Occupy tissue: the movement in cancer metastasis.

The critical role of migration and invasion in cancer metastasis warrants new therapeutic approaches targeting the machinery regulating cell migration and invasion. While 2-dimensional (2D) models have helped identify a range of adhesion molecules, cytoskeletal components and regulators that are potentially important for cell migration, the use of models that better mimic the 3-dimensional (3D) environment has yielded new insights into the physiology of cell movement. For example, studying cells in 3D models has revealed that invading cancer cells may switch between heterogeneous invasion modes and thus evade pharmacological inhibition of invasion. Here we summarize published data in which the role of cell adhesion molecules in 2D vs. 3D migration have been directly compared and discuss mechanisms that regulate migration speed and persistence in 2D and 3D. Finally we discuss limits of 3D culture models to recapitulate the in vivo situation.

Introduction

Nearly all adherent cells are able to crawl and migrate under standard 2D cell culture conditions. Time-lapse recordings of migrating cells never fail to excite a sense of awe about the complexity of the migration process, as lamellipodia and filopodia form, extend and make contact with the matrix, as cells change their shape, lunge forward, contract and detach. Cell migration depends on a variety of temporally and spatially orchestrated parameters, with several hundreds of proteins involved. Although changes in cell migration after knockout or mutation of a protein may not be the ultimate proof of a causal relationship; watching cells migrate and measuring parameters of cell speed and directional persistence have proven invaluable in elucidating the functional components regulating intrinsic cell migration in 2D. A number of studies have now begun to examine these same parameters using 3D culture �systems.

Interstitial �Dissemination

The ability of cancer cells to invade tissue surrounding the primary tumor site, leading to the development of metastatic tumors, is one of the hallmarks of cancer. Disseminating tumor cells encounter a variety of tissue architectures through which they must transmigrate. While the majority of tumor cells are confronted by collagen-rich connective tissue found around most organs in the body, invasion in the brain represents a specialized form of transmigration due to the unique makeup of the brain interstitium. For more detail we direct readers to an excellent, recent review that provides a comprehensive description of tissue structures through which cancer cells transmigrate in vivo. The structure and composition of the interstitial tissues determines the mechanisms that must be employed by the cancer cells for successful navigation. Gritsenko and colleagues suggest three generic mechanisms employed by invasive cancer cells: contact guidance along stiff matrix elements through integrin mediated mechanisms, directed migration toward chemokines and growth factors and physical pushing through small �spaces.

Cell Migration In �Vivo

The process of cell migration and invasion is, of course, not restricted to metastatic cancer. Cell migration and invasion occurs during embryonic development and continues throughout adulthood as it is critical for the correct execution of a variety of biological programs. During embryonic development there are precise migration and invasion events that are necessary for the final correct organization of the adult tissue. A particularly well-studied example of this is during the development of the mammalian brain, where cells undergo precisely timed migration events that are essential for the final tissue organization and wiring of the fully developed brain (for review, see ref. 3). The migration speed and persistence of distinct cell populations appear to be important parameters for determining their final destination in the maturing brain tissue. Moreover, cell populations change speed as they migrate throughout different zones of the brain. Mutations that alter speed and directional persistence of migrating neuronal cell populations can therefore perturb the organization of the brain tissue into discrete �layers.,

The consequences for changes in speed and persistence of invasive cancer cells are currently less well known. Approaches that abrogate cell movement in environments that mimic the tissue organization of the in vivo tumor are generally considered to be strong candidates for novel therapies to treat metastatic cancer. Similarly, approaches that do not completely abrogate movement but instead slow down the rate of cell migration are expected to restrict the dissemination of cells from the primary tumor and increase the window of opportunity for localized and targeted delivery of therapies. In the particular case of highly invasive primary brain tumors such as the deadly grade IV gliomas (also known as glioblastoma), there is extensive collateral damage to the brain tissue as a result of the invading tumor cells, leading to significant loss of quality of life. Thus, even a partial reduction of tumor invasion may ameliorate some of the collateral damage that occurs in the brains of these patients. Moreover, targeting the cell’s migration machinery may in turn sensitize the cells to different therapies. One example of such sensitization is tumor cells lacking expression of the invasion/metastasis promoter NEDD9. These cells appear to concomitantly become sensitized to the effects of Src kinase inhibition. Similarly, decreasing the persistence of the cancer cell movement will necessarily decrease the numbers of successful escapers from the primary tumor site. But to confirm these speculations and expectations, any anti-invasive therapy must first demonstrate its effectiveness in models that mimic the natural in vivo 3D organization of the relevant tissue surrounding a particular �tumor.

3D Culture �Models

The interactions between cells and the external environment, either with the ECM or with neighboring cells, are in a literal sense vitally important. Once thought to solely provide physical and structural support, the ECM is now known to regulate and influence survival, proliferation, migration and adhesion,, morphology, internal cellular structures and signaling pathways. During tumorigenesis, cellular interaction with the ECM can become deregulated not only as a result of altered integrin and focal adhesion protein expression, but also because tumor cells can rearrange the local native tissue architecture, alter its chemical composition, structure and stiffness, which in turn often promotes cell migration and invasion., For example, in breast carcinomas the formation of the distinctive palpable “lump,” known as desmoplasia, is attributed to tumor cell secretion of platelet derived growth factor (PDGF). This induces the deposition of large quantities of collagen and collagen cross-linkers by surrounding fibroblasts, thus increasing local tissue stiffness and �rigidity.,

In vivo, most invading cancer cells encounter two major ECM groups. The first is the basement membrane that directly interacts with both the epithelium and endothelium layers. The second is the interstitial matrix. It harbors numerous collagen isoforms and fibronectin and is thought to contribute to the overall mechanical strength of the tissue. In vitro, the extracellular matrix is most commonly planar, tissue culture-treated plastic or sometimes glass coated with collagen, fibronectin or other ECM-derived proteins. Even then, comparative analysis of cell migration and morphology has revealed striking inconsistencies between cells grown on 2D surfaces vs. those seeded within 3D matrices., Therefore, in attempting to recreate a cell’s in vivo environment, it is not sufficient to only consider which ECM group the cell type in question is most likely to interact with; it is equally important that the 3D structure and mechanical properties of the matrix closely mirror that of the in vivo environment. Three of the most commonly used in vitro 3D matrix models—fibrous gels, basement membrane extracts (or matrigel) and cell-derived matrices—are discussed in the following �sections.

Fibrous �gels

Fibrous gels, namely reconstituted collagen and fibrin gels, aim to replicate the major architectural and structural components of the extracellular matrix. Collagen and fibrin emerged as ideal candidates for single-protein 3D matrix models as both are major structural components of the extracellular matrix. It is worth noting, however, that interstitial tissue is composed of heterogenous mixtures of ECM components, which these gels do not recapitulate. Nevertheless, due to the abundance of collagen and fibrin in vivo, reconstituted fibrous gels are physiologically compatible with numerous cells types. Fibrin gels are particularly suitable to study migratory events involved in wound healing,, while collagen gels are more suitable to study interstitial cell migration and metastatic �invasion.,,

Not all fibrous gels are created alike, however. Seemingly minor protocol variations can have major effects on the gel properties. For instance, an increase in fiber concentration enhances fiber density,, resulting in an overall increase in the mechanical strength of the gel. Additionally, the effective pore size of the fibrous meshwork decreases at higher fiber concentrations, thus increasing steric hindrance for migrating cells., The gelation temperature will also affect the mechanical and structural details of the gel; for instance, the pore size and fibril diameter decreases and the number of fibrils formed increases at higher polymerization �temperatures.,

The production details of harvesting, dissolving and purifying collagen or fibrin from different animal tissues are also a source of substantial variability. This is especially true in the case of collagen, where fibers harvested for example from bovine Achilles tendon demonstrate a higher denaturation temperature and a thicker fiber diameter when compared with collagen harvested from rat tail tendon.,, The method for dissolving the collagen, whether with acid or by proteolytic digestion (i.e., pepsin or trypsin) similarly influences the structure and polymerization kinetics of the reconstituted gels. Acid-dissolved collagen will begin the fibrillogenesis process more rapidly,,, showing characteristic fiber lengths and diameters that closely resemble those observed in vivo. In comparison, collagen extracted by proteolytic enzyme digestion shows significantly smaller fiber diameters and reduced fibril formation, thus reducing the gels’ strength and �rigidity.,,

Basement membrane �extracts

Matrigel is a reconstituted basement membrane matrix product extracted from Engelbreth-Holm-Swarm (EHS) mouse sarcoma cells. When polymerized, matrigel mimics the microenvironment of in vivo extracellular basement membrane.- Composed chiefly of laminin, collagen IV isoforms and heparan sulfate proteoglycans, basement membranes provide a unique signaling platform for cells that adhere to it., Matrigel 3D substrates can restore the normal morphological characteristics and specific cell functions that would otherwise be lost under standard cell culture conditions, especially in epithelial-,, endothelial- and Schwan-derived cells. But this is not generally true for all cells; for instance fibroblasts that are not normally in contact with a basement membrane in vivo adopt a noticeably uncharacteristic rounded morphology and exhibit a non-migratory phenotype when cultured in �matrigel.,

Cell-derived �matrix

When cultured at high density, fibroblasts “bioengineer” their own extracellular matrix. This cell-derived matrix (CDM) is composed of fibronectin, collagen types I and II, hyaluronic acid and heparin sulfate proteoglycans., CDMs mirror the flexibility and malleability of an in vivo ECM, as cells seeded onto a CDM can reorganize and modulate the matrix, recapitulating the migration and invasion events of both physiological and metastatic processes observed in vivo.,, However, as with all protocols, there are limitations to this method also. Somewhat problematic is the poorly defined and highly variable composition of fibroblastic CDMs. Moreover, although CDMs display a 3D matrix organization, these matrices are typically thin and therefore subject to the rigidity of the underlying 2D surface. Regardless, the use of CDM has permitted insights into the signaling pathways that govern proliferation, morphology, cell-matrix attachments and individual cell migration, all of which impact and promote �tumorigenesis.,,

Cell �Adhesion

The role of focal adhesions has been thoroughly studied in 2D cultures and has provided important insight into the mechanisms of cell movement in 2D. However, as investigators have turned to the analysis of focal adhesions in 3D, there has been conflicting reports regarding their presence and detection in 3D.,, We direct readers to the excellent review by Harunaga for a comprehensive analysis of the issue of cell-ECM adhesions in 3D culture models. The upshot of which is that focal adhesions do exist in 3D, but the composition, mechanical properties and structure of the 3D matrix profoundly affects the appearance and distribution of focal �adhesions.

In Table 1 we summarize data from studies in which migration parameters of speed and persistence have been directly compared between 2D and 3D culture conditions in the same cell background. Treatments that are expected to promote invasion and metastasis in vivo such as stimulation with EGF, complete oncogenic transformation following combined expression of ErbB2 and 14-3-3ζ and high α5β1 expression all promoted faster migration speed in 3D cultures, yet had variable effects in 2D (Table 1). For example, breast epithelial cells induced to undergo oncogenic transformation displayed reduced migration speed in 2D, while EGF treatment of glioblastoma cells increased 2D speed. Similarly, cells from a metastatic progression series showed increasing speeds in 3D collagen gels, but decreasing speeds in 2D. Without exception, the knock-down or loss of cytoplasmic members of focal adhesion sites (p130Cas, NEDD9, vinculin, talin, FAK, VASP, paxillin and Hic-5) inhibited cell migration speed in 3D cultures, yet again varied results are seen in 2D cultures. Depletion of these molecules variously increased 2D speed (p130Cas, NEDD9 and vinculin knockout fibroblasts), decreased 2D speed (talin and FAK) or did not alter speed relative to controls (vinculin shRNA in HT1080 fibroblasts and VASP). With the exception of EGF stimulation and high α5β1 expression, all treatments resulted in reduced intrinsic persistence of migration in 3D. As was seen for the speed data, there is no clear correlation between effects on migration persistence in 2D and effects seen in �3D.

Table 1. Studies reporting cell speed and persistence in 3D cultures compared with 2D migration parameters in the same cell �background

Adhesionmolecule	Cell line	2D		3D
		Speed	Persistence	Speed	Persistence
p130Cas shRNA52	HT1080fibrosarcoma	↑	↑	3D collagen↓	↓
NEDD9 knockout60,†	Fibroblast	↑	ND	3D collagen↓	↓
Vinculin knockout87	Fibroblast	↑	↑	3D collagen↓	ND
Vinculin shRNA52	HT1080fibrosarcoma	—	—	3D collagenNo change	↓
β3 integrinknockout59,*	Fibroblast	↑	↓	CDM#↑	↓
EGF stimulation56	U87MGglioblastoma	↑	↓	3D collagen||↑	3D collagen↑
β1 integrininhibition52	DU-145prostate cancer	Biphasic	ND	Low matrigel %:↑High matrigel %:↓	ND
Talin shRNA52	HT1080fibrosarcoma	↓	↑	3D collagen↓	↓
FAK shRNA52	HT1080fibrosarcoma	↓	↑	3D collagen↓	↓
VASP shRNA52	HT1080fibrosarcoma	—	—	3D collagen↓	↓
Transformationprogressionseries:55(1) +ErbB2(2) +14-3-3ζ(3) +ErbB2+14-3-3ζ	MCF10a non-transformedbreast epithelia	(1) ↑(2) ↓(3) ↓	(1) ↓(2) ↓(3) ↓	“Compliant”3D collagen‡:(1) —(2) —(3) ↑“Stiff”3D collagen§:(1) ↓(2) ↓(3) ↑	“Compliant” 3D collagen‡:(1) —(2) —(3) ↓“Stiff”3D collagen§:(1) ↓(2) ↓(3) ↓
Metastasis progression series:57(1) 67NR—tumorigenic(2) 168FARN—micrometastases(3) 4T07—secondary micrometastases(4) 66c14—secondary Tumors	NmuMg mousebreast epithelialline	↓(relativeto67NR)	↓(relativetoNmuMG)	Collagen gel:↑(relative to67NR)	Collagen gel:66c14↓
Paxillin siRNA94,¶	MDA-MB-231breast cancer	ND	ND	CDM#↓	↓
Hic-5 siRNA94	MDA-MB-231breast cancer	ND	ND	CDM#↓	↓
High α5β162	MDA-MB-231breast cancer	ND	ND	3D collagen↑	↑

Reduced focal adhesion length and increased rates of focal adhesion turnover; †increased rates of focal adhesion disassembly; ‡elastic modulus = 103 Pa; §elastic modulus = 391 Pa; ||effects were compared across a range of collagen concentrations (2, 3 and 4 mg/mL); ¶measured dynamics of GFP-tagged talin adhesions in CDM and showed multiple roles—paxillin depletion caused increased numbers of highly dynamic, peripheral, short-lived adhesions, increased stability of more centrally located adhesions accompanied by reduced rates of adhesion assembly and disassembly. #cell-derived matrix; ND, not determined; —, no change.

Forces in Cell �Migration

The disparate results for 2D and 3D speed and persistence in exactly the same cell background highlight critical differences between these different model systems. Cell migration in 2D differs in one fundamental aspect from 3D migration in that steric hindrance is absent. In 2D, the cell needs to overcome only the frictional (drag) forces from the surrounding liquid and the 2D surface. At a speed of only a few microns per minute, liquid drag can be neglected, as the force of a single myosin motor would be sufficient to propel the cell forward. Thus, the better a cell adheres to its substrate, the greater the contractile force needed to overcome the adhesive friction from cell-matrix adhesions; as a general rule, a cell with poorer adhesion can migrate faster on a 2D surface. Poorer adhesion can either be a consequence of a lower matrix protein density on the surface, or a consequence of a reduced expression level of adhesion proteins on the cell (for examples, see Table 1). Of course, a minimum amount of adhesion is essential for cell movement as otherwise the cell cannot polarize or activate its contractile machinery. Thus, in a 2D context, changes in focal adhesion dynamics, dominantly effect cell migration speed and likely explain the disparate effects on 2D cell migration seen following depletion of focal adhesion-associated molecules (Table �1).

In 3D, in addition to frictional forces, the cell also has to overcome forces that arise from the steric hindrance of the matrix network, provided that the pores and crevices through which the cell migrates are smaller than the cell itself. Here, the cell has two options. It can either deform itself until it can fit through the pores; or it can deform the network until the pores are large enough for passage. Switching between cell body deforming vs. matrix deforming migration strategies is evidenced by changes in cell morphology from rounded cell shapes to elongated cell shapes. Deforming the cell body sufficiently to squeeze through small pores requires forces for overcoming the elastic and frictional forces of the cytoskeleton. The cell can decrease cytoskeletal elasticity (stiffness) and friction by depolymerizing cytoskeletal filaments, but although this would reduce the forces necessary to deform the cell until it fits through a pore, it also reduces the force-generating capacity of the acto-myosin contractile �apparatus.

Analogously, deforming the matrix network requires forces for overcoming the elastic and frictional forces of the matrix. By secreting matrix-digesting enzymes, the cell can soften the network structure or increase its porosity, thereby reducing the necessary forces for deforming the matrix network until the cell can fit through. But at the same time this also limits the ability of the cell to adhere to the matrix, which is a prerequisite for contractile force generation. Therefore, does the speed of intrinsic cell migration in 3D give clues as to whether cells are using the rounded (cell body deforming) vs. elongated (matrix deforming) cell migration strategies? Initial descriptions suggested that there is no loss of cell speed between the two modes. However, absolute cell speed depends on the make-up of the surrounding matrix. While partial transformation of breast epithelial cells (Erb2 or 14-3-3ζ overexpression) did not affect cell speed in “compliant” collagen gels (103 Pa elastic modulus), cells had reduced migration speed in “stiff” collagen gels (391 Pa elastic modulus). Similarly, increasing steric hindrance at high collagen concentrations can inhibit cell invasion. Thus the concentration of the matrix, extent of cross-linking and matrix pore size all critically determine cell speed and moreover, whether a matrix deforming or cell body deforming mode of invasion is possible. Given the ultimate goal of determining the molecular mechanisms that promote cancer cell invasion and metastasis, it will be increasingly important to adopt a range of 3D culture models that can mimic the various different extracellular environments encountered by invading cancer cells in vivo. Two examples from recent studies with opposing findings highlight the problem in interpreting data that are measured under only a single condition: Several studies have reported that invasive, metastatic cancer cells are softer than their nonmetastatic counterparts derived from the same tissue, when measured under non-adherent conditions. By contrast, in adherent cancer cells, the opposite behavior is seen, with more invasive cells tending to be more contractile and consequently �stiffer.,

Predictive Power of 3D Invasion Models for Cancer �Metastasis

The goal of the 3D cell migration models is to better mimic the physiological environment. An important question in cancer research therefore is how successfully data from the 3D models predicts in vivo metastatic behavior. To address this question we have searched the literature for examples where adhesion molecules that have been directly compared in 2D vs. 3D models (Table 1) have been analyzed using in vivo models of metastasis and invasion (Table 2). There is an extensive literature on mouse models of cancer cell invasion, but for our purposes we have limited our analysis to studies where adhesion molecules from Table 1 have been manipulated by overexpression or knockdown. It is noticeable that surprisingly few studies have been performed to directly test the effects of these molecules on in vivo tumor invasion and metastasis—the bulk of studies have focused on tumor initiation. Our search yielded examples for FAK, p130Cas, NEDD9, vinculin, talin, β1 integrin and α5β1 and αvβ3 integrin receptors (Table �2).

Table 2. Summary of in vivo metastasis data, targeting the adhesion molecules described in Table �1

Adhesionmolecule	In vivo model	Metastasis phenotype
FAK	Intravenous injectionof FAK−/− tumor cells82	Failed to form lung tumors; cells retained in the lung capillary bed were rounded and lacked membrane extensions into the vessel
	Activated ErbB2 mammary tumor model with epithelial-specific FAK deletion69	Metastatic lung tumors all negative for cells with homozygous deletion of FAK
	Mammary epithelium-specific deletion of FAK in MMTV-PyVmT Mouse Tumor Model70	Metastatic lung tumors all negative for cells with homozygous deletion of FAK
	Mammary epithelium-specific deletion of FAK in MMTV-PyVmT Mouse Tumor Model71	Reduced lung tumor metastases
	Orthotopic injection of pancreatic cancer cells treated with FAK siRNA73	Prevented formation of liver metastases
p130Cas	Mammary tumor model—injection of cells expressing inducible p130Cas shRNA77	Inhibits lung colonization
	Athymic nude mice injected sub-cutaneously with p130Cas−/− fibroblasts transformed with oncogenic Src and expressing p130Cas78	Exogenous p130Cas expression increased formation of metastatic lung tumors after surgical removal of primary tumors; authors comment that “the capacity of the cells to invade through matrigel was strongly correlated with their capacity to invade and metastasize in vivo”
Vinculin	Exogenous vinculin expression in highly metastatic rat adenocarcinoma injected into foot pad80	Highest levels of vinculin expression suppressed formation of lung metastases, low to moderate expressors formed tumors in lymph nodes close to injection site but failed to form lung metastases
NEDD9	Mammary epithelium-specific deletion of NEDD9 in MMTV-PyVmT Mouse Tumor Model67	Trend to fewer lung metastases
	Tail vein injection of NEDD9-null primary tumors8	Tumors formed of null-cell lines exhibited increased aggressiveness, with all injected mice generating secondary tumors
Talin	Tail vein injections of prostate cancer cell lines treated with talin shRNA84	Reduced numbers of metastatic lung lesions
αvβ3	Orthotopic injection into mammary fat pad with mammary carcinoma cell line expressing exogenous β376	Drove unique formation of bone metastases; authors show increased haptotactic and chemotactic response to bone-matrix proteins and soluble factors
	MDA-MB-435 breast cancer cells expressing constitutively active αvβ3 injected into mouse tail vein81	Enhanced lung colonization
	Intravenous injection of metastatic αvβ3 negative melanoma cells expressing exogenous β391	Re-expression of β3 in metastatic, β3 negative lines reduced lung colonization
β1 integrin	Conditional deletion of β1 integrin from mammary epithelia, crossed with MMTV/activated erbB272	Significantly reduced formation of lung metastases
	Orthotopic injection of pancreatic cancer cells treated with β1 integrin siRNA74	Absence of any metastatic tumors; controls treated with α2 or α3 integrin subunit siRNA displayed metastatic tumors
	Conditional deletion of β1 integrin from pancreatic β cells crossed with Rip1Tag2 mice68	Loss of β1 expression induced increased tumor cell emboli in the lymphatic vasculature but no metastasis formation; similarly, tail vein injections of β tumor cells lacking β1 integrin expression did not form metastases
	Ras-myc transformed β1 null ES cells injected sub-cutaneously79	Reduced numbers and size of metastatic foci in the lung
α5β1 integrin	HT-29 colon cancer cells expressing exogenous α5 integrin injected intravenously85	Significantly reduced lung and extrapulmonary metastases
	Lewis Lung Carcinoma cells expressing α5 shRNA injected into tail vein83	Fewer lung tumors

The in vivo models used in the summarized studies can be divided into three categories: (1) tissue-specific neo-oncogenesis that recapitulates carcinoma progression seen in human disease- and orthotopic injection: both have the advantage of mimicking the correct tissue context of the human disease,- (2) sub-cutaneous injection at non-orthotopic sites- and (3) intravenous injection.,- In the case of the first two categories successful metastasis requires the tumor cells to invade locally, intravasate into blood or lymph vessels, extravasate at secondary sites and form new tumors. Conversely, intravenous injection does not depend on the formation of a primary tumor, thus allowing separation of the effects of primary tumor formation from dissemination into secondary tissues. This approach is frequently used when the molecule or pathway of interest inhibits initial tumor formation. A limitation is that this does not address initial migration and invasion away from primary tumor sites. A further important caveat to all of these models is that none measure migration and invasion alone. The ability of the tumor cells to disseminate and form secondary tumors is a multifactorial process, which additionally requires survival in the circulation, adhesion to vessel endothelial layers and transmigration through the endothelium. Below we compare data for 3D migration speed and persistence with in vivo metastasis results in response to altered adhesion molecule expression and �activities.

FAK, p130Cas, talin and �vinculin

To date, a large number of studies have focused on the role of FAK for in vivo metastasis. FAK has long been known to be upregulated in a wide variety of invasive cancer types (reviewed in ref. 86). Cancer cells in which FAK is either homozygously deleted or downregulated via expression of a dominant negative form of FAK universally display reduced metastatic tumor formation (Table 2), matching the findings that FAK depletion reduces 3D cell migration speed (Table 1). FAK depletion inhibited metastatic tumor formation in all three categories of in vivo invasion models (see above). Congruent results were also seen between 3D culture migration speed and in vivo metastases for p130Cas and talin (Tables 1 and 2). However, the data are not so clear for vinculin. The depletion or absence of vinculin had no effect or led to reduced migration speed in 3D models. In contrast, a high level of exogenous vinculin expression reduced lung tumor formation in �vivo.

NEDD9

NEDD9 overexpression promotes invasion and metastasis in a range of cancer cell types. In apparent concordance with this, NEDD9 depletion reduces fibroblast migration in 3D collagen gels (Table 1). However, while NEDD9−/− tumor cells in a model of oncogenic transformation of mammary epithelia had a tendency to form fewer metastatic tumors, tail vein injection of these isolated cells revealed aggressive lung tumor formation. Potentially, this may represent a switch to the amoeboid/cell deforming invasion mode that occurs following NEDD9 depletion, which may promote lung colonization. By contrast, the dense, rigid matrix that is reported to characteristically surround breast tumors, which is presumably present in the transformed breast epithelia model, may obviate the rounded/cell deforming mode of invasion. Alternatively, increased metastatic lesions following tail vein injection may reflect changes in survival in circulation, adhesion and transmigration of the endothelial cell layer of the �vessel.

β1 �integrin

The effects of β1 integrin depletion on migration in matrigel are dependent on the matrigel concentration; at high concentrations when the gel is less deformable, the loss of β1 impaired the ability of the cells to invade. These results are in agreement with in vivo data where β1 depletion also caused reduced metastatic lesions (Table 2). In one of the studies, increased numbers of tumor emboli in the lymphatic vasculature were noted. This suggests that the cells in this model were competent to locally invade and intravasate, but were not able to execute the final stages required for secondary tumor �formation.

α5β1 and αvβ3 integrin �receptors

The finding that high level α5β1 is associated with enhanced migration speed and persistence of breast cancer cells in 3D collagen gels appears to agree with reports that α5 depletion reduced lung colonization in a tail vein injection model. However, when α5 and β1 are analyzed separately, the findings are conflicting. Exogenous α5 expression was shown to significantly reduce lung and extrapulmonary metastases arising from injected colon cancer cells. Upregulation of β3 expression increased bone and lung metastases,, yet decreased lung tumor formation in an invasive melanoma model. Fibroblasts lacking β3 integrin expression have significantly faster migration speeds and reduced migration persistence in a 3D CDM assay. It is difficult to estimate to what degree the usage of different cell types has contributed to the disparate findings. For instance, the differential expression of other integrin receptors in different cell types may have influenced the invasion and metastasis behavior. In particular, it has been suggested that cross-talk between recycling α5β1 and αVβ3 receptors may be an important mechanism of regulating migration persistence. These data highlight the need for direct comparison between 3D culture models and in vivo metastasis models for �validation.

Conclusion

Our comparison of the effects of a handful of studies that have investigated adhesion proteins in 2D vs. 3D culture models with in vivo measurements of metastasis has revealed a complex picture. For instance, it is notable that without exception, the depletion of the adhesion molecules summarized in Table 1 resulted in decreased migration persistence in 3D models but not always in reduced metastasis formation in vivo (Table 2). To accurately predict in vivo invasion behavior, 3D model systems are required that faithfully replicate the physical characteristics of the environments encountered by transmigrating cancer cells. In particular it will be critical to develop model systems mimicking the interstitial tissue organization both at the primary tumor site and at the metastatic niches that characterize different tumors. Such tailored 3D models to map the journey taken by metastasizing cancer cells are still lacking for interactions with the vasculature, host immune responses and other cell-cell interactions, and can generally mimic only short-term (< 1 week) conditions. But even with these limitations, 3D models are essential for gaining a basic understanding of cell migration and invasion through tissue and will ultimately lead to novel cancer therapies. Invasion away from the primary tumor site fundamentally underpins metastatic progression, and treatments that target invasion therefore hold the promise of significantly improving survival rates for most �cancers.