I recently described five biologically plausible models of relations between a genotype and an environmental exposure in terms of their effects on disease risk [13
] (). Each of the five models leads to a different set of predictions about disease risk in individuals classified by presence or absence of the high-risk genotype and environmental exposure (). In the table, “>1” denotes a relative risk exceeding 1.0, and “
1” denotes a greater increase in risk.
Five models of the relation between a high-risk genotype and an environmental exposure, in terms of their effect on disease risk.
Expected Relative Risks Under Different Models
In Model A, the effect of the genotype is to produce, or increase expression of, a “risk factor” that can also be produced nongenetically. An example is the relation of the autosomal recessive disorder, phenylketonuria (PKU), to high blood phenylalanine and mental retardation. Individuals who are homozygous for the PKU gene have a deficiency in the enzyme required to convert phenylalanine to tyrosine. If left untreated, a buildup of blood phenylalanine occurs after birth (before birth, the mother’s enzymes are used), and the high blood phenylalanine levels cause mental retardation. The retardation can be prevented if blood phenylalanine levels are kept low by dietary restriction. Mental retardation can also result from exposure to high blood phenylalanine in persons who are not homozygous for PKU: offspring of PKU mothers, who have intrauterine exposure to high blood levels because of their mother’s enzyme deficiency.
In epidemiologic terms, the high blood phenylalanine level in Model A is an intervening
], and the effect of the exposure is the same in persons with and without the high risk genotype. This is explicitly not
interaction, as defined above. It is an important model, however, because discovery of the mechanisms by which susceptibility genes influence disease is a central goal of genetic epidemiology. The same biologic mechanisms may apply to both genetic and nongenetic causal pathways.
In Model B, the genotype exacerbates the effect of the risk factor, but there is no effect of the genotype in unexposed persons. One example is the relation of xeroderma pigmentosum, an autosomal recessive disorder, to ultraviolet (UV) radiation and skin cancer. Excessive exposure to UV radiation increases risk for skin cancer in the general population, but individuals with xeroderma pigmentosum are deficient in an enzyme required for repair of DNA damage induced by UV radiation, and hence have even higher risk. If sun exposure could be prevented completely in these persons, they would not have increased risk for skin cancer.
In Model C, the exposure exacerbates the effect of the genotype, but there is no effect of the exposure in persons with the low-risk genotype. Individuals with porphyria variegata, an autosomal dominant disorder, have skin problems of varying severity, including unusual sun sensitivity and a tendency to blister easily. If they are exposed to barbiturates, an innocuous exposure in the general population, they experience acute attacks that may involve paralysis or even death.
In Model D, both the exposure and the genotype are required to increase risk. Most individuals with glucose-6-phosphate dehydrogenase (G6PD) deficiency, an X-linked recessive disorder, are asymptomatic. However, some persons with this genotype develop severe hemolytic anemia if they eat fava beans. Dietary exposure to fava beans does not produce this reaction in individuals without G6PD deficiency.
In Model E, the exposure and the genotype each have some effect on disease risk, and when they occur together risk is higher or lower than when they occur alone. An example is the relation between α-1-antitrypsin deficiency, smoking, and chronic obstructive pulmonary disease (COPD). Risk of COPD is increased both in nonsmokers with α-1-antitrypsin deficiency and in smokers without α-1-antitrypsin deficiency. Risk is increased to a greater extent in smokers with α-1-antitrypsin deficiency.
Note that in Models B, C, and D, interaction is always present, regardless of whether risks are measured on an additive or multiplicative scale. However, Model E encompaśses situations both with and without interaction, and moreover, the choice of the scale of measurement will determine whether or not interaction is said to exist. In the α
-1-antitrypsin example, Khoury et al. [15
] reported relative risks of COPD of RR10
= 3.8 in smokers with the low-risk genotype (PiM), RR01
= 1.6 in nonsmokers with the high-risk genotype (PiMZ), and RR11
= 4.7 in smokers with the high-risk genotype, using nonsmokers with the lowrisk genotype as the reference group. These relative risks are consistent with an additive model (i.e., no interaction on an additive scale), since 3.8 + 1.6 − 1 = 4.4 ≈ 4.7.
In a limited sense, Models B, C, D, and E form an exhaustive set of possible models of gene–environment interaction. There are four possible combinations of genotype and exposure, in terms of their individual effects on disease risk: (a) an effect of the exposure but not the genotype, (b) an effect of the genotype but not the exposure, (c) an effect of neither the genotype nor the exposure, and (d) an effect of both the genotype and the exposure. If we add interaction to each of these four possibilities, the result is Models B through E, respectively.
Model E can encompass antagonism, or a joint effect lower than expected from a multiplicative or additive model, as shown by the RR11 = ? in . In Models B, C, and D, however, the interactions are synergistic. If antagonistic interactions are included, there are three additional models: the genotype suppresses the effect of the environmental exposure, but has no effect when acting alone (Model B’), the exposure suppresses the effect of the genotype, but has no effect when acting alone (Model C’), and risk is reduced in persons with both exposure and genotype, but neither has an effect when acting alone (Model D’). With increasing emphasis on research on chemopreventive agents, the importance of these models is likely to increase.
As an illustration of the application of these models to disorders with complex genetic influences, consider the joint effects of genetic susceptibility and heavy alcohol drinking on risk for epilepsy (recurrent unprovoked seizures). There is strong evidence for a genetic component in some forms of epilepsy [16
]. However, the familial distribution does not follow a simple Mendelian pattern, suggesting that if susceptibility genes are important in some families, their phenotypic expression depends on unidentified environmental factors. Ng et al. [17
] reported an association between heavy drinking and risk of a first unprovoked seizure (i.e., a seizure not associated with an acute structural or metabolic insult to the central nervous system): the odds ratio increased from 3-fold in persons who drank 51–100 g of ethanol per day to almost 20-fold in those who drank more than 200 g per day, compared with nondrinkers.
The relations between heavy drinking and a genetic susceptibility to epilepsy are unclear, and the models described above provide a framework for considering several alternatives. A mechanism consistent with Model A, for example, might involve a genotype that increased risk of alcoholism. In this case, the genotype would not affect risk of epilepsy directly, but would lead to increased levels of exposure to alcohol, and thus to an indirect effect on epilepsy risk. With Model B, the effect of the susceptibility genotype would be to increase the brain’s sensitivity to the effects of alcohol; the genotype would then have no effect on risk for epilepsy in nondrinkers. With Model C, the susceptibility genotype would raise risk for epilepsy regardless of drinking behavior. Heavy drinking would raise risk further in those with the genotype, but would have no effect in those without the genotype. With Model D, the effect of the susceptibility genotype would be restricted to heavy drinkers, and at the same time, drinking would have no effect on epilepsy risk in those without the genotype. With Model E, the genotype and heavy drinking would each affect risk in the absence of the other, and the combined effects might fit a multiplicative model, an additive model, or neither.