N-cadherin-mediated cell–cell adhesion promotes cell migration in a three-dimensional matrix

Cell migration is a critical first step in the assembly and development of tissues and metastasis of cancer cells. Many motile cells migrate with a prototypical mesenchymal phenotype by grabbing onto the complex substrate surrounding primary tumor sites, thus cell-to-extracellular matrix adhesion is a key determinant of migration phenotype. Recent studies have focused on the properties of extracellular matrix, including the organization, stiffness, and dimensionality (Geiger and Yamada, 2011). Three-dimensional (3D) matrices provide unique external cues typically absent in two-dimensional (2D), stiff surfaces. Not surprisingly, cell migration in a 3D matrix is distinct from the typical, lamellipodia-driven cell migration on 2D substrates. Unlike the prominent focal adhesions formed on a stiff substrate, adhesive complexes in a soft 3D matrix have different molecular compositions and size (Cukierman et al., 2001; Kubow and Horwitz, 2011), and in some cases are absent (Fraley et al., 2010). These studies suggest that the matrix dimension can significantly alter cell-extracellular matrix adhesion, but little is known about cell–cell adhesion between migrating cells in a 3D matrix.

Cell–cell adhesion is required for coordinated multicellular movement. Cadherins, calcium-dependent adhesion receptors, play a central role during embryo compaction (Vestweber and Kemler, 1985), cell intercalation (Cavey et al., 2008; Rauzi et al., 2010), apical constriction (Martin et al., 2009; Sawyer et al., 2009), cell sorting (Steinberg and Takeichi, 1994), and purse-string wound healing (Danjo and Gipson, 1998; Jacinto et al., 2001). With cadherin associating proteins and the underlying actin cytoskeleton, cadherin-mediated cell–cell adhesion promotes a unique cytoskeletal structure to provide adhesive strength (Gomez et al., 2011; Leckband et al., 2011; Weis and Nelson, 2006).

Cellular movement is initiated by the activation of specific transcription factors, and the altered gene expression profile results in a phenotypic transition from an epithelial to mesenchymal morphology. This epithelial-to-mesenchymal transition (EMT) drives gastrulation and neural crest delamination in embryogenesis, but is thought to also initiate the cancer cell invasion and the progression of metastatic cancer (Thiery, 2002; Thiery et al., 2009). Similar to collective cell migration observed in embryogenesis, cancer cell invasion has been shown to occur as a multicellular process in vitro (Ilina et al., 2011; Wolf et al., 2007) and in vivo (Friedl and Gilmour, 2009; Friedl and Wolf, 2003). EMT alters the gene expression profile of cell–cell adhesion receptors: the down-regulation of epithelial (E)-cadherin and the up-regulation of neural (N)-cadherin. The down-regulation of E-cadherin is a hallmark of cancer development, and E-cadherin is thought to act as a tumor suppressor (Cavallaro and Christofori, 2004). Furthermore, the E-to-N cadherin switch is often observed in aggressive cancers (Wheelock et al., 2008). Therefore, a mechanistic understanding of N-cadherin in transformed epithelial cell migration has significant implications, not only in normal developmental processes, but also in cancer progression.

Using hepatocyte growth factor (HGF) as an EMT inducer of MDCK cells, we analyzed cell invasion of transformed epithelial cells. Although HGF acts in an upstream of snail, a transcription factor that regulates E-cadherin expression (Grotegut et al., 2006), whether HGF can induce complete EMT or 3D cell invasion has not been analyzed. Here, we demonstrate that HGF-treated MDCK cells undergo the E-to-N cadherin switch and develop a highly invasive phenotype in a 3D matrix. These transformed cells migrate collectively and N-cadherin is required for both pro-migratory signaling and cell–cell adhesion between invasive cells. Furthermore, the dynamic N-cadherin-actin linkage is an essential requirement for intercellular movement within a cluster during collective cell invasion in a 3D matrix. These results reveal the roles of newly up-regulated N-cadherin in collective cell invasion of transformed epithelial cells, and may provide the mechanistic understanding of N-cadherin during cancer progression.

To study the migration of epithelial cells that have undergone an EMT, MDCK epithelial cells were cultured in HGF containing media. Unlike partial EMT observed under short term HGF exposure (Leroy and Mostov, 2007), under prolonged HGF exposure, the protein level of cadherins switched from E-to-N cadherin (Fig. 1A), which localized prominently at the cell–cell contacts of pre and post EMT cells (Fig. 1B) respectively. HGF-treated cells showed increased expression of fibronectin, another mesenchymal marker (Fig. 1A), and reduced levels of desmoplakin, a component of desmosomes (supplementary material Fig. S1). These cells lost their typical epithelial cobblestone morphology and adopted a mesenchymal, spindle shape (Fig. 1B). On a coverslip, these transformed cells migrated without maintaining cell–cell contacts, and migrated faster than untransformed epithelial cells (supplementary material Fig. S2).

In a 3D collagen matrix, HGF-treated cells exhibited an elongated morphology, with thin membrane extensions, and were also highly migratory in all three dimensions. Invasive cells exerted significant traction forces that deformed the surrounding collagen matrix toward cell body (supplementary material Movie 1), which in turn suggests the presence of a linkage between the collagen and migrating cells, likely mediated by integrins. The addition of Y27632 (Rho kinase inhibitor) relaxed the anterior and posterior collagen network of migrating cells, and halted cell migration (supplementary material Fig. S3, Movie 2), suggesting that myosin II generated traction forces are required for 3D cell migration. Despite the significant collagen deformation (supplementary material Fig. S3), thus indicating high traction forces, cells were able to maintain cell–cell contacts with neighboring cells, and migrate collectively as an elongated cluster through the 3D matrix (Fig. 1C, supplementary material Movie 1). Therefore, in a 3D matrix, either cell–cell junctions are more resilient and/or traction forces are weaker than on a 2D surface, thus preventing the mechanical disruption of cell–cell contacts often observed for migrating cells on a coverslip (de Rooij et al., 2005). In a 3D matrix, the formation of cell–cell adhesion has important consequences on cell migration phenotype.

In a 3D matrix, HGF-treated cells migrated collectively using a combination of two techniques. Individual cells within a cluster migrated along cell–cell contacts, sliding past neighboring cells (Fig. 1D, supplementary material Movie 3). Alternatively, cell clusters translocated cohesively as a single unit, while individual cells maintained contacts with the same neighboring cells (Fig. 1E, supplementary material Movie 4). The advantage of collective cell migration is the directional persistence. Single cells in the matrix migrated randomly at a similar speed in both the parallel and perpendicular directions (relative to the initial axis of cell polarization, Fig. 1G). In contrast, individual cells in a cluster migrated faster along adjacent cells parallel to the elongated direction of the cluster, than in the perpendicular direction (Fig. 1G). Since the leader cells exhibited a similar speed as single isolated cells in a 3D matrix, the faster movement of cells in a cluster is due to the increased speed of the follower cells (Fig. 1H). Significantly, the leader cells of a cell cluster migrated in a more persistent direction than single isolated cells as indicated by end-to-end displacement (Fig. 1F,I), suggesting that cell-to-cell interactions between the leader cells and follower cells promote a linear migration path in a 3D matrix.

One possible explanation for why cells prefer to migrate along and with adjacent cells is that the path provides the least resistance compared to the extracellular matrix. For example, some leader cells establish a migration path by digesting extracellular matrix and creating a tunnel for other cells to follow (Gaggioli et al., 2007; Wolf et al., 2007). To test whether a migration track or cell–cell interactions are required for persistent cell migration, the cell clusters were treated with EDTA to chelate the calcium and disrupt cell–cell adhesion. After EDTA treatment, cell–cell contacts became disrupted, and the cell cluster dissociated (Fig. 2A, supplementary material Movie 5). The newly isolated single cells extended numerous protrusions, and migrated in random directions as opposed to migrating in the same direction (Fig. 2A, arrowheads), suggesting that the presence of such a track (if any) does not influence the migration direction. In addition, since there were increased thin membrane protrusions as cell–cell contacts dissociated (Fig. 2A), cell–cell adhesion inhibits the formation of random membrane extensions, thereby minimizing cell scattering. Altogether, these data suggest that cell–cell adhesion promotes directionally persistent collective cell migration in a 3D environment by suppressing random membrane extensions.

The newly up-regulated N-cadherin prominently localized to cell–cell contacts of highly migratory cells in a 3D matrix (Fig. 2B). The N-cadherin junctions colocalized with the actin cytoskeleton, which aligned parallel to cell–cell contacts (Fig. 2C, arrowheads). Interestingly, some actin bundles organized perpendicular to N-cadherin junctions (Fig. 2C, arrows), and in the direction of the contractile force generated by migrating cells (supplementary material Fig. S3). These actin bundles, reminiscent of actin stress fibers in cells that adhere to stiff 2D substrates, were only present in cells with an elongated, migratory morphology, and were absent from circular cells. The presence of N-cadherin at the ends of the actin bundles suggests that N-cadherin may be responsible for organizing actin fibers in a 3D matrix. However, N-cadherin proteins were not the only cadherin expressed in these cells. HGF-treated cells also expressed T-cadherin, K-cadherin and cadherin 11 (Fig. 2D). None of the other cadherin levels were HGF treatment dependent, except N-cadherin, suggesting that the newly acquired cell–cell interaction and the unique actin organization between invasive cells are primarily mediated by N-cadherin.

To test the roles of N-cadherin in HGF-treated post-EMT cells, we stably silenced N-cadherin in HGF-treated cells, which had 99% (KD1) and 90% (KD2) reduction of endogenous N-cadherin (Fig. 3A; supplementary material Fig. S4A), while other cadherin levels remained unchanged (supplementary material Fig. S4B). In suspension, wild-type HGF-treated cells aggregated, but N-cadherin depleted cells (KD1) did not (Fig. 3B). In contrast, partial knockdown (KD2) cells which expressed a low level of residual endogenous N-cadherin (∼10%), formed aggregates similar in size to wild-type HGF-treated cells (Fig. 3B), indicating that a small fraction of endogenous N-cadherin is sufficient for cell aggregation. Furthermore, in the presence of cytochalasin D, which disrupts the actin cytoskeleton, wild-type HGF-treated cells also did not aggregate (Fig. 3B). Therefore, in addition to N-cadherin, an intact actin network is required for the formation of cell–cell adhesion and cell aggregates in a matrix-free suspension.

In a 3D matrix, neither N-cadherin KD1 nor KD2 cells formed large cell clusters (Fig. 3C). Cells with almost complete N-cadherin depletion (KD1) remained mostly round, and isolated (Fig. 3C). The N-cadherin depleted cells (KD1) migrated slower and traveled a shorter distance than wild-type single and clustered cells in a 3D matrix (Fig. 3D and E), suggesting that N-cadherin expression is required for the development of an elongated, and therefore migratory cell morphology, in addition to the formation of cell–cell contacts. In support of this, N-cadherin knockdown cells, which still retained some residual N-cadherin (KD2), adopted an elongated morphology, and formed some contacts with neighboring cells. Despite the ability to form cell–cell adhesion in suspension (Fig. 3B), the partial knockdown cells did not develop large cell clusters in a 3D matrix. These cells migrated at a similar speed to wild-type HGF-treated cells (Fig. 3D), but did not migrate as far (Fig. 3E). Similarly, on a 2D coverslip, N-cadherin knockdown cells migrated slower than wild-type cells in an N-cadherin level dependent manner (supplementary material Fig. S4C). These data indicate that the N-cadherin level is critical for both promoting the formation of mechanically resilient cell–cell contacts and the migration potential on a 2D substrate and in a 3D matrix.

To examine the relative contribution of N-cadherin domains in cell–cell adhesion and cell migration, we generated knockdown cells transfected with tomato-tagged full-length N-cadherin (Ncad), N-cadherin with the cytoplasmic domain deleted (Δc746), N-cadherin lacking the β–catenin binding domain (Δc841), N-cadherin extracellular domain fused with the actin-binding domain of α-catenin (Δc746α509), or N-cadherin with the extracellular domain deleted (Ncyto) (Fig. 4A). These mutant constructs were stably expressed in HGF-treated MDCK cells with shRNA depleted endogenous N-cadherin (<3%, Fig. 4B, supplementary material Fig. S5A), and properly localized to cell–cell contacts (supplementary material Fig. S5B), and β1-integrin levels were similar in all cell lines analyzed except the Ncyto expressing cell line (supplementary material Fig. S6).

In the absence of extracellular matrix, the cells expressing exogenous N-cadherin Δc746, Δc841, or full-length N-cadherin aggregated in suspension similar to wild-type HGF-treated cells (Fig. 4C), suggesting that the N-cadherin extracellular domain alone is sufficient for cell aggregation. In contrast, cells expressing Ncyto did not aggregate in a suspension, due to the efficient knockdown of endogenous N-cadherin and the absence of the extracellular domain of N-cadherin. Despite the lower N-cadherin Δc746α509 protein level compared to N-cadherin Δc746 (Fig. 4B), cells expressing N-cadherin Δc746α509 formed larger aggregates than others.

In a 3D matrix, N-cadherin Δc746 and Δc841 expressing cells formed smaller clusters compared to wild-type HGF-treated cells or the full-length N-cadherin rescued cells (Fig. 5A,B). This indicates that neither the β-catenin binding domain, nor the entire cytoplasmic domain is required for the formation of small cell clusters, and for promoting migration in a 3D matrix. The Ncyto expressing cells remained as single cells in the matrix, but formed long, thin membrane extensions, and adopted an elongated, migratory morphology (Fig. 5A,B; a low circularity index indicates the elongated phenotype – see also Materials and Methods). Since cells expressing either the cytoplasmic or the extracellular domain of N-cadherin were both able to form an elongated cell shape in the 3D matrix, these two opposing domains of N-cadherin can independently regulate the formation of the elongated cell shape. N-cadherin Δc746α509 expressing cells formed large, elongated cell clusters similar to wild-type or exogenous full-length N-cadherin expressing cells (Fig. 5A,B), suggesting that the artificial N-cadherin-actin linkage promotes cell–cell adhesion in suspension and a 3D matrix.

In a 3D matrix, the leader cells of cell clusters expressing N-cadherin Δc746α509 and full-length N-cadherin migrated at a similar speed (Fig. 6A) and traveled a similar distance to wild-type cells (Fig. 6B). N-cadherin Δc746 or Δc841 expressing cells migrated slightly faster (Fig. 6A), but traveled a similar end-to-end displacement to the wild type, and N-cadherin Δc746α509 or full-length N-cadherin expressing cells (Fig. 6B). In contrast, single cells expressing Ncyto migrated at a similar speed to others (Fig. 6A), but had a shorter end-to-end displacement than the other clustered N-cadherin mutant rescued and wild-type cells (Fig. 6B). This is consistent with the observation that cell–cell adhesion between the leader cells and follower cells induces directional persistency (Fig. 1I). This migration analysis suggests that the expression of the N-cadherin cytoplasmic domain alone is sufficient to induce single cell migration, but in the absence of the adhesive extracellular domain of this construct, the expression of the N-cadherin cytoplasmic domain alone is not sufficient for collective, persistent cell migration.

Although the N-cadherin mutants expressing cells migrated at the similar speed as the wild-type cells, these mutants did not completely rescue the migration phenotype of wild-type cells. In the 3D matrix, these cells often moved away from the cell clusters with trailing thin membrane connections that often detached (Fig. 6D). Cell detachment rates of N-cadherin Δc746 and Δc841 expressing cells were higher than wild-type or full-length N-cadherin expressing cells (Fig. 6C,D; supplementary material Movie 6). Since β1 integrin levels remained constant in N-cadherin Δc746 and Δc841 expressing cells compared to wild-type HGF-treated cells (supplementary material Fig. S6), the higher rates of cell detachment observed in cells expressing the extracellular domain of N-cadherin (Fig. 6C) are unlikely due to increased cell-matrix adhesion. Rather, the high cluster detachment rates may be due to the lack of N-cadherin cytoplasmic domain. In contrast, N-cadherin Δc746α509 expressing cells formed tight clusters with a similar cell detachment rate from the cluster as wild-type or full-length N-cadherin expressing cells (Fig. 6C). While N-cadherin Δc746 and Δc841expression is sufficient to induce the formation of cell aggregates (Fig. 4C) and small cell clusters in the matrix (Fig. 5A,B), these cell–cell junctions are not strong enough to prevent cell detachment from the highly migratory cluster.

The artificial linkage between N-cadherin and the actin cytoskeleton in N-cadherin Δc746α509 expressing cells appears to be sufficient for collective cell migration, but with one key distinction. We compared the velocities of follower cells with that of leader cells in the same cell clusters in a 3D matrix. The follower-leader cell velocity ratio was about 1.5 for wild-type cells (Fig. 6E). This value indicates that follower cells typically migrated faster than leader cells of the same cell cluster by rapidly migrating along neighboring cells. The average ratios of the follower-leader cell velocities of exogenous N-cadherin expressing cells were all less than that of wild-type cells (Fig. 6E). The cytoplasmic deleted mutants form smaller cell clusters with higher cell detachment rates from the cluster compared to wild-type cells, thus these cells do not have other neighboring cells to crawl along, and the leader and follower cells migrate approximately at the same speed (Fig. 6E). In contrast, N-cadherin Δc746α509 expressing cells formed large cell clusters similar to wild-type cells, yet, the follower cells did not migrate as fast as wild-type cells, suggesting that the chimeric N-cadherin Δc746α509 proteins and its junctions are hindering follower cell movement along the neighboring cells.

The failure of N-cadherin mutants to rescue the collective cell migration is in part due to the inability to organize actin network. Wild-type cells were enriched with actin bundles propagating across cell–cell junctions (Fig. 2C; Fig. 7A). The actin bundles in N-cadherin Δc746 and Δc841 expressing cells were less frequently observed, and were typically thinner (Fig. 7A), therefore, the cytoplasmic domain of N-cadherin is required for robust actin bundle formation. Since N-cadherin Δc746 and Δc841 expressing cells frequently detached from the cluster in a 3D matrix (Fig. 6D), these actin bundles may be required for mechanically resilient cell–cell adhesion between migrating cells.

The actin bundles in N-cadherin Δc746α509 expressing cells were similar to wild-type cells (Fig. 7A). Despite the robust actin network in N-cadherin Δc746α509 expressing cells, the artificial N-cadherin and actin linkage hindered follower cell movement (Fig. 6E). These actin networks may be artificially anchoring the N-cadherin complex to the sites of cell–cell contacts and preventing the follower cells to move along the neighboring cells. To test this hypothesis, FRAP analysis was used to selectively photobleach tomato-tagged N-cadherin and probe the stability of N-cadherin at the sites of cell–cell adhesion. Both N-cadherin Δc746 and Δc841 proteins had similar mobile fractions as full-length N-cadherin (Fig. 7B,C,D), suggesting that the trans-interactions of N-cadherin extracellular domain restrict the diffusion of N-cadherin at cell–cell contacts. However, the N-cadherin Δ746α509 protein had a decreased mobile fraction compared to full-length N-cadherin (Fig. 7B,C,D), suggesting that the artificial interaction between N-cadherin and the actin cytoskeleton decreases the N-cadherin mobility. Together, these data highlight that a dynamic N-cadherin and actin linkage is essential for efficient collective cell migration.

Our analysis of 3D cell migration revealed that cell–cell adhesion is required for persistent, collective cell migration in a 3D matrix. On a stiff surface, transformed epithelial cells are highly migratory, and cell–cell adhesion appears to be too weak to maintain cell–cell contacts of migratory cells. In a 3D collagen matrix, however, the transformed epithelial cells are able to maintain cell–cell contacts through N-cadherin-mediated cell–cell adhesion, which in turn increases the directional persistence of cell migration. Our results suggest that, rather than hindering cell movement, N-cadherin mediated cell–cell adhesion promotes cell migration in a 3D matrix.

Cell polarity of migrating cells in a 3D matrix is established in part by N-cadherin-mediated cell–cell adhesion. We found that transformed cells develop membrane extensions from the contact free polarized cell edge (Fig. 1C–E), and loss of cell–cell contacts causes the formation of random cell protrusions (Fig. 2A). Membrane extensions are suppressed in cells that are in contact with at least two neighboring cells at the anterior and posterior ends, thus resulting in cell polarization, and collective directional migration. Similarly, during neural crest cell migration, N-cadherin-mediated cell–cell adhesion has been shown to suppress membrane extensions by inhibiting Rac1 activity, so that the protrusions develop from the contact free edge (Theveneau et al., 2010). In addition, N-cadherin-mediated cell–cell adhesion is required for cell polarization and directional migration of vascular smooth muscle cells (Sabatini et al., 2008). Therefore, this N-cadherin-mediated contact inhibition may be what distinguishes the roles of leader vs. follower cells during collective migration of N-cadherin expressing cells.

The role of N-cadherin during collective cell migration is not limited to the establishment of cell polarity. Since the follower cells in a cell cluster migrate with faster speed compared to the leader cells (Fig. 1H), N-cadherin junctions on the surface of neighboring cells serve as a migration track. This N-cadherin junction dependent cell movement is also observed in cortical neuronal locomotion along radial glial fibers (Shikanai et al., 2011). Previous studies have shown that cells exert similar magnitudes of traction forces through N-cadherin junctions as focal adhesions (Ganz et al., 2006), suggesting that N-cadherin junctions are not passive contacts that maintain multicellular integrity, but rather act as an active participant of cell migration.

Interestingly, the shRNA depletion of N-cadherin reduces cell migration (Fig. 3). This is consistent with previous observations that exogenous expression of N-cadherin induces migration of some cancer cells (Nieman et al., 1999). This N-cadherin induced cell migration is not solely due to the presence of N-cadherin, since the overexpression of N-cadherin in normal MDCK cells does not induce cell migration (our unpublished observations), but likely depends on other genes that are up-regulated during EMT. Previous studies have shown that N-cadherin interacts with the FGF-receptor via the EC4 domain of N-cadherin (Sanchez-Heras et al., 2006). The N-cadherin–FGF-receptor mediated signaling may explain why the cytoplasmic deleted mutant expressing cells are able to migrate in a 3D matrix, albeit with frequent cell detachment from the cell cluster (Fig. 6C). In contrast, N-cadherin depleted cells expressing only the N-cadherin cytoplasmic domain, migrate in a 3D matrix, albeit as single cells (Fig. 6A). Previous studies also have suggested that the cytoplasmic domain of cadherins, which are highly conserved among classical cadherins, may contain a signaling sequence, which promotes motility (Fedor-Chaiken et al., 2003; Pacquelet and Rørth, 2005). These data suggest that N-cadherin extracellular and intracellular domains each induce cell migration potential by independent mechanisms. Currently, how the cytoplasmic domain of N-cadherin induces cell migration is not known.

In post EMT cells, N-cadherin junctions organize the actin cytoskeleton into cortical actin at the plasma membrane and cell–cell contacts, and actin bundles that are oriented parallel to the long-axis of migratory cells. The actin bundles are aligned in the direction of traction force observed in a 3D matrix (supplementary material Fig. S3), and inhibition of ROCK resulted in traction force generation and the loss of actin bundles (supplementary material Fig. S3), therefore these actin bundles likely provide contractile forces. These actin bundles terminate at N-cadherin cell–cell junctions, similar to contractile actin bundles at cell–cell junctions in vitro (Vasioukhin et al., 2000; Yamada and Nelson, 2007; Yonemura et al., 1995), wound healing myofibroblasts in vivo (Hinz et al., 2001), or high fluid shear regions of blood vessel (Wong et al., 1983). Other cancer cells in a 3D matrix lacked the extensive actin bundles (Fraley et al., 2010; Hegerfeldt et al., 2002), and this is likely due to the absence or lower level of N-cadherin junctions as the levels of N-cadherin is critical for the actin organization and collective cell migration (Fig. 3, Fig. 7A).

The artificial actin recruitment at cell–cell contacts is sufficient for the organization of actin bundles. However, the N-cadherin Δc746α509 expressing cells are more prone to cell aggregation (Fig. 4C) and the N-cadherin Δc746α509 proteins at cell–cell junctions are less mobile than full-length N-cadherin (Fig. 7B–D), suggesting that the interaction between N-cadherin and the actin network is not a direct and stable linkage. Similar to previous observations of E-cadherin (Drees et al., 2005; Yamada et al., 2005), these data imply that N-cadherin junctions are dynamic complexes that transiently interact with the underlying actin network to promote cell migration within cell clusters. This artificial linkage by the N-cadherin Δc746α509 reduces the speed of follower cells in 3D cell clusters (Fig. 6E) and intercellular movement in a monolayer (Nagafuchi et al., 1994). The artificial recruitment of actin to N-cadherin Δc746α509 junctions, therefore, is not dynamic enough to support cell movement along neighboring cells within a cluster. Efficient collective migration depends on a delicate balance between cell–cell adhesion that is strong enough to prevent cell detachment from the cluster, yet dynamic enough to support cell movement between and along neighboring cells.

The up-regulation of N-cadherin in aggressive carcinomas suggests that the level of N-cadherin is a critical parameter for cancer cell invasion. Previous studies have shown that N-cadherin mediates transendothelial migration of cancer cells, a key step in metastasis (Drake et al., 2009; Qi et al., 2005). Our results demonstrate that N-cadherin is important for promoting migration even before cancer cells reach the vasculature. N-cadherin targeted therapeutic strategies have been shown to minimize N-cadherin positive tumor growth (Li et al., 2007) and metastasis in a xenograft of prostate cancers (Tanaka et al., 2010). Our results provide the mechanistic details of how N-cadherin promotes cancer cell invasion and the explanations for why N-cadherin targeted therapy might be effective. Although the therapeutic agents targeting the extracellular domain of N-cadherin may not prevent single cancer cell invasion, breaking down cell–cell adhesion between cancer cells will effectively minimize collective cell migration and may extend the time required for cancer cell dissemination.

MDCK GII cells were cultured in DMEM (Invitrogen, Carlsbad, CA) supplemented with 10% FBS (Atlanta Biologicals, Lawrenceville, GA) and 5 ng/ml of HGF (Sigma-Aldrich, St Louis, MO). The MDCK cells were treated with HGF for at least two weeks, and selected for complete EMT. Subsequent removal of HGF from post-EMT cells caused N-to-E-cadherin switch, thus HGF-induced EMT was reversible. The higher concentration of HGF (20 ng/ml) was added to a collagen gel to increase the penetration of HGF into the gel. For western blot or immunofluorescence applications, the antibodies used were E-cadherin (BD Biosciences, Franklin Lakes, NJ), N-cadherin (BD Biosciences, Franklin Lakes, NJ), α-catenin (Enzo Lifesciences, Farmingdale, NY), fibronectin (BD Biosciences, Franklin Lakes, NJ), tubulin (Sigma), T-cadherin (Anaspec, San Jose, CA), Cadherin 11 (R&D Systems, Minneapolis, MN), K-cadherin (a gift from James Nelson, Stanford University), desmoplakin I/II (a gift from James Nelson), DsRed (Invitrogen, Carlsbad, CA) and β1-integrin (Millipore, Billerica, MA). Filamentous actin was labeled with phalloidin (Invitrogen). For western blot, the signals on the nitrocellulose membrane were detected by chemiluminescence with an enhanced ECL reagent (Pierce Biotechnology). Pharmacological inhibitors used were 100 μM Y-27632 (EMD Chemicals) and 10 μM Cytochalasin D (EMD Chemicals).

Canine N-cadherin shRNA targeting the sequences: #1: 5′-AGTATTCCCAAGACAAGCG-3′; #2: 5′-TGAACGGGCAAATAACAAC-3′; #3: 5′-AACGCCGAGGTAAAGAACG-3′; and #4: 5′-GATCGATATATGCAGCAAA-3′, were inserted into the pSuper.neo+gfp vector (Oligoengine, Seattle, WA). Lipofectamine 2000 and G418 (both from Invitrogen, Carlsbad, CA) were used to generate 35 stably transfected HGF-treated MDCK clones. Rescue N-cadherin mutants were derived from GFP-tagged N-cadherin (a gift from Kathleen Green, Northwestern University) and the GFP tag was replaced with a tandem dimer Tomato fluorescent protein. We created the following constructs: mouse N-cadherin with only the extracellular domain AA1-746 (N-cadΔc746), N-cadherin lacking the β-catenin binding domain AA1-841 (N-cadΔc841), N-cadherin Δc746 fused with α-catenin (AA 509-906), N-cadherin cytoplasmic domain AA747-906 with a membrane targeting sequence from lyn protein (Ncyto), and the full-length N-cadherin. The rescue constructs were double transfected with a N-cadherin shRNA plasmid.

All samples were imaged on a Zeiss Axio Observer equipped with a Yokogawa spinning disk confocal system, a 10× and 40× objective, 488 and 561 nm solid-state lasers, and a CoolSNAP HQ camera. The microscope system was controlled and automated by Slidebook software (Intelligent Imaging Innovations, Denver, CO). Live cells were imaged on glass bottom dishes (MatTek, Ashland, MA) in a temperature-controlled chamber at 37°C.

FRAP analysis was performed on cells expressing tomato-tagged N-cadherin mutants. A small region (∼1 µm²) at cell–cell contacts was selectively photobleached, and the fluorescence intensity recovery was analyzed over time. Using Excel statistical analysis software (Microsoft, Redmond, WA), the recovery curve was fitted to an exponential equation, I = I_f (1−e^−kt), where I is the intensity, I_f is the final intensity, t is time. The mobile fraction of the protein is the ratio of the final and initial intensity. A one-way analysis of variance (ANOVA) with Dunnett's post-hoc test was performed to compare the mobile fraction of each N-cadherin mutant protein with that of full-length N-cadherin. The difference was assumed to be statistically significant when P<0.05.

For three-dimensional migration assays, cells were embedded in a 1 mg/ml collagen I matrix (BD Biosciences, Franklin Lakes, NJ) as previously described (Shih and Yamada, 2010). To chelate calcium, 0.5 mM EDTA was used to treat the cells. Note that the concentration of EDTA used did not have any effects on cell spreading and attachment to a collagen-coated 2D surface (Shih and Yamada, unpublished data), therefore this disruption of cell–cell contacts and random cell invasion are not the consequences of altered cell-ECM adhesion, but are unique effects of EDTA on calcium dependent cell–cell adhesion. Fluorescent tracer particles (1 µm diameter) were added to the matrix to visualize the matrix deformation (Shih and Yamada, 2010).

To track individual cells within a cluster, HGF-treated cells were stably transfected with GFP or GFP-tagged histone, and GFP-expressing cells were mixed with non-expressing cells before seeding in the matrix. Cells were imaged over 12 hours, and individual GFP-positive cells were tracked using Slidebook software (Intelligent Imaging Innovations, Denver, CO). Leader cell velocities were obtained by tracking the leading edge of the cluster, and the follower-leader velocity ratio was obtained by dividing the velocity of the follower cell by the velocity of the leader cell for that particular cluster. End-to-end displacements were calculated based on the initial and final cell position in 12 hour time-lapse images. All cell trajectories were quantified in Slidebook software (Intelligent Imaging Innovations, Denver, CO) and analyzed in Excel (Microsoft, Redmond, WA). The number of detachments per cell cluster was counted, and normalized to the cluster size and the imaging duration. A one-way analysis of variance (ANOVA) with Dunnett's post-hoc test was performed in (Fig. 1I), and the difference was assumed to be statistically significant when P<0.05. Statistical significance between wild-type and Ncyto in Fig. 6B was determined using z-test with P<0.05.

Cell movement was analyzed in the directions parallel and perpendicular to the primary axis of migration. For clusters, this reference direction was the long axis of the cluster, while for single cells, the reference direction was the initial long axis of the individual cell as determined by an ellipsoid fit. The cluster area of cells in the matrix was determined by thresholding the image in ImageJ (http://rsb.info.nih.gov/ij), and quantifying the resulting cluster area of each object. The circularity index for each thresholded object was calculated by taking the ratio of area and perimeter (4π area/perimeter²), in which an object with perfect circle has an index of 1.

For cell clustering analysis, cultured cells were dissociated with trypsin (Invitrogen, Carlsbad, CA) supplemented with 1.8 mM calcium to preserve cell surface receptors. Total volume of 25 µl containing approximately 250,000 cells were hanged upside-down from the lid of the culture dish, and allowed to aggregate for specified time. Some cell lines with strong cell–cell adhesion (e.g. wild-type N-cadherin or N-cadherin Δc746α509 rescue cells) were difficult to dissociate under these conditions and initially contained single cells and few cell clusters (Fig. 4C). Every hour, the cell solution was triturated 5 times, then imaged. The average aggregate size was determined by object thresholding using ImageJ. Hanging drop results were analyzed using the Kruskal-Wallis test statistic with Dunn's post-hoc test. The difference was assumed to be statistically significant when P<0.01.

We thank James Nelson (Stanford University) for sharing the K-cadherin and desmoplakin antibodies, Kathleen Green (Northwestern University) for the N-cadherin-GFP plasmid, Tony Ronco for help with cell aggregation analysis.