Mutagenesis vol.7 no.5 pp.335-341, 1992

Relationships between in vitro mutagenicity assays

Romualdo Benigni Laboratory of Comparative Toxicology and Ecotoxicology, Istituto Superiore di Saniti, Rome, Italy

Introduction In the 1970s and 1980s, the short-term tests (STTs) for mutagenicity underwent vigorous development; the number of new assays has progressively increased and an enormous range of different chemicals have been studied for their ability to induce genetic damage. In this research, an invaluable role is played by the comparative studies, where different assays are studied with a common set of rationally selected chemicals, thus permitting a direct comparison of the STT responses. Recently, the US National Toxicology Program (NTP) published the experimental results relative to a further 41 chemicals studied with the four in vitro STTs (Zeiger et al., 1990). The main concern of this study was the ability of the STTs to predict carcinogenicity. The study confirmed the results of the previous NTP study with 73 chemicals (Tennant et al., 1987): no battery of tests increased the ability of Salmonella alone to predict the rodent carcinogenicity; the chemicals positive in Salmonella have a high probability of being rodent carcinogens; no prediction about carcinogenicity or non carcinogenicity could be inferred from a negative Salmonella result. © Oxford University Press

Data and methods Database The experimental results of 41 chemicals studied with four in vitro STTs were generated by the US NTP (NTP41) (Zeiger et al., 1990). The other experimental results considered in this paper were produced by the US National Toxicology Program (NTP73) (Tennant et al., 1987), the International Program for the Evaluation of Short-Term Tests for Carcinogens (IPESTTC) (De Serres and Ashby, 1981), the International Program on Chemical Safety (IPCS) (Ashby et al., 1985) and Gene-Tox [for this database, the compilation of Palajda and Rosenkranz (1985) was used]. Compilations of the original data can also be found in Benigni (1986, 1989) and Benigni and Giuliani (1985). The classification of results into positive, negative or borderline was based on the evaluations of the original papers. Chart 1 reports the mutagenicity data generated within the framework of NTP73 and NTP41. The data are displayed according to the results of a multivariate analysis (see rationale and technical details in the text). Statistical analyses Principal component analysis (PCA), which belongs to the family of factorial analyses, was used to analyze and graphically display the data. The principal components (unrotated factors) were calculated with the BMDP4M program, from the BMDP Statistical Package Software (Dixon, 1981). An excellent introduction to PCA and other multivariate statistical analysis techniques can be found in Lebart et al. (1984). Rationale and details about the application of multivariate methods to genotoxicity data are presented in Benigni and Giuliani (1988a, 1991b). In the latter two papers, the transformation of qualitative into quantitative data—via calculation of Hamming distances—is also presented. The Hamming distance (Lee, 1981) is a measure of the dissimilarity between two objects based on the descriptors of interest. The Hamming distance between two cases A and B defined by binary variables is the following: (number of variables with different values in A and B)/(number of variables with non-missing values in both cases A and B). The equivocal results were also used to estimate Hamming distances between assays, with the difference between two variables considered 0.5 instead of 1.0.

Results and discussion Comparison of the 41 chemical NTP database with the other databases To compare the new NTP41 database with IPESTTC, IPCS, Gene-Tox and NTP73, a mathematical approach that permits a global comparison between databases including the same set of assays has been used (Benigni and Giuliani, 1991a). This approach also permits the distinction between the indications which have a general validity and those which are specific to the individual databases. The relationships between the four assays [Salmonella typhimurium (STY), Chromosomal aberrations in Chinese hamster ovary (CHO) cells (CHA), Sister chromatid exchange in CHO cells (SCE), mutation in L5178Y mouse lymphoma cells (MLY)] in each database were determined by calculating their Hamming distances. Table I gives the test distances in the NTP41 and in the whole range of NTP chemicals (73 + 41 chemicals). 335

Downloaded from http://mutage.oxfordjournals.org/ at Pennsylvania State University on May 10, 2012

This paper analyzes the mutagenicity results reported by the US National Toxicology Program (NTP), relative to 41 chemicals assayed with four in vitro short-term tests [Salmonella typhimurium (STY), Chromosomal aberrations in Chinese hamster ovary (CHO) cells (CHA), Sister chromatid exchange in CHO ceUs (SCE), mutation in L5178Y mouse lymphoma cells (MLY)] and puts this database in perspective with respect to other databases. It is shown that the test relationships pointed out by the experiments on the 41 chemicals are in substantial agreement with those indicated by a previous NTP report on 73 chemicals, and that the same test relationships were also indicated by the results on the International Program for the Evaluation of Short-Term Tests for Carcinogens (IPESTTC). The NTP and IPESTTC databases consistently indicated that there is a gradual increase in the sensitivity to the genotoxins in the following order: STY < CHA < SCE < MLY. On this scale, SCE and MLY show a great degree of similarity of responses to the chemicals, as does STY with CHA. The overall evidence provided by these results, and by the general pattern of IPESTTC and NTP genotoxicity profiles, does not support the notion that the genotoxic chemicals have genetic end point specificity. Moreover, a mathematical simulation analysis demonstrated that MLY and SCE—the two most sensitive assays of those studied by NTP—are not more subject to erratic results than other assays, and that they form—together with STY and CHA—a consistent family of genotoxicity assays.

The present paper analyzes a different aspect of the data on the 41 NTP chemicals: the relationships between assays. In particular, it examines (i) the global consistency of the results of the 41 chemical NTP database with those of the other databases, (ii) the similarities between the responses of the tests to the chemicals and (iii) the genotoxicity profiles of the chemicals. All these investigations were performed in a quantitative way, with multivariate data analysis techniques.

R.Benigni RELATIONSHIPS BETWEEN ASSAYS Table I. Hamming distance matrices NTP41 STY MLY

CHA

SCE

NTP STY

MLY

CHA

SCE

NTP CHA

MLY SCE STY MLY CHA SCE

0.000 0.512 0.268 0.427

0.000 0.488 0.256

0.000 0.232

0.000

0.000 0.399 0.268 0.430

STY A

A A

0.000 0.386 0.206

H0.000 0.303

-.5

0 .5 Component 1 scores (69%)

0.000

Hamming distances between STTs based on (i) the subset of 41 chemicals reported by Zeiger et al. (1990; NTP41) and (ii) the whole NTP database (NTP). A Hamming distance between two assays measures the dissimilarity of their responses to the chemicals.

NTP73 SCE

CHA

MLY

A

A A -.5

STY A

.5 Component 1 scores (69%)

Table II. Correlation coefficients between databases

NTP41

NTP

1.000 0.523 0.898 0.628 -0.014 0.465

1.000 0.845 0.825 0.287 0.422

1.000 0.821 0.141 0.511

IPESTTC

IPCS

Gene-Tox NTP41 MLY

SCE

A

1.000 -0.112 520

1.000 0.430

-.5 1.000

The distance matrices relative to the various databases under study were compared by calculating the correlation coefficients between each pair of them. These coefficients measure the global similarities of the databases, i.e. how similar the STT inter-relationship patterns are in the various databases.

CHA

STY

A

A

0 .5 Component 1 scores (65%)

IPESTTC MLY

SCE

CHA A

A Component 2

H-

1-

A IPCS

STY A

.5

-.5

Component 1 scores

(50%)

Fig. 2. Relationships between assays in four databases. The relationships are measured by the Component 1 scores (see details in the text). . 5A Gene-Tox

A NTP41 0A NTP NTP73

IPESTTC -.5-

-.5

Component 1 Fig. 1. Map of the similarities between databases. It displays the loadings obtained by PCA of Table II data. The percentages of original variance explained by Components 1 and 2 are 61 and 21%, respectively.

Test distance matrices for the other databases were reported in Benigni and Giuliani (1991a). The various distance matrices were compared to each other by calculating their correlation coefficients (Table II). Figure 1 shows the relationships between the databases: the display was obtained by PCA of Table II data. NTP41 is closehence similar—to NTP, NTP73 and IPESTTC, whereas IPCS and Gene-Tox express test inter-relationships different from those defined by the other two databases. Gene-Tox includes many 336

chemicals assayed before 1980, a period in which the chemicals were selected mainly because of suspicions concerning their genetic activity, and actually includes mostly positive results. IPCS essentially consists of eight carcinogens selected because they were negative in Salmonella. Thus, the latter two databases are biased towards specific goals and are not aimed at being representative of the whole universe of chemicals (as are NTP and IPESTTC). This feature was correctly demonstrated by the distances between databases in Figure 1. It should be noted that Gene-Tox is intermediate between IPCS and the cluster of other databases. Figure 1 also indicates a similarity between the two subsets of the NTP data base, as observed by Zeiger et al. (1990), which showed a similarity in terms of prediction of rodent carcinogenicity. The similarity between NTP41, NTP73, NTP and IPESTTC was investigated in more detail; to do this, the relationships between the four assays in each database were studied with PCA. Figure 2 shows the relationships between assays described by Component 1: it should be recalled that the first component always represents the most important effect underlying the data (Linsker, 1988). In the NTP41 two components were obtained, explaining 65 and 31 % of the variance, respectively. In the case of NTP73, PCA generated only one component, explaining 69% of the variance; also the analysis of the whole NTP pointed to only one component, explaining 69% of the variance. The analysis of IPESTTC generated two components: Component 1 explained 50% of the variance. The results in Figure 2 indicate that the relationships between the four assays—in their major

Downloaded from http://mutage.oxfordjournals.org/ at Pennsylvania State University on May 10, 2012

NTP73 NTP41 NTP IPESTTC IPCS Gene-Tox

NTP73

Relationships between in vitro mutagenicity assays

trends—can be represented on a linear scale. Moreover, MLY and SCE give similar profiles of responses to the chemicals (they are close to each other in the components), and so do STY and CHA.

Are the results of the most sensitive tests (MLY and SCE) consistent with those of the other genotoxicity assays? The Salmonella assay is commonly accepted as the first test in a testing strategy, thus constituting a kind of milestone for the

Conclusions The test relationships indicated by the NTP and IPESTTC give rise to the difficult question of how to explain these interrelationships in terms of biological mechanisms. The hypothesis that genotoxic chemicals act through mechanisms specific for the 337

Downloaded from http://mutage.oxfordjournals.org/ at Pennsylvania State University on May 10, 2012

Mutagenicity profiles The relationships between the mutation profiles induced by the 114 NTP chemicals were studied with PC A [a detailed presentation of the procedure for the analysis of mutation profiles is given in Benigni (1992)]. The analysis demonstrated that most of the differences between genetic damage profiles can be represented by just one linear scale (i.e. Component 1, explaining 62% of the variance). The other components explained 24, 11 and 3% of the variance, respectively. Chart 1 presents the NTP data with the chemicals placed according to the order of Component 1 scores; the tests have been ordered according to their similarities (Figure 2). It is evident from Chart 1 that STY and MLY are the tests with the most divergent results. Moreover, SCE gives the same results as MLY for most of the chemicals, whereas CHA has a considerable degree of similarity with STY. It is also apparent that the chemicals positive in all assays are at one end and the chemicals homogeneously negative fall at the other end. Between these two poles, there is an ordered grading of the genotoxic properties of the chemicals. A considerable group of chemicals is negative in STY and positive in the other three assays; another large group is negative in STY and CHA, and positive in SCE and MLY. A small proportion of the chemicals present peculiar genotoxicity profiles, which do not fall into these major classes. It should be stressed that this quantitative modulation of genotoxicity is the most important underlying effect (it corresponds to 62% of variance), whereas genetic damage profile specificities are not apparent. Chart 2 displays the genotoxicity profiles relative to IPESTTC results for 20 tests. Both tests and chemicals have been ordered according to the similarity scales provided by PCA (analytical details not shown). Chart 2 also reports the re-partition of the assays into three clusters of systems characterized by homogeneous responses to the chemicals (Benigni and Giuliani, 1985). The similarity between the patterns shown in Charts 1 and 2 is evident. In Chart 2, as in Chart 1, the main differences between tests are quantitative and not qualitative; similarly, the differences between genotoxicity profiles are—in their main trendquantitative. There are chemicals positive in most assays, then chemicals positive in the two groups of in vitro systems, then other chemicals positive especially in the cluster of the most sensitive in viro tests and, lastly, chemicals negative in all assays. Whereas the traditional picture of genotoxicity is based mainly on the idea of qualitative differences, according to which the chemicals specifically affect the various genetic end points, the insight provided by the IPESTTC and NTP experimental data is contrary to this. The genotoxicity seems to be a unitary property of the chemicals and is quantitatively modulated from much to little to nothing. Likewise, the tests can be divided into groups of systems highly sensitive to genotoxicity—as some of the in vitro assays—and other less sensitive to genotoxicity, until the extreme situation of the in vivo assays which are sensitive only to the most genotoxic chemicals.

in vitro STTs. The question may then arise as to whether the MLY- and SCE-positives, which are STY- (and CHA-) negatives, should be considered as truly genotoxic, or should be considered as erratic results, whose genotoxic character would be questionable. This problem is important not only from a basic research point of view, but also from a regulatory point of view. For example, the recently issued US Environmental Protection Agency testing guidelines for assessing the potential mutagenicity of a chemical included MLY within the proposed initial battery of assays, instead of complementing STY with the traditionally preferred CHA (Dearfield et al., 1991). In this paper, the degree of consistency of MLY and SCE results with the other assays results, was investigated by simulating a test which gives completely random responses to the 114 NTP chemicals. The 'fake' test was constrained to give 65% positive responses: in this way, it mimicked the global performance of MLY and SCE, which gave about 65% positive responses in the NTP. In practice, a computer program sorted various sets of 114 random numbers between 0 and 1 from a uniform distribution. The values below 0.35 were considered as negatives, and the other values as positives. After a series of 114 random results was generated, the Hamming distances between the fake test (i.e. the series of 114 random results) and the four real STTs was calculated. This procedure was repeated 10 000 times. Table HI reports the average distances of the fake test from the real tests; the small standard deviations indicate that the results are very homogeneous and stable. Moreover, a preliminary series of 1000 cycles of random extractions produced the same average distances as the series of 10 000 cycles (results not shown). The average distances of the fake test from the four real assays were used to construct a Hamming distance table, further studied by PCA (Figure 3). Whereas the inter-relationships between the four tests were summarized by one component only (Figure 2), the inclusion of the fake test resulted in a different structure, consisting of two components. This indicates that the results of the four assays are highly consistent: their differences can be represented on one linear scale, since they share to different extents a common property (i.e. the ability to respond to genotoxins). On the contrary, basically different information (an additional component) is represented by the fake test. This result is a demonstration that MLY and SCE belong to the same universe as STY and CHA, whereas a test with erratic results produces profiles of responses which are dramatically different from those of the real genotoxicity tests. Concerning the criticism that in some cases MLY and SCE responded positively to chemicals that apparently should not be mutagenic, it should be recalled that other in vitro systems have also experienced similar problems, especially when nonphysiological conditions are employed (high osmotic pressure, pH, etc). For example, increased chromosomal aberrations have been found due to increased osmotic pressure (Galloway et al., 1987). This simply indicates that a testing procedure cannot be an automatic process, and that an accurate selection of experimental conditions (doses, exposure conditions, metabolic activation system, etc.) is needed, together with a sensible judgement on the results of the experiments (also see Dearfield et al., 1991).

R.Benigni NTP data STY SCE CHA MLV

III • • I

III

Number of chemicals 25

2 12

NTP73:

9,15,23,26,27,29,30,34,35,44, 46,52,54,59,61,68,69 NTP41: 1 , 2 , 9 , 1 2 , 1 7 , 1 9 , 3 0 , 3 1

NTP73: NTP41: NTP73: NTP41:

1 26 2,16,25, 36, 39, 40, 56,58 8,21,29, 39

2

NTP7 3: 8,73

5

NTP73: NTP41: NTP73: NTP41: NTP41:

2 1 3

18

20,21,31 ,45 18 53 4 7

NTP73: 63,72 NTP41: 10 NTP41: 40 NTP73: 13,14,19 ,28 ,42 ,47 ,48,55,64,66, NTP41: 16,22,23 ,25 ,33 ,34 ,35

1

NTP73: 4

1

NTP73: 6

1

NTP41: 11

1

NTP73: 32

2 1

NTP7 3: 41 NTP41: 5 NTP41: 15

1

NTP7 3: 17

3

NTP73: 3,5,33

6 1

NTP73: 7,22 NTP41: 28,32,36 ,41 NTP7 3: 24

3

NTP73: 10,18,50

5

NTP41: 6,14,20, 37, 38

17

NTP73: 11,12,37 ,38 ,43 ,49 ,51,57,60,62, 65,67,71 NTP41: 3,13,24, 27

Chart 1. Results of the 114 chemicals studied by NTP with four in vitro assays. The chemicals are divided according to their profiles of induced genetic damage. The order of both tests and chemicals in the chart was established by PCA (see details in the text) in such a way as to be able to visualize as much as possible the relationships existing in the data. The size of the symbols is only approximately proportional to the number of chemicals showing a given profile. Codes of the chemicals: NTP73 (Tennant el at., 1987): (1) allyl isothiocyanate, (2) allyl isovalerate, (3) 11-aminoundecanoic acid, (4) /-ascorbic acid, (5) benzene, (6) benzoin, (7) benzyl acetate, (8) 2-biphenylamine HC1, (9) bis(2-chloro-l-methylethyl) ether, (10) bisphenol A, (11) butyl benzyl phtalate, (12) caprolactam, (13) chlorobenzene, (14) chlorodibromomethane, (15) 2-chloroethanol, (16) 3-chloro-2-methyl-propene, (17) CI acid orange 10, (18) CI acid red 14, (19) CI acid yellow 73, (20) CI disperse yellow 3, (21) CI solvent yellow 14, (22) cinnamyl anthranilate, (23) cytembena, (24) D & C red 9, (25) diallyl phthalate, (26) 1,2-dibromc-3-chloro-propane, (27) 1,2-dibromoethane, (28) 1,2-dichlorobenzene, (29) 2,6-dichloro-p-phenylenediamine, (30) 1,2-dichloropropane, (31) 1,3-dichloropropene, (32) di(2-ethylhexyl)adipate, (33) di(2-ethylhexyl)phthalate, (34) diglycidyl resorcinol ether, (35) dimethyl hydrogen phosphite, (36) dimethyl morpholinophosphoramidate, (37) dimethyl terephthalate, (38) ethoxylated dodecyl alcohol, (39) ethyl acrylate, (40) eugenol, (41) FD & C yellow no. 6, (42) geranyl acetate; (43) hamamelis water, (44) HC blue 1, (45) HC blue 2, (46) 8-hydroxyquinoline, (47) isophorone, (48) malaoxon, (49) :

••• :r r: 1 7

1 : . ? Li < •:

1

•! :•.;

1 1 •> 1 : •.. ;•• •:• •;• 1 1 1 ii i" ' iii ,; :. s ? r

• .1

•

II

I r i•

i ". i- Li 5' •': I

:: ,'i

• :::

I ?iI 7. r r- ii:" D y

Li

..: .' = = :-i : :

?

'i

-J :> -...

un

fl

&

B iii

::

"

Li b

iii

iii i : :

n

%i

L

b

-J

i", S !? i

J

if !" "•

M

20

I

.'.

j-j ;.)

28

: : t'

: • •

I

III

h Z l i

• .. 21

: :

• .

? • •1

II

23

i '

i l l

• ... •

4

II

.

J

3

1 n i l

I ?

III

I • 11

•

1

11

•

1

in

• : 29

I 1 I I I 13

11 ?

?1:

32

,:

'•'•

1

!

.:

r: •

:.: 3 1

n .i 1 :

1 1•

••

1 =;

=» i . 3 5

h

7

2 !!! V I':

:.l .. ' •? - i

r

2

.;

18

..! :!! f

I'I

22 6

j r.

i, Li

40

"'

M

• r B & • ! ? ri

5

i6

1 1.

••:

v b c r • & H• , ": ? ii

Downloaded from http://mutage.oxfordjournals.org/ at Pennsylvania State University on May 10, 2012

in

I? 37

!:i a £ c- 24

v 0

.;;

i.

i•

L

u

30 38

? = border-line or indeterminate; 1 = positive .-. = egative; n

Chart 2. Results of 20 STTs with 42 chemicals studied by IPESTTC (De Serres and Ashby, 1981). The order of both tests and chemicals in the chart was established by PCA (see details in the text). The tests are divided in clusters characterized by similar responses to the chemicals; the classification was obtained by mathematical methods (Benigni and Giuliani, 1985). Codes of the chemicals: (1) 4-dimethylamino azobenzene, (2) 3,3'-5,5'-tetramethylbenzidine, (3) 4-acetylaminofluorene, (4) 1-naphthylamine, (5) pyrene, (6) chloroform, (7) diethylstilbestrol, (8) 4-nitroquinoline W-oxide, (9) 2-naphthylaimne, (10) 4-dimethylazobenzene-4-sulfonic acid, (11) benzidine, (12) 2-acetylaminofluorene, (13) benzo[a]pyrene, (14) hydrazine sulfate, (15) 3-methyW-nitroquinoline yV-oxide, (16) hexamethylphosphoramide, (17) diphenylnitrosoamine, (18) anthracene, (19) dinitrosopentamethylene tetramine, (20) ethylenethiourea, (21) a-butyrolactone, (22) methionine, (23) safrole, (24) dimethylformamide, (25) DL-ethionine; (26) epichlorydrine, (27) W-nitrosomorpholine, (28) isopropyl-W-(3-chlorophenyl)carbamate, (29) /3-propiolactone, (30) 1,1,1-trichloroethane, (31) urethane, (32) dimethylcarbamoyl chloride, (33) azoxybenzene, (34) cyclophosphamide, (35) 3-aminotriazole, (36) 4,4'-methylenebis(2-chloroaniline) (MOCA), (37) methylazoxymethanolacetate, (38) sugar (sucrose), (39) 9,10-dimethylanthracene, (40) ascorbic acid, (41) o-toluidine, (42) auramine. Codes of assays: transformation BHK21 cells, BHK; S.cerevisiae XV185-14C, SCX; Bacillus subtilis rec~, BSR; E.coli polA, ECP; E.coli 343/113, EC3; UDS human fibroblasts/HeLa cells, UDS; Mutation TK±L5178Y cells, MLY; S.typhimurium his', STY; E.coli WP2, ECW; S.cerevisiae aneuploidy, SCA; S.pombe ode, SPA; chromosome aberrations CHO cells, CHA; E.coli rec~, ECR; SCE CHO cells, SCE; S.cerevisiae mitotic recombination, SCR; Mutation HPRT CHO/V79 cells, HGP; micronucleus mouse, MIC; SCE mouse, SCM; Drosophila melanogaster sex-linked recessive lethals, DRO; sperm morphology mouse, SPM. Hi = negative; ? = border-line or indeterminate; %\ = positive.

339

R.Benigni A Component 2

Table III. Simulation of a test with random responses: Hamming distances from the real tests

STY MLY CHA SCE

Average distance

Standard deviation

0.550 0.438 0.523 0.450

0.044 0.043 0.044 0.042

In light of the above evidence, it is the initial steps (absorption, transport and metabolic transformation) of the long series of events leading to the expression of the genetic damage which appear to be responsible for the observed differences between STT performances. These factors depend on the type of cells (e.g. nature of the cell wall or membrane, endogenous metabolizing enzymes, etc.); however, the experimental conditions used (incubation medium, osmolality, pH, exogenous metabolism, etc.) combine with and superimpose themselves on 340

A

A

SCE

. 5-

oA CHA A

Fake A STY

-.5-

-.5

.5

Component 1 Fig. 3. The figure displays the result of an experiment in which a test (Fake), with random responses to the chemicals, was simulated. This figure reports the loadings obtained by PCA of the distances between assays. The simulated assay is clearly far away (different) from the real assays. The percentages of original variance explained by Components 1 and 2 are 47 and 37%, respectively.

die 'intrinsic' characteristics of the cell systems. This suggests that each STT is characterized by its specific pattern of the above properties, should not be included in such general categories as genetic end point, etc., but should be classified on the basis of its actual responses to the different chemicals. Another important point is the implications of the present results concerning the practical utilization of STTs for risk assessment. These results indicate that MLY and SCE are greatly consistent with the other two genotoxicity assays, and should not be considered as subject to erratic results more than any other genotoxicity assays are. Hence, their indications should not be dismissed as insignificant. Moreover, the present trends in regulation and legislation rule out the chance that, in this way, the number of chemicals considered as genotoxic would increase unreasonably. The tendency now is to consider the in vitro STTs as tools for the primary screening of potential genotoxic chemicals; for the chemicals positive in vitro, a second phase of experimentation should ascertain whether the potential genotoxicity is actually expressed in the in vivo systems (Anonymous, 1989; Carere and Benigni, 1990; Dearfield etal., 1991). The combination of STY with another sensitive assay, such as MLY, in the first phase, is a powerful filter, able to minimize the probability that a genotoxic chemical is missed and, thus, not further scrutinized with the in vivo systems. Acknowledgements I would like to thank Dr Alessandro Giuliani for his helpful discussion and advice. I thank Eve Silvester for her editing of the manuscript. The advice and suggestions of Dr Edward Friedman (New York University Academic Computing Facility) are appreciated.

References Anonymous (1989) Guidelines for the Testing of Chemicals for Mutagenicity. UK Committee on Mutagenicity of Chemicals in Food, Consumer Products and Environment, Department of Health, Her Majesty's Stationery Office, London.

Downloaded from http://mutage.oxfordjournals.org/ at Pennsylvania State University on May 10, 2012

various genetic end points does not have experimental support. The other current hypothesis is that the differences in the performances of the assays should be due mainly to the type of cells used. Any quick scanning of Charts 1 and 2 indicates that this hypothesis, in this simple formulation, is not very helpful. For example, both CHA and SCE are cytogenetic assays based on the same type of cells (CHO cells); however, SCE is more similar to MLY and CHA is more similar to STY (bacterial mutation assay) (Figure 2 and Chart 1). In IPESTTC, the cluster of the most sensitive tests includes assays such as BHK21 cell transformation and S.cerevisiae XV185-14C mutation, which are based on completely different types of cells, whereas Escherichia coli 343/113 mutation and E.coli WP2 mutation (based on the same type of cells) are in two different clusters, since they give different responses to the chemicals (Chart 2). Hence, the explanation of the differences between test performances should be sought elsewhere. There is growing evidence that these differences greatly depend on the chemical absorption into, and transport within, the cells, and on the ability to metabolically transform premutagens into active mutagens. For example, quantitative structure-activity relationship (QSAR) studies performed in Professor Hansch's laboratory have indicated that the crucial step for the Salmonella mutagenic activity of aromatic nitro compounds is the metabolic reduction by one or more cytosolic nitroreductases (Lopez de Compadre et al., 1990) and that the Salmonella mutagenicity of triazenes depends on the same factors that favor hydroxylation by cytochrome P450 (Shusterman et al., 1989). Other interesting evidence was provided by the IPESTTC, where 12 different laboratories used the Ames test. Ten out of 12 laboratories showed a good reproducibility of the results, whereas two laboratories produced response patterns remarkably different from those of the other laboratories: these two laboratories used different experimental protocols, with much greater amounts of S9 metabolizing fraction (Benigni and Giuliani, 1988b). Moreover, a preliminary analysis of the physical chemical properties of the NTP chemicals pointed out a major difference in molar refractivity (MR) between chemicals that tend to be positive in all the four assays, chemicals that tend to be positive mainly in MLY and SCE, and negative chemicals (our unpublished results). Since MR parameterizes the bulkiness of chemicals, this may indicate that the different accessibility to the cellular target structures is a major determinant of the differences between the performances of the four in vitro assays.

MLY

Relationships between in vitro mutagenicity assays

Downloaded from http://mutage.oxfordjournals.org/ at Pennsylvania State University on May 10, 2012

Ashby,J., de Serres.F.J., Draper.M., Ishidate.M., Margolin,B.H., Matter.B.E. and Shelby,M.D. (eds) (1985) Evaluation of short-term tests for carcinogens, report of the International Program on Chemical Safety Collaborative Study on In Vitro Assays. Progress in Mutation Research. Elsevier/North-Holland, Amsterdam, Vol. 5. Benigni,R. (1986) The second international collaborative study on comparative mutagenesis re-visited: analysis by multivariate statistical methods. Mutagenesis, 1, 185-190. Benigni.R. (1989) Analysis of the National Toxicology Program data on the in vitro genetic toxicity tests using multivariate statistical methods. Mutagenesis, 4, 412-419. Benigni,R. (1992) Rational approach to the quantification of genotoxicity. Environ. Mol. Mutagen., 19, 83-89. Benigni,R. and Giuliani.A. (1985) Cluster analysis of short-term tests: a new methodological approach. Mutat. Res., 147, 139-151. Benigm.R. and Giuliani.A. (1988a) Predicting carcinogenicity with short-term tests: biological models and operational approaches. Mutat. Res., 205, 227-236. Benigni.R. and Giuliani.A. (1988b) Computer-assisted analysis of interlaboratory Ames test variability. J. Toxicol. Environ. Health, 1, 135 — 148. Benigni.R and Giuliani.A. (1991a) Simultaneous evaluation of genotoxicity data from different sources: a multivariate statistical approach. Mutat. Res., in press. Benigni,R. and Giuliani,A. (1991b) Multivariate analyses in genetic toxicology. In Devillers.J. and Karcher,W. (eds), Applied Multivariate Analysis in SAR and Environmental Studies. Kluwer, Dordrecht, pp. 347-376. Carere.A. and Benigni.R. (1990) Strategies and governmental regulations. Teratogen. Carcinogen. Mutagen., 10, 199—208. Dearfield.K.L., Auletta,A.E., Cimino,M.C. and Moore.M.M. (1991) Considerations in the US Environmental Protection Agency's testing approach for mutagenicity. Mutat. Res., 258, 259-283. De Serres,F.J. and Ashby.J. (eds) (1981) Evaluation of short-term tests for carcinogens, report of the International Collaborative Program. Progress in Mutation Research. Elsevier/North-Holland, Amsterdam, Vol. 1. Dixon.W.J. (ed.) (1981) BMDP Statistical Software. University of California Press, Berkeley. Galloway,S.M., Deasy.D.A., Bean.C.L., Kraynak.A.R., Armstrong,M.J. and Bradley,M.O. (1987) Effects of high ionic strength on chromosome aberrations, sister-chromatid exchanges and DNA strand breaks, and the relation to toxicity. Mutat. Res., 189, 15-25. Lebart.L., Morineau,A. and Warwick,K.M. (1984) Multivariate Descriptive Statistical Analysis. Wiley, New York. Lee,R.C.T. (1981) Clustering analysis and its applications. In Tou.J.T. (ed.), Advances in Information Science. Plenum Press, New York, pp. 168—292. Linsker.R. (1988) Self-organization in a perceptual network. Computer, March, 105-117. Lopez de Compadre.R.L., Debnath.A.K., Shusterman.A.J. and Hansch.C. (1990) LUMO energies and hydrophobicity as determinants of mutagenicity by nitroaromatic compounds in Salmonella typhimurium. Environ. Mol. Mutagen., 15, 44-55. Palajda,M. and Rosenkranz.H.S. (1985) Assembly and preliminary analysis of a genotoxicity data base for predicting carcinogens. Mutat. Res., 153, 79-134. Shusterman.A.J., Debnath.A.K., Hansch.C., Hom.G.W., Fronczek.F.R., Greene.A.C. and Watkins.S.F. (1989) Mutagenicity of dimethyl heteroaromatic triazenes in the Ames test. The role of hydrophobicity and electronic effects. Mol. Pharmacol., 12, 939-944. Tennant.R.W., Margolin.B.H., Shelby.M.D., Zeiger.E., Haseman.J.K., Spalding.J.W., Caspary.W., Resnick.M., Stasiewicz.S., Anderson,B. and Minor.R. (1987) Prediction of chemical carcinogenicity in rodents from in vitro genetic toxicity assays. Science, 236, 933-941. Zeiger,E., Haseman,J.K., Shelby.M.D., Margolin.B.H. and Tennant.R.W. (1990) Evaluation of four in vitro genetic toxicity tests for predicting rodent carcinogenicity: confirmation of earlier results with 41 additional chemicals. Environ. Mol. Mutagen., 16 (Suppl. 18), 1-14. Received on December 12, 1991; accepted on May II, 1992

341