Consistent with most previous studies and a recent meta-analysis was our failure to find any evidence for an association between the CYP17
genotype defined by the – 34 promoter region T→C nucleotide-substitution polymorphism and risk of breast cancer overall, or within premenopausal or postmenopausal women, whether the genotype be defined under a codominant or a recessive mode of inheritance. Specifically, we failed to confirm our own positive finding [23
] of a significantly increased risk associated with the CC (A2
) genotype in women reporting a positive family history.
When we examined associations between CYP17 genotypes and hormone-related risk factors in controls, the only significant finding was an increased risk for older age at menarche among premenopausal women with the C (A2) allele. This finding is not consistent with previous positive reports, or with the hypothesis that the C (A2) allele is associated with increased endogenous oestrogen levels and an earlier age at menarche. The point estimates for the effects on breast cancer risk of later age at menopause and HRT use were stronger among women carrying the TT (A1/A1) genotype, but the interaction terms were not statistically significant.
Other studies investigating the influence of CYP17
genotype on hormonal risk factors have found varying results. For example, the CYP17
variant was associated with earlier age at menarche in only two of eight reports [8
]. Positive association with early age at first birth was observed in two of three reports [5
], and only single studies have reported significant associations with decreased use of HRT [35
] and with decreased difficulty in becoming pregnant [5
Results from previous studies on the effect modification of reproductive risk factors by CYP17
genotypes have also been conflicting. The most consistent association reported was an effect modification of age at menarche. Four studies have presented evidence that the protective effect for a later age at menarche (at least 13 years) was mainly limited to women with the wild-type homozygous genotype [5
], but in two of these studies the effect was observed only in premenopausal women [5
], and seven other studies have failed to confirm these results [8
]. Reports of significant associations between risk and age at first birth within strata of CYP17
genotype were in opposing directions from two studies reporting such results [11
Our larger study found no association between breast cancer and age at menarche or age at first birth within any of the variant genotypes, either overall or within postmenopausal or premenopausal women. Although we have found at best marginally significant associations between breast cancer risk and both HRT use and age at menopausal status within some CYP17 genotype groups, these must be interpreted with caution. We considered both premenopausal and postmenopausal women, seven risk factors, and two genotype groups, so by chance alone we would expect to find a few significant results even if there were no real effects.
Perhaps more convincing evidence against a role of this CYP17
variant in modulating endogenous oestrogen levels and associated breast cancer risk factors is the results of a recent study of 1,975 postmenopausal women, which found no association between CYP17
genotypes defined by several polymorphisms and mean levels of sex hormones, in particular oestradiol, oestrone and sex-hormone-binding globulin [25
]. This suggests that there might be little if any functional effect of the common CYP17
polymorphisms, at least among postmenopausal women, although there is a possibility that there might exist other variants and haplotypes associated with hormone levels and risk of breast cancer.
Although we have conducted a relatively large study, there are some limitations. Because the functionality, if any, of the polymorphism we have studied is not well established, it was not possible to specify a priori hypotheses about the likely existence and direction of interactions with risk factors. However, the few positive associations with serum hormone levels reported in the literature would suggest that the C (A2) variant would be associated with increased endogenous hormone levels, and we have observed significant effects both contradicting and in support of such an association. In addition, in this and other similar studies, there are multiple tests being conducted; the quoted P values are only nominal and should be interpreted accordingly. Consequently, we cannot claim with confidence that any of our 'significant' findings represent true effects. As there seems to be no overall effect of CYP17 genotype on breast cancer risk, finding any true interactions with breast cancer risk factors (should they exist) will require massive individual studies, or pooling of studies. The marginally significant deviation from Hardy–Weinberg equilibrium is unlikely to be due to genotyping error because cases and controls were genotyped at the same time on the same PCR plates, and thus any genotyping bias (and deviation from Hardy–Weinberg equilibrium) would be expected to be seen equally in both cases and controls, but this was not so. Furthermore, our PCR success rate was more than 99.5% for both cases and controls, and results were fully concordant for a subset of 168 duplicate DNAs for which PCR was successful.
The lack of significant associations in our data could be a consequence of not having sufficient statistical power to detect real effects. In terms of detecting a real effect on breast cancer risk associated with the homozygote A2/A2 (CC) genotype (whose frequency in controls is 14%; see Table ), with the total sample sizes we studied we would have had 80% power at the 0.05 level of statistical significance to detect effects greater than the threshold of 1.5-fold. If analyses were restricted to postmenopausal women, this detection threshold would become about twofold, whereas if we were to consider only women with a family history the threshold would be about threefold to fourfold. Within the two genotype groups (for example A1/A1 compared with A1/A2 and A2/A2, which subdivides controls 39:61; see Table ), the detection thresholds for effects associated with the risk factors (most of which are divided about 40:60 into two groups; see Table ) would be a minimum of 1.8-fold, and much greater for the smaller subgroupings.
Our study sample was in general younger than that of the other studies reporting on possible effect modification by CYP17
genotype. Given that the younger the age at onset of breast cancer the stronger are the familial effects [41
], one might expect the effects of genetic factors to be more pronounced in earlier onset disease. We therefore interpret our essentially null results as further support for the increasing body of evidence suggesting that there are no true associations or effect modifications, or at most weak ones, associated with this specific genetic variant of CYP17