Polymorphisms in Nucleotide Excision Repair Genes and Endometrial Cancer Risk

The increased risk of endometrial cancer associated with estrogen exposure has been hypothesized to result, at least in part, via 2 possible mechanisms: (i) increased cellular proliferation (1, 2) and (ii) through the production of certain estrogen metabolites, which can result in DNA damage by directly binding to DNA and forming bulky DNA adducts, and/or indirectly through the production of reactive oxygen species (3–5). Estradiol and estrone are metabolized primarily via 2-, 4-, and 16α-hydroxylation (3, 6). The 2- and 4-hydroxylated metabolites can be further oxidized to semiquinones and quinones, which can form bulky DNA adducts and can undergo redox cycling, producing reactive oxygen species that may cause oxidative stress, lipid peroxidation, and DNA damage (4, 5).

The ability to repair DNA damage caused by estrogens could plausibly affect endometrial cancer risk. Nucleotide excision repair (NER) identifies and excises bulky adducts formed by estrogen, as well as by other substances such as components of cigarette smoke and well-cooked meat (7). NER also corrects UV-induced pyrimidine dimers and cross-links (8). As many as 30 proteins, including those encoded by the xeroderma pigmentosum (XP) genes, are involved in this process, in which a large multiprotein complex is assembled. First, the damaged DNA is recognized by XPC, centrin 2, and HHRAD23B. XPA and RPA (replication protein A) then bind to the lesion, allowing the proper positioning of the endonucleases ERCC1 (excision repair cross-complimenting 1)-ERCC4 (XPF) and ERCC5 (XPG; ref. 9). After the DNA in this region is unwound by the helicases ERCC3 (XPB) and ERCC2 (XPD), which are 2 subunits of the transcription factor IIH, ERCC1-ERCC4 and ERCC5 (XPG) excise a ∼30-bp oligonucleotide fragment (at the 5′ and 3′ ends, respectively) containing the bulky adduct. DNA polymerases and ligases use the intact strand as a template to fill the resulting gap (8).

In an earlier study, we examined whether 11 candidate variants in NER genes were associated with endometrial cancer risk (371 cases and 420 controls; ref. 10). We have since nearly doubled our study population, and herein, we report findings from a more systematic evaluation of single-nucleotide polymorphism (SNP) in the genes ERCC1, ERCC2 (XPD), ERCC3 (XPB), ERCC4 (XPF), ERCC5 (XPG), LIG1, XPA, and XPC in the combined studies.

We genotyped a total of 783 residents of western Washington who were diagnosed with endometrial cancer in 1994–1995, 1997–1999, and 2003–2005 and 795 population-based controls. The studies of women diagnosed during the 1990s have been described previously (11). Briefly, eligible cases (n = 582) comprised Caucasian and African-American female residents of western Washington state aged 50 to 69 years diagnosed with invasive endometrial cancer between January 1, 1994, and December 31, 1995 (King county only), and between July 1, 1997, and December 31, 1999 (King, Pierce, and Snohomish counties). These women were identified through the Cancer Surveillance System, a population-based tumor registry affiliated with the Surveillance, Epidemiology and End Results Program of the National Cancer Institute (12). A total of 472 (81.1%) women were successfully interviewed and 383 of them (81.1%) provided a blood sample.

Eligible control women included Caucasian and African American female residents of the 3-county area during the years the cases were diagnosed, with intact uteri and no prior history of endometrial cancer. Those selected by random-digit dialing (RDD; ref. 13; women aged 50–65 years) and random selection from Health Care Financing Administration data files (women aged 66–69 years) were frequency matched to the cases by 5-year age group and county of residence. The overall RDD response [the screening response (91.3%) multiplied by the interview response (83.6%)] was 76.3%, resulting in 297 interviewed women. Of the 175 eligible Health Care Financing Administration controls, 116 (66.3%) agreed to an interview. An additional source of population-based controls was the CARE study, which was conducted during the same period as the endometrial cancer case–control study using a similar questionnaire (14). The CARE study controls included Caucasian and African American women aged 35 to 64 years ascertained through RDD in 5 metropolitan areas of the United States, including King county, between 1994 and 1998. The screening and interview response for King county were 83.6% and 88.3%, respectively. We invited 132 King county CARE control women aged 50 to 64 years, with intact uteri, to provide a blood sample, and we successfully obtained a blood sample from 115. Overall, of the 929 eligible controls, 664 (71.5%) were interviewed and 449 provided a blood sample (67.7% of interviewed controls). The data from 2 controls in the earlier case–control study who were ascertained as cases in the later case–control study, were included in both case and control groups; and 1 case was excluded because of poor quality interview data. Thus, for the earlier 2 studies, there were 382 cases and 449 controls with blood samples available for genotyping.

For the most recent study (2003–2005; also described in Bodelon and colleagues; ref. 15), eligible case participants included all female residents of western Washington state aged 50 to 74 years diagnosed with invasive endometrial cancer between July 1, 2003, and November 31, 2005, in King, Pierce, and Snohomish counties, identified through the Cancer Surveillance System. Of 586 eligible cases, we were unable to locate 12, 34 were deceased before we could contact them, and 130 refused to participate (or their physician instructed us not to contact them). A total of 410 (70.0%) women were successfully interviewed. Of these, 401 provided a blood or buccal sample.

Control women were a subset of those enrolled in a population-based case–control study of ovarian cancer in western Washington that included women aged 35 to 74 years with reference dates of 2002 to 2005, described in detail in Rossing and colleagues (16). Controls were selected by RDD (13), and 82.0% of 17,768 residential telephone numbers were screened to determine whether an eligible woman resided there. Of the 1,561 women identified as eligible controls, 1,313 were interviewed (84.1%); the remaining women refused (n = 240) or were lost to follow-up (n = 8). The overall control response proportion (screening response × interview response) was 69.0%. From this set of population-based controls, we attempted to recruit women aged 50 to 74 years with intact uteri, who resided in King, Pierce, and Snohomish counties, and had reference dates comparable with the endometrial cancer cases. Of 365 eligible controls, 9 women refused to participate and 356 women were interviewed. A total of 347 control women provided a blood or buccal sample (though 2 of these samples were collected after the other samples had been plated and thus were not genotyped).

After informed consent, all participants were administered an in-person interview conducted according to a standard protocol. Each participant was asked only about events that occurred before her reference date, which is the date of diagnosis for cases. Controls were assigned a reference date on the basis of the distribution of diagnosis years for the cases. Data were collected on demographic factors; height; weight at different ages; reproductive, contraceptive, and menstrual history; family history of cancer; history of selected chronic conditions; and history of contraceptive and noncontraceptive hormone use. Color pictures of oral contraceptive and hormone replacement therapy pill packs were used to aid recall. The protocols of the studies were approved by the Institutional Review Board of the Fred Hutchinson Cancer Research Center.

We selected 64 tag and functional SNPs in ERCC1, ERCC2 (XPD), ERCC3 (XPB), ERCC4 (XPF), ERCC5 (XPG), LIG1, XPA, and XPC using the LDSelect algorithm on the National Institute of Environmental Health Sciences (NIEHS) Environmental Genome Project (17) sequencing data from 23 European Americans (>5% allele frequency; r² > 0.64).

Genotyping was conducted blinded to all characteristics of the study participants. In-house DNA of known genotypes was used for positive controls and negative controls were prepared identically but without template. Cases and controls were arranged randomly on plates including 56 duplicate pairs of samples used to evaluate genotyping concordance. We used several genotyping methods including SNPlex, TaqMan Assay On Demand, TaqMan Assay by Design, Snapshot (ABI), RFLP assays, and fragment analyses. The assay method used for each SNP and details about the assays (e.g., primers, probes, restriction enzymes where applicable) are included in Supplementary Table S1.

For each polymorphism, we calculated the proportion of samples that were successfully genotyped (the SNP genotyping success proportion). For SNPs genotyped using the SNPlex platform, we also assessed sample performance. The SNP genotyping success proportion was calculated after excluding samples that failed SNPlex (defined as those which were successfully genotyped for <90% of the SNPlex pool).These samples were included in calculating genotyping success for SNPs assayed with platforms other than SNPlex. For all of the SNPs, we evaluated genotype call concordance between the 56 pairs of duplicate samples, and we tested for deviation from Hardy–Weinberg equilibrium (HWE) among non-Hispanic white controls using Fisher's exact test. SNPs that were successfully genotyped in less than 95% of the samples or with duplicate concordance of less than 98% were excluded except as noted in the Results section. SNPs that deviated from HWE at P < 0.01 were also excluded.

Because we had too few women who were Hispanic or nonwhite (56 cases and 59 controls) to be able to conduct meaningful analyses of these subgroups, all analyses were restricted to non-Hispanic white women (727 cases and 736 controls). Per-minor allele ORs and 95% CIs for each SNP and endometrial cancer risk were calculated using unconditional logistic regression and were adjusted for age, county, and reference year. We also calculated ORs and 95% CIs for homozygotes and heterozygotes compared with homozygotes for the common allele. These analyses were conducted using Stata SE 11 statistical software. The Max(T) permutation test implemented in PLINK v1.07 (18, 19), with 10,000 permutations, was used to evaluate whether there were gene-level associations with endometrial cancer risk, after taking into account linkage disequilibrium (LD) between SNPs and the number of SNPs in the gene. We used HPlus (20–22) to estimate gene-specific haplotype frequencies and calculated ORs and 95% CIs to estimate their associations with endometrial cancer risk, using the most common haplotype as the reference category. Utilizing the logistic regression model as the penetrance function, HPlus directly assesses associations between haplotypes and outcome (e.g., endometrial cancer), after taking into account uncertainties of haplotype inference from unphased genotype data.

Because endometrial cancer risk is so strongly influenced by obesity and menopausal use of high-risk estrogen regimens, we explored whether the magnitude of the associations with the SNPs varied by exposure to these factors. Because of the particularly high risk of endometrial cancer associated with a body mass index (BMI) ≥30 kg/m², we dichotomized BMI as <30 kg/m² and ≥30 kg/m². High-risk estrogen use was dichotomized as use or nonuse of greater than 6 months of either unopposed estrogen or estrogen used with less than 10 days of progestin per month. We also examined associations stratified by parity (nulliparous vs. parous) and age (<55 years and ≥55 years). Tests for interaction between SNPs and these factors were conducted using a likelihood ratio test comparing a model with the SNP of interest to a model with an additional cross-product term for the SNP and the exposure of interest.

For the set of SNPs that were associated with endometrial cancer in our study, we evaluated whether including them all in a single logistic regression model affected the magnitude of the association between each of the SNPs and endometrial cancer. We also evaluated whether carriage of a combination of the genotypes was associated with endometrial cancer risk by first dichotomizing the data for each SNP, combining heterozygotes and homozygotes for the minor allele. We defined “risk” genotypes as the following: if our observed OR for the minor allele genotypes was greater than 1, then the “risk” genotype included heterozygotes and homozygotes for the minor allele; and if our observed OR was less than 1, then the homozygous major allele genotype was assigned as the “risk” genotype.

Finally, we calculated false-positive report probabilities (FPRP) for each SNP, using the Excel spreadsheet provided by Wacholder and colleagues (23). The FPRP is “the probability of no association given a statistically significant finding.” It is determined by the P value, the prior probability for the association, and the statistical power (23). Prior probabilities are based on previous knowledge of the relevance of the gene/SNP of interest to the outcome and evolve over time as more becomes known. We therefore calculated the FPRP for a range of prior probabilities [high (≈0.1), moderate (≈0.01), and low (≈0.001)]. Wacholder and colleagues (23) recommend that for an initial study of an SNP–disease association such as ours, the FPRP should be set at less than 0.5 to designate findings that are “noteworthy.”

Compared with controls, endometrial cancer cases were slightly older, had higher BMI, had fewer births, were less likely to have used oral contraceptives, and were less likely to have ever smoked. Cases were also more likely than controls to have taken a high-risk hormone therapy regimen (Table 1). Distributions of risk factors were similar for women who were and were not genotyped (data not shown).

Assays failed for 1 SNP in ERCC2 (rs3810366), 2 in ERCC4 (rs1799797 and rs1799801), and 1 each in LIG1 (rs156641), XPA (rs2805834), and XPC (rs2227999). Genotyping success for ERCC2 rs50872 was 92.8%, but because the duplicate concordance was 100%, we did not exclude this SNP. The duplicate concordance for ERCC5 rs4150375 was 96.4%, but genotyping success was ≥95%, so it was also included. The genotype distribution for XPA rs3176689 deviated from that expected under HWE (P = 0.001), but the genotype calls were unambiguous, so we did not exclude this SNP from analyses.

Most of the genotypes that we investigated were not associated with endometrial cancer risk (Table 2 and Table 3). However, the presence of the rs4150386 C allele (ERCC5), the rs3730865 C allele (LIG1), the rs2808667 T allele (XPA), and the rs3731127 T allele (XPC) was associated with risk, with respective per-allele ORs, 95% CIs, and P values of 0.68 (0.53–0.87, P = 0.002), 1.46 (1.02–2.10, P = 0.04), 0.71 (0.52–0.97, P = 0.03), and 1.57 (1.13–2.17, P = 0.007; Table 3). The empirical gene-level P values generated from gene set permutation testing (which takes into account the LD structure of the gene and the number of SNPs tested in the gene) were 0.007 for ERCC1, 0.12 for LIG1, 0.08 for XPA, and 0.04 for XPC. The association between ERCC5 rs4150386 and endometrial cancer risk was considerably stronger among obese women (BMI ≥30 kg/m²: per-C-allele OR = 0.41, 95% CI: 0.25–0.68) than women with BMI < 30 kg/m² (per-C-allele OR = 0.85, 95% CI: 0.64–1.14, P_interaction = 0.03). For both LIG1 rs3730865 and XPA rs2808667, associations were stronger among users of high-risk estrogen regimens than never users. For LIG1 rs3730865, per-C-allele ORs and 95% CIs were respectively: 2.98, 1.06–8.37 and 1.27, 0.86–1.89 (P_interaction = 0.10), and for XPA rs2808667, per-T-allele ORs and 95% CIs were respectively: 0.39, 0.20–0.76 and 0.81, 0.57–1.16 (P_interaction = 0.08). The association with XPC rs3731127 did not vary by BMI or use of high-risk estrogen regimens. For several SNPs in XPA and XPC that were not associated with endometrial cancer risk overall, suggestive associations were observed only among women who had used high-risk estrogen regimens (e.g., rs3176748, rs3176658, and rs1800975 in XPA and rs3731151, rs2228000, and rs2733537 in XPC; Table 3). The magnitude of the observed associations did not vary by parity or age (data not shown).

For LIG1 rs3730865, XPA rs2808667, and XPC rs3731127, associations were restricted to the single haplotype in each respective gene that contained the variant allele (ORs and 95% CIs, respectively: 1.46, 0.99–2.15, P = 0.06; 0.70, 0.51–0.96, P = 0.03; and 1.48, 1.04–2.11, P = 0.03). The 2 haplotypes that contained the minor allele of ERCC5 rs4150386 were also associated with decreased risk (0.75, 0.53–1.05, P = 0.09; and 0.46, 0.22–0.97, P = 0.04; Table 4). In a logistic model including all 4 of these SNPs, results for ERCC5 rs4150386, LIG1 rs3730865, and XPC rs3731127 were unchanged. However, the magnitude of the association between the XPA rs2808667 T allele and endometrial cancer risk was closer to the null than the unadjusted result (data not shown). We therefore decided to exclude this SNP from the combined genotype analysis. In the combined genotype analysis, carriage of increasing numbers of “risk” genotypes was associated with increasing risk. Compared with carrying no “risk” genotypes, ORs and 95% CIs for 1, 2, and all 3 “risk” genotypes were respectively: 1.60 (1.20–2.13), 2.35 (1.59–3.46), and 6.54 (1.35–31.81; Table 5).

We observed associations between some polymorphisms in XPC, XPA, ERCC5, and LIG1 in the NER pathway and endometrial cancer risk. We previously reported that women who carried the minor alleles in XPC, both rs2228000 and rs2228001, or XPA rs1800975 had a decreased risk of endometrial cancer (10). In our larger, combined study population, we observed a borderline decreased risk associated with carrying the XPC rs2228000 C allele and no association with the other 2 SNPs. The SNPs that were associated with risk in our present study (XPC rs3731127 and XPA rs2808667) were not in LD with XPA rs1800975 or XPC rs2228000. The 4 SNPs that were associated with endometrial cancer risk in our study are intronic. On the basis of FastSNP bioinformatics data (24), ERCC5 rs4150386 and LIG1 rs3730865 may be intronic enhancers but XPA rs2808667 and XPC rs3731127 have no known function.

To evaluate the degree to which these findings were robust, we conducted gene set analyses for each gene and calculated FPRP (23) for each of the 4 SNPs. The gene set analysis P values were lower than 0.05 only for ERCC5 and XPC, which lends support to the observed associations with SNPs in those genes. With respect to FPRP, associations for the ERCC5 and XPC SNPs were noteworthy with a prior of 0.01 (defined as low) and above, and the LIG1 SNP was noteworthy with a prior of 0.1 (defined as moderate) and above. The results for the XPA SNP were not noteworthy at a reasonable prior. Therefore, the results for the ERCC5 and XPC SNPs may be slightly less likely that the others to be false positives.

DNA repair pathways are complicated, with many redundancies and interactions between proteins. It is biologically plausible that in a system with multiple redundancies, cumulative perturbations in component proteins might be necessary before a difference in risk is observed. It is therefore of particular interest to examine combined genotypes. Because the SNPs that were associated with endometrial cancer risk in our study have low minor allele frequencies, we combined heterozygous and homozygous minor allele genotype categories, and it is possible that this model is not correct for all of the SNPs. Nonetheless, we observed increasing risk associated with increasing numbers of “risk” genotypes carried in XPC, ERCC5, and LIG1. Only 8 cases and 2 controls carried all 3 “risk” genotypes. While we observed a particularly high risk of endometrial cancer among women in this subgroup, we view this observation as no more than suggestive and it requires further exploration in other studies.

With the exception of a single study that, like us, observed no association between ERCC1 rs11615 and endometrial cancer risk (25), to our knowledge, there are no other studies of NER genotypes and endometrial cancer risk to date. Many of the genes and SNPs included in our study have been examined in studies of cancers of the lung, head and neck, prostate, bladder, colon/rectum, breast, and esophagus, as well as glioma and melanoma (comprehensive reviews of most cancer types include Goode and colleagues, ref. 26; Neumann and colleagues, ref. 27; and Vineis and colleagues, ref. 28). While a recessive association between ERCC2 (XPD) rs13181 and lung cancer risk appears to be validated (28), no clear picture has emerged with respect to patterns of associations with NER genes and SNPs across cancer types.

Our study has several limitations. We attempted to characterize the NER genes (while also including “functional” SNPs), but we still had limited SNP coverage for some of the genes because we used an r² value of 0.64 to select tagSNPs. For example, if we had used an r² cutoff point of 0.8 to select tagSNPs, we would have needed an additional 10 SNPs in ERCC4, 9 in ERCC1, 7 in ERCC5, and 5 in ERCC2, but for XPC and LIG1, only a single additional SNP would have been needed and for XPA, the coverage would have been identical. We selected our SNPs using data from sequenced individuals from the Environmental Genome Project (EGP) prior to the availability of HapMap data. Because many of the EGP SNPs were not genotyped in HapMap, it is difficult to compare SNP coverage with HapMap data. Also, while we were not able to obtain a DNA sample from all women who were interviewed for our studies, the distributions of characteristics (e.g., age, parity, oral contraceptive use, smoking, and hormone use) are similar between women who did and did not provide a blood sample. Because of the small number of Hispanic and nonwhite women in the study (56 cases and 59 controls), we were not able to assess their risk associated with the genotypes. Finally, small numbers of women had ever taken a high-risk estrogen regimen, limiting our ability to see differences by this exposure.

While we observed suggestive associations between SNPs in XPC, XPA, ERCC5, and LIG1, because of the large number of possible associations examined in the NER pathways, there is the potential for false-positive findings, both overall and in subgroups. Broader exploration of SNP associations in NER genes using data from existing (29) and pending genome-wide association studies may help to clarify whether genetic variation in NER is associated with endometrial cancer risk. Conclusions about the influence of these genotypes on risk of endometrial cancer must await the results of additional studies and pooled analyses.

No potential conflicts of interest were disclosed.

The authors thank Dr. Kathleen Malone (Fred Hutchinson Cancer Research Center) for facilitating the use of the CARE data (N01 HD 2 3166). They also thank the participants in our series of endometrial cancer studies, as well as the participants in the National Institute of Child Health and Development CARE study.

This work was supported by grants R01 CA 105212, R01 CA 87538, R01 CA 75977, R03 CA 80636, N01 HD 2 3166, R35 CA 39779, K05 CA 92002, and funds from the Fred Hutchinson Cancer Research Center.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.