| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
© The Author 2007. Published by Oxford University Press.
CORRESPONDENCE |
Response: Re: Commonly Studied Single-Nucleotide Polymorphisms and Breast Cancer: Results From the Breast Cancer Association Consortium
On behalf of the Breast Cancer Association Consortium
Correspondence to: Paul Pharoah, MD, PhD, Strangeways Research Laboratory, Department of Oncology, Worts Causeway, University of Cambridge, Cambridge CB1 8RN, UK (e-mail: paul1{at}srl.cam.ac.uk).
We thank the correspondents for their comments. Ambrosone et al. suggest that our interpretation of our results represents a growing divergence in perspective between investigators with a statistical genetic viewpoint and those with a molecular epidemiology viewpoint. We do not subscribe to this view. The Breast Cancer Association Consortium (BCAC) represents a wide spectrum of researchers, from statistical geneticists to clinical oncologists, and includes investigators who regard themselves as molecular epidemiologists. Moreover, in several respects, our interpretation of the data is not fundamentally different from that of Ambrosone et al. Our main conclusion was that for 11 of the 16 single-nucleotide polymorphisms (SNPs) tested there was no main effect, and that for the other five there was weak evidence for a main effect. We completely agree that susceptibility is likely to be the result of a complex interplay between genetic variation and environmental/lifestyle factors, and we did not rule out the possibility that such interactions may occur, either for these specific SNPs or for any other variants. Indeed, we stated that "...it is possible that these polymorphisms alter risk in subgroups of the population that have been exposed to specific environmental and lifestyle factors, for example, if an association of ADH1C I350V with breast cancer risk were limited to women with high alcohol consumption." Given that the main effect for any given risk factor is a weighted average of all possible interactions of that risk factor with other factors, we went on to state that "...where no main effect has been detected, such subgroup effects must be small, the at-risk subgroup must represent a small proportion of the population under study, or there must be true crossover effects (i.e., genotype associations of different direction among subgroups), which are unlikely." Moreover, because the number of possible interacting factors is very large the opportunities for type I statistical errors are greater, and even more stringent levels of statistical significance are required for subgroup and interaction analyses. Consequently, the power to detect a modest interaction effect is extremely small, even with large sample sizes such as in our dataset. Although the likelihood of detecting interactions may be increased by careful selection of the exposure of interest and the genetic variant(s), this approach is most likely to be productive where main effects have already been established for both the exposure and variant. This was the topic of earlier commentaries by several BCAC collaborators (1,2).
The examples of interactions between environmental exposures and genotype cited by Ambrosone et al. are not directly relevant to the issue of performing underpowered subgroup/interaction analyses when there are no main effects. In the first example (3), the putative interaction between the genetic variant of catalase (CAT-262 c>t) and fruit and vegetable consumption on catalase activity occurred when there was a statistically significant main effect for both genotype and environmental factor. Similarly, the possible interaction between the glutathione peroxidase variant GPX1 P198L and alcohol consumption on catalytic activity of this enzyme occurred in the context of statistically significant main effects (4).
We agree that, in general, heterogeneity of effects might be seen between studies because of differences in the prevalence of important interacting exposures. However, such differences are unlikely to explain the heterogeneity we observed for several SNPs, given that there were not substantial differences between the findings of studies carried out in populations of Asian origin and those based on subjects of European descent.
Baker suggests that one conclusion that might be drawn from our data is that they "support other evidence of very weak, or no association between common genetic variants and cancer." This conclusion seems premature given that we have studied just 16 common variants out of several million across the genome, of which five showed some evidence, albeit weak, of association. Several genome-wide association studies in breast cancer are ongoing, and over the next year or so, it should become clearer whether or not this conclusion is justified. Baker additionally suggests that it is not worth detecting alleles with small effect, even if they do exist, because of their limited clinical significance. We agree that the clinical significance of individual markers is likely to be small. However, the combined effect of multiple alleles may well be substantial and of clinical importance (5). Furthermore, although practical application in prevention is an important consideration, it is not the only justification for such research. Important insights into cancer biology are likely to come from identifying novel genes, and these may have wider implications for prevention and cure beyond simply identifying individuals at high risk.
NOTES
This response has been seen and approved by all Breast Cancer Association Consortium contributors.
REFERENCES
(1) Pharoah PD, Dunning AM, Ponder BA, Easton DF. (2005) The reliable identification of disease-gene associations. Cancer Epidemiol Biomarkers Prev 14:1362.
(2) Byrnes G, Gurrin L, Dowty J, Hopper JL. (2005) Publication policy or publication bias? Cancer Epidemiol Biomarkers Prev 14:1363.
(3) Ahn J, Nowell S, McCann SE, Yu J, Carter L, Lang NP, et al. (2006) Associations between catalase phenotype and genotype: modification by epidemiologic factors. Cancer Epidemiol Biomarkers Prev 15:1217–22.
(4) Ravn-Haren G, Olsen A, Tjonneland A, Dragsted LO, Nexo BA, Wallin H, et al. (2006) Associations between GPX1 Pro198Leu polymorphism, erythrocyte GPX activity, alcohol consumption and breast cancer risk in a prospective cohort study. Carcinogenesis 27:820–5.
(5) Pharoah PDP, Antoniou A, Bobrow M, Zimmern RL, Ponder BAJ, Easton DF. (2002) Polygenic susceptibility to breast cancer: implications for prevention. Nat Genet 31:33–6.[CrossRef][Web of Science][Medline]
Related Correspondence
CiteULike 
Connotea 
Del.icio.us What's this?
J Natl Cancer Inst 2007 99: 487.
This article has been cited by other articles:
|
|
 |
|
  P. A. Thompson Counterpoint: Genetic Risk Feedback for Common Disease Time to Test the Waters Cancer Epidemiol. Biomarkers Prev., September 1, 2007; 16(9): 1727 - 1729. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||