In this study we characterized coexpression of the functionally linked prognostic markers Sdc1, E-cad and c-met in 200 DCIS using TMA technology and in different human breast cancer cell lines. Our initial objective was to identify a group of functionally interrelated markers that have prognostic significance in breast cancer. In this regard we observed significant correlations between expression levels for the markers c-met, Sdc1 and E-cad in DCIS. The markers c-met and Sdc1 were significantly more frequently expressed in pure DCIS than in DCIS with a coexistent invasive carcinoma. In the case of Sdc1, this finding may indicate downregulation during epithelial-mesenchymal transition [40, 41], which has been demonstrated in mammary epithelial cells in vitro [23, 24] and during development . Regarding the association between the expression pattern of Sdc1, E-cad and c-met with disease progression, future studies could investigate the correlation with expression for coexisting DCIS and invasive carcinoma. However, based on our cell culture data and on current knowledge on the prognostic value and biological function of the markers (see Table 1), we speculate that a reduction in E-cad expression and an upregulation of Sdc1 expression may occur in the invasive component of mixed lesions. In the case of c-met, our finding that c-met expression is highest in MDA-MB231 cells, and exhibits a positive correlation with other features of aggressiveness such as HER2 overexpression seemingly contradicts our observation that c-met expression is higher in pure DCIS than in DCIS associated with invasive cancer. We recently observed significantly increased expression of prognostically unfavourable proangiogenic factors in pure DCIS compared with coexistent DCIS . This is consistent with findings reported by Teo and coworkers , who described different vascular density and phenotypes in pure versus coexistent DCIS. Therefore, the preferential association of c-met expression with pure DCIS may be indicative of the increased angiogenic activity in pure DCIS [11, 29].
The metastatic origin of the investigated human breast cancer cell lines  has inherent limitations with respect to a direct comparison of marker expression in DCIS tissue. However, the positive correlation of c-met, Sdc1 and E-cad in an early stage of tumour progression at the tissue level was largely supported by our colocalization studies in these cell lines, in which partial colocalization was observed between c-met and Sdc1 and E-cad, respectively, in the less aggressive MCF-7 and MDA-MB468 cell lines. In addition, Sdc1 and E-cad exhibited a similar cell surface distribution in these cells. Moreover, progressive loss of E-cad expression and a progressive increase in c-met expression were observed with increasing de-differentiation and higher metastatic potential (MCF-7 < MDA-MB468 < MDA-MB231) of the breast cancer cell lines at the mRNA level.
These findings are similar to the expression patterns of E-cad and c-met observed in clinical studies. Reduced expression of E-cad in breast carcinoma is associated with shortened disease-free survival and high histological grade [43–46], whereas high expression of c-met is associated with more aggressive disease and decreased disease-free survival in node-negative breast cancer [15, 47] (see Table 1). In addition, coordinated overexpression of c-met and the oncogene RON is observed in MBA-MB231 but not in MCF-7 cells . Our observation of a synchronous downregulation of E-cad and Sdc1 mRNA in highly aggressive MDA-MB231 cells is in accordance with previous in vitro experiments on mammary epithelial cells [23, 24]. Expression of Sdc1 can be regulated at the post-transcriptional level [48, 49], which may explain the strong Sdc1 protein expression observed by confocal immunofluorescence microscopy in MDA-MB231 cells (Figure 1). The heterogeneity in Sdc1 and E-cad coexpression in breast cancer cell lines of different degrees of de-differentiation demonstrates that the concept of synchronous regulation of Sdc1 and E-cad in epithelial cells [23, 24] cannot be fully adapted to breast cancer. Our data suggest that the degree of coexpression is higher in breast cells that exhibit a more benign phenotype, such as DCIS, whereas coexpression may be lost in more advanced stages of tumour progression. These findings could help to explain why no correlation was found between E-cad and Sdc1 expression in a patient collective containing ductal invasive, lobular and tubular breast cancer tissues , whereas a correlation was found in our more uniform DCIS collective.
We observed no significant association of c-met, E-cad and Sdc-1 expression with hormone receptor status in DCIS. In patient collectives including more advanced stages of breast cancer progression, clinical and experimental data suggest a possible link between loss of E-cad expression and ER-negative status [20, 43, 50]. Moreover, recent studies have linked Sdc1 expression to an aggressive, ER-negative breast carcinoma subtype [6–8]. In order to distinguish between low and high expression levels of possibly synchronously regulated Sdc1 and E-cad, both low expression and absence of staining for the markers investigated were scored as 'negative'. Although application of this scoring method was necessary to detect synchronous coregulation of marker expression, it might have lead to failure to demonstrate significant associations of E-cad expression with hormone receptor status and grade of DCIS in the present study. Although it may occasionally be difficult to distinguish between DCIS and LCIS in cases where LCIS may involve extralobular ducts, we can largely exclude the possibility that the analyzed DCIS cases were erroneously scored as LCIS (see Materials and methods, above).
We found a significant correlation between HER2 and c-met expression. This correlation may be of functional importance, because both molecules are tyrosine kinase receptors that promote cell proliferation [11, 51]. Of note, Khoury and coworkers  recently demonstrated that HGF-mediated c-met signalling in the presence of constitutively active HER2 leads to a downregulation of E-cad and promotes epithelial-mesenchymal transition in MDCK cells. Thus, signalling downstream of the c-met receptor synergizes with HER2 to enhance a malignant phenotype. In a similar manner, additional receptors could contribute to this synergism in signalling, and our finding of c-met correlating with FGFR1 expression does support this view. Moreover, this hypothesis is supported by a study on the expression of c-met and the tyrosine kinase receptor RON in node-negative breast cancer patients . Ten-year disease-free survival was significantly decreased in patients with tumours expressing only one of the two markers, and it was even worse in patients with RON-positive/c-met-positive tumours. Because Sdc1 is an established coreceptor for several receptor tyrosine kinases, including FGFR1 [1, 3] and c-met , it could also contribute to synergistic signalling. We observed a trend toward a correlation of Sdc1 and HER2 expression in DCIS (P = 0.057). This is in accordance with the study conducted by Barbareschi and coworkers , which demonstrated positive correlation of Sdc1 and HER2 in a collective of 254 invasive breast carcinomas. In accordance with its role as a signalling coreceptor, a recent study conducted in 207 breast cancer patients demonstrated a significant association of Sdc1 overexpression with the Ki67 proliferation index .
In DCIS, an increase in periductal blood vessels has been correlated with recurrence of invasive disease , demonstrating the relevance of angiogenesis to tumour progression. Expression of the angiogenic factor VEGF-A in tumour cells from patients with DCIS was shown to correlate with the degree of angiogenesis . In the present study, we found a significant correlation of both c-met and E-cad with VEGF-A expression in DCIS. In addition, c-met expression exhibited a negative correlation with Flt-1 expression; Flt-1 is a negative regulator of VEGF availability, which is regarded as an antiangiogenic receptor. Our findings constitute the first characterization of a proangiogenic role for c-met, Sdc1 and E-cad in DCIS, but they are in accordance with published work on the biological role of these molecules. Ribozyme targeting of c-met in mammary cancer cells reduced mammary cancer and tumour-associated angiogenesis in a xenograft model . In addition, the HGF-antagonist NK4 inhibits c-met signalling and angiogenesis in vitro  and reduces intratumoural microvessel density in a hepatocellular carcinoma xenograft model . Sdc1 acts as a coreceptor for several angiogenic and growth factors, including multiple VEGF isoforms and bFGF [1, 3]. In vivo, we recently demonstrated increased corneal angiogenesis in Sdc-1 deficient mice and the formation of abnormally dilated blood vessels in skin wounds of Sdc-1 overexpressing mice, indicating a regulatory role for Sdc1 in angiogenesis [2, 4]. Moreover, in a nude mouse xenograft model, coinjection of MDA-MB-231 breast cancer cells with Sdc-1 overexpressing and mock-transfected mouse fibroblasts resulted in significantly elevated microvessel density and a larger vessel area in tumours containing Sdc1 overexpressing stroma cells . A TMA analysis of 207 human breast cancer samples by the same authors revealed that stromal Sdc1 expression correlated with vessel density and total vessel area, demonstrating a role for Sdc1 in breast tumour angiogenesis.
Although we observed a trend toward an association of Sdc1 expression with markers of lymphangiogenesis, we could not demonstrate a statistically significant correlation of Sdc1 expression and proangiogenic markers in DCIS (Table 5). This finding may suggest that the angiogenesis-associated functions of Sdc1 primarily contribute to more advanced stages of tumour progression. We have recently demonstrated that VEGF-C and Flt-4, which are established mediators of lymphangiogenesis [55, 56], were expressed in the tumour cells of a large proportion (88% to 95%) of DCIS specimens . In the present study, we found a significant association of E-cad with VEGF-C expression (P = 0.048). Althuogh very little is known about the role of E-cad in lymphangiogenesis, by virtue of its biological role an involvement in this process could easily be envisaged. Moreover, the correlations of VEGF-A/C with E-cad and of VEGF-A with c-met could be a sign of an angiogenic stimulation of the stroma by the tumour cells . Furthermore, the preferential expression of Sdc1 and c-met in pure versus coexistent DCIS parallels the significantly more common expression of proangiogenic factors in pure versus coexistent DCIS .
ET-1 and its G-protein-coupled receptors ETAR and ETBR (the endothelin axis) are overexpressed in breast carcinomas, and influence angiogenesis, tumourigenesis and tumour progression [34, 36, 58]. Moreover, we previously demonstrated overexpression of the ET axis in DCIS and observed an association with different angiogenic and lymphangiogenic factors . An important novel finding of the present study is the highly significant association of c-met and E-cad expression with ET receptor expression. Similar to ETAR , c-met expression is upregulated by hypoxia , thus promoting invasive growth of tumour cells. In addition, the c-met ligand HGF has been shown to inhibit ET-1 release in endothelial cells, indicating a possible link between c-met mediated signalling and the ET axis . However, the coexpression of c-met and the ET receptors may simply indicate an upregulation of parallel signalling pathways that utilize different upstream ligands and converge intracellularly. Sdc1 and E-cad have been associated with epithelial-mesenchymal transition [22, 40], and it has recently been shown that ET-1 mediated signalling is required during epithelial-mesenchymal transition in ovarian cancer progression [61, 62]. In ovarian cancer [61, 62] and melanoma [63, 64], activation of the ET axis results in downregulated E-cad expression. In DCIS, coexpression of ET receptors, E-cad, c-met and Sdc1 may constitute a specific expression signature indicative of the transition of an early stage to later stages of tumour progression.
One possible caveat associated with the present study is the large number of statistical tests that were performed (n = 17). Although the statistical analysis employed in this study permits easier comparison and interpretation of P values within the context of the results from other studies, the reader must keep in mind that some significant associations may be falsely positive. Using a Bonferroni adjustment, the corrected significance level for the tests employed in this study would be P < 0.003. However, although corrections for multiple comparisons on the one hand reduce the chances of making a type I error, the chances of making a type II error are increased. Moreover, the relevance of the null hypothesis underlying Bonferroni adjustments has recently been questioned (see the report by Perneger  for a discussion). Therefore, there remains a lack of consensus within the scientific community regarding whether corrections for multiple comparisons are applicable.