A Comparison of Reactivity Schemes for the Prediction Skin Sensitization Potential Grace Patlewicz,* ,† David W. Roberts, ‡ and E. Uriarte § European Commission, DG Joint Research Centre, Institute for Health and Consumer Protection, European Chemicals Bureau TP582, 21027 Ispra (VA), Italy, School of Pharmacy and Chemistry, LiVerpool John Moores UniVersity, LiVerpool L3 3AF, United Kingdom, and Department of Organic Chemistry, Faculty of Pharmacy, UniVersity of Santiago de Compostela, 15872 Santiago de Compostela, Spain ReceiVed September 14, 2007 Skin sensitization is an important toxic end point for both regulatory frameworks and safety assessment. There are many hurdles for a chemical to overcome in terms of inducing skin sensitization, although the binding of chemicals to skin protein is thought to be the rate-determining step. Current strategies to predict the skin sensitization potential of chemicals in silico is through the identification of electrophilic characteristics. A number of predictive schemes have been developed in recent years, some based on broad structural rules and some with a reaction chemistry mechanistic basis. This work compares two schemes that are based on reaction chemistry. The first scheme comprises a set of rules that characterize reaction mechanistic domains as proposed by Aptula and Roberts [(2006) Chem. Res. Toxicol. 19, 1097–1105]. The second is a set of structure-toxicity and structure-metabolism pathways that are encoded and embedded into the TIssue MEtabolism Simulator skin sensitization model (TIMES-SS) [(2005) Int. J. Toxicol. 24, 189–204]. Here, a comparison of these schemes has been made using a recently published data set of 210 chemicals that have been tested in the local lymph node assay. The similarities and differences of the schemes are highlighted, together with modifications that could be made to TIMES-SS to harmonize the two approaches. Introduction Skin sensitization is an important end point that needs to be evaluated in the context of regulatory frameworks such as Registration, Evaluation, Authorisation, and Restriction of Chemicals (REACH) 1 (1) as well as for risk management purposes within company safety evaluation assessments. Current strategies in the prediction of skin sensitization rely on in vivo testingsprincipally the local lymph node assay (LLNA), which provides a measure of relative sensitizing potency, a critical requirement for risk management purposes. Alternative approaches encompass in vitro technologies and in silico techniques such as (Q)SARs, read-across, or chemical categories (2). It is worth mentioning here in brief the context of the skin sensitization mechanism that shapes the factors that are thought to be important in the induction of skin sensitization. Clearly, for induction to occur, a chemical must overcome a number of hurdles (3). These include penetration into the viable epidermis and binding to skin proteins followed by lymphocyte prolifera- tion as part of the immune response. In this framework, the critical factor is really the binding step; that is, a chemical has to react with the skin protein to form an immunogenic complex in order for the rest of the immune response to occur. The nature of this binding is not fully established, although the current paradigm states that the binding is covalent; that is, the chemical behaves as an electrophile, and the protein behaves as the nucleophile. No doubt, there are other pathways to consider, such as free radical binding and so on, but it is believed that in the majority of cases the binding is covalent; hence, a range of electrophilic-nucleophilic reaction pathways play a role in how a chemical will react (4). The early strategies in silico that attempt to predict the likely sensitizing behavior of new chemicals have very much relied on this paradigm of covalent bonding. The types of approaches used for predicting skin-sensitizing effects span both qualitative and quantitative methods. The first quantitative method was that proposed by Roberts and Williams in 1982 when the so-called relative alkylation index (RAI) was defined (5). The fundamental basis of the approach is simple; sensitization is treated as a function of hydrophobicity and reactivity. The approach has been successfully applied to evaluate a number of data sets of chemicals including sultones, aldehydes, ketones, sulfonates, etc (6). The approach has been traditionally geared toward specific chemical classes and hence, until recently, been of limited predictive coverage. The identification of electrophilic features in chemicals has been the focus of an array of work where qualitative associations between structure and sensitization response have been inves- tigated. Workers such as Barratt et al. (7) derived structural alerts for sensitizers that were subsequently encoded into the Derek for Windows (DfW) expert system (see later). Work conducted at the same time by Payne and Walsh (8) identified similar alerts that were also incorporated into the same expert system. Ashby et al. (9) defined broad structural relationships based on an evaluation of a large data set of LLNA skin sensitization data. Other workers such as Gerner et al. (10) and Zinke et al. (11) * To whom correspondence should be addressed. Tel: +39 0332 789616. Fax: +39 0332 786717. E-mail: grace.patlewicz@ec.europa.eu or patlewig@ hotmail.com. † European Commission. ‡ Liverpool John Moores University. § University of Santiago de Compostela. 1 Abbreviations: REACH, Registration, Evaluation, Authorisation, and Restriction of Chemicals; RAI, relative alkylation index; LLNA, local lymph node assay; GP, guinea pig; EC3, estimated concentration required to produce a SI of 3; SI, stimulation index. Chem. Res. Toxicol. 2008, 21, 521–541 521 10.1021/tx700338q CCC: $40.75  2008 American Chemical Society Published on Web 01/12/2008