Gene Expression Patterns Associated with the Metastatic Phenotype in Rodent and Human Tumors

INTRODUCTION

A generally accepted principle is that tumor cells need to acquire altered cellular properties to metastasize. The nature and number of such properties will determine the range of secondary sites to which the tumor cells can spread. During progression to metastatic cancer, the development of heterogeneity within the primary tumor, driven largely by genomic instability (1) , generates populations of tumor cells with the properties needed for invasion, migration, and metastatic colony formation in other organs (2) . Not every tumor cell develops all these properties because the ability to enter the circulatory system or even to form micrometastases is insufficient for progressive metastatic growth (reviewed in Refs. 2 and 3 ). The critical factors determining overt metastasis formation are whether disseminated tumor cells die, survive at distant sites but remain dormant, or have the properties required to grow progressively as metastases. Thus, tumor cells that have the necessary properties to metastasize are selected for during their metastatic journey.

The mechanisms leading to the dissemination of tumor cells, the routes metastasizing tumor cells take through the organism, and the selection pressures tumor cells have to overcome are similar for many different types of cancer (reviewed in Ref. 3 ). It is therefore reasonable to assume that tumor cells from different tumor types require common properties and that they express similar metastasis-promoting genes. Attempts have been made to identify such genes using differential screening techniques, each identifying one gene or a few such genes (4, 5, 6, 7, 8, 9) . However, many outstanding fundamental questions remain to be answered. How many genes are involved in the metastatic process? Are any of these genes absolutely required for metastasis formation? Conversely, can the same property required for metastasis be provided by several different genes or groups of genes? Due to the effects of genomic instability, to what extent are genes that do not play a role in the metastatic process up-regulated during progression to the metastatic phenotype? To begin to find answers to these important questions, it would be necessary to describe and compare the repertoire of genes specifically expressed in metastasizing cells but not in their nonmetastasizing counterparts.

New technologies allow global descriptions of complex transcriptional changes associated with different tumor properties (10, 11, 12) . We have used SSH 5 coupled with a sensitive high-throughput screening protocol (13) to define the profile of genes whose expression is up-regulated during progression from a locally growing tumor to metastatic competence. Using two rat tumor progression model systems, we have isolated and identified 268 different cDNAs representing individual genes whose expression is exclusive to or up-regulated in the metastatic phenotype (279 total clones, 11 of which were found in both screens). These target cDNAs were further screened for expression in other tumorigenic cell lines of both rat and human origin. In addition, human primary tumors and metastases of different tissue origin were monitored for expression of selected cDNAs by in situ hybridization studies. From our data, we conclude that subsets of genes are commonly up-regulated during metastatic progression and that several of the genes we have newly identified as being up-regulated in metastasizing cells may be useful diagnostically or as therapeutic targets in the treatment of cancer.

MATERIALS AND METHODS

Tumor Samples and Cell Lines.

The rat pancreatic cell lines BSp73-1AS, BSp73-10AS, and BSp73-ASML (14) ; the rat prostatic tumor cell lines G, AT-1, AT-3 AT-6, Mat-Lu, and Mat-LyLu (15) ; and the rat mammary carcinoma cell lines MTPa, MTC, MTLY, MTLN-2, and MTLN-3 (16) were cultured as described previously. NM081 and MT450, nonmetastasizing and metastasizing rat mammary carcinoma cell lines, respectively (17) , were both maintained in DMEM supplemented with 10% FCS and antibiotics. Two human cell lines, MDA-MB-231 and MDA-MB-468, were obtained from American Type Culture Collection (American Type Culture Collection numbers: MDA-MB-231, HTB-26; MDA-MB-468, HTB-132). All human tumor samples and corresponding normal tissue used for the in situ hybridization were obtained from the Institute of Pathology, University of Graz (Graz, Austria).

Subtractive Library Construction and Screening.

The construction and screening method of the subtracted mammary carcinoma library used in this study was essentially identical to that used in Ref. 13 , using the Clontech SSH kit (PCR-Select cDNA Subtraction Kit; Clontech Laboratories, Palo Alto, CA). Briefly, the subtracted cDNA library was amplified and TA cloned. Clones were picked, and their inserts were amplified by colony PCR. Equal amounts of the PCR reactions were electrophoresed on duplicate high-density gels. Gels were blotted and individually probed with driver and tester cDNA to identify differentially expressed clones.

PCR Amplification.

All clones were amplified by colony PCR using 96-well microtiter plates as follows. Five μl of each bacterial culture were added to 95μ l of sterile water and denatured for 5 min at 94°C. Amplification of 5 μl of this lysate was performed in a total volume of 100 μl using standard PCR buffer (Eurobio), 250 μm deoxynucleotide triphosphates, 10 pm each of primer PN1 and PN2, and 1 unit of Taq (Eurobio). The primers used were the inner primers of the Clontech SSH-Kit: (a) PN1, 5′-TCGAGCGGCCGCCCGGGCAGGT-3′; and (b) PN2, 5′-AGGGCGTGGTGCGGAGGGCGGT-3′.

Cycle conditions were as follows: 30 cycles of 94°C for 20 s, 68°C for 12 s, and 72°C for 30 s. Products were purified from agarose gels using DNAeasy (Biozyme) and used for Northern analysis.

Northern Analysis.

Poly(A)⁺ RNA was isolated from cell lines by standard methods, and 2-μg aliquots were size-fractionated on a 1.0% formaldehyde-agarose gel. Ethidium bromide staining was used to ensure equivalent loading in each lane. Gels were blotted onto Hybond N⁺ membrane filters (Amersham). The filters were then cross-linked (UV Stratalinker 2400; Stratagene) and hybridized at 65°C in Church buffer [1 mm EDTA, 0.5 m NaHPO₄ (pH 7.2), and 7% SDS]. Probes were generated by PCR amplification of the target clones, followed by gel purification and ³²P-labeling of the cDNA fragment (ReadyPrime; Amersham). Unincorporated label was removed before hybridization using an Elutip (Schleicher & Schüll) according to the manufacturer’s specifications. After hybridization with the labeled probes, membranes were washed twice in 2× SSC and 0.1% SDS and twice in 1× SSC and 0.1% SDS at 64°C, after which they were exposed to film for between 2 h and 8 days. Filters were stripped and subsequently probed with another gene to be analyzed. Alternatively, stripped filters were finally hybridized with a glyceraldehyde-3-phosphate dehydrogenase probe to ensure the presence of equivalent amounts of poly(A)⁺ RNA in each lane of a filter and to ensure that the hybridization patterns of different filters could be compared.

Computational Analysis.

All computational analyses concerning the identity of the clones obtained through the subtracted libraries were performed using the blastn sequence similarity search. 6 The clones were compared with the sequences contained in the human and rodent public domain databases, including all of the nonredundant GenBank, European Molecular Biology Laboratory, DDBJ, and PDB sequences, and the DDBJ EST divisions. Before comparative blast screens, we developed a program that automatically recognizes and removes the flanking nested primers to compare only the inserts and thus avoids false blastn results. In the first round of comparison, all of the sequences were compared with each other to identify fragments isolated several times (18) . In the next round, the sequences were compared with human and rodent sequences in the public domain databases (19) .

In Situ Hybridization.

In situ hybridization was performed to detect S7 and CD24 expression in human tumor material. The inserts to be tested were cloned into the bidirectional TA vector pGEM-Teasy (Promega) and transcribed in vitro using SP6 or T7 RNA polymerase (Life Technologies, Inc.). Sections (4 μm thick) of paraffin-embedded human tumor samples were mounted on silanized glass slides, deparaffinized, and postfixed in 4% paraformaldehyde in PBS for 20 min at room temperature. After rinsing with TBS [50 mm Tris-HCl (pH 7.5) and 150 mm NaCl], sections were treated with 0.2 m HCl for 10 min, and, after washing in TBS, incubated in 20 μg/ml proteinase K (Sigma, St. Louis, MO) in TBS containing 2 mm CaCl₂ for 15 min at 37°C. The reaction was stopped with TBS for 5 min at 4°C. After treatment with 0.5% acetic anhydride in 100 mm Tris (pH 8.0) for 10 min, sections were rinsed with TBS, dehydrated in graded ethanol, and air dried. Hybridization with fluorescence-labeled (RNA color kit; Amersham Life Science) sense and antisense RNA probes was performed according to the manufacturer’s instructions. After H&E counterstaining, the sections were analyzed by light microscopy. The in situ hybridizations were performed with 10 mammary carcinomas (invasive ductal carcinomas including lymph node metastases from four of the tumors), 5 samples with benign fibrocystic disease, and 10 colorectal carcinomas (including the corresponding nonneoplastic mucosa from the resection margins and lymph node metastases from four tumors). The carcinomas were staged according to UICC guidelines (20) .

RESULTS

Choice of Experimental System.

We set out to define sets of genes up-regulated or exclusively expressed in metastatic human tumors. Instead of starting directly from human tumor material, we chose to analyze defined clonal rodent tumor cell lines (14 , 16) . The disadvantage of this approach is that contributions to metastasis by stromal cells and transcripts induced by the interaction of tumor cells with their environment are lost. This is outweighed by numerous advantages. Tumor cell lines often exhibit a reproducible metastatic or nonmetastatic phenotype that can be retested at any stage of the analysis (e.g., Ref. 17 ). Moreover, tumor cell lines are accessible to genetic manipulation and functional tests in experimental animals. Rat tumor cells have the advantage of being able to be passaged in syngeneic animals, whereas human tumor cells have to be passaged in the rather artificial setting of an immunodeficient host. Furthermore, the expected high degrees of homology between rodent and human sequences should permit subsequent isolation of human homologues of candidate tumor progression genes and evaluation of their expression in primary human tumor material.

For the molecular comparisons, we used two rat carcinoma models. The rat pancreatic adenocarcinoma model is comprised of several clones that differ in their metastatic potential in vivo and have been derived from a common primary tumor (14) . For example, BSp73-1AS cells form primary tumors but do not metastasize, whereas BSp73-ASML cells are highly metastatic and, after s.c. injection, disseminate via the lymphatics to finally colonize the lungs. The rat mammary adenocarcinoma cell system 13762NF (16) is composed of a number of cell lines derived from a parental mammary tumor and its corresponding spontaneous lung and lymph node metastases. For example, the cell line MTPa has been reported to be nonmetastatic in vivo in syngeneic animals, whereas MTLY is highly metastatic, giving rise to multiple metastases in the lymph nodes and lungs (16) .

Before subtractions, the phenotypes of the cell lines were verified by s.c. injection of each cell line into syngeneic animals (Table 1) ⇓ . In all cases, 100% of animals developed tumors after injection of 5 × 10 5 cells. MTPa and BSp73-1AS did not metastasize at all; the MTLN cell lines and BSp73-ASML grew in both the lymph nodes and lungs, whereas MTLY and BSp73-10AS appeared to prefer the lymphoid organs. Despite extensive investigation of many organs, metastases were only observed in the lymph nodes or lungs.

Subtractive Cloning of Differentially Expressed Genes.

From the cell lines examined and documented in Table 1 ⇓ , we decided to compare the gene expression profiles of pairs of tumor cells with the most extreme metastatic behavior. Transcripts of the nonmetastatic mammary tumor cell line MTPa were subtracted from those of the metastatic cell line MTLY (this library was denoted ML). The nonmetastatic pancreatic cell line BSp73-1AS was subtracted from its metastatic counterpart BSp73-ASML (this library was denoted PL). Both subtractive libraries were generated and screened as described previously (10 , 13) . A total of 1985 and 5000 clones were screened from the mammary and pancreatic tumor libraries, respectively, resulting in the isolation of 487 and 494 differentially expressed cDNA clones, which, through sequence analysis, equated to 160 and 119 individual genes, respectively 7 (Table 2 ⇓ A). Eleven of these genes were found in both the mammary and pancreatic tumor library screens. Thus, a total of 268 different cDNAs were isolated in the two screens. A complete list of all clones obtained will be accessible via an internet web page. 8

As indicated in Table 2 ⇓ , the clones of both carcinoma-specific libraries were grouped according to their degree of sequence homology to the described public domain human and rodent genes (18 , 19) . Sixty-five clones identified from the ML subtraction and 43 clones identified from the PL subtraction represent described rat genes (see Table 2B ⇓ ); an additional 25 ML and 16 PL clones share significant homologies with known genes from other species, and some of them show homology to rat genes and could therefore encode new rat family members. All sequences with a homology of less than 70% to previously described genes can be separated into those that demonstrate significant homology to already existing ESTs and those that show little or no significant match to any sequence in the public databases. Clones with less than 70% homology to previously described genes have been classified as novel. On this basis, 70 of 160 (ML) and 60 of 119 (PL) clones represent previously unidentified sequences. 7

By probing Northern blots with randomly selected clones, we convincingly verified that the SSH method does indeed identify differentially expressed genes (Fig. 1) ⇓ . The normalization step of SSH enriches for rare transcripts and thereby allows the identification of both high and low abundance transcripts. As indicated in Fig. 1 ⇓ , clones C and B isolated from the PL and ML, respectively, represent genes expressed at low levels. Conversely, clones F (PL) and L (ML) represent transcripts of high abundance.

Cancer cells from different tumor types likely require common properties during metastasis, and therefore recurrent patterns of metastasis-related gene expression should be found. Indeed, within the list of metastasis-related cDNAs for which sequences are already in the databases, we find gene products that have been described previously by others as associated with tumor progression (21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51) . 8 Furthermore, several genes were identified as being metastasis associated in both the PL and ML subtractions (Fig. 2) ⇓ . Moreover, several genes isolated in SAGE screens to identify human colonic and pancreatic tumor-specific genes were also found in the ML SSH and PL SSH libraries (Fig. 2) ⇓ .

Expression of cDNA Sequences in Several Metastatic Tumor Cell Lines.

To determine to what extent metastatic tumor cells of different origin share a basic set of gene products expressed in the metastatic phenotype, we analyzed the expression of randomly selected clones in several tumor progression models (Fig. 3) ⇓ . For this purpose, two additional tumor systems were used, namely, a pair of mammary tumor cell lines and a group of related prostatic cancer cell lines (15 , 17) . The metastatic behavior of these tumor cell lines was verified by in vivo passaging (Table 1) ⇓ . Additionally, two human MDA-MB cell lines reported to be metastatic in immunocompromised mice (52 , 53) were included in the analysis.

By direct Northern analysis, we found that several of the selected clones were differentially expressed in metastasizing cell lines from tumor models other than the model from which the clones were originally isolated. cDNA clones whose expression correlates with metastatic potential in more than one tumor model are the most likely candidates for general relevance in the metastatic process. Genes shown to be differentially expressed in all four systems tested were CD24, S7, B23, oncomodulin, and HB2. Genes clearly differentially expressed but not correlating with metastatic potency in all tumor models tested were choline kinase, megakaryocyte potentiating factor, metastasis-associated factor, c-Met, mirf 6, cytokeratin 8, uPAR, HMG-1, and the novel clones 23, 25, 56, 62, 63, 82, 154, 162, and 163. The 26 clones used for this analysis represent only 10% of all of the clones identified and were randomly selected. The remaining 242 as yet untested genes are expected to show a similar degree of correlation between expression and metastatic potential.

Selected Examples of the Metastatic Gene Expression Pattern in Human Primary and Metastatic Carcinomas.

The ultimate test for the validity of our differential isolation approach is the exclusive or strongly enhanced expression of candidate clones in human cancer, which we tested by in situ hybridization analysis. We concentrated our study on two cDNA clones, CD24 and S7, which showed the best cross-hybridization to human sequences (Fig. 3) ⇓ and are highly expressed in the metastatic rat cell lines tested.

In situ hybridizations were performed on tumor sections from primary tumors as well as from corresponding metastases and nonneoplastic tissue from colorectal and breast carcinomas. In the colorectal carcinomas, CD24 mRNA expression was weakly detectable in normal colonic mucosa but highly abundant in tumor cells and, to a lesser extent, in the cells of the tumor stroma (Fig. 4) ⇓ . No marked difference in the expression levels was found between the primary tumors and the corresponding lymph node metastases. S7 was very weakly expressed in normal colorectal mucosa and was highly abundant in colorectal tumor cells. There were no marked differences in the expression levels of S7 between the primary tumors and the corresponding lymph node metastases. There was, however, a striking difference in the expression between S7 and CD24 in that CD24 was not only overexpressed in the invasive carcinoma but was also overexpressed in the adjacent noninvasive areas with mild atypia (compare Fig. 4, j–l ⇓ ). Furthermore, the expression of CD24 in the tumor stroma was more pronounced than the expression of S7.

In the breast carcinomas, CD24 expression levels were essentially comparable in the nonneoplastic mammary parenchyma and in the carcinoma cells of the primary tumors as well as the corresponding metastases (Fig. 5) ⇓ . In general, the hybridization signals obtained for S7 were much weaker than those obtained for CD24. In comparison with nonneoplastic mammary gland parenchyma and primary carcinoma, a moderate overexpression of S7 was seen in lymph node metastases (Fig. 5) ⇓ .

Thus, genes whose expression we identified as being metastasis associated in several rodent tumor systems also show a tumor stage-dependent expression behavior in the human cancers investigated, particularly those of colorectal origin, proving the validity of our approach.

DISCUSSION

It is clear from many studies that tumor cells that can metastasize have gained properties not possessed by their nonmetastasizing counterparts. These additional properties must be reflected in changed gene expression patterns. However, it is completely unknown how many and which genes are differentially expressed between metastasizing and nonmetastasizing cells from a given tumor. Furthermore, there is no indication as to what extent these differentially expressed genes are absolutely required for metastasis formation by other tumor types or to what extent there is redundancy in the gene expression patterns that lead to metastatic competence. Here we present the first reported attempt to describe and compare the complete sets of genes expressed in metastasizing but not in nonmetastasizing tumor cells.

From our experiments, we conclude that metastasizing tumor cells from different tumor types up-regulate many common genes. Firstly, many of the known genes isolated in the PL and ML screens have previously been associated with tumor progression. Secondly, the Northern blots in Fig. 3 ⇓ show that many of the differentially expressed clones from the PL and ML screens are also expressed in a metastasis-specific fashion in other model systems. Thirdly, the two genes selected for in situ hybridization analysis are strongly up-regulated in human tumor material (Figs. 4 ⇓ and 5) ⇓ . Fourthly, several genes were identified as being metastasis associated by both the PL and ML subtractions.

We also conclude that metastasizing tumor cells up-regulate large pools of genes. Genes identified as being metastasis associated in both the PL and ML subtractions were comparatively few in number (a total of 11). This could be interpreted to mean that mammary and pancreatic cancer cells use widely divergent subsets of genes to metastasize or that there is extensive redundancy in the requirement of genes whose expression is associated with metastasis formation, either because the same property can be provided by many different genes or groups of genes or because many genes are up-regulated but not necessary for metastasis formation. However, other data suggest that the yield of clones differentially expressed in our subtractions was not exhaustive. Randomly selected clones identified as being expressed in a metastasis-specific manner in the ML but not PL subtraction were used to screen Northern blots of the pancreatic tumor cells (Fig. 3) ⇓ . Here, 12 of 16 clones proved to be up-regulated in the metastatic BSp73-ASML but not in the BSp73-1AS nonmetastatic pancreatic carcinoma cells. Similarly, seven of seven clones from the PL screen that were not isolated in the ML screen were nevertheless more highly expressed in the metastasizing MTLY mammary carcinoma cells than in the nonmetastasizing MTPa cells. Thus, although we isolated many genes as being expressed in a metastasis-specific manner in the PL (119 genes) and ML (160 genes) screens, we anticipate that many more genes are also differentially expressed in metastatic cells of the rat mammary and pancreatic carcinoma models. This suggests that metastasizing tumor cells up-regulate a large pool of genes compared with their nonmetastasizing counterparts.

On the basis of our current data, we cannot draw any firm conclusions about the degree of redundancy in gene expression required for metastatic competence. There clearly is redundancy in the genes that are up-regulated during metastatic progression because several of the clones from the PL and ML screens were not significantly up-regulated in a metastasis-specific manner in all tumor models or were even down-regulated (e.g., uPAR, novel genes 23 and 82; see Fig. 3 ⇓ ). This is to be expected: for every qualitative genetic change that is advantageous for metastatic progression, there will be many others that are either deleterious for the tumor cell or neutral in terms of conferring metastatic properties on the tumor cell. Furthermore, some genes are likely to be up-regulated in a cell type-specific manner.

Several laboratories have made attempts to screen for cancer-associated transcripts using different methods [differential display, representational differentational analysis, SAGE, and SSH (10 , 54, 55, 56, 57)] . Of those techniques capable of efficient expression profiling, SAGE has the advantage that even moderate changes in transcription can be detected, but it suffers from the fact that the sequence information it produces is only 13-bp long. Not only is this problematic for nucleotide sequence comparison algorithms, but it is also not ideal for use in techniques such as in situ hybridization and Northern blotting. On the other hand, subtractive cloning techniques such as SSH yield cDNA fragments of 200–800 bp, which overcomes the problems associated with SAGE. Furthermore, SSH enriches differentially expressed genes with high efficiency irrespective of their relative abundance. The combination of SSH with high-throughput screening methods, for example, using microarrays (58) , provides a powerful method for the rapid identification of differentially expressed genes. SSH suffers from the drawback that it does not readily detect transcriptional differences less than 5-fold. SAGE and SSH have never been compared directly using the same cell system. However, the data presented in this study verify that SSH is a valuable tool for identifying differentially expressed genes in tumors and other cellular contexts.

Because the data presented here represent the first extensive description of genes expressed specifically in metastasizing cells, comparison with previous differential screens is informative but needs to be interpreted with caution. The only other extensive descriptions of genes up-regulated in tumors used differential hybridization to arrayed cDNAs, SAGE, or SSH. These studies either compared normal tissue with tumor material or ER-positive cells with ER-negative cells, and no investigation of metastasis-related gene expression has been made (11 , 58, 59, 60) When the differentially expressed sequences in these analyses are compared with the sequences identified in the ML and PL screens, few of the sequences are identical. This is illustrated by a comparison of human pancreatic tumor-specific genes identified by SAGE with metastasis-specific genes expressed in rat pancreatic tumor cells identified by SSH (Fig. 2) ⇓ . At first sight, this would suggest that many of the genes identified by SAGE as being up-regulated in the tumor material do not contribute to the metastatic phenotype but rather are concerned with tumorigenesis. However, it should be borne in mind that the different screens do not constitute an exhaustive list of up-regulated genes. Furthermore, these comparisons are preliminary because ESTs are not always made accessible in the databases, and identities may be overlooked if the ESTs are short.

With respect to molecular functions of the genes we report to be expressed in a metastasis-specific fashion, several of the known sequences isolated in the PL and ML screens have previously been reported to be involved in the process of metastasis. For example, c-Met is the receptor for hepatocyte growth factor/scatter factor. A large body of evidence implicates this receptor-ligand pair in the processes of tumorigenesis, invasion, migration, and metastasis (reviewed in Ref. 61 ). Interestingly, another gene identified as being up-regulated in metastasizing cells is ezrin, a cytoplasmic substrate for c-Met required for hepatocyte growth factor-mediated morphogenesis whose expression is associated with the metastatic phenotype (62 , 63) . We also isolated uPAR in our screens, which binds urokinase-type plasminogen activator and is believed to be essential for activation of numerous proteases involved in invasion (64) . As a further example, elevated caveolin protein levels have been associated with human prostate carcinoma progression (65) , and we report here that caveolin is a gene up-regulated in metastatic cells.

The data presented here would suggest that some of the clones we have identified as being expressed in a tumorigenesis-related fashion could also prove to serve as new markers for further elucidation of the molecular events in the dysplasia-carcinoma sequence. Others may serve as targets for therapeutic intervention. An impressive correlation between expression and metastatic potential was seen with CD24, S7, and several novel clones. CD24 is expressed on B-cell precursors, neutrophils, neurons, and keratinocytes (66 , 67) , and its expression has not previously been associated directly with metastasis. However, it is expressed on leukemic cells of the B-cell lineage and on various solid tumors such as small cell lung carcinoma, neuroblastoma, rhabdomyosarcoma, and renal cell carcinoma (68, 69, 70, 71) . CD24 was also identified by a SSH screen to identify genes differentially expressed between ER-positive and -negative cells lines (58) . Significantly, reduction or absence of CD24 expression on acute lymphoblastic lymphoma cells predicts a better prognosis for patients (72) . As a surface protein able to bind to P-selectin, CD24 could promote metastasis formation by mediating binding of circulating tumor cells to endothelium (32) .

Although several ribosomal components have been reported to be up-regulated in cancer, S7 has not previously been associated with tumor growth and metastasis. It is a component of the translation machinery, and its RNA binding properties are required for the correct folding of the 16S rRNA (73) . Our observations call for a novel function for S7. Enhanced S7 expression in tumor cells is not simply a correlate of enhanced proliferation rate because only very weak expression is observed in the highly proliferative colonic mucosa. In colorectal carcinomas, we observed that S7 is homogeneously expressed and up-regulated specifically in invasive carcinomas and metastases. Moreover, S7 expression is also clearly enhanced in metastases of the mammary carcinomas we investigated. Our data suggest that S7 may be useful in certain instances as a marker for tumor staging and indicate that additional experiments are warranted to investigate whether S7 plays a functional role in invasive tumor growth.

In conclusion, this study is the first reported attempt to describe the profile of genes expressed specifically in metastasizing tumor cells using a direct comparative method and proves that SSH is a useful tool in identifying differentially expressed genes. We demonstrate that a large number of genes are up-regulated during progression to metastatic competence, and we have confirmed the association of the expression of certain genes with the metastatic phenotype. More significantly, we have connected the expression of already characterized genes that have not previously been linked with tumor metastasis with the metastatic phenotype. Most importantly, we have isolated totally novel metastasis-associated genes. These genes represent a rich source of candidates for application in the diagnosis and therapy of human cancer.

Note Added in Proof:

While this manuscript was under review, Clark et al.(Nature 406: 532–535, 2000) reported the isolation of genes using microarray screening whose expression correlates with metastatic potential.