Simultaneous CXCL12 and ESR1 CpG island hypermethylation correlates with poor prognosis in sporadic breast cancer

Breast tumour cell lines were all obtained from the Ludwig Institute for Cancer Research, (São Paulo, Brazil). The following cell lines were used: MDA-MB-436, MDA-MB-435, MDA-MB-231, MCF-7, PMC42, HB4a (control immortalized normal cells) [12] and HB4aC3.6 [13]. The cell lines were cultured in RPMI 1640 medium [14] containing 10% foetal bovine serum (supplemented with 0.2 mM glutamine, and 40 μg/mL garamycin, 10 μg/mL insulin, if necessary) at 37°C in a humidified incubator with 5% CO₂.

Frozen samples of breast tumours (n = 69) used for methylation analysis were obtained from patients treated by primary surgery for breast cancer at the Nossa Senhora das Graças Hospital, Curitiba, PR, Brazil with institutional approval (Process number 25000.007020/2003-93; CONEP register 7220 opinion number 251/2003). The study included only female patients with invasive breast tumours. All patients gave informed consent for the study to retain and analyse their tissue for research purposes. The ages of the patients ranged from 27 to 84 years (mean 57.8 ± 14.7). Histological types were either infiltrative ductal carcinoma (IDC) (n = 51, 73.9%) or infiltrative lobular carcinoma (ILC) (n = 18, 26.1%). The lymph node status of the patients was positive (n = 35, 51.5%) or negative (n = 33, 48.5%). Histologic grade was determined according to the modified Bloom-Richardson criteria. Of the patients, 27.6% were Grade I, 47.8% were Grade II and 24.6% were Grade III. TNM staging was done according to official classification methods [15]. Other clinicopathological data (tumour size, local recurrence, metastasis and death) are summarised in Table 1.

Standard immunohistochemical (IHC) detection was performed on sections from archival paraffin embedded breast tumour tissues. Protein expression in malignant breast tissues was detected with specific antibodies against estrogen receptor (ER) and progesterone receptor (PR); clones 1D5 and PgR 636 (DAKO), respectively. Monoclonal mouse anti-human antibodies were pre-diluted and incubated for 18 h at 4°C according to the manufacturer's instructions. The rabbit polyclonal antibody against HER2 detection was performed by the HercepTest™ (DAKO CYTOMATION code K5204). In addition, positive and negative controls for each marker were routinely performed during experiments.

Sections were then processed using the EnVision™FLEX Target Retrieval Solution (DAKO) according to the manufacture's recommendations. Immunohistochemical staining of the samples was evaluated and scored by two pathologists who were responsible for clinicopathological data. The cutt-of value for ER and PR status values was 10% of cells. Fifty-seven ERα-positive tumours were scored based on the number of positive cells present in IHC staining. The slides were scored following the criteria: <10% tumour cells showing nuclear staining scored as 0, 10-30% tumour cells showing nuclear staining scored as 1, 30-50% tumour cells showing nuclear staining scored as 2, >50% tumour cells showing nuclear staining scored as 3.

Total RNA was isolated using the TRIzol Reagent (Life Technologies, USA) according to the protocol supplied by the manufacturer. Reverse transcription reactions were performed using 500 ng of DNA-free RNA, an oligo (dT)_12-18 primer and Superscript II Reverse Transcriptase (Gibco, BRL). PCR was performed using CXCL12-specific primers, and GAPDH was used as a housekeeping control (Table 2). The PCR was performed in a volume of 20 μl containing 1× PCR buffer (Invitrogen), 1.5 mM MgCl₂ (Invitrogen), 200 μM dNTPs, 0.3 μM of each primer and 1 U of Taq Platinum (Invitrogen). The PCR conditions were as follows: 95°C for 10 min, 94°C for 45 s, the appropriate annealing temperature for 45 s, 72°C for 1 min and a final extension of 72°C for 5 min. PCR products were resolved on 1% agarose gels and visualised by ethidium bromide staining.

The cell lines MDA-MB-231, MDA-MB-435 and MDA-MB-436 were analysed using this technique. Cells were plated (10⁶ cells/ml) and treated for 7 days with 1 μM 5-aza-CdR (Sigma Aldrich, Deisenhein, Germany) or left untreated. The medium was changed every day, and no significant cell death was observed. After 7 days of treatment, total RNA was isolated. The expression of CXCL12 in breast tumour cells was analysed by RT-PCR using the housekeeping gene GAPDH as an internal control. PCR products were resolved on 1% agarose gels and visualised by ethidium bromide staining.

Genomic DNA was prepared from breast cancer cell lines or frozen tumour samples by the phenol/chloroform protocol [16]. They were then subjected to sodium bisulphite treatment using the EpiTect^® Bisulphite Kit (Qiagen) according to the manufacturer's instructions.

Previous studies used the MSP technique to evaluate the regulation of CXCL12 expression by DNA methylation and in this study we have utilized the designed primers for this region (island 2), as described [7].

In an attempt to determine all CpG islands and all potential transcription start sites (TSS) for CXCL12, we first proceeded with the identification of the promoter sequence [17]. The analyses were initiated using an identified RefSeq by GenBank accession number, after which we submitted the gene sequence to a Genome BLAT Search through the UCSC Genome bioinformatics website http://genome.ucsc.edu. We selected 2000 bps of sequence extending from the 5' upstream region to 1000 bps downstream of the region of the TSS. The BLAT program returned a sequence of 5677 bps that was first submitted to the CpGPLOT program from the European Bioinformatics Institute website http://www.ebi.ac.uk/emboss/cpgplot. This program defines a CpG island as ≥ 200 bps of sequence with ≥ 50% C + G content and ≥ 0.6 CpG observed/CpG expected. The 5677 bps from CXCL12 that we have analysed contained five CpG islands in a region of 3447 bps (Figure 1A). The 5677 bps sequence was also submitted to computational analysis to predict transcription factor binding sites using TESS http://www.cbil.upenn.edu/cgi-bin/tess/tess and MatInspector http://www.genomatix.de/ [18]. The DNA region we refer to as island 4 (Figure 1A) is positioned next to an estrogen responsive element (ERE) binding site that could be involved in breast cancer. This CpG island was also selected for methylation status analyses in this study.

Bild vergrößern

Island 4 was amplified from bisulphite-treated DNA samples using a nested-PCR amplification protocol. The two sets of primers were used for the nested reactions at their appropriate annealing temperatures, and are shown in Table 2. The first PCR reactions were performed as described below: 1 cycle of 95°C for 10 min, 94°C for 3 min, the appropriate annealing temperature for 3 min, 72°C for 2 min; 5 cycles of 94°C for 3 min, annealing temperature for 3 min, 72°C for 2 min; 35 cycles of 94°C for 1 min, annealing temperature for 1 min, and 72°C for 5 min. Amplified products were purified using the Qiaquick Gel Extraction Kit (Qiagen) and cloned into a pCR2.1 cloning vector (Invitrogen). Eight clones were sequenced for each cell line using the universal or reverse primers. DNA sequencing reactions were performed using Big Dye Terminator technology (Applied Biosystems) on an ABI 377 sequencer (Applied Biosystems) according to the manufacturer's instructions. One hundred percent methylation was obtained if a methylated cytosine in the CpG dinucleotides was present in eight sequenced clones. The methylation percentage for each tumour cell line (global methylation pattern) was calculated by dividing the number of methylated CpG dinucleotides by the total number of CpGs analysed.

Methylation-specific PCR was performed as previously described [19] and the CXCL12 primer sequences for island 2 [7] can be visualized in the Table 2. gDNA from the primary breast tumours was bisulphite-modified and amplified with CXCL12 primers specific for methylated (M) and unmethylated (U) DNA. Analysis of methylation of the ESR1 promoter was performed using the set of primers described by Li et al. [20]. PCR reactions were performed with 1 μL of modified DNA, 1× PCR Buffer (Invitrogen), 1.5 mM (for CXCL12) or 2.5 mM (for ESR1) of MgCl₂ (Invitrogen), 200 μM dNTPs, 0.3 μM of each primer and 1 U of Taq Platinum (Invitrogen). The PCR protocol was 95°C for 10 min; 38 cycles of 94°C for 45 s, the appropriate annealing temperature for 30 s and 72°C for 45 s, followed by a final extension of 72°C for 5 min. DNA modification was confirmed by a nested-PCR reaction with a set of primers for a previously described satellite region (Table 2). This reaction was used as a control for bisulphite modification quality [21]. This PCR reaction, the nested-PCR and temperature conditions are described in Table 2. The amplification products were separated on 2% agarose gels and stained with ethidium bromide.

The statistical analysis was carried out using SPSS program (version 16.0, SPSS Inc., Chicago, Illinois, USA). Associations between specific clinicopathological parameters were analyzed using Chi square test and Fisher's exact test. Statistical significance was assumed for p < 0.05. The overall survival was calculated from the time of diagnoses of disease to the occurrence of death. Survival data were censored on 30^th June, 2009, which was the date in which the survival data were correlated with the death registry for the last time of a means of 91 months after onset of the disease. Kaplan-Meier estimates are presented for the survival functions, and differences in survival were analyzed using the log rank test. Multivariate analysis was conducted with a backward application of Cox proportional hazards regression analysis was used to estimate hazards ratio (HR) and 95% confidence intervals (95% CI) for overall survival and metastasis-free survival. All covariates with p < 0.25 were retained in the final model.

We first analysed the CXCL12 expression pattern in seven breast tumour cell lines using RT-PCR. A transcript of 383 bps corresponding to the CXCL12 gene was detected in the less aggressive carcinoma cell lines MCF-7 and PMC42, in the normal cell line HB4a and in the modified normal cell line HB4aC3.6 with ERBB2 overexpression (Figure 2A). In contrast, no detectable levels of CXCL12 expression were observed in the highly aggressive tumour cell lines MDA-MB-435, MDA-MB-436 or MDA-MB-231. To definitively determine if CXCL12 expression was, in fact, lost, all analyses were repeated at least twice. GAPDH expression was detected in all samples (Figure 2A). These data are in agreement with the literature [6], except in the case of the normal cell lines and the tumour cell lines MDA-MB-436 and PMC42, where such data have not been previously described.

Bild vergrößern

To confirm the epigenetic transcriptional silencing of CXCL12, we treated the MDA-MB-435, MDA-MB-436 and MDA-MB-231 breast tumour cell lines with the demethylating agent 5-aza-2'-deoxycytidine (5-aza-CdR). The expression of CXCL12 was restored in MDA-MB-435 cells upon treatment as previously observed [6]. Expression of CXCL12 was also restored in MDA-MB-231 cells, but not in MDA-MB-436 cells (Figure 2B).

The CpGPlot program analysis of 5677 bps of the 5' region of the CXCL12 gene identified five putative islands in the CXCL12 promoter region. The CpG island is positioned in an area extending from -2594 to +853 (Figure 1A). In this region, five distinct CpG-rich regions were identified in the positions +512 to +853 (island 1), -840 to +506 (island 2), -1118 to -893 (island 3), -1872 to -1575 (island 4) and -2594 to -2323 (island 5). The region comprised of island 2 studied in the breast and colon carcinomas [6, 7] was -493 to +168 relative to the TSS and promoter activity was found in the region from -1010 to +122 [22]. The search for transcription factor binding sites by computational analysis revealed eight oestrogen responsive element (ERE) consensus sites, which are defined as 13 bps perfect palindromic inverted repeats with a 3 bps spacing of variable bases (n) 5'GGTCAnnnTGACC 3' [23]. Five of these sites are very distant from the TSS, two are inside island 2 and the essential GACC sequence is absent. Only one site with the sequence aGctgctggGACC (capital letters for those that match with the consensus), which lies 19 bps away from island 4, might be an oestrogen-binding site (Figure 1B). Therefore, we complemented our tumour sample analysis with MSP for CXCL12 CpG island 4.

Sodium bisulphite sequencing was carried out on a 297 bps DNA fragment containing 27 CpG dinucleotides (-1872 to -1575) (Figure 1B). We analysed the methylation pattern of eight independent alleles (eight clones) from HB4a and HB4aC3.6 normal cell lines, as well as the PMC42, MCF-7, MDA-MB-436, MDA-MB-435, MDA-MB-231 breast tumour cell lines. The methylation pattern varied among the different cell lines as shown in Figure 1C. The CXCL12-negative cell lines, MDA-MB-436 (71%) MDA-MB435 (84%) and MDA-MB-231 (78%) showed CpG dinucleotide hypermethylation. In contrast, the CXCL12-positive cell lines, HB4a (11%), HB4aC3.6 (4%), PMC42 (12%) and MCF-7 (5%), had lower levels of CpG dinucleotide methylation (Figure 1C). These results were in agreement with our expression analysis (Figure 2A). On the other hand, the breast tumour cell line MDA-MB-436 had a dense area of methylated DNA and did not regain expression of the CXCL12 gene when cultured with the demethylating agent 5-aza-CdR (Figure 2B). The importance of DNA methylation in cancer has been well established [24], and the focus in the field is now changing to include the mechanisms by which other chromatin modifications play a role in cancer development. Among these changes are the covalent modifications of histones that can control gene activity. Histone acetylation and methylation of specific lysine residues, such as lysine 9 in histone H3 or lysine 27 in histone H3, clearly participate in the silencing of genes [25].

We also evaluated the CpG dinucleotide sequence for island 4 in the primary breast tumour samples 1, 3 and 8 (Figure 1D). These samples were chosen from the MSP data analysis, and we have confirmed the pattern that was described.

Dinucleotides 4 and 5, and 21-24, which were chosen for the design of methylation-specific PCR (MSP) primers, lie within a region that is differentially methylated (Figure 1C). The MSP technique was tested with CXCL12 (island 2 and island 4) and ESR1 genes in DNA from the tumour cell lines in order to confirm the results from RT-PCR and data from the literature (ESR1). The CXCL12 MSP results from breast tumour cell lines were identical for the two islands (Figure 3A). The MDA-MB-436, MDA-MB-435 and MDA-MB-231 breast tumour cell lines showed methylation of the two islands, which is in agreement with the observed lack of CXCL12 expression. In contrast, the HB4a, HB4aC3.6, MCF-7 and PMC42 cell lines expressing CXCL12 showed no methylated bands (Figure 3A).

Bild vergrößern

The ESR1 MSP results showed hypermethylation in the MDA-MB-436, MDA-MB-435 and MDA-MB-231 breast tumour cell lines as well as in the HB4aC3.6 normal cell line. The HB4a normal cell line and the MCF-7 and PMC42 breast tumour cell lines did not show any methylated bands (Figure 3A). These data corroborate the literature [10, 26] except in the case of cell lines MDA-MB-436, PMC42 and HB4aC3.6, which have not been previously described.

In this work, we did not evaluate normal breast tissue, but Zhou et al. [27] used twenty normal samples and all of them showed CXCL12 expression without methylation, which is in agreement with a possible tumoral methylation specificity observed in our results.

The MSP assay was subsequently used to analyse primary breast tumour samples. All samples showed the band corresponding to the unmethylated state for islands 2 and 4 (data not shown). Based on these results, we can conclude that the samples have cellular heterogeneity due to the presence of normal tissue or infiltrating lymphocytes.

For the methylated condition, nine representative tumour samples are shown (Figure 3B). Of all the samples tested (69), only four were methylated at both islands, and three of these samples methylated at both genes (Figure 3B). Moreover eighteen samples were methylated in both, CXCL12 island 4 and ESR1 genes (p = 0.0220). This finding indicates a strong correlation between the promoter methylation of ESR1 and CXCL12 island 4. In other words, methylation of the ESR1 promoter could be involved in directing the methylation of island 4 (see more details in discussion).

We also wanted to confirm that the presence or absence of island 2 or island 4 methylation observed in MSP (Figure 3B) was not just a PCR artefact. Therefore, we sequenced island 4 isolated from tumour samples 1, 3 and 8 (representative for banding pattern). Island 4 displayed 6% methylation in samples 1 and 3, and 89% methylation in sample 8 (Figure 1D). These results are in agreement with the PCR bands observed in the MSP.

The correlation between the CpG islands of CXCL12 and ESR1 methylation and the clinicopathological parameters of our samples are shown in Table 1. For island 2, CXCL12 methylation was observed in ten of 69 (14.5%) samples and was not significantly associated with any clinicopathological data (Table 1a). For island 4, CXCL12 methylation was observed in 37 of 69 (54%) samples. The clinicopathological parameters are shown in Table 1b and hypermethylation in this region was significantly associated with histological grade SBR (p = 0.0180), metastasis (p = 0.0008) and death (p = 0.0019). These results suggested that CpG island 4 might be important as a factor for poor prognosis of disease. In order to test this hypothesis we evaluated all of the clinicopathological data for their prognostic value in a univariate analysis for overall survival (OS) and metastasis-free survival (MFS) using Kaplan-Meier analysis (p value for log rank test). The overall survival (OS) was significantly worse for many clinicopathological parameters, which are shown in additional file 1. The graphical results for OS and MFS are shown in Figure 4 for ESR1 and CXCL12. The OS correlated with ESR1 (p = 0.0009) and CXCL12 island 4 (p = 0.0071) methylation (Figure 4A and 4B respectively). The metastasis-free survival correlated with ESR1 and CXCL12 island 4 (for both, p < 0.0001) (Figure 4C and 4D, respectively). The Kaplan-Meier analysis showed that the overall survival and metastasis-free survival were significantly shorter when silencing of CXCL12 island 4 and ESR1 occurred by gene hypermethylation. The effects of the covariables on OS and MFS were examined in the Cox proportional hazard regression model. The results for the univariate analysis are shown in Table 3. In the multivariate analysis, all variables presenting p < 0.2 from the univariate analysis were selected to build the multiple model (Table 3). For overall survival, the presence of metastasis (p = 0.0022) was considered a prognostic factor and ESR1 hypermethylation showed a statistical trend (p = 0.0555). Besides ERα, the hypermethylation of CXCL12 island 4 and ESR1 genes (p = 0.0302, p = 0.0089 and p = 0.0046, respectively) was considered an independent prognostic factor for metastasis-free survival. Patients with ERα protein present had the lowest risk of metastasis development (HR 0.3343; 95% CI 0.1247-0.8963). Otherwise, patients with methylation at both CXCL12 island 4 and ESR1 genes had a risk of developing metastasis, with CXCL12 island 4 having the highest risk (HR 7.1355; 95% CI 1.6480-30.8958).

Bild vergrößern

ESR1methylation was observed in 28 of 69 (41%) tumour samples (Table 1c) and was not significantly associated with age, HER2 or progesterone receptor status, recurrence or histological type. On the other hand, the stage (p = 0.0003), tumour size (p = 0.0029), histological grade SBR (p = 0.0011), lymph node status (p = 0.0288), oestrogen receptor status (p = 0.0054), metastasis (p < 0.0001) and death (p = 0.0105) were significantly associated with ESR1 gene methylation. Statistical analyses demonstrated a correlation between the intensity of ERα IHC staining and the frequency of ESR1 promoter methylation. Nine out of eleven (82%) ERα-negative cases were hypermethylated. Based on the results from the immunohistochemical analyses, it should be noted that of the samples scored as 0, 15 out of 25 (60%) were methylated; of the samples scored as 1, 8 of the 22 (36%) were methylated; of the samples scored as 2, 5 of the 16 (31%) were methylated; and of the samples scored as 3, none of the six samples were methylated (Figure 5). The number of unmethylated and methylated samples for the scores were submitted to a statistical Fisher's exact test, and statistical significance (p < 0.0001) was obtained (Figure 5). This result with ERα-negative status could be due to many mechanisms, including epigenetic mechanisms such as DNA methylation silencing.

Bild vergrößern

These results suggest that in breast cancer patients, silencing of the CXCL12 island 4 and ESR1 genes by promoter hypermethylation can be used as a complementary marker for metastasis risk assessment.