Environmental and Molecular Mutagenesis 34:297–304 (1999)

Prediction of Rodent Carcinogenicity Utilizing a Battery

of In Vitro and In Vivo Genotoxicity Tests
Byung Soo Kim1* and Barry H. Margolin2
1
Department of Applied Statistics, Yonsei University, Seoul, South Korea
2
Department of Biostatistics, University of North Carolina,
Chapel Hill, North Carolina

The primary purpose of this study is to investigate
the degree to which we can improve the prediction
of rodent carcinogenicity (CA) by combining re-
sults from an in vitro and two in vivo genotoxicity
tests. We used the Ames Salmonella assay (SAL)
for the in vitro test and the micronucleus assay
(MNC) and chromosome aberration assay (ABS) in
mouse bone marrow cells for the two in vivo tests.
We collected complete assay data for 82 chemi-
cals (55 carcinogens and 27 noncarcinogens)
from the NTP data base and the IARC monograph
series. Our results indicate that: (1) only SAL affects
the predictivity of CA, (2) MNC has a strong asso-
ciation with ABS, and (3) SAL predicts ABS. It has
been known for some time that once the SAL assay
result is available for prediction, other in vitro mu-
tation tests provide little additional information for
predicting CA. Our study indicates that the same
conclusion holds for CA, SAL, MNC, and ABS.
Environ. Mol. Mutagen. 34:297–304, 1999.
© 1999 Wiley-Liss, Inc.

Key words: carcinogenicity; contingency table; genotoxicity; predictivity; short-term tests

INTRODUCTION Shelby [1988], updated by Shelby and Zeiger [1990],

Recently, Zeiger and Tennant [1996] wrote in a commen-
tary that “If an (in vitro) bacterial mutation test has a high
positive predictivity for CA, an in vivo rodent mutation test
should, according to all we know about biology, have an
even greater predictivity. The in vivo test should theoreti-
cally offer the additional advantage of predicting the tissue-,
sex-, and species-specific responses. Data to address these
correlations between mutagenicity and carcinogenicity are
currently being developed in a number of laboratories.” As
an initial trial to elucidate these correlations, the Ames
Salmonella mutation assay (SAL) [Maron and Ames, 1983],
the micronucleus assay (MNC) [Hayashi et al., 1994], and
the chromosome aberration assay (ABS) [Shelby et al.,
1989] in mouse bone marrow cells were considered as a set
of predictors of the long-term CA assay. For this purpose we
were able to collect data on all four assays for 82 chemicals
from the National Toxicology Program (NTP) data base,
from the International Agency for Research on Cancer
(IARC) monograph series, or from literature sources.
There have been several reports [Tennant et al., 1987;
Haseman et al., 1990; Zeiger et al., 1990] on the evaluation
of four widely used in vitro genotoxicity tests for predicting
CA. These four tests were SAL, mouse lymphoma L5178Y,
sister chromatid exchanges, and chromosome aberrations in
Chinese hamster ovary cells. Major conclusions of these
studies were: (1) SAL was the most predictive single assay
for CA; (2) there was no complementarity among the four in
vitro tests; and (3) no combination of the four tests was
more effective than any single test for predicting CA.
investigated to what extent the 23 IARC human carcinogens
could be predicted by the activities of genetic toxicity in
vitro and in vivo. He considered SAL for the in vitro assay
and either MNC or ABS for the in vivo assay. He reported
that 20 of the 23 human carcinogens were active in either
short-term test, and the combination of these two assays
might serve as a screening procedure for chemicals that
present a carcinogenic hazard to humans. Shelby et al.
[1993] reported that in a study of 49 chemicals, among
which 25 were rodent carcinogens and 24 were noncarcino-
gens, MNC did not show association with CA. However,
they did observe that a combination of SAL and MNC could
provide important information on the genotoxicity of test
chemicals.
Recently, Shelby and Witt [1995] reported on the com-
parative outcomes of the MNC and the ABS based on the
65-chemical data base that had been accumulated by the
NTP in the past 15 years. They observed 80% concordance
between MNC and ABS, although these tests were not
conducted with the purpose of comparing the outcomes of
these two in vivo genetic toxicity endpoints.
In this study, we combine the approaches employed by
Grant sponsor: Korea Research Foundation.
*Correspondence to: Dr. Byung Soo Kim, Department of Applied Statis-
tics, Yonsei University, Seoul, 120-749, South Korea.
E-mail: bskim@yonsei.ac.kr
Received 18 February 1998; provisionally accepted 16 May 1999; and in
final form 7 July 1999

© 1999 Wiley-Liss, Inc.

298 Kim and Margolin
Shelby et al. [1993], Shelby [1988], and Shelby and Witt
[1995] and investigate whether an in vitro test, here SAL,
and two in vivo tests, here MNC and ABS, can be combined
to improve the predictivity of CA. We try to provide an-
swers to three questions regarding the relation between CA,
SAL, MNC, and ABS.
1. For prediction of CA, what are the important differ-
ences in performance among SAL, MNC, and ABS,
and is one test better than the others?
2. What does the statistical analysis indicate on the bio-
logical relations among SAL, MNC, ABS, and CA?
3. Do MNC and ABS complement SAL for predicting
CA?
In the following discussion we employ contigency table
analyses to answer these three questions.
MATERIALS AND METHODS
Data
There are 82 chemicals in our data base for which test outcomes of CA,
SAL, MNC, and ABS are available. In Table I we list the test outcomes of
the 82 chemicals. For 75 chemicals we obtained these test outcomes from
the NTP data base through Central Data Management (CDM) of NIEHS.
For the remaining seven chemicals that have not been tested for CA by
NTP, we collected the test outcomes from the literature [IARC, 1987;
Shelby and Witt, 1995; Zeiger et al., 1990]. For some chemicals, whose
genotoxicity assay results were not available in the NTP publications, we
cited by way of footnotes to Table I the literature sources from which we
obtained these results.
In at least one of the four assays—SAL, MNC, ABS, and CA—there
were 10 chemicals that had multiple outcomes. We refer multiple outcomes
to a set of outcomes where different conclusions are reached by different
authors on the same chemical in basically the same assay. These are listed
in Table II. We considered the SAL results for acrylamide, allyl isothio-
cyanate, L-ascorbic acid, and 3-chloro-2-methylpropene as negative for
statistical analysis. For the MNC results for 1,3-dichloropropene, diglyci-
dyl resorcinol ether, propyl gallate, and 2,6-toluenediamine 2HCl, we
followed Shelby and Witt’s [1995] result for the statistical analysis.
There were a few chemicals whose assay results needed some explana-
tion. Geranyl acetate was found to be negative in ABS in one laboratory.
The highest dose that allowed survival was 1000 mg/kg. In a second
laboratory, in each of two trials, geranyl acetate gave a positive response in
ABS. The highest dose tested in each of these two trials was 1700 mg/kg.
The overall call was taken to be positive in ABS for geranyl acetate. Benzyl
acetate had positive and negative responses in the CA in NTP tech report
(TR) numbers 250 (gavage study) and 431 (feed study), respectively.
Dichlorvos yielded a negative response in NTP TR 10 (feed study) and a
positive response in NTP TR 342 (gavage study). Both 3-chloro-2-meth-
ylpropene and 1,2-dichlorobenzene had two independent tests in two
different labs for ABS. One lab found the chemical to be positive and the
other negative. For each of these four cases, we considered the overall call
the strongest response for the statistical analysis. We did not include
Aroclor 1254 (NTP TR 038) in Table I because its CA test was performed
for only one species.
The chemicals giving equivocal CA responses were treated as noncar-
cinogens and equivocal results in SAL, MNC, or ABS were also treated as
negative results. This was done because in both the carcinogenicity and
genetic toxicity testing programs, equivocal responses are more generally
thought of as negative [Tennant et al., 1987; Zeiger et al., 1990].
Statistical Methods
For each of three short-term tests (SAL, MNC, ABS), we tested the null
hypothesis of no association against positive association with CA in a 2 3
2 contingency table using Fisher’s exact test, and calculated the sensitivity,
specificity, positive predictivity, negative predictivity, and concordance.
We used Yates’s continuity correction factor for calculating the 95%
confidence interval [Fleiss, 1981].
To investigate the complementarity of SAL, we explored the CA–MNC
and CA–ABS associations for 30 SAL-positive chemicals and for 52
SAL-negative chemicals, respectively. We stratified the 82 chemicals by
the CA outome and investigated whether three short-term tests had the
same dependence structure for 55 carcinogens and for 27 noncarcinogens.
We also studied the relationships between SAL, MNC, ABS, and CA. For
example, an in vitro test, say SAL, may plausibly predict the activities of
in vivo tests, say MNC and ABS. For the relation between MNC and ABS
we note that MNC has replaced ABS over the past several years as a
routine screening assay because of the relative ease and speed of the MNC.
MNC data are easier to handle, although they are less informative than
ABS data [Shelby and Witt, 1995]. Finally, the three short-term tests can
be combined to predict the activity of CA. These biological relations
among SAL, MNC, ABS, and CA allow us to hypothesize the following
possible causal ordering of these assays:
SAL may predict MNC, which may predict ABS, which may predict CA
or, in arrow notation,
SAL 3 MNC 3 ABS 3 CA. (1)
We may provide a path diagram that schematically illustrates the struc-
tural relationship among these four assays by combining the statistical
analysis results with the plausible predictive ordering of assays in notation
(1).
RESULTS
In Tables III to X, we provide the descriptive nature of
the 24 contingency table formed from yes/no results for CA,
SAL, MNC, and ABS. Estimates of sensitivity, specificity,
positive and negative predictivities, and concordance, with
respect to CA for the three short-term tests, are presented in
Table III. It is clear that SAL outperforms MNC and ABS
and that there is no dominance between MNC and ABS. We
note here that only SAL has a significant positive associa-
tion with CA. We may further emphasize that neither of the
two in vivo assays has a significant positive association with
CA.
An initial way to approach the question of complemen-
tarity is to stratify the 82 chemicals by the qualitative
(1 or 2) results obtained with SAL. The data in Table IV
indicate that when one considers only the 30 SAL-positive
chemicals, CA results show significant association with
ABS results. However, as we can see in Table V, neither
MNC nor ABS shows significant association with CA for 52
SAL-negative chemicals. The results in Tables IV and V
exhibit a feature labeled the lack of conditional indepen-
dence, and indicate that ABS, when combined with SAL,
 Predicting Carcinogenicity From In Vitro and In Vivo Tests 299

TABLE I. Summary Results of CA, Salmonella Mutagenicity, In Vivo Mouse Bone Marrow Micronucleus, and Chromosome
Aberration Tests
No. Chemical name CAS no. NTP tech report no. CA SAL MNC ABS

1 Acrylamide 79-06-1 IARC [1986] v. 39 1 1w, 2 a

1 b
1
2 Allyl isothiocyanate 57-06-7 234 1 1w, 2 2 2
3 11-Aminoundecanoic acid 2432-99-7 216 1 2 2 2
4 L-Ascorbic acid 50-81-7 247 2 1w, 2 1 1
5 Benzene 71-43-2 289 1 2 1 1b
6 Benzidine 2HCl 531-85-1 IARC [1987] Suppl. 6 1 1 1 1
7 Benzoin 119-53-9 204 2 2 2 2
8 Benzyl acetate 140-11-4 250 (Gavage) 1 2 2 2
431 (Feed) 2
9 2-Biphenylamine HCl 2185-92-4 233 1 1 2 2
10 1,3-Butadiene 106-99-0 434 1 1 1 1
11 Butyl benzyl phthalate 85-68-7 458 1 2 2 1
12 Butylhydroquinone 1948-33-0 459 2 2 2 2
13 C.I. Acid Orange 10 1936-15-8 211 2 2 2 2b
14 C.I. Acid Red 14 3567-69-9 220 2 2 2 2
15 C.I. Disperse Yellow 3 2832-40-8 222 1 1 2 2
16 Caprolactam 105-60-2 214 2 2 2 2
17 Chlorobenzene 108-90-7 261 Ec 2 2 1
18 Chlorodibromomethane 124-48-1 282 1 2 2 2b
19 2-Chloroethanol 107-07-3 275 2 1 2 2
20 Chloroform 67-66-3 000 1 2 1 2
21 3-Chloro-2-methylpropene 563-47-3 300 1 1w, 2, 2 2 1, 2
22 2-Chloromethyl-pyridine HCl 6959-47-3 178 2 1 2 2b
23 3-Chloromethyl-pyridine HCl 6959-48-4 095 1 1 2 2
24 4-Chloro-o-phenylenediamine 95-83-0 063 1 1 1 1b
25 Chloroprene 126-99-8 467 1 2 2 2
26 Cinnamyl anthranilate 87-29-6 196 1 2 2 2
27 C.I. Solvent Yellow 14 842-07-9 226 1 1 E 2
28 Cyclophosphamide 6055-19-2 IARC [1987] Suppl. 6 1 1 1b 1b
29 Cytembena 21739-91-3 207 1 1 2 2
30 Di(2-ethylhexyl) adipate 103-23-1 212 1 2 2 2
31 Diallyl phthalate 131-17-9 242 (Mouse) E 2 2 1
284 (Rat) E
32 1,2-Dibromo-3-chloropropane 96-12-8 206 1 1 2 2
33 1,2-Dibromoethane 106-93-4 210 1 1 2 2
34 1,2-Dichlorobenzene 95-50-1 255 2 2 2 1, 2
35 1,4-Dichlorobenzene 106-46-7 319 1 2 2 2
36 2,6-Dichloro-p-phenylenediamine 609-20-1 219 1 1 2 1
37 1,3-Dichloropropene 542-75-6 269 1 1 1, 2 1
38 Dichlorvos 62-73-7 010 (Feed) 2 1 2 2
342 (Gavage) 1
39 Diglycidyl resorcinol ether 101-90-6 257 1 1 1, 2 1
40 7,12-Dimethylbenz (a)-anthracene 57-96-6 441 1 1 1b 1b
41 Dimethyl hydrogen phosphite 868-85-9 287 1 1 1 1
42 Dimethylterephthalate 120-61-6 121 E 2 2 E
43 5,5-Diphenylhydantoin 57-41-0 404 1 2 2 2
44 Eugenol 97-53-0 223 E 2 2 E
45 FD&C Yellow No. 6 2783-94-0 208 2 2 2 2
46 Geranyl acetate 105-87-3 252 2 2 2 1, 2d
47 8-Hydroxyquinoline 148-24-3 276 2 1 2 2
48 Isobutyraldehyde 78-84-2 472 2 2 2 1
49 Isoprene 78-79-5 486 1 2 1 2
50 Lead(II) sulfide IARC, [1987] Suppl. 6 1 2 2 2
51 D-Mannitol 69-65-8 236 2 2 2 2
52 Melamine 108-78-1 245 1 2 2 1
53 Melphalan 148-82-3 IARC, [1987] Suppl. 6 1 1 1 1
54 DL-Menthol 15356-70-4 098 2 2 2 E
55 5-Methoxypsoralen 484-20-8 IARC, [1987] Suppl. 6 1 1 2 1
56 8-Methoxypsoralen 298-81-7 359 1 1 E 1
57 4,49-Methylene-dianiline 2HCl 13552-44-8 248 1 1 1b 1b
58 Mitomycin C 50-07-7 IARC, [1973] v. 10 1 1e 1 1b
300 Kim and Margolin

TABLE I. Continued
No. Chemical name CAS no. NTP tech report no. CA SAL MNC ABS

59 Monuron 150-68-5 266 1 2 1 2

60 4,49-Oxydianiline 101-80-4 205 1 1 1 1
61 Pentachloroethane 76-01-7 232 1 2 2 2
62 Phenol 108-95-2 203 2 2 1 1
63 Phenolphthalein 77-09-8 465 1 2 1 2
64 Polybrominated biphenyl mixture 67774-32-7 398 1 2f 2 2
65 1,2-Propylene oxide 75-56-9 267 1 1 2 1
66 Propyl gallate 121-79-9 240 E 2 1,2 1
67 Pyridine 110-86-1 470 1 2 2 2
68 Reserpine 50-55-5 193 1 2 2 1
69 Salicylazosulfapyridine 599-79-1 457 1 2 1 2
70 Selenium sulfide 7446-34-6 194 1 1 2 1
71 Sodium fluoride 7681-49-4 393 E 2 2 2
72 Stannous chloride 7772-99-8 231 E 2 2 1
73 Sulfisoxazole 127-69-5 138 2 2 2 2
74 1,1,1,2-Tetrachloroethane 630-20-6 237 1 2 1 1
75 Tetrachloroethylene 127-18-4 311 1 2 2 2
76 Tetrahydrofuran 109-99-9 475 1 2 2 2
77 Theophylline 58-55-9 473 2 2 1 2
78 Titanium dioxide 13463-67-7 097 2 2 1 2
79 2,6-Toluenediamine 2HCl 15481-70-6 200 2 1 1,2 2
80 Tribromomethane (Bromoform) 75-25-2 350 1 1 2 2
81 Trichloroethylene 79-01-6 243 1 2 2 2
82 Tris(2-ethylhexyl) phosphate 78-42-2 274 1 2 2 2
a
The chemical has both weak positive and negative responses.
b
From Shelby and Witt [1995].
c
E means equivocal.
d
Only one time point was investigated for the negative response.
e
From IARC [1973], Vol. 10, p. 175.
f
From Zeiger et al. [1990].

TABLE II. List of Chemicals That Have Multiple Outcomes

Considered for the
Chemical name Assay NTP statistical analysis Remark

Acrylamide SAL 1w, 2 2

Allylisothiocyanate SAL 1w, 2 2 E in Zeiger et al. [1990]
L-Ascorbinc acid SAL 1w, 2 2 Following Zeiger et al. [1990]
3-Chloro-2-methyl-propene SAL 1w, 2, 2 2 Following Zeiger et al. [1990]
ABS 1, 2 1
1,2-Dichlorobenzene ABS 2, 1 1
1,3-Dichloropropene MNC 1, 2 1 Following Shelby and Witt [1995]
Diglycidyl resorcinol ether MNC 1, 2 1 Following Shelby and Witt [1995]
Geranly acetate ABS 1, 2a 1
Propyl gallate MNC 1, 2b 1 Following Shelby and Witt [1995]
2,6-Toluenediamine 2HCl MNC 1, 2 1 Following Shelby and Witt [1995]
a
Only one time point was investigated for the negative response.
b
No positive control was tested for the negative response.

may lead to the improvement of the prediction of CA. We
explore the possibility of improving the predictivity of CA
by combining a battery of an in vivo assay with SAL. There
are two ways of making a decision regarding CA when we
combine two short-term tests, say SAL and ABS. We may
predict the test chemical to be CA if either SAL or ABS
yields a positive response. We refer to this combination as
“SAL or ABS.” We may also predict it to be CA when both
SAL and ABS yield positive responses, of which the com-
bination is referred to as “SAL and ABS.” The other battery
combinations, “SAL or MNC” and “SAL and MNC,” are
similarly defined. Based on the results in Tables IV and V,
we can calculate the sensitivity, specificity, and positive and
negative predictivities of each of these four battery combi-
nations with SAL, as shown in Table VI. It is readily seen
that none of the battery combinations improves the predic-
 Predicting Carcinogenicity From In Vitro and In Vivo Tests 301

TABLE III. Operational Characteristics of SAL, MNC, and ABS as Predictors of the CA
SAL MNC ABS
Rodent carcinogenicity
(CA) 1 2 1 2 1 2

1 26 29 20 35 24 31
2 4 23 6 21 9 18

Total 30 52 26 56 33 49

P value of the Fisher’s

exact test for positive
association .003 .149 .258
a
Sensitivity .48 (.34–.61) .36 (.24–.50) .44 (.31–.58)
Specificity .85 (.65–.95) .78 (.57–.91) .67 (.46–.83)
Positive predictivity .87 (.68–.96) .77 (.56–.90) .73 (.54–.86)
Negative predictivity .44 (.31–.59) .38 (.25–.51) .37 (.24–.52)
Concordance .60 (.48–.70) .50 (.39–.61) .51 (.40–.62)
a
95% confidence intervals are in parentheses.

TABLE IV. Association of CA With MNC and ABS for 30 TABLE VI. Operational Characteristics of SAL Alone and
Chemicals That Are SAL Positive Four Batteries as Predictors of CA
MNC ABS SAL “SAL or “SAL and “SAL or “SAL and
alone ABS” ABS” MNC” MNC”
CA 1 2 1 2
Sensitivity .47 .60 .31 .62 .22
1 12 14 17 9
Specificity .85 .52 1.00 .67 .96
2 1 3 0 4
Positive
Total 13 17 17 13 predictivity .87 .72 1.00 .79 .92
Negative
P value of Fisher’s exact test predictivity .44 .39 .42 .46 .38
for positive association .409 .026

TABLE VII. 24 Contingency Table of SAL, MNC, ABS, and

TABLE V. Association of CA With MNC and ABS for 52 CA
Chemicals That Are SAL Negative
CA
MNC ABS
SAL MNC ABS 1 2
CA 1 2 1 2
1 1 1 12 0
1 8 21 7 22 1 1 2 0 1
2 5 18 9 14 1 2 1 5 0
1 2 2 9 3
Total 13 39 16 36
2 1 1 3 3
P value of Fisher’s exact test 2 1 2 5 2
for positive association .439 .929 2 2 1 4 6
2 2 2 17 12

tivity of CA achieved by SAL alone. However, we may note
that the “SAL and ABS” combination provides 100% for
both specificity and positive predictivity, which may be
quite useful for the screening procedure of CA assay.
We may further propose the following decision rule (2)
for CA prediction based on the three-test battery:
If ~SAL, MNC, ABS! 5 ~ 1 , 1 , 1 !,
~ 1 , 2 , 1 !, ~ 1 , 2 , 2 !,
we decide the test chemical to be a carcinogen;
otherwise, we label it a noncarcinogen.
Total 55 27
The sensitivity, specificity, and positive and negative pre-
dictivities of the three-test battery decision rule (2) are .47,
.89, .90, and .45, respectively, based on the 24 contingency
table of Table VII. We may conclude that the decision based
on SAL is at least as good as the decision rule (2), due to its
simplicity and cost-effectiveness.
Tables VIII to X may provide us some insight into the
(2)
structural relationship among SAL, MNC, ABS, and CA.
As we can see in Table VIII, SAL has a weak association
with MNC and a significant positive association with
ABS. We may note that MNC and ABS are themselves
302 Kim and Margolin

TABLE VIII. Associations Among SAL, MNC, and ABS

MNC ABS
SAL 1 2 1 2 Total

1 13 17 17 13 30
2 13 39 16 36 52

P value of Fisher’s exact test for positive association .071 .019

ABS
MNC 1 2 Total

1 18 8 24
2 15 41 58

P value of Fisher’s exact test for positive association .0003

TABLE IX. Associations Among SAL, MNC, and ABS for 55 Carcinogens
MNC ABS
SAL 1 2 1 2 Total

1 12 14 17 9 26
2 8 21 7 22 29

P value of Fisher’s exact test for positive association .125 .002

ABS
MNC 1 2 Total

1 15 5 20
2 9 26 35

P value of Fisher’s exact test for positive association .0005

TABLE X. Associations Among SAL, MNC, and ABS for 27 Noncarcinogens

MNC ABS
SAL 1 2 1 2 Total

1 1 3 0 4 4
2 5 18 9 14 23

P value of Fisher’s exact test for positive association .659 1.000

ABS
MNC 1 2 Total

1 3 3 6
2 6 15 21

P value of Fisher’s exact test for positive association .305

strongly associated with concordance 0.72, which is close tures among SAL, MNC, and ABS for carcinogens and
to the 0.8 reported in Shelby and Witt [1995]. When we noncarcinogens, respectively.
do the same analyses for the 55 carcinogens and 27 From the results in Tables III to X we may note the
noncarcinogens, respectively, we have quite different following conclusions:
results in each case. In Table IX we have significant
associations between SAL and ABS, and between MNC 1. Only SAL affects CA.
and ABS for 55 carcinogens. For 27 noncarcinogens in 2. MNC affects ABS.
Table X, we observe statistical independence between 3. SAL affects ABS.
any pair of SAL, MNC, and ABS. This result strongly
indicates that we have quite different dependence struc- These three conclusions and the plausible causal ordering of
 Predicting Carcinogenicity From In Vitro and In Vivo Tests
Fig. 1. Path diagram of CA, SAL, MNC, and ABS.
four assays of notation (1) can be schematically represented
as in Figure 1.
We also note that the decision rule (2) differs from the
decision based on SAL alone by the configuration (SAL,
MNC, ABS) 5 (1, 1, 2), which has only one observation,
that is, 2,6-toluenediamine 2HCl. Even though 2,6-tolu-
enediamine gave no evidence of carcinogenicity in rodents,
Shelby et al. [1993] considered it a hazardous chemical
because it was a mutagen and it was a structural analogue of
the carcinogen, 2,4-toluenediamine.
Even in the later experiments it is more likely for a
carcinogen to belong to the event (SAL, MNC, ABS) 5 (1,
1, 2) than for a noncarcinogen, because SAL has almost
90% positive predictivity. Therefore, it is not likely that the
decision rule (2) will improve on the decision rule based on
SAL only. It also remains to be seen whether chemicals
belonging to this class can be classified into hazardous
chemicals like 2,6-toluenediamine in our data set.
DISCUSSION
It has been clear for some time now that additional in
vitro tests provide little useful predictive information once
the SAL assay result is in hand [Shelby, 1988; Tennant et
al., 1987; Haseman et al., 1990]. Our study indicates that the
same conclusion can be drawn for in vivo tests, because the
decision rule (2), based on SAL, MNC, and ABS, does not
provide significant improvement for the prediction of CA
over the decision rule based on SAL alone. We leave it for
further study to investigate whether modern transgenic ap-
proaches may prove more useful in the prediction of CA.
Shelby [1988, p. 5] wrote: “The use of an in vivo STT in
a primary screening role may be indicated by such in vivo
clastogens where the information of genetic intermediates is
not anticipated by structural considerations and the chemi-
cal is not detected by the Salmonella assay.” We showed
that two in vivo STTs, that is, MNC and ABS, did not
complement SAL for predicting CA. However, structural
303
consideration can be combined with these two in vivo tests
to determine CA of chemicals not detected by the SAL
assay. We have 29 carcinogens and 23 noncarcinogens in
our data set that had negative results in the SAL assay. It is
left for further study to see how structural considerations
can improve the prediction of CA among SAL-negative
chemicals.
The 82 chemicals that we evaluated in this study do not
likely constitute a random sample of all chemicals evaluated
for CA, in that two thirds (55/87) are carcinogens, whereas
only approximately half of all chemicals evaluated in CA
are carcinogens.
We may limit the targeted carcinogenic response to the
stronger effects that represent multisite, multispecies, and/or
“clear evidence” carcinogenic effects. This approach may
explain low sensitivity of SAL and the poor predictive
performance of MNC and ABS. Our study did not address
the issue that in vivo test outcomes might have advantages
in predicting tissue-, sex-, and species-specific responses.
The exploration of these issues is left for further study.
It has been noted recently by Piegorsch et al. [1992] that
summary measures such as specificity or concordance can
vary among different chemicals as their potencies vary. The
primary cause of the dependence of observed specificity or
concordance on the potency is that in long-term animal
carcinogenicity assays, a small sample size (n 5 50) is used,
in which case the low-potency chemical is liable to yield a
false positive outcome; however, in the short-term assay,
this sample size problem is not as serious as in the long-term
animal assay. The incorporation of the chemical potency
factor in the construction of confidence intervals of various
summary measures in Table III will pose an interesting
research topic.
ACKNOWLEDGMENTS
We thank Kristine Witt and Michael Shelby at NIEHS for
their help in many aspects of this research. We gratefully
acknowledge two anonymous referees and the editor-in-
chief for their invaluable comments and suggestions. Byung
Soo Kim’s research was partially supported by the Korea
Research Foundation through its 1995 Overseas Research
Program for University Professors.
REFERENCES
Fless JL. 1981. Statistical methods for rates and proportions, 2nd Ed. New
York: Wiley.
Haseman JK, Zeiger E, Shelby MD, Margolin BH, Tennant, RW. 1990.
Predicting rodent carcinogenicity from four in vitro genetic toxicity
assays: An evaluation of 114 chemicals studied by the National
Toxicology Program. J Am Stat Assoc 85:964 –971.
Hayashi M, Tice RR, MacGregor JT, Anderson D, Blakey DH, Kirsh-
Volders K, Oleson FB Jr, Pacchierotti F, Romagna F, Shimada H,
Sutou S, Vannier B. 1994. In vivo rodent erythrocyte micronucleus
assay. Mutat Res 312:293–304.
304 Kim and Margolin
IARC. 1973. IARC monographs on the evaluation of the carcinogenic risk
of chemicals to man: some naturally occurring substances, Vol. 10.
Lyon: IARC.
IARC. 1986. IARC monographs on the evaluation of the carcinogenic risk
of chemicals to humans: some chemicals used in plastics and
elastomers, Vol. 39. Lyon: IARC.
IARC. 1987. IARC monographs on the evaluation of the carcinogenic risk
of chemicals to humans: overall evaluations of carcinogenicity, an
updating of IARC Monographs from Vols. 1– 42, Suppl. 6. Lyon:
IARC.
Maron DM, Ames BN. 1983. Revised methods for the Salmonella muta-
genicity test. Mutat Res 113:173–215.
Piegorsch WW, Carr GJ, Portier CJ, Hoel DG. 1992. Concordance of
carcinogenic response between rodent species: Potency dependence
and potential underestimation. Risk Anal 12:115–121.
Shelby MD. 1988. The genetic toxicity of human carcinogens and its
implications. Mutat Res 204:3–15.
Shelby MD, Witt KL. 1995. Comparison of results from mouse bone
marrow chromosome aberration and micronucleus tests. Environ
Mol Mutagen 25:302–313.
Shelby MD, Zeiger E. 1990. Activity of human carcinogens in the Salmo-
nella and rodent bone-marrow cytogenetic tests. Mutat Res 234:
257–261.
Shelby MD, Gulati DK, Tice Rr, Wojciechowski JP. 1989. Results of tests
for micronuclei and chromosomal aberrations in mouse bone mar-
row cells with the human carcinogens 4-aminobiphenyl, treosul-
phan, and mephalan. Environ Mol Mutagen 13:339 –342.
Shelby MD, Erexson GL, Hook GJ, Tice RR. 1993. Evaluation of a
three-exposure mouse bone marrow micronucleus protocol: Results
with 49 chemicals. Environ Mol Mutagen 21:160 –179.
Tennant RW, Maroglin BH, Shelby MD, Zeiger E, Haseman JK, Spalding
J, Caspary W, Resnick M, Stasiewicz S, Anderson B, Minor R.
1987. Prediction of chemical carcinogenicity in rodents from in
vitro genetic toxicity assays. Science 236:933–941.
Zeiger E. Haseman JK, Shelby MD, Maroglin BH, Tennant RW. 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.
Zeiger E, Tennant RW. 1996. Commentary. Environ Mol Mutagen 28:3– 4.
Accepted by—
E. Zeiger