Epithelial to Mesenchymal Transition in Human Breast Cancer Can Provide a Nonmalignant Stroma

Establishment of the HBFL-1 Cell Line

A breast carcinoma was obtained from the Department of Pathology, Rigshospitalet, from residual material. Histologically, the carcinoma was classified as metaplastic. Phenotypically normal tissue from the same patient was dissected from the mastectomy specimen at a site remote from the cancer. The use of such tissue material has been reviewed by the Regional Scientific-Ethical Committees for Copenhagen and Frederiksberg, Denmark and found consistent with Laws no. 503 of 24th June 1992 and no. 499 of 12th June 1996 (KF) 01–161/98. The tissue was collected in culture medium (Dulbecco’s modified Eagle’s medium-Ham’s F12 (DME-F12) supplemented with 20% fetal calf serum (FCS) and processed as previously described. 12 For zymography and TGF-β experiments, a subline (S2) was cultured in chemically defined medium alone (DME-F12). Clonal cultures for studies of multipotency were performed as previously described 13 using a subline (W2331) propagated in the presence of 20% FCS. One clonal subline (S4) was generated by limited dilution and isolation of the clonal cell islet by trypsinization under the microscope within a Perspex ring. The cells were split 1:3 once a week, and have been kept in culture for more than 3 years. They are referred to as human breast fibroblast-like-1 cells (HBFL-1).

Other Cell Lines

Purified primary breast fibroblasts were obtained from reduction mammoplasties or from within a normal appearing area of the HBFL-1 biopsy and cultured in the presence of 20% FCS as previously described to obtain experimentally generated myofibroblasts. 12 In some instances the cells were immortalized by transduction with a retroviral construct containing the HPV16 E6 and E7 oncogenes (ATCC) or the catalytic subunit of human telomerase hTERT ( 14 kindly provided by Dr. Judy Campisi, Lawrence Berkeley National Laboratory). For zymography MCF-7 subline 9 cells, kindly provided by Dr. Per Briand, 15 were cultured alone or recombined with HBFL-1 or primary breast fibroblasts in DME-F12. HMT3909S1, HMT3522T4-2 and MCF-7 (5% FCS) cells were cultured as previously described. 16-18 For induction of α-sm actin 20% fetal calf serum and/or 100 pg/ml TGF-β1 (Sigma Chemical Co., T-1654, Sigma-Aldrich, Vallensbeck Strand, Denmark) were added to the media for 6 days (up to 17 days for the MCF-7 cells), and the medium was changed every second day. For zymography (see below) HT-1080 cells were cultured in RPMI supplemented with 10% FCS. As controls for characterization of HBFL-1 cells we also used primary luminal epithelial cells and myoepithelial cells cultured as previously described. 19

Immunocytochemistry

The original routine formalin-fixed, paraffin-embedded biopsy was sectioned at a 4 μm thickness and stained for α-sm actin (1A4, Sigma), Vimentin (3B4 Boehringer Mannheim Roche, Hvidovre, Denmark), keratin CK7 (OV-TM, DAKO, Glostrup, Denmark) and Wide range keratins (MNF116, DAKO). The antibodies were visualized by streptavidin-biotin (DAKO 5004).

Cell cultures were stained for immunofluorescence as previously described, and double stainings were performed with isotype specific antibodies. 20,21 Immunoperoxidase cytochemistry and quantification of immunostainings were performed as previously described. 13,19 The following primary antibodies were used: cytokeratin K14 (C-8791, Sigma), EDA-fibronectin (DH1; Locus Genex, Helsinki, Finland), K17 (M-7046, DAKO), Ki-67 (M722, DAKO), P53 (M7001, DAKO), uPA (kindly provided by Dr. Boye Schnack Nielsen, The Finsen Laboratory, Copenhagen, Denmark), K18 (F3006, Monosan, Am Uden, The Netherlands), K19 (M772, DAKO), pankeratins (which stained also myoepithelial cells; LP34, M717, DAKO, 22 ); vimentin (6400-301, Labsystems, Helsinki, Finland), BG3C8 (kindly provided by Dr. J. E. Celis; the Danish Cancer Society, Denmark 23 ), β4-integrin (MAB 1964, Chemicon International Inc., Temecula, CA), α1-integrin (VLA-1, T Cell Diagnostics, Cambridge, MA), αV-integrin (MAB 1953, Chemicon), P- and E-Cadherin (NCC-CAD-299 and HECD-1; Kindly provided by Dr. Atsushi Ochiai, Tokyo, Japan), desmoplakin I+II (clone DP1&2–2.15; Boehringer Mannheim, GmbH), maspin (M45620, Transduction Laboratories, Lexington), connexin 43 (03-6900, Zymed Laboratories Inc., South San Francisco, CA), smooth muscle myosin heavy chain (kindly provided by Dr. V. E. Koteliansky 21 ), laminin (M638, DAKO), sialomucin (MAM6, clone 115D8, kindly provided by Dr. J. Hilgers, The Netherlands), estrogen receptor (M7047, DAKO), Thy-1 (ASO-2, Dianova, Hamburg, Germany; 24 ), 1B10 (MC-48, Sigma), α-sm actin, (A-2547, Sigma), type IV collagen (PHM12, Silenus, Victoria, Australia), HGF-receptor (c-met; 05-237, Upstate Biotechnology, Lake Placid, NY), and CALLA (J5, Coulter Immunology, Miami, FL). Immunofluorescence was visualized using a Zeiss LSM 510 laser scanning microscope (Carl Zeiss, Jena, GmbH).

Karyotyping

Chromosome preparations were made from exponentially growing cultures treated for 1.5 hours with colcemid (Gibco BRL, Invitrogen, Tåstrúp, Denmark) at final concentrations of 0.1 μg/ml. Cells were further harvested, centrifuged, and treated with hypotonic solution as previously described. 25 The cell pellet was fixed in methanol:glacial acetic acid (3:1) dropped on cold slides and stained for Giemsa bands. Chromosome counts were done on well-spread metaphases.

Telomeric Repeat Amplification Protocol (TRAP) Assay

Experimentally generated myofibroblasts and HBFL-1 cells were grown to 70 to 80% confluence, trypsinized and counted. Cells (10⁶) were washed twice in PBS, pelleted, and stored at −80°C. The telomerase activity was determined with the TRAP assay 26 using the TRAPeze Telomerase Detection Kit (Intergen, Oxford, UK, purchased from AHdiagnostics, Aarhus, Denmark) according to the manufacturer’s instructions.

Human Androgen Receptor (HUMARA) Clonality Assay

Laser pressure catapulting was performed as described using 500 to 1000 cells for lysis. 27 To isolate DNA, the lysate was phenol-chloroform extracted with 200 μl of 25:24:1 phenol:chloroform:isoamyl alcohol (Sigma, P2069), followed by ethanol precipitation in the presence of glycogen (Roche, Hvidovre, Denmark). DNA extraction from HBFL-1 and tumor and normal breast specimens was performed with the Qiagen, QIAamp Tissue Kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions.

The HUMARA clonality assay was performed in the original formulation with a few modifications. 28 The extracted DNA was divided in two, and incubated over night at 37°C in a final volume of 20 μl with or without 10 U HpaII (Roche) according to the manufacturer’s recommendations. It was heat inactivated for 10 minutes at 95°C, and 5 μl was PCR-amplified in a 15 μl volume. HPLC-purified primers were as follows: sense: 5′tcc aga atc tgt tcc aga gcg tgc-3′; and FAM-labeled antisense: 5′FAM-gct gtg aag gtt gct gtt cct cat-3′ 28 (Applied Biosystems). The cycling conditions were 15 minutes at 95°C, followed by 28 to 32 cycles of 95°C (1 minute), 60°C (1 minute), 72°C (1 minute), and extension (7 minutes). The PCR product (3 μl), together with a 350 TAMRA internal size marker for assessment of allelic sizes, was electrophoresed on an ABI PRISM 310 Genetic Analyzer using the Performance Optimized Polymer-4 (Applied Biosystems, Perkin Elmer). Fluorescent peaks were identified using the Genescan software version 3.1 (Applied Biosystems). Clonality ratios were calculated according to Willman et al, 29 and a clonal ratio of less than 0.5 was considered indicative of a clonal composition. 30,31

Two-Dimensional Gel Electrophoresis

Cells were treated and submitted to two-dimensional gel electrophoresis as previously described. 32

Tumorigenic Potential

HBFL-1 cells between passages 21–38 were cultured in T-75 flasks for transplantation. Confluent cultures were trypsinized and resuspended in DME-F12. Approximately 10⁷ cells were inoculated adjacent to the fourth mammary gland of athymic female mice. A total of six mice were inoculated and were observed weekly for tumor formation for more than 3 months. In another series of experiments, HBFL-1 cells were co-inoculated with MCF-7 cells (n = 6 mice). Finally, two series of experiments (a total of 18 mice) were conducted to see the possible effect of implantation of cells in laminin-rich gels. The gels consisted of 300 μl of 20% reconstituted basement membrane (Matrigel, Engelbreth-Holm-Swarm matrix, Collaborative Biomedical Products, purchased from Becton Dickinson, Catalog no. 40230A, lot no. 913156) and 80% Vitrogen (Collagen Corporation, Fremont, CA 33 ) containing either HBFL-1 cells, cloned HBFL-1 cells (S4), MCF-7 cells, HBFL-1+MCF-7 cells, primary experimentally generated myofibroblasts or primary experimentally generated myofibroblasts from reduction mammoplasties+MCF-7 cells. Each cell type was added to the gel at a concentration of 10⁶ cells. The gels were implanted in triplicate subcutaneously in the right flank adjacent to the fourth mammary gland following 24 hours in culture. The tumors were observed weekly for 1 month and the final size determined by weight. Significance was assessed by Student’s t-test.

Response to Cytokines

HBFL-1 cells cultured in DME-F12 without serum were exposed to 100 pg/ml TGF-β1 (T-1654, Sigma) or vehicle (4 mmol/L HCl with 1 mg/ml BSA), and the level of induction of α-sm actin was analyzed after 6 days by immunoperoxidase cytochemistry. 34 The effect of FGF-2 (basic fibroblast growth factor, bFGF, Pepro Tech EC Ltd., London, UK) on this induction was tested by concurrent exposure to 1 ng/ml FGF-2. TGF-β1 experiments are described above (see other cell lines and biopsies used).

Reconstituted Basement Membrane Assay

HBFL-1 cells or experimentally generated myofibroblasts (10⁵) were embedded as single cells in 300 μl reconstituted basement membrane (Matrigel) and allowed to gel for 30 minutes. at 37°C. The cultures were kept for 2 to 3 weeks before the gels were frozen and stored. Some gels were sectioned and stained with immunoperoxidase for laminin, keratin K19, β4-integrin, and type IV collagen.

Tumor Environment Assay

Collagen gels (Cellon S. A., Strassen, Luxemburg) were prepared at a concentration of 2 mg/ml at a volume of 2 ml per well in 6-well dishes (Nunc) as previously described. 17,21 2.2 × 10⁵ HBFL-1 cells or primary breast fibroblasts and 5 × 10⁵ MCF-7S9 cells, either separately or together. For substrate gels (MMP2, see below) medium was collected and frozen for later use. At the end of the experiments, gels were snap frozen in −80°C N-hexane, mounted, and sectioned for histological examination. 17

Zymography

Zymography was performed using 7.5% SDS-PAGE (BioRad) containing 0.1% gelatin as substrate for gelatinase activity. The serum-free supernatants from the tumor environment assay were concentrated using Centriplus concentrator, 10 kd cut-off (Amicon, Bedford, MA), and the protein concentration of each sample was determined spectrometrically (Beckman), adjusted to 1000 μg/ml. MMP bands were developed by incubation for 24 hours in 50 mmol/L Tris-HCl, 0.2 mol/L NaC1, 5 mmol/L CaCl₂, pH 7.4.

Reverse Transcription-PCR

Total RNA from HBFL-1 cells cultured in DME-F12 with or without 1 ng/ml FGF-2 (see above), was isolated with the TRIZOL reagent (Life Technologies, Roskilde, Denmark) according to the manufacturer’s instructions. RNA (2 μg) was DNase-treated (DNase I Amp Grade, Gibco BRL) and served as a template for first strand cDNA synthesis with an oligo dT primer (Superscript Preamplification System, Gibco BRL). One μl of cDNA was subsequently PCR-amplified using Expand High Fidelity PCR System (Roche, Hvidovre, Denmark) and synthetic oligonucleotide primers for urokinase plasminogen activator (uPA) 5′-AAAATGCTGTGTGCTGCTGAC-3′ and 5′-CCCTGCCCTGAAGTCGTTAGTG-3′, product size 704 bp, 35 stromelysin-3 (STR-3) 5′-CCATGGCAGTTGGTGCAGGAGCAG and 5′-CTGCAGTCATCTGGGCTGAGACTC, product size 401 bp, and β-actin 5′-CACTGGCATCGTGATGGACTC-3′ and 5′-TCTGCATCCTGTCGGCAATGC-3′, product size 500 bp (TAG Copenhagen, Copenhagen, Denmark). PCR conditions for uPA were 94°C, 58°C, and 72°C for 1 minute each, 26 cycles for uPA and 94°C, 64°C and for β-actin, 72°C for 1 minute each, 18 cycles. Amplifications without reverse transcription were included as controls.

HBFL-1 Cells Resemble Normal Fibroblasts and Myofibroblasts

A new cell line, HBFL-1, was established permanently in culture from a breast carcinoma. Initial morphological characterization showed that under serum-free conditions, the cells resembled non-activated primary fibroblasts (Figure 1A, a and b) ▶ . 17 In the presence of serum, they took up an appearance very similar to that of activated primary fibroblasts (Figure 1A, c and d) ▶ . 34 Two-dimensional gel electrophoresis of total proteins and staining for a panel of breast cell markers revealed a protein fingerprint which clearly resembled that of resident fibroblasts, the most common precursor of myofibroblasts (Figure 1A, e and f ▶ , and Table 1 ▶ ). 21 For lists of markers see also Refs. 21, 32, 36, 37 .

We next determined how closely the tumor-derived myofibroblasts functionally resembled experimentally generated myofibroblasts derived from normal stroma. First, a hallmark of normal breast fibroblasts is their ability to convert to myofibroblasts on TGF-β1 stimulation, and the ability of bFGF (FGF-2) to inhibit this conversion. 34 HBFL-1 cells mimicked this behavior in a serum-free medium: the baseline level of α-sm actin positive cells was 14 ± 8%. TGF-β1 induced this level to 45 ± 8%, and FGF-2 reduced it even below the baseline to 8 ± 2%. Second, others have produced evidence that the addition of FGF-2 to the typical peritumoral myofibroblast induces uPA. 38 In this respect also the cell line conformed with the definition of myofibroblasts both at the mRNA (Figure 1B) ▶ and protein levels of uPA (not shown). Third, as assessed by RT-PCR analysis, HBFL-1 produced stromelysin-3 as do typical myofibroblasts 39 (data not shown). Fourth, the tumor-derived myofibroblasts were non-tumorigenic (0/6 mice) when inoculated under standard conditions.

Finally, as another evidence of the typical myofibroblast behavior, we analyzed the interaction of HBFL-1 cells with cancer cells in a tumor environment assay. 17,21,40 Cross-talk between the two cell types was clearly taking place, because tumor cells (MCF-7) could induce both inactive and active isoforms of MMP2 in HBFL-1 cells (Figure 1C) ▶ . The targeting of MMP2 activity to HBFL-1 cells was confirmed immunocytochemically (Figure 1D, a and b) ▶ . Similarly, peritumoral myofibroblasts expressed EDA-fibronectin and α-sm actin (Figure 1D, c–f ▶ ). Reciprocally, the fibroblasts allowed tumor cells to spread in the collagen gel (Figure 1C) ▶ and induced strong and widespread staining of Ki-67 in the tumor cells (Figure 1D, g and h) ▶ .

To ascertain a functional significance of HBFL-1 also in vivo, inoculates were made subcutaneously in the region of the fourth mammary gland of nude mice in the presence of Matrigel 33 with or without MCF-7 cells. Both HBFL-1 and experimentally generated myofibroblasts significantly augmented tumor formation by MCF-7 cells (3.5-fold, P < 0.01 in one experiment, Figure 2 ▶ , and 7.0 in another experiment, not shown) as evaluated by weight of the dissected tumors. Staining of tumors with human specific antibodies revealed a histological appearance of HBFL-1 cells in the peritumoral stroma very similar to that seen with experimentally generated myofibroblasts (Figure 2) ▶ . Double staining with human specific vimentin and α-sm actin revealed that the inoculated HBFL-1 cells and experimentally generated myofibroblasts resembled typical myofibroblasts also in vivo (Figure 2) ▶ . It is concluded that tumor-derived myofibroblasts facilitate epithelial tumor growth in vivo.

Evidence for a Neoplastic Origin of HBFL-1

That HBFL-1 cells were in fact from a neoplastic origin, and were thus derived by EMT, was suggested by the following observations.

The cells were spontaneously immortal as evidenced by the presence of telomerase activity (Figure 3A) ▶ 41 and viability for more than 3 years (70+ passages) in culture. Also, the karyotype was highly aneuploid with a modal value of 87 already in passage 9 (not shown; earlier passages not done). Moreover, in the carcinoma of origin we found a frequent and intimate association of regular nests of epithelial cancer cells with elongated fibroblastoid cells which stained for keratins with CAM 5.2 (Figure 3B) ▶ . All fibroblastoid cells stained for vimentin and some stained for α-sm actin (not shown). Normal stroma in the periphery of the cancer biopsy was completely keratin-negative (not shown). The epithelial stromal relationship was further substantiated by the fact that HBFL-1 and laser pressure catapulted carcinoma cells from the tissue of origin were clonal with an identical non-random X-chromosome inactivation pattern (ratio less than 0.5) 30,31 as assessed with the HUMARA clonality assay (Figure 3C) ▶ . A rudimentary epithelial phenotype was revealed in the majority of cells also by double staining with antibodies against keratins and α-sm actin (Figure 4a) ▶ . Identical data were obtained in clonal cultures (not shown). Two-dimensional gel electrophoresis of cytoskeletal extracts confirmed a weak expression of simple keratins (Figure 4b) ▶ which was not seen in normal fibroblasts (not shown). To show a broader evidence that a combined induction of α-sm actin and loss of keratin could arise as a consequence of an EMT phenomenon was demonstrated by use of two additional basaloid breast carcinoma cell lines (HMT-3909S1 and HMT-3522T4). Under standard culture conditions these cell lines were vimentin- and keratin-positive and entirely α-sm actin-negative (not shown). In the presence of serum (HMT-3522T4)- or TGF-β (HMT3909S1) cells, they down-regulated keratin and induced α-sm actin (Figure 4, c and d) ▶ . Luminal epithelial MCF-7 cells failed to respond to both serum and TGF-β treatments (not shown). To exclude the possibility that the keratin expression was an artifact of immortalization induced in culture, we compared HBFL-1 cells with two breast fibroblast cell lines we established from primary breast fibroblasts after infection with a retroviral construct containing either E6/E7 or hTERT. While these cells had extended life spans, neither expressed keratins (not shown).

Finally, inside a reconstituted basement membrane (Matrigel), HBFL-1 differed from normal breast fibroblasts taken from the same patient by the ability to form cellular clusters. The cells further stained for the epithelial markers β4-integrin, type IV collagen (not shown) and keratin and laminin (Figure 5A, a ▶ inset). We found no staining for E-cadherin or desmoplakin. The clusters produced elongated cells which projected into the gel (Figure 5A, a) ▶ . Normal fibroblasts from the same specimen, either formed only elongated cells, or remained as rounded single cells (Figure 5A, b) ▶ , a morphology also observed in experimentally generated myofibroblasts 17 (data not shown). Also, when transplanted into nude mice in the absence of MCF-7 cells, HBFL-1 cells inside a laminin-rich gel did not form solid tumors per se but mostly resembled resident fibroblasts in an interstitial stroma (compare Figure 2d and 5B, b ▶ ▶ ). Nevertheless, evidence for a rudimentary neoplastic epithelial phenotype of HBFL-1 cells in vivo was provided by the occasional clonal appearance of microfoci of vimentin-negative epithelial cancer cells surrounded by keratin-negative fibroblast-like cells analogous to those obtained in Matrigel cultures (Figure 5B, a) ▶ . Experimentally generated (normal) myofibroblasts which showed no signs of epithelial plasticity in mice inoculates (n = 3 mice) served as controls (Figure 5B, b) ▶ .