Polymorphisms in Nucleotide Excision Repair Genes and DNA Repair Capacity Phenotype in Sisters Discordant for Breast Cancer

Elevated levels of DNA damage caused by excessive exposure to carcinogens or inherited deficiencies in repair capacity may increase an individual's susceptibility to many kinds of human cancers (1, 2). Using different in vitro phenotype assays, several recent studies have shown that DNA repair capacity (DRC) has considerable interindividual variation and that individuals with reduced DRC are at increased risk of developing breast cancer (3-6). DRC measured by the host-cell reactivation assay using both luciferase (luc) and chloramphenicol acetyltransferase (cat) as reporter genes, indicated that women with breast carcinoma had a 22% to 36% reductions in DRC compared with controls (4-6). Those with DRC lower than the median level of controls had a 3-fold increased risk for breast cancer compared with those with higher DRC (5). Our own previous finding also indicated not only that do breast cancer patients have a 24.5% decrease in average DRC compared with their sisters without breast cancer but also that there is an increase in risk with decreasing DRC (odds ratios ranged from 1.2 to 2.4 to 3.0, by quartiles; P_trend < 0.01) using an assay measuring the removal of in vitro–induced benzo[a]pyrene diolepoxide (BPDE)-DNA adducts in lymphoblastoid cells (3). For several other types of human cancers (lung, prostate, skin, head, and neck cancers), increased risk is also associated with deficient DRC (7). These phenotype studies are usually good at predicting cancer risk but have relatively large variations in risk estimates because of the large assay variation in addition to the relatively small numbers of subjects compared with genotype association studies (2). Because the assays are very time consuming and it is not easy to develop large-scale, high-throughput methods for molecular epidemiologic studies, few subjects are assayed, a major limitation of phenotype assays (3, 5, 8-12).

It is likely that single nucleotide polymorphisms (SNP) in coding and regulatory sequences of genes in the nucleotide excision repair (NER) pathway may result in subtle structural alterations in DNA repair enzymes and modulate cancer susceptibility by affecting individual DRC (13-19). Previous studies showed that DRC for removal of BPDE-induced DNA damage measured by the host-cell reactivation assay was modulated by two XPD/ERCC2 polymorphisms (Asp312Asn and Lys751Gln; ref. 20). In other studies, DRC was consistently lower in subjects homozygous for XPC intron 9 poly(AT) or XPD 156Arg, 312Asn, and 751Gln than in subjects with other genotypes, suggesting these polymorphic alleles might have a recessive effect on the DRC phenotype (9, 21). But results for XPD Asp312Asn are inconsistent (22). It may be difficult to detect subtle differences in DRC by studying only one or two SNPs in a very complex pathway. Genotype studies might be improved by using a robust study design and including multiple polymorphisms in the same pathway to estimate risks associated with individual DRC.

In the present study, we used sisters discordant for breast cancer from the Metropolitan New York Registry, one of six international collaborating sites (http://www.metronyregistry.org/) of the Breast Cooperative Family Registry (23) to predict the influence of seven SNPs in the NER pathway genes (XPA 5′ untranslated region +62T>C, XPC Ala499Val and Lys939Gln, XPD Asp312Asn and Lys751Gln, XPG His1104Asp, and ERCC1 8092 C>A) on DRC phenotype. Our aim was to evaluate whether these genetic polymorphisms in NER pathway genes could partly explain the variation of individual DRC phenotype. If so, then sets of polymorphisms in the DNA repair pathway, which are easier to measure in large-scale epidemiologic studies, may be a proxy for DNA repair phenotype, which is less easily measured in epidemiologic settings.

The Breast Cooperative Family Registry is an international consortium established in 1995 as a resource for research on the epidemiologic, clinical, and genetic aspects of breast cancer (http://epi.grants.cancer.gov/CFR/; ref. 23). The description of the sources of study participants and recruitment and data collection methods as part of the parent Metropolitan New York Registry project have been described in detail elsewhere (3, 24). Briefly, the Metropolitan New York Registry has been recruiting high-risk breast and/or ovarian cancer families from clinical and community settings within the metropolitan New York area since 1995 who met one of the following criteria: (a) one or more members with breast cancer or ovarian cancer diagnosed at <45 years of age; (b) one or more members with both breast and ovarian cancer; (c) two or more first-degree relatives with breast or ovarian cancer diagnosed at age ≥45 years; (d) any male with breast cancer; and (e) a known BRCA mutation. The participants are interviewed either in person or by phone using epidemiology and family history questionnaires that collect information on demographics, ethnicity, history of cancers, smoking, and alcohol consumption, reproductive history, hormone use, weight, height, and physical activity. A self-administered dietary questionnaire is also provided with return by mail. In addition, a sample of blood or buccal cells is collected from participants. The present study includes 160 sister pairs with available lymphoblastoid cell lines for phenotype.

Genomic DNA was extracted from WBC by salting out and SNPs were determined by one of two different methods: fluorescence polarization and Taqman assay. A commercial AcycloPrime-FP SNP Detection kit obtained from Perkin-Elmer Life Sciences (Boston, MA) was used for detection of five SNPs detection. The Taqman system using a 7500 Real-time PCR system (ABI Applied Biosystems, Foster City, CA) with allelic discrimination software supplied by the manufacturer was used for the remaining two SNPs. Table 1 displays the genes and polymorphisms included in this study, with their National Center for Biotechnology Information reference sequence, chromosome location, and polymorphic site as well as the PCR and single nucleotide extension primers for the fluorescence polarization assay and the Taqman probes. Genotyping data were complete for 99.5% of breast cancer cases and 98.9% of controls. Ten percent of samples were reassayed after relabeling to keep laboratory staff blinded. Concordance for reassayed samples was 99.6% (279/280).

DRC in the NER pathway was determined using a phenotype assay that measured the removal of in vitro–induced BPDE-DNA adducts in lymphoblastoid cells essentially as described previously (3). Data were available for 156 cases and 153 controls who were 97.5% and 95.6% of the eligible cases (160) and controls (160) with lymphoblastoid cell lines. All assays were done with the laboratory blinded to the subject's case control status.

Hardy-Weinberg equilibrium was tested to compare the observed and expected genotype frequencies among cases and controls. DRC was calculated from the formula % DRC = 100 × [(I₀−I_C) − (I₄−I_C)] / (I₀−I_C), where I₀ is the intensity of treated cells immediately after treatment, I₄ after 4 hours of repair, and I_C is the value for control, untreated cells (3). Paired t test was used for comparisons of continuous variables and McNemar's test for comparison of categorical variables between cases and sister controls. One-way ANOVA was used to compare the means of DRC across the various genotypes for each gene.

A conditional logistic regression model was used to predict risk for low DRC defined as having a DRC ≤35.1% (the median DRC value of the controls; ref. 25). The likelihood ratio test was used to assess the effect of variables of interest, such as case-control status (yes and no), genotype status, age at time of blood donation (<50, 50-59, and ≥60 years old), smoking status (yes and no), and body mass index (BMI) level (<25, 25-29, and ≥30). Because age at time of blood donation and BMI, as continuous variables, were not linearly related to DRC, they were modeled as categorical variables for the analysis. The final model was developed in the following way. First, we ran all models separately among cases and controls and we also pooled the data modeling multiplicative interaction terms between the case-control status and the predictors described above. If the interaction term was not significant at the 0.20 level, we reassessed the effect of the variable on low DRC without adjusting for its interaction term. If the interaction term was significant, we included it in the final model when we assessed the model fit. Second, predictors were considered as potential predictors in the final model if they were significant at the 0.20 level. Age group and case-control status, however, were considered to be important risk factors of low DRC a priori and therefore adjusted for in all models. Those variables that were no longer significant at the 0.20 levels were removed one at a time based on their corresponding likelihood ratio test statistic. Next, variables that had been initially omitted in step 2 are included one at a time to reassess their importance in the presence of the remaining variables. The final model includes those predictors significant at the 0.20 level along with age and case-control status. The predictability of the model was assessed using receiver operator characteristic (ROC) curves, and sensitivity and specificity were calculated (26). All of the statistical analyses were carried out using Statistical Analysis System version 9.0 (SAS Institute, Cary, NC).

Selected characteristics of the sisters discordant for breast cancer are presented in Table 2. Overall, there were no statistically significant differences between breast cancer cases and sister controls in the distribution of age at blood donation, smoking status, number of cigarettes daily, and BMI. Similar to our previous results, in this slightly larger sample, DRC was statistically significantly lower in the cases than in the controls (26.5% versus 35.4%; P < 0.01), representing an average 33.5% reduction in DRC among the cases. Low DRC, defined as having a value below the median of the controls, was found to be statistically significantly associated with elevated breast cancer risk (adjusted odds ratio, 2.3; 95% confidence interval, 1.4-3.8). This is consistent with our previous observation (3).

We compared the average DRC among different genotypes in the NER pathway genes separately by case-control status (Table 3). Overall, there were no statistically significant differences in mean DRC by allele for most of the SNPs. A significantly higher DRC was found among controls with one or two copies of the XPC codon 499 variant T allele compared with those carrying the wild-type CC genotype (P = 0.04). Among breast cancer cases, subjects with the XPD codon 312 heterozygous GA genotype had significantly higher DRC (31.1%) compared with those with the wild-type GG genotype (21.4%). But there was no significant difference in DRC between those with the codon 312 homozygous AA genotype (22.5%) and with GG genotype among breast cancer cases.

We conducted an analysis to determine whether the seven SNPs in the NER pathway genes and factors (age group of blood donation, cigarette smoking status, and BMI levels) predicted low DRC phenotype in a multivariate conditional logistic model (Table 4). The final model included three genes (XPA +62T>C, XPC Ala499Val, and XPG His1104Asp) and ever smoking after adjusting for case-control status and age group of blood donation.

ROC curve analysis was used to identify the predictive accuracies of the forecasted model for low DRC. ROC curves were separately obtained from the full model (including all genetic and environmental variables available) and the final model. The overall predictive accuracy (defined as the percentage of patients correctly classified as case and controls correctly distinguished as noncase using the model) was 61% for the probability cut point of 0.50 in the final forecasted model with a sensitivity of 78% and a specificity of 39%. This is similar to the predictive accuracy (63%) of the full model that has a sensitivity of 74% and a specificity of 48% (Fig. 1). In the full model of environmental variables only, the overall predictive accuracy was 58% for the probability cut point of 0.50 with a sensitivity 73% and a specificity 38% (data not shown), which are slightly lower than that in the final model.

This molecular epidemiologic study used seven genetic polymorphisms in the NER pathway genes to predict individual DRC phenotype. SNP profiles in the NER pathway and ever smoking status only modestly predict variation of individual DRC after adjusting for case-control status and age of blood donation (Table 4). The SNPs selected in the final predictive model included XPA +62T>C, XPC Ala499Val, and XPG His1104Asp. The overall predictive accuracy of the final model was 61% with a moderate sensitivity (78%), but the specificity was relative low (39%). Our failure to predict individual DRC phenotype with these variables suggests that there are other unknown factors (including genetic, epigenetic, and nongenetic factors) that contribute to individual differences in DRC.

Until now, few studies have examined the effects of genetic polymorphisms in the NER pathway genes on individual DRC phenotype. A small number of SNPs are usually analyzed and the results have not always been concordant. Wu et al. (12) observed that the XPA-A23G variant was associated with more efficient DRC in control subjects compared with those with the homozygous A allele but not in cases. Qiao et al. (9) found that DRC was lower in subject homozygous for the XPC intron 9 poly(AT) variant, and Spitz et al. (20, 21) reported that two XPD polymorphisms (Asp312Asn and Lys751Gln) were associated with reduced NER capacity and increased risk of lung cancer. But, other studies did not show the same associations with cancer risk (22, 27, 28). Genotyping in cancer association studies has also provided some evidence that these polymorphisms might be involved in cancer risk (12, 29, 30). Our ability to predict only a small part of individual DRC phenotype was not surprising because we included only ∼3.5% of the total known SNPs in the NER pathway. There are >60 other SNPs in 30 NER genes unanalyzed in the present study that might potentially influence function (7, 31). The NER is a pathway composed of four major consecutive steps: damage recognition, unwinding around the damaged sites, excision of the oligonucleotide containing the site of damage, and gap filling by DNA repair synthesis (31, 32). The DRC phenotype assay used in the current study measures the ability to carry out all four steps in NER, whereas the seven genotypes studied here are primarily involved in the first three steps.

In addition to genetic polymorphisms, previous study suggested that cancer status, age, and cigarette smoking might affect DRC phenotype. Reports using direct or indirect measurement of DNA damage repair found a decline in DRC with donor age in human dermal fibroblasts, lymphocytes, and transformed lymphoblastoid cells (33-36). A study in Japan found that cells from elderly subjects recovered slower than cells from young subjects after UV stimulation (31). The variable treatment history for cancer patients, either radiation or chemotherapy, may also lead to decreased DRC (37-39). To control for the potential effect of these factors on DRC, we adjusted for breast cancer status and age group of blood donation in the process of model selection. In the final model, consistent with previous observations from cancer patients (37-39), breast cancer status was significantly associated with low DRC phenotype. We reported previously (3) that DRC decreased with age among the unaffected sisters as expected but had an opposite pattern in the sisters with cancer. In these analyses, we did not observe a linear association with age; women >60 years were more likely to have poorer DNA repair, although the point estimate was not statistically significant. We observed a positive association between smoking and DRC phenotype (Table 4) in agreement with another study finding smoking associated with DRC (40).

Compared with previous studies, our study has several advantages. One strength was the family-based design using sisters from the same families as controls, which can potentially reduce confounding from population admixture as well as lifestyle factors that cluster within families (41). The study included nuclear families that were not selected based on a population-based scheme and may limit the generalizability of the findings (24). The study sample is of moderate size limiting our statistical power. Given our sample size and the range of frequencies for our genotypes, we were only able to detect associations 2-fold or higher between genotype and phenotype. Other factors possibly predicting individual DRC phenotype, such as dietary antioxidants, were not examined in the current study (42). These results should therefore be viewed as preliminary until confirmed in larger studies.

In conclusion, specific SNPs in the NER pathway genes (XPA +62T>C, XPC Ala499Val, and XPG His1104Asp) and ever smoking only modestly predict individual variation in DRC; overall predictive accuracy (61%) was relatively low (sensitivity, 78%; specificity, 39%). These NER pathway SNPs that we selected are commonly assessed in epidemiologic studies and their poor prediction for DNA repair phenotype may offer some evidence of why studies only using SNPs have been inconsistent. It may still be possible that an extensive panel of DNA repair SNPs may predict phenotype, but our findings suggest that the small subsets of SNPs are likely a poor proxy for phenotype. The results obtained from discordant sister pairs in a breast cancer family registry should serve as a stimulus for further studies on the complex molecular events that affect individual DRC phenotype.

We thank Irina Gurvich for the processing, storage, and inventory of biospecimens for the study.