quantitative structure–activity (qsar) relationships of mutagenic aromatic and heterocyclic amines

10
Ž . Mutation Research 376 1997 87–96 ž / Quantitative structure–activity QSAR relationships of mutagenic aromatic and heterocyclic amines F.T. Hatch a, ) , M.E. Colvin b a Biology and Biotechnology Research Program, Lawrence LiÕermore National Laboratory, LiÕermore, CA 94551, USA b Center for Computational Engineering, Sandia National Laboratory, LiÕermore, CA 94551, USA Abstract We extended our previous studies of mutagenicrcarcinogenic heterocyclic aromatic amines formed during the cooking of foods to 66 aromatic and 99 heterocyclic amines for which mutagenic potency data are available. The amines require activation by enzymes to form metabolites reactive with DNA and exhibit an enormous range of potency as frameshift mutagens in the AmesrSalmonella assay. To ascertain factors that might influence potency, structural features and quantum Ž mechanical parameters calculated by the Huckel method and, for a subset of 20 amines, by semi-empirical AM1, and ab ¨ . Ž. initio methods were analyzed by multiple linear regression. The major findings were: 1 earlier findings on cooked food Ž. mutagens and their synthetic congeners can be extended to other amines; 2 mutagenic potency is directly related to the Ž . Ž number of fused aromatic rings size of the aromatic system , the number of ring nitrogen atoms participation of lone . Ž. electron pairs in the pi-cloud , and presence of a methyl substituent on a ring nitrogen; 3 potency is inversely related to the Ž . Ž . energy of the lowest unoccupied molecular orbital LUMO of the parent amine. Ford and Griffin 1992 and Sabbioni and Ž . Ž Wild 1992 showed that the LUMO energy of the derived nitrenium ion is closely related to its stability calculated with . reference to aniline . Increased stability has been hypothesized to enhance the probability of adduct formation with DNA by avoiding detoxifying side reactions and increasing the lifetime of the ion. In the large heterogeneous series of amines in our Ž . present study the Huckel method energy of the highest occupied molecular orbital HOMO , rather than the LUMO energy, ¨ of the nitrenium ion was marginally related to the potency of the parent amine. However, in the selected subset of 20 amines with ab initio calculation, the LUMO energy of the ion confirmed the previous reports. The contribution of quantum chemical factors to mechanistic insight on the mutagenicity and carcinogenicity of aromatic and heterocyclic amines is still under development. Keywords: Cooked food mutagen; Molecular orbital; Quantum mechanics; Huckel; Semi-empirical; Ab initio ¨ ) Corresponding author. Present address: 27 Pease Road, Ž . Meredith, NH 03253-5506, USA. Tel.: q1 603 279-5142; Fax: Ž . q1 603 279-5057; e-mail: [email protected] 1. Introduction Aromatic amines have had a bad image for a Ž . century, since Rehn 1895 described bladder cancers in workers in a Swiss dye factory. The first thorough Ž . epidemiologic study was by Case et al. 1954 in the United Kingdom. The most hazardous chemicals 0027-5107r97r$17.00 Copyright q 1997 Elsevier Science B.V. All rights reserved. Ž . PII S0027-5107 97 00029-8

Upload: ft-hatch

Post on 15-Sep-2016

214 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Quantitative structure–activity (QSAR) relationships of mutagenic aromatic and heterocyclic amines

Ž .Mutation Research 376 1997 87–96

ž /Quantitative structure–activity QSAR relationships ofmutagenic aromatic and heterocyclic amines

F.T. Hatch a,), M.E. Colvin b

a Biology and Biotechnology Research Program, Lawrence LiÕermore National Laboratory, LiÕermore, CA 94551, USAb Center for Computational Engineering, Sandia National Laboratory, LiÕermore, CA 94551, USA

Abstract

We extended our previous studies of mutagenicrcarcinogenic heterocyclic aromatic amines formed during the cooking offoods to 66 aromatic and 99 heterocyclic amines for which mutagenic potency data are available. The amines requireactivation by enzymes to form metabolites reactive with DNA and exhibit an enormous range of potency as frameshiftmutagens in the AmesrSalmonella assay. To ascertain factors that might influence potency, structural features and quantum

Žmechanical parameters calculated by the Huckel method and, for a subset of 20 amines, by semi-empirical AM1, and ab¨. Ž .initio methods were analyzed by multiple linear regression. The major findings were: 1 earlier findings on cooked food

Ž .mutagens and their synthetic congeners can be extended to other amines; 2 mutagenic potency is directly related to theŽ . Žnumber of fused aromatic rings size of the aromatic system , the number of ring nitrogen atoms participation of lone

. Ž .electron pairs in the pi-cloud , and presence of a methyl substituent on a ring nitrogen; 3 potency is inversely related to theŽ . Ž .energy of the lowest unoccupied molecular orbital LUMO of the parent amine. Ford and Griffin 1992 and Sabbioni and

Ž . ŽWild 1992 showed that the LUMO energy of the derived nitrenium ion is closely related to its stability calculated with.reference to aniline . Increased stability has been hypothesized to enhance the probability of adduct formation with DNA by

avoiding detoxifying side reactions and increasing the lifetime of the ion. In the large heterogeneous series of amines in ourŽ .present study the Huckel method energy of the highest occupied molecular orbital HOMO , rather than the LUMO energy,¨

of the nitrenium ion was marginally related to the potency of the parent amine. However, in the selected subset of 20 amineswith ab initio calculation, the LUMO energy of the ion confirmed the previous reports. The contribution of quantumchemical factors to mechanistic insight on the mutagenicity and carcinogenicity of aromatic and heterocyclic amines is stillunder development.

Keywords: Cooked food mutagen; Molecular orbital; Quantum mechanics; Huckel; Semi-empirical; Ab initio¨

) Corresponding author. Present address: 27 Pease Road,Ž .Meredith, NH 03253-5506, USA. Tel.: q1 603 279-5142; Fax:

Ž .q1 603 279-5057; e-mail: [email protected]

1. Introduction

Aromatic amines have had a bad image for aŽ .century, since Rehn 1895 described bladder cancers

in workers in a Swiss dye factory. The first thoroughŽ .epidemiologic study was by Case et al. 1954 in the

United Kingdom. The most hazardous chemicals

0027-5107r97r$17.00 Copyright q 1997 Elsevier Science B.V. All rights reserved.Ž .PII S0027-5107 97 00029-8

Page 2: Quantitative structure–activity (QSAR) relationships of mutagenic aromatic and heterocyclic amines

( )F.T. Hatch, M.E. ColÕinrMutation Research 376 1997 87–9688

were 2-naphthylamine and benzidine. Recent studieshave shown broad evidence of genotoxicity in many

Ž .test systems Adamson et al., 1995 ; and a newsreport revealed excess bladder cancer in workers

Žmanufacturing a benzidine substitute Union Leader,.1995 .

Over the past 15 years our laboratory has studiedextensively a series of heterocyclic aromatic aminesthat are produced during cooking or other heat pro-cessing of protein-rich foods or pyrolysis of amino

Ž .acids Hatch et al., 1991, 1992 . These compoundsare bacterial promutagens that, when activated enzy-matically, are mutagenic in Salmonella frameshift-

Ž .sensitive strains TA1538 and TA98 with a potencyrange covering more than six orders of magnitude.

Ž .Quantitative structure–activity QSAR relationshipsof the heat-derived mutagens and their synthetic

Ž .congeners 61 chemicals , and of a subset of theseplus other aromatic amines that are both frameshiftmutagens and suspected or confirmed animal or hu-

Ž .man carcinogens 34 chemicals were reported.We have now compiled a much broader database

Žof 66 aromatic and 99 heterocyclic amines hereafter.noted as AHA165 , for which mutagenic potencies

over nine orders of magnitude have been determinedŽ .Table 1 . QSAR relationships, which now includeboth structural factors and Huckel quantum mechani-¨cal calculations, are presented for AHA165 and for a

Ž .selection of 20 of these amines noted as SEL20spanning the range of mutagenic potency at half-logintervals. SEL20 preserves much of the structuralheterogeneity of the full series but permits addingthe more elaborate semi-empirical and ab initioquantum mechanical calculations.

In our previous studies structural molecular fea-tures that appeared to be associated with mutagenic

potency were identified. In general these involvedthe number of fused rings, methylation on ring nitro-gen or carbon atoms, presence and number of ringnitrogen atoms, and number and location of the

Ž .exocyclic amine group s on the ring system. Interms of organic chemical mechanism we hypothe-sized that the structural factors are related to the sizeof the approximately planar aromatic system, thepresence of lone electron pairs on ring nitrogenatoms, the blocking by substitution of potential sitesfor deactivation reactions or diversion from the mu-tagenic pathway, and either steric or electronic influ-ences on electron delocalization from the aromaticsystem to the exocyclic nitrogen atom.

The objective of the QSAR studies is to gaininsight into the mechanism whereby these aminescause mutations. The most common mechanism ofbacterial mutagenesis by these amines is frameshifts,which may relate to the presumed ability of theseplanar or near-planar aromatic compounds to interca-late into DNA. The exocyclic amino group is re-quired; compounds containing the cyclic nuclei butlacking the amino group are either mutagenicallyinactive or active through some other mechanism.The frameshift-sensitive strains of Salmonella areextremely sensitive to many of these chemicals andthis has greatly facilitated their isolation and purifi-cation from complex media such as cooked foodsŽ .Knize et al., 1987 . Further objectives are to deter-mine the generality of the QSAR relationships shownin our recent study of a rather homogeneous series of16 aminoimidazo-azaarenes related to mutagens

Ž .found in cooked foods Hatch et al., 1996 , and toconfirm the good correlations among three quantummechanics-based computation methods of varied so-phistication and computational intensity.

Table 1Structural types of amines

Type Compounds Examples AHA165 SEL20

aa Aromatic amines 1–2 ring 33 3sep Linked ring aromatics biphenyl, benzidine 10 1paa Polycyclic aromatic amines 3–4 ring 23 2ha Heterocyclic amines quinoline, carbazole, phenazine 17 1ap Aminopyridines, carbolines glu-P, trp-P groups 24 4ai2 Aminoimidazo-azaarenes 2 fused rings: benzimidazole, imidazopyridine 22 2ai3 Aminoimidazo-azaarenes 3 fused rings: iq, iqx groups 36 7

Page 3: Quantitative structure–activity (QSAR) relationships of mutagenic aromatic and heterocyclic amines

( )F.T. Hatch, M.E. ColÕinrMutation Research 376 1997 87–96 89

2. Methods

2.1. AbbreÕiations and definition of parameters

Abbreviations and definitions of the propertiescalculated, together with their units, are given inTable 2.

2.2. Source of amines

The chemicals selected for this database wereŽaromatic or heterocyclic primary amines unsub-

.stituted on the amino nitrogen atom , except for2-acetlyaminofluorene. Ring substitutions were lim-ited to amino, methyl, methoxy, and in the case ofseparated rings, methylene groups.

2.3. Database and calculations

The computer system used was an IBM PSr1DX2-40 with Microsoft DOS 5 and Windows 3.1

Ž .operating systems Microsoft Corp., Redmond, WA .All data were maintained in Paradox ver. 5 for

Ž .Windows Borland International, Scotts Valley, CA .

The ab initio quantum mechanical calculations wereperformed on an 1840 processor Intel Paragon paral-lel computer and an 8-processor Silicon GraphicsPower Challenge shared-memory parallel computer.

2.4. Mutagenic potency

The mutagenicity data for these chemicals wereobtained mainly from a series of reports of studiessupporting the National Toxicology Program. Addi-tional sources were published and unpublished datafrom Lawrence Livermore National Laboratory andreports by de Meester, Ford and Herman, Debnath etal., and Vikse et al. Data and references are given in

Ž .the Appendix see footnote 1 . The logarithm ofrevertants in AmesrSalmonella strains TA98 or

Ž .TA1538 per nanomole of chemical Log MP wastaken from data in our laboratory where availableand otherwise from literature data as specified in

Ž .Table A1. Mutagenicity data strain TA98 for sev-eral chemicals were taken from multiple sources andthe median was used in our calculations. In ourlaboratory and usually elsewhere mutagenic potencywas calculated from the linear portion of dose–re-

Table 2Abbreviations and definitions of parameters used in the tables and text

Abbrev. Meaning

AHA165 Series of 66 aromatic and 99 heterocyclic aminesSEL20 Twenty amines selected from AHA165 at half-log intervals of potencyLog MP Log of revertants in AmesrSalmonella strain TA98 or TA1538 per nanomoleFus Number of fused aromatic ringsNme Presence of methyl substituent on an endocyclic nitrogenHet Number of endocyclic heteroatoms

Ž .LUMOab Hartree Fock ab initio LUMO energy of parent amine in electron voltsŽ .LUMOse AM1 semiempirical LUMO energy of parent amine in electron volts

LUMOhu Huckel LUMO energy of parent amine in beta units¨Ž .HOMOab Hartree Fock ab initio HOMO energy of parent amine in electron volts

Ž .HOMOse AM1 semiempirical HOMO energy of parent amine in electron voltsHOMOhu Huckel HOMO energy of parent amine in beta units¨

Ž .LUMOabnh Hartree Fock ab initio LUMO energy of nitrenium ion in electron voltsŽ .LUMOsenh AM1 semiempirical LUMO energy of nitrenium ion in electron volts

LUMOhunh Huckel LUMO energy of nitrenium ion in beta units¨Ž .HOMOabnh Hartree Fock ab initio HOMO energy of nitrenium ion in electron volts

Ž .HOMOsenh AM1 semiempirical HOMO energy of nitrenium ion in electron voltsHOMOhunh Huckel HOMO energy of nitrenium ion in beta units¨

Ž .nchrgenh NPA atomic charge on exocyclic nitrogen of nitrenium ion electronsŽ .DDH ab Ab initio heat of reaction for nitrenium-aniline hydride shift kcalrmolef

Ž . Ž .dipoleab Hartree-Fock ab initio dipole moment of parent amine Debye

Page 4: Quantitative structure–activity (QSAR) relationships of mutagenic aromatic and heterocyclic amines

( )F.T. Hatch, M.E. ColÕinrMutation Research 376 1997 87–9690

Ž .sponse curves Moore and Felton, 1983 . Becausethis large series of chemicals contained many weakmutagens, raw data were examined in the originalsources without regard for the authors’ opinion aboutpositivernegative results. If any combination of doserange and S9 speciesrconcentration appeared to givesome linearity at the low end of the concentrationcurve, the potency calculated from the slope wasincluded in the database. The spectra of mutagenicpotencies for each type-class of amine are presentedin Fig. 1.

2.5. Structural factors

Structural factors that appeared to differ mostbetween compounds of low and high mutagenic po-

Ž .tency were chosen Hatch et al., 1991 . This selec-tion method does not exhaustively test all possible

structural factors, but in our earlier study it yieldedgood predictors of Log MP. The following structural

Ž .factors were examined. a Size of the aromatic ringŽ .system Fus : the number of fused rings from one to

Ž . Ž .four; b N-methyl Nme : this factor is based onŽ . Ž .absence 0 or presence 1 of a methyl substituent

Ž .on an imidazole ring nitrogen in Ring 1. c No. ofŽ . Ž .ring heteroatoms Het from zero to four; d C-Ž . Ž .methyl substitution on Ring 2 or 3 ; e Equatorial

Ž .or axial orientation of the exocyclic amino group; fNumber of exocyclic amino groups. Factors d, e andf never entered a significant multiple regressionmodel and are not discussed further.

2.6. Quantum mechanical calculations

In all cases identical calculations were carried outon the parent amines and on the derived nitrenium

Fig. 1. Frequency distributions of mutagenic potency for each type of aromatic and heterocyclic amines.

Page 5: Quantitative structure–activity (QSAR) relationships of mutagenic aromatic and heterocyclic amines

( )F.T. Hatch, M.E. ColÕinrMutation Research 376 1997 87–96 91

ions. Although nitrenium ions are considered bysome authors to be the ultimate mutagenic agents for

Žreaction with DNA Miller and Miller, 1981; Kad-.lubar, 1987 , the experimental evidence for this is

Ž . Ž .weak Ford and Griffin, 1992 . Novak et al. 1993have shown that nitrenium ions derived from 4-aminobiphenyl or its acetyl derivative may have along enough solution lifetime that reaction with DNAbases is credible. We believe, however, that theseions may be suitable models for the final reactionstep in DNA binding, even if they are not the actual

Žagent. Frontier orbital energies highest occupieds.HOMO and lowest unoccupiedsLUMO were cal-

culated by the Huckel method. For the SEL20 amines¨frontier orbital energies, charge on the exocyclicnitrogen atom, and dipole moment were calculatedusing ab initio and semi-empirical AM1 methods.The ab initio heats of formation were used to calcu-late the DDH for the exchange of hydride withf

aniline:

w x qMutagen yNH qPhenylyNH2

w x qs Mutagen yNH qPhenylyNH 2

Ž .following the method of Ford and Griffin 1992 .This provides a measure of the relative stabilities ofnitrenium ions.

The Huckel molecular orbital calculations were¨Žperformed with HMO, ver. 2.0, Trinity Software,

.Campton NH . Structures were drawn on the screenand arbitrary geometry was assigned by the programusing Streitweiser parameters. Note that the Huckel¨method depends only on the connectivity of theatoms.

The semi-empirical calculations were performedŽwith HyperChem, ver. 4.0 Hypercube, Inc., Water-

.loo, Ontario, Canada . Starting with hand-sketchedstructures, optimization was performed initially by

A: LogMPs1.52 fusedq1.19 nmethylq0.31 heteroy3.11;LogMP s5.31 LUMOhuy7.79 HOMOhunhq0.507 heteroq10.85.B: LogMPs1.85 fusedq2.70 nmethyly4.26;LogMPs8.54 LUMOhuy5.96 HOMOhunhq2.14 nmethylq11.60;LogMPsy1.95 LUMOabq1.45 LUMOabnhq0.89 dipolabq12.19.

Table 3Multiple regression models

A: Multiple regression models for AHA165Structural factors Dependent variable: LogMP

Model Predictor variables

Fus Nme Het

Coefficients 1.43 1.10 0.28Pcoeff. 0.000 0.000 0.003Variance fractions 50.8% 10.1% 1.5%

2Adjusted R 62.4%Fmodel 91.7Pmodel 0.0000

Q.M. and structural factorsLUMOhu HOMO-hunh Het

Coefficients 5.31 y7.79 0.507Pcoeff. 0.000 0.000 0.000Variance Fractions 40.3% 11.1% 8.6%

2Adjusted R 59.9%Fmodel 82.7Pmodel 0.0000

B: Multiple regression models for SEL20

Structural factors Dependent variable: LogMP

Model Predictor variables

Fus Nme

Coefficient 1.85 2.7Pcoeff. 0 0.001Variance fraction 56.0% 18.8%

2Adjusted R 74.8%Fmodel 29.2Pmodel 0.0000

Structural and Huckel Q.M. factors¨LUMOhu Nme HOMOhunh

Coefficients 8.45 2.14 y5.96Pcoeff. 0.000 0.005 0.070Variance fract. 66.8% 10.0% 3.3%

2Adjusted R 80.0%Fmodel 26.4Pmodel 0.0000

Ab initio Q.M. factorsLUMOab LUMOabnh dipoleab

Coefficients y1.95 1.45 0.89Pcoeff. 0.002 0.052 0.001Variance fract. 55.8% 21.7% 3.7%

2Adjusted R 81.2%Fmodel 28.4Pmodel 0.0000

Page 6: Quantitative structure–activity (QSAR) relationships of mutagenic aromatic and heterocyclic amines

( )F.T. Hatch, M.E. ColÕinrMutation Research 376 1997 87–9692

molecular mechanics, MM2, and then by the AM1variant. After optimization, single point calculationsof orbital data were done with AM1. Solvation ef-fects were not included. Of the two rotamers of theexocyclic N-H bond in the nitrenium ions, one wasselected in which the bond was in the least hindered

Ž .position Sabbioni and Wild, 1992 .The ab initio quantum mechanical calculations

were performed using a custom-written parallelŽHartree-Fock energy and gradients program Colvin

. Ž .et al., 1993 and Gaussian 92 Frisch et al., 1992 .All of the SEL20 parent compounds and their nitre-nium derivatives were optimized at the Hartree-Focklevel of theory using a 6-31G ) basis set. Analyticalfrequencies were calculated at the optimized geome-tries to verify that these were true minima. HOMOand LUMO energies, dipole moments, and Natural

Ž . ŽPopulation Analysis NPA charges Reed et al.,.1985 were calculated with the Hartree-Fockr6-

31G ) wavefunction.

2.7. Statistical analysis

Stepwise multiple linear regression models werecalculated with Log MP as the dependent variable

Ž .and groups of predictor independent variables. Theprogram used for regression and all other statistics

Žwas Small STATA for Windows Stata Corp., Col-.lege Station, TX 77840 . Forward stepwise regres-

sion incorporates independent variables into modelsin order of their ability to reduce the residual sum ofsquares. In addition to the multiple regression coeffi-cients for each predictor, the results include thefollowing statistical parameters: Pcoeff. which is thestatistical significance of each coefficient; the coeffi-

Ž 2 .cient of determination Adj R for the model as awhole, adjusted for the number of predictor vari-ables; Fmodel which is the ratio of the varianceestimate due to regression to the residual varianceestimate; Pmodel which is the overall statistical sig-nificance of the model derived from the probabilitydistribution of F, i.e., the probability that the nullhypothesis is true. Models were calculated stepwise;with the addition of each new predictor variable, theincrement in Adj R2 is the variance fraction whichindicates the relative importance of the added predic-tor.

Linear interrelationships between certain pairs of

Table 4Predictor variables for dependent variable log MP

Major Minor

Structural factorsSEL20 Fus NmeAHA165 Fus Nme, Het

Huckel Q. M. factors¨SEL20 LUMOhu HOMOhunh?AHA165 LUMOhu HOMOhunh?

Ab initio Q.M. factorsSEL20 LUMOab LUMOabnh, dipoleab

structural and quantum mechanical factors are notuncommon in these datasets. Allowing a pair ofinterdependent predictors to enter a regression modelresults in multicollinearity and leads to a faulty

Ž .model Glantz and Slinker, 1990 . Thus, no morethan one factor of such a pair was included in amodel. A discussion of the defenses against multi-

Žcollinearity is found in our previous paper Hatch et.al., 1996 .

3. Results

The values of the qualitative structural and quan-tum mechanical factors described in Section 2 arelisted in an Appendix for AHA165 and SEL20 1.The structural factors represent the size of the aro-matic ring system, the presence of a methyl sub-

Ž .stituent if any on a nitrogen atom of an imidazolering, and the number of ring nitrogen atoms. Thesefactors were previously shown to be correlated withthe mutagenic potencies of a congeneric series of

Ž .aminoimidazo-azaarenes Hatch et al., 1991 . Theelectronic and energetic properties for parent aminesand nitrenium ions were calculated by the Huckel¨method for AHA165 and by the Huckel, AM1 and¨ab initio methods for SEL20.

1 The compound-specific data could not be published with thispaper. An Appendix is available from Dr. Hatch in hard copy or

Ž .3.5" disk Word Perfect for Windows , or from Dr. Colvin on theW eb at the URL http:rrmidway.ca.sandia.govr ;

mecolvrqsar_appendixr and by e-mail at [email protected]

Page 7: Quantitative structure–activity (QSAR) relationships of mutagenic aromatic and heterocyclic amines

( )F.T. Hatch, M.E. ColÕinrMutation Research 376 1997 87–96 93

Table 5Correlations for AHA165; only coefficients with p-0.001 are given

LogMP Fus Nme Het HOMOhu LUMOhu HOMOhunh

LogMP 1.0000Fus 0.7078 1.0000Nme 0.4542 1.0000Het 0.5738 0.4293 0.6338 1.0000HOMOhu –0.3848 –0.5833 1.0000LUMOhu 0.6373 0.7476 0.3329 0.4300 –0.4218 1.0000HOMOhunh –0.3989 –0.4534 0.2896 1.0000LUMOhunh –0.4442 –0.6420 0.9794 –0.4860 0.3314

Regression models were calculated to relate theŽ .mutagenic potency Log MP to the qualitative struc-

tural factors and the quantum mechanical energeticand electronic properties. Table 3A lists the resultsof these regression models for AHA165 and Table3B for SEL20, with LogMP as dependent variableand sets of independent variables. Table 4 summa-rizes, for the Huckel and ab initio calculation meth-¨ods, the major and minor predictors of Log MP forAHA165 and SEL20.

The regression models for the large series

Ž .AHA165 Table 3A show that three structural fac-tors, Fus, Nme and Het, contribute about 62 percentof the variance in the dependent variable Log MP.The model as a whole and each regression coeffi-cient are highly significant. The size of the aromatic

Ž .system Fus is responsible for the major share ofvariance, Nme contributes a moderate share and Heta very minor share. A model including both quantummechanical and structural factors for AHA165 is alsosignificant. The major contributor to variance is LU-

ŽMOhu i.e., the energy of the lowest unoccupied

Table 6Correlations for SEL20; only coefficients with p-0.01 are given

LogMP Fus Nme Het LUMOhu HOMOhunh LUMOhunh

LogMP 1.0000Fus 0.7684 1.0000Nme 0.6596 1.0000Het 0.6644 0.6504 1.0000LUMOhu 0.8280 0.8465 0.6121 1.0000HOMOhunh –0.6484 1.0000LUMOhunh –0.5737 –0.7513 0.7017 0.5995 1.0000LUMOab –0.7625 –0.9113 –0.9339 0.7020HOMOabnh 0.6888 0.5781 –0.8804 –0.8460LUMOabnh 0.6746 0.7419 0.5960 -0.7699 –0.7847nchrgenh –0.8196 –0.731 –0.7317 –0.7636 –0.6756 0.6152dipoleab 0.6677 0.8136 0.6305DDH ab –0.7189 –0.7158 –0.6440 –0.5697 0.7471 0.7153f

LUMOab HOMOabnh LUMOabnh nchrgenh dipoleab DDH abf

LUMOab 1.0000HOMOabnh 1.0000LUMOabnh –0.5937 0.8085 1.0000nchrgenh 0.5713 –0.7306 1.0000dipoleab –0.7636 1.0000DDH ab –0.7445 –0.9559 0.7892 1.0000f

Page 8: Quantitative structure–activity (QSAR) relationships of mutagenic aromatic and heterocyclic amines

( )F.T. Hatch, M.E. ColÕinrMutation Research 376 1997 87–9694

.orbital of the parent amine ; and moderate contribu-Žtions come from HOMOhunh i.e., the energy of the

.highest occupied orbital of the nitrenium ion and thenumber of ring heteroatoms.

The regression models for the small representativeseries SEL20 shows that two structural factors, Fusand Nme contribute about 75% of the variance inLog MP, with the major share coming from FusŽ .Table 3B . For SEL20 the model including both

Ž .structural Nme and Huckel quantum mechanical¨Ž .factors LUMOhu and HOMOhunh accounts for

80% of the variance in Log MP, with the majorcontribution from LUMOhu and a small contributionfrom HOMOhunh. The regression coefficient forHOMOhunh is of marginal significance. The regres-sion models for the AM1 method are similar to the

Ž .Huckel model data not shown . HOMOnh and¨LUMO by either Huckel or AM1 methods are not¨correlated with each other so that inclusion ofHOMOnh in the models is not due to collinearity.However, a comparable model for ab initio calcula-tions includes LUMOab, a moderate contribution

Ž .from LUMOabnh with a borderline coefficient , anda small contribution from the dipole moment; but itdoes not incorporate HOMOhunh. This model isvery similar to one found earlier for the series of 16

Fig. 2. Scatter plot of the relationship between mutagenic potencyand Huckel LUMO energy of the parent amines in AHA165.¨Because LUMOhu is expressed in units of b , which is always anegative number, the x-axis is in terms of -LUMO so that thesense of the energy level is like that for semi-empirical and abinitio calculations. The four outliers at about 0.5 on the x-axis areall aminoquinolines, which may be subject to unidentified influ-ences on potency. The outlier near 0.4 is 9-aminoacridine, whoseamino group is sterically hindered.

Fig. 3. Scatter plot of the relationship between mutagenic potencyand Huckel LUMO energy of the parent amines in SEL20. The¨LUMO scale is inverted as explained in the legend to Fig. 2.Compound numbers in the figure are identified in the list below.

No. Chemical name Abbrev.

1 benzenamine aniline2 3-aminotoluene 3toluidin3 2-aminopyridine 2apyridn4 3-methyl-4-methoxybenzenamine 34cresidin

X X5 2,2 -diaminobiphenyl 22 ab6 1-aminocarbazole 1acarbzl

X Xw x7 4-amino-dipyrido 1,2-a:3 ,2 -d - 4aglup2imidazole

w x8 2-amino-3H-imidazo 4,5-b pyridine dsmipw x9 2-amino-3-methylnaphtho 2,1-d - 3mni

imidazole10 2-amino-3-methyl-9H-pyrido- amac

w x Ž .2,3-b indole AMACw x11 2-amino-3H-naphtho 3,2-d imidazole lindsmni

12 1-aminophenanthrene 1aphen13 2-amino-1,5,6-trimethylbenzo- 156mbi

w x1,2-d imidazole14 2-amino-3,4,5,7,8-pentamethyl- 34578miqx

w ximidazo 4,5-f quinoxaline15 2-amino-4-methyldipyrido- 4mglup2

X Xw x1,2-a:3 ,2 -d imidazole16 8-aminofluoranthene 8aflan

w x17 2-amino-3-methylimidazo 4,5-f - 3miqxquinoxaline

w x18 2-amino-3-methylimidazo 4,5-f - 3miqŽ .quinoline IQ

19 2-amino-3,4-dimethylimidazo- 34miqxw x4,5-f quinoxaline

w x20 2-amino-1-methylimidazo 4,5-f - 1miqquinoline

Ž .aminoimidazo-azaarenes Hatch et al., 1996 . Wehave therefore placed a question mark besideHOMOnh in the summary of the multiple regression

Ž .analysis Table 4 .

Page 9: Quantitative structure–activity (QSAR) relationships of mutagenic aromatic and heterocyclic amines

( )F.T. Hatch, M.E. ColÕinrMutation Research 376 1997 87–96 95

Fig. 4. Scatter plot of the relationship between mutagenic potencyand the stability of the nitrenium ions in SEL20, calculated with

Ž .reference to aniline by the method of Ford and Griffin 1992 . Thex-axis label ddHf corresponds to DDH in Table 2. Numbers inf

the figure are identified in the legend to Fig. 3.

The pairwise Pearson correlation relationshipsamong the structural and quantum chemical factorsabove are shown in Table 5 for AHA165 and Table6 for SEL20. The LUMO energies of the parentamines are strongly correlated with the Log MP.Although the sign of the coefficient for LUMOhu bythe Huckel method is positive, the sense of the¨relationship is that its correlation with Log MP isinverse, because LUMOhu is expressed in units ofthe Huckel matrix element b , which is always a¨negative number. Higher mutagenic potency is asso-

Ž .ciated with lower LUMO energy Fig. 2 and Fig. 3 .Of particular interest for SEL20 are the correlationsof Log MP with nchrgenh, the charge on the exo-cyclic nitrogen of the nitrenium ion, y0.82, andwith DDH , the calculated stability of the nitreniumf

Ž .ion, y0.72 Fig. 4 . Other relationships in thesetables indicate linear dependencies among potentialpredictor variables which should be avoided in mul-tiple regression models.

4. Discussion

The major result of this study is the considerabledegree to which the findings of our previous study of

Ž16 congeneric aminoimidazo-azaarenes Hatch et al.,.1996 can be generalized to a large heterogeneous

series of aromatic and heterocyclic amines. Muta-genic potency is related in similar fashion to struc-tural factors characterizing the aromatic system and

to the LUMO energy of the parent amines. In thecase of the nitrenium ion derivatives, this relation-ship is not as clear, although the same multipleregression model was obtained for the ab initiocalculation in the aminoimidazo-azaarenes and theSEL20 series. Close linear relationships are also seenbetween the LUMOnh energy and both the calcu-

Ž . Ž .lated stability of the nitrenium ion DDH Fig. 4fŽ .and the NPA charge nchrgenh on the exocyclic

nitrogen atom of the nitrenium ion. The latter twoparameters are also significantly correlated withLog MP. Thus, mutagenic potency is at least in partrelated to low electrophilicity of the nitrenium ion,p-electron delocalization onto the exocyclic N-atom,and the calculated stability of the nitrenium ion. Forthe nitrenium ions of the aminoimidazo-azaarenes,an intramolecular rearrangement might contribute to

Ž .ion stabilization Colvin et al., 1997 . These featuresare compatible with the hypothesis expressed earlier.If the nitrenium ion is either the ultimate mutagenicmetabolite or a model for a transition state in thereaction binding an amine to a DNA base, the ionwill be most mutagenic if it has only limited reactiv-ity, so that it is not diverted into side reactions withnon-DNA nucleophiles before reacting with DNA.

4.1. Congruence of quantum mechanical methods

A special feature of this study is the variety oftheoretical methods applied to the SEL20 amines,ranging from simple Huckel calculations running on¨a PC in the blink of an eye, through semi-empiricalcalculations running a few hours on a 486 computer,to ab initio quantum chemical calculations performed

Table 7Pairwise correlation coefficients among three quantum mechanicalcalculation methods

HOMOab HOMOse LUMOab LUMOse

HOMOse 0.58HOMOhu –0.84 –0.60LUMOse 0.95LUMOhu –0.93 –0.97

HOMOabnh HOMOsenh LUMOabnh LUMOsenh

HOMOsenh 0.97HOMOhunh –0.88 –0.84LUMOsenh 0.92LUMOhunh –0.78 –0.81

Page 10: Quantitative structure–activity (QSAR) relationships of mutagenic aromatic and heterocyclic amines

( )F.T. Hatch, M.E. ColÕinrMutation Research 376 1997 87–9696

on a massively parallel supercomputer or a highlevel workstation. Agreement among the all threemethods for the orbital energies was remarkablygood. Correlation coefficients were calculated pair-wise for frontier orbital energies of the parent aminesand nitrenium ions derived from the three quantum

Ž .chemical methods Table 7 . The coefficients wereremarkably high, and all were significant at thep-0.01 level. We call attention to this good perfor-mance by the Huckel method, because it may prove¨to be an inexpensive and rapid method for use byother investigators.

Acknowledgements

This work was performed under the auspices ofthe US Department of Energy for Lawrence Liver-more National Laboratory under contract W-7405-ENG-48 and at Sandia National Laboratory undercontract SE-AC04-94AL85000. Partial funding was

Žprovided by the National Cancer Institute NIH Grant.no. P01 CA55861 . We thank the Massively Parallel

Computer Research Laboratory at Sandia for com-puter time on the Intel Paragon. We also thank MarkKnize for providing mutagenicity data on syntheticintermediates, Amal Moulik for researching CASnumbers, Keiji Wakabayashi for permitting use ofunpublished data on a series of aminoimidazo-azaarene isomers, and Errol Zeiger for providing uswith the database of Ames tests performed in supportof the National Toxicology Program.

References

˚Adamson, R.H., J.-A. Gustafsson, N. Ito, M. Nagao, T. Sugimura,Ž . Ž .K. Wakabayashi and Y. Yamazoe Eds. 1995 Heterocyclic

Amines in Cooked Foods: Possible Human Carcinogens, 23rdIntl. Symp. Princess Takamatsu Cancer Res. Fund, PrincetonScientific Publishing Co., Princeton, NJ.

Ž .Case, R.A., M. Hosker, D.B. McDonald and J.T. Pearson 1954Tumors of the urinary bladder in workmen engaged in themanufacture and use of certain dyestuff intermediates in theBritish chemical industry, Parts I and II, Br. J. Indust. Med.,11, 75, 213.

Colvin, M.E., O.L. Janssen, R.A. Whiteside and C.H. TongŽ .1993 Parallel direct SCF for large-scale calculations, Theor.Chim. Acta, 84, 301–314.

Ž .Colvin, M.E., E.T. Seidl, I.M. Nielsen and F.T. Hatch 1996Hydride shifts in aminoimidazole-azaarene mutagens, Chem.-Biol. Interact., submitted.

Ž .Ford, G.P. and G.R. Griffin 1992 Relative stabilities of nitre-nium ions derived from heterocyclic amine food carcinogens:Relationship to mutagenicity, Chem.-Biol. Interact., 81, 19–33.

Frisch, M.J., G.W. Trucks, M. Head-Gordon, P.M. Gill, M.W.Wong, J.B. Foresman, B.G. Johnson, H.B. Schlegel, M.A.Robb, E.S. Replogle, R. Gomperts, J.L. Andres, K.Raghavachari, J.S. Binkley, C. Gonzalez, R.L. Martin, D.J.

Ž .Fox, D.J. Defrees, J. Baker, J.J. Stewart and J.A. Pople 1992Gaussian 92, Gaussian Inc., Pittsburgh, PA.

Ž .Glantz, S.A. and B.K. Slinker 1990 Primer of applied regressionand analysis of variance, pp. 181–236, McGraw-Hill, NewYork.

Ž .Hatch, F.T., M.G. Knize and J.S. Felton 1991 Quantitativestructure-activity relationships of heterocyclic amine mutagensformed during the cooking of foods, Environ. Mol. Mutagen.,17, 4–19.

Ž .Hatch, F.T., M.G. Knize, D.H. Moore and J.S. Felton 1992Quantitative correlation of mutagenic and carcinogenic poten-cies for heterocyclic amines from cooked foods and additionalaromatic amines, Mutation Res., 271, 269–287.

Ž .Hatch, F.T., M.E. Colvin and E.T. Seidl 1996 Structural andquantum chemical factors affecting mutagenic potency ofaminoimidazo-azaarenes, Environ. Mol. Mutagen., 27, 314–330.

Ž .Kadlubar, F.F. 1987 Metabolism and DNA binding of carcino-genic aromatic amines, ISI Atlas of Science: Pharmacology,129–132.

Knize, M.G., N.H. Shen, S.K. Healy, F.T. Hatch and J.S. FeltonŽ .1987 The use of bacterial mutation tests to monitor chemical

Ž .isolation of the mutagens in cooked food, in G. Pierce Ed. , J.Indust. Microbiol. Suppl. 2, Developments in Industrial Micro-biology, pp. 171–180.

Ž .Miller, E.C. and J.A. Miller 1981 Searches for ultimate chemicalcarcinogens and their reactions with cellular macromolecules,Cancer, 47, 2327–2345.

Ž .Moore, D.H. and J.S. Felton 1983 A microcomputer program foranalyzing Ames test data, Mutation Res., 119, 95–102.

Novak, M., M.J. Kahley, E. Eiger, J.S. Helmick and H.E. PetersŽ .1993 Reactivity and selectivity of nitrenium ions derived

Ž .from ester derivatives of carcinogenic N- 4-biphenylyl hy-droxylamine and the corresponding hydroxamic acid, J. Amer.Chem. Soc., 115, 9453–9460.

Ž .Reed, A.E., R.B. Weinstock and F. Weinhold 1985 Naturalpopulation analysis, J. Chem. Physics, 83, 735–746.

Ž .Rehn, L. 1895 Blasengeschwulste bei fuchsin arbeitern, Arch.¨Klin. Chir., 50, 588–600.

Ž .Sabbioni, G. and D. Wild 1992 Quantitative structure–activityrelationships of mutagenic aromatic and heteroaromatic azidesand amines, Carcinogenesis, 13, 709–713.

Ž .Union Leader 1995 Study: Upjohn Chemical Workers HaveŽ .Cancer Risk, newspaper article, Sept. 21 AP , Union Leader

Corp., Manchester, NH.