Greater than 40% of patients with squamous cell carcinoma (SCC) of the oral cavity have lymph node metastasis at the time of diagnosis and a 5-year survival rate of less than 50%. Changes in gene expression that regulate metastasis of SCC to lymph nodes have not been identified.
To study metastasis of oral SCC, highly metastatic oral SCC cell lines from a poorly metastatic oral SCC cell line were established by in vivo selection using a lymph node metastatic mouse model. The metastatic potential of the cells was studied using Matrigel invasion and cell surface protein adhesion assays. mRNA and protein encoded from metastasis-related genes in the metastatic derivatives and in their parental cells were examined using Northern blot analysis, immunoblotting, rapid analysis of gene expression, and a cDNA microarray technique.
The in vivo selected metastatic cells showed much higher Matrigel invasion capability than the parental cells. They also showed alterations in their adhesion properties to three cell surface proteins. Comparison of metastatic and nonmetastatic cells revealed several significant alterations in the expression of metastasis-related genes, including up-regulation of the urokinase-type plasminogen activator receptor, integrin β1, membrane type 1-matrix metalloproteinase, and down-regulation of protease-activated receptor-1.
In the United States, 15,600 people died of squamous cell carcinoma (SCC) of the oral cavity in 2000.1 Nearly 50% of patients with oral SCC present with clinical or pathologic evidence of lymph node metastases. The 5-year survival rate is less than 50% for patients with a single unilateral lymph node metastasis and less than 25% for patients with bilateral metastases.1, 2 Unfortunately, the overall survival rate among SCC patients has not improved much over the past 30 years.
One of the difficulties in studying the metastasis of human oral SCC is the lack of an appropriate animal model. The oral cavity provides a unique physiologic environment for tumor cell growth. One of the characteristics of human oral SCC is that it frequently spreads to cervical lymph nodes. Although nude mouse models have been developed for studying the formation and propagation of human oral SCC, cervical lymph node metastasis is rare in these models.
Another difficulty in studying oral SCC is that it is highly heterogeneous.3 However, it is possible to select highly metastatic cells from a population of poorly metastatic tumor cells using an animal model.4 For example, Clark et al.5 have established highly metastatic melanoma cell lines from a parental cell line with a low metastatic potential, using in vivo selection followed by intravenous injection of the melanoma cells and isolation of the pulmonary metastases. The poorly metastatic parental cell line and its metastatic derivatives became valuable material for analyzing the expression of metastasis-related genes.
Metastasis is a multistep process requiring a variety of genetic alterations in the tumor cells. Abundant experimental findings indicate that genes encoding basement membrane-degrading enzymes and cell surface receptors, such as matrix metalloproteinases (MMPs), urokinase-type plasminogen activator (uPA) and its receptor (uPAR), integrins, and protease-activated receptor-1 (PAR-1), may play an important role in tumor cell invasion and metastasis. However, differential expression of these genes in metastatic oral SCCs compared with genetically related nonmetastatic human oral SCCs has not been reported. It is a challenge to determine the crucial alterations of gene expression that regulate the metastasis of human carcinoma.
To gain insight into the alterations in gene expression that govern metastasis of oral SCC, we established highly metastatic oral SCC cell lines from a poorly metastatic oral SCC cell line using in vivo selection in a floor-of-mouth (FOM)-lymph node mouse model. We describe the lymph node mouse model, in vitro invasion, and adhesion capabilities of the highly metastatic cell lines selected by this model, as well as the differences in the expression of genes encoding uPAR, integrin β1, PAR-1, MMP-2, MMP-9, and membrane type 1-MMP (MT1-MMP) between the poorly metastatic parental cells and the highly metastatic derivatives, as shown by Northern blot analysis and immunoblotting. Expression of the genes encoding PAR-1, proMMP-2, and proMMP-9 was reduced significantly in the highly metastatic cell lines compared with their poorly metastatic parental cell line. However, the most active forms of MMP-2 and MMP-9 were detected only in the metastatic derivatives. Furthermore, the expression of genes encoding uPAR, integrin β1, and MT1-MMP was elevated in the metastatic cells, supporting the observed high invasive activity of these cells. Some of the results were corroborated by cDNA microarray analysis or rapid analysis of gene expression (RAGE), suggesting that certain alterations in metastasis-related gene expression may favor metastasis of oral SCC and that these cell surface proteins may play major roles in the process.
MATERIALS AND METHODS
The oral SCC cell line 686LN was established from a human lymph node metastasis and was kindly provided by Dr. Peter G. Sacks (Memorial Sloan-Kettering Cancer Center, New York, NY).6 The cell line was maintained as a monolayer culture in Dulbecco's modified Eagle's medium (DMEM)/F12 medium (1:1) supplemented with 10% fetal bovine serum (FBS).
In Vivo Selection of Highly Metastatic Cell Lines by a Modified FOM Mouse Model
All animal experiments were approved by the Animal Care and Use Committee of The University of Texas M. D. Anderson Cancer Center. Nude mice, aged 4–6 weeks, were injected with 1–5 × 106 686LN cells suspended in 0.05 mL of Hanks-buffered saline into the submandibular to mylohyoid muscle as described by Simon et al.7 When the primary tumor reached a length of 2.0 cm, it was removed by survival surgery. The mice were killed only when the tumor attained its original size. Cervical lymph nodes were collected and divided into two sections. One section was fixed in 10% buffered formalin and embedded in paraffin. The tissue blocks were cut and stained with hematoxylin-eosin. The tissue sections were examined microscopically by two individuals and reviewed by a pathologist. The number of tumor metastases was recorded for each animal. The second half of the lymph node was cultured in DMEM/F12 medium supplemented with 10% FBS. The individual cell line recovered from each mouse lymph node metastasis was reinjected into another mouse as described above and the whole selection process was repeated until the injected cell line produced a significantly higher incidence of lymph node metastasis than the parental cell line.
Matrigel Invasion Assay
The Matrigel invasion assay was performed using the Matrigel basement membrane matrix according to the manufacturer's protocol (Becton Dickinson Biosciences Discovery Labware, Bedford, MA). Briefly, 2.5 × 104 cells in 0.5 mL of serum-free medium were seeded in the invasion chamber containing the Matrigel membrane (27.2 μg per chamber) in triplicate and allowed to settle for 8 hours. NIH3T3-conditioned medium was added as a chemoattractant in the lower compartment of the invasion chamber. The chambers were incubated for 36 hours at 37 °C in a 5% CO2 atmosphere. The invading cells appeared at the lower surface of the membrane. The upper surface of the membrane was scrubbed with a cotton swab and the absence of cells in the upper surface was confirmed under the light microscope. After the cells were fixed and stained with Baxter Diff-Quik stain kit (Allegiance Healthcare Corp., McGaw Park, IL), the membrane was placed on a microscope slide with the bottom side up and covered with immersion oil and a cover slip. Cells were counted under a microscope as a sum of 10 high power fields that were distributed randomly on the central membrane. The experiment was repeated four times.
Cell Surface Protein Adhesion Assay
The cell surface protein adhesion assay was performed using CytoMatrix cell adhesion strips coated with human fibronectin, vitronectin, laminin, collagen I, and collagen IV (Chemicon International, Temecula, CA). Cells were dissociated with 3 mM ethylenediaminetetraacetic acid in phosphate-buffered saline (PBS) and resuspended in a regular medium as a single-cell suspension. Next, approximately 1.5 × 105 cells were seeded into each well in seven repeats and incubated at 37 °C for 3 hours. After washing the cells with PBS containing Ca2+ (1 mM) and Mg2+ (1 mM), the adhesive cells were visualized with crystal violet, the absorbency of which was determined at 540 nm on a microplate reader. The experiment was repeated twice.
Northern Blot Analysis
mRNAs were prepared using the Oligotex mRNA purification kit (Qiagen, Valencia, CA) from cell lines established from the primary tumors (T3a and T3b) and from metastases (M3a2, M3a3, M3b2, and M3b3) generated in the third round of in vivo selection. Two micrograms of mRNA from each of the cell lines was used for Northern blot analysis. The 1.0-kb par-1 cDNA fragment was obtained from Dr. Shaun R. Coughlin (The University of California at San Francisco, San Francisco, CA). The 800-bp uPAR cDNA was provided by Dr. Douglas D. Boyd (M. D. Anderson Cancer Center). The 1.5-kb glyceraldehyde-3-phosphate dehydrogenase (G3PDH) cDNA fragment was used as a control. These cDNAs were labeled with 32P-dCTP and used as probes for the Northern hybridization. The mRNA was quantified by image analysis using an Alpha Imager 2000 system (Alpha Innotech Corporation, San Leandro, CA). The relative levels of mRNA in each sample were determined by normalization with the G3PDH in the same sample. Each experiment was repeated twice.
Cells were cultured in the absence of serum for 24 hours before collecting the medium for immunoblotting analysis of secreted MMP protein. Other proteins were prepared from whole-cell lysates. Five to fifty micrograms of protein was used for immunoblotting analysis. The immunoblotting was performed using primary antibodies at a dilution of 1:500 or 1:1000. Protein signals on the blotting membrane were detected by an enhanced chemiluminescence detection reagent (Amersham Life Sciences, Arlington Heights, IL). Monoclonal antibodies used in the experiment were PAR-1-SPAN 12 (Immunotech-Coulter, Miami, FL); uPAR-10G7 (Santa Cruz Biotechnology, Santa Cruz, CA); MMP-2-Ab-4, MMP-9-Ab-2, and MT1-MMP-Ab-3 (Oncogene Research Products, Boston, MA); polyclonal anti-integrin β1 antibody clone 18 (BD Transduction Laboratories, San Diego, CA); and β-actin as an equal loading control. The protein level was obtained by image analysis using the Alpha Imager 2000 system. Relative levels of the protein in each cell line were determined by normalization with β-actin in the same cell line. Each analysis was repeated twice.
RAGE was performed as described by Wang et al.8 In brief, double-stranded cDNA was synthesized from mRNA using a kit from Gibco-BRL Life Technologies (Rockville, MD), but biotinylated oligo-dTn was substituted as the primer for first-strand synthesis. The cDNA was first digested with DpnII. The 3′-most DpnII fragment of each cDNA was adsorbed to stretavidin-magnetic beads (Dynal, Lake Success, NY) and ligated to polymerase chain reaction (PCR) linker B. The nonbiotinylated fragments were removed and the preparation was digested with NlaII. The released fragment was ligated with PCR linker A. These fragments of cDNA containing the gene-specific targets ligated to the B and A linkers were referred to as B/A bitags. The second preparation, A/B bitags, was obtained by performing NlaII digestion with the linker A ligation before DpnII digestion and the linker B ligation. The B/A and A/B bitags prepared from the two comparable samples were standardized together using known housekeeping genes before RAGE was performed to determine the practical amount of the bitags to be used in the experiments. The target gene-specific primers for RAGE-PCR were selected from a database at http://www.capitalgenomix.com (KPL/Capital Genomix, Gaitherburg, MD). Therefore, the size of each amplified DNA fragment had been defined. The fluorescent image of the PCR products was digitized with a FluorImager (Storm 860; Molecular Dynamics, Sunnyvale, CA). Each analysis was performed in duplicate and repeated once.
cDNA Microarray Analysis
Total RNA was prepared using Trizol Reagent (Gibco-BRL Life Technologies). Five micrograms of total RNA was then used for single-stranded cDNA preparation using a cDNA cycle kit (Invitrogen, San Diego, CA). The 33P-dCTP–labeled cDNA was hybridized to the human cDNA microarray filter GF204 following the manufacturer's protocol (Research Genetics, Huntsville, AL). The hybridization signals were recorded by a Phosphor Imager (Storm 860; Molecular Dynamics) and analyzed by a computer program (Pathway™) provided by Research Genetics. One pair of samples was hybridized individually to a pair of filters. To eliminate false-positive results caused by the uneven distribution of the cDNA spots, the hybridization was repeated after the two filters were swapped. Only the signals that appeared in both hybridization studies were considered positive.
Establishment of Highly Metastatic Oral SCC Cell Lines and a FOM-Lymph Node Metastatic Model
The xenograft procedure is illustrated in Figure 1A. First, the oral SCC cell line 686LN was injected transcutaneously into the mylohyoid muscle near the FOM in four mice. Tumors appeared about 10 days after the injection. The tumors were removed surgically in the third week to keep the mice alive. The mice were killed 5–8 weeks after the implantation. In the first round of selection, no metastases were found in the cervical lymph nodes using microscopic histopathologic evaluation. However, when the lymph node tissue was cultured, the 686LN-M1 cell line was isolated and examined by the trypsin-Giemsa–banding technique to ensure the human origin of the cells. All cells contained human chromosomes in metaphase (data not shown). In the second round of selection, injection of 686LN-M1 cells resulted in microscopically detectable lymph node metastases (one lymph node metastasis among five tumors). One of the cultured cell lines (686LN-M2) from the metastasis, obtained at the second in vivo selection, was injected into mice for the third round of selection. Tumors grew in eight of nine mice and lymph node metastases developed in four mice (Fig. 1B,C; Table 1). Parental cells from 686LN were also injected into five mice. Although tumors developed at the injection site in all of these mice, no metastasis was observed histopathologically (Table 1). The lymph node metastases from different mice (mouse-a or mouse-b) were cultured further and labeled alphabetically as 686LN-M3a, 686LN-M3b, and so on. The primary tumors were also cultured and identified as 686LN-T3a, 686LN-T3b, and so on. Because their metastatic capacity was greater than that of the poorly metastatic parental cells, cell lines such as 686LN-T3a, 686LN-T3b, 686LN-M3a2, 686LN-M3a3, 686LN-M3b2, and 686LN-M3b3 were used for the gene expression analysis compared with their parental cells.
Table 1. Spontaneous Lymph Node Metastases in Mice Injected with In Vivo Selected 686LN Cells
The number of lymph node metastases was recorded independently by two individuals.
The difference between the two groups is statistically significant (P < 0.04) by t test.
Matrigel Invasion and Adhesion Capabilities of Metastatic Sublines
The invasion and adhesion capabilities of the metastatic sublines were examined in vitro by Matrigel invasion and cell surface protein adhesion assays as surrogate measures for their in vivo metastatic potential. The results showed that the invasion capabilities of all four metastatic cell lines (686LN-M3a2, 686LN-M3a3, 686LN-M3b2, and 686LN-M3b3) were 13-fold higher on average than that of their nonmetastatic parental cell line 686LN (Fig. 2). Cell line 686LN-T1, established from 686LN tumor tissue in the first round of in vivo selection, showed an invasion activity similar to 686LN. This suggests that the enhanced invasion capability of metastatic cells resulted from intrinsic genetic alterations, not from interactions with the mouse tissue environment.
To further check the interaction between tumor cells and their surrounding cellular matrix proteins, cell surface protein adhesion assays were performed (Table 2). The metastatic cells showed 5–10-fold lower adhesion activity to laminin than nonmetastatic parental cells 686LN and 686LN-T1. A slight reduction of adhesion to collagen IV was also observed in the metastatic sublines. In contrast, adhesion of the metastatic cells to fibronectin was about fourfold higher than that of 686LN-T1 cells, but it was not significantly different from that of 686LN cells. There was no significant alteration in adhesion of the metastatic sublines to vitronectin and collagen I compared with 686LN and 686LN-T1 (data not shown).
Table 2. Analysis of Squamous Cell Carcinoma Adhesion to Cell Surface Proteinsa
OD: optical density.
The P value was calculated by t test.
OD/cells seeded × 105 ± SD
1.97 ± 0.51
2.20 ± 0.17
0.18 ± 0.09
0.45 ± 0.15
0.36 ± 0.18
0.01 ± 0.007
P to 686LN
7 × 10−6
4 × 10−5
P to 686LN-T1
3 × 10−6
3 × 10−7
3 × 10−6
4 × 10−9
OD/cells seeded × 105 ± SD
2.95 ± 0.09
2.63 ± 0.56
1.80 ± 0.02
2.37 ± 0.25
1.81 ± 0.05
2.01 ± 0.34
P to 686LN
4 × 10−9
5 × 10−5
4 × 10−8
5 × 10−7
P to 686LN-T1
OD/cells seeded × 105 ± SD
1.11 ± 0.36
0.38 ± 0.16
1.46 ± 0.33
1.76 ± 0.04
1.24 ± 0.05
1.46 ± 0.07
P to 686LN
P to 686LN-T1
8 × 10−5
2 × 10−6
Expression of Metastasis-Related Cell Surface Receptors
Production of three metastasis-related cell surface receptors, PAR-1, uPAR, and integrin β1, was examined in two primary tumor cell lines and in four metastasis cell lines obtained from the third round of in vivo selection and compared with 686LN parental cells. Both Northern blot and immunoblot analyses demonstrated that PAR-1 levels decreased by 60–80%, whereas the expression of uPAR increased about fivefold in the highly metastatic derivatives (Fig. 3). These observations were consistent with the finding from a reverse transcription-PCR–based RAGE analysis (Fig. 4), which showed that the ratios of 686LN-M3a2 mRNA to 686LN mRNA were 0.28 ± 0.02 for PAR-1 and 6.12 ± 0.01 for uPAR. Immunoblotting analysis showed increased levels of integrin β1 in the metastatic derivatives (Fig. 5), as did cDNA microarray analysis. The integrin β1 mRNA ratios of 686LN-M3a2 to 686LN from two independent experiments were 3.4 and 4.7.
Expression of Metastasis-Related Cellular Matrix-Degrading Enzymes
MMPs are major cellular matrix-degrading enzymes involved in cancer cell invasion. We used immunoblotting to measure the secreted levels of MMP-2 and MMP-9. Surprisingly, the secreted levels of proMMP-2 (72 kilodaltons [kD]) and proMMP-9 (92 kD) were much lower in the metastatic derivatives than in their parental cells (Fig. 6). This was also observed by gelatin zymography (data not shown). However, the most active products of MMP-2 (62 kD) and MMP-9 (67 kD) were not observed in the parental cells. We also used immunoblotting to check expression of MT1-MMP, an activator of MMP-2 and MMP-9. The results showed that MT1-MMP expression was significantly higher in the metastatic cells than in their parental cells (Fig. 7).
We established an in vivo selection procedure for identifying the more biologically aggressive phenotype of cells within a heterogeneous SCC tumor cell population. This has allowed us to further investigate the potential genes that regulate aggressive biologic behavior such as metastasis, which is the single most predictive factor of mortality in patients with SCC of the head and neck.
Dinesman et al.9 initially developed the FOM human tumor model in nude mice to imitate the physiologic environment of the oral cavity. Human laryngeal SCC cells were implanted into the FOM of the mice via an extramural route, providing an excellent model in which to study tumor invasion and evaluate therapeutic agents. However, lymph node metastasis was observed in only 2 of the 42 animals in that study. Simon et al.7 also described an orthotopic FOM cancer model for the quantification of tumor invasion. Consistent with the findings from Dinesman et al., they found lymph node metastasis in only 1 of the 24 animals injected with SCC cells. Because of its unique ability to mimic the physiologic environment of head and neck carcinoma, the FOM mouse model has been used by many researchers to study head and neck SCC. For example, the model was used to study the effect of suicide gene therapy for the treatment of head and neck carcinoma.10 The model was also modified to study cell invasion and metastasis by using an immunocompetent mouse.11 However, a high rate of lymph node metastasis was only accomplished using a murine SCC cell line. The rarity of lymph node metastases from human SCC in mice may be due to natural killer cell activity or other immune responses of the mouse and a balance between tumor cell proliferation and apoptosis.
The behavior of SCC cells in lymph nodes mimicks the seed and soil concept first promoted by Stephen Paget (1889).12 Tumor cells need a compatible environment to survive. Because we achieved a metastasis rate of 50% for human oral SCC in our FOM-lymph node metastatic model, the alterations in gene expression seen in the cell lines established from the lymph nodemetastases indicate that the tumor cells are compatible with the lymph node tissues. Using immunohistochemistry, Saito et al.13 reported that expression of the gene encoding MMP-9 was significantly lower in stromal cells and macrophages along the invasive margin of metastatic colorectal carcinoma than the expression in those along the margin of colorectal tumors that had not metastasized for more than 5 years. In contrast, the expression of the uPAR gene was greater in the metastatic colorectal carcinoma cells than in the nonmetastatic control cells. Consistent with these observations, we also demonstrated that the expression of the genes encoding proMMP-2 and proMMP-9 decreased in metastatic cells compared with their nonmetastatic parental cells, but the uPAR level increased. Although our observation was made in cultured cancer cells whose interaction with the surrounding cellular environment was not existent, we believe that the genetic alteration in the SCC cells that made them compatible with lymph node tissue likely occurs during the in vivo selection. The reduction in proMMP-2 and proMMP-9 in the metastatic cells might also result from a decreasing expression of PAR-1. Liu et al.14 reported that both thrombin and the PAR-1 activation peptide induced MMP-2 and MMP-9 expression in oral SCC cells.
It is generally accepted by researchers in the field that MMPs promote cell invasion and metastasis. However, our data showed that the pro-forms of MMP-2 and MMP-9 were reduced, whereas the most active forms of MMP-2 and MMP-9 existed only in the metastatic cell lines, suggesting that MMP activation enzymes must be very efficient in metastatic cells. On the basis of these findings, we speculate that MMPs have multiple physiologic functions in metastasis and that certain MMPs as well as their modulators, such as plasmin and MT-MMPs, must exist on the surface of different cancer cells for metastasis to occur. In fact, we did find higher expressions of MT1-MMP and uPAR in metastatic cells than in their poorly metastatic parental cells. Both contribute to activation of several MMPs including MMP-2 and MMP-9. Furthermore, these metastatic cells showed a much higher Matrigel invasion capability than their parental cells. Laminin is one of the major components in Matrigel and it is a substrate of many MMPs and plasmin. Cleavage of laminin by cell matrix-degrading enzymes stimulates tumor cell motility and invasion.15 The fact that these selected metastatic cells showed low adhesion to laminin and a high invasion rate through Matrigel further supports that activation of proMMPs may be one of the key steps in metastasis.
Both uPAR mRNA and protein levels were elevated significantly in the metastatic derivatives, suggesting that uPAR protein may be one of the key players in the metastasis of SCC. uPAR is a glycosyl-phosphatidylinositol-linked cell surface protein. As a major receptor of uPA, uPAR promotes uPA activation, which in turn converts plasminogen to plasmin and activates several cell matrix-degrading MMPs, including MMP-2 and MMP-9.16, 17 uPAR also serves as a ligand to integrin α4β1 and induces integrin-mediated signaling transduction via the mitogen-activated protein kinase pathway.18 The elevated level of integrin β1 in metastatic cells supports the interaction between uPAR and integrin.
PAR-1 is a thrombin-activated G-protein–coupled receptor that is involved in multiple cellular signaling pathways and facilitates the proliferation of platelets and some cancer cells.19–22 The expression of PAR-1 was observed in oral SCC and was found to facilitate the proliferation of oral SCC with high levels of PAR-1 in vitro.14 However, we and others14, 23, 24 have also observed that PAR-1 inhibits cell growth, suggesting that it has a dual function in SCC metastasis. The significant reduction in PAR-1 levels in the metastatic cells suggests the importance of this protein in the metastasis of oral SCC. To support our observation, Kamath et al.25 reported that PAR-1–mediated signaling inhibited migration and invasion of breast carcinoma cells. The exact functions of uPAR and PAR-1 in oral SCC metastasis are not yet clear and are being investigated in our laboratory.
The establishment of metastatic cell lines from a poorly metastatic parental cell line in our lymph node metastatic mouse model provided strong evidence for the heterogeneity of oral SCC. Solid tumors contain cell populations with different genetic variations or with different potentials for further genetic alterations during tumor progression. It is important to consider the heterogeneity of oral SCC when defining molecular prognostic markers or searching for crucial genes in oral SCC. Leethanakul et al.26 reported the successful use of laser capture microdissection to procure specific cell populations for cDNA array analysis. Using this method, they identified distinct patterns of gene expression in SCC. Similarly, the in vivo selection in our model also identified a cell population with a specific phenotype, lymph node metastatic potential in this case. The cell lines established from this metastatic cell population provide precise materials for analyzing metastasis-related gene expression.
The authors thank Dr. Michael C. MacLeod for his assistance with the RAGE technique and Dr. Yaan Kang for his assistance with the animal studies.