EPA - 600/D-84-186
PB-84-223478
RISK ASSESSMENT OF COMPLEX MIXTURES
Herman J. Gibb and Chao W. Chen1
^Carcinogen Assessment Group, Office of Health and Environmental
Assessment, U. S. Environmental Protection Agency, 401 M Street,
S.W., Washington, D. C. 20460. The views expressed in this
article are those of the authors and not necessarily those of the
U. S. Environmental Protection Agency.
INTRODUCTION
Risk assessment of suspected carcinogens involves both a quali-
tative and quantitative evaluation. The qualitative evaluation
evaluates the relevant animal, epidemiologic, mutagenic, and cell
transformation studies as to the likelihood that the agent is a
human carcinogen.
The quantitative evaluation is an estimate of the carcinogenic
potency of the suspected carcinogen. Potency is derived by fitting
a mathematical model to the dose-response data from either animal or
epidemiologic studi.es in an attempt to describe what an estimate of
the risk would be at low doses. Short-term genetic bioassay data,
which is the focus of this symposium, are not used for carcinogenic
risk estimation, although such data have been used in some instances
to provide an idea of the comparative carcinogenic potency of dif-
ferent compounds. Several types of risk estimation models have been
used for animal dose-response data. Most of the models that have
-------
been used for the epidemiclogic data are derivatives of the multi-
stage model.
The use of animal data for human risk assessment has several
limitations:
1. Animals may respond differently than humans as a result
of metabolic or other species-specific differences.
2. The animals are tested at high doses, usually doses that
humans would not encounter.
3. The lifetime testing to which animals are subjected is not
the equivalent of a human lifetime.
Many limitations are also encountered in the use of epidemiologic
data for quantitative evaluation, however. These include:
1. Lack of exposure data for the time period of concern.
2. Small sample sizes and short follow-up periods in the case
of cohort studies.
3. Confounding exposures to other carcinogens.
In mo's't cases, however, epidemiologic or animal data on complex
mixtures simply does not exist. In lieu of such data, a comparative
potency approach in short-term bioassays, as indicated earlier, has
been proposed. By this approach, a unit cancer risk estimate, or in
other words, the risk at unit dose (e.g., 1 jj g/L), is calculated
-------
for a mixture based on its potency in a short-term bioassay relative
to that of a mixture for which a unit risk has been calculated.
Albert et al. (1983) found that the relative potencies by skin
tumor initiation in SENCAR mice of coke oven emission extracts,
roofing tar emission extract, and cigarette smoke condensate appeared
to correlate well with the relative carcinogenic potencies based on
epidemiologic data. The limitation of this bioassay is that potency
as determined by a skin tumor initiation bioassay may not correlate
well with the potency of the mixture in a cancer bioassay (e.g.,
the mixture may have both initiation and promotion potential such
that the complete carcinogenic potency of the mixture is quite
different from its tumor initiation potential). The mathematical
implications of the multistage theory with regard to the carcinogenic
action of a complex mixture are explored here. Actual data from
both epidemiologic studies and animal investigations are used for
illustration. Regulatory ramifications of this discussion are
also addressed.
RISK ASSESSMENT OF COMPLEX MIXTURES
The main problem associated with doing risk assessments of
complex mixtures is that the chemical profiles, and thus the car-
cinogenic interaction of the mixtures, may vary from source to
source. To eventually be able to assess the risk of a population
-------
exposed to a complex mixture with a reasonable amount of confidence,
it is necessary that we better understand some of the carcinogenic
mechanisms involved. Several studies have examined the synergistic
and antagonistic effects of chemicals in a mixture with regard to
carcinogenicity. Some examples of these studies, both human and
animal, are reported in Tables 1 and 2.
In an attempt to explain these phenomena, we will apply the
theory of multistage carcinogenesis to interpret and evaluate the
data obtained from the animal and human studies. The multistage
theory of carcinogenesis. though oversimplified in our example, does
offer considerable plausibility for interpreting the dose-response
data obtained from animal experiments and epidemiologic studies.
However, it is important to understand the underlying assumptions
and limitations. For instance, the multistage model assumes that
the transition rate from one stage to the next stage is independent
of age. While this assumption has been shown by Peto et al.
(1975) to be true for B[a]P-induced skin cancer in animals, it may
not be true for other carcinogen-induced cancers. For instance,
with regard to human breast cancer data, the transition rates may
vary with hormone levels which are closely related to the age of
the individuals. Another possibility that is not included in the
simple multistage theory, although the model can be extended to
accommodate, is that the promotion and/or inhibition activity of
-------
C"
environmental factors other than the factor under study may have
an effect on the proliferation rates of partially or completely
transformed cells.
The simple multistage model assumes that a cell is capable of
generating a malignant neoplasm when it has undergone k changes in
a certain order. The rate, r^, of the i change is assumed to be
linearly related to D(t), the dose at age t, i.e., r^ = a^ + b^D(t),
where a^ is the background rate and b^ is the proportionality con-
stant for the dose (Figure 1). It can be shown (Crump and Howe,
1984) that the probability of cancer by age t is given by
P(t) = 1 - exp [-H(t)]
where
H(t) = /*/^k.../u2 { [a,
is the cumulative incidence rate by time t.
When H(t) or the risk of cancer is small, P(t) is approximately
equal to H(t). When only one stage is dose-related, all proportion-
ality constants are zero except for the proportionality constant
for the dose-related stage. The implications of the model when
one stage is carcinogen-affected has been summarized by Brown and
-------
Chu (1983) as follows:
For exposure at a near constant level to a carcinogen,
the multistage theory predicts the following patterns
of excess risk: (1) For any affected stage, excess risk
will increase with increasing level and/or duration of
exposure; (2) if only the first stage is affected, for
fixed exposure duration, excess risk is independent of
age at start of exposure and is an increasing function
of time since exposure stopped; and (3) if only the
penultimate stage is affected, for fixed exposure
duration, excess risk is an increasing function of age
at start of exposure and is independent of time since
exposure stopped.
Since a complex mixture often contains more than one carcinogen, the
likelihood is increased that the mixture will act on more than one
stage of the carcinogenic process. Without loss of generality in
our discussion, assume that two stages, the mtn and n*-" (1 _<_ m <
n < k), are dose-related. Then,
H(t) = H0 + H! + H2 + H12
where
HQ = (a1a2...ak)tk/k!
H! = (a1a2...ak)(bm/am)/ J/ uk.../ U2 D(um)du1. . .duk
H2 is similar to HI except that subscript m is replaced by n
H12 = (a1a2...ak)(bmbn/aman)/ J/ uk.../ ^2
D(um)D(un)du1. . .
That is, the cumulative incidence function H(t) can be decomposed
into four components: HQ is the background cumulative incidence,
-------
7
HI and H2 are cumulative incidences when only one stage is dose-
related, and Hi2 is the multiplicative term related to the multiple
of the two dose-related rates of change. When the two stages are
affected separately by two different carcinogens (e.g., B[a]P and
a non-B[a]P carcinogen in the mixture) then the multiplicative term
reflects the synergism due to the two carcinogens. Obviously, the
multiplicative effect would not exist if one of the two compounds
were removed. When the same stage is affected by different agents,
the synergistic effect does not occur under the simple multistage
theory, but antagonism may occur due to the competition of
carcinogens for the partially transformed cells of a particular
stage. These conclusions may not hold if the transition rate from
one stage to the next is modified due to external influences (e.g.,
breast cancer associated with hormonal change).
The above theoretical discussion suggests that exposure to a
complex mixture may produce synergistic and/or antagonistic effects.
Thus, the multistage theory can be used to interpret the synergistic
and antagonistic observations in humans and animals described
earlier.
An illustrative example would be that of Doll and Hill's dose-
response data (1956, 1964) for lung cancer and cigarette smoking in
British doctors. These data were analyzed by Doll (1971a) and were
-------
presented before the Royal Statistical Society in December 1970.
Doll found that the age-specific lung cancer mortality rate for
smokers is approximately proportional to the 5*-** power of duration
since the start of exposure and is linearly related to the amount
smoked. This implies that smoking affects an early stage of lung
carcinogenesis. Following Doll's presentation, Armitage raised
the issue of whether smoking affects the early or late stage of
carcinogenesis. Armitage stated:
In this connection, I have always been somewhat puzzled
about the effect of cigarette smoke as a carcinogen. The
dose-response relationship seems to be linear, which sug-
gests that the carcinogen affects the rate of occurrence
of critical events at one stage, and one only, in the
induction period. ... On the other hand, the halt
in risk quite soon after smoking stops suggests that a
late stage is involved. . . .
The Armitage view can best be seen from Figure 2, which was reported
by Doll (1971b) after his presentation of the earlier paper. The
fact that the lung cancer rate for the ex-smokers decreased and
then increased again approximately 15 years after smoking stopped
suggests that an early stage and a late stage are affected by the
cigarette smoke. In a subsequent analysis, Doll and Peto (1978)
suggested that a quadratic dose-response relationship seems to be
preferred to the linear dose-response relationship, suggesting
that more than one stage is dose-affected as was observed by
Armitage 14 years ago.
-------
Pershagen (1982) recently found that cigarette smoking and
exposure to arsenic had a synergistic effect with regard to carcino-
genesis. Brown and Chu (1983) concluded from studies of smelter
workers that arsenic is a late-stage carcinogen in the multistage
model. Brown and Chu found that excess lung cancer risk among
smelter workers was an increasing function of age at the start of
exposure, and for individuals greater than or equal to 55 years of
age, the risk was independent of the time since exposure stopped.
This follows the pattern for a late-stage carcinogen as discussed
earlier. It is theorized that older individuals are at a greater
risk of lung cancer mortality from exposure to a late-stage
carcinogen since they have had time to accumulate more cells in
the earlier stages of the cancer process, such cells being partic-
ularly susceptible to a late-stage carcinogen such as arsenic.
Since a late-stage carcinogen cannot increase the number of cells
in the early stages of carcinogenesis, the individual's risk remains
constant after cessation of exposure. Following the simple multi-
stage model, we would then explain the synergistic effect of
cigarette smoke and arsenic observed by Pershagen as an interaction
between the effect on a late-stage of carcinogenesis by arsenic and
the effect on an early stage of carcinogenesis by components of
cigarette smoke.
-------
The simple multistage model that we have discussed would
conclude:
1. That carcinogenic synergism in mixtures is a result of con-
stituents of the mixture acting on separate stages of the
multistage process of carcinogenesis.
2. If all constituents of the mixture act on a single stage
of the multistage process of carcinogenesis, there will
be no synergism. However, an antagonistic effect could
result due to the availability of partially transformed
cells.
CONCLUSION
Obviously, the simple model cannot explain all of the syner-
gistic or antagonistic carcinogenic effects observed in animal or
human studies. As stated earlier, the model does not consider
changes in transition rates between stages that may be brought on
by age or environmental factors. It is a mathematical model, how-
ever, that could certainly explain some of the data. Thus, we feel
that implications from the multistage model should be considered
in the design of future animal studies or even short-term bioassay
studies. In regard to animal bioassays, we would suggest that
mixtures be fractionated and administered to the animals, varying
the age at which the dose is given and, perhaps, the duration of
-------
the dose. In addition, epidemiologic data should be reported when-
ever possible to facilitate analysis with regard to the affected
carcinogenic stage or stages on which the complex mixture may be
acting. The data reported by Doll (1971a) and Brown and Chu (1983)
have provided insight with regard to the effects of carcinogens on
different stages.
Perhaps one final point should be offered with regard to the
understanding of mixtures and components of mixtures by their car-
cinogenic stage of action. The effects of a late-stage carcinogen
would be seen in a relatively short period of time, whereas the
effects of an early-stage carcinogen may take many years to be
detected. These effects may affect the way we regulate complex
mixtures and certainly, we hope, should affect the way in which we
study such mixtures.
REFERENCES
Albert, R.E., J. Lewtas, S. Nesnow, T.W. Thorslund, and E.L. Ander-
son. 1983. Comparative potency method for cancer risk assess-
ment: Application to diesel particulate emissions. J. Risk
Analysis 3(2):101-117.
-------
Bock, F.G., S.K. Crouch, and G.E. Moore. 1964. Tumor-promoting
activity of extracts of unburned tobacco. Science 145:831-833.
Brown, C.C., and K.C. Chu. 1983. Implications of the multistage
theory of carcinogenesis applied to occupational arsenic
exposure. J. Natl. Cancer Inst. 70(3):455-463.
Crump, K., and R. Howe. 1984. The multistage model with a time-
dependent dose pattern: Applications to carcinogenic risk
assessment. J. Risk Analysis (to be published in 1984).
Doll, R., and A. Hill. 1956. Lung cancer and other causes of death
in relation to smoking. Br. Med. J. 2:1071-1076.
Doll, R., and A. Hill. 1964. Mortality in relation to smoking:
Ten year's observation of British doctors. Br. Med. J. 1:1399-
1410, 1460-1467.
Doll, R. 1971a. The age distribution of cancer: Implications for
models of carcinogenesis (with discussion). J. Royal Stat. Soc.
Series A., 134:133-166.
-------
/3
Doll, R. 1971b. Cancer and aging: The epidemiologic evidence.
In: Oncology. Tenth International Cancer Congress. Vol. V.
R.L. Clark, R.W. Cumley, T.E. McCoy, et al., eds. Chicago,
IL: Chicago Year Book Medical Publishers, pp. 1-28.
Doll, R., and R. Peto. 1978. Cigarette smoking and bronchial car-
cinoma: Dose and time relationship among regular smokers and
lifelong non-smokers. J. Epidemiol. Community Health 32:303-313.
Falk, H.L., P. Kotin, and S. Thompson. 1964. Inhibition of carcino-
genesis. Arch. Environ. Health 9:169-179.
Grimmer, G. 1977. Anaylsis of automobile exhaust condensate. In:
U. Mohr, D. Scha'hl, and L. Tomatis, eds. Air pollution
and cancer in man. Proceedings of the Second International
Carcinogenesis Meeting, Hanover, Germany, October 22-24, 1975.
International Agency for Research on Cancer, IARC Scientific
Publication No. 16. Lyon, France: pp. 29-39.
Hammond, E.G., I.J. Sellikoff, and H. Seidman. 1979. Asbestos
exposure, cigarette smoking and death rates. Ann. NY Acad.
Sci. 330:473-490.
-------
Lundin, F.E. , Jr., J.W. Lloyd, E.M. Smith, et al. 1969. Mortality
of uranium miners in relation to radiation exposure, hard rock
mining and cigarette smoking—1950 through Sept. 1967. Health
Phys. 16:571-578.
Pershagen, G. 1982. Arsenic and lung cancer with special reference
to interacting factors—epidemiological and experimental
evidence. Submitted as a doctoral thesis at the Karolinska
Institute, Stockholm, Sweden.
Peto, R., F. Rose, P. Lee, L. Levy, and J. Clack. 1975. Cancer and
aging in mice and men. Br. J. Cancer 32:411-426.
Sellikoff, I.J., E.G. Hammond, and J. Churg. 1968. Asbestos
exposure, smoking, and neoplasia. J. Am. Med. Assoc. 204:
106-112.
Van Duuren, B.L., and B.M. Goldschmidt. 1976. Cocarcinogenic and
tumor-promoting agents in tobacco carcinogenesis. J. Natl.
Cancer Inst. 56(6):1237-1242.
Wanebo, C.K., K.G. Johnson, K. Sato, et al. 1968. Lung cancer
following atomic radiation. Am. Rev. Respir. Dis. 98:
778-787.
-------
o
age
ri
-------
Figure 1. Schematic view of the transition rate of the ith change
in the simple multistage model.
Assumptions: ri = ai + bjDCu), transition rate from (i-l)t stage
to i^ stage, where a, is the background rate and b^ is the
proportionality constant for the dose.
-------
Continuing smokers
5 10 15
-------
Figure 2. The rate of lung cancer among people who have stopped
smoking cigarettes, those who continue to smoke, and
)
those who have never smoked.
-------
Table 1. Examples of Synergism with regard to Carcinogen Response
in Human Studies
Agents
Involved
Smoking and asbestos
Uranium mining and
cigarette smoking
Radiation and smoking
Arsenic and smoking
Type of
Study
Cohort
Cohort
Cohort
Cohort
Tumor
Site
Lung
Lung
Lung
Lung
Authors
Selikoff et al. (1968)
Hammond et al. (1979)
Lundin et al. (1969)
Wanebo et al. (1968)
Pershagen (1982)
-------
Table 2. Examples of Synergism and Antagonism with regard
to Carcinogen Response in Animal Studies
Agents Involved
Type of Study
Authors
Synergism
7,12-DMBA and extracts
of unburned cigarette
tobacco
7,12-DMBA and each of
the following:
catechol, pyrogallol,
decane, indecane,
pyrene, benzo[e]pyrene,
and fluoranthene
Automobile exhaust
condensate without
particulate matter
and Benzo[a]pyrene
(B[a]P)
Antagonism
Automobile exhaust
condensate with
particulate matter
and B[a]P
B[a]P and 10 different
non-carcinogens
•Mouse skin painting Bock et al. (1964)
Mouse skin painting
Mouse subcutaneous
injection
Mouse subcutaneous
injection
B[a]P and esculin,
quercetin and squalene,
and oleic acid
(tobacco smoke components)
Mouse subcutaneous
injection
Mouse skin painting
Van Duuren and
Goldschmidt (1976)
Grimmer (1977)
Grimmer (1977)
Falk et al. (1964)
Van Duuren and
Goldschmidt (1976)
------- |