Initial steps of metastasis: cell invasion and endothelial transmigration.

Metastasis is the leading cause of cancer mortality. The metastatic cascade represents a multi-step process which includes local tumor cell invasion, entry into the vasculature followed by the exit of carcinoma cells from the circulation and colonization at the distal sites. At the earliest stage of successful cancer cell dissemination, the primary cancer adapts the secondary site of tumor colonization involving the tumor-stroma crosstalk. The migration and plasticity of cancer cells as well as the surrounding environment such as stromal and endothelial cells are mandatory. Consequently, the mechanisms of cell movement are of utmost relevance for targeted intervention of which three different types have been reported. Tumor cells can migrate either collectively, in a mesenchymal or in an amoeboid type of movement and intravasate the blood or lymph vasculature. Intravasation by the interaction of tumor cells with the vascular endothelium is mechanistically poorly understood. Changes in the epithelial plasticity enable carcinoma cells to switch between these types of motility. The types of migration may change depending on the intervention thereby increasing the velocity and aggressiveness of invading cancer cells. Interference with collective or mesenchymal cell invasion by targeting integrin expression or metalloproteinase activity, respectively, resulted in an amoeboid cell phenotype as the ultimate exit strategy of cancer cells. There are little mechanistic details reported in vivo showing that the amoeboid behavior can be either reversed or efficiently inhibited. Future concepts of metastasis intervention must simultaneously address the collective, mesenchymal and amoeboid mechanisms of cell invasion in order to advance in anti-metastatic strategies as these different types of movement can coexist and cooperate. Beyond the targeting of cell movements, the adhesion of cancer cells to the stroma in heterotypic circulating tumor cell emboli is of paramount relevance for anti-metastatic therapy.

Introduction

Metastasis is a multi-step process encompassing the (i) local infiltration of tumor cells into the adjacent tissue, (ii) transendothelial migration of cancer cells into vessels known as intravasation, (iii) survival in the circulatory system, (iv) extravasation and (v) subsequent proliferation in competent organs leading to colonization [1]. Metastasis is known as a very inefficient process [2,3], because a coordinated choreography of the multiple events is required to prevent failure of the complex process that otherwise runs into the elimination of emigrating cancer cells at any of the many steps on the way [2]. Although the dissemination from a primary tumor of 1 cm size (roughly corresponding to 1 × 109 cancer cells) can infiltrate the circulatory system with one million cancer cells per day [4], subsequent colonization is very limited due to incompatible distal sites. Consequently, <0.1% of disseminated cancer cells successfully develop distal metastasis [5,6]. Albeit, dormant solitary tumor cells or micrometastases, which represent undetectable cancer cell populations due to either a cell cycle arrest or a balance between proliferation and apoptosis [6], might eventually outgrow to clinically detectable macrometastases many years post anti-cancer treatment [6-8].

The discussion whether metastasis is an early or late event in tumor progression is ongoing and still remains an open issue [9]. On the one hand, the “linear progression model” suggests cancer cell dissemination after extensive expansion of primary tumors, whereas on the other hand, the “parallel progression model” claims early tumor cell dissemination of small tumors with 1–4 mm in size. In line with linear progression, an increased mutation frequency was shown for the mutation of the tumor suppressor TP53 at advanced stages (T3) of breast cancer as compared to smaller tumors (T1 stage) [10]. In contrast, the parallel progression model accounts for distinct genetic alterations of tumor cells at primary and distal sites due to early separation and independent development. In fact, early tumor cell dissemination and formation of micrometastasis in the bone marrow and the lung were shown for ERBB2 (HER2/neu) mutant breast cancer cells prior to the morphological invasion of the primary tumor [11]. Furthermore, amplification of ERBB2 was observed in disseminated esophageal cancer cells and independent of the ERBB2 status in the primary tumors [12,13]. In this line, numerous studies indicate that disseminated tumor cells (DTCs) can localize in lymph nodes or in the bone marrow prior to the establishment of metastases [9,13]. The early spreading of DTCs to distant sites as described by the parallel progression model allows to develop concepts of cancer diagnosis and therapy by monitoring and molecularly characterizing DTCs that could not be attributed in linear cancer progression. Moreover, the dormancy of DTCs is of particular relevance for tumor progression and manifestation of metastases [14]. CD8+ T cells were recently found to provide cytostatic rather than cytotoxic effects to DTCs which keep them in a dormant state, whereas reduction of immunosurveillance by depletion of CD8+ T cells caused faster metastatic colonization [15]. This immune-mediated dormancy of DTCs by the inhibition of cell cycle progression could be the reason for the comparable genetic signature of early DTCs and established macrometastases at late stage of tumor progression [16]. Interestingly, various studies revealed that the dormancy of DTCs required interferon (IFN)-γ and tumor necrosis factor (TNF)-mediated signaling which was not correlated with the death of tumor cells [16]. Therefore, immunotherapies employing IFNs are suggested to reduce the metastatic burden.

An important issue is whether tumor cells prefer distinct organs for distant metastatic colonization. Stephen Paget described an organ-specific pattern of metastasis already in 1889 [17]. Employing his “seed and soil” theory, he proposed that secondary growth of cancer cells (the “seed”) is dependent on the competence of the distal organ (the “soil”). Up to now, this theory has been approved as distinct cancer types metastasize at different and tumor-specific sites [6,18]. For example, the major sites of metastasis of breast carcinoma are the bone, the lung and the brain [2]. Colorectal and pancreatic carcinoma show preferential sites of distal colonization in the liver and the lung. However, some cancer cells are highly invasive but rarely seed other organs. For instance, hepatocellular carcinoma (HCC) and gliomas most frequently show proximal metastasis leading to intrahepatic and intracerebral colonization, respectively [19,20]. While the dense hepatic vasculature mainly favors the intrahepatic metastasis of HCC, physical barriers and systemic hurdles avoid colonization of gliomas because the patients mostly die prior to the establishment of metastasis at distal sites [21]. Besides the influence of the primary tumor on site-specific metastasis, a plethora of biological, chemical and physiological constraints are governing metastasis [22]. Extrinsic chemical barriers are represented by e.g. pH, reactive oxygen species or hypoxia. The cellular microenvironment regulates tumor dissemination by regulatory cytokine and chemokine feedback loops and by the secretion of matrix metalloproteinases (MMPs). Physiological constraints include intratumoral tensional forces, the composition of the basement membrane and the anatomy of capillary walls. Fenestrated bone marrow sinusoids are rather easy to penetrate in comparison to the lung capillaries or the blood brain barrier [18]. Thus, the anatomical differences and the density of lymphatic or blood vessels have a huge impact on cancer cell spreading at distal sites of metastasis [2,23]. However, the metastatic process can be ambiguous and unpredictable as the individual genetic disposition and constitution of the patient play a pivotal role.

Prevention of cancer cell dissemination and secondary tumor formation must be a major goal of cancer therapy as about 90% of all cancer deaths are related to metastasis [24]. In this review, we focus on the initial steps of the metastatic process which is invasion of cancer cells and their entry into the circulatory system by endothelial transmigration. Elucidation of the cellular and molecular mechanisms underlying these critical steps in tumor progression will facilitate new strategies to combat metastasis.

Mechanisms of cell invasion

Cancer cells invade other tissues either by moving collectively as epithelial sheets or detached clusters, or as single cells via mesenchymal or amoeboid cell types [25]. During cancer progression, a variety of tumor cells show changes in their plasticity by morphological and phenotypical conversions (Fig. 1), including the epithelial to mesenchymal transition (EMT) [26], the collective to amoeboid transition (CAT) [27] and the mesenchymal to amoeboid transition (MAT) [28].

Fig. 1

Plasticity of invading cancer cells. (A) In collectively invading cell strands or sheets, the tumor cell at the invasive front is designated as tip cell and executes the driving force by promoting focal adhesions and ECM rearrangement. Cells following the tip cells maintain their epithelial characteristics such as desmosomes, adherence, tight and gap junctions. (B) A single cancer cell can detach from the epithelial cluster by undergoing an EMT through e.g. TGF-β induced downregulation of E-cadherin and rearrangement of the cytoskeleton. This mesenchymal-like cell is able to move freely by forming focal adhesions and rearranging the ECM by secretion of proteases. EMT-transformed cells sustain a rear–front cell polarity. (C) The fastest way of cell invasion is the amoeboid cell movement, which leads to a total loss of cell polarity and paracrine directed chemotaxis in a protease-independent fashion. Cancer cells were shown to acquire these characteristics in two ways: either by inhibition of β1-integrin leading to CAT or by inhibition of proteases which subsequently induce MAT. CAT, collective to amoeboid transition; ECM, extracellular matrix; EMT, epithelial to mesenchymal transition; MAT, mesenchymal to amoeboid transition; MMP, matrix metalloproteinase.

EMT has been increasingly recognized as a crucial event in cancer progression and metastasis during the last decade [26,29]. EMT represents a highly conserved process which occurs during embryogenesis, chronic inflammation and fibrosis as well as upon cancer progression, referring to type I, type II and type III EMTs, respectively [30]. Type III EMT denotes a hallmark in the malignant progression to aggressive carcinoma [26,30,31], and has been described in the development of breast tumor [32], colorectal cancer [33], HCC [34,35], lung cancer [36], prostate carcinoma [37] and pancreatic cancer cells [38] (Table 1). Molecularly, EMT is characterized by loss of epithelial characteristics and the concomitant gain of a mesenchymal gene expression program [30]. One hallmark of EMT is the downregulation or even loss of epithelial (E-)cadherin, which is an essential component of adherence junctions. E-cadherin binds with its extracellular domain to an E-cadherin molecule of the neighbouring epithelial cell which stabilizes cell-to-cell contacts. Intracellularly, E-cadherin binds to β-catenin, α-catenin and p120-catenin which mediates intracellular signaling and links adherence junctions to the actin cytoskeleton [39]. Downregulation of E-cadherin by the transcriptional repressors Snail/SNAI1, Slug/SNAI2, SIP1/ZEB2 or Twist leads to the disassembly of adherence junctions and translocation of membrane-bound β-catenin to the cell nucleus where it modulates transcription of numerous genes such as c-myc or cyclin D1 [26,40-42]. The subsequent upregulation of mesenchymal markers such as vimentin and neuronal (N-)cadherin is one of the hallmarks for mesenchymal cells [43,44]. Expression of N-cadherin leads to the rearrangement of the cytoskeleton by mediating Rho-induced stress fibers and the formation of lamellopodia and filopodia by Rac1 and Cdc42 activation, respectively [45]. This change from E- to N-cadherin expression, termed cadherin-switch, leads to enhanced motility of EMT-transformed cells [46]. Thus, cells that have undergone EMT lose their epithelial organization and gain the ability to detach from epithelial cell clusters in order to move as single cells in a mesenchymal fashion (Fig. 1A and B). Numerous factors have been identified to induce EMT including integrins, hepatocyte growth factor (HGF), fibroblast growth factor (FGF), vascular endothelial growth factor (VEGF) and platelet derived growth factor (PDGF) [26]. Transforming growth factor (TGF)-β was shown to play a central role in tumor progression [47,48], leading to Smad activation and the subsequent upregulation of transcriptional repressors of E-cadherin [49,50]. Recently, novel inducers of EMT in hepatocarcinogenesis have been described such as chromodomain helicase/ATPase DNA binding protein 1-like gene [51] and connective tissue growth factor [52].

Table 1

Cell invasion in cancers.

Type of cell invasion	Cancer type	Reference
Collective	Oral squamous cell carcinoma	[82,180]
	Colorectal carcinoma	[153,181]
	Melanoma	[27]
	Breast cancer	[63,86,180]
	Endometrial carcinoma	[74]
	Pancreatic cancer	[182]
Single
Mesenchymal	Fibrosarcoma	[28,91]
	Glioblastoma	[25,183]
	Melanoma	[68,184]
Amoeboid	Breast cancer	[65]
	Lymphoma	[25]
	Small-cell lung carcinoma	[185]
	Prostate carcinoma	[25]
	Melanoma	[68,184,186]
	Sarcoma	[73]
EMT	Hepatocellular cancer	[31,122,187]
	Breast cancer	[32,188,189]
	Colorectal cancer	[33,60]
	Lung cancer	[36,190]
	Prostate cancer	[37,191,192]
	Pancreatic cancer	[38,193]
MAT	Fibrosarcoma	[28,91,183]
	Breast carcinoma	[28,194]
	Melanoma	[69,72]
CAT	Melanoma	[27]

EMT, epithelial to mesenchymal transition; MAT, mesenchymal to amoeboid transition; CAT, collective to amoeboid transition.

EMT is not only associated with an increase in cell motility but also described as a trait to maintain stem cell properties, to prevent apoptosis and senescence, to suppress immune reactions and to acquire resistance against radiotherapy and chemotherapy [26]. The concomitant occurrence of stem cell characteristics and EMT features was suggested in colorectal cancer cells at the tumor–host border [53] and demonstrated in breast cancer [54]. Notably, therapeutic compounds have been identified which specifically target EMT-transformed cells that harbor cancer stem cell characteristics [55,56]. Recently, numerous microRNAs (miRs) have been demonstrated to modulate EMT. miR-9 was shown to directly interact with CDH1 (E-cadherin), which leads on the one hand to increased cell motility and invasiveness, and on the other hand to elevated VEGF levels resulting in enhanced angiogenesis [57]. Similarly, amplified miR-151 targeting RhoGDIA correlated with aggravated intrahepatic metastasis of HCC by its synergistic function with the focal adhesion kinase (FAK) and the subsequent activation of Rac1, Cdc42 and Rho [58]. In contrast, the miR-200 family including miR-200a, miR-200b, miR-200c, miR-429 and miR-141 leads to epithelial differentiation. The zinc finger enhanced binding (ZEB1 and ZEB2) transcription factors inhibit the miR-200 family at the transcriptional level, whereas these miRNAs themselves post-transcriptionally repress the EMT-inducers ZEB1 and ZEB2 [59]. This negative feedback loop between ZEBs and miR-200 is not only proposed to be the molecular driver of cellular plasticity but additionally affects the balance between stemness and differentiation, longevity and senescence, cell cycle arrest and proliferation and between survival and apoptosis. It has been suggested to occur in various human carcinomas including pancreatic and colorectal cancers. Interestingly, ZEB initiated EMT and stem cell properties by upregulation of Klf4, Sox2 and Bmi1 which are normally repressed by the miR-200 family [60].

The reversal of EMT, referred to as mesenchymal to epithelial transition (MET), is characterized by epithelial reorganization and is also described as individual to collective transition [26,50]. This process allows mesenchymal cells to regain the epithelial cell-to-cell junctions for colonization at secondary sites [30,61]. In this context, EMT and MET have been discussed as frequent events among many others in the metastatic cascade of epithelial cancers, where carcinoma cells initially start to invade collectively, followed by incomplete or complete EMT, both of which are reversible by MET for effective metastatic colonization at distal sites [62].

CAT known as collective to amoeboid transition has been described as the detachment of individual cells from cell clusters by the use of amoeboid migration, particularly in melanoma (Fig. 1A and C; Table 1) [27,63]. Amoeboid cells show reduced cell–ECM interaction which allows them to squeeze through gaps in the ECM barriers independently of ECM proteolysis [64]. Intravital imaging by multiphoton microscopy revealed that single breast carcinoma cells with amoeboid morphology migrated along ECM fibers with high velocity [65]. CAT most frequently occurs after inhibition with β1 integrin in multicellular clusters of melanoma cells that induces single cell movement in an amoeboid fashion [27]. The escape of single cells from the collective cell organization is associated with the disassembly of E-cadherin-dependent cell adhesion complexes and integrin-linked focal sites [66]. It is currently an open issue whether CAT occurs directly or in an indirect fashion by employing an intermediate mesenchymal cell phenotype.

MAT, referred to as the transition of mesenchymal cells to amoeboid cells, was detected in breast cancer, melanoma and fibrosarcoma (Table 1) and depends on Rac and Rho/ROCK signaling and is independent on protease activities such as MMPs, serine proteases and cathepsins [25,67,68]. Indeed, blocking of extracellular proteolysis was shown to induce MAT (Fig. 1B and C). Besides regulation by the tumor microenvironment, MAT can be induced by regulatory proteins such as EphA2 kinase. Re-expression of EphA2 in mesenchymal melanoma cells triggers changes in the plasticity to an amoeboid motility that is accompanied by the activation of a nonproteolytic invasive program along with a Rho-mediated cell rounding [69,70]. Furthermore, both the tumor-suppressor proteins p27kip1 and p53 are involved in controlling the MAT of fibroblasts and melanoma cells in 3-dimensional matrices [71,72]. Loss of these tumor suppressors therefore enhances the aggressiveness of these cancer cells by switching their mode of invasion. In addition, overexpression of the microtubule destabilizing protein stathmin has been described to induce MAT of sarcoma cells [73].

Collective cell invasion

The bulk invasion of cancer cells occurs in epithelial cancers such as breast, endometrial and colorectal cancers and in melanoma (Table 1). Three hallmarks characterize the collective cell invasion [74]. First, cell–cell junctions remain intact during movement [75]. Second, the multicellular coordination of polarity and cytoskeletal activity generates the traction force required for collective cell movement [27]. Third, collective cell invasion involves the remodeling of extracellular matrix (ECM) and rearrangement of the basement membrane (Fig. 1A) [63]. The types of collective cell invasion are manifold including formation of a monolayer which allows invading two dimensionally or building up cell strands for three dimensional invasion or even the detachment of a collectively invading cell cluster from the initial tumor. For instance, luminal epithelial cells exhibit reduced polarity by the loss of myoepithelial cells in mammary tumors which causes their collective cell migration and the transition from in situ to invasive breast carcinoma [76]. Another example includes the formation of leading endothelial tip cells showing filopodia during sprouting angiogenesis that depends on VEGF-induced Delta-like 4 expression, whereas trailing endothelial stalk cells generate the base of the emerging sprout showing activated Notch signaling and decreased VEGF receptor II (VEGF-R2) and VEGF-R3 levels [77].

Collectively invading cells maintain their cell-to-cell contacts such as adherence, tight and gap junctions as well as desmosomes (Fig. 1A) [74,75]. However, leader or tip cells in the front and the following cells at the back create a sort of front–rear asymmetry. Rho GTPases and myosin II mediate asymmetric stiffening of cortical actomyosin filaments that affect tip cells. Differences in gene expression lead to distinct cell morphologies between tip and following cells. Whereas the tip cell exhibits a rather mesenchymal phenotype, the following cells are organized with intact epithelial cell-to-cell contacts (Fig. 1A). Signals at the leading edge include chemokines and growth factors secreted in either paracrine, juxtacrine or autocrine fashion which affects the asymmetry of collectively invading cells. Collective migration can be induced by the paracrine secretion of stromal-cell derived factor (SDF1/CXCL12) or members of the FGF and the TGF-β family [74,78]. For example, the induction of the Wnt/β-catenin pathway by members of the FGF family results in the differential expression of membrane-bound CXCR4 at the front and CXCR7 at the rear of the migrating lateral line primordium in zebra fish which maintains front–rear-asymmetry [79].

To mediate collective cell movement, force generation is necessary for pulling cells from the front or pushing them from the rear. The generation of traction force is provided by substrate binding integrins in leading cells (Fig. 1A). For this task, the leading edge expresses β1 and β3 integrins to mediate focal adhesion complexes in order to connect to ECM components such as fibronectin [74]. Furthermore, the tip cell produces α2β1 and αvβ3 integrins to attach to collagen and fibrin-rich surfaces, respectively. In two dimensional migrating cell clusters, following cells even form basolateral lamelopodia which are protruded underneath the cell in front [80]. Hereby, α6β1 integrins help in attaching to the basement membrane, which was produced by the invading tip cells. Integrin-mediated attachment to the ECM results in the activation of cytoskeletal adaptor proteins such as cortactin, vinculin, paxilin and talin. For example, β1 integrin can cooperate with discoid domain receptor 1 to activate FAK and prolin rich tyrosin kinase (PYK)2/FAK2 [81], which subsequently leads to upregulation of N-cadherin and thus to an increased cell motility [27]. Similarly, integrins α3 and α5 in fibroblasts were described to be important for force generation and formation of tube-like tracks through which squamous cell carcinoma cells follow collectively [82]. The rearrangement of actin is crucial for both, the formation of the filopodia and the maintenance of VASP/Mena-dependent cell–cell adhesion during collective migration [83,84]. Pseudopodia and filopodia are controlled by Rac and Cdc42 [85], respectively, whereas Rho signaling is more important for single than collective cell movement [86].

Remodeling of the extracellular matrix (ECM) is essential in initiating track formation and depends on (i) the invading cell itself and (ii) the ECM determinants including its dimension, density and gap size, its stiffness and orientation [64]. Tip cells were shown to secrete MT1-MMP (membrane type 1 matrix metalloproteinase; MMP14) to initiate track formation [63,74,87,88] and following cells enlarge this track by further degradation of the ECM and subsequent deposition of laminins, nidogen 1, perlecan and type IV collagen (Fig. 1A). Interestingly, MT1-MMP represents the rate limiting factor of multi-cellular invasion [89].

Mesenchymal cell invasion

Mesenchymal cells employ a five-step migration cycle including pseudopod protrusion, formation of focal contacts, focalized proteolysis, actomyosin contraction, and finally, the detachment of the trailing edge [25]. Invasion of single mesenchymal cells was detected in fibrosarcoma, glioblastoma and melanoma (Table 1). In carcinoma, mesenchymal cells mostly arise from epithelial cell clusters through EMT (Fig. 1A and B; Table 1) [74]. The rearrangement of F-actin leads to partially polarized cancer cells. Focal adhesions in the front tightly attach the cell to the ECM, whereas the tail moves by contractions of the retraction fibers (Fig. 1B) [65].

Transient TGF-β signaling was shown to be responsible for the dedifferentiation of cancer cells and the following detachment of single cells from collectively moving cell strands [86]. Cells of different motilities are heterogeneously distributed within the tumor [65,90], and actually mobile ones only account for 5% [61]. After discrimination between collective clusters and single moving cells, time-lapse studies revealed that single mesenchymal cells show an activated TGF-β signaling as indicated by nuclear accumulation of Smad2 [86]. In contrast, collectively moving cells or non-motile cells retained Smad2 in the cytoplasm. Moreover, disseminated cell clusters into lymph nodes and large pulmonary metastatic colonies were devoid of nuclear Smad2 accumulation. Interestingly, intravasation of single mesenchymal cells into blood vessels was dependent on TGF-β/Smad2 signaling as interference with TGF-β type II receptor impeded hematogenous metastasis. However, collective invasion was not altered and cancer cell clusters could be still detected in the lymph system [86]. This study demonstrated that (i) TGF-β is essential to switch cancer cells from a cohesive to a single cell motility by activating Smad4, epidermal growth factor (EGF) receptor, Nedd4, M-RIP, FARP and RhoC and (ii) the subsequent downregulation of TGF-β at secondary sites is inevitable for distal metastasis.

Amoeboid cell invasion

The hallmarks of amoeboid cell invasion are the loose attachment to the ECM, complete loss of cell polarity and the ability of chemotaxis (Fig. 1C) [65]. Invasion of single amoeboid cells occurs in breast cancer, lymphoma, small-cell lung and prostate carcinomas as well as in melanoma and sarcoma (Table 1). In contrast to mesenchymal cells, amoeboid cancer cells are roundish, grow in suspension, show no formation of stress fibers, do not remodel the ECM and are devoid of focalized integrins [25]. The amoeboid cell invasion is described as the fastest migratory phenotype by reaching a velocity of up to 20 μm/min as compared to the one of the mesenchymal cells showing a movement of 0.1–1 μm/min [25]. Whereas mesenchymal and collective cell invasion depend on protease activities, amoeboid-like motility is protease-independent as cells rather employ actomyosin-based mechanical forces to physically displace matrix fibrils instead of degrading them [89]. In particular, protease-independent amoeboid cell invasion is suggested to occur when the structural pores formed in collagen networks fail to show stabilizing covalent crosslinks that determine ECM stiffness. Structural constraints limit the mobility of amoeboid cancer cells as collagen pores smaller than the size of the invading cell's nucleus cannot be negotiated [89]. Naturally crosslinked collagen in vitro and in vivo even requires MT1-MMP (MMP14) dependent proteolytic activity for amoeboid cell movement [44,89]. This provides evidence that the different types of movements are not mutually exclusive but can be entertained simultaneously and cooperatively.

Amoeboid cells most frequently develop after the treatment of cancer with integrin-blocking antibodies or with protease inhibitors. For example, blocking of β1 integrin in cancer cell clusters resulted in loss of cell–cell adhesion, cell detachment and transition to an amoeboid single-cell type of melanoma and fibrosarcoma cells [27,91]. Similarly, MMP inhibitors failed to prevent cancer progression in clinical trials showing the physiological relevance of amoeboid cell movement in vivo [89]. Yet, most of the studies dealing with amoeboid cell invasion have been performed in 3-dimensional ECM settings in vitro that do not recapitulate the native collagen composition in vivo. Thus, the protease-independent type of cancer cell movement is a current matter of debate.

Endothelial transmigration

The blood system has been considered as the main route of the metastatic spread, but there is increasing evidence that the lymphatic system might be the key player in cancer cell dissemination [92,93]. The lymphatic capillaries are thin-walled single layers of endothelial cells lacking inter-endothelial tight junctions (Fig. 2A). In contrast to blood capillaries, they are not covered by smooth muscle cells and lack a basement membrane [94]. Interstitial fluids, macromolecules and even bacteria can easily enter the lymphatic drain through valve-like openings. Anchoring filaments are connected to the lymphatic capillaries to prevent vessel collapse in condition of high lymph flow. Whereas VEGF-A and the activation of VEGF-R1 and -R2 are widely known to induce blood vessel angiogenesis, signaling through VEGF receptor III after activation by VEGF-C and VEGF-D is necessary for lymphangiogenesis (Fig. 2A) [95]. Tumoral expression of VEGF-C and VEGF-D was not only correlated with lymphatic cell invasion [96], distant metastasis and poor survival [97,98], but also additionally thought to actively induce vessel sprouting towards the tumor [99,100]. Moreover, tumor cells might be even attracted by lymphatic vessels through secretion of chemokines such as CCL21 (Fig. 2A) [94,101]. It is still an open discussion whether tumor cells are actually passively trapped in the lymph nodes or whether lymph nodes promote spreading of cancer cells and their distal metastasis [23]. Interestingly, a most recent study demonstrated that aggregates of breast cancer cells directly penetrate lymphatic vessels through ruptures in the vascular wall [102]. This process is mediated by the induction of the hypoxia-inducible enzyme ALOX12 or ALOX15 in mammary carcinoma cells which metabolizes arachidonic acid to 12(S)-hydroxy-eicosatetraenoic acid (12(S)-HETE). Interestingly, exposure of lymphatic endothelial cells to 12(S)-HETE transiently reduces their VE-cadherin expression, resulting in the migration of these cells and subsequent formation of circular defects in the integrity of the lymphatic endothelial cell layer [102].

Fig. 2

The tumor microenvironment and its impact on transendothelial migration. (A) Passive migration of epithelial cancer cells into lymph vessels may occur through intercellular gaps and active attraction of lymph endothelial cells upon secretion of VEGF-C, VEGF-D and CCL21. (B) (Transient) TGF-β induced EMT and the secretion of VEGF-A mediates hematogeneous dissemination. (C) Once transmigrated, tumor cells can use platelets as a shield against shear forces and natural killer (NK) cell attacks. (D) Two prominent cell types of the tumor microenvironment are tumor associated macrophages (TAMs, green color) and cancer associated fibroblasts (CAFs, red color). Both TAMs and CAFs secrete components which stimulate cancer progression. (E) CAFs, also referred to as myofibroblasts, rearrange the ECM and are able to form a track through which epithelial tumor cells can migrate. (F) In close contact with tumor cells, macrophages are able to promote hematogenous transmigration. CCL, chemokine (C–C motif) ligand; CSF, colony stimulating factor; CXCL12 (SDF1), stromal cell derived factor; ECM, extracellular matrix; EGF, epidermal growth factor; MMP, matrix metalloproteinase; NF-κB, nuclear factor-kappa B; TGF, transforming growth factor; VEGF, vascular endothelial growth factor.

It is a matter of debate whether cancer cells actively migrate through blood and lymph vessels in response to e.g. growth factor gradients or even passively crawl into the vasculature without the involvement of an active cell migration machinery [103]. Such a passive mechanism implicates that cancer cells impinge on fragile tumor endothelial cells that do not exhibit intact cell-to-cell structures. While physiological blood vessel formation is formed in a balanced process between pro- and anti-angiogenic factors, tumor angiogenesis generates poorly organized and immature blood vessels that might allow to pass the shedding cancer cells [104]. Evidence for a passive mechanism is provided by the observation that shed tumor cells form cell clumps in the lumen of blood vessels due to intravascular proliferation of surviving cells [105]. One major obstacle for the successful intravasation of shedding cancer cells is provided by their ability to survive as most of these cells undergo cell death [106]. Yet, phenotypical changes of cancer cells such as the expression and secretion of growth receptors and its ligands strongly support the idea of active intravasation during metastasis. In this scenario, the microenvironment-controlled migration of breast carcinoma cells towards blood vessels has been reported to depend on the secretion of EGF by colony stimulating factor (CSF) receptor-expressing macrophages and the CSF1 release of EGF-R expressing mammary tumor cells [107,108] (see also Section 4). Thus, we might assume that both active and passive mechanisms exist that contribute to the entry of cancer cells into the vasculature [103].

The ability for hematogenous intravasation of metastatic and non-metastatic mammary carcinoma cells showed no correlation with the size of the primary tumor [90]. However, metastatic cells were oriented towards the blood vessels and showed protrusions (Fig. 2B), whereas non-metastatic cells rather underwent fragmentation when interacting with blood vessels [65]. Interestingly, TGF-β type II receptor/Smad4 activation has been shown to cause hematogenous dissemination of breast carcinoma (Fig. 2B) [86], although prolonged TGF-β activation hampered lung metastasis formation suggesting a transient TGF-β activation for successful metastasis. Interestingly, highest TGF-β activation was monitored at blood vessels and tumor margins including 30–50% of cancer cells. Disseminated cells found in the lymphatic vasculature were not dependent on TGF-β signaling and exclusively organized in clusters (Fig. 2A). An alternative model proposes metastasis being a cooperation between EMT cells and non-EMT cells [109]. It is hypothesized that hematogenous intravasation is only feasible in cooperation between those two cell types, whereas subsequent metastasis at distal sites is solely provided by non-EMT cells. The latter phenotype was also observed by extravasation of tumor cells from lymphatics [86].

Once disseminated, cancer cells have to survive the challenges of the blood stream including physical constraints and the immune system. Surface shielding by platelets was shown to protect cancer cells from (i) shear forces (Fig. 2C), (ii) NK-cell mediated lysis [110] and (iii) facilitate extravasation at the distal site [111]. Remarkably, cancer cells express membrane bound tissue factor (Tf) which is the receptor for coagulation factors VIIa and X, which is the main initiator of blood coagulation and tumor-associated microthrombi [112,113]. The Tf-VIIa complex is capable of stimulating proteinase activated receptor (PAR)2, leading to immune regulation, angiogenic stimulation and anti-apoptotic signaling in cancer cells by activating G-protein-coupled signaling [114]. Furthermore, PAR2 recruits the intracellular adaptor protein β-arrestin to mediate cancer cell migration. In clinics, high concentrations of platelets have been associated with decreased survival in a variety of cancers including breast, colorectal and lung cancer. Noteworthy, treatment with anti-coagulants was shown to reduce metastasis in patients [115].

Role of the tumor microenvironment

The microenvironment comprises various cell types including endothelial cells forming lymph and blood vessels, pericytes, stromal fibroblasts and bone marrow derived cells such as macrophages, neutrophils, mast cells and mesenchymal stem cells [107,116]. In general, the tumor microenvironment contributes to tumor progression by secretion of growth factors, cytokines and chemokines, and by the rearrangement of ECM (Fig. 2D).

TGF-β, HGF, FGF, EGF and insulin-like growth factor (IGF) are potent inducers of EMT [26,44,46,50,117]. These factors including interleukin (IL)-1α also play a role in the stimulation of collective cell invasion by the particular influence on the tip cell. TGF-β has an exceptional role in the tumor–stroma interaction. The tumor-promoting effects of TGF-β are numerous [118], and include the (i) induction of EMT [119], (ii) generation of myofibroblasts [120,121], (iii) production of autocrine mitogens such as PDGF [122] and (iv) evasion of immunity by targeting CD8+ T cells [123]. Even more, TGF-β in cooperation with HGF/c-met signaling can induce hypoxia-inducible factor (HIF)-1A driven VEGF-A secretion which stimulates angiogenesis. Furthermore, TGF-β affects distal metastasis of breast carcinoma cells by priming pulmonary metastasis through upregulation of angiopoietin-like protein (ANGPL)4 [124]. Interestingly, ANGPL4 dissociates endothelial cell–cell junctions which allows to infiltrate ANGPL4-secreting tumor cells into the lung, thus facilitating pulmonary metastasis of breast cancer [18]. In the bone marrow, mammary carcinoma cells do not benefit from ANGPL4 for extravasation as vascular capillaries are fenestrated and TGF-β works via a different mechanism to cause organ-specific metastasis. Notably, TGF-β stored in the bone matrix stimulates the production of parathyroid hormone-related protein, IL-11, and connective tissue growth factor of circulating cancer cells [125]. These factors stimulate osteoblasts to release receptor activator of NF-κB ligand (RANKL) that mobilizes osteoclasts for inducing and maintaining an osteolytic cycle and outgrowth of bone macrometastasis. Recently, it has been proposed that an additional potential mechanism of switching TGF-β from anti- to pro-oncogenic functions is the attraction of host immune cells [126,127]. Interestingly, attraction of myeloid Gr-1+CD11b+ progenitor cells which express high levels of TGF-β1 and MMPs facilitate tumor invasion and cancer cell dissemination into the circulation [128].

Besides providing growth factors and cytokines, the tumor microenvironment promotes tumor progression by ECM rearrangement (Fig. 2D). MMPs are upregulated in almost every type of human cancer and were shown to associate with enhanced cell proliferation, migration, angiogenesis, metastasis and poor survival [129]. MMPs can be secreted from all cancer cells or tip cells of collective cell clusters, myofibroblasts and by almost all immune cells [129]. The functions of MMPs are (i) cleaving cell adhesion molecules such as E-cadherin, (ii) the degradation of ECM proteins, and (iii) the processing and activation of cytokines and growth factors [107]. Their functions are even more far-reaching as MMPs co-regulate inflammation and contribute to the generation of the metastatic niche [130].

Cancer associated fibroblasts (CAFs) represent essential cellular components of the tumor microenvironment. CAFs are able to secrete MMPs, cytokines (e.g. IL-8 and VEGF) and chemokines (e.g. CXCL12) and thus promote proliferation, cancer cell invasion, and neoangiogenesis (Fig. 2D) [118]. They were shown to communicate with cancer cells by a CXCL12/CXCR4 loop contributing to a controlled cancer cell migration [107]. Moreover, CAFs are able to form invading channels through which carcinoma cells can follow by maintaining their epithelial characteristics (Fig. 2E) [82].

Previous studies demonstrated that immune cells of the tumor microenvironment execute a tumor-promoting rather than a tumor-suppressing role [116,131-134]. It is well established that chronic infections and inflammation frequently lead to cancer development and enhancement of tumor progression [134,135]. For example, chronic infections with hepatitis B or hepatitis C virus are well described etiological factors for the generation of liver inflammation and HCC [20]. Thus, the impact of the inflammatory microenvironment has been proposed for inclusion as an additional hallmark in a renowned model of cancer [136,137]. Interestingly, NF-κB has been identified as a key activating signaling pathway in both cancer cells and tumor-associated immune cells [138,139]. Expression of cytokines such as IL-1, IL-6, TNF and RANKL induced by NF-κB has been shown to cause inflammation and to induce tumor cell dissemination [140,141]. In a multidrug resistance (Mdr)2 knock out model, TNF-α induced NF-κB activation is responsible for HCC development [142]. Interestingly, sex steroidal hormones such as estrogens can interfere with NF-κB activity and IL-6 production and thus can protect against liver cancer development in female mice [141,143,144].

Tumor associated macrophages (TAMs) produce growth factors and matrix-degrading enzymes and thus promote angiogenesis, cell invasion and hematogenous intravasation (Fig. 2D) [145-148]. The interaction between tumor cells and TAMs can be stimulated by EGF produced by macrophages and subsequent EGF receptor signaling in tumor cells which promotes migration and invasion. Cancer cells themselves secrete CSF1, which is a potent chemoattractant for CSF1 receptor-expressing macrophages [149,150]. Using this paracrine cytokine loop, TAMs were shown to directly assist hematogenous dissemination of mammary cancer cells [145], which is further supported by the secretion of proteases from both cancer cells and macrophages at the site of entry (Fig. 2F) [108]. Moreover, TAMs contribute to ECM rearrangement by MMP secretion [129] and the production of cysteine cathepsins [151] and serine proteases [152]. For example, the production of MMP2 and MMP9 of immature myeloid cells at the invasive front promotes collective invasion of colorectal cancer cells [153].

The impact of the tumor microenvironment is crucial for cancer progression and even goes beyond the location of the primary tumor. The preparation of the “soil” at the putative distal site of metastasis, referred to as pre-metastatic niche, has been shown to be under the control of the tumor microenvironment [154]. Importantly, metastasizing cancer cells bring their own soil such as activated fibroblasts, endothelial cells and macrophages from the primary tumor to the distant site i.e. the lung or the brain which protects traveling cancer cells from apoptosis [155]. Stromal cell passengers dramatically increase the metastatic colonization and interestingly, the metastatic capacity is lost when heterotypic cell clumps are dissociated. Studying the relevant adherence junction molecules is therefore of utmost importance.

Secreted growth factors from the primary tumor are suggested to affect bone marrow derived hematopoietic progenitor cells (BMDCs), which can form cellular clusters at the tumor-specific pre-metastatic site. Notably, VEGF-R1+VLA-4(integrin α4β1)+ BMDCs interacting with fibronectin-secreting fibroblasts were shown to play an essential role in the adherence and growth of disseminating cancer cells at the distal site [156]. Yet, the sufficiency of blocking VEGF-R1 in this context is a matter of considerable debate [157]. Although inhibition of VEGF-R1 was effective against metastasis formation, the effect on lung metastasis was lost after primary tumor resection [158]. Numerous studies revealed many more regulatory components which have an important role in forming the pre-metastatic niche. For example, tissue inhibitor of metalloproteinase 1 (TIMP1) was shown to promote colon cancer metastasis into the liver by inducing HGF signaling [159]. Prostate cancer is known to prepare a metastatic niche by the crosstalk with bone cells [160]. In particular, prostate cancer cells bind to cell surface expressed annexin II of osteoblasts in the bone marrow which induce expression of the growth arrest-specific (Gas6) receptor Axl in tumor cells [161]. Gas6 produced by osteoblasts induces metastastic tumor cell dormancy of prostate cancer cells and protects them from chemotherapy-induced apoptosis.

Concluding remarks and future perspectives

Local cancer cell invasion and endothelial transmigration represent the initial steps of metastasis that dramatically worsen prognosis and patient's survival after successful completion. Numerous attempts have tried to interfere with these early events and to eradicate metastasis at its initiation. Interference with VEGF-A using the anti-VEGF monoclonal antibody Bevacizumab (Avastin) aimed to hinder blood supply for the tumor and was found to be effective in clinical trials of many cancers such as non-small-cell lung and breast cancer [162]. This led to its approval by the FDA (Food and Drug Administration) for first-line treatment of metastatic colorectal cancer in combination with chemotherapy [163]. Similarly, interference with VEGF-R3 or blocking VEGF-C signaling reduced lymphatic tumor cell dissemination of breast carcinoma cells in mice [99,164].

VEGF-R2 is targeted in clinical trials to inhibit angiogenesis but indirectly also tumor cell intravasation into the vasculature. However, most of the clinical trials that intend to target angiogenesis or metastasis are combinations of a multitude of clinically used drugs or antibodies, which are not specifically designed against pro-metastatic targets. Interestingly, two clinical trials target semaphorin-4D (Clinical Trials.gov Identifier: NCT01313065; phase II) and neuropilin-1 (NCT00747734; phase I) with monoclonal antibodies. Semaphorins and their receptors that are expressed on endothelial cells trigger cell motility of cancer cells and of the microenvironment including the blood- and lymphendothelium. Blocking the contact between cancer cells and the microenvironment and the resulting signaling along this axis could provide specificity for anti-metastatic intervention. Semaphorins-3E and -3F are repellent molecules and could destroy the integrity of endothelial vascular walls and generate gates for invading tumor cells as observed with 12(S)-hydroxy-eicosatetraenoic acid that is secreted by tumor cells [102,165-168]. In addition, this could make the metabolizing enzymes i.e. ALOX12/15 and the migratory machinery involving cell adhesion molecules such as E-selectin to potential targets in clinical trials for combating early metastatic processes. The timing of this anti-metastatic therapy is of course crucial as it aims to block the very early steps of tumor intravasation into the vasculature. Therefore, it has a more preventive than curing character and attempts disease management. Of note, this intervention fails to work as soon as sentinel/post-sentinel lymph node colonization or even distant metastases have formed.

Alternatively, inhibition of integrins was considered as a promising attempt to block invasion. Unfortunately, these efforts resulted in the development of amoeboid cell types and thus caused an increase of invasion velocity and tumor progression rather than an inhibition of tumor dissemination [27,91,169]. Nevertheless, targeting adherence molecules, particularly those facilitating the hitch hike of tumor stroma cells such as activated fibroblasts that protect cancer cells from apoptosis and prepare a niche at the distant metastatic site [155,170], will be a very promising approach to control tumor spreading in the blood or lymph circulation. Another possible intervention at this stage seems to be the specific modulation of gene expression of passenger fibroblasts, as they apparently can express anti-metastatic gene products such as fibulin-5, which downregulates MMP9, and that itself becomes downregulated by cancer cell-secreted factors [171].

Furthermore, MMPs that are abundantly secreted by cancer cells and the tumor microenvironment were also targeted for intervention of tumor cell invasion and metastasis. However, the outcome of clinical trials was disappointing as the treatment with MMP inhibitors is particularly delicate and depends on the tumor stage [172]. For instance, the inhibition of MMP2 or MMP9 (NCT00001683, NCT00064142 and NCT00695851), even with monoclonal antibodies, does not seem to have the anticipated effect. The employment of MMP inhibitors resulted in the protease-independent mode of amoeboid cell invasion which squeezes through the ECM, showing that this intervention fails to block cell invasion [130]. Clinical studies employing MMP inhibitors are still in the process of recruiting patients (http://clinicaltrials.gov) and others are still running. The outcome of these trials has to be awaited. It seems that MMPs are too ubiquitous and hence unspecific targets, which may cause intolerable side effects particularly in combination with standard therapies [130]. Yet, the specific inhibition of MT1-MMP might be auspicious in targeting cancer cell invasion as this protease exhibits a dominant role in tumor cell dissemination [89]. Refinement of the specificity of MMP inhibitors for individual MMPs and the development of selective inhibitors of proteolytic and non-proteolytic MMP functions may provide further advance in the pharmacological interference with cancer progression and metastasis [130].

Another approach of novel anticancer treatment employs histone deacetylase (HDAC) inhibitors. A few are currently in development for clinical use (entinostat, belinostat, mocetinostat, panobinostat) and vorinostat (suberoylanilide hydroxamic acid, SAHA) is already FDA-approved for the treatment of relapsed cutaneous T-cell lymphoma [173]. SAHA changes the behavior of highly invasive MDA-MB231 breast cancer cells, which build up extensive actin-rich cell extensions [174] and inhibits brain metastasis of a derivative of this cell line in immuno-suppressed mice [175]. Also the invasion of tumor spheroids derived from the pleura effusion of patients with inflammatory breast cancer and locally advanced breast cancer was blocked by SAHA, which was accompanied by a relocation of E-cadherin from the membrane to the cytoplasm [176]. TNF-α and TGF-β treated esophageal carcinoma cell spheroids that were exposed to SAHA plus the proteasome inhibitor bortezomib decreased in vitro cell invasion due to restoration of E-cadherin expression and attenuation of TGF-β induced EMT [177]. Moreover, SAHA increases the expression of paxillin in human leukemic cell lines and facilitates their attachment to fibronectin [178]. In our hands, upregulation of paxillin and attachment of MCF-7 mammary tumor spheroids to endothelial cells correlated with the increased formation of gates into the lymphatic vasculature, termed circular chemorepellent-induced defects (CCID), which facilitates the transmigration of cancer cells. This is considered as a measure for potential tumor cell intravasation into the lymphatic vasculature ([168]; our unpublished data). SAHA is tested in 194 clinical trials (status April 2011) with 29 completed but with only a handful reporting data (NCT00486702, NCT00106626 and NCT00373490), and some were terminated for various reasons. The other trials are still running or recruiting.

Nowadays, researchers aim to cope with the complex influence of the immune system on tumor progression by developing novel anti-inflammatory drugs including antibodies, small RNAs and peptides. These strategies raise new hopes to design novel and effective anti-cancer therapies [179]. Intravital imaging to monitor disseminating cancer cells [65] and its combination with functional genomics is a promising attempt to elucidate pathways which enable the early steps of metastasis in vivo. There is still an urgent need for intense research of the initial steps of metastasis in order to investigate the underlying cellular and molecular mechanisms. In particular, the amoeboid type of cancer cell invasion, which seems to be an ultimate exit strategy of cancer cells, needs to be studied in more detail. Furthermore, future concepts of metastasis intervention must simultaneously address the collective, mesenchymal and amoeboid mechanisms of cell invasion. Undoubtedly, more insights into the complexity of the regulatory networks will allow the development of novel strategies to efficiently combat cancer cell dissemination.

Conflict of interest statement

None.