Polysialic acid controls NCAM signals at cell–cell contacts to regulate focal adhesion independent from FGF receptor activity

The balanced regulation of cell adhesion is crucial for coordinated cell motility in development, whereas dysregulated adhesion is a hallmark of tumour progression (Christofori, 2003). In particular, the cross-talk between signals induced by cell–cell contact and the control of cell–matrix interactions is essential for the different forms of multicellular streaming and collective cell migration (Friedl and Wolf, 2010). However, little is known on mechanisms linking cell–cell and cell–matrix adhesion to adjust cellular motility.

The neural cell adhesion molecule NCAM, a recognition molecule of the immunoglobulin superfamily plays a pivotal role in cell–cell interactions, as studied mainly in nervous system development (Maness and Schachner, 2007), but also modulates matrix adhesion of tumour cells (Cavallaro et al., 2001). Alternative splicing generates three major NCAM isoforms (NCAM120, NCAM140 and NCAM180) (Cunningham et al., 1987). These differ in their transmembrane and intracellular domains, but have identical extracellular parts composed of five amino-terminal immunoglobulin-like domains (Ig1–Ig5) followed by two fibronectin type 3 modules (FnI, FnII), and therefore can interact with the same extracellular binding partners. Initially, NCAM was thought to exert only homophilic binding (Hoffman and Edelman, 1983), but later numerous heterophilic cis and trans interactions were identified, e.g. with other CAMs of the immunoglobulin superfamily such as TAG1 or L1, cellular prion protein and ligands of the GDNF family and the GDNF family receptor alpha-1, but also with heparan and chondroitin sulphate proteoglycans of the extracellular matrix through heparin-binding sites localised to the Ig2 domain of NCAM (for a review, see Nielsen et al., 2010).

The most studied extracellular interaction partners in terms of NCAM function are members of the FGF receptor family (Maness and Schachner, 2007; Kiselyov, 2010). Activation of FGF receptors is mainly implicated in neurite outgrowth in response to homophilic NCAM trans interactions (Saffell et al., 1997; Niethammer et al., 2002; Kiselyov et al., 2003), but also contributes to the cell-autonomous modulation of matrix adhesion by an NCAM-dependent signalling complex in pancreatic tumour cells (Cavallaro et al., 2001). More recently, FGF-receptor-dependent promotion of cell migration was demonstrated by application of soluble NCAM to NCAM-negative cells (Francavilla et al., 2009). Other functions of NCAM as a signalling receptor are independent of interactions with FGF receptors. Activation of the src-family kinase Fyn and subsequent recruitment of the focal adhesion kinase (FAK) to NCAM140 depends on lipid raft association of NCAM and complements FGF receptor signalling in neuritogenesis induced by homophilic NCAM interactions (Beggs et al., 1997; Niethammer et al., 2002). Similarly, translocation of NCAM from FGF receptor complexes to lipid rafts and activation of Fyn was observed after up-regulation of NCAM in response to a loss of E-cadherin, whereas knockdown of NCAM caused a loss of focal adhesion and enhanced migration of cells with a mesenchymal phenotype (Lehembre et al., 2008).

Polysialic acid (polySia) is a major determinant of NCAM binding but also a general modulator of cell–cell interactions (Rutishauser, 2008; Hildebrandt et al., 2010). This unusual polymeric sugar can be added to N-glycosylation sites within the fifth immunoglobulin domain of NCAM by two polysialyltransferases, ST8SiaII (STX) and ST8SiaIV (PST), which exhibit a high specificity for the acceptor protein (Colley, 2010). PolySia, therefore, is confined to a small subset of proteins, with NCAM being by far the most abundant carrier of polySia in most mammalian cells (Colley, 2010; Galuska et al., 2010; Hildebrandt et al., 2010). Although polySia is diminished in the majority of tissues during development, various tumours are known to re-express polySia on NCAM and high polySia levels have been correlated with malignant potential and poor prognosis of small cell lung carcinoma, neuroblastoma, glioblastoma, medulloblastoma and rhabdomyosarcoma (Scheidegger et al., 1994; Figarella-Branger et al., 1996; Glüer et al., 1998a; Glüer et al., 1998b; Hildebrandt et al., 1998; Tanaka et al., 2000; Daniel et al., 2001; Amoureux et al., 2010). In the developing nervous system, polySia is crucially involved in the migration of neuronal precursors (Ono et al., 1994; Hu et al., 1996; Chazal et al., 2000; Weinhold et al., 2005; Angata et al., 2007; Burgess et al., 2008) and modulates the responsiveness of oligodendrocyte precursor to chemotactic migration cues (Barral-Moran et al., 2003; Zhang et al., 2004; Glaser et al., 2007). In both settings, polySia seems to affect motility independently of specific functions of its protein carrier NCAM. In these situations polySia might act as a non-specific steric inhibitor of cell–cell apposition or by modulation of chemotactic growth factor sensing. In pancreatic carcinoma cells, enhanced polysialylation of NCAM has been correlated with facilitated migration as a result of reduced E-cadherin-mediated cell–cell aggregation (Schreiber et al., 2008). So far, however, the impact of polySia on NCAM-induced or NCAM-independent modulation of motility and cell–matrix interactions of tumour cells has not been directly addressed. Here, we demonstrate that loss of polysialic acid leads to reduced migration and increased focal adhesion of tumour cells. Both effects were found to require that cells were in contact with each other and occurred in an NCAM-dependent manner but independent from FGF receptor activity.

The role of polySia in migration of tumour cells was studied in vitro with two-dimensional scratch wound assays. In a first approach, an EGFP-transfected clone of the polySia- and NCAM-positive neuroblastoma cell line SH-SY5Y was used (SH-SY5Y_EGFP) (Seidenfaden et al., 2003). As illustrated in Fig. 1A, the number of cells populating the cell-free area within 6 hours was significantly reduced in the presence of endo-N-acetylneuraminidase (endosialidase; endo), which reliably removes polySia from the surface of living cells (Seidenfaden et al., 2003) (supplementary material Fig. S1). This observation was confirmed by analysing the area covered by newly arrived cells in the scratch wound (Fig. 1B; for details, see Materials and Methods and supplementary material Fig. S2). In contrast to these results with cells that move in close contact with each other, we found that removal of polySia had no influence on the motility of single cells (Fig. 1C). Tracing of control and endo-treated cells was performed from time-lapse recordings with frames taken every 2 minutes (supplementary material Movies 1 and 2, and Fig. S3). Because this calculation of cell velocity strongly depends on the time interval between observations, evaluation was repeated using time intervals of 20 and 100 minutes for tracing the cells. Owing to a low persistence of direction in both the control and endo-treated group, the calculated velocities strongly decreased with increasing observation intervals. However, as for the 2 minute interval (Fig. 1C), no difference between the control and endo-treated group was obtained for the two longer observation intervals [velocities were 70.4±6.9, 69.7±6.3, and 38.1±4.0, 38.6±5.1 μm/hour (means ± s.e.m.) for control and endo-treated cells (n=16) measured at 20 and 100 minute intervals, respectively].

As described before, SH-SY5Y cells express the two transmembrane isoforms of NCAM, NCAM140 and NCAM180, and the entire NCAM pool is polysialylated (Seidenfaden et al., 2000; Seidenfaden and Hildebrandt, 2001). To corroborate that the effect of endo treatment was indeed caused by removal of polySia from NCAM, polysialylation of two recently identified alternative polySia acceptors, neuropilin-2 and SynCAM 1 (Curreli et al., 2007; Galuska et al., 2010) was analysed. Affinity isolation revealed that both proteins were present in SH-SY5Y cells, but neither of them was immunopositive for polySia (supplementary material Fig. S4).

Further scratch wounds assays were performed with non-transfected SH-SY5Y and other cell lines using image analysis of high contrast bright-field images (see Materials and Methods for details and supplementary material Movies 3 and 4 and Fig. S5 for an example). Consistent with the results described above, a significant reduction of cells repopulating a scratch wounded, cell-free area indicated reduced migration after removing polySia from SH-SY5Y (Fig. 2A) as well as from Kelly (neuroblastoma) and TE671 (rhabdomyosarcoma, Fig. 2B) cells, two other polySia- and NCAM-positive tumour cell lines (Seidenfaden et al., 2000). As with SH-SY5Y_EGFP (Fig. 1), analyses were restricted to a 6-hour time window to avoid a major bias due to the slightly reduced cell proliferation in response to polySia removal or NCAM exposure (Seidenfaden et al., 2003).

To determine whether changes in NCAM binding abilities might be responsible for reduced migration in response to endo treatment, we used the NCAM peptide ligand C3d (Ronn et al., 1999) and a non-polysialylated NCAM-Fc chimera to interfere with or to mimic interactions of polySia-free NCAM. C3d was used at 1 μM, a concentration that has been shown to either mimic NCAM functions or to inhibit ongoing NCAM interactions (Ronn et al., 1999; Ronn et al., 2000). In particular, C3d is known to bind to NCAM in the absence of polySia and thus prevents the formation of other NCAM contacts (Seidenfaden et al., 2003; Röckle et al., 2008). In the presence of endo, i.e. after loss of polySia, C3d improved the migration of SH-SY5Y cells (Fig. 2A). By contrast, C3d had no effect on migration if added to polySia–NCAM-positive SH-SY5Y (Fig. 2A) or to native LS neuroblastoma cells (Fig. 2C), which are negative for polySia and NCAM (Seidenfaden et al., 2003). Migration of LS cells, however, was attenuated by the addition of NCAM-Fc indicating that these cells respond to heterophilic NCAM contacts (Fig. 2C; supplementary material Movies 3 and 4, and Fig. S5). The assumed modulation of migration by polySia-negative NCAM was substantiated by the observation that addition of C3d to two different NCAM140-expressing, polySia-negative LS clones significantly enhanced the repopulation of a scratch wound (Fig. 2D). Finally, reduced migration after endo treatment and its reversal by the C3d peptide was confirmed with LSAM1PST transfected to express polysialylated NCAM140 (Fig. 2E). Current and previous observations indicate that polysialylated NCAM and, after endo treatment, non-polysialylated NCAM are concentrated at sites of cell–cell contact (Seidenfaden et al., 2003) (see Fig. 3D,E). This localisation and the results from the scratch wound assays suggest that reduced migration after removal of polySia is due to enhanced NCAM-mediated cell–cell contacts.

Cell migration requires the continuous formation and disassembly of adhesions to transmit motion generated by the actin cytoskeleton to the extracellular environment (Webb et al., 2002; Geiger et al., 2009; Parsons et al., 2010). We therefore analysed changes of actin-associated focal adhesions as a measure for altered cell–substrate adhesiveness and a potential cause of altered cell motility after endo treatment or exposure to soluble NCAM-Fc. The polySia–NCAM-positive clone LSAM1PST and parental, polySia- and NCAM-negative LS cells appeared particularly suited for such analyses. These cells display a clearly discernible pattern of peripheral focal adhesions, characterized as FAK, and paxillin immunoreactive spots located at the tip of actin fibres (see Fig. 3A,G for FAK, and supplementary material Fig. S6 for paxillin).

Consistent with the data obtained by migration assays, LSAM1PST cells responded to endo treatment with a significant increase in the number of peripheral focal adhesions per cell (Fig. 3A–C). Application of a mutant, enzymatically inactive variant of endo had no effect. Notably, the endo-induced increase was observed only in cells that were in contact with each other and not in isolated cells, which showed the same amounts of focal adhesions as cells in contact under control conditions (Fig. 3C). Independent of the presence or absence of polySia, NCAM was enriched at cell–cell contacts and never colocalised with focal adhesions (Fig. 3D,E). Application of the C3d peptide, which interferes with NCAM binding, completely blocked the cellular response to the enzymatic removal of polySia by endo (Fig. 3F). The observation that NCAM-induced modulation of cell motility is tightly linked to altered focal adhesion was further substantiated by experiments with NCAM-negative LS cells. Both isolated LS cells and LS cells in contact with each other, showed an increased focal adhesion upon exposure to soluble NCAM-Fc (Fig. 3G–I). Finally, mixed co-cultures of EGFP-transfected, polySia–NCAM-positive LSAM1PST cells (LSAM1PSTegfp) and NCAM-negative LS cells were treated with endo followed by an evaluation of focal adhesions of EGFP-negative LS cells in contact with EGFP-positive LSAM1PSTegfp, and vice versa (Fig. 3J–L). Under these conditions, focal adhesion was increased exclusively in NCAM-negative cells. This outcome provides direct evidence that unmasking NCAM by enzymatic removal of polySia instructs neighbouring cells to form more focal adhesions. Since these neighbouring cells do not need to express NCAM themselves to respond to the NCAM signal, focal adhesion must be induced by heterophilic NCAM binding.

The data presented so far strongly argue that NCAM interactions at cell–cell contacts induce reduced cell motility by increased focal cell–substrate adhesion. This raises the question of which matrix components might be involved. When scratch wound migration assays were performed on standard cell culture plastic, LS cells efficiently organized a fibronectin-containing extracellular matrix (Fig. 4), which assembled independently of any additional fibronectin coating of the plastic surface (compare Fig. 4B,C). Cell-associated fibronectin did not colocalise with peripheral focal adhesions, visualized by FAK immunoreactivity, but frequently aligned with FAK-positive streaks, indicative of fibrillar adhesions (for a review, see Parsons et al., 2010) (Fig. 4D–F). Consistent with the observation that substrate coating of plastic surfaces had no significant effects on the assembly of cell-associated fibronectin, only minor alterations in cell migration were seen in scratch wound assays that were performed on fibronectin-coated plates and migration could still be attenuated by addition of NCAM-Fc to NCAM-negative LS cells and by enzymatic removal of polySia from polySia–NCAM-positive LSAM1PST cells (Fig. 4G,H). Thus, exposure of polySia-free NCAM attenuates cell migration on fibronectin, but the segregation of fibronectin-containing adhesions and peripheral focal adhesions suggests a modulation of fibronectin-independent cell-substrate interactions.

NCAM interactions with heparin-like molecules are involved in neuronal cell–cell and cell–substrate adhesions (Cole et al., 1986; Cole and Glaser, 1986) and NCAM-induced inhibition of glioma cell motility is modulated by interference with heparin or heparan sulphate proteoglycan (HSPG) binding (Prag et al., 2002). Moreover, polySia promotes NCAM binding to HSPGs and these interactions are sensitive to digestion of heparan sulphates by combined heparinase I and III treatment (Storms and Rutishauser, 1998). Indeed, migration of the polySia–NCAM-positive LSAM1PST cells was significantly enhanced by application of heparinase I and III, but the same effect was observed with parental, polySia- and NCAM-negative LS cells (Fig. 4I,J). In addition, heparinase treatment did not prevent either the inhibition of migration (Fig. 4I,J) or the increase of focal adhesion after treatment with polySia-negative NCAM-Fc or in response to enzymatic removal of polySia (Fig. 4K,L). The migration promoting effects of heparinase, therefore, were not related to altered migration and focal adhesion induced by NCAM application or polySia removal.

The absence of NCAM from the sites of focal adhesion together with the promotion of focal adhesion by soluble NCAM indicates that these effects are not caused by direct adhesive interactions of NCAM but are mediated by a cellular signalling cascade. Together with the dynamic modulation of the actin cytoskeleton, the src-family kinase Fyn is a well-established determinant of focal adhesion turnover (Webb et al., 2002; Mitra et al., 2005; Schaller, 2010). To demonstrate that focal adhesions of LS cells depend on both components we used the kinase inhibitor PP2 for pharmacological inhibition of Fyn and the Rho-dependent protein kinase (ROCK) inhibitor Y27632 to disrupt actin fibre assembly. As expected, both treatments effectively reduced the number of focal adhesions in LS cells (Fig. 5A–D). Furthermore, co-immunoprecipitation experiments indicated the recruitment of Fyn to the focal adhesion scaffolding protein paxillin after treatment of LS cells with NCAM-Fc or LSAM1PST with endo (Fig. 5E,F).

Engagement of ERK1/2 is a hallmark of NCAM-induced signalling initiated by the loss of polySia (Seidenfaden et al., 2003; Seidenfaden et al., 2006) and triggered, at least in part, by activation of FGF receptors (Kolkova et al., 2000; Cavallaro et al., 2001; Niethammer et al., 2002; Francavilla et al., 2009). In addition, the ERK1/2 pathway is in intimate cross-talk with the regulation of focal adhesion in a stimulus- and cell-type-specific manner (for reviews, see Schwartz and Ginsberg, 2002; Huang et al., 2004; Mitra et al., 2005). We therefore wondered whether the NCAM-induced increase of focal adhesion in LS cells depends on ERK1/2 and FGF receptor activity. Confirming previous findings, exposure of LS cells to soluble NCAM-Fc enhanced, and pre-incubation with the MEK inhibitor PD98059 completely abolished, the fraction of dually phosphorylated, active ERK1/2 (Fig. 5G). Consistent with the observed tyrosine phosphorylation of Fyn associated with paxillin (Fig. 5E, lower panel) NCAM-Fc also induced an increase in tyrosine phosphorylation of Fyn immunoprecipitated from LS cell lysates (Fig. 5H). Surprisingly, however, inhibition with PD98059 was not able to prevent this increase, indicating that activation of ERK1/2 occurs either downstream or independently of Fyn phosphorylation (Fig. 5H).

To investigate the possible involvement of FGF receptor activity in the upregulation of focal adhesion after NCAM exposure or polySia removal we used PD173074, which specifically inhibits signalling of the FGF receptor, leaving other tyrosine kinases unaffected (Skaper et al., 2000). Consistent with recent results of Francavilla et al., pre-incubation with PD173074 inhibited the NCAM-Fc-induced stimulation of ERK1/2 (Fig. 6A) (Francavilla et al., 2009). By contrast, the same protocol of PD173074 pre-treatment was not able to prevent the increase in focal adhesion induced by NCAM-Fc (Fig. 6B). Along the same line, the previously described activation of ERK1/2 after polySia removal with endo (Seidenfaden et al., 2003), but not the increase of focal adhesion, was prevented by pre-incubation of LSAM1PST with PD173074 (Fig. 6C,D).

These results indicate that FGF receptor activity is involved in ERK1/2 activation but not in the modulation of focal adhesions in response to polySia removal or NCAM-induced signals. According to the prevailing model, homophilic NCAM binding (in trans) involves the first two immunoglobulin domains (Ig1, Ig2) (Kiselyov, 2010) and activates FGF receptors by cis interactions involving the two fibronectin (Fn) modules of NCAM (Kiselyov et al., 2003; Christensen et al., 2006). Peptides in either FnI or FnII are able to bind to the FGF receptor and activate ERK1/2 (Kiselyov et al., 2003; Neiiendam et al., 2004; Anderson et al., 2005; Jacobsen et al., 2008; Palser et al., 2009), whereas NCAM-Fc lacking the second Fn module fails to elicit various FGF receptor-dependent responses induced by NCAM-Fc containing the entire NCAM extracellular domain (Francavilla et al., 2007; Francavilla et al., 2009). To elucidate which of these NCAM modules are necessary for specifically stimulating focal adhesion in addition to ERK1/2, a number of different NCAM fragments were created (summarized in Fig. 7A,B). In addition to the entire NCAM extracellular domain (ecd) comprising the five immunoglobulin domains (Ig1–Ig5) and the two Fn modules (FnI and FnII), deletion constructs encoding the following NCAM modules were expressed in insect cells: Ig3 to FnII (3–II), Ig3 to FnI (3–I), Ig5 and FnI (5–I), FnI and FnII (I–II), Ig1 to Ig5 (1–5), Ig3 to Ig4 (3–4). Purified recombinant proteins were adjusted to approximately equimolar concentrations as determined by immunodetection with either the NCAM-specific monoclonal antibody 123C3, which associates with a region including FnI and therefore is able to bind to all fragments containing FnI (Fig. 7A), or with anti-His antibody detecting the N-terminal His tag (Fig. 7B). In a first step, activity of NCAM_ecd was confirmed by dose-dependent activation of ERK1/2 (Fig. 7C) before the other constructs were tested at concentrations of approximately 0.1 μM (corresponding to 10 μg/ml NCAM_ecd; Fig. 7D,E). As summarized in Fig. 7F, all NCAM fragments containing either of the two Fn domains were equally able to activate ERK1/2. In stark contrast, only the fragments including Ig3–Ig4, but not those consisting of only FnI–FnII or Ig5–FnI induced an increase of focal adhesion (Fig. 7G). Thus, Ig1 and Ig2 were not needed for NCAM-induced activation of both ERK1/2 and focal adhesion in NCAM-negative LS cells, whereas FnI or FnII were able to specifically activate ERK. This outcome was expected, as previous work clearly identified peptide motifs in each of the two Fn domains that are capable of activating FGF receptor signalling (Kiselyov et al., 2003; Anderson et al., 2005). Conversely, FnI and FnII were dispensable, and the Ig3–Ig4 region of NCAM was necessary and sufficient to trigger enhanced focal adhesion. On the one hand, this suggests that interactions with chondroitin- or heparan sulphate-containing proteoglycans through the heparin-binding domain in Ig2 were not involved, which is consistent with the inability of heparinase treatment to interfere with the effects of NCAM exposure or polySia removal (see Fig. 4I–L). On the other hand, the observation that FnI-FnII stimulated ERK1/2 but not focal adhesion is in perfect agreement with the assumption that enhanced focal adhesion in response to NCAM is not mediated by activation of the FGF receptor and ERK1/2 signalling.

NCAM-induced activation of the FGF receptor and ERK1/2 signalling can be mimicked by the FG loop region of the NCAM FnII module involved in FGF receptor binding (FGL; Kiselyov et al., 2003; Neiiendam et al., 2004; Francavilla et al., 2009). We therefore used this peptide to further dissect the effects of NCAM on focal adhesion and FGF receptor signalling. A dose-dependent, saturating activation of ERK1/2 was achieved by FGL application (Fig. 8A,B), and in line with previous data (Neiiendam et al., 2004; Francavilla et al., 2009) the FGL-induced ERK1/2 activation was efficiently prevented by pre-incubation with PD173074 (Fig. 8C). The capacity of FGL to promote focal adhesion of LS and LSAM1PST cells was compared with NCAM_ecd. Both reagents were used at concentrations that clearly activated ERK1/2 (1 μg/ml FGL, 10 μg/ml NCAM_ecd). Unlike NCAM_ecd, which promoted focal adhesion in both cell types, the application of FGL had no such effect (Fig. 8D,E). In the absence of polySia-free NCAM, the NCAM-binding C3d peptide (1 μM) also had no effect on focal adhesion. These distinct activities indicate that FGL recapitulates FGFR-dependent functions of NCAM, but is unable to mimic NCAM-induced focal adhesion.

Numerous studies describe a close correlation of polySia–NCAM expression with increased tumour invasion and metastatic potential (Scheidegger et al., 1994; Figarella-Branger et al., 1996; Figarella-Branger et al., 1996; Tanaka et al., 2000; Daniel et al., 2000; Daniel et al., 2001; Trouillas et al., 2003; Suzuki et al., 2005; Amoureux et al., 2010). In all these studies, a role of polySia in promoting tumour cell motility has been inferred from its known role in modulating adhesiveness and migration in the nervous system (Sadoul et al., 1983; Hoffman and Edelman, 1983; Ono et al., 1994; Wang et al., 1994; Hu et al., 1996; Chazal et al., 2000) but the NCAM-dependent or NCAM-independent modulation of tumour cell migration by polySia has been elusive. In the current study we have demonstrated that migration of neuroblastoma and rhabdomysarcoma cells in a two-dimensional scratch assay is promoted by the presence of polySia. This outcome was not unexpected because it has been shown before in similar scratch assays that migration of oligodendrocyte precursors is attenuated after removal of polySia (Barral-Moran et al., 2003) and that overexpression of polySia after transfection with ST8SiaII (STX) promotes migration of Schwann cells and embryonic stem cell-derived glial precursors (Lavdas et al., 2006; Glaser et al., 2007). However, it is surprising that loss of polySia promotes focal adhesion in an NCAM-dependent manner although NCAM is enriched at cell–cell contacts and not localised to focal adhesions. The implication is that removal of polySia promotes focal adhesions by initiating NCAM-induced cellular signalling. This is supported by the observed recruitment of phosphorylated Fyn to the focal adhesion scaffolding protein paxillin in response to polySia removal or application of polySia-negative NCAM. Together, these results suggest a novel role of polySia-regulated NCAM signalling in the crosstalk between cell–cell and cell–matrix adhesion to control cell migration.

Removing polySia from NCAM caused reduced migration and promoted focal adhesion only if cells were in contact with each other. Both effects were prevented by the NCAM-binding peptide C3d, a potent inhibitor of NCAM interactions (Ronn et al., 1999; Ronn et al., 2000; Kiryushko et al., 2003; Kiselyov et al., 2009), and could be recapitulated by exposing NCAM-negative cells to NCAM. Moreover, the loss of polySia from NCAM-positive cells enhanced focal adhesion of adjacent NCAM-negative cells. These data provide strong evidence that removal of polySia initiates interactions of NCAM as a heterophilic ligand at cell–cell contacts. This mechanism is consistent with earlier studies showing that polySia controls instructive NCAM signals and that heterophilic NCAM interactions can direct the differentiation of NCAM-negative neuroblastoma cells and of neural progenitors derived from NCAM-deficient mice (Amoureux et al., 2000; Seidenfaden et al., 2003; Röckle et al., 2008). Along the same lines, some of the phenotypic traits of mice with complete or partial ablation of polySia are not caused by the reduced amounts of polySia itself but by the untimely appearance of polySia-negative NCAM (Weinhold et al., 2005; Hildebrandt et al., 2009). By contrast, the NCAM-dependent mode by which polySia regulates tumour cell motility clearly differs from the role of polySia as a permissive factor in neuronal migration. Streaming of interneuron precursors from the subventricular zone towards the olfactory bulb is disturbed by either ablation of polySia alone or by a loss of polySia as a result of NCAM deficiency (Ono et al., 1994; Hu et al., 1996; Chazal et al., 2000; Weinhold et al., 2005) and therefore has been attributed to the general anti-adhesive properties of polySia caused by steric inhibition of membrane–membrane apposition and independent of NCAM-mediated interactions (Fujimoto et al., 2001; Rutishauser, 2008). The outcome of the current study is also not compatible with the possibility that polySia affects migration by modulating functions of NCAM as a signalling receptor (Beggs et al., 1997; Kolkova et al., 2000; Niethammer et al., 2002; Hinsby et al., 2004; Bodrikov et al., 2005; Cassens et al., 2010; Kleene et al., 2010a; Kleene et al., 2010b) or by adjusting cell-autonomous NCAM functions, such as the assembly of an NCAM-dependent signalling complex (Cavallaro et al., 2001; Prag et al., 2002; Lehembre et al., 2008). Moreover, the dependence on cell–cell contacts indicates that the observed effects are not caused by altered sensing of soluble factors as suggested for the modulation of chemotactic migration by polySia (Zhang et al., 2004; Glaser et al., 2007; Rey-Gallardo et al., 2010). Notably, constitutive shedding of soluble NCAM extracellular domain fragments was held responsible for enhanced migration of rat B35 neuroblastoma cells transfected with NCAM-140, whereas the inhibition of shedding reduced migration and increased attachment to fibronectin (Diestel et al., 2005). This raises the possibility that polySia removal affects migration and adhesion by modulating NCAM shedding. In this scenario, the concurrent effects of soluble NCAM and cell contact-dependent interactions after polySia removal would predict that loss of polySia increases cell surface NCAM by inhibition of shedding. There is, however, no reason to assume that NCAM shedding is facilitated by the presence of polySia and all data so far indicate that shed NCAM fragments are not polysialylated (Diestel et al., 2005; Hübschmann et al., 2005; Hinkle et al., 2006; Kalus et al., 2006).

Unlike activation of the ERK MAP kinase pathway, the stimulation of focal adhesion by removal of polySia or exposure to polySia-negative NCAM was independent of FGF receptor activity. The Ig3–Ig4 region of NCAM, which was not capable of inducing ERK activation, was found to be necessary and sufficient to induce enhanced focal adhesion in NCAM-negative cells. By contrast, increased focal adhesion could not be achieved by the NCAM modules FnI or FnII containing the sites that were identified to mediate FGF receptor interaction. This again was unexpected, because most of the previously described NCAM signalling functions depend on FGF receptor activation, either directly or through FGF receptor co-signalling (Kolkova et al., 2000; Cavallaro et al., 2001; Niethammer et al., 2002; Hinsby et al., 2004; Francavilla et al., 2009). Furthermore, a study on the role of NCAM in regulating the motility of glioma cells indicated a crucial function of the intracellular NCAM domain as well as the involvement of heterophilic interactions of the first two immunoglobulin modules with membrane-associated heparan sulphate proteoglycans (Prag et al., 2002). Heterophilic binding of NCAM to heparan sulphate proteoglycans is well known (Cole and Akeson, 1989), but this binding is promoted by the presence of polySia (Storms and Rutishauser, 1998), i.e. in contrast to the results of the current study, removal of polySia and NCAM exposure should have contrary effects. In fact, our results argue that the polySia-dependent effects observed by us are not caused by interactions of polySia or NCAM with glycosaminoglycans. First, the second immunoglobulin domain containing the heparin or chondroitin sulphate binding site of NCAM was dispensable for NCAM-induced focal adhesion. Second, heparinase treatment promoted migration per se, but did not interfere with either the inhibition of migration or the enhanced focal adhesion induced by NCAM treatment or polySia removal.

In summary, polySia promotes tumour cell motility by acting as a negative regulator of NCAM-induced but FGF-receptor-independent signalling from cell–cell contacts to focal adhesions. The heterophilic NCAM interactions responsible for activating the cellular response initiated by a loss of polySia remain to be defined. Nevertheless, the proposed mechanism allows for several important predictions concerning future development and assessment of polySia- or NCAM-directed tools aiming at reducing motility of the highly metastatic polySia–NCAM-positive tumours. First, attempts to reduce polySia levels should target the glycan itself without interfering with the expression of NCAM or with trans-interactions of polySia-free NCAM. Second, NCAM mimetic peptides targeted at FGF receptor interaction and activation are not likely to directly enhance cell–matrix adhesiveness but, as shown by others (Kiselyov et al., 2003; Neiiendam et al., 2004; Francavilla et al., 2009) and confirmed in the current study, to exert potent effects through activation of ERK MAP kinase signalling and possibly other pathways. Third, motifs contained within the Ig3 and Ig4 modules of the NCAM extracellular domain are necessary and sufficient for the heterophilic trans-interactions of NCAM that cause enhanced focal adhesion, whereas Ig1 and Ig2 or Ig5 to FnII are dispensable. Fourth, strategies to mimic these heterophilic NCAM contacts seem to be as promising as approaches to prevent or abolish polysialylation. It remains a major challenge for future studies to define these interactions and to determine the role of polySia in regulating NCAM binding to FGF receptors. An anti-apoptotic effect of ERK activation in response to heterophilic NCAM interactions after polySia removal has been demonstrated before (Seidenfaden et al., 2003). Assuming that the underlying NCAM contacts correspond to those responsible for the FGF-receptor-mediated activation of ERK, by applying NCAM-Fc or FGL peptide to NCAM-negative neuroblastoma (this study) or HeLa cells (Francavilla et al., 2009) it might be possible to develop different NCAM mimetic peptides in order to dissect the survival-promoting function of NCAM from its inhibitory effect on tumour cell motility, through the stimulation of focal adhesions.

The following commercial reagents were used: FGF receptor inhibitor PD173074, Rho-associated protein kinase inhibitor Y-27632, fluorescein isothiocyanate (FITC)-labelled phalloidin, goat IgG (all from Sigma-Aldrich, St Louis, MO), MFP647 Phalloidin (MoBiTec, Göttingen, Germany), src-family kinase inhibitor PP2 (Merck, Darmstadt, Germany), MEK inhibitor PD98059 (Alexis, San Diego, CA), bovine fibronectin (Biomol, Hamburg, Germany) and heparinases I and III from Flavobacterium heparinum (Sigma-Aldrich).

Mono- and polyclonal primary antibodies (mAb, pAb) were: anti-ERK1/2 rabbit pAb, anti-dually phosphorylated ERK1/2 mouse mAb, clone E10 (New England Biolabs, Ipswich, MA), anti-fibronectin goat pAb, anti-fibronectin mouse mAb, clone FN-15, anti-SynCAM 1 rabbit pAb (all from Sigma-Aldrich), anti-SynCAM 1 chicken mAb 3E1 (MBL, Woburn, MA), anti-FAK rabbit pAb, anti-Fyn rabbit pAb, anti-neuropilin2 rabbit pAb, anti-phosphorylated tyrosine mouse mAb, clone PY99 (all from Santa Cruz Biotechnology, Santa Cruz, CA), anti-paxillin mouse mAb, clone 394 (BD Biosciences, San Jose, CA), anti-penta-His mAb (Qiagen, Hilden, Germany), anti-polySia mouse mAb, clone 735 (IgG2a; Frosch et al., 1985) and anti-NCAM mouse mAb, clone 123C3 (IgG1), directed against a membrane-proximal region of NCAM comprising the first fibronectin type III module and reactive with all isoforms of human NCAM [(Gerardy-Schahn and Eckhardt, 1994) kindly provided by R. Gerardy-Schahn].

Endo-N-acetylneuraminidase F (endoNF) specifically degrading polySia was isolated as described previously (Stummeyer et al., 2005) and used in cell culture medium at a concentration of 200 ng per ml to remove polySia from the cell surface. As a control for endoNF, the inactive double mutant endoNF-R596A/R647A, that binds to polySia but does not degrade it (Stummeyer et al., 2005), was used in some of the experiments. Human IgG₁-Fc fragments and secreted, polySia-free NCAM-Fc chimera, consisting of the extracellular domain of human NCAM (amino acids 1–705) fused to the constant (Fc) part of human IgG₁, were produced as described previously (Röckle et al., 2008) and used at a concentration of 1 μg/ml. Soluble NCAM extracellular domain fragments were produced as described below. C3d, a synthetic dendrimeric undeca peptide, which binds to the first immunoglobulin-like module of NCAM, its inactive variant C3d2ala (Ronn et al., 1999) and a dimeric form of the FGL peptide derived from the second fibronectin type III module of NCAM, which is capable of binding to and activating FGF receptors (Neiiendam et al., 2004), were kindly provided by E. Bock (Panum Institute, Copenhagen, Denmark).

NCAM fragments were PCR amplified from human NCAM-140 with the following primer pairs: 5′-GCAGGGATCCCTGCAGGTGGATATTG-3′, 5′-ATCGCGGCCGCCGAGGTCCTGAACAC-3′ (NCAM ecd); 5′-GATCGGCGCCATGAGAACCATCCAGGCCAGGCAG-3′, 5′-GTGGGAAGCTTTTAGGTCCTGAACAC-3′ (NCAM Ig3–FnII); 5′-GATCGGCGCCATGAGAACCATCCAGGCCAGGCAG-3′, 5′-GTTAAAGCTTTTATGGCTGCGTCTTGAAC-3′ (NCAM Ig3–FnI); 5′-ACCGGATCCCAGGACTCCCAGTC-3′, 5′-GTTAAAGCTTTTATGGCTGCGTCTTGAAC-3′ (NCAM Ig5–FnI); 5′-GATCGGCGCCGACACCCCCTCTTCACCAT-3′, 5′-CAACAATTGCATTCATTTTAT-3′ (NCAM FnI–FnII); 5′-TGACCTCGAGTGATGAAATTCTTAGTCAACGTT-3′, 5′-TAGCAAGCTTTTATGCTTGAACAAGGATGAATTCC-3′ (NCAM Ig1–Ig5); 5′-GATCGGCGCCATGAGAACCATCCAGGCCAGGCAG-3′, 5′-TAGCAAGCTTTTATTGCACTTCAAGGTACATGG-3′ (NCAM Ig3–Ig4). The PCR products were cloned into a modified pFastBac HT A vector (Invitrogen, Paisley, UK) containing a honey bee melitin secretion signal and an N-terminal His tag using the following restriction sites: NotI–BamHI (NCAM Ig1–FnII), BamHI–HindIII (NCAM Ig5–FnI) and KasI–HindIII (NCAM Ig3–FnII, NCAM Ig3–FnI, NCAM FnI–FnII, NCAM Ig1–Ig5, NCAM Ig3–Ig4). The constructs code for the following NCAM protein fragments (according to UniProt 13591, but lacking the small alternatively spliced exons VASE and AAG): NCAM ecd: Ser19–Thr702; NCAM Ig3–FnII: Thr213–Thr702; NCAM Ig3–FnI: Thr213–Pro607; NCAM Ig5–FnI: Gln403–Pro607; NCAM FnI-FnII: Ala507–Thr702, NCAM Ig1–Ig5: Ser19–Ala507; NCAM Ig3–Ig4: Thr213–Gln413. Baculoviruses coding for the respective NCAM fragments were generated using the Bac-to-Bac® baculovirus expression system (Invitrogen) according to the manufacturer's instructions. Sf9 insect cells were grown in suspension culture at a density of 0.5–5×10⁶ cells/ml in protein-free Insect Xpress medium (Lonza, Basel, Switzerland) rotated at 70–90 r.p.m. and 27°C. Sf9 cell cultures (1–4 litres) were infected at a density of 1.7–2.0 cells/ml with P3 baculoviral stock, cell culture supernatants were harvested 72 hours post infection and secreted proteins were purified by Ni²⁺ chromatography using HisTrap columns (GE Healthcare, Munich, Germany) followed by size-exclusion chromatography (Superdex 200 HR 10/300 GL or HiLoad 16/60 Superdex 200; Amersham Biosciences, Freiburg, Germany) in 10 mM Tris-HCl buffer pH 7.5, containing 100 mM NaCl.

The human neuroblastoma cell lines SH-SY5Y (ATCC-no. CRL-2266), Kelly (ECACC no. 92110411) and LS (Rudolph et al., 1991), and the rhabdomyosarcoma cell line TE-671 (ATCC no. CRL-7774) were used. SH-SY5Y_EGFP, stably transfected to express cytosolic EGFP, and LS cell clones stably transfected to express either polySia-negative NCAM-140, (LS_AM1), polysialylated NCAM-140 (LS_AM1PST) or polysialylated NCAM-140 plus cytosolic EGFP LS_AM1PSTegfp were generated as described previously (Seidenfaden et al., 2003; Seidenfaden et al., 2006). Cells were cultured at 37°C and 9% CO₂ in DMEM–Ham's F12 medium containing 10% (v/v) heat-inactivated fetal bovine serum and 2mM glutamate (all from Biochrom, Berlin, Germany). Medium was changed every 2 days and cells were re-plated before confluency.

Time-lapse videomicroscopy and measurements of single cell motility were performed as described elsewhere (Röckle et al., 2008). For two-dimensional scratch wound migration assays, cells were grown to confluency in 35 mm Petri dishes (uncoated, if not stated otherwise). A plastic pipette tip was used to produce a scratch wound of approximately 10 mm length and 500 μm width. Medium was replaced to remove dead cells and to apply reagents, as specified for each experiment. After acute cellular reactions at the wound edge had abated (time point t0) and 6 hours later (t6) the entire scratch was documented using the MosaiX module of AxioVision software (see Image acquisistion below). To assess the number of cells that invaded the cell-free area a mask outlining the edges of the scratch at t0 was superimposed on the image of the same region at t6 using CorelDraw X3 software and the area newly covered by cells was determined using the thresholding and particle analyses tools of NIH ImageJ software (for an example, see supplementary material Fig. S2).

Immunoprecipitation of neuropilin-2 and SynCAM 1 was performed from cells lysed in 50 mM Tris-HCl, pH 7.4 containing 1% Triton X-100, 0.5% sodium deoxycholate, 150 mM NaCl, 5 mM EDTA, 1 mM phenylmethylsulfonylfluoride (PMSF), 10 pg/ml leupeptin and 10 mg/ml aprotinin. After centrifugation at 20,000 g (15 minutes, 4°C), supernatants were pre-cleared by incubation with normal IgG for 30 minutes and protein-G–Sepharose beads (GE Healthcare, Freiburg, Germany) for 30 minutes followed by centrifugation at 9,000 g. Primary antibody (1 μg) was added to extracts containing 1 mg protein in a 1 ml volume. After overnight incubation at 4°C with gentle inversion, immune complexes were recovered by incubation with 15 μl bed volume of protein-G–Sepharose (4 hours, 4°C). Pellets were collected by centrifugation at 9,000 g for 1 minute, washed three times with lysis buffer and reacted with 1 μg endoNF for 30 minutes on ice, where indicated. Proteins were eluted in reducing SDS sample buffer and separated using 10% SDS polyacrylamide gel electrophoresis.

Immunoprecipitation of Fyn and paxillin was conducted in the same way, but cells were lysed in Brij 96 lysis buffer consisting of 20 mM Tris-HCl, pH 7.4, 1% Brij 96, 150 mM NaCl, 10 mM NaF, 1 mM EDTA, 1 mM EGTA, 1 mM Na₃VO₄, 1 mM PMSF, 10 pg/ml leupeptin and 10 mg/ml aprotinin. Protein-A–Sepharose was used with the mouse anti-paxillin mAb. Immunoblotting and analysis of ERK1/2 mitogen activated protein (MAP) kinase or protein tyrosine phosphorylation was performed as described previously (Seidenfaden et al., 2003). Except for analysis of Fyn immunoprecipitation and detection of SynCAM 1, which were performed using enhanced chemiluminescence (ECL) (Seidenfaden et al., 2003; Galuska et al., 2010), the Odyssey Infrared imaging system (LI-COR Biosystems, Homburg, Germany) was used for semi-quantitative evaluation. To assess ERK1/2 phosphorylation, cells were washed with ice-cold PBS and harvested with a cell scraper in ice-cold Brij 96 lysis buffer. After 10 minutes of incubation on ice, the lysates were centrifuged at 20,000 g (15 minutes, 4°C) and supernatants were mixed with reducing electrophoresis buffer. Separation of 20 or 40 μg of protein was performed on 10% SDS polyacrylamide gels. After transfer to PVDF membranes, double-immunostaining was performed by combined incubation with rabbit anti-ERK and mouse anti-phosphorylated ERK primary antibodies followed by IRDye 680- and 800-labelled secondary antibodies diluted in Odyssey blocking buffer according to the manufacturer's instructions. Signals were detected and quantified with the Odyssey infrared imaging system. If antibodies from different species were available, the presence of two antigens was assessed by double-immunolabelling and simultaneous detection with the Odyssey infrared imaging system, as described for the analysis of ERK phosphorylation. For further detection or application of a second primary antibody from the same species, antibodies were washed off with NewBlot stripping buffer (LI-COR Biosystems). Before re-probing, membranes were scanned to ensure complete antibody removal.

Immunostaining was performed as described before (Seidenfaden and Hildebrandt, 2001; Schiff et al., 2009). Briefly, cells were fixed with 4% paraformaldehyde (PFA) for 30 minutes, blocked with 2% bovine serum albumin, and incubated with primary antibodies overnight at 4°C. For detection of intracellular epitopes, cells were permeabilized with 0.1% Triton X-100. Rabbit and mouse IgG-specific and subtype-specific Cy3-(Chemicon, Temecula, CA), Alexa-Fluor-488-, Alexa-Fluor-568- and Alexa-Fluor-647-(Molecular Probes/Invitrogen) conjugated antibodies and phalloidin conjugates were used as suggested by the suppliers. In double-stained immunofluorescence samples, cross reactivity of secondary antibodies was controlled by omitting either of the two primary antibodies. Coverslips were placed over the cells in Vectashield mounting medium with 4′,6′-diamidino-2-phenylindole (DAPI; Vector Laboratories, Burlingame, CA).

Microscopy was performed using a Zeiss Axiovert 200 M equipped with an ApoTome device for near confocal imaging, an AxioCam MRm digital camera and AxioVison software (Carl Zeiss Microimaging, Göttingen, Germany). For evaluation of scratch migration, high-contrast bright-field micrographs covering the entire area of the approximately 1 cm scratch wound were acquired using the MosaiX module. For colocalisation studies, optical sections of 0.81 μm thickness were obtained using a 63× Plan-Apochromat oil immersion objective with 1.4 numerical aperture (Zeiss).

For evaluation of peripheral focal adhesions, stained cultures and micrographs were coded and randomized to ensure that the observer was blind to the experimental conditions. Per culture, a minimum of 30 micrographs were acquired at 63× magnification. Positions of frames were selected using only the channel for nuclear stain (DAPI) first, before the frame was adjusted to image the entire cell of interest using the channel for actin staining (FITC- or MFP647–phalloidin). Typical peripheral focal adhesions were identified as sites of FAK immunofluorescence colocalised with actin staining and counted by visual inspection assisted by the interactive event counting tool in the AxioVision software.

Statistical analyses were performed using Graphpad Prism software. Differences between two groups were evaluated with Student's t-test (two-tailed). To compare more than two groups, one-way ANOVA with Newman–Keuls multiple comparison post-hoc test (two-tailed) was applied.

We thank E. Bock (University of Copenhagen), P. Claus and R. Gerardy-Schahn (Hannover Medical School) for kindly providing peptides and reagents.