Reply to Comments by Kraft and Wilson and by Weinberg and Mitchell on “Parental Genotypes in the Risk of a Complex Disease”

To the Editor:

Kraft and Wilson (2002 [in this issue]) point out that there are other analytical options to a joint application of case-control and TDT analysis in our study of the effect of parental genetics in the risk of a complex disease. They propose a “pseudo-sibling controls” design as an alternative to the approach proposed earlier by Weinberg and colleagues (1998) to study parental effects in case-parent trios. However, these tests are directed to evaluate the effect within a presumed model and are not designed to estimate joint effects of both parents’ genotypes, which appeared to be the case with our data. Our study, inspired by original experimental observations, led us to understand the underlying genetic effects that did not follow established paradigms. We concluded that a number of complementary strategies will need to be used simultaneously to dissect genetic predisposition to complex disorders (Labuda et al. 2002). In this regard, we are in agreement with Kraft and Wilson (2002 [in this issue]) that additional collecting of control-parent trios would extend possibilities of testing the observed effects under a greater variety of genetic and statistical models.

In the context of simple Mendelian disorders, fig. 1 could be troubling, but our paper was intended to divert the reader from this paradigm. Indeed, in a highly penetrant autosomal recessive condition, such as cystic fibrosis, in which two defective gene copies mean disease, collecting patients obviously identifies heterozygous parents, who otherwise would be difficult to find in such numbers in a control population. The example of Huntington chorea used by Weinberg and Mitchell (2002 [in this issue]) is rather unfortunate, since, at time of diagnosis, the case carrier-parent would already succumb to this disease. In a complex, multifactorial, and multilocus disease, in which the effect of a given allele is likely due to gene-gene (i.e., the presence of another variant at a different locus) and/or gene-environment interaction (e.g., exposure in only a fraction of carriers), one does not necessarily expect the enrichment in the at-risk alleles among patients’ parents, expected in turn to transmit these alleles “preferentially” to their case-offspring (fig. 1A). In other words, we believe that figure 1 provides a good illustration of the experimental situation we faced.

We apologize for not giving satisfactory credit to earlier developments, which was pointed out by Weinberg and Mitchell (2002 [in this issue]). The fact that reference to Lande and Price (1989) is also absent among articles cited by Weinberg et al. (1998) is not an excuse. Rather, it reflects the fact that these excellent methodological contributions were reported in the absence of experimental data, in contrast, for example, to a recent paper by Infante-Rivard et al. (2002b) where both the sampling scenarios include control-parent trios as well as testing for maternally mediated effects.

Obviously, the mechanisms through which the maternal and paternal genotypes could influence child phenotype might be very different, but their net effect relevant to cancer risk, such as an increased mutation burden, need not be. In respect to this, different pathways controlling the metabolism of carcinogens, the level of oxidative stress, or the efficiency of DNA repair may have their unique contributions to the increase in the level of DNA lesions and, consequently, cancer risk. The observed effect with CYP2E1*5 is, therefore, not at all unlikely. It is, however, possible that, for the effect to occur, the CYP2E1*5 carriers would have to undergo a particular environmental exposure (Infante-Rivard et al. 2002a). However, as with all such results, this is the first report that will have to be confirmed by other studies that include different populations.

Here, our population of case and control subjects, both of French-Canadian origin, seems to be excellent for association studies because of the common genetics and lifestyle. Moreover, as presented in our report, we independently tested this population for a possibility of stratification that, in light of the recent results of Ardlie et al. (2002), appears to be less of a problem in appropriately designed studies. Kraft and Wilson (2002 [in this issue]) evaluated an underestimation of 11% in the odds ratio, related to the use of “surrogate” parental controls. This 11% arises from the elevated disease probability in the chosen numerical example and actually corresponds only to 1 SD in the variant frequency (0.250±0.023) estimated in a sample of 350 chromosomes. Such extent of variation is expected under experimental conditions.

Weinberg and Mitchell (2002 [in this issue]) in their comments were also concerned by the effect of the CYP2E1*5/*5 homozygotes. Because of the rarity (see table 1 in Labuda et al. 2002) of the variant in question, we did not need to consider the effect of its homozygotes. The dominant effect is, therefore, the one to be assumed to be consistent with presumed phenotypic outcome of this allele, leading to higher inducibility and, therefore, to higher activity of the enzyme (see references in Labuda et al. 2002).

For the reasons discussed above, we believe that our study provides solid evidence for the parental effects. It provides also an experimental illustration of genetic effects that, although escaping a simple Mendelian paradigm, were anticipated in earlier studies such as that of Weinberg and her colleagues (Weinberg et al. 1998). There is, therefore, no reason to believe that these effects should not be expected in other complex diseases.