Metastasis is a complex, multistep process that begins with the epithelial–mesenchymal transition (EMT). Circulating tumor cells (CTC) are believed to have undergone EMT and thus lack or express low levels of epithelial markers commonly used for enrichment and/or detection of such cells. However, most current CTC detection methods target only EpCAM and/or cytokeratin (CK) to enrich epithelial CTCs, resulting in failure to recognize other, perhaps more important, CTC phenotypes that lack expression of these markers. Here, we describe a population of complex aneuploid CTCs that do not express CK or CD45 antigen in patients with breast, ovarian, or colorectal cancer. These cells were not observed in healthy subjects. We show that the primary epithelial tumors were characterized by similar complex aneuploidy, indicating conversion to an EMT phenotype in the captured cells. Collectively, our study provides a new method for highly efficient capture of previously unrecognized populations of CTCs.
Significance: Current assays for CTC capture likely miss populations of cells that have undergone EMT. Capture and study of CTCs that have undergone EMT would allow a better understanding of the mechanisms driving metastasis. Cancer Discovery; 1(7); 580–86. ©2011 AACR.
This article is highlighted in the In This Issue feature, p. 539
An epithelial–mesenchymal transition (EMT) in cancer is characterized by loss of cell adhesion, repression of E-cadherin, acquisition of mesenchymal markers, increased cell motility, and invasive potential (1). Though several approaches describing alternative strategies for recovery and detection of circulating tumor cells (CTC) have been reported, the only U.S. Food and Drug Administration (FDA)–approved technique for CTC detection relies on the use of antibodies targeting the epithelial cell adhesion molecule (EpCAM), followed by cytokeratin (CK) and CD45 staining to confirm an epithelial phenotype (2, 3). These modern enrichment techniques have characterized such CTCs, showing correlation with survival (4–6) and response to treatment (6, 7). CTCs can also express biomarkers reflective of the primary tumor and could be useful as a surrogate for tumor biopsy (8, 9). However, only a small proportion of CTCs are capable of establishing distant metastasis (10, 11) as evidenced by the phenotypic heterogeneity observed among CTCs within patient blood samples (8). Biologic mechanisms, such as EMT, may also result in a spectrum of epithelial marker shedding and a more progressive metastatic phenotype. Yet, these important EMT-derived CTC populations are likely missed by current techniques (12). In fact, one of the initial reports describing EpCAM levels in various types of cancer found that nearly all epithelial tumors have high expression, whereas tumors of mesenchymal lineage, such as melanomas and sarcomas, have no expression (13). It is unknown whether the most invasive and metastasis-primed cells that have undergone EMT (resembling a more mesenchymal phenotype) no longer express EpCAM and/or CK and thus evade detection with traditional techniques. More important, the relative paucity of clinical evidence that such a process exists has called the theory of EMT into question (14). Thus, the ability to capture post-EMT cells in circulation could have implications for the study of the metastatic process as well as clinical management of patients with cancer.
Capture of Carcinoma Cells Independent of EpCAM Expression
We have developed a microfluidic-based system for capture and analysis of rare cells in circulation, including CTCs (Supplementary Fig. S1A). The platform is capable of capturing rare cells from blood for subsequent molecular characterization directly within a uniquely designed microchannel (Supplementary Fig. S1B and S1C). Each microchannel consists of a roughly rectangular chamber (40 mm × 12 mm × 55 μm) in which approximately 9,000 variable diameter posts are randomly placed to disrupt laminar flow and maximize the probability of contact between target cells and the streptavidin-derivatized posts, resulting in their capture (15). The platform is versatile, permitting assay optimization with flexibility in the number of antibodies selected for capture and detection as well as allowing immediate and direct single-cell microscopic analyses through the transparent microchannel without the need to manipulate cells onto glass slides. This approach utilizes an antibody cocktail that is added directly to cells prior to capture, enabling recovery of variable CTC phenotypes. Briefly, this mixture contains antibodies directed toward a variety of epithelial cell surface antigens (EpCAM, HER2, MUC1, epidermal growth factor receptor, folate-binding protein receptor, TROP-2) and mesenchymal or stem cell antigens (c-MET, N-cadherin, CD318, and mesenchymal stem cell antigen). Each of these antibodies was tested by flow cytometry for reactivity to several well-characterized cancer cell lines (e.g., SKOV3, LNCaP, SKBR3) and shown to be additive in binding to cells (Supplementary Fig. S2). Antibodies were tested to have minimal cross-reactivity to nucleated cells in healthy control blood. Analytical control samples, namely blood from healthy donors spiked with cancer (SKOV3) cells, were run using this antibody mixture to ensure optimal performance in detecting tumor cells (based on CK staining).
Analytical validation of the platform to show precision, reliability, and reproducibility in recovery and detection of CTCs based on CK+/CD45−/DAPI+ (4′,6-diamidino-2-phenylindole) staining was done with ex vivo spiking experiments of carcinoma cells into whole blood. Tumor cell capture efficiency was validated through a series of cell-spiking experiments. We show analytic precision in recovery and detection of low- to high-EpCAM-expressing target cells, independent of the number of cells spiked (Fig. 1A). We further show >95% reproducibility in tumor cell capture with several cell lines (Fig. 1B). Capture efficiency using the antibody cocktail was shown with low (T24) and medium (SKOV3) EpCAM-expressing cell lines (16). While SKOV3 cells had high capture efficiency with EpCAM alone as well as with cocktail (>80%), T24 capture efficiency was markedly improved from <10% capture with EpCAM alone to >80% with antibody cocktail (P = 0.0001; Fig. 1C).
Next, clinical verification of capture efficiency was tested by examining CTCs in patient samples. In the first cohort of 21 patients with advanced-stage disease (colon, breast, lung, and prostate cancers), we compared CTC detection with our platform using either EpCAM alone or antibody cocktail (Supplementary Table S1). Improved recovery was observed with higher CTC numbers in 14 of 21 samples with the antibody cocktail (P = 0.07). In 100 normal blood controls, only 1 CK+/CD45−/DAPI+ cell was detected, demonstrating high specificity of the staining procedure. In the second, larger cohort of 93 patients with advanced-stage lung, breast, colorectal, or prostate cancer, we compared CellSearch® and our platform using the antibody cocktail for capture. Based on standard CK+/CD45−/DAPI+ stain criteria, our platform was found to be significantly more sensitive for CTC enumeration in colorectal, prostate, and lung cancers (Fig. 1D).
Capture and Detection of CK− CTCs in Patients with Breast, Ovarian, or Colorectal Cancer
Given that CK expression levels can often be variable among epithelial cells and absent among other nonepithelial cell types, we examined CK staining efficiency within the microchannel using a third cohort of HER2-neu-positive breast cancer samples (n = 19) for which FISH analysis of the HER2 locus enabled confirmation of CTC recovery independent of CK/CD45 staining (Fig. 2A). CK+ cells were located in each of the 19 cases and used as target cells for analysis of HER2 by FISH. In addition, all CK−/CD45− cells were classified as “possible” CTCs for subsequent analysis of HER2 signals. In 18 of 19 cases (94.7%), a range of 0.04 to 2.4 HER2-amplified CTCs/mL of blood were detected among the cells classified as CK+/CD45− and CK−/CD45−, suggesting that the CK− HER2-amplified CTCs originated from the primary tumor. When excluding CK− HER2-amplified CTCs, only 12 (63.2%) of the patient samples were found to have CK+ HER2-amplified cells, consistent with the findings of other similar reports (17). Interestingly, only 24.3% of all CK+ CTCs were found to have HER2-neu amplification (P = 0.007; Fig. 2B). Surprisingly, 49.7% of the HER2-amplified cells were CK− (Fig. 2B). These results show the inefficiency in detection of CTCs based solely on CK+/CD45− stain criteria, resulting in failure to detect a significant population of HER2-amplified cells. Thus, given the unique design and amenity to sequential staining and FISH, the platform enabled incorporation of FISH directly within the microchannel as a valuable independent method in confirming recovery of both CK+ and CK− CTCs immediately after CK staining.
To confirm further the use of the FISH-based strategy, we applied the same antibody cocktail capture approach and CTC detection (CK+/CD45−/DAPI+ staining) with samples from patients with advanced ovarian or colorectal cancer. To verify detection of CTCs, we again used FISH, but with the aim to detect aneuploidy (unbalanced genomic copy number changes). We rarely found cells stained with CK+/CD45−; roughly 0.1 cells/mL of blood from ovarian cancer patients regardless of stage, tumor grade, or CA-125 levels were CK+/CD45− (Supplementary Table S2). FISH probes targeted to frequently gained genomic regions in ovarian and colorectal carcinoma were chosen based on array comparative genomic hybridization data (18, 19). Each probe was first tested on a total of 2,500 normal peripheral blood cells from five donors and in CK+ cultured SKOV3 cells to determine hybridization efficiency and scoring accuracy in interphase cells captured within the microchannel (see Methods section). Cells that displayed only a monosomic signal for any one of the three probes were excluded (potential signal overlap). Therefore, only cells classified as either trisomic or complex aneuploid (at least one locus gain and another gain or loss at a second locus) were enumerated. As observed with the HER2-positive breast cancer cohort, for both ovarian and colorectal cancer, we found that not only captured CK+/CD45− cells (Fig. 2C), but also co-captured CK−/CD45− staining cells (Fig. 2D; Supplementary Tables S3 and S4) had complex aneuploidy. Complex aneuploidy was not observed among normal control blood samples. Similar to breast cancer patients, when assessing for frequency of complex aneuploidy, colorectal and ovarian cancer patients had nearly the same number of CK− as CK+ CTCs (Fig. 2E, F). Thus, the presence of CK− complex aneuploid cells further shows the inefficiency of CK as a marker to detect all candidate CTCs.
CK− CTCs Found within the Primary Tumor
We hypothesized that if the isolated CK− CTCs are the consequence of EMT, then similar cells should be present within the primary tumor. We sought to verify that the circulating aneuploid cells identified in patients with ovarian cancer had molecular features reflective of the primary tumor and, thus, are a useful population of CTCs. Blood was collected from 7 patients just before cytoreductive surgery. Multiple regions in the matched tumor and CTCs shared similar complex aneuploid patterns (Fig. 3A). Interestingly, about 20% of these aneuploid regions within the tumor were CK− and heterogeneous in distribution (Fig. 3B). Although similar findings using FISH in CTCs and circulating endothelial cells compared to primary tumors have been described (20), these were based on circulating CK+ cells.
Linking CK− CTCs to an EMT
One possible mechanism giving rise to the presence of CK− CTCs is EMT, a biologic process reported to play a significant role in tumor progression and metastasis. Thus, there is growing interest in methods that enable capture and analysis of EMT-derived CTCs. To determine whether our assay captures cells that have undergone EMT, we studied the effect of TGF-β (an inducer of EMT) treatment on SKOV3 ovarian carcinoma cells. After 72 hours of treatment, CK staining was lost in about 20% of cells, correlating with an EMT morphologic change (Fig. 4A). Quantitative PCR analysis of these cells before and after TGF-β treatment showed an increase in expression of mesenchymal markers (Fig. 4B). Cells, before and after treatment, were spiked ex vivo into mouse blood and run through the microchannel. All untreated cells captured were CK+; however, after TGF-β treatment, 16% of the cells captured were CK− and had complex aneuploidy (Fig. 4C). Direct visualization of the FISH staining within the microchannel shows these CK+ and CK− cells had nearly identical complex aneuploid patterns (Fig. 4D).
Next, to examine the utility of our CTC detection system for capturing tumor cells with EMT features in an in vivo setting, we studied the effect of TGF-β treatment on HeyA8 ovarian carcinoma cells. After 72 hours of TGF-β treatment, approximately 50% to 60% of cells lost their CK staining (Fig. 4E). To determine whether these CK− cells can be captured in circulation, we established a metastatic orthotopic model with HeyA8 cells. Ten tumor-bearing mice were monitored for signs of morbidity, at which point approximately 350 μL of blood was obtained per mouse by cardiac puncture prior to sacrifice. All sites of metastatic tumor were carefully removed and weighed. Both CK+ and CK− complex aneuploid CTCs were isolated from circulation within the microchannel (Fig. 4F). Enumeration of complex aneuploid CK− CTCs correlated with aggregate tumor burden (Fig. 4G).
Collectively, the results confirm the utility of our microfluidics platform as a reliable method for assay development and efficient recovery of CTCs. We observed that a potentially important population of cancer cells is present in circulation that would likely be missed by standard detection criteria. Identification of the full spectrum of CTCs would permit more efficient and directed analysis among patient specimens in which heterogeneous CTC populations are expected. Namely, detection of EMT-derived CTCs has been widely hypothesized as a population of cells that is missed by current platforms (12, 21). Our data suggest that these complex aneuploid CK− CTCs isolated in clinical samples may represent EMT-derived CTCs.
There are some recent reports of isolated CTCs expressing markers of EMT. For example, in metastatic breast cancer patients receiving standard therapies, CTCs correlated with more frequent expression of EMT markers (TWIST1, AKT2, and PI3Kα) in those who were resistant to treatment (22). Similar reports have found that CTCs can coexpress both epithelial (CK and E-cadherin) and mesenchymal (vimentin and N-cadherin) markers (23, 24). Likewise, a higher incidence of CK+vimentin+ and CK+TWIST+ CTCs was found in metastatic breast cancer patients versus women with earlier stage disease (25). These studies suggest a continuum in the spectrum of epithelial differentiation to mesenchymal phenotype, suggesting that CTCs may have a partial EMT phenotype (12). However, these platforms may miss clinically relevant populations of CTCs because they rely on CK for CTC detection. To our knowledge, we provide the first report linking CK− CTCs to EMT.
Although the antibodies used in our study detect a wide range of CKs, it is possible that these antibodies may not detect all CKs. Thus, it is possible that the complex aneuploid CK− CTCs we detected may still express other CKs. There is increasing evidence that different CK antibodies detect different subsets of CTCs. Currently, different groups are targeting a variety of CKs for detection of CTCs, such as CK8, CK18 and CK19 (26–28), and CK7 and CK8 (2, 29). However, there is increasing concern that targeting only a few CKs for CTC detection, such as CK18 (30) or CK7, CK8, and CK18 (31), may fail to detect certain tumor cells. To target many different CTC phenotypes, our CK mix detects luminal (CK7, CK8, CK18, CK19) and basal (CK5, CK6, CK14, CK17) CTCs. Although the pan-CK antibody used in our study has been successfully utilized by other groups (32), we added three additional CK antibodies (CK7/17, CK18, and CK19) to the detection mix.
Interestingly, we observed similar ratios of total CK+ CTCs to complex aneuploid CK+ and CK− CTCs in three common cancer types. For example, only 24.3% of CK+ CTCs in HER2/neu-positive breast cancer patients were found to have HER2 amplification. There have now been multiple reports assessing the HER2 status in CTCs from HER2-amplified breast cancer patients, most of which show a large proportion of captured CK+ CTCs being negative for HER2 amplification (33). This suggests that the classically defined CTC (CK+/CD45−/DAPI+) may overrepresent cells that are of tumor origin. The FISH probes used in colorectal and ovarian cancer patients are unlikely to detect all CTCs with complex aneuploidy, suggesting that some complex aneuploid CK+ and CK− cells may still be missed. However, our platform allows for adaptability in selection of capture antibodies as well as use of different FISH probes and detection antibodies for immunofluorescence. Other investigators are now also exploring methods beyond the use of EpCAM and CK-based antibody platforms for capturing and detecting previously unrecognized CTC populations (34).
The platform presented here enables efficient assay development given the ability to interrogate all recovered cells for confirmation of success in targeted cell capture (i.e., detection of HER2-amplified CK− cells). Although using only a limited number of FISH probes is unlikely to find all CK+ and CK− cells with complex aneuploidy, more sophisticated interphase analysis may be considered. To the extent that CTCs represent a window into a patient's tumor, we present a new method for capturing and studying previously unrecognized CTC populations. Studies are clearly warranted to further classify distinct subtypes of CTCs. Additional study of CK− CTCs may provide new insights into mechanisms of EMT and contribute to development of a clinically useful biomarker.
Blood samples were drawn from patients with ovarian, peritoneal, fallopian tube, breast, or gastrointestinal cancer or benign pelvic tumors according to an Institutional Review Board–approved protocol at the MD Anderson Cancer Center, where patients were being treated. Diagnoses of cancers and benign pelvic tumors were based on pathologic review. All blood samples from healthy donors who had no history of cancer were drawn after obtaining informed consent. Blood samples were collected into 8.5-mL Vacutainer tubes containing 1.5 mL of acid-citrate-dextrose (ACD Solution A Vacutainers; Becton, Dickinson and Company). Within 60 minutes of blood collection, 250 μL of anticlumping reagent (CEE-Sure™; Biocept) was injected into each tube. The samples were shipped to Biocept, where they were stored at room temperature and processed within 24 hours of collection in Biocept's College of American Pathologists (CAP)–accredited Clinical Laboratory Improvement Amendments (CLIA) laboratory.
In the enumeration comparison of the CEE versus CellSearch® platforms, three tubes (one CellSave tube and two ACD tubes containing CEE-Sure) of blood were collected from each of the 93 patients. Given that the CellSearch assay is FDA-approved for CTC enumeration using only 7.5 mL of blood, only one tube was obtained and delivered to an independent medical laboratory (Genoptix Medical Laboratory). For the microchannel platform, the overall assay recovery was interpolated between two tubes of blood. Each tube of blood was used to generate a cell pellet for capture within one microchannel. Each microchannel was scored for the presence of CTCs based on the standard stain criteria (CK+/CD45−/DAPI+). For the enumeration comparison between the CEE and CellSearch® platforms, the higher total number of CTCs detected in one of two tubes of blood was selected.
A detailed description of CTC capture and detection; FISH; immunofluorescence imaging; statistical analysis; and in vitro, ex vivo, and in vivo experiments is provided in the Supplementary Methods.
Disclosure of Potential Conflicts of Interest
F. Z. Bischoff, J.A. Mayer, T. J. Pircher, S. D. Mikolajczyk, K. L. Wong, and T. Pham are employees of Biocept Incorporated. None of the other contributors have any potential conflicts of interest.
Portions of this work were supported by the NIH (CA016672, CA110793, CA109298, P50 CA083639, P50 CA098258, CA128797, RC2GM092599, U54 CA151668), the Ovarian Cancer Research Fund, Inc. (Program Project Development Grant), the Department of Defense (DOD) (OC073399, W81XWH-10-1-0158, BC085265), the Marcus Foundation, the Blanton-Davis Ovarian Cancer Research Program, and the Betty Anne Asche Murray Distinguished Professorship. C.V. Pecot is supported by a grant from the National Cancer Institute (T32 training grant CA009666). J. Bottsford-Miller, R.L. Stone, and A.M. Nick were supported by NCI-DHHS-NIH T32 training grant (T32 CA101642).