Chemistry-Based Risk Assessment for Skin Sensitization: Quantitative

ARTICLE pubs.acs.org/crt

Chemistry-Based Risk Assessment for Skin Sensitization: Quantitative Mechanistic Modeling for the SNAr Domain D. W. Roberts,*,† A. O. Aptula,‡ and G. Y. Patlewicz§ †

School of Pharmacy and Biomolecular Sciences, Liverpool John Moores University, Byrom Street, Liverpool L3 3AF, England SEAC, Unilever Colworth, Sharnbrook, Bedford MK44 1LQ, England § DuPont Haskell Global Centers for Health and Environmental Sciences, Newark, Delaware 19711, United States ‡

ABSTRACT: There is a strong impetus to develop nonanimal based methods to predict skin sensitization potency. An approach based on physical organic chemistry, whereby chemicals are classified into reaction mechanistic domains and quantitative models or read-across methods are derived for each domain, has been the basis of several recent publications. This article is concerned with the SNAr reaction mechanistic domain. Electrophiles able to react by the SNAr mechanism have long been recognized as skin sensitizers and have been used extensively in research studies on the biology of skin sensitization. Although qualitative discriminant analysis approaches have been developed for estimating the sensitization potential for SNAr electrophiles on a yes/no qualitative basis, no quantitative mechanistic model (QMM) has so far been developed for this domain. Here, we derive a QMM that correlates skin sensitization potency, quantified by murine local lymph node assay (LLNA) EC3 data on a range of SNAr electrophiles. It is based on the Hammett σ values for the activating groups and the Taft σ* value for the leaving group. The model takes the form pEC3 = 2.48 Σσ + 0.60 σ* 4.51. This QMM, generated from mouse LLNA data, provides a reactivity parameter 2.48 Σσ + 0.60 σ*, which was applied to a set of 20 compounds for which guinea pig test results were available in the literature and was found to successfully discriminate the sensitizers from the nonsensitizers. The reactivity parameter correctly predicted a known human sensitizer 2,4-dichloropyrimidine. New LLNA data on two further SNAr electrophiles are consistent with the QMM.

’ INTRODUCTION Skin sensitization is an important toxicological end point. The possibility that chemicals used in the workplace and in consumer products might cause skin sensitization is a major concern for individuals and for employers. The mouse local lymph node assay (LLNA) developed in the 1990s is an in vivo test used for skin sensitization hazard identification and characterization.13 The assay is usually carried out over a range of dosages of the test chemical and, from doseresponse analysis, it is usually possible to derive an EC3 value, this being the dose (expressed as % concentration by weight) giving a Stimulation Index, SI = 3. For modeling purposes, it is usual to express potency as pEC3, the negative logarithm of the EC3 in mol % units. The LLNA is now accepted as an approved test method for evaluating skin sensitization hazard for risk assessment including under various regulatory use programs, e.g., OECD Test Guidelines 429.3 It also provides a potential boost to skin sensitization model development since it gives a single-application-based quantitative end point. Recent changes in the European Union (EU) legislation will forbid the marketing of cosmetic product ingredients which have been tested on animals.4 Consequently, there is a strong impetus to develop nonanimal based methods to predict skin sensitization potency. The most successful so far have been based on relating sensitization potency to chemical properties. The underlying concept is that to induce sensitization a compound must be r 2011 American Chemical Society

able, either directly or after metabolic or abiotic activation, to react covalently with skin protein. Attempts to develop “global” quantitative structureactivity relationships (QSARs), covering a wide diversity of sensitizers, either by RAI-related (RAI = relative alkylation index) approaches5 or by statistical approaches have not been met with sufficient success to be predictively useful.6,7 A more promising approach is to develop separate quantitative mechanistic models (QMMs),8 each based on reactivity9 or a combination of reactivity and hydrophobicity parameters,10 for the major reaction mechanisms by which compounds can react with nucleophiles.11 One of the reasons why this separation into domains12 is necessary is that the proportionality between reactivity with the skin protein and reactivity with a model nucleophile (or reactivity represented by computational indices or substituent constants) differs according to the reaction mechanism. The ability to group electrophilic chemicals into mechanisticbased structural domains for protein binding13 has the potential to facilitate the filling of data gaps by read-across or quantitative mechanistic model predictions for several regulatory end points12,1416 without the need for additional in vivo testing. Simple structure-based classifications (e.g., aldehydes, phenols, and organic halides) do not map onto the mechanistic classification, Received: December 5, 2010 Published: June 14, 2011 1003

dx.doi.org/10.1021/tx100420w | Chem. Res. Toxicol. 2011, 24, 1003–1011

Chemical Research in Toxicology

ARTICLE

Table 1. Reactivity Parameters, LLNA Potency, and Hydrophobicity Parameters for SNAr Electrophilesa EC3, name

Figure 1. General reaction for the SNAr domain.

and therefore, molecular structure per se is not a very useful basis for domain definition. However, structural alerts in the form of coupled substituents (e.g., polarized R,β-unsaturates) have been shown to be useful in identifying electrophilic toxicity.17 Quantitative mechanistic modeling of skin sensitization potency based on reactivity parameters has already been demonstrated for several reaction domains with LLNA data: SN2 domain,10 Schiff base,9 Michael acceptor,8 as well as for guinea pig data (summarized in ref 18). However, although qualitative discriminant analysis approaches have been developed for estimating sensitization potential for SNAr electrophiles on a yes/no qualitative basis,19,20 no QMM or QSAR has so far been developed for this domain. To induce sensitization, a compound must be able, either directly or after metabolic or abiotic activation, to react covalently with skin protein. Early evidence for this fundamental chemistry activity relationship was reported in 1936 by Landsteiner and Jacobs.21 They investigated 20 benzene derivatives, variously substituted with halogeno and/or nitro groups. Ten of these compounds were found to be reactive toward methanolic aniline, used as a simple model for nucleophilic groups on proteins, and ten were found to be unreactive. In their guinea pig tests, all ten aniline-reactive compounds (now recognized as SNAr electrophiles, a term which was not current at that time) were found to be sensitizers, and all ten nonreactive compounds failed to sensitize. Our aim in this study is to gain further evidence regarding the role of reaction chemistry in determining skin sensitization potency, by developing a quantitative mechanistic model for LLNA data on compounds in the SNAr domain. Most biological mechanistic work on sensitization has been and continues to be done with compounds in this domain. 2,4Dinitrochlorobenzene (DNCB) has been familiar for very many years as a standard contact allergen.2225 Several other SNAr compounds are used commercially (e.g., tetrachloroisophthalonitrile (TCPN)).26 In this article, we investigate the applicability of reactivity parameters to model quantitative skin sensitization potency data from the LLNA for the SNAr domain. The general features of SNAr domain reaction chemistry are presented in Figure 1. Electrophilic reactivity is determined by a combination of the effects of the leaving group X and the activating groups Y. To develop a predictive nonanimal model for SNAr sensitization, ideally a QSAR-type approach would be adopted, in which a set of compounds with a range of X’s and a range of Y’s, selected to give even coverage of the domain, would be studied, and multilinear regression would be applied to quantify the dependence of LLNA sensitization potency on physicochemical parameters relating to the X and Y groups. In practice, we aim to avoid further animal tests for the purpose of producing data to develop further risk assessment capability, and therefore, we choose to work only with the LLNA data that are already available. The X and Y groups in the available data set are far from ideally distributed for application of the QSAR-type multiple regression approach, and instead,

CAS

σ*

Σσ (% weight) log(1/EC3b) f (X)c

1 2

DNFB 70-34-8 DNCB 97-00-7

3.55 2.48 2.96 2.48

0.032 0.0765

3.76 3.42

0.37 0.94

3

DNBB 584-48-5

2.84 2.48

0.085

3.46

1.09

4

DNIB

2.46 2.48

0.17

3.24

1.35

5

DNTB 1594-56-5 3.43 2.48

0.047

3.68

0.64

6

DNBS

1.9

2.12

4.53

7

TCPN 1897-45-6 2.96 3.18

0.0035

4.88

8

DCNB 611-06-3

20

0.98

0.05 0.3

3.69 2.99

709-49-9 89-02-1

0.81 2.48 2.96 1.6

9 TNCB 88-88-0 2.96 3.72 10 TNBS 2508-19-2 0.81 3.72 a

Structures shown in Figure 2. b EC3 in mol/100 g. c The log P fragment value for the aromatic X group in DNXB.

Table 2. Single Substituent Constants, σ, Used to Calculate Σσ Values substituent

σ

reference

p-NO2

1.24

o-NO2

1.24

see text

p-CN o-CN

0.88 0.88

28 see text

p-Cl

0.24

28

m-Cl

0.37

28

o-Cl

0.68

28

m-OH

0.13

28

28

m-O

0.47

28

p-N in ring

1.351.45

19

o-N in ring

0.711.19

19

we apply a QMM approach8 in which we separately determine the effects of the X and the Y groups and apply a simple mathematical approach to arrive at an integrated model for the domain.

’ MATERIALS AND METHODS LLNA data on a range of 10 SNAr compounds are taken from the updated compilation of Kern et al.,27 except where otherwise referenced in the text. To the best of our knowledge, these are the only SNAr electrophiles for which LLNA data have been published. Many other SNAr electrophiles are known but have not been tested for skin sensitization potency. Kinetic data were taken from publications referenced in the text, and substituent constants were taken from Perrin et al.28 These are summarized in Tables 1 and 2, the LLNA data being shown in Table 1. The structures are given in Figure 2. The reaction site is indicated by an arrow. The reproducibility of LLNA data has been investigated by Basketter et al.29 For DNCB, representing the SNAr domain, the standard deviation (SD) on log EC3 is 0.17 (13 tests). Other sensitizers showed similar SD values, e.g., isoeugenol (31 tests, SD 0.15).

’ QUANTITATIVE MECHANISTIC MODELING OF SKIN SENSITIZATION POTENCY FOR THE SNAR DOMAIN Linear Free Energy Relationship (LFER) Modeling of Reactivity Using Substituent Constants. The physical organic reaction

chemistry of DNCB and other SNAr electrophiles has been intensively 1004

dx.doi.org/10.1021/tx100420w |Chem. Res. Toxicol. 2011, 24, 1003–1011


ARTICLE

Table 3. Adjusted σ* for 2,4-DNFB σ* with F adjustment corresponding to name 1 2,4-DNFB

no adjustment methoxide ion piperidine thiophenoxide ion 3.21

4.30

4.70

3.55

Figure 3. Transition state for the SNAr reaction.

Figure 2. Chemicals investigated in this work.

Figure 4. Plot of pEC3 vs σ* for compounds 16 using 4.3, derived from the methoxide kinetics data, as the σ* value for DNFB.

studied, particularly by Bunnett’s and Miller’s groups in the late 1950s and early 60s,30,31 and the salient points relevant to skin sensitization were discussed in some detail by Roberts in 1995.19 The SNAr reaction as shown in Figure 1 can occur when X is any halogen or pseudohalogen (e.g., SCN), and with a range of other X groups which are not usually considered as leaving groups, such as NO2, SO2Ph, SOPh, and SO3. The full scope of “X” and “Y” in the SNAr reaction, is quite difficult to define rigorously. The SNAr mechanism applies for a wide range of X groups that are capable of being displaced with the negative charge on X increased by 1. Even when all the Y groups are hydrogen (e.g., chlorobenzene), the SNAr mechanism probably applies, although forcing conditions much more severe than apply in vivo are required. Reactivity toward a given nucleophile depends on the ability of X to stabilize the negative charge in the ring in the transition state but in most cases not on the ability of the C-X bond to dissociate heterolytically. This is because the formation of the intermediate is usually the rate-determining step.32 The transition state is similar to the intermediate, but the bond between the ring and the attacking nucleophile is not fully formed, and the charge in the ring is not fully developed (see Figure 1). The ability of X to stabilize the negative charge depends on its inductive effect, which is modeled by the Taft σ* constant.19 If this were the only effect to be considered, we might expect a reasonable straight line when the reactivity (log of the rate constant) is plotted against the σ* value. Plots of this type have been presented previously19 for reactions of a variety of 2,4-DNXB compounds with three nucleophiles:

methoxide ion, piperidine, and thiophenoxide ion. Most of the DNXB compounds fit logk vs σ* lines quite well, with the exception of X = F (dinitrofluorobenzene, DNFB). In this case, the compound is more reactive than predicted by the line based on the other DNXB compounds: this is strikingly so for the methoxide and piperidine nucleophiles but only marginally so for the thiophenoxide nucleophile. The reason why DNFB is anomalously more reactive is that F has a steric advantage over the other X groups, being very small and presenting less of an obstruction to the approach of the nucleophile. To allow for this steric advantage, we can modify the σ* value for F, increasing it to what it would need to be to fit the line. The modification is done by calculating (modified σ*(F)) from the rate constant for DNFB and the slope of the line of log k vs σ* for other DNXB compounds.19 Table 3 shows the magnitude of the steric advantage effect for each nucleophile and how the σ* value for the F substituent would need to be modified to allow for this effect. The reason why the steric advantage effect is so much smaller for the thiophenoxide nucleophile is thought to be that the soft sulfur nucleophile, being more polarizable, produces a transition state with a greater distance between the nucleophile and the carbon atom of DNFB,33 as shown in Figure 3. Consequently, there is less steric interaction between the nucleophile and the leaving group in the transition state. Correlation of Sensitization Potency with σ* Values for DNXB Compounds. To develop a quantitative model for the sensitization of DNXBs (16, Figure 2), it is necessary to consider which σ* value should be used for DNFB. If the in 1005



ARTICLE

Figure 5. Plot of pEC3 vs σ* for compounds 16 with modified σ* (for DNFB) in line with the thiophenoxide reaction data (modified σ* = 3.55).

cutaneo protein binding reaction involves attack of a relatively hard oxygen or nitrogen nucelophile, then a σ* value in the range 4.34.7, correcting for the steric advantage effect with hard nucleophiles, would be appropriate for DNFB (see Table 3). If the protein binding reaction involves a soft sulfur nucleophile (for example, a cysteine unit) then a value of 3.55 would be more appropriate. It is not obvious a priori which is the most appropriate model nucleophile: DNCB has been found to react readily with both cysteine-based and lysine-based peptides 34,35 in in chemico reactivity measurements, and it had earlier been shown that DNCB can react in cutaneo with lysine units of proteins22 (we note, however, that there is no reason to assume that these proteins are the ones involved in the skin sensitization process). Figure 4 shows a plot of pEC3 vs σ*, using 4.3, derived from the methoxide kinetics data, as the σ* value for DNFB. It can be seen that with this σ* value, the potency of DNFB is highly overpredicted. An analogous plot (not shown) with the σ* value for DNFB based on the piperidine kinetics similarly shows DNFB to be highly overpredicted. When we modify σ* (for the F substituent) in line with the thiophenoxide reaction data (modified σ* = 3.55), DNFB is on line, and the overall fit is very good (see Figure 5). This implies that the step which determines the level of sensitization produced by DNXBs involves attack of a soft (thiophenoxide-like) nucleophile, such as the thiolate group of an ionized cysteine unit, rather than attack of a hard nitrogen- or oxygen-centered nucleophile. Regression analysis for the plot in Figure 5 (using the modified σ* value for F based on thiophenoxide kinetics) gives the following equation: pEC3 ¼ 0:60ð(0:03Þσ + 1:68ð(0:07Þ

Figure 6. Plot of pEC3 vs Σσ for compounds 2, 7, and 8.

Figure 7. pEC3 observed vs pEC3 calculated from eq 3.

earlier paper,19 it was assumed that the reaction takes place in a hydrophobic environment, for which it was appropriate to use a special σ value for the ortho-NO2 substituent. We now consider, in view of the lack of dependence of sensitization potency on hydrophobicity for the DNXB compounds, and the similar situation for Michael acceptors,9 that the in vivo reaction occurs in an aqueous medium, for which an ortho σ value for the nitro-substituent identical to the para σ value (1.24) is more appropriate. On the same basis, for the cyano groups in compound 8, we use the same σ value (0.88) for both the ortho and para positions. The Σσ values calculated on this basis are shown in Table 1. A plot of pEC3 against Σσ for compounds 2, 7, and 8 is shown in Figure 6. The equation of the line is as follows:

∑

pEC3 ¼ 2:48ð(0:03Þ σ 2:91ð(0:49Þ

ð1Þ

n ¼ 6, R 2 ¼ 0:988, R 2 ðadjÞ ¼ 0:985, s ¼ 0:07, F ¼ 323 The very high value of R2 is due partly to the inclusion of the compound, DNBS (6 in Table 1), corresponding to X = SO3 (σ* = 0.81), which has also been tested in the LLNA.36 It is substantially less potent and has a substantially lower σ* value than the other five DNXB compounds, and consequently, it has a high statistical leverage. Extension to Other SNAr Electrophiles. Referring to Figure 2, all of the compounds considered so far have the activating groups Y1 and Y2 as ortho- and para-NO2. To extend the model to include other combinations of activating groups, we can use data on two further compounds, 7 and 8 (Figure 2), together with DNCB, 2. The leaving group for all these three compounds is Cl. As discussed earlier,19 the influence of the activating groups can be modeled by the sum of their σ values, Σσ. However, a slight modification needs to be made to the calculation method. In the

ð2Þ

n ¼ 3, R 2 ¼ 0:994, R 2 ðadjÞ ¼ 0:988, s ¼ 0:22, F ¼ 162 We are able to combine all 8 chemicals in one equation, using the fact that DNCB is described both by eqs 1 and by eq 2. The mathematical operation is as follows: from eq 1, the dependence on σ* is represented by the coefficient 0.60, and from eq 2, the dependence on Σσ is represented by the coefficient 2.48. We can therefore write the general equation as follows:

∑

pEC3 ¼ 2:48 σ + 0:60σ + C Substituting the pEC3 value (3.42), the Σσ value (2.48), and the σ* value (2.96) of DNCB and solving it gives C = 4.51. The combined equation, covering variation in both leaving groups and activating groups, is as follows:

∑

pEC3 ¼ 2:48 σ + 0:60σ 4:51

ð3Þ

Equation 3 is a QMM for LLNA skin sensitization potency of 1006



ARTICLE

SNAr electrophiles. It is not a regression equation and therefore does not have statistical indices directly associated with it. However, the overall fit to the data can be quantified by plotting the experimental pEC3 values against pEC3 values calculated from eq 3, as shown in Figure 7. The statistics for this plot are as follows:

Table 4. LLNA Potency, Predicted by eq 3 for Compounds Negative in the Landsteiner and Jacobs21 Test

pEC3ðobsÞ ¼ 1:00ð(0:15ÞpEC3ðcalcÞ 0:03ð(0:18Þ ð4Þ n ¼ 8, R 2 ¼ 0:984, R 2 ðadjÞ ¼ 0:981, s ¼ 0:16, F ¼ 365 Although, as stated above, the various values of Σσ and σ* are not ideally distributed among the data set for conventional QSAR analysis, it is instructive to compare eq 3 with the result of multiple linear regression analysis of pEC3 against Σσ and σ*: pEC3 ¼ 2:50ð(0:15Þ + 0:57ð(0:08Þσ 4:52ð(0:44Þ ð5Þ n ¼ 8, R 2 ¼ 0:984, R 2 ðadjÞ ¼ 0:978, s ¼ 0:17, F ¼ 157 Although eq 5 is far from an ideal QSAR, and we would not put great weight on its apparent good statistics, it is relevant to note that the coefficients and the constant of eq 5 are not significantly different from those of eq 3. Compounds with Three Strong Activating Groups. For present purposes, we define a strong activating group as an electronegative substituent, ortho or para to the leaving group, that is able to stabilize a negative charge both by resonance and inductive effects. An ortho or para nitro group is an example of a strong activating group. LLNA sensitization data have been published for two SNAr electrophiles which have three activating nitro groups in the ortho and para positions. These are 9 (X = Cl) and 10 (X = SO3) in Table 1 and Figure 2. They may be regarded as derivatives of 2 and 6, respectively, with an extra nitro group in the second ortho position. The additional nitro group should make the compounds more reactive (data confirming this are plentiful for 9, e.g., ref 37) and hence more potent than their 2,4dinitro- counterparts 2 and 6; therefore, if eq 3 is applicable to these compounds, then they should be more potent as sensitizers than any of the compounds discussed so far. Although they are strong sensitizers, and each is more potent than its 2,4-dinitrocounterpart, they are substantially less potent than eq 3 predicts: for 9, EC3 observed,38 0.05%; EC3 calc (from eq 3), 0.00008%; for 10, EC3 observed,39 0.30%; EC3 calc (from eq 3), 0.002%. Although the formation of the intermediate (Figure 1) is usually the rate determining step in SNAr reactions, this is not always the case.32 The simplest explanation for the overprediction of the potency of 9 and 10 is that for these compounds with three strong activating groups, the intermediate is stabilized to such an extent that the departure of the leaving group from the intermediate becomes the rate determining step, i.e., this step occurs more slowly than the reversible formation of the intermediate. Since the reactivity parameter 2.48Σσ + 0.60σ* models the rate of formation of the intermediate, it overpredicts the overall rate if the departure of the leaving group is rate determining, and consequently, in this situation eq 3 overpredicts the potency in the LLNA.

On the basis of this reasoning, we would expect any SNAr electrophile with 3 or more strong activating groups to be a more potent sensitizer than its counterpart with only 2 such groups, but to be less potent than predicted by eq 3. More generally, any SNAr electrophile with 2.48Σσ + 0.60σ* values above about 9.7 (the value for TNBS, 10, is 9.71) could be overpredicted by eq 3 since a high 2.48Σσ + 0.60σ* value implies a high degree of stabilization of the intermediate by a combination of resonance and inductive effects and consequently the possibility that departure of the leaving group is rate determining. It is not possible to define a fixed value for 2.48Σσ + 0.60σ* above which eq 3 underpredicts since the value at which departure of the leaving group becomes rate determining will depend on the 1007



ARTICLE

Figure 8. Reaction possibilities for DNTB with cysteine-based proteins.

effectivenesss of the leaving group. For example, TNBS, 10, has a relatively poor leaving group, the sulphite dianion, and is overpredicted by eq 3, whereas TCPN, 7, has an almost identical 2.48Σσ + 0.60σ* value (9.66 as compared with 9.71 for 10) but a better leaving group, chloride ion, and is well predicted by eq 3. Test of Predictive Applicability of eq 3 for Historical Guinea Pig and Human Data. If eq 3 is mechanistically robust, then the weighted combination of Σσ and σ*, (2.48Σσ + 0.60σ*), should be able to serve as a discriminate reactivity parameter to rank the potency of SNAr electrophiles in any sensitization test. The guinea pig sensitization data set produced by Landsteiner and Jacobs in the 1930s21 can be used to assess the predictive capability of eq 3 in this way. Landsteiner and Jacobs tested 20 compounds with various combinations of activating groups and leaving groups (only 5 of these compounds having 2,4-dinitro as the activating combination), reporting the results simply as positive or negative. This data set has been the subject of previous discriminant analysis exercises,19,20 and the compounds are listed in those papers. Of these 20 compounds, 10 were found to be positive in the guinea pig test, and 10 were found to be negative. For the positive compounds, the lowest value of the discriminant parameter (2.48Σσ + 0.60σ*) in the data set is 7.62 (for 4, dinitroiodobenzene, DNIB). For the negative compounds, the highest value of the discriminant parameter (2.48Σσ + 0.60σ*) in the data set is 7.58 (for hexachlorobenzene). This could be interpreted as implying that there is a boundary value or range of (2.48Σσ + 0.60σ*), between 7.58 and 7.62, below which all compounds are negative and above which all compounds are positive, in the guinea pig method used by Landsteiner and Jacobs.21 However, this band of uncertainty seems unrealistically narrow, and we cannot exclude the possibility that hexachlorobenzene, 11, owes its lack of sensitization potency more to its hydrophobicity (log P = 5.5) and nonpolar nature than to low reactivity. For such a compound, mass transfer across lipid phases to an aqueous site of reaction could be the rate-limiting step. The next most reactive nonsensitizer in the Landsteiner and Jacobs data set is 3,4-dichloronitrobenzene, 12 (2.48Σσ + 0.60σ* = 6.53). Thus, of the 19 compounds after the

exclusion of hexachlorobenzene, all 10 sensitizers have 2.48Σσ + 0.60σ* g 7.62, and all nonsensitizers have 2.48Σσ + 0.60σ* e 6.53. We conclude that the reactivity boundary dividing sensitizers from nonsensitizers, in the guinea pig test used by Landsteiner and Jacobs, is between 2.48Σσ + 0.60σ* values of 6.53 and 7.62. We note that the Landsteiner and Jacobs guinea pig test appears to have been much less sensitive than the LLNA as an assay for detecting sensitization potential, at least for SNAr sensitizers. Seven out of the 10 compounds that were negative in the Landsteiner and Jacobs test are predicted by eq 3 to be positive in the LLNA, as listed in Table 4. As far as we know, the compounds listed in Table 4 have not been tested in the LLNA. As a further test of predictivity, we consider 2,4-dichloropyrimidine, 21, which is a known human sensitizer.20,40 For this compound, the activating groups are an ortho ring nitrogen (σ in the range 0.711.1919), a para ring nitrogen (σ in the range 1.351.4519), and a meta chloro substituent (σ = 0.37 28), and the leaving group is Cl (σ* = 2.96). Using the lower values of the σ ranges for the ring nitrogens, eq 3 gives a minimum estimated value for pEC3 of 3.29. This corresponds to an EC3 value of 0.08, similar to that of DNCB. Thus 2,4-dichloropyrimidine is predicted to have potency similar to that of DNCB. Since DNCB is a known potent human sensitizer, 2,4-dichloropyrimidine would be correctly predicted as a human sensitizer by eq 3. New LLNA Data. The compounds listed in Table 1 are the only SNAr electrophiles for which, as far as we are aware, LLNA data have been published. However, while this article was in preparation, data on two further compounds became available to us.41 These compounds (Figure 2) are 1,3,5-trichloro-2,4,6-triazine, 22 (TCT), and pentachlorophenol, 23 (PCP). TCT (22). This compound has three strong activating groups, the three ring nitrogens. Using the minimum values listed in Table 1 for the σ values of ortho and para ring nitrogens, the 2.48Σσ + 0.60σ* value is calculated to be 10.5. This is in the range where eq 3 is likely to overpredict (cf. TNCB, 9. and TNBS, 10). Consistent with this expectation, and similarly to TNCB, the potency of TCT is high but less than that predicted by eq 3: pEC3 (calc) 6.49; pEC3 (obs) 3.31 (EC3 (calc) 0.0002%; and EC3 (obs) 0.09%). 1008



ARTICLE

Thus, the data on TCT give no information on the general predictive performance of eq 3 but support the argument that for compounds with 3 strong activating groups eq 3 is overpredictive. PCP (23). The title compound is acidic, having a pKa value of 4.74,42 and at physiological pH, much of it will exist as the anion, pentachlorophenoxide. The neutral form of PCP is calculated by eq 3 to have an EC3 value of 0.89%, and the anion is calculated to have an EC3 value of 27.5%. The pH in skin is usually slightly acidic, but the cytosol within cells is usually neutral or even slightly alkaline.43 Using the HendersonHasselbach equation, we can calculate the fractions (f) of the neutral (n) and anionic (a) forms of PCP at various pH values, and from these, we can calculate the overall EC3 by means of the mixture toxicity equation: 1=EC3ðmixtureÞ ¼ f n =EC3n + f a =EC3a

ð6Þ

where EC3n and EC3a are the EC3 values calculated from eq 3 for the neutral and anionic forms, respectively. On this basis, the overall EC3 for PCP is calculated to be: 18% if the pH at the reaction site is 6.5; 24% if the pH at the reaction site is 7.0; 26% if the pH at the reaction site is 7.5. The observed EC3 is 20%, in good agreement with the above figures. Thus, eq 3 gives a good prediction for the LLNA potency of PCP. Conflicting Evidence for DNTB in Human Tests. 2,4-Dinitrothiocyanatobenzene, DNTB, is an extreme sensitizer (defined as EC3 < 0.1% w/w) in the LLNA, and its potency is well modeled by the combination of Σσ and σ* of eq 3 (#5, Table 1). However, human data conflicting with this finding have been reported by Pickard et al.44 These authors report that unlike DNCB, which is similarly highly potent in both humand and mice (LLNA), DNTB behaves as a weak sensitizer when applied topically to healthy human volunteers. They attribute their finding of weak human potency to DNTB being, to a large extent, intercepted by sulfur-rich proteins in the stratum corneum. This interpretation is supported by experimental evidence that levels of dinitrophenylated proteins observable in the epidermis are much lower in the case of DNTB than those produced by DNCB. They conclude that the sulfur-rich layer of the stratum corneum constitutes a previously undescribed outer epidermal biochemical barrier, a chemical component of the innate immune defense mechanisms that defend against sensitization by highly reactive environmental chemicals. It seems unlikely to us that this sulfur-rich layer is present as a result of natural selection in the evolution of the human immune system since its protective role apparently extends to few allergens other than DNTB. We consider it more likely that the sulfur-rich proteins in human stratum corneum originate as residues from dying skin cells and are of relatively low nucleophilic reactivity but are able to react with DNTB because of the special features of the chemistry of the carbon-bound SCN group. This group can react with nucleophiles in 3 ways:45 (i) attack at the carbon to which SCN is bound, with the thiocyanate ion acting as a leaving group; (ii) attack at the central carbon atom, with the thiolate anion acting as a leaving group; (iii) attack at the sulfur atom, with the cyanide ion acting as a leaving group. These reactions are shown in Figure 8. We therefore interpret the findings of Pickard et al. as indicating that the stratum corneum proteins are relatively rich

in thiol groups that react with DNTB preferentially by reactions (ii) and/or (iii), but lack sufficient SNAr reactivity to prevent DNCB, and related compounds lacking the SCN group, from reaching the epidermis. We note that the sulfur-rich stratum corneum protein layer must be absent, or of low effectiveness, in mice, such that DNTB is of similar potency to DNCB in the LLNA. The same appears to apply for guinea pigs since DNTB is reported to show potency similar to that of DNCB in guinea pig tests.46

’ DISCUSSION Insignificant Role of Hydrophobicity. Figure 5 and eq 1 indicate that even though with the inclusion of DNBS the log P range covered is over 6 log units, LLNA sensitization potency for these SNAr electrophiles is dependent on reactivity alone, and no hydrophobicity parameter is needed to model the data. It has previously been observed that the same applies for the Michael acceptor domain: LLNA sensitization potency is dependent on reactivity but not on hydrophobicity.9 The implication is that for both of these reaction mechanistic domains, the biological nucleophile involved is in an aqueous environment, for example, cytosol, rather than in a membrane.22 Comparison with the Michael Acceptor Domain. It has recently been reported9 that for skin sensitizers belonging to the Michael acceptor domain, pEC3 is correlated with experimental reactivity (expressed as log k) toward a model sulfur nucleophile, on the basis of a cysteine containing peptide, with the following equation:

pEC3 ¼ 0:24 log k + 2:11

ð7Þ

Equations 1 and 7 can be compared, using the log k vs σ* correlation for reactions of DNXBs with thiophenoxide.19 This correlation (excluding DNFB) has the following equation: log k ¼ 2:13σ 6:12

ð8Þ

Substituting for σ* using eq 8, eq 1 can be expressed in terms of the thiophenoxide rate constant: pEC3 ¼ 0:28 log k + 3:40

ð9Þ

Although the two nucleophiles represented in eqs 7 and 9 are not identical, they are both based on ionized SH as the nucleophilic center and should be similar in their sensitivity to changes in the electrophile, although not in their absolute reactivities. Therefore, if sensitization by Michael acceptors and SNAr electrophiles involves the same biological nucleophile, we would expect the log k coefficients in eqs 4 and 6 to be similar. This is what is found; therefore, the inference is that the same or very similar, biological nucleophile is involved in both cases. As has been pointed out previously,9 the low values of the log k coefficients suggest that the biological nucleophile(s) is/are about 6 times less selective than the model nucleophiles. (However this is not the only possible interpretation; see the following paragraph.) This does not necessarily mean that the in vivo nucleophile is more reactive than the model nucleophiles, but we consider it likely that it is. It is not immediately obvious why the in vivo nucleophile, if it is thiol-based, should be so much less selective than the model thiol nucleophiles. One possibility is that the relevant in vivo nucleophile is based on selenocysteine, which is known to be present in skin as a protein subunit.47 The ionized SeH group is an even softer nucleophile than ionized SH 1009


Chemical Research in Toxicology groups; therefore, its involvement would imply only a marginal steric advantage effect for DNFB, as discussed above (see Figures 4 and 5). The log k coefficient of eq 9 being only 0.28 tells us definitely that the sensitization mechanism is about 6 times less selective than the model nucleophile. There being no evidence that any other step in the sensitization process is dependent on electrophilic reactivity, the simplest interpretation is, as discussed above, that the biological nucleophiles are about 6 times less selective. However, we cannot exclude the possibility that the relatively low selectivity arises downstream of the protein covalent modification step. Thus, a cell process (CP) that is part of the cascade of events leading to sensitization might be dependent on the degree of protein modification (DPM) by a relationship such as CP = 0.3 log DPM + c. Reactivity Parameters for Skin Sensitization Modeling. The reactivity parameter used in this QMM is based on substituent constants, which are ultimately derived from rate constants for the hydrolysis of X-substituted aliphatic esters and pKa values of Y-substituted phenols.48 Bearing this in mind, the overall good fit of the QMM to the in vivo data demonstrates that when generating experimental reactivity data for nonanimal potency prediction, it is not necessary to use model nucleophiles designed to closely resemble the in vivo nucleophiles. A similar conclusion can be drawn from the findings of Chipinda et al. on the applicability of p-nitrothiophenoxide reaction kinetics to rationalizing skin sensitization potency.49 Although large compilations of substituent constants are available, not all conceivable SNAr electrophiles are covered. Further development of quantum-chemical (QC) parameters enabling reactivity to be expressed in QC terms is desirable to broaden the general scope of models relating toxicity to reactivity.

’ SUMMARY AND CONCLUSIONS 1. A QMM has been developed that correlates LLNA EC3 data on a range of SNAr electrophiles. It has the equation pEC3 = 2.48 Σσ + 0.60 σ* 4.51. It is applicable up to (2.48 Σσ + 0.60 σ*) values of about 9.7, above which it overpredicts the potency. 2. The QMM, generated from mouse LLNA data, provides a reactivity parameter that successfully discriminates sensitizing from nonsensitizing compounds in a set of 20 compounds for which guinea pig test results are available in the literature. New LLNA data on two further compounds are consistent with the QMM. 3. The QMM is based on a reactivity parameter only, this being a composite function of the electronic effects of the leaving group and the activating groups. 4. DNBS fits well to the QMM, although it is over 5 orders of magnitude less hydrophobic than the other compounds. It is inconceivable that such a large hydrophobicity difference would not lead to a large difference in ability to penetrate the stratum corneum. It follows that the good fit of DNBS to the QMM is further confirmation of our earlier conclusion18,50 that bioavailability at the site of biological action is a constant proportion of the applied dose and is not a function of structure-dependent penetration ability. 5. The reactivity parameter used in this QMM is based on substituent constants, which are ultimately derived from rate constants for the hydrolysis of X-substituted aliphatic esters and pKa values of Y-substituted phenols.48 Bearing

ARTICLE

this in mind, the overall good fit of the QMM, based on reactivity parameters ultimately derived from rate constants for the hydrolysis of X-substituted aliphatic esters and pKa values of Y-substituted phenols,48 demonstrates that when generating experimental reactivity data for nonanimal potency prediction, it is not necessary to use model nucleophiles designed to closely resemble the in vivo nucleophiles. A similar conclusion can be drawn from the findings of Chipinda et al. on the applicability of p-nitrothiophenoxide reaction kinetics to rationalizing skin sensitization potency.49 However, this does not mean that information or insight as to the nature of biological nucleophiles should be ignored when modeling toxicological end points. As discussed in detail by Schwoebel et al.,51 a model nucleophile (whether this be real or computational) that matches the in vivo nucleophile in terms of hardness/softness should give more robust QMMs and QSARs.

’ AUTHOR INFORMATION Corresponding Author

*Phone: + 44 151 231 2422. Fax: + 44 151 231 2170. E-mail: [email protected]. Funding Sources

This study was supported in part by the United Kingdom Department of Environment, Food and Rural Affairs Grant LK0984.

’ REFERENCES (1) Dearman, R. J., Basketter, D. A., and Kimber, I. (1999) Local lymph node assay: use in hazard and risk assessment. J. Appl. Toxicol. 19, 299–306. (2) Basketter, D. A., Gerberick, G. F., and Kimber, I. (2007) The local lymph node assay: current position in the regulatory classification of skin sensitizing chemicals. J. Cutaneous Ocul. Toxicol. 26, 293–301. (3) OECD (2002) Guidelines for Testing of Chemicals, Guideline No. 429, Skin Sensitization: The Local Lymph Node Assay, Organisation for Economic Cooperation and Development, Paris, France. (4) EU (2003) Directive 2003/15/EC of the European Parliament and of the Council of 27 February 2003 amending Council Directive 76/768/EEC on the approximation of the laws of the Member States relating to cosmetic products. Off. J. Eur. Union L66, 2635. (5) Roberts, D. W., and Williams, D. L. (1982) The derivation of quantitative correlations between skin sensitization and physico-chemical parameters for alkylating agents and their application to experimental data for sultones. J. Theor. Biol. 99, 807–825. (6) Patlewicz, G., Aptula, A. O., Uriarte, E., Roberts, D. W., Kern, P. S., Gerberick, G. F., Kimber, I., Dearman, R. J., Ryan, C. A., and Basketter, D. A. (2007) An evaluation of selected global (Q)SARs/ expert systems for the prediction of skin sensitization potential. SAR QSAR Environ. Res. 18, 515–541. (7) Roberts, D. W., Aptula, A. O., Cronin, M. T. D., Hulzebos, E., and Patlewicz, G. (2007) Global (Q)SARs for skin sensitization assessment against OECD principles. SAR QSAR Environ. Res. 18, 343–365. (8) Roberts, D. W., Patlewicz, G., Kern, P., Gerberick, F., Kimber, I., Dearman, R., Ryan, C., Basketter, D., and Aptula, A. O. (2007) Mechanistic applicability domain classification of a Local Lymph Node Assay dataset for skin sensitization. Chem. Res. Toxicol. 20, 1019–1030. (9) Roberts, D. W., and Natsch, A. (2009) High throughput kinetic profiling approach for covalent binding to peptides: Application to skin sensitization potency of Michael acceptor electrophiles. Chem. Res. Toxicol. 22, 592–603. (10) Aptula, A. O., Roberts., D. W., and Patlewicz., G. (2006) Mechanistic applicability domains for non-animal based toxicological 1010


Chemical Research in Toxicology endpoints. QSAR analysis of the Schiff Base applicability domain for skin sensitization. Chem. Res. Toxicol. 19, 1228–1233. (11) Roberts, D. W., Aptula, A. O., and Patlewicz, G. (2007) Electrophilic chemistry related to skin sensitization. Reaction mechanistic applicability domain classification for a published dataset of 106 chemicals tested in the mouse local lymph node assay. Chem. Res. Toxicol. 20, 44–60. (12) Aptula, A. O., and Roberts, D. W. (2006) Mechanistic applicability domains for non-animal based toxicological endpoints. General principles, and application to reactive toxicity. Chem. Res. Toxicol. 19, 1097–1105. (13) Organization for Economic Co-operation and Development (OECD) (2007) OECD Environment Health and Safety Publications Series on Testing and Assessment No. 80: Guidance Document on Grouping of Chemicals ENV/JM/MONO(2007) 28. (14) Enoch, S. J., Cronin, M. T. D., Schultz, T. W., and Madden, J. C. (2008) Quantitative and mechanistic read across for predicting the skin sensitization potential of alkenes acting via Michael addition. Chem. Res. Toxicol. 21, 513–520. (15) Roberts, D. W., Aptula, A. O., Patlewicz, G., and Pease, C. (2008) Chemical reactivity indices and mechanism-based read-across for non-animal based assessment of skin sensitization potential. J. Appl. Toxicol. 28, 443–454. (16) Schultz, T. W., Rogers., K., and Aptula, A. O. (2009) Readacross to rank skin sensitization potential: Subcategories for the Michael acceptor domain. Contact Dermatitis 60, 21–31. (17) Schultz, T. W., Yarbrough, J. W., Hunter, R. S., and Aptula, A. O. (2007) Verification of the structural alerts for Michael acceptors. Chem. Res. Toxicol. 20, 1359–1363. (18) Roberts, D. W., and Aptula, A. O. (2008) Determinants of skin sensitisation potential. J. Appl. Toxicol. 28, 377–387. (19) Roberts, D. W. (1995) Linear free energy relationships for reactions of electrophilic halo- and pseudohalobenzenes, and their application in prediction of skin sensitization potential for SNAr electrophiles. Chem. Res. Toxicol. 8, 545–551. (20) Mekenyan, O., Roberts, D. W., and Karcher, W. (1997) Molecular orbital parameters as predictors of skin sensitization potential of halo- and pseudohalobenzenes acting as SNAr alectrophiles. Chem. Res. Toxicol. 10, 994–1000. (21) Landsteiner, K., and Jacobs, J. (1936) Studies on the sensitization of animals with simple chemical compounds. J. Exp. Med. 64, 643–655. (22) Eisen, H., and Tabachnick, M. (1958) Elicitation of allergic contact dermatitis in the Guinea pig. The distribution of bound dinitrobenzene groups within the skin and quantitative determination of the extent of combination of 2,4-dinitrochlorobenzene with epidermal protein in vivo. J. Exp. Med. 108, 773–796. (23) Salvin, S. B., and Smith, R. F. (1961) The specificity of allergic reactions. III. Contact hypersensitivity. J. Exp. Med 114 (2), 185–194. (24) Heylings, J. R., Clowes, H. M., Cumberbatch, M., Dearman, R. J., Fielding, I., Hilton, J., and Kimber, I. (1996) Sensitization to 2,4dinitrochlorobenzene: influence of vehicle on absorption and lymph node activation. Toxicology 109 (1), 57–65. (25) Basketter, D., Dearman, R., Hilton, J., and Kimber, I. (1997) Dinitrohalobenzenes: evaluation of relative skin sensitization potential using the local lymph node assay. Contact Dermatitis 36 (2), 97–100. (26) Huang, J., Aoyama, K., Ueda, A., and Matsushita, T. (1995) Respiratory effects and skin allergy in workers exposed to Tetrachloroisophthalonitrile. Bull. Environ. Contam. Toxicol. 55, 320–324. (27) Kern, P. S., Gerberick, G. F., Ryan, C. A., Kimber, I., Aptula, A., and Basketter, D. A. (2010) Local lymph node data for the evaluation of skin sensitization alternatives: A second compilation. Dermatitis 21, 8–32. (28) Perrin, D. D., Dempsey, B., and Serjeant, E. P. (1981) pKa Prediction for Organic Acids and Bases, Chapman and Hall, London. (29) Basketter, D. A., Gerberick, G. F., and Kimber, I. (2007) The local lymph node assay and the assessment of relative potency: status of validation. Contact Dermatitis 57, 70–75. (30) Bunnett, J. F. (1958) Mechanism and reactivity in aromatic nucleophilic substitution reactions. Quart. Rev. 12, 1–16. (31) Miller, J. (1956) Electrophilic and nucleophilic substitution in the benzene ring and the Hammett equation. Aust. J . Chem. 9, 61–73.

ARTICLE

(32) Smith, M. B., and March, J. (2001) March’s Advanced Organic Chemistry, 5th ed., p 850 Wiley, New York. (33) Bartoli, G., and Todesco, P. E. (1977) Nucleophilic substitution. Linear free energy relationships between reactivity and physical properties of leaving groups and substrates. Acc. Chem. Res. 10, 125–132. (34) Gerberick, G. F., Vassallo, J. D., Foertsch, L. M., Price, B. B., Chaney, J. G., and Lepoittevin, J. P. (2007) Quantification of chemical peptide reactivity for screening contact allergens: a classification tree model approach. Toxicol. Sci. 97, 417–427. (35) Aleksic, M., Thain, E., Roger, D., Saib, O., Davies, M., Li, J., Aptula, A., and Zazzeroni, R. (2009) Reactivity profiling: Covalent modification of single nucleophile peptides for skin sensitization risk assessment. Toxicol. Sci. 108 (2), 401–411. (36) Ryan, C. A., Cruse, L. W., Skinner, R. A., Dearman, R. J., Kimber, I., and Gerberick, G. F. (2002) Examination of a vehicle for use with water soluble materials in the murine local lymph node assay. Food Chem. Toxicol. 40 (11), 1719–25. (37) Bacaloglu, R., Blasko, A., Bunton, C., Dorwin, E., Ortega, F., and Zucco, C. (1991) Mechanism of reaction of hydroxide ion with dinitrochlorobenzenes. J. Am. Chem. Soc. 113 (1), 238–246. (38) Smith, C. K., and Hotchkiss, S. A. M. (2001) Allergic Contact Dermatitis: Chemical and Metabolic Mechanisms, Taylor & Francis, U.K. (39) Peiser, M. (2010) Background on skin sensitization hazard identification: Aspects in the development of in-vitro based alternatives for sensitization testing LRI workshop, Brussels, 02.02.2010, www.cefic-lri. org/uploads/.../LRI_LLNA-Workshop_100202-Peiser.pdf (40) Graul, E. H., and Patzschke, K. (1952) Klinischer und experimenteller Beitrag zur Frage der gewerblichen Hautsensibilisierung durch Synthese-Zwischen-und-Nebenproduckte. (Clinical and experimental contribution to the question of industrial skin sensitization by synthesis intermediates and byproducts.). Arch. Dermatol. Syphilis 193, 582–591. (41) Safford, R. J., Aptula, A. O., and Gilmour, N. (2011) Refinement of the Dermal Sensitisation Threshold (DST) approach using a larger dataset and incorporating mechanistic chemistry domains. Regul. Toxicol. Pharmacol. 60, 218–224. (42) EPA technical factsheet on: pentachlorophenol, http://www. epa.gov/ogwdw000/pdfs/factsheets/soc/tech/pentachl.pdf, part of EPA 816-F-09-004, National Drinking Water Regulations, May 2009. (43) Roos, A, and Boron, W. F. (1981) Intracellular pH. Physiol. Rev. 61 (2), 296–434. (44) Pickard, C., Louafi, F., McGuire, C., Lowings, K., Kumar, P., Cooper, H., Dearman, R. J., Cumberbatch, M., Kimber, I., Healy, E., and Friedmann, P. S. (2009) The cutaneous biochemical redox barrier: a component of the innate immune defenses against sensitization by highly reactive environmental xenobiotics. J. Immunol. 183, 7576–84. (45) Oae, S., Yamada, N., Fujimori, K., and Kikuchi, O. (1983) Nucleophilic substitution on 4-Methylbenzyl Thiocyanate with nucleophiles. Bull. Chem. Soc. Jpn. 56, 248–256. (46) Godfrey, H. P., and Baer, H. (1971) The effect of physical and chemical properties of the sensitizing substance on the induction and elicitation of delayed contact sensitivity. J. Immunol. 106, 431. (47) Hatfield, D. L. (2001) Selenium: Its Molecular Biology and Role in Human Health, Kluwer Academic Publishers, Norwell, MA. (48) Williams, A. (2003) Free Energy Relationships in Organic and BioOrganic Chemistry, pp 2083, Royal Society of Chemistry, Cambridge, U.K. (49) Chipinda, I., Ajibola, R., Morakinyo, M., Ruwona, T., Simoyi, R., and Siegel., P. (2010) Rapid and simple kinetics screening assay for electrophilic dermal sensitizers using nitrobenzenethiol. Chem. Res. Toxicol. 23, 918–925. (50) Roberts, D., and Patlewicz, G. (2010) Updating the skin sensitization in vitro data assessment paradigm in 2009: a chemistry and QSAR perspective. J. Appl. Toxicol. 30, 286–288. (51) Schwoebel, J., Koleva, Y., Enoch, S., Bajot, F., Hewitt, M., Madden, J., Roberts, D., Schultz, T., and Cronin, M. (2011) Measurement and estimation of electrophilic reactivity for predictive toxicology. Chem. Rev. 111, 2562–2596.

1011


Chemistry-Based Risk Assessment for Skin Sensitization: Quantitative

Recommend Documents