Research ArticleCancer

Hematogenous dissemination of glioblastoma multiforme

See allHide authors and affiliations

Science Translational Medicine  30 Jul 2014:
Vol. 6, Issue 247, pp. 247ra101
DOI: 10.1126/scitranslmed.3009095


Glioblastoma multiforme (GBM) is the most frequent and aggressive brain tumor in adults. The dogma that GBM spread is restricted to the brain was challenged by reports on extracranial metastases after organ transplantation from GBM donors. We identified circulating tumor cells (CTCs) in peripheral blood (PB) from 29 of 141 (20.6%) GBM patients by immunostaining of enriched mononuclear cells with antibodies directed against glial fibrillary acidic protein (GFAP). Tumor cell spread was not significantly enhanced by surgical intervention. The tumor nature of GFAP-positive cells was supported by the absence of those cells in healthy volunteers and the presence of tumor-specific aberrations such as EGFR gene amplification and gains and losses in genomic regions of chromosomes 7 and 10. Release of CTCs was associated with EGFR gene amplification, suggesting a growth potential of these cells. We demonstrate that hematogenous GBM spread is an intrinsic feature of GBM biology.


Gliomas are intrinsic tumors of the central nervous system, and among them, astrocytoma WHO grade IV or glioblastoma multiforme (GBM) is the most common tumor in adults. It is associated with very limited life expectancy after diagnosis, ranging in median between 12 and 15 months (1, 2). These tumors are considered to be confined to the central nervous system because extracranial metastasis of glioblastomas rarely occurs (0.4 to 0.5% of patients), affecting mostly bone, lymph nodes, lungs, and liver (35). Several reasons for this low incidence of extracranial metastasis have been discussed, including short survival periods (2), the blood-brain barrier (6), suppression of extracranial growth of glioblastoma cells by the immune system (3), the absence of lymphatic channels in the central nervous system (7), or the inability of GBM cells to invade and loosen connective tissue in the extracranial space (8).

Clinically dormant, extracranial metastasis is suspected to occur more frequently than reported in GBM patients because neuro-oncologists do not systematically search for systemic disease during the short survival period (3, 9). Because of new therapeutic approaches, survival of some GBM patient subgroups has been increasing over the past decades, potentially giving extracranial tumor cells more time to develop into distant metastases (3). Considering that GBM cells may need years to adapt to the different microenvironmental conditions of extracranial organs, the question arises whether GBM cells circulate in peripheral blood (PB) more frequently than the reported incidence of extracranial metastases implies. Thus far, there is only one report addressing this issue, which was published 10 years ago and failed to detect GBM cells in PB, probably because of insufficient technology and the low number of patients analyzed (10).

In the present investigation, we established a sensitive immunocytochemical detection system and analyzed PB samples applying glial fibrillary acidic protein (GFAP) as a marker for GBM cells. To further prove the GBM origin of GFAP-positive cells in PB, we identified genomic aberrations by comparative genomic hybridization (CGH), sequence analysis, and fluorescence in situ hybridization (FISH) in putative circulating tumor cells (CTCs) and corresponding tumor tissues of two patients. We screened blood samples for GFAP-positive cells before, during, and/or after tumor resection to assess whether hematogenous tumor cell dissemination is an intrinsic property of GBM biology or is associated with tissue manipulation during surgical intervention.


Incidence of GFAP-positive cells in PB in relation to patient and tumor characteristics

In total, 141 patients with GBM were enrolled in the study, including 30 patients suffering from first or second GBM recurrence. Regardless of the time of blood drawing, GFAP-positive cells were detected in 29 of 141 patients (20.6%; Fig. 1, A to D). The number of these cells varied between 1 and 22 among 2.1 × 106 mononuclear cells (MNCs) (table S1). The incidence and concentration of GFAP-positive cells did not significantly differ between blood samples from patients presenting with primary tumors versus recurrences (P = 0.6, Table 1). Within a median of 17.3 months (mean, 13.1 months) of follow-up and/or life span after diagnosis of this cohort of patients, no extracranial metastases were observed. Furthermore, the presence of GFAP-positive cells did not correlate to the overall survival of the patients, irrespective of the time point of blood collection (fig. S1).

Fig. 1. Detection of CTCs and EGFR status in corresponding primary tumors.

(A and B) Putative CTC from patient #63 (red: positive for GFAP, tetramethyl rhodamine isothiocyanate; green: negative for CD45, fluorescein isothiocyanate; blue: DAPI-stained nucleus; right images: GFAP/CD45/DAPI composition) before (A) and after (B) tumor resection. Scale bars, 20 μm. (C) GFAP-positive CTC from patient #3 [red, alkaline phosphatase anti–alkaline phosphatase (APAAP) immunostaining with New Fuchsin as chromogenic substrate, counterstaining of nuclei with hemalum]. Scale bar, 10 μm. (D) Dark blue CTC detected by APAAP immunostaining using NBT-BCIP (5-nitro blue tetrazolium–bromo-4-chloro-3-indolyl-phosphate) as substrate for alkaline phosphatase. CTC was transferred with a custom-made capillary. (E and F) Primary tumor tissues presented with a strong amplification of the EGFR gene detected by FISH [EGFR-specific hybridization signals: (E) patient #3, green; (F) patient #63, red; DAPI, blue]. Scale bars, 10 μm. (G and H) EGFRvIII expression in the primary tumors detected by immunohistochemistry on paraffin-embedded tissue sections: (G) patient #3, (H) patient #63. Scale bars, 20 μm.

Table 1. Detection of CTCs and patient/tumor characteristics.
View this table:

Only one GFAP-positive/CD45-negative cell in total could be detected in the blood of control patients (23 noncancer controls and 5 cancer patients with brain metastasis), demonstrating the high specificity of our blood assay and indicating that GFAP-positive cells are putative CTCs. GFAP-positive cells were also detected in PB from patients whose tumors displayed a partial loss of GFAP expression (n = 11, Table 1), supporting the high sensitivity of our CTC assay.

We did not observe significant differences between detection rates and counts of GFAP-positive cells before, during, or after tumor resection. Detection of GFAP-positive cells before and after tumor resection was observed in 9 of 67 cases with blood collection at both time points (13.4%). In 4 of 67 cases (6.0%), GFAP-positive cells could be detected before, but not after, tumor resection, whereas in 5 of 67 cases (7.5%), those cells were only found after tumor resection (table S1).

Because EGFR gene abnormalities, including amplification and rearrangements, frequently occur in GBM (11, 12), we subsequently analyzed whether the EGFR gene status of GBM was related to the release of tumor cells into the circulation. Recent work indicates that EGFR expression might identify tumor-initiating cells in GBM, and is therefore pivotal for gliomagenesis (13). EGFR gene amplification was assessed in tumor tissues by quantitative real-time polymerase chain reaction (qPCR) and additionally by FISH in tumor tissues from patients with GFAP-positive cells in PB. GFAP-positive cells in PB were more frequently detected in patients with EGFR gene amplification in the corresponding tumor tissues than in patients with non–EGFR-amplified tumors [P = 0.041, for >2.2-fold EGFR amplification (Table 1 and Fig. 1, E and F)]. Detection of the EGFRvIII deletion variant by immunohistochemistry (Fig. 1, G and H) was significantly associated with EGFR gene amplification higher than 2.2-fold in tumor tissue (P < 0.0001). Whereas for 61 of 83 analyzed tumor tissues (73.5%), concordant results were obtained for the presence or absence of EGFR gene amplification and the EGFRvIII-deleted protein, EGFRvIII protein could be detected in only 3 of 83 cases (3.6%) without EGFR gene amplification, and 19 of 83 cases (22.9%) demonstrated EGFR amplification without detectable EGFRvIII.

Characterization of circulating GFAP-positive cells

To further consolidate the hypothesis that GFAP-positive cells in blood are indeed tumor cells, we characterized single GFAP-positive cells (also called CTCs in the following text) enriched from two GBM patients by sequencing and/or CGH. We first searched for DNA mutations in the primary tumor of patient #63 by next-generation deep sequencing of 370 cancer-associated genes. From more than 1200 identified variants, 671 were located in exonic regions [+2 base pairs (bp)]. A total of 24 variants were prioritized as potential pathogenic variants. After visual evaluation of sequence alignments, 15 of these variants (table S2) were chosen for validation with Sanger sequencing and further analyzed in corresponding putative CTCs and constitutional DNA. The p.P18S mutation in the MECOM gene was also detected in two of these GFAP-positive cells (CTC1 and CTC1A, Fig. 2A), and the p.E1892D mutation in the MYH11 gene was found in another GFAP-positive cell from the same patient (CTC4, Fig. 2B). These mutations were somatic because they were not detected in constitutional DNA of pooled PB cells from the same patient (Fig. 2, A and B).

Fig. 2. Characterization of CTCs and corresponding primary tumors.

Comparative validation of two mutations on three CTCs (CTC1 and CTC1A before surgery; CTC4 after surgery) on the primary tumor (PT) and on a pool of PMNCs (peripheral mononuclear cells) from patient #63. (A and B) Mutations (position indicated by blue arrows; red arrows indicate wild-type nucleotide) in the MECOM gene (A) in the primary tumor, CTC1, and CTC1A, but not in CTC4 or PMNC, and mutations in the MYH11 gene (B) in the primary tumor and CTC4, but not in CTC1, CTC1A, or PMNC. (C and D) Selected profiles for chromosomes 7, 12, and 22 after chromosomal CGH (C) of a CTC detected after surgery and (D) of the primary tumor (patient #3, green: gains of chromosomal regions; red: losses of chromosomal regions). The number of chromosomes included in the evaluation is given in parentheses. (E) Chromosomal aberrations detected by array-CGH of WGA products from CTC1 and CTC4 and of DNA from the primary tumor (patient #63, green: gains of chromosomal regions; red: losses of chromosomal regions). (F and G) EGFR gene status detected by FISH (EGFR, red; centromere 7, green) in one CTC before (F) and one CTC after (G) surgery, surrounded by leukocytes. (H to J) Nucleated CTC (H; DAPI, blue) with EGFR (I; green) and GFAP (J; red) immunostaining. (K) DAPI/GFAP/EGFR composition. (L and M) More than three chromosome 3–specific signals in individual tumor cells were detected by FISH in a paraffin-embedded specimen (centromere 3, green; centromere 15, red). Scale bars, 10 μm.

To find out whether MECOM and MYH11 are detectable in GBM, we performed Western blot analysis in a collection of glioblastoma cell lines and found that both MECOM and MYH11 are expressed (fig. S2). Mutations in the other selected genes, such as GBM-related NF1 and PDGFAR, could not be found in the corresponding putative CTCs (table S2).

We further performed chromosomal and array-CGH on whole genome amplification (WGA) products from individual GFAP-positive circulating cells and DNA isolated from the corresponding primary tumors of cases #3 (Fig. 2, C and D) and #63 (Fig. 2E). A GFAP-positive cell from PB (Fig. 2C) and the primary tumor (Fig. 2D) from case #3 shared common genomic aberrations, such as gains in chromosomes 7 and 12 or losses in chromosome 22. Gains in chromosome 7p are consistent with a strong EGFR gene amplification that was observed in the primary tumor of patient #3 (Fig. 1E) by FISH analysis. Common gains in chromosomes 3, 7, and 12 and losses in chromosomes 10, 13, and 22—all characteristic for gliomas—were also found in CGH profiles from CTC1 and/or CTC4 and the corresponding primary tumor (Fig. 2E) of patient #63.

Individual CTCs and tumor tissue cells of the same patient, however, were heterogeneous with respect to the rate of EGFR gene amplification (Figs. 1, E and F, and 2, F and G). Strong EGFR protein expression in a GFAP-positive cell from PB of patient #63 is shown in Fig. 2, H to K. Heterogeneity for amplification of chromosome 3 was also observed between different circulating GFAP-positive cells and the primary tumor from the same patient (Fig. 2E). To investigate whether the primary tumor consists of cells with and without chromosome 3 amplifications, we performed a FISH analysis of the respective primary tumor tissue from patient #63. FISH analysis allowed us to determine the amplification status at the single-cell level in the primary tumor. Indeed, we detected a small subset of tumor cells with an amplified chromosome 3 in the primary tumor tissue (Fig. 2, L and M), which was missed by our initial CGH analysis of bulk tumor tissue, and this subset might be the source of CTC1 and CTC4.


This study demonstrated that CTCs occur much more frequently in PB from GBM patients than expected from the occurrence of overt extracranial metastases. We used immunostaining against GFAP as a GBM marker and detected CTCs in 29 of 141 (20.6%) GBM patients. Noteworthy, we did not find evidence for a CTC release induced by surgery. The incidence and number of GFAP-positive cells observed in GBM patients is comparable to CTC counts detected in carcinoma patients without overt metastases (14). Because GFAP-positive cells harbored glioblastoma-associated genomic aberrations, we conclude that these cells in the PB of GBM patients are CTCs. We observed an association between EGFR amplification and release of CTCs, indicating that EGFR signaling might support the extracranial spread of GBM.

For tracking circulating GBM cells, we selected GFAP as a marker distinguishing putative CTCs from normal blood cells. At present, GFAP, the main intermediate filament in mature and developing astrocytes (15), is the most specific diagnostic marker for GBM cells, and serum concentrations of GFAP as measured by enzyme-linked immunosorbent assay have been suggested as a diagnostic marker for GBM (16, 17). However, GFAP expression is not totally restricted to glial cells (18, 19), and it was therefore important to exclude that a minor subpopulation of GFAP-positive cells exists in the PB of non-GBM patients by the analyses of blood from healthy volunteers and non-GBM patients used as negative control groups.

To find out whether GFAP-positive cells detected in PB are indeed CTCs and to get insights into their genomic characteristics, we applied next-generation sequencing analysis specifically developed for molecular screening of individual CTCs (20). We observed common genomic aberrations in putative CTCs and GBM tumors, including mutations in the MECOM and MYH11 genes. Mutations and/or gene fusions of MECOM (EVI-1) mostly have been associated with the development of leukemia (21). However, there are also reports linking genetic aberrations and overexpression of the proto-oncogene MECOM to solid tumors such as ovarian (22) or colorectal carcinomas (23, 24) and mutations in the MYH11 gene to the hereditary nonpolyposis colorectal cancer syndrome (25). Thus far, no mutations of MECOM and MYH11 were described in glioblastoma, but Koos et al. (26) analyzed 19 GBM by immunohistochemistry and found that 2 cases were MECOM-positive. We could also demonstrate that both MECOM and MYH11 are expressed at least in some glioblastoma cell lines, although without detected mutations. However, the detection of these mutations in tumor tissue and CTCs simultaneously within our study mainly served the purpose of proving the tumor cell origin of the detected CTCs.

Nevertheless, even our current CGH and FISH analyses already showed that CTCs in GBM patients harbor several GBM-associated genomic aberrations, such as amplification of the EGFR gene, encoding a putative but still elusive therapeutic target (27). Consistent with the observation that EGFR-amplified cells are preferentially located at infiltrating edges of the GBM tumors (28), early observations already linked EGFR amplification to biological aggressiveness (29). We found an association between EGFR gene amplification in GBM tissues and the presence of GFAP-positive cells in PB. Previous studies already reported heterogeneity of receptor tyrosine kinase gene amplification in GBM cells (30). Consistently, in our study, individual single cells within the tumor tissues showed a very heterogeneous pattern of EGFR amplification, whereas putative CTCs from the same patients frequently presented with high EGFR gene amplification. Thus, EGFR signaling might support the release and/or survival of CTCs in the circulation, and antibodies against EGFR could become suitable tools to capture CTCs in future studies. In our study, EGFR amplification in tumor tissue was associated with the presence of the EGFRvIII variant that is supposed to drive transformation and tumorigenesis in a cell-intrinsic manner (31). Further characterization of EGFR-amplified CTCs with regard to the presence of the EGFRvIII deletion variant might also identify GBM cells that are resistant to chemotherapy (32).

Increased expression and genomic aberrations of EGFR are characteristic features of the classical subtype of GBM. Next-generation sequencing of tumor tissue–derived DNA from our patient #63 revealed subsets of cells that also had mutations in NF1 and PDGFRA (table S2), defining the mesenchymal and proneural subtypes, respectively (12). Moreover, loss of genomic regions of chromosome 10, containing the tumor suppressor gene PTEN that is also important for subtype analysis (33), was found in both analyzed cases and the corresponding CTCs analyzed by array-CGH. As described previously, intratumoral heterogeneity of the current markers applied to distinguish subtypes is frequent (34, 35), and our results imply that small GBM subclones can also contribute to tumor cell dissemination. Comprehensive analysis of these subclones might be pivotal for understanding tumor cell dissemination and the fate of CTCs outside the brain.

Mechanisms controlling the outgrowth of GBM cells at extracranial sites are unclear and subject to future investigations. Because transmission of GBM cells and outgrowth into fatal overt metastases has been observed as a complication of organ transplantation (3638), it is suggested that GBM cells may lodge in extracranial organs of donors in a dormant nonproliferative state. The results from previous clinical reports suggest that immunosuppression after organ transplantation suffices to allow these cells to survive in an active proliferative state in patients receiving organs from GBM donors (39, 40). Although the exact incidence is not known, it is estimated that between 10 and 20% of recipients from GBM organ donors develop extracranial tumors in organs such as kidney, liver, pancreas, and heart (36). This incidence almost matches the CTC detection rate observed in the present study, suggesting that at least a subset of circulating GBM cells might survive and home to a secondary organ, where their outgrowth might be controlled by the immune system. Glioma patients present with intracranial immunosuppression mediated by multiple cell types and mechanisms (41). Understanding the mechanisms of an apparently more effective extracranial control of the outgrowth of GBM cells might be valuable for designing more effective immunotherapies.

However, the present study has some limitations. Unfortunately, no common specific markers for GFAP-negative GBM tumor cells are available yet. Therefore, with new reliable marker-free approaches for CTC detection on the horizon, it will be interesting to investigate whether GFAP-positive and GFAP-negative tumor cells in tumors heterogeneous for GFAP expression have different capacities to disseminate and especially to enter the PB circulation. We also do not know if EGFR amplification correlates with GFAP expression and whether GFAP-negative CTCs (if they exist) express EGFR. Moreover, new approaches to improve the yield of CTCs in GBM patients, such as by diagnostic leukapheresis or in vivo capture of CTCs (14, 42), are pivotal to improve the number of CTCs available for molecular characterization. This will also allow performance of functional studies that are warranted to validate the role of the genes identified in CTCs and hematogenous GBM cell dissemination. Consequently, validating our findings in experimental animal models would be an essential next step to corroborate the hypothesis generated by the present proof-of-principle study.

The present report might lead to a paradigm shift and induce increased awareness that extracranial metastases may occur in GBM patients. Even without any significant correlation to survival, our key finding on a cohort of 141 patients that intact GBM cells can enter the circulation has important translational implications both for organ transplantation and for studies on prevention of distant metastasis in GBM patients. Notably, routine screening for extracranial metastases in GBM patients is not performed in clinical practice. Thus, metastases that do not cause clinical symptoms within the limited life span of GBM patients can be easily overlooked. A rapid metastatic progression within months after initial diagnosis is also infrequent in other types of solid tumors known to disseminate through the blood circulation, such as breast cancer. Thus, it cannot be excluded that in particular, the CTCs characterized by the presence of EGFRvIII or harboring an amplification and overexpression of EGFR might contribute to extracranial metastases once the life span of GBM patients will be increased by more efficient locoregional therapies. Checking the status of metastasis beyond the central nervous system might be indicated for GBM patients who are long-term survivors, and GBM patients without CTCs might be considered as transplant donors. Because of the urgent demand of organs for transplantation, the suitability of patients with GBM as organ donors is being discussed (4345). To date, the UNOS (United Network of Organ Sharing) claims that the use of organs from GBM patients is reasonable, but should be decided on a case-by-case basis (45). The detection of GBM cells in various distant organs is cumbersome and nearly impossible. This problem might be circumvented by the detection of CTCs in PB, a so-called liquid biopsy, to identify patients who should be excluded from donor lists. The CTC detection approach developed in this study is easily applicable as a companion diagnostic to identify patients with extracranial tumor cell spread, and it might be advisable to exclude these patients as organ donors.


Study design

PB samples (n = 147, including 30 GBM recurrences, 6 of them with matched primary tumors) from 141 patients with histologically confirmed GBM grade IV, treated at the Department of Neurological Surgery, University Medical Center Hamburg-Eppendorf, from May 2005 to August 2013 (Table 1) were drawn from central intravenous lines, which are part of the regular instrumentation for general anesthesia. Samples were taken during the initial phase of surgery, craniotomy, and dural opening but before tumor removal (“before resection,” n = 78), during the dissection/removal (“during resection,” n = 32), or after hemostasis was achieved and dural closure began (“after resection,” n = 106). Sections from paraffin-embedded tumor tissues were provided by the Institute of Neuropathology, University Medical Center Hamburg-Eppendorf. Fresh-frozen tumor tissue was available from 141 cases.

Furthermore, PB samples from 23 healthy volunteers provided by the Department of Transfusion Medicine, University Medical Center Hamburg-Eppendorf, and from 5 patients with carcinoma-derived brain metastases, treated in the Department of Neurological Surgery, University Medical Center Hamburg-Eppendorf, were collected. The investigation was conducted according to institutional and ethical guidelines, including written consent, and approved by the Ethics Committee of the Chamber of Physicians, Hamburg, Germany within the ERC-2010-AdG_20100317 DISSECT project.

Sample preparation

Blood (10 ml) was drawn into Monovettes containing EDTA (1.6 mg/ml) (Sarstedt). MNCs were enriched by Ficoll density gradient centrifugation. Cytospins were prepared by spinning every 7 × 105 MNCs onto Superfrost/Plus slides (Karl Hecht KG). Subsequently, slides were air-dried overnight and stored at −80°C until further processing.

Tumor cell enrichment

Each blood sample was diluted with cold Hanks’ balanced salt solution (Biochrom) and centrifuged at 400g for 10 min at 4°C. After removing the supernatant, cold phosphate-buffered saline (PBS; Life Technologies) was added to a final volume of 30 ml. The mixture was layered over 20 ml of Ficoll-Paque Plus (GE Healthcare). After centrifugation at 400g for 30 min without brake, MNCs were collected from the interphase layer and transferred together with the supernatant into a new 50-ml tube. Centrifugation at 400g for 10 min at 4°C was followed by resuspending the MNC pellet in PBS. After lysis of erythrocytes with H-lyse buffer (Human Erythrocyte Lysing Kit, R&D Systems), cells were counted and 7 × 105 viable MNCs were centrifuged onto Superfrost Plus glass slides (Karl Hecht KG). Supernatants were discarded, and the slides were allowed to air-dry overnight. The slides were stored at −80°C until staining.

Detection of GFAP-positive cells in PB

Fluorescence immunocytochemistry was performed on three slides from each case, resulting in a total screened amount of 2.1 × 106 MNCs. For detection of putative CTCs, a polyclonal rabbit anti-GFAP antibody (29.9 μg/ml, Dako) was applied. An Alexa Fluor 546 goat anti-rabbit immunoglobulin (IgG) antibody (heavy + light chain, 10 μg/ml, Invitrogen) was used as the detection antibody. Additionally, an Alexa Fluor 488–labeled anti-human CD45 monoclonal antibody (0.13 μg/ml, clone HI30, BioLegend) was used to exclude hematopoietic MNCs. Counterstaining of nuclei and mounting of cells were performed with Vectashield mounting medium (Vector Laboratories) containing 4′,6-diamidino-2-phenylindole (DAPI; 0.5 μg/ml). A circulating glioma cell was defined as a nucleated cell that was GFAP-positive and CD45-negative.

Alternatively, circulating glioma cells were detected by immunocytochemistry with a chromogenic substrate. Here, after applying the primary anti-GFAP antibody, an alkaline phosphatase goat anti-mouse/anti-rabbit immunoglobulin cocktail from the Envision System (Dako) was used. After visualization with the New Fuchsine substrate solution (46) and counterstaining of nuclei with Mayer’s hemalum solution (Merck), cytospins were analyzed by the automated cellular imaging system ACIS (ChromaVision).

For micromanipulation and subsequent genetic analysis of GFAP-positive cells, we performed immunostaining with the Alkaline Phosphatase Conjugate Substrate Kit (Bio-Rad Laboratories). We used the primary anti-GFAP antibody mentioned above, followed by the polyclonal goat anti-rabbit–alkaline phosphatase IgG (21 μg/ml, Dako) as secondary antibody. To visualize the antibody binding, we used BCIP in conjunction with NBT for colorimetric detection of alkaline phosphatase activity. For specificity control of each staining procedure, we used cells of the glioblastoma cell line GS-22. Positive staining was independently evaluated by two observers.

Cell line

The glioblastoma cell line GS-22 (47), established and provided by the Department of Neurological Surgery, was cultured in Neurobasal medium supplemented with 2% B-27 supplement, 2 mM l-glutamine, 1% penicillin/streptomycin (all Invitrogen), 1% amphotericin B (fungizone, Atlanta Biologicals Inc.), human fibroblast growth factor 2 (20 ng/μl; Miltenyi Biotec GmbH), and Heparin-Natrium-25000-ratiopharm (5 μl/ml; Ratiopharm).

Immunohistochemistry to detect EGFR expression in tumor tissue

After dewaxing of sections from paraffin-embedded tumor tissue for 2 hours at 60°C, then washing steps in xylene (J.T.Baker) and a descending alcohol series, cells were incubated for 6 min in proteinase K solution (Dako). Afterward, H2O2 blocking was performed (Real Protease Blocking Solution, Dako), and slides were incubated with the primary anti-EGFR antibody (6.14 μg/ml, Dako) overnight at 4°C. Bound antibody was detected by the Dako Real EnVision Detection System (Dako) following the manufacturer’s instructions. After counterstaining of nuclei with Mayer’s hemalum solution (1:5 in distilled water, Merck), cells were subsequently dehydrated in graded alcohol and xylene and finally embedded in Eukitt (Kindler GmbH).

Immunohistochemistry to detect EGFRvIII expression in tumor tissue

After dewaxing of sections from paraffin-embedded tumor tissue for 2 hours at 60°C, then washing steps in xylene (J.T.Baker) and a descending alcohol series, slides were pretreated with citrate buffer (BioGenex) for 5 min at 120°C. Afterward, the slides were incubated with the EGFRvIII-specific mouse IgG1 monoclonal antibody (clone L8A4) (47) overnight at 4°C. Bound antibody was detected by the Dako Real Detection System Peroxidase/DAB+, Rabbit Mouse (Dako) following the manufacturer’s instructions. After counterstaining of nuclei with Mayer’s hemalum solution (1:5 in distilled water, Merck), cells were subsequently dehydrated in graded alcohols and xylene and finally embedded in Eukitt (Kindler GmbH).

Immunocytochemistry to detect EGFR expression in CTCs

To detect EGFR expression in CTCs, double immunocytochemistry was performed on cytospins containing 7 × 105 MNCs, applying the anti-GFAP antibody (29.9 μg/ml, Dako) as well as the anti-EGFR antibody (6.14 μg/ml, Dako). For visualization of GFAP-expressing cells, we used the Alexa Fluor 546 goat anti-rabbit IgG antibody, and for EGFR-expressing cells, we used the Alexa Fluor 488 goat anti-mouse IgG antibody (both heavy + light chain, 10 μg/ml, Invitrogen) as the secondary antibody. Counterstaining of nuclei and mounting of cells were performed with Vectashield mounting medium (Vector Laboratories) containing DAPI (0.5 μg/ml).

Isolation of DNA from fresh-frozen tumor tissue

After homogenization of frozen tumor tissue in liquid nitrogen, we used the DNA Mini Kit (Qiagen) to isolate DNA according to the manufacturer’s instructions.

Transfer of CTCs by micromanipulation

Isolation of GFAP-positive single cells was performed by micromanipulation with the CellTram Vario Microinjector and Eppendorf TransferMan NK 2 micromanipulator equipment (Eppendorf) and the inverted Axiovert 40 C microscope (Carl Zeiss MicroImaging). CustomTips Type III capillaries (Eppendorf) with a beveled end (Fig. 1D) and an inner diameter of 40 μm were applied to transfer these cells into caps of 200-μl Eppendorf tubes containing 2 μl of deoxyribonuclease-free water (PCR grade). Finally, after centrifugation, the Eppendorf tube was stored at −20°C until WGA (48).

Whole genome amplification

DNA from single cells was amplified with the illustra GenomiPhi V2 DNA Amplification Kit (GE Healthcare) as described elsewhere (49). In brief, after thawing the single cell–containing water drop, we added 10 μl of GenomiPhi V2 Sample Buffer, including 1 μl of protease (1:10, 7.5 Anson unit, Qiagen). In a GeneAmp PCR System PE cycler (Applied Biosystems), the samples were consecutively incubated at 50°C for 15 min, at 70°C for 15 min, and at 95°C for 2 min. After cooling the samples on ice, we added 9 μl of GenomiPhi V2 reaction buffer and 1 μl of GenomiPhi V2 enzyme mix to each sample mix and then incubated the samples at 30°C for 2.5 hours and at 65°C for 10 min to inactivate the protease. A positive control DNA (Control Human Genomic DNA) included in the amplification kit (Sigma-Aldrich) and a negative control (double-distilled water) were processed in parallel in each experiment. DNA purification was performed with the NucleoSeq clean-up kit (Macherey-Nagel) according to the manufacturer’s instructions. DNA concentrations were measured with NanoDrop ND-1000 (Thermo Fisher Scientific). Alternatively, the GenomePlex Single Cell Whole Genome Amplification Kit (Sigma-Aldrich) was used for single-cell DNA amplification. All steps were performed according to the manufacturer’s instructions. The amplified DNA was stored at −20°C until further analysis. To evaluate the quality of WGA products derived from single-cell DNA, different multiplex PCRs were performed.

Multiplex PCR to evaluate DNA quality of WGA products

One PCR was conducted with four primer sets (50) that generate products of 100, 200, 300, and 400 bp [100F (forward): gttccaatatgattccaccc, 100R (reverse): ctcctggaagatggtgatgg; 200F: aggtggagcgaggctagc, 200R: ttttgcggtggaaatgtcct; 300F: aggtgagacattcttgctgg, 300R: tccactaaccagtcagcgtc; 400F: acagtccatgccatcactgc, 400R: gcttgacaaagtggtcgttg] from nonoverlapping target sites in the GAPDH gene. The protocol has been described in detail elsewhere (50).

The second PCR was accomplished with four primer sets generating products from target sites of different genes (EGFR, AURKA, HER2, and LGR5) using the following primers: EGFR-F: gtcataagtctccttgttg, EGFR-R: gtccaggctaatttggtg; HER2-F: cagcccagccttcgacaac, HER2-R: gttctctgccgtaggtgtc; LGR5-F: tcagcgtcttcacctcctacctg, LGR5-R: acgctactccccgagacac; AURKA-F: gtgactcagcaatttccttg, AURKA-R: ggcagacactgttacttcaac.

Quantitative PCR analysis on tumor tissue

EGFR copy numbers were determined by quantitative SYBR Green PCR as described previously (47). In brief, genomic DNA was extracted from frozen tumor specimens and from cells with the NucleoSpin Tissue Kit (Macherey-Nagel). qPCR analysis was performed using the 7500 Fast Real-Time PCR System (Applied Biosystems) with standard conditions for the 9600 emulation run mode, including a dissociation stage. Amplification reactions contained 10 ng of genomic DNA and 10 pmol of forward and reverse primers for EGFR or IFN-γ as an internal control (MWG Biotech). Reactions also contained 10 μl of 2× QuantiTect SYBR Green PCR Master Mix (Qiagen) in a total volume of 20 μl. Relative amounts of amplified DNA were determined by the ΔΔCT method, with normal human leukocyte DNA as a calibrator.

Gene amplification detected by FISH

The EGFR-specific FISH probe was prepared from the Homo sapiens PAC clone RP5–1091E12, containing a 180-kb insert spanning almost the entire EGFR gene as described previously (47). Additionally, SpectrumGreen- and SpectrumOrange-labeled centromere probes were used (Abbott Molecular).

FISH on paraffin sections of tumor specimens was performed as described previously (47). For FISH on mononuclear cells, cytospins were incubated for 2 hours at 60°C and for 2 min at 120°C in Dako REAL Target Retrival Solution (1×, Dako). Subsequently, the cells were fixed in 1% formalin in PBS for 10 min and subsequently dehydrated in a graded alcohol series. After being air-dried, the slides were incubated in denaturation buffer (70% formamide, 0.6× SSC, pH 7.4) for 5 min at 73°C, washed for 2 min in 70% alcohol at 4°C, and subsequently dehydrated in graded alcohols. After hybridization, treatment and mounting were carried out as described before (49). Finally, the slides were evaluated with a fluorescence microscope (Zeiss) by two blinded observers, and results were discussed until consensus was reached.

Comparative genomic hybridization

Classical CGH of purified WGA products and DNA from the corresponding primary tumors was performed as described previously (51). To obtain DNA suitable for CGH, DNA derived from single cells was amplified by the GenomePlex Single Cell Whole Genome Amplification Kit (Sigma-Aldrich) according to the manufacturer’s instructions.

Array-CGH of purified WGA products and DNA derived from the corresponding primary tumors was carried out as published elsewhere (20). For this application, DNA derived from single cells was amplified with the illustra GenomiPhi V2 DNA Amplification Kit (GE Healthcare). Evaluation of array-CGH was based on a previously published algorithm (52).

Targeted enrichment of the primary tumor

We enriched a total of 370 cancer-associated genes included in the Cancer Gene Census from the Cancer Genome Project (Wellcome Trust Sanger Institute,, accounting for about 1 Mb, using the TruSeq Custom Trial Kit from Illumina. The Cancer Gene Census is a constantly developing catalog of genes for which mutations have been causally implicated in cancer. Briefly, a standard shotgun library was made from genomic DNA of the primary tumor extracted from formalin-fixed paraffin-embedded material with the TruSeq DNA Sample Preparation Kit v2 (Illumina). For quantification of the library, we used the Quant-iT PicoGreen dsDNA Kit (Invitrogen). Five hundred nanograms of the library was hybridized to the biotinylated probe library of the TruSeq Trial Kit (95°C for 10 min, 18 cycles of 1-min incubations, starting at 93°C, then decreasing 2°C per cycle, and finally 58°C for 16 to 20 hours). Subsequently, streptavidin beads were used to pull down the complex of captured oligos and genomic DNA fragments, whereas unbound fragments were removed by three washing steps. The enriched library was then eluted from the beads and prepared for a second hybridization to make sure that the targeted regions are further enriched. Another three washing steps were performed to remove nonspecific binding from the beads. The enriched library was then eluted from the beads, amplified by PCR for 10 cycles, and purified with AMPure XP beads (Beckman Coulter). For quality control and quantification, the enriched and amplified library was analyzed on an Agilent Bioanalyzer DNA 7500 chip (Agilent Technologies). The library was then sequenced on an Illumina MiSeq (Illumina) using a half lane of the flow cell. Reads were aligned to the human reference genome (UCSC hg19) using the BWA algorithm (53), and PCR duplicates were removed for subsequent analyses using SAMtools (54). We also performed local realignment around known insertions/deletions and recalibrated precomputed base quality scores with the Genome Analysis Tool Kit (GATK) (54). For variant calling, we used Unified Genotyper provided by the GATK. Variant prioritization steps included removing of variants that are not located within the protein-coding regions of the genome with the exception of splice sites. We retained variants that have a predicted functional effect on the protein and that are unique in the patient or at least very rare in the general population.

Validation of sequence variants identified with next-generation sequencing

For amplification of genomic regions harboring putative mutations identified in the primary tumor, we designed primers using Primer3 software (54, 55). Amplification was carried out in a reaction volume of 12 μl using the HotStarTaq Master Mix Kit (Qiagen). The template for PCR consisted of 50 to 100 ng of DNA, and reaction mixtures were subjected to 35 cycles of amplification in a thermocycler (15 min at 95°C, 35 cycles of 45 s at 95°C, 45 s at 57°C, and 45 s at 72°C, followed by 7 min at 72°C). The amplified products were then checked by gel electrophoresis on a 1% agarose gel. Sequencing reactions were performed using 0.5 μl of the PCR products with a Big Dye Terminator v3.1 Cycle Sequencing Kit (Life Technologies). After purification of the reactions using Sephadex G50 columns (Sigma-Aldrich), amplicons were separated in an ABI3130 system (Life Technologies). DNA sequence analysis was performed on a PC equipped with SeqScape software (Life Technologies).

Western blot analysis

Cell lines derived from glioblastoma were cultured as described elsewhere (5658). For Western blot analysis of MECOM and MYH11 proteins, whole-cell extracts from these cultured cells were lysed directly in SDS sample buffer containing proteinase inhibitors, with sonication. Proteins were separated on denaturing 6% polyacrylamide gels and were then subjected to Western blot analysis as described previously (59), using the polyclonal rabbit anti-MECOM antibody (Abcam) or the rabbit anti-MYH11 antibody (Sigma-Aldrich).

Statistical analysis

Statistical analysis was performed with IBM SPSS Statistics 21. Two-sided Fisher’s exact test was used to compare the presence of CTCs and patient characteristics. To do this, we classified each characteristic into subgroups (qPCR-EGFR amplification: 0, 0 to ≤2.2-fold; 1, >2.2-fold; EGFRvIII immunohistochemistry: 0, negative; 1, positive; gender: male/female; age: ≤50 and >50 years; GFAP expression: +, positive; +/−, heterogeneous). Estimation of survival curves was performed by Kaplan-Meier analysis, and survival curves were compared by the log-rank test.


Fig. S1. Overall survival of CTC-positive and CTC-negative GBM patients.

Fig. S2. Western blot analysis of a panel of glioblastoma cell lines.

Table S1. Patient/tumor characteristics for all patients with GFAP-positive cells detected in PB.

Table S2. Sequence variants chosen for validation with Sanger sequencing of GBM.


  1. Acknowledgments: The excellent technical assistance by S. Zapf and S. Hoppe is acknowledged. We thank E. Vettorazzi (Department of Medical Biometry and Epidemiology, University Medical Center Hamburg-Eppendorf) for statistical analysis. We thank D. Bigner (Duke University Medical Center, Durham, NC) for providing the EGFRvIII-specific mouse IgG1 monoclonal antibody (clone L8A4). Funding: This work was supported by the ERC-2010-AdG_20100317 DISSECT to K.P. Author contributions: K.P., S.R., M.W., M.R.S., J.B.G., K.L., and M.G. contributed to the design of the study. S.R., K.P., M.W., C.M., K.L., M.R.S., A.S., J.H., C.G., J.B.G., M.G., E.H., J.M., and S.P. participated in drafting and revising the manuscript. All authors participated in the acquisition, interpretation, and/or analysis of data or in a specialized role in a research field. C.M., J.H., C.G., M.S., O.M., K.P., and S.R. were responsible for the detection and characterization of CTCs; M.A., E.H., C.M., S.L.-F., C.G., O.M., J.B.G., and M.R.S. for molecular single-cell analysis including WGA, sequencing, and CGH; C.M., J.H., A.S., K.L., M.S., S.R., M.A., E.H., M.R.S., and J.B.G. for molecular and phenotypic analysis of tumor tissues and cell lines; M.W., M.G., J.M., K.L., and A.S. for providing clinical tumor material as well as morphological and clinical data; S.P. for providing blood samples from healthy controls; and O.M. and C.M. for statistical analysis. All authors contributed to the writing and review of the manuscript. Competing interests: The authors declare that they have no competing interests.
View Abstract

Stay Connected to Science Translational Medicine

Navigate This Article