- Email: [email protected]
Carcinogenicity of the aromatic amines: from structure–activity relationships to mechanisms of action and risk assessment
Mutation Research 511 (2002) 191–206
Review
Carcinogenicity of the aromatic amines: from structure–activity relationships to mechanisms of action and risk assessment Romualdo Benigni∗ , Laura Passerini Laboratory of Comparative Toxicology and Ecotoxicology, Istituto Superiore di Sanita’, Viale Regina Elena 299, 00161 Rome, Italy Received 18 January 2002; received in revised form 12 March 2002; accepted 12 March 2002
Abstract Aromatic amines represent one of the most important classes of industrial and environmental chemicals: many of them have been reported to be powerful carcinogens and mutagens, and/or hemotoxicants. Their toxicity has been studied also with quantitative structure–activity relationship (QSAR) methods: these studies are potentially suitable for investigating mechanisms of action and for estimating the toxicity of compounds lacking experimental determinations. In this paper, we first summarized the QSAR models for the rodent carcinogenicity of the aromatic amines. The gradation of potency of the carcinogenic amines depended firstly on their hydrophobicity, and secondly on electronic (reactivity, propensity to be metabolically transformed) and steric properties. On the contrary, the difference between carcinogenic and non-carcinogenic aromatic amines depended mainly on electronic and steric properties. These QSARs can be used directly for estimating the carcinogenicity of aromatic amines. A two-step prediction is possible: (1) estimation of yes/no activity; (2) if the answer from step 1 is yes, then prediction of the degree of potency. The QSARs for rodent carcinogenicity were put in a wider context by comparing them with those for: (a) Salmonella mutagenicity; (b) general toxicity; (c) enzymatic reactions; (d) physical–chemical reactions. This comparative QSAR exercise generated a coherent global picture of the action mechanisms of the aromatic amines. The QSARs for carcinogenicity were similar to those for Salmonella mutagenicity, thus pointing to a similar mechanism of action. On the contrary, the general toxicity QSARs (both in vitro and in vivo systems) were mostly based on hydrophobicity, pointing to an aspecific mechanism of action much simpler than that for carcinogenicity and mutagenicity. The oxidation of the amines (first step in the main metabolic pathway leading to carcinogenic and mutagenic species) had identical QSARs in both enzymatic and physical–chemical systems, thus providing evidence for the link between simple chemical reactions and those in biological systems. The results show that it is possible to generate mechanistically and statistically sound QSAR models for rodent carcinogenicity, and indirectly that the rodent bioassay is a reliable source of good quality data. © 2002 Elsevier Science B.V. All rights reserved. Keywords: QSAR; Amine; Carcinogen; Structure–activity; Toxicity
1. Introduction
∗ Corresponding author. Tel.: +39-64990-2579; fax: +39-64938-7139. E-mail address: [email protected] (R. Benigni).
Aromatic amines represent one of the most important classes of industrial and environmental chemicals. Many aromatic amines have been reported to be powerful carcinogens and mutagens, and/or hemotoxicants. Exposure to aromatic amines occurs in
1383-5742/02/$ – see front matter © 2002 Elsevier Science B.V. All rights reserved. PII: S 1 3 8 3 - 5 7 4 2 ( 0 2 ) 0 0 0 0 8 - X
192
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
Nomenclature B1 -STM B5 -STM ES
F HOMO I IL KM log A log C log K log Kcat log Krel log P log TA log Vmax LUMO n omit pKa PI
sterimol width of the first atom of the substituent sterimol overall width of the substituent steric Taft parameter for substituents (for intramolecular steric effects) Swain–Lupton inductive/field effect for aromatic systems energy of the Highest Occupied Molecular Orbital substituent position indicator is 1, if three or more fused rings are present Michaelis–Menten constant A, activity C, molar concentration (mol/l or mol/kg) K, rate or equilibrium constant Kcat , catalytic rate constant Krel , relative to H P, octanol/water partition coefficient TA (revertants/mol) Vmax , maximum reaction rate energy of the Lowest Unoccupied Molecular Orbital number of datapoints omitted datapoints acidic dissociation constant hydrophobic parameter (π ) for substituents correlation coefficient standard deviation
r s suffixes 1, 3, 4, etc. ring substitution position S Hammet electronic (σ ) constant for substitutents
different industrial and agricultural activities as well as in tobacco smoking. Substantial worker exposure to aromatic amines with subsequent induction of bladder cancer occurred before preventive measures were instituted. Owing to their hazard potential, aromatic amines have been the subject of many in vivo and in vitro experimental studies, as well as biomonitoring
investigations. For an updated review on the toxicology of aromatic amines and their mechanisms of action, see [1]. Given their importance and the large amount of data available, the toxicity of the aromatic amines has been studied also with methods based on structure–activity relationship (SAR) and quantitative SAR (QSAR) concepts.1 The foundation of the modern QSAR science came about in the 1960s, after many attempts essentially qualitative. At present QSAR is one of the basic tools of modern drug and pesticide design [2,3], and has an increasing role in environmental sciences [4–8]. Its strength derives from the fact that it permits the identification of the molecular determinants of the biological action of the chemicals, and provides mathematical models to predict the activity of chemicals not tested experimentally provided that data on similar chemicals, acting with the same mechanism, are available. Several QSAR studies on the aromatic amines have been reported, mainly regarding their mutagenic properties. Surprisingly, very sporadic and limited QSAR studies of their carcinogenic properties existed until recently, in spite of the fact that several of them had been bioassayed thus providing the necessary database [9]. We have recently presented the first detailed QSAR analysis of the carcinogenicity of the aromatic amines, by investigating: (a) the molecular determinants that rule the gradation of the carcinogenic potency [9]; (b) the molecular determinants that make some of the aromatic amines non-carcinogenic [10]. The availability of a range of QSARs on different aspects of the toxicology of aromatic amines permits now to put the various pieces of evidence into perspective, and provides new insights into the mechanisms of action of the amines in different biological and chemical–physical systems, as well as on the use of QSAR for risk assessment. This paper presents: (a) a summary of the QSAR models for the carcinogenicity of the aromatic amines and implications for the mechanisms of action; (c) a comparative analysis of the QSARs of the aromatic amines for various toxicity end-points, as well as for enzymatic and chemical–physical reactions; (d) the implications of our results for the risk assessment of the aromatic amines. 1 QSAR models retrieved from the C-QSAR database. For more details on the chemical parameters (see [2,33]).
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
1.1. The QSAR models for the rodent carcinogenicity of the aromatic amines In our first QSAR analysis of the carcinogenicity of the aromatic amines, we considered only the carcinogenic aromatic amines, and we investigated the structural factors that influence the gradation of carcinogenic potency in rodents [9]. The study focused on the homogeneous class of non-heterocyclic amines. The following are the QSAR models emerging from the analysis of the bioassay data (BRM = carcinogenic potency in mice; BRR = carcinogenic potency in rats) BRM = 0.88(±0.27) log P × I (monoNH2 ) + 0.29(±0.20) log P × I (diNH2 ) + 1.38(±0.76)EHOMO − 1.28(±0.54)ELUMO − 1.06(±0.34) MR2,6 − 1.10(±0.80)MR3 − 0.20(±0.16)ES (R) + 0.75(±0.75)I (diNH2 ) + 11.16(±6.68) n = 37,
r = 0.907,
F = 16.3,
r 2 = 0.823,
s = 0.381,
P < 0.001
(1)
BRR = 0.35(±0.18) log P + 1.93(±0.48) I (Bi) + 1.15(±0.60)I (F) − 1.06(±0.53)I (BiBr) + 2.75(±0.64)I (RNNO) − 0.48(±0.30) n = 41, F = 47.4,
r = 0.933, P < 0.001
r 2 = 0.871,
s = 0.398, (2)
where BRM = log(MW/TD50 )mouse and BRR = log(MW/TD50 )rat . TD50 is the daily dose required to halve the probability for an experimental animal of remaining tumorless to the end of its standard life span [11]. The chemical parameters in the equations are: log P, which is a measure of hydrophobicity; energy of the highest occupied molecular orbital (EHOMO), energy of the lowest unoccupied molecular orbital (ELUMO); MR2,6 , sum of molar refractivity of substituents in the ortho-positions of the aniline ring; MR3 , molar refractivity of substituents in the meta-position of the aniline ring; Es (R), Charton’s substituent constant for substituents at the functional amino group;
193
I (monoNH2 ) = 1 for compounds with only one amino group; I (diNH2 ) = 1 for compounds with more than one amino group; I (Bi) = 1 for biphenyls; I (BiBr) = 1 for biphenyls with a bridge between the phenyl rings; I (RNNO) = 1 for compounds with the group N(Me)NO; I (F) = 1 for fluoroamines. N(Me)NO is a nitroso group, with a methyl substitution at the amino nitrogen. EHOMO and ELUMO were calculated with the program SYBYL (Tripos) after optimization by AM1; log P was calculated with the program TSAR (Oxford Molecular, now Accelrys). The terms in the equations point to the physical– chemical and structural determinants that govern the carcinogenic potency gradation, whereas the signs (+ and −) indicate the direction of the effects (increasing or decreasing). The key factor for carcinogenic potency is hydrophobicity (log P). Both BRM and BRR increase with increasing hydrophobicity. In the case of BRM (mouse) the influence of hydrophobicity is stronger for compounds with one amino group (characterized by the indicator variable I(monoNH2 )) in comparison with compounds with more than one amino group (characterized by the indicator variable I(diNH2 )) (see the different coefficients 0.88 and 0.29). For BRM, electronic factors also play a role: potency increases with increasing EHOMO and with decreasing ELUMO. Such effects seem to be less important for BRR (rat): no electronic terms occur in (Eq. (2). Carcinogenic potency also depends on the type of the ring system: aminobiphenyls (indicator variable I(Bi)) and, in the case of BRR, also fluoroamines (indicator variable I(F)) are intrinsically more active than anilines or naphthylamines. A bridge between the rings of the biphenyls decreases potency I(BiBr). Steric factors are involved in the case of BRM but cannot be detected in the case of BRR. BRM strongly decreases with bulk in the positions adjacent to the functional amino group, and bulky substituents at the nitrogen and in position 3 also decrease potency. The latter effects are, however, not so important. In the case of BRR, R = (Me)NO strongly enhances potency (compounds with this substituent have no measured value for BRM). Eqs. (1) and (2) were derived from the analysis of the carcinogenic aromatic amines only, and have a high explanatory power for the gradation of their carcinogenic potency (see r2 values). However, when we
194
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
applied the equations to the non-carcinogenic amines, we found that the equations did not predict well their lack of carcinogenic effects (the non-carcinogens were predicted as having a certain even though low-degree of activity). This means that the molecular determinants that rule the gradation of carcinogenic potency are not the same that make the difference between carcinogens and non-carcinogens. Thus, in a subsequent work we specifically studied the differences in molecular properties between the two classes of carcinogenic and non-carcinogenic aromatic amines [10]. Four equations were derived, one for each of the experimental groups (rat and mouse, male and female). The two classes were coded as: 1 = inactive; 2 = active compounds. The following discriminant function achieves a highly significant separation of classes for female rat carcinogenicity: w = 0.65L(R) + 0.79EHOMO − 1.54ELUMO +0.76MR2 − 0.50MR5 + 1.32I (An) −0.53I (o-NH2 ) + 0.99I (BiBr) + 0.99I (diNH2 ) −1.08 log P × I (diNH2 ) w(mean, class 1) = 1.05,
(3)
N1 = 30
w(mean, class 2) = −1.21, N2 = 26 where L(R) is the length of the substituent at the amino group; I (An) = 1 for anilines; I (o-NH2 ) = 1 if non-substituted amino group occurs in ortho-position to the functional amino group. The w(mean, class 1) is the mean of the w values of the class 1 chemicals; w(mean, class 2) is the mean of the w values of the class 2 chemicals. Chemicals with calculated w values closer to 1.05 are reclassified (predicted) as inactives; chemicals with calculated w values closer to −1.21 are reclassified as actives. The correct reclassification rate of discriminant function (3) amounts to 91.1% (class 1: 93.3%; class 2: 88.5%) with a fairly stable cross validation (all compounds: 80.4%; class 1: 76.7%; class 2: 84.6%). Cross-validation is a tool to assess the robustness of the model, and is performed by constructing a model on two thirds of the compounds, and checking the ability of the model to correctly predict the activity of the remaining one-third.
For male rat carcinogenicity a good separation of classes is achieved by discriminant function (4). w = 0.48L(R) + 0.90EHOMO − 1.43ELUMO +0.72MR2 + 1.13I (An) − 0.54I (o-NH2 ) −0.45MR5 + 0.70I (diNH2 ) −0.80 log P × I (diNH2 ) + 0.65I (BiBr) w(mean, class 1) = 1.15,
(4)
N1 = 28
w(mean, class 2) = −1.01, N2 = 32 The correct reclassification rate amounts to 91.7% (class 1: 92.9%; class 2: 90.6%) with a good result for cross validation (all compounds: 83.3%; class 1: 82.1%; class 2: 84.4%). The results obtained for male and female rat resemble each other. Of key importance for class separation are electronic properties as expressed by EHOMO and ELUMO, the type of ring system, and substitution in the ortho-position as well as at the amino nitrogen. The probability of a compound to be assigned to the active class increases with increasing values of ELUMO, decreasing values of EHOMO, decreasing bulk of substituents in position 2 (ortho-position), decreasing length (or bulk) of substituents at the amino nitrogen, and increasing number of aromatic rings (anilines have a distinctively lower probability to be active than biphenyls, fluorenes, or naphthalenes). An important feature promoting carcinogenic potency also is the occurrence of an amino group in ortho-position to the functional amino group. Of lesser importance are the variables I(diNH2 ), I(BiBr), MR5 , and the cross product log P × I (diNH2 ). It appears that the key factors differentiating between active and inactive compounds on the one hand, and governing potency within the group of active compounds are different. The most pronounced differences are with respect to the importance of hydrophobicity and the directionality of electronic effects. For female mouse carcinogenicity, the following discriminant function reclassifies 85.7% of the compounds correctly (class 1: 87.9%; class 2: 83.3%) and is of acceptable stability in cross validation (all compounds: 81.0%; class 1: 84.8%; class 2: 76.7%): w = −0.47I (NR) + 1.38 log P × I (monoNH2 ) +1.68 log P × I (diNH2 ) − 0.37I (An) (4.3.1) +0.33I (o-NH2 ) − 0.55MR5 −0.45I (BiBr)
(5)
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
w(mean, class 1) = −0.92, N1 = 33 w(mean, class 2) = 1.01,
N2 = 30
where I (NR) = 1 if the amino nitrogen is substituted. For male mouse discriminant function (6) is obtained: w = −1.96L(R) + 1.69B5 (R) − 0.83EHOMO +0.97ELUMO − 1.22I (An) + 0.73I (o-NH2 ) +0.59MR3 + 0.69MR5 + 0.77MR6 −0.76I (diNH2 ) + 1.09 log P × I (diNH2 ) −0.79I (BiBr)
(6)
w(mean, class 1) = −1.11, N1 = 25 w(mean, class 2) = 1.16,
N2 = 24
where B5 is the maximal width of the substituent at the amino group. It should be noted that the difference in sign of the average w values for the two classes in Eqs. (3)–(6) is only formal, and does not have any relevance on mechanisms. Discriminant function (6) shows a good reclassification rate (all compounds: 89.8%; class 1: 96.0%; class 2: 83.3%) and stability in cross validation (all compounds: 83.7%; class 1: 96.0%; class 2: 70.8%). The results for the mouse were similar to those found for the rat. Thus, hydrophobicity is a key factor for the gradation of the carcinogenic potency (Eqs. (1) and (2)) but only of small importance for yes/no activity (Eqs. (3)–(6)). The reverse is true for electronic properties (EHOMO, ELUMO), which show a minor effect for the gradation of potency, but a pronounced effect for yes/no activity. Eqs. (4)–(6) also point out the importance of steric (shape, size) factors for yes/no activity: for example, in all four equations the first term indicates that the probability of being non-carcinogenic increases with increasing length of the substituent L(R) or simply with the presence of a substituent I(NR) at the amino nitrogen. Hydrophobicity is a force involved in the absorption and transport of the drugs in the cells and organisms, as well as in the interaction between drugs and the metabolizing enzymes. The electronic parameters are measures of chemical reactivity, hence of the ability of undergoing metabolic transformations. It should be noted that the results of the QSAR analyses agree with the
195
notion that the aromatic amines require metabolic activation to become carcinogenic [1]. For amines and amides, this typically involves an initial N-oxidation to N-hydroxylamine and N-hydroxylamide. In particular, EHOMO is a parameter for oxidation reactions. For a more detailed discussion of the chemical structural parameters involved in the carcinogenicity of the aromatic amines and how they relate to the mechanisms of action, see [10]. Here, we would like to emphasize two points. First, the QSAR analyses help to give a firm and quantitative basis to many notions which were previously only qualitative: this happens through the establishment of equations where the contribution of each physical–chemical and structural parameter to the biological activity is quantitatively defined. Second, the QSAR analyses can provide entirely new information. For example, an original contribution of this study is the notion that the carcinogenic activity of the aromatic amines is a non-linear phenomenon. The non-carcinogenic amines are not simply “weakly active”. On the contrary there is a physical–chemical and structural border between the amines that can be carcinogenic and those that cannot. This border, or threshold, can be imagined as a surface in the space of the chemical descriptors: Eqs. (3)–(6) permit the calculation of the border as a combination of the different parameters present in the equations. 1.2. Comparative QSAR: learning from models for different end-points The importance and widespread use of aromatic amines has permitted the accumulation of a large amount of experimental data on the effects of these compounds in a range of chemical, physical and biological systems. This in turn has generated a variety of QSARs. This allows us to compare the QSARs obtained, to check their consistency and to highlight general trends. This approach is what Corwin Hansch has called “comparative QSAR”: “unless we can show that there are meaningful relationships among QSARs for different chemicals acting on the same or different systems we cannot call it a science. Establishing lateral correlations among QSARs is the only path to developing such a science” [12]. To this aim a database of QSARs (now including around 17,000 equations) and a software to interrogate the database (C-QSAR) has been developed at Pomona College
196
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
under the direction of Hansch, who generously permitted us to search for the existing QSARs on the aromatic amines. The enormous repertoire of equations contained in the C-QSAR database is divided into two main sections: chemical–biological interactions and physical–chemical reactions. A major point of interest in the development of the interrogation computerized system is the ability to compare QSARs corresponding to rather simple chemical reactions with those of biological systems. In many cases such comparisons were of definite value in understanding the more complex biological processes [12,13]. From the C-QSAR database we selected the QSARs which were of interest for this paper; they are listed in Appendix A. The high r fitting parameter of the equations should be remarked. The following is a discussion of the salient points of this comparison. A first comparison is with the QSARs for the mutagenic potency of the aromatic amines in Salmonella (Appendix A, Section A.1) (see also [14]). Interestingly, the QSARs for Salmonella mutagenic potency pointed to a pattern similar to that for the rodent carcinogenic potency. In both cases hydrophobicity (log P) is the major molecular determinant. Second in order of importance are the reactivity parameters, which are present in the equations for TA98 and TA100 Salmonella strains, as well as in the model for the mouse carcinogenicity (Eq. (1)). Thus, there is a high degree of qualitative similarity between the two sets of equations, hence between the most important molecular determinants of the mutagenic and carcinogenic activities. Eqs. (3)–(6) apply to the difference between carcinogenic and non-carcinogenic aromatic amines. No similar equations exist in the C-QSAR database for mutagenicity. However, in our previous studies [15] we found that also the difference between mutagenic and non-mutagenic amines is determined mainly by electronic and steric factors, and that hydrophobicity plays—if any—a minor role. Overall, this evidence represents a support for the similarity in mechanism of the aromatic amines in Salmonella and in rodents, and for their genotoxic mechanism of carcinogenicity. Appendix A, Section A.2 reports the QSARs for general toxicity of aromatic amine datasets. The range of systems and organisms is impressive: it includes single cells (both procaryotic and eukaryotic), cells
in culture, and animals. In spite of such a variety, the simplicity and homogeneity of these QSARs is equally impressive. The total number of QSAR equations is 36. Out of them, 20 equations are based exclusively on log P, and in 30 out of 36 equations log P is the first (most important) parameter. Only two equations do not contain log P. The comparison with the QSARs for the rodent carcinogenicity and the Salmonella mutagenicity points to a lower degree of complexity and to different underlying mechanisms of action of general toxicity. This result can be better understood by recalling the research of Koneman [16] on inert narcotic pollutants, that led to the so-called baseline toxicity concept: chemicals with toxicities in line with the baseline concept are classified as inert and are not interacting with specific receptors in an organism. They are presumed to interact with cellular membranes, producing an aspecific effect (narcosis) that can be parametrized by log P only. However, some classes of compounds are more toxic than predicted by the baseline concept, indicating the existence of additional specific components in the toxicity mechanisms. In these cases, the QSAR models contain additional parameters (electronic and/or steric) [5,17]. It appears that the QSARs for the general toxicity of the aromatic amines follow the above scheme: hydrophobicity is largely the main (or unique) determinant, thus, pointing mainly to an aspecific mode of action. It should be remarked that this evidence is in agreement with independent assignments of the aromatic amines to mode-of-action categories based on a range of experimental determinations [18]. Thus, the operation of comparing QSARs at different complexity levels and for different end-points reveals details of general scientific relevance. The picture changes again with the enzymatic reactions (Appendix A, Section A.3). Among the many types of enzymatic reactions contained in the C-QSAR database, we have selected those relevant for the mechanism of action of the aromatic amines in carcinogenicity and mutagenicity. The most relevant is the oxidation of the amino group, which is considered to be the first step in the main metabolic pathway [1]. As for general toxicity, the equations for enzymatic reactions are simpler than in the case of carcinogenicity and mutagenicity; however the parameters included are different from those predominant in general toxicity. It appears that the oxidation
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
QSARs rely systematically on σ (Hammett electronic descriptor), which is the first parameter even in the few cases in which it is complemented by π (hydrophobicity parameter for the substituent groups, whereas log P is a parameter for the entire molecule) or by steric parameters. The constant presence of σ in the equations is consistent with the fact that the oxidation reactions can take place only when the electronic characteristics of the amine group ensure sufficient reactivity, whereas π and the steric parameters may modulate the correct interaction between enzyme and substrate. Positive σ values represent electron withdrawal from the ring by the substituents; negative σ values indicate electron release to the ring. As reflected by the negative sign of the π coefficients in the QSAR models, the oxidation of the amino nitrogen is favored by electron releasing substituents on the ring ([2], chapter 2). This makes more electrons available both for the chemical process of oxidation and for the nucleophilic attack to the enzyme (which is a hypothesized step of the enzymatic oxidation). Appendix A, Section A.3 also shows the QSARs for p-hydroxylation. There is evidence that in addition to the cytochrome P450-mediated N-hydroxylation pathway, some anilines may induce tumors via the formation of a reactive quinoneimine metabolite after bioactivation by peroxidases [1]. This involves ring hydroxylation as first step. At odds with the QSARs for the N-hydroxylation, five out six QSARs for p-hydroxylation are based on log P. This can be explained by the fact that the process does not depend on the reactivity at some specific reaction center but on the general quality of the interaction between enzyme and substrate. Regarding the QSARs for the physical–chemical reactions, we selected those relative to general oxidations, as they are potentially relevant for the mechanism of action in carcinogenicity and mutagenicity (Appendix A, Section A.4). The picture is consistent with that for the enzymatic reactions: the electronic parameters (σ ) are the unique or main determinants of the reactions. The only exception are the oxidations involving halogen compounds. The halogens are characterized by inductive and resonance effects with opposite signs. The net balance between two opposite reactions accounts for the positive sign in the equations. Moreover, it should be remarked that most of the equations have negative
197
coefficients for σ , as in the case of the enzymatic oxidations. No hydrophobic parameter appears, and only steric parameters sometimes modulate the reaction. This short discussion on the comparative QSAR of the aromatic amines, and the consistencies found, confirm the strength of the scientific approach of lateral comparison of QSARs for the investigation of mechanisms and for providing generalizations. More specifically, the comparative QSAR analysis has highlighted the similarity in the mechanisms of mutagenicity and carcinogenicity of the aromatic amines, and the difference with the general toxicity mechanisms. In addition, the mechanistic coherence between chemical–biological reactions and their physical–chemical basis has been pointed out (see, for example, the similarity between enzymatic and physical–chemical oxidations). Obviously, there are limitations to the amount and quality of the knowledge that can be derived from the comparison of the QSARs. In the present case, these limitations largely depend on the type of information on which the comparison relies. For example, the bioassay experiments, whose results were used in our QSAR studies, were not designed specifically for QSAR modeling, but for the assessment of the risk posed by the individual aromatic amines. This means that the selection of the chemicals did not follow the criteria that maximize the ability to identify the contributions of the different chemical parameters to the activity (e.g. balanced selection of chemicals representative of the extent of the chemical parameters space, orthogonality of the chemical parameters) [2]. In addition, the QSAR models in Appendix A were generated by different authors based on different subsets of aromatic amines. The strength of the comparative approach is demonstrated by the fact that, in spite of this, a clear and informative general picture emerges. However, the discussion of the results should respect the imperfect nature of the data analyzed, and speculations should not go too far. For example, Eqs. (3)–(6) point to different models for rat and mouse, and also to differences between male and female mice. Before drawing conclusions on the underlying mechanisms based only on these four equations, further QSAR models relative to the carcinogenicity of other classes of chemicals should be examined.
198
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
1.3. QSAR models: implications for risk assessment The discussion in this paper has focused on the QSAR models for the rodent carcinogenicity of the aromatic amines. The QSARs for the yes/no carcinogenic activity and for the modulation of the carcinogenic potency have been summarized. It should be recalled that the rodent bioassay provides a large amount of information, namely: (a) yes/no carcinogenic activity of the tested compound; (b) the carcinogenic potency of those chemicals determined to be active; (c) information on the target organs. The yes/no activity is the most important information: all the human carcinogens were positive in the bioassay when adequately tested, and many human carcinogens were first indicted by positive carcinogenicity test results in rodents. Thus, the yes/no result from a rodent bioassay is very relevant to the human risk assessment [19–22]. The information on the potency is also important. There is evidence for a strong correlation: (a) between the potency ranking of the carcinogens in rat and mouse [23]; (b) between carcinogenic potency in humans estimated from epidemiologic data, and that estimated from the rodent bioassay [24]. The similar ranking of the carcinogenic potency in different species suggests that potency is an intrinsic property of a chemical carcinogen that is more directly derived from its chemical reactivity. Thus, both yes/no activity and carcinogenic potency are relevant for the extrapolation of animal data to assessing human risk. In this context, the QSARs generated in our laboratory can be used directly for estimating the carcinogenicity of non-heterocyclic aromatic amines for which the experimental carcinogenicity data are not available. A two-step prediction of carcinogenicity of aromatic amines is possible: (1) estimation of yes/no activity from the Eqs. (3)–(6); (2) if the answer from step 1 is yes, then prediction of the degree of potency from Eqs. (1) and (2). Thus, the QSAR models can contribute to the: synthesis of safer chemicals; estimation of the risk posed by aromatic amines in the environment; setting priorities for experimentation; reducing the use of animals. Any new animal bioassay results will, in turn, allow for further refinement and validation of these models. The pattern of target organs is the third piece of evidence provided from the rodent bioassay. The tumor types induced appear to be highly variable
from species to species [23], and also vary with the condition of use (e.g. age of the host, dose and route of administration) of the carcinogen [25]. For example, we calculated that in the present aromatic amines database the overlap of tumor types induced in rat and mouse by the individual chemicals was a mere 18%. Thus, this information is not suitable as a basis for extrapolation to human risk. It is interesting that, whereas we obtained quite satisfactory QSAR models for the yes/no activity and for carcinogenic potency of the aromatic amines, we were not successful in modeling the carcinogenic potency for the individual organs (our unpublished results). A parallel line of evidence is that the aromatic amines, like other classes of chemical carcinogens, do not induce in animals one or a few specific patterns of tumor types, but can induce the widest spectrum of tumors [26]. This indicates that the species differences in the tumor profiles may result not only from differences in targeted reactions of the ultimate carcinogens, but also in the myriad of events that mediate and surround these reactions (e.g. toxicokinetics): this makes particularly difficult the modeling of tumor profiles and—what is most important—makes the tumor profiles not particularly useful for species extrapolation in risk assessment. The QSAR models obtained for the carcinogenicity of the aromatic amines are all statistically robust and provide good fits to existing data. Both QSARs for the carcinogenic potency (Eqs. (1) and (2)) have r 2 = 0.8; the models for the separation of carcinogens from non-carcinogens (Eqs. (3)–(6)) have discriminant accuracy in the range 80–90%. This is a very encouraging result, and it looks even better when one takes into account that we modeled animal data, which have an intrinsic experimental variability and— most important—were generated in different laboratories in different periods. Approximately two-thirds of the chemicals used to build our QSAR models were generated by the NCI/NTP experimentation with well controlled protocols, whereas the remaining one-third had different origins [10]. In light of the critical importance of the experimental protocols in the rodent bioassay results, the goodness of fit of the QSARs found is quite high, and deserves some comments. A first comment comes from the comparison of the above QSARs with the limited accuracy shown by various systems devised in the recent years for the prediction of chemical carcinogenicity. Chemical
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
carcinogenicity has been the target of a more diverse array of structure-based predictive modeling efforts than perhaps any other toxicological end-point. This has been encouraged, in large part, by the historical successes of the structure–activity paradigm, particularly when guided by mechanistic considerations applied to well defined classes of chemicals and end-points. In addition, it has been encouraged by the prominent role that a positive rodent carcinogenicity result plays in driving regulatory action on chemicals, and by the enormous investment of time, resources and animal lives required for testing a single chemical. Thus, many “all-purposes” models, hopefully aimed at predicting the carcinogenicity of any kind of chemical, irrespective of the chemical class, have been devised [27,28]. In the last decade, unique opportunities to objectively assess the relative performance of model systems for carcinogenicity prediction have been provided by two prospective prediction toxicity exercises held under the aegis of the US National Toxicology Program (NTP). Both exercises invited the modeling community to submit predictions of rodent carcinogenicity for chemicals that were in the process of being bioassayed by the NTP: the predictions were published before the bioassay results were known and made public [29,30]. The prediction systems had very different rational basis and implementation (biologically-, structure–activity-based, computerized systems, human experts, etc.). Detailed analyses of the results of the comparative exercises are in [8,31,32]. It appeared that the best performance in prediction was around 70%, but many prediction systems had very low performance. This is quite inferior to the 80–90% accuracy obtained when modeling the carcinogenicity of the aromatic amines. This gap makes sense. The all-purpose models attempt to model at the same time many different mechanisms of action (each class of chemical carcinogens has its specific
199
mechanism); moreover, they rely on the database of bioassay results that is well represented only for a few classes of carcinogens (e.g. the aromatic amines) and very under-represented for most classes. Thus, two sources of difficulty combine together: the difficulty in modeling simultaneously many mechanisms adds to the limited database from where the predictive approaches can learn. This explains the limited performance of the “all-purpose” approaches. On the other hand, the application of the powerful QSAR methods to the mechanistically well defined and sufficiently represented class of the aromatic amines permitted the attainment of a much higher modeling accuracy. Thus, the results obtained with the aromatic amines can be considered as an estimate of the best performance that can be reached in modeling the bioassay data. As final consideration, it should be remarked that the successful modeling of in vivo data is a strong argument in favor of the reliability of the traditional rodent carcinogenicity assay. In spite of the different sources (laboratories) of the rodent data used in our work, the data were coherent enough to allow the QSAR modeling to highlight with precision the underlying chemical determinants of the carcinogenicity of the aromatic amines. The QSAR models were robust from a statistical point of view, and were coherent and meaningful from a mechanistic point of view. Decades of experience in the field of QSAR indicates that this could not have happened without a solid basis of good quality data.
Acknowledgements Prof. Corwin Hansch is gratefully acknowledged for generously providing us the access to the C-QSAR database. We thank Dr. Marina Cotta-Ramusino for her continued interest and for the helpful discussions.
Appendix A A.1. Mutagenicity Salmonella typhimurium TA98 + S9 log TA98 = 1.08 log P + 1.28HOMO − 0.73LUMO + 1.46IL + 7.20 TA100 + S9 log TA100 = 0.92 log P + 1.17HOMO − 1.18LUMO + 7.35
n = 88; s = 0.860; r = 0.898; omit = 7 n = 67; s = 0.708; r = 0.877; omit = 6
200
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
201
202
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
203
204
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
References [1] Y.T. Woo, D.Y. Lai, Aromatic amino and nitro-amino compounds and their halogenated derivatives, in: E. Bingham, B. Cohrssen, C.H. Powell (Eds.), Patty’s Toxicology, Wiley, New York, 2001, pp. 969–1105. [2] C. Hansch, A. Leo, Exploring QSAR. Part 1. Fundamentals and Applications in Chemistry and Biology, American Chemical Society, Washington, DC, 1995. [3] T. Fujita, Recent success stories leading to commercializable bioactive compounds with the aid of traditional QSAR procedures, Q. Struct. Act. Relat. 16 (1997) 107–112. [4] C. Hansch, D. Kim, A.J. Leo, E. Novellino, C. Silipo, A. Vittoria, Toward a quantitative comparative toxicology of organic compounds, Crit. Rev. Toxicol. 19 (1989) 185– 226. [5] H.J.M. Verhaar, J. Solbe, J. Speksnijder, C.J. van Leeuwen, J.L.M. Hermens, Classifying environmental pollutants. Part 3. External validation of the classification system, Chemosphere 40 (2000) 875–883. [6] H. Hong, W. Tong, H. Fang, L.M. Shi, Q. Xie, J. Wu, R. Perkins, J.D. Walker, W. Branham, D.M. Sheehan, Prediction of estrogen-receptor binding for 58,000 chemicals using an integrated system of a tree-based model with structural alerts, Environ. Health Perspect. 110 (2002) 29–36. [7] R. Benigni, A. Giuliani, Quantitative structure–activity relationship (QSAR) studies of mutagens and carcinogens, Med. Res. Rev. 16 (1996) 267–284. [8] A.M. Richard, R. Benigni, AI and SAR approaches for predicting chemical carcinogenicity: survey and status report, SAR QSAR Environ. Res. 13 (1) (2002) 1–19. [9] R. Benigni, A. Giuliani, R. Franke, A. Gruska, Quantitative structure–activity relationships of mutagenic and carcinogenic aromatic amines, Chem. Rev. 100 (2000) 3697–3714. [10] R. Franke, A. Gruska, A. Giuliani, R. Benigni, Prediction of rodent carcinogenicity of aromatic amines: a quantitative structure–activity relationships model, Carcinogenesis 22 (2001) 1561–1571. [11] L.S. Gold, N.B. Manley, T.H. Slone, L. Rohrbach, Supplement to the carcinogenic potency database (CPDB): results of animal bioassays published in the general literature in 1993–1994 and by the National Toxicology Program in 1995–1996, Environ. Health Perspect. Suppl. 107 (Suppl. 4) (1999) 527–600. [12] C. Hansch, D. Hoekman, H. Gao, Comparative QSAR: toward a deeper understanding of chemical–biological interactions, Chem. Rev. 96 (1996) 1045–1075. [13] C. Hansch, H. Gao, Comparative QSAR: radical reactions of benzene derivatives in chemistry and biology, Chem. Rev. 97 (1997) 2995–3060. [14] A.K. Debnath, G. Debnath, A.J. Shusterman, C. Hansch, A QSAR investigation of the role of hydrophobicity in regulating mutagenicity in the Ames test. Part 1. Mutagenicity of aromatic and heteroaromatic amines in Salmonella typhimurium TA98 and TA100, Environ. Mol. Mutagen. 19 (1992) 37–52.
205
[15] R. Benigni, L. Passerini, G. Gallo, F. Giorgi, M. CottaRamusino, QSAR models for discriminating between mutagenic and non-mutagenic aromatic and heteroaromatic amines, Environ. Mol. Mutagen. 32 (1998) 75–83. [16] H. Koneman, Quantitative structure–activity relationships in fish toxicity studies. I. Relationships for 50 industrial pollutants, Toxicology 19 (1981) 209–221. [17] J.L.M. Hermens, Quantitative structure–activity relationships of environmental pollutants, in: O. Hutzinger (Ed.), Handbook of Environmental Chemistry, Springer, Berlin, 1989, pp. 111–162. [18] C.L. Russom, S.P. Bradbury, S.J. Broderius, D.E. Hammermeister, R.A. Drummond, Predicting modes of toxic action from chemical structure: acute toxicity in fathead minnow (Pimephales promelas), Environ. Toxicol. Chem. 16 (1997) 948–967. [19] J. Huff, Value, validity and historical development of carcinogenesis studies for predicting and confirming carcinogenic risks to humans, in: K.T. Kitchin (Ed.), Carcinogenicity: Testing, Predicting, and Interpreting Chemical Effects, Marcel Dekker, New York, 1999, pp. 21–123. [20] V.A. Fung, J.C. Barrett, J. Huff, The carcinogenesis biossay in perspective: application in identifying human cancer hazards, Environ. Health Perspect. 103 (1995) 680–683. [21] L. Tomatis, J. Huff, I. Hertz-Picciotto, D.P. Sandler, J. Bucher, P. Boffetta, O. Axelson, A. Blair, J. Taylor, L. Stayner, J.C. Barrett, Avoided and avoidable risks of cancer, Carcinogenesis 18 (1997) 97–105. [22] L. Tomatis, J. Huff, Evolution of cancer etiology and primary prevention, Environ. Health Perspect. 109 (2001) 5–7. [23] R. Benigni, A. Giuliani, Tumor profiles and carcinogenic potency in rodents and humans: value for cancer risk assessment, Environ. Carcinogen. Ecotoxicol. Rev. C17 (1999) 45–67. [24] B.C. Allen, K.S. Crump, A.M. Shipp, Correlation between carcinogenic potency of chemicals in animals and humans, Risk Anal. 8 (1988) 531–544. [25] T.J. Bucci, Profiles of induced tumors in animals, Toxicol. Pathol. 13 (1985) 105–109. [26] R. Benigni, A. Pino, Profiles of chemically-induced tumors in rodents: quantitative relationships, Mutat. Res. 421 (1998) 93–107. [27] J.C. Dearden, M.D. Barratt, R. Benigni, D.W. Bristol, R.D. Combes, M.T.D. Cronin, P.N. Judson, M.P. Payne, A.M. Richard, M. Tichy, A.P. Worth, J.J. Yourick, The development and validation of expert systems for predicting toxicity, Alternat. Lab. Anim. (ATLA) 25 (1997) 223–252. [28] R. Benigni, A.M. Richard, Quantitative structure-based modeling applied to characterization and prediction of chemical toxicity, Methods 14 (1998) 264–276. [29] R.W. Tennant, J. Spalding, S. Stasiewicz, J. Ashby, Prediction of the outcome of rodent carcinogenicity bioassays currently being conducted on 44 chemicals by the National Toxicology Program, Mutagenesis 5 (1990) 3–14. [30] D.W. Bristol, J.T. Wachsman, A. Greenwell, The NIEHS predictive—Toxicology Evaluation Project: chemcarcinogenicity biossay, Environ. Health Perspect. 104 (1996) 1001–1010.
206
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
[31] R. Benigni, The first US National Toxicology Program exercise on the prediction of rodent carcinogenicity: definitive results, Mutat. Res. 387 (1997) 35–45. [32] R. Benigni, Structure–activity relationships in mutagenesis and carcinogenesis, in: M. Balls, A.M. van Zeller, M.E. Halder (Eds.), Progress in the Reduction, Refinement
and Replacement of Animal Experimentation, Elsevier, Amsterdam, 2000, pp. 469–478. [33] C. Hansch, A. Leo, D. Hoekman, Exploring QSAR. Part 2 Hydrophobic, Electronic and Steric Constants, American Chemical Society, Washington, DC, 1995.
Review
Carcinogenicity of the aromatic amines: from structure–activity relationships to mechanisms of action and risk assessment Romualdo Benigni∗ , Laura Passerini Laboratory of Comparative Toxicology and Ecotoxicology, Istituto Superiore di Sanita’, Viale Regina Elena 299, 00161 Rome, Italy Received 18 January 2002; received in revised form 12 March 2002; accepted 12 March 2002
Abstract Aromatic amines represent one of the most important classes of industrial and environmental chemicals: many of them have been reported to be powerful carcinogens and mutagens, and/or hemotoxicants. Their toxicity has been studied also with quantitative structure–activity relationship (QSAR) methods: these studies are potentially suitable for investigating mechanisms of action and for estimating the toxicity of compounds lacking experimental determinations. In this paper, we first summarized the QSAR models for the rodent carcinogenicity of the aromatic amines. The gradation of potency of the carcinogenic amines depended firstly on their hydrophobicity, and secondly on electronic (reactivity, propensity to be metabolically transformed) and steric properties. On the contrary, the difference between carcinogenic and non-carcinogenic aromatic amines depended mainly on electronic and steric properties. These QSARs can be used directly for estimating the carcinogenicity of aromatic amines. A two-step prediction is possible: (1) estimation of yes/no activity; (2) if the answer from step 1 is yes, then prediction of the degree of potency. The QSARs for rodent carcinogenicity were put in a wider context by comparing them with those for: (a) Salmonella mutagenicity; (b) general toxicity; (c) enzymatic reactions; (d) physical–chemical reactions. This comparative QSAR exercise generated a coherent global picture of the action mechanisms of the aromatic amines. The QSARs for carcinogenicity were similar to those for Salmonella mutagenicity, thus pointing to a similar mechanism of action. On the contrary, the general toxicity QSARs (both in vitro and in vivo systems) were mostly based on hydrophobicity, pointing to an aspecific mechanism of action much simpler than that for carcinogenicity and mutagenicity. The oxidation of the amines (first step in the main metabolic pathway leading to carcinogenic and mutagenic species) had identical QSARs in both enzymatic and physical–chemical systems, thus providing evidence for the link between simple chemical reactions and those in biological systems. The results show that it is possible to generate mechanistically and statistically sound QSAR models for rodent carcinogenicity, and indirectly that the rodent bioassay is a reliable source of good quality data. © 2002 Elsevier Science B.V. All rights reserved. Keywords: QSAR; Amine; Carcinogen; Structure–activity; Toxicity
1. Introduction
∗ Corresponding author. Tel.: +39-64990-2579; fax: +39-64938-7139. E-mail address: [email protected] (R. Benigni).
Aromatic amines represent one of the most important classes of industrial and environmental chemicals. Many aromatic amines have been reported to be powerful carcinogens and mutagens, and/or hemotoxicants. Exposure to aromatic amines occurs in
1383-5742/02/$ – see front matter © 2002 Elsevier Science B.V. All rights reserved. PII: S 1 3 8 3 - 5 7 4 2 ( 0 2 ) 0 0 0 0 8 - X
192
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
Nomenclature B1 -STM B5 -STM ES
F HOMO I IL KM log A log C log K log Kcat log Krel log P log TA log Vmax LUMO n omit pKa PI
sterimol width of the first atom of the substituent sterimol overall width of the substituent steric Taft parameter for substituents (for intramolecular steric effects) Swain–Lupton inductive/field effect for aromatic systems energy of the Highest Occupied Molecular Orbital substituent position indicator is 1, if three or more fused rings are present Michaelis–Menten constant A, activity C, molar concentration (mol/l or mol/kg) K, rate or equilibrium constant Kcat , catalytic rate constant Krel , relative to H P, octanol/water partition coefficient TA (revertants/mol) Vmax , maximum reaction rate energy of the Lowest Unoccupied Molecular Orbital number of datapoints omitted datapoints acidic dissociation constant hydrophobic parameter (π ) for substituents correlation coefficient standard deviation
r s suffixes 1, 3, 4, etc. ring substitution position S Hammet electronic (σ ) constant for substitutents
different industrial and agricultural activities as well as in tobacco smoking. Substantial worker exposure to aromatic amines with subsequent induction of bladder cancer occurred before preventive measures were instituted. Owing to their hazard potential, aromatic amines have been the subject of many in vivo and in vitro experimental studies, as well as biomonitoring
investigations. For an updated review on the toxicology of aromatic amines and their mechanisms of action, see [1]. Given their importance and the large amount of data available, the toxicity of the aromatic amines has been studied also with methods based on structure–activity relationship (SAR) and quantitative SAR (QSAR) concepts.1 The foundation of the modern QSAR science came about in the 1960s, after many attempts essentially qualitative. At present QSAR is one of the basic tools of modern drug and pesticide design [2,3], and has an increasing role in environmental sciences [4–8]. Its strength derives from the fact that it permits the identification of the molecular determinants of the biological action of the chemicals, and provides mathematical models to predict the activity of chemicals not tested experimentally provided that data on similar chemicals, acting with the same mechanism, are available. Several QSAR studies on the aromatic amines have been reported, mainly regarding their mutagenic properties. Surprisingly, very sporadic and limited QSAR studies of their carcinogenic properties existed until recently, in spite of the fact that several of them had been bioassayed thus providing the necessary database [9]. We have recently presented the first detailed QSAR analysis of the carcinogenicity of the aromatic amines, by investigating: (a) the molecular determinants that rule the gradation of the carcinogenic potency [9]; (b) the molecular determinants that make some of the aromatic amines non-carcinogenic [10]. The availability of a range of QSARs on different aspects of the toxicology of aromatic amines permits now to put the various pieces of evidence into perspective, and provides new insights into the mechanisms of action of the amines in different biological and chemical–physical systems, as well as on the use of QSAR for risk assessment. This paper presents: (a) a summary of the QSAR models for the carcinogenicity of the aromatic amines and implications for the mechanisms of action; (c) a comparative analysis of the QSARs of the aromatic amines for various toxicity end-points, as well as for enzymatic and chemical–physical reactions; (d) the implications of our results for the risk assessment of the aromatic amines. 1 QSAR models retrieved from the C-QSAR database. For more details on the chemical parameters (see [2,33]).
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
1.1. The QSAR models for the rodent carcinogenicity of the aromatic amines In our first QSAR analysis of the carcinogenicity of the aromatic amines, we considered only the carcinogenic aromatic amines, and we investigated the structural factors that influence the gradation of carcinogenic potency in rodents [9]. The study focused on the homogeneous class of non-heterocyclic amines. The following are the QSAR models emerging from the analysis of the bioassay data (BRM = carcinogenic potency in mice; BRR = carcinogenic potency in rats) BRM = 0.88(±0.27) log P × I (monoNH2 ) + 0.29(±0.20) log P × I (diNH2 ) + 1.38(±0.76)EHOMO − 1.28(±0.54)ELUMO − 1.06(±0.34) MR2,6 − 1.10(±0.80)MR3 − 0.20(±0.16)ES (R) + 0.75(±0.75)I (diNH2 ) + 11.16(±6.68) n = 37,
r = 0.907,
F = 16.3,
r 2 = 0.823,
s = 0.381,
P < 0.001
(1)
BRR = 0.35(±0.18) log P + 1.93(±0.48) I (Bi) + 1.15(±0.60)I (F) − 1.06(±0.53)I (BiBr) + 2.75(±0.64)I (RNNO) − 0.48(±0.30) n = 41, F = 47.4,
r = 0.933, P < 0.001
r 2 = 0.871,
s = 0.398, (2)
where BRM = log(MW/TD50 )mouse and BRR = log(MW/TD50 )rat . TD50 is the daily dose required to halve the probability for an experimental animal of remaining tumorless to the end of its standard life span [11]. The chemical parameters in the equations are: log P, which is a measure of hydrophobicity; energy of the highest occupied molecular orbital (EHOMO), energy of the lowest unoccupied molecular orbital (ELUMO); MR2,6 , sum of molar refractivity of substituents in the ortho-positions of the aniline ring; MR3 , molar refractivity of substituents in the meta-position of the aniline ring; Es (R), Charton’s substituent constant for substituents at the functional amino group;
193
I (monoNH2 ) = 1 for compounds with only one amino group; I (diNH2 ) = 1 for compounds with more than one amino group; I (Bi) = 1 for biphenyls; I (BiBr) = 1 for biphenyls with a bridge between the phenyl rings; I (RNNO) = 1 for compounds with the group N(Me)NO; I (F) = 1 for fluoroamines. N(Me)NO is a nitroso group, with a methyl substitution at the amino nitrogen. EHOMO and ELUMO were calculated with the program SYBYL (Tripos) after optimization by AM1; log P was calculated with the program TSAR (Oxford Molecular, now Accelrys). The terms in the equations point to the physical– chemical and structural determinants that govern the carcinogenic potency gradation, whereas the signs (+ and −) indicate the direction of the effects (increasing or decreasing). The key factor for carcinogenic potency is hydrophobicity (log P). Both BRM and BRR increase with increasing hydrophobicity. In the case of BRM (mouse) the influence of hydrophobicity is stronger for compounds with one amino group (characterized by the indicator variable I(monoNH2 )) in comparison with compounds with more than one amino group (characterized by the indicator variable I(diNH2 )) (see the different coefficients 0.88 and 0.29). For BRM, electronic factors also play a role: potency increases with increasing EHOMO and with decreasing ELUMO. Such effects seem to be less important for BRR (rat): no electronic terms occur in (Eq. (2). Carcinogenic potency also depends on the type of the ring system: aminobiphenyls (indicator variable I(Bi)) and, in the case of BRR, also fluoroamines (indicator variable I(F)) are intrinsically more active than anilines or naphthylamines. A bridge between the rings of the biphenyls decreases potency I(BiBr). Steric factors are involved in the case of BRM but cannot be detected in the case of BRR. BRM strongly decreases with bulk in the positions adjacent to the functional amino group, and bulky substituents at the nitrogen and in position 3 also decrease potency. The latter effects are, however, not so important. In the case of BRR, R = (Me)NO strongly enhances potency (compounds with this substituent have no measured value for BRM). Eqs. (1) and (2) were derived from the analysis of the carcinogenic aromatic amines only, and have a high explanatory power for the gradation of their carcinogenic potency (see r2 values). However, when we
194
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
applied the equations to the non-carcinogenic amines, we found that the equations did not predict well their lack of carcinogenic effects (the non-carcinogens were predicted as having a certain even though low-degree of activity). This means that the molecular determinants that rule the gradation of carcinogenic potency are not the same that make the difference between carcinogens and non-carcinogens. Thus, in a subsequent work we specifically studied the differences in molecular properties between the two classes of carcinogenic and non-carcinogenic aromatic amines [10]. Four equations were derived, one for each of the experimental groups (rat and mouse, male and female). The two classes were coded as: 1 = inactive; 2 = active compounds. The following discriminant function achieves a highly significant separation of classes for female rat carcinogenicity: w = 0.65L(R) + 0.79EHOMO − 1.54ELUMO +0.76MR2 − 0.50MR5 + 1.32I (An) −0.53I (o-NH2 ) + 0.99I (BiBr) + 0.99I (diNH2 ) −1.08 log P × I (diNH2 ) w(mean, class 1) = 1.05,
(3)
N1 = 30
w(mean, class 2) = −1.21, N2 = 26 where L(R) is the length of the substituent at the amino group; I (An) = 1 for anilines; I (o-NH2 ) = 1 if non-substituted amino group occurs in ortho-position to the functional amino group. The w(mean, class 1) is the mean of the w values of the class 1 chemicals; w(mean, class 2) is the mean of the w values of the class 2 chemicals. Chemicals with calculated w values closer to 1.05 are reclassified (predicted) as inactives; chemicals with calculated w values closer to −1.21 are reclassified as actives. The correct reclassification rate of discriminant function (3) amounts to 91.1% (class 1: 93.3%; class 2: 88.5%) with a fairly stable cross validation (all compounds: 80.4%; class 1: 76.7%; class 2: 84.6%). Cross-validation is a tool to assess the robustness of the model, and is performed by constructing a model on two thirds of the compounds, and checking the ability of the model to correctly predict the activity of the remaining one-third.
For male rat carcinogenicity a good separation of classes is achieved by discriminant function (4). w = 0.48L(R) + 0.90EHOMO − 1.43ELUMO +0.72MR2 + 1.13I (An) − 0.54I (o-NH2 ) −0.45MR5 + 0.70I (diNH2 ) −0.80 log P × I (diNH2 ) + 0.65I (BiBr) w(mean, class 1) = 1.15,
(4)
N1 = 28
w(mean, class 2) = −1.01, N2 = 32 The correct reclassification rate amounts to 91.7% (class 1: 92.9%; class 2: 90.6%) with a good result for cross validation (all compounds: 83.3%; class 1: 82.1%; class 2: 84.4%). The results obtained for male and female rat resemble each other. Of key importance for class separation are electronic properties as expressed by EHOMO and ELUMO, the type of ring system, and substitution in the ortho-position as well as at the amino nitrogen. The probability of a compound to be assigned to the active class increases with increasing values of ELUMO, decreasing values of EHOMO, decreasing bulk of substituents in position 2 (ortho-position), decreasing length (or bulk) of substituents at the amino nitrogen, and increasing number of aromatic rings (anilines have a distinctively lower probability to be active than biphenyls, fluorenes, or naphthalenes). An important feature promoting carcinogenic potency also is the occurrence of an amino group in ortho-position to the functional amino group. Of lesser importance are the variables I(diNH2 ), I(BiBr), MR5 , and the cross product log P × I (diNH2 ). It appears that the key factors differentiating between active and inactive compounds on the one hand, and governing potency within the group of active compounds are different. The most pronounced differences are with respect to the importance of hydrophobicity and the directionality of electronic effects. For female mouse carcinogenicity, the following discriminant function reclassifies 85.7% of the compounds correctly (class 1: 87.9%; class 2: 83.3%) and is of acceptable stability in cross validation (all compounds: 81.0%; class 1: 84.8%; class 2: 76.7%): w = −0.47I (NR) + 1.38 log P × I (monoNH2 ) +1.68 log P × I (diNH2 ) − 0.37I (An) (4.3.1) +0.33I (o-NH2 ) − 0.55MR5 −0.45I (BiBr)
(5)
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
w(mean, class 1) = −0.92, N1 = 33 w(mean, class 2) = 1.01,
N2 = 30
where I (NR) = 1 if the amino nitrogen is substituted. For male mouse discriminant function (6) is obtained: w = −1.96L(R) + 1.69B5 (R) − 0.83EHOMO +0.97ELUMO − 1.22I (An) + 0.73I (o-NH2 ) +0.59MR3 + 0.69MR5 + 0.77MR6 −0.76I (diNH2 ) + 1.09 log P × I (diNH2 ) −0.79I (BiBr)
(6)
w(mean, class 1) = −1.11, N1 = 25 w(mean, class 2) = 1.16,
N2 = 24
where B5 is the maximal width of the substituent at the amino group. It should be noted that the difference in sign of the average w values for the two classes in Eqs. (3)–(6) is only formal, and does not have any relevance on mechanisms. Discriminant function (6) shows a good reclassification rate (all compounds: 89.8%; class 1: 96.0%; class 2: 83.3%) and stability in cross validation (all compounds: 83.7%; class 1: 96.0%; class 2: 70.8%). The results for the mouse were similar to those found for the rat. Thus, hydrophobicity is a key factor for the gradation of the carcinogenic potency (Eqs. (1) and (2)) but only of small importance for yes/no activity (Eqs. (3)–(6)). The reverse is true for electronic properties (EHOMO, ELUMO), which show a minor effect for the gradation of potency, but a pronounced effect for yes/no activity. Eqs. (4)–(6) also point out the importance of steric (shape, size) factors for yes/no activity: for example, in all four equations the first term indicates that the probability of being non-carcinogenic increases with increasing length of the substituent L(R) or simply with the presence of a substituent I(NR) at the amino nitrogen. Hydrophobicity is a force involved in the absorption and transport of the drugs in the cells and organisms, as well as in the interaction between drugs and the metabolizing enzymes. The electronic parameters are measures of chemical reactivity, hence of the ability of undergoing metabolic transformations. It should be noted that the results of the QSAR analyses agree with the
195
notion that the aromatic amines require metabolic activation to become carcinogenic [1]. For amines and amides, this typically involves an initial N-oxidation to N-hydroxylamine and N-hydroxylamide. In particular, EHOMO is a parameter for oxidation reactions. For a more detailed discussion of the chemical structural parameters involved in the carcinogenicity of the aromatic amines and how they relate to the mechanisms of action, see [10]. Here, we would like to emphasize two points. First, the QSAR analyses help to give a firm and quantitative basis to many notions which were previously only qualitative: this happens through the establishment of equations where the contribution of each physical–chemical and structural parameter to the biological activity is quantitatively defined. Second, the QSAR analyses can provide entirely new information. For example, an original contribution of this study is the notion that the carcinogenic activity of the aromatic amines is a non-linear phenomenon. The non-carcinogenic amines are not simply “weakly active”. On the contrary there is a physical–chemical and structural border between the amines that can be carcinogenic and those that cannot. This border, or threshold, can be imagined as a surface in the space of the chemical descriptors: Eqs. (3)–(6) permit the calculation of the border as a combination of the different parameters present in the equations. 1.2. Comparative QSAR: learning from models for different end-points The importance and widespread use of aromatic amines has permitted the accumulation of a large amount of experimental data on the effects of these compounds in a range of chemical, physical and biological systems. This in turn has generated a variety of QSARs. This allows us to compare the QSARs obtained, to check their consistency and to highlight general trends. This approach is what Corwin Hansch has called “comparative QSAR”: “unless we can show that there are meaningful relationships among QSARs for different chemicals acting on the same or different systems we cannot call it a science. Establishing lateral correlations among QSARs is the only path to developing such a science” [12]. To this aim a database of QSARs (now including around 17,000 equations) and a software to interrogate the database (C-QSAR) has been developed at Pomona College
196
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
under the direction of Hansch, who generously permitted us to search for the existing QSARs on the aromatic amines. The enormous repertoire of equations contained in the C-QSAR database is divided into two main sections: chemical–biological interactions and physical–chemical reactions. A major point of interest in the development of the interrogation computerized system is the ability to compare QSARs corresponding to rather simple chemical reactions with those of biological systems. In many cases such comparisons were of definite value in understanding the more complex biological processes [12,13]. From the C-QSAR database we selected the QSARs which were of interest for this paper; they are listed in Appendix A. The high r fitting parameter of the equations should be remarked. The following is a discussion of the salient points of this comparison. A first comparison is with the QSARs for the mutagenic potency of the aromatic amines in Salmonella (Appendix A, Section A.1) (see also [14]). Interestingly, the QSARs for Salmonella mutagenic potency pointed to a pattern similar to that for the rodent carcinogenic potency. In both cases hydrophobicity (log P) is the major molecular determinant. Second in order of importance are the reactivity parameters, which are present in the equations for TA98 and TA100 Salmonella strains, as well as in the model for the mouse carcinogenicity (Eq. (1)). Thus, there is a high degree of qualitative similarity between the two sets of equations, hence between the most important molecular determinants of the mutagenic and carcinogenic activities. Eqs. (3)–(6) apply to the difference between carcinogenic and non-carcinogenic aromatic amines. No similar equations exist in the C-QSAR database for mutagenicity. However, in our previous studies [15] we found that also the difference between mutagenic and non-mutagenic amines is determined mainly by electronic and steric factors, and that hydrophobicity plays—if any—a minor role. Overall, this evidence represents a support for the similarity in mechanism of the aromatic amines in Salmonella and in rodents, and for their genotoxic mechanism of carcinogenicity. Appendix A, Section A.2 reports the QSARs for general toxicity of aromatic amine datasets. The range of systems and organisms is impressive: it includes single cells (both procaryotic and eukaryotic), cells
in culture, and animals. In spite of such a variety, the simplicity and homogeneity of these QSARs is equally impressive. The total number of QSAR equations is 36. Out of them, 20 equations are based exclusively on log P, and in 30 out of 36 equations log P is the first (most important) parameter. Only two equations do not contain log P. The comparison with the QSARs for the rodent carcinogenicity and the Salmonella mutagenicity points to a lower degree of complexity and to different underlying mechanisms of action of general toxicity. This result can be better understood by recalling the research of Koneman [16] on inert narcotic pollutants, that led to the so-called baseline toxicity concept: chemicals with toxicities in line with the baseline concept are classified as inert and are not interacting with specific receptors in an organism. They are presumed to interact with cellular membranes, producing an aspecific effect (narcosis) that can be parametrized by log P only. However, some classes of compounds are more toxic than predicted by the baseline concept, indicating the existence of additional specific components in the toxicity mechanisms. In these cases, the QSAR models contain additional parameters (electronic and/or steric) [5,17]. It appears that the QSARs for the general toxicity of the aromatic amines follow the above scheme: hydrophobicity is largely the main (or unique) determinant, thus, pointing mainly to an aspecific mode of action. It should be remarked that this evidence is in agreement with independent assignments of the aromatic amines to mode-of-action categories based on a range of experimental determinations [18]. Thus, the operation of comparing QSARs at different complexity levels and for different end-points reveals details of general scientific relevance. The picture changes again with the enzymatic reactions (Appendix A, Section A.3). Among the many types of enzymatic reactions contained in the C-QSAR database, we have selected those relevant for the mechanism of action of the aromatic amines in carcinogenicity and mutagenicity. The most relevant is the oxidation of the amino group, which is considered to be the first step in the main metabolic pathway [1]. As for general toxicity, the equations for enzymatic reactions are simpler than in the case of carcinogenicity and mutagenicity; however the parameters included are different from those predominant in general toxicity. It appears that the oxidation
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
QSARs rely systematically on σ (Hammett electronic descriptor), which is the first parameter even in the few cases in which it is complemented by π (hydrophobicity parameter for the substituent groups, whereas log P is a parameter for the entire molecule) or by steric parameters. The constant presence of σ in the equations is consistent with the fact that the oxidation reactions can take place only when the electronic characteristics of the amine group ensure sufficient reactivity, whereas π and the steric parameters may modulate the correct interaction between enzyme and substrate. Positive σ values represent electron withdrawal from the ring by the substituents; negative σ values indicate electron release to the ring. As reflected by the negative sign of the π coefficients in the QSAR models, the oxidation of the amino nitrogen is favored by electron releasing substituents on the ring ([2], chapter 2). This makes more electrons available both for the chemical process of oxidation and for the nucleophilic attack to the enzyme (which is a hypothesized step of the enzymatic oxidation). Appendix A, Section A.3 also shows the QSARs for p-hydroxylation. There is evidence that in addition to the cytochrome P450-mediated N-hydroxylation pathway, some anilines may induce tumors via the formation of a reactive quinoneimine metabolite after bioactivation by peroxidases [1]. This involves ring hydroxylation as first step. At odds with the QSARs for the N-hydroxylation, five out six QSARs for p-hydroxylation are based on log P. This can be explained by the fact that the process does not depend on the reactivity at some specific reaction center but on the general quality of the interaction between enzyme and substrate. Regarding the QSARs for the physical–chemical reactions, we selected those relative to general oxidations, as they are potentially relevant for the mechanism of action in carcinogenicity and mutagenicity (Appendix A, Section A.4). The picture is consistent with that for the enzymatic reactions: the electronic parameters (σ ) are the unique or main determinants of the reactions. The only exception are the oxidations involving halogen compounds. The halogens are characterized by inductive and resonance effects with opposite signs. The net balance between two opposite reactions accounts for the positive sign in the equations. Moreover, it should be remarked that most of the equations have negative
197
coefficients for σ , as in the case of the enzymatic oxidations. No hydrophobic parameter appears, and only steric parameters sometimes modulate the reaction. This short discussion on the comparative QSAR of the aromatic amines, and the consistencies found, confirm the strength of the scientific approach of lateral comparison of QSARs for the investigation of mechanisms and for providing generalizations. More specifically, the comparative QSAR analysis has highlighted the similarity in the mechanisms of mutagenicity and carcinogenicity of the aromatic amines, and the difference with the general toxicity mechanisms. In addition, the mechanistic coherence between chemical–biological reactions and their physical–chemical basis has been pointed out (see, for example, the similarity between enzymatic and physical–chemical oxidations). Obviously, there are limitations to the amount and quality of the knowledge that can be derived from the comparison of the QSARs. In the present case, these limitations largely depend on the type of information on which the comparison relies. For example, the bioassay experiments, whose results were used in our QSAR studies, were not designed specifically for QSAR modeling, but for the assessment of the risk posed by the individual aromatic amines. This means that the selection of the chemicals did not follow the criteria that maximize the ability to identify the contributions of the different chemical parameters to the activity (e.g. balanced selection of chemicals representative of the extent of the chemical parameters space, orthogonality of the chemical parameters) [2]. In addition, the QSAR models in Appendix A were generated by different authors based on different subsets of aromatic amines. The strength of the comparative approach is demonstrated by the fact that, in spite of this, a clear and informative general picture emerges. However, the discussion of the results should respect the imperfect nature of the data analyzed, and speculations should not go too far. For example, Eqs. (3)–(6) point to different models for rat and mouse, and also to differences between male and female mice. Before drawing conclusions on the underlying mechanisms based only on these four equations, further QSAR models relative to the carcinogenicity of other classes of chemicals should be examined.
198
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
1.3. QSAR models: implications for risk assessment The discussion in this paper has focused on the QSAR models for the rodent carcinogenicity of the aromatic amines. The QSARs for the yes/no carcinogenic activity and for the modulation of the carcinogenic potency have been summarized. It should be recalled that the rodent bioassay provides a large amount of information, namely: (a) yes/no carcinogenic activity of the tested compound; (b) the carcinogenic potency of those chemicals determined to be active; (c) information on the target organs. The yes/no activity is the most important information: all the human carcinogens were positive in the bioassay when adequately tested, and many human carcinogens were first indicted by positive carcinogenicity test results in rodents. Thus, the yes/no result from a rodent bioassay is very relevant to the human risk assessment [19–22]. The information on the potency is also important. There is evidence for a strong correlation: (a) between the potency ranking of the carcinogens in rat and mouse [23]; (b) between carcinogenic potency in humans estimated from epidemiologic data, and that estimated from the rodent bioassay [24]. The similar ranking of the carcinogenic potency in different species suggests that potency is an intrinsic property of a chemical carcinogen that is more directly derived from its chemical reactivity. Thus, both yes/no activity and carcinogenic potency are relevant for the extrapolation of animal data to assessing human risk. In this context, the QSARs generated in our laboratory can be used directly for estimating the carcinogenicity of non-heterocyclic aromatic amines for which the experimental carcinogenicity data are not available. A two-step prediction of carcinogenicity of aromatic amines is possible: (1) estimation of yes/no activity from the Eqs. (3)–(6); (2) if the answer from step 1 is yes, then prediction of the degree of potency from Eqs. (1) and (2). Thus, the QSAR models can contribute to the: synthesis of safer chemicals; estimation of the risk posed by aromatic amines in the environment; setting priorities for experimentation; reducing the use of animals. Any new animal bioassay results will, in turn, allow for further refinement and validation of these models. The pattern of target organs is the third piece of evidence provided from the rodent bioassay. The tumor types induced appear to be highly variable
from species to species [23], and also vary with the condition of use (e.g. age of the host, dose and route of administration) of the carcinogen [25]. For example, we calculated that in the present aromatic amines database the overlap of tumor types induced in rat and mouse by the individual chemicals was a mere 18%. Thus, this information is not suitable as a basis for extrapolation to human risk. It is interesting that, whereas we obtained quite satisfactory QSAR models for the yes/no activity and for carcinogenic potency of the aromatic amines, we were not successful in modeling the carcinogenic potency for the individual organs (our unpublished results). A parallel line of evidence is that the aromatic amines, like other classes of chemical carcinogens, do not induce in animals one or a few specific patterns of tumor types, but can induce the widest spectrum of tumors [26]. This indicates that the species differences in the tumor profiles may result not only from differences in targeted reactions of the ultimate carcinogens, but also in the myriad of events that mediate and surround these reactions (e.g. toxicokinetics): this makes particularly difficult the modeling of tumor profiles and—what is most important—makes the tumor profiles not particularly useful for species extrapolation in risk assessment. The QSAR models obtained for the carcinogenicity of the aromatic amines are all statistically robust and provide good fits to existing data. Both QSARs for the carcinogenic potency (Eqs. (1) and (2)) have r 2 = 0.8; the models for the separation of carcinogens from non-carcinogens (Eqs. (3)–(6)) have discriminant accuracy in the range 80–90%. This is a very encouraging result, and it looks even better when one takes into account that we modeled animal data, which have an intrinsic experimental variability and— most important—were generated in different laboratories in different periods. Approximately two-thirds of the chemicals used to build our QSAR models were generated by the NCI/NTP experimentation with well controlled protocols, whereas the remaining one-third had different origins [10]. In light of the critical importance of the experimental protocols in the rodent bioassay results, the goodness of fit of the QSARs found is quite high, and deserves some comments. A first comment comes from the comparison of the above QSARs with the limited accuracy shown by various systems devised in the recent years for the prediction of chemical carcinogenicity. Chemical
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
carcinogenicity has been the target of a more diverse array of structure-based predictive modeling efforts than perhaps any other toxicological end-point. This has been encouraged, in large part, by the historical successes of the structure–activity paradigm, particularly when guided by mechanistic considerations applied to well defined classes of chemicals and end-points. In addition, it has been encouraged by the prominent role that a positive rodent carcinogenicity result plays in driving regulatory action on chemicals, and by the enormous investment of time, resources and animal lives required for testing a single chemical. Thus, many “all-purposes” models, hopefully aimed at predicting the carcinogenicity of any kind of chemical, irrespective of the chemical class, have been devised [27,28]. In the last decade, unique opportunities to objectively assess the relative performance of model systems for carcinogenicity prediction have been provided by two prospective prediction toxicity exercises held under the aegis of the US National Toxicology Program (NTP). Both exercises invited the modeling community to submit predictions of rodent carcinogenicity for chemicals that were in the process of being bioassayed by the NTP: the predictions were published before the bioassay results were known and made public [29,30]. The prediction systems had very different rational basis and implementation (biologically-, structure–activity-based, computerized systems, human experts, etc.). Detailed analyses of the results of the comparative exercises are in [8,31,32]. It appeared that the best performance in prediction was around 70%, but many prediction systems had very low performance. This is quite inferior to the 80–90% accuracy obtained when modeling the carcinogenicity of the aromatic amines. This gap makes sense. The all-purpose models attempt to model at the same time many different mechanisms of action (each class of chemical carcinogens has its specific
199
mechanism); moreover, they rely on the database of bioassay results that is well represented only for a few classes of carcinogens (e.g. the aromatic amines) and very under-represented for most classes. Thus, two sources of difficulty combine together: the difficulty in modeling simultaneously many mechanisms adds to the limited database from where the predictive approaches can learn. This explains the limited performance of the “all-purpose” approaches. On the other hand, the application of the powerful QSAR methods to the mechanistically well defined and sufficiently represented class of the aromatic amines permitted the attainment of a much higher modeling accuracy. Thus, the results obtained with the aromatic amines can be considered as an estimate of the best performance that can be reached in modeling the bioassay data. As final consideration, it should be remarked that the successful modeling of in vivo data is a strong argument in favor of the reliability of the traditional rodent carcinogenicity assay. In spite of the different sources (laboratories) of the rodent data used in our work, the data were coherent enough to allow the QSAR modeling to highlight with precision the underlying chemical determinants of the carcinogenicity of the aromatic amines. The QSAR models were robust from a statistical point of view, and were coherent and meaningful from a mechanistic point of view. Decades of experience in the field of QSAR indicates that this could not have happened without a solid basis of good quality data.
Acknowledgements Prof. Corwin Hansch is gratefully acknowledged for generously providing us the access to the C-QSAR database. We thank Dr. Marina Cotta-Ramusino for her continued interest and for the helpful discussions.
Appendix A A.1. Mutagenicity Salmonella typhimurium TA98 + S9 log TA98 = 1.08 log P + 1.28HOMO − 0.73LUMO + 1.46IL + 7.20 TA100 + S9 log TA100 = 0.92 log P + 1.17HOMO − 1.18LUMO + 7.35
n = 88; s = 0.860; r = 0.898; omit = 7 n = 67; s = 0.708; r = 0.877; omit = 6
200
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
201
202
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
203
204
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
References [1] Y.T. Woo, D.Y. Lai, Aromatic amino and nitro-amino compounds and their halogenated derivatives, in: E. Bingham, B. Cohrssen, C.H. Powell (Eds.), Patty’s Toxicology, Wiley, New York, 2001, pp. 969–1105. [2] C. Hansch, A. Leo, Exploring QSAR. Part 1. Fundamentals and Applications in Chemistry and Biology, American Chemical Society, Washington, DC, 1995. [3] T. Fujita, Recent success stories leading to commercializable bioactive compounds with the aid of traditional QSAR procedures, Q. Struct. Act. Relat. 16 (1997) 107–112. [4] C. Hansch, D. Kim, A.J. Leo, E. Novellino, C. Silipo, A. Vittoria, Toward a quantitative comparative toxicology of organic compounds, Crit. Rev. Toxicol. 19 (1989) 185– 226. [5] H.J.M. Verhaar, J. Solbe, J. Speksnijder, C.J. van Leeuwen, J.L.M. Hermens, Classifying environmental pollutants. Part 3. External validation of the classification system, Chemosphere 40 (2000) 875–883. [6] H. Hong, W. Tong, H. Fang, L.M. Shi, Q. Xie, J. Wu, R. Perkins, J.D. Walker, W. Branham, D.M. Sheehan, Prediction of estrogen-receptor binding for 58,000 chemicals using an integrated system of a tree-based model with structural alerts, Environ. Health Perspect. 110 (2002) 29–36. [7] R. Benigni, A. Giuliani, Quantitative structure–activity relationship (QSAR) studies of mutagens and carcinogens, Med. Res. Rev. 16 (1996) 267–284. [8] A.M. Richard, R. Benigni, AI and SAR approaches for predicting chemical carcinogenicity: survey and status report, SAR QSAR Environ. Res. 13 (1) (2002) 1–19. [9] R. Benigni, A. Giuliani, R. Franke, A. Gruska, Quantitative structure–activity relationships of mutagenic and carcinogenic aromatic amines, Chem. Rev. 100 (2000) 3697–3714. [10] R. Franke, A. Gruska, A. Giuliani, R. Benigni, Prediction of rodent carcinogenicity of aromatic amines: a quantitative structure–activity relationships model, Carcinogenesis 22 (2001) 1561–1571. [11] L.S. Gold, N.B. Manley, T.H. Slone, L. Rohrbach, Supplement to the carcinogenic potency database (CPDB): results of animal bioassays published in the general literature in 1993–1994 and by the National Toxicology Program in 1995–1996, Environ. Health Perspect. Suppl. 107 (Suppl. 4) (1999) 527–600. [12] C. Hansch, D. Hoekman, H. Gao, Comparative QSAR: toward a deeper understanding of chemical–biological interactions, Chem. Rev. 96 (1996) 1045–1075. [13] C. Hansch, H. Gao, Comparative QSAR: radical reactions of benzene derivatives in chemistry and biology, Chem. Rev. 97 (1997) 2995–3060. [14] A.K. Debnath, G. Debnath, A.J. Shusterman, C. Hansch, A QSAR investigation of the role of hydrophobicity in regulating mutagenicity in the Ames test. Part 1. Mutagenicity of aromatic and heteroaromatic amines in Salmonella typhimurium TA98 and TA100, Environ. Mol. Mutagen. 19 (1992) 37–52.
205
[15] R. Benigni, L. Passerini, G. Gallo, F. Giorgi, M. CottaRamusino, QSAR models for discriminating between mutagenic and non-mutagenic aromatic and heteroaromatic amines, Environ. Mol. Mutagen. 32 (1998) 75–83. [16] H. Koneman, Quantitative structure–activity relationships in fish toxicity studies. I. Relationships for 50 industrial pollutants, Toxicology 19 (1981) 209–221. [17] J.L.M. Hermens, Quantitative structure–activity relationships of environmental pollutants, in: O. Hutzinger (Ed.), Handbook of Environmental Chemistry, Springer, Berlin, 1989, pp. 111–162. [18] C.L. Russom, S.P. Bradbury, S.J. Broderius, D.E. Hammermeister, R.A. Drummond, Predicting modes of toxic action from chemical structure: acute toxicity in fathead minnow (Pimephales promelas), Environ. Toxicol. Chem. 16 (1997) 948–967. [19] J. Huff, Value, validity and historical development of carcinogenesis studies for predicting and confirming carcinogenic risks to humans, in: K.T. Kitchin (Ed.), Carcinogenicity: Testing, Predicting, and Interpreting Chemical Effects, Marcel Dekker, New York, 1999, pp. 21–123. [20] V.A. Fung, J.C. Barrett, J. Huff, The carcinogenesis biossay in perspective: application in identifying human cancer hazards, Environ. Health Perspect. 103 (1995) 680–683. [21] L. Tomatis, J. Huff, I. Hertz-Picciotto, D.P. Sandler, J. Bucher, P. Boffetta, O. Axelson, A. Blair, J. Taylor, L. Stayner, J.C. Barrett, Avoided and avoidable risks of cancer, Carcinogenesis 18 (1997) 97–105. [22] L. Tomatis, J. Huff, Evolution of cancer etiology and primary prevention, Environ. Health Perspect. 109 (2001) 5–7. [23] R. Benigni, A. Giuliani, Tumor profiles and carcinogenic potency in rodents and humans: value for cancer risk assessment, Environ. Carcinogen. Ecotoxicol. Rev. C17 (1999) 45–67. [24] B.C. Allen, K.S. Crump, A.M. Shipp, Correlation between carcinogenic potency of chemicals in animals and humans, Risk Anal. 8 (1988) 531–544. [25] T.J. Bucci, Profiles of induced tumors in animals, Toxicol. Pathol. 13 (1985) 105–109. [26] R. Benigni, A. Pino, Profiles of chemically-induced tumors in rodents: quantitative relationships, Mutat. Res. 421 (1998) 93–107. [27] J.C. Dearden, M.D. Barratt, R. Benigni, D.W. Bristol, R.D. Combes, M.T.D. Cronin, P.N. Judson, M.P. Payne, A.M. Richard, M. Tichy, A.P. Worth, J.J. Yourick, The development and validation of expert systems for predicting toxicity, Alternat. Lab. Anim. (ATLA) 25 (1997) 223–252. [28] R. Benigni, A.M. Richard, Quantitative structure-based modeling applied to characterization and prediction of chemical toxicity, Methods 14 (1998) 264–276. [29] R.W. Tennant, J. Spalding, S. Stasiewicz, J. Ashby, Prediction of the outcome of rodent carcinogenicity bioassays currently being conducted on 44 chemicals by the National Toxicology Program, Mutagenesis 5 (1990) 3–14. [30] D.W. Bristol, J.T. Wachsman, A. Greenwell, The NIEHS predictive—Toxicology Evaluation Project: chemcarcinogenicity biossay, Environ. Health Perspect. 104 (1996) 1001–1010.
206
R. Benigni, L. Passerini / Mutation Research 511 (2002) 191–206
[31] R. Benigni, The first US National Toxicology Program exercise on the prediction of rodent carcinogenicity: definitive results, Mutat. Res. 387 (1997) 35–45. [32] R. Benigni, Structure–activity relationships in mutagenesis and carcinogenesis, in: M. Balls, A.M. van Zeller, M.E. Halder (Eds.), Progress in the Reduction, Refinement
and Replacement of Animal Experimentation, Elsevier, Amsterdam, 2000, pp. 469–478. [33] C. Hansch, A. Leo, D. Hoekman, Exploring QSAR. Part 2 Hydrophobic, Electronic and Steric Constants, American Chemical Society, Washington, DC, 1995.