ToxSci Advance Access originally published online on March 30, 2007
Toxicological Sciences 2007 97(2):417-427; doi:10.1093/toxsci/kfm064
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Quantification of Chemical Peptide Reactivity for Screening Contact Allergens: A Classification Tree Model Approach
* The Procter & Gamble Company, Miami Valley Innovation Center, Cincinnati, Ohio 45252
Laboratoire de Dermatochimie, UMR 7123, Université Louis Pasteur, Strasbourg, France
1 To whom correspondence should be addressed. Fax: (513) 627-0400. E-mail: gerberick.gf{at}pg.com.
Received February 6, 2007; accepted March 13, 2007
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
In the interest of reducing animal use, in vitro alternatives for skin sensitization testing are under development. One unifying characteristic of chemical allergens is the requirement that they react with proteins for the effective induction of skin sensitization. The majority of chemical allergens are electrophilic and react with nucleophilic amino acids. To determine whether and to what extent reactivity correlates with skin sensitization potential, 82 chemicals comprising allergens of different potencies and nonallergenic chemicals were evaluated for their ability to react with reduced glutathione (GSH) or with two synthetic peptides containing either a single cysteine or lysine. Following a 15-min reaction time with GSH, or a 24-h reaction time with the two synthetic peptides, the samples were analyzed by high-performance liquid chromatography. UV detection was used to monitor the depletion of GSH or the peptides. The peptide reactivity data were compared with existing local lymph node assay data using recursive partitioning methodology to build a classification tree that allowed a ranking of reactivity as minimal, low, moderate, and high. Generally, nonallergens and weak allergens demonstrated minimal to low peptide reactivity, whereas moderate to extremely potent allergens displayed moderate to high peptide reactivity. Classifying minimal reactivity as nonsensitizers and low, moderate, and high reactivity as sensitizers, it was determined that a model based on cysteine and lysine gave a prediction accuracy of 89%. The results of these investigations reveal that measurement of peptide reactivity has considerable potential utility as a screening approach for skin sensitization testing, and thereby for reducing reliance on animal-based test methods.
Key Words: allergens; alternatives; skin sensitization; peptide reactivity; prediction model.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Allergic contact dermatitis (ACD) resulting from skin sensitization is a critical toxicological endpoint evaluated for all new chemicals developed for consumer and/or occupational use. The acquisition of skin sensitization and the subsequent elicitation of an ACD reaction in the skin are processes dependent upon recognition of chemical allergens in the skin by Langerhans cells (LC) and the induction of specific T lymphocyte responses (Kimber et al., 2000
). The local lymph node assay (LLNA) is viewed as the most appropriate skin sensitization test method for the evaluation of chemicals that have potential to come in contact with the skin (Cockshott et al., 2006). The LLNA is based upon characterization of induced proliferative responses in draining lymph nodes following topical exposure of mice to chemicals (Gerberick et al., 2000; Kimber et al., 2002). However, there is a critical need to develop non-animal–based methods for the evaluation of new chemicals that will reduce significantly or eliminate the need for animals in skin sensitization testing in the future (Jowsey et al., 2006; Ryan et al., 2001, 2005). This is of particular importance in view of the forthcoming European Union ban on in vivo testing of cosmetic and toiletry ingredients following the publication of the Seventh Amendment to the Cosmetic Directive (European Union Seventh Amendment to Cosmetic Directive) and for Registration, Evaluation, and Authorization of Chemicals that requires evaluation of a large number of chemicals.
Fortunately, the underlying chemical and cellular mechanisms of ACD are relatively well understood to aid scientists in the development of alternative methods for skin sensitization testing. It is believed that for a chemical to function as a contact sensitizer (or allergen), it must be capable of penetrating into the viable epidermis, react with protein, induce local trauma, and be recognized by the immune system. Thus, characterization of skin sensitization must integrate various sources of information from a battery of assays representing the key steps of skin allergy (Jowsey et al., 2006). For example, investigators have undertaken recently the development of chemical reactivity screening methods for aiding in the assessment of a chemical's skin sensitization potential (Aptula et al., 2006; Divkovic et al., 2005; Gerberick et al., 2004; Kato et al., 2003; Natsch et al., in press).
The correlation of skin protein reactivity and skin sensitization is well established and has been known for many years (Dupuis and Benezra, 1982; Landsteiner and Jacobs, 1936; Lepoittevin et al., 1998). It is accepted that if a chemical is capable of reacting with protein directly or after appropriate biotransformation, then it has the potential to act as an allergen. While a variety of mechanisms contribute to protein reactivity, it is generally recognized that this process involves the reaction of a small molecule, having electrophilic properties, with a nucleophilic amino acid on a protein. The majority of chemical allergens (or their metabolites) have electrophilic properties and are able to react with various nucleophiles to form covalent bonds. In proteins, the side chains of many amino acids contain electron-rich groups, nucleophiles, capable of reacting with electrophilic allergens. Lysine and cysteine are those most often cited, but other amino acids containing nucleophilic heteroatoms, such as histidine, methionine, and tyrosine, can also react with electrophiles (Ahlfors et al., 2003; Dupuis and Benezra, 1982; Lepoittevin et al., 1998).
Measuring chemical reactivity on nucleophile-containing peptides has potential utility for evaluating the skin sensitization potential of chemicals (Gerberick et al., 2004). Specifically, it was demonstrated that peptides containing either cysteine or lysine along with glutathione (GSH) served as surrogate nucleophiles to quantitatively measure chemical reactivity. The purpose of this work was to examine the reactivity of a large set of test chemicals (38 from original study and 44 new chemicals for a total of 82) using lysine, cysteine, and GSH peptides at different peptide to chemical molar ratios to determine whether the degree of reactivity correlated with the compound's sensitization potency. The data were analyzed using a classification tree model approach to develop a pragmatic prediction model for assessing and interpreting the peptide reactivity assay data.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Test chemicals.
The rationale for selecting chemicals for evaluation was based on chemical diversity and on the availability of robust LLNA data representing a good distribution of weak, moderate, strong, and extreme allergens along with nonallergens. The following chemicals with accompanying purity and CAS numbers were purchased from Aldrich Chemical Company (Milwaukee, WI): 2-acetylcyclohexanone, 97% (874-23-7); a-amylcinnamaldehyde, 85% (122-40-7); benzaldehyde, 95% (100-52-7); 1,2-benzisothiazolin-3-one, 97% (2634-33-5); benzyl benzoate, 99% (120-51-4); benzylideneacetone, 99% (122-57-6); 1-butanol, 99.5% (71-36-3); chlorobenzene, 99% (108-90-7); cinnamaldehyde, 99% (14371-10-9); coumarin (91-64-5); cyclamen aldehyde, 90% (103-95-7); diethyl maleate, 97% (141-05-9); diethyl phthalate, 99.5% (84-66-2); diphenylcyclopropenone, 98% (886-38-4); ethyl acrylate, 99% (140-88-5); ethyl vanillin, 99% (121-32-4); ethyleneglycol dimethacrylate, 98% (97-90-5); farnesal, 85% (19317-11-4); formaldehyde (50-00-0); 2,4-heptadienal, 90% (5910-85-0); hexenal, 98% (6728-26-3); a-hexylcinnamaldehyde, 85% (101-86-0); 4-hydroxybenzoic acid, 99% (99-96-7); hydroxycitronellal, 95% (107-75-5); 2-hydroxyethyl acrylate, 96% (818-61-1); 2-hydroxypropyl methacrylate, 97% (923-26-2); 2-mercaptobenzothiazole, 98% (149-30-4); 6-methyl coumarin, 99% (92-48-8); methyl salicylate, 99% (119-36-8); methyl-2-nonynoate, 99% (111-80-8); methylparaben, 99% (99-76-3); metol, 99% (55-55-0); nonanoyl chloride, 96% (764-85-2); oxalic acid, 99% (144-62-7); oxazolone, 90% (15646-46-5); perillaldehyde, 92% (2111-75-3); phenylacetaldehyde, 90% (122-78-1); 2-phenylpropionaldehyde, 98% (93-53-8); phthalic anhydride, 99% (85-44-9); propyl gallate, 98% (121-79-9); propyl paraben, 99% (94-13-3); resorcinol, 99.5% (108-46-3); salicylic acid, 99% (69-72-7); squaric acid, 99% (2892-51-5); vanillin, 99% (121-33-5); vinylidene dichloride, 99% (75-35-4); vinyl pyridine (1337-81-1).
The following chemicals with accompanying purity and CAS numbers were purchased from Sigma Chemical Company (St Louis, MO): p-benzoquinone, 98% (106-51-4); CD3 (25646-71-3); 2,4-dinitrochlorobenzene, 99% (97-00-7); glutaraldehyde, 70% (111-30-8); imidazolidinyl urea, 95% (39236-46-9); isopropanol, 99% (67-63-0); isopropyl myristate, 98% (110-27-0); lactic acid, 85% (50-21-5); 1-(4-methoxyphenyl)-1-penten-3-one (104-27-8); nonanoic acid, 97% (112-05-0); octanoic acid, 98% (124-07-2); sulfanilamide, 99% (63-74-1); sulphanilic acid, 99% (121-57-3); trimellitic anhydride, 97% (552-30-7).
The following chemicals with accompanying purity and CAS numbers were purchased from Fluka Chemical Company (Milwaukee, WI): 4-allylanisole, 98% (140-67-0); benzoyl peroxide, 97% (94-36-0); 1-bromobutane, 99% (109-65-9); 2,3-butanedione, 99% (431-03-8); 5-chloro-2-methyl-4-isothiazolin-3-one (26172-55-4); ethylbenzoylacetate, 97% (94-02-0); fluorescein isothiocynate, 98% (3326-32-7); glyoxal (107-22-2); lilial, 95% (80-54-6); 4-methoxyacetophenone, 99% (100-06-1); palmitoyl chloride, 98% (112-67-4); propylene glycol, 99.7% (57-55-6); 2,2,6,6-tetramethyl-3,5-heptanedione, 98% (1118-71-4).
Hexane (110-54-3) was purchased from EM Science (Gibbstown, NJ). Bandrowski's base was purchased from ICN (Costa Mesa, CA). Glycerol, 99% (56-81-5) was purchased from J.T Baker (Phillipsburg, NJ).Tetracholorosalicylanilide (1154-59-2) was purchased from Eastman Kodak Company (Rochester, NY). Lauryl gallate, 98% (1166-52-5) was purchased from Alfa Aesar (Ward Hill, MA). 5-Methyl-2,3-hexandione (13706-86-0) was purchased from Penta MFG (Livingston, NJ). Kathon CG (55965-84-9) was purchased from Rohm & Haas (Philadelphia, PA). 2-Methyl-2H-isothiazol-3-one (2682-20-4) was supplied by J.-P. L.
LLNA protocol and chemicals tested.
The LLNA data reported in this manuscript are derived from previously conducted studies. The LLNA studies were conducted as described elsewhere (Gerberick et al., 2000; Kimber et al., 2002). Briefly, groups of CBA female mice (7–12 weeks of age) were exposed topically on the dorsum of both ears to 25 µl of test material or to an equal volume of the relevant vehicle alone. Treatment was performed daily for three consecutive days. Five days following the initiation of exposure, all mice were injected via the tail vein with 250 µl of phosphate-buffered saline containing 20 µCi of tritiated thymidine. Mice were sacrificed 5 h later, and the draining lymph nodes were excised for each experimental group. The incorporation of tritiated thymidine measured by ß-scintillation counting was reported in disintegrations per minute (dpm). A stimulation index (SI) was calculated for each allergen-treated group as the ratio of the dpm of the treated group over the dpm of the concurrent vehicle control. A substance was classified as a skin sensitizer if at one or more test concentrations it induced a threefold or greater increase in local lymph node proliferative activity compared with concurrent vehicle-treated controls.
Potency estimation in the LLNA.
The method used to determine the relative skin sensitization potency of a chemical has been previously described and is based upon the mathematical estimation of the concentration necessary to induce a threefold increase in the proliferative activity in the draining lymph nodes relative to vehicle-treated mice (Basketter et al., 1999). This estimated concentration, known as the EC3 value, is calculated by conducting a linear interpolation of coordinates above and below the value of three on the LLNA dose-response plot. The EC3 value for chemicals which had an SI greater than three for the lowest concentration tested was extrapolated from the two lowest doses evaluated (Ryan et al., in press). EC3 values extrapolated by this method were calculated by log-linear interpolation between these two points on a plane in which the dose level and SI are represented on the x-axis and y-axis, respectively. Existing dose-response data from previously conducted LLNA experiments have been used to calculate the EC3 values for the chemicals used in this manuscript. An arbitrary classification scheme based on EC3 values was used for categorizing the relative skin sensitization potency of chemicals evaluated in this study (Kimber et al., 2003). This system classifies the sensitization potency of a chemical as extreme (EC3 < 0.1), strong (EC3 
0.1 to < 1), moderate (EC3 1 to < 10), weak (EC3 10 to 
100), and nonsensitizing are not calculated.
The specific EC3 potency data used in this paper for the majority of chemicals are found in a recently published LLNA database paper (Gerberick et al., 2005). References for the other chemicals used are as follows: diphenylcyclopropenone (Ryan et al., 2000), phthalic anhydride (Dearman et al., 1992), oxazolone (Loveless et al., 1996), propyl gallate (Ashby et al., 1995; Basketter and Scholes, 1992), metol (Ashby et al., 1995; Basketter and Scholes, 1992), benzoyl peroxide (Kimber et al., 1998), squaric acid (Ryan et al., 2000), 2-methyl-2H-isothiazol-3-one (Estrada et al., 2003), lilial (Basketter et al., 2001), and nonanoic acid (P&G, unpublished data).
GSH, cysteine, and lysine peptide depletion assays.
A method to measure reactivity of a test chemical with reduced GSH was recently developed (Gerberick et al., 2004) which is based on a previously described method (Farriss and Reed, 1987). Briefly, 50 µl of a 2mM GSH stock solution prepared in oxygen-free 100mM sodium phosphate buffer (pH 7.4) and 50 µl of a 200mM test chemical prepared in dimethyl sulfoxide (DMSO) were added to 400 µl of oxygen-free 100mM sodium phosphate buffer (pH 7.4). The final reaction, containing 0.2mM GSH and 20mM of the test chemical, representing 1:100 molar ratio, was mixed and incubated for 15 min at 25°C with agitation. Control samples and standards used for defining the calibration curve for each analysis were prepared without test chemical for GSH (0.05–200mM) and glutathione disulfide (GSSG) (0.025–100mM). All samples were prepared in triplicate. Following incubation, GSH and GSSG in the samples and standards were derivatized with iodoacetic acid and 2,4 dinitrofluorobenzene. Derivatized GSH and GSSG were separated and quantitated by reverse-phase high-performance liquid chromatography (HPLC) on a Waters Alliance 2695 system (Waters Corporation, Milford, MA) using a Waters UV detector (365 nm) and a Waters Spherisorb NH2 analytical column (3 µm, 2.0 x 100 mm) under gradient conditions. Total GSH (GSH equivalents as GSH or GSSG) in each sample was determined from the calibration curve and used to calculate the percent peptide depletion relative to the mean concentration of total GSH in the control sample (no test chemical).
A method to measure reactivity of a test chemical with model heptapeptides containing lysine (Ac-RFAAKAA-COOH) or cysteine (Ac-RFAACAA-COOH) was recently developed (Gerberick et al., 2004). Peptides were prepared and purified by the SynPep Corporation (Dublin CA, USA) to > 90% purity as measured by HPLC, and molecular weight confirmation was determined by flow injection positive-ion electrospray mass spectrometry. Briefly, 400 µl of a 1.25mM peptide stock solution prepared in buffer and a 100mM test chemical stock solution prepared in either acetonitrile or DMSO/acetonitrile were added to 100mM ammonium acetate buffer (pH 10.2) for the lysine peptide or 100mM sodium phosphate buffer (pH 7.5) for the cysteine peptide. The final reaction, containing 0.5mM of the peptide and 5 or 25mM of the test chemical, representing 1:10 and 1:50 molar ratios, was mixed and incubated in the dark for 24 h at 25°C. Control samples and standards used for defining the calibration curve for each analysis were prepared without test chemical for each peptide and ranged from 0.0156 to 1.0mM. All samples were prepared in triplicate. Following incubation, the peptide was quantified by reverse-phase HPLC (Waters 2695 Alliance) on a Zorbax SB-C18 column (3.5 µm, 100 x 2.1 mm) with UV detection at 220 nm (Waters 996 PDA detector) using an external standard linear calibration curve. The UV spectrum was collected from 210 to 400 nm to permit verification of the peptide peak identity. Peptide reactivity was reported as percent depletion based on the decrease in nonreacted peptide concentration in the sample relative to the average concentration measured in the control.
Classification tree model development.
The goal was to develop a prediction model that would quantify in some way peptide depletion as related to level of reactivity. Various models were developed using classification tree methodology (Brieman et al., 1983) and the recursive partitioning routines implemented in S-Plus 7.0 statistical software (2003, Insightful Corp., Seattle, WA). During model development, each peptide at each concentration (GSH 1:100, cysteine 1:10, cysteine 1:50, lysine 1:10, and lysine 1:50) was considered as a potential predictor. For each of the 82 chemicals examined, three measurements were taken as the percentage of peptide depletion for each peptide/concentration, and the average depletion was determined. Some potential models were developed using these peptide depletion averages as predictors, while other potential models were developed using peptide depletion percentages averaged further across various peptides/concentrations (e.g., the average of the cysteine 1:10 and lysine 1:50 percentages). Specifically, models were developed using the following peptide depletion values as potential predictors in each model: model #1, average of all five peptides as the only predictor; model #2, cysteine 1:10, cysteine 1:50, GSH 1:100, lysine 1:10, lysine 1:50; model #3, average of cysteine 1:10, GSH 1:100, and lysine 1:10 as the only predictor; model #4, average of cysteine 1:10, cysteine 1:50, lysine 1:10, and lysine 1:50 as the only predictor; model #5, average of cysteine 1:10 and lysine 1:50 as the only predictor; model #6, cysteine 1:10, cysteine 1:50, lysine 1:10, lysine 1:50.
Classification tree building begins with the root node, which includes all of the chemical compounds in the learning data set (a total of 56). Beginning with this node, if more than one variable is considered, S-Plus software finds the best possible variable (peptide and concentration) to split the node into two child nodes. In order to find the best peptide/concentration, the software checks all possible peptides/concentrations as well as all possible values of the peptide/concentration used to split the node. For example, suppose that an attempt is made to build a tree using cysteine 1:10 and lysine 1:10. For each of these peptides, the individual chemicals are rank ordered based on their depletion values as potential predictors. The root node is then split into two child nodes using the average of two adjacent values of one of the peptides. The rank-ordered cysteine 1:10 values are – 10, – 3.8, – 3.7, – 1.9, – 1.3, ..., 100, so child nodes are created by splitting the root node based on cysteine 1:10 at – 6.9 (the average of – 10 and – 3.8), – 3.75 (the average of – 3.8 and – 3.7), etc. One child node represents all chemicals with peptide depletion less than the specified value, and the other node represents all chemicals with peptide depletion greater than the specified value. The software then seeks to maximize the average "purity" of the two child nodes. In other words, a pair of child nodes in which one node contains all nonsensitizers and weak sensitizers and the other node contains all moderate and strong sensitizers would be superior to a pair of child nodes in which each node contains a mix of chemicals from each sensitization category. Once the best pair of child nodes is determined, the process that was used on the root node is repeated on each child node. The splitting of nodes into child nodes continues in an iterative manner until the level of purity in the child nodes reaches a reasonable level or until a minimum sample size per node is reached.
In the various models fit to peptide reactivity data, splits were made until there were a total of four child nodes. Once these nodes were determined, each node was named based on the sensitization category most often represented in each node. The names assigned to each node include "minimal reactivity," "low reactivity," "moderate reactivity," or "high reactivity" (corresponding to the prediction of non-, weak, moderate, and strong/extreme sensitizers, respectively). Chemicals of a different sensitization category than the category of the node in which they are included are considered to be misclassified.
This modeling procedure was conducted six times (once for each potential model previously mentioned) based on all 56 chemicals in the learning data set. Once the six models were determined, they were tested on 26 additional chemicals. Some advantages to using classification trees for prediction include ease of variable selection and model interpretation, no assumptions regarding the distribution of the data, predictor interaction effects are taken into account, models can discriminate on one or more variables, differing costs can be taken into account for different types of misclassification, and more than two response levels are easily handled.
The final classification tree model was assessed via scatterplots and by calculating Cooper statistics (Cooper et al., 1979). Cooper statistics were used to determine how well the model distinguished sensitizers (of any strength) and nonsensitizers. The Cooper statistics calculated include sensitivity (the proportion of true sensitizers predicted as having low, moderate, or high reactivity), specificity (the proportion of true nonsensitizers predicted as having minimal reactivity), positive predictivity (the proportion of chemicals classified as having low, moderate, or high reactivity that are true sensitizers), negative predictivity (the proportion of chemicals classified as having minimal reactivity that are true nonsensitizers), and accuracy (the overall proportion of correct predictions). Cooper statistics were computed on the entire set of chemicals (up to 56 training set chemicals and 26 validation set chemicals).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Peptide Reactivity Data with GSH, Lysine, and Cysteine
Peptide depletion results on 38 chemicals using GSH, cysteine, and lysine peptides were previously published (Gerberick et al., 2004
). The ratios of peptide to chemical used were 1:100 for GSH, 1:50 for lysine, and 1:10 for cysteine. The results indicated a strong correlation between allergen potency and depletion of the nonreacted peptide. In this study, we have expanded the number of chemicals evaluated to 82 and added two experimental conditions: cysteine at 1:50 and lysine at 1:10. The results for the 82 test chemicals are presented in Table 1. The chemicals are listed in the order of lowest EC3 values (i.e., the most potent allergens) through nonsensitizers and include 18 extreme/strong sensitizers; 19 moderate sensitizers; 15 weak sensitizers; and 30 nonsensitizers based on an existing LLNA categorization scheme (Kimber et al., 2003). The LLNA EC3 data reported in this manuscript are derived from previously conducted studies (Ashby et al., 1995; Basketter and Scholes, 1992; Basketter et al., 2001; Dearman et al., 1992; Estrada et al., 2003; Gerberick et al., 2005; Kimber et al., 1998; Loveless et al., 1996; Ryan et al., 2000). Generally, it is evident that the more potent the allergen, the more peptide depletion that is observed (Table 1), specifically for the GSH and cysteine peptides. For the majority of the extreme and strong allergens, greater than 75% depletion was observed for the 1:50 cysteine peptide. Less peptide depletion was noted for the GSH and 1:10 cysteine peptides but again generally more depletion was observed with the more potent allergens. With 1:10 and 1:50 lysine peptides, peptide depletion was greater also with the more potent allergens but not to levels of the cysteine-containing peptides. Interestingly, phthalic anhydride and trimellitic anhydride demonstrated significant depletion with GSH and lysine 1:50 peptides but not with the others. Finally, only a few of the nonsensitizers (e.g., 2-hydroxypropyl methacrylate) demonstrated peptide depletion values similar to those observed with the allergens suggesting good specificity for peptide reactivity assays. It is important to note that for a few test compounds (1,2-benzisothiazolin-3-one; 2-hydroxypropyl methacrylate; ethyl vanillin; and salicylic acid), one or two peptide depletion values are missing due to either an incompatibility with the solvent system or the test compounds coeluted with peptide in the HPLC analysis.
|
Prediction Models Based on Classification Tree Model Analysis
One requirement for using the peptide reactivity assay data for screening the skin sensitization potential of chemicals was to find a robust method to analyze and categorize the data. Another need was to determine if each of the five peptides were necessary for screening since the possibility of reducing the amount of work necessary to analyze each chemical would increase throughput and reduce the cost and amount of test material required for testing. To address these two needs, we chose to use a classification tree model approach which is a form of binary recursive partitioning that is used when observations need to be assigned to a category based on a number of predictor variables (Brieman et al., 1983
). Specifically, the classification tree approach used an algorithm to evaluate all of the peptide reactivity depletion data for each chemical in the context of its known LLNA potency category. Table 2 lists six prediction models that were developed based on use of all of the peptide data or limited to the use of specific peptide data (e.g., exclusion of GSH data). For each model generated, the model predictors used for evaluation of the chemical data set is given along with the model's accuracy and number of misclassifications. Cooper statistics were used to determine how well the model distinguished sensitizers (of any strength) and nonsensitizers. The Cooper statistics calculated accuracy based on chemicals predicted as sensitizers if they were categorized as having low, moderate, or high reactivity versus chemicals predicted as nonsensitizers if they were categorized as having minimal reactivity. It is clear from use of Cooper statistics analysis that model #1, which incorporates all of the peptides and their ratios, delivers the highest accuracy and fewest number of misclassifications. However, it is important to note that the delivered accuracy for the other models (#2–#6), which incorporate fewer peptides (e.g., model #5), is not that dissimilar to the more peptide comprehensive model #1. The similarity between the models suggest that for screening purposes it might be adequate to go with a model that uses fewer peptides and thus requires less material and less time for analysis. Comparison shows that the model #1 (sum of GSH; cysteine 1:10 and 1:50; and lysine 1:10 and 1:50) yields an accuracy of 94% and five misclassifications, whereas a model based only on cysteine 1:10 and lysine 1:50 yields a reasonable accuracy of 89% and nine misclassifications. Thus, we chose model #5 which includes only cysteine 1:10 and lysine 1:50 as predictors for analyzing further our peptide reactivity data.
|
Cysteine 1:10 and Lysine 1:50 Classification Decision Tree Model (Model #5)
The decision tree model that incorporates cysteine 1:10 and lysine 1:50 as predictors is presented in Figure 1. The model is based on making decisions on the average of peptide depletion data for cysteine 1:10 and lysine 1:50. As indicated in the methods, the classification decision tree model was developed using the peptide depletion data along with the LLNA potency data. By incorporating different cutoffs generated by the model, we have chosen to name the peptide depletion reactivity categories as minimal, low, moderate, and high reactivity. Generally, chemicals with moderate to high reactivity are associated with moderate to strong allergenicity, while those categorized as having minimal to low reactivity include weak and nonsensitizers (Table 3). However, it would be inappropriate to consider that a simple peptide reactivity assay would have the capability to predict a chemical's sensitization potency. It is believed that to accomplish this task, additional assay data will be needed to make an accurate prediction of a chemical's skin sensitization potential (Jowsey et al., 2006
). As far as the capability of using this model for classifying a chemical as a sensitizer or nonsensitizer, the Cooper statistics show that this model performs very well with an accuracy of 89% (Table 2 and Fig. 2). The high values for the sensitivity (88%), specificity (90%), positive predictivity (94%), and negative predictivity (81%) suggest that this peptide reactivity model would perform well as a screening assay, especially if used along with other physiochemical or biological data (Fig. 2). The nine chemicals that are misclassified include six sensitizers (
-hexylcinnamaldehyde, -amylcinnamaldehyde, benzyl benzoate, 2,2,6,6-tetramethyl-3,5-heptanedione, oxalic acid, and nonanoyl chloride) and three nonsensitizers (2-acetylcyclohexanone, 2-hydroxypropyl methacrylate, and 1-bromobutane).
|
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Our understanding of the chemical and biological processes associated with skin sensitization and ACD has advanced significantly in recent years. This knowledge is providing the foundation for the development of numerous alternative methods for skin sensitization testing. One particular area of development has been to apply our knowledge of how chemical reactivity plays an important role in the initiation of skin sensitization response (reviewed in Lepoittevin et al., 1998
). Specifically, we (Gerberick et al., 2004) as well as others (Aptula et al., 2006; Divkovic et al., 2005; Gerberick et al., 2004a; Kato et al., 2003; Natsch et al., in press) have addressed the development of chemical reactivity screening methods for assessing the skin sensitization potential of chemicals.
It is believed that the formation of hapten-protein complexes is a prerequisite for the initiation of skin sensitization and which occurs prior to the processing of the complexes by antigen-presenting LC in the skin for the eventual presentation of the chemical to antigen-specific T cells. Chemical allergens (haptens) or their metabolites are small molecular weight compounds (generally less than 500 Da) with electrophilic properties. They are able to react with nucleophiles to form covalent bonds. In proteins, the side chains of many amino acids contain electron-rich groups capable of reacting with allergens. Lysine and cysteine are those most often cited but other amino acids containing nucleophilic heteroatoms, such as histidine, methionine, and tyrosine, can react with electrophiles (Ahlfors et al., 2003; Dupuis and Benezra, 1982; Lepoittevin et al., 1998). Thus, electrophilic allergens are believed to react with nucleophilic amino acids to form a stable covalent bond which is critical to the initiation of a skin sensitization response. However, it must be realized that other mechanisms for hapten interaction need to be considered as well (Divkovic, 2006).
Since reactivity is one key step in the induction of skin sensitization, we have been interested in pursuing whether measuring a chemical's reactivity could be used to develop a quantitative peptide-based reactivity assay that would have utility for screening a chemical's skin sensitization potency as defined in the LLNA. We evaluated 38 chemicals representing allergens of different potencies (weak to extreme) and nonsensitizers for their ability to react with GSH or three synthetic peptides containing either cysteine, lysine, or histidine (Gerberick et al., 2004). The results demonstrated that a significant correlation exists between allergen potency and the depletion of GSH, lysine, and cysteine but not histidine. It is important to note that our intent in developing a peptide reactivity approach was not for the purpose of reproducing the physiological conditions of reactivity. For example, the lysine peptide assay must be run at pH 10.5 for optimal reactivity of the amine group. Moreover, we have chosen to focus on only two nucleophiles, lysine and cysteine, for use in developing a screening assay for determining a chemical's reactivity potential. Although lysine has been demonstrated to be an important nucleophile for allergens such as sultones and methylisothiazolone derivaties (Alvarez-Sanchez et al., 2003; Meschkat et al., 2001) and cysteine for ,ß-unsaturated allergens (Ahlfors et al., 2003), it is probable that nucleophiles other than lysine and cysteine are critical for the initiation of a sensitization response (Divkovic, 2006; Divkovic et al., 2005). Thus, our approach will yield minimal information on how a specific chemical reacts with protein in vivo but does provide a means of quantifying reactivity for the purpose of screening skin sensitization potential.
Using GSH as a cysteine-containing peptide and two synthetic hepapeptides, one with lysine and other with cysteine, we expanded our analysis of chemical reactivity from 38 to 82 chemicals. The chemicals represented in the data set comprise weak (n = 15), moderate (n = 19), strong and extreme sensitizers (n = 18), as well as nonsensitizing materials (n = 30), as based on potency categorization criteria that have been developed by a European Centre for Ecotoxicology and Toxicology of Chemicals Task Force (Kimber et al., 2003). The LLNA EC3 values listed in Table 3 show a range of potency from 0.0003% for the extreme allergen, diphenylcyclopropenone, to 33% for the weak allergen, hydroxycitronellal.
The results, as summarized in Table 3, demonstrate that an association between the degree of peptide reactivity (as measured by nonreacted peptide depletion) and sensitization potency is evident. To help with interpretation of the peptide depletion data, we examined the utility of using classification tree methodology for development of a prediction model. Classification tree methodology involves an algorithm to group data based on one or more predictors. In this particular case, we used the peptide depletion data for each of the peptides (predictors) to see which ones would be used to subgroup the data. Although we used LLNA potency data for each of the 82 compounds to build the model, we chose to use high, moderate, low, and minimal reactivity as the category names for groups determined by the model. To evaluate each of the models for their hazard identification ability, we considered any compound that was categorized as high, moderate, or low as a skin sensitizer and those categorized as minimal as nonsensitizers. Cooper statistics were used to determine how well the different models distinguished sensitizers from nonsensitizers. All of the models generated are listed in Table 2. Model #1, that incorporated all of the peptide depletion data for each of the peptides, demonstrated a prediction accuracy of 94%. In addition, this model yielded only five misclassifications. Although the performance of this model is outstanding, it has the limitation of requiring the use of five different peptides. Moreover, model #1 involves the use of the GSH assay which involves a multiple-step procedure which can be challenging to transfer to other laboratories (data not shown). Thus, we were interested to see if a robust model could be developed that did not incorporate GSH into the decision tree. Table 2 summarizes the classification models and lists for each model the accuracy and number of misclassifications obtained with the model. Although the accuracy values are lower for the "simpler" models, they still show a very good ability to distinguish sensitizers from nonsensitizers. The one model we think demonstrates a good compromise between requiring fewer peptide ratios for analysis and no GSH is model #5, which includes use of cysteine at 1:10 and lysine at 1:50. Model #5 has a prediction accuracy of 89% with nine misclassifications. Of the six sensitizers classified as nonsensitizers, five of them are weak sensitizers (e.g., a-hexylcinnamaldehyde, benzyl benzoate). Nonanoyl chloride, a moderate sensitizer, was classified as a nonsensitizer. For chemicals that are misclassified, it is important to consider the chemical's water solubility as related to compatibility with the assay conditions as well as the possibility that the chemical is a prohapten and might require bioactivation prior to it reacting with nucleophile-containing peptides. Moreover, in some instances it might be prudent to review the LLNA data used to categorize a chemical as a sensitizer or nonsensitizer. For example, oxalic acid is categorized as weak sensitizer in the LLNA but the chemical does not contain apparent alerts or does human data exist to classify it as a sensitizer. Interestingly, a few of the nonsensitizers identified as sensitizers are compounds believed to have reactive properties (e.g., 2-hydroxypropyl methacrylate, 1-bromobutane). Moreover, two anhydride compounds are identified with the use of the lysine 1:50 peptide which support the use of using both cysteine and lysine for screening unknown chemicals. In addition to providing good assistance for hazard identification, the cysteine 1:10 and lysine 1:50 model provides quantifying data on reactivity that has potential for use, along with other data, for predicting the skin sensitization potency of an unknown chemical. Consistent with what we have observed in the past (Gerberick et al., 2004), the amount of peptide depletion corresponds closely with the allergenic potency of the compound. Generally, moderate, strong, and extreme sensitizers show moderate to high reactivity, while weak and nonsensitizers show minimal to low reactivity (Table 3).
Of course, one would not expect an extremely high correlation between reactivity and potency since other factors, such as skin penetration and immune recognition by T cells, are critical for the acquisition of skin sensitization. It is not possible to say which event is most critical and it is likely unique for each chemical (e.g., reactivity for one chemical versus bioavailability for another chemical). Thus, it is very important to point out that it is not expected that the peptide reactivity assay alone should have the ability to predict a compound's sensitization potential. It is believed that to replace the LLNA, a battery of assays will be needed to reproduce the complex chemistry and biology that are involved in the induction of skin sensitization. Jowsey et al. (2006) have described this need in a very informative way by showing how different assays could provide quantitative information on different aspects known about the mechanism of ACD. For example, the peptide reactivity assay could serve as a first tier screening assay and also provide in time information needed to complete a holistic assessment of a chemical's skin sensitization potential.
One potential challenge for developing alternative methods for skin sensitization testing is that it is well known that some chemical allergens are prohaptens and as such require biotransformation prior to initiating a skin sensitization response in vivo (Smith and Hotchkiss, 2001). The need for biotransformation has been demonstrated with many chemicals, such as the formation of benzoquinonediimine from azo hair dyes (Basketter and Goodwin, 1988), or orthoquinone from isoeugenol (Bertrand et al., 1997). Based on the knowledge that some chemical allergens need to be biotransformed prior to reacting with proteins/peptides, it will be critical to incorporate a metabolism component to address these types of molecules. We are currently evaluating a peroxidase/peroxide oxidizing system for use in a modified peptide reactivity assay.
The goal of this work was to evaluate the use of chemical reactivity as a means for screening the skin sensitization potential of chemicals. A prediction model was developed using a classification tree approach which allowed ranking the reactivity as minimal, low, moderate, or high as well as for assessing skin sensitization hazard. The results presented show clearly that using a cysteine- and lysine-based peptide depletion, assay demonstrates a good, but not perfect, association between chemical reactivity and allergenic potency. Generally, moderate to extreme allergens demonstrate high peptide depletion whereas weak and nonsensitizers demonstrate significantly less peptide depletion. It is hoped that with additional information from other in vitro assays and modification of existing peptide reactivity assays (e.g., addition of metabolism component), this methodology will be even more helpful in reducing our reliance on animals for skin sensitization testing in the future.
|
ACKNOWLEDGMENTS |
|---|
The authors wish to acknowledge George Daston, Don Bjerke, and Petra Kern for their excellent review of the manuscript and Roy Dobson, Mike Quijano, Pam Riley, and John Troutman for technical contributions to this work. None of the P&G authors will benefit financially from this work other than their salaries. J.-P.L. works in collaboration with the P&G scientists and is not receiving any financial support from P&G for this work. This research was funded in part by a grant from the European Cosmetic Toiletry and Perfumer Association. Conflicts of interest: None declared.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Ahlfors SR, Sterner O, Hansson C. Reactivity of contact allergenic haptens to amino acid residues in a model carrier peptide, and characterization of formed peptide-hapten adducts. Skin Pharmacol. Appl. Skin Physiol. (2003) 16:59–68.[CrossRef][Web of Science][Medline]
Alvarez-Sanchez R, Basketter DA, Pease C, Lepoittevin JP. Studies of chemical selectivity of hapten, reactivity, and skin sensitization potency. 3. Synthesis and studies on the reactivity toward model nucleophiles of the 13C-labeled skin sensitizers, 5-chloro-2-methylisothiazol-3-one (MCI) and 2-methylisothiazol-3-one (MI). Chem. Res. Toxicol. (2003) 16:627–637.[Web of Science][Medline]
Aptula AO, Patlewicz G, Roberts DW, Schultz TW. Non-enzymatic glutathione reactivity and in vitro toxicity: A non-animal approach to skin sensitization. Toxicol. In Vitro (2006) 20:239–247.[CrossRef][Web of Science][Medline]
Ashby J, Basketter DA, Patton D, Kimber I. Structure activity relationships in skin sensitization using the murine local lymph node assay. Toxicology (1995) 103:177–194.[CrossRef][Web of Science][Medline]
Basketter DA, Goodwin BFJ. Investigation of the prohapten concept cross reaction between 1,4-disubstituted benzene derivatives in the guinea pig. Contact Derm. (1988) 19:248–253.[CrossRef][Medline]
Basketter DA, Lea LJ, Cooper K, Dickens A, Briggs D, Pate I, Dearman RJ, Kimber I. A comparison of statistical approaches to derivation of EC3 values from local lymph node assay dose responses. J. Appl. Toxicol. (1999) 19:261–266.[CrossRef][Web of Science][Medline]
Basketter DA, Scholes EW. Comparison of the local lymph node assay with the guinea pig maximization test for the detection of a range of contact allergens. Food Chem. Toxicol. (1992) 30:65–69.[Web of Science][Medline]
Basketter DA, Wright ZM, Warbrick EV, Dearman RJ, Kimber I, Ryan CA, Gerberick GF, White IR. Human potency predications for aldehydes using the local lymph node assay. Contact Derm. (2001) 45:89–94.[CrossRef][Medline]
Bertrand F, Basketter DA, Roberts DW, Lepoittevin J-P. Skin sensitization to eugenol and isoeugenol in mice: Possible metabolic pathways involving ortho-Quinone and quinone methide intermediates. Chem. Res. Toxicol. (1997) 10:335–343.[CrossRef][Web of Science][Medline]
Brieman L, Friedman JH, Olshen RA, Stone CJ. Classification and Regression Trees (1983) Wadsworth, Belmont, CA.
Cockshott A, Evans P, Ryan CA, Gerberick GF, Betts CJ, Dearman RJ, Kimber I, Basketter DA. The local lymph node assay in practice: A current regulatory perspective. Hum. Exp. Toxicol. (2006) 25:387–394.
Cooper JA, Saracci R, Cole P. Describing the validity of carcinogen screening tests. Br. J. Cancer (1979) 39:87–89.[Web of Science][Medline]
Dearman RJ, Spence LM, Kimber I. Characterization of murine immune responses to allergenic diisocyanates. Toxicol. Appl. Pharmacol. (1992) 112:190–197.[CrossRef][Web of Science][Medline]
Divkovic M. Hapten-protein binding: What do we know? ALTEX (2006) 23:229–233.[Medline]
Divkovic M, Pease CK, Gerberick GF, Basketter DA. Hapten-protein binding: From theory to practical application in the in vitro prediction of skin sensitization. Contact Derm. (2005) 53:189–200.[CrossRef][Medline]
Dupuis G, Benezra C. Allergic Contact Dermatitis to Simple Chemicals: A Molecular Approach (1982) New York: Marcel Dekker Inc.
Estrada E, Patlewicz G, Chamberlain M, Basketter DA, Larbey S. Computer-aided knowledge generation for understanding skin sensitization mechanisms: The TOPS-MODE Approach. Chem. Res. Toxicol. (2003) 10:1226–1235.
Farriss MW, Reed DJ. High-performance liquid chromatography of thiols and disulfides: Dinitrophenol derivatives. Meth. Enzymol. (1987) 143:101–109.[Web of Science][Medline]
Gerberick GF, Ryan CA, Kimber I, Dearman RJ, Lea LJ, Basketter DA. Local lymph node assay: Validation assessment for regulatory purposes. Am. J. Contact Dermat. (2000) 11:3–18.[CrossRef][Medline]
Gerberick GF, Ryan CA, Kern PS, Schlatter H, Dearman RJ, Kimber I, Patlewicz GY, Basketter DA. Compilation of historical lymph node data for evaluation of skin sensitization alternative methods. Dermatitis (2005) 16:157–202.[Web of Science][Medline]
Gerberick GF, Vassallo JD, Bailey RE, Chaney JG, Morrall SW, Lepoittevin J-P. Development of a peptide reactivity assay for screening contact allergens. Toxicol. Sci. (2004) 81:332–343.
Jowsey IR, Basketter DA, Westmorland C, Kimber I. A future approach to measuring relative skin sensitizing potency: A proposal. J. Appl. Toxicol. (2006) 26:341–350.[CrossRef][Web of Science][Medline]
Kato H, Okamoto M, Yamashita K, Nakamura Y, Fukumori Y, Nakai K, Kaneko H. Peptide-binding assessment using mass spectrometry as a new screening method for skin sensitization. J. Toxicol. Sci. (2003) 28:19–24.[CrossRef][Medline]
Kimber I, Basketter DA, Butler M, Gamer A, Garrigue J-L, Gerberick GF, Newsome C, Steiling W, Vohr H-W. Classification of contact allergens according to potency: Proposals. Food Chem. Toxicol. (2003) 41:1799–1809.[CrossRef][Web of Science][Medline]
Kimber I, Cumberbatch M, Dearman RJ, Bhushan M, Griffiths CE. Cytokines and chemokines in the initiation and regulation of epidermal Langerhans cell mobilization. Br. J. Dermatol. (2000) 142:410–441.
Kimber I, Dearman RJ, Basketter DA, Ryan CA, Gerberick GF. The local lymph node assay: Past, present and future. Contact Derm. (2002) 47:315–328.[CrossRef][Medline]
Kimber I, Hilton J, Dearman RJ, Gerberick GF, Ryan CA, Basketter DA, Lea L, House RV, Ladics GS, Loveless SE, et al. Assessment of the skin sensitization potential of topical medicaments using the local lymph node assay: An interlaboratory evaluation. J. Toxicol. Environ. Health (1998) 53:563–579.[CrossRef][Web of Science]
Landsteiner K, Jacobs J. Studies on the sensitization of animals with simple chemical compounds. J. Exp. Med. (1936) 64:625–639.[Abstract]
Lepoittevin J-P, Basketter DA, Goossens A, Karlberg A-T. Allergic Contact Dermatitis: The Molecular Basis (1998) Springer, Berlin.
Loveless SE, Ladics GS, Gerberick GF, Ryan CA, Basketter DA, Scholes EW, House RV, Hilton J, Dearman RJ, Kimber I. Further evaluation of the local lymph node assay in the final phase of an international collaborative trial. Toxicology (1996) 108:141–152.[CrossRef][Web of Science][Medline]
Meschkat E, Barratt MD, Lepoittevin J-P. 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. Chem. Res. Toxicol. (2001) 14:110–117.[CrossRef][Web of Science][Medline]
Natsch A, Gfeller H, Rothaupt M, Ellis G. Utility and limitations of a peptide reactivity assay to predict fragrance allergens in vitro. In: Toxicol. in Vitro. (in press).
Ryan CA, Chaney JG, Kern PS, Dearman RJ, Kimber I, Basketter DA, Gerberick GF. Extrapolating local lymph node assay EC3 values to estimate relative sensitizing potency. In: J. Toxicol. Cutaneous Ocul. Toxicol. (in press).
Ryan CA, Gerberick GF, Cruse LW, Basketter DA, Lea L, Blaikie L, Dearman RJ, Warbrick EV, Kimber I. Activity of human contact allergens in the murine local lymph node assay. Contact Derm. (2000) 43:95–102.[CrossRef][Medline]
Ryan CA, Gerberick GF, Gildea LA, Hulette BC, Betts CJ, Cumberbatch M, Dearman RJ. Interactions of chemicals with dendritic cells—a novel approach for identification of potential allergens. Toxicol. Sci. (2005) 88:4–11.
Ryan CA, Hulette BC, Gerberick FG. Review: Approaches for the development of cell based in vitro methods for contact sensitization. Toxicol. In Vitro (2001) 15:43–55.[CrossRef][Web of Science][Medline]
Smith CM, Hotchkiss SAM. Allergic Contact Dermatitis: Chemical and Metabolic Mechanisms (2001) London: Taylor and Francis.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  G. F. Gerberick, J. A. Troutman, L. M. Foertsch, J. D. Vassallo, M. Quijano, R. L. M. Dobson, C. Goebel, and J.-P. Lepoittevin Investigation of Peptide Reactivity of Pro-hapten Skin Sensitizers Using a Peroxidase-Peroxide Oxidation System Toxicol. Sci., November 1, 2009; 112(1): 164 - 174. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. Aleksic, E. Thain, D. Roger, O. Saib, M. Davies, J. Li, A. Aptula, and R. Zazzeroni Reactivity Profiling: Covalent Modification of Single Nucleophile Peptides for Skin Sensitization Risk Assessment Toxicol. Sci., April 1, 2009; 108(2): 401 - 411. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. A. Basketter Nonanimal Alternatives for Skin Sensitization: A Step Forward? Toxicol. Sci., March 1, 2008; 102(1): 1 - 2. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||