All known human carcinogens that have been studied adequately
for carcinogenicity in experimental animals have produced
positive results in one or more animal species (Wilbourn
et al., 1986; Tomatis et al., 1989). For
several agents (e.g. aflatoxins, diethylstilbestrol, solar
radiation, vinyl chloride), carcinogenicity in experimental
animals was established or highly suspected before epidemiological
studies confirmed their carcinogenicity in humans (Vainio
et al., 1995). Although this association cannot
establish that all agents that cause cancer in experimental
animals also cause cancer in humans, it is biologically
plausible that agents for which there is sufficient
evidence of carcinogenicity in experimental animals
(see Part B,
Section 6b) also present a carcinogenic hazard to
humans. Accordingly, in the absence of additional scientific
information, these agents are considered to pose a carcinogenic
hazard to humans. Examples of additional scientific information
are data that demonstrate that a given agent causes cancer
in animals through a species-specific mechanism that does
not operate in humans or data that demonstrate that the
mechanism in experimental animals also operates in humans
(see Part B,
Section 6).
Consideration is given to all available long-term studies
of cancer in experimental animals with the agent under
review (see Part A, Section
4). In all experimental settings, the nature and extent
of impurities or contaminants present in the agent being
evaluated are given when available. Animal species, strain
(including genetic background where applicable), sex,
numbers per group, age at start of treatment, route of
exposure, dose levels, duration of exposure, survival
and information on tumours (incidence, latency, severity
or multiplicity of neoplasms or preneoplastic lesions)
are reported. Those studies in experimental animals that
are judged to be irrelevant to the evaluation or judged
to be inadequate (e.g. too short a duration, too few animals,
poor survival; see below) may be omitted. Guidelines for
conducting long-term carcinogenicity experiments have
been published (e.g. OECD, 2002).
Other studies considered may include: experiments in
which the agent was administered in the presence of factors
that modify carcinogenic effects (e.g. initiation-promotion
studies, co-carcinogenicity studies and studies in genetically
modified animals); studies in which the end-point was
not cancer but a defined precancerous lesion; experiments
on the carcinogenicity of known metabolites and derivatives;
and studies of cancer in non-laboratory animals (e.g.
livestock and companion animals) exposed to the agent.
For studies of mixtures, consideration is given to the
possibility that changes in the physicochemical properties
of the individual substances may occur during collection,
storage, extraction, concentration and delivery. Another
consideration is that chemical and toxicological interactions
of components in a mixture may alter dose-response relationships.
The relevance to human exposure of the test mixture administered
in the animal experiment is also assessed. This may involve
consideration of the following aspects of the mixture
tested: (i) physical and chemical characteristics, (ii)
identified constituents that may indicate the presence
of a class of substances and (iii) the results of genetic
toxicity and related tests.
The relevance of results obtained with an agent that
is analogous (e.g. similar in structure or of a similar
virus genus) to that being evaluated is also considered.
Such results may provide biological and mechanistic information
that is relevant to the understanding of the process of
carcinogenesis in humans and may strengthen the biological
plausibility that the agent being evaluated is carcinogenic
to humans (see Part
B, Section 2f).
(a) Qualitative aspects
An assessment of carcinogenicity involves several considerations
of qualitative importance, including (i) the experimental
conditions under which the test was performed, including
route, schedule and duration of exposure, species, strain
(including genetic background where applicable), sex,
age and duration of follow-up; (ii) the consistency of
the results, for example, across species and target organ(s);
(iii) the spectrum of neoplastic response, from preneoplastic
lesions and benign tumours to malignant neoplasms; and
(iv) the possible role of modifying factors.
Considerations of importance in the interpretation and
evaluation of a particular study include: (i) how clearly
the agent was defined and, in the case of mixtures, how
adequately the sample characterization was reported; (ii)
whether the dose was monitored adequately, particularly
in inhalation experiments; (iii) whether the doses, duration
of treatment and route of exposure were appropriate; (iv)
whether the survival of treated animals was similar to
that of controls; (v) whether there were adequate numbers
of animals per group; (vi) whether both male and female
animals were used; (vii) whether animals were allocated
randomly to groups; (viii) whether the duration of observation
was adequate; and (ix) whether the data were reported
and analysed adequately.
When benign tumours (a) occur together with and originate
from the same cell type as malignant tumours in an organ
or tissue in a particular study and (b) appear to represent
a stage in the progression to malignancy, they are usually
combined in the assessment of tumour incidence (Huff et
al., 1989). The occurrence of lesions presumed to
be preneoplastic may in certain instances aid in assessing
the biological plausibility of any neoplastic response
observed. If an agent induces only benign neoplasms that
appear to be end-points that do not readily undergo transition
to malignancy, the agent should nevertheless be suspected
of being carcinogenic and requires further investigation.
(b) Quantitative aspects
The probability that tumours will occur may depend on
the species, sex, strain, genetic background and age of
the animal, and on the dose, route, timing and duration
of the exposure. Evidence of an increased incidence of
neoplasms with increasing levels of exposure strengthens
the inference of a causal association between the exposure
and the development of neoplasms.
The form of the dose-response relationship can vary widely,
depending on the particular agent under study and the
target organ. Mechanisms such as induction of DNA damage
or inhibition of repair, altered cell division and cell
death rates and changes in intercellular communication
are important determinants of dose-response relationships
for some carcinogens. Since many chemicals require metabolic
activation before being converted to their reactive intermediates,
both metabolic and toxicokinetic aspects are important
in determining the dose-response pattern. Saturation of
steps such as absorption, activation, inactivation and
elimination may produce non-linearity in the dose-response
relationship (Hoel et al., 1983; Gart et al.,
1986), as could saturation of processes such as DNA repair.
The dose-response relationship can also be affected by
differences in survival among the treatment groups.
(c) Statistical analyses
Factors considered include the adequacy of the information
given for each treatment group: (i) number of animals
studied and number examined histologically, (ii) number
of animals with a given tumour type and (iii) length of
survival. The statistical methods used should be clearly
stated and should be the generally accepted techniques
refined for this purpose (Peto et al., 1980; Gart
et al., 1986; Portier & Bailer, 1989; Bieler
& Williams, 1993). The choice of the most appropriate
statistical method requires consideration of whether or
not there are differences in survival among the treatment
groups; for example, reduced survival because of non-tumour-related
mortality can preclude the occurrence of tumours later
in life. When detailed information on survival is not
available, comparisons of the proportions of tumour-bearing
animals among the effective number of animals (alive at
the time the first tumour was discovered) can be useful
when significant differences in survival occur before
tumours appear. The lethality of the tumour also requires
consideration: for rapidly fatal tumours, the time of
death provides an indication of the time of tumour onset
and can be assessed using life-table methods; non-fatal
or incidental tumours that do not affect survival can
be assessed using methods such as the Mantel-Haenzel test
for changes in tumour prevalence. Because tumour lethality
is often difficult to determine, methods such as the Poly-K
test that do not require such information can also be
used. When results are available on the number and size
of tumours seen in experimental animals (e.g. papillomas
on mouse skin, liver tumours observed through nuclear
magnetic resonance tomography), other more complicated
statistical procedures may be needed (Sherman et al.,
1994; Dunson et al., 2003).
Formal statistical methods have been developed to incorporate
historical control data into the analysis of data from
a given experiment. These methods assign an appropriate
weight to historical and concurrent controls on the basis
of the extent of between-study and within-study variability:
less weight is given to historical controls when they
show a high degree of variability, and greater weight
when they show little variability. It is generally not
appropriate to discount a tumour response that is significantly
increased compared with concurrent controls by arguing
that it falls within the range of historical controls,
particularly when historical controls show high between-study
variability and are, thus, of little relevance to the
current experiment. In analysing results for uncommon
tumours, however, the analysis may be improved by considering
historical control data, particularly when between-study
variability is low. Historical controls should be selected
to resemble the concurrent controls as closely as possible
with respect to species, gender and strain, as well as
other factors such as basal diet and general laboratory
environment, which may affect tumour-response rates in
control animals (Haseman et al., 1984; Fung et
al., 1996; Greim et al., 2003).
Although meta-analyses and combined analyses are conducted
less frequently for animal experiments than for epidemiological
studies due to differences in animal strains, they can
be useful aids in interpreting animal data when the experimental
protocols are sufficiently similar.
Posted 23 January 2006
