Relationship between allergic contact dermatitis and electrophilicity.

To evaluate the role of electrophilicity in the induction of allergic contact dermatitis (ACD) in humans, we compared the structure-activity relationship (SAR) model of ACD with those of electrophilic and nonelectrophilic subsets of chemicals in the ACD database. For these analyses, electrophilicity was defined as the potential of a chemical to induce mutations in Salmonella. It was found that electrophilicity accounted for approximately 30-40% of ACD-inducing ability, and the remainder was associated with nonelectrophilic structures. The identification of these moieties opens the possibility for studying their role in ACD.

To evaluate the role of electrophilicity in the induction of allergic contact dermatitis (ACD) in humans, we compared the strcture-activity relationship (SAR) model of ACD with those of electrophilic and noneectrophilic subsets of chemicals in the ACD database. For these analses, electrophilicity was defined as the potential of a chemical to induce mutations in Salmonellz It was found that eleophilicity an d for appromately 30-40% of ACD-inducing ability, and the remainder was associated with nonectrophiic structures. The identification of these moieties opens the possibility for studying their role in ACD. Key work allergic contact dermatitis, electrophilicity, mutagenicity, SAR, structure-activity relationship. Environ Health Perpect 107: 129-132 (1999). [ Online 13 January 1999] bttp://ehpnetl.nies.nibhgov/do /1999107p129-132rosenranlabsract.btmI One of the mechanistic paradigms associated with the disorder allergic cutaneous dermatitis (ACD) is that the agent (i.e., electrophile or proelectrophile) responsible for eliciting this response attacks a nucleophilic cellular target (1). Others suggest that it is the overall reactivity that determines the response (2). Electrophilicity is also associated with mutagenicity and carcinogenicity (3,4). Mutagenic (genotoxic) carcinogens are considered to pose a greater risk to humans than nongenotoxic ones (5). The relationship between ACD and other health risks has recently been addressed by Albert (6), who suggested a relationship between ACD and nongenotoxic carcinogens. The present study was undertaken to evaluate this hypothesis, to determine the extent to which electrophilicity contributes to ACD, and possibly to identify other, nonelectrophilic, mechanisms that also contribute to ACD. Our approach was based on our earlier observation that the structural basis of ACD in humans could be modeled successfully by an expert structure-activity relationship (SAR) system (7-9) yielding a predictivity of 84%. The approach we have taken here was to derive SAR submodels for electrophilic and nonelectrophilic chemicals in the ACD database and to compare features of the two submodels.
Methods CASE/MULTICASE. The algorithms employed within CASE and MULTICASE (Multicase, Inc., Beachwood, OH) have been described previously (10)(11)(12). Briefly, a set of chemicals and their respective experimentally determined biological activities (quantitative or qualitative) were input into a database. Both programs provide a means to identify descriptors consisting of molecular fragments, ranging from 2 to 10 heavy atoms along with their associated hydrogens, which account for the biological activity of the compounds under study. The molecular fragments are generated as a result of breaking down each individual chemical structure within the database into its constituent parts. Each fragment is "labeled" with respect to its origin within active or inactive structures. Fragments of relevance are those that exhibit a statistically significant nonrandom distribution among the active and inactive classes of compounds. In addition to using molecu-. lar fragments, MULTICASE identifies relevant two-dimensional distances between atoms within a chemical structure.
At this point within the analysis, CASE uses the statistically significant fragments in order to classify compounds as active or inactive. Predictions for submitted compounds are expressed as percent probabilities (0-100%) of being active or inactive. In addition, CASE utilizes the molecular fragments of the total database to derive a "global" quantitative structure-activity relationship (QSAR), which is used to predict the potency or extent of activity. These two predictions are derived independently of one another.
MULTICASE, on the other hand, uses the set of statistically significant descriptors (fragment and/or distance) to find a descriptor (biophore) that has the highest probability of being responsible for the observed biological activity. Compounds within the database that contain the primary biophore are removed from the analysis, and subsequent biophores that explain the activity of the remaining compounds are selected. This iterative process of selection is continued until either all of the active compounds are accounted for or no statistically significant descriptors remain. The presence of biophores determines the likelihood of a compound to exhibit activity. Predictions made on compounds submitted for SAR analysis consist of the identification of the biophore(s) responsible for activity and the percent probability of the compound being biologically active due to such occurrences. A compound is presumed to be inactive if it contains no biophores.
In MULTICASE, attempts are also made to derive a "local" QSAR within each group of compounds containing a particular biophore in order to identify molecular features that control the degree of activity. These features, termed modulators, are selected from the pool of molecular fragments, distance descriptors, calculated electronic indices (molecular orbital energies, charge densities), and calculated transport parameters (octanol/water partition coefficient, water solubility). These local QSARs are used to predict the potency of chemicals containing the specific biophore.
Database. The database used for SAR modeling was described previously (7,8). With respect to the present analyses in the model of ACD in humans, an "active" chemical elicits a response in human patch testing or has a sensitization rate of at least 16% in human maximization tests, when given at a dose greater than 10%. A chemical with "marginal" activity has a sensitization rate of 4-15% under the same conditions. Results and Discussion An "inactive" chemical is one that has a sen-To classify the chemicals in the ACD datasitization rate of <4% in human maximiza-base into electrophiles/proelectrophiles and tion testing.
nonelectrophiles, we chose their potential ACD+, ACDM, and ACDrefer to chemicals shown experimentally to induce, marginally induce, and not induce allergic contact dermatitis, respectively. Potential electrophiles and potential nonelectrophiles refer to chemicals predicted to be mutagens and nonmutagens, respectively.
to induce mutations in Salmonella as a surrogate for electrophilicity. This is based upon the recognition 1) that mutagenicity in Salmonella derives from an electrophilic attack and 2) that the Salmonella mutagenicity protocol includes testing in the presence of an exogenous metabolic activation mixture, thereby making it possible to identify proelectrophiles. Accordingly, we used an SAR model of the induction of mutations in Salmonella to classify chemicals with known ACD-inducing activity as potential electrophiles and nonelectrophiles. The predictivity of the model for Salmonella mutagenicity assay is 81%; however, because its sensitivity and specificity are approximately the same, the chance of false positive and false negative predictions is equal. bA nitro substituent is on the second carbon atom from the left. cThis carbon atom is common to two rings.

Articles * Allergic contact dermatitis and electrophilicity
Overall, it should be noted that while 28.0% of the 384 ACD-inducing agents were predicted to be mutagens, only 6% of the 383 noninducing chemicals were predicted to be mutagens. This suggests that electrophilicity is associated with ACD (p.O.OOOl; X2 test), but that it is only one of a number of factors because it cannot explain the basis of the allergenicity of the remaining 72% ofACD-inducing agents.
The ACD data available consisted of 917 chemicals tested in humans; however, only 767 of these were used to construct the SAR model. Recent studies have shown that a ratio of active to inactive molecules in the learning set of unity was optimal for developing SAR models (15). Because inactive chemicals exceeded active ones, we randomly deleted some of them (using a computerized random number generator). Our final model consisted of 355 chemicals that were active in the human patch test or in the human maximization test, 29 marginally active and 383 inactive chemicals (8). As a result, we had a surplus of ACD noninducing chemicals that were not used to construct the ACD SAR model. A total of 652 of the 767 chemicals in the ACD database (85.0%) were predicted to be Salmonella nonmutagens (nonelectrophiles). Of these, 266 (40.8%) were inducers of ACD, 26 were marginal inducers, and 366 (56.1%) were noninducers of ACD (database: ACD-NE) ( Table 1).
One hundred fourteen of the chemicals in the ACD database (14.9%) were predicted to be Salmonella mutagens (electrophiles). Of these, 88 (77.2%) were ACD inducers, 3 were marginal inducers, and 23 (20.2%) were noninducers. To increase the number of potential electrophilic ACD noninducers in the SAR model, we screened the "excess" ACD noninducers (n = 150) that had not been used to construct the original ACD SAR model. An additional 13 potential electrophilic non-ACD inducers were thus identified. To bring the ratio of ACD-active to ACD-inactive to Table 3. Major CASE biophores associated with the allergic contact dermatitis electrophilic (ACD-EL) structure-activity relationship submodel   unity, we included an additional 52 ACD noninducers randomly selected from the data set (Table 1). The inclusion of potentially nonelectrophilic non-ACD inducers should not affect the SAR analyses to be performed, as we were interested only in activating moieties (biophores). Thus, this database (ACD-EL) consisted of 88 ACD inducers, 3 marginal inducers, and 88 noninducers (Table 1). Using CASE/MULTI-CASE we developed SAR models for the ACD, ACD-EL, and ACD-NE databases.
The major biophores associated with the "parent" ACD SAR model (Table 2) are, as expected, also present in the putative electrophilic (ACD-EL, Table 3) and nonelectrophiic (ACD-NE, Table 4) SAR submodels. Indeed, this is further evidenced by the extent of the structural overlap (i.e., commonality of fagmnents) between ACD on one hand and ACD-EL (42%) and ACD-NE (73%) on the other (e.g., Fragment 5 in Table 3 is identical to Fragment 4 in Table 4, etc.) ( Table 5). This clearly suggests that the induction of ACD is not solely due to electrophiicity. However, as expected, the ACD-EL SAR model dearly contains typically electrophiic biophores (Table 3, biophores 7, 8, 18; e.g., arylamines, epoxides, and ortho-substituted nitroarenes) that are absent from the ACD-NE SAR model (Table 4). Further indication that ACD-EL and ACD-NE models reflect common, nonelectrophilic mechanisms of ACD induction is the substantial structural overlap between ACD-NE and ACD-EL, 45-63% (Table 5). Moreover, the ACD-NE model is highly predictive of the chemicals in ACD-EL and vice versa (i.e., 85 and 78% respectively, Table 6).
The findings described in this paper indicate that while electrophilicity is associated with the potential for inducing ACD, it is ,'The ACD-NE SAR model was used to predict the ACD-inducing ability of the chemicals used to derive the ACD-EL model and vice versa.
*Statistical significance of the concordance between experimental and predicted results.
probably only one of the mechanisms involved. Clearly, the nonelectrophilic biophores identified in the two SAR submodels provide dues of alternate mechanisms whereby ACD may be induced. They need to be investigated further with respect to their origin and possible mechanistic significance. Based on an examination of the activity of chemicals in the NTP rodent carcinogenicity bioassay and as ACD-inducing agents, Albert (6) suggested that there was a component among nongenotoxic carcinogens that was shared by human ACDinducing agents. He reasoned that the results of ACD testing and mutagenicity testing in Salmonella could be complementary and could be used to identify carcinogens. Our results indicate that ACD is driven, to a large part, by nonelectrophilic mechanisms. Our findings, however, do not address the question of whether activity as an ACD-inducing agent can complement mutagenicity in Salmonella in predicting potential carcinogens.