The computational prediction of toxicity.

Recent developments in the prediction of toxicity from chemical structure have been reviewed. Attention has been drawn to some of the problems that can be encountered in the area of predictive toxicology, including the need for a multi-disciplinary approach and the need to address mechanisms of action. Progress has been hampered by the sparseness of good quality toxicological data. Perhaps too much effort has been devoted to exploring new statistical methods rather than to the creation of data sets for hitherto uninvestigated toxicological endpoints and/or classes of chemicals.

Studies of the chemical selectivity of hapten, reactivity, and skin sensitization potency. 1. Synthesis and studies on the reactivity toward model nucleophiles of the (13)C-labeled skin sensitizers hex-1-ene- and hexane-1,3-sultones.

The potent skin sensitizers hex-1-ene- and hexane-1,3-sultone have been synthesized isotopically labeled with (13)C at reactive sites. The reactivity of 2-[(13)C]- and 3-[(13)C]hex-1-ene-1,3-sultones and of 3-[(13)C]hexane-1,3-sultone toward a series of model nucleophiles for protein amino acid residues, i.e., butylamine, diethylamine, imidazole, propanethiol, and phenol, was followed by (13)C NMR spectroscopy. The reactivity in water of hex-1-ene-1,3-sultone toward model nucleophiles follows the hard and soft acid and base theory with the hard nucleophiles (primary and secondary amine and phenate) mainly reacting at position 3 by S(N) substitution, and the soft nucleophiles (thiolate and imidazole) mainly reacting at position 2 by a Michael addition reaction. Hexane-1,3-sultone reacts with model nucleophiles at position 3 by S(N) substitution. Both saturated and unsaturated sultones are sensitive to hydrolysis when reacted in water.

Studies of the chemical selectivity of hapten, reactivity, and skin sensitization potency. 2. nmr studies of the covalent binding of the (13)c-labeled skin sensitizers 2-[13C]- and 3-[13C]hex-1-ene- and 3-[13C]hexane-1,3-sultones to human serum albumin.

3-[(13)C]- and 2-[(13)C]hex-1-ene-1,3-sultones (1a and 1b, respectively) and 3-[(13)C]hex-1-ene-1,3-sultone 2a were incubated with human serum albumin in phosphate buffer at pH 8.1. In both cases, the main reaction was a hydrolysis via an S(N) reaction at position 3, but several adducts were also formed. Hex-1-ene-1,3-sultone, which is a strong skin sensitizer, appears also to be a strongly oxophilic molecule reacting mainly at position 3 through an S(N) reaction to give adducts on tyrosines. This sultone was also able to react with a single lysine residue, also via an initial S(N) reaction at position 3, followed by an intramolecular Michael addition at position 2 to form a mixture of aziridinium intermediates which were subsequently hydrolyzed to give an amino alcohol derivative as the final product. The same reaction carried out on acetylated human serum albumin seems to indicate that the target lysine could be Lys199, which is known to be easily acetylated. Hexane-1,3-sultone, which is a weak sensitizer, appears to be an even more oxophilic molecule, making adducts on tyrosines through an S(N) reaction at position 3. No reaction was observed on Lys199. The difference in skin sensitization potential seems therefore to be more related to the selective ability of modifying lysine residues than to the more general ability to modify tyrosine residues.

Prediction of toxicity from chemical structure.

The basis for the prediction of toxicity from chemical structure is that the properties of a chemical are implicit in its molecular structure. Biological activity can be expressed as a function of partition and reactivity, that is, for a chemical to be able to express its toxicity, it must be transported from its site of administration to its site of action and then it must bind to or react with its receptor or target. This process may also involve metabolic transformation of the chemical. The application of these principles to the prediction of the toxicity of new or untested chemicals has been achieved in a number of different ways covering a wide range of complexity, from computer systems containing databases of hundreds of chemicals, to simple "reading across" between chemicals with similar chemical/toxicological functionality. The common feature of the approaches described in this article is that their starting point is a mechanistic hypothesis linking chemical structure and/or functionality with the toxicological endpoint of interest. The prediction of toxicity from chemical structure can make a valuable contribution to the reduction of animal usage in the screening out of potentially toxic chemicals at an early stage and in providing data for making positive classifications of toxicity.

Comparison of an in vitro cellular phototoxicity model against controlled clinical trials of fluoroquinolone skin phototoxicity.

Many therapeutic drugs induce phototoxic skin responses following exposure to solar or artificial ultraviolet radiation sources. Several in vitro model systems have been developed to predict drug phototoxicity but none have been conducted in parallel with controlled clinical phototoxicity studies on systemically administered pharmaceuticals. The in vitro phototoxicity of eight fluoroquinolone (FQ) antibiotics (ciprofloxacin, grepafloxacin, lomefloxacin, norfloxacin, ofloxacin, trovafloxacin, BAYy3118, moxifloxacin) was determined by exposing Chinese hamster fibroblasts to UVA radiation. Cell damage was quantified with standard MTT or neutral red assays and an in vitro phototoxic index calculated (PI(vit)=% cell viability with UVA alone /% cell viability with UVA+FQ) for each endpoint. Clinical photosensitizing ability of the eight systemically administered FQ was investigated using double-blind, placebo and positive controlled, clinical skin phototesting of normal subjects. Minimal erythema doses at 365+/-30nm were determined before and after 6-7 days of FQ ingestion and PI(clin) (minimal erythema dose without FQ/minimal erythema dose with FQ) calculated. Linear regression analysis of PI(vit) vs PI(clin) gave correlations of up to 0.893. Principal components analysis of PI(vit), daily dose, plasma levels and photophysical (absorption) properties of the eight FQ showed that phototoxic (arbitrarily defined as PI(clin)> or =2) and non-phototoxic (PI(clin)<2) FQ could be completely discriminated using these parameters, and that the in vitro models were able to rank the relative phototoxic potential of the eight FQ.

A QSAR model for the eye irritation of cationic surfactants.

A QSAR model for the eye irritation of cationic surfactants has been constructed using a dataset consisting of the maximum average scores (MAS-accordance to Draize) for 29 in vivo rabbit eye irritation tests on 19 different cationic surfactants. The parameters used were logP (log [octanol/water partition coefficient]) and molecular volume (to model the partition of the surfactants into the membranes of the eye), logCMC (log critical micelle concentration-a measure of the reactivity of the surfactants with the eye) together with surfactant concentration. The model was constructed using neural network analysis. MAS showed strongly positive, non-linear correlations with surfactant concentration and logCMC and a strongly negative, non-linear correlation with logP. The Pearson correlation between the actual and predicted values of MAS was 0.838 showing that around 70% (r(2)=0.702) of the variance in the dataset is explained by the model. This value is consistent with levels of biological variability reported historically for the Draize rabbit eye test. The relationship provides a potentially useful prediction model for the eye irritation potential of new or untested cationic surfactants with physicochemical properties lying within the parameter space of the model.

Development of an expert system rulebase for the prospective identification of photoallergens.

Relationships between the structure and properties of chemicals can be programmed into knowledge-based systems such as DEREK (an acronym for 'Deductive Estimation of Risk from Existing Knowledge'). The DEREK knowledge-based computer system contains a sub-set of over 50 rules describing chemical substructures (toxophores) responsible for skin sensitization. This rulebase, based originally on Unilever historical in-house guinea pig maximisation test data, is largely complete and is undergoing refinement as the next stage of its development. As part of an ongoing program of validation and testing, the predictive ability of the sensitization rule set was assessed by processing the structures of over 100 chemical substances in the list of contact allergens identified by the BgVV (German Federal Institute for Health Protection of Consumers). The exercise highlighted areas of chemistry where further development of the rulebase was required, either by extension of the scope of existing rules or by generation of new rules where a sound mechanistic rationale for the biological activity could be established. Several chemicals likely to be acting as photoallergens were identified and rules for photoallergenicity were written covering three classes of chemicals. This paper describes work to extend the DEREK rules for photoallergenicity as part of the European Phototox Project.

Validation and subsequent development of the DEREK skin sensitization rulebase by analysis of the BgVV list of contact allergens.

The DEREK knowledge-based computer system contains a subset of approximately 50 rules describing chemical substructures (toxophores) responsible for skin sensitization. This rulebase, based originally on Unilever historical in-house guinea pig maximization test data, has been subject to extensive validation and is undergoing refinement as the next stage of its development. As part of an ongoing program of validation and testing, the predictive ability of the sensitization rule set has been assessed by processing the structures of the 84 chemical substances in the list of contact allergens issued by the BgVV (German Federal Institute for Health Protection of Consumers). This list of chemicals is important because the biological data for each of the chemicals have been carefully scrutinized and peer reviewed, a key consideration in an area of toxicology in which much unreliable and potentially misleading data have been published. The existing DEREK rulebase for skin sensitization identified toxophores for skin sensitization in the structures of 71 out of the 84 chemicals (85%). The exercise highlighted areas of chemistry where further development of the rulebase was required, either by extension of the scope of existing rules or by generation of new rules where a sound mechanistic rationale for the biological activity could be established. Chemicals likely to be acting as photoallergens were identified, and new rules for photoallergenicity have subsequently been written. At the end of the exercise, the refined rulebase was able to identify toxophores for skin sensitization for 82 of the 84 chemicals in the BgVV list.

The Integrated Use of Alternative Methods in Toxicological Risk Evaluation - ECVAM Integrated Testing Strategies Task Force Report 1.

The ECVAM Task Force on Integrated Testing Strategies was established in December 1996, with the remit of assessing the current status of integrated toxicity testing, and of making proposals regarding the design and implementation of integrated testing strategies. The first step in an integrated testing strategy is usually to determine the chemical functionality of a substance, on the basis of its structure and physicochemical properties. The biokinetic and dynamic behaviours of the chemical in various in vitro systems are then assessed. The various elements are then integrated, in either a parallel or a stepwise fashion, to make predictions of the local or systemic toxicity of the chemical of interest. In this report, a generic scheme for local/systemic toxicity, and a specific scheme for target organ toxicity, are proposed. The scope and limitations of the approaches are discussed. The task force hopes that its proposals will stimulate a discussion on the feasibility of this type of approach and it welcomes any feedback. It is planned that the discussion points will be elaborated in a second task force report.

Skin irritation potential of mixed surfactant systems.

Virtually all current detergent formulations contain mixtures of surfactants. Our experience and test data on these formulations, which is in agreement with that of many others, has shown that in use the formulations exhibit lower acute irritation potential than predicted by simple summation of the irritation potential of the individual actives. Using the criteria of the Dangerous Preparations Directive (EC Directive 88/379/EEC), many of these formulations classify as irritant in the neat state, with consequent labelling requirements. Such classification is based on addition of irritant components giving a total concentration which exceeds a nominal threshold. In this study, mixtures of surfactants were tested by application to a panel of 31 human volunteers for up to 4 hr, using the technique established for the assessment of acute skin irritation potential. The positive control, sodium dodecyl sulfate (SDS) at 20% concentration, gave an 84% positive response. Dimethyl dodecyl amido betaine (DDAB) at the same concentration gave a 94% response. However, a combination of 20% of each of these surfactants in the same panellists gave a response of only 44%--a significant reduction in the irritation potential. A further test conducted with a mixture of 10% SDS and 10% DDAB in a second panel gave a 31% positive response compared with a 94% positive response to the 20% SDS control in that panel. These results clearly demonstrate that the acute irritation potential of mixed surfactants cannot be predicted by simple summation of the irritation potential of the component substances. Initial results of the mechanistic investigation indicate that the reduced irritation induced by the mixed surfactant systems correlates with a reduced critical micelle concentration (CMC). However, the reduced CMC itself seems not to be responsible for the lowered irritation, since these experiments were conducted at concentrations well above the CMC. It is proposed that the critical event leading to skin irritation is binding to skin protein and that in mixed surfactant systems, the individual surfactants exhibit less affinity for this protein.

Integration of QSAR and in vitro toxicology.

The principles of quantitative structure-activity relationships (QSAR) are based on the premise that the properties of a chemical are implicit in its molecular structure. Therefore, if a mechanistic hypothesis can be proposed linking a group of related chemicals with a particular toxic end point, the hypothesis can be used to define relevant parameters to establish a QSAR. Ways in which QSAR and in vitro toxicology can complement each other in development of alternatives to live animal experiments are described and illustrated by examples from acute toxicological end points. Integration of QSAR and in vitro methods is examined in the context of assessing mechanistic competence and improving the design of in vitro assays and the development of prediction models. The nature of biological variability is explored together with its implications for the selection of sets of chemicals for test development, optimization, and validation. Methods are described to support the use of data from in vivo tests that do not meet today's stringent requirements of acceptability. Integration of QSAR and in vitro methods into strategic approaches for the replacement, reduction, and refinement of the use of animals is described with examples.

Integrating computer prediction systems with in vitro methods towards a better understanding of toxicology.

Structure Activity Relationships (SARs) or Quantitative Structure Activity Relationships (QSARs) form the basis of most computer prediction systems in toxicology. The underlying premise of SARs and QSARs is that the properties of a chemical are implicit in its molecular structure. For an SAR or QSAR to be valid and reliable, the dependent property for all of the chemicals covered by the relationship has to be elicited by a mechanism which is both common to the set of chemicals as well as relevant to that dependent property. Similar principles must also be applied to the development of in vitro alternatives to animal tests if those methods are to be reliable. A number of ways in which computer prediction systems and in vitro toxicology can complement each other in the development of alternatives to live animal experiments are described.

The ECVAM International Validation Study on In Vitro Tests for Skin Corrosivity. 1. Selection and Distribution of the Test Chemicals.

An international validation study on in vitro tests for skin corrosivity was conducted during 1996 and 1997 under the auspices of the European Centre for the Validation of Alternative Methods (ECVAM). The main objectives of the study were to assess the performances of selected in vitro tests in discriminating between: (a) corrosives (C) and non-corrosives (NC), for selected groups of chemicals (e.g. organic acids, phenols) and/or for all chemicals (single chemical entities only); and (b) known R35 (UN packing group I) and R34 (UN packing groups II & III) chemicals. Each test was evaluated for reliability and relevance by using a test set of 60 coded chemicals. In this paper, the test chemicals used in the validation study are identified; they include organic acids (6C/5NC), organic bases (7C/3NC), neutral organics (9NC), phenols (2C/3NC), inorganic acids (6C/1NC), inorganic bases (2C/2NC), inorganic salts (1C/2NC), electrophiles (3C/5NC) and soaps/surfactants (3NC). The in vivo classifications and important physicochemical properties (e.g. logP, pKa) of the test chemicals are given. The main criterion for including chemicals in the test set was that their corrosivity classifications were based on unequivocal animal data. Where available, structure-activity information was also used to support the corrosivity classifications. Despite the small numbers of chemicals in some of the categories, it was felt that the test set chosen represented the best possible for evaluating the performances of the in vitro tests for predicting skin corrosivity, given the limited availability of unequivocal animal data. The prediction of skin corrosivity from pH data was also investigated for those chemicals with extreme pH values (i.e. pH2 or 11.5). Nine of the 12 strongly acidic or alkaline chemicals in the test set, which were predicted to be C on the basis of their pH values, had also been found to be C in vivo.

QSARS for the eye irritation potential of neutral organic chemicals.

A quantitative structure-activity relationship (QSAR) was derived previously relating European Community (EC) eye irritation classification data of a set of neutral organic chemicals, to log(octanol/water partition coefficient), to the minor principal inertial axes (Ry and Rz) and to dipole moment. Eye irritation scores on a scale of 1-10 for a set of aliphatic alcohols (from the work of Smyth and Carpenter) have been shown to correlate well with the same four physicochemical parameters by means of neural network analysis. The original classification dataset of neutral organic chemicals has been augmented by the addition of a number of the aliphatic alcohols from the Smyth and Carpenter data that could unequivocally be assigned the EC classifications of irritant (those with eye irritation scores of 8 and 9) or non-irritant (scores of 1). Analysis of the extended dataset by both principal components and neural network analysis showed a clear discrimination between irritant and non-irritant chemicals using the same four physicochemical parameters. Predictions of EC eye irritation classifications for aliphatic alcohols with eye scores of 2-7, using the neural network model, showed that alcohols with eye scores of 2 and 3 lie on the classification boundary between irritant and non-irritant whereas those with scores of 4 and above are classified as irritant. These analyses support the validity of the original four-parameter eye irritation QSAR model for neutral organic chemicals. Furthermore, they provide a method for interrelating sets of in vivo data in which the biological response parameters are expressed in quite different formats, providing a means of utilizing historical data and thereby extending the availability of in vivo data suitable for the validation of in vitro alternative methods.

Computer prediction of possible toxic action from chemical structure: an update on the DEREK system.

Computer-based assessment of potential toxicity has become increasingly popular in recent years. The knowledge-base system DEREK is developed under the guidance of a multinational Collaborative Group of expert toxicologists and provides a qualitative approach to toxicity prediction. Major developments of the DEREK program and knowledge-base have taken place in the last 3 years. Program developments include improvements in both the user interface and data processing. Work on the knowledge-base has concentrated on the areas of genotoxicity and skin sensitisation. DEREK's predictive capabilities for these toxicological end-points has been demonstrated. In addition to the continued expansion of the knowledge-base, a number of enhancements are planned in the DEREK program. In particular, work is in progress to develop further DEREK's ability to report the reasoning behind its predictions.

Quantitative structure-activity relationships (QSARs) for skin corrosivity of organic acids, bases and phenols: Principal components and neural network analysis of extended datasets.

Quantitative structure-activity relationships (QSARs) relating skin corrosivity data of organic acids, bases and phenols to their log(octanol/water partition coefficient), molecular volume, melting point and pK(a). have been extended to substantially larger datasets. In addition to principal components analysis, as used in earlier work, the datasets have also been analysed using neural networks. Plots of the first two principal components of the four independent variables, which broadly model skin permeability and cytotoxicity, for each of the extended datasets confirmed that the analysis was able to discriminate well between corrosive and non-corrosive chemicals. Neural networks using the same parameters as inputs, were trained to an output in the range 0.0 to 1.0, with non-corrosive chemicals being assigned the value 0 and corrosive chemicals the value 1. As well as yielding classification predictions in agreement with those in the training sets, predicted outputs in the 0 to 1 range gave a useful indication of the confidence of the predicted classification. These QSARs are useful (a) for the prediction of the skin corrosivity potentials of new or untested chemicals and (b) for determining the confidence of predictions in regions of 'biological uncertainty' which exist at the classification threshold between corrosive and non-corrosive chemicals.

Skin corrosivity potential of fatty acids: In vitro rat and human skin testing and QSAR studies.

The corrosive potential of a series of fatty acids-propanoic acid (C3), butanoic acid (C4), hexanoic acid (C6), octanoic acid (C8), decanoic acid (C10) and dodecanoic acid (C12)-was investigated in the in vitro skin corrosivity test (IVSCT) using both rat skin and human skin. All the fatty acids with alkyl chain lengths up to and including C8 were found to be corrosive to rat skin. When human skin was used, the corrosive/non-corrosive threshold was shifted to around the C6 fatty acid. The results are discussed in the context of a QSAR for the corrosivity of organic acids, with the putative mechanism that corrosivity is a function of the ability of the chemical to permeate the skin together with its cytotoxicity, expressed in this case as acidity (pK(a)). This mechanistic interpretation is consistent with the known differences in barrier properties between rat and human skin.

In vitro investigations of the direct effects of complex anions on thyroidal iodide uptake: Identification of novel inhibitors.

Iodide uptake (IU) by thyrocytes from the plasma against chemical and electrical gradients is by a specific iodide transporter or 'pump'. Perchlorate (ClO(4)(-)) and other univalent, symmetrical anions are competitive inhibitors of iodide uptake (IU), and apparent K(i) for individual anions can be correlated with ion size. This study uses cultured thyrocytes and a broad range of anion size, in particular a series of spherical hexafluoride ions, in order to understand more about the parameters governing the activity of competitive inhibitors of IU. (125)I uptake and organification by cultured porcine thyrocytes was combined with biochemical enzyme inhibition studies on thyroid peroxidase in order to identify specific effects on IU. Known inhibitors were used in a validation phase and demonstrated that a combination of the in vitro thyrocyte (125)I assay and thyroid peroxidase inhibition assay could be used to identify selective inhibitors that can be difficult to identify using thyrocytes alone. Anions of less than, or similar, volume to I(-) (35.0 A(3)) were weak inhibitors with potency increasing proportional to ion size up to an apparent maximum for AsF(6)(-) (94.45 A(3)); this correlation was strong (r = 0.96). PF(6)(-), AsF(6)(-) and SbF(6)(-) were identified as novel inhibitors of IU, showing that the size range of anions active in IU inhibition is greater than that previously identified. The biological significance in vivo of the inhibitory action of the hexafluoride ions is not known. Their potency in this study suggests that these anions may have the potential to affect thyroid function in vivo if they were available systemically.

Quantitative structure-Activity relationships for skin irritation and corrosivity of neutral and electrophilic organic chemicals.

Quantitative structure-activity relationships (QSARs) have been derived by relating skin irritation and corrosivity data of neutral and electrophilic organic chemicals to their log(octanol/water partition coefficient) (logP), molecular volume, dipole moment and 1/molecular weight. Datasets were analysed using stepwise regression, discriminant and principal components analysis. Discriminant analysis between irritant and non-irritant neutral and electrophilic organic chemicals using the above parameters, which broadly model skin permeability (logP and molecular volume), 'reactivity' (dipole moment) and l/molecular weight to compensate for the fact that skin irritation/corrosivity testing is carried out using a fixed mass or volume of chemical, was found to discriminate well for only 73.1% of the dataset (67.3% cross-validated). The poor discrimination at the irritant/non-irritant classification boundary is attributed largely to biological variability. Stepwise regression analysis of the Primary Irritation Index (PII) for the same dataset showed a poor correlation (r(2) = 0.422; cross-validated r(2) = 0.201) with a positive dependence on logP and dipole moment and a negative dependence on molecular volume; l/molecular weight was not a significant variable. While this QSAR for PII has little value as a predictive model, mainly because of the large biological variability evident in PII values, it is useful in confirming the putative model for skin irritation. Discriminant analysis using logP, molecular volume and dipole moment, was able to discriminate reasonably well (92.9% well-classified; 92.9% cross-validated) between corrosive and non-corrosive electrophiles. A plot of the first two principal components of the same parameters showed a clear demarcation between corrosive and non-corrosive electrophiles. In contrast to the QSARs for skin irritation, increasing skin corrosivity was found to correlate with decreasing molecular volume, with increasing dipole moment, and with decreasing logP. The predominant parameter in determining the skin corrosivity of electrophilic organic chemicals appears to be the molar dose at which they are tested; this arises because skin corrosivity testing is conducted using a fixed mass or volume of chemical. A stepwise approach to the skin corrosivity/irritation classification of neutral and electrophilic organic chemicals is outlined. The derived QSARs should be useful for the prediction of the skin corrosivity potential of new or untested electrophiles. (Non-electrophilic neutral organic chemicals, as a category, do not generally appear to be corrosive.) Discrimination between some non-irritant and irritant neutral and electrophilic organic chemicals using these techniques is also possible. For a large number of chemicals whose irritation potentials lie in a fairly broad band around the irritant/non-irritant classification boundary, no firm prediction of classification is possible.