Mark Cronin is a Professor of Predictive Toxicology at Liverpool John Moores University, and has over 20 years of experience in the development of in silico approaches to predict toxicity including the use of QSAR, and read-across. He has published three books and over 160 papers in these areas. Judith Madden is a reader in Molecular Design at Liverpool John Moores University and has over 15 years experience in the application of QSAR and in silico techniques to predict the ADMET properties of xenobiotics. She has published one book and over 40 papers in these areas. Steven Enoch is a Post-Doctoral Research Fellow in Predictive Toxicology at Liverpool John Moores University. He is internationally recognised for his research in mechanistic categorisation for toxicology.

Chemical Toxicity Prediction

Category Formation and Read-Across

By Mark T. D. Cronin, Judith C. Madden, Steven J. Enoch, David W. Roberts

The Royal Society of Chemistry

Copyright © 2013 M.T.D. Cronin, J.C. Madden, S.J. Enoch, D.W. Roberts
All rights reserved.
ISBN: 978-1-84973-384-7

Contents

Chapter 1 An Introduction to Chemical Grouping, Categories and Read-Across to Predict Toxicity M. T. D. Cronin, 1,
Chapter 2 Approaches for Grouping Chemicals into Categories S. J. Enoch and D. W. Roberts, 30,
Chapter 3 Informing Chemical Categories through the Development of Adverse Outcome Pathways K. R. Przybylak and T. W. Schultz, 44,
Chapter 4 Tools for Grouping Chemicals and Forming Categories J. C. Madden, 72,
Chapter 5 Sources of Chemical Information, Toxicity Data and Assessment of Their Quality J. C. Madden, 98,
Chapter 6 Category Formation Case Studies S. J. Enoch, K. R. Przybylak and M. T. D. Cronin, 127,
Chapter 7 Evaluation of Categories and Read-Across for Toxicity Prediction Allowing for Regulatory Acceptance M. T. D. Cronin, 155,
Chapter 8 The State of the Art and Future Directions of Category Formation and Read-Across for Toxicity Prediction M. T. D. Cronin, 168,
Subject index, 180,

CHAPTER 1

An Introduction to Chemical Grouping, Categories and Read-Across to Predict Toxicity

M. T. D. CRONIN

School of Pharmacy and Chemistry, Liverpool John Moores University, Byrom Street, Liverpool L3 3AF, England E-mail: m.t.cronin@ljmu.ac.uk

1.1 Introduction – Ensuring the Safety of Exposure to Chemicals

Modern society requires safe chemicals. However, nothing is without risk and there is increasing pressure to identify hazardous chemicals and replace them with those that are more benign. In order to ensure the well-being of their population and the environment, governments enforce legislation to determine the effects of chemicals and ensure that every day accidental or occupational exposure will not cause harm. This is desirable for all substances that man comes into contact with, or that may be released into the environment, whether the substance is in foods, medicines, pesticides, fertilisers, or cosmetic ingredients (amongst many other types of chemicals that are in use). Different regulations are applicable to each type of chemical associated with a particular use.

In order to determine the risk associated with the use of a chemical, a certain amount of information is required. Firstly, a means of defining risk is a pre-requisite. In this context, risk is a function of the intrinsic hazard of a chemical and the exposure. Considering hazard, this can be considered as the ability to cause harm to a species, be that organisms that are deliberately exposed to the chemical or a non-target (for instance, environmental) species. Exposure can be simplistically considered as the quantity of a chemical to which the target and non-target species are exposed.

Within the current definition of risk, information is required regarding the hazards of chemicals; this is provided by the science of toxicology. Assessing the toxicity of chemicals involves determining what the harmful effects of a chemical may be, i.e. toxicity to particular organs, effects to the skin, lethality, tumour promotion and countless others. Assessment normally involves testing for these effects and being able to use the test results in a manner that is protective of man and the environment. The tests and the information they provide need to be scientifically credible and satisfy the needs of the manufacturer, government or regulatory agency that has to interpret them and, ultimately, the user or consumer for whom safety must be assured. The information must be reliable, trustworthy and protective, i.e. precautionary. The need to determine the hazard of a chemical has resulted in the toxicological testing of chemicals for a wide range of specific effects, e.g. the ability to promote tumours. These effects can then be reported and interpreted to identify hazard. The most accepted paradigm for the identification of the majority of toxic effects has been the use of animal testing, through a series of standardised assays. However, the use of animals to identify hazard has received much criticism as being unethical, difficult to extrapolate results and findings to humans, costly and not always capable of identifying subtle or idiosyncratic toxicities. Therefore, for decades, alternatives to animal testing have been sought. Amongst these are the so-called computational, or in silico, models which attempt to draw conclusions regarding the toxicity of a chemical from existing knowledge and/or its chemical structure. It is a selection of these techniques, those involving grouping similar chemicals together and reading across (or interpolating) activity, that form the focus of this volume.

With regard to exposure, a number of issues must be considered. The first is how much of the material is the organism in question exposed to? Also of importance is the time period over which the organism will be exposed, the route and manner of administration, i.e. the formulation (that may affect uptake). Consideration must also be given to whether local, i.e. at the site of exposure (if applicable), or systemic effects are of concern. Assessment of exposure therefore requires appreciation of uptake and bioavailability within the organism. A key principle to remember is that if there is no exposure to a chemical, or it is at a level below that which can cause harm (as defined by the toxicological assessment), there will be no risk.

In silico or computational toxicity prediction methods cover a very wide range of techniques and approaches, some of which are described in Sections 1.1.1 and 1.1.2. However, the main focus of this volume is to describe in detail category formation and read-across.

1.1.1 In Silico Predictions of Toxicity – Grouping, Category Formation and Read-Across

Similar objects tend to have similar properties. Applied to chemistry, this means that for chemicals that can be classed as being similar to other chemicals, we can understand and predict their properties without the need for testing. This fundamental concept has been applied to the prediction of properties and harmful effects of compounds for decades. Thus, being able to form groups of similar compounds (also called categories) becomes a powerful approach. If a compound belongs to a group of compounds with a well categorised toxicological profile, it can be possible to interpolate its activity. These interpolations, (predictions) of toxicity may, when utilised properly, provide hazard information that can be used in the assessment procedure described above. The process of prediction is termed “read-across” as it assumes that activities, toxicities or properties can be read across between compounds within a category.

Two hypothetical examples of read-across are provided in Figures 1.1 and 1.2 — these use data obtained from the OECD QSAR Toolbox version 3.1 (see Section 4.3 for more details). In the first example, Figure 1.1, a read-across prediction of Salmonella typhimurium gene mutation is made for 2-(3-ethylphenyl)oxirane. No mutagenicity data are available for this chemical. However, S. typhimurium gene mutation data are available for four closely related chemicals — termed analogues 1–4. These chemicals are considered “similar” as they all contain an aromatic ring, an epoxy group and limited alkyl substitution. The epoxy group allows the chemical to act as a direct acting electrophile by the SN2 mechanism. All the analogues to the target chemical are positive in the S. typhimurium gene mutation assay, they share the same structural features to the target, hence the read-across prediction for the target is that it will also share the same mechanism and be positive in this assay. This is therefore an example of a “qualitative” read-across.

The second hypothetical example is shown in Figure 1.2. This is a quantitative read-across in that a prediction is made for the acute fish toxicity of 3,4-dimethyl-1-pentanol. 96 hour LC50 values to the fathead minnow (Pimephales promelas) were retrieved for six analogues. These analogues are similar in that they are all simple saturated aliphatic molecules with a hydroxy group. As such they are assumed to act by the same mechanism of action — termed non-polar narcosis — and a good relationship is expected with the logarithm of the octanol-water partition coefficient (log P). Figure 1.2 actually demonstrates the development of a local Quantitative Structure-Activity Relationship (QSAR), the line of best fit between toxicity and log P has the following equation:

Toxicity = 0:963 log P + 0:769 (1)

Where:

Toxicity is the inverse logarithm of the 96 hour LC50 values to Pimephales promelas (millimoles per litre).

Equation (1) has very good statistical fit (the correlation coefficient is 0.99). The target chemical has a log P value of 2.17, hence toxicity is calculated to be 2.86 (log units).

The purpose of these grouping and read-across techniques is described in detail below (Section 1.2), the goal being to predict the effects of compounds directly from chemical structure. The area of “predictive” toxicology, including computational techniques, has seen rapid growth and development over several decades, fuelled by the desire to know more about the properties of chemicals. The history of this area is described in Section 1.3. Read-across is one of the most simplistic approaches to predict toxicity. There are several other levels of computational techniques, increasing in complexity and (probably) sophistication, all of which are well established, which can be applied to predict toxicity and other endpoints; these are summarised in Section 1.1.2.

It is true to say that read-across has grown in popularity due to the realisation that other types of modelling are not likely to be predictive for some sub-acute endpoints, especially those associated with repeat dose toxicity. Some of the key phrases and concepts with regard to read-across are defined in Table 1.1. Whilst read-across is simplistic and, in theory at least, easy to apply, there are a number of drawbacks; such advantages and disadvantages of these approaches are described in Section 1.5.

1.1.2 In Silico Predictions of Toxicity – (Quantitative) Structure-Activity Relationships ((Q)SARs)

Read-across is a fundamental and empirical approach to predict activity. There is also a wide variety of techniques where models have been developed from larger groups of data, which use more detailed descriptors of molecules. These techniques are broadly termed Quantitative Structure-Activity Relationships (QSARs) and include aspects of statistical modelling. Good overviews of the types of QSAR models widely used are available.

Whilst this book focuses on category formation it is wrong to exclude it from the other techniques of toxicity and property prediction, i.e. SARs, QSARs and expert systems. Read-across can be considered to be a simplistic form of QSAR analysis. Indeed, within a category simple QSARs may be formed, should sufficient data be available. Some of the basic definitions relevant to these quantitative approaches are provided in Table 1.2.

1.2 Purpose of Category Formation and Read-Across

The grouping of similar objects together, forming patterns and attempting to make a rational and reasoned interpretation is a sign of an intelligent and numerate mind. The human brain has instinctive capabilities to understand the relationship between objects and develop knowledge. For many years attempts have been made to capture both the processes of knowledge gathering and the knowledge. It is easier to recreate the knowledge than the process of achieving it.

With regard to chemical structure, it is almost instinctive to begin to group “similar” molecules together. Medicinal chemists have for many years been familiar with the concept of identifying similar molecules in terms of pharmacological activity. More recently, development of this concept has been necessitated by the global need to assess the properties and safety of new and, more urgently, existing chemicals. Most significantly, these methods will be used to identify similar chemicals with regard to toxicity. The purpose therefore is to provide information for chemicals where it is missing. These missing data, or data gaps, are most prevalent for existing industrial chemicals, particularly those produced in low tonnage.

The purpose of forming categories and performing read-across extends to the prediction of the effects of chemicals to humans and the environment. It should not be overlooked that grouping and read-across may also be used to predict other effect data, e.g. physico-chemical properties (log P) and many others and properties relating to ADME (such as skin permeability).

The use of category formation and read-across as a technique is seeing growth for a number of reasons, including the following:

• A realisation that many chemicals have missing toxicological and physicochemical data (data gaps) and that these may be crucial for understanding the risk posed by chemicals.

• Chemicals legislation that has forced the need for rapid non-test methods to assess chemical safety.

• Acceptance (or at least partial acceptance) in the past by regulatory agencies of read-across to provide information for regulatory submissions.

• The development of new software e.g. the OECD QSAR Toolbox as well as easier access to toxicological databases and methods to assess and determine the similarity of chemicals.

• Category formation and read-across is seen as a clear, simple and transparent technique to make in silico predictions, thus increasing use by allowing for ease of regulatory acceptance.

• Traditional (Q)SAR methods have often performed poorly for complex toxicological endpoints e.g. repeat dose toxicity and reproductive effects in humans, as well as chronic toxicity in environmental species. Even fundamental properties such as water solubility have proved difficult to predict. Read-across has been shown to provide a more reasonable solution to these problems than (Q)SAR.

• Read-across can allow for predictions to be made with a small number of data (even when there may only be one data point, should that be of high quality and the category very robust).

1.3 History: From Structure-Activity to Grouping

It is almost impossible to provide a detailed history of the use of category formation for toxicity prediction. The reason for this is that little of it is documented in the public literature. However, it is certain that ad hoc grouping and read-across has been undertaken for decades in areas such as medicinal chemistry, toxicity and ADME property prediction. Often this was simply termed structure-activity, or the development of structure-activity relationships.

Considering the literature available, what is certain is that there are only a handful of publications (possibly fewer than twenty) dealing with read-across for toxicity prediction prior to 2005 — see for instance the publications noted in Table 1.3. It is also true that since then the number of publications dealing with category formation and read-across is growing rapidly year-on-year with no sign of a cessation of growth. There is no coincidence in these dates or growth in interest.

Structure-activity has been a cornerstone of understanding and rationalising the toxic effects of chemicals for the best part of a century. For instance, by the 1950s there was a clear appreciation of the relationship between carcinogenic activity of molecules and their shape, structure and properties (cf. Lacassagne et al.) with much fundamental work being performed several decades before then. These early ventures into SAR were not captured electronically until computational technology caught up with the science in the (late) 1980s and early 1990s. Examples of computational methods for predicting toxicity include: the DEREK system which later became DEREK for Windows and more recently DEREK NEXUS, and the United States Environmental Protection Agency’s (US EPA’s) ECOSAR and Oncologic systems. Other good examples of the application of chemistry to explain toxicity were the books by Dupuis and Benezra and Lien. The book by Dupuis and Benezra was remarkably visionary and is often overlooked, it was probably a decade before its contents were taken up again. Basically, the book sets out the SAR behind skin sensitisation (in terms of protein reactivity) which is still the basis of grouping in this area to this day. The other well known and still used example of toxicological SAR from this era was the mutagenic “supermolecule” devised by Ashby and Tennant.

The 1990s saw the growth in computational technology that allowed SAR knowledge to become common in the workplace and eventually on the desktop. At this time however, grouping and read-across (particularly for regulatory purposes) were undertaken, but little recognised, at least in the European Union. Interest in computational methods to predict toxicity grew following the adoption in 2001, by the European Commission, of a White Paper setting out the strategy for a future Community Policy for Chemicals. The main objective of the new chemical strategy was to ensure a high level of protection for human health and the environment. One of the consequences was to require more information regarding the safety and potential harmful effects of chemicals. In the absence of extant toxicological data, there appeared to be few options to obtain such data other than testing or the use of predictions from computational toxicology. Interestingly, as the legislation progressed into regulation, use of all non-test data through the application of Integrated Testing Strategies (ITS), as well as exposure-based waiving, also came to the fore. With regard to computational testing, a group of approximately 60 industry, regulatory and academic scientists met in Setubal, near Lisbon, Portugal in March 2002 to set out the approaches for what became the framework for the regulatory use and acceptance of (Q)SARs. At the time of the Setubal Workshop, reviews of the regulatory use of QSAR made little mention of the existing category formation and read- across methods. Therefore, it can be concluded that the subsequent ten years were the realisation of this approach and its routine application.

(Continues…)Excerpted from Chemical Toxicity Prediction by Mark T. D. Cronin, Judith C. Madden, Steven J. Enoch, David W. Roberts. Copyright © 2013 M.T.D. Cronin, J.C. Madden, S.J. Enoch, D.W. Roberts. Excerpted by permission of The Royal Society of Chemistry.
All rights reserved. No part of this excerpt may be reproduced or reprinted without permission in writing from the publisher.
Excerpts are provided by Dial-A-Book Inc. solely for the personal use of visitors to this web site.