REVIEWS

TUMOUR-CELL INVASION AND

MIGRATION: DIVERSITY AND
ESCAPE MECHANISMS
Peter Friedl and Katarina Wolf
Cancer cells possess a broad spectrum of migration and invasion mechanisms. These include
both individual and collective cell-migration strategies. Cancer therapeutics that are designed to
target adhesion receptors or proteases have not proven to be effective in slowing tumour
progression in clinical trials — this might be due to the fact that cancer cells can modify their
migration mechanisms in response to different conditions. Learning more about the cellular and
molecular basis of these different migration/invasion programmes will help us to understand how
cancer cells disseminate and lead to new treatment strategies.

INVASION The ability of a cancer cell to undergo migration and Cell protrusions that initiate ECM recognition and
Penetration of tissue barriers, INVASION allow it to change position within the tissues. For binding can be quite diverse in morphology and dynam-
such as basement membrane example, these processes allow neoplastic cells to enter ics. These are termed LAMELLIPODA, FILOPODA, pseudopods or
and interstitial stroma, by cells.
Invasion requires adhesion,
lymphatic and blood vessels for dissemination into the invadopods5. Other cell extensions include RUFFLES (early
proteolysis of extracellular- circulation, and then undergo metastatic growth in dis- pseudopods) or spikes (early filopods within lamellae, as
matrix components and tant organs1. To spread within the tissues, tumour cells well as PODOSOMES. These different cell protrusions all con-
migration. It occurs during use migration mechanisms that are similar, if not identi- tain filamentous actin, as well as varying sets of structural
normal cell morphogenesis and
cal, to those that occur in normal, non-neoplastic cells and signalling proteins, and lead to dynamic interactions
wound healing, and also in
malignant cells. during physiological processes such as embryonic mor- with ECM substrates. Cell extensions are the prerequisite
phogenesis, would healing and immune-cell trafficking2. for the onset and maintenance of cell motility in normal
PSEUDOPOD The principles of cell migration were initially investi- and cancer cells, which form either spontaneously or
Cylindrical finger-like gated in non-neoplastic fibroblasts, keratinocytes and can be induced by chemokines and growth factors.
protrusions that protrude and
retract. Pseudopods are thicker
myoblasts3,4, but additional studies on tumour cells show Observation of cell extensions is therefore a useful way to
than filopodia. They are termed that the same basic strategies are retained. monitor the onset of cell motility.
‘invadopodium’ when To migrate, the cell body must modify its shape and
proteolytic matrix degradation stiffness to interact with the surrounding tissue struc- Molecular mechanisms of cell migration
is executed.
tures. Hereby, the extracellular matrix (ECM) provides The protein–protein interactions and signalling events
the substrate, as well as a barrier towards the advanc- that underlie shape change and regulate cell migration
ing cell body. Cell migration through tissues results are integrated in the concepts of focalized adhesion
from a continuous cycle of interdependent steps2–4. dynamics6 and actomyosin polymerization and contrac-
Department of
Dermatology, First, the moving cell becomes polarized and elongates. tion4,7. Initial propulsion and elongation of leading
University of Würzburg, A PSEUDOPOD is then formed, via extension of the cell’s pseudopods are driven by actin polymerization and
Josef-Schneider-Str. 2, 97080 leading edge, which attaches to the ECM substrate. assembly to filaments7,8 (BOX 1a), whereas little or no
Würzburg, Germany. Subsequently, regions of the leading edge or the entire adhesion (integrin binding to the ECM) or traction
Correspondence to P.F.
e-mail: peter.fr@mail. cell body contract, thereby generating traction force that are required at this step. Growing cell protrusions
uni-wuerzburg.de leads to the gradual forward gliding of the cell body and then touch the adjacent ECM and initiate binding via
doi:10.1038/nrc1075 its trailing edge. adhesion molecules, most notably transmembrane

362 | MAY 2003 | VOLUME 3 www.nature.com/reviews/cancer

© 2003 Nature Publishing Group
 REVIEWS

Summary advancing cell body gains volume towards the ECM

scaffold and is likely to provide the space required for
• The process of tumour-cell invasion and metastasis is conventionally understood as cell expansion and migration, leaving behind tube-like
the migration of individual cells that detach from the primary tumour, enter lymphatic matrix defects along the migration track19,20.
vessels or the bloodstream and seed in distant organs. Before and while focal contacts develop, ACTIN FILA-
• Novel imaging techniques (both in vitro and in vivo), together with re-evaluation of MENTS locally elongate and assemble, through the action
histopathological pattern formation in tumours, have provided a detailed view of of crosslinking proteins such as α-actinin, myosin II
cellular and molecular migration dynamics in cancer cells. and others7,21 (BOX 1d). Branched actin networks below
• Cancer cells disseminate from the primary tumour either as individual cells, using the inner leaflet of the plasma membrane are termed
21
amoeboid- or mesenchymal-type movement, or as cell sheets, strands and clusters CORTICAL ACTIN , whereas cytoplasmic bundles and
using collective migration. elongated cables of actin filaments are designated
21
• Cancer-cell migration is typically regulated by integrins, matrix-degrading enzymes, STRESS FIBRES . The contraction of actin filaments is

cell–cell adhesion molecules and cell–cell communication. provided by myosin II — the main motor protein in
• Cancer therapeutics designed to target adhesion receptors or proteases have not yet eukaryotic non-muscle cells (BOX 1d; arrows)22,23.
been show to be effective in clinical trials. This might be due to the fact that the cancer Stress-fibre assembly and contraction, which are con-
cell’s migration mechanisms can be reprogrammed, allowing it to maintain its invasive trolled by myosin II, are predominantly induced by
properties via morphological and functional de-differentiation. the small G-protein RHO and its important down-
• These adaptation responses include the epithelial–mesenchymal transition (EMT), stream effector, the RHO-associated serine/threonine
the mesenchymal–amoeboid transition (MAT) and the collective–amoeboid kinase (ROCK)22,24. By contrast, the cortical actin net-
transition (CAT). work is regulated by the myosin light-chain kinase
(MLCK), but not by RHO25–27. This allows the cell to
• Further studies are required to identify the factors that are involved in each type of cell
migration, as well as related escape strategies that are used by cancer cells after
separately control cortical actin dynamics from con-
pharmacotherapeutic intervention. tractions in inner regions. Actomyosin contraction
promotes the shortening of the cell’s length axis and
generates inward tension towards focal contacts that
are located at outward edges28,29. By several mecha-
receptors of the integrin family9 (BOX 1b). Integrins cou- nisms, which are incompletely understood, cell-sub-
ple to the actin cytoskeleton via adaptor proteins, then strate linkages then resolve preferentially in the back
become locally enriched, cluster and develop into an of the cell, whereas the leading edge remains attached
initial small FOCAL COMPLEX, which can grow and stabilize to the ECM and further elongates30,31 (BOX 1e).
within minutes to form a FOCAL CONTACT6,9–11. Focal Following focal contact disassembly, the trailing edge
contacts span approximately 20 nm between cell mem- — including the main cell body and nucleus — slowly
LAMELLIPOD brane and the substrate, vary in length (1–4 µm) and glide forward13,14,31.
A flat broad sheet of membrane width (0.5–2 µm), and cover an area of 1 to several µm2 A given cell type might preferentially use one partic-
and polymerized actin filaments (REFS 10,11). They are dynamic in assembly and composi- ular adhesion and migration mechanism, yet the num-
that flows forward at the front of
tion11,12; therefore, they stably adhere to or slowly glide ber and size of focal contacts can vary from cell to cell
moving cells on planar substrate.
It is a 2D variant of the along the substrate as the cell moves13,14. — and even within the same cell type — in response to
pseudopod. Depending on the cell type and ECM substrate, different environmental conditions. The speed gener-
focal contact assembly and migration can be regulated ated by the migration cycle is limited by the turnover
FILOPOD by different integrins. These include α5β1 integrin, rates of adhesion and de-adhesion events4, yielding an
A finger-like, relatively long-
lived dynamic protrusion of up
which binds fibronectin15; α6β1 or α6β4, which bind inverse relationship between focal contact strength and
to 50 µm length or more. It laminin16; αvβ3, which binds fibronetin or vit- migration rates. Stabilization of focal contacts increases
contains a core of actin filaments ronectin17; and α2β1, which binds fibrillar collagen18. attachment, reduces detachment and impairs migra-
that are bundled in parallel. It is Other, non-integrin receptors, such as CD44, discoidin tion rates, whereas weakening of adhesion strength, to a
most prominent in sprouting
receptors, CD26, immunoglobulin superfamily recep- certain degree, propels migration4,30,32.
axons, dendritic cells and some
cancer cells. Early forms are tors, and surface proteoglycans, also interact with Fully mature focal contacts have only been
termed ‘spikes’. ECM components and signal cell motility. The precise observed in cells that are firmly attached to 2D sub-
role of these factors in force generation, however, is not strates6. A total of 100 or more focal contacts can be
RUFFLE clearly established18. The engagement of integrins and formed by highly adhesive cells that express high levels
A small, short-lived dynamic
membrane protrusion that
other adhesion receptors leads to the recruitment of of integrins11,14,33. Conversely, when cells are placed in
forms at the cell’s leading edge. It surface proteases towards attachment sites, which, 3D substrates, clustered integrins tend to couple to
contains filamentous actin, and in turn, degrade ECM components that are in close less-completely assembled focal interactions and a pre-
retracts or contributes to proximity to the cell surface (BOX 1c). Soluble proteases dominantly cortical actin cytoskeleton, whereas stress-
pseudopod or lamellipod
can directly bind to integrins. For example, seprase (a fibre formation is rare15,29. Low-adhesion cells, such as
growth.
gelatinolytic enzyme) binds to α3β1 integrin158, lymphocytes and lymphoma cells, express low levels of
PODOSOME MMP1 (a collagenase) binds to α2β1 (REF. 160) and ECM-binding integrins. In these cells, migration is
Small, dot-like adhesive actin- MMP2 (a gelatinase) binds to αvβ3 (REF. 161). Similarly, sustained by cortical actin networks, and no focal con-
containing protrusions that form membrane-type matrix metalloproteinase-1 (MT1- tacts or stress fibres form34,35. So, similar to the mor-
along the lower cell axis towards
2D substrata. They are detected
MMP), an important collagenase, and MMP2 co-local- phological diversity of cell protrusions, a broad range
in osteoclasts, macrophages and ize with β1 or β3 integrins as they adhere to collagen of cytoskeletal structures can be formed in motile cells
some cancer cells. fibres58,162,163. ECM degradation occurs while the — from fully matured focal contacts to diffuse cortical

NATURE REVIEWS | C ANCER VOLUME 3 | MAY 2003 | 3 6 3

© 2003 Nature Publishing Group
REVIEWS

Box 1 | Cell migration in 3D ECM

A five-step model of cell migration in 2D was established over the past 30 years (reviewed in REF. 4).When
cells move in 3D, such as in tissues, an additional component is the proteolytic remodelling of the
extracellular matrix (ECM) (step c).
Step 1: protrusion of the leading edge. Growing actin filaments connect to adaptor proteins and push the
cell membrane in an outward direction.Actin polymerizes by coupling to the actin-nucleating ARP2/3
5 4 3 1,2
complex, which binds to the Wiscott–Aldrich syndrome protein (WASP), a multifunctional adaptor
1 Pseudopod protusion at the protein8. The ARP2/3/WASP complex connects to the inner leaflet of the plasma membrane via clustered
leading edge
phosphoinosites (PIPs) inserted therein8.ARP2/3 can further interact laterally with pre-existing actin
filaments and induce branching of pre-existing filaments to actin networks147. PIPs also bind and activate
RAC, RHO, guanine-nucleotide exchange factors (GEFs) that regulate the activity of the small GTPases RAC, CDC42
CDC42
and RHO, as well as RAS24. In one pathway, CDC42 binds WASP and PIPs, which activates WASP and
thereby induces filopodial ruffling and pseudopod extension8,148. Chemokines and growth-factor receptors
(CK-Rs, GFRs) activate phosphatidylinositol 3-kinase (PI3K) and phosphatidylinositol 4,5-kinase (PI45K),
CDC42 which are the key enzymes that generate PIPs149,150, and further engage RAC, CDC42 and RHO24,149,151.
Step 2: Cell-matrix interaction and formation of focal contacts. Integrins come into contact with ECM
ligands and cluster in the cell membrane12,152. Clustered integrins recruit adaptor and signalling proteins via
their intracellular domains, thereby inducing phosphorylation and dephosphorylation signals into the cell
ECM fibre (‘outside in’ signalling)9,152. The integrin cytoplasmic tail directly interacts with α-actinin, talin, the focal
adhesion kinase (FAK), and other proteins such as tensin12,152–154.All these proteins can bind adaptor
proteins — through SH2, SH3 or proline-rich domains — to recruit actin-binding proteins (vinculin,
paxillin and more α-actinin) as well as regulatory molecules (PI3K and RHO-family GTPases) to focal
2 Formation of focal contact contacts12,152,154.A link between the actin/ARP2/3 complex and emerging focal contacts is provided by
vinculin155. The assembly of focal contacts is directly and indirectly induced by various signalling pathways,
such as active PI3K, protein kinase C (PKC) and RHO GTPases (‘inside-out’ signalling)9,154,156,157.
Step 3: recruitment of surface proteases to ECM contacts and focalized proteolysis. Surface proteases
become concentrated near substrate binding sites158. In close proximity to the cell surface, proteases
cleave ECM components, such as collagen, fibronectin and laminins, as well as pro-MMPs, to create active
soluble MMPs, such as MMP2 (REFS 20,39,159). MT1-MMP, MMP1 and other collagenases cleave native
collagens, along with other ECM macromolecules, into smaller fragments, which, in turn, are accessible
to subsequent degradation by gelatinases (MMP2 and MMP9)43,159,164 or serine proteases.
Step 4: cell contraction by actomyosin. Active myosin II binds to actin filaments (then termed
actomyosin) and generates actomyosin contraction7,22,28 (contraction direction is indicated by arrows).
The Ca2+- and calmodulin-dependent myosin light-chain kinase (MLCK) phosphorylates the myosin light
chain (MLC), which activates myosin II. This is counteracted by dephosphorylation via the MLC
phosphatase (MLCPtase)26. RHO regulates actomyosin contraction predominantly through its effector,
ROCK, which phosphorylates and inhibits MLCPtase22,165.
3 Focalized proteolysis Step 5: detachment of the trailing edge. At the trailing edge, focal contact dissassembly occurs through
several mechanisms.Actin binding and severing proteins (for example, gelsolin and cofilin) cap actin
filaments and cause actin filament strand breakage, respectively, thereby promoting filament turnover166.
Phosphatases limit the assembly of cytoskeletal proteins167. The cytoplasmic protease calpain cleaves focal
contact components (talin, cytoplasmic tail of β1 and β3 integrins)32,168. FAK causes focal contact
disassembly by yet unknown mechanisms157,167. Focal contacts are further weakened through proteolytic
cleavage of adhesion receptors by sheddases169 and the accumulation of collagen fragments that are
generated while the cell moves forward170. Following focal contact disassembly, integrins detach from the
substrate and become internalized via endocytotic vesicles for recycling towards the leading edge171 or
deposited onto the substrate13,19.
4 Actomyosin contraction 5 Detachment of the trailing edge

Actin filament Vinculin, paxillin,

ARP2/3 complex α-actinin
Recycling WASP FAK
Clustered PIPs MT1-MMP
PI3K, PI45K Soluble MMPs
(MMP1, 2, 9)
β1, β3 integrins
seprase
α-actinin,
talin, tensin, Partially degraded
MLCK MLCPtase ICAP, ILK ECM

ECM fragments Phosphatases

ROCK Myosin II Calpain
Sliding
Ca2+ Actin-capping Sheddase
Calmodulin proteins
RHO

364 | MAY 2003 | VOLUME 3 www.nature.com/reviews/cancer

© 2003 Nature Publishing Group
 REVIEWS

Table 1 | Promigratory factors of the tumour microenvironment

Factor type Factor name Tumour cells affected Effect
Chemokines AMF Melanoma151 Activation of migration through heterotrimeric
G-proteins, PIPs, RAC and RHO
SDF1/CXCR4 Ovarian carcinoma173 Activation of migration through heterotrimeric
G-proteins, PI3K and RHO/ROCK
Growth factors EGF Ovarian cancer52 Activation of autocrine signalling loops
at the leading edge that induce ruffling and focal
contact formation
Breast cancer174 Activation of PI3K and PLC
LPA Ovarian cancer175 Activation of migration through heterotrimeric
Hepatoma48 G-proteins and RHO/ROCK
IGF1 Breast cancer176 Engagement of α2β1 integrin
Breast cancer177 Activation of migration through PI3K, FAK and paxillin
Pancreatic carcinoma, Activation of PLCγ, PI3K, RAC and RHO;
melanoma54 facilitated detachment through activation of
FOCAL COMPLEX phosphatases
Small and transient cell Partially degraded Collagenases Engagement of αvβ3 and α2β1 integrins
interactions between the cell and collagen (MMP-1, 13,
the extracellular-matrix MT1-MMP)
substrate that contain integrins,
FAK and talin. They interact Factors that cleave MMP2, MT1-MMP Breast cancer cells51 Engagement of αv integrins;
with the diffuse actin network at laminin-5 (γ2 chain) Melanoma53 release of chemotactic peptides
high turnover rates. They can AMF, autocrine motility factor; CXCR4, CX-chemokine receptor-4; FAK, focal adhesion kinase; IGF1, insulin-like growth factor-1; LPA,
resolve within seconds to lysophosphatidic acid; MMP, matrix metalloproteinase; MT-MMP, membrane-type matrix metalloproteinase; PI3K, phosphatidylinositol
minutes, or mature into a focal 3-kinase; PIPs, phosphoinosites; PLC, phospholipase C; ROCK, RHO-associated kinase; SDF1, stromal-cell-derived factor-1.
contact.

FOCAL CONTACT assemblies of integrins and cytoskeletal proteins12,33. fragments as well as promigratory neoepitopes that
(syn. focal adhesion)
This provides a molecular basis for the diversity that is engage specific sets of integrins51,53 (TABLE 1). Many of
Stable cell–substrate interactions
that evolve from a focal complex. observed in cell-migration strategies. these migration-enhancing factors, such as the epider-
They contain integrins, FAK, Many studies confirm that the multistep model of mal growth factor (EGF), also promote other functions,
talin, vinculin, paxillin and cell migration (BOX 1) applies to cancer cells. Cancer-cell such as cell proliferation and survival. Factors such as
many other proteins that couple motility involves integrin signalling, focal-contact insulin-like growth-factor-1 (IGF1), alternatively, seem
to the actin filament network.
Turnover rates are in the range
formation and actomyosin-dependent contractil- to preferentially activate cell migration54.
of at least minutes and longer. ity16,18,31,36–38. ECM-degrading enzymes, such as matrix
They are the insertion place of metalloproteinases (MMPs) and cathepsins, are fre- Diversity in tumour-cell migration
organized actin filaments, which quently upregulated in tumour cells39–42, and facilitate But how do migrating cancer cells move through tissues
disassemble or mature into
migration in vitro 39,40,43, as well as dissemination and and become organized into invasive tumours, as deter-
stable adhsion sites.
metastasis in vivo44,45. Similarly, the overexpression or mined by histopathology? In vitro and in vivo observations
ACTIN FILAMENT activation of the RAC, RHO, ROCK or MLCK signalling have shown that tumour cells infiltrate neighbouring
Elongated polymers of pathways have been correlated with in vitro tumour- tissue matrices in diverse patterns. They can disseminate
aggregated actin monomers. cell migration, as well as in vivo invasion and pro- as individual cells, referred to as ‘individual cell migration’,
Filaments aggregate to networks
or thicker strands through
gression46–49. Therefore, pharmacological inhibitors that or expand in solid cell strands, sheets, files or clusters,
intercalation of crosslinking block integrins, MMPs, ROCK or MLCK are being called ‘collective migration’. In many tumours, both single
proteins. developed to interfere with cancer-cell invasion27,46,48–50. cells and collectives are simultaneously present. Whereas
These and other findings indicate that the basic leukaemias, lymphomas and most solid stromal tumours,
CORTICAL ACTIN
migration machinery of normal cells is retained in such as sarcomas, disseminate via single cells, epithelial
Meshwork of branched actin
filaments that form along the tumour cells. Unlike physiological processes of cell inva- tumours commonly use collective migration mecha-
inner leaflet of the plasma sion, however, the migration of tumour cells seems to be nisms. In principle, the lower the differentiation stage, the
membrane. It provides stiffness activated by a dominance of promigratory events in the more likely the tumour is to disperse via individual cells55.
and contractility, and interacts absence of counteracting stop signals51,52. This imbalance Such differences in cellular patterning putatively reflect
with integrins and signalling
molecules. It can be rapidly
of signals allows cancer cells to become continuously variations in the molecular repertoire used by a cancer cell
remodelled, which is correlated migratory and invasive, leading to tumour expansion to migrate.
with higher migration dynamics. across tissue boundaries, followed by metastasis.
Single-cell migration
STRESS FIBRE
Modulation by environmental factors In vitro and in vivo studies in cell lines led to the original
Highly organized, thick fibres of
actin filaments that are Multiple environmental factors (TABLE 1) can propel, observations that individual tumour cells are motile56,57.
organized in parallel by direct and regulate tumour-cell motility and thereby Individual motile tumour cells usually originate from the
crosslinking proteins. They contribute to invasion. Motility-inducing chemokines interstitial stroma or bone marrow. Alternatively, cells
extend between focal adhesions and growth factors induce and maintain migration by that originated from a multicellular compartment, such
to the cytoplasm. Their
formation is correlated with less-
promigratory signal transduction via phosphatidyl- as epithelium, lose their cell contacts, detach and migrate
dynamic cell anchoring and inositol 3-kinase (PI3K), RAC and RHO signalling as individual cells through the adjacent connective
contractility. (TABLE 1). Matrix proteases generate chemotactic ECM tissue55. Based on cell type, integrin engagement,

NATURE REVIEWS | C ANCER VOLUME 3 | MAY 2003 | 3 6 5

© 2003 Nature Publishing Group
REVIEWS

Migration strategy Tumour type Mesenchymal migration. Mesenchymal cells move via
the five-step migration cycle (BOX 1). Mesenchymal
movement is predominantly found in cells from con-
Lymphoma nective-tissue tumours, such as fibrosarcomas 58,
Leukaemia
SCLC gliomas59 and in epithelial cancers following progres-
Ameoboid sive dedifferentiation60,61 (FIG. 1). It represents an effi-
Individual cient mechanism for tumour-cell dissemination and
metastasis 60–62. Cells that undergo mesenchymal
Fibrosarcoma
Glioblastoma migration (TABLE 2) have a fibroblast-like spindle-
Mesenchymal (single cells) Anaplastic tumours shaped morphology that is dependent on integrin-
Mesenchymal (chains) mediated adhesion dynamics and the presence of
high traction forces on both cell poles 15–18,31,33,59.
Integrins, MT-MMPs and other proteases co-localize
Epithelial cancer at fibre binding sites to execute pericellular proteoly-
Melanoma
Cadherins, gap junctions

sis39,58,63. The activation of MMPs and uPA is required

Cluster/cohorts for maintenance of the phenotype and mesenchymal
Integrins, proteases

Collective migration in vitro 64,65 and in vivo 61,66. Focal contacts

form and are turned over in the range of 10–120
Epithelial cancer minutes 11,31, resulting in relatively slow migration
Vascular tumours velocities of 0.1–2 µm/min in 3D models33. At least
Multicellular strands/sheets part of the invasiveness of the mesenchymal pheon-
type can be attributed to the presence of focal con-
Figure 1 | Diversity of tumour invasion mechanisms. Individual or collective tumour-cell tacts and actomyosin-mediated contractility, which
migration strategies are determined by different molecular programmes (triangles). From
individual (top) to collective (bottom) movements, increased control of cell–ECM interaction is
are controlled by RHO and ROCK, or MLCK27,48,49,67.
provided by integrins and matrix-degrading proteases. Cell–cell adhesion through cadherins Consequently, mesenchymal migration is impaired
and other adhesion receptors, as well as cell–cell communication, via gap junctions, are by inhibiting the function of integrins, RHO GTPases
specific characteristics of collective cell behaviour. Haematopoietic neoplasia (leukaemia and or MLCK18,33,48,49.
lymphoma) and small-cell lung carcinoma (SCLC) cells have been shown to undergo amoeboid
behaviour. By contrast, mesenchymal-type migration occurs in sarcomas and glioblastomas. Amoeboid migration. Many established tumour cell
Detached and disseminating cell collectives (clusters or cohorts) are observed in epithelial
lines do not follow these mesenchymal characteris-
cancers that retain high or intermediate levels of differentiation, such as breast and colon
carcinoma, prostate cancer, as well as melanoma. Multicellular strands and sheets that do not tics, but use a less adhesive, amoeboid type of migra-
detach are invasive, yet rarely metastatic. These occur in some epithelial cancers, including tion56,68. The characteristics of amoeboid movement
basal-cell carcinomas and benign vascular tumours. have been established through studies of the single-
cell amoeba, Dictyostelium discoideum (BOX 2).
Dictyostelium is an elliptoid cell that translocates via
cytoskeletal structure and protease production, single-cell rapidly alternating cycles of morphological expan-
migration can occur in different morphological variants. sion and contraction, extraordinary deformability
These variants include MESENCHYMAL and amoeboid types, and relatively low-affinity substrate binding that is
as well as cell chains (FIG. 1). integrin independent69–71. In higher eukaryotes, signs

Table 2 | Differences in cellular and molecular migration mechanisms

Characteristic Mesenchymal Amoeboid
Cell shape Elongated, fibroblast-like Roundish /elliptoid
(length 50–200 µm) (length 10–30 µm)
Growth in culture Adhesive Growth in suspension
Migration velocity Low (0.1–1 µm/min) Low to high (0.1–20 µm/min)
Cell-matrix interactions Integrins and proteases focalize Integrins and proteases are
non-focalized
Structure of actin cytoskeleton Cortical and stress fibres Cortical
Adhesion force generated High, fibre pulling and bundling Low, minor fibre bending
Proteolytic extracellular-matrix remodelling Present to extensive Not present
Cellular migration mechanism Traction dependent Propulsive, cytoplasmic streaming
MESENCHYME
Synonomous with mesoderm, a Mechanism overcoming matrix barriers Path generation, formation of Path finding, propulsion and
3D network of undifferentiated proteolytic ECM defects cytoplasmic forward flow
migratory fibroblast-like cells (‘streaming’); squeezing through
and interstitial stroma that narrow regions (constriction ring)
forms in the notochord of the Prototypic non-neoplastic cell Fibroblast, smooth-muscle cell Lymphocyte, neutrophil
developing embryo. It gives rise
to all connective tissues, Neoplastic cells, carcinoma Fibrosarcoma, glioblastoma, Lymphoma, small-cell lung
including muscle and bone.
defifferentiated epithelial cancer carcinoma, small-cell prostate cancer

366 | MAY 2003 | VOLUME 3 www.nature.com/reviews/cancer

© 2003 Nature Publishing Group
 REVIEWS

(FIG. 1). Amoeboid tumour cells have a low degree of

Box 2 | Dictyostelium discoideum as a model for cell migration
adhesiveness for collagens (compared to mesenchymal
The amoeba Dictyostelium discoideum undergoes single cell, as well as collective, cell cells), due to their low levels of β1 and β3 integrin expres-
movements at different stages of its life cycle. It represents an important experimental sion80–83. Amoeboid dissemination allows cancer cells to
model for studying the diversity and plasticity in eukaryotic cell migration172. In undergo early detachment and metastatic spread from a
contrast to mammalian cells, neither integrins nor surface proteases, such as matrix small primary tumour, and is most commonly observed
metalloproteinases, are expressed by Dictyostelium71,146. These cells therefore use in lymphomas and small-cell lung carcinomas80.
alternate mechanisms to interact with their extracellular environments.
Dictyostelium amoeba spend most of their life cycle as individual cells that migrate
Chain migration. Chain migration occurs in non-neo-
through soil, leaf mold or across litter stratum, where they phagocytose bacteria and
plastic neural crest cells84, myoblasts85 and melanomas.
yeast to obtain nutrients. Under good nutrient conditions, amoebae move individually,
These cells ‘stream’ one after another in a strand-like
chemotax towards their food supply (they are attracted by bacterial products), and
divide every few hours. Under starvation conditions, low humidity or inappropriate fashion. When clusters of melanoma cells were
external osmolarity, amoebae stop dividing and initiate a developmental programme implanted into 3D fibrillar collagen, cell ‘strands’ devel-
that leads to the formation of a multicellular organism. Initially, 104–105 cells aggregate oped within a matrix defect. The strand was generated
together by chemotaxis to form a tipped mound. The mound develops a spore head, by the first invading cell, or the ‘guerilla cell’19(C. Mayer
which elongates, falls onto the substrate, and generates a multicellular slug or and P. F., unpublished observations). This represents
pseudoplasmodium — that is, a worm-like migratory cell cluster. Following slug the migratory alignment of cells along tracks of remod-
formation, two different cell types (prestalk and prespore) differentiate, sort and elled ECM19,84. While the cells move in a chain-like
position themselves within the slug. Prestalk cells form the faster-moving front of the fashion, they form cell–cell contacts at ‘tip- like’ junc-
slug, whereas prespore cells are located in rear parts. Slug cells also secrete an outer sheet tions, so some cell–cell adhesion and communication
of extracellular-matrix proteins and cellulose. Attracted by light and heat, multicellular mechanisms seem to be preserved (FIG. 1).
slug migration can be maintained for days. Improving growth conditions induce slug Chains of single tumour cells aligned between stro-
maturation and differentiation into a fruiting body, which consists of a head that mal fibres, termed ‘Indian files’, are a characteristic histo-
contains spores, a stalk tube and a basal disk that attaches to the substrate. Finally, the logical feature of epithelial neoplasias. These chains have
spores germinate, dissociate and form a new generation of individual amoeba, thereby been observed in infiltrating lobular or metaplastic
completing the life cycle. breast carcinoma86,87, ovarian cancer 88 and melanoma of
a vascular-type pattern89. The arrangement of invading
tumour cells in chains seems to represent a particularly
of amoeboid movement (TABLE 2) are retained in leuko- effective penetration mechanism that confers high
cytes and some tumour cells68,70,72,73. The migration of metastatic capacity and poor prognosis86,89.
these cells differs in several ways from mesenchymal
migration (TABLE 2). Lymphocytes and neutrophils use Collective migration and invasion. COLLECTIVE MOVEMENT
a fast ‘gliding’ type of movement that is driven by of cells is a well-described phenomenon that occurs
short-lived and relatively weak interactions with the during embryological development, such as during the
substrate. Movement is generated by cortical filamen- migration of cell sheets in the blastoderm or the ecto-
tous actin, whereas mature focal contacts, stress fibres derm following closure of the neural tube90, or during
and focalized proteolytic activity are lacking 34,73,74 the development of glands and ducts of mammary tis-
(K. Wolf et al., unpublished observations). sue (termed ‘branching morphogenesis’)91,92. Other
In lymphocytes and neutrophils, integrin-medi- processes that involve collective migration include the
ated adhesion is completely or partially dispensable migration of myoblasts after ortotopic injection86 and
for cell migration within connective tissue, both the sprouting of endothelial cells during the formation
in vitro and in vivo 35,75,76. Similar to Dictyostelium, of new blood vessels93,94.
T lymphocytes use protease-independent, physical The biomechanics of collective movements were pri-
mechanisms to overcome matrix barriers. Physical marily investigated in explant cultures in vitro, revealing
migration strategies include adaptation to preformed important differences from individual cell movement.
matrix structures, extension of lateral footholds77,78 Studies of the in vitro migration of keratinocyte sheets95,
and the formation of constriction rings at regions of fish melanocyte clusters96 and cell collectives from pri-
narrow space74 (K. Wolf et al., unpublished observa- mary tumour explants or tumour cell lines97,98 showed
tions). Such shape-driven migration allows cells to that aggregated cells can move as a functional unit (see
circumnavigate, rather than degrade, their ECM bar- Movie 1 at http://cancerres.aacrjournals.org/cgi/content/
riers. These cells are highly deformable, due to their full/62/7/2125/DC1; REF. 99). In contrast to individual
lack of focal contacts, allowing them to move at migrating cells, cell–cell adhesion that occurs in cell
10–30-fold higher velocities than cells that use mes- groups leads to a specific form of cortical actin filament
enchymal migration mechanisms33,73 (K. Wolf et al., assembly along cell junctions99. This allows the formation
unpublished observations). of a larger-sized, multicellular contractile body.
COLLECTIVE MOVEMENT Amoeboid migration is a characteristic feature of V2 A subset of highly mobile cells at the front of the
The migration of a coherent cell kidney carcinoma cells injected into the dermal stroma of body, designated ‘path-generating cells’, generates
group or mass that move as the rabbit ear57, mammary carcinoma cells transformed migratory traction via pseudopod activity 97,99. Cells in
strands, sheets or clusters of up
to several hundred cells. Cell
by oncogenic RAS72, lymphoma and myeloic leukaemia inner and trailing regions are passively dragged
junctions are maintained by cells79, as well as cells of certain neuroendocrine tumours, behind97. Cells at the leading margin engage and cluster
cell–cell adhesion. including small-cell carcinoma of the lung and prostate80 β1 integrins in anterior protrusions towards the ECM

NATURE REVIEWS | C ANCER VOLUME 3 | MAY 2003 | 3 6 7

© 2003 Nature Publishing Group
REVIEWS

Mesenchymal Amoeboid

Mesenchymal–amoeboid
transition

Proteases

Epithelial–mesenchymal
Cadherins
transition
Collective–amoeboid
β1 integrins
transition
Strand
Cluster

Figure 2 | Plasticity of tumour invasion mechanisms. Disseminating cancer cells can undergo a variety of adaptation reactions in
response to changes in their molecular migration programme. In cancer-cell collectives, such as epithelial sheets or strands, individual
cells that have lost cell–cell interactions (such as by inhibition of cadherin function) can detach, and use integrins and proteases to
develop a mesenchymal-type migration (epithelial–mesenchymal transition). When proteases such as matrix metalloproteinases
(MMPs), serine proteases and cathepsins are blocked in cells that are undergoing mesenchymal migration, these cells can adapt by
using protease-independent amoeboid crawling (mesenchymal–amoeboid transition). Amoeboid single-cell migration can also result
from cell collectives and clusters after treatment with adhesion-disrupting anti-β1 integrin antibodies. By directly and indirectly interfering
with cell–cell junctions, as well as adhesion to collagen fibres, single cells detach and migrate by β1-integrin-independent amoeboid
mechanisms. All these transitions can be induced by drugs that are designed to inhibit cancer-cell migration, such as integrin or
cadherin antagonists and protease inhibitors. Similar transitions might also occur spontaneously during tumour progression.

substrate91,99, and show an increased expression and superfamily (for example, ALCAM), as well as
activity of MT1-MMP and MMP-2, leading to polar- connexins, which are involved in communication
ized ECM degradation100. Collective movements are through gap junctions104,105,108,109.
therefore sensitive to integrin antagonists, which block Comparative studies have identified individual
traction and migration91,99. They are also susceptible to invading tumour cells, observed during neoplastic
protease inhibitors — this sensitivity has been shown differentiation and in haematological neoplasias, as
for myoblast collectives85 during branching morpho- the main cause for systemic dissemination and
genesis92 and during angiogenesis93,94. metastasis 55,110. Collective-cell invasion, however,
In tumours, two morphological and functional predominates in highly differentiated tumours, such
variants of collective migration have been described as lobular breast cancer, epithelial prostate cancer and
(FIG. 1). The first is protruding sheets and strands that large-cell lung cancer86,101,106. Histological studies have
maintain contact with the primary site, yet generate indicated that tissue infiltration by individual cells
local invasion. These characteristics are histologically is rarely detected (or absent) in these tumour types.
detectable in invasive epithelial cancer such as oral So, collective-cell movement could be a primary
squamous-cell carcinoma and mammary carci- mechanism for invasion and metastasis by highly
noma86,101, colon carcinoma102, basal-cell carcinoma differentiated tumours.
and others. The second results from detached cell clus- What are the advantages of collective-cell movement?
ters. Groups of cells, histologically seen as ‘nests’, The large cell mass can produce high autocrine concen-
detach from their origin and frequently extend along trations of promigratory factors and matrix proteases,
interstitial tissue gaps and paths of least resistance, as and protect inner cells from immunological assault by
well as along perineural structures, as seen in epithelial lymphocytes and natural-killer cells. Because heteroge-
cancer, melanoma and rhabdomyosarcoma86,101–103. neous sets of cells move as one functional unit, cells of
In vivo, such tumour collectives can be detected at any different clonal origin and with different biological abili-
stage of metastasis. Tumour clusters enter lymphyatic ties can function together. For example, the more migra-
vessels104–106 and can be isolated from peripheral tory cells can promote the invasion of less mobile, yet
blood107. It is obvious that the coordination of integrin possibly apoptosis-resistant clones, thereby increasing
and cytoskeletal activity within the cell collective tumour invasion efficiency and survival probability.
requires not only cell–cell adhesion, but also commu-
nication among cells (FIG. 1). In epithelial cancer and Plasticity of tumour-cell migration
melanoma samples, homotypic cell–cell interactions In vitro cancer-cell migration studies using 2D sub-
within multicellular strands and sheets have been strata have reproducibly shown the importance of inte-
shown to express cadherins (E-, N-, P-, VE-cadherin grin-mediated adhesion to the underlying substrate.
and cadherin-11), members of the immunoglobulin 3D culture and in vivo studies of the contribution of

368 | MAY 2003 | VOLUME 3 www.nature.com/reviews/cancer

© 2003 Nature Publishing Group
 REVIEWS

integrins to cancer-cell migration, however, have deliv- cancers of breast, colon, lung and prostate61,110. As a pre-
ered inconsistent results. A positive correlation between requisite for EMT, the cells lose their cell–cell junctions
integrin function and invasion was shown for via different mechanisms, while retaining expression of
melanoma (integrins α2β1 and β3)111,112, ovarian can- migration-promoting molecules, such as integrins55,130.
cer (integrins β1 and αvβ3)113, colorectal carcinoma These changes can occur through loss-of-function
(integrin α6β4)114, mammary and prostate carcinoma mutations in cadherins or catenins, or by the upregula-
cells115, and glioma (integrin α6β1)59. Other investiga- tion of proteases that cleave cadherins55,61,65,131. Loss of
tions, however, have shown that integrins can serve as cell–cell adhesion can also occur through production of
negative effectors that impede or counteract invasion cytokines by the tumour microenvironment. For exam-
and progression, such as integrin αvβ3 in melanoma116, ple, hepatocyte growth factor/scatter factor (HGF/SF)
integrin α2β1 in breast carcinoma cells117, and integrin downregulates cadherins and activates promigratory
α5β1 in colon carcinoma cells118. small GTPases64,66,132. These changes in cell morphology
Similarly, data on the function of proteases in and function are accompanied by changes in protein
tumour invasion and metastasis are not completely expression profiles, including the loss of the epithelial
consistent. The upregulation of MMPs, uPA and CYTOKERATINS and the de novo expression of VIMENTIN.
cathepsins is a uniform process in many different can- The EMT is considered to be a significant step in the
cer cell types, and has been positively correlated with invasive cascade. Because it marks the transition from a
tumour progression and metastasis41,42,45. In several collective to a single-cell migration mechanism, EMT
epithelial cancer models, inhibition of MMPs and represents an example of phenotypic and functional
serine proteases impairs tumour-cell migration plasticity that spontaneously occurs during the natural
in vitro119–122 and metastasis after ortotopic implanta- course of tumour progression. Once the tumour has
tion44,123,124. In lymphoma125 or oesophageal and ovar- achieved the dedifferentiated stage of single-cell dis-
ian carcinoma models126, however, cell migration was semination, metastatic spread is increased, resulting in
not inhibited by protease inhibitors. Similarly, clinical poor prognosis55,110.
trials on MMP inhibitors (MMPIs) in late-stage can-
cer patients have yielded an inconsistent outcome — Mesenchymal–amoeboid transition
in most cases, significant progression occurs despite Under certain circumstances, cancer cells can undergo
MMPI treatment127–129. conversion from a mesenchymal cell type towards an
These diverse observations could be attributed to the amoeboid cell type, which is termed the mesenchy-
varying degree of integrin and protease expression and mal–amoeboid transition (MAT)58. This transition is
function in different cancer cells. In addition, however, not only accompanied by a change in cell morphology
tumour cells could compensate for the loss of a particu- (from fibroblast-like spindle-shaped towards roundish
lar motile ability by developing migratory escape strate- and eliptoid), but also results in altered integrin distri-
gies. Such adaptation responses, known as ‘plasticity’, bution, organization of the actin cytoskeleton, and
can be experimentally induced in several ways (FIG. 2). changes in molecular strategies to overcome tissue bar-
These include modifying the amount of cell–cell adhe- riers (TABLE 2). Several factors can lead to MAT, includ-
sion, the traction force generated via integrins and ing abrogation of pericellular proteolysis by protease
cytoskeleton, the level of substrate adhesiveness (by inhibitors, weakening of cell–ECM linkages, and
changing ECM composition and density), or the inhibition of RHO signalling pathways.
EPITHELIAL–MESENCHYMAL requirement for proteolysis of the ECM. As a conse-
TRANSITION
Detachment of individual
quence, cells can shift from highly adhesive to low adhe- Abrogation of pericellular proteolysis. MAT can serve as
fibroblast-like moving cells from sive migration, from proteolytic to non-proteolytic, a compensatory mechanism of tumour-cell migration
an epithelial collective. It from collective to individual migration, and so on. It has after the inhibition of pericellular proteolysis (FIG. 2).
requires the downregulation of been established that during the natural course of Highly invasive and metastatic HT1080 fibrosarcoma
cell–cell junctions, such as
tumour progression, changes in protein expression and cells and MDA-MB-231 breast cancer cells use a mes-
cadherins.
function, such as through mutations in adhesion mole- enchymal migration strategy in 3D collagen matrices —
CYTOKERATIN cules, can alter the cellular phenotype and invasiveness. that is, they execute all five steps of migration, including
Cytoskeletal proteins that Similar changes in cell motility can also occur after pericellular proteolysis and the generation of tube-like
assemble to form the treatment with biological response modifiers — a matrix defects58. The cellular capacity to degrade and
intermediate filament
cytoskeleton in sessile, epithelial
response called ‘drug-induced plasticity’. remodel the pericellular ECM environment can be
cells. Cytokeratins anchor to blocked by treatment with pharmacological inhibitors
adhesive structures, such as Epithelial–mesenchymal transition of MMPs, serine proteases and cathepsins. After pro-
desmosomes and The most well-established example of changes in can- tease function is virtually abolished, these cancer cells
hemidesmosomes.
cer-cell pattern and function is termed the EPITHELIAL– did not, however, become trapped in the tissue scaffold
55
MESENCHYMAL TRANSITION (EMT) . Following tumour and cease migration, but instead acquired the ability to
VIMENTIN
Intermediate filaments in progression and dedifferentiation, epithelial cancer cells undergo non-proteolytic amoeboid migration. The
moving non-epithelial cells that can undergo a transition from a collective invasion pat- resulting protease-independent migration was sustained
lack cell–cell junctions. tern towards a detached and disseminated cell migra- by cell alignment along pre-existing fibre strands, shape
Vimentin is a (not entirely
specific) histological marker for
tion mechanism. These cells move in a fibroblast-like change and the ability to squeeze through narrow
cells of mesenchymal fashion (FIG. 2). Histologically, this process corresponds matrix regions (see movies 4–6 at http://www.jcb.org/
phenotype. to the sarcoma-like lesions that might arise in epithelial cgi/content/full/jcb.200209006/DC1). In a manner sim-

NATURE REVIEWS | C ANCER VOLUME 3 | MAY 2003 | 3 6 9

© 2003 Nature Publishing Group
REVIEWS

Box 3 | Impacts of methodology on migration strategies shown that different cell types retain an amazing capac-
ity to migrate in the absence of MMPs and other pro-
The type of cell-migration assay used in a cell motility study has considerable teases. This capacity has been shown for fibroblasts93,
impact on the results and conclusions. As cells migrate across a 2D substrate, activated T lymphocytes and lymphoma cells (K. Wolf
blocking proteins that maintain adhesion (such as integrins) will lead to the loss of et al., unpublished observations), and tumour cells58.
cell-substrate binding and ultimately result in cell detachment into the supernatant. One way in which cells could continue to migrate in the
In 2D models, migration is therefore dependent on adhesion. When migrating cells absence of pericellular proteolysis might involve the
are placed in a 3D fibrillar or non-fibrillar extracellular matrix (ECM), they become
induction of amoeboid crawling58 (K. Wolf et al., unpub-
embedded in the scaffold, and passively undergo contacts with surrounding matrix
lished observations). MAT could therefore represent a
structures. They therefore cannot escape from being touched by ECM. Whereas
drug-resistance mechanism, although the molecular
basic migration frequently involves integrin-mediated traction2,75, the loss of
pathways and clinical implications of this mechanism
integrin-mediated adhesion can then be partially or completely compensated by
other non- or low-adhesion mechanisms, such as cell shape change, amoeboid clearly require further investigation.
propulsion, formation of lateral footholds or cytoplasmic streaming 35,77,78. These
morphodynamic mechanisms have been observed during adhesion-independent Weakening of cell–ECM linkages. Because integrins are
lymphocyte migration through collagen lattices35,77, and in neutrophils migrating required for the generation of spindle-shaped cell elon-
within the amnionic membrane78. Such alternate mechanisms of motility cannot be gation, disruption of integrin-mediated adhesion causes
detected in 2D migration models, yet are relevant to the cell’s physiological cells to acquire a spherical morphology. Cytoskeletal
environment, as tissue scaffolds contain gaps and spaces that are confined by non- dynamics and the generation of ruffles and pseudopods,
randomly aligned ECM strands. So, differences in assay conditions could account however, persist (see movie 1B at http://www.jcb.org/
for our current lack of understanding about cell-adhesion mechanisms that involve cgi/content/full/jcb.200209006/DC1). In tumour cells
very low adhesive forces. migrating within 3D fibrillar collagen (BOX 3), integrin
blocking agents have been shown to induce the loss of
mesenchymal elongation, followed by the gain of amoe-
Spontaneous migration Integrins blocked boid morphodynamics (N. Daryab and P. F., manuscript
2D
substrate
in preparation). Uncoupled from β1 integrin function,
Integrins Loss of clustering
clustered
such morphological plasticity, connected to cytoplasmic
streaming and amoeboid propulsion, allows tumour
Detachment
cells to slowly travel over several hundred micrometers in
the course of days or weeks (N. Daryab, unpublished
observations), similar to integrin-independent migra-
tion of lymphocytes or Dictyostelium. It should be noted
3D Mesenchymal Amoeboid that, for mostly technical reasons, few examples for inte-
fibrillar
ECM 3D fibrillar matrix grin-independent migration in non-neoplastic and neo-
plastic cells have been reported (BOX 3). In vivo, several
different cell types from chimeric β1-integrin-deficient
mice, such as embryonic stem cells, lymphocytes,
myoblasts and neural-crest cells, all undergo normal
migration and positioning76,137,138. This implicates alter-
native mechanisms of cell translocation that might also
be retained in tumour cells, including compensation by
ilar to that of leukocytes and Dictyostelium, this amoe- other integrins or adhesion receptor systems, as well as
boid type of tumour-cell movement was mediated by physical migration mechanisms (BOX 3).
an exclusively cortical actin cytoskeleton and non-focal-
ized β1 integrins58. Traction forces on collagen fibres Inhibition of RHO-related pathways. Migratory plastic-
were reduced (K. W. and P. F., unpublished observa- ity of tumour cells can further be induced by inhibiting
tions), indicating a switch to a less adhesive type of the RHO-effector ROCK. In some RAS-transformed
tumour-cell crawling. cancer cells, including epithelial cancer and transformed
Because proteases produced by cancer cells have been fibroblasts, lack of stress-fibre formation has been corre-
shown to mediate invasive procedures such as pericellular lated with the induction of amoeboid-type behaviour,
ECM breakdown, growth-factor activation and the such as roundish morphology and amoeboid shape
induction of EMT, pharmacological protease inhibitors change72. This yields high migration velocities both
have been developed as cancer therapeutics. These target in vitro and in vivo72,139–141. Induction of this phenotype
MMPs129,133,134, serine proteases135 and cathepsins136. The depends on RAC activation (for polarization, cortical
efficacy of MMP and serine protease inhibitors has actin and pseudopod dynamics)142, whereas ROCK-
recently been tested in clinical trials of patients with late- dependent stress-fibre formation and adhesivity are
stage cancers. Results have shown that these compounds reduced or lost 46.
are largely ineffective in slowing late-stage tumour pro- In conjunction, these findings indicate that changes
gression and metastasis. The possible explanations for the at different positions of the multistep migration model
clinical failure of protease inhibitors are complex, and can lead to a stereotypic conversion of cells towards
require further investigations (reviewed in REFS 127–129). amoeboid tumour-cell migration.
However, basic research and clinical studies have both

370 | MAY 2003 | VOLUME 3 www.nature.com/reviews/cancer

© 2003 Nature Publishing Group
 REVIEWS

Collective–amoeboid transition (CAT) Future directions

The process of collective–amoeboid transition is similar It will be important to diversify and further develop
to the conversion that occurs in EMT. The important dif- our current reductionist picture of tumour invasion
ference, however, is that after breakage of cell–cell link- and migration. Mechanisms such as EMT, MAT and
ages, β1-integrin-independent amoeboid migration CAT could contribute to the de-differentiation of
occurs. Several lines of evidence indicate that β1 integrins, tumours and their respective dissemination patterns.
most notably α2β1, α3β1 and α5β1, directly or indirectly Both individual and collective migration modes are
maintain cell–cell adhesion and cooperation in cell col- presumably further regulated — and complicated —
lectives143–145. In migrating cell clusters that eminate from by heterologous interactions between tumour cells and
primary melanoma explants, anti-β1 integrin antibodies reactive stromal cells of the tumour microenvironment.
induce not only the loss of collective movement, but also A more comprehensive understanding of the molecular
the dissemination of individual cells (see movies 3 and 4 basis of diversity and adaptation of cell migration —
at http://cancerres.aacrjournals.org/cgi/content/full/ possibly through gene expression profiling and pro-
62/7/2125/DC1). In these studies, detaching cells under- teomic analysis — is therefore required to efficiently
went β1-independent amoeboid migration, as deter- target cancer-cell motility and invasion.
mined by cell shape change, appearance of small ruffles As a supramolecular process, cell motility has turned
and pseudopods, uropod formation, diffuse integrin dis- out to be unexpectedly durable, and difficult to safely yet
tribution and the presence of constriction rings99. So, in a effectively block with pharmacological compounds.
manner that is similar to spontaneous dedifferentiation Stringently regulated cell-adhesion signalling pathways
in EMT, the abrogation of β1 integrin function can gen- are exclusive to cells of higher multicelluar organisms146.
erate single-cell dissemination. Mechanisms of cell motility and positioning that devel-
Plasticity of tumour-cell patterning is likely to occur oped earlier in evolution, however, are likely to be
in a bi-directional manner. The spontaneous reversion retained in higher eukaryotic cells. It is possible that
from amoeboid to a more collective phenotype was disruption of components that maintain the highly dif-
recently reported for small-cell lung carcinoma cells, ferentiated cell–cell and cell–ECM interactions of verte-
leading to increased resistance to cytostatic drugs and brate cells, such as integrins, cadherins, proteases and
irradiation83. This reverse transition occurred in non- some cytoskeletal components, induce cancer cells to
adhesive, amoeboid single cells that expressed low levels undergo a ‘step backwards in evolutionary time’, towards
of α2β1 and α3β1 integrins. Following continuous a more primitive, amoeboid-type cell migration.
subculture in vitro, these cells spontaneously developed Although it is unclear at present how such adapta-
an epithelial phenotype, as characterized by newly tion responses contribute to the overall process of
acquired cell–cell contacts, growth in cell islands, signif- tumorigenesis, it is conceivable that activation of
icant upregulation of α2β1, α3β1 and CD44, as well as amoeboid-cell behaviour could allow cancer cells to
increased substrate adhesion83. Together, these studies escape therapeutic agents that are designed to target
indicate that the conversion of cancer cells from an integrins, proteases and RHO signalling pathways. A
individual to a collective behaviour might occur better understanding of amoeboid migration, as well
through an aggregation and detachment process that is as the development of therapeutic combination regi-
reminiscent of morphodynamic events that occur in mens that target multiple motility pathways, are
Dictyostelium and during vertebrate development. therefore warranted27.

1. Chambers, A. F., Groom, A. C. & MacDonald, I. C. 10. Otey, C. A. & Burridge, K. Patterning of the membrane 17. Leavesley, D. I., Ferguson, G. D., Wayner, E. A. & Cheresh, D. A.
Dissemination and growth of cancer cells in metastatic sites. cytoskeleton by the extracellular matrix. Semin. Cell Biol. 1, Requirement of the integrin beta 3 subunit for carcinoma cell
Nature Rev. Cancer 2, 563–572 (2002). 391–399 (1990). spreading or migration on vitronectin and fibrinogen. J. Cell
2. Friedl, P. & Bröcker, E.-B. The biology of cell locomotion 11. Zamir, E. et al. Dynamics and segregation of cell-matrix Biol. 117, 1101–1107 (1992).
within three-dimensional extracellular matrix. Cell. Mol. Life adhesions in cultured fibroblasts. Nature Cell Biol. 2, 18. Maaser, K. et al. Functional hierarchy of simultaneously
Sci. 57, 41–64 (2000). 191–196 (2000). expressed adhesion receptors: integrin alpha2beta1 but not
3. Abercrombie, M., Dunn, G. A. & Heath, J. P. The shape and 12. Zamir, E. & Geiger, B. Molecular complexity and dynamics CD44 mediates MV3 melanoma cell migration and matrix
movement of fibroblasts in culture. Soc. Gen. Physiol. Ser. of cell-matrix adhesions. J. Cell Sci. 114, 3583–3590 reorganization within three-dimensional hyaluronan-
32, 57–70 (1977). (2001). containing collagen matrices. Mol. Biol. Cell 10, 3067–3079
4. Lauffenburger, D. A. & Horwitz, A. F. Cell migration: a References 11 and 12 explain the dynamic nature and (1999).
physically integrated molecular process. Cell 84, 359–369 diversity of cell-matrix interactions. They provide the 19. Friedl, P. et al. Migration of highly aggressive MV3
(1996). molecular basis for plasticity and adaptation melanoma cells in 3-dimensional collagen lattices results in
The ‘bible’ for cell migration researchers. responses in cell migration. local matrix reorganization and shedding of alpha2 and
5. Adams, J. C. Cell-matrix contact structures. Cell. Mol. Life 13. Regen, C. M. & Horwitz, A. F. Dynamics of beta 1 integrin- beta1 integrins and CD44. Cancer Res. 57, 2061–2070
Sci. 58, 371–392 (2001). mediated adhesive contacts in motile fibroblasts. J. Cell Biol. (1997).
6. Burridge, K. & Chrzanowska-Wodicka, M. Focal adhesions, 119, 1347–1359 (1992). 20. Friedl, P. & Wolf, K. Proteolytic and non–proteolytic migration in
contractility, and signaling. Annu. Rev. Cell Dev. Biol. 12, 14. Smilenov, L. B., Mikhailov, A., Pelham, R. J., Marcantonio, tumor cells and leukocytes. Biochem. Soc. Symp. (in the press).
463–519 (1996). E. E. & Gundersen, G. G. Focal adhesion motility 21. Byers, H. R. & Fujiwara, K. Stress fibers in cells in situ:
7. Cramer, L. P. Organization and polarity of actin filament revealed in stationary fibroblasts. Science 286, immunofluorescence visualization with antiactin,
networks in cells: implications for the mechanism of 1172–1174 (1999). antimyosin, and anti-alpha-actinin. J. Cell Biol. 93, 804–811
myosin-based cell motility. Biochem. Soc. Symp. 65, 15. Cukierman, E., Pankov, R., Stevens, D. R. & Yamada, K. M. (1982).
173–205 (1999). Taking cell-matrix adhesions to the third dimension. Science 22. Katoh, K. et al. Rho-kinase-mediated contraction of isolated
8. Rohatgi, R. et al. The interaction between N-WASP and 294, 1708–1712 (2001). stress fibers. J. Cell Biol. 153, 569–584 (2001).
the Arp2/3 complex links Cdc42-dependent signals to 16. Rabinovitz, I. & Mercurio, A. M. The integrin alpha6beta4 23. Chew, T. L., Wolf, W. A., Gallagher, P. J., Matsumura, F. &
actin assembly. Cell 97, 221–231 functions in carcinoma cell migration on laminin-1 by Chisholm, R. L. A fluorescent resonant energy transfer-
(1999). mediating the formation and stabilization of actin- based biosensor reveals transient and regional myosin light
9. Hynes, R. O. Integrins: bidirectional, allosteric signaling containing motility structures. J. Cell Biol. 139, 1873–1884 chain kinase activation in lamella and cleavage furrows.
machines. Cell 110, 673–687 (2002). (1997). J. Cell Biol. 156, 543–553 (2002).

NATURE REVIEWS | C ANCER VOLUME 3 | MAY 2003 | 3 7 1

© 2003 Nature Publishing Group
REVIEWS

24. Kaibuchi, K., Kuroda, S. & Amano, M. Regulation of the 48. Itoh, K. et al. An essential part for Rho-associated kinase in 71. Fey, P., Stephens, S., Titus, M. A. & Chisholm, R. L. SadA, a
cytoskeleton and cell adhesion by the Rho family GTPases the transcellular invasion of tumor cells. Nature Med. 5, novel adhesion receptor in Dictyostelium. J. Cell Biol. 159,
in mammalian cells. Annu. Rev. Biochem. 68, 459–486 221–225 (1999). 1109–1119 (2002).
(1999). 49. Kaneko, K., Satoh, K., Masamune, A., Satoh, A. & 72. Farina, K. L. et al. Cell motility of tumor cells visualized in
25. Totsukawa, G. et al. Distinct roles of ROCK (Rho-kinase) and Shimosegawa, T. Myosin light chain kinase inhibitors can living intact primary tumors using green fluorescent protein.
MLCK in spatial regulation of MLC phosphorylation for block invasion and adhesion of human pancreatic cancer Cancer Res. 58, 2528–2532 (1998).
assembly of stress fibers and focal adhesions in 3T3 cell lines. Pancreas 24, 34–41 (2002). Fascinating in vivo imaging of amoeboid tumour-cell
fibroblasts. J. Cell Biol. 150, 797–806 (2000). 50. Allman, R., Cowburn, P. & Mason, M. In vitro and in vivo movement.
26. Kamm, K. E. & Stull, J. T. Dedicated myosin light chain effects of a cyclic peptide with affinity for the alpha(nu)beta3 73. Friedl, P., Borgmann, S. & Brocker, E. B. Leukocyte crawling
kinases with diverse cellular functions. J. Biol. Chem. 276, integrin in human melanoma cells. Eur. J. Cancer 36, through extracellular matrix and the Dictyostelium paradigm
4527–4530 (2001). 410–422 (2000). of movement: lessons from a social amoeba. J. Leukoc.
27. Somlyo, A. V. et al. Rho kinase and matrix metalloproteinase 51. Gianelli, G., Falk-Marzillier, J., Schiraldi, O., Stetler- Biol. 70, 491–509 (2001).
inhibitors cooperate to inhibit angiogenesis and growth of Stevenson, W. G. & Quaranta, V. Induction of cell migration 74. Lewis, W. H. On the locomotion of the polymorphonuclear
human prostate cancer xenotransplants. FASEB J. 17, by matrix metalloprotease-2 cleavage of laminin-5. Science neutrophiles of the rat in autoplasma cultures. Bull. Johns
223–234 (2003). 277, 225–228 (1997). Hopkins. Hosp. 4, 273–279 (1934).
28. Verkhovsky, A. B., Svitkina, T. M. & Borisy, G. G. Myosin II 52. Alper, O. et al. Epidermal growth factor receptor signaling One of the first papers showing shape change as a
filament assemblies in the active lamella of fibroblasts: their and the invasive phenotype of ovarian carcinoma cells. mechanism to bypass ECM barriers.
morphogenesis and role in the formation of actin filament J. Natl Cancer Inst. 93, 1375–1384 (2001). 75. Werr, J., Xie, X., Hedqvist, P., Ruoslahti, E. & Lindbom, L.
bundles. J. Cell Biol. 131, 989–1002 (1995). 53. Seftor, R. E. et al. Cooperative interactions of laminin 5 Beta1 integrins are critically involved in neutrophil
29. Friedl, P. & Brocker, E. B. in Image Analysis. Methods and gamma2 chain, matrix metalloproteinase-2, and membrane locomotion in extravascular tissue in vivo. J. Exp. Med. 187,
Applications 2nd edn (ed. Hader, D. P.) 9–21 (CRC Press, type-1-matrix/metalloproteinase are required for mimicry of 2091–2096 (1998).
London, 2001). embryonic vasculogenesis by aggressive melanoma. 76. Brakebusch, C. et al. Beta1 integrin is not essential for
30. Palecek, S. P., Loftus, J. C., Ginsberg, M. H., Lauffenburger, D. A. Cancer Res. 61, 6322–6327 (2001). hematopoiesis but is necessary for the T cell-dependent
& Horwitz, A. F. Integrin-ligand binding properties govern cell 54. Brooks, P. C. et al. Insulin-like growth factor receptor IgM antibody response. Immunity 16, 465–477 (2002).
migration speed through cell-substratum adhesiveness. cooperates with integrin alpha v beta 5 to promote tumor cell 77. Haston, W. S., Shields, J. M. & Wilkinson, P. C. Lymphocyte
Nature 385, 537–540 (1997). dissemination in vivo. J. Clin. Invest. 99, 1390–1398 (1997). locomotion and attachment on two-dimensional surfaces
31. Ballestrem, C., Hinz, B., Imhof, B. A. & Wehrle-Haller, B. 55. Thiery, J. P. Epithelial–mesenchymal transitions in tumour and in three-dimensional matrices. J. Cell Biol. 92, 747–752
Marching at the front and dragging behind: differential progression. Nature Rev. Cancer 2, 442–454 (2002). (1982).
alphaVbeta3-integrin turnover regulates focal adhesion Comprehensive review on a classic example of Provides the first example of ‘non-adhesive’ migration
behavior. J. Cell Biol. 155, 1319–1332 (2001). plasticity during tumour invasion. in 3D collagen lattices.
32. Potter, D. A. et al. Calpain regulates actin remodeling during 56. Enterline, H. T. & Cohen, D. R. The ameboid motility of 78. Mandeville, J. T., Lawson, M. A. & Maxfield, F. R. Dynamic
cell spreading. J. Cell Biol. 141, 647–662 (1998). human and animal neoplastic cells. Cancer 3, 1033–1038 imaging of neutrophil migration in three dimensions:
33. Friedl, P., Zanker, K. S. & Bröcker, E.-B. Cell migration (1950). mechanical interactions between cells and matrix.
strategies in 3-D extracellular matrix: differences in 57. Wood, S. Jr. Pathogenesis of metastasis formation J. Leukoc. Biol. 61, 188–200 (1997).
morphology, cell matrix interactions, and integrin function. observed in vivo in the rabbit ear chamber. Arch. Pathol. 66, 79. Verschueren, H., de Baetselier, P. & Bereiter-Hahn, J.
Microsc. Res. Tech. 43, 369–378 (1998). 550–568 (1958). Dynamic morphology of metastatic mouse T-lymphoma
34. Entschladen, F., Niggemann, B., Zänker, K. S. & Friedl, P. 58. Wolf, K. et al. Compensation mechanism in tumor cell cells invading through monolayers of 10T1/2 cells. Cell Motil.
Differential requirement of protein tyrosine kinases and migration: mesenchymal–amoeboid transition after Cytoskeleton 20, 203–214 (1991).
protein kinase C in the regulation of T cell locomotion in blocking of pericellular proteolysis. J. Cell Biol. 160, 80. Rintoul, R. C. & Sethi, T. The role of extracellular matrix in
three-dimensional collagen matrices. J. Immunol. 159, 267–277 (2003). small-cell lung cancer. Lancet Oncol. 2, 437–442 (2002).
3203–3210 (1997). A novel escape mechanism after the blocking of 81. Falcioni, R. et al. Expression of beta 1, beta 3, beta 4, and
35. Friedl, P., Entschladen, F., Conrad, C., Niggemann, B. & pericellular proteolysis. By converting to a more beta 5 integrins by human lung carcinoma cells of different
Zanker, K. S. CD4+ T lymphocytes migrating in three- ‘primitive’ amoeboid migration mode, previously histotypes. Exp. Cell Res. 210, 113–122 (1994).
dimensional collagen lattices lack focal adhesions and utilize proteolytic tumour cells continue to move non- 82. Jaspars, L. H., Bonnet, P., Bloemena, E. & Meijer, C. J.
beta1 integrin-independent strategies for polarization, proteolytically by path-finding migration strategies. Extracellular matrix and beta 1 integrin expression in nodal
interaction with collagen fibers and locomotion. Eur. J. 59. Paulus, W., Baur, I., Beutler, A. S. & Reeves, S. A. Diffuse and extranodal T-cell lymphomas. J. Pathol. 178, 36–43
Immunol. 28, 2331–2343 (1998). brain invasion of glioma cells requires beta 1 integrins. Lab. (1996).
36. Yamada, K. M. et al. Monoclonal antibody and synthetic Invest. 75, 819–826 (1996). 83. Kraus, A. C. et al. In vitro chemo- and radio-resistance in
peptide inhibitors of human tumor cell migration. Cancer 60. Polette, M. et al. Association of fibroblastoid features with small cell lung cancer correlates with cell adhesion and
Res. 50, 4485–4496 (1990). the invasive phenotype in human bronchial cancer cell lines. constitutive activation of AKT and MAP kinase pathways.
37. Filardo, E. J., Brooks, P. C., Deming, S. L., Damsky, C. & Clin. Exp. Metastasis 16, 105–112 (1998). Oncogene 21, 8683–8695 (2002).
Cheresh, D. A. Requirement of the NPXY motif in the integrin 61. Tester, A. M., Ruangpani, N., Anderso, R. L. & 84. Jacques, T. S. et al. Neural precursor cell chain migration
beta 3 subunit cytoplasmic tail for melanoma cell migration Thompson, E. W. MMP-9 secretion and MMP-2 and division are regulated through different beta1 integrins.
in vitro and in vivo. J. Cell Biol. 130, 441–450 (1995). activation distinguish invasive and metastatic sublines of Development 125, 3167–3177 (1998).
38. Flanagan, L. A. et al. Filamin A, the Arp2/3 complex, and the a mouse mammary carcinoma system showing 85. El Fahime, E., Torrente, Y., Caron, N. J., Bresolin, M. D. &
morphology and function of cortical actin filaments in human epithelial–mesenchymal transition traits. Clin. Exp. Tremblay, J. P. In vivo migration of transplanted myoblasts
melanoma cells. J. Cell Biol. 155, 511–517 (2001). Metastasis 18, 553–560 (2000). requires matrix metalloproteinase activity. Exp. Cell Res.
39. Sameni, M., Moin, K. & Sloane, B. F. Imaging proteolysis by 62. Putz, E. et al. Phenotypic characteristics of cell lines derived 258, 279–287 (2000).
living human breast cancer cells. Neoplasia 2, 496–504 from disseminated cancer cells in bone marrow of patients 86. Page, D. L., Anderson, T. J. & Sakamoto, G. in Diagnostic
(2001). with solid epithelial tumors: establishment of working Histopathology of the Breast 219–222 (Churchill
40. Rosenthal, E. L., Hotary, K., Bradford, C. & Weiss, S. J. Role models for human micrometastases. Cancer Res. 59, Livingstone, New York, 1987).
of membrane type 1-matrix metalloproteinase and 241–248 (1999). 87. Pitts, W. C. et al. Carcinomas with metaplasia and sarcomas
gelatinase A in head and neck squamous cell carcinoma 63. d’Ortho, M. P. et al. MT1-MMP on the cell surface causes of the breast. Am. J. Clin. Pathol. 95, 623–632 (1991).
invasion in vitro. Otolaryngol. Head Neck Surg. 121, focal degradation of gelatin films. FEBS Lett. 421, 159–164 88. Sood, A. K. et al. Molecular determinants of ovarian cancer
337–343 (1999). (1998). plasticity. Am. J. Pathol. 158, 1279–1288 (2001).
41. Koblinski, J. E., Ahram, M. & Sloane, B. F. Unraveling the 64. Jeffers, M., Rong, S. & Vande, W. G. Enhanced 89. Seftor, E. A. et al. Molecular determinants of human uveal
role of proteases in cancer. Clin. Chim. Acta 291, 113–135 tumorigenicity and invasion-metastasis by hepatocyte melanoma invasion and metastasis. Clin. Exp. Metastasis
(2000). growth factor/scatter factor-Met signalling in human cells 19, 233–246 (2002).
42. Hofmann, U. B., Westphal, J. R., van Muijen, G. N. & Ruiter, concomitant with induction of the urokinase proteolysis 90. Davidson, L. A. & Keller, R. E. Neural tube closure in
D. J. Matrix metalloproteinases in human melanoma. network. Mol. Cell. Biol 16, 1115–1125 (1996). Xenopus laevis involves medial migration, directed
J. Invest. Dermatol. 115, 337–344 (2000). 65. Pulyaeva, H. et al. MT1-MMP correlates with MMP-2 protrusive activity, cell intercalation and convergent
43. Deryugina, E. I., Bourdon, M. A., Reisfeld, R. A. & activation potential seen after epithelial to mesenchymal extension. Development 126, 4547–4556 (1999).
Strongin, A. Remodeling of collagen matrix by human transition in human breast carcinoma cells. Clin. Exp. 91. Klinowska, T. C. et al. Laminin and beta1 integrins are crucial
tumor cells requires activation and cell surface Metastasis 15, 111–120 (1997). for normal mammary gland development in the mouse. Dev.
association of matrix metalloproteinase-2. Cancer Res. 66. Sternlicht, M. D. et al. The stromal proteinase Biol. 215, 13–32 (1999).
58, 3743–3750 (1998). MMP3/stromelysin-1 promotes mammary carcinogenesis. 92. Simian, M. et al. The interplay of matrix metalloproteinases,
44. Maekawa, K., Sato, H., Furukawa, M. & Yoshizaki, T. Inhibition Cell 98, 137–146 (1999). morphogens and growth factors is necessary for branching
of cervical lymph node metastasis by marimastat (BB-2516) in 67. Yoshioka, K., Nakamori, S. & Itoh, K. Overexpression of of mammary epithelial cells. Development 128, 3117–3131
an orthotopic oral squamous cell carcinoma implantation small GTP-binding protein RhoA promotes invasion of tumor (2001).
model. Clin. Exp. Metastasis 19, 513–518 (2002). cells. Cancer Res. 59, 2004–2010 (1999). 93. Hiraoka, N., Allen, E., Apel, I. J., Gyetko, M. R. & Weiss, S. J.
45. Rudolph-Owen, L. A., Chan, R., Muller, W. J. & Matrisian, L. M. 68. Condeelis, J., Jones, J. & Segall, J. E. Chemotaxis of Matrix metalloproteinases regulate neovascularization by
The matrix metalloproteinase matrilysin influences early- metastatic tumor cells: clues to mechanisms from the acting as pericellular fibrinolysins. Cell 95, 365–377 (1998).
stage mammary tumorigenesis. Cancer Res. 58, Dictyostelium paradigm. Cancer Metastasis Rev. 11, 55–68 94. Collen, A. et al. Membrane-type matrix metalloproteinase-
5500–5506 (1998). (1992). mediated angiogenesis in a fibrin-collagen matrix. Blood
46. Sahai, E., Olson, M. F. & Marshall, C. J. Cross-talk between 69. Yumura, S., Mori, H. & Fukui, Y. Localization of actin and 101, 1810–1817 (2003).
Ras and Rho signalling pathways in transformation favours myosin for the study of ameboid movement in Dictyostelium 95. Vaughan, R. B. & Trinkaus, J. P. Movements of epithelial cell
proliferation and increased motility. EMBO J. 20, 755–766 using improved immunofluorescence. J. Cell Biol. 99, sheets in vitro. J. Cell Sci. 1, 407–413 (1966).
(2001). 894–899 (1984). 96. Kolega, J. The movement of cell clusters in vitro:
47. Clark, E. A., Golub, T. R., Lander, E. S. & Hynes, R. O. 70. Devreotes, P. N. & Zigmond, S. H. Chemotaxis in eukaryotic morphology and directionality. J. Cell Sci. 49, 15–32 (1981).
Genomic analysis of metastasis reveals an essential role for cells: a focus on leukocytes and Dictyostelium. Annu. Rev. Provides a detailed description of how clustered cells
RhoC. Nature 406, 532–535 (2000). Cell Biol. 4, 649–686 (1988). move as a functional unit.

372 | MAY 2003 | VOLUME 3 www.nature.com/reviews/cancer

© 2003 Nature Publishing Group
 REVIEWS

97. Friedl, P. et al. Migration of coordinated cell clusters in activation by modulation of membrane-type 1 matrix 141. Khosravi-Far, R. et al. Dbl and Vav mediate transformation
mesenchymal and epithelial cancer explants in vitro. Cancer metalloproteinase activity in high and low invasive melanoma via mitogen-activated protein kinase pathways that are
Res. 55, 4557–4560 (1995). cell lines. J. Biol. Chem. 274, 21056–21062 (1999). distinct from those activated by oncogenic Ras. Mol. Cell.
98. Nabeshima, K. et al. Ultrastructural study of TPA-induced 121. Ntayi, C., Lorimier, S., Berthier-Vergnes, O., Hornebeck, W. Biol. 14, 6848–6857 (1994).
cell motility: human well-differentiated rectal & Bernard, P. Cumulative influence of matrix 142. Qiu, R. G., Chen, J., McCormick, F. & Symons, M. A role for
adenocarcinoma cells move as coherent sheets via localized metalloproteinase-1 and -2 in the migration of melanoma Rho in Ras transformation. Proc. Natl Acad. Sci. USA 92,
modulation of cell–cell adhesion. Clin. Exp. Metastasis 13, cells within three-dimensional type I collagen lattices. Exp. 11781–11785 (1995).
499–508 (1995). Cell Res. 270, 110–118 (2001). 143. Whittard, J. D. & Akiyama, S. K. Activation of beta1 integrins
99. Hegerfeldt, Y., Tusch, M., Brocker, E. B. & Friedl, P. 122. Hotary, K., Allen, E., Punturieri, A., Yana, I. & Weiss, S. J. induces cell–cell adhesion. Exp. Cell Res. 263, 65–76 (2001).
Collective cell movement in primary melanoma explants: Regulation of cell invasion and morphogenesis in a three- 144. Robinson, E. E., Zazzali, K. M., Corbett, S. A. & Foty, R. A.
plasticity of cell-cell interaction, β1-integrin function, and dimensional type I collagen matrix by membrane-type matrix Alpha5beta1 integrin mediates strong tissue cohesion.
migration strategies. Cancer Res. 62, 2125–2130 (2002). metalloproteinases 1, 2, and 3. J. Cell Biol. 149, 1309–1323 J. Cell Sci. 116, 377–386 (2003).
First example of collective–amoeboid transition after (2000). 145. Whittard, J. D. et al. E-cadherin is a ligand for integrin
blocking of β1-integrins — another escape Together with references 63 and 159, this paper alpha2beta1. Matrix Biol. 21, 525–532 (2002).
mechanism. shows that MT1-MMP is the most important 146. Hynes, R. O. & Zhao, Q. The evolution of cell adhesion.
100. Nabeshima, K. et al. Front-cell-specific expression of collagenase in invading tumour cells. They formally J. Cell Biol. 150, F89–F96 (2000).
membrane-type 1 matrix metalloproteinase and gelatinase A prove the concept that pericellular proteolysis Insightful overview of how adhesion receptors
during cohort migration of colon carcinoma cells induced by contributes to invasion. evolved in the context of increasing tissue complexity
hepatocyte growth factor/scatter factor. Cancer Res. 60, 123. Wang, X., Fu, X., Brown, P. D., Crimmin, M. J. & Hoffman, R. M. and segregation.
3364–3369 (2000). Matrix metalloproteinase inhibitor BB-94 (batimastat) inhibits 147. Blanchoin, L. et al. Direct observation of dendritic actin
101. Bell, C. D. & Waizbard, E. Variability of cell size in primary human colon tumor growth and spread in a patient-like filament networks nucleated by Arp2/3 complex and
and metastatic human breast carcinoma. Invasion orthotopic model in nude mice. Cancer Res. 54, 4726–4728 WASP/Scar proteins. Nature 404, 1007–1011 (2000).
Metastasis 6, 11–20 (1986). (1994). 148. Nobes, C. D. & Hall, A. Rho, Rac, and Cdc42 GTPases
102. Nabeshima, K., Inoue, T., Shimao, Y., Kataoka, H. & Koono, M. 124. Maki, H. et al. Augmented anti-metastatic efficacy of a regulate the assembly of multimolecular focal complexes
Cohort migration of carcinoma cells: differentiated colorectal selective matrix metalloproteinase inhibitor, MMI-166, in associated with actin stress fibers, lamellipodia, and
carcinoma cells move as coherent cell clusters or sheets. combination with CPT-11. Clin. Exp. Metastasis 19, filopodia. Cell 81, 53–62 (1995).
Histol. Histopathol. 14, 1183–1197 (1999). 519–526 (2002). 149. Ren, X. D. et al. Physical association of the small GTPase Rho
103. Ackerman, A. B. & Ragaz, A. in The Lives of Lesions (ed. 125. Kruger, A. et al. Hydroxamate-type matrix metalloproteinase with a 68-kDa phosphatidylinositol 4-phosphate
Ackerman, A. B.) 147–158 (Masson Publishers, New York, inhibitor batimastat promotes liver metastasis. Cancer Res. 5-kinase in Swiss 3T3 cells. Mol. Biol. Cell 7, 435–442 (1996).
1984). 61, 1272–1275 (2001). 150. Leopoldt, D. et al. Gbetagamma stimulates
104. Hashizume, R., Koizumi, H., Ihara, A., Ohta, T. & Uchikoshi, T. 126. Della, P. P. et al. Combined treatment with serine protease phosphoinositide 3-kinase-gamma by direct interaction with
Expression of beta-catenin in normal breast tissue and inhibitor aprotinin and matrix metalloproteinase inhibitor two domains of the catalytic p110 subunit. J. Biol. Chem.
breast carcinoma: a comparative study with epithelial Batimastat (BB-94) does not prevent invasion of human 273, 7024–7029 (1998).
cadherin and alpha-catenin. Histopathology 29, 139–146 esophageal and ovarian carcinoma cells in vivo. Anticancer 151. Tsutsumi, S., Gupta, S. K., Hogan, V., Collard, J. G. & Raz, A.
(1996). Res. 19, 3809–3816 (1999). Activation of small GTPase Rho is required for autocrine
105. Madhavan, M. et al. Cadherins as predictive markers of 127. Zucker, S., Cao, J. & Chen, W. T. Critical appraisal of the use motility factor signaling. Cancer Res. 62, 4484–4490 (2002).
nodal metastasis in breast cancer. Mod. Pathol. 14, of matrix metalloproteinase inhibitors in cancer treatment. 152. Miyamoto, S. et al. Integrin function: molecular hierarchies of
423–427 (2001). Oncogene 19, 6642–6650 (2000). cytoskeletal and signaling molecules. J. Cell Biol. 131,
106. Byers, S. W., Sommers, C. L., Hoxter, B., Mercurio, A. M. & 128. Coussens, L. M., Fingleton, B. & Matrisian, L. M. Matrix 791–805 (1995).
Tozeren, A. Role of E-cadherin in the response of tumor cell metalloproteinase inhibitors and cancer: trials and 153. Calderwood, D. A., Shattil, S. J. & Ginsberg, M. H. Integrins
aggregates to lymphatic, venous and arterial flow: tribulations. Science 295, 2387–2392 (2002). and actin filaments: reciprocal regulation of cell adhesion
measurement of cell–cell adhesion strength. J. Cell Sci. 108, 129. Overall, C. M. & Lopez-Otin, C. Strategies for MMP inhibition and signaling. J. Biol. Chem. 275, 22607–22610 (2000).
2053–2064 (1995). in cancer: innovations for the post-trial era. Nature Rev. 154. Degani, S. et al. The integrin cytoplasmic domain-associated
107. Brandt, B. et al. Isolation of prostate-derived single cells and Cancer 2, 657–672 (2002). protein ICAP-1 binds and regulates Rho family GTPases
cell clusters from human peripheral blood. Cancer Res. 56, References 127–129 summarize and explain the during cell spreading. J. Cell Biol. 156, 377–387 (2002).
4556–4561 (1996). failure of clinical trials of matrix metalloproteinase 155. DeMali, K. A., Barlow, C. A. & Burridge, K. Recruitment of the
108. Pishvaian, M. J. et al. Cadherin-11 is expressed in inhibitors in late-stage cancer patients. It sets the Arp2/3 complex to vinculin: coupling membrane protrusion
invasive breast cancer cell lines. Cancer Res. 59, scene for future work in the MMP field. to matrix adhesion. J. Cell Biol. 159, 881–891 (2002).
947–952 (1999). 130. Lochter, A., Navre, M., Werb, Z. & Bissell, M. J. alpha1 and 156. Schwartz, M. A. & Shattil, S. J. Signaling networks linking
109. Hsu, M., Andl, T., Li, G., Meinkoth, J. L. & Herlyn, M. alpha2 integrins mediate invasive activity of mouse integrins and Rho family GTPases. Trends Biochem. Sci. 25,
Cadherin repertoire determines partner-specific gap mammary carcinoma cells through regulation of stromelysin- 388–391 (2000).
junctional communication during melanoma progression. 1 expression. Mol. Biol. Cell 10, 271–282 (1999). 157. Ilic, D. et al. Reduced cell motility and anhanced focal
J. Cell Sci. 113, 1535–1542 (2000). 131. Sternlicht, M. D. & Werb, Z. How matrix metalloproteinases adhesion contact formation in cells from FAK-deficient mice.
110. Moll, R., Mitze, M., Frixen, U. H. & Birchmeier, W. Differential regulate cell behavior. Annu. Rev. Cell Dev. Biol. 17, Nature 377, 539–544 (1995).
loss of E-cadherin expression in infiltrating ductal and 463–516 (2001). 158. Mueller, S. C. et al. A novel protease-docking function of
lobular breast carcinomas. Am. J. Pathol. 143, 1731–1742 132. Sakkab, D. et al. Signaling of hepatocyte growth integrin at invadopodia. J. Biol. Chem. 274, 24947–24952
(1993). factor/scatter factor (HGF) to the small GTPase Rap1 via the (1999).
111. Klein, C. E., Steinmayer, T., Kaufmann, D., Weber, L. & large docking protein Gab1 and the adapter protein CRKL. 159. Ohuchi, E. et al. Membrane type 1 matrix
Brocker, E. B. Identification of a melanoma progression J. Biol. Chem. 275, 10772–10778 (2000). metalloproteinase digests interstitial collagens and other
antigen as integrin VLA-2. J. Invest. Dermatol. 96, 281–284 133. Quirt, I. et al. Phase II study of marimastat (BB-2516) in extracellular matrix macromolecules. J. Biol. Chem. 272,
(1991). malignant melanoma: a clinical and tumor biopsy study of 2446–2451 (1997).
112. Trikha, M. et al. The high affinity alphaIIb beta3 integrin is the National Cancer Institute of Canada Clinical Trials Group. 160. Dumin, J. A. et al. Pro-collagenase-1 (matrix
involved in invasion of human melanoma cells. Cancer Res. Invest. New Drugs 20, 431–437 (2002). metalloproteinase-1) binds the alpha(2)beta(1) integrin upon
57, 2522–2528 (1997). 134. Bonomi, P. Matrix metalloproteinases and matrix release from keratinocytes migrating on type I collagen.
113. Strobel, T. & Cannistra, S. A. Beta1-integrins partly mediate metalloproteinase inhibitors in lung cancer. Semin. Oncol. J. Biol. Chem. 276, 29368–29374 (2001).
binding of ovarian cancer cells to peritoneal mesothelium 29, 78–86 (2002). 161. Brooks, P. C., Silletti, S., von Schalscha, T. L., Friedlander, M.
in vitro. Gynecol. Oncol. 73, 362–367 (1999). 135. Harbeck, N., Kates, R. E. & Schmitt, M. Clinical relevance & Cheresh, D. A. Disruption of angiogenesis by PEX, a
114. Chao, C., Lotz, M. M., Clarke, A. C. & Mercurio, A. M. of invasion factors urokinase-type plasminogen activator noncatalytic metalloproteinase fragment with integrin
A function for the integrin alpha6beta4 in the invasive and plasminogen activator inhibitor type 1 for binding activity. Cell 92, 391–400 (1998).
properties of colorectal carcinoma cells. Cancer Res. 56, individualized therapy decisions in primary breast cancer is 162. Ellerbroek, S. M., Wu, Y. I., Overall, C. M. & Stack, M. S.
4811–4819 (1996). greatest when used in combination. J. Clin. Oncol. 20, Functional interplay between type I collagen and cell surface
115. Cress, A. E., Rabinovitz, I., Zhu, W. & Nagle, R. B. The alpha 1000–1007 (2002). matrix metalloproteinase activity. J. Biol. Chem. 276,
6 beta 1 and alpha 6 beta 4 integrins in human prostate 136. Leung-Toung, R., Li, W., Tam, T. F. & Karimian, K. Thiol- 24833–24842 (2001).
cancer progression. Cancer Metastasis Rev. 14, 219–228 dependent enzymes and their inhibitors: a review. Curr. 163. Galvez, B. G., Matias-Roman, S., Yanez-Mo, M., Sanchez-
(1995). Med. Chem. 9, 979–1002 (2002). Madrid, F. & Arroyo, A. G. ECM regulates MT1-MMP
116. Danen, E. H. J. et al. Regulation of integrin-mediated 137. Fassler, R. & Meyer, M. Consequences of lack of beta 1 integrin localization with beta1 or alphavbeta3 integrins at distinct cell
adhesion to laminin and collagen in human melanocytes and gene expression in mice. Genes Dev. 9, 1896–1908 (1995). compartments modulating its internalization and activity on
in non-metastatic and highly metastatic human melanoma 138. Fassler, R. et al. Lack of beta 1 integrin gene in embryonic human endothelial cells. J. Cell Biol. 159, 509–521 (2002).
cells. Int. J. Cancer 54, 315–321 (1993). stem cells affects morphology, adhesion, and migration but 164. Tam, E. M., Wu, Y. I., Butler, G. S., Stack, M. S. & Overall, C. M.
117. Zutter, M. M., Santoro, S. A., Staatz, W. D. & Tsung, Y. L. not integration into the inner cell mass of blastocysts. J. Cell Collagen binding properties of the membrane type-1 matrix
Re-expression of the alpha 2 beta 1 integrin abrogates the Biol. 128, 979–988 (1995). metalloproteinase (MT1-MMP) hemopexin C domain. The
malignant phenotype of breast carcinoma cells. Proc. Natl Together with references 35, 76 and 77, this paper ectodomain of the 44-kDa autocatalytic product of MT1-
Acad. Sci. USA 92, 7411–7415 (1995). shows that β1 integrins can be dispensable for cell MMP inhibits cell invasion by disrupting native type I collagen
118. Schirner, M. et al. Integrin alpha5beta1: a potent inhibitor of movement and positioning in 3D tissue structures. cleavage. J. Biol. Chem. 277, 39005–39014 (2002).
experimental lung metastasis. Clin. Exp. Metastasis 16, 139. Lombardi, L. et al. Ultrastructural cytoskeleton alterations 165. Fukata, Y., Amano, M. & Kaibuchi, K. Rho-Rho-kinase
427–435 (1998). and modification of actin expression in the NIH/3T3 cell line pathway in smooth muscle contraction and cytoskeletal
119. Mignatti, P., Robbins, E. & Rifkin, D. B. Tumor invasion after transformation with Ha-ras-activated oncogene. Cell reorganization of non-muscle cells. Trends Pharmacol. Sci.
through the human amniotic membrane: requirement for a Motil. Cytoskeleton 15, 220–229 (1990). 22, 32–39 (2001).
proteinase cascade. Cell 47, 487–498 (1986). 140. Dartsch, P. C., Ritter, M., Haussinger, D. & Lang, F. 166. Wear, M. A., Schafer, D. A. & Cooper, J. A. Actin dynamics:
120. Kurschat, P. et al. Tissue inhibitor of matrix Cytoskeletal reorganization in NIH 3T3 fibroblasts expressing assembly and disassembly of actin networks. Curr. Biol. 10,
metalloproteinase-2 regulates matrix metalloproteinase-2 the Ras oncogene. Eur. J. Cell Biol. 63, 316–325 (1994). R891–R895 (2000).

NATURE REVIEWS | C ANCER VOLUME 3 | MAY 2003 | 3 7 3

© 2003 Nature Publishing Group
REVIEWS

167. Zeng, L. et al. PTP alpha regulates integrin-stimulated FAK breast cancer cell migration through a phosphatidylinositol Online links
autophosphorylation and cytoskeletal rearrangement in 3′-kinase and phospholipase C-dependent mechanism.
cell spreading and migration. J. Cell Biol. 160, 137–146 Cancer Res. 59, 5475–5478 (1999). DATABASES
(2003). 175. Sawada, K. et al. Alendronate inhibits lysophosphatidic The following terms in this article are linked online to:
168. Pfaff, M., Du, X. & Ginsberg, M. H. Calpain cleavage of acid-induced migration of human ovarian cancer cells by Cancer.gov: http://www.cancer.gov/cancer_information/
integrin beta cytoplasmic domains. FEBS Lett. 460, 17–22 attenuating the activation of Rho. Cancer Res. 62, breast cancer | colon cancer | glioma | melanoma | oesophageal
(1999). 6015–6020 (2002). cancer | ovarian cancer | prostate cancer | rhabdomyosarcoma |
169. Moss, M. L. & Lambert, M. H. Shedding of membrane proteins 176. Doerr, M. E. & Jones, J. I. The roles of integrins and small-cell lung cancer
by ADAM family proteases. Essays Biochem. 38, 141–153 extracellular matrix proteins in the insulin-like growth factor LocusLink: http://www.ncbi.nih.gov/LocusLink/
(2002). 1-stimulated chemotaxis of human breast cancer cells. β1 integrin | α2β1 integrin | β3 integrin | α5β1 integrin | α6β1
170. Carragher, N. O., Levkau, B., Ross, R. & Raines, E. W. J. Biol. Chem. 271, 2443–2447 (1996). integrin | α6β4 integrin | αvβ3 integrin | actin | ALCAM | cadherin-11 |
Degraded collagen fragments promote rapid disassembly of 177. Guvakova, M. A. & Surmacz, E. The activated insulin-like CD26 | CD44 | collagen | E-cadherin | EGF | fibronectin | HGF | IGF1 |
smooth muscle focal adhesions that correlates with growth factor 1 receptor induces depolarization in breast laminin | MLCK | MMP2 | MT1-MMP | N-cadherin | P-cadherin |
cleavage of pp125(FAK), paxillin, and talin. J. Cell Biol. 147, epithelial cells characterized by actin filament disassembly PI3K | RAC | ROCK | uPA | VE-cadherin | vitronectin
619–630 (1999). and tyrosine dephosphorylation of FAK, Cas, and paxillin.
171. Bretscher, M. S. Getting membrane flow and the cytoskeleton Exp. Cell Res. 251, 244–255 FURTHER INFORMATION
to cooperate in moving cells. Cell 87, 601–606 (1996). (1999). Amoeboid migration in T cells in collagen lattices and across
172. Firtel, R. A. & Meili, R. Dictyostelium: a model for regulated dendritic cells:
cell movement during morphogenesis. Curr. Opin. Genet. Acknowledgements http://www.immunity.com/cgi/content/full/13/3/323/DC1
Dev. 10, 421–427 (2000). We gratefully acknowledge E.-B. Bröcker for continuous support. Collective-amoeboid transition:
173. Scotton, C. J. et al. Multiple actions of the chemokine The work from the authors’ laboratory is supported by current or http://cancerres.aacrjournals.org/cgi/content/full/62/7/2125/DC1
CXCL12 on epithelial tumor cells in human ovarian previous grants from the Deutsche Forschungsgemeinschaft, the Dicty web site: http://dictybase.org/tutorial/
cancer. Cancer Res. 62, 5930–5938 (2002). Ministry for Education and Research, the Wilhelm-Sander Mesenchymal–amoeboid transition:
174. Price, J. T., Tiganis, T., Agarwal, A., Djakiew, D. & Thompson, Foundation, Evangelisches Studienwerk e.V., Haus Villigst and the http://www.jcb.org/cgi/content/full/jcb.200209006/DC1
E. W. Epidermal growth factor promotes MDA-MB-231 Interdisciplinary Center for Clinical Research (IZKF), Würzburg. Access to this interactive links box is free online.

374 | MAY 2003 | VOLUME 3 www.nature.com/reviews/cancer

© 2003 Nature Publishing Group