A QSAR for the mutagenic potencies of twelve 2-amino- trimethylimidazopyridine isomers: Structural, quantum chemical, and hydropathic factors

Article abstract of DOI:10.1002/em.20177

An isomeric series of heterocyclic amines related to one found in heated muscle meats was investigated for properties that predict their measured mutagenic potency. Eleven of the 12 possible 2-amino-trimethylimidazopyridine (TMIP) isomers were tested for

Full text of DOI:10.1002/em.20177

Environmental and Molecular Mutagenesis 47:132^146 (2006)
A QSAR for the Mutagenic Potencies of Twelve
2-Amino-trimethylimidazopyridine Isomers: Structural,
Quantum Chemical, and Hydropathic Factors
1
2
3
1
4
M.G. Knize, * F.T. Hatch, M.J.Tanga, E.Y. Lau, and M.E.Colvin
1
Biosciences Directorate, Lawrence Livermore National Laboratory,
University of California, Livermore, California
2
Meredith, New Hampshire
3
Biosciences Division, SRI International, Menlo Park, California
4
School of Natural Sciences, University of California, Merced, California
An isomeric series of heterocyclic amines related to amines are: (1) a small dipole moment, (2) the com-
one found in heated muscle meats was investigated bination of b-face ring fusion and N3-methyl group,
for properties that predict their measured mutagenic (3) a lower calculated energy of the p electron sys-
potency. Eleven of the 12 possible 2-amino-trimethy- tem, (4) a smaller energy gap between the amine
limidazopyridine (TMIP) isomers were tested for HOMO and LUMO orbitals (Pearson ‘‘softness’’),
mutagenic potency in the Ames/Salmonella test with and (5) a more stable nitrenium ion. Based on pre-
bacterial strain TA98, and resulted in a 600-fold dicted potency from the average of six regression
range in potency. Structural, quantum chemical, and models, the isomer not yet synthesized and tested is
hydropathic data were calculated on the parent mol- expected to have a mutagenic potency of 0.77
ecules and the corresponding nitrenium ions of all of revertants/lg in tester strain TA98, which is near
the tested isomers to establish models for predicting the low end of the potency range of the isomers.
the potency of the unknown isomer. The principal Environ. Mol. Mutagen. 47:132–146, 2006.
determinants of higher mutagenic potency in these
VV C 2005 Wiley-Liss, Inc.
Key words: QSAR; TMIP; heterocyclic amine; cooking mutagen
INTRODUCTION
[Novak et al., 1998; Novak and Rajagopal, 2001], which
is a reactive electrophile that covalently binds and dam-
ages DNA, primarily at guanine bases by forming a bulky
adduct [Schut and Snyderwine, 1999]. A purely chemical
production of putative nitrenium ions has been reported
by Sabbioni and Wild [1992].
Aromatic and heterocyclic aromatic amines have been
considered to be actual or potential mutagens and carcino-
gens for over a century [Rehn, 1895; Case et al., 1954].
For the past quarter-century, a variety of heterocyclic aro-
matic amines have been found in cooked foods, primarily
well-done or over-cooked meats, and have been shown to
be extensively genotoxic in many test systems and carci-
nogenic in rodents and monkeys [Adamson et al., 1995;
Sugimura, 1995; Felton et al., 1995, 1999; Weisburger,
Since 1978, our laboratory has engaged in the identifi-
cation, isolation, synthesis, and assessment of the genetic
toxicology of the thermic food mutagens. We have pub-
lished a series of quantitative structure-activity relation-
ships (QSARs) directed at gaining insight into the chemical
2002]. These amines are genotoxic only after activation
by a series of enzymatic or biochemical reactions that
convert the parent compound into an electrophilic deriva-
tive. Cytochrome P450 1A1 and 1A2 are catalysts for the
Grant sponsor: U.S. Department of Energy, Lawrence Livermore Nati-
first activation step [Guengrich, 1995; Guengrich et al., onal Laboratory, University of California; Grant number: W-7405-Eng-48;
995; Turesky et al., 2002; Kim and Guengrich, 2004]. A Grant sponsor: NCI; Grant number: CA55861.
1
quantum chemical computational study of the mechanism *Correspondence to: Mark Knize, Biosciences Directorate, Lawrence
of this step, which is an oxidation involving removal of Livermore National Laboratory, University of California, 7000 East Avenue,
Livermore, CA 94550, USA. E-mail: knize1@llnl.gov
electrons from the parent amine has been carried out by
Sasaki et al. [2002]. The first stable intermediate in the
Received 29 August 2005; and in final form 16 September 2005
activation sequence is a hydroxylamine. This product is DOI 10.1002/em.20177
then esterified to an acetoxy or sulfate ester. Departure of
the ester group is postulated to leave a nitrenium ion wiley.com).
Published online 28 October 2005 in Wiley InterScience (www.interscience.
VVC 2005 Wiley-Liss, Inc.

QSAR of TMIP Isomers
133
mechanisms responsible for their genotoxic effects [Hatch tionship is stronger. Finally, the earlier study found an
et al., 1991; Hatch et al., 1992; Hatch et al., 1996; Hatch extreme correlation (r ¼ 0.98) between the highest occu-
and Colvin, 1997; Colvin et al., 1998; Hatch et al., 2001]. pied molecular orbital (E
) of the parent TMIPs and
homo
This work has broadly covered mutagenic aromatic and the energy to form the nitrenium ion, which reflects the
heterocyclic amines whose structures bear a resemblance oxidative loss of two electrons from the HOMO orbital of
to the thermic mutagens.
the parent amines. This is confirmed in the present study
An unknown mutagenic compound thought to be an but not elaborated upon, since Ehomo is not significantly
amino-trimethylimidazopyridine (TMIP) was isolated from correlated with mutagenic potency and was therefore not
well-done beef [Felton, 1984]. Many years later, the one iso- included in the predictor variables.
mer found in a meat is the 2-amino-1,5,6-trimethylimi-
Vracko et al. [2004] have reported a QSAR study of the
dazo[4,5-b]pyridine (156bmip) in heated chicken breast, TMIP isomers. Mutagenic potency data in Salmonella
and in the meat drippings from beef, chicken thigh, and pork strains TA98 and YG1024 were taken from Felton et al.
[
Pais et al., 1999]. A model system based on the composi- [1999] for 10 isomers (the 11th was synthesized and
tion of meat formed the same 156bmip isomer [Borgen assayed more recently). In their study, several machine
et al., 2001]. learning techniques were applied to identify strong predic-
Interest in determining which of the 12 possible iso- tors of potency and the results pooled as a training set to
mers of TMIP would be mutagenic or the most potent predict the potency of the remaining two isomers. Specifi-
mutagen led to the present work to determine properties cally, the following four approaches were reported: (1)
that could predict mutagenic potency. We have obtained multiple linear regression was applied to series of orbital
11 of the 12 possible isomers of TMIP (Table I) by chem- energies below HOMO and above LUMO and the best set
ical synthesis. The 12th isomer has eluded all synthesis of orbitals determined from the regression statistics. For
efforts to date, giving only trace amounts for the several unstated reasons the main HOMO and LUMO energies
synthesis routes attempted. The isomers differ structurally were not included; (2) counter Propagation Neural Network
only in the positions of methyl group substitution on the software was also applied to the orbital energy data; (3) a
imidazole and pyridine rings and in the fusion face geometric ‘‘spectrum-like representation’’ of the 3D molec-
between the imidazole and pyridine rings. These mole- ular structures of the isomers that was developed in the
cules have structural correlations with an important cook- authors’ laboratory was applied to produce the 3D position
ing mutagen/carcinogen PhIP (2-amino-1-methyl-6-phe- and polar angle for each atom leading to a Lorentzian
nylimidazo[4,5-b]pyridine) and other studied congeners. curve, which could then be analyzed as a spectrum; (4) the
These features seemed ideal for a QSAR study, although same geometric method in method 3 was supplemented
the mutagenic potency range (3 orders of magnitude) and with Mulliken atomic charges for each atom.
sample size are rather limited for statistical analyses.
The results of methods 2, 3, and 4 were subjected to
Three of the isomers are of moderate potency when cluster analysis producing Kohonen maps, which showed
compared with many other heterocyclic amines, and eight good relationships to mutagenic potency for methods 2
others are weak to very weak mutagens; the one not yet and 4, but not for method 3. Finally, all four methods
synthesized is of unknown potency. We have used struc- were pooled to predict the potency of the two untested
tural, quantum chemical, and hydropathic data calculated isomers. The prediction for 167bmip was in good agree-
on the parent molecules and their nitrenium ions (which ment with our recently measured value. The prediction
are hypothesized to be the penultimate or ultimate muta- for the still untested isomer 347cmip was about one log10
gens after metabolic activation) of all the isomers to higher than the prediction reached in this article.
establish models for predicting the potency of the un-
known. Limited data were also calculated for the imine quantum chemical, and hydropathic data on parent amines
tautomers and the hydroxylamines. and nitrinium ions to establish models that correlate prop-
In our article, we calculate a complete set of structural,
In an earlier article on the status of our food mutagen erties with potency, and use these to predict the potency
studies, we reported briefly [Felton et al., 1999] on the of the one remaining not-yet synthesized isomer of the set
mutagenic potencies and preliminary semi-empirical and of mutagenic TMIP.
ab initio quantum chemical calculations on the TMIP iso-
mers that are now being more fully reported. At that time,
MATERIALS AND METHODS
we showed that the strongest correlation with mutagenic
potency came from the dipole moment, although this rela-
tionship was not significant in earlier studies on more
diverse amines. Another difference from earlier work was
that the lowest unoccupied molecular orbital (Elumo) of
the parent TMIPs showed only a borderline correlation
Abbreviations and definitions are presented in Table II.
Chemicals
Eleven of the 12 possible TMIP isomers were synthesized in the labo-
with potency, whereas in other amine QSARs, this rela- ratory of co-author Dr. Mary Tanga at SRI International (Menlo Park,

TABLE I. Structure, CAS Number, Name, Synthesis Reference, Ring Fusion, N-Methyl Position, and Mutagenic
Potency of TMIP Isomers
Structure abbreviation
CAS NO.
Name
Synthesis
See Methods
b_c face Nme LogMP98 98 rev/lg LogMP1024
4
01560-72-3
2-Amino-3,6,7-
trimethylimidazo[4,5-b]-
pyridine
b
3
1.7
317.0
2.0
5
7667-51-3
2-Amino-3,5,6-
trimethylimidazo[4,5-b]-
pyridine
[Tanga et al., 1994]
[Tanga et al., 1997]
b
b
3
3
1.6
0.7
233.0
28.0
2.2
1.3
1
32898-06-7
2-Amino-3,5,7-
trimethylimidazo[4,5-b]-
pyridine
2
-Amino-1,6,7-
[Ahiberg et al., 2000]
[Tanga et al., 2003]
[Tanga et al., 1994]
b
b
b
1
1
1
0.2
0.0
10.0
—
trimethylimidazo[4,5-b]-
pyridine
4
1
01560-75-6
61091-56-0
2-Amino-1,5,7-
trimethylimidazo[4,5-b]-
pyridine
5.7
0.3
0.4
2-Amino-1,5,6-
trimethylimidazo[4,5-b]-
pyridine
ꢀ0.3
3.1
1
4
93690-74-3
01560-75-4
2-Amino-3,4,6-
trimethylimidazo[4,5-c]-
pyridine
[Tanga et al., 1997]
[Tanga et al., 2003]
c
c
3
1
ꢀ0.4
ꢀ0.5
2.3
1.6
0.2
0.4
2-Amino-1,4,6-
trimethylimidazo[4,5-b]-
pyridine
1
93690-65-2
2-Amino-1,4,7-
trimethylimidazo[4,5-c]-
pyridine
[Tanga et al., 1997]
c
c
1
3
ꢀ0.8
0.9
0.5
0.0
4
1
01560-74-5
93690-71-0
2-Amino-3,6,7-
trimethylimidazo[4,5-c]-
pyridine
See Methods
ꢀ1.0550
0.0911
2-Amino-1,6,7-
trimethylimidazo[4,5-c]-
pyridine
[Tanga et al., 1997]
c
1
3
ꢀ1.0550
0.5
0.0721
2
-Amino-3,4,7-
Not made
38c
—
—
—
trimethylimidazo[4,5-c]-
pyridine

QSAR of TMIP Isomers
TABLE II. Abbreviations and Definitions of Variables Used in the Tables and Text
135
Abbreviation
Meaning
Notation
AHA80
nnn
Aromatic and heterocyclic amines studied previously [Hatch et al., 1992–2001]
Positions of first methyl group on the imidazole ring followed by the two methyl
groups on the pyridine ring, shown in the abbreviation for each molecule
þ
Suffixes
NH2, parent amine; NHOH, hydroxylamine derivative; NH , nitrenium ion; tau,
imine tautomer of parent
Main dependent variable
LogMP98
Log10 of revertants per nanomole in Ames/Salmonella strain TA98 or TA1538
LogYG1024
Log10 of revertants per nanomole in O-acetylase over-producing strain YG1024
Predictor variables:
Structural factors
Nme
Methyl group on imidazole ring N atom at position 1 or 3; in Table III, pos 3¼1, pos 1¼0
Position of fusing imidazole and pyridine rings: pyridine face 2-3¼b; face 3-4¼c,
shown in the name of each molecule; in Table III, face b¼1, c¼0
Face
Quantum chemical factors
RHF-
Ford-
Restricted Hartree-Fock calculation of total energy in a.u. (Hartrees)
Simulation of reaction energy calculation (Eproducts-Ereactants) vs aniline [Ford and
Griffin, 1992] in kcal/mol
DE_
DipoleNH2
Difference in RHF energies between two suffixed derivatives
Dipole moment of parent amine from Hartree-Fock (ab initio) SpartanES
calculation in Debyes
Dipoletau
DDipole
Dipole moment of imine tautomer
Difference: DipoleNH2–Dipoletau
EhomoNH2
ElumoNH2
SoftNH2
Energy of Highest Occupied Molecular Orbital in electron volts
Energy of Lowest Unoccupied Molecular Orbital in electron volts
Reciprocal of Pearson calculation (ElumoNH2-EhomoNH2)/2 of Hardness in reciprocal
electron volts [Pearson, 1963; Pearson and Songstad, 1967; Jensen, 1999]
Total energy of p electrons in parent molecule calculated by simple H u¨ ckel
theory with HMO software in b units; a units are constant and are omitted.
Natural Population Analysis charge on an atom, calculated by Spartan [Reed at al., 1985]
NPA charge on imidazole N atom containing the methyl substituent
Ep_b
NPA
qNme_NH2
Hydropathic factors
LogP_GC
Ghose-Crippen calculation of LogP from constituent atoms and atomic connectivity
[
Ghose et al., 1988] by Spartan software
LogP_HL
LogP predicted by MMP software from molecular fragments [Hansch and Leo, 1979]
CA). All were greater than 98% purity as judged by HPLC using two Two milligrams of Aroclor-induced rat liver S9 protein was added per
different chromatographic systems. Their structures and references for
synthesis are shown in Table I.
plate for metabolic activation, and the compounds were tested in doses to
give a linear response covering a limited range from 1 to 100 lg/plate. A
positive control, 2-amino-3-methylimidazo[4,5-f]quinoline (IQ), gave
Synthetic routes for two of the isomers were not published. Eight steps
were used to make the 367cmip isomer from 2,3-lutidine via nitration 1200–1500 revertants/5 ng dose. Dimethylsulfoxide was included in the
at the 4-position, reduction, bromination at the 5-position, dehalogenation
negative controls (spontaneous revertant counts) and gave TA98 values of
5–60 revertant colonies per plate. Strain YG1024 gave higher spontane-
1
with methylamine, and cyclization with cyanogen bromide ( H NMR
2
(
CD
67bmip isomer was made from 3,4-lutidine N-oxide reacted with N-
methylchlorophenylmethanimine (made from N-methylbenzamide) to form
-methylamino-4,5-dimethylpyridine, which was nitrated at the 2-position,
3 4
O-d ): d2.40 (s, 3H), 2.53 (s, 3H), 3.61 (s, 3H), 8.15 (s, 1H)). The
ous reversion rates of 100–150 colonies per plate.
3
A minimum of four dose points from duplicate platings was used, and
the linear portion of the curve was used to calculate the number of rever-
tant colonies per microgram [Moore and Felton, 1983]. The standard
error of the linear fits for all samples was less than 15%.
2
1
reduced, and cyclized with cyanogen bromide ( H NMR (DMSO-d
6
):
d2.20 (s, 3H), 2.28 (s, 3H), 3.45 (s, 3H), 6.62 (br s, 2H), 7.64 (s, 1H)).
It is important to point out that the Ames test was carried out with S.
typhimurium strain TA98 bacteria. This strain and a closely related strain
TA1538 are commonly used for studying aromatic and heterocyclic
amines because they are especially sensitive to frame-shift mutations,
which are caused by large planar molecules. The basis for the sensitivity
of the assay are the mutations engineered into the strains by Dr. Ames
Salmonella Mutagenicity Assay
The mutagenic activity of the samples was determined using the
standard plate incorporation assay described by Ames et al. [Ames et al., as follows: hisD3052 mutation results in a strain that cannot grow with-
975; Maron and Ames, 1983], with Salmonella typhimurium strain TA98 out added histidine unless reverted by a frame-shift mutation [Fuscoe
a gift of Professor Bruce Ames, University of California, Berkeley). Sam- et al., 1988; Felton et al., 1995]; uvrB mutation results in a deficiency in
1
(
ples were also assayed with strain YG1024 [Watanabe et al., 1990; Einisto DNA repair; and rfa mutation causes a partial loss of the lipopolysac-
et al., 1991], an O-acetyltransferase overproducing derivative of TA98 (a charide surface that increases permeability to large molecules [Ames et al.,
gift of Dr. T. Nohmi, National Institute of Hygienic Sciences, Tokyo).
1973].

136
Knize et al.
TABLE III. TMIP Variables Database
Name LogMP98 Log1024 Face Nme DipoleNH2 Dipoletau DDipole EhomoNH2 ElumoNH2 SoftNH2 LogP_GC Ep_b qNmeNH2
3
3
3
1
1
1
3
1
1
3
1
3
67bmip
56bmip
57bmip
67bmip
57bmip
58bmip
46cmip
46cmip
47cmip
67cmip
67cmip
47cmip
1.7471 1.9503
1.6134 2.1845
0.6932 1.2588
0.2460
1
1
1
1
1
1
0
0
0
0
0
0
1
1
1
0
0
0
1
0
0
1
0
1
3.030
2.990
2.960
7.070
6.020
6.390
4.850
6.450
6.780
5.670
6.870
5.830
2.380
2.474
2.444
2.555
2.370
2.348
1.046
0.450
0.130
0.390
0.304
0.708
0.650
0.516
0.516
4.515
3.650
4.042
3.804
6.000
6.650
5.280
6.566
5.122
ꢀ0.2742
ꢀ0.2681
ꢀ0.2688
ꢀ0.2740
ꢀ0.2828
ꢀ0.2689
ꢀ0.2949
ꢀ0.2717
ꢀ0.2738
ꢀ0.2924
ꢀ0.2746
ꢀ0.2932
0.1407
0.1419
0.1436
0.1361
0.1359
0.1426
0.1409
0.1492
0.1530
0.1475
0.1451
0.1522
4.8208
4.8784
4.8490
4.8774
4.7776
4.8613
4.5886
4.7516
4.6861
4.5462
4.7650
4.4905
2.159
1.873
1.873
2.159
1.873
1.873
0.871
0.871
1.157
1.157
1.157
1.157
17.667
17.665
17.695
17.662
17.687
17.661
17.774
17.781
17.769
17.755
17.755
17.768
ꢀ0.530
ꢀ0.529
ꢀ0.527
ꢀ0.522
ꢀ0.517
ꢀ0.518
ꢀ0.517
ꢀ0.520
ꢀ0.520
ꢀ0.517
ꢀ0.522
ꢀ0.515
0.0019 0.3142
ꢀ0.2626 0.4391
ꢀ0.3923 0.2098
ꢀ0.5499 0.3763
ꢀ0.7998 0.0242
ꢀ1.0550 0.0911
ꢀ1.0550 0.0721
avgtop3
2.993
6.263
0.0000
2.433
1.199
0.0129
0.785
0.561
5.063
ꢀ0.2704
ꢀ0.2791
0.0548
0.425
0.1421
0.1438
0.4768
ꢀ0.484
4.8494
4.7317
0.0294
0.637
1.968
1.390
0.0183
0.741
17.676
17.731
0.0260
ꢀ0.735
ꢀ0.5287
ꢀ0.5191
0.0003
avgnext8
T-testprob
Correl.
0.0000
ꢀ0.913
1.000
0.971 0.770 0.493 ꢀ0.799
ꢀ0.836
LogMP98
The bacteria are added to a buffered suspension of microsomes (the file. LogP_HL was calculated on minimized structures with Molecular
9000g supernatant of homogenized rat or hamster liver cells), which sup- Modeling Pro ver. 5 (MMP, ChemSW, Fairfield, CA).
ply CYP 450 and other enzymes. Microsomes are tiny membranous
vesicles derived from the endoplasmic reticulum of the mammalian cells.
When the bacteria are incubated in this suspension with certain supple- Dependent and Predictor Variables
ments, reactive molecules produced in the activation process diffuse or
The dependent variable LogMP98 was derived from the Ames test data
are transported through the bacterial membrane into the interior, where
they may react with bacterial DNA, or with other enzymes or biochemi-
cals that may divert them from genotoxic effects. Despite the rfa muta-
tion, the hydrophobicity of the molecules could still affect their rate of
transport. In a QSAR for mammalian systems, the activated molecules
encounter several membrane layers on the way from the site of mutagen
exposure to the DNA within the nuclear membrane. In the mammalian
case, the consideration of hydrophobic properties and possible detoxifica-
tion reactions becomes more complicated.
by conversion of revertants/microgram of mutagen into log10 revertants/
nanomoles. The predictor variables were derived from the software calcu-
lations described above and are defined in Table II.
RESULTS
Mutagenic Potency
The database of the main dependent variable of Ames/
Salmonella mutagenic potency in strains TA98 and the
independent or predictor variables are presented for the
set of 12 TMIP isomers in Table III. Potency data for O-
acetylase-overproducing strain YG1024 are also given but
Computational System and Software
The computer used was an HP Vectra VL420 1.9 GHz Pentium 4
(
(
(
Hewlett-Packard, Mountain View, CA) running Windows XP Pro
Microsoft, Redmond, WA). Database management was with Excel 97
Microsoft) supplemented with Accord 3 for Excel (Accelrys, formerly
Synopsys, San Diego, CA) and with [Excel 2002]. Structures for Table I were not analyzed in detail. Below the data are the aver-
were drawn with Isis/Draw (MDL, San Leandro, CA) and incorporated ages for the more potent three isomers, the next eight iso-
into the table with Accord. Graphics and statistical analysis were done
mers, and the probability of a t-test for the significance of
with Excel 2002 and Small Stata 8 (Stata, College Station, TX); 3-D
the difference, using unequal variances. A significant dif-
graphics were generated with PSI-Plot, ver. 7.8 (Poly Software Intl.,
ference between the isomer subsets suggests a predictor
Pearl River, NY). Forward stepwise multiple linear regression models
were calculated with Small Stata ver. 8. Probability for a variable to that should be examined further for influence on potency.
enter a model was set at 0.20 and to be removed at 0.21. Ab initio quan-
The bottom line of this table gives the correlation coeffi-
cient for each predictor variable with LogMP98.
tum chemical calculations, made with SpartanES 04 (Wavefunction,
Irvine, CA), begin with optimization of geometry using in sequence
Whereas our previously reported series of 80 aromatic
and heterocyclic amines (AHA80) of heterogeneous struc-
MMFF, AM1, and Hartree-Fock (basis set 6-31G**) levels of theory.
Following optimization, the Restricted Hartree-Fock (RHF) energies,
molecular orbital energies, dipole moment, LogP_Ghose-Crippen, and ture covered a potency range of about 10 orders of mag-
natural population atomic charges (NPA) on several atoms were calcu-
nitude [Hatch et al., 2001], the 11 measured TMIPs had a
range of nearly three orders of magnitude between the
lated. Simple H u¨ ckel calculation of total p electron energy was per-
formed with the HMO program ver. 3.0a (Trinity Software, Plymouth,
30th and 76th percentiles of the AHA80 series. The nar-
rower potency range, as well as extensive multicollinear-
NH). This program had an extra line added to the parameter file includ-
ing parameters for a positively charged N-atom and auxiliary induction
and methyl group correction factors were added to the HMO parameter ity among many pairs of predictor variables, place limita-

QSAR of TMIP Isomers
137
tions on our ability to produce reliable statistical models
relating potency to predictor variables.
Predictor Variables
General
Independent or predictor variables in this study are sep-
arated into three types: structural, quantum chemical, and
hydropathic, as described in Table II. Because there are
many substantial correlations among the predictor varia-
bles, it is essential to avoid combining highly correlated
variable pairs (r > ꢁ0.7) in multivariate regression mod-
els to prevent artifacts created by multicollinearity [Glantz
and Slinker, 1990] pp 181 ff.]. Carefully selected combi-
nations of predictors were incorporated into models that
are believed to be free of artifact.
The relatively small size of the data set poses a chal-
lenge in developing a statistically significant QSAR from
a sizable number of potential predictor variables. Topliss
and Edwards [Topliss, 1979] used Monte Carlo simula-
tions to investigate the probability of significant correla-
tions occurring by chance in models built from sets of
random numbers. Although no simple rule of thumb
emerges from that study, it suggests that to achieve a P
<
0.05 level of confidence that a regression model is not
spurious, 11 observations will support up to six or seven
variables. Recently, Livingstone and Salt published a new
analysis of the question of ‘‘selection bias’’ in multiple
regression analysis [Livingstone and Salt, 2005]. They
provide an approximate equation for Fmax that is cor-
rected for this bias. The significance of the models in this
article was tested using this adjusted Fmax (see later).
The database for the isomers is presented in Table III,
showing mutagenic potency in Salmonella and only those
predictor variables that contributed to satisfactory multi-
variate regression models or were of potential mechanistic
interest. Other variables were calculated with the several
software programs described under Methods, but did not
appear to be significant for prediction of potency or for
mechanistic insight; and these data are omitted.
Table IV presents the RHF energies of the amine, tau-
tomer, hydroxylamine, and nitrenium ion forms of the
TMIPs and for aniline derivatives to facilitate calculation
of reaction energies. In addition, the RHF energy differen-
ces between the reaction stages and the relative energies
for hydroxylamine and nitrenium ion formation are pro-
vided (calculated relative to aniline, following Ford and
Griffin [1992]). Table V contains a correlation matrix for
the dependent and all relevant independent variables of
the isomers database (Tables III and IV). Values are
printed only if their significance level (shown on the sec-
ond line of each entry) is below 0.1. Clearly, all of the
predictors are at least moderately well-correlated pairwise

138
Knize et al.
TABLE V. Correlation Matrix
Dipole
NH2
LogMP98
DDipole qNmeNH2 LogP_GC
Ep_b
FordNHþ_NHOH
LogMP98
1.000
DipoleNH2
ꢀ0.799
1.000
0
.003
DDipole
ꢀ0.913
.000
ꢀ0.836
.001
0.741
.009
ꢀ0.735
.010
ꢀ0.657
.028
0.637
.035
0.918
0.000
0.707
0.010
1.000
0
qNmeNH2
LogP_GC
Ep_b
0.721
0.008
ꢀ0.676
0.016
0.686
0.014
0.561
0.058
1.000
0
ꢀ0.577
0.050
0.549
0.065
0.700
0.011
ꢀ0.680
0.015
1.000
0
ꢀ0.970
0.000
1.000
0
FordNHþ_NHOH
SoftNH2
ꢀ0.677
0.016
0.712
0.009
1.000
0
0.746
ꢀ0.788
0.002
ꢀ0.930
0
0.005
0.000
Fig. 1. 3-D graph showing the segregation of the mutagenic potencies
of the TMIP isomers by the face of pyridine–imidazole ring fusion and
the position of the imidazole N-methyl substituent.
Fig. 2. Scatter graph of the relationship between LogMP98 and DDipole,
which is the difference between the dipole moments of the parent amine
and the imine tautomer. The correlation coefficient is highly significant at
with mutagenic potency and these form the source for ꢀ0.913. Also shown are the linear predicted fit line and its 95% confi-
regression models.
dence interval bands. Actual data points lie under the ‘‘b’’ or ‘‘c’’ of the
isomer names.
Structural Predictors
þ
NH2, LogP_GC, Ep_b, FordNH _NHOH, and SoftNH2
Two structural variables, Face and Nme, show extraordi-
nary relationships to potency (Fig. 1). The six isomers of in descending order of correlation with LogMP98.
During study of the various activation intermediates, we
became interested in the imine tautomer that could be
higher potency all have the pyridine ring fused to the imida-
zole ring at the b-face, with the pyridine ring nitrogen at
position 4. The more potent three of these have the N- formed by transfer of a hydrogen from the exocyclic N-
atom of the parent amine to the nonmethylated nitrogen of
tion 1. The five isomers of lower potency all have a c-face the imidazole ring and shifting of the adjacent double bond
to form an exocyclic imine. The RHF energies to form the
Three of these have the N-methyl at position 1 and two at tautomers from the parent amines are fairly low (1.5–7.0 kcal/
methyl group at position 3 and the less potent three at posi-
ring fusion, with the pyridine ring nitrogen at position 5.
mol), indicating that reactions involving tautomer formation
could be quite fast at room temperature. There is little overall
correlation between the energy of the tautomers and muta-
genic potency, but the tautomer formation energy of the three
position 3; the unsynthesized isomer is also at position 3.
Quantum Chemical Predictors
The quantum chemical variables found to be potential most mutagenic isomers is much higher than that of most of
predictors of potency are DDipole, qNmeNH2, Dipole the other isomers. The tautomer dipole moment, Dipoletau, is

QSAR of TMIP Isomers
139
their complexity, found that only the reaction to form the
nitrenium ion from the parent amine showed any correlation
between reaction energy and potency [Ford and Griffin,
1992].
Our calculations, using ab initio (RHF) energies and
omitting the ester intermediates, also show a slight correla-
tion between the mutagenic potency and the energy to form
the nitrenium ion from the parent amine (r ¼ 0.52), as well
as slightly stronger correlations with potency for both the
formation of the hydroxylamine derivative from the parent
amine (r ¼ 0.60) and the formation of the nitrenium ion
from the hydroxylamine (r ¼ ꢀ0.66). Note that the energy
to form the nitrenium ion from the parent amine correlates
very strongly with the energy to form the nitrenium ion from
the hydroxylamine (r ¼ 0.96), but not with the energy to
Fig. 3. 3-D graph showing the segregation of the mutagenic potencies form the hydroxylamine from the parent (r ¼ ꢀ0.10).
of the TMIP isomers by the predictors Ep_b and DipoleNH2. Ep_b cre-
ates a major separation based on Face, and DipoleNH2 creates a modest
separation based on Nme position.
FordNHOH_NH2 is the energy (E
ꢀ Ereactants)
products
of the isodesmic reaction between AnilineNHOH and
TMIPNH2, yielding anilineNH2 and TMIPNHOH (Fig. 4).
The slope is positive, indicating that the more potent three
isomers have less negative reaction energies, and therefore
well correlated with potency. The difference between the their hydroxylamine derivatives will require more energy
parent dipole and the tautomer dipole, DDipole, has the high- to form from the parent amines (Note that all of these
est correlation with potency of the variables we have studied reaction energies are calculated relative to aniline, so their
(
r ¼ ꢀ0.91) (Fig. 2).
sign does not indicate whether the reaction is exothermic
The variable qNmeNH2 is the NPA charge on the imida- or endothermic in absolute terms).
þ
zole N-atom containing the methyl substituent. This varia-
FordNH _NHOH is the energy of the reaction between
þ
ble is strongly correlated (r ¼ ꢀ0.89) with LogMP98.
anilineNH and TMIPNHOH, yielding anilineNHOH and
þ
DipoleNH2 exhibits a striking relation to potency; the TMIPNH (Fig. 5). The slope is negative and the more
three isomers of highest potency have the lowest dipole potent three isomers have relatively large negative reac-
moments by a wide margin, and these are b-face and N3- tion energies for the formation of more stable nitrenium
methyl molecules (Fig. 3). Among the c-face isomers, the ions from the N-hydroxyl species.
3
-Nmethyls also have lower dipole moments than the 1-
Nmethyls, but by a lesser margin. DipoleNH2 is the main which signifies that the more potent isomers are generally
contributor to four of the regression models (Table VI).
softer in the Pearson sense (i.e., have narrower energy dif-
SoftNH2 shows a modest correlation with potency,
Ep_b is the total energy of p electrons calculated using ferences between HOMO and LUMO orbitals and are
simple H u¨ ckel theory. Although the number of p electrons more readily polarizable). See discussion of softness as a
is uniformly 12 in all isomers, the total energies show QSAR predictor in Arulmozhiraja and Morita [2004].
slightly smaller b values in the more potent isomers. In fact,
Hydropathic Predictors
the p-electronic energies are less negative for all b-face iso-
mer energies than for all c-face isomers. This variable is
moderately correlated with potency and makes a moderate
contribution to one regression model (Table VI and Fig. 3).
The hydropathic variable LogP_GC is moderately cor-
related with mutagenic potency (r ¼ 0.74). This Ghose-
Crippen calculation is atom-based, in contrast to the more
Reaction energies for some steps in the activation proc-
familiar Hansch method that is based on the summation
ess were calculated according to the method of Ford and
of values for molecular fragments. The Hansch calcula-
Herman [1991, 1992]. This scheme, as applied to hetero-
tion (LogP_HL) was somewhat correlated with potency
cyclic amine food mutagens by Ford and Griffin [1992],
but did not enter significant regression models (data not
calculates the relative energy of formation of the TMIP
shown).
nitrenium ions from the parent amines, using the analo-
gous reaction for aniline as a reference. The energy to
form the nitrenium ion may relate to the energy to form
reactive intermediates that ultimately react with DNA to
Regression Models
Satisfactory regression models, based on potency as the
form adducts, as measured by LogMP98. Ford and Griffin, dependent variable, require care in design because of the
using the semi-empirical AM1 method and omitting the number of predictor variables and the frequency of multi-
esterification and deesterification reaction steps because of collinearity among them. Six acceptable model equations

140
Knize et al.
TABLE VI. Multiple Linear Regression Models
Model 1
LogMP98¼
Pcoeff.
ꢀ0.3304
0.000
*DDipole
þ2.1818
0.106
*SoftNH2
*LogP_GC
*Ep_b
ꢀ1.8472
Beta
ꢀ0.7906
81.60%
0.3751
85.30%
39.06
0.2522
3.70%
Variance %:
RMSE
2
Adjusted R :
Fmodel:
Pmodel:
LogMP98¼
Pcoeff.
0.0002
ꢀ0.3624
0.002
Model 2
Model 3
Model 4
Model 5
Model 6
*DipoleNH2
*DipoleNH2
*DipoleNH2
*qNmeNH2
*DipoleNH2
1.0049
0.005
þ0.4086
Beta
ꢀ0.6091
59.80%
0.3946
83.80%
26.78
0.5178
24.00%
Variance %:
RMSE
2
Adjusted R :
Fmodel:
Pmodel:
LogMP98¼
Pcoeff.
0.0003
ꢀ0.364
0.002
ꢀ9.8782
0.006
þ176.9349
ꢀ17.6514
þ80.4331
þ1.8472
Beta
ꢀ0.6118
59.80%
0.4034
83.00%
25.44
ꢀ0.511
23.20%
Variance %:
RMSE
2
Adjusted R :
Fmodel:
Pmodel:
LogMP98¼
Pcoeff.
0.0003
ꢀ0.4128
0.001
þ4.1743
0.007
*SoftNH2
Beta
ꢀ0.6937
59.80%
0.4035
83.01%
25.43
0.487
Variance %:
RMSE
2
Adjusted R :
23.20%
Fmodel:
Pmodel:
LogMP98¼
Pcoeff.
0.0003
ꢀ128.0435
0.005
ꢀ8.2513
0.032
*Ep_b
Beta
ꢀ0.6257
66.50%
0.4423
79.59%
20.49
ꢀ0.427
13.09%
Variance %:
RMSE
2
Adjusted R :
Fmodel:
Pmodel:
LogMP98¼
Pcoeff.
0.0007
ꢀ0.3819
0.007
þ
ꢀ0.0619
0.049
*FordNH _NHOH
Beta
ꢀ0.3417
59.80%
0.5097
72.90%
14.44
ꢀ0.4121
13.10%
Variance %:
RMSE
2
Adjusted R :
Fmodel:
Pmodel:
0.0022
were found, involving LogMP98 as the dependent varia- the dependent and independent variables in test QSARs
ble and a variety of predictor variables. The coefficients with random numbers and showed that stepwise multiple
2
of determination, adjusted R , are equal to or greater than regressions produced significant responses by chance
8
0% for five of the models and slightly lower for the when the ratio of number of variables to number of obser-
sixth model. In each case, no other variables deemed to vations exceeded certain values. Since our study has
be relevant could be entered into the models without vio- seven variables and 11 observations, it could be subjected
lating multicollinearity criteria for a proper model. Thus, to the foregoing problem. A recent article [Livingstone
a large fraction of the total variance in potency is and Salt, 2005] extends the work of Topliss and Edwards
‘‘explained’’ by the models.
by producing an algorithm to predict Fmax values from
There is a further limitation to the regression analysis large numbers of simulations with random number input at
of the TMIP isomers. This is based on the work of Topliss various numbers of observations, total predictors con-
and Edwards [1979], who replaced either or both of sidered, and predictors in the model. From a rank-ordered

QSAR of TMIP Isomers
141
Fig. 5. Scatter graph of the relationship between LogMP98 and the
Fig. 4. Scatter graph of the relationship between LogMP98 and the value of FordNHþ_NHOH, which is calculated as follows:
value of FordNHOH_NH2, which is calculated as follows:
+
+
Reaction: TMIP_NHOH + PhNH = TMIP_NH + PhNHOH
Reaction: TMIP_NH2 + PhNHOH = TMIP_NHOH + PhNH2
+
Calculation: ((RHF_NH + RHF_PhNHOH) - (RHF_NHOH + RHF_
+
Calculation: ((RHF_NHOH + RHF_PhNH2) - (RHF_NH2 + RHF_ PhNH ))*627.51 kcal/mol. The correlation coefficient (r ¼ ꢀ0.657) is
PhNHOH))*627.51 kcal/mol. The correlation coefficient (r ¼ 0.646) is significant. Also shown are the linear predicted fit line and its 95% confi-
significant. Also shown are the linear predicted fit line and its 95% confi- dence interval bands. Actual data points lie under the ‘‘b’’ or ‘‘c’’ of the
dence interval bands. Actual data points lie under the ‘‘b’’ or ‘‘c’’ of the isomer names. ‘‘Ph’’ equals Phenyl [Ford and Griffin, 1992]. Note that
isomer names. ‘‘Ph’’ equals Phenyl [Ford and Griffin, 1992].
the slopes of Figures 4 and 5 are opposite.
list of the resulting Fmodels, Fmax values can be obtained create all models presented. Thus, b gives the normalized
for a series of probability levels. A power function equa- value of each coefficient, or the relative weight of each
tion was fitted to a large amount of simulation data. Both predictor variable in the model. The fourth line shows the
the simulation calculation and the equation output are fraction of the total variance in the dependent variable
freely available on the authors’ website. When applied to that is ‘‘explained’’ by each predictor variable. The final
our six regression models for 11 observations, seven total three rows give the value of the coefficient of determina-
2
predictors, and two predictors per model, the Fmax for tion for the model, R (adjusted for the number of predic-
chance significance of the model at 5% probability by sim- tor variables in the model); the value of the F-test for the
ulation was 11.8 and by equation was 13.0. For five of our significance of the model; and the Pmodel, which is the
regression models, the Fmodel ranged from 20 to 30, far probability that the null hypothesis (i.e., no credible pre-
above the chance result. For our weaker sixth model, Fmo- diction) can be accepted.
del is 14.4, which gives a probability of a chance result of
2
The first five models give adjusted R from 80 to 85%
somewhat less than 5%; so chance is not completely ruled of total variance. Model 1 admits DDipole with a large
out. contribution and a small addition from SoftNH2. In models
Table VI details the regression models in descending 2–4, DipoleNH2 is the major contributor with additions of
2
order of the adjusted R . The first line of each model 23–24% from LogP_GC, Ep_b or SoftNH2. Only one
shows the equation derived from the data: the dependent other variable was allowed per model by forward stepwise
variable is listed first, followed by the major predictor regression. Model 5 admitted qNmeNH2 at 66.5% of var-
2
variable and other predictors, if any, and finally the con- iance and Ep_b at 13.1%. Model 6, at a lower adjusted R
stant. The second line gives the significance level of the of 72.9%, again had DipoleNH2 as the main predictor and
þ
coefficient above it; in several instances, significance FordNH _NHOH with a smaller contribution.
greater than P ¼ 0.05 was allowed for minor variables
because the model was otherwise highly significant potency of the not-yet synthesized isomer 347cmip by
Fmodel, Pmodel). The third line gives the b values for entering the values of the predictors for that molecule into
All six regression models were used to predict the
(
each coefficient, which are the coefficients that would be the equation (Table VII). The predicted potencies from
obtained if all variables had been normalized to a mean these model equations were averaged and gave a value of
of 0 and a standard deviation of 1, thus correcting the LogMP98 ¼ ꢀ0.866 or 0.77 revertants/lg for 347cmip,
coefficients for differences in scale among the calculated which is near the bottom of the potency range of the 11
variables. This normalization was used by the software to measured isomers.

142
Knize et al.
TABLE VII. Regression Models for Predicting Mutagenic Potency (LogMP98) of 347cmip
Predictions
Model 1 LogMP98¼
ꢀ0.3304
0.3304
*DDipole
5.1223
þ2.1618
þ2.1618
*SoftNH2
4.49055
ꢀ9.01425 ꢀ0.9990
ꢀ
ꢀ9.01425 0.569rev/lg
Model 2 LogMP98¼
Model 3 LogMP98¼
Model 4 LogMP98¼
ꢀ0.3624 *DipoleNH2 þ1.0049 *LogP_GC
þ0.4086
ꢀ0.5415
ꢀ
0.3624
ꢀ0.364
0.364
ꢀ0.4128 *DipoleNH2 þ4.1743
0.4128 5.83
þ4.1743
5.83
*DipoleNH2 ꢀ9.8762
5.83
ꢀ9.8762
þ1.0049
1.157
þ0.4086
1.631rev/lg
*Ep_b
17.768
þ176.9349
þ176.9349
ꢀ0.6675
ꢀ
1.22rev/lg
*SoftNH2
4.49055
ꢀ17.6514
ꢀ1.3131
ꢀ
ꢀ17.6514
0.276rev/lg
Model 5 LogMP98¼ ꢀ126.0435 *qNmeNH2 ꢀ8.2513
*Ep_b
17.768
þ80.4331
þ80.4331
ꢀ1.2636
ꢀ
126.0435 ꢀ8.2513
ꢀ0.515
0.309rev/lg
Model 6 LogMP98¼
ꢀ0.3819 *DipoleNH2 ꢀ0.0619 *FordNHþ
NHOH
9.3286
þ1.8472
ꢀ0.957
_
ꢀ0.3819
5.83
ꢀ0.0619
þ1.8472
0.627rev/lg
Predicted ꢀ0.8664
LogMP98
Average 0.772rev/lg
Relationships Among Predictor Variables
potency variation is a continuum without a sharp distinction
based on Face. Concomitantly, in the more potent three
isomers, the imidazo-substituted methyl group is on the
As shown in Table V, DDipole is moderately and inver-
sely correlated with Face, LogP_GC, and Ep_b; and it is
positively correlated with qNme_NH2. DipoleNH2 makes
a substantial contribution to determining the variance in
mutagenic potency of the isomers. DipoleNH2 is nega-
tively correlated with Nme, whereby the three highest
potency isomers (all N3-methyl) have very low dipole mo-
ments and the remaining N3-methyl isomers are slightly
less potent than the N1-methyl isomers. LogP_GC is corre-
lated strongly with b-face over c-face, and inversely with
Ep_b. LogP_GC also correlates moderately with Elu-
moNH2 (inversely) and SoftNH2. Ep_b is very strongly
correlated with b-face over c-face and shows a sizeable
gap between them (Fig. 3).
3-position and on the 1-position of the next most potent
three isomers; all of these are b-face. In the remaining less
potent six isomers (all c-face), the N1- and N3-methyl sub-
stitutions are scattered, with the unsynthesized isomer being
at N3-methyl and c-face. The relation between potency and
methyl position on the imidazole ring (N1 vs. N3) contrasts
with our previous AHA80 series, where N1-methyl was
usually associated with higher potency than N3-methyl for
several N-methyl-2-aminoimidazopyridine compounds [Hatch
et al., 2001].
A low dipole moment is characteristic of the three more
potent isomers, all of which have N3-methyl groups and b-
face, and to a lesser extent of the remaining two less potent
c-face N3-methyl isomers. This signifies a smaller asymme-
try of atomic charge distribution. Separately including the
three Cartesian components of the dipole moment (using a
standard Cartesian reference frame for all isomers) in the
regression did not provide any additional relation to
potency; however, the TMIPs can be classified according
to the direction of their dipole moment vector. The dipole
moments of the N1-methyl-b-Face isomers average 6.51 D
and lie on average 24.78 clockwise from a reference line
drawn along the bond shared by the imidazole and pyridine
rings (and oriented with the exocyclic amino group at the
right [as in Table I]). The N1-methyl-c-Face isomers have
an average dipole magnitude of 6.68 D and point 47.78
In summary, our results indicate that the most impor-
tant determinants of relatively high mutagenic potency in
the TMIP amines are: (1) a small dipole moment, (2) the
combination of b-face and N3-methyl group, (3) a lower
calculated energy of the p electron system, (4) a smaller
energy gap between the amine HOMO and LUMO, and
(
[
5) a more stable nitrenium ion by the Ford and Griffin
1992] calculation.
DISCUSSION
Observations
An intriguing characteristic of the TMIP isomers is that clockwise from this line. The most mutagenic N3-methyl-
the more potent six members have the pyridine nitrogen b-Face subset average 2.99 D and 124.78 clockwise, and
atom at the 4-position (b-face ring fusion) and the less the N3-methyl-c-Face isomers average 5.72 D and 120.08
potent six at the 5-position (c-face ring fusion). The clockwise. Thus, the positions of the N-methyl substituent

QSAR of TMIP Isomers
143
and the pyridine N-atom affect both the magnitude and ori- between them (Fig. 3). Thus, the shift of N4 to the N5
entation of the dipole moment. The position of the pyridine position increases the p-energy of the aromatic system
N-atom (Face) also obviously affects whether methyl sub- significantly. A combination of Ep_b and DipoleNH2 is
stitutions can occur at positions 4 or 5.
capable of segregating the isomers into subgroups both by
The position of the pyridine ring N-atom at the 4 (b- Face and Nme (Fig. 3).
face fusion) or 5 (c-face fusion) position is strongly cor-
related with potency, in that all six N4 isomers are
Mechanistic Interpretations
more potent than the five measured N5 isomers. The N4
isomers correspond to the configuration found in all
This study is focused on finding relationships between
known thermic mutagens containing a pyridine ring that the observed structural and calculated electronic proper-
have been isolated from cooked foods. Furthermore, b- ties of the TMIP mutagens and their measured mutagenic
vs. c-face fusion is correlated positively with LogP_GC potencies. These properties presumably relate to the meta-
and SoftNH2, and inversely with Ep_b. Thus, the fusion bolic activation process that is required to convert the
mode is related to a constellation of quantum chemical promutagenic TMIPs into active mutagens whose potency
parameters that can contribute to an hypothesis of chemical is measured by the Ames/Salmonella assay. However, one
mechanism.
must be aware that the assay outcome includes several
The calculated LogP values produce interesting but dis- additional potential sources of variation for which data
parate results in terms of relationships to each other and are not available: differential distribution or detoxification
to other predictors. The classical fragment-hydrophobicity of the different isomers or their metabolic intermediates,
method LogP_HL of Hansch and Leo [1979] is not signif- interaction of the activated mutagen with DNA, structure-
icantly correlated with potency (r ¼ 0.54), but is signifi- specific distortion of the DNA helix by bulky adducts and
cantly correlated with LogP_GC (r ¼ 0.68) [Ghose et al., its effect on mutation, and any residual repair capacity of
1
988]. However, LogP_HL is strongly correlated with the repair-defective assay organism.
SoftNH2 (r ¼ 0.91), moderately with Ep_b, and Face.
LogP_GC is significantly correlated with potency (r ¼ there has been no isolation or direct spectroscopic obser-
.74), strongly correlated with Ep_b (r ¼ ꢀ0.97) and vation of a nitrenium form of the thermic mutagens,
In chemical systems related to the thermic mutagens,
0
Face (r ¼ 0.96), and moderately with SoftNH2 (r ¼ although there is indirect evidence, such as reactant selec-
0
1
.75). This method is connectivity-based [Ghose et al., tivity and reasonably long measured lifetimes for related
4
ꢀ1
988], but upon examination appears to create many tiny species (e.g., up to 2 3 10 sec for the 2-fluorenyl-
fragments consistent with a refinement of the Hansch-Leo nitrenium ion) (Anderson and Falvey [1993], and see
method. The modest correlation of LogP_GC with Novak and Rajagopal [2001] for review). Such a product
LogP_HL and the different correlation relationships of the should be very reactive and prone to interaction with
two methods with other predictors signify important dif- scavengers such as SH compounds or free radicals that
ferences in what they actually measure, at least when would divert the ion from a mutagenic pathway. This
applied to the TMIP isomers. These differences should be would also be true for an N-ester with an excellent leav-
noted when evaluating the hydropathic properties of ing group. Furthermore, the chemistry involved in the for-
chemicals or drugs.
SoftNH2 is defined here as the reciprocal of Pearson’s ble to a concerted S 2 reaction (bimolecular nucleophilic
mation of the bulky DNA adduct may be equally amena-
N
Hardness, which makes the correlation with mutagenic substitution in which a DNA base nucleophile attacks an
potency of the TMIPs positive (i.e., greater SoftNH2 cor- electrophilic center, forcing out a leaving group in a con-
responds to greater mutagenic potency). Although moder- certed reaction) or to the putative stepwise S 1 pathway
N
ately correlated with potency (r ¼ 0.64), SoftNH2 is (unimolecular nucleophilic substitution in which a leaving
rather well correlated with LogP_HL, Ep_b, and Face; ester group departs resulting in formation of an electro-
and moderately with LogP_GC; and strongly but inver- philic nitrenium ion, followed by attachment of a DNA
þ
sely with FordNH _NHOH. A high SoftNH2 implies a base nucleophile in a two-step reaction). This activation
narrow HOMO-LUMO gap in the parent amine. The fur- step may take place near the DNA molecule. For our pur-
ther correlation of SoftNH2 with two of the LogP calcu- pose of developing a QSAR for mutagenicity, the use of
lations is consistent with Softness, facilitating the cross- the hypothetical nitrenium ion seems suitable, as this
ing of the cell membrane and/or access to membrane- structure can either be an intermediate in an S 1 pathway
N
bound CYP450 enzymes.
or a surrogate for a transition state in an S 2 pathway.
N
Ep_b represents the simple H u¨ ckel calculation of total However, the conventional nitrenium ion structure with the
p-electron energy, in b units. EB_b is correlated inversely positive charge at the exocyclic N-atom, which we used in
strongly with Face and LogP_GC, moderately with po- our Ford calculations, is now believed to be correctly rep-
tency (r ¼ ꢀ0.73), and SoftNH2. EB_b completely se- resented by a canonical imine structure with the charge
gregates b-face and c-face isomers, with a sizable gap distributed on the proximal ring [Novak and Lin, 1999].

144
Knize et al.
Therefore, this type of calculation may be more appropri- an isomer or its ion into the active site of a CYP450 1A
ately done in accordance with Scheme 1 of that reference, enzyme catalyzing an early activation step, or the active
which then produces an hydrated derivative in solution. site of an ester transferase enzyme catalyzing a late acti-
However, the correct imine-type structures for the two-ring vation step. There is a strong suggestion from organic
TMIP molecules (with a five-membered proximal ring) chemical mechanisms that the key properties tune each
would have to be determined. [Note that the quantum che- TMIP isomer to its own propensity for kinetic and/or
mical methods will automatically form the lowest energy thermodynamic reactivity in the activation reactions of
electronic structure (nitrene, imine, etc.) for any given the Ames/Salmonella assay. This study has presented re-
atomic connectivity].
gression models for the mutagenic potency of the 12
All of the predictor variables analyzed here are well cor- unique isomers of TMIP using a variety of structural,
related with the mutagenic potency of the TMIP isomers: electronic, and hydropathic properties. Such a study is
DDipole, qNmeNH2, DipoleNH2, Face, LogP_GC, Ep_b, inherently limited by the choice and predictability of the
þ
FordNH _NHOH, and SoftNH2 in descending order of regression properties, although this study has reduced one
pairwise correlation coefficients. Within this series of pre- typical source of uncertainty in QSAR studies—the error
dictors, there are many significant pairwise correlations in the predictor variables—by using accurate ab initio
affecting multicollinearity, but the challenge remains to quantum chemical methods to predict most of the proper-
elucidate the mechanistic insight provided by these predic- ties used in the regression. This study produced a number
tors. As in all studies of enzyme-mediated pathways, an of regression models that reasonably account for the ex-
important hypothesis is that the most potent compounds dif- perimentally measured mutagenic potencies of the TMIP
ferentially bind into the active site based on idiosyncratic isomers. Finally, these regression models predict rather
structural and electronic properties of the substrate. Clearly, consistently the potency of the single unassayed isomer.
several of the predictors of mutagenic potency, such as the We await the synthesis of sufficient material for the assay,
dipole moment of the parent amine (DipoleNH2), the loca- and hope for confirmation of the QSAR.
tion of the pyridine endocyclic nitrogen (Face), and
LogP_GC (affecting access to membrane-bound CYP450
enzymes), could plausibly affect the binding to the enzyme
ACKNOWLEDGMENTS
active site. The correlations to other properties, such as the
energy of the p electron system (Ep_b) and the reaction
The authors thank Rebekah Wu for performing the
Ames/Salmonella tests, Dr. Felice Lightstone for helpful
discussion of quantum chemistry, and Dr. John Figueras
for modification of his HMO software.
energy to form the nitrenium ion from the N-hydoxy spe-
cies, are intriguing because they suggest that differences in
the reactivity of the different isomers may also contribute
to the observed differences in mutagenic potency. Hence,
other compounds with the same binding affinity to the
enzyme, but with different electronic properties, may
exhibit widely varying mutagenic potency.
REFERENCES
Adamson RH, Farb A, Virmani R, Snyderwine EG, Thorgeirsson SS,
Takayama S, Sugimura T, Dalgard DW, Thorgeirsson UP. 1995.
Studies on the carcinogenic and myocardial effects of 2-amino-3-
methylimidazo [4,5-f] quinoline (IQ) in nonhuman primates. In:
Adamson RH, Gustafsson J-A, Ito N, Nagao M, Sugimura T,
Wakabayashi K, Yamazoe Y, editors. Heterocyclic Amines in
Cooked Foods: Possible Human Carcinogens. (23rd Intl Symp
Princess Takamatsu Cancer Res Fund) Princeton, NJ: Princeton
Scientific Publishing Co. Inc. pp 260–267.
CONCLUSIONS
The objective of this article has been to apply QSAR
technology to a series of mutagenic and potentially carcino-
genic heterocyclic amine isomers. We have attempted to
obtain some insights into the chemical features and calcula-
ble parameters that contribute to their significant variation
in mutagenic potency. The findings may be limited in their
applicability to the isomers and may not be directly relevant
to molecules of greater structural diversity.
Ahlberg LA, Danthi SN, Tanga MJ. 2000. Synthesis of the potential food
mutagen 1,6,7-trimethyl-2-aminoimidazo[4,5-b]pyridine. American
chemical Society 219:376 [Abstract]
Ames BN, Lee FD, Durston WE. 1973. An improved bacterial test sys-
tem for the detection and classification of mutagens and carcino-
gens. Proc Natl Acad Sci USA 70:782–786.
It seems amazing that a dozen isomers of nearly identi-
cal structure can display a 634-fold variation in mutagenic
potency that is probably based on subtle differences in
properties, most of which can only be calculated by use
of quantum chemical theory. In addition to the calculated
variables, unmeasured variables such as hyperconjugation
from methyl substituents, variation in surface hydropathy,
or details of molecular shape may influence the fitting of
Ames BN, McCann J, Yamasaki E. 1975. Methods for detecting carcino-
gens and mutagens with the Salmonella/mammalian microsomal
mutagenicity test. Mutat Res 31:347–364.
Anderson GB, Falvey DE. 1993. Photogenerated arylnitrenium ions:
absorption spectra and absolute rate constants for tert-butyl(4-
halo-2-acetylphenyl)nitrenium ions measured by time-resolved
laser spectroscopy. J Am Chem Soc 115:9870–9871.
Arulmozhiraja S, Morita M. 2004. Structure-activity relationships for the
toxicity of polychlorinated dibenzofurans: approach through den-

QSAR of TMIP Isomers
145
sity functional theory-based descriptors. Chem Res Toxicol 17: Hatch FT, Colvin ME. 1997. Quantitative structure-activity (QSAR) rela-
3
48–356.
tionships of mutagenic aromatic and heterocyclic amines. Mutat
Res 376:87–96.
Borgen E, Solyakov A, Skog K. 2001. Effects of precursor composition
and water formation of heterocyclic amines in meat model sys- Hatch FT, Knize MG, Felton JS. 1991. Quantitative structure-activity
tems. Food Chem 74:11–19.
relationships of heterocyclic amine mutagens formed during the
cooking of foods. Environ Mol Mutagen 17:4–19.
Case RAM, Hosker M-E, McDonald DB, Pearson JT. 1954. Tumors of
the urinary bladder in workmen engaged in the manufacture and Hatch FT, Knize MG, Moore DH, II, Felton JS. 1992. Quantitative cor-
use of certain dyestuff intermediates in the British chemical industry,
Parts I, II. Br J Ind Med 11:75, 213
relation of mutagenic and carcinogenic potencies for heterocyclic
amines from cooked foods and additional aromatic amines. Mutat
Res 271:269–287.
Colvin ME, Hatch FT, Felton JS. 1998. Chemical and biological factors
affecting mutagen potency. Mutat Res 400:479–492.
Hatch FT, Colvin ME, Seidl ET. 1996. Structural and quantum chemical
factors affecting mutagenic potency of aminoimidazo-azaarenes.
Environ Mol Mutagen 27:314–330.
Einisto P, Watanabe M, Ishidate M, Nohmi T. 1991. Mutagenicity of 30
chemicals in Salmonella typhimurium strains possessing different
nitroreductase or O-acetyltransferase activities. Mutat Res 259: Hatch FT, Knize MG, Colvin ME. 2001. Extended quantitative structure-
9
5–102.
activity relationships for 80 aromatic and heterocyclic amines:
structural, electronic, and hydropathic factors affecting mutagenic
potency. Environ Mol Mutagen 38:268–291.
Felton JS, Knize MG, Wood C, Wuebbles BJ, Healy SK, Stuermer DH,
Bjeldanes LF, Kimble BJ, Hatch FT. 1984. Isolation and charac-
terization of new mutagens from fried ground beef. Carcinogene-
sis 5:95–102.
Jensen F. 1999. Introduction to Computational Chemistry. New York:
Wiley. p 353.
Felton JS, Wu R, Knize MG, Thompson LH, Hatch FT. 1995. Hetero- Kim D, Guengerich FP. 2004. Selection of human cytochrome P450 1A2
cyclic amine mutagenicity/carcinogenicity: influence of repair,
metabolism, and structure. In: Adamson RH, Gustafsson J-A, Ito
mutants with enhanced activity for heterocyclic amine N-hydrox-
ylation. Biochemistry 43:981–988.
N, Nagao M, Sugimura T, Wakabayashi K, Yamazoe Y, editors. Livingstone DJ, Salt DW. 2005. Judging the significance of multiple lin-
Heterocyclic Amines in Cooked Foods: Possible Human Car- ear regression models. J Med Chem 48:661–663.
cinogens. (23rd Intl Symp Princess Takamatsu Cancer Res Maron D, Ames BN. 1983. Revised methods for the Salmonella test.
Fund) Princeton, NJ: Princeton Scientific Publishing Co. Inc. pp
0–58.
Felton JS, Knize MG, Hatch FT, Tanga MJ, Colvin ME. 1999. Heterocy-
Mutat Res 113:173–215.
5
Moore D, Felton JS. 1983. A microcomputer program for analyzing
Ames test data. Mutat Res 119:95–102.
clic amine formation and the impact of structure on their mutage- Novak M, Lin J. 1999. Correlation of azido/solvent selectivities for nitre-
nicity. Cancer Lett 143:127–134.
nium ions with ab initio hydration energies: understanding the
kinetic lability of nitrenium ions in aqueous solution. J Org Chem
64:6032–6040.
Ford GP, Herman PS. 1991. Comparison of the relative stabilities of
polycyclic aryl nitrenium ions and arylmethyl cations: ab initio
and semiempirical molecular orbital calculations. J Mol Struct Novak M, Rajagopal S. 2001. N-arylnitrenium ions. In: Tidwell TT,
THEOCHEM) 236:269–282. Richard JP, editors. Adv Phys Org Chem 36:167–253.
Ford GP, Griffin GR. 1992. Relative stabilities of nitrenium ions derived Novak M, Xu L, Wolf RA. 1998. Nitrenium ions from food-derived het-
(
from heterocyclic amine food carcinogens: relationship to muta-
genicity. Chem Biol Interact 81:19–33.
erocyclic arylamine mutagens. J Am Chem Soc 120:1643–1644.
Pais P, Salmon CP, Knize MG, Felton JS. 1999. Formation of mutagenic/
carcinogenic heterocyclic amines in dry-heated model systems,
meats, and meat drippings. J Agric Food Chem 47:1098–1108.
Pearson RG. 1963. Hard and soft acids and bases. J Am Chem Soc
85:3533–3539.
Ford GP, Herman PS. 1992. Relative stabilities of nitrenium ions derived
from polycyclic aromatic amines: relationship to mutagenicity.
Chem Biol Interact 81:1–18.
Fuscoe JC, Wu R, Shen NH, Healy SK, Felton JS. 1988. Base change anal-
ysis of Salmonella his gene revertant alleles. Mutat Res 201: Pearson RG, Songstad J. 1967. Application of the principle of hard and soft
41–251. acids and bases to organic chemistry. J Am Chem Soc 89:1827.
Ghose AK, Pritchett A, Crippen GM. 1988. Atomic physicochemical Reed AE, Weinstock RB, Weinhold F. 1985. Natural population analy-
parameters for three dimensional structure directed quantitative sis. J Chem Phys 83:735–746.
structure-activity relationships III: modeling hydrophobic interac- Rehn L. 1895. Blasengeschw u¨ lste bei fuchsinarbeitern. Arch Klin Chir
tions. J Comput Chem 9:80–90. 50:588.
Glantz SA, Slinker BK. 1990. Primer of Applied Regression and Ana- Sabbioni G, Wild D. 1992. Quantitative structure-activity relationships
2
lysis of Variance. New York: McGraw-Hill. pp 181–236, 262–
67.
of mutagenic aromatic and heteroaromatic azides and amines.
Carcinogenesis 13:709–713.
2
Guengerich FP. 1995. Human cytochrome P450 enzymes. In: Ortiz de
Montellano PR, editor. Cytochrome P450: Structure, Mechanism,
and Biochemistry. 2nd edition. New York: Plenum. pp 473–
Sasaki JC, Fellers RS, Colvin ME. 2002. Metabolic oxidation of carcino-
genic arylamines by P450 monooxygenases: theoretical support
for the one-electron transfer mechanism. Mutat Res 506:79–89.
Schut HAJ, Snyderwine EG. 1999. DNA adducts of heterocyclic amine
food mutagens: implications for mutagenesis and carcinogenesis.
Carcinogenesis 20:353–368.
535.
Guengerich FP, Humphreys WG, Yun C-H, Hammons GJ, Kadlubar FF,
Seto Y, Okazaki O, Martin MV. 1995. Mechanisms of cyto-
chrome P450 1A2-mediated formation of N-hydroxy arylamines Sugimura T. 1995. History, present and future, of heterocyclic amines,
and heterocyclic amines and their reaction with guanyl residues.
In: Adamson RH, Gustafsson J-A, Ito N, Nagao M, Sugimura T,
Wakabayashi K, Yamazoe Y, editors. Heterocyclic Amines in
Cooked Foods: Possible Human Carcinogens. (23rd Intl Symp
Princess Takamatsu Cancer Res Fund) Princeton, NJ: Princeton
Scientific Publishing Co. Inc. pp 78–84.
cooked food mutagens. In: Adamson RH, Gustafsson J-A, Ito N,
Nagao M, Sugimura T, Wakabayashi K, Yamazoe Y, editors. Het-
erocyclic Amines in Cooked Foods: Possible Human Carcinogens.
(23rd Intl Symp Princess Takamatsu Cancer Res Fund) Princeton,
NJ: Princeton Scientific Publishing Co. Inc. pp 214–231.
Tanga MJ, Bupp JE, Tochimoto TK. 1994. Syntheses of 1,5,6-trimethyl-
2-aminoimidazo[4,5-b]pyridine and 3,5,6-trimethylimidazo[4,5-b]
pyridine. J Heterocycl Chem 31:1641–1645.
Hansch C, Leo A. 1979. Substituent Constants for Correlation Analysis
in Chemistry and Biology. New York: Wiley.

146
Knize et al.
Tanga MJ, Bupp JE, Tochimoto TK. 1997. Syntheses of five potential het- Vracko M, Szymoszek A, Barbieri P. 2004. Structure-mutagenicity study
erocyclic amine food mutagens. J Heterocycl Chem 34:717–
27.
Tanga MJ, Bradford WW, Bupp JE, Kozocas JA. 2003. Syntheses of
of 12 trimethylimidazopyridine isomers using orbital energies and
‘‘spectrum-like representation’’ as descriptors. J Chem Inf Com-
put Sci 44:352–358.
7
two potential food mutagens.
73.
J
Heterocycl Chem 40:569– Watanabe M, Ishidate M, Jr, Nohmi T. 1990. Sensitive method for
the detection of mutagenic nitroarenes and aromatic amines: new
5
Topliss JG, Edwards RP. 1979. Chance factors in studies of quantitative derivatives of Salmonella typhimurium tester strains possessing
structure-activity relationships. J Med Chem 22:1238–1244. elevated O-acetyltransferase levels. Mutat Res 234: 337–348.
Turesky RJ, Guengerich FP, Guillouzo A, Langou e¨ t. 2002. Metabolism Weisburger JH. 2002. Comments on the history and importance of aro-
of heterocyclic amines by human hepatocytes and cytochrome
P4501A2. Mutat Res 506/507:187–195.
matic and heterocyclic amines in public health. Mutat Res 506/
507:9–20.