Our study is the largest case–control study reported on
ERBB2 genetic variation. To our knowledge, this is also the first study on
ERBB2 reporting results for more than two polymorphisms and looking for involvement of haplotypes in breast cancer predisposition. We performed a study of five common SNPs and found no evidence for association with breast cancer risk. Four of the polymorphisms may be functional: SNP 1 near the promoter region and SNP 2 in intron 1 could be involved in regulatory processes whereas SNP 4 and SNP 5 are nonsynonymous coding SNPs that could affect tyrosine kinase activity or protein structure [
7]. Two association studies have previously reported a positive association between SNP 4 (I655V) and breast cancer risk [
8,
14]. Both genotyped about 700 individuals and showed a similarly increased risk for carriers of the Val allele (OR = 1.4). We were not able to replicate these findings. We have over 90% power to detect a risk of this magnitude at the 10
-4 level of significance. This suggests that previous positive findings may have been due to type I statistical errors. Neither could we replicate findings associating I655V with low-stage breast cancer or with breast cancer in younger women [
12,
13]. Positive results from stratified analyses should be treated with caution; very large sample sizes are required to obtain reliable results, the number of possible analyses that can be undertaken is large, and there is a strong possibility that one or more tests will be statistically significant simply by chance [
24]. We could not carry out analyses within cases stratified by family history, because we only had incomplete family history data [
15]. To investigate the possibility that a common polymorphism not included in our study might be involved in breast cancer predisposition, we constructed multilocus haplotypes and observed similar frequencies in cases and controls. We found six common haplotypes. Recently, the NIEHS Environmental Genome Project at the University of Washington released resequencing data based on 90 individuals (the PDR90 population; individual genotypes are available on line:
http://egp.gs.washington.edu/genes.html) and identified nine common SNPs (frequency ≥ 5%) in
ERBB2. All the common haplotypes (frequency ≥ 5%) were tagged by our set of five SNPs, even though, as expected given the multiethnicity of PDR90, differences in frequencies were seen between the two populations (data not shown). Crawford and colleagues resequenced 100 candidate genes involved in inflammation, lipid metabolism, and blood pressure regulation and showed that in a population of European descent the average number of common haplotypes per gene was 4.5, with a maximum number of 8 observed in only two genes [
25]. We are therefore confident that we have detected all common
ERBB2 haplotypes present in our population. We limited our study to common polymorphisms. A larger study set would be needed to identify a rarer polymorphism involved in disease predisposition. For example, dominant alleles with a frequency of 2% would require more than 4000 cases and 4000 controls to detect a relative risk of 1.5 significant at the 1% level with 90% power. We cannot exclude the possibility that a common SNP might have a differential effect in another ethnic group via gene–gene or gene–environment interactions, or that a predisposing SNP might be present exclusively in another population [
26]. In summary, we conducted a large case–control study of
ERBB2 and breast cancer. We genotyped five common SNPs, including the much-studied I655V polymorphism, and saw no association with the disease. Our set of SNPs generated all common haplotypes, and no differences in haplotype frequencies were seen between cases and controls.