Correlation between CpG methylation profiles and hormone receptor status in breast cancers

Human breast cancer cell lines SKBr3, MDA-MB-435, MDA-MB-468, BT-20, MDA-MB-231, and MCF-7 were maintained in RPMI 1640 medium supplemented with 10% fetal bovine serum. Normal breast epithelial cells, HMEC 231 and HMEC234, were cultured in a 1:1 solution of MCDB 105 and medium 199 with 15% fetal bovine serum and 10 ng/ml epithelial growth factor (Sigma, St Louis, MO, USA), as described elsewhere [21].

We used 90 samples, consisting of paired tissues and associated clinicopathologic data from the Breast Tumor Bank at The University of Texas MD Anderson Cancer Center (Houston, TX, USA). The samples of breast tumors and corresponding adjacent normal-appearing tissues (from tissues located at least 3 cm away from the site at which the tumor was sampled) came from 80 patients who had undergone surgery at The University of Texas MD Anderson Cancer Center in 2004 or 2005 and 10 patients who had been diagnosed with breast cancer between 1995 and 2003. All tissue samples had been fresh-frozen and stored at -80°C. No patient was recruited specifically for this study. All patients gave written informed consent permitting the use of their breast tissue for research at the time specimens were collected. This study was approved by The University of Texas MD Anderson Cancer Center's Institutional Review Board.

All breast specimens were reviewed by experienced pathologists. Slides were subject to immunoperoxidase staining for estrogen receptors (ERs; clone 6F11, Novocastra Laboratories Ltd, Benton Lane, UK) and progesterone receptors (PRs; clone1A6, Novocastra Laboratories Ltd), according to the manufacturer's recommendations. Cancers were considered receptor-positive if > 10% of malignant cells showed nuclear staining. Cancers were classified as HR-positive if the ER and/or PR status was positive. HER-2/neu gene amplification was assessed by interphase fluorescence in situ hybridization (FISH) using the PathVysion HER-2/neu probe kit (Vysis Inc., Downers Grove, IL, USA), according to the manufacturer's recommendations. Levels of the HER-2/neu:CEP17 signal ratio were considered normal if FISH detected ≤ 2.0 copies per cell.

To avoid contamination with normal tissues in the methylation analysis, we isolated breast cancer cells and paired normal breast epithelial cells from tissues by manual microdissection, following the National Cancer Institutes (Bethesda, MA, USA) protocols [22]. In brief, 5–10 μm sections were cut from each archival fresh-frozen tissue block. For each pair of tissues, the presence of tumor cells in malignant tissues and absence of cancer cells in normal tissues were confirmed by histopathologic examination. Frozen tissue sections were fixed with ethanol, stained with H & E, and microdissected using a needle. Each tumor sample contained > 70% tumor cells after microdissection. Genomic DNA was extracted from patient samples, breast cancer cell lines, and normal breast epithelial cells using the Dneasy tissue kit (Qiagen, Valencia, CA, USA). Bisulfite treatment of 1–2 μg of genomic DNA was performed, as previously described [3].

Twelve tumor-suppressor genes were selected for this study. ARHI (CpG I and II), RASSF1A, HIN-1 (SCGB3A1), RARβ2, hMLH1, the 14-3-3 σ gene, RIZ1, p16 (CDKN2A), and the E-cadherin gene (CDH1) were selected on the basis of previous reports of elevated methylation rates in breast cancers [3‐12]. RIL (PDLIM4) [23], CDH13, and NKD2 were identified as hypermethylated genes using genome-wide methylated CpG island amplification (MCA) from multiple tumors (24). Global methylation was estimated by testing methylation levels of LINE1 repetitive elements [25]. We also assessed the methylation of ERα and PGRB genes.

Bisulfite pyrosequencing was used to detect methylation of all 15 genes. Pyrosequencing primers were designed using Assay Design software 1.0 (Biotage, Westborough, MA, USA). For each gene, we selected the CpG island region flanking the transcription start site at the 5'UTR. Two to six CpG sites were studied for each particular CpG island. The primers for pyrosequencing and PCR conditions are listed in Additional file 1. Bisulfite-treated DNA (1 μl) was amplified in 50 μl of reaction mixture, containing primers and 0.2 U of Taq polymerase (New England Biolabs, Ipswich, MA, USA). For the amplification of HIN-1, RARβ2, hMLH1, the 14-3-3 σ gene, RIZ1, the E-cadherin gene, NKD2, and PGRB, we used a universal primer approach [18]. The PCR product was purified and methylation was quantitated using the PSQ HS 96A pyrosequencing system and Pyro gold reagents (Biotage). Methylation data are presented as the percentage of average methylation in all observed CpG sites. To set the controls for pyrosequencing, we used cancer cell lines and normal cells that were consistently positive or negative with stable levels of methylation. In this study, each PCR assay included a positive control (the MDA-MB-231 breast cancer cell line, which is highly methylated in most genes) and a negative control (normal breast epithelial cells, HMEC231, which are unmethylated in all genes, except LINE1, ARHI, and the 14-3-3 σ gene). RKO, a colon cancer cell line that is highly methylated in hMLH1 and p16, was also used as a positive control.

In the studied cohort, adjacent normal breast tissue was taken from each of 90 patients during surgery. Taking advantage of paired normal/tumor samples in this study, we chose the value of normal samples as the reference. If using the sample mean plus two times the standard deviation of the pooled normal samples (and a minimum of 10% methylation) as a cut-off point, there is > 97% probability that the methylation level for a normal tissue will be lower than the cut-off point. It is reasonable to assume that a value larger than the cut-off point is likely to be abnormal (or positive). Panels of genes, that is to say, HIN-1/RASSF1A and RIL/CDH13 were considered positive if both markers in each panel were positive. Descriptive analyses were performed first for exploratory purposes. Pair-wise scatter plots are presented to show the correlations among the genes methylated. Heat maps were plotted to show levels of gene expression or methylation using hierarchical clustering to visually represent the association of different genes or samples by histopathologic tumor characteristics (ERs, PRs, and HER-2/neu). Chi-square or Fisher's exact test were used to assess the dependence between two categorical variables. Pearson's correlation coefficient was used to assess the relationship between two continuous variables. The Wilcoxon rank-sum test was used to compare either continuous or categorical variables between two groups. Analysis of variance (ANOVA) was applied to compare the values of gene methylation with tumor characteristics. Multivariate logistic regression was used to assess the ability of various levels of gene methylation to predict the ER, PR, HR, or HER-2/neu status. All reported p values are two-sided and considered statistically significant if p < 0.05. Analyses were performed using S-PLUS 2000 software (Insightful Corp., Seattle, WA, USA).

Patient characteristics are presented in Table 1. Normal paired tissues were unavailable in two cases. Information regarding the grade, HER-2/neu status, and ER and PR status were unavailable for one, ten, and three patients, respectively.

To confirm the reliability of bisulfite pyrosequencing for the quantitation of methylation levels, we measured the methylation of ARHI CpG islands I and II using pyrosequencing and COBRA and compared the resulting data. COBRA detected only one CpG site, whereas pyrosequencing detected four to eight CpG sites in one CpG island (Figure 1). From six breast cancer cell lines and two cultures of normal breast epithelial cells, methylation data were matched in CpG island II but not completely matched in CpG island I. COBRA showed two breast cancer cell lines were partially methylated in CpG island I (50% and 54% in one CpG site), but they were shown to be hypermethylated in CpG island I by pyrosequencing (the mean of four CpG sites per island was 92% and 91%, respectively). On the basis of these results, pyrosequencing was more sensitive for methylation detection than COBRA.

To confirm the reproducibility of pyrosequencing, the methylation of HIN-1 and RIL was measured repeatedly in 22 samples using bisulfite treatment and pyrosequencing assays. The correlation between repeated and initial values was high (for HIN-1, Pearson's correlation coefficient R = 0.99 and p < 0.0001; for RIL, R = 0.99 and p < 0.0001). In addition, the variation between individual assays was < 6%: the mean differences and 95% confidence intervals (CIs) for HIN-1 and RIL were 1.9% with 95% CIs of -0.1% to 3.9% and 3.0% with 95% CIs of 0.1% to 6.0%, respectively.

We evaluated the methylation status of 12 putative tumor-suppressor genes (ARHI, RASSF1A, HIN-1, RARβ2, hMLH1, the 14-3-3 σ gene, RIZ1, p16, the E-cadherin gene, RIL, CDH13, and NKD2), LINE1, ER, and PGRB in both culture cell lines and primary cancer tissues. The results from the analysis of six breast cancer cell lines and two normal breast epithelial cell samples (Additional file 2) indicated that 7 of 12 genes (RIL, HIN-1, RASSF1A, ARHI, CDH13, RARβ2, and NKD2) were densely methylated (marked bold) in cancer cell lines but not in normal breast epithelial cell lines. We also tested the methylation profiles of the 12 genes in breast cancer and the adjacent normal breast tissues from the same patients. In our pilot study, we screened 12 tumor-suppressor genes in 32–37 paired tissues. Five tumor-suppressor genes (RASSF1A, RIL, HIN-1, CDH13, and RARβ2) were frequently methylated in breast cancer tissues (the positive rate ranged from 17% to 58%) but not in normal breast tissues (the positive rate ranged from 0% to 4.5%). Another five genes (RIZ1, the E-cadherin gene, p16, hMLH1, and NKD2) were not frequently methylated in either malignant or normal breast tissue (the positive rate ranged from 0% to 9%, all < 10%). The 14-3-3 σ gene is highly methylated in both malignant and normal breast tissues. ARHI, as an imprinted gene, is at least partially methylated in all cases (Figure 2 and Table 2). Consequently, seven genes were unsuitable for development of methylation profiles. To correlate the methylation profile to clinical outcomes, we focused on the five highly methylated genes by assaying 53 more paired samples. PGRB was not frequently methylated in the 48 cases of breast tumor or normal breast tissue. ER was not highly methylated in the 40 cases of breast tumor or normal breast tissue, but higher methylation levels were observed in normal tissues. LINE1, a global methylation marker, was highly methylated in all normal and malignant tissues, but the methylation levels were significant lower in tumors (60.0 ± 8.8) than normal tissues (68.2 ± 3.5; p < 0.001).

On the basis of data from continuous marker methylation analyses in normal tissues, the methylation levels of RASSF1A and HIN-1 exhibited a strong correlation to each other (R = 0.66, p < 0.001). Some moderate correlations were found between RARβ2 and CDH13 (R = 0.40, p < 0.001) and between RIL and CDH13 (R = 0.30, p = 0.004). Similarly, in tumor tissues, a moderate correlation existed between RASSF1A and HIN-1 (R = 0.51, p < 0.001) and a weak correlation existed between RIL and CDH13 (R = 0.19, p = 0.068). In addition, the global methylation marker LINE1 was negatively correlated to RIL (R = -0.25, p = 0.019) and RASSF1A (R = -0.26, p = 0.014) in breast cancer samples.

In normal tissues, RIL methylation levels increased with age (R = 0.336, p = 0.001), whereas other markers had little correlation to age. In the cancer tissues, HIN-1 levels exhibited a weak correlation to age (R = 0.229, p = 0.03), whereas other markers had little correlation to age.

To study whether gene methylation in cancer would affect adjacent normal-appearing mammary gland tissues, the pair-wise correlation between paired tumor and adjacent normal breast tissues was estimated for each marker. Methylation of RIL, HIN-1, RASSF1A, and CDH13 exhibited a positive correlation between breast cancer and normal tissues. The Pearson correlation values ranged from 0.25 to 0.38 (p values ranged from 0.02 to 0.0002).

Using the ANOVA model, no statistically significant associations were found between clinical stage, tumor size, or node status. Tumors that were poorly differentiated (grade 3) according to Black's nuclear grade (BNG) had lower methylation levels of HIN-1 than well- and moderately differentiated tumors (grade 1 + 2; p = 0.01). Similarly, RASSF1A methylation levels were lower in poorly differentiated cancers than well- and moderately differentiated tumors (p = 0.053). In addition, CDH13 and RIL methylation levels cancers were marginally higher in poorly differentiated than well- and moderately differentiated cancers (p = 0.059 and 0.096, respectively).

In univariate analyses, the ER status was positively associated with high HIN-1 and RASSF1A methylation levels but negatively correlated to high RIL methylation levels (p < 0.001, 0.002, and 0.01, respectively). Specifically, patients with a positive ER status had higher methylation levels of HIN-1 (mean of 29% in ER-positive tumors versus 9% in ER-negative tumors), whereas patients with a negative ER status had higher methylation levels of RIL (mean of 36% in ER-negative tumors versus 22% in ER-positive tumors). In addition, the PR status was positively associated with high HIN-1 methylation levels (p < 0.001) and negatively associated with high CDH13 methylation levels (p = 0.03). By defining the HR status as positive if the status of ERs and/or PRs is positive, the associations between the methylation markers and HRs are similar to those between the methylation markers and ERs. In multivariate logistic models, high levels of both HIN-1 and RIL methylation are independent predictors of the status of ERs and HRs, and high levels of HIN-1 and CDH13 methylation are independent predictors of PR status (Table 3). Higher methylation levels of HIN-1 were more likely to be associated with ER-positive than ER-negative tumors, whereas higher methylation levels of RIL were more likely to be associated with ER-negative than ER-positive tumors. Using multivariable logistic models, we have assessed the effect of each gene (LINE1, RIL, HIN-1, RASFF1A, CDH13, and RARβ2) on HR (adjusted for age), ER (adjusted for age and PR status), or PR (adjusted for age and ER status) status (Additional file 3). The results shown in Additional file 3 are similar to those in Table 3. Among the markers studied, the higher methylation levels of HIN-1 were significantly correlated to a lack of HER-2/neu amplification (p = 0.014).

Because of a correlation between the methylation levels of several of the genes studied, we assembled four potential markers into two panels: HIN-1/RASSFIA and RIL/CDH13. A univariate analysis was performed and a multivariate logistic model was fitted to the two panels. The results of the univariate analysis for each panel indicated that the panels had a greater power to predict ER, PR, and HR status (Table 4 and Figure 3) than the methylation of individual genes (Table 3). Methylation of the HIN-1/RASSFIA panel was strongly correlated to positive ER, PR, and HR expression; 88% of HIN-1/RASSFIA-positive breast cancers were ER-positive and only 12% of breast cancers were ER-negative (p < 0.001). Conversely, methylation of the RIL/CDH13 panel was strongly correlated to negative ER, PR, and HR expression, because 69% of methylation-positive tumors were ER-negative and only 31% were ER-positive (p = 0.001).

We summarize the fitted multivariate logistic models of the HIN-1/RASSF1A and RIL/CDH13 panels by categorical variables (Table 3). Specifically, a patient with a positive value in the RIL/CDH13 panel is ten times more likely to have an ER-negative tumor than a patient with a negative value. By contrast, a patient with a positive value in the HIN-1/RASSF1A panel is 17 times more likely to have an ER-positive tumor than a patient with a negative value.

Interestingly, we found that triple-negative breast cancers (in other words, those with negative ER, PR, and HER-2/neu status) were positively correlated to methylation of the RIL/CDH13 panel and negatively correlated to methylation of the HIN-1/RASSFIA panel (Table 4). In particular, 58% (15 out of 26) of triple-negative cancers exhibited methylation of the RIL/CDH13 panel, whereas 17% (11 out of 64) of triple-negative cancers failed to methylate in this panel (p < 0.001). Only 9% (3 out of 34) of triple-negative cancers exhibited HIN-1/RASSFIA panel methylation, whereas 41% (23 out of 56) of triple-negative cancers failed to methylate in this panel (p = 0.001). In addition, triple-negative cancers were significantly associated with advanced tumor grade (24% in grade 1 and 2 versus 76% in grade 3; p = 0.001).

We studied the mutation status of p53 in the 84 tumors in which DNA sequences were available for analysis. Overall, p53 mutations were found in 9 out of 84 (11%) cases. Seven cases with p53 mutations belonged to the group that lacked methylation in both panels; one case was positive in the HIN-1/RASSF1A panel and one case was positive in the RIL/CDH13 panel. In the methylation-negative group (both panels were negative), 18% (7 out of 38) of the specimens had p53 mutations; in the methylation-positive group (either panel or both panels were positive), only 4% (2 out of 46) of specimens had p53 mutation (p = 0.072 in Fisher's exact test). Interestingly, five of the cases with p53 mutations occurred in triple-negative (ER-, PR-, and HER2-negative) cancers.