Predicting Chemical Toxicity and Fate - Section 1 ppsx

Predicting
Chemical Toxicity
and Fate
© 2004 by CRC Press LLC
CRC PRESS
Boca Raton London New York Washington, D.C.
Mark T.D. Cronin
Liverpool John Moores University
Liverpool, England
David J. Livingstone
University of Portsmouth
Portsmouth, England
Predicting
Chemical Toxicity
and Fate
© 2004 by CRC Press LLC
This book contains information obtained from authentic and highly regarded sources. Reprinted material is quoted with
permission, and sources are indicated. A wide variety of references are listed. Reasonable efforts have been made to publish
reliable data and information, but the author and the publisher cannot assume responsibility for the validity of all materials
or for the consequences of their use.
Neither this book nor any part may be reproduced or transmitted in any form or by any means, electronic or mechanical,
including photocopying, microfilming, and recording, or by any information storage or retrieval system, without prior
permission in writing from the publisher.
All rights reserved. Authorization to photocopy items for internal or personal use, or the personal or internal use of specific
clients, may be granted by CRC Press LLC, provided that $1.50 per page photocopied is paid directly to Copyright Clearance
Center, 222 Rosewood Drive, Danvers, MA 01923 USA. The fee code for users of the Transactional Reporting Service is
ISBN 0-415-27180-0/04/$0.00+$1.50. The fee is subject to change without notice. For organizations that have been granted
a photocopy license by the CCC, a separate system of payment has been arranged.
The consent of CRC Press LLC does not extend to copying for general distribution, for promotion, for creating new works,
or for resale. Specific permission must be obtained in writing from CRC Press LLC for such copying.
Direct all inquiries to CRC Press LLC, 2000 N.W. Corporate Blvd., Boca Raton, Florida 33431.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for
identification and explanation, without intent to infringe.
Visit the CRC Press Web site at www.crcpress.com
© 2004 by CRC Press LLC
No claim to original U.S. Government works
International Standard Book Number 0-415-27180-0
Library of Congress Card Number 2004043999
Printed in the United States of America 1 2 3 4 5 6 7 8 9 0
Printed on acid-free paper
Library of Congress Cataloging-in-Publication Data
Predicting chemical toxicity and fate / edited by Mark T.D. Cronin and David J.
Livingstone.
p. cm.
Includes bibliographical references and index.
ISBN 0-415-27180-0 (alk. paper)
1. Molecular toxicology. 2. Toxicological chemistry. 3. QSAR (Biochemistry).
I. Cronin, Mark T.D. II. Livingstone, D. (David). III. Title.
RA1220.3.P74 2004
615.9—dc22 2004043999
© 2004 by CRC Press LLC
Dedications
From MC
To AMC and CCFC, for the pleasure and the pain.
© 2004 by CRC Press LLC
Table of Contents
Section 1 Introduction.
Chapter 1 Predicting Chemical Toxicity and Fate in Humans and the Environment — An
Introduction
Mark T.D. Cronin
Section 2 Methodology .

Chapter 2 Toxicity Data Sources
Klaus L.E. Kaiser
Chapter 3 Calculation of Physicochemical Properties
Mark T.D. Cronin and David J. Livingstone
Chapter 4 Good Practice in Physicochemical Property Prediction
Peter R. Fisk, Louise McLaughlin, Rosalind J. Wildey
Chapter 5 Whole Molecule and Atom-Based Topological Descriptors
Tatiana I. Netzeva
Chapter 6 Quantum Chemical Descriptors in Structure-Activity Relationships —
Calculation, Interpretation, and Comparison of Methods
Gerrit Schüürmann
Chapter 7 Building QSAR Models: A Practical Guide
David J. Livingstone
Section 3 QSARs for Human Health Endpoints.
Chapter 8 Prediction of Human Health Endpoints: Mutagenicity and Carcinogenicity
Romualdo Benigni
Chapter 9 The Use of Expert Systems for Toxicity Prediction: Illustrated with Reference
to the DEREK Program
Robert D. Combes and Rosemary A. Rodford
Chapter 10 Computer-Based Methods for the Prediction of Chemical Metabolism and
Biotransformation within Biological Organisms
Martin P. Payne
Chapter 11 Prediction of Pharmacokinetic Parameters in Drug Design and Toxicology
Judith C. Duffy
© 2004 by CRC Press LLC
Section 4 QSARs for Environmental Toxicity and Fate.
Chapter 12 Development and Evaluation of QSARs for Ecotoxic Endpoints: The Benzene
Response-Surface Model for Tetrahymena Toxicity
T. Wayne Schultz and Tatiana I. Netzeva
Chapter 13 Receptor-Mediated Toxicity: QSARs for Estrogen Receptor Binding and

Priority Setting of Potential Estrogenic Endocrine Disruptors
Weida Tong, Hong Fang, Huixiao Hong, Qian Xie, Roger Perkins, and
Daniel M. Sheehan
Chapter 14 Prediction of Persistence
Monika Nendza
Chapter 15 QSAR Modeling of Bioaccumulation
John C. Dearden
Chapter 16 QSAR Modeling of Soil Sorption
John C. Dearden
Chapter 17 Application of Catabolic-Based Biosensors to Develop QSARs for
Degradation
Graeme I. Paton, Jacob G. Bundy, Colin D. Campbell, and Helena Maciel
Section 5 Application.
Chapter 18 The Tiered Approach to Toxicity Assessment Based on the Integrated Use of
Alternative (Non-animal) Tests
Andrew P. Worth
Chapter 19 The Use by Governmental Regulatory Agencies of Quantitative
Structure-Activity Relationships and Expert Systems to Predict Toxicity
Mark T.D. Cronin
Chapter 20 A Framework for Promoting the Acceptance and Regulatory Use of
(Quantitative) Structure-Activity Relationships
Andrew P. Worth, Mark T.D. Cronin, and Cornelius J. Van Leeuwen
© 2004 by CRC Press LLC
Contributors
Romualdo Benigni
Istituto Superiore di Sanità
Rome, Italy
Jacob G. Bundy
School of Biological Sciences
University of Aberdeen

Aberdeen, Scotland
Colin D. Campbell
Macaulay Institute
Aberdeen, Scotland
Robert D. Combes
Fund for the Replacement of Animals in
Medical Experiments
Russell & Burch House
Nottingham, England
Mark T.D. Cronin
School of Pharmacy and Chemistry
Liverpool John Moores University
Liverpool, England
John C. Dearden
School of Pharmacy and Chemistry
Liverpool John Moores University
Liverpool, England
Judith C. Duffy
School of Pharmacy and Chemistry
Liverpool John Moores University
Liverpool, England
Hong Fang
Logicon ROW Sciences
Jefferson, AR, U.S.A.
Huixiao Hong
Logicon ROW Sciences
Jefferson, AR, U.S.A.
Peter R. Fisk
Peter Fisk Associates
Whitstable, England

Klaus L.E. Kaiser
TerraBase, Inc.
Hamilton, Ontario, Canada
David J. Livingstone
ChemQuest, Isle of Wight, England
Centre for Molecular Design, University of
Portsmouth, Portsmouth, U.K.
Helena Maciel
School of Biological Sciences
University of Aberdeen
Aberdeen, U.K.
Louise McLaughlin
Peter Fisk Associates
Whitstable, England
Monika Nendza
Analytisches Laboratorium
Luhnstedt, Germany
Tatiana I. Netzeva
School of Pharmacy and Chemistry
Liverpool John Moores University
Liverpool, U.K.
Graeme I. Paton
School of Biological Sciences
University of Aberdeen
Aberdeen, Scotland
Martin P. Payne
LHASA Ltd.
Department of Chemistry
University of Leeds
Leeds, England

Roger Perkins
Logicon ROW Sciences
Jefferson, AR, U.S.A.
Rosemary A. Rodford
SoloSTAR Ltd.
Bedford, England
© 2004 by CRC Press LLC
T. Wayne Schultz
College of Veterinary Medicine
University of Tennessee
Knoxville, TN, U.S.A.
Gerrit Schüürmann
Department of Chemical Ecotoxicology
UFZ Centre for Environmental Research
Leipzig, Germany
Daniel M. Sheehan
Food and Drug Administration’s National
Center for Toxicological Research,
Jefferson, AR, U.S.A.
Daniel M. Sheehan and Associates,
Little Rock, AR, U.S.A.
Weida Tong
Food and Drug Administration’s National
Center for Toxicological Research
Jefferson, AR, U.S.A.
Cornelius J. Van Leeuwen
Institute for Health and Consumer Protection,
Joint Research Centre
European Commission
Ispra, Italy

Gilman D. Veith
International QSAR Foundation to Reduce
Animal Testing
Duluth, MN, U.S.A.
Rosalind J. Wildey
Peter Fisk Associates
Whitstable, England
Andrew P. Worth
European Center for the Validation of
Alternative Methods
Institute for Health and Consumer Protection,
Joint Research Center
European Commission
Ispra, Italy
Qian Xie
Logicon ROW Sciences
Jefferson, AR, U.S.A.
© 2004 by CRC Press LLC
Foreword
When Corwin Hansch and Al Leo encouraged me in applying quantitative structure-activity
relationships (QSARs) to the screening of environmental hazards, the U.S. Toxic Substances Control
Act was still only a concept, and most QSAR calculations were still being made with a pencil.
Their encouragement included two principles for QSAR along with a word of caution. The principles
were that QSAR ought to be based on well-defined endpoints of intrinsic chemical activities as
well as on molecular descriptors that could be interpreted mechanistically. The word of caution
was that bureaucracies founded on laboratory testing, whether private or a regulatory agency, will
only begrudgingly accept QSAR as a strategic tool in designing chemicals and managing chemical
risks. Looking back over the last three decades, the Hansch/Leo principles for QSAR development
have been largely ignored, if not disputed, by the growing QSAR community, with the possible
exception in Europe where QSAR acceptance criteria will require transparency and a mechanistic

foundation. Only the skepticism toward QSAR itself by our testing-oriented society seems to have
been steadfast over three decades. The increasing costs of testing have produced renewed interest
in more strategic in silico methods at a time when QSAR has been freed from many early
computational barriers. Now more than ever, the scientific community needs an expert summary
of QSAR methods like this book.
The guiding principles for QSAR development were intended to aid in the discovery of useful
and robust models. The literature is replete with more than 10,000 QSAR correlations and models,
yet few of them are useful enough to sway the skeptics. Still, progress in QSAR research can be
measured by its own critics and the changing nature of their skepticism. The “yes-but” skeptics are
particularly instructive to me. In 1974, our research plans faced the criticism, “yes, QSAR may be
able to predict some chemical properties, but it will never be able to predict bioaccumulation of
chemical residues.” In 1981, we faced, “okay, QSAR may be able to predict bioconcentration
potential, but it will never be able to predict toxicity.” When the acute toxicity models appeared,
we were confronted by “yes, QSAR may be able to predict some ecotoxicity endpoints, but it will
never predict chronic toxicity in mammals.” Today, as the first mechanistic QSAR models are
emerging for chronic reproductive effects and mutagenicity, this historical perspective on the QSAR
skeptics serves as benchmarks for progress, if not encouragement.
Chemical reactivity in biological systems is far more complex than 20th century computational
capabilities could have allowed one to address in quantitative terms. The rapid progress in computing
power over the last decade enabled a steady stream of new computational methods in QSAR to
emerge. Unfortunately, these new capabilities were not matched with the generation of high-quality
biological databases needed to reveal systematic variation within heterogeneous chemical invento-
ries. While many combinatorial problems in QSAR are likely to challenge computer sciences for
years, present computer capabilities are sufficient to make future QSAR progress limited mostly
by the databases for relevant, well-defined endpoints.
Our QSAR program at the Duluth, MN, U.S.A., laboratory focused on well-defined ecotoxi-
cological endpoints that could be used directly in regulatory decisions. Our proof-of-concept paper
in 1979 for estimating the bioconcentration potential required only a minimal database. Since then,
many researchers have contributed to the evolution of bioaccumulation models and to extend them
from simple screening-level methods for new chemicals to more exact estimates of tissue residues

for risk assessments. In contrast to the bioconcentration database, the creation of the Duluth
ecotoxicity database involved a multimillion dollar investment and dozens of scientists over most
of a decade. Finding chemicals with common toxicity pathways to build mechanistic structure-
toxicity relationships required better diagnostic bioassays, including behavioral symptomology (fish
acute toxicity syndromes) and joint-toxicity studies. Our first paper on acute toxicity in 1983 was
delayed almost 3 years due to rejections from toxicological journals based on our use of the term
“narcosis” in describing reversible, baseline lethality — a criticism that lingers today in the health
© 2004 by CRC Press LLC
research community. The dozens of more recent supporting papers on baseline toxicity and the
even larger toxicity database created by Terry Schultz at the University of Tennessee (Knoxville)
should be sufficient to overcome the skeptics of acute toxicity predictions so that the full attention
of effects research can focus on important chronic toxicity endpoints.
The European Chemical Industry Council-led analysis of the state of QSAR in Setubal, Portugal
(March 2002), concluded that QSARs for biodegradability were still the largest research gap in
exposure research. Developing QSARs for important chemical properties progressed rapidly in the
1980s, but developing structure-biodegradability models has been paralyzed by a lack of systematic
databases. Fortunately, in 1985 Hiroshi Tadakoro at the Hita laboratory in Japan recognized the
need for a biodegradation database, and his team devoted more than a decade to systematically
testing chemicals using activated sludge. Almost immediately after the Hita database was made
available, the first QSAR screening models for biodegradability began to appear at scientific
meetings. Again, these advances illustrate the importance of generating systematic data on crucial
endpoints in the overall progress of predictive methods. Finding such endpoints and understanding
how they can be reliably used in risk management is the central research challenge for QSAR.
Once identified, QSAR progress seems to depend only on government funding to generate the
systematic data needed to build acceptable QSARs for the respective endpoints.
The estimation of lethality and biodegradability directly from chemical structure has been one
of the important first steps in applying QSAR to risk management. Shifting our focus to chronic
effects and persistence of chemicals will require us to cross some exciting new frontiers, not the
least of which will be the merger of metabolism and effects models as QSAR is incorporated into
systems biology. To meet these challenges, scores of chronic toxicity pathways will have to be

described, and “-omics” technology promises to open new doors in clustering chemicals by common
toxicity pathways for QSAR modeling. With metabolic activation a critical step in many pathways,
metabonomics offers unprecedented capability for identifying the key molecular initiating events
for chronic effects, many being the new well-defined endpoints QSAR needs for chronic hazard
identification. It is hoped that this book will play an important role in advancing QSAR in the face
of healthy skepticism, and will bring greater attention to the need for high-quality data in strategic
testing.
Dr. Gilman Veith
Duluth, MN
© 2004 by CRC Press LLC
Preface
The motivation for this book was stimulated by a one-day meeting, “Modelling Environmental
Fate and Toxicity,” organized by the BioActive Sciences Group of the Society of Chemical Industry.
The meeting was chaired by Drs. Mark Cronin and Dave Livingstone and held in London on March
27, 2001. The speakers at the meeting were drawn from industry and academia and described how
computational methods could be applied to predict the toxicity and fate of chemicals in the
environment. The meeting itself was well attended and was particularly timely. It coincided with
an upsurge of interest in this area due both to legislative changes and the commercial possibilities
of predicting toxicity and fate.
We are moving into a new era that is computationally rich and data poor. Modeling of toxicity
is much easier than it was a decade ago because of increased computational power and greater
availability of software to calculate descriptors of molecules (some of which is freely download-
able). However, we must never lose sight of the fact that good models require high quality input
data, and preferably large amounts of it. Neither should we forget that predictive techniques are
empirical models to be used; they should not be seen as an academic exercise. In commissioning
this book we attempted to bring together a collection of chapters that would assist future modelers
develop meaningful predictive techniques. This was always hoped to be a practical and didactic
book, there are plenty of published reviews in all areas covered in the book. All authors were
encouraged to make recommendations for the use of the methods and techniques described. The
editors support the recommendations and hope they will be applied and useful to the next generation

of modelers.
Mark Cronin and Dave Livingstone
July 2003
© 2004 by CRC Press LLC
Acknowledgments
The editors wish to thank the BioActive Sciences Group of the Society of Chemical Industry
(London, England) for the original opportunity to put on the one-day meeting that stimulated this
volume. Without the group’s foresight, encouragement, and organization, none of this would have
been achievable. We also wish to thank the authors who have cheerfully contributed to the book,
accepted our criticism, and made helpful comments. Finally we are extremely grateful to Taylor
and Francis for originally commissioning the book and CRC Press for the final opportunity to
publish it.
© 2004 by CRC Press LLC
List of Abbreviations
␹ Randi
´
c branching index, or molecular connectivity
␬ Kappa shape index
␴ Hammett constant
␮ Dipole moment
⌿ Wave function characterizing the state of a system state
AAR Activity-Activity Relationship
ADME Absorption, Distribution, Metabolism, and Excretion
AFP a-Feto Protein
AM1 Austin Model 1
A
max
Maximum acceptor superdelocalizability
ANN Artificial Neural Network
AO Atomic Orbital

AR Androgen Receptor
ARI Automated Rule-Induction
ATSDR Agency for Toxic Substances and Disease Registry
BBB Blood-Brain Barrier
BCF Bioconcentration factor
BESS Biodegradability Evaluation and Simulation System
BgVV (German) Federal Institute for Health Protection of Consumers and
Veterinary Medicine
BMD BenchMark Dose
BMF Biomagnification Factor
BRM Carcinogenic potency in mice
BRR Carcinogenic potency in rats
B3LYP Hybrid density functional theory ab initio calculation method
c Cluster (molecular connectivity)
C Concentration (of a drug or toxicant)
C Corrosive
CADD Computer-Aided Drug Design
CAS Chemical Abstract Service
CASE Computer Automated Structure Evaluation
CCOHS Canadian Center for Occupational Health and Safety
CDER Center for Drug Evaluation and Research
Cl Clearance
CM Classification Model
CM1 Charge Model 1
CM2 Charge Model 2
CODESSA COmprehensive DEscriptors for Structural and Statistical Analysis
CoMFA Comparative Molecular Field Analysis
COMPACT Computerized Optimized Parametric Analysis of Chemical Toxicity
COREPA Common REactivity PAttern
CPSA Charged Partial Surface Area

CRADA Cooperative Research and Development Agreement
CT Classification Tree
CV Cross Validation
D Distribution coefficient
DBP Disinfection By Products
DEREK Deductive Estimation of Risk from Existing Knowledge
© 2004 by CRC Press LLC
DES Diethylstilbestrol
DFT Density Functional Theory
DSC Differential Scanning Calorimetry
DSL Domestic Substance List
DSS Decision Support System
DSSTox Distributed Structure-Searchable Toxicity
E Hepatic extraction ratio
EA Electron Affinity
E
1/2
Half-wave oxidation potential
⌬E Difference in the energies of the highest occupied and lowest unoccupied
molecular orbitals
EC Extended connectivity
ECB European Chemical Bureau
EC
50
Concentration causing 50% reduction in a specified effect
ECOSAR Syracuse Research Corporation program to predict environmental toxicities
ECVAM European Centre for the Validation of Alternative Methods
EDC Endocrine Disrupting Chemical
EDKB Endocrine Disruptor Knowledge Base
EDPSD Endocrine Disruption Priority Setting Database

EDSTAC Endocrine Disruptors Screening and Testing Advisory Committee
E
HOMO
Energy of the Highest Occupied Molecular Orbital
E
kin
Kinetic energy of a system
E
LUMO
Energy of the Lowest Unoccupied Molecular Orbital
EN Electronegativity
EPIWIN Estimations Programs Interface for Windows
E
pot
potential energy of a system
ER Estrogen Receptor
e-state Electrotopological state index
E
tot
Total energy of a system
EU European Union
FDA Food and Drug Administration
GI Gastrointestinal
GST Glutathione S-Transferase
FIRM Formal Inference-based Recursive Modeling
H Harary index
H Hamilton operator
HD Hardness
HENRYWIN Syracuse Research Corporation program to predict Henry’s law constant
HESI Health and Environmental Sciences Institute

HF Hartree-Fock
⌬HF Heat of Formation
HPLC High Performance Liquid Chromatography
HPV High Production Volume
HPVC High production volume chemical
HQSAR Hologram Quantitative Structure-Activity Relationship
ICG
50
Concentration causing 50% inhibition of growth
ILSI International Life Sciences Institute
IP Ionization potential
Ipb Isopropylbenzene
Ind Induction
ITC Interagency Testing Committee
© 2004 by CRC Press LLC
IUCLID International Uniform Chemicals Information Database
JME Java Molecular Editor
JRC Joint Research Centre of the European Commision
K Partition coefficient
K
a
Equilibrium acid ionisation
KBS Knowledge-Based Systems
K
i
Inhibition constant
K
m
Binding constant
K

mxa
Cuticular Matrix-Air partition coefficient
KNN K-Nearest Neighbors
K
oa
Octanol-Air partition coefficient
K
oc
Soil-Water partition coefficient normalised for organic carbon content
K
om
Soil-Organic matter partition coefficient
K
ow
Octanol-Water partition coefficient
K¢¢
¢¢
ow
Apparent Octanol-Water partition coefficient
KOWWIN Syracuse Research Corporation program to predict octanol-water partition
coefficient
K
p
Skin permeability coefficients
K
pa
Plant-Air partition coefficient
K
vs
Vegetation-Soil partition coefficient

LCAO Linear Combinations of Atomic Orbitals
LC
50
Lethal Concentration for 50% of animals
LD
50
Lethal Dose for 50% of animals
LFER Linear Free Energy Relationship
LSER Linear Solvation Energy Relationship
LNO Leave-N-Out
Log Logarithm to base 10
LOO Leave-One Out
LRA Linear Regression Analysis
MHBP Molecular Hydrogen Bond Potential
MLP Multilayer Perceptron
MLPot Molecular Lipophilicity Potential
MLR Multiple Linear Regression
MNDO Modified Neglect of Diatomic Overlap
MO Molecular Orbital
MOPAC Molecular Orbital PACkage
MP Melting Point
MPBPVP Syracuse Research Corporation program to predict melting point, boiling
point, and vapor pressure
MR Molar Refractivity
MS-WHIM Molecular-Surface Weighted Holistic Invariant Molecular
MultiCASE Multiple Computer Automated Structure Evaluation
MW Molecular Weight
NC Non-corrosive
NCI National Cancer Institute
NCTR National Center for Toxicological Research

NDDO Neglect of Diatomic Differential Overlap
NIOSH National Institute for Occupational Safety and Health
NN Neural Network
NOEC No Observed Effect Concentration
NR Nuclear Receptor
© 2004 by CRC Press LLC
NTP National Toxicology Program
OECD Organization for Economic Co-operation and Development
OM1 Orthogonalization Model 1
OM2 Orthogonalization Model 2
OPPT Office of Pollution Prevention and Toxics
OPS Optimum Prediction Space
ORMUCS Ordered MUlticategorical Classification method using the Simplex technique
p Path (molecular connectivity)
P Partition coefficient
PAH PolyAromatic Hydrocarbon
P
alkb
Promoter of the alkb gene
PBPK Physiologically Based PharmacoKinetic
PBT Persistent, Bioaccumulative, and toxic
pc Path-cluster (molecular connectivity)
PC Principal Component
PCA Principal Component Analysis
PCB PolyChlorinated Biphenyl
PCR Regression on Principal Components
P-gp P-Glycoprotein
pH Negative logarithm of the hydrated proton concentration
PLS Partial Least Squares
PM Prediction Model

PM3 Parameterized Model 3
PM5 Parameterized Model 5
PMN PreManufacture Notification
PNN Probabilistic Neural Network
PNSA Partial Negative Surface Area
PPAR Peroxisome Proliferator Activated Receptor
PPSA Partial Positive Surface Area
PSA Polar Surface Area
Q
2
Leave-one out or cross-validated R
2
Q
A
Net atomic charge on atom A
Q
h
Hepatic blood flow
QM Quantum Mechanical
QSAR Quantitative Structure-Activity Relationship
QSBR Quantitative Structure-Biodegradability Relationship
QSPKR Quantitative Structure-Pharmacokinetic Relationship
QSPR Quantitative Structure-Property Relationship
R and R
2
Multiple correlation coefficient and its square
RBA Relative Binding Affinity
REACH Registration, Evaluation, and Authorization of CHemicals
rms Root-mean-square
RTECS Registry of Toxic Effects of Chemicals

SA (Sub-)Structural Alerts
SAS Statistical Analysis System
S
aq
Aqueous solubility
SAR Structure-Activity Relationship
SCF Self-Consistent Field
SDF Structure Data File
SHBG Sex Hormone Binding Globulin
SMILES Simplified Molecular Line Entry System
© 2004 by CRC Press LLC
SRC Syracuse Research Corporation
t ½ Half-life
TD
50
Dose required to halve the probability of animals remaining tumourless
TGD Technical Guidance Document
TI Topological Indices
TOPKAT TOxicity Prediction by Komputer-Assisted Technology
TPSA Topological Polar Surface Area
TSCA Toxic Substances Control Act
UNIFAC Uniquac Functional-group Activity Coefficient (where UNIQUAC = Universal
Quasi-Chemical)
U.S. EPA United States Environmental Protection Agency
V
d
Volume of distribution
V
m
Molecular Volume

vol molar volume in cm
3
mol
–1
W Wiener index
WLN Wiswesser Line Notation
WMPT Waste Minimization Prioritization Tool
WSKOWIN Syracuse Research Corporation program to predict water solubility
© 2004 by CRC Press LLC
SECTION 1
Introduction
© 2004 by CRC Press LLC
C
HAPTER
1
Predicting Chemical Toxicity and Fate
in Humans and the Environment —
An Introduction
Mark T.D. Cronin
CONTENTS
I.Introduction
A.History of Predictive Methods for Toxicology and Fate
B.Motivation for Predicting Toxicity and Fate
1.Computer Models Provide a Prediction of Toxicity and Fate
2.Public Pressure to Reduce Animal Testing
3.Legislation
4.Filling Data Gaps
5.Cost of Testing — Finance and Time
6.Reaction to New Toxicological Problems
7.Designing New Compounds

8.Increased Understanding of the Biology and Chemistry
C.The Cornerstones of Predictive Methods for Toxicity and Fate
1.Biological Activity
2.Description of the Compounds
3.Statistical Techniques
D.How to Use Predictions
E.General Reference Sources
References
I. INTRODUCTION
Chemicals are able to bring about both desirable and undesirable effects on organisms to which
they are exposed, and the actions of medicines and poisons and toxic agents have been recognized
for thousands of years. As a result of industrialization, modern man and the environment is now
exposed to increasing numbers of chemicals. Because of their potential hazard, there is an appreci-
ation of the requirement to assess the effects of these chemicals. Since chemical structure was
elucidated (for a very brief history see Table 1.1), the relationship between chemical structure and
© 2004 by CRC Press LLC
Table 1.1 Summary of the Key Historical Events (Scientific and Sociological) That Have Given Rise
to the Modern Science of Predictive Toxicology
Year/Era Event
c. 5000 B.C.
and earlier
Knowledge of animal venoms and poisonous plants
3000 B.C. Egyptians recognized the toxic effects of some substances; Menes, first of the Pharaohs, studied
and cultivated poisonous plants
1550 B.C. Ebers Papyrus describes over 800 recipes for poisons
c. 300 B.C. Theophrastus, a pupil of Aristotle, referenced poisons and was later executed by poison
Early 1500s Paracelsus determined that specific chemicals were actually responsible for the toxicity of a plant
or animal poison
Middle Ages Poisoning an accepted fact of life; Shakespeare’s Romeo begs to the apothecary, “A dram of
poison, such soon-speeding gear, As will disperse itself through all the veins, That the life-weary

taker may fall dead ”
Early 1800s Orfila is credited with founding toxicology (i.e., the correlation between the chemistry and biology
of poisons)
1860’s Transition from alchemy to chemistry: structure of benzene proposed, the periodic table determined
1863 Observation by Cros that the toxicity of alcohols decreased with their water solubility
1868 Crum-Brown, and Fraser concluded that physiological activity is a function of chemical constitution
1893 Richet observed toxicity to be inversely related to solubility
1899–1901 Meyer and Overton independently proposed that narcosis is related to partitioning between oil
and water phases; assessment of partitioning using olive oil and water
1939 Ferguson proposed solubility cutoff for acute toxicity
1940 Hammett published Physical Organic Chemistry showing that the effects of substituents could be
quantified
1964 Hansch and coworkers used regression analysis and descriptors for the hydrophobic, electronic,
and steric properties of molecules to formulate a QSAR
1970s Growth in the development of QSARs
1976 The U.S. Toxic Substances Control Act acts as the spur to find methods of predicting toxicity
1981 Könemann demonstrated that the acute toxicity to fish of non-reactive narcotic compounds may
be modeled by hydrophobicity
1980s Rapid expansion of computational power makes molecular graphics and modeling, as well as
multivariate statistical analysis, practical; computational power allows for the commercial
development of software for descriptor calculation, molecular modeling, and toxicity prediction
1980s 3D QSAR techniques allow for the modeling of receptor-mediated effects, including toxicity and
metabolism
1980s Development of mechanism-of-action-based QSARs for acute toxicity, progression from narcotic
mechanisms to reactive mechanisms
1980s Application of QSAR techniques to a wide variety of toxic and fate endpoints such as
carcinogenicity, irritation, biodegradation, and bioaccumulation
1980s Creation of the fathead minnow toxicity database
Late 1980s SMILES for molecular description developed and becomes widely used
1990s Application of QSAR techniques to a broader range of toxicological endpoints such as skin

sensitisation, percutaneous absorption, skin and eye irritation, and carcinogenicity
1990s A growth in concern over animal usage, and resultant public pressure, increases the commercial
and legislative requirement for alternative methods to animal testing
Early 1990s The Internet becomes a reality. Storage, searching and analysis of large amounts of chemical
and biological information become trivial. Desktop computing becomes the standard.
Mid 1990s The National Toxicology Program’s carcinogenicity prediction challenge highlights the difficulty of
estimating this endpoint
Mid 1990s Concern over endocrine disruption brings receptor-mediated modeling techniques into mainstream
toxicity prediction
Late 1990s User-friendly software to calculate large numbers of molecular descriptors from 2D structure
becomes widely available both commercially and from the Internet (e.g. DRAGON)
2000 Release of the EPISUITE software freely downloadable from the Internet
2000 Solving of the human genome and application of toxicogenomics
2000 Tetrahymena database reaches 2000 compounds tested
2000 Pharmaceutical development regularly utilizes combinatorial chemistry, high throughput screening
and virtual library design; the interest in predictive ADMET grows
2001 European Union’s White Paper on the Strategy for a Future Chemicals Policy stimulates further
interest in the validation and application of QSARs to predict toxicity and fate
2001 Bioterrorism, including the use of toxic agents, becomes a reality
© 2004 by CRC Press LLC
biological activity has intrigued scientists. Latterly, it has been recognized that the investigation of
chemical structure — biological activity relationships (or structure-activity relationships [SARs]) is
more than an academic exercise. They may provide useful tools to solve real world problems, such
as the requirement for information regarding the effects of chemicals on man and the environment.
This book intends to provide a starting point for those interested in the prediction of the toxicity
and fate of chemicals to humans and the environment. SARs and, more frequently, quantitative
structure-activity relationships (QSARs) provide methods to predict these endpoints. A brief history
of the area, the driving forces, and basis of the topic is provided in this chapter. Further chapters
(2 to 7) describe the methods to develop predictive models; the application of models to human
health endpoints (Chapters 8 to 11); their application to environmental toxicity and fate (Chapters 12

to 17); and the use of predictive models (Chapter 19), adoption by the regulatory authorities (Chapter
19), and validation (Chapter 20).
A. History of Predictive Methods for Toxicology and Fate
It would be wrong to consider the history of predictive toxicology in complete isolation from
other scientific and sociological events. The interest and ability to predict the toxicity and fate of
chemicals has a number of drivers and has been influenced by key individuals, organizations, com-
mercial and welfare pressures, and legislation. The prediction of effects has relied on advances in all
the areas of biology, chemistry, and informatics, as well as benefiting in particular from the substantial
advance in computer technology. To describe these events in detail is clearly beyond the scope of this
book, and certainly exceeds the capability of this author! As a starting point for historians, various
papers provide a good overview of past achievements (Kubinyi, 2002; Lipnick, 1999; Rekker, 1992;
Schultz et al., 2003; van de Waterbeemd, 1992). From a personal point of view, some of the key
events that influenced the science are summarized in Table 1.1 and commented upon below.
It is often overlooked that much of the basis of modern drug design can be traced back to
toxicological research performed in the 1890s. Indeed a cynic might suggest that there has been
little progress since the work of Richet, Meyer and Overton! Certainly the finding of Könemann
(1981) reinforced and quantified these findings, but this was 80 years after the original work. From
this reinvention of acute toxicological QSAR there has been progress through class-based to
mechanism-of-action modeling, leading to the development of more global approaches to toxicity
prediction (see Chapter 12). This progress has been underpinned by the development of reliable
and diverse databases of toxicity values, those having been developed for the fathead minnow
(Pimephales promelas) by the U.S. Environmental Protection Agency (EPA; Mid-Continent Ecol-
ogy Division, Duluth) (Russom et al., 1997) and freshwater ciliate (Tetrahymena pyriformis) at the
University of Tennessee, Knoxville (Schultz, 1997), being of considerable importance.
In other areas of predictive toxicology and fate, progress has been steady, and spurred on in
recent years by many of the legislative and commercial pressures mentioned in Table 1.1. Progress
and interest in the prediction of human effects and pharmacokinetics has been complemented by
advances in chemo-informatics. This has resulted in a large number of commercially available
expert system approaches to toxicity prediction (see Chapter 9) and algorithms for the prediction
of absorption, distribution, metabolism, and excretion (ADME; see Chapters 10 and 11).

B. Motivation for Predicting Toxicity and Fate
There is no single motivation for wishing to predict the toxicity and fate of chemicals — the
desire to do so varies from user to user. The following describe some of the key reasons for which
models have been developed (in no particular order). Naturally, the drivers for predicting these
endpoints are related closely to the historical development of the science and most of the criteria
listed below are related in some manner.
© 2004 by CRC Press LLC
1. Computer Models Provide a Prediction of Toxicity and Fate
It may seem too obvious to state, but it is fundamental that computer models allow for the
effects of chemicals (i.e., physicochemical properties, toxicological activity, distribution, fate, etc.)
to be predicted. These predictions may be obtained from a knowledge of chemical structure alone.
For most methods, provided that the chemical structure can be described in two (or occasionally
three) dimensions, the effects may be predicted. Information regarding the chemicals may be gained
without chemical testing, or even the need to synthesize the chemical.
2. Public Pressure to Reduce Animal Testing
For several decades there has been growing public concern regarding the use of animals in
testing, especially in toxicology and medical research. The concern over animal welfare has been
concentrated in Europe and in the U.K. in particular. This has resulted in the boycotting of
companies, organizations, and individuals associated with animal testing. At the most radical edge,
terror campaigns have been mounted to target particular workers and even financial institutions.
Campaigners for animal welfare cite a number of approaches to reduce and ultimately replace
animal tests. These include the use of validated alternative tests such as in vitro and cell culture
techniques, as well as the computer-aided prediction of toxicity. The status of alternative techniques
is well reviewed by Worth and Balls (2002).
There is clearly a role for predictive techniques in the replacement of animal tests, either as
stand-alone methods, or more commonly as part of a tiered assessment strategy. Chapter 18
describes the integration of computational methods, in combination with the judicious use of
physicochemical properties, as viable alternatives to animal testing. The strategies described in
Chapter 18 are being integrated into international guidelines for the toxicological assessment of
endpoints such as irritation.

3. Legislation
Much of the legislation that has underpinned the use of computational methods to predict
toxicity is summarized in Chapter 19. Governmental policy in both the European Union (EU) and
North America has encouraged and, in some cases, mandated the use of computational techniques
to predict toxicity. For instance, the EPA has utilized QSARs to assist in the pre-manufactory
notification of new chemicals, especially where no toxicity data exist. This requirement for models
has stimulated considerable progress in the prediction of acute toxicity for environmental endpoints.
Elsewhere, EU directives decree that animal tests should not be used if a suitable, validated,
alternative (which includes computer models) is available. Generally national and international
regulatory agencies and other competent bodies require and utilise predictive techniques to prior-
itize, classify, and label compounds for testing.
4. Filling Data Gaps
Approximately 100,000 separate chemicals may be released into the environment annually; it
is frightening to consider that reliable toxicity data exist for only a tiny proportion of these
chemicals, probably less than 5%. The percentage of chemicals with a complete set of reliable
toxicity data (i.e., across a broad spectrum of environmental and human health effects) is consid-
erably less than 5%. Computer-aided prediction of toxicity has the capability to assist in the
prioritisation of chemicals for testing, and for predicting specific toxicities to allow for labeling.
Chapter 19 describes these activities in more detail. As the reliability of models for toxicity
prediction increases, there will undoubtedly be increased use for the filling of data gaps.
© 2004 by CRC Press LLC
5. Cost of Testing — Finance and Time
Toxicological testing is costly financially as well as in terms of the animals used and time taken.
Even a simple ecotoxicological assay may cost several thousand dollars, and a two-year carcino-
genicity assay may cost several million dollars. The cost of testing impacts business in a number
of ways. For existing chemicals, cost (and obtaining resources) is clearly prohibitive to the filling
of data gaps for the many compounds that have not been tested. With regard to the development
of new chemicals (e.g., pharmaceuticals), the cost of toxicological testing of large numbers of lead
compounds is prohibitive both financially and in the time it may take to obtain a full profile of a
chemical. In both these areas, there are clear advantages to the use of methods to predict toxicity

and fate (i.e., costs are greatly reduced). This should allow for faster and less expensive product
development, as well as assessment of environmental effects.
6. Reaction to New Toxicological Problems
As we become exposed to more xenobiotic chemicals, and we learn more about the human
genome, we will almost certainly become aware of an increasing number of toxic effects. A good
example is provided by the issues associated with endocrine disruption, which came to the forefront
during the 1990s. The use of computer-aided modeling in this area is well described in Chapter 13.
This shows that the development of computational techniques not only allows for the prediction
of the potential for estrogenicity to be made, but also allows for rational direction to be given to
testing programs. The modeling of estrogenicity is an excellent example of the application of tools
developed primarily for drug design (i.e., 3D QSAR and Comparative Molecular Field Analysis
[CoMFA]) for the purpose of toxicological prediction. As an integral part of the modeling process,
hypotheses have been built about the receptor (in this case the estrogen receptor) and have been
tested in a rational manner. This has greatly extended the knowledge available from testing above
that which would be gained from testing of random compounds.
7. Designing New Compounds
Drug and pesticide design has sought for many years to optimize activity and efficacy. As well
as designing “in” attractive features of molecules, it is now possible to design “out” toxic features,
or those associated with an unwanted ADME profile. To achieve this there is increasing use of
commercial expert systems (described in Chapter 9) in various industries. There are obvious benefits
to this process, such as savings in time and money throughout product development, as well as
making lead optimization more relevant and directed.
8. Increased Understanding of the Biology and Chemistry
An often-ignored spin-off from the development of QSARs is the increased understanding they
can provide in both the biology and chemistry of active compounds. In the modeling of acute
toxicological endpoints much has been gained regarding mechanisms of action. For many modeling
approaches, it may be assumed that compounds fitting the same QSAR are acting by the same
mechanism of action (see Chapter 12 or Schultz et al. [2003] for more details). This has allowed
workers to define the chemical domain of certain mechanisms. Compounds that do not fit a particular
model also become of interest. Such compounds, known as outliers, suggest that they are acting

by a different mechanism of action, or may be broken down rapidly either by chemical effects (i.e.,
degradation) or biological effects (i.e., metabolism). There are countless examples where knowledge
of biology and chemistry has been advanced by modeling in the field of toxicological and fate
effects.
© 2004 by CRC Press LLC
C. The Cornerstones of Predictive Methods for Toxicity and Fate
One of the interesting aspects of predictive toxicology is that it brings together a large number
of interfacing sciences and challenges. Figure 1.1 outlines the three key areas, and some of the
important factors.
There are various definitions of what constitutes a computer-aided toxicity prediction method.
It is true to say that all techniques are based upon the relationship of the biological activity of one
or more molecules with some aspect of chemical structure. This broadest of definitions normally
requires three components for a prediction method:
1. Some measure of the activity (i.e., toxicity or fate) in a biological or environmental system
2. A description of the physicochemical properties and/or structure of a molecule
3. A form of statistical relationship to link activity and descriptors
The relationship between these three areas is given in Figure 1.2 and a brief introduction to each
is provided below.
1. Biological Activity
There are a remarkable number and diversity of activities that have been modeled successfully.
The activity to be modeled may be a toxicity to an environmental organism or to man, the fate of
a pollutant in an ecosystem, or the pharmacokinetic properties of a xenobiotic in man. To model
any of these activities, relevant biological data for the endpoint are required. Chapter 2 describes
how toxicological and fate information for chemicals may be obtained from external sources such
as the open literature, databases, and the Internet. QSAR developers may also have their own data
to model.
A key to the successful development of a predictive model for toxicity is the use of high-quality
data. A definition of the quality of data is provided in Chapter 2 and discussed further by Cronin
and Schultz (2003) and Schultz and Cronin (2003). However the quality of data is described, it is
Figure 1.1 Interfacing sciences behind predictive toxicology.

CHEMISTRY
Molecular Modeling
Measurement and Calculation of
Molecular Descriptors
STATISTICS
CHEMO- & BIO-INFORMATICS
Multivariate Analysis
Derivation of Expert Systems
Data Storage and Retrieval
BIOLOGY / ACTIVITY
Assessment of Toxicology and Fate
Mechanism of Action
Toxicogenomics
© 2004 by CRC Press LLC
clear that reliable data from a single, well-defined protocol and measured to a high standard are
required to build high-quality models. While the use of poor quality data is not precluded in
modeling, an appreciation of the limitations of the data must be apparent to the modeler and
ultimately to the user of the model.
From a broad (statistical) point of view, there are two types of activity that may be modeled.
These are continuous and categorical data — as such they are statistical terms, rather than relating
specifically to biological activity. Continuous data are numeral values that typically describe a
concentration that elicits a particular effect. Most often this is a 50% effect concentration such as
LC
50
, LD
50
, EC
50
, IGC
50

, etc. Occasionally it may be an effect brought about by a particular
concentration such as the number of revertants in the Ames test. It must be stressed that when
dealing with effect concentration, or equi-concentration, data, concentrations in molar units must
be used. The use of molar values allows for the comparison of the effect of one molecule with
another, rather than the weight of molecules. A subgroup of continuous data for modeling is not
necessarily biological in nature. It includes physicochemical properties themselves (such as partition
coefficient, melting point, etc.), diffusion characteristics (such as the flux through a particular
membrane), fate descriptors (including persistence and bioaccumulation), pharmacokinetic prop-
erties (bioavailability and metabolism), and a multitude of other activities. As with the modeling
of effects, if any of these data include a concentration, that concentration will be required in molar
units.
The second type of activity data to be modeled are those data that are described as being
categoric or ordinal. Typically these are yes/no data that classify a chemical as being active or
inactive in a particular toxicological assay (e.g., carcinogenic or non-carcinogenic), or may be
descriptive such as high or low bioavailability, or rapidly or slowly degraded. Occasionally there
may be further grading (e.g., non-active, weakly active, moderately active, and highly active). For
some activities and endpoints a classification may be assigned on the basis of a particular number
of criteria, which may originally be developed from quantitative measures of toxicity. It is possible
that a chemical may be classified differently according to its toxicity on the basis of national
regulations (e.g., EU vs. U.S.). For such data it is essential that models are based on consistent and
reliable data, and that the basis for the classification is consistent and understood.
Figure 1.2General scheme for a computer-aided toxicological prediction method (as developed from Figure 1.1).
External Validation
(Chapter 20)
Internal Evaluation
(Chapter 7)
Topological Indexes
(Chapter 5)
Physicochemical
Properties

(Chapter 3, 4)
Molecular Orbital
Properties
(Chapter 6)
Description of
the Molecules
Statistical
Technique
Predictive
Method
Activity to be
Modeled
Environmental
Toxicity
(Chapters 2, 12, 13)
Environmental
Fate
(Chapters 14-17)
Human Health
Effects
(Chapters 8, 9)
Human
Pharmacokinetics
(Chapters 10, 11)
© 2004 by CRC Press LLC