The Integrated Use of Alternative Approaches for Predicting Toxic Hazard

The Report and Recommendations of ECVAM Workshop 81,2

Reprinted with minor amendments from ATLA 23, 410-429

Martin D. Barratt3, José V. Castell4, Mark Chamberlain3, Robert D. Combes5, John C. Dearden6, Julia H. Fentem7, Ingrid Gerner8, Alessandro Giuliani9, Tim J.B. Gray10, David J. Livingstone11, W. McLean Provan12, Fons A.J.J.L. Rutten13, Henk J.M. Verhaar14 and Peter Zbinden15
3Environmental Safety Laboratory, Unilever Research, Colworth House, Sharnbrook, Bedford MK44 1LQ, UK; 4Unidad de Hepatologia Experimental, Hospital Universitario La Fe, Avda de Campanar 21, 46009 Valencia, Spain; 5FRAME, Russell & Burch House, 96-98 North Sherwood Street, Nottingham NG1 4EE, UK; 6School of Pharmacy, Liverpool John Moores University, Byrom Street, Liverpool L3 3AF, UK; 7ECVAM, JRC Environment Institute, 21020 Ispra (Va), Italy; 8Bundesinstitut für gesundheitlichen Verbraucherschutz und Veterinärmedizin (BgVV), Thielallee 88-92, D-14195 Berlin, Germany; 9Istituto di Ricerca sulla Senescenza, Sigma-Tau, Via Pontina, km 30.400, 00040 Pomezia, Italy; 10Sanofi Research Division, Alnwick Research Centre, Alnwick, Northumberland NE66 2JH, UK; 11ChemQuest, Cheyney House, 19-21 Cheyney Street, Steeple Morden, Herts. SG8 0LP, UK; 12ZENECA Central Toxicology Laboratory, Alderley Park, Macclesfield, Cheshire SK10 4TJ, UK; 13TNO Nutrition and Food Research Institute, Division of Toxicology, P.O. Box 360, 3700 AJ Zeist, The Netherlands; 14Research Institute of Toxicology (RITOX), Utrecht University, P.O. Box 80.176, Yalelaan 2, 3508 TD Utrecht, The Netherlands; 15SIAT, Missionsstrasse 60, CH-4055 Basel, Switzerland

1ECVAM - The European Centre for the Validation of Alternative Methods. 2This document represents the agreed report of the participants as individual scientists.

Address for correspondence: Martin D. Barratt, Environmental Safety Laboratory, Unilever Research, Colworth House, Sharnbrook, Bedford MK44 1LQ, UK

Address for reprints: ECVAM, TP 580, JRC Environment Institute, 21020 Ispra (VA), Italy

Preface

Preface

This is the report of the eighth of a series of workshops organised by the European Centre for the Validation of Alternative Methods (ECVAM). ECVAM's main goal, as defined in 1993 by its Scientific Advisory Committee, is to promote the scientific and regulatory acceptance of alternative methods which are of importance to the biosciences and which reduce, refine or replace the use of laboratory animals. One of the first priorities set by ECVAM was the implementation of procedures which would enable it to become well-informed about the state-of-the-art of non-animal test development and validation, and the potential for the possible incorporation of alternative tests into regulatory procedures. It was decided that this would be best achieved by the organisation of ECVAM workshops on specific topics, at which small groups of invited experts would review the current status of various types of in vitro tests and their potential uses, and make recommendations about the best ways forward (1).

The workshop on The Integrated Use of Alternative Approaches for Predicting Toxic Hazard was held in Angera, Italy on 23-27 January 1995, under the chairmanship of Martin Barratt. Ways in which quantitative structure-activity relationships (QSARs) might be used in the design, evaluation, and validation of in vitro tests, and in the selection of appropriate test chemicals for validation studies, were discussed in depth at the workshop. The conclusions and recommendations of the workshop participants are summarised in the final section of this report.

Introduction

The principles underlying the development of QSARs are based on the premise that the properties of a chemical are implicit in its molecular structure. It therefore follows that, if the mechanism for the activity of a group of chemicals can be elucidated, and relevant parameters can be measured or calculated, then, in theory, a quantitative structure-activity relationship (QSAR) can be established. For a QSAR to be valid and reliable, the activities of all the chemicals covered must be elicited by a mechanism which is both common and relevant. Attempts to derive QSARs for data sets where this is not the case have not always been successful.

The same principles need to be applied to the development of in vitro toxicity tests. However, in many cases, such principles are overlooked. As a result, some alternative tests predict endpoints which are different from those which they claim to predict (because the wrong mechanism has been identified) or they cannot predict endpoints accurately for all classes of chemicals (because a common mechanism is lacking).

In terms of developing integrated approaches to the use of alternative methods for the prediction of toxic hazard, it is important to consider how QSARs, computer modelling techniques (for example, physiologically-based pharmacokinetic [PBPK] models) and in vitro methods might be used in combination. Thus, the main objective of the workshop was to bring together experts from these, typically separate, areas, to initiate preliminary discussions on what may be possible, and to make suggestions and recommendations for future actions considered to be necessary for the development of such integrated testing strategies.

In contrast to the other ECVAM workshops held up until now, which have mainly concerned alternatives to animal testing for the prediction of specific toxicological endpoints, the participants at the workshop on The Integrated Use of Alternative Approaches for Predicting Toxic Hazard were from diverse scientific backgrounds, spanning mathematics, computing, chemistry, biochemistry, cell biology, and toxicology. Thus, to establish common ground for the subsequent discussions, the first two days of the workshop took the form of an informal symposium, consisting of five sessions. The presentations given by the participants are summarised after the introductory sections on QSAR, PBPK modelling, and in vitro tests.

Quantitative Structure-Activity Relationships

A QSAR is a model which relates the biological activities of a series of similar compounds to one or more physicochemical or structural properties of the compounds (2-4). In this definition, "similar" means having the same mechanism of action, but not necessarily having a related chemical structure. However, it is often difficult (if not impossible) to determine the mechanism of action, whereas it is usually less difficult to establish chemical similarity. Hence, QSARs are generally developed for sets of chemically-similar compounds (congeneric series), in the hope that they will also have the same mechanism of action. Any compounds which do not act by the same mechanism are likely to fit the correlation only poorly and to appear as "outliers".

When a chemical is administered to an organism, two events must occur for a biological response to be triggered. Firstly, the compound has to be transported to the site of action (the "receptor"); secondly, it must interact with the target in an appropriate manner.

In some biological systems, for example fish, transport has been found to be largely a process of partitioning between aqueous and lipid compartments, and can often be modelled by the partition coefficient (P). Typically, the octanol-water partition coefficient (sometimes called Kow) is used, although P values obtained with other solvent pairs have been used (5).

Interaction with the target ("receptor") is governed largely by two factors:

  1. the size and shape of the xenobiotic, which will control how well the molecule fits the receptor site; and
  2. the nature and relative positions of appropriate functional groups on the molecule, which will affect the type and strength of the interaction with complementary groups on the receptor.

Many physicochemical properties can be used to model receptor interaction; for example, molar volume (MV) is an overall measure of molecular size, and the energy of the lowest unoccupied molecular orbital (ELUMO) is a crude measure of the electron-accepting ability of a compound.

It is important to note that, in effect, a QSAR is concerned with the change in biological activity brought about by a change in chemical structure. The latter may alter the rate at which a xenobiotic arrives at the receptor site and its ability to interact with the receptor. Hence, extremely complex biological endpoints can usually be correlated with simple physicochemical properties. An example of this is given by Könemann (6), concerning the acute toxicities (LC50 values, in mM/litre) to the guppy of 50 industrial chemicals, of various chemical classes:

log (1/LC50) = 0.871(log P) - 4.87
(n = 50; r2 = 0.976; s = 0.237)

Logarithmic expressions are usually used because QSARs represent free energy changes. The fraction of the variation in biological activity that is accounted for by the term(s) on the right-hand side of the equation is indicated by the square of the correlation coefficient (r2); s is the standard error of the estimate.

For chemicals whose aquatic toxicities can be described by log P alone, it is believed that their toxic effects generally arise from membrane perturbation. For chemicals which are more reactive, some term(s) modelling receptor interaction or even metabolism can be incorporated into the equation. An example is the QSAR developed by Jäckel and Klein (7) to describe the toxicity of anilines to the rat:

log (1/LD50) = 1.061 log P - 0.21 (log P)2
- 0.305 ELUMO - 0.038 MV + 2.46

(LD50 in mols/kg; n = 29; r2 = 0.901; s = 0.108)

QSARs like the one given above can be developed only if the biological data are on a continuous scale. Sometimes this is not the case; for example, a chemical may be classified as either carcinogenic or non-carcinogenic. In such cases, other statistical techniques (for example, discriminant analysis) must be applied (8), in which the physicochemical properties of the compounds are used to discriminate between activity and inactivity. If more than two such properties are used, they can be combined into principal components, and a plot of two major principal components may distinguish active from inactive compounds. An example of this, taken from the work of Barratt on skin corrosivity (9, 10), is shown in Figure 1.

Figure 1: Plot of the First Two Principle Components of log P Values, Molecular Volumes, Melting Points and pKa Values, for 27 Organic Acids

Corrosive chemicals , non-corrosive chemicals
P = octanol/water partition coefficient.

There are a number of caveats to the use of QSARs. It is important that these are borne in mind; otherwise, the application of QSARs may not be valid.

  1. Currently QSARs can be applied only to pure compounds. Some work has been undertaken on their application to mixtures but, as yet, there are no firm guidelines for their use in this respect.
  2. The set of compounds used to derive the QSAR (the "training set") should be selected from knowledge, or assumptions, of a common mechanism of action. The training set should also be chosen to cover appropriate ranges of parameter values (for example, a suitable range of log P values).
  3. The parameters ("descriptors") used in the QSAR equation should, if possible, be selected on the basis of mechanistic considerations. Alternatively, they should be amenable to mechanistic interpretation.
  4. For comparative purposes, concentrations or doses must be in molar, not weight, units.
  5. Each QSAR should be validated by investigating its predictive ability using a different set of compounds (the "test set"), which should cover the same ranges of parameter space.
  6. The QSAR must not be applied outside of its domain of validity (i.e. outside of the parameter space covered by the training set).

Physiologically-based Pharmacokinetic Modelling

To develop a full understanding of the biological responses resulting from exposure to toxic or therapeutic doses of chemicals, it is necessary to understand the factors which determine the tissue-delivered dose of the active substance, in addition to the interactions of the chemical with tissues. There are two types of pharmacokinetic models currently available which predict the time-dependent concentration of a xenobiotic in animals and man.

The classical approach fits exponential functions to time-dependent plasma concentration data to produce a single- or multi-compartment model. Although these models are useful in many clinical situations, this procedure does not describe a physiological system, and hence such models are poor for extrapolating beyond the experimental range, or between species.

PBPK models separate the body into a number of tissues, or groups of tissues, interconnected by the vascular system. Therefore, the whole body can be modelled by performing mass balances on each organ. In practice, however, besides the cost of developing such a complex model, this level of detail is often not required. For example, some models have been used successfully where those tissues with similar perfusion rates and xenobiotic solubilities are combined, leaving only three major tissue compartments: a richly perfused compartment (liver, kidney, etc.), a slowly perfused compartment (skin, muscle, etc.), and a fat compartment. Particular organs of interest, such as those which are capable of metabolism, or which represent the target of a toxicological or pharmacological effect, may be removed from these compartments and treated separately. In addition, PBPK models can be constructed for different routes of xenobiotic exposure, and thereby lend themselves to the comparison of tissue-delivered xenobiotic doses for several exposure routes.

As long ago as 1924, Haggard proposed the use of PBPK models to describe the disposition of volatile chemicals (11). Models for volatile chemicals have since been refined and extended (12-14). Physiological modelling of drugs was proposed by Teorell in 1937 (15), and has been extended and modified by Bischoff (16), Dedrick (17), and Roland (18). Several of these early workers were unable to test the usefulness of their proposed models, since too many complex equations were involved which could not be solved. Nowadays, powerful desktop computers are capable of performing these calculations (19), which should encourage the more widespread use of such models in toxicology and risk assessment.

In general, the data required to develop PBPK models are of three types:

  1. species-specific physiological parameters (for example, organ volumes and perfusion rates for laboratory animals and man);
  2. partition coefficients (for example, partitioning between the blood and tissues) for the chemical for each species; and
  3. species-specific metabolic parameters (for example, Km and Vmax) for each metabolic pathway.

The physiological information required is readily available from the literature. Partition coefficients can be measured by vial equilibration techniques (20), and metabolic constants can be estimated from experiments with isolated cells, tissue fractions, etc. or from in vivo studies.

PBPK models can thus incorporate a large number of variables. The precision with which any one variable needs to be defined is dependent upon the intended use of the PBPK model. Sensitivity analysis is a useful technique for guiding the investigator to those variables which are most likely to affect the chosen endpoint of a PBPK model, and provides an understanding of the strengths and limitations of the PBPK model being studied. For example, consider the situation where estimations are required of the hepatic metabolism of a volatile chemical which is metabolised rapidly but has low solubility in blood. At low doses, the rate of metabolism in the liver will be perfusion-limited and, therefore, will be highly dependent upon the ventilation-perfusion rate and the solubility of the chemical in blood, rather than on the Km and Vmax for the metabolic pathway. In contrast, at the highest doses, when the perfusion limitation is removed, the critical variable will be Vmax.

PBPK models have several advantages over classical pharmacokinetic models, including the ability to scale physiological parameters between species, to predict time-dependent organ concentrations across doses and species, to incorporate species-specific metabolic rates, and to compare target organ bioavailability between different exposure routes. The benefits of using PBPK modelling for estimating the target organ concentrations of chemotherapeutic agents intended for clinical use, and the target organ concentrations of industrial chemicals for risk assessment, have been well documented (21, 22). Thus, PBPK modelling is a particularly valuable tool for evaluating tissue concentrations under various exposure conditions in different animal species, and has potential applications in many areas of toxicology and pharmacology.

In Vitro Tests

The use of in vitro systems (subcellular fractions, cell lines, primary cell cultures, tissue slices, organ cultures, etc.) as research tools in toxicology is widespread (23, 24). In recent years there has also been a move towards standardising the various techniques and procedures available, with the concomitant development and evaluation of numerous in vitro 'tests' which have been proposed as potential replacements for various animal tests currently required by regulatory authorities (25, 26).

There are four main factors driving the development of in vitro toxicology (27):

  1. the need for relatively simple, inexpensive, and efficient systems for testing the large number of chemicals for which toxicological data are required;
  2. public and legislative pressures to reduce animal experimentation;
  3. a need for a better understanding of the mechanisms of chemical-induced toxicity, in order to improve the scientific basis of current risk assessment procedures; and
  4. the desire to use human cells and tissues whenever possible, for example to undertake inter-species comparisons of xenobiotic metabolism and toxicity.

In vitro systems are ideally suited to investigations of the molecular, cellular, and physiological mechanisms of chemical-induced toxicity (which cannot readily be studied in vivo), for known target organ and target species toxicity studies, and for answering specific questions about toxic effects. The main justification for developing in vitro toxicity tests is that they will make toxicology a more scientifically-based practice (28). Understanding the mechanisms by which chemicals cause cell and tissue damage, and the reasons for the increased susceptibility of certain species, individuals or tissues to particular chemicals, will markedly improve our ability to predict the possible consequences of human and/or environmental exposure to them. Perhaps the greatest advantage of in vitro toxicity tests is that human cells and tissues can be used (29), thereby obviating the need to extrapolate data from laboratory animals to man. Most progress in the use of in vitro test systems has occurred in the areas of local (ocular and dermal) toxicity (30, 31) and target organ toxicity, especially in testing for hepatotoxicity (32), nephrotoxicity (33), and neurotoxicity (34). Despite numerous comparisons of in vitro cytotoxicity data with rodent LD50 values (35), there has been little progress in the use of in vitro tests for predicting overall systemic toxicity (27), for which the need to incorporate data on biokinetics, in particular, is a critical issue.

It is becoming increasingly apparent that the development and incorporation of step-wise testing strategies, combining experimental data from a range of alternative methods (physicochemical techniques, QSAR, metabolic and kinetic modelling, and in vitro tests), provides the most effective way forward for trying to predict toxicity while at the same time reducing the number of laboratory animals used for testing purposes (36). Several examples of such hierarchical testing schemes have been published recently (37-39).

Principles and Applications of QSAR in Toxicology

The first session of the informal symposium was opened by John Dearden (QSAR in Toxicology: A Historical Perspective), who reviewed the development of QSAR from 1839 to the present day. He illustrated how hydrophobic, then electronic and steric, parameters were recognised as important in controlling the biological effects of xenobiotics. The pivotal role of Hansch's 1962 paper (40) was discussed in some detail. Examples were presented which demonstrated the importance of hydrophobicity parameters, such as log P, in a range of mammalian and aquatic toxicities. In aquatic systems, the relationship between toxicity and hydrophobicity is generally linear for unreactive chemicals (those acting by non-polar narcosis, 41), whilst in mammals the correlation tends to be parabolic, such as that for the acute toxicity of barbiturates to the mouse (42):

log (1/LD50) = 1.02 log P - 0.27 (log P)2 + 1.86
(n = 13; r2 = 0.852; s = 0.113)

For more reactive chemicals, additional electronic terms need to be incorporated.

David Livingstone (Problems and Pitfalls in QSAR) drew attention to one of the earliest publications on QSAR, which contained the equation ø = f (C), where ø is a biological response and C is some descriptor of chemical structure (43). The paper stated that, at the time (1868), it was not possible to describe changes in chemical structure with sufficient precision to make the function in the equation the subject of calculation. The same is still largely true today. Examples were given of the use of molecular modelling and computational chemistry to calculate a range of physicochemical descriptors, which can be used to augment the information available in tables of substituent constants for characterising chemicals which have no 'standard' common parent (44). However, such an approach can raise problems, since it requires the selection of a single conformation for the molecular structure (45).

As well as problems which may be encountered in the description of molecular structure, there are pitfalls governing the choice of techniques to be used in generating QSARs (46). Regression analysis is still probably one of the most common methods used in the establishment of QSARs, although there are several alternative approaches (47). To date, the application of QSAR methodology to mammalian toxicological data has not been particularly widespread, but there does not appear to be any compelling reason why QSAR approaches could not be applied more extensively in areas of toxicity testing other than ecotoxicology (4). For example, computational chemistry techniques have been applied to the analysis of teratogenic effects (48).

In the final presentation of the first session, Mark Chamberlain (The Application of QSAR Models to In Vitro Testing) demonstrated how QSAR models can be used in various ways, in addition to their use for predicting toxic effects, employing a QSAR describing the eye irritancies of neutral organic chemicals (49) as an example. From the identification of molecular features or parameters associated with the toxic endpoint, a QSAR can be used to help select an appropriate set of chemicals for testing the robustness of the model, based on the full range of those parameters. The uncertainty arising from the biological variability in a particular toxic endpoint, and its effect on the predictive value of a QSAR, can be evaluated.

Proposals for the use of QSAR models in evaluating the appropriateness of in vitro methods, in terms of the operation of relevant mechanisms, were described. A QSAR simulation was presented, in which the omission of one key independent variable resulted in a marked decrease in the correlation obtained. The analogy was made between this model and the expected, similar, behaviour of a hypothetical in vitro assay in which only part of the in vivo mechanism was represented. It was concluded that the integration of QSARs with in vitro methods would also help in designing in vitro tests which had better predictive powers, as well as enabling a more rational selection of test chemicals for use in validation studies (50).

Short-term Toxicity

Ingrid Gerner (Development of SAR Models for the Introduction of Alternative Methods into Acute Toxicity and Local Irritancy Testing Strategies) described the use of physicochemical and toxicological data on approximately 600 new chemicals submitted under the Chemical Substances Act 1982 (based on Directive 67/548/EEC) in developing SAR models and a prototype expert system for predicting local irritancy and acute toxicity. The aim is to evaluate whether the local irritant/corrosive properties of chemicals can be predicted using SARs based on physicochemical properties only, or using a combination of SARs and validated alternative methods (51, 52). The testing strategy developed is shown in Figure 2. Current EU and OECD regulations permit the use of alternative methods for labelling corrosives or severe irritants only (53, 54).

Figure 2: Testing Strategy for Acute Toxicity, Corrosivity and Irritancy

Strategy for the introduction of alternative methods (SAR models, in vitro tests) into acute toxicity, corrosivity and local irritancy testing, developed following evaluatiuon of data submitted to the Chemicals Department of the German Federal Institute for Health Protection of Consumers (I. Gerner, BgVV, Germany).

I-VII: animal tests conducted according to EU/OECD test guidelines.

EU risk phrases: R34 = causes burns; R35 = causes severe burns; and R41 = risk of serious damage to eyes.

= alternative approaches

Chemicals are characterised by specific partial structures, empirical formulae, and molecular weights; thermodynamic properties are characterised by vapour pressures, melting points, and boiling points; toxicokinetic properties are characterised by lipid solubilities, water solubilities, and octanol/water partition coefficients; and the physical interactions and chemical reactions with water (as a typical biological medium) are characterised by the surface tensions of saturated solutions at 20°C, the pH values of saturated solutions at 20°C, and the potential for hydrolysis upon contact with water. Corrosivity or severe local irritancy leading to labelling with R35 ('causes severe burns'), R34 ('causes burns') or R41 ('risk of serious damage to eyes'). EU risk phrases can be predicted on the basis of these properties; less irritant effects, or the absence of effects on skin or eyes, can also be predicted.

Martin Barratt (SAR Models for Skin Sensitisation, Skin Corrosivity and Skin Irritation) discussed a method for the identification of potential contact allergens, based on the premise that, for a chemical to be a contact allergen, it must be able to penetrate the skin and be able to react covalently with skin proteins, either directly or after metabolic activation. A set of around 50 structure-activity rules (called "structural alerts" or "toxicophores"), relating to the protein reactivities of chemicals, have been programmed into the knowledge-based expert system DEREK (Deductive Estimation of Risk from Existing Knowledge; 55). These structural alerts are used to identify the reactivity component of skin sensitisation. The skin permeability of the chemical is estimated using a QSAR (56), in which the logarithm of the permeability coefficient is related to log P (positive dependence), molecular volume (negative dependence), and melting point (negative dependence). At present, there are no suitable in vitro methods available for the identification of skin sensitisers.

QSARs were presented for the skin corrosivities of organic acids, bases, and phenols, based on the mechanistic hypothesis that, for the manifestation of corrosivity, chemicals must penetrate the skin and then cause a cytotoxic response. Skin permeability is again modelled by log P, molecular volume, and melting point, whilst the cytotoxicity component is represented by pKa (49). Preliminary analysis of skin irritation data (obtained from the European Centre for the Ecotoxicology and Toxicology of Chemicals [ECETOC]) for a set of neutral and electrophilic organic chemicals suggests that irritancy can also be modelled on the basis of skin permeability represented by log P (positive dependence) and molecular volume (negative dependence), together with dipole moment (positive dependence) as a reactivity parameter. However, the predictive power of this QSAR model is poor, due mainly to high variability in the biological data (M.D. Barratt, unpublished observation).

The presentation by Henk Verhaar (The Distinction of Chemical Classes in Acute Toxicity via Expert Systems and Multivariate Analysis) outlined the methods used in aquatic toxicity testing for dividing pollutants into four broad classes. The primary reference point is the so-called 'baseline toxicity' or non-polar narcosis, which assumes a non-specific mode of toxic action (57). The internal chemical concentration in fish at the LC50 for baseline toxicity (class I chemicals), normalised for lipid content, is essentially constant at about 50mM (58-60). For these chemicals, the external median lethal concentration (LC50) is governed by the partitioning of the chemical between water and the fish, and can be adequately predicted by log P. Chemicals causing toxicity which exceeds the baseline level can be separated roughly into three classes using a rule-based system (61): class II -- those causing polar narcosis; class III -- reactive chemicals (mainly small electrophiles); and class IV -- chemicals, such as pesticides, which act through specific targets.

Of 2000 EU High Production Volume (HPV) chemicals, about 1000 could not be classified on the basis described above because they were inorganics, mixtures, or were not characterised properly. Of the remainder, 27% were assigned to class I, 8% to class II, and 42% to classes III or IV; 23% are not yet classified (62). A preliminary separation of 172 class I and class II chemicals by partial least squares (PLS) discriminant analysis (63), using hydrogen bond donor/acceptor properties and properties from electrostatic potential-fitted charges (64) as the prime descriptors, was achieved with no misclassifications (65). These methods are considered useful both for the classification of chemicals in aquatic toxicology and in the elucidation of modes of toxic action.

In Vitro Methods

Julia Fentem's presentation (In Vitro Tests for Skin and Eye Irritation) reviewed the status of alternative methods for predicting dermal and ocular toxicity. The majority of in vitro tests for skin corrosivity, and skin and eye irritation are, in reality, cytotoxicity tests. A recent prevalidation study of in vitro methods for skin corrosivity involved comparison and further evaluation of the transcutaneous electrical resistance assay (using rat skin), CORROSITEX™ and an assay using Skin2 (66). To varying extents, these tests model the dermal penetration of chemicals and their subsequent cytotoxicity, both of which appear to be key events in corrosivity as judged by the successful application of QSARs to discriminating between corrosives and non-corrosives (9).

In vitro alternatives for skin irritation testing include physicochemical methods (for example, determination of the pH-acid/alkaline reserve and SKINTEX™), cell culture methods, and the use of epidermal strips, organ cultures, and human skin models (67). Keratinocytes are believed to be the initial key targets in the complex series of reactions which constitute contact irritant dermatitis. They produce, and can be stimulated to release, a large number of inflammatory mediators (68). Although the biochemical mechanism of skin irritation remains to be elucidated, it is believed that epidermal cytokines, for example, interleukin-1(alpha), play a pivotal role at an early stage (68). These phenomena are still insufficiently understood to provide the basis of a model system for predicting human skin irritation hazard accurately.

A similar picture exists with alternative methods for the prediction of eye irritation. Over 60 in vitro tests have been proposed as replacements for the Draize rabbit eye test (69). A number of these alternatives have been included in recent validation studies, for example those conducted under the auspices of the EC/British Home Office and COLIPA. The complexity of the ocular irritation process, and our limited knowledge of its mechanistic basis in vivo, limit our ability to develop relevant and reliable in vitro tests for eye irritation. The inadequacies of many of the existing alternative methods have prompted the use of batteries of in vitro tests (70). It has been suggested that, as a minimum, the test battery should be able to detect chemical-induced cytotoxicity, non-cytotoxic inflammation, and recovery from damage (71).

Future requirements for the development of in vitro tests include more attention being focused on the elucidation of mechanisms of toxicity in vivo, the use of QSAR in the rational selection of test chemicals for the development and evaluation of in vitro models, and the development of integrated testing strategies which combine information on toxicological effects from different sources, including human studies where appropriate.

Tim Gray (Screening for Systemic Toxicity) drew attention to the fact that, whilst most efforts to develop alternatives have been directed towards the replacement of skin and eye irritation testing, the vast majority of experimental animals (about 95%) are used in evaluating systemic toxicity (72). In order to make a real impact on animal usage, the balance needs to be redressed. The questions asked in testing for systemic toxicity are broader than for skin and eye irritation; for example, what will be the target organ, what types of effects will be produced and at what exposure levels, and what is the role of metabolism? Frequently, effects will be produced only following chronic exposure. The use of in vitro approaches to address these questions poses major challenges from a primary screening perspective. Currently, the best opportunities for employing in vitro methods successfully are in 'secondary screening', i.e. within chemical classes having known target organ specificities. Good examples are the use of bone marrow cultures for assessing the relative myelotoxic potentials of anticancer agents (73, 74), and the application of primary hepatocytes in screening for peroxisome proliferators (75). In these cases, enough is known about the in vivo toxicity relative to the response in vitro to make a clear and meaningful interpretation of the in vitro data.

The greatest benefits in this area are likely to be realised through bringing in vitro toxicology into closer liaison with chemical and drug discovery programmes. In vitro screening of new chemicals is much more difficult because of the multiplicity of potential target sites, and the lack of information to guide or interpret the in vitro approach. Much more work is needed to elucidate the factors which determine the target sites of toxicity, in particular on the dependence of biokinetic factors on physicochemical and structural considerations. A greater understanding of the mechanisms underlying the induction of cellular injury is required, together with better indices of tissue- and cell-specific toxicity in vitro. Data held by companies which could enable structure-target relationships to be developed should be utilised for this purpose. Research in these areas needs to be given priority over attempts to validate existing 'imperfect' in vitro tests for systemic toxicity, and greater priority relative to the continuing effort expended on developing and validating alternative tests for eye and skin irritation.

QSAR and Modelling of Genotoxicity

Bob Combes (Mechanisms of Mutagenesis and Carcinogenesis) described tests for the detection of DNA damage arising in both somatic and germ line cells, which has the potential to give rise to human diseases such as cancer, atherosclerosis, and inherited syndromes (76). Various types of DNA damage were described, such as the covalent binding of chemicals to DNA bases via interactions with several different nucleophilic sites of attack (77). Genotoxins can be either electrophilic per se or following metabolic activation. DNA damage can give rise to microlesions -- so-called 'point mutations' (changes in base sequence which do not impair replication), and macrolesions -- visible changes to chromosomes, many of which are lethal to cells and which are used to indicate the induction of DNA damage (78). Such lesions can occur either before or following various complicated repair processes. Since some chemicals induce different types of lesions preferentially, regulatory testing generally entails the use of a tiered system of tests, in which in vitro screens for gene mutation and chromosomal damage are used to establish genotoxic potential, the toxicological significance of which is confirmed (or otherwise) in one or more in vivo assays (79, 80).

The existence of both genotoxic and non-genotoxic carcinogens is well-established from extensive studies in rodents (81). A defined mechanistic framework for genotoxicity testing is emerging, and most, if not all, genotoxic carcinogens can be detected using a combination of salmonella mutagenicity testing and SARs (82). Most chemicals belonging to the non-genotoxic category are not detected by any of the in vitro assays and exert species-, strain-, sex- and tissue-specific effects in rodent assays, often only at very high dose levels (83, 84).

With genotoxins, considerable progress has been made in identifying structural alerts which contribute to toxicity (85, 86). Their activity is dependent on several factors (the stability of the ultimate reactive metabolite, its accessibility to DNA, and the subsequent repair of the resulting DNA adduct), all of which are related intrinsically to chemical structure. Such information, together with known pathways for metabolic activation, is being used to identify toxicophores responsible for the carcinogenicity and mutagenicity of several compound groups, in conjunction with the expert system DEREK (87). The resulting rules are being used to predict the activities of related chemicals possessing the identified toxicophores, but with other structural features which possibly modulate their activities (88).

Alessandro Giuliani's presentation (Modelling of Mutagenicity and Carcinogenicity: QSAR Studies and Test Classification) highlighted some of the problems associated with the application of QSAR methods to toxicological endpoints such as carcinogenicity. The application of QSAR beyond its classical domain of a strictly congeneric series, with only one mechanism of action, is necessary because of the particular nature of risk assessment. However, exploration outside of these boundaries highlights some crucial theoretical facets of QSAR which have important implications for the use of SARs in toxicology in general, and for assessing the validity and relevance of in vitro procedures.

The first of these is exemplified by the difference between the asymptotic character of the "no-activity" concept, implicit in QSAR as the lowest limit of activity, and the constitutive absence of activity typical in toxicology (89). This discrepancy causes a lack of agreement between QSARs which model potency and those which discriminate between active and non-active compounds. This is because it is impossible to distinguish between substances which show no activity for different reasons, either because of very low, undetectable, activity or due to a true absence of activity. A second, more general, point is the distinction between the knowledge obtainable on the large variability scale relative to the small variability scale, i.e. the relationships between the same sets of variables investigated at two different levels of variability can lead to mutually exclusive views of the phenomenon (90), which are caused by operating within different frames of reference.

The practical consequence of this for test validation is that results from validation studies in which large sets of chemicals are tested (the general relationship on a large scale) are likely to be very different from results carried out on sets of congeneric, mechanistically-related chemicals (specific relationships on a small scale). Nevertheless, both approaches point to biologically relevant phenomena -- the former to the intrinsic basis of the relationships between different tests, and the latter to the mapping of individual mechanisms of action.

Modelling Metabolism and Receptors

José Castell (On the Way to Predictive Hepatotoxicity) described the difference between intrinsic and idiosyncratic hepatotoxins; the former cause liver damage in a concentration-dependent manner in all individuals, whilst the latter elicit their effects only in certain individuals, because of differences in metabolism or due to some immunological response. In vitro models can provide valuable information to aid the early identification of intrinsic hepatotoxins. The key parameters are not just those which determine cytotoxicity, but also those which indicate impairment of specific hepatic functions, for example, gluconeogenesis, synthesis of plasma proteins, ureogenesis, biotransformation, and transport of bile acids (91).

While some intrinsic hepatotoxins are active per se, most hepatotoxins require biotransformation to elicit their toxic effects. Consequently, it is of utmost importance when using hepatocytes as an in vitro model to assess their ability to metabolise different xenobiotics. Rat hepatocytes are commonly used, but they lose most of their xenobiotic biotransformation activity during the first 24 hours in culture (92). To some extent, the co-culture of rat hepatocytes with either non-permanent liver-derived epithelial cells or cell-lines reduces this loss of cytochrome P450-mediated activity (93).

One area in which the use of hepatocyte cultures has become widely accepted is the study of species differences in metabolism. The use of human hepatocyte cultures at an early stage of drug development helps the selection of the most appropriate animal model to use subsequently (32). In vitro experiments are a source of valuable information for the generation of QSARs for predicting the hepatotoxic hazards of new chemicals. However, it must be remembered that several factors need to be taken into account when interpreting in vitro data in terms of in vivo hazard: the relevance of any cellular function which is altered, the reversibility of the observed effect, and the maximum non-toxic in vitro concentration in relation to the concentrations which are likely to be reached in vivo.

The presentation by Fons Rutten (Species and Organ-specific Toxicity of Chemical Compounds applying Fresh and Cryopreserved Precision-cut Tissue Slices and Organ Cultures) described the use of precision-cut slices of liver, kidney, and lung, and cultures of skin and trachea, to assess the in vitro organ-specific and species-specific toxicities of a range of chemicals (menadione, vitamin A, lidocaine, HgCl2, CdCl2, K2Cr2O7, paraquat, benzo[a]pyrene, sodium dodecyl sulphate, propoxur, and aflatoxin B1). To reduce, refine and replace animal tests for skin toxicity, a two-compartment organ culture model has been used to undertake species comparisons of the dermal penetration, irritancy, and metabolism of chemicals, employing various quantitative endpoints (94). The effects of the chemicals were investigated using cryopreserved and fresh tissue slices. The precision-cut tissue slices were viable and functionally-active in vitro for several days in the dynamic organ culture system (95). A broad spectrum of endpoints (MTT reduction, leakage of lactate dehydrogenase, DNA damage and repair, cell proliferation, protein synthesis, morphology, and ion transport) were measured. Chemicals requiring bioactivation (benzo[a]pyrene and aflatoxin B1) were toxic after a 24-hour exposure period (95, 96). Both lung and kidney slices remained viable in vitro for at least 24 hours when cultured following cryopreservation (W.R. Leeman et al, submitted for publication).

PBPK modelling (Mac Provan) predicts the disposition of xenobiotics and their metabolites by integrating three types of information: species-specific physiological parameters, partition coefficients for the chemical, and metabolic parameters. The physiological information required for developing such models is readily available from the literature. Partition coefficients can be measured by vial equilibration techniques (20); metabolic constants can be estimated from experiments with isolated cells, tissue fractions, etc., or from in vivo metabolism studies.

Based upon this information, PBPK models can simulate the disposition of xenobiotics and their metabolites, and thus predict tissue exposure to these chemicals for different doses and species. In addition, some PBPK models include different routes of administration (inhalation, oral, intravenous, dermal, etc.) of the test chemicals and may, therefore, be of value in characterising the tissue concentrations achieved in each case. These models have a clear potential for refining absorption, distribution, metabolism, and excretion (ADME) studies, and may also be valuable in focusing predictive in vitro studies of xenobiotic toxicity to concentration ranges which are likely to be found in vivo.

Peter Zbinden (Pseudoreceptor Modelling: The Construction of Three-dimensional Pseudoreceptor Surrogates; Genetic Function Approximation in QSAR) described how receptor surrogates can be constructed for structurally uncharacterised enzymes or receptors based upon the structures of known ligand molecules. Although, in general, pseudoreceptors bear little structural resemblance to their natural counterparts, they are designed to accommodate a series of ligand molecules in a similar manner. The pseudoreceptor model is then validated by predicting the relative free energies of binding for a test set of ligand molecules.

The pseudoreceptor concept known as 'Yak' (97, 98) allows the construction of a peptidic pseudoreceptor model around a set of superimposed ligand molecules (i.e. the pharmacophore model). Three types of vectors are generated for the pharmacophore model: hydrogen-extension, ion-pair, and hydrophobicity vectors. The preferred orientations of these vectors can be derived from crystal structures of small molecules and proteins. The vectors are then subjected to cluster analysis, using frequently observed types of ligand-receptor interaction pairs and molecular lipophilicity potentials to select appropriate amino acids to place around specific functional groups of the ligand molecules. This process is repeated until all the functional groups of the ligands are saturated, or until spatial constraints prevent the further attachment of receptor residues. The concept has been validated by constructing models for human carbonic anhydrase, the dopaminergic receptor, the β2-adrenergic receptor (99), and the aromatic hydrocarbon receptor.

Genetic function approximation (GFA) offers an alternative to regression analysis for generating QSARs. The user specifies the types of molecular descriptors and the types of functions (linear, quadratic, spline, step) to be used in GFA for a given training set. The algorithm then starts with a population of models (usually 100-200) constructed randomly from the descriptors and functions, and uses genetic algorithms (crossovers and mutations; usually 3000-5000 crossovers) to yield a family of improved QSAR models (100).

Discussion and Recommendations

The discussion sessions at the workshop focused on three main, but overlapping, topics:

  1. the quality and availability of data;
  2. ways to achieve a better integration of QSAR and in vitro methods; and
  3. methodology, validation, and the selection of test chemicals.

Availability of high quality data

A major problem governing the development and subsequent validation of QSAR and in vitro models is the apparent lack of reliable and relevant in vivo data which are readily accessible. The generation of high quality data requires the unambiguous specification of relevant experimental conditions and the conduct of experiments according to good scientific practice. Ideally, the methods used to generate the data should be specified in international guidelines. Well-established protocols and standard operating procedures should be used. Ideally, experiments should be conducted under GLP conditions, intralaboratory and interlaboratory experimental reproducibility should be established, and any statistical treatment of the data should be clearly defined. For data to be used in QSAR studies, or for the validation of in vitro methods, some statistical considerations need to be taken into account, such as the adequate coverage of parameter space. It is recognised that data generated from in vitro or in vivo experiments are typically less precise than those obtained in physicochemical experiments.

There is a need to distinguish between data and information. Information is structured data. Most data in the public domain really constitute information, and may be sub-optimal (i.e. insufficiently detailed and informative) for the development of QSARs, or for the validation of in vitro tests. Individual data points, not just mean values, need to be made available; these must define the smallest relevant unit, for example, separate tissue scores for individual animals in a Draize eye test. It is recognised that some journals undertake to hold and distribute, on request, the raw data from which the published information was derived.

Large quantities of data are held by companies and by regulatory authorities; the majority of these data are regarded as confidential. Several of the following recommendations relate to ways in which these data might be made accessible.

  1. Raw data should be archived in a readily retrievable form.
  2. In all cases, data on the specification (including chemical purity) of the material tested must be provided. Information on the nature of any impurities or contaminants, and the age of the sample when tested, should be included where known.
  3. Companies should be encouraged to make non-confidential data available to external groups, perhaps via an independent organisation such as ECVAM. For confidential data, they should be encouraged to review the need to maintain that confidentiality on a regular (continual) basis.
  4. Regulatory agencies should be encouraged formally to establish QSARs utilising submission data and to work with groups developing QSARs. This would facilitate greater use of data submitted in confidence. Companies should also be encouraged to develop QSARs using their confidential data. Such QSAR models should then be placed in the public domain, along with supporting non-confidential data. This is particularly pertinent for the development of QSARs for predicting target organ toxicity. Information on likely organ distribution and metabolism would also be required for predicting systemic toxicity.
  5. ECVAM should instigate a contract with an appropriate group(s) to review all existing sources of data and information on the toxicological effects of chemicals, with the following objectives:

    1. to define the contents of existing databanks, includingthe number and type of chemicals, the nature of thedata, etc.; and
    2. to evaluate critically the data for their utilityfor the development of predictive models.

Ways to achieve a better integration of QSAR and in vitro methods

In vitro tests fall into three categories: empirical -- those for which no clear mechanistic basis can be identified; mechanistic -- those with a clear mechanistic basis; and analogous -- those in which the in vivo test system is essentially reproduced in vitro. QSAR methods may help to identify the mechanisms operating in in vitro assays. These mechanisms should be defined in operational terms, i.e. in terms of the level of organisation (molecular, cellular, tissue, organ, whole organism) at which they are working in the animal. From this information it should be possible to judge the 'mechanistic relevance' of the in vitro test.

Results from appropriate in vitro tests may also be used as molecular descriptors to develop QSAR equations. For example, in vitro cytotoxicity data, rather than computed reactivity parameters, could be used to model a toxic endpoint at the level of the whole organ. Although data from in vitro tests can be used to construct QSARs, it is also important to use existing animal data (where available); the predictions obtained from such QSARs could then be used to improve the in vitro tests via the definition of relevant mechanisms.

A single QSAR or in vitro test cannot be used directly to extrapolate from one level of organisation to another. However, families of QSARs or in vitro tests may be used to define rate-limiting parameters which govern effects at different levels or in different compartments (PBPK models), for example, to relate the inhibition of acetylcholinesterase activity to lethality in the whole organism.

In principle, QSARs can be used to improve testing schemes (batteries and hierarchical strategies), via the identification of endpoints which should be incorporated from a consideration of the mechanisms of toxicity. The correspondence between the performances of different in vitro tests should be assessed, to minimise redundancy and to maximise the coverage of chemical parameter space during test selection. New tests with mechanistically relevant endpoints can also be introduced.

Some toxicologists are inclined to apply only a descriptive rather than a quantitative approach to their work. There is a need for improved communication between toxicologists and those applying mathematical, chemical, and statistical methods to describe mechanisms of toxicity. The use of simpler ways of presenting results and conclusions for non-specialists should assist an appreciation of the applicability and usefulness of quantitative methods. One way that this could be achieved is via the publication of worked examples, using data and analyses from collaborative studies between toxicologists and physical scientists. There is also a need for the clarification of some potentially confusing terminology and concepts, for example, the roles of SAR (identification of structure-activity relationships) and QSAR (identification of SAR using quantitative descriptions of chemical structure) in predicting chemical reactivity.

Within companies, the integration of the approaches used, for example, in drug design (mathematical and computer modelling techniques) with those used for toxicity assessments should be encouraged, by bringing together scientists from both areas. The aim of such a process would be to obtain a toxicity profile at an early stage of compound development, although it is recognised that this information should be used with caution (i.e. for test or chemical prioritisation rather than for rejecting new chemicals).

It is recommended that:

  1. An ECVAM contract should be established to address the mechanistic bases of currently available in vitro tests, in order to:

    1. rationalise their predictive power,
    2. aid in the design of test batteries and hierarchicaltesting strategies, and
    3. help in the design of new in vitro tests.

    It is envisaged that these activities would facilitate the regulatory acceptance of alternative methods.

  2. As a matter of urgency, funding should be made available for the development of QSAR models which incorporate mammalian in vitro test data to try to extrapolate between different levels of biological organisation (molecular, cellular, tissue, organ, organism).

Methodology, validation and selection of test chemicals

Physiologically-based pharmacokinetic modelling
PBPK models are very useful for predicting systemic toxicity and their use should be encouraged. Consideration should be given to how their more widespread use and acceptance can be achieved. PBPK models can reduce, refine and possibly replace conventional animal procedures for predicting systemic toxicity, including ADME studies. PBPK modelling and SAR predictions could be used in combination with in vitro models before, or instead of, using animals. PBPK models can help predict those organs in which, depending on the dose and route of administration, a xenobiotic is likely to accumulate. Thus, in combination with in vitro toxicity assays, they could provide a first indication of those tissues/cells which may be susceptible to the effects of a particular chemical. The exact strategy adopted would depend on the type of chemical under consideration and the information available on the likely plasma levels from exposure.

Use of appropriate statistical methods and the implications for validation
Statistical methods appropriate to the particular type of data (continuous or discrete) should be used for their analysis, to avoid the generation of misleading relationships.

Current attempts to validate in vitro methods as alternatives to animal tests are based on the premise that in vitro potency can be related directly to in vivo potency. However, it is of fundamental importance to recognise the general inapplicability of this premise, since the relationship between in vitro and in vivo potencies can be expressed only in probabilistic terms. The practical significance of this for the validation of alternative methods is that, in the absence of a clear mechanistic relationship between in vitro and in vivo potencies, such validation must proceed in two stages:

Stage 1: Establishment of the general statistical relationship between the responses in the in vitro and in vivo tests. This stage must be performed using a wide range of chemicals, with different physicochemical properties, acting via different mechanisms of toxicity.

Stage 2: Establishment of the specific relationship between the in vitro and in vivo responses using separate models for each chemical class (defined by a common mechanism of toxicity).

It follows that, where a clear mechanistic relationship between in vitro and in vivo potencies has already been established, validation can begin at Stage 2.

Selection of chemicals
The selection of appropriate chemicals as training or test sets for both QSAR and in vitro toxicology studies is of paramount importance. A 'set' consists of those chemicals which exert a given toxic effect via a common mechanism that can be modelled by a single QSAR equation or in vitro test.

Extreme care should be exercised in the selection of training sets of chemicals to be used for QSAR studies since this will affect the validity of the results obtained. A variety of selection procedures have been proposed, including non-statistical methods (for example, 101), cluster analysis (102), fractional factorial design (103), principal components analysis (104), and D-optimal design (105). Unfortunately, there is no general agreement on which, if any, of these methods represents the best strategy to adopt for a particular set of compounds. Pleiss and Unger (106) have published an extensive review on chemical selection methods, but it may be beneficial to those involved in developing and validating testing strategies if some guidelines could be produced, perhaps by ECVAM.

It is recommended that:

  1. The development of integrated testing strategies for predicting the systemic toxicity of chemicals should be given high priority, due to animal welfare and economic considerations.
  2. The most appropriate statistical techniques must be used to analyse the data, taking into account whether the data are continuous or discrete.
  3. Depending on the claims made for the applicability of an in vitro test, as wide a range as possible of relevant chemicals must be selected to assess its utility (i.e. an even spread across appropriate mechanistic groups and ranges of in vivo activities). Where possible, known boundary conditions in test performance/interpretation must be taken into account during the selection of the test chemicals.
  4. The development and validation of expert systems for predicting toxicity and metabolism should be encouraged. An ECVAM workshop should be organised to discuss this topic.

 

References

  1. Anon. (1994). ECVAM News & Views. ATLA 22: 7-11.
  2. Turner, L., Choplin, F., Dugard, P., Hermens, J., Jaeckh, R., Marsmann, M. & Roberts, D. (1987). Structure-activity relationships in toxicology and ecotoxicology: an assessment. Toxicology In Vitro 1: 143-171.
  3. Karcher, W. (1992). Basic concepts and aims of QSAR studies. In Quantitative Structure/Activity Relationships (QSAR) in Toxicology. Ed. T. Coccini, L. Giannoni, W. Karcher, L. Manzo & R. Roi, pp. 5-25. Luxembourg: Commission of the European Communities.
  4. Livingstone, D.J. (1994). Computational techniques for the prediction of toxicity. Toxicology In Vitro 8: 873-877.
  5. Leahy, D.E., Taylor, P.J. & Wait, A.R. (1989). Model solvent systems for QSAR. Part I. Propylene glycol dipelargonate (PGDP). A new standard solvent for use in partition coefficient determination. Quantitative Structure Activity Relationships 8: 17-31.
  6. Könemann, H. (1981). Quantitative structure-activity relationships in fish toxicity. Part 1. Relationship for 50 industrial pollutants. Toxicology 19: 209-221.
  7. Jäckel, H. & Klein, W. (1991). Prediction of mammalian toxicity by quantitative structure-activity relationships: aliphatic amines and anilines. Quantitative Structure-Activity Relationships 10: 198-204.
  8. Devillers, J. (1992). Statistical analyses in drug design and environmental chemistry: basic concepts. In Quantitative Structure/Activity Relationships (QSAR) in Toxicology. Ed. T. Coccini, L. Giannoni, W. Karcher, L. Manzo & R. Roi, pp. 27-41. Luxembourg: Commission of the European Communities.
  9. Barratt, M.D. (1995). Quantitative structure-activity relationships for skin corrosivity of organic acids, bases, and phenols. Toxicology Letters 75: 169-176.
  10. Barratt, M.D. (1995). The role of structure-activity relationships and expert systems in alternative strategies for the determination of skin sensitisation, skin corrosivity and eye irritation. ATLA 23: 111-122.
  11. Haggard, H.W. (1924). The absorption, distribution, and elimination of ethyl ether. II. Analysis of the mechanism of the absorption and elimination of such a gas or vapour as ethyl ether. Journal of Biological Chemistry 59: 753-770.
  12. Mapleson, W.W. (1963). Quantitative predictions of anaesthetic concentrations. In Uptake and Distribution of Anaesthetic Agents. Ed. E.M. Papper & R.J. Kitz, pp. 104-119. New York: McGraw-Hill.
  13. Riggs, D.S. (1963). The Mathematical Approach to Physiological Problems: A Critical Primer. Cambridge, MA, USA: MIT Press.
  14. Fiserova-Bergerova, V. & Holaday, D.A. (1979). Uptake and clearance of inhalation anaesthetics in man. Drug Metabolism Reviews 9: 43-60.
  15. Teorell, T. (1937). Kinetics of distribution of substances administered to the body. I. The extravascular modes of administration. Archives of Internal Pharmacodynamics 57: 205-240.
  16. Bischoff, K.B., Dedrick, R.L., Zaharko, D.S. & Longstreth, J.A. (1971). Methotrexate pharmacokinetics. Journal of Pharmaceutical Science 60: 1128-1133.
  17. Dedrick, R.L. (1973). Animal scale up. Journal of Pharmacokinetics and Biopharmacy 1: 435-461.
  18. Roland, M. (1984). Physiologic pharmacokinetic models: relevance, experience, and future trends. Drug Metabolism Reviews 15: 55-74.
  19. Ramsey, J.R. & Andersen, M.E. (1984). A physiological based description of the inhalation pharmacokinetics of styrene in rats and humans. Toxicology & Applied Pharmacology 73: 159-175.
  20. Sato, A., Fujiwara, Y. & Hirosawa, K. (1971). Solubility of benzene, toluene, and m-xylene in blood. Japanese Journal of Industrial Health 14: 3-8.
  21. Dedrick, R.L., Forrester, D.D. & Ho, D.H.O. (1971). In vitro/in vivo correlation of drug metabolism - deamination of 1-β-D-arabinofuranosylcytosine. Biochemical Pharmacology 21: 1-16.
  22. Green, T., Provan, W.M., Dugard, P.H. & Cook, S.K. (1988). Methylene chloride (Dichloromethane): Human Risk Assessment using Experimental Animal Data. ECETOC Technical Report No. 32. 64pp. Brussels, Belgium: European Centre for Ecotoxicology and Toxicology of Chemicals.
  23. Jolles, G. & Cordier, A. (1992). Editors. In Vitro Methods in Toxicology. 607pp. London: Academic Press.
  24. Watson, R.R. (1992). Editor. In Vitro Methods of Toxicology. 298pp. Boca Raton, Florida: CRC Press.
  25. Frazier, J.M. (1992). Editor. In Vitro Toxicity Testing. Applications to Safety Evaluation. 299pp. New York: Marcel Dekker.
  26. Gad, S.C. (1993). Alternatives to in vivo studies in toxicology. In General and Applied Toxicology, Volume 1. Ed. B. Ballantyne, T. Marrs & P. Turner. pp.179-206. Basingstoke: Macmillan.
  27. Fry, J.R. (1993). Development in in vitro toxicology. Comparative Haematology International 3: 4-7.
  28. Roberfroid, M. (1994). Alternatives in safety testing: progress or uselessness? ATLA 22: 438-444.
  29. Fentem, J.H. (1994). The use of human tissues in in vitro toxicology. Stirling, 28/29 April 1993. Summary of general discussions. Human & Experimental Toxicology 13: 445-449.
  30. Wilcox, D.K. & Bruner, L.H. (1990). In vitro alternatives for ocular safety testing: an outline of assays and possible future developments. ATLA 18: 117-128.
  31. DeLeo, V.A. (1992). Cutaneous irritancy. In In Vitro Toxicity Testing. Applications to Safety Evaluation. Ed. J.M. Frazier. pp.191-203. New York: Marcel Dekker.
  32. Blaauboer, B.J., Boobis, A.R., Castell, J.V., Coecke, S., Groothuis, G.M.M., Guillouzo, A., Hall, T.J., Hawksworth, G.M., Lorenzon, G., Miltenburger, H.G., Rogiers, V., Skett, P., Villa, P. & Wiebel, F.J. (1994). The practical applicability of hepatocyte cultures in routine testing. The report and recommendations of ECVAM workshop 1. ATLA 22: 231-241.
  33. Hawksworth, G.M., Bach, P.H., Dekant, W., Diezi, J.E., Harpur, E., Lock, E.A., MacDonald, C., Morin, J-P., Nagelkerke, J.F., Pfaller, W., Rutten, A.A.J.J.L., Ryan, M.P., Toutain, H.J. & Trevisan, A. (1995). Nephrotoxicity testing in vitro. The report and recommendations of ECVAM workshop 10. ATLA, in press.
  34. Atterwill, C.K., Bruinink, A., Drejer, J., Duarte, E., McFarlane Abdulla, E., Meredith, C., Nicotera, P., Regan, C., Rodríguez-Farré, E., Simpson, M.G., Smith, R., Veronesi, B., Vijverberg, H., Walum, E. & Williams, D.C. (1994). In vitro neurotoxicity testing. The report and recommendations of ECVAM workshop 3. ATLA 22: 350-362.
  35. Garle, M.J., Fentem, J.H. & Fry, J.R. (1994). In vitro cytotoxicity tests for the prediction of acute toxicity. Toxicology In Vitro 8: 1303-1312.
  36. Balls, M. & Fentem, J.H. (1992). The use of basal cytotoxicity and target organ toxicity tests in hazard identification and risk assessment. ATLA 20: 368-388.
  37. Basketter, D.A., Whittle, E. & Chamberlain, M. (1994). Identification of irritation and corrosion hazards to skin: an alternative strategy to animal testing. Food & Chemical Toxicology 32: 539-542.
  38. Spielmann, H., Lovell, W.W., Hslzle, E., Johnson, B.E., Maurer, T., Miranda, M.A., Pape, W.J.W., Sapora, O. & Sladowski, D. (1994). In vitro phototoxicity testing. The report and recommendations of ECVAM workshop 2. ATLA 22: 314-348.
  39. Basketter, D.A., Scholes, E.W., Chamberlain, M. & Barratt, M.D. An alternative strategy to the use of guinea pigs for the identification of skin sensitisation hazard. Food & Chemical Toxicology, in press.
  40. Hansch, C., Maloney, P.P., Fujita, T. & Muir, R.M. (1962). Correlation of biological activity of phenoxyacetic acids with Hammett substituent constants and partition coefficients. Nature 194: 178-180.
  41. van Leeuwen, C.J., van der Zandt, P.T.J., Aldenberg, T., Verhaar, H.J.M. & Hermens, J.L.M. (1992). Application of quantitative structure-activity relationships, extrapolation and equilibrium partitioning in aquatic effects assessment. 1. Narcotic industrial pollutants. Environmental Toxicology and Chemistry 11: 267-282.
  42. Hansch, C. & Clayton, J.M. (1973). Lipophilic character and biological activity of drugs. 2. The parabolic case. Journal of Pharmaceutical Science 62: 1-21.
  43. Crum Brown, A. & Frazer, T. (1868/9). On the connection between chemical constitution and physiological action. Part I. On the physiological action of the salts of the ammonia bases derived from Strychnia, Brucia, Thebaia, Codeia, Morphia, and Nicotia. Transactions of the Royal Society of Edinburgh 25: 151-203.
  44. Hyde, R.M. & Livingstone, D.J. (1988). Perspectives in QSAR: computer chemistry and pattern recognition. Journal of Computer-Aided Molecular Design 2: 145-157.
  45. Livingstone, D.J., Hudson, B.D., George, A. & Ford, M.G. (1991). The use of molecular dynamics in QSAR studies. In QSAR: Rational Approaches to the Design of Bioactive Compounds. Ed. C. Silipo & A. Vittoria, pp. 557-560. Amsterdam: Elsevier.
  46. Livingstone, D.J. The trouble with chemometrics ...? In Proceedings of the 10th European Symposium on QSAR. Ed. F. Sanz. Barcelona: J.R. Prous, in press.
  47. Livingstone, D.J. (1991). Pattern recognition methods in rational drug design. In Molecular Design and Modeling: Concepts and Applications, Part B. Methods in Enzymology, Volume 203. Ed. J.J. Langone, pp. 613-638. San Diego: Academic Press.
  48. Ridings, J.E., Manallack, D.T., Saunders, M.R., Baldwin, J.A. & Livingstone, D.J. (1992). Multivariate quantitative structure-toxicity relationships in a series of dopamine mimetics. Toxicology 76: 209-217.
  49. Barratt, M.D. A quantitative structure-activity relationship for the eye irritation potential of neutral organic chemicals. Toxicology Letters, in press.
  50. Chamberlain, M. & Barratt, M.D. Practical applications of QSAR to in vitro toxicology by consideration of eye irritation. Toxicology In Vitro, in press.
  51. Spielmann, H., Gerner, I., Kalweit, S., Moog, R., Wirnsberger, T., Krauser, K., Kreiling, R., Kreuzer, H., Lüpke, N.P., Miltenburger, H.G., Müller, N., Mürmann, P., Pape, W., Siegemund, B., Spengler, J., Steiling, W. & Wiebel, F.J. (1991). Interlaboratory assessment of alternatives to the Draize eye irritation test in Germany. Toxicology In Vitro 5: 539-542.
  52. Spielmann, H., Kalweit, S., Liebsch, M., Wirnsberger, T., Gerner, I., Bertram-Neis, W., Krauser, K., Kreiling, G., Miltenburger, H., Pape, W. & Steiling, W. (1993). Validation study of alternatives to the Draize eye irritation test in Germany. Cytotoxicity testing and HET-CAM assay with 136 industrial chemicals. Toxicology In Vitro 7: 505-510.
  53. Anon. (1992). Guidelines for Testing of Chemicals No. 404. Acute dermal irritation/corrosion. 6pp. Paris, France: OECD.
  54. Anon. (1987). Guidelines for Testing of Chemicals No. 405. Acute eye irritation/corrosion. 9pp. Paris, France: OECD.
  55. Barratt, M.D., Basketter, D.A., Chamberlain, M., Admans, G.D. & Langowski, J.J. (1994). An expert system rulebase for identifying contact allergens. Toxicology In Vitro 8: 1053-1060.
  56. Barratt, M.D. (1995). Quantitative structure-activity relationships for skin permeability. Toxicology In Vitro 9: 27-37.
  57. Hermens, J.L.M. (1989). Quantitative structure-activity relationships of environmental pollutants. In Handbook of Environmental Chemistry, Volume 2E. Ed. O. Hutzinger, pp. 111-162. Berlin, Germany: Springer-Verlag.
  58. McCarty, L.S. (1987). Relationship between toxicity and bioconcentration for some organic chemicals. I. Examination of the relationship. In QSAR in Environmental Toxicology - II. Ed. K.L.E. Kaiser, pp. 207-220. Dordrecht, The Netherlands: D. Reidel Publishing Company.
  59. McCarty, L.S. & Mackay, D. (1993). Enhancing ecotoxicological modelling and assessment. Environmental Science and Technology 27: 1719-1728.
  60. van Hoogen, G. & Opperhuizen, A. (1988). Toxicokinetics of chlorobenzenes in fish. Environmental Toxicology and Chemistry 7: 213-219.
  61. Verhaar, H.J.M., van Leeuwen, C.J. & Hermens, J.L.M. (1992). Classifying environmental pollutants. 1. Structure-activity relationships for prediction of aquatic toxicity. Chemosphere 25: 471-491.
  62. Verhaar, H.J.M., van Leeuwen, C.J., Bol, J. & Hermens, J.L.M. (1994). Application of QSARs in risk management of existing chemicals. SAR & QSAR Environmental Research 2: 39-58.
  63. Wold, S., Albano, C., Dunn, W.J. III, Edlund, U., Esbensen, K., Geladi, P., Hellberg, S., Johansson, E., Lindberg, W. & Sjöström, M. (1984). Multivariate data analysis in chemistry. In Chemometrics - Mathematics and Statistics in Chemistry. Ed. B.R. Kowalski, pp. 17-95. Dordrecht, The Netherlands: D. Reidel Publishing Company.
  64. Cramer, C.J., Famini, G.R. & Lowrey, A.H. (1993). Use of calculated quantum chemical properties as surrogates for solvatochromic parameters in structure-activity relationships. Accounts of Chemical Research 26: 599-605.
  65. Verhaar, H.J.M. (1995). Predictive Methods in Aquatic Toxicology. PhD Thesis. Utrecht University, The Netherlands.
  66. Botham, P.A., Chamberlain, M., Barratt, M.D., Curren, R.D., Esdaile, D.J., Gardner, J.R., Gordon, V.C., Hildebrand, B., Lewis, R.W., Liebsch, M., Logemann, P., Osborne, R., Ponec, M., Régnier, J-F., Steiling, W., Walker, A.P. & Balls, M. (1995). A prevalidation study on in vitro skin corrosivity testing. The report and recommendations of ECVAM workshop 6. ATLA 23: 219-255.
  67. Chamberlain, M. & Earl, L. (1994). Use of cell cultures in irritancy testing. In In Vitro Skin Toxicology. Irritation, Phototoxicity, Sensitisation. Alternative Methods in Toxicology, Volume 10. Ed. A. Rougier, A.M. Goldberg & H.I. Maibach. pp.59-67. New York: Mary Ann Liebert.
  68. Shivji, G.M., Gupta, A.K. & Sauder, D.N. (1994). Role of cytokines in irritant contact dermatitis. In In Vitro Skin Toxicology. Irritation, Phototoxicity, Sensitisation. Alternative Methods in Toxicology, Volume 10. Ed. A. Rougier, A.M. Goldberg & H.I. Maibach. pp.13-22. New York: Mary Ann Liebert.
  69. Atkinson, K.A., Fentem, J.H., Clothier, R.H. & Balls, M. (1992). Alternatives to ocular irritation testing in animals. Lens and Eye Toxicity Research 9: 247-258.
  70. Rougier, A., Cottin, M., De Silva, O., Catroux, P., Roguet, R. & Dossou, K.G. (1994). The use of in vitro methods in the ocular irritation assessment of cosmetic products. Toxicology in Vitro 8: 893-905.
  71. McCulley, J.P. & Stephens, T.J. (1993). Draize eye testing alternatives - a perspective. In In Vitro Toxicology: Tenth Anniversary Symposium of CAAT. Alternative Methods in Toxicology, Volume 9. Ed. A.M. Goldberg. pp.107-119. New York: Mary Ann Liebert.
  72. Anon. (1992). The Use of Animals in Research, Development and Testing, p.30. London, UK: Parliamentary Office of Science and Technology (POST).
  73. Deldar, A. & Stevens, C.E. (1993). Development and application of in vitro models of hematopoiesis to drug development. Toxicologic Pathology 21: 231-240.
  74. Parchment, R.E., Huang, M. & Erickson-Miller, C.L. (1993). Roles for in vitro myelotoxicity tests in preclinical drug development and clinical trial planning. Toxicologic Pathology 21: 241-250.
  75. Lake, B.G., Lewis, D.F.V., Gray, T.J.B. & Beamand, J.A. (1993). Structure-activity relationships for induction of peroxisomal enzyme activities in primary rat hepatocyte cultures. Toxicology in Vitro 7: 605-614.
  76. Tweats, D. (1994). Mutagenicity. In General and Applied Toxicology. Ed. B. Ballantyne, T. Marrs & P. Turner, pp. 871-936. London, UK: Macmillan.
  77. Lawlet, P. (1989). Mutagens as carcinogens: development of current concepts. Mutation Research 213: 3-26.
  78. Brusick, D. (1994). Genetic toxicology. In Principles and Methods of Toxicology. Ed. A.W. Hayes, pp. 545-577. New York: Raven Press.
  79. Combes, R.D. (1992). Trends in genotoxicity testing. Chemistry in Industry 24: 950-954.
  80. Aardema, M.J., Ed. (1993). Updated worldwide regulatory guidelines for genotoxicity testing. Workshop proceedings from the 1992 Environmental Mutagen Society Meeting. Environmental and Molecular Mutagenesis 21: 1-57.
  81. Haseman, J.K. & Clark, A.M. (1990). Carcinogenicity results for 114 laboratory animal studies used to assess the predictivity of four in vitro genetic toxicity assays for rodent carcinogenicity. Environmental and Molecular Mutagenesis 16: Suppl. 18, 15-31.
  82. Ashby, J. & Tennant, R.W. (1988). Chemical structure, salmonella mutagenicity and extent of carcinogenicity as indicators of genotoxic carcinogenesis among 222 chemicals tested in rodents by the U.S. NCI/NTP. Mutation Research 204: 17-115.
  83. Ashby, J. & Tennant, R.W. (1991). Definitive relationships among chemical structure, carcinogenicity, and mutagenicity for 301 chemicals tested by the U.S. NTP. Mutation Research 257: 229-306.
  84. Ashby, J. & Paton, D. (1993). The influence of chemical structure on the extent and sites of carcinogenesis for 522 rodent carcinogens and 55 different human carcinogen exposures. Mutation Research 286: 3-74.
  85. Combes, R.D. & Judson, P. The use of artificial intelligence systems for predicting toxicity. Pesticide Science, in press.
  86. Klopman, G. & Rosenkranz, H.S. (1994). Approaches to SAR in carcinogenesis and mutagenesis. Prediction of carcinogenicity/ mutagenicity using MULTI-CASE. Mutation Research 305: 33-46.
  87. Sanderson, D.M. & Earnshaw, C.G. (1991). Computer prediction of possible toxic action from chemical structure: the DEREK system. Human and Experimental Toxicology 10: 261-273.
  88. Long, A. & Combes, R.D. Using DEREK to predict the activity of some carcinogens/mutagens found in foods. Toxicology In Vitro, in press.
  89. Benigni, R., Andreoli, C. & Giuliani, A. (1994). QSAR models for both mutagenic potency and activity: application to nitroarenes and aromatic amines. Environmental and Molecular Mutagenesis 24: 208-219.
  90. Benigni, R. & Giuliani, A. (1994). Quantitative modelling and biology: the multivariate approach. American Journal of Physiology 266: R1697-1704.
  91. Castell, J.V. & Gómez-Lechón, M.J. (1992). The in vitro evaluation of the potential risk of hepatotoxicity of drugs. In In Vitro Alternatives to Animal Pharmaco-toxicology. Ed. J.V. Castell & M.J. Gómez-Lechón, Chapter 8. Madrid, Spain: Farmaindustria.
  92. Gómez-Lechón, M.J., Montoya, A., López, P., Donato, T., Larrauri, A. & Castell, J.V. (1988). The potential use of cultured hepatocytes in predicting the hepatotoxicity of xenobiotics. Xenobiotica 18: 725-735.
  93. Donato, M.T., Castell, J.V. & Gómez-Lechón, M.J. (1994). Cytochrome P450 activities in pure and co-cultured rat hepatocytes. Effects of model drug inducers. In Vitro Cellular and Developmental Biology 30: 825-832.
  94. Rutten, A.A.J.J.L. & van de Sandt, J.J.M. (1994). In vitro dermal toxicology using skin organ cultures. Toxicology In Vitro 8: 703-705.
  95. Leeman, W.R., van de Gevel, I.A. & Rutten, A.A.J.J.L. (1995). Cytotoxicity of retinoic acid, menadione and aflatoxin B1 in rat liver slices using Netwell inserts as a new culture system. Toxicology in Vitro 9: 291-298.
  96. Wolterbeek, A.P.M., Roggeband, R., Steenwinkel, M-J.S.T., Baan, R.A. & Rutten, A.A.J.J.L. (1993). Formation and repair of benzo[a]pyrene-DNA adducts in cultured hamster tracheal epithelium determined by 32P-postlabelling analysis and unscheduled DNA synthesis. Carcinogenesis 14: 463-467.
  97. Vedani, A., Zbinden, P. & Snyder, J.P. (1993). Pseudo-receptor modelling: a new concept for the three-dimensional construction of receptor binding sites. Journal of Receptor Research 13: 163-177.
  98. Snyder, J.P., Rao, S.N., Koehler, K.F. & Vedani, A. (1993). Pseudoreceptors. In 3D QSAR in Drug Design. Ed. H. Kubinyi, pp. 336-354. Leiden, The Netherlands: ESCOM Science Publishers.
  99. Vedani, A., Zbinden, P., Snyder, J.P. & Greenidge, P.A. Pseudoreceptor modelling: the construction of three-dimensional receptor surrogates. Journal of the American Chemical Society, in press.
  100. Rogers, D. & Hopfinger, A.J. (1994). Application of genetic function approximation to quantitative structure-activity relationships and quantitative structure-property relationships. Journal of Chemical Informatics and Computer Science 34: 854-866.
  101. Topliss, J.G. (1975). Utilization of operational schemes for analog synthesis in drug design. In Drug Design, Volume V. Ed. E.J. Ariens (Medicinal Chemistry Volume 11), pp. 1-21. New York & London: Academic Press.
  102. Hansch, C., Unger, S.H. & Forsythe, A.B. (1973). Strategy in drug design. Cluster analysis as an aid in the selection of substituents. Journal of Medicinal Chemistry 16: 1217-1222.
  103. Austell, V. (1982). A manual method for systematic drug design. European Journal of Medicinal Chemistry 17: 9-16.
  104. Streich, W.J., Dove, S. & Franke, R. (1980). On the rational selection of test series.1. Principal component method combined with multidimensional mapping. Journal of Medicinal Chemistry 23: 1452-1456.
  105. Baroni, M., Clementi, S., Cruciani, G., Kettaneh-Wold, N. & Wold, S. (1993). D-Optimal designs in QSAR. Quantitative Structure Activity Relationships 12: 225-231.
  106. Pleiss, M.A. & Unger, S.H. (1990). The design of test series and the significance of QSAR relationships. In Quantitative Drug Design. Ed. C.A. Ramsden, pp. 561-587. Oxford, UK: Pergamon Press.