Association of progesterone receptor gene (PGR) variants and breast cancer risk in African American women.

Baseline characteristics of cases and controls are presented in Table 1 . In both the EA and AA cohorts, the only characteristic that differed significantly between cases and controls was parity, with cases having a much higher proportion of nulliparous women. Measures of environmental hormonal exposure, including parity, age at first term pregnancy, age at menarche, age at menopause, and BMI) were all considered as potential confounders. None of these variables were associated with any of our exposures (PGR variants) or our outcome (breast cancer), and thus, by definition, were not confounders. Because there is ample data demonstrating that combined estrogen/progesterone-containing hormone replacement therapy (chrt) increases risk of breast cancer (compared to estrogen-only hormone replacement therapy), we decided a priori to adjust for CHRT in our models. CHRT use was defined as use of combined estrogen and progesterone hormone replacement therapy for greater than 3 years. Tables 2a and 2b show the associations between each PGR variant and risk of breast cancer in the study population. In EAs, no significant main effects were observed in any of the PGR variants studied. In AAs, two SNPs ( rs590688 and rs10895054 ) were found to be statistically significantly associated with breast cancer. For rs590688 , each addition of the variant (C allele) was found to be protective compared to the wild type allele (G allele) (OR 0.56 (95% CI:0.39–0.82), p value 0.003, corrected p value 0.03). For rs10895054 , each addition of the variant (T allele) increased the risk of breast cancer compared to the wild type allele (A) (OR 2.9 (95% CI: 1.47–6.02), p value 0.002, corrected p value 0.04). To explore further why these two variants ( rs590688 and rs10895054 ) were found to be significantly associated with breast cancer in the AA population but not in the EA population, we investigated whether environmental exposures modified the effect of the gene variant on breast cancer risk. Because the number of homozygous variants was small, we combined heterozygotes and variant homozygotes into the logistic regression analysis and used wildtype homozygotes as the reference category. Of note, we did not correct for multiple testing for this portion of the analysis. In EAs, the presence of the rs590688 variant (C allele) decreases breast cancer risk (see Table 3 ), however this finding is not significant. However, the variant is associated with increased breast cancer risk in the presence of nulliparity (OR 3.07 (95% CI: 1.25–7.57, p value 0.014) or a BMI ≥ 30.0 (OR 5.7 (95% CI: 1.3–24, p value 0.018). In EAs with the rs590688 variant, CHRT use and menarche ≥ 14 years were associated with a decreased risk of breast cancer (OR 0.33 (95% CI: 0.14–0.755, p value 0.003) and OR 0.297 (95% CI: 0.11–0.799, p value 0.009) respectively). In AAs with the rs590688 variant, menarche ≥ 14 years was also associated with a decreased risk of breast cancer (OR 0.177 (95% CI: 0.37–0.857, p value 0.031)) (See Table 4 ). No other statistically significant interactions were observed in AAs with the rs590688 variant, possibly due to insufficient power in this subgroup. In EAs, the presence of the rs10895054 variant increases breast cancer risk, albeit not significantly so. However, for EAs with the rs10895054 variant, nulliparity and a BMI of ≥30.0 was associated with a decreased risk of developing breast cancer (OR 0.29 (95% CI: 0.125–0.67, p value 0.004) and OR 0.17 (95% CI: 0.05–0.67, p value 0.01) respectively) (See Table 4 ). There were no statistically significant interactions seen in the AA subgroup, again, possibly due to limited power in this subgroup. Because both the EA and AA subgroups have relatively small sample sizes, these results should be interpreted with caution. For EAs with either variant ( rs590688 or rs10895054 ), the presence of nulliparity alters the association with breast cancer, so that it is in the opposite direction of the variant’s main effect. One could conclude that the main effect of the variant is ‘diluted’ in EAs by the increased proportion of nulliparious women in that subgroup (see Table 1 ). However, the presence of a BMI ≥30.0, also alters both variant’s association with breast cancer so that it is in the opposite direction of the main effect, and BMI ≥30.0 is more common in AAs (see Table 1 ). Thus, the discrepancy in the variants effect in the EA population and AA population is currently unexplained by interactions with environmental factors that were measured in this study. In AA, three haplotype blocks (all containing rs590688 ) were found to be significantly associated with breast cancer risk (see Table 4 ). However, it appears that the effect of each of these haplotypes on our outcome is being driven by the main effect of the individual SNP, rs590688 that happens to be in strong linkage disequilibrium with rs601040 and rs537681 (See Figure 1 ). The variant, rs10895054 , was not found to be in strong linkage disequilibrium with any other SNP’s investigated in this study. Again, we did not correct for multiple testing in this portion of the analysis.

We report that two PGR tagging SNPs ( rs590688 and rs10895054) are significantly associated with breast cancer in our AA cohort. Why this association was found in AAs, but not EAs is unclear. Confounding by ethnicity (population stratification bias) refers to false-positive associations found in racially heterogeneous populations where both the disease and allele frequency vary by race. To avoid the potential for confounding by ethnicity, we stratified our analysis based on race. However, AAs are a highly admixed population, composed of both African and EA ancestry. Breast Cancer incidence varies by race, with breast cancer being less common in AA women, then in women of European descent [ 37 ]. The control allele frequencies of rs590688 (EAs: C 0.5054, G 0.4946; AAs C 0.5209, G 0.4791) and rs10895054 (EAs: A 0.8337, T 0.1663; AAs: A 0.9725, T0.0275) were similar in AAs and EAs. Because a requirement for population stratification is that differences in allele frequencies exist across populations [ 38 ], confounding by ethnicity is unlikely to explain the different results obtained in each race. Another possible explanation for the discrepancy in results is due to gene-environment interactions. In this study, we examined environmental factors related to hormone exposure. We do report a few gene-environment interactions (see Table 4 ), however these results should be interpreted with caution due to our small sample size. Regardless, even if these interactions are real, they do not explain the discrepancy in findings between AAs and EAs. It is possible that the discrepancy in our results may be due the interaction between rs590688 and rs10895054 and environmental or genetic factors that we did not measure in our study. To minimize false-positive associations, we adjusted our p values in our main analysis for multiple comparisons using the Benjamini and Hochberg method [ 35 ]. Despite this correction, we still cannot rule out the possibility that our findings are due to chance alone, and thus need to be confirmed in larger epidemiologic studies. One candidate SNP, rs1042838, was associated with breast cancer in prior studies [ 26 , 27 , 24 ] , but was not found to be significantly associated with breast cancer in our study. In these prior studies, the magnitude of the effect was small (OR < 1.5); our study is only powered to detect odds ratios greater that 1.9. Thus, there may be a real association between rs1042838 and breast cancer, but our study was under powered to detect it in either subgroup. Likewise, there may be associations between other SNPs that we studied and breast cancer, however, if the magnitude of the association was small (OR < 1.9), we were not powered to detect this. Our study is a candidate gene association study, and thus has many of the limitations associated with these studies. Breast cancer is a complex disease and is likely the result of interplay between many genes and the environment, limiting our power to find an association between just one gene and this outcome. In addition, breast cancer is a heterogeneous disease composed of many biological subtypes that behave differently and respond differently to therapy. Different biologic subtypes may have different genetic origins. We limited our study population to women who were 50 to 79 years old, capturing mainly postmenopausal women. Breast cancer in pre-menopausal women is more likely to be hormone receptor negative, behave more aggressively, and is associated with a poorer prognosis when compared to post-menopausal breast cancer. In addition, a recent study looking at gene expression profiling in breast cancers, identified over-expression of 367 gene sets that could differentiate tumors from pre-menopausal and post-menopausal women, indicating that breast cancer in pre-menopausal women may be a distinct entity [ 39 ]. Thus, by limiting our study to post-menopausal, we eliminate some disease heterogeneity, but certainly not all. In conclusion, we report that two SNPs in PGR ( rs590688 and rs10895054) are significantly associated with breast cancer in our AA cohort. This is the first study that reports this association. These findings need to be confirmed in larger epidemiologic studies.

Prolonged exposure to estrogen is clearly implicated in the development of breast cancer. Early menarche, late menopause, reduced parity, later maternal age of first live birth, and obesity are all associated with prolonged estrogen exposure and increased breast cancer risk [ 1 ]. The role of progesterone and progesterone receptors in the development of malignancy is not as clear. Interestingly, the Women’s Health Initiative demonstrated that prolonged exposure to combined hormone replacement therapy including estrogen plus progestin increases a post-menopausal woman’s risk of breast cancer, whereas estrogen-only hormone replacement therapy does not [ 2 , 3 ]. This finding has been reported in several other studies [ 4 – 9 ]. This suggests that progestins (a synthetic form of progesterone) may play a role in breast carcinogenesis. Interestingly, physiologic exposure to progesterone does not appear to increase and may actually reduce breast cancer risk [ 10 , 11 ]. The role of progesterone in breast tissue is not well defined, although it appears that the hormone may stimulate progesterone receptor mediated transcription of genes that both stimulate and inhibit breast tissue differentiation [ 12 ]. Because the progesterone receptor is crucial to the function of progesterone and progestins, it has been hypothesized that a progesterone receptor variant could alter the effect of progesterone on breast cells and thus, influence breast cancer risk. The progesterone receptor is coded by PGR located on chromosome 11q22.1. PGR encodes two progesterone receptor isoforms with different promoters and different translational start sites. This results in two gene products: PRA is 94KDa and acts as a transcriptional repressor; PRB is 114 Kda and acts as a transcriptional activator [ 12 ]. PGR is polymorphic with greater than 800 described SNPs [ 13 ]. Candidate gene association studies have reported associations between PGR variants and various hormonally related diseases such as endometrial and ovarian cancer [ 14 – 19 ], endometriosis [ 20 ], impaired implantation after in vitro fertilization [ 21 ], and recurrent miscarriages [ 22 , 23 ]. Some studies have found an association with PGR variants and breast cancer, but the results have not been consistent [ 24 – 26 , 14 , 27 – 31 , 16 ]. To further explore the role of PGR in steroid hormone-related disease, we conducted a case control study to examine whether variability in PGR is related to breast cancer risk, whether this relationship varies by race and whether environmental factors related to hormone exposure modify the effect of PGR on breast cancer risk.

The Women’s Insights and Shared Experiences (WISE) study is a population based case-control study of postmenopausal women conducted in a 3-county area of metropolitan Philadelphia in which incident breast and endometrial cancer cases were identified through hospitals and validated using the Pennsylvania State Cancer Registry. Frequency-matched controls were identified from the community using random digit dialing. Additional details of the WISE study design can be found in studies by Bunin et al. [ 32 ] and Strom et al. [ 33 ] The source population for this study was Philadelphia County (PA), Delaware County (PA), and Camden County (NJ). Potentially eligible cases were women residing in these counties at the time of diagnosis who were ages 50 to 79 years old and were newly diagnosed with breast cancer between July 1, 1999 and June 30, 2002. Genomic DNA was obtained from buccal swabs as previously described. Additional details of our study design, which included ascertainment of both breast and endometrial cancer cases and matched controls, have been previously reported [ 28 ]. Buccal swabs were obtained by mail from each participant. Extraction of genomic DNA was done using QIAamp 96 DNA Buccal Swab Biorobot kit and done on a 9604 Biorobot (Qiagen, Inc., Valencia, CA). No whole genome amplification was done before genotype analysis. We identified 8 candidate single nucleotide polymorphisms (SNPs) in the PGR gene with a minor allele frequency (MAF) greater than or equal to 5% which were reported in the literature to have some clinical relevance using NCBI Entrez Gene SNP Geneview. The variants selected were rs1042838, rs1042839, rs11571171, rs3740753, rs500760, rs566351, rs660149 , and rs7116336 . In addition to identifying candidate SNPs, we also selected nineteen haplotype tagging SNPs (htSNPs) using SNPbrowser 4.0. htSNPs are representative SNPs in regions of high linkage disequilibrium. The identification and genotyping of a set of tag SNPS allows for the identification of genetic variation within a chromosomal region without genotyping every SNP in that region [ 34 ]. In this study, htSNPs were considered if they had a haplotype R 2 > 95% and a MAF greater than or equal to 5%. The tag SNPs selected were as follows: rs11224561, rs10895054, rs11224580, rs471767, rs492457, rs506487, rs518382, rs538915, rs553272 , rs635984, rs11224580, rs1379131, rs1824126, rs495997, rs507141, rs537681, rs590688, rs601040, and rs653752. Genotyping was performed via Taqman Open Array Platform. A total of 1,364 subjects underwent genotyping. Adequate genotype data could not be obtained from thirty-four (34) participants. Four SNPs ( rs11571171, rs566351, rs11224561, and rs1824126 ) were excluded from the analysis because of high failure rates (>10%). Two additional SNPs ( rs495997 and rs3740753 ) were excluded from the analysis due to significant deviations from Hardy-Weinberg Equilibrium (p <10 −5 ) in the control population. To assess the impact of SNP genotypes on breast cancer risk, we used the homozygous wild type genotype (wt/wt) based on the common allele as the referent category and assessed the relative risk associated with heterozygous and homozygous variant carriages (var/*). We conducted separate analyses for EA and AA women to avoid the potential for confounding based on race (population stratification bias). A study participant’s race was determined by self report. Differences in baseline characteristics between cases and controls were assessed using chi-squared tests for categorical variables and t-tests for continuous variables. In order to assess the association of genetic variants with breast cancer risk, we estimated odds ratios, their corresponding 95% confidence intervals, and associated p-values using logistic regression. We assessed potential confounding due to endogenous or exogenous hormone exposures such as combined hormone replacement use, parity, age at first live birth, age of menopause/menarche, and BMI by including them in the multivariable logistic regression model. We adjusted for multiple comparisons using the false discovery rate method described by Benjamini and Hochberg [ 35 ]. We also explored the possibility of interaction between all genotypes and these hormone exposures. All analyses were performed using STATA (version 11.0, STATA Corporation, College Station, TX We used Haploview to generate haplotype blocks and visualize linkage disequilibrium from our genotype dataset. To investigate haplotype associations, we first used the EM algorithm to estimate haplotype frequencies and then used a generalized linear models framework to obtain haplotype odds ratios using the R programs Haplo.em and Haplo.stats (R version 2.8.0) [ 36 ].