Decreased expression of angiogenesis antagonist EFEMP1 in sporadic breast cancer is caused by aberrant promoter methylation and points to an impact of EFEMP1 as molecular biomarker

Authors


Abstract

EGF-containing fibulin-like extracellular matrix protein 1 (EFEMP1) was recently described as an antagonist of angiogenesis. Motivated by a strong dependence of tumor growth and metastasis on angiogenesis, we investigated the role of EFEMP1 in human breast cancer. We applied RNA microarray expression analysis and quantitative real-time PCR (QRT) in a total of 45 sporadic breast cancer tissues and found EFEMP1 down-regulation in 59% and 61% of the analyzed tissues, respectively. This down-regulation was confirmed on protein level. Immunohistochemistry in 211 breast cancer tissues resulted in reduced or even abolished EFEMP1 expression in 57–62.5% of the tumors. Bisulphite genomic sequencing in breast cancer cell lines and primary breast cancer tissues revealed promoter methylation as the major cause of this down-regulation. Furthermore, analysis of 203 clinically well characterized primary breast cancers displayed a significant correlation of reduced EFEMP1 protein expression with poor disease-free (p = 0.037) and overall survival (p = 0.032), particularly in those node-positive patients who received adjuvant anthracycline-based chemotherapy, but not in those treated by either cyclophosphamide-methotrexate-5-fluorouracil (CMF) or Tamoxifen. In summary, the presented data demonstrate for the first time the reduced EFEMP1 expression on RNA and protein level in a substantial number of sporadic breast carcinomas and its correlation with epigenetic alterations. Furthermore, these data point towards a possible predictive impact of EFEMP1 expression in primary breast cancer. © 2008 Wiley-Liss, Inc.

EFEMP1 (Fibulin-3) belongs to the Fibulin gene family, a newly characterized family of 6 extracellular matrix (ECM) proteins that are localized at basal membranes, stroma and ECM fibers mediating cell to cell and cell to matrix communication. They are thought to provide organization and stabilization to ECM structures during organogenesis and vasculogenesis.1 While the potential role of Fibulin 1, 2, 4 and 5 in tumor development has been described in more detail,2–8 only little information has so far been available on the role of EFEMP1.9, 10 Klenotic et al.11 described a strong interaction of EFEMP1 with TIMP-3, the tissue inhibitor of metalloproteinases-3, an ECM-bound protein, which regulates matrix composition and affects tumor growth, invasion and angiogenesis. Recently, EFEMP1 was described to be an antagonist of angiogenesis.12 Among other results, Albig et al.12 showed that EFEMP1 prevents angiogenesis and vessel infiltration into bFGF-supplemented matrigel plugs implanted in genetically normal mice, and that EFEMP1 also decreases growth and blood vessel density in tumors produced by MCA102 fibrosarcoma cells implanted into syngeneic mice.

To characterize angiogenesis antagonist EFEMP1 in sporadic breast cancer and to correlate its expression with epigenetic alterations, we performed array-based RNA expression profiling, quantitative real-time PCR (QRT-PCR), immunohistochemistry, loss of heterozygosity (LOH) analysis, mutation screening and methylation studies in several breast cancer tissue panels. Because of the strong dependence of tumor growth and metastasis on angiogenesis, several components of the angiogenic pathway such as VEGF, VEGFR2 and HIF1A have been already analyzed for their prognostic and predictive impact in breast cancer.13 We therefore also aimed to evaluate the potential of EFEMP1 as a prognostic and/or predictive biomarker for this cancer entity.

Material and methods

Clinical tissue samples

The clinico-pathological characteristics of the tumors analyzed by microarray expression analysis, QRT-PCR, LOH analysis, mutation screening and methylation studies represent normal occurrence in clinical practice. Menopausal status: 81% post-, 19% pre- or perimenopausal; tumor size: 77% ≤ 2 cm, 23% > 2 cm; nodal status: 54% node-negative, 46% node-positive; Grade: 7% G1, 39% G2, 54% G3; histologic type: 78% ductal, 13% lobular, 9% others. For tissue microarray (TMA) analysis, malignant tumor samples were obtained from 203 clinically well characterized breast cancers (clinico-pathological characteristics given in Table I). Appropriate informed consent was obtained from all patients. The ethics committees of all participating centers approved the use of the tissues and the corresponding clinical data.

Table I. Clinico-Pathological Data of Tumor Samples Analyzed by TMA
CharacteristicCategoryn
Median age at diagnosis58 years (range, 28–89)203
Menopausal statusPre55
Post143
Peri5
Tumor size≤ 2cm142
> 2cm60
Unknown1
Nodal statuspN017
pN1151
pN235
Tumor gradeG14
G279
G3115
Unknown4
Her-2 statusnegative162
positive37
unknown4
Estrogen receptor statusnegative42
positive129
unknown32
Progesterone receptor statusnegative55
positive116
unknown32
Adjuvant systemic therapyAnthracycline-containing chemotherapy31
CMF chemotherapy66
Tamoxifen106

Cell lines

Human mammary epithelial cell line MCF-12A as well as breast cancer cell lines BT-20, MCF-7, SK-BR-3, T-47-D and ZR-75-1 were obtained from ATCC (Rockville, MD) and cultured under recommended conditions.

RNA amplification and expression analysis using Affymetrix microarray technology

Tumor material was snap frozen in liquid nitrogen immediately after surgery. Only samples with >70% tumor cells were selected for analysis. Isolation of total RNA was performed using TRIZOL reagent (Gibco-BRL, Glasgow, UK). cRNA was synthesized using 800–2000 ng purified RNA, linearly amplified, labeled with biotin and hybridized to the Human Genome U133A array (Affymetrix, UK) according to the recommendations of the manufacturer. Signal intensities were scanned using an Agilent Gene Array Scanner G2500A (Agilent Technologies, Waldbronn, Germany). Data analysis was performed using “Microarray Suite 5.0,” “MicroDB 3.0,” Data Mining Tool (Affymetrix) and BioConductor software (http://www.bioconductor.org/).

Quantitative real-time PCR

Cancer tissues were microdissected as described by Rhiem et al.14 Total RNA isolation was performed using TRIZOL (Gibco-BRL). Reverse transcription of RNA, PCR amplification and detection was performed using TaqMan EZ RT-PCR Core Reagents (Applied Biosystems, Weiterstadt, Germany) on a Sequence Detection System (ABI SDS 7700, PE Applied Biosystems) applying a standard one-step–protocol recommended by ABI. The sequences of the specific TaqMan primers and probe for EFEMP1 are given in Table II. The housekeeping gene GAPDH was used as a reference.

Table II. Primer Sequences Used in this Study
LOH-primer F25′-GTAACTGCTTAATGCCTTACC-3′
5′-Cy5-CTGGGCAACAGAGTGAGAC-3′
LOH-primer F35′-GTAGTCCCAGCTACTCAGAG-3′
5′-Cy5-AAGCTGGCTTCAAAATTTGGG-3′
Bisulphite genomic sequenicng A25′-TATTTGGATTTTATAGGAGTTGGTTAGA-3′
5′-CTCTTTTTTCTTATCAATCTAAATCCC-3′
Bisulphite genomic sequencing B25′-AGGGAGGTGGAGGTGTGTAGTTT-3′
5′-ATAAAAACCCCTTTCTTAACAACAAAC-3′
TaqMan-primer5′-TTGGCACACGTTGTGTTCAC -3′
5′-GAACCCAGCTGACCCTCA-3′
TaqMan-probe5′-TTCCCACCGTATCCAGTGTGCAGCA-3′
LightCycler-primer5′-GAT GAA TGC AGA ACC TCA AGC-3′
5′-TCG TGG ATA ACA ACG GAA GC-3′

LOH analysis

LOH analysis was carried out as described previously.15, 16 Primer sequences for the used microsatellite markers F2 and F3 are listed in Table II.

Mutation analysis

Genomic mutation screening was performed by DHPLC analysis (WAVE, Transgenomics, Omaha, NE) and exon sequencing with the adjacent intronic sequences using Big Dye chemistry (Perkin Elmer, Heidelberg, Germany) with separation of the fragments on an ABI capillary sequencer (ABI3100). Primer sequences for DHPLC analysis and exon sequencing as well as conditions for DHPLC analysis are given in Supporting Information 1.

5-Aza-2′-deoxycytidine treatment

Cells were seeded at a density of 1 to 3 × 105 cells/cm2 in a 6-well plate on day 0. The demethylating agent 5-Aza-2′-deoxycytidine (DAC) (Sigma-Aldrich, Steinheim, Germany) was added to a final concentration of 0.1–7.5 μM in fresh medium on day 1. Cells were harvested on day 4 for DNA and RNA extraction. Control cells were incubated without the addition of DAC. Restoration of EFEMP1 expression was analyzed applying real-time PCR using the LightCycler system together with the LightCycler DNA Master SYBR Green I Kit (Roche Diagnostics, Mannheim, Germany). Primer sequences are given in Table II. Gene expression was quantified by the comparative CT method, normalizing CT values to the housekeeping gene GAPDH and calculating relative expression values.16, 17

Bisulphite modification and bisulphite genomic sequencing

Genomic DNA was isolated using the DNA Extraction Mini Kit (Qiagen). Bisulphite reactions were performed using the CpGenome DNA Modification Kit according to the manufacturer's recommendations (Intergen, New York). For each bisulphite reaction, 0.5–3 μg of DNA was used. For bisulphite genomic sequencing-PCR reactions, 50 ng of bisulphite-treated DNA was used in a final reaction volume of 50 μL. Primer sequences are listed in Table II (PCR product A2, −739 to −962; PCR product B2, −868 to −1,132 relative to start codon). The amplified fragments were subcloned using the TOPO-TA cloning kit (Invitrogen, Karlsruhe, Germany). Inserts were sequenced using M13 primers.

In silico screening for the promoter region of EFEMP1 was performed using the following software tools: “Methprimer” (http://www.urogene.org/methprimer/index1.html), FirstEF (http://rulai.cshl.org/tools/FirstEF/), “WEBGENE” (http://www.itb.cnr.it/sun/webgene/) and “Genomatix.”

Immunohistochemistry

Tissue microarray construction

A total of 203 tissue specimens of consecutive primary breast carcinomas were used for TMA construction. Median follow-up in patients still alive at the time of analysis was 92 months (range, 1–191). All clinico-pathological characteristics are given in Table I. Tumors had originally been fixed in neutral buffered formalin 4% and then paraffin embedded. TMAs were prepared using 2 peripheral and 1 central 1 mm2 tumor areas of each tumor.

Immunohistochemistry on tissue arrays

A specific polyclonal rabbit antibody against recombinant mouse Efemp1 protein was used for immunohistochemistry.18 Cross-reactivity of this antibody with human EFEMP1 was shown by western blot using recombinant human EFEMP119 and by specific immunostaining of human tissues (unpublished data). After dewaxing, endogenous peroxidase was blocked by 1% hydrogen peroxide (10 min). Antigen retrieval was performed using Protease XXIV (50 mg diluted in 15 mL of 0.1 M Tris/HCl, pH 8.0, 20 min at room temperature). Standard immunostaining used antibody goat anti-rabbit K 4002/4003 for EnVision-method (DAKO, Copenhagen, Denmark) on TechMate Horizon autostainer (DAKO) according to supplier's recommendations. Primary anti-Efemp1 polyclonal antibody was diluted 1:100 (antibody dilution and storage buffer, DAKO, F 3022) and incubated (room temperature, 2 × 25 min). Hematoxylin counterstaining was performed.

Immunohistochemistry scoring

Interpretation of the immunohistochemistry results was performed by 2 independent scientists blinded to the corresponding clinicopathological data. Slides were evaluated using light microscopy and a semiquantitative immunoreactivity score (IRS). By recording the percentage of positive ECM staining (PP-value: 0 = negative, 1 = <10%, 2 = 10–50%, 3 = >50%) and staining intensity (SI-value: 0 = no, 1 = weak, 2 = moderate, 3 = strong) for each sample, IRS was calculated by multiplying PP with SI. If different scores were obtained for the central and the 2 peripheral areas of a tumor, a mean IRS value was calculated.

Statistical analysis of tissue microarray data

Statistical analysis was performed using the SPSS software version 14.0 (SPSS, Chicago, IL). Fisher's exact tests and χ2-tests were applied to analyze the correlation between clinico-pathological and immunohistochemical variables. All tests were performed two-sided. Disease-free and overall survival curves were calculated using the Kaplan-Meier method and compared by log-rank testing. Multivariate Cox proportional hazard models were used to define the potential prognostic significance of individual parameters. The significance level was set to 5%.

Results

Analysis of EFEMP1 expression in primary breast tumors applying RNA microarray expression analysis

RNA microarray analysis was carried out in 27 macro-dissected primary breast cancer samples (>70% tumor cells) in comparison to 6 normal breast tissue samples. Clinical characteristics for the tumors represented normal occurrence in clinical practice and are given in the Material and Methods section “Clinical tissue samples.” For EFEMP1, down-regulation was observed in 59% (16/27) of cancer tissues with a 3.5-fold average change. Down-regulation was defined as an at least 2-fold reduction of expression compared with normal tissue.

Quantitative real-time PCR experiments in micro-dissected primary breast tumors

To confirm the results of the array analyzes, expression of EFEMP1 was also analyzed by quantitative real-time PCR in matched tumor/normal tissues of 18 micro-dissected primary breast cancer samples (>90% tumor cells). These analyzes revealed that 61% (11/18) of the tumors displayed down-regulation ranging between at least 50% and almost 100% compared with corresponding normal breast tissue (Fig. 1, black columns).

Figure 1.

Logarithmic presentation of differential EFEMP1 expression of 18 micro-dissected primary breast cancer samples (>90% tumor cells) in comparison to their corresponding normal breast tissues analyzed by real-time PCR. Sixty-one percent (11/18) of the tumors display a down-regulation ranging between at least 50% and almost 100% as compared with normal tissue (black columns).

LOH and mutation analysis

To assess the allelic loss of the EFEMP1 gene locus, microsatellite markers F2 and F3 (Table II), located in close vicinity of the EFEMP1 locus, were used for LOH analysis. One hundred and one matched tumor/normal DNA sample pairs were analyzed with marker F2, 57 matched samples with marker F3. Allelic loss was observed in 32% (F2) and 21% (F3) of informative cases, respectively.

Twenty-one LOH positive tumors were subsequently screened for mutations in EFEMP1. Besides 3 known intronic polymorphisms (rs2277886, rs10496056, rs3748959 [dbSNP, NCBI]), only 1 deletion (1818 delT) (NM_004105) in the 3′ UTR in 1 of the tumors and 1 unknown missense mutation c.476A>C, p.Asp49Ala (NM_004105) in a second tumor were detected.

Analysis of epigenetic alterations by expression studies after treatment with demethylating agents in cell lines

The results of mutation analysis indicated that the major reason for the reduced gene expression is unlikely to be a result of LOH, associated with pathogenic point mutations on the remaining allele. We therefore analyzed whether epigenetic alterations may be the cause of the observed reduced expression. One nontumorous cell line (MCF-12A) and five breast cancer cell lines (BT-20, MCF-7, SK-BR-3, T-47 D, ZR-75-1) were treated with the demethylating agent DAC. Endogenous EFEMP1 expression levels in these cell lines before and after DAC treatment were assessed by quantitative real-time PCR. As shown in Figure 2a, the analyzed tumor cell lines displayed different levels of EFEMP1 expression before DAC treatment. Relative to the nontumorous cell line MCF-12A with an expression level set to 1.000, MCF-7 showed the lowest expression value (0.016) followed by BT-20 (0.177) and SK-BR-3 (0.290). In contrast, ZR-75-1 and T-47-D displayed higher expression with values of 2.037 and 7.127 relative to MCF-12A. Thus, similar to the above described expression results in native tumor tissues, 60% (3/5) of the tumor cell lines displayed a reduced expression compared with the nontumorous cell line.

Figure 2.

(a) Endogenous level of EFEMP1 expression in 1 nontumorous cell line (MCF-12A) and 5 breast cancer cell lines (MCF-7, ZR-75-1, BT-20, SK-BR-3, T-47-D) before and after DAC treatment as revealed by real-time PCR. Values of expression induction (fold change) after DAC treatment of the different cell lines are marked with asterisks. (b) Methylation status in 1 nontumorous cell line (MCF-12A) and 5 breast cancer cell lines (MCF-7, ZR-75-1, BT-20, SK-BR-3, T-47-D) as revealed by bisulphite genomic sequencing. Columns indicate CpG dinucleotides 1-14, horizontal lines analyzed clones 1–16. Black dots symbolize methylated CpGs, white dots unmethylated CpGs. Values describe the percentage of methylated CpGs.

After DAC treatment, the 2 cell lines with the lowest expression levels, MCF-7 and BT-20, displayed the most pronounced up-regulation with a 124-fold increase of expression in MCF-7 and a 19.4-fold increase in BT-20. In contrast, the nontumorous cell line MCF-12A together with tumor cell lines SK-BR-3 and T-47-D showed only moderate up-regulation with 8.9-, 6.7-, and 5.9-fold changes, whereas ZR-75-1 did not show any up-regulation.

To further investigate the influence of the EFEMP1 promoter methylation status on EFEMP1 expression in the analyzed cell lines, we applied bisulphite genomic sequencing in a genomic interval ranging between −739 and −1132 relative to the position of the start codon of EFEMP1. Per bisulphite-treated DNA sample, 16 clones with 25 CpG dinucleotides each were investigated. None of the 11 CpG dinucleotides located within interval −962 to −1132 displayed any methylation in any of the investigated cell lines. However, differential methylation within the different cell lines was observed for the 14 CpG dinucleotides within interval −739 to −963.

As expected, the nontumorous cell line MCF-12A displayed no methylation. The methylation status of tumor cell lines MCF-7 and BT-20 with the lowest expression levels showed the highest methylation levels with 91% and 76% of methylated CpGs (Fig. 2b). In contrast, cell lines ZR-75-1 and T47-D with high expression levels showed no significant methylation (6.7% and 0%). Thus, 40% (2/5) of the tumor cell lines displayed significant methylation in the promoter region of EFEMP1 which was correlated with a significant reduction of expression compared with the nontumorous cell line and a pronounced up-regulation after DAC treatment. The moderate increase of expression after DAC treatment in MCF-12A, SK-BR-3 and T-47-D may be explained by transgenic effects because of demethylation of other genes rather than by demethylation of the EFEMP1 promoter.

Methylation analysis of the putative EFEMP1 promoter in primary breast cancer samples

We next analyzed whether EFEMP1 promoter methylation also contributes to the reduction of EFEMP1 expression in primary breast cancer samples. Therefore, we performed methylation analysis by applying bisulphite genomic sequencing in the promoter region described before in 14 macro-dissected breast cancer samples with LOH at 1 allele, 4 micro-dissected tumor samples without LOH, and 8 micro-dissected normal breast tissues.

As described before, per bisulphite-treated DNA sample, 16 clones with 25 CpG dinucleotides (position −739 to 1,132, relative to position of start codon) each were investigated. Again, none of the 11 CpG dinucleotides located within interval −962 to −1,132 displayed any methylation neither in tumor samples nor in normal breast tissues. However, differential methylation was observed for the 14 CpG dinucleotides within interval −739 to −963 (Fig. 3). The endogenous methylation levels in the primary breast cancer and normal breast tissues analyzed are shown in Figure 4. Based on the expression and methylation results in the analyzed cell lines, a tissue was defined as methylation-positive, if at least 25% of the CpG dinucleotides were methylated (Table III, Fig. 4). Based on this definition, 57% (8/14) of the macro-dissected LOH-positive tumors (T1–T8) and 50% (2/4) of the micro-dissected LOH-negative tumors (T15, T16) were defined as methylation-positive with an average of 55% and 32.8% of methylated CpG dinucleotides, respectively (Table III). In contrast, none of the 8 normal breast tissues (N1–N8) was found to be methylation-positive, since their highest percentage of methylated CpG dinucleotides was 8.9% (average 3.4%) (Table III, Fig. 4).

Figure 3.

Three examples of sequences of the differentially methylated interval (−739 to −962) as revealed by bisulphite genomic sequencing. Panel A displays a fully methylated sequence with 14 methylated CpGs. The cytosine nucleotides of the 14 CpGs are not converted into thymines during bisulphite treatment since they are protected by a methyl group. Panel B shows a partly methylated sequence. Only CpGs 1, 2, 3, 4, 6, 12, 13 and 14 are methylated. Panel C displays a fully unmethylated sequence since all cytosine nucleotides of the 14 CpGs are converted into thymines.

Figure 4.

Endogenous level of methylation amongst the analyzed primary breast cancer (T1–T18) and normal breast tissues (N1–N8) as revealed by bisulphite genomic sequencing. T1–T14 represent macro-dissected, LOH positive tumors and T15–T18 represent micro-dissected, LOH-negative tumors. Normal tissues N5–N8 are corresponding tissues to tumor tissues T15–T18 (indicated by arrows). Columns indicate CpG dinucleotides 1-14, horizontal lines analyzed clones 1-16. Black dots symbolize methylated CpGs, white dots unmethylated CpGs. A tissue was defined as methylation positive (+) if at least 25% of CpG dinucleotids displayed methylation.

Table III. Extent of Methylation Within Interval −739 to −969 of the 5′ Region of EFEMP1 in 18 Breast Tumor Tissues and 8 Normal Breast Tissues
inline image

Thus, these results strongly indicate that promoter methylation contributes to a major extent to the reduced EFEMP1 expression in the investigated sporadic breast cancer tissues.

EFEMP1 expression on protein level

After observing reduced EFEMP1 expression in sporadic breast tumors on the RNA-level, we investigated EFEMP1 expression on the protein level. A specific polyclonal rabbit antibody against recombinant mouse Efemp1 protein was used for immunohistochemistry (IHC).18 The amino acid sequences between human EFEMP1 and mouse Efemp1 are highly conserved (Supporting Information 2) and cross-reactivity of this antibody with human EFEMP1 was shown by western blot using recombinant human EFEMP119 and by specific immunostaining of human tissues (unpublished data).

Localization and expression of EFEMP1 in normal breast tissue

Localization of EFEMP1 was done by IHC, applying the mouse Efemp1 antibody described before. As shown in Figure 5a, EFEMP1 was detectable in normal breast tissues; faint staining was detected in myoepithelial and luminal cells of lobules and ducts, whereas intense staining was observed in the ECM as a linear pattern in the vicinity of the basal membrane of lobules (terminal duct-lobular-units) and as a diffuse pattern in the ECM of intralobular, interlobular and periductal stroma.

Figure 5.

(a) IHC staining of EFEMP1 in normal breast tissue. Faint staining is detectable in myoepithelial and luminal cells of lobules and ducts, intense staining can be seen in the ECM in a linear pattern in the vicinity of the basal membrane of lobules (terminal duct-lobular-units) and in a diffuse pattern in the ECM of the intralobular, interlobular and periductal stroma. Magnification 400×. (b) IHC staining of ECM of different breast cancer tissues with different EFEMP1 expression. Panel A and B: Weak EFEMP1 expression with Immune Reactivity Score (IRS) 1 and IRS 3. Panel C and D: EFEMP1 expression with IRS 6 and IRS 9. Magnification 200×.

EFEMP1 expression in breast cancer tissues

IHC staining of 8 matched breast cancer/normal breast tissues revealed reduced or even abolished expression of EFEMP1 in the ECM in 62.5% (5/8) of the tumor samples compared with the ECM of normal breast tissues. Immunohistochemically, EFEMP1 was not detectable in the tumor cells themselves in most cases; faint tumor cell staining was only observed in few cases, thus, only ECM staining was evaluable by this method.

Tissue microarray analysis

After observing reduced EFEMP1 expression in 5 of 8 tumor samples as compared with matched normal breast tissues, we investigated a set of 203 clinically well characterized primary breast cancers using TMA analysis.

Strong EFEMP1 expression (IRS 6.5–9) was only seen in 9% of the tumor tissues, moderate expression (IRS 4.5–6.0) in 34%, whereas 57% of the tumors displayed weak or absent expression (IRS 0–4.0) (Fig. 5b).

Correlation of EFEMP1 protein expression with methylation status

EFEMP1 protein expression was then analyzed in selected tumor tissues with different methylation status (tumor 2 [T2] with 91.5% methylation; tumor 4 [T4] with 41.5% methylation; tumor 14 [T14] with 0% methylation, see also Fig. 4). Figure 6 displays the results of IHC staining. While the tumor with the highest methylation status, T2, showed the lowest protein expression (IRS 3.0), tumor T4 with 41.5% methylation displayed moderate expression (IRS 6.0), whereas the tumor without any methylation, T14, showed strong expression (IRS 9.0). Thus, these results strongly support the correlation between high methylation status and low levels of gene expression.

Figure 6.

EFEMP1 expression on protein level by applying IHC staining in tumors of known methylation status. Magnification 100×. Panel A: Weak expression (IRS 3) was detected in tumor T2, which had displayed a very high methylation status (91.5%). Panel B: Moderate expression (IRS 6) was found in tumor T4 displaying a medium methylation level of 41.5%. Panel C: High expression (IRS9) was detected in tumor T14 which had displayed no methylation at all.

Correlation of EFEMP1-expression with patient outcome

To asses a potential clinical relevance of reduced EFEMP1 protein expression regarding patient outcome, stepwise Cox regression for multivariate survival modeling was performed for the above described 203 breast cancer samples. The following variables were included: tumor grade, steroid hormone receptor status, IRS for EFEMP1 protein expression, lymph node status, Her-2 status, tumor size and type of adjuvant systemic therapy.

We first analyzed the total cohort of 203 breast cancer samples, consisting of 17 node-negative and 186 node-positive cases. Looking at the total cohort, no obvious clinical relevance of EFEMP1 expression was seen. On the basis of assumed strong influence of EFEMP1 on tumor progression, we then analyzed the collective of the high-risk node-positive cases separately. This analysis revealed that next to tumor size and grade, EFEMP1 expression (applying an optimized IRS cut off value of 3.5) remained in the survival model as a relevant factor influencing disease-free and overall survival at borderline significance (DFS: p = 0.14; OS: p = 0.077).

Keeping in mind the fundamental differences between node-negative and node-positive patients regarding their primary therapy where particularly frequency and type of adjuvant systemic therapy may strongly influence their outcome results,20 we therefore investigated the relevance of EFEMP1 expression in patient subgroups with homogeneous adjuvant systemic therapy. Interestingly, this analysis revealed a significant correlation of low EFEMP1 expression with poor disease-free and overall survival (p = 0.037 and p = 0.032, respectively) in those node-positive patients who had received adjuvant anthracycline-containing chemotherapy (n = 31) (Fig. 7), whereas no significant impact was seen in node-positive patients treated either by cyclophosphamide-methotrexate-5-fluorouracil (CMF) chemotherapy (n = 49) (DFS: p = 0.605; OS: p = 0.934) or by adjuvant endocrine therapy (tamoxifen) alone (n = 106) (DFS: p = 0.735; OS: p = 0.275). While median disease-free survival in anthracycline-treated patients exceeded 10 years, if tumors showed high EFEMP1 expression, it was only 3.1 years in cases with low EFEMP1 expression (Fig. 7, Panel B). Median overall survival in these patients showed similar results with a median overall survival of more than 10 years for tumors with high EFEMP1 expression in contrast to only 4.5 years for those with low EFEMP1 expression (Fig. 7, Panel A).

Figure 7.

Kaplan-Meier analysis of survival in node-positive patients treated by adjuvant anthracycline-containing chemotherapy (n = 31). Overall (OS) (Panel A) and disease-free survival (DFS) (Panel B) according to EFEMP1 expression applying an optimized IRS cut off value of 3.5. Survival time in years is given on the X-axis and cumulative survival rate on the Y-axis.

Discussion

In contrast to other members of the fibulin gene family,2–8 only little information has been available so far on the function of EFEMP1 (Fibulin 3) regarding tumor development and progression.1, 9–11 The recent description of EFEMP1 as an angiogenesis antagonist12 prompted us to further characterize EFEMP1 in several panels of sporadic breast carcinomas.

Our expression studies on RNA level in a total of 45 cancer tissues revealed down-regulation of EFEMP1 in ∼60% of the tumors and, thus, supporting our preliminary results on differential EFEMP1 expression in breast carcinomas by electronic northern approaches and CPA analysis21 as well as the results of Albig et al.12 On the basis of cDNA dot blot analysis, these authors had described down-regulation of EFEMP1 in 6 of 9 (67%) breast cancer tissues. On the protein level, we assessed EFEMP1 expression by immunohistochemical staining in a total of 211 primary breast cancers and found down-regulation in 57–62.5% of these tumor samples. Thus, we identified comparable levels of EFEMP1 loss both on RNA and protein level. Interestingly, there was no correlation between the amount of desmoplastic tumor stroma and intensity of EFEMP1 expression shown by immunohistochemistry.

Using DNA methylation analysis, we were then able to show that EFEMP1 down-regulation in primary breast tumors is very likely caused by hypermethylation of the EFEMP1 promoter. Bisulphite genomic sequencing of 14 macro-dissected and 4 micro-dissected breast cancer tissues revealed significant methylation with at least 25% of methylated CpG dinucleotides in 57% and 50% of the tumors, while none of the 8 investigated normal breast tissues was found to be significantly methylated. These observations are supported by the methylation and expression results in 5 breast cancer cell lines. Those cell lines with the lowest expression levels compared with a nontumorous cell line displayed the highest methylation levels. Moreover, after DAC treatment they showed the most pronounced up-regulation. In contrast, the 2 cell lines with high expression levels compared with 1 nontumorous cell line showed only low or even no methylation and therefore only moderate or no up-regulation after DAC treatment. These results strongly indicate that promoter methylation contributes to a major extent to the reduced EFEMP1 expression in the investigated sporadic breast cancer tissues. Furthermore, we were also able to prove the correlation between the EFEMP1 methylation status and its protein expression by applying IHC staining in tumor tissues of known methylation status (Fig. 6).

To our knowledge, these are the first data on altered EFEMP1 methylation in sporadic breast cancer. Our data on hypermethylation of EFEMP1 in sporadic breast carcinomas are also supported by very recent results of Yue et al.22 These authors reported on hypermethylation of the same 5′ regulatory region of EFEMP1 in lung cancer.

Angiogenesis is a physiological process involving the growth of new blood vessels from preexisting ones. Besides normal growth and development, this process is of great importance for growth, invasion and metastasis of malignant tumors, since it enables them to initiate recruitment of their own blood and nutrient supply by shifting the balance between specific pro- and antiangiogenic factors.13, 23 Most studies on the relationship of angiogenesis and patient outcome reveal a clear relationship between high levels of angiogenesis and poor prognosis.24 To evaluate the correlation between the expression of the angiogenesis antagonist EFEMP1 and patient outcome in sporadic breast carcinomas, we also investigated the correlation of clinicopathological variables to immunohistochemical EFEMP1 expression. By performing multivariate Cox regression analysis, we were indeed able to show that next to tumor size and grade, EFEMP1 expression had a clinically meaningful impact on disease-free and overall survival in a substantial cohort of 186 node-positive patients with a long-term median follow-up of 92 months. Moreover, taking adjuvant systemic therapy into account, we saw significant correlations only in those node-positive patients who had received anthracycline-containing adjuvant chemotherapy, but not in those treated by adjuvant CMF or endocrine therapy alone.

In our retrospective cohort, patients who had received anthracycline-containing adjuvant chemotherapy were also those patients with a clinically perceived high-risk of relapse. Thus, at present we cannot clearly distinguish between a potential predictive impact of EFEMP1 expression regarding response to anthracycline chemotherapy from a mere prognostic impact of this factor regarding survival in high-risk breast cancer. However, keeping the biological role of EFEMP1 in mind, it is not unreasonable to assume that its antiangiogenic properties may enhance an antiangiogenic chemotherapy effect which has, in fact, been attributed to anthracyclines.24, 25

In conclusion, the presented data provide evidence for a down-regulation because of promoter methylation of the angiogenesis antagonist EFEMP1 on RNA and on protein level in a substantial number of sporadic breast carcinomas, thus indicating EFEMP1 as a new candidate tumor suppressor gene in this cancer entity. Moreover, multivariate regression analysis revealed a correlation of reduced EFEMP1 expression with poor disease-free and overall survival in node-positive breast cancer. Taking adjuvant systemic therapy into account, the impact of EFEMP1 expression was most pronounced in patients with adjuvant anthracycline chemotherapy. Thus, these results point to a possible predictive value of EFEMP1 expression regarding anthracycline response which needs to be further validated in larger collectives of homogeneously treated breast cancer patients. In view of clinically emerging angiogenesis inhibitors, identification and characterization of components of the angiogenic pathway as specific prognostic as well as predictive markers is of great relevance for the success of this treatment option to overcome drug resistance and identify correct target populations.13 EFEMP1, with its antiangiogenic properties, may serve here as an important molecular marker for defining an adequate tumor-biology oriented therapeutic strategy.

Acknowledgements

The authors thank Mrs. R. Busch, Institute of Medical Statistics and Epidemiology (IMSE), Technical University, Munich, for her expert statistical analysis and Dr. Günter Kostka (deceased on January 19, 2006) for providing the Efemp1 antibody.

Ancillary