|Home | About | Journals | Submit | Contact Us | Français|
The relevance of progesterone to breast carcinogenesis is highlighted by evidence indicating that use of combined estrogen-progesterone therapy (EPT) is more strongly related to breast cancer risk than is use of unopposed estrogen therapy. However, few investigators have assessed how genetic variation in progesterone-related genes modifies the effect of EPT on risk. In an analysis combining data from 2 population-based case-control studies of postmenopausal breast cancer (1,296 cases and 1,055 controls) conducted in Washington State in 1997–1999 and 2000–2004, the authors evaluated how 51 single nucleotide polymorphisms in 7 progesterone-related genes (AKR1C1, AKR1C2, AKR1C3, CYP3A4, SRD5A1, SRD5A2, and PGR) influenced breast cancer risk. There was no appreciable association with breast cancer risk overall for any single nucleotide polymorphism. For rs2854482 in AKR1C2, carrying 1 or 2 A alleles was associated with a 2.0-fold increased breast cancer risk in EPT users (95% confidence interval: 1.0, 4.0) but not in never users (Pheterogeneity = 0.03). For rs12387 in AKR1C3, the presence of 1 or 2 G alleles was associated with a 1.5-fold increased risk among EPT users (95% confidence interval: 1.1, 2.2) but not in never users (Pheterogeneity = 0.02). Interpretation of these subgroup associations must await the results of similar studies conducted in other populations.
Combined estrogen-progesterone therapy (EPT) is a well-established risk factor for breast cancer (1–5). Multiple observational studies document 1.6- to 3.0-fold elevated risks of breast cancer associated with current, long-term use of EPT (4, 6–11). In the Women's Health Initiative randomized trial, investigators observed a 1.3-fold (95% confidence interval (CI): 1.00, 1.59) increased risk of breast cancer associated with EPT use after a mean follow-up period of 5.2 years (2, 12). In contrast, use of unopposed estrogen therapy was not related to breast cancer risk in the Women's Health Initiative trial, which is consistent with several recent observational studies (6, 9, 13–15).
The mechanisms by which EPT, particularly progestogen, influences the risk of breast cancer remain unclear. With progesterone appearing to exhibit both anticarcinogenic and procarcinogenic qualities in the female body (16–21), there is a potential for mechanisms involved in this duality to be involved in breast carcinogenesis. First, the progesterone receptor (PGR) gene codes for 2 PGR isoforms, PGR-A and PGR-B, which have opposing transcriptional targets (17). Additionally, several enzymes involved in the prereceptor metabolism of progesterone, including the enzymes steroid 5-α reductase (SRD5A) and aldo-keto reductase 1C (AKR1C), metabolize progesterone into the 5α-pregnanes and 4-pregnenes, respectively (16). These metabolites have opposing effects within breast tissue: The 5α-pregnanes have been observed to increase cellular proliferation and detachment, while the 4-pregnenes are associated with weak activation of PGR (16, 22–24). Lastly, the cytochrome P-450 3A4 (CYP3A4) enzyme is included in this pathway because the vast majority of EPT used in the United States contains a synthetic progestogen, medroxyprogesterone acetate, which is predominantly metabolized in the liver by the CYP3A4 enzyme (25, 26).
Thus, single nucleotide polymorphisms (SNPs) in the AKR1C1, AKR1C2, AKR1C3, CYP3A4, PGR, SRD5A1, and SRD5A2 genes could influence progesterone's regulation and carcinogenic potential. On the basis of evidence accumulated to date, we hypothesized that the presence of 1 or more specific alleles in polymorphic genes that regulate PGR or lie within the progestogen metabolism pathway influence the risk of developing breast cancer and, more specifically, interact with EPT to affect breast cancer risk.
The population for this study was accrued from 2 population-based case-control studies of incident breast cancer carried out in western Washington State. The methods used for the Puget Sound Area Breast Cancer Evaluation (PACE) Study (4) and the Seattle Area Hormone and Reproductive Epidemiology (SHARE) Breast Cancer Study (27) have been described in detail previously. Briefly, this ancillary study was conducted within participants from each study who: 1) were aged 65–79 years when diagnosed with primary invasive breast cancer; 2) were residents of the 3-county Seattle, Washington, metropolitan area at diagnosis; 3) had no previous history of in situ or invasive breast cancer; and 4) had the ability to communicate in English. All cases were ascertained through the Cancer Surveillance System, which is the population-based tumor registry serving western Washington and is part of the Surveillance, Epidemiology, and End Results Program of the National Cancer Institute. For the PACE Study, all breast cancer cases diagnosed in this region during the period of April 1, 1997, through May 31, 1999, that were reported to the Cancer Surveillance System were eligible for this study. For the SHARE Study, all cases of invasive lobular carcinoma diagnosed in this region during the period January 1, 2000, through March 31, 2004, and reported to the Cancer Surveillance System were eligible for this study, along with age-matched randomly selected invasive ductal carcinoma cases. Overall, 1,450 (81.5%) of the 1,780 eligible cases identified through the Cancer Surveillance System were interviewed. Of the 1,450 cases interviewed, 1,347 (92.9%) provided a blood sample, of which 1,296 were available for genotyping.
The controls were women of all races between the ages of 65 and 79 years with no previous history of in situ or invasive breast cancer who were randomly selected from Social Security rolls (PACE Study) or through random digit dialing (SHARE Study). The controls were frequency-matched to the cases by 5-year age group. Of the 1,634 eligible controls, 1,218 (74.5%) were interviewed. Of the 1,218 controls interviewed, 1,084 (89.0%) provided blood samples, of which 1,055 were available for genotyping.
Data collection was performed in an identical manner for cases and controls in both studies. For each participant, upon the attainment of informed consent, a trained study interviewer conducted a structured, in-person interview on established and suspected breast cancer risk factors, including: demographic characteristics, reproductive history, menstrual history, hormonal contraception history, medical history, use of certain medications, weight and height history, lifestyle factors, family history of cancer, and menopausal hormone therapy. The questions on hormone therapy sought information on lifetime use, including drug name(s), starting and stopping dates of use, separate and combined uses of estrogen and progestin, strength, and monthly pattern of pill use. All interview questions were limited to events occurring before each participant's reference date (diagnosis date for cases). Reference dates were set for controls in a distribution similar to cases’ diagnosis dates. Data on tumor characteristics were obtained from the Cancer Surveillance System.
The protocols for this study and the 2 parent studies were approved by the Fred Hutchinson Cancer Research Center's institutional review board.
DNA was extracted from buffy coats using phenol chloroform. Using the Genome Variation Server (http://gvs.gs.washington.edu/GVS/), tagSNPs were selected for each gene, with the reference group being the CEPH (Centre d'Etude du Polymorphisme Humain) subjects from the Utah (CEU) HapMap population (28). Seventy SNPs were selected to be genotyped; 66 of those SNPs were tagSNPs, and 4 SNPs were nonsynonymous (rs3740753 and rs2854482) or synonymous (rs3207909 and rs3736316) SNPs that did not tag other SNPs.
Genotyping was performed using Sequenom's iPLEX Gold assay (Sequenom, Inc., San Diego, California) by the Translational Genomics Research Institute (Phoenix, Arizona). Polymerase chain reaction and extension primers for these SNPs were designed using MassARRAY Assay Design 3.0 software (Sequenom, Inc.) and are available upon request. Polymerase chain reaction amplification and single base extension reactions were performed according to the manufacturer's instructions. Extension product sizes were determined by mass spectrometry using Sequenom's Compact MALDI-TOF (matrix-assisted laser desorption/ionization time-of-flight) mass spectrometer. The resulting mass spectra were converted to genotype data using SpectroTYPER-RT software (Sequenom, Inc.). Genotype calls were classified according to the assay status values; these values indicate the level of confidence associated with each genotype call. The assay status values ranged from conservative (high confidence) to aggressive (low confidence) for genotypes included in the analysis.
Negative and positive controls were included on each genotyping plate. Genotype data from 30 CEPH trios (Coriell Cell Repository, Camden, New Jersey) were used to confirm the reliability and reproducibility of the genotyping. One SNP (rs1937883) was removed from analysis because Mendelian inheritance errors were detected. Four SNPs were removed because genotypes were called as monomorphic (rs1824126, rs9804392, rs11252946, and rs4646437), and 4 SNPs were removed because more than 15% of the genotyping data were missing (rs4344395, rs9329316, rs39848, and rs16877779). Hardy-Weinberg equilibrium was evaluated among controls in our study population, and 10 SNPs that were not in Hardy-Weinberg equilibrium (rs1379130, rs4478959, rs635984, rs555653, rs613120, rs2904802, rs10904387, rs2854466, rs4242785, and rs3207909) were excluded from analysis. Intra- and interplate replicates chosen at random from the study population at a rate of 10% (n = 222) were included at randomly assigned locations on all plates and in all batches, with laboratory staff blinded. Among the SNPs that met these quality control criteria (n = 51), the concordance rates between replicates for each SNP ranged from 93.6% to 100.0%, with a 99.5% concordance on average. The average percentage of missing genotype data per SNP among the 51 SNPs was 4.1.
Testing for deviation from Hardy-Weinberg equilibrium was performed using a χ2 goodness-of-fit test. The Mantel-Haenszel χ2 test was used in bivariate analyses.
Odds ratios and corresponding 95% confidence intervals were calculated using logistic regression. The homozygous wild-type genotype, as determined by the more common of the homozygous genotypes, served as the reference category, with the heterozygous genotype and homozygous variant genotypes being collapsed into 1 category. Variables which altered the magnitude of a risk estimate by 10% and were not likely to be consequences of genotype were included in the final model. We considered body mass index, race/ethnicity, and reproductive factors, including age at menopause and cause of menopause, as potentially confounding factors, but none of these factors met our criteria for inclusion in the model. Additionally, we included study (because of the differences in sampling methods for the 2 studies comprising these analyses) and age at diagnosis in the model.
The Cochran-Armitage test for trend was used to evaluate trends in models containing variables with multiple levels. Heterogeneity was evaluated using the likelihood ratio test for analyses including 2 strata and Wald's test statistic for analyses including more than 2 strata.
We estimated haplotype frequencies using Phase, version 2.1 (29). We evaluated haplotypes using logistic regression while taking into account phase ambiguity; the most common haplotype served as the reference category.
We evaluated the potential for false-positive results due to multiple testing of 51 gene-EPT interactions using the false-positive reporting probability. We preset the criterion for false-positive reporting probability at 50%, based on the guidelines set forth by Wacholder et al. (30) for initial investigation of SNPs within studies of common cancers. We calculated the false-positive reporting probability for prior probabilities ranging from 0.1 to 0.01.
Compared with controls, cases were more likely to have an older age at menopause, to have a higher body mass index, and to be users of EPT (Table 1).
Similar to findings previously described in these study populations (4, 27), ever use of EPT was associated with an increased risk of breast cancer (odds ratio (OR) = 1.4, 95% CI: 1.1, 1.8; data not shown). With respect to the use of unopposed estrogen therapy, ever use was not associated with breast cancer risk (OR = 1.1, 95% CI: 0.9, 1.3; data not shown).
Fifty-one SNPs in 7 genes were included in our assessment of breast cancer risk. No genotypes of these single SNPs were observed to be more than weakly associated with breast cancer risk (see Web Table 1, which is posted on the Journal’s Web site (http://aje.oupjournals.org/)). No differences in association with SNPs by histologic type or race/ethnicity were observed to any appreciable degree in the main analyses (data not shown). In addition, we did not observe associations between breast cancer and haplotypes within any of these genes (data not shown).
We investigated the potential for gene-EPT interactions for each of the 51 SNPs. The risk of breast cancer varied according to EPT use for 2 SNPs in the AKR1C gene family (Table 2). For rs2854482 in AKR1C2, an increased risk of breast cancer among carriers of the A allele was limited to ever users of EPT (OR = 2.0 (95% CI: 1.0, 4.0); in never users, OR = 1.0 (95% CI: 0.7, 1.4); Pheterogeneity = 0.03). For rs12387 in AKR1C3, an increased risk of breast cancer among carriers of the G allele was also limited to ever users of EPT (OR = 1.5 (95% CI: 1.1, 2.2); in never users, OR = 1.0 (95% CI: 0.8, 1.2); Pheterogeneity = 0.02). Given the possibility of obtaining false-positive results due to multiple testing, we calculated false-positive reporting probabilities for these interactions using prior probabilities of 0.1, 0.05, and 0.01. For rs2854482, we observed false-positive reporting probabilities equal to 0.39, 0.58, and 0.88 at these 3 prior probability levels, respectively; and for rs12387, we observed false-positive reporting probabilities equal to 0.31, 0.49, and 0.83, respectively. These findings indicated that the false-positive reporting probability was within the 50% criterion set a priori for rs2854482 at a prior probability of 0.1 but not lower, and for rs12387 at a prior probability of 0.05 but not at 0.01.
Also included in the 51 SNPs examined for potential interaction with EPT were several for which investigations of potential gene-EPT interactions have been reported in the literature, namely PGR +331 (rs10895068), PGR Val660Leu (rs1042838), and CYP3A4*1B (rs2740574) (31–33). We observed no evidence of an altered risk of breast cancer in women with 1 or more G alleles in CYP3A4*1B or in women with 1 or more T alleles in PGR Val660Leu, regardless of EPT use. While there was a suggestion of an increased risk in women with at least 1 A allele for PGR +331 among those who received EPT (OR = 1.5, 95% CI: 0.8, 2.6), the 95% confidence interval included 1 in this instance.
Additionally, we conducted a subset analysis to repeat the above analyses with results stratified by histologic type, but we observed no evidence for a difference in breast cancer risk by histologic type (Table 3). Our sample size was somewhat constrained for the estimation of odds ratios and evaluation of gene-EPT interactions for histology-specific outcomes. Results did not differ by estrogen receptor/progesterone receptor status or by stage.
This study was designed to investigate contributions of genetic variation in the PGR and progesterone metabolism pathway to breast cancer risk. Our study observed little evidence of such a contribution, though there was a suggestion of an increased risk in women with certain AKR1C2 and AKR1C3 (and possibly PGR) genotypes if they had received EPT.
To our knowledge, the AKR1C gene family has not previously been studied with respect to breast cancer risk in epidemiologic research. Thus far, Ji et al. (24) have described a putative role in breast cancer based on the AKR1C enzyme's metabolizing progesterone into a 4-pregnene metabolite, which is believed to be a weaker activator of PGR. Of the 2 AKR1C SNPs observed to have an association with breast cancer risk among EPT users, the AKR1C2 SNP rs2854482 represents a nonsynonymous change at position 46 (tyrosine to phenylalanine) of the protein, while rs12387 in AKR1C3 represents a synonymous mutation in amino acid 104. In a study evaluating the functionality of the rs2854482 variant of the AKR1C2 enzyme with respect to dihydrotestosterone, the variant enzyme was observed to have significantly decreased maximal velocity Vmax and intrinsic clearance of dihydrotestosterone in comparison with wild-type enzymes (34). To our knowledge, no studies have investigated the effect of this variant on progestogen metabolism.
We evaluated the probability that these findings represented false-positive associations. While these associations were not statistically significant after adjustment for multiple comparisons using standard methods (the Bonferroni method and the method of Benjamini and Hochberg (35)), these methods may be overly conservative. Thus, we also calculated the false-positive reporting probability because of the ability of this method to incorporate information on the a priori likelihood of finding an association, which is particularly salient for SNPs that may have a functional impact on the protein. The likelihood of these results’ representing false-positive findings was calculated to be 39% and 31% for the SNPs in AKR1C2 and AKR1C3, respectively, at a prior probability of 0.1. In other words, were we to assume that the probability of an association between the variant allele in AKR1C2 was 10% or more, the probability of this finding's being a false-positive association is less than 40%. Based on the criterion set a priori at 50%, these results fall within the range considered “noteworthy.” Because our study is the first to report on the relation of breast cancer risk to the combined influence of EPT and these SNPs (and, as a result, our a priori criterion for determining noteworthiness was set higher than in confirmatory studies), we recommend caution in the interpretation of these findings.
The PGR gene codes for the 2 PGR isoforms PGR-A and PGR-B (36). Multiple studies have evaluated SNPs in the PGR gene in relation to breast cancer risk. With respect to the +331 allele, located in the promoter region of the gene, our results are consistent with 3 of the 4 previous studies in finding no association between this SNP and breast cancer risk (32, 37–39). With the Val660Leu allele, our findings are consistent with those of a recent meta-analysis that found no increased risk associated with carrying the T allele (OR = 1.05, 95% CI: 0.8, 1.5) (39).
A few epidemiologic studies have investigated the potential for breast cancer risk associated with genes in this pathway to be modified by EPT use. Rebbeck et al. (31) observed a suggestion of EPT-gene interaction for the +331A allele (P = 0.19), with the greatest breast cancer risk associated with carrying the A allele being observed among EPT users of at least 3 years, while 2 other studies found no evidence of an interaction with EPT (32, 33). Although we observed a suggestion of an increased risk associated with carrying the +331A allele among ever users, further investigation of a potential gene-environment interaction by duration of use did not yield any stronger evidence of an interaction. Carriers of the +331A allele who had used EPT for at least 3 years were at a modestly greater risk of breast cancer (OR = 1.5, 95% CI: 0.7, 3.2) than EPT users of less than 3 years (OR = 1.3, 95% CI: 0.4, 3.8), and no association was observed among EPT nonusers (Pheterogeneity = 0.27). Taken together, the lack of an association between +331A and overall breast cancer risk and the mixed findings on interactions with EPT suggest that this allele plays, at most, a modest role in breast cancer risk.
Diergaarde (33) observed a 3.1-fold increased risk of breast cancer associated with carrying the T allele in PGR Val660Leu among EPT users of at least 10 years and a 1.9-fold increased risk among EPT users of 1–10 years, while no increased risk was seen in never users (Pheterogeneity = 0.01). We did not observe the Val660Leu allele to interact with EPT overall or by duration of use (for EPT use of ≥10 years, OR = 1.2 (95% CI: 0.7, 2.0); for EPT use of 1–10 years, OR = 1.1 (95% CI: 0.6, 1.9); Pheterogeneity = 0.59).
In a study of postmenopausal women, Rebbeck et al. (31) also observed a borderline statistically significant interaction between CYP3A4*1B and EPT in breast cancer (P = 0.08); carriers of the A allele in never EPT users were at a 2-fold increased risk of breast cancer (OR = 2.08, 95% CI: 1.03, 4.20), users of less than 3 years were at a 57% reduced risk (OR = 0.43, 95% CI: 0.13, 1.45), and users of 3 or more years were at no increased risk (OR = 1.1, 95% CI: 0.50, 2.65). We did not find evidence that breast cancer risk was related to CYP3A4*1B genotype overall, by ever use of EPT, or by duration of EPT use.
In reviewing our study's findings, it is important to consider our study's limitations. First, with a portion of eligible cases (18.5% for both studies combined) and controls (25.5%) not participating in the parent studies, it is possible that our participants were not entirely representative of the general population. Another possible source of selection bias could arise from the portion of interviewed cases (7.1%) and controls (11.0%) who did not donate a blood specimen. Comparison of the subjects who donated blood with all subjects interviewed is reassuring in that there were no discernible differences in risk-factor distributions or in terms of stage and other disease features.
Secondly, for analyses of genotype according to EPT use, which utilized interview data, the ability of women to recall past exposures is a potential limitation. However, recall aids were employed in the parent studies, including a lifetime calendar and colored photographs of medications. As a check on the quality of recall in PACE participants, we previously compared self-reported data on several categories of medications with pharmacy data and found that recall did not differ by case/control status (40). Thus, any recall error would have been nondifferential and would have biased the results towards the null.
Thirdly, genotyping errors could have introduced bias into our results. Our laboratory protocol included a quality assurance check, and, as reported above, we excluded any SNPs for which there was evidence of genotyping errors. Further, among the 51 SNPs for which results are presented here, the correlation in replicate samples was quite high and the amount of missing data was reasonably low. Additionally, we investigated the potential impact of laboratory errors by removing genotypes with aggressive genotype calls and did not find a difference in the risk estimates. Any misclassification due to genotyping is likely to have been nondifferential, since laboratory staff were blinded to case/control status.
Lastly, breast cancer is a heterogeneous disease, and the ability to detect an association in breast cancer overall can be hindered by potential differences in effects by subtype (i.e., histologic types). We attempted to address this by repeating our analyses within lobular and ductal carcinomas, the histologic subtypes which comprised 91.4% of the breast cancers in our data set. Because of the oversampling of invasive lobular carcinomas in the SHARE Study, our study included a relatively large proportion and absolute number of invasive lobular carcinomas; however, we were still somewhat constrained by the small number of invasive lobular carcinomas.
Because we combined data from 2 studies with different sampling methods (and because invasive lobular carcinomas constitute a distinct histologic group with a potentially unique etiologic profile), there is also a possibility that the breast cancer cases in our study were not representative of breast cancer in the population. However, as Table 3 shows, we observed no difference in the risk estimates according to histologic type. Additionally, the estimates observed within the PACE Study, which recruited cases without any sampling restrictions, did not differ to any appreciable degree from the results presented above (although the PACE-specific analyses lacked the precision of the combined analysis because of the smaller sample size). Because we selected only a subset of cases from the SHARE Study, we lacked the precision to reliably estimate odds ratios within the SHARE Study alone.
The results of our study suggest that certain genotypes in polymorphisms in the AKR1C2 and AKR1C3 genes could increase the risk of breast cancer among women who have used EPT, potentially through their influence on progesterone metabolism. Since this is, to our knowledge, the first study to report on potential interactions in these particular SNPs, our results should be interpreted with caution.
Author affiliations: Department of Epidemiology, School of Public Health, University of Washington, Seattle, Washington (Kerryn W. Reding, Christopher I. Li, Noel S. Weiss, Chu Chen, Christopher S. Carlson, Kenneth E. Thummel, Kathleen E. Malone); Public Health Sciences Division, Fred Hutchinson Cancer Research Center, Seattle, Washington (Kerryn W. Reding, Christopher I. Li, Noel S. Weiss, Chu Chen, Christopher S. Carlson, Janet R. Daling, Kathleen E. Malone); and Population Genetics Research Lab, Translational Genomics Research Institute, Phoenix, Arizona (David Duggan).
This work was supported by grants R01 CA72787 and R01 CA85913 (data and blood specimen collection) and R03 CA119746 (genotyping) from the National Cancer Institute. K. W. R. was supported by a Cancer Epidemiology and Biostatistics training grant (T32 CA09168) from the National Institutes of Health and a Department of Defense Breast Cancer Research Program predoctoral training grant (06-1-0312). K. E. M. was supported by National Cancer Institute grant R01 CA098858.
Conflict of interest: none declared.