Introduction
The National Cancer Institute (NCI) and the National Toxicology Program
(NTP) have designed, carried out, and evaluated more than 400 long-term
carcinogenicity studies in laboratory rodents (1-3). The purpose
of these studies is to identify chemicals that may pose a carcinogenic hazard
to humans. The results of these studies have been presented in a series
of technical reports and in the scientific literature. These data are used
by the international scientific community and by various governmental agencies
in making regulatory decisions affecting public health.
The focus of the NCI/NTP studies is the evaluation and interpretation
of site-specific carcinogenic effects (4) . A recent publication
(2) presented a tabular summary of site-specific carcinogenicity
results for 379 NCI/NTP studies. The purpose of this paper is to investigate
correlations in chemically related site-specific carcinogenic responses
in this database both within and between species for a variety of different
target sites.
Other investigators have considered certain aspects of these associations.
For example, Gold et al. (5,6) investigated the predictive value
of site-specific carcinogenic responses for chemicals in the Carcinogenic
Potency Database. Their evaluation differed from ours in that their site-specific
analyses involved primarily the same target site in the other species, and
the evaluation used data from a variety of sources and thus included studies
that did not employ the standardized protocol used by the NCI/NTP.
The primary objective of our investigation is to identify associations
between site-specific carcinogenic effects that may increase our understanding
of the value and limitations of long-term rodent studies. The implication
of these associations on the possible use of a reduced protocol for rodent
carcinogenicity evaluation is also discussed.
Methods
The carcinogenicity database used in our analysis consisted of the 379
long-term NCI/NTP studies in rats and/or mice considered by Huff et al.
(2). The site-specific carcinogenic effects observed in each sex-species
group were evaluated for possible associations. The specific target organs
used in our evaluation were the 13 most frequent sites of carcinogenicity
in these studies (Table 1). Overall, 89% of the rat carcinogens and 91%
of the mouse carcinogens in the NCI/NTP database produced carcinogenic effects
in at least one of these sites.
Initially, experimental outcomes were divided into three categories:
positive, negative, and equivocal. To simplify the analysis, equivocal outcomes
were subsequently combined with negative results, because in our judgment
chemicals producing equivocal effects are closer to being negative than
positive (7). However, alternative analyses (results not shown) in
which equivocal responses were considered positive or were excluded altogether
produced similar results to those reported in this paper.
We considered several types of associations: first, for each target site,
we evaluated the correlation in carcinogenic response between males and
females within a species for those studies with adequate experiments in
both sexes. Second, we evaluated correlations between species for a given
target site for those 313 studies with experiments adequate for evaluating
carcinogenicity in all four sex-species groups. In these analyses, a species
was considered positive if a carcinogenic response was observed in males,
in females, or in both sexes. Finally, we evaluated correlations between
different target sites both within and between species. Our analyses included
noncarcinogens as well as carcinogens.
The strengths of the associations were assessed by Fisher's exact test.
Because the analyses involved a large number of comparisons, p<0.01
(rather than p<0.05) associations were considered significant.
This reduces the likelihood of false positive outcomes.
Results
Within a species, most target sites showed a significant correlation
between males and females with respect to site-specific carcinogenic responses.
Table 2 summarizes the most striking of these associations for the 349 rat
studies and the 339 mouse studies with adequate experiments for both sexes.
The greatest consistency was found for the forestomach. For example,
in rats, all 11 studies showing forestomach carcinogenicity in female rats
also showed these effects in males rats (Table 2). Similarly, in mice, 14
of the 15 studies showing carcinogenic forestomach effects in one sex also
showed similar effects in the other sex (Table 2). Other sites showing a
strong consistency between males and females include liver (rats and mice),
Zymbal gland (rats), and lung (mice). In contrast, mammary gland tumors
showed relatively low association between males and females because they
were induced almost exclusively in females. Also, there were far more carcinogens
that produced kidney tumors in male rats than in female rats.
There was also a strong correlation between species for certain site-specific
carcinogenic effects. For example, a chemical carcinogenic to the liver
of male and/or female rats was four times more likely to be a mouse liver
carcinogen than was a chemical not carcinogenic to rat liver (76% = 25/33
versus 19% = 53/280; see Table 3). In analogous comparisons, the corresponding
probabilities for forestomach, thyroid, and mammary gland tumors were 57%
versus 2%, 44% versus 1%, and 22% versus 1%, respectively (Table 3). All
of these differences are highly significant (p<0.001).
Equally interesting were the sites showing no significant correlation
between species. For example, there were no chemicals in the database that
produced adrenal pheochromocytoma or preputial/clitoral gland tumors in
both rats and mice (Table 3). The lung was another site showing no significant
interspecies correlation in chemically related carcinogenicity: only 2 of
the 23 chemicals producing lung tumors showed these effects in both rats
and mice (Table 3). If all target sites are considered collectively, the
overall interspecies concordance in carcinogenic response is 74% (233/313;
see Table 3). This interspecies association is consistent with other published
estimates (5,8).
Within a sex-species group, certain target sites tended to show carcinogenic
effects for the same chemicals. For example, in rats, tumors of the skin,
Zymbal gland, preputial/clitoral gland, and (to a lesser extent) mammary
gland were intercorrelated (Table 4). Eight chemicals (C.I. acid red 114,
C.I. direct blue 15, 3,3´-dimethylbenzidine dihydrochloride, 3,3´
dimethylbenzidine dihydrochloride, 2,4-diaminoanisole sulfate, glycidol,
5-nitroacenaphthene, and 5-nitro-o-anisidine) produced carcinogenic
effects at 4/4 or 3/4 of these sites, and 8 other chemicals [3-amino-9-ethylcarbazole
HCl, benzene, C.I. basic red 9, 2,4-dinitrotoluene, 3,3´-dimethoxybenzidine-4,4´-diisocynate,
hydrazobenzene, nithiazide, and tris(azi-ridinyl)-phosphine sulfide] produced
effects at 2/4 of these sites.
Two other target sites that showed an association both within and across
species were liver and thyroid gland. The likelihood of a chemical being
a rodent liver carcinogen was much higher for those chemicals that were
thyroid follicular cell carcinogens than for those chemicals that were not
(Table 5). For example, of the 18 chemicals showing thyroid carcinogenicity
in rats or mice, 61% (11/18) also produced liver tumors in one or both species.
In contrast, for those chemicals not producing thyroid follicular cell tumors,
only 25% (75/295) produced rodent liver tumors. This association between
liver and thyroid neoplasms has been previously noted (9).
There were relatively few other notable correlations between different
sites of carcinogenicity, and the most significant of these are summarized
in Table 6. Chemically related lung tumors in mice were associated with
forestomach and hematopoietic system tumors in mice and with mammary gland
tumors in both rats and mice. All six chemicals producing adrenal pheochromocytomas
in mice were also mouse liver carcinogens. Liver and Zymbal gland tumors
in rats were also correlated.
Kidney and urinary bladder tumors in rats showed no significant association
with any other tumor type in rats or mice. Circulatory system tumors in
mice did not show strong associations with other tumor types. Interestingly,
there were no significant negative correlations in chemically related site-specific
carcinogenic responses in the database. That is, there were no target sites
for which chemicals producing carcinogenic effects at that site were significantly
less likely to produce carcinogenic effects at another site.
Although mesotheliomas were not part of our primary analysis because
of the relatively low frequency of carcinogenic effects, it is noteworthy
that all six chemicals producing mesothelioma in male rats (cytembena, 1,2-dibromoethane,
3,3'-dimeoxybenzidine dihydrochloride, 3,3´-dimethylbenzidine dihydrochloride,
glycidol, and O-toluidine hydrochloride) also produced mammary gland
tumors in female rats. This association is highly significant (p<0.0001).
Another tumor site that was not part of our primary analysis was oral
cavity, which showed chemically related neoplastic effects for eight chemicals
in rats (none in mice). These tumors are strongly associated with the skin-Zymbal
gland-clitoral/pre-putial gland-mammary gland "syndrome" noted
in Table 4: 6/8 chemicals producing oral cavity tumors in rats also produced
2 or more of the other tumors in the syndrome.
Approximately 10% (32/313) of the chemicals adequately evaluated in all
four sex-species groups produced single site, single sex-species carcinogenic
effects. The organs most often involved were liver (five chemicals for female
mice; four for male mice) and kidney (four for male rats). Approximately
58% of the carcinogens in the NTP database are positive in Salmonella,
compared to 30% of the noncarcinogens (10). Gold et al. (11)
found similar results for chemicals in the Carcinogenic Potency Database.
Both investigators (10,11) reported that multisite carcinogenicity
was associated with Salmonella mutagenicity, a finding consistent
with our own evaluation. For example, consider the chemicals that produced
the multisite syndrome of carcinogenic effects noted in Table 4. Of the
16 chemicals showing site-specific effects in at least 2 of the 4 target
sites (skin, Zymbal gland, clitoral/preputial gland, and mammary gland)
all but 2 (benzene and C.I. direct blue 15) were positive in Salmonella.
Similarly, all six chemicals producing mesotheliomas in male rats and mammary
gland tumors in female rats were positive in Salmonella.
In contrast, chemicals producing kidney tumors (which were not associated
with other tumor types) tended to be nonmutagens. Of the 25 chemicals producing
male rat kidney tumors, only 28% (7/25) were mutagenic in Salmonella.
Further, only 31% (4/13) of the chemicals noted above with single sex-species
liver or kidney tumor effects were mutagenic in Salmonella. These
percentages are similar to those observed for the noncarcinogens in the
database.
Among the liver carcinogens in the database, 70% (16/23) of the chemicals
that produced these neoplasms in both species were mutagenic in Salmonella
compared to only 30% (7/23) for those chemicals whose only carcinogenic
effect was liver tumors in mice (2,12). This difference in Salmonella
mutagenicity (70% versus 30%) is statistically significant (p<0.05).
One possible explanation for this difference is that a relatively high proportion
of mouse-liver-only carcinogens are chlorinated compounds, which tend to
be nonmutagenic (5,11).
Our evaluation also permits a comparison of the male-female correlation
with the interspecies association in site-specific carcinogenicity. This
evaluation is summarized in Table 7. Although the interspecies correlation
varies for each site of carcinogenicity, if the 13 major target sites are
considered collectively, the overall probability is 35% (61/173) that a
site-specific carcinogenic effect in rats will also be produced by that
chemical in mice at the same site (Table 7). Similarly, 37% (61/167) of
the site-specific carcinogenic effects in mice are also produced by the
same chemical in rats (Table 7).
For these same target sites, 66% (97/146) of the site-specific carcinogenic
effects observed in female rats were also observed in male rats (Table 7).
For mice, 65% (103/159) of the site-specific effects observed in female
mice also occurred in males. Thus, the agreement between sexes within a
species is considerably higher than the corresponding agreement between
species.
Discussion
When evaluating associations in site-specific carcinogenic responses,
we considered all experimental outcomes, negative as well as positive. For
example, there were 61 chemicals that produced liver carcinogenicity in
a single species and 25 that produced a liver tumor response in both rats
and mice (Table 3). However, this information alone is insufficient to evaluate
interspecies correlation in liver tumor carcinogenicity. If there is no
interspecies correlation, the likelihood of observing a mouse liver carcinogen
should be the same regardless of whether a chemical is a rat liver carcinogen.
Thus, had there been only 17 chemicals not carcinogenic to the liver of
rats or mice, then the proportion of mouse liver carcinogens would have
been 76% (25/33) for rat liver carcinogens and 76% (53/70) for chemicals
not carcinogenic to rat liver, indicating no association. However, the actual
database contained 227, not 17, chemicals not carcinogenic to the liver
of either species (Table 3). Thus, only 19% of the rat liver noncarcinogens
were mouse liver carcinogens, and this difference in response (76% versus
19%) implies a strong association, as noted above.
Our evaluation revealed a number of target sites showing significant
associations in chemically related carcinogenic responses between species.
However, such agreement was far from complete, and the overall probability
that a chemical carcinogenic at a particular site in rats will be carcinogenic
at the same site in mice (and vice versa) is approximately 36%.
Consideration of different target sites produced similar results. For
example, despite the significant (p<0.01) correlation between
chemically induced liver and thyroid tumors, the vast majority of liver
carcinogens (in either species) did not produce thyroid tumors. Similarly,
many chemicals produced thyroid tumors without increasing the incidence
of liver neoplasms (see Table 5).
There was much better agreement between sexes within a species, and this
strong correlation in carcinogenic response between males and females has
been noted previously (3,5,8). However, these earlier analyses were
based on overall carcinogenicity rather than on site-specific effects. Our
evaluation shows that this same strong correlation also holds for site-specific
carcinogenicity.
These results may have implications for the experimental design of long-term
rodent studies. Some investigators have noted that from the standpoint of
detecting carcinogenic effects, relatively little sensitivity would be lost
by restricting the study to two rather than four sex-species groups (5,8).
Because the male-female correlation in carcinogenic response is much stronger
than the interspecies correlation, a reduced protocol of one sex of each
species would be more sensitive for detecting rodent carcinogenicity than
using both sexes of either species. The best choice for a reduced protocol
may be male rats and female mice (8).
For example, of the 162 NCI/NTP chemicals adequately evaluated in all
four sex-species groups and found to be carcinogenic in at least one group
(see Table 3), only 15 (9%) would not have been detected as rodent carcinogens
in a reduced protocol of male rats and female mice. These 15 chemicals,
listed in Table 8, generally produce single-site, single sex-species carcinogenic
effects. Moreover, the toxicities of these chemicals as measured by the
maximum tolerated dose (MTD; 13) are somewhat lower on average than
those of the other carcinogens in the database. More than half of the chemicals
(8/15) are negative in the Salmonella assay. None are listed as "reasonably
anticipated to be" human carcinogens in the NTP Sixth Annual Report
on Carcinogens (14) or as being "possible" or "probable"
human carcinogens by the International Agency for Research on Cancer (IARC).
Thus, it appears that a reduced protocol would not miss many of the carcinogens
likely to be of public health significance.
However, reduced sensitivity involves more than not detecting carcinogens.
Many national and international organizations (e.g., IARC, the NTP in its
Annual Report on Carcinogens) generally require evidence of carcinogenicity
in two species before a chemical is considered "possibly," "probably,"
or "reasonably anticipated to be" a human carcinogen. A reduced
protocol would miss some of these two-species carcinogens, declaring them
to be one-species carcinogens. For the NCI/NTP database, 12 two-species
carcinogens would have been reduced to one-species carcinogens using a reduced
protocol of male rats and female mice, and these chemicals are summarized
in Table 9.
Unlike the chemicals in Table 8, most of these chemicals produced carcinogenic
effects at multiple target sites. The toxicities (MTDs) of these chemicals
were similar on average to those of other carcinogens in the NCI/NTP database,
and half of these chemicals that were tested in Salmonella (4/8)
were positive. Even so, only 1 of these12 chemicals (chlorendic acid) is
listed in the NTP Sixth Annual Report on Carcinogens as 1 of the 150 substances
and medical treatments "reasonably anticipated to be [human] carcinogens"
(14). Chlorendic acid is also the only 1 of these 12 chemicals considered
by IARC to be a "possible" human carcinogen (15).
If these two types of sensitivity loss are considered collectively, less
than 10% (27/313) of the chemicals evaluated by the NCI/NTP (17% of the
carcinogens) would have shown some sensitivity loss by the use of a reduced
protocol of male rats and female mice. Moreover, of these 27 chemicals,
only one is currently considered by the NTP or IARC to be a likely or possible
human carcinogen.
It must be noted that the NTP Annual Report on Carcinogens (14)
has exposure and production criteria that may not have been met by certain
chemicals in Tables 8 and 9. Further, less than half of the chemicals in
these two tables has been formally evaluated by IARC in its series of monographs.
Nevertheless, our evaluation suggests that the carcinogens missed by using
a reduced protocol of male rats and female mice may not be those chemicals
most likely to pose a carcinogenic threat to humans.
A further consideration in the decision of whether to use a reduced protocol
is the possible loss of supporting information in one sex when interpreting
results in the other sex. That is, because of the high correlation in carcinogenic
response between males and females, there may be instances in which a borderline
carcinogenic effect in one sex-species group would be called positive rather
than equivocal or negative because of a similar carcinogenic effect at that
same site in the other sex of that species. This supporting information
would be lost in a reduced protocol of one sex of each species. However,
an evaluation of the magnitude of the carcinogenic responses in recent NCI/NTP
studies suggests that this type of sensitivity loss would not be a frequent
occurrence.
The loss of sensitivity associated with a reduced protocol must be balanced
against the possible advantages of evaluating more chemicals and/or reallocating
resources to basic mechanistic studies. Although our evaluation suggests
that this more limited experimental design is an option that may be appropriate
in some instances, the decision of whether to use such a reduced protocol
is best made on a case-by-case basis.
In conclusion, our evaluation 1) shows that the previously reported high
male-female and interspecies correlations in carcinogenic response also
hold for many site-specific neoplasms, 2) identifies significant associations
in site-specific carcinogenic responses that may be useful in further research
efforts; and 3) suggests that because of the high correlation in carcinogenic
response between males and females, in many cases a reduced protocol of
one sex of each species may be sufficient to detect rodent carcinogenicity.
