769 A Systematic Approach to Analysing -Gene Interactions: Polymorphisms at the Microsomal Epoxide Hydrolase EPHX and S-transferase GSTM1, GSTT1, and GSTP1 Loci and Breast Cancer Risk

Amanda B. Spurdle,1 Jiun-Horng Chang,2 Graham B. Byrnes,2 Xiaoqing Chen,1 Gillian S. Dite,2 Margaret R.E. McCredie,4 Graham G. Giles,3 Melissa C. Southey,2 Georgia Chenevix-Trench,1 and John L. Hopper2 1Cancer and Cell Biology Division, The Queensland Institute of Medical Research and The University of Queensland, Brisbane, Queensland, Australia; 2Centre for Molecular, Environmental, Genetic and Analytic Epidemiology, The University of Melbourne and 3The Cancer Council Victoria, Carlton, Victoria, Australia; and 4University of Otago, Dunedin, New Zealand

Abstract

Objective: We undertook a case-control study in an Australian all interactions gave a significantly better fit than a main- Caucasian population-based sample of 1,246 cases and 664 effects-only model (P < 0.001), providing evidence for gene- controls to assess the roles of detoxification gene poly- gene interactions. The most parsimonious model included morphisms EPHX T>C Tyr113His, GSTT1 deletion, GSTM1 main effects for EPHX, GSTT1, and GSTM1; a two-way deletion, and GSTP1 A>G Ile105Val on risk of breast cancer. interaction between EPHX and GSTM1; and a three-way Methods: We systematically addressed the main effects and interaction between EPHX, GSTM1, and GSTT1. Predicted possible gene-gene interactions using unconditional logistic risks were greatest for women carrying deletions of both regression to estimate odds ratios (OR) adjusted for potential GSTT1 and GSTM1, with either the EPHX TC genotype (OR, confounders and using standard model building approaches 2.02; 95% CI, 1.19-3.45; P = 0.009) or EPHX CC genotype (OR, based on likelihood theory. 3.54; 95% CI, 1.29-9.72; P = 0.14). Results: There was a decreased risk associated with the EPHX Conclusion: Detoxification gene polymorphisms may interact CC genotype [OR, 0.60; 95% confidence interval (95% CI), with each other to result in small groups of individuals at 0.43-0.84; P = 0.003], marginally significant evidence of modestly increased risk. We caution against overinterpreta- increased risk with GSTM1 null genotype (OR, 1.21; 95% tion and suggest that pooling of similarly large studies is CI, 1.00-1.47; P = 0.05), but no association with GSTT1 null needed to clarify the possible role of such complex gene- genotype (OR, 1.12; 95% CI, 0.86-1.45; P = 0.4) or GSTP1 (OR, gene interactions on breast cancer risk. (Cancer Epidemiol 0.95; 95% CI, 0.82-1.10; P = 0.5) genotype. The full model with Biomarkers Prev 2007;16(4):769–74)

Introduction

Genetic risk factors for breast cancer include dominantly aggregation (2). It has been argued that these effects may be inherited deleterious mutations in the BRCA1 and BRCA2 evident only when considering the joint effect of several . However, carriers of such mutations are rare at the genetic and environmental factors that act within a biological population level and account for little of the familial pathway (3), although searching for these interactions requires aggregation observed for breast cancer (1). Thus, it is possible some care to reduce the frequency of making false-positive that more commonly occurring ‘‘low-risk’’ genetic factors conclusions (4). explain a substantial proportion of the residual familial Candidate breast cancer predisposition genes include those that are involved in the metabolism of carcinogens that contain allelic variants that affect or protein function. Received 9/14/06; revised 1/17/07; accepted 1/29/07. These include the phase I microsomal epoxide hydrolase Grant support: The Australian Breast Cancer Family Study was funded by the National Health gene EPHX and the phase II glutathione S-transferase genes and Medical Research Council, the Victorian Health Promotion Foundation, the New South Wales Cancer Council, the Peter MacCallum Cancer Institute, the Inkster-Ross Memorial Fund, GSTT1, GSTM1,andGSTP1. EPHX encodes the epoxide and by the NIH, as part of the Cancer Family Registry for Breast Cancer Study grant CA 69638. hydrolase protein mEH, which catalyses the hydrolysis of This work used data collected from the Breast Cancer Family Registry Studies, funded by the trans National Cancer Institute under RFA CA-95-003 and through cooperative agreements with arene and aliphatic epoxides to -dihydrodiols (5), and the Fox Chase Cancer Center, Huntsman Cancer Institute, Columbia University, Northern typically results in detoxification and preparation for phase II California Cancer Center, Cancer Care Ontario, and University of Melbourne. This work was GST funded by the National Health and Medical Research Council. A.B. Spurdle is funded by a conjugation reactions. The genes catalyze the glutathione- National Health and Medical Research Council Career Development Award, G.B. Byrnes is mediated reduction of exogenous and endogenous electro- funded by a National Health and Medical Research Council Capacity Building Grant in Population Health, and G. Chenevix-Trench and J.L. Hopper are National Health and Medical philes with broad and overlapping substrate specificity (6, 7), Research Council Principal and Senior Principal Research Fellows, respectively. generally producing readily excretable water-soluble com- The costs of publication of this article were defrayed in part by the payment of page charges. pounds. Thus, allelic variants associated with altered (faster This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. or slower) detoxification rates of potential carcinogens may Note: J.L. Hopper is a Victorian Breast Cancer Research Consortium Group Leader. The confer an increased susceptibility to cancer, perhaps more so in content of this article does not necessarily reflect the views or policies of the National Cancer the presence of environmental stresses such as smoking and Institute or any of the collaborating centers in the Cancer Family Registry, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. UV light exposure (8). Government or the Cancer Family Registry Centers. A number of candidate functional variants exist for the Requests for reprints: John L. Hopper, Centre for Molecular, Environmental, Genetic, and EPHX and GST genes. The EPHX exon 3 T to C Tyr113His Analytical Epidemiology, The University of Melbourne, Carlton, Victoria, Australia. Phone: 61-3-8344-0697; Fax: 61-3-9349-5815. E-mail: [email protected] amino acid substitution variant was shown to have functional 113 Copyright D 2007 American Association for Cancer Research. significance in vitro (9), with the His variant demonstrating 113 doi:10.1158/1055-9965.EPI-06-0776 60% activity relative to the wild type Tyr . Deletion variant or

Cancer Epidemiol Biomarkers Prev 2007;16(4). April 2007 Downloaded from cebp.aacrjournals.org on September 25, 2021. © 2007 American Association for Cancer Research. 770 GST and EPHX Polymorphisms and BreastCancer null alleles exist for the GSTT1 and GSTM1 genes (10-12) and Councils of Victoria and New South Wales, and the Queens- present biochemically as a failure to express protein. These land Institute of Medical Research. All subjects provided alleles are common, and GSTT1 and GSTM1 null genotypes written informed consent. occur in f10% to 20% and 50% of the Caucasian population, Individuals included in genetic analyses (86% of participat- respectively (13). The GSTP1 exon 5 A to G Ile105Val ing subjects) differed from the remainder of participating polymorphism is one of several variants reported for this subjects with respect to certain factors shown previously to be locus (14, 15), and the Val-containing isoform was shown associated with breast cancer (25), namely family history to have altered specific activity and decreased heat stability (defined as having at least one first- or second-degree relative (16-18). with breast cancer), oral contraceptive use, and parity. These and other polymorphisms in the above-mentioned Ethnicity was self-reported. Subjects with any Australian genes have been investigated as risk factors for numerous aboriginal, Torres Strait Islander, or Maori heritage; or country cancers, either independently or in combination. Results from of birth in the South Pacific, Indian Ocean, Caribbean islands, individual case-control studies assessing the effect of GST or Asia were classified as non-Caucasian. Preliminary analyses polymorphisms on risk of breast cancer have been conflicting. indicated that genotype distributions for GSTT1, GSTP1, and Two meta-analysis studies (see Discussion) found a significant EPHX differed between individuals reporting Caucasian and association of breast cancer with the GSTM1 deletion genotype non-Caucasian ethnicity, consistent with other studies (see for postmenopausal women (19, 20), with stronger evidence Discussion). Further analyses were thus restricted to partic- and higher odds ratios (OR) from studies in populations with a ipants who identified themselves as Caucasian. In addition, low frequency of the GSTM1 null genotype (20). One study molecular analyses to date (24)5 have identified 59 Caucasian reported a slight, marginally significant increase in risk cases carrying a deleterious germ line mutation in BRCA1 or associated with the GSTT1 deletion genotype (19). However, BRCA2, and these participants were also excluded from the a recent pooled analysis (see Discussion) by Vogl et al. (21) analyses. Final analyses included 1,246 cases and 664 controls. found no significant association with breast cancer risk overall Genotype Measurement. Genotyping was as described for GSTT1 or GSTM1, for either premenopausal or postmen- previously (26, 27). Briefly, PCR-agarose methodology was opausal women. Meta-analyses of the GSTP1 exon 5 A>G used to detect the GSTT1 and GSTM1 homozygous deletion Ile105Val polymorphism found no evidence that this variant genotypes, and the ABI Prism 7700 Taqman Sequence was associated with risk for either the heterozygous genotype Detection System methodology was used for genotyping the (21) or the homozygous genotype (19, 21). Furthermore, in the EPHX T to C Tyr113His polymorphism (rs1051740), and the pooled analysis (21), there was no evidence that the GSTT1, GSTP1 A to G Ile105Val polymorphism (rs1695). In addition, GSTM1,orGSTP1 genotypes acted in combination to increase because preliminary analysis indicated deviation from Hardy- breast cancer risk, or interacted with smoking or reproductive Weinberg equilibrium, all individuals with EPHX CC and a history to modify breast cancer risk. subset of individuals with TC genotype were regenotyped The only published study investigating the effects of EPHX using standard denaturing high-performance liquid chroma- on breast cancer (22) reported a not significant 1.5-fold tography and sequencing methodology. Primer sequences were increased risk associated with the EPHX exon 3 CC genotype as follows: forward, GCTTCCACTATGGCTTC, and reverse, compared with the wild-type TT genotype and a borderline TTGGGTTCTGAATCTCTCCAA, and PCR was done using a significant 2.2-fold increased risk for EPHX exon 3 CC touchdown program with final annealing temperature of 55jC, genotype in combination with GSTM1 null genotype. How- and the denaturing high-performance liquid chromatography ever, this hospital-based study of mostly postmenopausal melting temperature of 59jC. This methodology was used to women was relatively small—consisting of 238 cases and 313 confirm genotype at the Tyr113His position and also to establish controls. that no other variation, such as the reported Trp97Stop variant, In view of the uncertainty in the literature, we have might be compromising assay results. Sequence confirmation undertaken a case-control study in Australia to assess the role of all aberrant denaturing high-performance liquid chroma- of GSTT1, GSTM1, GSTP1, and EPHX genetic polymorphisms tography profiles detected during this rescreening revealed no as risk factors for breast cancer. We specifically undertook a evidence of additional variation under Taqman primer and systematic analysis of the possible interactions between these probe binding sites, and identified genotype misclassification genetic polymorphisms. in 0.5% of cases and 0.2% of controls, with overcalling of the CC genotype for the Taqman genotyping. Materials and Methods Statistical Analysis. Hardy-Weinberg equilibrium was tested by Pearson’s m2 test. Association of case-control status Subjects. The Australian Breast Cancer Family Study, a with genotype and environmental factors was modeled using population-based, case-control-family study of breast cancer in unconditional logistic regression. EPHX and GSTP1 were women less than 60 years old, was carried out in Melbourne initially modeled as codominant genotypes. Subsequently, and Sydney with an emphasis on early-onset disease (23). EPHX and GSTP1 were included as linear terms coded 0, 1, or Details of data collection methods have been described 2, with the homozygote wild-type (TT for EPHX, AA for previously (23, 24). Briefly, cases were women diagnosed with GSTP1) coded as 0. The likelihood ratio test was used to assess a first primary breast cancer, identified through the Victoria any improvement of fit between the codominant genotype and New South Wales cancer registries, and controls were model and linear model for EPHX and GSTP1. GSTM1 and women with no previous breast cancer selected from the GSTT1 were each coded so that the homozygote deletion was electoral rolls (adult registration for voting is compulsory in contrasted with the heterozygote and wild-type pooled, Australia) by a stratified random sampling, frequency- reflecting the sensitivity of the gel-based genotyping methods matched for age. Questionnaires were used to measure risk used. factors; family history of cancers was systematically collected In addition to the genotype, a number of environmental for each case and control from multiple sources in the family; factors considered to be potential confounders were also and blood samples were collected at the time of interview included: reference age categories (V30, 31-40, 41-50, >50 from subjects agreeing to participate in genetic studies. Genotyping was done on the basis of DNA availability. Approval for the study was obtained from the ethics committees of The University of Melbourne, The Cancer 5 Unpublished data.

Cancer Epidemiol Biomarkers Prev 2007;16(4). April 2007 Downloaded from cebp.aacrjournals.org on September 25, 2021. © 2007 American Association for Cancer Research. Cancer Epidemiology, Biomarkers & Prevention 771 years), menopausal status, age at menarche quartile, number the null model, the next level shows models with 1 degree of of live births (0, 1, 2, 3, z4), family history (any first- or second- freedom (df), the next with 2 df, and so on. Each level considers degree relative reported to have had breast cancer), previous all possible combinations of variables. Because including an benign breast disease, body mass index [weight (kg)/height effect for GSTP1 made no significant difference to any of the (m)2] quartile, smoking (never/ever), ever drinking alcohol fitted models (see Results below), we have omitted it from the regularly, oral contraceptive use (never/ever), hormone lattice to simplify the presentation (see Fig. 1). Differences in G replacement therapy ever, education (three categories), coun- statistics between nested models are compared with the m2 try of birth, and current Australian state of residence. distribution with df = difference in the df of the nested models. A systematic model building procedure was used to include Whenever evidence of a gene-gene interaction with cancer or exclude the genetic polymorphisms and their interactions, risk was found, we tested for nonindependent assortment using the likelihood ratio test to calculate the statistical separately in cases and controls using a log-linear model (29). significance of nested models (28). Both forward selection The same approach was used to test for associations between and backward elimination were used. In addition to the main genotype and smoking, and between genotype and regular effects of EPHX, GSTT1, GSTM1, and GSTP1, we considered drinking. Finally, the analyses were repeated with the all possible two-, three-, and four-way gene-gene interactions. environmental factors omitted. After estimating the final Each genetic model was considered without including any models, the correlation matrix of the estimates was checked environmental factors, including all those listed above, and for evidence of collinearity. A significance level of 5% was including only those that remained significantly associated chosen a priori and all tests were two-sided. Stata software was according to the Wald test. Due to the role of these candidate used for the analysis (version 9.0, StataCorp). genes in detoxification, we also considered the interaction of each allele with smoking and regular drinking. Results Results were presented as a lattice showing, for each model, the G statistic defined as twice the difference in log likelihoods The genotype distributions for GSTT1, GSTP1, and EPHX between that from the fitted model and that from the null differed between individuals reporting Caucasian and non- model (hence, G for the null model = 0). The first level shows Caucasian ethnicity (P < 0.001; P = 0.013; P = 0.002,

Figure 1. Model building lattice. A, EPHX; B, GSTM1; C, GSTT1; underlined, interac- tion terms. Likelihood ratio test: number of variables (df) in models increases with each row. Differences in G = 2 ln (likelihood of fitted model /likelihood of null model) between nested models (connected by lines) can be compared with the m2 distribution with df = difference of model df. Note that the GSTP1 genotype was not found to be associated with breast cancer, nor did addition of terms related to this genotype lead to more than minimal change to any of the other estimates. Therefore, for simplic- ity, GSTP1 is not shown in the lattice diagram.

Cancer Epidemiol Biomarkers Prev 2007;16(4). April 2007 Downloaded from cebp.aacrjournals.org on September 25, 2021. © 2007 American Association for Cancer Research. 772 GST and EPHX Polymorphisms and BreastCancer

Table 1. Crude and adjusted univariate ORs for association codominant model for EPHX and GSTP1 did not provide a between breast cancer and EPHX, GSTP1, GSTM1, and significantly better fit than the linear model (P = 0.1 and P = GSTT1 genotypes 0.9, respectively). There was marginally significant evidence of increased breast cancer risk with GSTM1 null genotype (OR, Genotype Controls Cases Adjusted* 1.21; 95% CI, 1.00-1.47; P = 0.05), but not with GSTT1 null (%) (%) P OR (95% CI) P genotype (OR, 1.12; 95% CI, 0.86-1.45; = 0.4). A marginally significant interaction was found between EPHX drinking (never/ever) and GSTP1 (interaction OR, 1.36; 95% TT 316 (47.7) 639 (51.6) 1.00 (Reference) CI, 1.01-1.83; P = 0.04) and a suggestive association between TC 262 (39.5) 496 (40.1) 0.94 (0.76-1.15) 0.5 CC smoking (never/ever) and GSTP1 (interaction OR, 1.30; 95% 85 (12.8) 103 (8.3) 0.60 (0.43-0.84) 0.003 P As linear term 0.83 (0.72-0.96) 0.01 CI, 0.97-1.73; = 0.09). Because a weak interaction could be T allele frequency 0.674 0.716 the result of a population-level association between behaviors C allele frequency 0.326 0.284 and genotype, these associations were tested for the controls P P Hardy-Weinberg = 0.01 =0.6 only and found to be not significant for smoking (OR, 0.88; equilibrium 95% CI, 0.65-1.21; P = 0.4) and drinking (OR, 0.83; 95% CI, 0.61- GSTP1 1.15; P = 0.3). Interactions between genotype and menopausal AA 283 (43.6) 539 (43.8) 1.00 (Reference) status and other reproductive factors such as parity and age at AG 286 (44.1) 545 (44.2) 0.95 (0.77-1.17) 0.6 menarche were also not significant (data not shown). GG 80 (12.3) 148 (12.0) 0.90 (0.66-1.25) 0.5 The results of both forward selection and backward As linear term 0.95 (0.82-1.10) 0.5 A elimination excluded all main effects and gene-gene inter- allele frequency 0.656 0.659 GSTP1 P G allele frequency 0.344 0.341 actions with ( from 0.3 to 0.95). Therefore, for Hardy-Weinberg P =0.6 P =0.6 simplicity, the raw data for cases and controls in Table 2, equilibrium and results from the model building procedure illustrated by GSTP1 GSTM1 Fig. 1, was shown excluding . The main-effects-only on model gave a significantly better Present 333 (50.2) 571 (45.8) 1.00 (Reference) m2 df P Null 331 (49.9) 675 (54.2) 1.21 (1.00-1.47) 0.05 fit than the null model ( =9.94on3 ; < 0.02), even after adding the nonsignificant effect for GSTP1 (P < 0.05; data not GSTT1 shown). The full model also gave a significantly improved fit Present 555 (83.6) 1032(82.8) 1.00 (Reference) compared with the null model (m2 = 31.15 on 7 df; P < 0.0001), Null 109 (16.4) 214 (17.2) 1.12 (0.86-1.45) 0.4 even after adding eight nonsignificant effects for all inter- GSTP1 P *Adjusted for reference age, education, family history (any reported first degree actions with ( < 0.01; data not shown). The full model or second degree), previous benign breast disease, state, country of birth, also gave a significantly better fit than the main-effects-only hormone replacement therapy ever, and menopause status. Initially also model (m2 = 21.21 on 4 df; P < 0.001), even after allowing for adjusted for body mass index, age at menarche, oral contraceptive ever, number GSTP1 (P < 0.02). Therefore, there was evidence for the of live births, smoking (never/ever), and drinking (never/ever); however, none of these were statistically significant and did not alter any of the estimates of presence of gene-gene interactions. GST or EPHX and interaction terms by 10%, and thus were dropped. There was The most parsimonious fit was provided by the highlighted no difference between the crude ORs and adjusted ORs. model in Fig. 1 with 5 df, with G = 30.26. The fit was significantly better than the best models with 4 df (P = 0.02), 2 df (P = 0.003), respectively). Therefore, further analyses were restricted to or 1 df (P = 0.0001). The best 3 df model was not nested within women who identified themselves as Caucasian. This included the most parsimonious model and cannot be formally com- 663 controls and 1,238 cases genotyped for EPHX; 649 controls pared, but an AIC difference of 11.2suggests the final model and 1,232 cases genotyped for GSTP1; 664 controls and 1,246 gave a superior fit. More complex models did not provide better GSTT1 GSTM1 df G P df cases genotyped for and , respectively (Table 1). fits: the best 6 gave 6 =30.55( = 0.6), whereas the 7 G P Evidence of deviation from Hardy-Weinberg equilibrium in model with all possible interactions gave 7 =31.15( = 0.6). controls (fewer heterozygotes than expected) was observed for As described in Table 3, the most parsimonious model that EPHX and persisted after confirmation of genotypes using arose from this procedure consisted of main effects for EPHX alternative genotyping methodology (P = 0.01). There was no (OR, 0.92; P = 0.5), GSTT1 (OR, 0.85; P = 0.3), and GSTM1 (OR, evidence of deviation from Hardy-Weinberg equilibrium in 1.36; P = 0.02), a two-way interaction between EPHX and controls for GSTP1 (P = 0.6). GSTM1 (interaction OR, 0.72; P = 0.03), and a three-way In no instance were substantial differences found between interaction between EPHX, GSTM1 and GSTT1 (interaction the unadjusted and adjusted estimates. Adjusted estimates are OR, 2.63 per EPHX C allele, P = 0.001). This implies an OR of reported in the text. As shown in Table 1, a significant decrease 2.02 (95% CI 1.19-3.45; P = 0.009) for those without a valid copy in breast cancer risk was associated with the EPHX CC of GSTM1 or GSTT1, and the TC genotype for EPHX. When genotype when compared with the TT genotype [OR, 0.60; combined with the EPHX CC genotype, this OR becomes 3.54 95% confidence interval (95% CI), 0.43-0.84; P = 0.003; trend (95% CI, 1.29-9.72; P = 0.01). When GSTP1 was added back to OR, 0.83; 95% CI, 0.72-0.96; P = 0.01]. There was no association the final model as a main effect, there was neither evidence of of breast cancer risk with GSTP1 genotype (GG versus AA OR, association as a main effect (OR, 0.95; P = 0.5) nor modification 0.90; 95% CI, 0.66-1.25; trend OR, 0.95; 95% CI, 0.82-1.10). The of the effects of the other genes (results not shown).

Table 2. Number of cases and controls by GSTM1, GSTT1, and EPHX genotype status

GSTM1 present GSTM1 null

GSTT1 present GSTT1 null GSTT1 present GSTT1 null

Controls Cases Controls Cases Controls Cases Controls Cases EPHX TT 123 239 37 44 124 295 31 60 TC 108 185 22 35 118 210 12 64 CC 36 56 4 4 41 36 3 7

Cancer Epidemiol Biomarkers Prev 2007;16(4). April 2007 Downloaded from cebp.aacrjournals.org on September 25, 2021. © 2007 American Association for Cancer Research. Cancer Epidemiology, Biomarkers & Prevention 773

Table 3. Estimates of main and interaction effects for EPHX, GSTM1, and GSTT1 from the most parsimoniuos model

Crude Adjusted*

OR (95% CI) P OR (95% CI) P EPHX 0.92(0.75-1.13) 0.4 0.92(0.75-1.14) 0.5 GSTT1 null 0.77 (0.58-1.02) 0.07 0.85 (0.63-1.15) 0.3 GSTM1 null 1.35 (1.04-1.75) 0.021.36 (1.04-1.78) 0.02 EPHX*GSTM1-null 0.70 (0.52-0.93) 0.02 0.72 (0.53-0.97) 0.03 EPHX*GSTM1-null* GSTT1-null 3.07 (1.75-5.40) <0.001 2.63 (1.48-4.67) 0.001

*Adjusted for reference age, education, family history (any reported first degree or second degree), previous benign breast disease, state, country of birth, hormone replacement therapy ever, and menopause status. Initially also adjusted for body mass index, age at menarche, oral contraceptive ever, number of live births; however, none of these were statistically significant and did not alter any of the estimates of GST or EPHX and interaction terms by 10% and thus were dropped. EPHX genotype was considered as a linear term.

When the most parsimonious model was fitted to cases and no association for GSTM1 (OR, 0.98), GSTT1 (OR, 1.11), or controls separately by Poisson regression, the two-way GSTP1 (OR for CC genotype, 0.93), and no interactions with interaction was significant only for the cases (P = 0.035). There smoking exposure, hormonal risk factors, or combinations of was no evidence of nonindependent assortment of EPHX and these genes. Last, a meta-analysis of seven case-control studies GSTM1 for the controls (P = 0.2). However, there seemed to be (2,815 cases, 3,170 controls) specifically evaluating the risk a three-way association between EPHX, GSTT1, and GSTM1 associated with cigarette smoking by GSTM1 genotype for the controls. Although this dependence was also seen in the reported that smoking was associated with risk only when cases, the association was greater for the controls (estimated GSTM1 is deleted (30). Although the overall sample size coefficient = À0.6; P = 0.002) than for the cases (estimated included in these meta-analyses would have given sufficient coefficient = 0.2; P = 0.05). power to detect small alterations in risk, these meta-analyses are likely to have been flawed by several factors. These include design issues (hospital-based versus population- Discussion based) and failure to adjust for ethnicity. In relation to ethnicity, differences between studies are traditionally Detoxification genes have long been considered plausible thought to reflect confounding by ethnic genetic background candidates genes for predisposition to breast and other or unmeasured exposures which differ between study cancers. Our case-control study of a sample of relatively populations. Alternatively, such differences may relate (at early-onset breast cancer found evidence for a protective effect least in part) to the fact that the GSTM1 deletion genotype associated with the EPHX CC genotype, suggestive evidence differs in frequency across ethnic groups (ref. 20; this study), for increased risk associated with the GSTM1 deletion and even within Caucasians (37.5-64.0%). This has the genotype, and no association with the GSTT1 deletion potential to confound the results due to issues associated genotype or GSTP1 variant genotypes overall. These associa- with assay design—the gel-based assay design used almost tions were not modified by menopausal status or other exclusively to date for genotyping this deletion variant has reproductive risk factors, and there was no strong evidence precluded detection of heterozygote null carriers separately for interaction with cigarette smoking or regular alcohol from homozygote wild-type. Consequently, the proportion of drinking exposure. heterozygotes within this pool will vary with allele frequency, As these genes act in a common pathway, we systemat- as will the extent to which any codominant effect effectively ically tested a priori for gene-gene interactions using an mask the risk associated with the homozygous deletion. It is hierarchical statistical model. We found evidence that the interesting to note that from the meta-analysis of Sull et al. addition of gene-gene interactions gave a better fit than main (20), the risk for breast cancer was significantly increased effects only, as models that included interactions gave a for populations with a lower GSTM1 null allele frequency significantly better fit than one that included only main (< 50.4%), but not for populations with GSTM1 null allele effects. Of all the interaction models, the most parsimonious frequency z50.4%. However, most importantly, these previ- one predicted that the GSTM1 and GSTT1 deletions were ous studies did not systematically evaluate the role of within associated with a 2-fold increased risk of breast cancer when pathway interactions as we have. jointly inherited with the EPHX TC genotype, and 3.5-fold This is the first large study investigating the associations with the CC genotype. between EPHX and GST gene variants and breast cancer risk. The involvement of GST genes in breast cancer etiology has Our findings of associations with the GSTT1 and GSTP1 been widely investigated by a large number of small studies genotypes individually are in line with the meta-analysis with conflicting results, and joint analyses of published data studies, but not our findings of the GSTM1 deletion geno- sets have provided no firm conclusions regarding their role in type alone. We also found evidence of statistical interaction breast cancer risk. Meta-analysis of at least 2,000 cases and between genotypes, with a three-way interaction between 2,000 controls from 10 to 15 studies of predominantly EPHX, GSTT1, and GSTM1 genotypes, with an estimate that Caucasian women by Egan et al. (19) found a borderline was much larger than for any individual genotypes or other significant association with GSTM1 in postmenopausal epidemiologic/demographic variables included in the same women only (OR, 1.14), a borderline significant increasing adjusted model. This suggests that the risk of breast cancer risk for the GSTT1 deletion (OR, 1.11), and no association with from such gene-gene interactions may be higher than for many GSTP1 GG genotype (OR, 1.04). The meta-analysis of GSTM1 of the known epidemiologic risk factors of breast cancer. data from more than 2,000 cases and 2,000 controls from 10 Similar evidence for a contribution to breast cancer suscepti- studies by Sull et al. (20) also found no overall association for bility by two-way interactions between polymorphisms of the deletion genotype, but a significant increased risk for likely functional relevance, including those within the carcin- postmenopausal women only (pooled OR, 1.19), particularly ogen metabolism, has recently been presented (31). for populations with a low frequency of the GSTM1 null Although there is a strong biological rationale for inter- allele. The smaller pooled analysis of Vogl et al. (21) of seven actions between polymorphisms in the same pathway, we case-control studies (558-1,899 controls, 921-2,033 cases) found recognize that we have tested multiple hypotheses. Great

Cancer Epidemiol Biomarkers Prev 2007;16(4). April 2007 Downloaded from cebp.aacrjournals.org on September 25, 2021. © 2007 American Association for Cancer Research. 774 GST and EPHX Polymorphisms and BreastCancer caution should be exercised when assessing interactions, 11. Board PG. Biochemical genetics of glutathione-S-transferase in man. Am J Hum Genet 1981;33:36 – 43. especially those based on small cell sizes, even though they 12. Seidegard J, Vorachek WR, Pero RW, et al. Hereditary differences in the have been derived from a moderately large study. The expression of the human glutathione transferase active on trans-stilbene importance of very large sample sizes in providing sufficient oxide are due to a gene deletion. Proc Natl Acad Sci U S A 1988;85: statistical power to investigate with confidence possible effect 7293 – 7. 13. Rebbeck TR. Molecular epidemiology of the human glutathione modifications or interactions has been highlighted by studies S CYP17 -transferase genotypes GSTM1 and GSTT1 in cancer susceptibility. Cancer of the hormone-metabolizing gene (32). We thus Epidemiol Biomarkers Prev 1997;6:733 – 43. present the data in a way that others can combine with their 14. Ahmad H, Wilson DE, Fritz RR, et al. Primary and secondary structural own so as to try to replicate or refute our findings. analyses of glutathione S-transferase k from human placenta. Arch Biochem Biophys 1990;278:398 – 408. This study further highlights the need for pooling of 15. Harries LW, Stubbins MJ, Forman D, et al. Identification of genetic comparable data from large population-based case-control polymorphisms at the glutathione S-transferase k locus and association studies to have any hope of establishing definitive gene-gene with susceptibility to bladder, testicular and prostate cancer. Carcinogenesis interactions in breast cancer predisposition. These studies 1997;18:641 – 4. 16. Zimniak P, Nanduri B, Pikula S, et al. Naturally occurring human should use consistent or comparable exposure measures to be glutathione S-transferase GSTP1-1 isoforms with isoleucine and valine able to assess effect modification by common environmental in position 104 differ in enzymic properties. Eur J Biochem 1994;224: carcinogen exposures, and to adjust for factors such as 893 – 9. socioeconomic status as a surrogate for unmeasured expo- 17. Ali Osman F, Akande O, Antoun G, et al. Molecular cloning, characteriza- tion, and expression in Escherichia coli of full-length cDNAs of three human sure(s). The role of the Breast Cancer Family Registry (33), and glutathione S-transferase k gene variants. Evidence for differential catalytic the recently established Breast Cancer Association Consortium activity of the encoded proteins. J Biol Chem 1997;272:10004 – 12. (34), will be integral to confirmation of speculative findings 18. Johansson AS, Stenberg G, Widersten M, et al. Structure-activity relation- such as those reported here. ships and thermal stability of human glutathione transferase P1-1 governed by the H-site residue 105. J Mol Biol 1998;278:687 – 98. 19. Egan KM, Cai Q, Shu XO, et al. Genetic polymorphisms in GSTM1, GSTP1, Acknowledgments and GSTT1 and the risk for breast cancer: results from the Shanghai Breast Cancer Study and meta-analysis. Cancer Epidemiol Biomarkers Prev 2004; We thank the many women who participated in this research; the 13:197 – 204. interviewing staff; physicians, surgeons, and oncologists who en- 20. Sull JW, Ohrr H, Kang DR, et al. Glutathione S-transferase M1 status and dorsed this project; and R. Minchin of University of Queensland for breast cancer risk: a meta-analysis. Yonsei Med J 2004;45:683 – 9. helpful comments on the manuscript. 21. Vogl FD, Taioli E, Maugard C, et al. Glutathione S-transferases M1, T1, and P1 and breast cancer: a pooled analysis. Cancer Epidemiol Biomarkers Prev 2004;13:1473 – 9. 22. Sarmanova J, Susova S, Gut I, et al. Breast cancer: role of polymorphisms in References biotransformation . Eur J Hum Genet 2004;12:848 – 54. 1. Hopper JL, Southey MC, Dite GS, et al. Population-based estimate of the 23. Hopper JL, Chenevix-Trench G, Jolley DJ, et al. Design and analysis issues in average age-specific cumulative risk of breast cancer for a defined set of a population-based case-control-family study of the genetic epidemiology of protein-truncating mutations in BRCA1 and BRCA2. Cancer Epidemiol breast cancer and the Co-operative Family Registry for Breast Cancer Biomarkers Prev 1999;8:741 – 7. Studies (CFRBCS). J Natl Cancer Inst Monogr 1999;26:95 – 100. 2. Easton D, Peto J. The contribution of inherited predisposition to cancer 24. Dite GS, Jenkins MA, Southey MC, et al. Familial risks, early-onset breast incidence. Cancer Surv 1990;9:395 – 416. cancer, and BRCA1 and BRCA2germline mutations. J Natl Cancer Inst 2003; 3. Rebbeck TR, Martinez ME, Sellers TA, et al. Genetic variation and cancer: 95:448 – 57. improving the environment for publication of association studies. Cancer 25. McCredie MR, Dite GS, Giles GG, et al. Breast cancer in Australian women Epidemiol Biomarkers Prev 2004;13:1985 – 6. under the age of 40. Cancer Causes Control 1998;9:189 – 98. 4. Byrnes G, Gurrin L, Dowty J, et al. Publication policy or publication bias? 26. Spurdle AB, Purdie DM, Webb PM, et al. The microsomal epoxide hydrolase Cancer Epidemiol Biomarkers Prev 2005;14:1363. Tyr113His polymorphism: association with risk of ovarian cancer. Mol 5. Oesch F. Mammalian epoxide hydrases: inducible enzymes catalysing the Carcinog 2001;30:71 – 8. inactivation of carcinogenic and cytotoxic metabolites derived from aromatic 27. Spurdle AB, Webb PM, Purdie DM, et al. Polymorphisms at the glutathione and olefinic compounds. Xenobiotica 1973;3:305 – 40. S-transferase GSTM1, GSTT1 and GSTP1 loci: risk of ovarian cancer by 6. Adang AE, Meyer DJ, Brussee J, et al. Interaction of rat glutathione histological subtype. Carcinogenesis 2001;22:67 – 72. S-transferases 7-7 and 8-8 with g-glutamyl- or glycyl-modified glutathione 28. Clayton D, Hills M. Statistical models in epidemiology. Oxford; New York: analogues. Biochem J 1989;264:759 – 64. Oxford University Press; 1993. 7. Hengstler JG, Arand M, Herrero ME, et al. Polymorphisms of N- 29. Agresti A. Categorical data analysis. New York; Brisbane: Wiley; 1990. acetyltransferases, glutathione S-transferases, microsomal epoxide hydro- 30. Terry PD, Goodman M. Is the association between cigarette smoking and lase and sulfotransferases: influence on cancer susceptibility. Recent Results breast cancer modified by genotype? A review of epidemiologic studies and Cancer Res 1998;154:47 – 85. meta-analysis. Cancer Epidemiol Biomarkers Prev 2006;15:602 – 11. 8. Strange RC, Fryer AA. The glutathione S-transferases: influence of 31. Onay VU, Briollais L, Knight JA, et al. SNP-SNP interactions in breast cancer polymorphism on cancer susceptibility. In: Vineis P, editor. Metabolic susceptibility. BMC Cancer 2006;6:114. polymorphisms and susceptibility to cancer no. 148. Lyon: IARC Scientific 32. Little J, Simard J. CYP17 and breast cancer: no overall effect, but what about Publications; 1999. p. 231 – 49. interactions? Breast Cancer Res 2005;7:238 – 42. 9. Hassett C, Aicher L, Sidhu JS, et al. Human microsomal epoxide hydrolase: 33. John EM, Hopper JL, Beck JC, et al. The Breast Cancer Family Registry: an genetic polymorphism and functional expression in vitro of amino acid infrastructure for cooperative multinational, interdisciplinary and transla- variants. Hum Mol Genet 1994;3:421 – 8. Erratum in: Hum Mol Genet tional studies of the genetic epidemiology of breast cancer. Breast Cancer 1994;3:1214. Res 2004;6:R375 – 89. 10. Pemble S, Schroeder KR, Spencer SR, et al. Human glutathione S-transferase 34. Breast Cancer Association Consortium. Commonly studied single-nucleo- u (GSTT1): cDNA cloning and the characterization of a genetic polymor- tide polymorphisms and breast cancer: results from the Breast Cancer phism. Biochem J 1994;300:271 – 6. Association Consortium. J Natl Cancer Inst 2006;98:1382 – 96.

Cancer Epidemiol Biomarkers Prev 2007;16(4). April 2007 Downloaded from cebp.aacrjournals.org on September 25, 2021. © 2007 American Association for Cancer Research. A Systematic Approach to Analysing Gene-Gene Interactions: Polymorphisms at the Microsomal Epoxide Hydrolase EPHX and Glutathione S-transferase GSTM1, GSTT1 , and GSTP1 Loci and Breast Cancer Risk

Amanda B. Spurdle, Jiun-Horng Chang, Graham B. Byrnes, et al.

Cancer Epidemiol Biomarkers Prev 2007;16:769-774.

Updated version Access the most recent version of this article at: http://cebp.aacrjournals.org/content/16/4/769

Cited articles This article cites 31 articles, 11 of which you can access for free at: http://cebp.aacrjournals.org/content/16/4/769.full#ref-list-1

Citing articles This article has been cited by 2 HighWire-hosted articles. Access the articles at: http://cebp.aacrjournals.org/content/16/4/769.full#related-urls

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://cebp.aacrjournals.org/content/16/4/769. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.

Downloaded from cebp.aacrjournals.org on September 25, 2021. © 2007 American Association for Cancer Research.