Sir-We think that Gerard te Meerman and Elisabeth de Vries' points require some clarification. In particular, their question is whether necessary factors (smoking) can be compared numerically with sufficient factors (eg, genetic factors in combination with smoking). This question addresses the possibility of correctly apportioning the relative contribution of genes and the environment. However: smoking is not a necessary cause of lung cancer, nor is it the only known risk factor (others are asbestos, ionising radiation, BCME, arsenic, and other occupational exposures), and non-smokers clearly have a finite, although small, risk of lung cancer; the statement that single gene-environment interactions represent sufficient causal complexes is probably true, but it is a hypothesis that requires much additional research. The issue of interactions is relatively complex. The fact that one factor, such as smoking, explains 90% of the incidence of a cancer does not mean that the other factors cause the remaining 10%. The sum of single-factor attributable risks typically exceeds 100%, and the excess over 100% is due to interactions. This conclusion may seem counterintuitive, but an example can help. In the general population there are smokers, workers exposed to asbestos, and individuals with genetic susceptibility, with hypothetical attributable risks in the order of 90% (smoking), 5% (asbestos in some populations), and 12% (hypothetical estimate for GSTM1, based on a relative risk of 1·3 and a frequency of 50% of the null genotype). These are reasonable estimates, whose sum exceeds 100%; the excess will increase if we add other types of genetic susceptibility, other environmental risk factors, or both. However, subgroups in the population share more than one environmental or genetic risk factor; in fact, we expect that the simultaneous presence of both environmental and genetic factors is the basis for "sufficient" causal complexes. The attributable risk in excess of 100% is due to the fact that we count twice, in the population, factors that contribute together to a sufficient causal complex in the single individual. There is much evidence, indeed, to think that the main causes of cancer are interactions. Another major issue is whether by the means of the new high-throughput tools of molecular biology we can identify people with a combination of low penetrance polymorphisms, the sum of which substantially increases the risk of cancer. This, again, is probably true, but does not contradict the view that we express. Unless they are in linkage disequilibrium, different genes are independently inherited. Therefore, the probability of having two or more polymorphisms that predispose to cancer is given by the product of separate probabilities. So, even if the level of susceptibility associated with a certain haplotype (due to, say, six different genes) is very high, the probability of having that haplotype is low. In general, the inverse relation that applies between the frequency of a genetic allele in the population and its penetrance seems to apply also to haplotypes.
Unit of Cancer Epidemiology, Dipartimento di Scienze Biomediche, e Oncologia Umana and CPO-Piemonte, University of Torino, via Santena 7, 10126 Torino, Italy