Interactions between genes involved in the antioxidant defence system and breast cancer risk

Abstract

The aim of the study is to examine the association between multilocus genotypes across 10 genes encoding proteins in the antioxidant defence system and breast cancer. The 10 genes are SOD1, SOD2, GPX1, GPX4, GSR, CAT, TXN, TXN2, TXNRD1 and TXNRD2. In all, 2271 cases and 2280 controls were used to examine gene-gene interactions between 52 single nucleotide polymorphisms (SNPs) that are hypothesised to tag all common variants in the 10 genes. The statistical analysis is based on three methods: unconditional logistic regression, multifactor dimensionality reduction and hierarchical cluster analysis. We examined all two- and three-way combinations with unconditional logistic regression and multifactor dimensionality reduction, and used a global approach with all SNPs in the hierarchical cluster analysis. Single-locus studies of an association of genetic variants in the antioxidant defence genes and breast cancer have been contradictory and inconclusive. It is the first time, to our knowledge, the association between multilocus genotypes across genes coding for antioxidant defence enzymes and breast cancer is investigated. We found no evidence of an association with breast cancer with our multilocus approach. The search for two-way interactions gave experiment-wise significance levels of P=0.24 (TXN [t2715c] and TXNRD2 [g23524a]) and P=0.58 (GSR [c39396t] and TXNRD2 [a442g]), for the unconditional logistic regression and multifactor dimensionality reduction, respectively. The experiment-wise significance levels for the three-way interactions were P=0.94 (GPX4 [t2572c], TXN [t2715c] and TXNRD2 [g23524a]) and P=0.29 (GSR [c39396t], TXN [t2715c] and TXNRD2 [a442g]) for the unconditional logistic regression and multifactor dimensionality reduction, respectively. In the hierarchical cluster analysis neither the average across four rounds with replacement of missing values at random (P=0.12) nor a fifth round with more balanced proportion of missing values between cases and controls (P=0.17) was significant.