Most immunocytochemical CTC detection technologies are based on a separation of CTCs from normal blood cells with EpCAM antibodies. The exact biological function of EpCAM is not fully understood and remains controversial. In some publications EpCAM is argued to act as an intercellular adhesion molecule, and loss of EpCAM expression therefore reduces cell-cell adhesion, thereby promoting dissemination of tumor cells . In contrast, Osta et al.  report that silencing EpCAM gene expression in vitro decreases the proliferation, migration and invasion potential of breast cancer cell line MD-MB-231. EpCAM might also have a dual role as cell adhesion molecule and receptor involved in the regulation of gene transcription and cell proliferation .
Our CTC detection method is based on a commercially available Carcinoma Cell Enrichment and Detection Kit which uses CK7/CK8 MicroBeads supplemented by EpCAM-MicroBeads. In former, very limited experiments with the commercially available kit based on CK-MicroBeads alone, a considerable amount of CTCs missed the magnetic attachment and was found in the flow through cell fraction (mean efficiency 44%). The objective of the CK/EpCAM combination was to improve detection rates for CTCs with low target antigen densities. Because the expression of EpCAM and cytokeratins in epithelial tumor cells can vary per se, phenotypes with low or missing expression of these epithelial specific antigens exist [28–30]. Based on this heterogeneous expression pattern CTCs can be classified into EpCAM+/CK-/low, EpCAM-/low/CK+ and EpCAM+/CK+[15, 16, 31]. E.g., Rao et al. investigated the expression of EpCAM and its co-expression with the epithelial cell specific markers CK 8/18 or CK 19 and Muc-1 in carcinoma cells present in blood . They found CK- CTCs in about 23% of blood samples of 30 carcinoma patients but only 0.2% EpCAM- CTCs. Yet, they demonstrated that the efficiency of immunomagnetic recovery with anti-EpCAM coated ferrofluids is rapidly declining with decreasing EpCAM antigen density. Furthermore, Mego et al.  give preliminary data on breast cancer patients where the loss of epithelial antigen on CTCs due to epithelial mesenchymal transition might be responsible for partially missed CTCs by the CellSearch System. Deng et al. processed blood samples from 49 metastatic breast cancer patients with the CellSearch™ system and in parallel by a combined anti-EpCAM- and anti-cytokeratin magnetic cells separation method, comparable to our method. They obtained a significantly higher CTC positive rate (49% vs. 29%) and a larger dynamic CTC detected range (1 to 571 vs. 1 to 270) than that of the CellSearch™ system . Another recent study indicated that the so-called normal genotype of invasive breast cancer, which accounts for approximately 10% of all cases, is typically negative for EpCAM expression and may thus be a cause of false-negative CTC determinations .
To estimate the accuracy, linearity and precision of our CTC detection method, the breast cancer cell line HCC1937 was spiked into the blood of healthy donors. The average percentage of cell recovery was 84% which is well within the recovery range of 60 to 85% of comparable published CTC detection methods [1, 15, 32–34]. The within-run coefficients of variation (CV) for the spiking of cells (12, 18, 12 and 30% for 185, 85, 30 and 10 cells, respectively) were comparable to CVs detected by Allard et al.  (8.2, 15.4, 22 and 47% for 319, 58, 18 and 4 cells, respectively). The variation of recovered cell numbers can partially be explained by the difficulty to accurately and reproducibly spike low numbers of cells that tend to form adherent clusters. Although a large volume of the buffy coat was collected, we can not entirely exclude the loss of cells due to this step.
For our assay, criteria for classification as circulating tumor cells were round or oval morphology, positive staining for cytokeratin and negative corresponding isotype control. Because the standard kit does not include CD45 labeling to specify leukocytes and our method is a nonfluorescent immunocytochemical method that allows the chromogenic detection of only one marker simultaneously, the unambiguous characterization of CTCs as nucleated, CD45 negative cells was not possible. However, after optimizing our assay in proof-of-concept experiments we were able to reduce the non-specific binding of antibodies to 0%, fulfilling the criteria for the evaluation of rare immunocytochemically identifiable cancer cells, defined by the European ISHAGE Working Group 1999 .
Using blood samples of 59 MBC patients and a standard threshold of 5 CTCs to define 2 risk groups, we can confirm the high prognostic value of this dichotomization. But, we hypothesized that there is more information in low CTC levels than generally recognized and defined a medium risk group CTC-M between the low risk group CTC-L with 0 CTC and the high risk group CTC-H with ≥5 CTC. The 3-valued marker CTC-LMH proved to be significant as well and in a statistical comparison of the model with 3 vs. 2 risk groups, the model with 2 risk groups was rejected. It is theoretically possible that the medium survival risk of patients with medium CTC-level is only because this group contains more patients with events falsely classified as CTC than the high risk group. Tibbe et al.  used a theoretical model with equal death risk for all patients with true CTC in their blood circulation (regardless of CTC level) and got rather extreme (unrealistic) results. The only analyses with more than 2 risk groups we know are described in [6, 19] where 1, 5 and 20 were used as cut-off values. In their study the corresponding survival curves also exhibited low, medium and high risk corresponding to CTC-level, therefore suggesting a continuous nature of association although after approximately 5 CTCs the increase rate lessened. Fehm et al. and Botteri et al. recently discussed the relationship between CTC as a count variable and clinical outcome [19, 20]. But yet, to our knowledge there has not been a broad debate on the existence of more than 2 risk groups or on the kind of relationship of CTC-level as a count variable to survival. We believe that such a debate would help to assess the clinical impact of medium changes in CTC level and it could provide a basis for a better understanding of disease stages or activity.
It is a limitation of our study that due to the retrospective collection of baseline data from medical records a considerable proportion of this data is missing. But, the analysis of demographic and baseline data in our study confirms known relationships of CTCs in peripheral blood to other variables. Similar to our study Müller et al.  report about an elevation of CA 15-3 in patients with detectable CTCs, probably a reflection of metastatic tumor burden. According to Nole et al.  the presence of ≥5 CTCs at baseline was associated with a higher number of positive lymph nodes, with an elevated CA 15-3 value, with non-overexpressing Her2/neu tumors and with the presence of bone metastases at baseline. The association of high CTC counts with the presence of bone metastases was shown in patients with different metastatic cancers [6, 38–42]. As most patients with bone metastases were found in the high risk group it is coherent that this group most frequently receives Bisphosphonate treatment which is a standard treatment of malignant bone diseases. In our study patients with measurable CTCs had also elevated systemic CRP levels. Elevated biomarkers of inflammation like CRP are associated with reduced survival among breast cancer patients  which suggest that CRP may be related to tumor burden or progression and that chronic inflammation promotes mammary tumor development.