Phenotypic Characteristics of Cell Lines Derived from Disseminated Cancer Cells in Bone Marrow of Patients with Solid Epithelial Tumors: Establishment of Working Models for Human Micrometastases¹

The set of molecules required for early micrometastatic spread, a major prognostic determinant in patients with operable solid tumors, is largely unknown (1). It is difficult to obtain this information from analysis of a fully established metastatic lesion because the increasing genetic instability of solid metastases may have masked the initial events (2, 3). On the other hand, the direct study of micrometastatic tumor cells has been hampered for many years by the lack of suitable markers to identify these cells in secondary organs. For patients with epithelial tumors, which represent the majority of cancers in industrialized countries, the detection level of micrometastatic tumor cells is now improved. Immunocytochemical screening with mAbs³ against CKs now appears to be a reliable method to detect isolated tumor cells in BM. BM is a prognostically relevant indicator organ for micrometastatic spread, even in those types of cancer in which overt skeletal metastases are rare events, such as carcinomas of the colon or head and neck region (4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14). One explanation for this surprising finding is that BM might function as an important reservoir for disseminated epithelial tumor cells, which can be the source for further dissemination into other organs. This assumption is also supported by the clinical observation that the release of carcinoma cells from BM into the peripheral blood can be induced by cytokine treatment (15, 16).

It would, therefore, be of great interest to identify the set of molecules that determine the affinity of epithelial tumor cells to BM. Immunocytochemical double marker analyses have indicated that CK⁺ cells in BM are a selected but still heterogeneous population of tumor cells with regard to the expression of cell adhesion receptors, oncogenes, proliferation markers, or MHC class I antigens (17, 18, 19). However, these evaluations were limited by the extremely low frequency of CK⁺ cells (∼10⁻⁵–10⁻⁶). This obstacle has been overcome by the development of appropriate cell culture conditions and an experimental immortalization strategy that allows the generation of permanent cell lines from BM cancer cells (20). Here, we report the phenotypic characterization of nine of these unique cell lines with a focus on histogenetic marker molecules and metastasis-associated cell adhesion receptors. Our data indicate that these cell lines represent good working models of human BM micrometastases because they demonstrate many properties of in situ epithelial tumor cells. Implementation of these models into the development of new drugs might improve the efficacy of current anticancer therapy because micrometastatic tumor cells in BM should be better therapeutic targets than overt metastases, because these cells are easily accessible by i.v. injected drugs, and because the total tumor burden is significantly lower (i.e., in the range of 10⁶–10⁸ tumor cells per total BM; Ref. 21).

BM aspirates were obtained from the upper iliac crest of patients with primary carcinomas of the breast, colon, lung, or prostate without overt metastasis (stage M₀). The exact tumor stages of these patients are listed in Table 1. The BM cells were washed with Hank’s medium (Biochrom, Berlin, Germany) at 170 × g for 10 min, and MNCs were obtained by Ficoll density centrifugation at 1230 × g for 30 min. MNCs were washed at room temperature at 520 × g for 10 min, and an aliquot of 2 × 10⁶ MNCs was then cytocentrifuged onto glass slides at room temperature at 130 × g for 5 min (10⁶ cells per slide). After overnight air drying, slides were either stained immediately or stored at −80°C before use; after such preparation and storage, antigens were found to be preserved for at least 2 years⁴. From each aspirate, one slide (10⁶ MNCs) was examined, whereas one additional slide served as an immunoglobulin (immunoglobulin) isotype control.

For tumor cell detection in BM cytospin preparations, mAb A45-B/B3 (IgG1; Micromet GmbH, München, Germany) was used at 2.5 μg/ml according to our previous work (20, 21). A45-B/B3 detects a common epitope of a variety of CK components, including CK8/18 and CK8/19 heterodimers (22). Appropriate dilutions of unrelated mouse myeloma proteins served as IgG1 isotype control (MOPC21; Sigma Chemical Co., Deisenhofen, Germany). The antibody reaction was developed with the alkaline phosphatase anti-alkaline phosphatase technique combined with Neufuchsin stain (23). Briefly, after incubation with the primary antibody, a polyvalent rabbit antimouse immunoglobulin antiserum (Z259; DAKO, Hamburg, Germany) and preformed complexes of alkaline phosphatase and monoclonal antialkaline phosphatase antibodies (D651; DAKO) were used at the dilutions recommended by the manufacturer (DAKO). To allow a fast screening for mAb-positive cells on the slides, no counterstaining was performed.

Between 8 × 10⁶ and 150 × 10⁶ MNCs per flask were initially plated in T25 culture flasks coated with extracellular matrix (Paesel & Lorei, Frankfurt, Germany; Table 1). The extracellular matrix material is similar in organization and composition to naturally occurring basement membranes. Cells were incubated in 5–10% CO₂ (incubator B 5061 EK/CO₂; Heraeus, München, Germany). The culture medium contained RPMI 1640 supplemented with 10% FCS, 10 μg/ml transferrin, 5 μg/ml insulin, 2 mm glutamine, 10 ng/ml recombinant human epidermal growth factor (Boehringer Mannheim GmbH, Mannheim, Germany), synthetic androgen (R 1881, in case of prostate cells) and 10 ng/ml recombinant human basic fibroblast growth factor (PBH, Hannover, Germany). The cells were cultured under reduced oxygen, the medium was changed twice a week, and fresh growth factors were added to the cultures. At confluence, the adherent cells (including the epithelial tumor cells) were removed by trypsinization and passaged in new extracellular matrix-coated flasks.

The plasmid construct pUC In wt was cloned by integration of the SV40 genome fragments PstI/BstXI (2471 bp) and BstXI/BamHI (2226 bp) from the plasmid pSV In1 into the plasmid vector pUC12. The fragments from pSV In1 contain a disrupted origin of replication generated by an insertion of 1 bp in the center of the 27-bp palindrome, thereby destroying the BglI site. The plasmid contains the complete SV40 promoter/enhancer (nucleotides 1988–ori–2533). The pSV40 plasmid (5 μg per T25 culture flask; Ref. 20) was transfected at passage 2–4 when the cells were grown to 50–70% confluency using either microinjection, as described recently (20), or 12.5 μl of the lipophilic transfection reagent lipofectin (Life Technologies, Inc., Eggenstein, Germany) in serum containing medium, according to the manufacturer’s instructions.

At different time points (Table 1 and Fig. 1) adherent cells were detached by trypsin-EDTA treatment. The expression of SV40 T-Ag was monitored by immunocytochemical analysis (alkaline phosphatase anti-alkaline phosphatase alkaline technique, as described above), using anti-SV40 T-Ag mAbs Pab 101, 220, 416, and 419 (provided by Dr. Ellen Fanning, Vanderbilt University, Nashville, TN). For flow cytometric analysis, cells were fixed for 10 min at 4°C with either 80% methanol (staining of α-smooth muscle actin, vimentin, E-cadherin, or androgen receptor) or 80% acetone (staining of CK) diluted in 19.8% PBS and 0.2% BSA; all other stainings were performed with unfixed cells. The primary antibodies were used at optimal concentrations of 2.5 to 10 μg/ml or as undiluted hybridoma supernatants. The following antibody clones were used: 1A4, anti-smooth muscle actin (DAKO); F39.4.1, anti-androgen receptor (Biogenex, San Ramon, CA); MEM48, anti-CD-18 (anti-β₂-integrin, Southern Biotechnology Associates, Birmingham, AL); 2A4, anti-CD29 (anti-β₁-integrin; Southern Biotechnology Associates); BIRMA-K3, anti-CD34 (DAKO); SAM-1, anti-CD49e (anti-α₅-integrin, Southern Biotechnology Associates); GoH3, anti-CD49f (anti-α₆-integrin, PharMingen, Hamburg, Germany); 23C6, anti-CD51 (anti-α_v-integrin, Southern Biotechnology Associates); BL-E6, anti-CD61 (anti-β₃-integrin; Sigma); 450–9D, anti-CD104 (anti-β₄-integrin, PharMingen); IOL44, anti-CD44s (Immunotech, Hamburg, Germany); VFF-7, anti-CD44v7+v8 (Bender, Vienna, Austria); VFF-8, anti-CD44v5 (Bender); VFF-18, anti-CD44v6 (Bender); 5B1, anti-CD45 (Boehringer Mannheim); A45B/B3, anti-pan CK (Micromet, Munich, Germany); CK2, anti-CK-18 (Boehringer Mannheim); anti-pan-CK, AB1/AE3 (DAKO); anti-CK5/6/8/18 LP34 S03 (Medac, Hamburg, Germany) AICD58, anti-CD58 (Immunotech); 50F11, anti-CD82 (PharMingen); D6.3.10, anti-desmoglein (Progen, Heidelberg, Germany); 6F9, anti-E-cadherin (provided by Dr. W. Birchmeier, Max-Delbrueck-Center, Berlin, Germany); 29.1.1, anti-epidermal growth factor receptor (Yeda, Rehovet, Israel); EN-4, anti-endothelial cells (Monosan, Uden, the Netherlands); erbB-7C1, anti-p185^erbB2 (provided by J. P. Johnson); 4F9, anti-factor VIII (Immunotech); P3.58BA.3, anti-ICAM-1 (provided by J. P. Johnson); MLuc5, anti-M_r 67,000 laminin receptor (kindly provided by Dr. Maria Colnaghi, Institute for Experimental Cancer Research, Milan, Italy); 55–2LL10, anti-Lewis Y (provided by Dr. Peter Rieber, Institute of Immunology, Dresden, Germany); A76, anti-MUC1 (provided by Dr. U. Karsten, Max-Delbrueck-Center, Berlin, Germany); MUC18, MUC18BA.4; MAd5D7 and MAd1B4, anti-MUC18 (provided by J. P. Johnson); A67B/E3, anti-PSA (Connex, Munich, Germany); CYT-351, anti-PSM (Cytogen, Princeton, NJ); 1G11, anti-VCAM (Immunotech); and V9, anti-vimentin (DAKO). The antibody reaction was developed with a secondary FITC-conjugated rabbit antimouse immunoglobulin antibody (F0313; DAKO), used at a 1:50 dilution. Appropriate IgG and IgM isotype control antibodies were used to ensure specificity of staining. Undiluted culture supernatant of a hybridoma against murine H-2 (141-11, anti-H2; G. Hämmerling, DKFZ, German Cancer Research Centre, Heidelberg, Germany) served as unrelated IgM antibody control.

Total RNA from micrometastacic cell lines was isolated according to the method of Chirgwin (24), and purified to poly(A) RNA using the Oligotex mRNA Mini Kit (Qiagen, Hilden, Germany). The reverse transcription of poly(A) RNA was performed using the Superscript II reverse transcriptase (Life Technologies, Inc.) and random hexameric primers according to manufacturer’s instructions. The following primer pairs were used to amplify prostate cancer-associated cDNAs: androgen receptor, 5′-TGGAAGCCATTGAGCCAGG-3′ and 5′-GGCTTGACTTTCCCAGAAAGG-3′; human prostate-specific transglutaminase, 5′-aagcttcatATGATGGATGCATCAAAAGAGCGCAAGTTCTCCACAT-3′ and 5′-AAAGCTCATCATAGCACACATGG-3′; prostate cancer tumor antigen, 5′-CCTGGAACTTTGATTGTGATACG-3′ and 5′-TCCTTGAACGACGACAGTTCG-3′; PSA, 5′-GATATGAGCCTCCTGAAGAATCG-3′ and 5′-TGTACAGGGAAGGCCTTTCG-3′; PSM, 5′-AGTGCTCCCTTTTGATTGTCG-3′ and 5′-GTGACATACCACACAAATTCAATACG-3′; PTI-1 5′-TGCATCCTACCACCAACTCG-3′. 5′-CTTCCAGCTTTTTACCAGAACG-3′; kallikrein 2, 5′-ACAGGCCAGAGGGTCCCTG-3′ and 5′-CTGTCACCTTCTCAGAGTAAGCTCTAG-3′; and cytosine-5 DNA methyltransferase, 5′-GCACTGGAGATCTCCTACAACG-3′ and 5′-CAC-TCGAGCCTTCCATAGAGG-3′ (uppercase letters of primer sequence indicate gene sequence specificity). Prolonged primers were designed to create restriction site for appropriate restriction enzymes simplifying eventual subcloning. In a 50-μl reaction mix, 20 pmol of the primers were used together with 2 μl of the cDNA for PCR. The cDNA was amplified for 40 cycles (delay, 1 min at 94°C; 94°C for 30 s, 50°C for 30 s, and 72°C for 30 s), separated by agarose gel electrophoresis, and stained with ethidium bromide. The integrity of transcribed RNAs from all cells was controlled with the amplification of the housekeeping gene cytosine-5 DNA methyltransferase.

For amplification of MAGE, a PCR mixture (10 μl) was composed of 1 μl of cDNA, 1 μl of 10× PCR buffer [100 mm Tris (pH 8.3), 500 mm KCl, and 10 mm MgCl₂], 40 μm dNTP, 0.4 μm each of the two primers, 5 μg of BSA (Boehringer Mannheim), and 0.6 units of Taq DNA-Polymerase (Boehringer Mannheim), overlaid with 12.5 μl of mineral oil. The primers for the MAGE genes were designed from previously published sequences (25) and were selected to maximize mismatches between the different MAGE sequences, particularly in the 3′ region, to avoid cross-amplification⁵ Specific PCR assays were possible for MAGE-1, -2, -4, and -12 because the percentage of identity of these genes varies between 64 and 85%. In contrast, the sequence of MAGE-6 was found to be 99% identical to that of MAGE-3. Therefore, our primers for MAGE-3/-6 hybridize to both genes, and PCR can detect MAGE-3 and/or MAGE-6 expression. To confirm the presence of cDNA template, a control amplification of cytosine-5 DNA methyltransferase was performed. Oligonucleotide primers were synthesized and purified at Genset (Paris, France). For the external fragment amplification of the MAGE, a PCR assay with the following cycling profile was used: denaturation at 94°C for 6 min, annealing at 60°C for 30 s, and extension at 72°C for 2 min for the first cycle; denaturation at 93°C for 40 s, annealing at 60°C for 30 s, and extension at 72°C for 20 s for 14 cycles; denaturation at 93°C for 40 s, annealing at 60°C for 30 s, and extension at 72°C for 30 s for 50 cycles; terminal extension at 72°C for 2 min; and cooling at 30°C for 1 min. One μl of the reaction was then transferred into a second tube containing the PCR mixture described above. For the nested fragment amplification, 30 more cycles were run at 93°C (40 s), 58°C (30 s), and 72°C (30 s), with a final extension at 72°C for 2 min and cooling at 30°c for 1 min. The amplification was performed on a Hybaid Thermal Reactor (Biometra, Göttingen, Germany) using a plate control. Importantly, all pipetting was performed on ice under a laminar air-flow bench with filtered pipette tips; external and nested fragment amplification were performed in different rooms to avoid potential DNA cross-contamination. Negative control reactions (mock reactions) for the RNA preparation and RT-PCR were performed routinely. PCR products were analyzed on 1.8% agarose by gel electrophoresis and by direct visualization after ethidium bromide staining under UV light.

Genomic DNA of cell lines or peripheral blood lymphocytes was isolated following the protocol of Miller et al. (26). Genomic DNAs from cytospin preparations were isolated as followed: Hettich centrifuge carriers were cleaned with 2 m HCl, and the cytospins were fixed. The cells were overlaid with 80 μl of solution A [100 mm KCl, 10 mm Tris-HCl (pH 8.3), and 2.5 mm MgCl₂], 80 μl of solution B [10 mm Tris-HCl (pH 8.3), 2.5 mm MgCl₂, 1% Tween 20, and 1% NP40]. Proteinase K was added to a final concentration of 100 μg/ml, and centrifuge carriers were sealed with Parafilm and incubated for 2 h at 56°C. The liquid was transferred into an Eppendorf tube and extracted with phenol/chloroform, and the aqueous DNA solution was frozen at −20°C.

HLA-DRB1* typing was performed using either genomic DNA of cell lines and peripheral blood lymphocytes or DNA isolated from cytospin preparations of the patients, following the oligonucleotide typing system protocol described by Nevinny-Stickel et al. (27).

The analyzed cell lines were established from BM aspirates of nine patients who presented with carcinomas of the prostate (PC-E1, PC-H1, PC-R1, and PC-S1), breast (MC-H1 and MC-K1), lung (LC-D1 and LC-M1), and colon (CC-B1). Remarkably, all of these patients were free of histopathological lymph node metastases (pN₀) as well as clinical signs of overt systemic metastases in distant organs (M₀) at the time of BM analysis. However, after a short median observation period of 2 years (range, 1–5 years), tumor relapse was observed in six of eight evaluable patients, and three patients died. Most remarkably, patient BC-K1, who was staged as early pT₁N₀M₀ breast cancer, presented already with distant overt metastases 2 years after complete resection of the primary tumor.

In seven patients, the initial concentration of CK⁺ tumor cells in BM was below 1 in 10⁶ MNCs, which was the detection limit of our immunocytochemical assay (Table 1). Considering that we plated between 8 × 10⁶ and 150 × 10⁶ MNCs per culture, the estimated initial number of CK⁺ cells in each culture varied between <9 and 75 (Table 1). The establishment of a permanent cell line from these CK⁺ cells is exemplified for LC-M1 in (Fig. 1). CK⁺ cells started to grow exponentially and were transfected with SV40 T-Ag cDNA after the third culture passage. Crisis occurred in SV40-positive cells after a mean of 14 culture passages (range, 9–19 passages). The clones emerging from the crisis continued to grow exponentially and established permanent cell lines; all of these cells permanently expressed the SV40 T-Ag (Table 1). In contrast, BC-K1 and PC-H1, which were also expanded from BM but did not survive crisis, were negative for SV40 T-Ag (Table 1), indicating that the transfection was insufficient or only transient. Subsequent analyses were performed on SV40 T-Ag-positive cells after their crisis and on SV40 T-Ag-negative cells (PC-H1 and BC-K1) before their crisis (Table 1).

To verify the origin of our established cell lines from the respective patients and to rule out cross-contaminations in cell culture handling, HLA-DRB1* typing was performed using the genomic DNA of the cell lines in comparison to genomic DNA isolated from the original BM cytospins or, when available, DNA isolated from peripheral blood lymphocytes of the respective patient. In all cases, the established cell lines showed the same typing as their autologous peripheral blood lymphocytes (PC-R1) or BM cells (all other cell lines). The HLA-DRB1* type of all cell lines differed from each other, with the exception of CC-B1 and PC-H1. For DRB1*, the following typing results were obtained: PC-E1, 04/07; PC-H1, 03/6; PC-R1, 04/−; PC-S1, 04/12; BC-H1, 07/11; BC-K1, 03/14; LC-D1, 03/13; LC-M1, 09/11; and CC-B1, 03/6 (data not shown).

To confirm the epithelial nature of the established cell lines, we analyzed the expression of CKs as the major constituents of the epithelial cytoskeleton. As shown in Table 2, all cell lines expressed detectable levels of CK if they were incubated with mAbs A45-B/B3 or CK2. However, staining for CK was only weak in two of these lines (BC-K1 and CC-B1) and heterogeneous in PC-E1 and BC-H1 cells. The presence of CK-negative subpopulations in these lines was confirmed by additional immunostaining with other anti-CK mAbs (AE1/AE3, LP34, and 5D3; data not shown). There was no significant difference between the staining pattern obtained with the broad spectrum anti-CK mAbs A45-B/B3 and mAb CK2 against CK component 18 (data not shown). Interestingly, Lehr and Pienta (28) recently established a human BM endothelial cell line by transfection with SV40 T-Ag, which may serve as a control to this analysis. These cells lacked expression of CKs, indicating that this typical feature of epithelial tumor cells in BM is not induced by the SV40 oncogene. The epithelial nature of our cell lines was further supported by the consistent expression of E-cadherin, the most prominent epithelial cell adhesion molecule (Table 2).

The ectopic nature of the newly established cell lines was shown by negative staining of marker proteins typical for cells residing in BM. None of the cell lines expressed the hematopoietic stem cell antigen CD34, the common leukocyte antigen CD45, smooth muscle actin, or the endothelial cell markers factor VIII and EN4 (Table 2). As positive controls, BM cells were stained under the same conditions, and all of the above antigens could be detected on subpopulations of these cells (data not shown). Interestingly, the mesenchymal cytoskeleton protein vimentin was coexpressed with CK in all cell lines (Table 2), indicating an epithelial to mesenchymal transition of micrometastatic cells.

The tumor origin of seven of nine cell lines was supported by immunodetection of the ErbB2 proto-oncogene gene product (Table 2), which usually results from an overexpression of the p185^ErbB2 protein due to an amplification of the ErbB2 gene (29). Immunodetection of ErbB2 is a common feature of many epithelial tumors, including cancer of the breast (29, 30, 31), colon (32), lung (33), and prostate (34), 35). Using immunocytochemical double labeling, we showed previously that BM micrometastases express ErbB2 more frequently than the corresponding primary tumors; e.g., almost 70% of breast cancer patients presented with ErbB2⁺/CK⁺ cells in the BM, suggesting a selection bias for ErbB2⁺/CK⁺ cells in the BM environment (17).

Another common characteristic of cancer cells is the expression of genes of the MAGE family, which consists of at least 12 members (36, 37). Here, we analyzed MAGE mRNA expression, using RT-PCR with primers specific for MAGE-1, -2, -3/-6, -4, and -12. As shown in Table 3, at least one of these MAGE genes was expressed in all of eight cell lines analyzed (BC-K1 could not be analyzed because of the limited number of cells available). Two lines, PC-S1 and LC-M1, expressed all of the screened MAGE genes, whereas the expression pattern in the remaining lines was more heterogeneous. Interestingly, all of the cell lines expressed MHC class I molecules, as determined by staining with mAb W6/32 against a monomorphic determinant of these molecules (Table 3). The cell lines might, therefore, be able to present MAGE-derived peptides in the context of MHC class I molecules. To exclude that MAGE expression was only the result of our immortalization strategy, we analyzed BM cultures from seven cancer patients before transfection with SV40 T-Ag DNA. Consistent with the results on micrometastatic cell lines, MAGE mRNA expression was revealed by RT-PCR analysis in all of these cultures (data not shown).

In the case of the established prostate carcinoma cell lines, the expression of prostate-specific genes was assessed (Table 4). Three established American Type Culture Collection prostate cancer cell lines (Du145, PC-3, and LNCaP) were used as positive controls. The prostatic marker antigens were heterogeneously expressed in those lines as well as our micrometastatic cancer cell lines. PTI-1 mRNA was detected in all prostate carcinoma cell lines, indicating their prostatic origin. All micrometastatic cells stained negative for androgen receptor protein or PSA, whereas PSM seemed to be weakly expressed on a subpopulation of PC-E1 (5%) and PC-H1 (4%) cells, but this expression could not be confirmed at the mRNA level. However, RT-PCR analysis revealed that PC-S1 weakly expressed PSA mRNA and PCTA mRNA, whereas PC-E1 expressed androgen receptor mRNA (Table 4). Moreover, PC-R1 cells expressed mRNA encoding for human prostate-specific transglutaminase, a new marker for prostatic cells (38), whereas mRNA encoding kallikrein 2, which belongs to the same protease family as PSA, was only found in the control cell line LNCaP, not in any of our micrometastatic cell lines.

Homotypic and heterotypic cell adhesion play a crucial role in the metastatic cascade of solid tumor cells from invasive growth to manifest metastasis. We, therefore, assessed the expression of adhesion molecules known to be involved in invasion and metastasis. Micrometastatic cells of different tumor entities showed a remarkable homogeneity with regard to the expression of adhesion molecules (Table 5). Among the integrins, β₁ (CD29), β₃ (CD61), and α_V (CD51) integrin subunits stained positive on all cell lines. All lines with the exception of LC-M1 also expressed the α₅ integrin (CD49e). On the other hand, the expression of α₆ (CD49f) and β₄ (CD104) was more heterogeneous (expression in five and two cell lines, respectively), and none of the cells analyzed were positive for β₂ integrin (CD18).

With regard to the adhesion molecules of the immunoglobulin superfamily, the expression patterns were even more homogeneous. As shown in Table 5, all cell lines expressed MUC18/MCAM (CD146), ICAM-1 (CD54), and LFA-3 (CD58). Flow cytometric analysis of 17 cell lines derived from primary tumors and metastases of patients with cancers of the breast, colon, prostate, and lung revealed that MUC18/MCAM was only detected in four cell lines (Table 6) even in those cell lines, MUC18/MCAM was only present on subpopulations of tumor cells consisting of <10% to >50% of the total population. Thus, expression of MUC18/MCAM is not a common phenomenon of cultured epithelial tumor cells. MUC18/MCAM expression also appeared to be not only a consequence of the immortalization with SV40TAg, because both SV40TAg-negative cell lines (PC-H1 and BC-K1), which were not immortal, expressed MUC18/MCAM at the same level as the remaining SV40 T-Ag-positive cell lines (Table 5). Moreover, MUC18/MCAM was already expressed on LC-M1 tumor cells before transfection with SV40 T-Ag cDNA (Fig. 1). The intensity of MUC18/MCAM staining varied among the cell lines analyzed, with no clear differences between the different tumor entities (data not shown). MUC18/MCAM was not detectable by flow cytometric analysis on normal BM cells (data not shown). However, Filshie et al. (39) recently demonstrated that MUC18 is expressed on a rare subpopulation (<1%), containing >90% of the stromal BM precursors.

Vascular cell adhesion molecule-1 (CD 106), which mediates binding of tumor cells to endothelium, was expressed only on LC-D1 lung cancer cells and on minor subpopulations (7%) of PC-E1 and PC-S1 prostate cancer cells (Table 5). Similarly, CD44 variant isoforms CD44v5 and CD44v6, which have been implicated in metastasis formation (40) were only detected on CC-B1 and PC-E1 cells, and CD44v7/8 were not expressed at all (Table 5), whereas CD44s, the standard form of CD44, was expressed on all cell lines analyzed (Table 5).

Among the molecules specifically mediating epithelial cell adhesion, the M_r 67,000 laminin receptor plays a prominent role in the interaction of tumor cells and the basement membrane (41). The M_r 67,000 receptor was present on eight cell lines (Table 5); however, in seven cell lines, the laminin receptor was only expressed on certain subsets of cells (Table 5). Further indication for a disturbance of epithelial cell adhesion was derived from analysis of homotypic adhesion receptors. Although E-cadherin was consistently present in all cell lines tested (Table 2), the desmoglein was only expressed in two lines (PC-R1 and PC-S1) and plakoglobin, the common intracytoplasmatic binding protein of both E-cadherin and desmoglein, was only found in two of eight cell lines analyzed (Table 5).

Here, we performed a detailed phenotypic analysis of nine unique cell lines established from micrometastatic tumor cells present in BM of patients with epithelial primary tumors. The epithelial nature of the established cell lines was supported by the expression of epithelial proteins (CK and E-cadherin) and the lack of markers of BM cells (CD34 and CD45). The prostatic origin of four cell lines was supported by the expression of PTI-1. The reduced or absent expression of androgen receptor and PSA in three of our prostate cancer-derived cell lines was not surprising because most of the other prostate tumor cell lines, such as PC3, DU145, or TSU-pr1, are androgen-insensitive (42) and do not express PSA (43). Although CK was expressed in all of our cell lines, two lines consisted of subpopulations that lacked a detectable expression, using different mAbs against various CK components. The lack of CK expression does not necessarily argue against the epithelial nature of these cells. Down-regulation of CK expression has been described as indicator of an unfavorable prognosis in breast cancer (44, 45), suggesting that changes in the cytoskeletal composition of tumor cells may favor metastases. In this context, it is also noteworthy that the mesenchymal cytoskeleton protein vimentin was coexpressed with CK in all cell lines analyzed (Table 2), indicating an epithelial to mesenchymal transition of micrometastatic cells that is associated with an increased in vitro motility, plasticity and growth rate of epithelial tumor cells (46, 47). At present, the in situ role of an epithelial to mesenchymal transition in tumor progression and metastasis is still under debate (48, 49). Indirect evidence has been derived from the clinical observation that neoexpression of vimentin on primary breast carcinomas is associated with an increased rate of metastatic relapse (50).

Positive staining for the receptor tyrosine kinase p185^ErbB2 in the majority of our cell lines might reflect a selective growth advantage of tumor cells expressing this growth factor receptor. ErbB2 expression is also a frequent characteristic of dormant BM carcinoma cells in situ (17), which suggest that ErbB2-positive tumor cells have been selected during the dissemination process. One possible mechanism by which activated ErbB2 might support early dissemination of tumor cells is loss of epithelial cell adhesion through phosphorylation of catenins, the cytoplasmatic binding proteins of E-cadherin (51). In addition, this analysis indicates that down-regulated expression of other homophilic adhesion receptors, such as desmoglein or plakoglobin, might also contribute to the loss of epithelial cell-cell adhesion required for the onset of tumor cell dissemination.

Originally described as melanoma-associated genes (MAGE), genes of the MAGE family are widely expressed in various epithelial tumors and cell lines derived thereof (36). With the exception of normal testis, MAGE antigens are not expressed in normal tissues (36). Thus, expression of these antigens on all of the micrometastatic cell lines analyzed provides further proof for their oncogenic nature (Table 3). Moreover, the concomitant expression of these antigens and MHC class I molecules on these cell lines (Table 3) supports the view that these lines may be used to generate a tumor cell vaccine directed against minimal residual cancer (52). Because all of our cell lines consistently lacked a detectable expression of B7.1 protein (data not shown), it might be necessary to transfect these cells with this important costimulatory molecule together with cytokine genes, such as interleukin-2 or granulocyte macrophage colony-stimulating factor, to enhance the T cell-activating properties of such a future vaccine.

Among the cell surface molecules that mediate cell-substratum adhesion, integrins play a predominant role in cancer metastasis (53, 54, 55, 56). Our present data indicate that α₅α_v, β₁, and β₃ integrin subunits are common characteristics of micrometastatic cells. Expression of the α₅ subunit is associated with malignant progression in colon cancer (57), and it suppresses apoptosis in HT29 colon carcinoma cells (58). The α₅β₁ fibronectin receptor is usually not expressed in nonmalignant tissues, but it plays an important role in the anchorage and regulation of fibronectin-mediated tumor cell adhesion in BM (59). Another important function of β₁ integrins is to suppress apoptosis in mammary epithelial cells through negative regulation of interleukin-1β-converting enzyme expression (60). On the other hand, the α_V subunit is frequently expressed in epithelial cells (61), and in the form of an α_Vβ₃ heterodimer, it mediates osteoclast-bone recognition (62) and is overexpressed by bone-residing breast cancer cells (63). Thus, it is conceivable that the observed pattern of integrin expression on micrometastatic cancer cells might favor their homing and survival in BM. Targeting of these receptors might be efficient to prevent homing of tumor cells reinfused into cancer patients during an autologous stem cell transplantation.

Besides integrins, the M_r 67,000 laminin receptor plays an additional role in tumor cell-substratum interaction. Expression of this receptor protein is associated with an increased rate of metastatic relapse in breast cancer (64, 65, 66). Indirect evidence that this observation might be due to an early onset of micrometastasis is provided by the consistent expression of the M_r 67,000 receptor on at least subsets of micrometastatic cancer cells. Interestingly, the M_r 67,000 receptor was coexpressed with the α₆ integrin subunit in five of our cell lines, which might reflect the physical and regulatory association between both adhesion molecules (41).

Among the adhesion molecules of the immunoglobulin superfamily, the consistent expression of MUC18/MCAM on micrometastatic cells was unexpected because these molecules are usually undetectable on normal epithelial cells and infrequently expressed on primary carcinomas (67, 68). Both MUC18/MCAM and ICAM-1 have been implicated in the metastasis of melanoma cells (68, 69, 70), but their role in carcinoma progression is still unclear. Neoexpression of ICAM-1 is found in ∼30% of epithelial tumors, whereas expression of MUC18/MCAM in primary carcinomas is even more infrequent (e.g., breast, 11%; colon, 0%; lung, 40%; and stomach, 7%). It might be speculated that a special selective pressure of the BM microenvironment could be responsible for the high frequency of MUC18/MCAM⁺ tumor cells seeding in this particular “soil.” Evidence that the observed expression of MUC18/MCAM was not merely a cell culture artifact is derived from negative staining of several other cell lines established from primary or metastatic carcinomas and cultured under similar conditions as our micrometastatic cell lines. However, it cannot be excluded that MUC18/MCAM expression is up-regulated in cultured micrometastatic cells.

One of the prominent molecules involved in the interaction between cancer cells and the extracellular matrix is the glycoprotein CD44. In general, expression of CD44 variant isoforms has been positively correlated with human cancer progression [reviewed in Günthert (71)]; however, in prostate cancer patients treated by radical prostatectomy, loss of CD44s and CD44v6 predicted an adverse prognosis (72). Using mAbs against CD44v5–8, we observed staining of only two cell lines. Because CD44v6 expression on carcinoma cell lines cannot be easily modulated in vitro, it is unlikely that CD44v6 expression was down-regulated under our culture conditions (73). The infrequent finding of the variant isoforms CD44v5–8 on our cell lines argues against an important role of CD44 as facilitator of BM micrometastasis, whereas the consistent expression of CD44 points to its potential role in cancer micrometastasis, as demonstrated recently in a murine fibrosarcoma model (74).

Because of the few tumor cells initially present in each of the BM samples, we analyzed cell lines established from these cells. Although the in vitro expansion and subsequent immortalization with SV40 T-Ag cDNA might have altered the phenotype of tumor cells, we did not observe consistent differences between SV40 T-Ag-positive and -negative cells. Furthermore, the expression of CKs, ErbB2, and ICAM-1 antigens and the down-regulation of certain epithelial adhesion molecules (e.g., plakoglobin), which were observed on our cell lines, are also characteristics of in situ micrometastatic cancer cells (17, 19, 75). This is consistent with a number of reports demonstrating retention of phenotypic properties following SV 40 T-Ag-induced immortalization of different epithelial cells (reviewed in Ref. 20). Interestingly, six of eight of our patients relapsed after a short postoperative period. Although this evaluation is very preliminary, it suggests that the common antigens found on our lines might be invasive markers.

The established cell lines might be useful models to study the biology of BM micrometastases. Moreover, these lines may help to identify potential target molecules for adjuvant anticancer therapy and to test new anticancer agents specifically directed against minimal residual cancer.

We thank Diana Kiefer and Tanja Siart for excellent technical assistance. We gratefully acknowledge the following persons for providing BM samples from carcinoma patients: Dr. B. Passlick and Dr. S. Braun (Chirurgische Klinik und Poliklinik, and I. Frauenklinik, Klinikum Innenstadt, Ludwig-Maximilians-Universität München, München, Germany); Dr. G. Schlimok (Medizinische Klinik II, Zentralkrankenhaus, Augsburg, Germany); and Dr. S. Thorban (Chirurgische Klink und Poliklinik, Technische Universität München, München, Germany). We thank Drs. M. Osborn (Max Planck Institut, Göttingen, Germany) and H. Bodenmüller (Boehringer Mannheim, Tutzing, Germany) for mAb CK2 and Micromet GmbH (Martinsried, Germany) for the mAb A45-B/B3.