vitro SITS and rodent carcinogenicity tests were highly encouraging. (4-7); sensitivities ... derived study of the performance of SITS conducted by the Gene-. Tox Program of the ... gum, locust bean gum, propylene, tara gum, vinylidene chloride).
Science 1987; 236:933-41. - (CREATED FROM SCANNED DOCUMENT, PLEASE REFER TO ORIGINAL)
Prediction of Chemical Carcinogenicity in Rodents from in Vitro Genetic Toxicity Assays R A Y M O N D W . TENNANT, BA R R Y H . MA R G O L I N , MI C H A E L D . S H E L B Y, E R R O L Z E I G E R , JO S E P H K . HASEMAN, JU D S O N GALDING, WI L LI A M C A S P A R Y , M I C H A E L R E S N I C K , ST AN L E Y S T A S I E W I C Z, BE T H A ND E R S O N, RO B E R T M I N O R
Four widely used in vitro assays for genetic toxicity were evaluated for their ability to predict the carcinogenicity of selected chemicals in rodents. These assays were mutagenesis in Salmonella and mouse lymphoma cells and chromosome aberrations and sister chromatid exchanges in Chinese hamster ovary cells. Seventy-three chemicals recently tested in 2-year carcinogenicity studies conducted by the National Cancer Institute and the National Toxicology Program were used in this evaluation. Test results from the four in vitro assays did not show significant differences in individual concordance with the rodent carcinogenicity results; the concordance of each assay was approximately 60 percent. Within the limits of this study there was no evidence of complementarity among the four assays, and no battery of tests constructed from these assays improved substantiaIIy on the overah performance of the Salmonella assay. The in vitro assays which represented a range of three ceII types and four end points did show substantial agreement among themselves, indicating that chemicals positive in one in vitro assay tended to be positive in the other in vitro assays.
S
thorough evaluation of the ability of these tests to predict rodent carcinogenicitv: for most SITS there was a dearth of results for documented noncarcinogens (IO), and too few chemicals had been tested in multiple SITS to permit meaningful comparisons of the ability of different SITS and SIT combinations to predict carcinogens. In the early 1970s the National Cancer Institute (NCI) developed a protocol for rodent carcinogenicity studies that specified long-term exposure of both sexes of two species of rodents, generally F344 rats and B6C3Fi mice, to high doses of chemicals in 2-year studies, with the use of 50 animals per treatment group (11). This protocol, adopted and refined by the National Toxicology Program (NTP), has been used during the last 12 years to study over 300 chemicals (12). Results from these studies constitute the primary database available today for the evaluation of mammalian carcinogenesis. In 1984, the NTP initiated a project to develop a database that , would permit evaluation of the ability of four of the most commonly used in vitro SITS to predict rodent carcinogenicity: the Ames Salmonellalmicrosome (SAL) mutagenesis assay (13), the assays for chromosome aberration (ABS) and sister chromatid exchange (SCE) induction in Chinese hamster ovary cells (14), and the mouse lymphoma L5 178Y (MOLY) cell mutagenesis assay (15). In this article, we present results and conclusions derived from this study.
HORT-TERM TESTS (s?TS) F O R G E N O T O X I C C H E M I C A L S were
originally developed to study mechanisms of chemically induced DNA damage and to assess the potential genetic hazard of chemicals to humans. The role of these tests has increased, however, because of accumulating evidence in support of the somatic mutation theory of carcinogenesis (1, 2) and because of reports that many rodent carcinogens are genotoxic in in vitro SlTs (3). The in vitro SITS have the advantages that they can be conducted relatively quickly and inexpensively compared to longterm carcinogenicity assays with rodents and do not involve testing in animals. Early studies of concordance between results from in vitro SITS and rodent carcinogenicity tests were highly encouraging (4-7); sensitivities (percentages of carcinogens identified as mutagens) and specificities (percentages of noncarcinogens identified as nonmutagens) of 90% or better were reported, especially for the Ames Sahnonella mutagenesis assay (5, 62. As a consequence of these reported concordances and because of concern fo,r heritable damage to future generations, many countries drew up regulatory guidelines requiring submission of SIT data for the registration of new chemicals (8). On the basis of a literaturederived study of the performance of SITS conducted by the GeneTox Program of the Environmental‘ Protection Agency (9), it became apparent that there were two major impediments to a
Study
Design
A number of characteristics of the design of this study distinguish it from previous attempts to evaluate SlTs as predictors of rodent carcinogenicity, especially those based on results compiled from the scientific literature. First, standard protocols for the four SITS were developed by the NTP and shown to yield reproducible results in interlaboratory trials with coded chemicals (13, 14, 16). The literature results, by contrast, arise from a highly diverse set of protocols. Second, because literature-based evaluations often reflect the bias of the publication of results on strongly positive mutagens and mammalian carcinogens, a major design concern was selection of test chemicals by a procedure that would minimally bias the evaluation. The 83 chemicals initially selected for this project were those tested for rodent carcinogenicity by NC1 and NTP with studies ending R. W. T~M~Ix, M. D. Shelbv, E. Zei et, J. S alding, W. Caspay, M. Resnick, S. Stasiewin, and B. Andetson’.ue in 2 e Cell $ ar and Genetic Toxicology Branch, Toxicology Research and Testing Division, Natiwal InSUNte of Environmental Health Sciences, Research Triangle Park, NC 27709. B. H. Margolin and J. K. Haseman are in the Statistics and Biomathematics Branch. Biometry and Risk Assessment Division, National Institute of Environmental Health Sciences, Research Triangle Park, NC 27709. R. Minor is with Informanon Syems and Networks Gxporauon, Research Triangle Park, NC 27709. ARTICLES 933
22 MAY 1987
Prediction of chemical carcinogenicity in rodents from in vitro genetic toxicity assays. X-2
Science 1987; 236:933-41. - (CREATED FROM SCANNED DOCUMENT, PLEASE REFER TO ORIGINAL)
..
I.-- -
i r
December 1976 or later, and final NTP peer review approval dates of 1 January 1985 or earlier. Ten chemicals (hexachlorodibenzo-pdioxin mixture, 1,3-butadiene, agar, gilsonite, gum arabic, guar gum, locust bean gum, propylene, tara gum, vinylidene chloride) were excluded from further consideration because the first was not available in the same mixture tested for carcinogenicity, and the physical properties of the remaining nine indicated that they could not be adequately tested with SIT protocols used herein. Only the first two of the ten are rodent carcinogens. The remaining 73 chemicals, well-characterized for carcinogenicity in both sexes of , two rodent species, were then tested under code with each of the four SITS. Where possible, we used the same chemical lot as was used in the rodent carcinogenicity study. Finally, the SIT experiments generally were replicated within each laboratorv and evaluated with the aid of statistical analyses (17-19).
Carcinogenicity Results Forty-four chemicals (60%) were tumorigenic in at least one site in one of the four combinations of sex and species, the NTP criterion for a determination of a chemical carcinogen. Twenty chemicals (27%) showed no evidence of carcinogenicity. and nine (12%) were reported as “equivocal,” neither clearly positive nor clearly negative. These nine chemicals are treated as noncarcinogens in the statistical analysis. Typically, an equivocal carcinogenic response was described in the NT’P reports as follows: “[Chemical name] was not carcinogenic, . but [certain tumors] may have been related to chemical administration.” Such an outcome was considered substantially closer to negative than positive. To evaluate comprehensively the performance of SITS, we included all available ,carcinogenicity data, even those studies that were difficult to interpret; however, the omission of the equivocal studies would not have markedly affected our qualitative conclusions. The patterns of tumorigenicity exhibited by the 44 carcinogens were highly varied. The most frequent site of tumor induction was the liver (26 out of 44), and this was the only site of activity for 12 of the 44. The concordance, or observed agreement, between rat and mouse carcinogenicity determinations was 67% (49 out of 73) (that is, 12 chemicals were positive onlv in mice and 12 were positive only in rats, whereas 20 were positive and 29 were negative in both species). This agreement is significantly lower (P < 0.01) than the concordance of 85% (211 out of 249) for these species, reported by Purchase (20). Historicallv, the interspecies concordance for the NTP rodent carcinogenic&y assay has been approximately 74% (198 out of 266) (12). Elimination from the 83 chemicals of the 10 not tested in SITS served to depress the interspecies concordance because 9 of those eliminated were concordant and the tenth (1,3_butadiene) was tested only in mice. Nine of the 24 interspecies disagreements are attributable solely to tumors of the liver. For prediction of mouse carcinogenicity, the rat carcinogenicity assay has an estimated sensitivity of 63% (20 out of 32) and specificity of 71% (29 out of 41); when the roles of the rat and mouse in prediction are interchanged, these estimates remain unchanged. These values may represent an approximate upper bound on the concordance that can be achieved between rodent tumorigenicitv and SIT results.
STTs as Predictors of Carcinogenicity \ I
,
The data in Table 1 provide answers to three questions regarding the relation between rodent carcinogenicity and in vitro tests for genotoxicitv.
1) What, if any, are the important differences in performance among the four SITS, and is one test clearly better than the others? 2) If SAL is central to in vitro testing schemes, as has often been proposed, which, if am, SIT best complements this assay in the sense of raising sensitivity without significant loss in specificity? 3) Can these four SITS in any combination form a battery (multiple concurrent tests) that outperforms SAL in discriminating between carcinogens and noncarcinogens? The frequencies of positive responses for the four SITS were as follows: SAL, 33% (24 of 73); ABS, 45% (33 of 73); SCE, 66% (48 of 73); and MOLY, 64% (47 of 73). Estimates (and definitions) of sensitivity, specificity, positive and negative predictivity, and concordance, with respect to rodent carcinogenicity for the four SITS, are presented in Table 2. It is apparent that SAL and ABS performed similarly overall, as did SCE and MOLY. These two pairs, however, did d@er; SAL and ABS were reasonably specific but relatively insensitive regarding rodent carcinogenicity, while the reverse was true for SCE and MOLY. One should note that the 0.45 sensitivity for SAL is significantly (P < 0.001) below the 0.90 reported earlier (5, 6). The negative predictivity of each of the ,four SITS is essentially the same, approximately 0.50, which should be judged against a prevalence of noncarcinogens in this database of 0.40. The positive predictivities of the SITS range from a high of 0.83 for SAL to a low of 0.66 for MOLY; this needs to be judged against a 0.60 prevalence of carcinogens among the 73 chemicals. As an illustration of the value of a positive SIT result, the a priori odds for carcinogenicity among the 73 compounds are 3 : 2, but a positive SAL result shifts these odds to nearly 5 : 1. In terms of concordance or percent agreement with the rodent carcinogenicity results, all four SITS had scores of approximately 60% (21, 22). Increasing the stringency of the evaluation criteria for positive SIT results produced the classic trade-off between sensitivity and specificity, with little if any consequent overall gain in concordance. On the basis of the current evaluation, there is no single test that is clearly superior to any of the other three SITS studied. The SAL assay does enjoy advantages when compared to the other three SITS: technical ease of conduct, wide availability, a sizable literature, and low cost. For all these reasons, the SAL test is generally central to any scheme that is intended to screen for carcinogens (23). It has long been recognized, however, that this test does not detect all carcinogens (7, 24); in the current study it missed over one-half (24 of 44). Thus, a pressing question is whether am of the other in vitro tests can serve as a complementa? assay to SAL (25); that is, can am assav detect the SAL-negative carcmogens without also detecting as positive an unacceptable number of noncarcinogens? A way to approach the question of complementarity is to strati6 the 73 chemicals by the qualitative ( + or -) results obtained with SAL. The data in Table 3 indicate that when one considers only the 49 SAL-negative chemicals, rodent carcinogenicity results show no association with the results obtained with MOLY, ABS, or SCE. For example, consider the use of ABS to complement SAL, with a positive result in either assay predicting a carcinogen. When the combined results are compared to predictions with SAL alone, an additional eight carcinogens are correctly identified but an additional six noncarcinogens are incorrectly predicted to be carcinogenic. Thus, sensitivity is improved somewhat, but at the expense of specificity, while the overall concordance is barely altered. The data in Table 4 show that ABS, SCE, and MOLY, however, do confirm positive SAL results very effectively; that is, the large majority of the 24 SAL-positive chemicals are also positive in ABS (79%), SCE (88%), and MOLY (96%). From a statistical viewpoint, the results with ABS, SCE, and MOLY lack association with rodent carcinogenicity when they are stratified bv the SAL outcome, a feature labeled SCIENCE. VOL. 236
Prediction of chemical carcinogenicity in rodents from in vitro genetic toxicity assays. X-2
Prediction of chemical carcinogenicity in rodents from in vitro genetic toxicity assays. X-8 19.8 1188.0
FE FE
226 196
842-07-9
87-29-6
E (II C y t e m b e n a
3.0
198.1
FE
222
2832-40-s
IP
1634.2
WA
265
518-47-8
207
1238.0
FE
220
3567-69-9
C.I. Acid Red 14
21739-91-3
772.5
FE
211
1936-15-8
C.I. Acid Yellow 73
70.7
GA
300
563-47-3
3-Chloro-2-methylpropene C.I. Acid Orange 10
70.7
SP
275
107-07-3
71.4
1931.3
FE
GA
124-48-I
Chlorodibromomethane 2Chloroethanol
594.2
FE
282
261
108-90-7
Chlorobenzene
1287.5
FE
84.9
214
105-60-2
70.7
386.3
353.7
650.0
17.7
6437.5
300.0
43.9
17.7
GA
FE
GA
FE
GA
FE
FE
GA
GA
RoRoutet d e n t dose*
GA
213
85-68-7
191 239 215
Butyl benzyl phthalate Caprolactam
so-057
10%60- 1
250
140-l l-4
Benzyl acetate
bis(2-Chloro-lmethylethyl) ether Bisphenol A
204
119-53-9
Benzoin
233
289
71-43-2
Benzene
2185-92-4
247
50-81-7
2-Biphenylamine HCI
216
2432-99-7
1lAminoundecanoic acid L-Ascorbic acid
253
234
Report numher*
2835-39-4
57-06-7
Chemical Abstract Services number
AUyI isovalerate
AlIyl isothiocyanate
Chemical
C.I. Disperse Yellow 3 C.I. Solvent Yellow 14 $ 8 Cinnamyl anthranilate
G ep
M: F: M: E F: M: F: M: F: M: E F: M: I F: HS M: F: M: E F: M: F: M: F: M: ES F: FS M: F: M: F: M: E F: M: L F: M: L F: L M: P,K F: M: MS F: MG
F: ZG,OC
M: UB F: E M: HS F: M: L,UB F: M: F: M: ZG,OC,S
Rat activityJ M: F: M: F: HS M: E F: M: F: M: ZG,HS,LU, HGJ’G F: ZG,HS,LU, 0, MG M: F: M: L,FS F: L,FS M: F: CS M: LU,L F: LU M: F: M: F: M: F: M: F M: E F: I. M: F: M: FS F: ES M: F: M: F: M: F: M: F: L M: F: M: L F: L M: F:\
Mouse activity§
100
3,333
N,A,3
_
A,2
0.3
_
-
-
_
A,2
-
_
-
-
-
A,2
A,3
_
E
_
E
-
-
E
Activity7
N,A,3
I
IO
10,000
5,000
10,000
10,000
3,333
10,000
3,333
10,000
11,550
333
33
IO
10,000
10,000
1,000
10,000
10,000
10,000
1,000
Dose I
SAL
25.0
10.0
N
A
25.3
40.0
250.0
A 4.0
1,500.o
A
10.0
5,000.0
5,000.0
1,250.o
N
_
-
120.0
200.0
5,000.0
5,000.0
N
10,100.0
A 80.0 23.2
2,540.O
510.0
5,000.0
1,250.O
50.0
124.0
200.0
5,000.0
2,000.0
5,000.0
3,000.0
l,ooo.o
BOO.0
5.0
N
IN
_
-
N
A
N
A
-
E
-
N
N
30.0
_
1.0
8.2
-
N,A
5.0
50.1
5,000.0
5,000.0
16.0
1,200.o
740.0
300.0
5,000.0
1,250.O
50.0
375.0
200.0
5,000.0
2,000.0
750.0
500.0
500.0
500.0
0.2
Doset t
-
-
A
N,A
N,A
-
-
A
N
-
-
-
-
A
WA
A c t i v i t y 7 D o s e * * Activityl
ABS
100.0
100.0
5,000.0
67.2
50.0
250.0
110.0
700.0
10.0
905.0
2,000.0
320.0
100.0
0.4
Dose#
MOLY
N,A
_
N,A
N
N,A
E
_
N,A
N,A
A
N
_
_
E
N,A
-
-
E
N
N
N
N,A
A
Activiry(
SCE
chemicals. Conclusions regarding sites of induced tumors, taken directly from the NTP Technical Reports, are presented separated by sex and species. Derailed SIT results for each of the chemicals and evahiation criteria for each of the in vitro assays can be found in one of the following references: SAL (13, 37), ABS and SCE (14, 38), and MOLY (15, 3Y), or can be obtained on request from the senior author. A negative response is indicated by (-), an equivocal response by “E,” and an inadequate study by; “I”; all other responses are positive.
Table 1. Tumorigenicity and genetic toxicity results for 73 chemicals. This table presents the qualitative SIT results, together with the lowest positive or highest negative dose tested for each of the
Science 1987; 236:933-41. - (CREATED FROM SCANNED DOCUMENT, PLEASE REFER TO ORIGINAL)
Prediction of chemical carcinogenicity in rodents from in vitro genetic toxicity assays. X-8
! 2 r N x
ii
x ;;;
zm
FE GA GA
GA GA GA FE FE GA
219 263 269 212 217 257 287 298 121 264 259
609-20-l
78-87-5
542-75-6
103-23- 1
117-81-7
101-90-6
868-85-9
597-25-l
120-61-6
9002-92-o
140-88-5
97-53-o
2,6-Dichloro-pphenylenediamine 1,2-Dichloropropane
1,3-Dichloropropene
di( 2-Ethylhexyl) adipate di(2-Ethylhexyl) phthalatc Diglycidyl resorcinol ether Dimethyl hydrogen phosphite Dimethyl morpholino: phosphoramidate Dimethyl tcrephthalate
Ethoxylated dodccyl alcohol Ethyl acrylate
Eugenol
D-Mannitol
Malaoxon
Isophorone
69-65-8
1634-78-2
78-59-l
236
135
291
276
293
33229-34-4
HC Blue 2
148-24-3
271
2784-94-3
I-Hydroxyquinoline
286
252
FE
FE
GA
FE
FE
FE
SP
GA
FE
208
68916-39-2
105-87-3
FE
FE
223
Hamam& water (witch hazel) HC Blue 1
Geranyl acetate
2783-94-O
GA
255
95-50- 1
1,2-Dichlorobenzenc
FD & C Yellow No. 6
IN
210
106-93-4
FE
IN
96-12-8
GA
FE
1,2-Dibromo-3-chloropropane 1,2-Dibromoethane
-
Route?
242 284 206
225
Report num her*
131-17-9
5160-02-l
Chemical Abstract Services number
Diallyl phthalatc
D & C Red 9
Chemical
6437.5
130.0
176.8
386.3
2600.0
148.6
5536.0
1414.8
3218.8
772.5
70.7
1545.0
638.0
424.5
141.5
8.5
297.1
1609.0
21.4
88.4
356.5
84.9
4.0
0.3
212.2
39.6
Rodent dose+ M: SP,L F: E M: F: E M: N,T F: N,T,AG M: N,CS,TV F: N,CS,LU, MG M: F: M: F: M: F: E M: FS,L F: FS M: F: M: L F: L M: FS F: FS M: LU,FS F: E M: HS F: HS M: F: M: F: M: FS F: FS M: F: M: F: M: F: M: F: M: E F: LU M: F: M: F: M: K F: M: F: M: F: -
Rat activity$
\
M: F: M: E F: E M: N,LU F: N,LU M: LU F: N,CS,LU, SC, MG M: F: M: L F: L M: L F: L M: I F: UB,LU,FS M: L F: L M: L F: L M: FS F: FS M: F: M: F: M: E F: M: F: M: FS F: FS M: E F: E M: F: M: F: M: F: M: L,TG F: L M: F: M: F: M: E F: M: F: M: F: -
Mouse activitys
10,000
10,000
10,000
1
3,333
100
10,000
3,333
10,000
333
10,000
1,000
333
10,000
10,000
3.3
10,000
10,000
10
1,667
33
333
100
0.3
10,000
1,000
Dosell
MOLY
ABS
SCE
6.5
_
1,845.O
_
21.3
-
A2
N,A,3
N,A,3
-
20.0
-
5,000.0
123.5
400.0
0.4
30.0
20.0
5,000.0
18.3
l,ooo.o
40.0
-
75.0
1,700.o
A,1
0.1
122.6
N,A,2
5,000.0
3.0
11.6
-
N,A,2
N,L
A,4
25.0
W
68.1
A,3
3,500.o 5,000.0
-
1,600.O
7.5
2,500.O
960.0
5,000.0
150.0
5,000.0
300.0
299.0
N
N
N
N
N
-
A
A
N
N
50.0
10.0
-
5,000.0
1,600.O N
A
1.6
5,000.0
N
400.0
-
1,370.o
A
100.0
250.0
N
N
202.0
A
250.6
50.0
A4
N
250.0
83.7
200.0
67.2
8.0
NJ
NJ -
A
A
-
-
N
N,A
N,A
-
5.000.0
50.0
500.0
2.6
240.0
119.0
5,000.0
70.0
5,000.0
75.0
150.0
30.6
10.0
l,ooo.o
250.0
0.05
25.0
400.0
29.9
_ N
112.7
16.0
59.0
5.0
10.0
160.0
500.0
N,A
N,A
_
N,A
N,A
A
-
-
N,A
N
N,A
NJ
NJ
-
N,A
E
N,A
A
-
-
N
N,A
N,A
N
E
N,A
N,A
N,A
A
N,A
N,A
A
A c t i v i t y l l Dose+ Activityll Dose** Activityll Dosett Activity(i
SAL
\
’
,
Science 1987; 236:933-41. - (CREATED FROM SCANNED DOCUMENT, PLEASE REFER TO ORIGINAL)
232 2Ci3
101-80-4
76-01-7
108-95-2
4,4’-Oxydianiline
Pentachloroethane
Phenol
Ziram , 238
235
17924-92-4
137-30-4
274
78-42-2
243
79-01-6
tris(2-Ethylhexyl) phosphate Zearalenone
200
15481-70-6
2,6-Toluenediamine 2HCl Trichlcqethylene
251
97
13463-67-7
26471-62-5
237
630-20-6
2,4- and 2,6:Toluene diisocyanate
209
1746-01-6
2,3,7,&Tetrachlorodibenzo-p-dioxin 1,1,1,2-Tetrachioroethane Titanium dioxide
138
127-69-5
Suhisoxazole
Sodium (2-ethylhexyl)alcohol sulfate Stannous chloride 231
194
7446-34-6
7772-99-8
193
50-55-5
Reserpine
256
247
75-56-9
Propylene oxide
126-92-1
240
121-79-9
Selenium sulfide
244
67774-32-7
Polybrominated biphenyl mixture Propyl gallate
205
266
150-68-5
98 248
15356-70-4
245
13552-44-8
,
108-78-l
4,4’-Methylenedianihne 2HC1 Monuron
nL-Menthol
Melamine
178.3 M: UB F: 515.0 ‘M: FE F: WA 8.5 M: TG,L F: TG 59.4 M: K,L FE F: 7 . 9 M : L,TG FE F: L.TG 176.8 M : i GA F: -I 990.4 M: WA F: 0.2 M: L GA F: L FE 1545.0 M: E F: IN 48.4 M: N F: N 0.2 M: AG FE F: 14.9 M: L GA F: L FE 2575.0 M: F: FE 262.5 M: E F: 1980.8 M: GA F: GA 0.00001 M: TG F: L GA 176.8 M: E F: FE 6440.0 M: F: GA 43.3 M: SCJ F: SCJ’, L, MG FE 2418 M: F: GA 7 0 7 . 4 M: I F: 707.4 M: E GA F: FE 12.1 M: F: 23.8 M: TG FE F: FE
M: F: M: L F: L M: F: L M: PI F: P&L M: F: E
Prediction of chemical carcinogenicity in rodents from in vitro genetic toxicity assays. X-8 10
1,000
10,000
1,000
10
33
10,000
1,000
1.000
100
333
10,000
1,000
10,000
333
1,000
10,000
3,333
333
1,000
5,000
1
666
1,111
750.0 1.0
-
74.1
N,A,3
-
0.063
N
-
1,670.O
A
146.2
-
65.0
14,900.o
N
200.0 A,4
0.025
15.0
250.0
800.0 N,A
50.0 A2
25.0
506.0
0.8
5,000.0
25.0
5,OlO.O
-
A
_
N
-
5.0
200.0
_ N
160.0
N
5.0
500.0
_ N
2,000.0
206.0
160.0
1,300.o
800.0
250.0
300.0
A
N
N
E
N
-
-
1.6
-
SO.0
-
200.0
4,200.O
0.1
10.0
0.4
50.0
_
N,A,2
N,4,3
-
20.0
300.0
70.0
50.0
1,100.o
_ N,A,5
250.0
200.0
160.0
A,4
_
tested(nucrograms per milhbter).
M: F: M: F: M: TG,L,AG F: TG,L,HS M: F: M, L,HG F: L,TG,HG M: L F: L M: F: M: L F: L M: E F: M: N F: N M: SV F: MG M: F: L,LU M: F: E M: F: M: F: M: L F: L,TG M: L F: L M: F: M: F: L&S
N,A
N
-
N
N
1.8
12.5
839.0
700.0
50.0
50.0
25.0
158.0
_
0.8
100.0
25.0
4,980.O
1.0
200.0
5.0
5.0
500.0
-
-
N,A
-
N,A
-
N,A
N
-
300.0
100.0
_ A
50.0
100.0
16.0
167.0
300.0
N,A
A
WA
-
_
_
N
-
NA
NJ
N
_
N
-
NJ
NA
-
N
-
NA
NA
_
NA
N
N,A
NA
NJ
_
E
Science 1987; 236:933-41. - (CREATED FROM SCANNED DOCUMENT, PLEASE REFER TO ORIGINAL)
Science 1987; 236:933-41. - (CREATED FROM SCANNED DOCUMENT, PLEASE REFER TO ORIGINAL)
Table 2. Operational characteristics of each of the four SITS for predicting carcinogenicity. Carcinogenicity of chemical substances as tested in rodents (+, carcinogenic; -, not carcinogenic) may be compared with the results from four SITS; for example, 20 carcinogenic substtices tested positive by SAL test, 24 did not.
Measure
+
_,
+
-
+
20 4
24 25
24 9
0.004 4865 (30-61)t (68-96) 83 (63-95)
Significance of association (Fisher’s Exact Test) Sensitivity* (% Specificity* (%)) Positive predictivityJ (%) Negative predictiviry ConcordanceP (%)
ABS
SAL
Carcinogenicity
I (%)
62 51
*Percentage of carcinogens melding a positive SIT result. result. SPercentage of STT positives that are carcinogens.
(36-66) (50-73)
69 55
SCE -
0.041 (39-70) (49-85)
73 (54-87) (48-72) 60 (34-66) 50
MOLY
+
-
+
32 16
12 13
31 16
45 73
x. 13 13
0.139 (55-83) (26-64) 66 (51-79) 50 (30-70) 60 (48-72)
0.098 (26-64) (57-85)
45 70
67 (52-80) 62 52 (31-72) (50-73)
tNumbers in parentheses, 95% contidence mten&. Percentage of noncarclnogcns yel&g a negative S?T IlPercentages of SlT negatives that are noncarcinogens. Wercentage of quahtative agreements between SITS
and rodent carcinogenicity test results.
Table 3. Association of rodent carcinogenicitv results with ABS, SCE, and
Table 4. Association of rodent carcinogenicity results with ABS, SCE, and
MOLY for 49 chemicals that are SAL nega&e.
MOLY for 24 chemicals that are SAL positive.
carcinogenicity + -
ABS
MOLY
SCE
+
-
+
-
+
-
8 6
16 19
15 12
9 13
12 12
12 13
conditional independence (26) (P = 0.75, 0.42, and 0.98 for ABS,
SCE, and MOLY, respectively). In summary, within the limits of this study, none of the other three in vitro SITS studied is a satisfactory complement to SAL in predicting rodent carcinogenicity. Because no single SIT is adequate to detect all carcinogens, a battery approach to screening for carcinogens has been frequently proposed as an improvement over any single STI’. If there is no complementary in vitro assay among these four assays, it is not surprising that batteries of two or more SITS do not appreciably improve the overall predictive performance of SAL alone. In fact, for the carcinogenicity data in Table 5, it can be shown that the maximum concordance for any prediction based on the 16 possible STT outcomes is 0.67 (49 of 73), while SAL alone has a concordance of 0.62 (45 of 73). This incremental gain is not significant. Table 6 summarizes the performance of the four-test battery for predicting carcinogenicity. Concordances range from 0.55 to 0.66, depending on the criteria used in defining a “positive”; similar results hold for two- and three-test batteries. To summarize the evidence for carcinogenesis into a positive or negative result is to grossly simplify a complex process manifest in a fairly long-term experiment. To determine whether performance by the SITS would improve if the target of prediction were refocused on some specific aspect of carcinogenesis, we considered five aspects. The SITS were then evaluated for their ability to distinguish: (i) the 2 1 chemicals judged to be “high potency” carcinogens, as determined by the lowest dose (milligrams per kilogram per day) producing statistically significant (P < 0.05) increases in tumor incidence; (ii) the 32 chemicals showing evidence of carcinogenesis in more than one sex or species group; (iii) the 20 chemicals exhibiting carcinogenic effects at more than one organ site; (iv) the 32 chemicals judged carcinogenic when liver tumors are excluded; and (v) the 26 chemicals showing increased incidences of malignant neoplasms (all sites combined). Table 5 summarizes the qualitative data for these responses. These five aspects of carcinogenicity yield prevalence rates ranging from
carcinogenicity + -
ABS
SCE
MOLY
+
-
+
-
+
_
16 3
4 1
17 4
3 0
19 4
;
27% (20 of 73) to 44% (32 of 73), which are close to the frequency of positive SAL responses (24 of 73, 33%). If these aspects are considered to be the targeted response, then the overall performance of SAL improves. For example, concordance for SAL increases from 62% to 67-74%, depending on which aspect of carcinogenicity is considered. Conversely, the performance of the other three SITS tends to diminish. For example, from the data on malignancy in Table 5, it can be shown that the concordance observed for SAL (53 of 73, 73%) is significantly greater than that observed for ABS (550/o), SCE (53%), or MOLY (52%). Similar results were found for the other four aspects of carcinogenicity. Further, for the five aspects above, no batterv constructed from the STTs exhibited improvement in predictive performance over that of SAL alone. Indeed, in many cases the concordance of the battery strategy was actually lower than that of SAL. Thus, regardless of whether the targeted response is “carcinogenicity” or some aspect of the carcinogenic response, there is little evidence that the four SITS have any enhanced ability to predict carcinogenicity beyond that of SAL.
Implications for Testing Strategies For more than a decade, the dominant paradigm motivating the use of STTs to predict chemical carcinogenicity has been that carcinogens are mutagens and, by implication, that mutagens are carcinogens (4). On the basis of the results presented here, it is clear that strong qualifications to these associations are needed. No single in vitro S?T adequately anticipates the diverse mechanisms of carcinogenesis; and, more important, the advantage of a battery of in vitro SITS is not supported by results of the present studv. These conclusions have major implications for carcinogen screekg and regulation based on SlT results. They also call into question proposed testing strategies (27) based on results from earlier attempts to evaluate SITS. Before implementing any proposed battery, substantial empirical evidence must be available to document the battery’s claimed performance.
938
SCIENCE, VOL. 236
Prediction of chemical carcinogenicity in rodents from in vitro genetic toxicity assays. X-2
Science 1987; 236:933-41. - (CREATED FROM SCANNED DOCUMENT, PLEASE REFER TO ORIGINAL)
demonstration of the reproducibility of the rodent studies, the assumption that the carcinogenicitv findings are correct remains necessary for the purpose of determining the predictivity of SITS. However, because of health concerns apart from cancer, it seems prudent not to dismiss as insignificant the in vitro mutagenicity of the noncarcinogens (30, 31). Although point mutations are phenomenologically different from cytogenetic effects, the four SITS showed good interassay agreement. All four SITS agreed for 33 of the 73 chemicals (45%), whereas three of the four SITS concurred for an additional 26 (36%) chemicals. In fact, the pairwise associations among the STTs were highly significant (all P < 0.01) and uniformly greater than the association between any one SIT and the carcinogenicity assay. In short, chemicals that were positive in one in vitro ST’T tended to be positive in other in vitro SITS representing three cell types and four end points.
The crux of the ditliculty encountered by any battery drawn from SAL, ABS, SCE, and MOLY is best illustrated by Table 7. This contains the 12 most potent carcinogens in Table 1 as judged by the criterion of lowest effective dose. There is strong consistency among the carcinogenicity determinations for the four sex and species combinations, and equally strong consistency among the results of the four SITS. Nevertheless, the three most potent carcinogens produced no genetic toxicity in any of the four SITS studied. One may speculate that these three carcinogens do not operate primarily by direct interaction with DNA, as suggested by the tmnorpromoting capability of two of the three carcinogens in model systems for two-stage liver carcinogenesis (28, 29). From Table 5 it is apparent that three other weaker carcinogens produced no positive response in any SIT, and that three noncarcinogens were positive in all four SITS. Possible explanations for the mutagenic noncarcinogens include low sensitivity of the carcinogenicity assay, in vivo detoxification into innocuous metabolites, or rapid excretion. In vivo SIT and pharmacokinetic studies might clarify this point, although they could not “prove” that any carcinogenicity assay result is in error; at present, only a larger and more definitive carcinogenicity assay, or carcinogenicity studies in other species, could do that. Without more extensive carcinogenicity studies or
To help put this project into its proper context, we emphasize certain features of the study:
Table 5. Patterns of SIT and rodent carcinogenesis results. The qualitative results in Table 1 are summarized and, as with the tumorigenicity results, the equivocal SIT results were treated as negative.
SAL
Aspects of carcinogemcity+
Overall carcinogenicity*
STI results
Potency
>l group
Malignancyf
Without liver
>l site
ABS
SCE
MOLY
+
-
High
Low
+
-
+
-
+
-
-I-
+ + + + -
+ + +
+ + +
14 0 2 0
: 0 0
9 0 1 0
13 0 0 0
5 f
4 1 0
20 0 1 2 2 0
10 0 0 0 1 0 0 1 0 :
4 0 2 0 2 0 0 0 5 0 0
A
0
: 10 29
: 3 21
: 2 2 2 20
: 1 0 4 24
2 0 0 0 2 0 0 0 0 0 0 1 4 1
11 0
1 8
1 0 2 0 0 0 0 1 1 1 0 0 2 1 0 3 12
12 0 2 0 1
: 0
5 0 1 0 1 0 0 0 3 0 0 1 5 2 2 3 23
-I+ -I-
+ + t
+ + -
-
-
Totals
+ +
1
0
1 5 3 2 6 44
0
: 0 0 4 1 0 1 3 2 2 3 32
x : 2 0 0 1 2 2 4 32
9 12
6 0 2 0 1 0 0 0 7 3
2 1 4 26
l Carcmogenicity for each chemical as reported in Table 1. The em iricd de&on for each SlT result that produces maximum concordance with carcmogenicitv outcome is indicated m boldface This maximum concordance is 0.67 (49 o P73). tThc 44 carcinogens are subdivided on the basis of four d&rent cnteria: (i) potency (high-potency carcmogens defined ,T’ ermcals producing effects at doses ~60 mgkg per day); (ii) chemicals producing carcinogenic effects in more than one sex and species group; (iii) chemicals pr+ucmg carcinogenic effects at more than one organ site; and (iv) chemicals producing carcinogemc effects when liver tmnors are excluded. SRcsults based on mcrcased mcldenccs of malignant tumors (alI sites combined); not necessariIy a subset of the carcinogens evaluated on the basis of site-specific effects. Table 6. Operational characteristics of batteries of SlTs for predicting carcinogenicity. Number of posnive SITS required for prediction of carcinogenicity* Measure Sensitivityt (%) Specificity* (%) Positive predictivity§ (%) Negative predictivity I (%) ConcordanceT (%)
One or more
Two or more
Three or more
86 34 67 62 66
70 48 67 52 62
55 72 75 51 62
(38 (10 (38 (10 (48
of of of of of
44) 29) 57) 16) 73)
(31 (14 (31 (14 (45
of of of of of
44) 29) 46) 27) 73)
(24 (21 (24 (21 (45
of of of of of
44) 29) 32) 41) 73)
All four 32 90 82 46 55
(14 (26 (14 (26 (40
of of of of of
44) 29) 17) 56) 73)
ARTICLES 939
22 MAY 1987
Prediction of chemical carcinogenicity in rodents from in vitro genetic toxicity assays. X-2
Science 1987; 236:933-41. - (CREATED FROM SCANNED DOCUMENT, PLEASE REFER TO ORIGINAL) \
I .~_
Table 7. Carcmogenicity and S?T results for the 12 most potent rodent carcinogens as defined by lowest effective dose from Table 1. Carcinogenici? Chemical name
Rat M
/
2,3,7,8-Teuachlorodibenzo-p-dioxin Polybrominated biphenyl mixture Reserpine i,2-Dibromo-3-chloropropane Cytembena 1,2-Dibromoethane 4&‘-Oxvdiauiline Diglycidyl resorcinol ether 4,4’-Methylenedianie 2HCI Zearalenone Selenium sulfide Ally1 isothiocyanate
l
Milhgrams per kilogram per day.
+ + + + + + + + + + +
Lowest effective rodent dose*
Mouse F
+ + + + + + + +
&
M
F
+ + + + + + + + + -
+ + + +
STI SAL
MOLY
-
1.0 x 10-5 0.2 0.2 0.3 3.0 4.0
+ + + + + + _
ABS
SCE
-
“:
_ -
+
‘/
+
+ + +
;:: 8.5 12.1 14.9 17.7
+ ,
+ +
.+
+
+
+
+
+
+
+
+
+
‘E, equivocal.
1) Standard protocols were used to mimic the major use of SITS worldwide-screening for mutagens and carcinogens; no attempt was made to optimize protocols for specific chemicals. 2) The 73 NTP chemicals and their 60% incidence of carcinogenicity are probably not representative of the universe of chemicals but rather reflect the recent chemical selection process for the NTP carcinogenicitv assay. 3) The smail, diverse group of chemicals precludes a meaningfU1 evaluation of the predictive utility of chemical structure information. 4) The NTP is currently testing these same 73 chemicals in two in vivo SITS for chromosomal effects. 5) Complete data for an additional group of 30 to 40 NTP chemicals will be gathered on carcinogenicitv and the four in vitro SITS to attempt to confirm the current findings. The standard against which the performance of SITS is measured has changed dramatically in the past decade. The high levels of concordance published in the early 1970s were accurate at the time. Nearly all known carcinogens tested were genotoxic, and there was little experimental evidence on which to base a judgment of noncarcinogenicity which, taken together, restricted assessment of test performances with noncarcinogens. With the increasing availability of results from NC1 and NTP 2-year carcinogenicity studies in rodents, higher frequencies of nongenotoxic carcinogens and genotoxic noncarcinogens have been observed; this has resulted in the reduced concordance of the SIT results with carcinogenicity results. It is clear that even with a battery of assays, not all rodent carcinogens are in vitro mutagens nor are all in vitro mutagens rodent carcinogens. If current in vitro SITS are expected to replace long-term rodent studies for the identification of chemical carcinogens, then that expectation should be abandoned. SITS do, however, continue to offer an economical, rapid, and dependable means to detect genotoxic chemicals. There is a range of applications in which SITS have been used successfully, from the identification of mutagenic fractions in complex mixtures such as cooked meat (32,33) or air pollutants (34) to the early identification of genetic toxicity in the development of new chemical products (35). Requirements for the use of STT have not been consistent in both the national and international regulatory agencies. This is evident in the variety of testing requirements (8) and the different impacts that positive test results have on the registration or further testing requirements of chemicals. Consensus on these matters is not likely to occur in the near future, but agreement should be possible in certain areas. For instance, any time a new test or saategy is proposed, it is imperative that there be documentation by a substantial set of systematically acquired test results on well-defined rodent carcinogens
and noncarcinogens (36). The current study represents a prototype of the evaluative effort needed for such documentation. Results of the current stud!7 focus attention on two questions involving discordances between carcinogenicity and genotoxicity test results: (i) Do nongenotoxic rodent carcinogens pose the same carcinogenic risk to humans as those that are genotoxic? (ii) Can the apparent high frequency of in vitro genotoxic rodent noncarcinogens be explained as a combination of artifacts arising from extremel!; high dosing in in vitro tests or the failure of many bona fide in vitro genotoxins to express their genetic toxicitv in whole animals? Until these questions are resolved, chemicals that show mutagenic effects, particularlv if such effects are observed in vivo, must be initially considered to pose human health risks as long as the somatic mutation theory of cancer remains a viable explanation for the etiology of some chemically induced cancers. REFERENCES AND NOTES 1. D. S. Suaus, J. Natl. Cancer Inst. 67, 233 (1981). 2. B. D. Crawford, inAdvanus in Modem Envinmmental Tat&&v, W. G. Flamm and R. J. Lxxcntzen, Eds. (Princeton Scientific, Priiceton, NJ, 1985), pp. 13-59. 3. B. N. Ames, S&w 204, 587 (1979). 4. _ W. E. Durston. E. Yamasaki, F. D. Lee, Pnx. Natl. Acad. Sti. USA. 70, 2281 (1973). 5. J. McCann, E. Choi, E. Yamasaki, B. N. Ames. ibid. 72, 5135 (1975). 6. I. F. H. Purchase et al., Nature (London) 264. 624 (1976). 7. T. Su@mura et al., in FundamcntaL in Catam- Prtvmtion, P. N. Magee et al., Eds. (University Park Press, Baltimore, 1976), pp. 191-215. 8. D. J. Be- and M. H. Litchfield, in Emhtion of Short-Term Tcm fw Cwcinogcm:
Rqbort of the Intematiaal Pnyramnu on Chemical Sa cm+ Cdlabmatitr St$v a i n Vim Assq vol. 5 of Pmg’csr in Mutatm Restarc 6-Smcr, J. Ashby et al., Eds.
1::
11. 12.
13. 14. 15.
(Elsevier, Amsterdam. 1985), pp. 727-740. M. D. Waters and A. Auletta, J. Chm. If: Comput. SEi. 21. 35 (1981). M. D. Shelby and S. Stasiewicz, En&m. Mutagm. 6,871 (1984); L. E. Kier ct al., Mutat. Res. 168, 69 (1986); V. A. Ray ct nl., ibid. 185, 197 (1987). J. A. Sontag, N. I? Page, U. Saffiotti, U.S. Dep. Health Edzu. WelfircPubl. (NIH) Carcim~. Tcth. RQ. Ser. 1, 76 (1976). J. K. Hascman. J. E. Huff, E. Geiger, E. E. McConnell. En+vn. HcultbPqwt.. in ress J E Huff. J. K. Haseman, E. E. McConnell, J. A. Moore, in Safstl E valuuam ;: ofDqgr and Chtm&, W. E. Lloyd, Ed. (Hemisphere, Washington, DC. 1986), p. 41 l-423. S. Haworth, % Lwior, K. Mortelmans. W. Speck, E. Zeiger. Enmnm. Mutqtn. 5 (suppl. 1). 3 (1983). S. Galloway et al., ibid. 7. 1 (1985). B. Myhr, L. Bowers, W. Caspary, in Em&ztnm of Short-Tm Tcsts~% Carti~m: Rywt of the IntrmacMIIJ P9y-t em Chmd SJw&yyy-$~ Z-J fJg, Vitro Assays, vol. 5 of Pnyras m Mutatm Rcsca
(Elsevier, Amsterdam, 1985), p 555-568. 16. W. J. Caspaty et al., Enmm. &tagc”., in press. ’ 17. B. H. Matgolin, N. Kaplan, E. Geiger, Pm. Natl. Acad. Sci. U.SA. 78, 3779 (1981). 18. B. H. Margolin a a[., Entrimn. Mwwcn. 8,183 (1986). 19. W. J. Caspary n al., ibid., in press. 20. I. F. H. Purchase, Br. I. Cancer 41, 454 (1980). 21. V. C. Dunkcl cr al., Envinm. Mutap. 7 (so pl. 5), 1 (1985). 22. Two other Snlmonzlkz studies with 60 chcmics (21) and 224 chemicals [E. Zeigcr, Cancer Rer. 47, 1287 (1987)] from the NC1 and NTP database have been reported. The 224 chemicals studied by Geiger included the 60 from the study b\r
940
SCIENCE, VOL. 236
Prediction of chemical carcinogenicity in rodents from in vitro genetic toxicity assays. X-2
Science 1987; 236:933-41. - (CREATED FROM SCANNED DOCUMENT, PLEASE REFER TO ORIGINAL)
D,,&I cf al. (21) and 69 of the 73 from the present SF*: In both studies, a htghcr sen&vity and lower spcclficltv and poslave prcdtcawty than seen here were rcporccd. The b&m reported by Zciger were as s+ficity, 0.70; and positive pdkti~~~, 0.69.. sNda cm be attributed to &fferences m the relauve pro cher&al &sses rcprcxnted in the databases. In all three o the ConCo&CU w e r e s i m i l a r . 23. ,, H, W&burger aqd G. M. Williams, Scimcz 214,401 (1981). 24. ,, ~by, in +&&on of Short-Tmn Ttstrfbr Camngem: Rrpmt o the Inrmrarra alcdhbmafnu Prqqram, vol. 1 of Pmgm in Mz+ta&m Ram&&, F. J.‘d, m and J. Ashby, Eds. (Ekevicr, Amsterdam, 1981), p. 112-171. mqqm: Rqmt of rhc 25. J, Ashby et al., in Emfwtm of Short-Tm Tcrk@ c!. Inr# p*qp*atr#crc on Cbemiral %&v’s Coil&m&v6 SW a in Vim Asrays, voI. 5 of Pqqrm in Mut& Resazrcb Saks, J. Ashby et nl., Eds. (Elswier, 26. 27. 28. H. C. Pitot, T. Goldsworthy, H. A. d~pb& A. ~ol&i, C&Y Rcr. 40, 3616 (1980). 29. M. Nishizumi, CMccr Left. 2, 11 (1976). Committee 17, Scicnrc 187, 503 (1975). 30. 31. &mminec on Chemical Environmental Mutagens, I&ntz@q and Estimating rbe~ r&r& Inqwt of Chcmid Mntrhqnrr (National Academy Press, Washington, DC, 1983), pp. l-15.
32. F. T. Hatch. J. S. Fclton. D. H. Stuermer. L. F. Bjeldanes, tn ChnnrrrJMwagnu: Pnnnpks and Mrdd.r fm Thnr Lktunm, F. J. de Serres, Ed. (Plenum. New York,
35. J. P. Tassignon, Food Chem. Tmcic~l. 23, 5 (1985). 36. S. Nesnow et al., Mum. Rcs. 185, 1 (1986). 37. K. ~Morrekruns, S. Haworth, W. Spe& E. Zeiger, Tcmwl. Appl. Phannacol. 75, 137 (1984). E. Zeigcr and S. Haworth. in Evakuztwn of Shorr-Tm Tt+sa jir Cwcino~mm:‘Rqort
of the Inrmraciaal
Pmgr~~me cm Chmical
Safes Cdabora-
rive Stu& on tn Vihv Asrqvs. vol. 5 of Pmpreu in Mutarion Rmarch S&s, J. Ashby cr al., Eds. (Elsevier, Amsterdam, 1985), pp. 187-199: E. Zeigcr, S. Haworth, K. Mortelmans, W. Speck, Entim. Murngcn. 7,213 (1985); (21); K. Mortelmans cr al., ibid. 8 (su pl. 7). 1 (1986); D. A. Canterrr nl., Murat. Ru. 172, 105 (1986); E.ZcigerctJ , Enmrmt Mutagm., in press; E. Zci er et al., m preparation. 38. S. M. Galloway cr al., Entim. Murngm., in press; b. K. G&d, B. Anderson, E. Zeiger, M. D. Shelby, in preparanon; K. Loveday et al., in preparation; J. Ivett cf al.. * oreoaration. 39. D.‘Mc&gor. A. Brown, P. Cattanach, I. Edwards, D. McBride, W. J. Caspary, in preparation; B. Myhr and W. Caspary, in preparation; D. McGregor ef nl., in re aradon; B. Myhr and W. Caspary, in preparation. 40. !A: Gold cf al., Envim. Health Pmpat. 58, 9 (1984). 41. The contmued interest and encouragement of M. Mcndelsohn re arding this study are much appreciated. The critical commentx of B. Schwca, J. Se l&r k, and D. Hoc1 are also appreciated.
Prediction of chemical carcinogenicity in rodents from in vitro genetic toxicity assays. X-2