concern for toxicity, not only for patients directly taking thedrug, but also for ecological species exposed to these compounds in waste water [16].Despite large amounts of toxicity data submitted to the FDA during the drug-approvalprocess, including clinical data on humans if the drug reaches clinical trials, as wellas additional preclinical toxicity data generated within the pharmaceutical industry,little of these data see the light of day due to confidentiality concerns. As a result,public availability of toxicity data on pharmaceuticals is generally limited to what isavailable in the open literature.

1.2.5 Food Additives/Ingredients

This category includes both natural and synthetic small molecules that are inten-tionally added to food, often to enhance nutritional value (e.g., vitamins), to act aspreservatives, such as in food packaging, or to enhance color or texture. FDA regu-lates allowed tolerances for such chemicals and has the authority to require a batteryof in vitro (primarily genotoxicity) and in vivo toxicity testing to support such reviewswithin the Center for Food Safety and Nutrition (CFSAN) [17]. Such data can bemade publicly available, hence providing a potentially rich source of additional invivo data for computational toxicology modeling.

8 ToxCast

1.2.6 Water Contaminants

EPA regulates chemicals in surface and drinking water, and the relevant chemicalsinclude any of the above categories that enter the water system, as well as metabolitesor degradation products. One example of the latter is disinfection byproducts thatcan result from reactions of chlorine with organic molecules in a drinking watersystem to produce polychlorinated organic compounds. The regulatory authority inthis instance is reactive. First, a chemical has to be detected in water at sufficient levelsto cause some concern, and then sufficient scientific justification must be provided towarrant regulatory action. As a result, toxicity data is generally lacking for many ofthese chemicals, similar to the situation for industrial chemicals.

Because there are so many chemicals to which humans and ecological species arepotentially exposed, it is necessary to prioritize among them when setting up a large-scale screening program such as ToxCast or Tox21. The potential for exposure is onecritical aspect of this prioritization, and these and further chemical use-categories areimportant indicators of the potential for exposure. For instance, any chemical that isdirectly in food or water (e.g., food additives or pesticides that leave residues on cropsor chemicals found in drinking water) would have extra weight in a prioritizationscheme. More detailed “use-categories” are also available to help refine estimatesof potential exposure routes. For instance, if a chemical is found in products towhich children are exposed (e.g., baby bottles, clothing), that chemical would havea heightened priority for screening. There is no general mapping of chemicals touse-categories that is publicly available, but the ExpoCast project, affiliated withthe ToxCast project within EPA, is currently developing such a mapping based onmerging data from many different sources [18].

The lack of data availability on chemicals, whether it is use-category, exposurepotential, or toxicity data, is one of the major drivers of EPA’s HTS computationaltoxicology program [4]. However, the success of this effort also relies upon the abilityto collate as much available data as possible and systematize and format these datainto computable forms to enable modeling efforts to proceed. To provide a centralresource to support this effort, a large-scale database is being created to gather allpublicly available data on chemicals in the environment through the AggregatedComputational Toxicology Resource (ACToR) effort [19]. Thus far, varying amountsand types of data have been compiled on several hundred thousand chemicals col-lected from over 1,000 different sources, consisting of data types that, for example,include information on hazard (i.e., in vitro and in vivo toxicity data), exposure, use,and production.

The above discussion focuses on the chemical landscape of concern for testingfrom a regulatory and use or exposure perspective, but an equally important consid-eration for our long-range purposes is providing adequate coverage of the chemicalfeature and property landscape spanned by the various use-category inventories ofchemicals. Given the intimate relationship between the chemical structure and its bio-logical activity, building a computational toxicology approach capable of predictingtoxicity from HTS bioactivity profiles must provide for sufficient coverage of bio-logical pathways and toxicity mechanisms across the chemical landscape of interest.

THE CHEMICAL LIBRARIES 9

This means that a chemical testing library must also provide sufficient coverage ofthe diverse chemical property and features space capable of adequately probing thisbiological mechanism diversity.

1.3 THE CHEMICAL LIBRARIES

To generate the in vitro dataset required for the computational toxicology approach,a chemical library was assembled, with initial and later testing candidates largelydrawn from the chemical inventories described above. Meeting the initial objectives ofproviding proof-of-concept of the HTS computational toxicology approach requireda strong anchoring to in vivo animal toxicity studies. Hence, selection of the initialtesting set for ToxCast, which we refer to as the Phase I chemical library, wasprimarily driven by the availability of detailed, in vivo toxicity data. The existence ofhigh-quality regulatory guideline studies required for chemical safety evaluation ofpesticide active ingredients by EPA motivated the selection of these compounds tofulfill these data requirement needs. Thus, the Phase I library consisted of 309 uniquechemical substances, with more than 90% pesticides and the rest a mixture of invivo data-rich industrial chemicals such as bisphenol A (BPA) and perfluorooctanoicacid (PFOA).

In vitro HTS testing procedures additionally have a number of practical require-ments that affect the types of chemicals that can be tested using current technologies.Obvious concerns are the solubility of the chemical in aqueous buffer, which is themedium used to conduct HTS testing, as well as dimethyl sulfoxide (DMSO), whichis the near universal solvent used to solubilize test chemicals for testing. Addition-ally, volatility is a concern, since the chemicals are run in batch mode and attentioncannot be paid to special handling requirements for volatile or semivolatile chem-icals. A few physical–chemical property filters, primarily molecular weight (MW)and octanol/water partition coefficient (logP), were used to choose the Phase I chem-icals, but the structures of pesticides are such that most met the criteria for inclusionand were soluble in DMSO. The ToxCast Phase I chemical solutions that under-went the initial round of HTS testing were also post-analyzed by analytical qualitycontrol (QC) methods that are amenable to high-throughput application (primarilyliquid chromatography–mass spectrometry [LC/MS] with gas chromatography–massspectrometry [GC/MS] follow-up for compounds not suitable for LC/MS analysis).Identity and purity were confirmed for over 80% of the Phase I library, with the major-ity of the remaining compounds deemed unsuitable for analysis because they weremetal containing or of low MW. One class of pesticides, consisting of 14 sulfurons,was found to significantly dissociate in DMSO over time, motivating the removal ofthese compounds from further ToxCast testing.

The ToxCast Phase I chemical library, despite its relatively small size, containeda significant amount of chemical and functional diversity, spanning over 40 chem-ical functional classes (e.g., pyrazoles, sulfonamides, organochlorines, pyrethroids,carbamates, organophosphates) and 24 known pesticidal functional classes (e.g.,phenylurea herbicides, organophosphate insecticides, dinitroaniline herbicides). The

10 ToxCast

implication is that although the particular compounds included in this Phase I test setmay not be representative of the larger chemical universe of potential interest suchas antimicrobials, food-additives, drugs, and industrial compounds, the constituentfeatures of these chemicals are potentially capable of representing a much broaderset of chemicals from a wide range of use-categories.

Clearly, however, in order to meet the larger objectives of the ToxCast programfor modeling in vivo toxicity, it is necessary to test larger chemical inventories thatinclude greater representation of the various use-categories of high interest, as wellas the more varied chemical and biological interactions that must be probed andcharacterized in order to build general models for predicting toxicity. Following thetesting of the Phase I library, a much larger chemical collection was assembled basedon these considerations for the dual purposes of expanding the ToxCast test libraryand constructing the EPA contribution to the Tox21 library. Nominations for thislibrary were broadly drawn from the previously described inventories and initiallyexceeded 9,000 compounds. Given the much larger structural diversity of the chem-icals nominated, a greater number of compounds were excluded from considerationon the basis of calculated physical–chemical properties, such as MW, vapor pres-sure, boiling point, solubility, and logP. Finally, practical considerations pertainingto physical samples, such as cost, availability, actual solubility in DMSO, and con-firmed volatility, determined whether or not a compound was included in the finalEPA Tox21 inventory, consisting of more than 3,700 unique chemical substances.

The ToxCast Phase II chemical library, currently undergoing testing, consists of776 unique chemical substances, including nine Phase I compounds used as testingreplicates, drawn from the expanded EPA Tox21 chemical inventory, spanning a muchbroader range of use-cases and chemical structures than in Phase I. For the selection ofPhase II compounds, significant weight was given to those substances with extensivein vivo data available, as well as to toxicity reference substances with well-definedactivities and mechanisms of action. Pursuant to this goal, approximately 30% of thePhase II compounds have in vivo data available from the National Toxicology Programor were generated to meet EPA or FDA’s regulatory requirements for pesticide or foodadditives. However, due to the relative paucity of data for many of the use-categoriesdescribed previously, many of the chemicals in this expanded collection had relativelylittle or no such data available. In addition, higher weight was given to chemicalson high-interest EPA inventories (such as listed above), as well as to chemicalsthat appeared on multiple inventories or use-categories. The Phase II inventory alsobenefitted from an unprecedented collaboration between EPA and the pharmaceuticalindustry, whereby 135 “failed drugs” were donated by six pharmaceutical companies(Pfizer, Merck, GlaxoSmithKline, Sanofi, Roche, and Astellas), along with preclinicaland, in some cases, human clinical data reporting adverse effects. The value of thesedata in extending findings made on chemicals tested only in laboratory animals tothose tested in humans may be significant.

The ToxCast Phase I and Phase II inventories total 1,060 unique compounds.These compounds are being run in the full suite of more than 500 ToxCast assays.Both of these chemical inventories are fully contained within the EPA Tox21 chem-ical inventory, which in turn is a subset of the complete Tox21 collection, totaling

THE BIOLOGICAL ASSAYS 11

approximately 8,200 unique chemical structures. In addition to the failed pharmaceu-ticals, the Tox21 library contains an extensive collection of human pharmaceuticals[20]. Although the Tox21 inventory is much larger and spans much greater chemicaldiversity, this library will only be tested in HTS assays being run at the NCTT and,thus, will have more limited bioactivity profiling data available. On the other hand,the smaller ToxCast Phase I and II chemical inventories will be run in the full suite ofToxCast assays, as well as in the additional Tox21 assays, providing a rich chemicaland biological context for the interpretation of these data. Details of the chemicallibraries can be accessed at http://www.epa.gov/ncct/toxcast/chemicals.html.

An expanded analytical quality control process to ensure that the tested chemicalsare indeed what they are intended to be is accompanying the full Tox21 effort.Careful review and curation of chemical identifiers, including names and ChemicalAbstracts Service Registry Numbers (CASRN), as well as reported purity wereextracted from Certificates of Analysis at the time of procurement. Further reviewand chemical structure annotation of the full Tox21 inventory and component ToxCastinventories were carried out within EPA’s Distributed Structure-Searchable Toxicity(DSSTox) project (see http://www.epa.gov/ncct/dsstox/ for access to downloadablestructure files). Following solubilization in DMSO, the chemical identity, purity, and,concentration are determined by appropriate analytical techniques, including LC/MSand follow-up GC/MS. This analysis will be repeated over the course of the use ofthe library to assess compound stability during testing. While complex and costly,such efforts ensure that biological activity measured in an assay is associated with theappropriate chemical structure and, conversely, those negative results are associatedwith a chemical structure only if that chemical was indeed present.

1.4 THE BIOLOGICAL ASSAYS

Selection of in vitro assays for toxicity testing would be relatively straightforwardif the molecular targets underlying mechanisms of toxicity were known. Advancesin HTS technologies to support the drug discovery industry have provided the toolsto develop assays for large numbers of biological targets, ranging from receptors toenzymes to ion channels and more. If a protein has a defined function, it is safe tosay that an in vitro assay can be built to measure effects of chemicals on that function.Techniques such as surface plasmon resonance or LC–MS–MS exist that measurechemical–protein interactions even when the function is unknown [21]. Beyondassays focusing on specific molecular targets, many assays are available to probephenotypic changes induced in cells by chemical exposure including effects onorganelles and cellular structures such as mitochondria, nuclei, cytoskeleton, andcell membrane. Again, with advances in automated fluorescent microscopy screen-ing platforms and associated imaging algorithms, the ability to measure altered cel-lular phenotypes is almost unlimited. However, assays targeting specific proteinsor cellular phenotypes suffer from our lack of detailed knowledge with respect tomechanisms of toxicity that would guide high-confidence assay selection. Excep-tions to this, while clear, are relatively few and include molecular targets such as the

12 ToxCast

potassium ion channel hERG [22], acetylcholinesterase [23], cytochrome P450s [24],drug transporters [25], nuclear receptors including the androgen, estrogen, and arylhydrocarbon receptor (AhR) [26], as well as the 5-HT2b G-protein-coupled recep-tor (GPCR) [27]. In addition, cellular phenotypic assays for genotoxicity, oxidativestress, mitochondria energy homeostasis, calcium release from intracellular stores,and necrotic and apoptotic cell death can be used to determine toxicity, although withless specificity with respect to molecular target. Acceptance of these as valid toxic-ity targets usually resulted from many years of research, sometimes combined withserendipitous findings. Continuing with this model to complete our understanding oftoxicology would be a long, expensive, and arduous route.

As an alternative approach, a broadly based interrogation of important familiesof biological targets and cellular phenotypes can be conducted efficiently usinghigh-throughput in vitro screens, probing them with large chemical libraries withknown animal and human health effects. The reference in vivo toxicity data for thesechemicals are needed to correlate the in vitro findings with in vivo endpoints. Thetools of computational toxicology can then be applied to analyze, interpret, and modelthe results, ultimately generating predictive signatures of toxicity compatible withcost-efficient, high-throughput assays conducive to screening unknown chemicals.

Defined toxicity targets are usually members of large protein families such asenzymes (e.g., acetylcholinesterase), receptors (e.g., estrogen receptor), and ion chan-nels (e.g., voltage-gated sodium channels). These protein families make up the major-ity of what is called the “druggable genome”, molecular targets thought to providean opportunity for therapeutic intervention and of high interest to the pharmaceuticalindustry [28]. As a result, hundreds of HTS assays have been developed to supportthis drug discovery research. Since the vast majority of these potential drug targetshave been selected based on believed critical roles in various pathological processes,extension of this thinking suggests that such targets could also be involved in toxicitywhen inappropriately perturbed by xenobiotic chemicals. This served as the impetusfor developing a diverse suite of HTS assays to use for profiling the biological activ-ity of chemical libraries by several groups including ourselves through the ToxCastprogram [4, 11, 12].

In vitro HTS assays facilitate the rapid, parallel generation of large numbers ofindividual assay data points through the use of miniaturized assay formats, automatedliquid dispensers, and high-speed plate readers. The miniaturized assay formats areusually in multi-well plates with densities of 96, 384, or 1536 wells per plate ina single, standardized plate footprint, and use total assay volumes ranging from200 �L down to 5 nL. The assay components can be highly varied and depend to alarge degree on the biological target being measured. For instance, an assay measuringkinase activity would have a purified kinase, required cofactors, required substrates,appropriate buffer, and chemical to be tested. In addition, a means of measuring theassay endpoint, here the phosphorylation of the substrate, is required. This couldbe a radioactive or fluorescent technique, a means to detect the loss of ATP or theincrease in ADP, or a separation of the phosphopeptide from the nonphosphorylatedone by means of mobility shift microfluidics assay technology. Cellular phenotypicassays use in vitro cultured cells and automated, fluorescence microscopy to image