Pulmonary metastases frequently develop in patients with aggressive bladder cancer, yet investigation of this process at the molecular level suffers from the poor availability of human metastatic tumor tissue and the absence of suitable animal models. To address this, we developed progressively more metastatic human bladder cancer cell lines and an in vivo bladder-cancer lung-metastasis model, and we successfully used these to identify genes of which the expression levels change according to the degree of pulmonary metastatic potential. By initially intravenously injecting the poorly metastatic T24T human urothelial cancer cells into nude mice, and then serially reintroducing and reisolating the human tumor cells from the resultant mouse lung tumors, three derivative human lines with increasingly metastatic phenotypes, designated FL1, FL2, and FL3, were sequentially isolated. To identify the genes associated with the most lung-metastatic phenotype, the RNA complement from the parental and derivative cells was evaluated with oligonucleotide microarrays. In doing so, we found 121 genes to be progressively up-regulated during the transition from T24T to FL3, whereas 43 genes were progressively down-regulated. As expected, many of the genes identified in these groups could, according to the ascribed functions of their protein product, theoretically participate in tissue invasion and metastasis. In addition, the magnitude of gene expression changes observed during the metastatic transition correlated with the in vivo propensity for earlier lung colonization and decreased host survival. To additionally define which genes found in the experimental system were of relevance to human bladder cancer lung metastasis, we evaluated gene expression profiles of 23 primary human bladder tumors of various stages and grades, and then we compared these gene expression profiles to the altered profiles in our model cell lines. Here we found that the expression of epiregulin, urokinase-type plasminogen activator (uPA), matrix metalloproteinase (MMP)14, and tissue inhibitor of metalloproteinase (TIMP-2) were consistently and progressively up-regulated when viewed as a function of tumor stage in tissues of patients versus the metastatic potential seen in the mouse lung model. The strong correlation of these four markers between the experimental and clinical situations helps validate this system as a useful tool for the study of lung metastasis and defines targets of therapy that may reduce the incidence of this process in patients.
Transitional cell carcinomas of the bladder present as either superficial lesions or as tumors that invade the muscle of the bladder wall or beyond. Despite successful treatment of the primary lesion, approximately half of the patients with muscle-invasive cancers eventually develop lung metastases, leading to an estimated 12,710 deaths (1) in the United States in 2004. Genetic events associated with the development of muscle-invasive transitional cell carcinoma include diminished expression of the tumor suppressor genes retinoblastoma (2) and p53 (3) , as well as increased expression of epidermal growth factor receptor (EGFR; refs. 4 , 5 ) and the H-ras proto-oncogene (6 , 7) . However, there still remains insufficient molecular information surrounding the progression of bladder cancer from the muscle-invasive but nonmetastatic (curable) form to those tumors that are metastatic. This limitation stems, in part, from a lack of adequate animal models and available human tumor tissues from metastatic lesions for more comprehensive biological investigation.
To address this need, we developed an animal model of human bladder cancer lung metastasis, and we used it in conjunction with whole genome approaches of gene expression profiling (8) in human tumors to better understand the biology of lung metastasis. Implicit in this analysis was the hypothesis that genes, the expression of which is associated with lung metastasis in human cancers comprise a subset of those associated with progression to muscle-invasive disease.
By serially passaging the poorly metastatic T24T (9 , 10) human urothelial cancer cell line in immunocompromised mice, three derivative lines were serially generated, with each subsequent line exhibiting an increased capacity for lung metastasis over its progenitor. Using a species-specific quantitative PCR assay, we determined that in this model, the degree of early pulmonary colonization by tumor cells was a surrogate indicator of eventual lethal metastatic disease in the animal. Importantly, by using the novel approach of combining gene expression profiles of progressively more lung metastatic cells in the animal model with those of progressively more invasive human tumors, we were able to identify several candidate genes associated with lung metastasis even in the absence of human metastatic tumor tissue. The overall experimental strategy is graphically presented as a flowchart in Fig. 1 ⇓ .
MATERIALS AND METHODS
Development of the Lung Metastasis Animal Model.
T24T (9 , 11) is a tumorigenic but poorly metastatic variant of T24, a tumor derived from a grade 3 invasive transitional cell carcinoma in a female patient (12) . The T24T cell line was maintained in culture as described previously (9 , 11) . Six-week–old NCr nu/nu mice, purchased from Taconic (Germantown, NY), were strictly maintained according to NIH and institutional guidelines. To establish the primary tumors, mice were given an i.v. lateral tail vein injection with 1 × 106 human bladder tumor cells suspended in 0.1 ml of serum-free DMEM medium. Mice were examined and weighed weekly.
At various time points, the lungs were removed and examined grossly. Visual counting of lung metastases was carried out as described previously (11) . Identifiable metastatic lesions were then cut into 1- to 3-mm3 cubes and placed into culture containing DMEM/F-12 media supplemented with 10% fetal bovine serum containing 1× Antibiotic-Antimycotic solution (100 units penicillin G, 100 μg streptomycin sulfate, and 0.25 μg amphotericin B/ml; Invitrogen, Palo Alto, CA). The lung tumor explants were maintained in culture until murine stromal cells were no longer microscopically apparent and then passaged into DMEM/F-12 with 5% fetal bovine serum without Antibiotic-Antimycotic solution. After 3 days of Antibiotic-Antimycotic-free culture, the resulting FL1 (“From Lung” 1) cell line was again injected into a new cohort of mice as described above. Additional repetition of these steps led to the derivation of the FL2 and FL3 cell lines.
In vitro Proliferation, Soft-Agarose, Cytoskeletal, and Cell Migration Assays.
Suspensions of 1,000 cells per well of T24T, FL1, FL2, and FL3 were plated in quadruplicate into 96-well dishes and cultured in DMEM/F-12 with 5% fetal bovine serum at 37°C, 5% CO2. The following day (Day 0), the media from a baseline plate was removed, and the plate was stored at −80°C, whereas for the remaining plates, the media was also removed but replenished with 75 μL of fresh media containing either 5% or 0.5% fetal bovine serum. Each day thereafter, for eight more days (Day 1–8), an additional plate was similarly frozen, and the remaining plates were refed.
To assess proliferation rate, the DNA content of each well was fluorometrically assessed with a CyQUANT Cell Proliferation Assay Kit (Molecular Probes, Eugene, OR) according to the manufacturer’s instructions. Data from Days 1 to 8 were subtracted from the average baseline plate fluorescence (determined at Day 0) and then averaged and plotted as degree of fluorescence versus days in culture. Soft agar clonogenicity for the four cell lines was evaluated as described previously (9) with quadruplicate wells of each cell line; in this case, the plates were imaged and colonies counted after 3 weeks incubation. To evaluate the integrity of the cellular actin cytoskeleton, cells were plated on glass coverslips coated with 2% Matrigel and, after adhering overnight, fixed in paraformaldehyde, permeabilized in 0.1% Triton X-100, and stained with Alexa Fluor 594 Phalloidin conjugate (Molecular Probes) and Hoechst 33342 (Molecular Probes). Slides were mounted in Elvanol and analyzed with a Zeiss Axiovert 135M and IPlab software (Scanalytics, Inc., Fairfax, VA). Cell migration activity was evaluated with the wound assay as described previously (11) .
Molecular Lung Colonization Assay.
To determine differences in early cell line lung colonization after intravenous tumor inoculation, a novel species-specific molecular lung colonization assay was developed. Here, the T24T, FL1, FL2, and FL3 cells in culture were resuspended, and their concentration was adjusted to 1 × 106 cells/100 μL. For each cell line tested, 10 female NCr nu/nu mice were injected with 100 μL of the cell suspension into the tail vein. The mice were maintained for 5 additional weeks and then euthanized. The lungs at necropsy were harvested, and total DNA was isolated with the Puregene DNA purification system (Gentra Systems, Minneapolis, MN). A real-time PCR strategy, with a small portion of chromosome 12p as a target (selected because it is present in the cell lines of interest, does not encode any known gene products, and human specific primers could be readily designed), was used to quantify the human cancer cells within the murine tissue background.
To increase the sensitivity of real-time PCR and decrease the complexity of the background murine DNA, 50 μL nested PCR reactions were carried out before performing the real-time PCR. In this modification, 10 cycles of amplification were done on all of the experimental samples in addition to those of a complex, logarithmic, standard curve set composed of 500 ng murine DNA mixed with serial dilutions of human DNA ranging from 10 pg to 100 ng (effectively at single-cell sensitivity).
Human sequences were ultimately detected with a primer set flanking the primers used for the secondary amplification (forward primer, 5′-AGGGAGTGCAGTGTGTCACTCTAGC-3′ and reverse primer, 5′-CTGACAGACAACCTTGCTCACTCAC-3′). For the secondary amplification, primer set and probe used were from the MGB Eclipse system (Epoch Biosciences, Bothell, WA) and consisted of forward primer, 5′-CATGGTGATGCGGTTTTG-3′, reverse primer, 5′-ATGG*GGTGGAGACTTGGA-3′, and probe 5′-GGGCGTGGATAGCGGTT-FAM Amidite-3′. Initial denaturation/activation was done with a Bio-Rad iCycler thermocycler (Bio-Rad, Hercules, CA) at 95°C for 3 minutes followed by 50 cycles at 95°C for 5 seconds, 64°C for 20 seconds, and 76°C for 5 seconds. Data were acquired during each annealing step when the fluorescent probe was hybridized to the template and, therefore, unquenched. The data were analyzed with iCycler software (Bio-Rad).
In vivo Murine Survival Assay.
We sought to determine whether cell lines capable of generating an increased number of lung metastases are associated with a decrease in disease-specific survival. The T24T, FL1, FL2, and FL3 cells were harvested, and the concentrations were adjusted to 1 × 106 cells/100 μL. Then for each cell line, 10 female NCr nu/nu mice were injected with 100 μL (1 × 106 cells) by tail vein. A terminal endpoint was reached when either of the following occurred: (a) sudden death from tumor; or (b) imminent death (defined as the inability of the mouse to reach food and water), in which case the mice were then euthanized. Kaplan-Meier curves and the log-rank test were used in comparing the survival time distributions among the groups.
Gene Expression Profiling of Bladder Cell Lines in the Lung Metastasis Model and Human Primary Bladder Tumors.
Gene expression analysis of T24T, FL1, FL2, and FL3 was done with the HG-U133A GeneChip array (Affymetrix, Santa Clara, CA) on duplicate RNA samples generated from independent cell line cultures as described previously (11) . Image files from quality validated experimental outputs were processed with Microarray Analysis Suite 5.0 (MAS 5.0, Affymetrix) to generate raw data files. Initially, the detection of a particular gene and the call of “present,” “absent,” or “marginal” in the data sets was also made with MAS 5.0. Expression values for each cell line, as determined by the software algorithms, were then compared across the passage sets to ultimately allow patterns of changes in gene expression as a function of metastatic competence to emerge. The genes so identified were pinpointed with the local-pooled–error test (13) with a false discovery rate P of <0.05, indicating a statistical significance that their expression in the FL1, FL2, and FL3 lines either consistently increased or decreased when compared with the parental T24T line.
Two different approaches were used to explore the functional relationships among the genes with altered expression in the metastasis model. Gene ontology analyses were done with dCHIP1.3/ChipInfo (14 , 15) . The P generated by this method indicates the strength of the association of the cluster of genes to the gene ontology terms or pathways discovered. P < 0.001 was considered significant. The second approach, carried out with the Ingenuity Pathway Analysis (Ingenuity, Mountain View, CA) tool, examined functional associations between differentially expressed genes. To infer coassociations of encoded proteins, this latter method uses the gene identities in conjunction with a controlled, vocabulary data mining of literature associations, protein-protein interaction databases, and the Kyoto Encyclopedia of Genes and Genomes metabolism pathway database (www.genome.ad.jp/kegg/) for the knowledge of historically established, well-described pathways and biological networks.
To determine the genes of which the expression was significantly changed as a function of lung metastasis, which were also associated with aggressive disease in patients, 23 primary tumor samples of various pathological stages [5 Ta, 5 T1, 6 T2, 3 T3, and 4 T4 as reviewed by a single pathologist (Henry F. Frierson)] were obtained with consent and Institutional Review Board approval from patients undergoing surgery for bladder cancer. To serve as a negative control, macrodissected normal bladder mucosa was also available from patients undergoing bladder surgery for benign causes. Samples were processed and hybridized on the HG-U133A GeneChip array as described above, and the data was processed in MAS 5.0 with a scaling factor of 200. Differentially expressed genes identified in the T24T, FL1, FL2, and FL3 cell series were sought in the human tumor profiles, and their expression was assessed in tumors according to stage (stage Ta versus stage ≥T1) by fold change and one-tailed t test. Comparisons to normal mucosa were also made as discussed below.
Validation of Affymetrix GeneChip Results.
To validate the GeneChip expression data, real-time quantitative reverse transcriptase-PCR was done with the same RNA samples used in the GeneChip array experiment, and four genes having significant expression differences between T24T and FL3 were targeted. Suitable Assays-on-Demand primers and probe sets (Applied Biosystems, Foster City, CA) for 18S RNA and the four genes of interest were obtained. These consisted of two, unlabeled PCR primers and a FAM dye-labeled TaqMan Minor Groove Binder probe. Two-step real-time reverse transcriptase-PCR with 5′ nuclease chemistry was carried out by adding an aliquot of cDNA to TaqMan Universal PCR Master Mix, and quantitative gene expression data were acquired on an Applied Biosystems ABI PRISM 7900HT Sequence Detection System (Applied Biosystems). The 18S RNA reverse transcriptase-PCR result was used to standardize results.
Protein expression of selected genes was also carried out by Western blotting as described previously (11) . Primary antibodies used were mouse monoclonal antibody antiannexin I at 1:500 (BD Biosciences, San Diego, CA) and mouse monoclonal antibody anti-α-tubulin at 1:1,000 (Oncogene Research Products, San Diego, CA). Secondary antibodies used at 1:100,000 were antirat horseradish peroxidase (Amersham Biosciences, Piscataway, NJ) and antimouse horseradish peroxidase (Pierce Biotechnology, Rockford, IL). SuperSignal West Femto substrate (Pierce Biotechnology) was applied, and the blots were imaged and quantified with a FluorChem 8800 (α Innotech Corporation, San Leandro, CA).
RESULTS AND DISCUSSION
Development of the FL Series Human Bladder Cancer Lung Metastasis Model.
Because the lungs are a major site of bladder tumor metastasis, we sought to develop a model that would permit molecular evaluation of this phenotype, including the discovery of putative genes associated with bladder cancer lung metastasis in patients. Starting with T24T, a tumorigenic yet poorly metastatic human cell line derived from a grade 3 invasive transitional cell carcinoma (9 , 10 , 12) , a series of cell lines of increasing metastatic capacity were generated. As T24T and its derivatives were serially passaged through nude mice, the cellular character progressively changed, giving greatly increasing pulmonary metastatic load over a decreasing period of time.
Initially, the T24T cells developed visible tumors 11 to 13 weeks after tail vein injection. From the few metastases that developed, the “FL1” line was derived. After a short period in culture, FL1 cells were subsequently reinjected into nude mice, producing several metastases at weeks 7 and 8. The human tumor cells isolated from the lungs of these new animals and derived from the FL1 line, now designated “FL2,” were themselves cultured, again injected, and this time resulted in frequent metastases within 4 to 6 weeks. Finally, cells from the lungs of this second set of animals, termed “FL3” and derived from the FL2 line, when similarly reinjected, filled the lungs with metastases in <4 weeks (Fig. 2A) ⇓ . It is important to note that in order to minimize changes that might occur during prolonged culture, the derivative cells, when adapted to grow in vitro, were expanded only enough to give sufficient cell numbers that would allow freezing and passage in a subsequent host.
Although several models of human bladder cancer invasion (7) or metastasis (16) have been developed, these have been limited by the fact that they generally involved paired cell lines rather than a series of progressively more metastatic tumors selected in multistep fashion in vivo. Pair-wise analysis of such tumor phenotypes is substantially less powerful than analysis of progressively more metastatic sequentially isolated derivatives, because trends in physiology or genetics associated with the metastatic phenotype cannot be easily evaluated in the former group. In other tumor systems such as melanoma, experiments such as those described here are well known (17) and have been examined with gene array technology (8) .
In vitro Phenotypes Associated with Metastatic Competence.
Propensity of a cell line for metastasis could conceivably be because of an intrinsically faster growth rate (increased cell division) or decreased apoptosis (18) . Therefore, evaluation of the in vitro monolayer growth of cell lines was done in both normal serum (5%) and reduced serum (0.5%; Fig. 2B ⇓ ), the latter to simulate limiting growth conditions. The doubling time for these lines was ∼22 hours in full serum and ∼32 hours in reduced serum conditions. Under both conditions, cells exhibited continued growth and lacked contact inhibition with no measurable differences between the different metastatic lines noted.
Morphologic assessment of cultured cells by phalloidin staining revealed increasing dysmorphism as cells gained metastatic competency. T24T cells generally maintained characteristics of epithelial cells, with low nuclear to cytoplasmic ratios, ruffled edges, and uniform size and shape. Nuclear to cytoplasmic ratios were increased along the progression model, and cells became increasingly small and round or irregular (Fig. 2C) ⇓ . Such cellular features are increasingly seen in tumors of increasing stage and grade and are considered factors associated with increased metastatic potential (19) . Hence, our model appears to recapitulate important features of human cancer.
Despite the association between cell migration and metastatic competence, cell migration in the wound assay did not reveal any differences between the four cell lines (data not shown). In contrast, T24T cells grown on Matrigel tended to remain as single cells or cluster into small groups, whereas the more metastatic cells in the series showed increasingly complex aggregations, sending multiple, thin processes between both cells and colonies, creating more organized honeycombed shapes (Fig. 2D) ⇓ . The presence of these tubular structures is consistent with the notion that invasive and metastatic carcinoma cells are thought to be more sensitive to the motility cues that normally regulate epithelial cell movements such as tube or sheet migration (20) . It is also of interest that the prime components of the tumor microenvironment thought by experts as most likely to be factors involved in stimulating cancer cell motility (i.e., extracellular matrix macromolecules, metalloproteinases, and soluble factors) are the products of many of the genes discovered in our system to be universally associated with lung metastatic potential as shown below.
High Lung Colonization Rates at Early Time Points Translate to Decreased Host Survival.
Differences in the metastatic ability of tumors may be because of different rates of early tumor cell colonization in distant organs and/or different rates of tumor cell outgrowth/dormancy (18) at the metastatic site. Hence, we sought to determine whether the observed differences in lung metastatic potential between T24T and FL3 may be because of differences in early lung colony formation rates in the lung. To answer this question, we developed a novel quantitative assay to measure tumor burden in murine lungs after tail vein inoculation long before any gross or microscopic lung metastases appear. Furthermore, we sought to determine whether there was a correlation between early lung colonization, clinical lung metastasis, and subsequent death from these metastases. If such a correlation were to exist, this would lend support to the notion that differences in early lung colonization are important drivers in metastasis development in this model.
Relying on the sensitivity of real-time PCR, a primer/probe set located on a portion of human chromosome 12p was designed to detect the presence of the injected human bladder cancer cell DNA in a background of murine tissue DNA sequences. Lungs of the entire cohort of injected animals were harvested at 5 weeks. With this model, it was shown that no amplification of mouse genomic DNA had occurred (Fig. 3A) ⇓ . A statistically significant (P < 0.05, t test) increase in the amount of human DNA (proportionate to the number of cells) from T24T through FL3 was present in murine lungs after tail vein injection (Fig. 3B) ⇓ .
To determine whether the degree of early lung colonization reflects decreased survival from clinical lung metastatic disease, a survival study was completed with mice after tail vein injection of tumor cells. Fig. 2C ⇓ shows Kaplan-Meyer estimates of host survival after tumor inoculation. None of the mice in the parental T24T cell line died over the course of one year, whereas mice having FL1-FL3 tumors displayed statistically significant increased mortality when compared with them (P = 0.02 for FL1, P = 0.01 for FL2, and P = 0.005 for FL3 versus T24T, respectively). Both FL2 and FL3 tumor-bearing animals showed a clear trend toward greater mortality at study end. This data lends support to the notion that differences in early lung colonization are important drivers in metastasis development in this model. Furthermore, no gross metastatic disease was observed in other organs, such as the liver, at the time of euthanasia in the survival experiment (data not shown).
As with most murine metastasis models, death, significant morbidity, or the time to visible metastasis are endpoints that come at considerable time and expense. The PCR-based molecular metastasis assay developed for this study has been shown to correlate well with Kaplan-Meier survival data. The significant advantages to this assay are its simplicity, the ability to quantitate data objectively, confidence that the results reflect long-term outcome, and a significant decrease in the time required achieving a valid endpoint at ∼1 month. Additionally, this molecular assay can be used to quantify metastasis in any murine organ containing any human tumor cells with the chromosome 12p target. This novel tool allows for rapid, cost-effective (reduced animal housing costs), and more humane (by limiting animal distress because of gross tumor burden) analysis of therapeutic interventions in vivo, including the investigation of drug therapy and manipulation of gene expression.
Gene Expression Profiling of the Metastatic Phenotype.
Oligonucleotide microarrays were used to identify differentially expressed genes among the four cell lines comprising the progression model. Identifying genes involved in this metastatic process might enable additional investigations toward a mechanistic understanding of metastatic pathways as well as potential targets for therapeutic intervention and/or biomarkers of aggressive metastatic disease.
Before carrying out the analysis, which aimed at identifying genes of which the expression was progressively altered with metastasis, confirmation of the results obtained by microarray hybridization was done with real-time reverse transcriptase-PCR to quantitate RNA expression for several randomly selected genes, showing many fold changes between T24T and FL3. As with previous validations (21) , the expression dynamics when compared were concordant (Table 1) ⇓ . In addition, the protein products of one of these differentially expressed genes, Annexin A1, was additionally evaluated and was found concordant to its mRNA expression levels by steadily increasing during phenotypic transition from T24T through to FL3 (data not shown).
In determining which genes were either over- or underexpressed with metastatic progression, we found that 43 had decreased expression, whereas 121 had increased expression. The top 24 genes, ranked by fold differences between T24T and FL3 for each respective group, are shown in Table 2 ⇓ . To identify potential functional relationships among genes with altered expression in the metastasis model, term ontology was first used (based on the Genome Ontology consortium) and revealed several significant groups of genes with keywords as follows: cell adhesion receptor, cell adhesion molecule, hyaluronic acid binding, extracellular, extracellular matrix, extracellular space, cell motility, cell communication, cell adhesion, signal transduction, developmental processes, embryogenesis and morphogenesis, and fibronectin type I, II, and III domains (Table 3) ⇓ . Thus, a majority of the significant gene groups relate to those with a role in the extracellular matrix and cell-cell contact. These data are consistent with the results above, indicating that the cells with enhanced metastatic capacity were those that form tube-like structures on Matrigel, additionally suggesting that genes involved in cell-extracellular matrix interactions are important determinants in metastasis.
Secondly, the Ingenuity Pathways Analysis, a controlled, vocabulary-based pathway tool that seeks associations between genes and their encoded proteins based on a formal extraction of keywords in the peer-reviewed literature, as well as knowledge of protein-protein interaction, biochemical networks, and well-characterized pathways, was applied. This analysis revealed a series of putative pathway relationships of which five had significant scores (determined by the number of differentially expressed genes within each of the networks and the strength of the associations, such as binding or enzymatic activation among pathway members). Functional breakdown of the top two pathways revealed a strong overrepresentation of genes associated with cell invasion and migration, largely driven by the coassociation of fibronectin, tPA, uPA, MMP-14, TIMP-2, and interleukin 8 in pathway 1 (Fig. 4) ⇓ and integrins A3, A6, and B4, amphiregulin, and melanoma cell adhesion molecule in pathway 2 (data not shown). In addition, the next highest scored pathway revealed a significant coassociation of genes involved in the suppression of apoptosis, driven by BCL2-L1, IER3, and BIK. This latter observation supports the notion that suppression of apoptosis is an important determinant of metastatic efficiency (18) .
Epiregulin and Matrix Proteolysis Genes Are Associated with Human Bladder Cancer Lung Metastasis.
Because it is not feasible to obtain tissues from bladder cancer lung metastases in patients, the progressive metastasis model system was used as a surrogate to compare gene expression in pathologically well-characterized, primary human bladder carcinomas of various stages from patients with no evidence of metastatic disease. In doing this, the assumption was made that as tumor stage increases, primary tumors will be progressively populated by more metastatically competent cells, and this is consistent with the hypothesis that cells capable of metastasis exist and progressively dominate advanced primary tumors (22) . This hypothesis has recently been supported by gene expression microarray experiments in human cancers (23) in tumor systems, where both primary and metastatic tissue is available and can be compared. Furthermore, this model is substantiated by the observation that gene expression profiles of primary tumors can predict the proclivity for metastatic spread (24) of breast cancers and has been elegantly argued by Bernards and Weinberg (25) .
The same oligonucleotide array format (21,500 genes) was used to profile a series of 23 lesions, 5 superficial (Ta) and 18 microinvasive invasive tumors of stage T1 and above. The 164 genes, identified as progressively altered in the model system, were analyzed for differential expression in stage Ta versus stage ≥T1 lesions. Of 121 genes that showed progressively elevated expression in the lung metastasis model, 51 were also elevated with the transition from superficial to invasive disease (>2-fold, P < 0.05) in human tumors. When the search was limited to genes with higher expression in tumor tissue compared with normal bladder tissue, only 4 genes remained. Interestingly, several of these genes, epiregulin, uPA, MMP-14, and TIMP-2, were included in the most significant scoring pathways noted above (Fig. 4) ⇓ . A similar analysis of the genes progressively down-regulated in the T24T to FL3 cell series, however, did not identify any that met the same selection criteria.
Epiregulin, a member of the epithelial growth factor family, has been shown to directly bind EGFR as well as the ErbB family of receptors (26) , and it may elicit a more potent mitogenic signal than epidermal growth factor (27) . This is particularly important because EGFR expression has been shown to be increased as a function of malignant potential in bladder cancer (28) . Increased expression of epiregulin has been associated with the pathogenesis of colon (29) , pancreatic (30) , and androgen-independent prostate cancer (31) . Modjtahedi et al. (32) showed that the anti-EGFR antibodies ICR15, ICR16, ICR61, ICR62, and ICR80, which inhibit epidermal growth factor ligation, likewise inhibit epiregulin binding to EGFR. These authors also found that 125I-epiregulin was bound poorly to EGFR in EJ bladder carcinoma cells, indicating perhaps that the ErbB4 (33) receptor is dominant for epiregulin in bladder cancer. Interestingly, ErbB2 expression has also been shown to be an important factor in bladder cancer progression (34 , 35) . Because structural data as well as the absence of a direct ErbB2 ligand imply that its heterodimerization with a second, ligand-binding ErbB receptor is necessary for ErbB2 activation (36) , such a ligand could be EGFR or ErbB4, resulting in an enhanced autocrine loop between epiregulin and ErbB, leading to lung metastasis. We are currently evaluating this hypothesis.
Urokinase and urokinase plasminogen activator both act as ligands for the uPA receptor and have been implicated in cancer metastases. Interrupting surface uPA receptor activity has been shown to decrease metastases in vivo (37) and led to the investigation of inhibitory antibodies (38) and peptides (39) against uPA receptor that have shown therapeutic promise. Similarly, elevated expression of MMPs is associated with increased metastatic potential in many cancer types. Inhibition of the mitogen-activated protein kinase/extracellular signal-regulated kinase kinase pathway with the mitogen-activated protein kinase/extracellular signal-regulated kinase kinase inhibitor, PD184352, has been shown to inhibit the invasiveness of human tumor cells and suppress expression of MMP-14 as well as other MMPs (40) , implying a relationship and possible treatment strategy. Elevated TIMP-2 expression has been associated with high stage and poor outcomes in bladder cancer. Grignon et al. (41) studied TIMP-2 expression by immunohistochemistry in human invasive bladder cancer cystectomy specimens and found 62% of tumors were positive for TIMP-2, which correlated significantly with outcome (69% versus 25% mortality because of cancer). TIMP-2 expression has also been associated with metastasis in colon (42) and breast (43) carcinomas, as well as related to poor prognosis in breast (44) , cervical (45) , and ovarian (46) carcinomas.
Our study has limitations. As with any model derived from human cancers, adaptation of the cells to the in vitro culture condition has undoubtedly altered their biological behavior. Nevertheless, the parental cell line T24, from which this model was derived, maintains its invasive ability in an orthotopic model (47) and recapitulates the original biological behavior in the patient from which it was isolated (12) . In addition, when we examined gene expression profiles of human prostate cell lines and clustered these with human prostate tumors (48) with unsupervised methods, the cell lines clustered together and away from the human tissues. Taken together, these observations help explain our data that only a few genes are associated with tumor progression when cell line and human tumor profiles are compared. Nevertheless, despite these limitations, cell lines selected for a progressively more lung metastatic phenotype can serve as a caricature of human tumor progression, and by combination of their gene expression profiles with those of human tumors, allow for the identification of several candidate genes associated with lung metastasis, an otherwise impossible task given the limitation of harvesting lung metastatic tissue in bladder cancer.
In conclusion, the model system we describe here opens up an experimental avenue to help reveal some of the molecular and cellular changes underlying the progression of bladder cancer from superficial lesions to the lung-metastatic form. In trying to address the drawbacks that have previously prevented a more comprehensive biological investigation of this phenotypic and clinical transition, we generated preliminary information that is in accordance with the biological processes suspected to be at play in general phenomena of metastasis. It is our hope that additional comparison of the model systems behavior to the snapshots provided by human clinical samples will reveal key target pathways that will soon permit better diagnosis and therapeutic intervention.
The authors thank Drs. Jay Fox and Yongde Bao of the University of Virginia Array Core facility for their assistance with chip hybridization, data analysis, and helpful suggestions and Dr. Gary Davis for helpful suggestions.
Grant support: NIH CA075115.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Note: J. M. Seraj is currently in Cancer Therapeutics Research, Drug Discovery Division, Johnson & Johnson Pharmaceutical Research & Development, Raritan, New Jersey.
Requests for reprints: Dan Theodorescu, Department of Molecular Physiology and Biological Physics, University of Virginia Health Sciences Center, Charlottesville, VA 22908. Phone: (434) 924-0042; Fax: (434) 982-3652; E-mail:
- Received March 6, 2004.
- Revision received July 1, 2004.
- Accepted August 24, 2004.
- ©2004 American Association for Cancer Research.