DNA Methylation in Benign Breast Epithelium in Relation to Age and Breast Cancer Risk

Genes required for cell cycle regulation, differentiation, and apoptosis are frequently silenced in breast cancer by promoter region hypermethylation. Genes that are frequently and intensely methylated in breast cancer are also methylated, albeit at a lower frequency and intensity, in benign breast epithelium (1). We have shown previously that TSG methylation, particularly methylation of RASSF1A, occurs more frequently in benign breast epithelium from women at high risk for breast cancer than women at lower risk (2).

In vitro studies of benign breast cells escaping senescence suggest that CpG methylation is acquired gradually over many cell cycles (3). During DNA replication, DNA methylation patterns are copied from the parent strand to the daughter strand by the enzyme DNA methyltransferase (4). Methylated CpG patterns are transmitted to progeny with greater fidelity than unmethylated CpG patterns resulting in de novo CpG island methylation during cell division (5). CpG island methylation does not occur randomly but extends from previously methylated regions into unmethylated regions (6). Of note, the normally unmethylated regulatory region of the RASSF1A promoter is flanked by CpG islands that are densely methylated in normal breast epithelium (7).

DNA methylation in benign epithelium has been proposed as a somatic epigenetic molecular clock recording the number of lifetime cell cycles. This concept is well illustrated by studies of CSX and MYOD methylation in small intestinal crypt stem cells (8). Indeed, since Issa et al. initially reported that estrogen receptor methylation increases in colonic mucosa with increasing age (9), age-associated methylation has been documented for a variety of genes in a variety of epithelial tissues (8, 10-13). BRCA1 gene methylation has recently been shown to increase with increasing age in benign breast epithelium obtained by palpation-directed fine-needle aspiration (PD-FNA) biopsy (14).

Proliferation, measured at a single time point in benign breast disease, has been associated with breast cancer risk in a case-control study (15). The epigenetic molecular clock hypothesis suggests that promoter-region methylation in benign epithelium may reflect the proliferation history of that epithelium and may, therefore, provide a measure of cancer risk. Ovarian hormones exert a proliferative influence on benign breast epithelium and many established breast cancer risk factors are related to the timing and duration of endogenous and exogenous hormone exposure. Early age at menarche (16), late age at first live birth (17), low parity (18), late age at menopause (19), and use of estrogen and progestins after menopause (20) are each associated with increased breast cancer risk.

To assess age-associated changes in TSG methylation in benign breast epithelium and its relationship to breast cancer risk, benign breast epithelial cells were obtained by PD-FNA from women across a wide age range who were selected to represent a wide range of breast cancer risk. Tumor suppressor gene methylation was measured by quantitative multiplex methylation-specific PCR (QM-MSP) and correlated with a personal history of breast cancer, predicted risk of breast cancer, and specific breast cancer risk factors, including age.

This study was done after approval by the University of Texas Southwestern Medical Center Institutional Review Board and in accordance with an assurance filed with and approved by the Department of Health and Human Services. Informed consent was documented in writing for all subjects. Patients with incident breast cancer and unaffected women over age 18 presenting for breast cancer risk assessment were offered participation regardless of the calculated risk level. Women who had ever used the selective estrogen receptor modifiers tamoxifen or raloxifene were excluded.

Comprehensive breast cancer risk factor information was collected and breast cancer risk was calculated using custom software (BreastC.A.R.E.), which guides a structured, on-screen interview and uses the Gail model (17), Claus model (21), and BRCAPRO (22) to estimate age-specific and cumulative breast cancer probabilities (Table 1). Portions of this software are available in the cancer genetics risk counseling program, CancerGene (http://www4.utsouthwestern.edu/breasthealth/cagene). The Gail model, which is based on age, race, age at menarche, age at first live birth, number or breast biopsies, and family history of breast cancer in first-degree relatives, is well calibrated for estimating breast cancer risk (23, 24). Absolute risk calculated by the Gail model is highly dependent on race and age so high risk was defined as a 5-year Gail model risk > twice age- and race-matched general population risk (25). The distribution of patients by absolute 5-year Gail, Claus, and BRCAPRO risk is shown in Supplementary Figure.⁵

PD-FNA was done as described previously (2). For women with an untreated primary breast cancer, ipsilateral benign cells were obtained from the quadrant opposite the tumor and contralateral samples from the upper outer quadrant. Palpable tumors were also sampled by FNA. For women unaffected by breast cancer, palpably dense breast tissue was sampled near the upper-outer edge of the areola for each breast. Direct smears were prepared and stained using the Papanicolaou method. Cell yields were estimated from direct smears and reported as insufficient cellular material for diagnosis, scant cellularity but sufficient for cytologic classification (≤10 cells), 11 to 99, 100 to 999, or ≥1,000 cells. A Masood score was also calculated for each sample (26). Masood scores ≥15 were classified as “atypical.”

Separate needle passes were collected in PreserveCyt (Cytyc Health) for subsequent DNA extraction (Puregene, Gentra), sodium bisulfite treatment (27), and then QM-MSP (28). A detailed description of the conditions used for the QM-MSP assay has been published previously (29). Rigorous quality assurance standards were enforced and the assay was documented as linear between 0% and 100% of gene copies methylated with a measured sensitivity of 1 methylated gene copy per 100,000 unmethylated copies (29).

We selected five genes that are known to be highly methylated in breast cancer compared with benign breast tissue or that are differentially methylated in benign breast epithelium from high-risk women compared with lower risk women (2). Each gene is known to be regulated by promoter region methylation and is 100% unmethylated in lymphocytes. Primers were specifically chosen to amplify a region of the promoter known to silence gene expression when methylated. Publications supporting our marker and primer selection include cyclin D2 (30, 31), APC (promoter A1; refs. 32, 33), HIN1 (34), RASSF1A (35), and RAR-β2 (36). PCR primer and probe sequences are found in Supplementary Table S1.⁵

Methylation fraction is defined as the proportion of gene copies in a sample that are methylated. A sample was classified as methylation positive for a given gene if the methylation fraction was greater than the 90th percentile for samples from unaffected women at lower risk for breast cancer according to the Gail model. These values were 0.0001 for cyclin D2, 0.007 for APC, 0.031 for HIN1, 0.023 for RASSF1A, and 0.003 for RAR-β2. Composite measures of TSG methylation, based on all five genes, include methylation of ≥2 genes at greater than the aforementioned thresholds, and a methylation sum defined as positive if the sum of all five methylation fractions was greater than the 90th percentile of the sum for the Gail lower risk samples (0.169).

Unsupervised cluster analysis was done using GeneSpring, which uses Pearson's correlation to quantify similarity between groups. The prevalence of TSG methylation in benign samples was assessed in relation to age, race, menopausal status, sample cellularity, and risk level of the breast providing the sample using logistic regression in a series of univariate analyses and then by multivariate analysis that included all covariates generating a univariate P < 0.15. Although the methylation status of one breast did not predict the methylation status of the opposite breast, either for individual genes or for composite measures, each subject was treated as an independent observation for this analysis using Generalized Estimating Equations to account for multiple samples (left and right breasts) for most patients (37, 38). Because at most two samples were taken from the same patient, either compound symmetry or unstructured covariance matrix gave the same results.

A similar analysis including only breasts from women unaffected with breast cancer was done to identify Gail and other risk factors predicting TSG methylation. Risk factors included age, age at menarche, age at first live birth, number of breast biopsies, number of first-degree relatives with breast cancer, age at menopause, body mass index, and waist-to-hip ratio. The statistical analysis was done using SAS version 9.1.3 Service Pack 3.

Benign breast epithelial cells were sampled by PD-FNA in 298 breasts from 164 women. This includes 69 women with a primary breast cancer, 25 unaffected women at high risk for breast cancer according to the Gail model, and 70 unaffected women at lower risk according to the Gail model. FNA samples were also obtained from 62 of the primary breast cancers. Promoter-region methylation of cyclin D2, APC, HIN1, RASSF1A, and RAR-β2 was measured by real-time QM-MSP. QM-MSP was not run on the 46 (15%) samples that contained insufficient cellular material for cytologic classification. Quality standards were not met for ≥4 of the 5 markers in 16 (5%) of the samples and these have been excluded from the analysis. Methylation results for ≥4 of the 5 markers were obtained for benign samples from 236 breasts (79%) from 143 patients (Fig. 1). Eighty-four of these 143 patients (59%) had detectable methylation (methylation fraction > 0.0001) for at least one gene. Methylation results are available for malignant samples from 54 cancers from 54 patients.

Figure 2 shows the distribution of methylation fractions for RASSF1A and APC by sample source. Figure 3 shows the prevalence of increased methylation for each gene by sample source. Increased methylation is defined as a methylation fraction greater than the 90th percentile for lower risk samples. The prevalence of increased methylation was much greater for the cancer samples than the benign samples and, in general, was greater for the benign samples from high-risk women than those from lower-risk women. RASSF1A was the most frequently methylated gene in both malignant (67%) and benign (19%) samples and showed the greatest discrimination between benign samples from women with breast cancer, unaffected Gail high-risk women, and unaffected Gail lower-risk women. RAR-β2 was the least frequently methylated gene in malignant (35%) and benign (11%) samples and did not vary among the benign samples according to the cancer or risk status of the patient providing the sample. Only 14 of 236 (6%) benign samples from 12 women were classified as cytologically atypical. Atypia was diagnosed most commonly in breasts ipsilateral to a breast cancer followed by breasts contralateral to a breast cancer and then breasts from Gail lower risk women. Atypia was not observed in any of the 25 Gail high-risk women.

Quantitative methylation values for 236 benign PD-FNA samples and 54 cellular cancer samples were subjected to unsupervised cluster analysis (Fig. 4). Forty-one percent of the breast cancers belonged to clusters 1 or 3, which were characterized by methylation of cyclin D2 and multiple additional genes. An additional 20% of breast cancers were members of clusters 2 or 4, which included a disproportionate number of benign high-risk samples. Twenty-six percent of the benign high-risk samples (defined as those ipsilateral or contralateral to a breast cancer or those from unaffected women at high risk by the Gail model) belonged to clusters 2 or 4 compared with only 10% of the lower-risk samples [odds ratio (OR) for group membership, 2.6; P = 0.005]. RASSF1A methylation was a central feature of the cancer and high-risk clusters. There were 47 different patients represented among the 53 breasts occurring in clusters 2 and 4. Both the cancer and the ipsilateral benign breast from a single patient occurred in cluster 2 as well as three pairs of contralateral and ipsilateral benign samples from the same cancer patients. Benign samples from both the right and left breasts of one lower-risk patient and one high-risk patient occurred in cluster 4. This is consistent with a separate analysis indicating that methylation profiles in one breast do not correlate with methylation profiles in the opposite breast of the same patient.

Methylation of one gene in a breast predicted methylation of other genes in the same breast. For instance, 38% of breasts with RASSF1A methylation also showed methylation of APC compared with only 9% of breasts when RASSF1A was not methylated (P < 0.0001). In fact, RASSF1A methylation predicted methylation of every other gene (cyclin D2, P = 0.01; HIN1, P < 0.0001; RAR-β2, P < 0.0001). HIN1 methylation also predicted methylation of every other gene (cyclin D2, P = 0.009; APC, P = 0.009; RASSF1A, P < 0.0001; RAR-β2, P < 0.0001).

On univariate analysis, which included all of the benign samples, every marker showed a trend for increased methylation prevalence with increasing age, but this was statistically significant for cyclin D2 only and then only for the middle age tertile [OR, 4.54; 95% confidence interval (95% CI), 1.37-15.07; P = 0.013; Table 2]. On multivariate analysis adjusting for cytology and risk group classification, the prevalence of cyclin D2 methylation was greatest for women in the middle age tertile (OR, 4.33; 95% CI, 1.29-14.57; P = 0.018) but decreased for women in the upper age tertile (OR, 1.15; 95% CI, 0.30-4.33; P = 0.840). To visualize changes in methylation with increasing age, 10-year moving averages were calculated for the methylation fractions for each gene for the 89 unaffected women with methylation results for at least one breast (Fig. 5). Cancer patients were excluded from this analysis because, in general, the cancer patients were older than the unaffected women and showed more methylation in their benign samples producing strong age-related trends that may be due to selection bias. HIN1 methylation increased rapidly between ages 32 and 40 and remained stable thereafter. RASSF1A methylation increased linearly between ages 32 and 55 but began to decline thereafter. Increases in APC methylation lagged behind those of RASSF1A. Cyclin D2 and RARβ2 methylation trended together with a gradual increase between ages 35 and 45 but a gradual decline thereafter.

Race, menopausal status, and sample cellularity were not related to methylation of any individual gene or to composite measures of methylation by univariate analysis (Table 2). Methylation of APC, RASSF1A, and RAR-β2 were each significantly positively correlated with cytologic atypia by univariate analysis as was the composite measure, ≥2 genes. Among the benign samples, the factor most strongly predicting TSG methylation was the risk group of the breast providing the sample. A personal history of breast cancer or a high Gail model risk classification predicted methylation of RASSF1A, ≥2 genes, and a positive methylation sum.

Multivariate analysis that included age, cytology, and risk group classification was also done. The genes most strongly associated with cytologic atypia by multivariate analysis were APC, HIN1, and RASSF1A. The markers segregating most strongly with benign samples ipsilateral or contralateral to a breast cancer or samples from Gail high-risk women were cyclin D2, RASSF1A, ≥2 genes, and the methylation sum (Table 3).

One hundred fifty-one samples from 89 unaffected women with methylation results for at least one breast were assessed by univariate and multivariate analysis to measure the association between methylation and Gail risk classification and to identify risk factors associated with methylation. Gail risk classification (high risk or lower risk) independently predicted TSG methylation in multivariate analysis for ≥2 genes (OR, 3.35; 95% CI, 1.30-8.60; P = 0.012) and for the methylation sum (OR, 3.72; 95% CI, 1.42-9.74; P = 0.007). The genes primarily responsible for this association were APC (OR, 3.14; 95% CI, 1.14-8.62; P = 0.027) and RASSF1A (OR, 5.28; 95% CI, 1.95-14.32; P = 0.001).

A history of benign breast disease requiring biopsy was strongly and independently associated with TSG methylation, particularly methylation of RASSF1A and APC (Fig. 6). Increasing parity was associated with decreased methylation of APC. Family history of breast cancer was not associated with methylation of any of the genes or either of the composite measures by univariate analysis so was not included in the multivariate model (univariate OR for one first-degree relative, methylation sum, 1.18; 95% CI, 0.34-4.08; P = 0.794; univariate OR for ≥2 first-degree relatives = 1.09; 95% CI, 0.18-6.73; P = 0.923). Age at menarche, age at first live birth, age at menopause, BMI, and waist-to-hip ratio did not show any consistent associations with TSG methylation.

Tumor suppressor gene methylation is identifiable in benign breast epithelial cells obtained by PD-FNA in 59% of women. Increased methylation of some genes, especially RASSF1A, is associated with cytologic atypia, predicted breast cancer risk, and a personal history of breast cancer. PD-FNA samples from breasts ipsilateral to a breast cancer showed more methylation and more cytologic atypia than other benign samples. It is possible that PD-FNA may have sampled occult multicentric breast cancer in these women. This seems unlikely as the cytology was not interpreted as malignant and methylation profiles for benign ipsilateral samples did not clustered with the primary cancers, except for one case (Fig. 4). A family history of breast cancer did not predict an increase in the prevalence or level of TSG methylation. Claus and BRCAPRO risk calculations, which are based on family history, did not predict TSG methylation, although Gail model risk calculations did. This supports the notion that familial breast cancer has a different pathogenesis than sporadic breast cancer. Tumor suppressor gene methylation was highly related to a history of benign breast disease requiring biopsy, but it is unclear whether methylation is a cause or a consequence of benign epithelial proliferation. The association between TSG methylation and breast cancer risk has been reported previously (2). The current study validates this finding in an independent sample set.

Methylation of some TSGs was associated with cytologic atypia. However, this analysis may be limited by our low atypia rate. Cytologic atypia was diagnosed in 12 of 143 women (8%) compared with 21% in a previously published series that included only high-risk women (39). Our lower atypia rate is likely due to overrepresentation of lower-risk women in our study sample, evaluation of direct smears rather than cells collected on millipore filters, and more conservative interpretation practices of our cytopathologist.

Tumor suppressor gene methylation in benign breast epithelium varies with age in gene-specific ways. Figure 5A shows 10-year moving averages for methylation fractions. Because 41% of women show no TSG methylation at all, these averages may increase if older women show greater methylation fractions (that is, larger populations of methylated cells) or if a greater proportion of older women show methylation at any level. Figure 5B shows the moving average for the proportion of women with methylation fractions >1%. Taken together, these figures suggest that a few women acquire significant levels of HIN1 methylation in their thirties and that these levels are fairly stable for at least two decades. In comparison, the age-associated pattern of RASSF1A methylation suggests that aging is associated with the acquisition of low levels of methylation by more and more women. Only 14% of women between ages 28 and 37 (median age, 32.5) showed >1% methylation of RASSF1A compared with 39% of women between ages 49 and 58 (median age, 53.5). The increase is linear during this period and the average methylation curve parallels the signal prevalence curve, suggesting that the increases are due to more women acquiring methylation and not to expansion of methylated cell populations in a few women.

Of note, APC methylation shows a similar pattern but lags behind RASSF1A methylation by 5 to 10 years. Methylation of both RASSF1A and APC in the same breast is highly correlated, suggesting that methylation of RASSF1A is an early event that may predispose to methylation of other TSGs. The age distribution for unaffected women (Fig. 5C) limits the inferences that can be made for women beyond the 53- to 62-year decade, but the curves suggest that the proportion of women with RASSF1A methylation begins to decline in the decade between 50 and 59.

These data may support the concept of TSG methylation, especially methylation of RASSF1A, as an epigenetic molecular clock in benign breast epithelium recording the proliferative history of that epithelium. The linear increase in RASSF1A methylation between ages 33 and 54 is consistent with de novo methylation acquired during luteal-phase cell division (40, 41). It is not clear, however, why RASSF1A methylation appears to decline in the fifties. This could be related to apoptosis during menopausal involution or may be an artifact of the smaller number of older unaffected women available for analysis. Menopausal status was not well correlated with methylation of any specific gene or composite measure of methylation, but APC methylation was inversely correlated with parity, suggesting that involutional or differentiating events may affect cell populations with TSG methylation. Mouse studies have shown significant epithelial cell apoptosis after cessation of lactation (42).

In summary, measures of TSG methylation in benign breast epithelium obtained by PD-FNA may have value for breast cancer risk stratification. Some genes, such as RASSF1A, appear to have more value in this regard than other genes. Future work should focus on identifying the optimal marker panel for risk stratification, measuring variability in methylation results over time, and measuring variability in methylation profiles for samples obtained from different parts of the breast. Whether measures of TSG methylation accurately predict breast cancer risk can only be known with certainty by conducting prospective studies with adequate follow-up that include the most relevant markers and employ optimized sampling methodologies. In vitro studies show that promoter region methylation can be acquired during cell division (3, 5); consequently, it is unclear whether TSG methylation in benign breast epithelium is a cause or effect of proliferation. If it is a cause of proliferation, then interventions that reduce or reverse methylation may reduce breast cancer risk. If it is simply an effect of proliferation, then it would retain value as a marker of risk but not as a target for intervention. Well-designed in vitro and in vivo studies that assess proliferation and methylation in benign breast epithelium will be required to resolve this uncertainty.

No potential conflicts of interest were disclosed.

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.