| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Published online before print January 15, 2009
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
From the Departments of Molecular Physiology and Biological Physics,* Pathology, Public Health Sciences, Division of Biostatistics and Epidemiology, Medicine,|| and the Paul Mellon Urologic Cancer Institute,| University of Virginia Health System, Charlottesville, Virginia; and the Genomics Institute of the Novartis Research Foundation, San Diego, California
 |
Abstract |
|---|
 Top Abstract Materials and MethodsResultsDiscussionReferences |
|---|
Bladder cancer is the fifth most commonly diagnosed cancer in the United States,8 and the molecular lesions characterizing bladder carcinogenesis and progression are beginning to be better elucidated.9 However, little is known about the pathways that regulate general metastatic propensity or organ site-specific tropism to lung or liver, which are two common sites of dissemination in this and other tumor types. Our prior work has focused on the discovery and investigation of genes that regulate lung metastasis, such as the metastasis suppressor RhoGDI2, by comparing metastatic T24T cells to their isogenic nonmetastatic relative, T24.10-12 We have extended this work by generating progressively more lung metastatic T24T-derivative cell lines to indentify additional lung metastasis promoting or suppressing genes.13 Here, we further extend this latter model and additionally develop a new T24T isogenic liver metastasis model to identify genes associated with the general metastatic phenotype as well as those with specific organ tropism. Data derived from these models was then evaluated in the context of gene expression data from human bladder tumors, to identify genes with a high likelihood of clinical relevance in human urothelial cancer.
|
Materials and Methods |
|---|
|
TopAbstractMaterials and MethodsResultsDiscussionReferences |
|---|
The T24T cell line and culture protocols have been described.10,11,14 Six-week-old nude mice were purchased from National Cancer Institute (Bethesda, MD) and were maintained under University of Virginia Animal Care and Use Committee guidelines. To generate cell lines with propensity for liver metastasis, mice were injected with 1 x 106 T24T cells in 100 µl of serum-free Dulbecco’s modified Eagle’s medium into the splenic capsule. Mice were sacrificed when weight loss was >20% when compared with that for control animals. At necropsy, visible liver metastases were cultured in vitro as described.13 The first liver metastatic derivative cell line was named SL1 (spleen to liver), and cells were maintained in culture for as short a time as possible before splenic re-injection, usually <1 week. This protocol was repeated three times to yield SL2, SL3, and SL4. Sequential tail vein injections generated increasingly lung metastatic T24T derivatives FL1, FL2, and FL3 (from lung) as reported.13 Here we extended this protocol to generate a fourth-generation cell line, FL4, to enable comparison with the equivalent liver metastasis-derived line, SL4.
To examine whether these cell lines retained any organ site-specific growth properties, 1 x 106 FL4 or SL4 cells were injected into the tail vein of 6-week-old nude mice. At 5 weeks, mice were necropsied, organs harvested, and visible metastases in the lungs were counted. In a parallel experiment, the same cells were injected into the splenic capsule, with subsequent necropsy and evaluation for hepatic metastasis. These experiments were repeated and pooled metastasis counts averaged, with differences in average metastatic counts evaluated by Student’s t-test, two tailed. To scale out any nonorgan-specific difference in metastatic ability between the cell lines, we then normalized the counts of metastases to either the liver or lung to the same cell line’s average number of metastases across both the liver and lung sites. Lung and liver tissues were formalin-fixed for subsequent immunohistochemical study as described below.
Comparative Genomic Hybridization (CGH) Analyses
CGH analysis with confirmation using a 99% confidence interval was performed as reported.15
Peripheral blood from a healthy donor was used as a negative control, and a gastric tumor with known copy number changes was used as a positive control, with 1.17- and 0.85-fold cutoffs for gains or losses and a 
1.5-fold cutoff for high-level amplification.
Oligonucleotide Microarray Analysis of Cell Lines and Human Tumors
We used Affymetrix (Santa Clara, CA) HG-U133A to analyze mRNA expression in bladder cancer cell lines and human tissues including the T24T and FL1 to FL3 derivatives.11,13,16 Here, we used HG-U133A arrays in duplicate for each cell line, to analyze SL1 to SL4 cell lines. We also re-analyzed all of FL1 to FL3 and the newly generated FL4 derivative on the same platform, in parallel to the SL series. Ontological analysis of genes discovered in these models was performed using Ingenuity (Redwood City, CA) Pathway Analysis 5.5. HG-U133A profiling of human bladder cancers (n = 65) and normal bladder mucosa (n = 15), included 30 stage Ta, 5 Tis, 6 T1, 5 T2, 9 T3, and 10 T4 tumors as reported.17
Immunohistochemistry
Immunohistochemistry on human tumors was performed with approval of the University of Virginia Institutional Review Board. A tissue microarray of quadruplicate cores of 199 bladder cancers from patients who underwent radical cystectomy was constructed using standard techniques.13
This tissue microarray included 3 stage Ta, 7 stage T1, 61 stage T2, 93 stage T3, and 33 stage T4 samples. Tissue microarray sections were stained for Laminin V gamma 2 (LAMC2) using the monoclonal antibody clone D4B5 (Millipore, Billerica, MA) and enzymatic treatment with Subtilisin A (Sigma, St. Louis, MO), as described.18
This antibody detects dimeric LAMC2, monomeric LAMC2, and the processed protein.19
Staining was scored as low if 
30% of tumor cells stained positively or high if >30% cells stained positively. A superficial bladder tumor was used as a positive control, whereas normal bladder mucosa was used as a negative control for antibody workup. Associations between LAMC2 staining (high or low) on the tissue microarray and tumor grade (G2 to G4), nodal status (N0 versus N1+), and presence or absence of lymphovascular invasion as reported on the specimen’s original pathology report were tested with 
2 tests of association, and disease recurrence with the log-rank test. For immunohistochemical staining of in vivo metastases from FL and SL cells, formalin-fixed tissues from the murine metastasis experiments were paraffin-embedded, and the same LAMC2 staining procedure as above was used excepting the use of a mouse-on-mouse treatment kit (Vector Laboratories, Burlingame, CA) per the manufacturer’s instructions.
|
Results |
|---|
|
TopAbstractMaterials and MethodsResultsDiscussionReferences |
|---|
After generation of increasingly metastatic T24T-derived liver (SL1 to SL4) and lung (FL1 to FL4) metastatic cell lines by in vivo selection (Figure 1A)
, we sought to determine whether such cells could retain their organ-specific growth properties. We tested their metastatic competence side by side in the liver or the lung to see if the cell line not selected for growth in a specific organ would have reduced numbers of metastases at that site. These experiments involved injecting 1 million FL4 or SL4 cells via tail vein or splenic capsule of 10 mice, with animals necropsied at 5 weeks and visible metastases counted.13
Comparing the lung metastatic abilities of FL4 and SL4, we found that FL4 is 10.1-fold more lung metastatic than SL4 (P = 0.004). At the liver site, we found that FL4 and SL4 showed no difference in numbers of metastases (P = 0.86), both consistent with our observation that in vivo lung selection has resulted in a FL4 line that is, on average, 2.7-fold more metastatic than SL4 (P = 0.02). To correct for the different general metastatic behaviors of these cell lines and begin to evaluate the fold of organ-specific ability of these cell lines, we normalized average metastases for a cell line at a particular organ site by the cell line’s average over both lung and liver sites. Although this calculation does not explain why serial lung metastatic selection of FL4 resulted in a derivative cell line that has gained 2.7-fold in metastatic ability at either site compared with SL4, it does allow scaled comparison of the organ-specific metastatic ability of these cell lines. We thus found that FL4 is 3.8-fold more lung-specific than SL4, whereas SL4 is 2.4-fold more liver-specific than FL4. Figure 1, B and C
, illustrates these comparison experiments, whereas Supplementary Figure 1, see http://ajp.amjpathol.org plots the metastasis experiment data.
|
To determine whether differential expression of genes was associated with the in vivo selection and derivation of these cells, we evaluated our series of lung-selected FL and liver-selected SL cell lines, using the Affymetrix HG-U133A microarray platform.11
Proceeding from the hypothesis that progressive changes in gene expression as a function of progressive rounds of in vivo passage are likely to be biologically relevant, we searched for genes increasingly up-regulated or down-regulated through the four generations of in vivo metastatic selection that produced the FL and SL series. Using the criteria that probes are expressed at the end of the in vivo selection process (microarray expression value >100 in either FL4 or SL4); are differentially expressed after the first round of in vivo metastatic selection compared with the parental cells (FL1 or SL1 > or <T24T); and are further changed more than twofold in the same direction throughout the selection process (FL4/FL1 >2 or <0.5 and or SL4/SL1 >2 or <0.5), we found a union of 93 probes for the two models. This set included a shared core metastatic differential expression profile between the liver- and lung-selected cells (10 probes: 8 up, 2 down) as well as organ site-specific probes (48 up, 12 down in the SL series; 18 twofold up and 5 twofold down in the FL series). Tables 1a
, Table 1b
, Table 2a
, Table 2b
, Table 3a
, and Table 3b
list the top 15 probes with the largest fold changes and P 0.05 (for complete list see Supplementary Table 1, see http://ajp.amjpathol.org).
|
|
|
|
|
|
provides a comparison of the top 10 ontological classes for the liver-specific probes with comparison with lung-specific probes, indicating that, in some cases, similar types of (although not the same) genes were present in both models.
|
To determine whether similarities or differences existed between genomic changes present in the SL and FL models compared with their parental T24T cell lines, we performed chromosomal CGH analysis. We reported previously that the lung metastasis-derived FL cell lines exhibited similar baseline changes compared with T24T, with progressive accumulation of additional gains and high-level amplifications, including gains at 1q, 3q, 7q, 8q, 9q, 12q, 17q, and high-level amplification of 20q by the third generation FL3 derivative.15
By comparison, the liver metastasis-derived SL cell lines showed fewer total changes to that for the T24T cell line than the FL series, lacking gains at 8q and 9q. However, in several instances the same genomic changes occurred in the SL series, including gains at 1q, 3q, 7q, 12q, 17q, and 20q amplification. Figure 1E
shows the gains shared between the models.
We previously showed that the 17q gain reported in the lung metastasis model was associated with the differential expression of eight genes within the region, constituting a gene expression hotspot.15
Here, we found that several of the core metastatic genes common to both the liver and lung models (see Table 3a
) were located in regions of chromosomal gains, including the laminin V gamma 2 subunit (LAMC2) and the KiSS-1 metastasis suppressor (KISS1) in the 1q region, as well as zinc finger, BED-type containing 2 (ZBED2) at the 3q locus.
Expression of Metastasis-Associated Probesets in Human Bladder Tumors
To focus on genes likely to be functionally relevant in human bladder cancer, we screened the candidate probesets discovered in this study against microarray expression data from 24 stage T2 to T4 muscle invasive bladder cancers.17
Using the criterion that probes discovered in either the liver or lung metastasis models must also be significantly differentially expressed in the same direction for invasive bladder cancers as compared with normal bladder mucosa (P < 0.05), we found that 16 of the 48 probes up-regulated in the SL model and 3 of the 12 probes down-regulated were also significantly up- or down-regulated in muscle invasive bladder cancer. In the FL model, 5 of 18 probes up-regulated and 2 of 5 probes down-regulated were also similarly expressed in human tissues. Finally, of the eight probes up-regulated in both the liver and lung models, three probes met the aforementioned criterion for overexpression in invasive bladder tumors: annexin A3 (ANXA3), vesicle-associated membrane protein 8 (VAMP8), and LAMC2. Figure 1D
shows the overlaps between the liver and lung model probes and human tumors.
Immunohistochemical Evaluation of LAMC2 Expression in Human Bladder Cancer
The convergence of data from CGH and gene expression analysis in both xenograft models and human tumors identified LAMC2 as a potential biomarker of metastasis in human bladder cancer. Hence, we used immunohistochemistry18
to compare LAMC2 protein expression with salient clinicopathological parameters in human bladder cancer, including metastatic recurrence. First, we examined FL4 and SL4 lung and liver metastases and found that in all cases, xenograft tumors strongly expressed LAMC2 protein (Figure 2, A and B)
. In our tissue microarray composed of 199 human bladder cancers (Figure 2, C and D)
, we found that LAMC2 staining was significantly associated with tumor grade (P = 0.0003). Interestingly, despite our discovery of LAMC2 as an overexpressed gene shared between liver and lung metastasis, we found that LAMC2 staining was inversely associated with nodal metastasis (P = 0.005). Given this finding, we evaluated the association of LAMC2 staining level from the tissue microarray with presence or absence of lymphovascular invasion on the case’s original pathological examination, finding that it was also inversely associated, approaching statistical significance (P = 0.08). Lymphovascular invasion is considered a harbinger of nodal metastasis,19
thus further supporting the inverse correlation of LAMC2 staining with nodal status. Finally, for 23 of the 199 tumors, tissue samples of nodal metastases were available, and LAMC2 staining of these revealed that 19 of 23 (83%) exhibited low levels of LAMC2 (Figure 2E)
. This was again consistent with the inverse relationship of LAMC2 expression with nodal metastasis in primary tumors.
|
, we found that high LAMC2 staining was significantly associated with visceral metastatic recurrence of bladder cancer in patients after cystectomy (P = 0.040) (Figure 2F)
. Supplementary Table 2, see http://ajp.amjpathol.org shows microarray data for LAMC2 staining stratified by original tumor grade (A) or nodal status (B) (assigned at pathological review of the tissue block from which the tissue microarray cores were obtained). |
Discussion |
|---|
|
TopAbstractMaterials and MethodsResultsDiscussionReferences |
|---|
2.7-fold on average), although there was a preponderance of selection for lung metastatic ability (10-fold as compared with SL4). One could simply attribute this phenomenon to the fact that genes specifically associated with lung model exhibited many of the same ontological classes as genes from the liver model (Table 4)
. However, this is a compelling finding, potentially even clinically important, given both the pervasiveness of tumor cell dissemination through systemic circulation from even small tumors21
and the fact that cells growing (and shedding into circulation) from a patient’s lung metastasis have already been extensively selected and enriched for metastatic competence.22
Future experiments should examine whether this phenomenon is of any generality to bladder or other cancers that metastasize to the lung; speculatively, such a process could beg the question of more aggressive management of patients with limited pulmonary metastasis to prevent dissemination elsewhere. On the other hand, liver metastasis selection of parental T24T cells though might have resulted in a less stringent selection than for the lung. The liver has been classically considered to be a receptive parenchyma,23 favorable to metastatic seeding from many tumor types. Speculatively, this might be associated with a the liver’s capacity to regenerate, and at least one recent report has actually experimentally demonstrated this phenomenon in a mouse model of colorectal cancer metastasis.24 In this study, the authors found that a priori partial hepatectomy actually stimulated formation of hepatic (and even extrahepatic) experimental metastases. Although obviously our experiments did not use hepatectomy as part of the protocol, it is important to consider this unique property of liver parenchyma, and its signaling milieu,25 as part of the process of liver metastasis.
We found substantial differences between genes up-regulated or down-regulated as a function of several rounds of in vivo metastatic selection, including liver or lung selection-specific and overlapping genes between the two models. The comparative ontological analysis suggested that there were cases in which similar broad ontological classes of genes (but not the same genes) were enriched among liver or lung model genes (cellular movement, proliferation, and cancer). However, the differences observed suggest that the host organ microenvironment26 participates in selection of clones expressing substantially different patterns of transcripts.
Always, an important caveat for this model (and all experimental, as opposed to spontaneous metastasis models) is that the model only recapitulates the last stage of metastasis, the arrest and colonization of the second parenchyma, missing potentially important (or targetable) genes regulating processes of invasion, intravasation and survival in lymphatic and venous draining of the primary tumor. The technique itself bears consideration; based on the experiments done, we would be unable to discriminate between genes irrelevant to liver metastasis but that support survival through injection, or more specifically, splenic injection and venous drainage (an anatomical route not traveled, at least directly, by disseminated bladder cancer cells). At a minimum, such cells are then subsequently and terminally selected for growth in the liver parenchyma, just as are their counterparts through the tail vein lung metastasis model.
Importantly, as regards clinical relevance of these targets, we identified a significant overlap between genes identified by the models and genes similarly dysregulated in invasive bladder cancer. It bears discussion that this is a very stringent secondary screen for candidate metastasis genes: individual genes contributing to metastasis or organ-specific metastasis do not perforce have to be differentially expressed on average among all cells in the tumor bulk (as is required to be detected in microarray studies of human tumor tissues). Indeed, recent reports suggest that even rare subpopulations of cancer cells may exhibit differential metastatic tropism because of their unique gene expression,6 although our screening strategy was designed to identify targets (ie, LAMC2) potentially amenable to use as immunohistochemical biomarker in gross tissue.
Surprisingly, one of the largest increases in gene expression in both liver and lung models was the KISS1 metastasis suppressor gene,27 whose expression was observed to be induced after multiple rounds of in vivo cell line selection. One potential mechanism for this phenomenon might be that in vivo selection of cells might have resulted in selection of clones with down-regulation of the KISS1 receptor (KISS1R/GPR54), whose expression is not monitored on the HG-U133A microarray platform, or another necessary downstream signaling component in this metastasis-suppressor pathway. A similar observation has been made for the RhoGDI2 metastasis suppressor gene, in which repeated pulmonary metastatic selection of breast cancer cells resulted in the re-expression of the RhoGDI2 transcript.7
At the genomic level, chromosomal CGH analysis showed many changes, including gains at 1q, 3q, 7q, 12q, and 17q, and amplification of 20q, that were shared between the liver and lung models. To our knowledge, there is no published report of CGH performed on metastatic bladder cancer tissue for comparison with our results. However, this is not the case in colorectal cancer, in which metastases are routinely excised, and two recent reports show interesting parallels between our cell line CGH gains in lung and liver metastasis and actual primary human colon cancer lung and liver metastases. Although gain at 1q is rather common in metastasis across several tumor types,28 gains at 3q, 7q, 12q, 17q, and 20q have all been observed in colorectal cancer lung metastases, although the 7q, 12q, and 20q lesions were particularly noted to arise during the transition between the primary tumor and lung met.29 Another report actually compared CGH changes in patient liver and lung metastases from colorectal cancer, finding, just as we did in our in vivo selected bladder cell lines, a phenomenon of generally less total CGH changes in liver compared with lung metastases, although they did not identify in their lung metastases the 8q and 9q gains that we found in our lung selected cell lines.30 For the 1q and 3q loci, these regions appeared to contain genes that were differentially expressed as a function of in vivo selection, including LAMC2 and two others of the eight core genes discovered to be overexpressed in both metastatic models. Given the similarity of CGH changes evinced between our model and these prior colorectal cancer reports and the lack of genes identified in all of the areas of CGH gain by our microarray platform, we believe that analysis of our metastasis models by high-density array CGH or SNP arrays might uncover other targets that are associated our causally related to metastasis, potentially relevant to both of these diseases.
The metastasis-associated genes that emerged from our models suggest some potential therapeutic targets. For example, monoclonal antibodies against TACSTD1 (EpCAM), discovered here as a lung metastasis-specific target gene (Table 2a)
and recently shown to be a marker for cancer stem cell populations,31
have been used in the clinical trial setting.32
Our findings, in concert with a recent report of associating TACSTD1 IHC with poor prognosis for human bladder cancer33
might suggest the preclinical evaluation of such a therapy for patients with lung metastatic bladder cancer. In the case of LAMC2, its EGF-like repeats have been shown to be processed from the protein by matrix metalloproteases and they may agonize the EGFR,19
a receptor tyrosine kinase now targeted via several therapeutic modalities. Moreover, our models can serve as platforms for validation of the potential preclinical efficacy of therapeutic modalities targeting these genes and their pathways in vivo, as we have recently shown for Atrasentan, a drug targeting endothelin-1, which is a downstream transcriptional target of the RhoGDI2 metastasis suppressor.34
A novel finding in our study was the observation that LAMC2 expression is a marker or even possibly a mediator for hematogenous dissemination of cancer cells as opposed to lymphatic metastasis, at least in bladder cancer. High LAMC2 expression was inversely correlated with lymph node metastasis despite being detected as a core metastatic gene whose elevated expression was shared between muscle invasive bladder cancer and the liver and lung metastatic cell. Yet, LAMC2 staining was associated with development of metastatic recurrence at visceral sites after surgery in patients without lymph node metastases. This observation highlights the importance of validation and interpretation of findings from animal models in the context of clinical outcomes observed in patients.3,6
In conclusion, we have developed an in vivo model system for the discovery and future study of genes implicated in lung and liver metastases in bladder cancer. We found genes whose expression differed based on the metastatic organ site from which they were derived, but also include a core metastatic signature shared between the sites. Several of these suggest therapeutic targets, and some, such as LAMC2, provide novel insights regarding the different routes of metastasis.
Future studies using these models should not only generalize the findings of this T24T-based system to other cell lines but also use dissociated cells from primary human bladder tumors. Because of the marked heterogeneity of tumors of even the same stage, grade, and histology and concomitant epigenetic and genetic instability, regulation of organ tropism is likely affected by large constellations of genes (as opposed to the few we have started to characterize herein), which larger numbers of cell lines or dissociated tumor cells could more appropriately represent. Such cells have also not been subjected to the vagaries of in vitro culture adaptation. Given new protocols to prospectively identify highly tumorigenic stem cell-like subpopulations of cells within tumor bulk35 and an unknown relationship between these stem cell-like populations and populations leading to metastatic colonization, such studies will hopefully provide further insights into this most morbid and mortal stage of disease.
|
Footnotes |
|---|
Supported by the National Institutes of Health (grant CA075115 to D.T., Medical Scientist Training Program training grant T32GM007267 to S.C.S., and Cancer Training grant CA009109 to S.C.S.).
Supplemental material for this article can be found on http://ajp.amjpathol.org.
Accepted for publication November 6, 2008.
|
References |
|---|
|
TopAbstractMaterials and MethodsResultsDiscussionReferences |
|---|
Related Article
This article has been cited by other articles:
 |
|
 |
|
  Z. Tian, T. Hwang, and R. Kuang A hypergraph-based learning algorithm for classifying gene expression and arrayCGH data with prior knowledge Bioinformatics, November 1, 2009; 25(21): 2831 - 2838. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
