I. INTRODUCTION

Top Previous Next References

Cancer is malignant because cancer cells invade into neighboring tissues and survive in this ectopic site. The term invasion indicates penetration into neighboring territories and their occupation. Cancer cells invade beyond the constraints of the normal tissue from which they originate; this invasion permits them to enter into the circulation from where they can reach distant organs and eventually form secondary tumors, called metastases. Invasion and metastasis are not unique for cancer as they also occur during embryonic development, in healthy adult organisms, and in many noncancerous diseases. Ectodermal cells invade through the primitive streak and occupy the subectodermal space where they form the mesoderm (118, 248, 252, 399, 410). Neural crest cells emerge from the dorsal aspect of the neural tube, migrate to different sites in the body, and survive there to grow and differentiate under the influence of specific local factors (230, 400). Leukocytes leave their tissue of origin in the bone marrow, enter into the circulation, and home at specific sites (376). Microorganisms enter their host by invasion through the lining epithelia of the skin or the gastrointestinal or respiratory tracts; they eventually reach the circulation and produce secondary lesions (346). We have discussed previously similarities in the molecular mechanisms of invasion by various organisms taking as examples four families of molecules, namely, cadherins, integrins, hydrolases, and chemokines (225). More recently, evidence was published in favor of a role for bacteria like Helicobacter pylori in the progression of gastric cells toward a more invasive phenotype (170).

Noninvasive tumors are benign, because they are cured easily by simple removal. Invasive tumors, called cancer, invariably kill their host if untreated and, even with optimal treatment, such tumors are a frequent cause of death (105). In the case of primary brain tumors, such as astrocytomas, death is due almost uniquely to local invasion, since, for yet unknown reasons, these tumors rarely form metastases. Cancers of the head and neck, originating in the mucosa of the upper respiratory and alimentary tracts, kill mainly through local invasion and metastasis to locoregional lymph nodes. Death by colorectal cancer is due to locoregional spread in one half and to distant metastasis in the other half of patients. In the case of breast cancers and melanomas, death is usually the consequence of distant metastasis. It is quite obvious from the course of these diseases that invasion and metastasis are the hallmarks of cancer malignancy. Consequently, invasion and metastasis are major prognostic markers. The 5-year survival rate for bladder cancer that is limited to the epithelium is 60-80%, compared with 30-60% when invaded into the deeper muscle and 10-40% when invaded through the bladder wall into the fat, provided radical treatment is performed. Melanoma 10-year survival rates are ~80% when the tumor invades into the dermis but is <1.5 mm thick and ~40% when it invades into the subcutaneous fat. For primary brain tumors, the prognosis depends on the loss of differentiation rather than on the degree of invasion. Patients with well-differentiated astrocytomas, grade I and II, survive for 10 years or more, whereas poorly differentiated astrocytomas, grades III and IV, have a median survival of 1 year.

In infectious diseases, though they are treated much more succesfully than malignant tumors, spread through invasion and metastasis may also herald a bad prognosis. In listeriosis, a disease caused by the bacterium Listeria monocytogenes, spread from the site of entry in the intestine, to the meninges, or to the fetus in pregnant women causes a potentially fatal disease (354). Resident noninvasive Streptococcus viridans is harmless; invasion through wounded oral mucosa and metastasis to damaged heart valves is a cause of death in 21% of the patients (211). Similarly, the cystic noninvasive form of Entamoeba histolytica is harmless; invasion of E. histolytica trophozoites into the enteric mucosa causes amoebic enteritis with eventual spread to liver and brain (321).

Earlier experimental investigation of invasion and metastasis focused on the development of appropriate models to score invasion and invasion-related activities by morphological techniques (1, 102, 109, 117, 164, 355, 416, 448). Today, the study of invasion benefits from the enormous advances in genomics and proteomics, providing thousands of genes and proteins, all well characterized structurally and functionally. We review here the more recent data about cellular and molecular mechanisms of invasion, taking selected examples with emphasis on homotypic and heterotypic cross-talk between the invaders and the host. Noncancer invasion by normal cells or by prokaryotic and eukaryotic cells will be discussed for comparison with cancer invasion. For a review of the older literature, the reader is referred to Reference 249. A selected literature search for the present review was closed in November 2001.

	II. CANCER PATHOGENESIS

Top Previous Next References

Cancer disturbs the cellular activities that are crucial for the development and the maintenance of multicellular organisms, namely, growth, differentiation, programmed cell death, and tissue integrity. Clinically, cancer manifests itself through a tumor because of excessive growth, through pain and bleeding because of invasion into nerves and vessels, and through functional disturbances because of pressure on and replacement of normal tissues. These symptoms are not cancer specific, and the diagnosis is made by histological examination of a sample from the tumor. This diagnosis includes the origin and type of cancer, its extent of growth and invasion, and its grade of differentiation. Attention is paid also to the host cell reaction evidenced by the stroma, blood vessels, and leukocytes. Because cancers are known to metastasize, the physician will search for secondary tumors in the lymph nodes and in distant organs. Growth, at least to the minimum volume detectable by the actual diagnostic techniques, is a prerequisite to find secondary tumors. Qualitative and quantitative criteria are used to stage and grade cancers for therapeutic and prognostic purposes. Staging of tumors is done following the volume of the primary tumor and its depth of invasion (T stage), the number and the volume of occupied lymph nodes as well as invasion through their capsula (N stage), and the presence of distant metastases (M stage). This TNM system, propagated by the International Union Against Cancer, is widely used in Europe (372). For example, a T4, N1biii, M1 breast cancer has invaded into the skin, occupied axillary lymph nodes with invasion through their capsula, and has metastasized to distant organs such as bone, liver, brain, or lungs. A T2, N0, M0 breast cancer has a diameter not exceeding 5 cm and no metastasis are detected. Kaplan-Meyer survival curves show that patients with such T2 cancers, provided accurate treatment is given, have 70% chances of being alive 5 years after diagnosis. Attempts are made to refine staging by the identification at diagnosis of these 30% of cancers that are not controlled by this treatment. The actual efforts include the search for micrometastases that are not detected in the routine TNM system. Immunohistochemistry, with antibodies against epithelial marker proteins, of lymph nodes and bone marrow in breast cancers that were scored N0, M0 by the standard criteria showed immunopositive cells in the lymph nodes in 6.4%, in the bone marrow in 26.0%, and in both in 4.8% of the patients (144). It is tempting to speculate that the presence of such micrometastases might predict the above-mentioned 30% of lethal cases. One decade of careful follow up is, however, needed to know the answer to the crucial question whether or not such nests of cancer cells will survive, grow, reach clinically relevant volumes, spread to other organs, and eventually kill their host. Grading is based on the loss of differentiation, sometimes combined with mitotic activity. Individual cancers are currently portrayed by DNA, RNA, and protein microarray systems, covering as many characteristics of maligancy as possible. Whether or not this method will enter into clinical routine for staging and grading is an open question (194).

Noninvasive precursor lesions, i.e., histological abnormalities in which cancer is more likely to occur than in the normal tissue counterpart, are found in the vicinity of invasive cancers (71, 135). There is compelling evidence to accept that many types of cancers have benign precursor lesions, recognized by accumulation of cells, such as in hyperplasia and adenoma, or by loss of differentiation and nuclear abnormalities, such as in atypia and carcinoma in situ. There has been a long debate about whether a common initiated progenitor cell population would give rise to both noninvasive and invasive lesions (field theory) or a noninvasive precursor lesion would transit toward invasive lesions (progression theory). In favor of the progression theory is the concept that somatic mutations favoring continuous proliferation or low apoptosis led to clonal expansion and to continuous selection of progressively more malignant cell populations (334). Moreover, at least part of the genetic abnormalities of invasive cancers are also found in apparently normal and in preinvasive lesions, as exemplified in the breast (101), the prostate (100), the esophagus (199), and the bronchus (57). The scenario inferred from these clinical observations is confirmed in models of experimental carcinogenesis in the rat colon and in the mouse skin (384, 454).

The above-mentioned clinical and experimental observations indicate that cancer is a disease of growth, causing accumulation of cells, of differentiation, causing loss of structure and function, and of tissue organization, leading to invasion and survival in an ectopic environment (Fig. 1). A multistep process of invasion leads to metastasis: invasion from the tissue, in which the cancer has originated, into the surrounding tissues through barriers such as the epithelial basement membrane; entry into blood or lymph vessels; transport through the circulation; arrest and exit from the circulation at the putative site of metastasis; and invasion into the tissues of the occupied organ. Although cancer is generally thought to evolve from bad to worse, large variations in the rate of progression have been published. In a number of studies collected from the literature about patients with cervical lesions that remained untreated against advice but accepted follow up, progression from carcinoma in situ toward invasive carcinoma varied between 3 and 70% (71). For earlier stages of cancer development, the probability of progression is lower than for more advanced stages of the disease, as suggested in Barrett's esophagus (162). Such differences in progression between earlier and later stages of cancer development suggest a point of no return where precursor lesions transit into lesions that in most cases progress toward invasive cancer. This is certainly the case for most cancers that have reached the stage of invasion and metastasis. There are, nevertheless, observations indicating that reversion to a more normal stage of at least part of the cancer cell population is possible. Regression of metastasis from hypernephroma upon removal of the primary cancer without adjuvant therapy has been described but remains a rare event with an unknown biological mechanism (105). Metastases sometimes show a higher degree of differentiation and grow with less or no invasion compared with the primary cancer. This suggests that the progression model should not assume that invasion and metastasis-associated phenotypes are fixed by genetic alteration. Colorectal carcinoma metastases may resemble the organized epithelial and tubular structure of a well-differentiated primary cancer, whereas the invasive front of the actual primary cancer displays loss of the epithelioid morphotype and appearance of fibroblastoid, presumably invasive and metastatic, cancer cells. An experimental demonstration is provided by cocultures of human colon cancer cells with enteric lymphoid cells, in which the cancer cells transit to M cells, a differentiated type of absorptive enterocytes covering Peyer's patches (208).

View larger version (26K): [in this window] [in a new window]   Fig. 1. Schematic representation of genetic, epigenetic, and phenotypic aspects of cancer development. [Adapted from Mareel and Bracke (247).]

There are indications that abnormal growth and invasion are not necessarily associated during cancer development. Metastasis without primary tumor, called CUP for cancer with unknown primary, is not a rare event (5-10% of all cancer patients with metastases). In such cases, the primary tumor cannot be found at the time of clinically evident, hence growing, metastases. Primary tumors may appear later or not at all. Conversely, metastases may grow to reach clinically relevant volumes many years after removal of the primary cancer as exemplified by ocular melanoma. Both clinical observations indicate that invasion and growth at the primary or secondary site can be regulated independently, a conclusion that is confirmed by the experimental finding that pharmacological agents can arrest growth whilst permitting invasion, and vice versa (385). This growth-separate-from-invasion concept deals probably with the exception; in the majority of invasive cancers, a complex and probably coordinated program of invasion, growth, survival, and loss of differentiation is at the basis of the clinical manifestations of the disease. It is, indeed, logical to accept that proliferation is needed to provide a cohort of invaders and that inhibition of apoptosis keeps them alive in an ectopic matrix environment. Considering noncancer invasion, leukocytes do proliferate in the bone marrow and, thereafter, invade and metastasize as nondividing cells. Parasitic trypanosomes and leishmania pass through an obligatory nondividing stage when they invade from one host into another or from one tissue into another (279).

	III. INVASION PROMOTER AND SUPPRESSOR GENES

Top Previous Next References

A series of alterations in the genome of the cell population of origin forms the basis of tumor development (40, 130). The genes of interest are classified as oncogenes or tumor-promoter genes, one allele of which is activated leading to gain-of-function events, and tumor-suppressor genes or antioncogenes, both alleles of which are inactivated leading to loss-of-function events. Genomic instability, due either to impairement in DNA repair (microsatellite instability) or to dominant negative mutations in mitotic check-point genes (chromosomal instability), leads to activation of oncogenes and inactivation of tumor suppressor genes. The products of these genes belong to various classes of protein families, such as cytokines, cell surface receptors, signal transducers, and transcription factors. The list of oncogenes encoding cell surface receptors of the protein-tyrosine kinase family alone counts more than 40 members (42). Mechanisms of activation of oncogenes implicate mutation, gene amplification, and promoter activation. Mechanisms of tumor-suppressor inactivation are exemplified by loss of heterozygosity (LOH) plus silencing of the second allele genetically, through mutation, or epigenetically, through methylation. In familial cancers, one mutation is carried with the germline. Well-documented examples include RB in retinoblastoma, BRCA1 in breast cancer, and adenomatous polyposis coli (APC) in colon cancer of the familial adenomatous polyposis (FAP) type. Cancer-related genetic alteration are multiple and occur in varying orders so that it is difficult to ascribe defined genetic alterations to distinct stages of tumor development (18, 77, 456). The sequence of genomic alterations during tumor development might be of particular interest for the understanding of their role in the acquisition of invasion. Are the genes implicated in invasion different from those implicated in growth disturbance, loss of differentiation, and of sensitivity to death signals? The recognition of some stage specificity of genetic alterations has led us previously to believe that oncogenes and tumor-suppressor genes were implicated in growth disturbance and that they differed from genes promoting or suppressing differentiation, invasion, or survival (250). The growing list of cancer genes, however, comprises several examples of oncogenes and tumor-suppressor genes that are, either in the same or in different types of tumors, implicated in earlier stages of growth disturbance as well as in the later stages when invasiveness is acquired (27). Furthermore, the categorization of tumor phenotypes in growth, differentiation, survival, and invasion underestimates the mutual relationships between these phenotypes (125). To illustrate, growth is the basis for clonal expansion of somatic cells, differentiation and proliferation are inversely related, and survival signals are implicated in invasion because normal cells die inside foreign trophic microecosystems (358). The known function of the gene product, sometimes, makes the stage of appearance of the genetic alteration unexpected. Loss of p53, the guardian of the genome, is frequently found at the transition between the noninvasive, premalignant and the invasive, potentially malignant stage, and this is later than expected from the more general role of the p53 phosphoprotein functioning in the check-point control that arrests cells with damaged DNA. In line with this observation, p53 null mice are less susceptible to induction of papillomas, but once the papillomas arise, they transit rapidly to carcinomas. The effect of gene activation may depend on the stage of development at which it occurs. For example, transforming growth factor- (TGF-) acts as a tumor suppressor in early stages of tumor development, whereas it causes invasion and metastasis upon inducible transgenic expression in papillomas (439). The following discussion about invasion-suppressor and invasion-promoter genes and their alterations during tumor development chooses examples on the basis of the interest of the authors' laboratory. The more tumor-suppressor or -promoter genes are examined, the better it is realized that many of them affect invasion as well as growth and differentiation.

Epithelial (E)-cadherin is a transmembrane glycoprotein of the type I cadherin superfamily (299); its cytoplasmic part is linked to the actin cytoskeleton via the catenins, -catenin, -catenin, and plakoglobin (-catenin). The gene encoding E-cadherin (CDH1, on chromosome 16q22.1) was one of the first to be considered as an invasion-suppressor gene (30, 138, 432). The experimental strategy consisted of the isolation from heterogeneous cell lines of clones with an epithelioid (e-type, resembling epithelial cells) morphotype and a fibroblastic (f-type, resembling fibroblasts) morphotype. The e-type cells were E-cadherin positive, failed to invade into organotypically cultured embryonic chick heart, and formed a differentiated epithelial layer around the heart tissue. The f-type cells were E-cadherin negative, did invade, and showed no epithelial differentiation. Similarly, a positive correlation was found between the invasion into collagen type I of human cancer cell lines and the lack of E-cadherin. The invasive phenotype as well as the morphotype of these cells could be manipulated in both directions, from e-type noninvasive to f-type invasive, and vice versa, by transfection with sense or antisense E-cadherin cDNA. The e- to f-type conversion is reminiscent of the epithelial to mesenchymal transition (coined EMT) observed during gastrulation. Interestingly, loss of E-cadherin in immortalized cell lines of noncancerous origin did induce the invasive phenotype, only when the cells were transfected with an oncogene (Fig. 2). Conclusions from these experimental findings were confirmed by immunohistochemical changes in E-cadherin expression and localization in most human cancers (56, 92, 98, 275, 276, 305, 353, 366, 419, 421). The positive correlation between cancer aggressiveness as evidenced by poor survival and disturbance of E-cadherin provides clinical support for E-cadherin as an invasion suppressor (304). Such clinical evidence is, however, at best circumstantial since deficient expression of E-cadherin may also be due to posttranscriptional and posttranslational events. The causal relationship between E-cadherin expression and invasion in vivo was convincingly demonstrated in transgenic mice (319). Such mice, expressing the tumorigenic simian virus 40 (SV40) T antigen under the insulin promoter (Rip1Tag), developed pancreatic -cell adenomas in 74% and invasive adenocarcinomas in 26% of the mice. Overexpression of E-cadherin under the same promoter (Rip1E-cad) did not cause tumors. Rip1Tag × Rip1E-cad crosses had a lower (8%) ratio of adenocarcinomas, showing that overexpression of E-cadherin counteracted the acquisition of the invasive phenotype. In contrast, crosses of Rip1Tag mice with mice expressing dominant negative E-cadherin that lacks the extracellular domain (Rip1dnE-cad) had invasive and metastatic carcinomas in 50% of the cases.

View larger version (16K): [in this window] [in a new window]   Fig. 2. E-cadherin promoter with positive and negative transcriptional regulators. For detailed description, see section IIIA. [Adapted from Van Aken et al. (419).]

Mutations in CDH1 are the exception rather than the rule as they occur only in diffuse type gastric cancer, lobular breast cancer, and endometrial cancer (25-27, 35, 36, 196). In invasive lobular breast cancers, a subtype in which cancer cells invade as Indian files, total loss of E-cadherin expression is due to E-cadherin gene mutations combined with loss of the wild-type allele. The above-mentioned observations are compatible with inactivation of CDH1 either at the transition between the noninvasive to the invasive stage or earlier. In lobular breast cancer early inactivation of CDH1 was demonstrated, putting forward CDH1 also as a tumor-suppressor gene (435). The same truncating mutations associated with loss of heterozygosity were found in sporadic lobular carcinoma in situ as in the associated invasive components but not in atypical hyperplasia. In diffuse type gastric cancers, the amplification product of E-cadherin cDNA was shorter than the expected 630 bp due to skipping of exon 9 or 8. In favor of the tumor-suppressor function of the E-cadherin gene is the finding of inactivating germ line mutations in families with a higher incidence of diffuse gastric cancer. Interestingly, the mother of one of the gastric cancer patients suffered from metachronous lobular breast cancer and diffuse gastric cancer and had the same CDH1 germline mutation as her child (206).

The low frequency of E-cadherin mutations is in striking contrast to the almost ubiquitous disturbance of E-cadherin in invasive and even in preinvasive cancers, suggesting other mechanisms of transcriptional or posttranscriptional downregulation. Such forms of downregulation may be reversible. In an experimental tumor model with an immortalized normal kidney-derived epithelial cell line Madin-Darby canine kidney (MDCK) transformed by a mutated RAS oncogene and coined MDCK-ras, E-cadherin-positive variants had an e-morphotype and were noninvasive in vitro, but produced invasive and metastatic cancers after injection into nude mice (246). Immunohistochemistry of the nude mouse tumors revealed loss of E-cadherin, but ex vivo culture of the MDCK-ras tumors resulted in rapid reexpression of E-cadherin, acquisition of an e-morphotype, and loss of invasiveness. The host mouse context responsible for the changes in E-cadherin expression has, not yet, been identified. Reversible downmodulation of E-cadherin is suggested also by its reexpression in metastases from breast cancers (63). Similar observations with colorectal cancers led to the conclusion that, to grow at the metastatic site, disseminated f-type cancer cells must regain at least some of their epithelial functions (53).

Methylation of DNA is a common type of transcriptional modification in mammals. It normally occurs during genomic imprinting and X chromosome inactivation. Highly methylated DNA is found in genetically silent regions of chromosomes, and hypermethylation of CpG islands in the promoter region of a gene leads to transcriptional silencing. This mechanism of downregulation was observed in several tumor-suppressor genes such as APC, von Hippel-Lindau (VHL), RB, and also CDH1 (124). Germline mutations in CDH1 without loss of heterozygosity at the CDH1 locus are suggestive for hypermethylation. In hereditary diffuse gastric cancer, with a mutation in one CDH1 allele, hypermethylation constitutes the second hit eliminating the expression of E-cadherin (155). In this type of cancer and also in esophageal cancer, hypermethylation may occur as early as the intramucosal, i.e., noninvasive stage, of the disease, underscoring the tumor-suppressor function of E-cadherin (116, 391). In a gastric cancer cell line, the expression of E-cadherin could be restored by treatment with the demethylating agent 5-azacytidine. Some authors have cautioned interpreting hypermethylation in terms of tumor development, as they consider that the causal relationship between both phenomena is not firmly established (129).

The promoter of CDH1 contains positive regulatory elements, a CCAAT-box and GC-boxes, as well as two E-boxes (29, 150). Proteins acting directly or indirectly on the E-cadherin promoter are presented in Figure 2. Note that some of these proteins are encoded by genes that were classified as tumor-suppressor genes or as protooncogenes. In MDCK cells, coined MDCK(LT), the SV40 large-T antigen, encoded by a viral oncogene, inactivates RB and causes a transition from the e- to the f-morphotype that is associated with loss of epithelial markers, including E-cadherin, and with loss of expression of the oncogene MYC (258). The latter encodes two distinct Myc proteins, acting as transcription factor (22). Transfections of dog kidney MDCK and human skin HaCat cells show that RB and Myc specifically activate transcription of the E-cadherin promoter, a phenomenon that is mediated by the transcription factor AP-2 (23). Transactivation of the E-cadherin promoter is strongly dependent on the expression ratio between the two Myc proteins and is cell type specific. The finding that inactivation of RB by human papilloma virus HPV16E7 in primary human mammary epithelial tissue explanted in reconstituted extracellular matrix does not interfere with the correct expression of E-cadherin confirms this cell type specificity (374). In such cultures, viral inactivation of RB causes loss of the differentiation markers lactoferrin and cytokeratin-19. These experiments illustrate the complexity of oncogenic pathways: a viral tumor-promoter binds to and inactivates a cellular tumor-suppressor that indirectly counteracts invasion in one type of cells and maintains differentiation in another type. The Wilm's tumor 1 (WT1) may transactivate CDH1 directly through binding to the proximal GC-rich sequence in the promoter as evidenced by transfection of 3T3 fibroblasts (174). Snail, a member of a multi-zinc finger protein family of transcription factors, is a strong repressor of CDH1, interacting specifically with E-boxes in the E-cadherin promoter and, so, repressing transcription of CDH1; in this way, it causes an e- to f-morphotype transition and invasion (21, 68). Snail is expressed in fibroblasts, in some E-cadherin-deficient cell lines, and in invasive regions of experimental carcinomas. Smad interacting protein 1 (SIP1) belongs to the same zinc finger protein family as Snail and displays specific DNA binding activity (81). It interacts with several members of the Smad protein family. Conditional expression of SIP1 in MDCK-Tetoff cells, an MDCK-derivative stably expressing the Tetoff transactivator, abrogates the expression of E-cadherin and of cell-cell adhesion as well as unidirectional migration that are both sensitive to inhibition by E-cadherin-neutralizing antibodies; it simultaneously induces invasion into collagen gels. Members of the Smad protein family, which normally act in the TGF- signaling pathway cooperatively with other transcription factors (104, 450), are implicated in invasion in, yet, another way. SMAD2 genes with a mutation of Ser at position 465 are found in colon cancer and in lung cancer. When such genes, encoding an unphosphorylable form of Smad2, are transfected into MDCK cells or into human colon cancer HCT-8 cells, coined HCT-8/E11 for clonal selection of an epithelioid morphotype, they induce the invasive phenotype, and invasion is enhanced by addition of TGF- to the culture medium through an, as yet, unknown mechanism (328). In cells from the normal murine mammary gland (NMuMG) family (273) and in pancreatic cancer cells carrying an activating RAS mutation (121), TGF- caused an e- to f-morphotype transition and invasion, with downregulation of E-cadherin and of other junctional proteins.

Gain of N-cadherin in cancer cells accompanies loss of E-cadherin, acquisition of an f-morphotype, increased motility, and invasion both in vitro and in vivo as summarized in Reference 419. In some of these cancer cells, E- and N-cadherin are coexpressed. In such cells, the invasion-promoter potency of N-cadherin seems to dominate the invasion-suppressor potency of E-cadherin. Indeed, N-cadherin promotes invasion and motility of human breast cancer cells in a way that is not overcome by forced expression of E-cadherin (290, 291). A weakly metastatic E-cadherin expressing breast cancer cell-line of the MCF-7 family, yielded, upon successful transfection with N-cadherin, cells that coexpressed E- and N-cadherin and that were highly metastatic (167). Conversely, E-cadherin transfection in N-cadherin expressing breast cancer cells did not revert their invasive phenotype. The shifts from E-cadherin to N-cadherin raise the question whether the expression of both genes is coregulated. Transfection experiments yielded conflicting results. Decreased N-cadherin expression upon transfection with cDNA encoding L-CAM, the chicken homolog of E-cadherin, was ascribed to instability of the N-cadherin protein and not to reduced transcription (234). In squamous carcinoma cells, transfection of N-cadherin cDNA caused a decrease in E-cadherin expression; conversely, when N-cadherin expression was decreased by antisense transfection, E-cadherin expression increased (182). That repressors of E-cadherin may transactivate the gene encoding N-cadherin is demonstrated for the zinc finger protein Snail (21, 68). The retinal pigment epithelium (RPE) of the human eye may provide an interesting experimental model for the further analysis of the E- to N-cadherin shift (64; E. Van Aken, personal communication). In the eye, the RPE expresses mainly E-cadherin; shortly after explantation in vitro, RPE cultures show a majority of E-cadherin-negative, N-cadherin-positive cells with an f-morphotype. When such RPE cells are seeded on collagen gel, they extensively invade, and invasion can be blocked by addition of the N-cadherin neutralizing antibody GC-4. Clinical data do not substantiate unanimously the invasion promoter role of N-cadherin. Immunohistochemical analysis of 75 bladder cancers led to the conclusion that focal expression of N-cadherin in urothelial cancers is a frequent phenomenon, but its significance for invasion is unclear (335).

-Catenins are considered to be essential elements of the E-cadherin invasion suppressor complex (173, 415). The CTNNA1 gene encoding E-catenin is localized on chromosome 5q3.1 (300). In cultures of the human colon cancer clone HCT-8/E11, round cells (r-morphotype) can be observed either on top of the epithelioid (e-type) cell layer or floating in the culture medium (429, 431). Examination of the E-cadherin/catenin complex in harvested or cloned r-type cells showed that loss of E-catenin caused the transition from an e- to r-morphotype in an irreversible manner. Repeated screening of a series of colon cancer cell lines stored in our and in other laboratories as well as purchased from commercial stock, showed a similar morphotypic instability with spontaneous emergence of similar r-type variants in routine culture as observed with HCT-8/E11. DNA fingerprinting of cell lines coined HRT-18, DLD-1, or HCT-15 all had the same microsatellite instability DNA profile as HCT-8, strongly indicating that they all originated from the same patient (430). Such confusion about the identity of cell lines kept in various deposits was recently estimated to be ~36% (262). In the case of the HCT-8 family, the confusion was useful because it pointed toward a genetic background for the e- to r-type transition as it occurred in all the cell lines that had the same DNA profile. Cells of the HCT-8 family carry a heterozygous mutation in a CTNNA1 gene and mutation or loss of the remaining wild-type allele causes loss of -catenin and the above-mentioned e- to r-morphotype transition. CTNNA1 is, therefore, considered as a tumor-suppressor gene in accordance with Knudson's criteria (213). The conclusion about the cellular activity suppressed by E-catenin depended on the assay used to test the cells. In the chick heart invasion assay (Fig. 3), E-catenin-positive e-type HCT-8/E11 cells are not or poorly invasive with a few undifferentiated cells inside the peripheral rim of the heart tissue, in contrast to the E-catenin-negative r-type cells that massively occupy the heart tissue without any sign of differentiation. These experiments led to the conclusion that CTNNA1 is an invasion-suppressor gene (429). However, differentiation constituted another striking difference between the two cell types. When assayed on top of collagen type I gels, both types of cells fail to invade unless cancer-associated myofibroblasts are admixed to the collagen (106). In the presence of myofibroblasts, both e- and r-type cells do invade, the former as epithelioid strands and the latter as loose files of undifferentiated cells. It is, therefore, justified to consider CTNNA1 as a differentiation gene, influencing invasion only in a quantitative way and, possibly, as a consequence of changes in differentiation. When injected orthotopically into the wall of the cecum of nude mice, genotypic differences in CTNNA1 between e- and r-type cells were conserved, but phenotypic differences could not be seen any more; both variants produced moderately differentiated invasive adenocarcinomas that were undistinguishable from one another (426). Reexpression of -catenin in an E-cadherin-positive prostate cancer cell line PC-3 suppressed tumorigenicity in nude mice (127), suggesting a role for -catenin in ectopic survival. Recent data suggest that E-catenin not only functions through maintainance of cell-cell adhesion but also through interference with -catenin/Tcf/DNA complex formation and -catenin signaling in the nucleus (147). The above-mentioned data again illustrate the implication of a single gene in multiple tumor progression-associated phenotypes.

View larger version (36K): [in this window] [in a new window]   Fig. 3. In vitro assays for invasion of cancer cells and of bacteria. For further details about these and other assays for cancer cell invasion, see Ref. 61; for the bacterial invasion assay, see Ref. 349.

-Catenin, like the other catenins, was described first as an essential element of the E-cadherin/catenin complex (311). In normal cells, -catenin is associated not only with cadherins but also with the APC multiprotein complex (157). Here, it is phosphorylated by the serine/threonine kinase glycogen synthase kinase (GSK)-3 and directed to the ubiquitine proteasome pathway for degradation (see Fig. 8). The APC complex belongs to the Wnt signaling pathway, in the context of which -catenin may act as a tumor promoter (32, 341, 387, 446). Mutations in the serine/threonine phosphorylation sites of -catenin make it resistant to degradation. After saturation of the E-cadherin complex, the superfluous -catenin stays in the nucleus in association with lymphocyte enhancer factor (LEF)/T-cell factor (TCF) influencing transcription. Because TCF proteins possess no intrinsic ability to modulate transciption, coactivators such as the acetyltransferases p300 (169), Smad3 (220), and Pontin52 (24) are crucial. In contrast, Reptin52 (repressing pontin52) acts as a repressor. An overview of -catenin mutations in human tumors is listed in Reference 298. This list contains desmoid tumors, locally invasive outgrowths of mesenchymal cells that do not produce metastases (398). -Catenin mutations are limited to certain types of cancer; they are not found in squamous cell carcinomas of head and neck and esophagus, gastric carcinomas, or lobular and ductal breast carcinomas (67, 99, 153). In a series of 58 colorectal cancers without APC mutation, there were no CTNNB1 point mutations, but 7 tumors showed deletions of 234-760 bp, each of which included all or part of exon 3 (184). Retention of -catenin in the nucleus may result also from mutations in the -catenin binding site of APC, which is necessary for the formation of the complex in which GSK-3-mediated phosphorylation occurs (283). Mutation in the armadillo repeats is one of the oncogenic changes in APC, a tumor suppressor that is implicated in sporadic and familial colon cancer (326), as evidenced also in transgenic mice (133). The participation of APC at tumor development is probably not limited to growth, as the APC proteome also regulates morphogenesis (19). Target genes of the -catenin/TCF transactivator complex comprise the genes encoding matrix metalloproteinase-7 (MMP-7; matrilysin) (90), myc (168), cyclin D1 (236), multidrug resistance protein 1 (MDR1) (453), two components of the AP-1 transcription complex jun and fra-1 (245), and the putative transcriptional regulator AF17 (237). In melanoma cells, the acquisition of the TGF--dependent fibroblastic morphotype is accompanied by localization of -CTN in the nucleus and an increased expression of MMP-9, next to an increase of integrin-linked kinase (ILK), ₁- and ₃-integrins, and a decrease of E-cadherin (186). We would like to emphasize, here, that an element of an invasion-suppressor complex transactivates genes that are implicated not only in the modulation of growth, differentiation and response to therapy, but also in the stimulation of invasion and invasion-associated cellular activities. In cancers where mutations in -catenin, APC or conductin (equal to axin) (28) are uncommon, -catenin may be upregulated by the prolyl isomerase (Pin1) resulting in the transcription of several -catenin target genes (344). Pin1 interferes with -CTN/APC interaction by changing the conformation of the phosphorylated Ser/Thr-Pro bonds. There is less evidence in favor of CTNNB1 as an invasion-suppressor gene. An in-frame deletion in the E-catenin binding site of -catenin in a signet ring cell carcinoma cell line (HSC-39) causes disruption of the cadherin-dependent cell-cell adhesion (196, 201, 310). In normal mesenchymal cells and in uveal melanoma, cytoplasmic nonphosphorylated -catenin might well participate at invasion without translocating into the nucleus (210).

Plakoglobin, also called -catenin, is a close homolog of -catenin, sharing the 12 armadillo repeats; both may promote tumor formation when overexpressed (17). In adherens junctions, these two molecules bind independently to E-cadherin, and plakoglobin binds also to the desmosomal cadherins desmoglein and desmocollin. Unlike for -catenin, neoplastic transformation by plakoglobin does not implicate stabilizing mutations (215). It does, however, require transcriptional activation of the oncogene MYC via plakoglobin/TCF complexes. One possible difference between the two armadillo proteins may be the type of binding coactivator proteins as suggested by differences in their carboxy-terminal parts. In experimental systems, plakoglobin may also act as a tumor suppressor and as a tumor promoter (458). Overexpression of plakoglobin in cells from a human tongue squamous cell carcinoma SCC9 causes uncontrolled growth and inhibition of apoptosis. Here, plakoglobin exerts a growth regulatory function by induction of the antiapoptotic protein BCL-2, independently of its role in mediating cell-cell adhesion (160). In the above-mentioned observations of metastatic breast cancer by Bukholm et al. (63), most members of the E-cadherin/catenin complex, including -catenin, were reexpressed in the metastases, whereas plakoglobin was lost. Mutation of APC, as frequently observed in colon cancer, results in elevated levels of both -catenin and plakoglobin (216, 283). Taken together, plakoglobin and -catenin are best considered as separate players in tumor development with close links to each other as well as to other elements of the cadherin and the APC complexes.

Phosphorylation and dephosphorylation are key phenomena in intracellular signaling, and genes encoding kinases and phosphatases are on the list of oncogenes and tumor-suppressor genes. We discuss their putative roles in invasion, taking the examples of the nonreceptor tyrosine kinase SRC, the receptor tyrosine kinases c-MET and FGFR, the small GTPase RAS and the dual phosphatase PTEN. Doubtless, many more genes encoding other members of these protein families may act as invasion promoters or invasion suppressors (see list in Ref. 42).

SRC is the first oncogene detected (381) and a prototype showing many characteristics of the other oncogenes (259). The viral oncogene v-SRC has a cellular counterpart c-SRC that is activated by mutation to become an oncogene. Such activating mutations are found in human cancer both in early and late stages of development (72, 180). In some colorectal cancers, SRC activation, through truncating mutations in a critical carboxy-terminal tyrosine, probably has a role in malignant progression (180), supporting its invasion-promoter function. The Src protein is anchored to the plasma membrane through myristoylation, receives various signals, signals in its turn to many substrates directly and indirectly (see Fig. 11), and is implicated in numerous cellular functions, including proliferation, motility, as well as cell-cell and cell-substrate adhesion (402). Temperature-sensitive mutants of SRC were used to transform MDCK cells, the invasion of which into embryonic chick heart or into collagen gels could be switched on by changing the incubation temperature from 39.5 to 35°C (31). In these MDCKts.src cells, activation of Src leads to loss of E-cadherin functions, as evidenced by deficient cellular aggregation and gain of invasion. One of the molecular changes of interest in the E-cadherin/catenin complex is tyrosine phosphorylation of -catenin, weakening its binding with E-cadherin. In PC/AA/C1 cells, derived from a colon polyp in a FAP patient, introduction of activated SRC was not sufficient to induce invasion, but made the cells sensitive to stimulation of invasion by other factors (123). Indeed, the parental cells (PC/AA/C1) failed to invade into collagen in vitro, even after stimulation with scatter factor (SF)/hepatocyte growth factor (HGF), whereas their SRC-transfected derivatives (PCmsrc) did invade upon addition of SF/HGF. The membrane-associated polyoma middle T oncoprotein (py-MT), known to increase the tyrosine kinase activity of pp60c-src by preventing phosphorylation of Y527, mimicked some of the transforming effects of SRC but not its proinvasive activity (294). In rat bladder cells NBT-II, Src activity correlates with loss of epithelial differentiation and metastasis (51). During gastrulation movements in the Xenopus embryo, Src kinases are pivotal as evidenced by in vivo injection of mRNAs coding for dominant negative forms of ubiquitous members of the Src family.

The activation of receptor tyrosine kinases, such as c-Met and FGFR, contributes to invasion. The c-Met receptor for SF/HGF consists of a 50-kDa extracellular -subunit that is disulfide-linked to a 145-kDa -subunit having cytoplasmic tyrosine kinase domains and sites of tyrosine phosphorylation (48, 286). The effects of SF/HGF on motility, invasion, and proliferation are due to activation of the c-Met receptor as shown by transfection of hybrid cDNA encoding the ligand binding domain of nerve growth factor and the transmembrane and tyrosine kinase domains of c-Met (442). Mutation and overexpression of c-Met are associated with tumor progression in various human cancers, including kidney (356), thyroid, pancreas, colorectal (110, 111) and gastric cancers (231). In hereditary papillary renal cell carcinoma, missense mutations in the MET protooncogene lead to constitutive activation of the c-Met protein (356). Activated multifunctional docking sites recognize SH₂-containing adaptors like Grb2 and Shc, attract effector proteins such as phosphoinositide (PI) 3-kinase, Src, and phospholipase C (PLC)- and so stimulate diverse cellular functions (327). Phosphorylation of Y1349 and Y1356 is essential for scattering; point mutation in Y1349 abolishes the metastatic potential whilst enhancing the transforming activity. In NIH 3T3 cells, transfected with a mutated c-MET, tyrosine phosphorylation corresponds with transforming potential, focus formation, and tumorigenicity (189). Mutant c-Met induces motility in MDCK cells and metastatic potential in NIH 3T3 cells; transgenic mice develop metastatic mammary cancer (188). Interestingly, the expression pattern of the c-Met, and of other protooncogenic receptor tyrosine kinases, with their respective ligands during embryonic development suggests that they are involved in normal epithelial morphogenesis as well (39).

The FGFR family consists of four genes: FGFR-1 (flg gene), FGFR-2 (bek gene), FGFR-3, and FGFR-4. In the rat Dunning prostate cancer model and in human prostate cancer , progression is associated with alternative spicing in FGFR-2 gene: early stages express the III-b isoform dominantly binding keratinocyte growth factor (KGF) and later stages the III-c isoform with high affinity for basic fibroblast growth factor (bFGF) (142). FGFR1 and FGFR 2c/bek transfection into NBT-II cells leads to epithelial-mesenchymal transition in response to acidic fibroblast growth factor (aFGF) and bFGF (352). Similarly, transfection of FGFR-1 into less malignant prostate cells accelerated the progression toward a more maligant phenotype (131, 263). It is interesting to note the existence of an FGF-2 nuclear isoform conveying metastatic properties upon NBT-II rat bladder carcinoma cells without secretion and without the conventional FGFR-mediated signaling pathway (306).

The RAS protooncogenes encode membrane-associated guanine nucleotide binding proteins of 21 kDa. Constitutive activation by substitution of amino acid residues at various positions is frequently found in human invasive colorectal carcinomas and in other types of human cancer (45). Interestingly, activated Ras cooperates with TGF- to regulate invasion via the Raf-MAPK pathway (154, 185, 303, 399). Invasion into chick heart as well as into collagen gels was conveyed by Py-MT or by mutated (Val-12) Ha-ras upon SLC-44 rat intestinal epithelial cells immortalized by polyoma large T but not upon Caco-2 cells, derived from a human colonic adenocarcinoma cell line (75). Interestingly, the SLC-44 cells and its derivatives had weak or no expression of E-cadherin, whereas all Caco-2 cells were clearly positive at the cell-cell borders. In these experiments the effect of constitutive expression of the oncogenes was not limited to invasion, since it also increased growth as measured after subcutaneous transplantation into nude mice. Moreover, Ha-RAS transfected Caco-2 cells failed to perform enterocytic differentiation. Examples of the competition between the invasion-suppressor effect of E-cadherin and the invasion-promoter effect of activated Ras are shown in Figure 4. Clearly, in these and in other model systems, the invasion-suppressor potency of E-cadherin did neutralize the invasion-promoter potency of the mutated Ha-Ras (432).

View larger version (30K): [in this window] [in a new window]   Fig. 4. Genetic manipulation of apparently spontaneously immortalized cell lines from dog renal (MDCK) and mouse mammary (NMuMG) origin resulting in E-cadherin-dependent alterations of invasion in vitro. [Schematic representation of data from Vleminckx et al. (432).]

PTEN (phosphatase and tensin homolog deleted on chromosome 10), also termed MMAC1 (mutated in multiple advanced cancers) or TEP1 (TGF- regulated and epithelial cell enriched phosphatase 1), was the first tumor suppressor gene encoding a protein with phosphatase activity (69). The gene is located on chromosome 10q23 and is a candidate invasion suppressor because of its late inactivation during cancer development (233, 379, 392). Its mutation frequency in human cancers is very high and close to that of p53. Mutations were described in high-grade but not in low-grade gliomas, independent of p53 mutations (114, 330), in bladder cancer (65), in advanced prostate cancer, and cell lines derived therefrom (66, 444). PTEN acts on both proteins and lipids; as a protein phosphatase it has a dual specificity acting on tyrosine and on serine/threonine. Its main targets as a lipid phosphatase are phosphoinositides where PTEN dephosphorylates the three position and in this way counteracts PI 3-kinase. Clustering of mutations in the lipid phosphatase domain, e.g., G129E, suggests that this domain is critical for the tumor-suppressor activity (284). The involvement of the protein phosphatase domain and the loss of expression observed in some cancers led to the conclusion that PTEN has a dual role, regulating growth and survival through its lipid phosphatase activity and adhesion and invasion through its protein tyrosine phosphatase activity (392, 444). More recent experiments with RAS or SRC transformed MDCK (MDCK-ras and MDCKts.src) cells favor the opinion that the lipid phosphatase activity of PTEN is implicated in stabilization of junctional complexes and restriction of invasion (218). Indeed, successful transfection of these MDCK transformants with wild-type PTEN, but not with mutants deficient in lipid phosphatase activities, induces cellular aggregation and abolishes invasion. The implication of the E-cadherin/catenin complex in the expression of the noninvasive phenotype was demonstrated by the proinvasive action of antibodies that functionally neutralized E-cadherin. The same tranfections counteracted invasion also in PTEN-defective cell lines derived from neuroblastoma, melanoma, and prostate carcinoma. Like for the genes discussed above, the product of the PTEN gene is involved not only in migration but also in proliferation and survival (107).

The multistep invasion process of metastasis explains that invasion is a prerequisite for metastasis; it does, however, not account for the large differences in metastatic ability of invasive tumors. A working definition of metastasis genes, different from invasion genes, might be that their activation or inactivation changes the metastatic phenotype of invasive tumors. In the above-mentioned example of crosses of Rip1Tag mice with mice expressing dominant negative E-cadherin (Rip1dnE-cad), no distinction can be made between the acquisition of invasion and metastasis. MTS1 (metastasin 1) possibly meets the criteria of a metastasis promoter gene because it conveys metastatic capability upon invasive nonmetastatic tumors (4). Mice of the GRS/A strain carry a mouse mammary tumor virus (MMTV) provirus and have a high incidence of mammary tumors, due to the proviral activation of the oncogenes WNT and INT-2. Histologically, these tumors represent moderately differentiated invasive adenocarcinomas. The GRS/A mice were crossed with transgenic mice expressing the MTS1 gene in the lactating mammary gland under the control of an MMTV promoter. These MTS1 transgenics do not develop mammary tumors. Successful crosses between GRS/A and MTS1 mice have mammary cancers that are not only locally invasive but also form metastases in the lungs. Here, MTS1 acts as a metastasis-promoter gene, but the phenotypic alteration responsible for the formation of metastases from invasive primary cancers is not clear.

During normal embryonic development, spatiotemporal activation and inactivation of invasion-promoter genes and invasion-suppressor genes participate at the regulation of gastrulation and morphogenesis. The idea to consider some of the proteins encoded by these genes as promoters or suppressors of cancer invasion came from embryology (390). For example, E-cadherin is first expressed at the morula stage, hence its former name uvomoruline, where it serves compaction, the earliest form of epithelial organization (178, 224). At the onset of gastrulation, when cells start to migrate from the ectoderm undergoing an epithelial to mesenchymal conversion, E-cadherin is downregulated and N-cadherin is expressed. This switch of cadherin expression from DE- to DN-type (D for Drosophila) occurs downstream of the invasion promoter genes Twist and Snail (302). Twist encodes a nuclear protein containing a helix-loop-helix motif, which probably acts as a transcription factor. At gastrulation in Drosophila melanogaster, Twist-positive cells roll into the presumptive mesoderm, as beautifully illustrated in Reference 226; Twist / mutants, like Snail / mutants, fail to complete gastrulation (68). Morphogenetic activities that act through activation of the E-cadherin gene are found also in hepatocytes, through the CDH1-binding transcription factor hepatocyte nuclear factor (HNF)-4 (375) and in thyrocytes, through thyroid stimulating hormone (52). Reversion of an invasion-associated phenotype, namely, mesenchymal to epithelial transition, is observed during metanephrogenesis. This transition was mimicked in the human fetal kidney cell line HEK293 where expression of PAX-2, a member of the "paired-box" homeotic gene family, was associated with a gain of E-cadherin and -catenin expression (408).

In microorganisms, virulence genes are regulators of invasion. Historically, the first transformation from nonvirulent into virulent Streptococcus pneumoniae formed the basis for the identification of DNA as the genetic material by Avery et al. (12). Transfection was applied to the analysis of specific invasion genes first in bacteria (181). Transfer of a single genetic locus from the invasive bacterium Yersinia pseudotuberculosis made the noninvasive Escherichia coli invasive into cultured vertebrate cells. When the noninvasive L. innocua is transfected with a plasmid harboring the internalin A (inlA) gene, the bacterium becomes invasive into Caco-2 cells (141). The invasion assay took advantage of the fact that invaded, hence intracellular, bacteria are protected against antibiotics that fail to penetrate into the cells (see Fig. 3). In Listeria, virulence genes regulate not only entry into the vertebrate cell, but also intracellular multiplication and spreading. Six of these virulence genes are clustered on the bacterial chromosome (prfA, plcA, hly, mpl, actA, and plcB), the two others (inlA and inlB) form a distinct operon. All these genes are coordinately regulated by PrfA, the trancriptional activator encoded by the prfA gene (268). Such a gene, switching on and off coordinated invasion programs, has not been found in cancer cells, so far. The invasive, virulent, and pathogenic E. histolytica differs genetically from the noninvasive, avirulent, and nonpathogenic E. dispar as evidenced by restriction fragment length polymorphism and sequencing of single copy genes (320). Here, like in bacteria, virulence genes are crucial for invasion; downregulation of their expression by antisense RNA transfections causes a reduction of invasion-associated molecules such as amoebapore (54), the 35-kDa light subunit of the Gal/GalNac specific lectin (8) and the cysteine proteinase 5 (9). The life cycle of the parasite Schistosoma mansoni provides an example of spatiotemporal regulation of an invasion-promoter gene at the cercaria stage. Before the cercaria leaves its snail host to swim freely in the water, the gene encoding a serine protease is switched on. This powerful lytic enzyme is activated only upon contact with the human skin and in this way acts exclusively at the site of invasion. Once the cercaria has invaded into the dermis of its new host, the protease gene is switched off (132).

Taken together, the above-mentioned examples of invasion-suppressor and invasion-promoter genes clearly demonstrate that multiple genes are implicated in invasion and that they vary between different types of cells. Their activation or inactivation triggers programs of cellular activities that are rarely restricted to invasion and usually changes also the other cellular activities depicted for cancer cells in Figure 1.

	IV. CANCER CELLS, HOST CELLS, AND TUMOR CELLS: ALL INVADERS

Top Previous Next References

Cellular behavior and gene activation or inactivation are greatly influenced by the environment in normal as well as in pathological situations including cancer (see Fig. 1). The Lancet's first volume (313) launched the "seed and soil" hypothesis asking the question: "What is it that decides what organ shall suffer in a case of disseminated cancer?" His answer is still valid: "The microenvironment of each organ (the soil) influences the survival and growth of tumor cells (the seed)." Pathologists have since a long time recognized that tumors contain not only neoplastic cells, further called cancer cells, but also host cells. The host participation at the establishment of the tumor is described as desmoplasia, consisting of fibroblastic cells and extracellular matrix, as inflammation and immune response represented by lymphocytes, macrophages, and dendritic cells, and as angiogenesis evidenced by newly formed blood and lymph vessels. These host elements, although more abundant in some types of cancer than in others, are omnipresent. For example, less than half of most pancreatic cancers are occupied by cancer cells, the majority being host cells. In line with this histological observation is the detection of a cluster of invasion-specific expression of genes encoding molecules that participate at the reaction of the host (345).

There are clinical and experimental data to believe that host cells play a major role in invasion and metastasis. Metastasis may depend on the specific site of the primary cancer. The frequency of distant metastasis from squamous cell carcinomas of the head and neck region depends on the subsite of the primary cancer, varying from 3.1% for tumors situated in the larynx to 28.1% for tumors in the nasopharynx as summarized in Reference 249. More recently, the latter figure has increased to >40% as new treatment techniques such as intensity modulated radiotherapy (IMRT) have improved local tumor control and patient survival (388). Orthotopic, compared with paratopic (usually indicating subcutaneous), implantation into immunosuppressed mice provides an experimental demonstration of site specificty of invasion and metastasis. When human colon cancer cells are implanted in the wall of the cecum (325) or when oral squamous cell carcinoma cells are implanted into the tongue of nude mice (202), they invade and metastasize in contrast to their subcutaneous counterparts. Organ specificity indicates that some tumors metastasize more frequently to specific organs than could be expected from their transport in the circulation and their passage through capillary networks. Examples of preferentially affected organs are the brain for lung cancer and melanoma and bone for prostate and breast cancers (436). It is still a matter of debate whether organ specificity of metastasis is due to specific homing and extravasation or to specific survival and growth of the cancer cells at the site of extravasation. The type of host cells, e.g., endothelial cells, that participate at tumor development all are invasive themselves and some, e.g., leukocytes, are even metastatic. It is, therefore, justified to ask the question who is invading who (238). The recruitment of host cells is most likely the result of the production by the tumor microecosystem of stimulatory and inhibitory factors. Moreover, these host cells may proliferate in the tumor ecosystem, again governed by balances between inhibitory and stimulatory factors. Cancer cells may also cause transdifferentiation of host cells, e.g., fibroblasts into myofibroblasts.

The role of myofibroblasts, first described as smooth muscle-like fibroblasts by Gabbiani et al. (140), in cancer invasion has been recently reviewed (106). The emphasis of this review is on the continuous molecular cross-talk between the cancer cells and the host (Fig. 5). Cytokines, such as TGF- and platelet-derived growth factor (PDGF), are released from the cancer cells, probably at a proinvasive state of tumor development; they stimulate the transition of fibroblasts into myofibroblasts. The latter cells are found, indeed, more frequently in preinvasive lesions of the colon such as villous adenomas and FAP, that have a higher risk of transition into invasive carcinoma, than in lesions such as tubular adenomas, that have a lower risk of progression (260). Myofibroblasts do participate at numerous noncancerous pathological and physiological processes. During wound healing they assist at migration, proliferation, and contraction. When the wound is closed, myofibroblasts undergo apoptosis, quite in contrast to tumors where they persist as in a wound that does not close (115). This idea illustrates that cancer cells operate by noncancer specific activities, but they fail to regulate these activities properly. Myofibroblasts produce numerous molecules, growth and motility factors, angiogenic factors, extracellular matrix components, and proteinases, that all promote the invasion and also the growth of cancer cells. Other molecules of putative interest for invasion expressed by myofibroblasts include -smooth muscle actin, vimentin, c-MET (404), proteolytic FAP displaying also dipeptidyl peptidase activity (316), cyclooxygenases (COX)-1 and -2 and N-cadherin, associated with -catenin, -catenin, p120^CTN, and T-catenin (187, 425). An early demonstration of the invasion-stimulating activity of myofibroblasts resulted from the differential behavior in vitro compared with in vivo of PROb cancer cells that were isolated from a chemically induced rat colon tumor (108). PROb cells invaded neither into collagen nor into Matrigel nor into embryonic chick heart in culture. Upon subcutaneous injection into syngeneic rats, PROb cells did, however, produce invasive cancers, and numerous myofibroblasts were present at the front of invasion in line with observations on human cancers. PROb cells were stimulated to invade also in vitro, provided myofibroblasts were added to the culture system. These myofibroblasts themselves were also invasive. One interesting example of noncancerous myofibroblast invasion is described in experimental tubulointerstitial fibrosis (288, 386). There is good histological and ultrasctructural evidence to accept that tubular epithelial cells transdifferentiate into myofibroblasts invading through a disrupted basement membrane into the underlying stroma. Like in cancer, TGF- produced by the tubular epithelial cells stimulates fibrosis (322).

View larger version (47K): [in this window] [in a new window]   Fig. 5. Micro-ecosystem of a primary cancer. Cellular activities associated with invasion implicate reduced cell-cell adhesion, altered cell-matrix adhesion, migration, ectopic survival, and lysis of extracellular matrix. Most of these activities are modulated by the cross-talk between cancer cells and host cells, which are activated by cytokines released from the cancer cells. [Adapted from Mareel et al. (251) and De Wever and Mareel (106).]

Blood vessels and lymph vessels provide tumors with nutrients and cytokines, necessary for growth and invasion; they provide the routes for systemic spread of cancer cells; and they mediate the communication between the primary tumor and its metastasis (309). These vessels represent the response of existing blood and lymph vessels to balances between positive and negative angiogenic factors produced by the cancer cells. The type of vessels, hematogenic or lymphatic, that are invading the tumor might be determined by the type of vascular endothelial growth factor (VEGF) produced (368). Because invasion into vessels initiates metastasis, the type of VEGF might also determine the route of metastasis, lymphogenic or hematogenic. In exceptional cases, the cancer cells themselves may transdifferentiate into endothelioid cells and form the tumor vascular system, a phenomenon that is called vasculogenic mimicry (244). Several excellent reviews on tumor angiogenesis were produced (70, 134, 207). We would, therefore, limit the discussion to the observation that neoangiogenesis may be needed for primary invasion as well as it is for growth as evidenced by an in vivo mouse model (14, 15). Transplantation of collagen gels coated with malignant murine keratinocytes on the dorsal muscle fascia of wild-type mice resulted in invasive squamous cell carcinomas. When plasminogen activator inhibitor (PAI)-1 / knock-out mice were used, the cancer cells failed to invade, and there was no angiogenesis. Intravenous injection of a recombinant adenovirus vector carrying the human PAI-1 cDNA restored angiogenesis and invasion. The molecular explanation is that plasmin proteolysis must be tightly controlled to allow vessel stabilization and differentiation.

Tumor tissues are frequently infiltrated by host leukocytes, sent in by the immune system of the host in an attempt to reject the tumor. Indeed, some of these host cells are able to kill cancer cells or to secrete antiangiogenic factors. A recent example is provided by the high susceptibility to skin carcinogenesis of mice lacking T cells (149). It is, however, evident that such infiltrated cells can also have tumor-promoting effects. This may be illustrated by the earlier finding that nonmetastatic lymphoma cells become metastatic upon fusion with activated leukocytes (96). The countercurrent principle originally developed in chemistry to separate oil lovers from water lovers and, later, in Drosophila genetics of behavior to separate light lovers from dark lovers, was applied to the helper function of leukocytes in invasion (308). Cancer cells produce chemotactic cytokines (called chemokines), a family of small proteins that attract leukocytes from the circulation along a chemical gradient toward the tumor. These chemokines also stimulate the production of MMPs by the attracted leukocytes, that dissolve the extracellular matrix (ECM) on their way to the tumor. By doing so, a tunnel is created for the invading cancer cells (Fig. 6). Moreover, the chemokines act as growth factors for the cancer cells and are angiogenic, so providing vessels that mediate invasion and serve as routes for metastasis.

View larger version (22K): [in this window] [in a new window]   Fig. 6. Schematic of the countercurrent model of leukocyte-assisted cancer invasion, as proposed by Opdenakker and Van Damme (308).

In bone metastasis, a frequent complication in malignant tumors, cancer cells subvert the host dynamic homeostatic mechanisms that preserve the structure and function of the skeleton. The result is excessive breakdown of bone, pain, and eventual fractures. Cancer cells release osteoclast activating factors, such as interleukin-1, tumor necrosis factor, TGF-, epidermal growth factor (EGF), PDGF, and prostaglandins, and activated osteoclasts cause breakdown of bone matrix (Fig. 7). This breakdown is probably not limited to osteoclasts as it has been attributed also to cancer cells and even to osteoblasts and osteoblast-like cells (221). Bone matrix breakdown releases chemotactic factors attracting cancer cells and growth factors stimulating their proliferation. Osteoclasts are successful targets for treatment of bone metastasis by bisphosphonates acting through inhibition of osteolytic activity, fortifying bone matrix and interfering with the formation of osteoclasts from monocytes (172, 282). Bisphosphonate acts not exclusively on osteoclasts as it induces also apoptosis in human breast cancer cells in experimental animal models. The naturally occurring decoy receptor osteoprotegerin, a member of the tumor necrosis receptor family, also inhibited metastatic osteolysis an