Cryo-conserved clinical samples were obtained from breast cancer patients treated by primary surgery at the University Hospitals of Aachen, Düsseldorf and Regensburg. Patients receiving neo-adjuvant chemotherapy and patients with recurrent breast cancer were excluded. Resected tissue was snap-frozen in liquid nitrogen immediately after surgery. Only samples containing more than 70% of tumourous cells in haematoxylin/eosin-stained control sections were further processed (n = 128). For 17 samples, macroscopically normal breast tissues containing at least 30% of epithelial cells were available. In all cases, two board-certified pathologists agreed on the diagnosis of breast cancer. Tumour histology was determined according to the criteria of the World Health Organization (2003), whereas disease stage was assessed according to the UICC (Union Internationale contre le Cancer) 19 . Tumours were graded according to Bloom and Richardson, as modified by Elston and Ellis 20 . All patients gave informed consent for retention and analysis of their tissue for research purposes, and the institutional review boards of the participating centres approved the study. For 98 patients, follow-up data were available with a median time of 63 months (range 1 to 124 months). Patient characteristics of this cohort are summarised in Table 1.

A tissue microarray (TMA) was created as described previously by Bubendorf and colleagues 21 . The formalin-fixed paraffin-embedded (FFPE) tissue sections were obtained from the archive of the Institute of Pathology, University of Regensburg, Germany. In all cases, two board-certified pathologists agreed on the diagnosis of breast cancer. Patients receiving neo-adjuvant chemotherapy and patients with recurrent breast cancer were excluded. All patients gave informed consent for retention and analysis of their tissues for research purposes, and the institutional review board of the participating centre approved the study. The TMA consisted of 150 primary tumours from malignant breast tissue and 44 normal breast specimens. Follow-up data were available for 146 patients with a median time of 77 months (range 1 to 148 months). Detailed tumour characteristics of this cohort are listed in Table 2.

The non-cancerous breast cell lines MCF10A and MCF12A as well as the cancerous breast cell lines MCF7, SKBR3, MDA-MB231 and BT20 were obtained from the American Type Culture Collection (Manassas, VA, USA) and cultured as recommended by the vendor.

Expression of EDN3 mRNA in various human tissues was tested using the commercial multiple-tissue Northern (MTN) blots I and II (Clontech, Heidelberg, Germany), containing 2 μg of poly A⁺ RNA per lane from 16 different human tissues (that is, blot I: heart, whole brain, placenta, lung, liver, skeletal muscle, kidney and pancreas; blot II: spleen, thymus, prostate, testis, ovary, small intestine, colon [no mucosa] and peripheral blood lymphocytes). Hybridisation was performed using 25 ng of an EDN3-specific 722-base pair (bp) polymerase chain reaction (PCR) product derived from (GenBank accession number NM_000114.2) (position: 968 to 1,707), which was verified by sequence analysis. ³²P-labelling of the DNA probe was achieved using the Megaprime DNA Labeling System (Amersham Biosciences, now part of GE Healthcare, Little Chalfont, Buckinghamshire, UK), and hybridisation was performed in accordance with the recommendation the manufacturer. The cancer profiling array (CPA) I (Clontech) is a matched tumour/normal expression array consisting of cDNA synthesised from 50 breast carcinomas, 50 normal breast tissues and 3 breast cancer lymph node metastasis specimens. Hybridisation was performed in accordance with the recommendations the manufacturer as described above for the MTN blots. Hybridisation signals of both, MTN blots and the CPA, were evaluated by use of a STORM-860 phosphoimager (Molecular Dynamics, now part of GE Healthcare). Intensity ratios were calculated after normalising signals against the background.

Frozen tissue samples and cell line pellets were dissolved in lysis buffer for subsequent DNA isolation using the QIAmp DNA Mini kit (Qiagen, Hilden, Germany) or for total RNA isolation by using TRIzol reagent (Invitrogen Corporation, Carlsbad, CA, USA) in accordance with the protocol supplied by the manufacturers.

Of the extracted total RNA, 1 μg was reverse-transcribed using the Reverse Transcription System (Promega Corporation, Madison, WI, USA) by applying a mix of oligo-dT and pdN₍₆₎-hexamer primers (1:2). The obtained cDNA was diluted (20 ng/μL) and test-amplified using intron-spanning primers for glyceraldehyde-3-phosphate-dehydrogenase (GAPDH). Primer sequences are provided in Table 3. PCRs were initiated as 'Hot Start' PCR at 95°C for 5 minutes and a hold at 80°C before the addition of 1 unit of GoTaq DNA polymerase (Promega Corporation). Cycle conditions were 95°C for 5 minutes, 35 cycles of 95°C for 1 minute, 60°C for 1 minute, 72°C for 1 minute and a final extension at 72°C for 10 minutes. PCR analyses were carried out in a PTC-200 cycler (Bio-Rad Laboratories, Inc., formerly MJ Research, Hercules, CA, USA). Amplificates were evaluated under ultraviolet light after 2% agarose gel electrophoresis containing ethidium bromide. Only samples yielding a specific 510-bp amplificate were further subjected to real-time PCR.

The Roche LightCycler system was used for semi-quantitative light cycler analysis in combination with the LightCycler DNA Master SYBR Green I Kit (Roche, Mannheim, Germany) as previously described 22 . Gene expression was quantified by the comparative cycle threshold (C_T) method, normalising C_T values to the housekeeping gene GAPDH and calculating relative expression values 23 . A commercially available normal breast cDNA pool (Clontech) was used as a breast reference standard. Primer sequences are listed in Table 3.

Paraffin sections of 2 μm were deparaffinised in xylene followed by rehydration in a decreasing ethanol series. Antigen retrieval was performed by pre-treatment in boiling citrate buffer (pH 6.0) in a microwave oven for 30 minutes (200 W). Immunohistochemistry (IHC) was performed using an NEXES Immuno Stainer (Ventana Medical Systems, Inc., Tucson, AZ, USA) in accordance with the specifications of the manufacturer. A goat polyclonal EDN3-specific antibody (sc-21628; Santa Cruz Biotechnology, Inc., Santa Cruz, CA, USA) was used in a 1:150 dilution by using the ChemMate Envision Kit (DAKO, Hamburg, Germany). Counterstaining was performed by using Mayer's haematoxylin. The incubation with primary antibodies was omitted in negative controls. All analysed samples were stained without the knowledge of histopathological data. Cytoplasmic protein staining was semi-quantitatively scored by an experienced breast pathologist according to the well-established scoring system developed by Remmele and Stegner 24 . To verify staining specificity, the primary antibody was incubated with a 200 molar excess of blocking peptide (sc-21628 P; Santa Cruz Biotechnology, Inc.) for 2 hours prior to its application on test samples.

EDN1, EDN2 and EDN3 genomic nucleotide sequences were taken from the Ensembl database and analysed for promoter CpG (cytosine-phosphate-guanine dinucleotide) islands in accordance with the method of Li and Dahiya 25 . A fragment of 2 kb in size, beginning 1 kb 5'-upstream from the annotated transcription start site (TSS) and ending 1 kb 3'-downstream, was analysed by applying the following criteria: island size of greater than 200 bp, guanine/cytosine content of greater than 60% and observed/expected CpG ratio of greater than 0.6.

Of the genomic DNA, 1 μg was bisulphite-modified using the EZ DNA Methylation Kit (Zymo Research Corporation, Orange, CA, USA) and eluted in 20 μL of Tris buffer (10 mM). Methylation-specific PCR (MSP) was performed in accordance with the method of Herman and colleagues 26 . One microlitre of bisulphite-treated DNA was amplified using MSP primers that specifically recognise either the unmethylated or methylated EDN3 promoter sequence after bisulphite conversion (Table 3). To achieve high accuracy, each primer was designed to cover three CpG sites of template DNA. Commercially available universal poly-methylated DNA and unmethylated DNA (EpiTect Control DNA; Qiagen) were used as positive controls for methylated and unmethylated EDN3 sequences, respectively. MSP products were visualised under ultraviolet light after 3% low-range ultra agarose gel electrophoresis containing ethidium bromide (Bio-Rad Laboratories, Inc.). Promoter methylation status was interpreted in a binary qualitative fashion.

Cells were seeded at a density of 3 × 10⁴ cells/cm² in a six-well plate. The demethylation agent 5-aza-2'-deoxycytidine (DAC) (Sigma-Aldrich, Deisenheim, Germany) was added to a final concentration of 1 μM in fresh medium at days 1, 2 and 3 after seeding. Additionally, cells were exposed to 300 nM trichostatin A (TSA) (Sigma-Aldrich) on day 3 for 24 hours. Control cells without DAC/TSA were supplied with fresh medium on days 1, 2 and 3. DNA and RNA were extracted on day 4 as mentioned above.

SPSS version 14.0 (SPSS Inc., Chicago, IL, USA) was used for statistical analyses. All tests were performed two-tailed, and P values of below 0.05 were considered statistically significant. The non-parametric Mann-Whitney U test and the Student t test (paired and unpaired) were used to compare expression results between cancer tissues and normal tissues or between EDN3 mRNA expression and EDN3 methylation status. Contingency table analysis and Fisher exact tests were used to study the statistical association between clinicopathological factors and EDN3 protein expression or EDN3 promoter methylation status. Survival curves comparing patients with or without any of the factors were calculated using the Kaplan-Meier method, with significance evaluated by log-rank statistics. Breast cancer-specific survival (CSS) was measured from the day of surgery until tumour-related death and was censored for patients alive at last contact or in case of death unrelated to the tumour. Disease-free survival (DFS) was measured from surgery until disease relapse and censored for patients alive without evidence of relapse at the last follow-up. For EDN3 protein expression, a multivariate Cox proportional hazard model was employed to assess the relative risks on patient CSS and to test for independent prognostic relevance of clinical/investigational factors. Only patients for whom the status of all selected variables was known were included in the proportional hazard model (n = 121). The limit for reverse-selection procedures was P = 0.2. The proportionality assumption for all variables was assessed with log-negative-log survival distribution functions. For analyses, the following variables were categorised into binary values: small-sized (pT1) versus large-sized (pT2 to pT4), node-negative (pN0) versus node-positive (pN1 to pN3) and low-grade (G1 and G2) versus high-grade (G3) tumours.

As previously reported, we performed in silico Northern blot analysis and RNA array-based expression profiling to identify novel candidate genes differentially expressed in human breast cancer 27 28 . In the latter approach, EDN3 was detected as one of the most frequently downregulated genes, showing substantial expression loss in 63% of breast carcinomas (data not shown). Therefore, we started a detailed analysis on EDN3 expression and its potential implication in human breast cancer.

An initial Northern blot analysis of normal human tissues demonstrates that EDN3 mRNA is abundantly expressed in a variety of non-malignant tissues, including pancreas, spleen, prostate, testis, small intestine and colon, with two major mRNA transcripts of 2.4 and 2.7 kb in size (Figure 1a). A breast cancer dot blot cDNA array was then hybridised with the same EDN3-specific probe. This showed a clear loss of EDN3 mRNA expression, as defined by a fold change (tumour versus normal) of greater than 2 (FC2), in 96% (48 of 50) of the analysed breast carcinoma samples and also in all three corresponding lymph node metastases (P < 0.001) (Figure 1b, c). To confirm this result, EDN3 mRNA expression was also assessed by real-time PCR in a set of 77 breast tumour tissues and 17 corresponding normal breast tissues. In breast carcinomas, downregulation of EDN3 expression by FC2 could be detected in 60 of 77 cases (78%; median: 21-fold) as compared with the normal breast reference standard (Figure 1d). This downregulation was still evident when comparing EDN3 expression among all tumour and normal breast tissues as illustrated by box plot analysis (P < 0.001) (Figure 1e) as well as when comparing fold changes of EDN3 expression among the 17 matched pairs (Figure 1f). In the latter analysis, 14 of 17 pairs (82%) showed EDN3 downregulation by FC2, with a median expression change of 13-fold.

To analyse whether loss of EDN3 expression in breast cancer is also evident on the protein level, we used a TMA comprising 150 invasive breast carcinomas and 44 normal breast tissue specimens. The specificity of the antibody applied was determined by the simultaneous use of blocking peptide against EDN3 in normal breast samples, which showed a clear decrease of overall EDN3 staining (Additional data file 1). EDN3 protein was clearly detectable in 75.0% (32 of 44) of normal breast tissue samples analysed (Figure 2a, b), as defined by an immunoreactivity score (IRS) of at least 8. The mean normal expression was determined to be IRS = 8.2 (± 3.8 standard deviations [SDs]). Expression was predominantly localised in luminal and basal epithelial cells of the normal breast and was weakly detectable in normal stromal compartments. In contrast, invasive breast carcinomas showed complete loss or reduced EDN3 expression (IRS < 8) in 45.3% (68 of 150) of cases (Figure 2c, d) and abundant EDN3 expression in 54.7% (82 of 159) (Figure 2e, f). Mean EDN3 expression in invasive breast carcinomas was determined to be IRS = 6.7 (± 4.0 SDs). EDN3 protein was rarely observed in tumour stroma (that is, was detectable only in those stromal cells adjacent to tumour cells with strong EDN3 expression). The difference of EDN3 expression between tumours and normal breast tissues was statistically significant (Mann-Whitney U test: P = 0.037; Student unpaired t test: P = 0.039).

To investigate a potential clinical relevance of EDN3 in breast cancer, we first analysed whether EDN3 protein expression is associated with clinicopathological parameters. In a bivariate analysis, differential EDN3 expression was not associated with patient age at diagnosis, tumour size, lymph node metastasis, histological grade, histological type, tumour focality or oestrogen or progesterone receptor status (Table 4). Interestingly, in a univariate survival analysis, we found a significant association between low EDN3 expression and unfavourable CSS (P = 0.022) (Figure 2g and Table 5). Patients with low EDN3 expression in the tumour had a mean CSS of 99 months (95% confidence interval [CI] 85 to 113 months) as compared with patients with tumours showing high EDN3 expression, which had a prolonged mean CSS of 116 months (95% CI 106 to 126 months). The visual impression of the Kaplan-Meier curves suggests also an association of EDN3 expression with DFS (Figure 2h), but this was statistically not significant (P = 0.167) (Table 5). Next, we performed a multivariate Cox regression analysis to test for independent significance of EDN3 expression as a prognostic factor in patient CSS (Table 6). Patient age at diagnosis, tumour size, lymph node metastasis, histological grade and EDN3 expression were included in the model. After reverse selection, patient age, lymph node status, grade and EDN3 expression remained significant in the Cox model, with low EDN3 expression displaying a twofold elevated risk of dying from breast cancer (hazard ratio 1.98, 95% CI 1.09 to 3.61; P = 0.025).

The NH₂-terminal peptide structure of EDN3 differs considerably in essential amino acids between residues 2 and residues 4 to 7 from those of EDN1 and EDN2 (Figure 3a). This region bulges out of the basic EDN structure and has been reported to represent a major domain for binding specificity to EDN receptors and thus is critical in EDN signalling activity 29 . Because we found that EDN3 expression is frequently lost in breast cancer tissues, we next compared EDN1, EDN2 and EDN3 mRNA expression in breast cell lines in parallel. EDN1 was much more strongly expressed in all cancerous cell lines (MCF7, SKBR3, MDA-MB231 and BT20) than in non-cancerous MCF12A cells (Figure 3b), showing a mean upregulation of 13.9-fold (± 7.8 SDs). EDN2 was more strongly expressed in SKBR3 cells but less expressed in MCF7, MDA-MB231 and BT20 as compared with MCF12A cells. EDN3 expression, however, was clearly downregulated in all cancerous cell lines as compared with non-cancerous cells, revealing a mean expression of 0.008 (± 0.009 SDs) of that found in MCF12A cells (set to 1). Therefore, upregulation of EDN1 appears to coincide with downregulation of EDN3 in breast cancer cell lines.

Since promoter hypermethylation is responsible for transcriptional silencing of important tumour suppressor genes in various human cancer types 30 , we searched all three EDN genes for the presence of CpG islands in their promoter region. A region of high CpG density in the EDN3 nucleotide sequence was identified as a CpG island (Figure 4a), whereas the CpG density in EDN1 and EDN2 is lower and does not define a CpG island under the selected criteria. To analyse the methylation status of the EDN3 CpG promoter (Figure 4b), we performed MSP with DNA after bisulphite treatment from non-malignant human tissues and three non-malignant and four malignant breast cell lines. The utilised MSP primers were tested on a dilution series of poly-methylated DNA, which revealed a sensitivity of 0.01 (1%) in detecting methylated EDN3 DNA molecules in a background of unmethylated EDN3 DNA molecules (Figure 4c). We observed no methylation of the EDN3 promoter in HMEC cells, human placental tissue, peripheral blood lymphocytes (Figure 4c) or non-malignant MCF10A cells (Figure 4e). A weak methylation signal was detected in non-malignant MCF12A cells. Of the malignant breast cell lines, MDA-MB231 and MCF7 harboured a methylated EDN3 promoter. In BT20 cells, both unmethylated and methylated EDN3 promoter sequences could be detected, whereas EDN3 was unmethylated in SKBR3 cells. Next, we analysed the association between EDN3 expression and promoter methylation in six breast cell lines by in vitro demethylating their DNA and assessing EDN3 mRNA expression after the treatment. A clear conversion of methylation could be observed in cell lines that were originally methylated in the EDN3 promoter region (that is, in MDA-MB231 and MCF7 cells) (Figure 4e), resulting in 47-fold and 28-fold increases, respectively, in EDN3 mRNA expression (Figure 4f). In BT20 cells, however, the demethylating effect was weaker and led to a 3-fold induction of EDN3 transcription. In weakly methylated MCF12A cells, the conversion of methylated alleles induced EDN3 expression by 5-fold, whereas no substantially altered change of EDN3 expression could be detected in unmethylated MCF10A (1.6-fold) or SKBR3 (1.0-fold) cells.

Next, we analysed EDN3 promoter methylation in primary breast cancer as well. In total, 89 of 128 breast carcinoma samples (69.5%) showed EDN3 promoter methylation (for example, #5 in Figure 5a). The remaining breast carcinoma samples (39 of 128; 30.5%) were not affected by this epigenetic modification (for example, #13 in Figure 5b). None of the 17 normal breast tissues exhibited EDN3 promoter methylation. Cancerous tissues yielded a PCR product with primers specific for the unmethylated EDN3 promoter sequence in all cases, due to non-malignant contaminants (stromal cells and endothelial cells) present in the bulk tumour tissue, as has also been described by Suzuki and colleagues 31 .

We next aimed to analyse whether EDN3 promoter methylation was associated with EDN3 mRNA expression in these tissues. For 71 breast cancer specimens, both EDN3 methylation and EDN3 mRNA expression have been investigated in parallel. Figure 5b shows the distribution of EDN3 mRNA expression among these two groups. While carcinoma samples with unmethylated EDN3 promoter exhibited similar EDN3 mRNA expression as compared with normal breast tissues (Figure 1e), breast carcinomas with EDN3 methylation exhibited a significant downregulation of EDN3 mRNA expression as compared with EDN3 unmethylated samples (P = 0.005, Mann-Whitney U test).

Finally, we asked whether EDN3 promoter methylation may be of clinical relevance in human breast cancer, as we have previously found for EDN3 protein expression. In a univariate analysis, the EDN3 methylation status in breast carcinomas was not associated with patient age at diagnosis, tumour size, lymph node metastasis, histological grade, histological type or oestrogen or progesterone receptor positivity (Table 7). In contrast to EDN3 protein expression, EDN3 promoter methylation was significantly associated neither with patient CSS (P = 0.703) nor with patient DFS (P = 0.632) (data not shown).