Expression of sialyl-Tn epitopes on β1 integrin alters epithelial cell phenotype, proliferation and haptotaxis

Altered glycosylation of cell surface components is a recurrent feature of cancer cells often associated with a metastatic phenotype and unfavorable prognosis (Le Pendu et al., 2001). Sialylation represents one of the most frequently occurring terminations of the oligosaccharide chains of glycoproteins and glycolipids with sialic acid commonly α2,3- or α2,6-linked to galactose (Gal), α2,6-linked to N-acetylgalactosamine (GalNAc) or α2,8-linked to another sialic acid. The biosynthesis of the various linkages is mediated by different members of the sialyltransferase family (Harduin-Lepers et al., 2001). O-glycan biosynthesis starts with the addition of a GalNAc on serine or threonine residues of the polypeptide backbone of various glycoproteins. The GalNAc-O-Ser/Thr motif can be substituted by a Gal residue to give the core 1 of O-glycans (Galβ3GalNAc-O-Ser/Thr). This disaccharide is also called T antigen. Addition of N-acetylglucosamine residues gives the core 2 (Galβ3[GlcNAcβ6]GalNAc-O-Ser/Thr) or core 3 (GlcNAcβ3GalNAc-O-Ser/Thr) structures. These core structures can then be either sialylated (core 1) or further extented and sialylated. However, premature sialylation in α2,6 of the GalNAc residue prevents further extention of the O-glycan. The resulting structure NeuAcα2,6GalNAc-O-Ser/Thr (where NeuAc is neuraminic acid) corresponds to the Sialyl-Tn antigen (STn or CD175s), which is a characteristic tumor-associated antigen (Brockhausen, 1999). This antigen is not strictly tumor-specific since it is found in a variety of normal tissues either at an intracellular level, at cell surfaces or on secreted mucins. Thus it has been detected in the skin, larynx, small intestine, stomach, salivary glands, cervix, testis and activated lymphocytes (Cao et al., 1996; Carrilho et al., 2000; Fernsten et al., 1998; Stein et al., 1991). However, its expression is largely increased in many types of cancers including kidney, prostate, urinary bladder, cervix, ovary, breast, liver, pancreas, salivary glands, esophagus, stomach and colorectal carcinomas (Le Pendu et al., 2001). Several groups have investigated whether abnormal expression of STn could be predictive of patients outcome and a significant association with unfavorable prognosis has been reported for patients with breast, ovary, lung, urinary bladder, stomach or colorectal carcinomas (Le Pendu et al., 2001). However, the meaning of these associations between expression of STn and bad prognosis is unclear as the biological role of STn at the cellular level has not been studied. In addition, contradictory data have been reported regarding the prognostic value of STn. Indeed, various studies failed to observe a significant association between expression of STn and prognosis in colon, stomach, salivary glands, esophagus and urinary bladder carcinoma (Baldus et al., 1998; Ikeda et al., 1993; Lundin et al., 1999; Miles et al., 1995; Therkildsen et al., 1998). Some authors showed that STn expression in the colon is higher in precancerous lesions than in carcinomas and one study even reported that the presence of STn in keratoacanthoma was associated with tumor regression (Jensen et al., 1999; Yoshida et al., 1998).

The sialyltransferase ST6GalNAc I was the first enzyme recognized to be able to synthesize the STn antigen (Ikehara et al., 1999). It was then found that the ST6GalNAc II enzyme can also catalyze the transfer of a sialic acid residue in α2,6 position on a GalNAc O-linked to a peptide (Kono et al., 2000). As premature addition of the α2,6-linked sialic acid impairs further elongation of the carbohydrate chain, we reasoned that overexpression of one of these two enzymes should generate cell surface STn expression by competition with the other enzymes that participate to the biosynthesis of O-glycans. Therefore, to define the cellular consequences of STn cell surface expression, we transfected TS/A cells with the cDNA encoding the human ST6GalNAc I. TS/A cells originate from a spontaneously arising mouse mammary carcinoma and were chosen because they are metastatic and because their strong labeling with the peanut agglutinin (PNA) and Jacalin (JAC) lectins indicated that they expressed high levels of T antigen or core 1 O-glycans. By contrast, they did not show any detectable expression of STn epitopes.

In the present report, we show that the ST6GalNAc I transfected TS/A cells express cell surface STn epitopes while the presence of T antigen was strongly decreased as expected. This was accompanied by major morphological changes and impaired proliferation, as well as decreased migration on fibronectin and hyaluronic acid. We also show that the presence of a STn motif on the integrin β1 subunit (CD29) results in a signalling defect involved in the loss of motility on fibronectin.

Fetuin, rhodamine-labeled phalloidin, fibronectin (FN), collagen I, IV, VI, laminin, hyaluronic acid were purchased from Sigma (St Louis, CO). Two monoclonal antobodies against STn were used in this study. The antibody TKH2 was a kind gift from E. Clausen (Copenhagen, Denmark) and B72.3 was purchased from NeoMarkers (Fremont, CA). Both were of the IgG1 subclass and the IgG1 MOPC 21 (Sigma, St Louis, MO) was used as a control irrelevant antibody. The anti-α5 integrin (CD49e) clone 5H10-27 (MFR5) and the anti-β1 integrin (CD29) clone KMI6 were obtained from Pharmingen. The anti-CD44 clone 1M7.8.1 was from Zymed (San Francisco, CA) and the anti-vinculin was purchased from Sigma. The fluorescein (FITC) conjugated anti-rabbit, anti-rat and anti-mouse IgG were from Jackson ImmunoResearch Laboratories (West Grove, PA).

The cDNA encoding for constitutively active RhoA (RhoA^Val-14) and dominant negative form of RhoA (RhoA^Asp-19) cloned into the bacterial expression vector pTAT-HA (Chellaiah et al., 2000) (kindly provided by S. Dowdy, St Louis, MO) have been used to produce the TAT-RhoA^Val-14 fusion protein and TAT-RhoA^Asp-19. Purification and transduction of proteins into TS/A cells were performed as previously described (Nagahara et al., 1998). TAT-RhoA^Val-14 and TAT-RhoA^Asp-19 fusion proteins were added to the cells 30 minutes before seeding on fibronectin for motility analysis and 16 hours before fixation for actin cytoskeleton observation.

For this work, we used cells derived from the TS/A cell line, a metastasing mouse cell line originated from a mammary adenocarcinoma which arose spontaneously in a Balb/c female (Nanni et al., 1983). Cells were cultured in RPMI 1640 supplemented with 10% FCS, 2 mM L-glutamine, 100 units/ml penicillin and 100 mg/ml streptomycin (Life Technologies). Cells were subcultured at confluency after dispersal with 0.025% trypsin in 0.02% EDTA. For experiments cells were used at 70%-80% confluency. They were routinely checked for mycoplasma contamination using HOECHST 33258 (Sigma) labeling.

To minimize variation due to heterogeneity of the parental cell line, a clone called B10 was first isolated from the TS/A line. TS/A B10 was transfected by lipofection with the cDNA encoding for the human ST6GalNac I inserted into the pCXN2 vector (Ikehara et al., 1999). Cells expressing the STn antigen were sorted by flow cytometry. The resulting population was cloned by limiting dilutions on a feeder of irradiated rat primary fibroblasts. Three positive clones, called D4, C2 and G9, were randomly selected among the positive ones. Control cells were obtained after transfection with the empty vector, followed by cloning as above. Two clones called F4 and C15 were selected.

Viable cells, 2.5 10⁵ per well of 96-well culture microtitre plates (Falcon) were incubated either with primary mAbs, TKH2 as a culture supernatant diluted 1/2 and B72.3 at 1 μg/ml or with the biotinylated PNA and JAC lectins at 1 μg/ml and 0.25 μg/ml, respectively, in PBS containing 0.1% gelatin for 30 minutes at 4°C. After three washes with this same buffer, a 30 minute incubation with the secondary anti-mouse IgG FITC-labeled antibody (Sigma) or with FITC-labeled streptavidin (Sigma) was performed at 4°C. After washing, fluorescence analysis was performed on a FACSCalibur (Becton Dickinson, Heidelberg, Germany).

Confluent cells were rinsed with ice-cold PBS, pH 7.2, then recovered by scrapping. After two washes with ice-cold PBS, cells were solubilized in 50 mM potassium phosphate pH 7.0 containing 2% (v/v) Triton X-100 on ice for 30 minutes. Following a centrifugation at 10,000 g for 10 minutes, the supernatant was collected and used as a crude enzyme preparation. Protein concentration was determined using bicinchoninic acid. The reaction mixture contained 0.06 mM CMP-neuraminic acid [¹⁴C] (273 mCi/mmole NEN, Dreieichendain, Germany), MgCl₂ 10 mM, CaCl₂ 5 mM, M.E.S. buffer 50 mM at pH 6, 30 μg fetuin as exogenous acceptor and 16 μg cell lysate in a final volume of 10 μl. After incubation at 37°C for 1 hour, the reaction mixtures were mixed with 10 μl reducing buffer (Tris-HCl 0.5 M, pH 6.8; glycerol 10%; SDS 10%; β-mercaptoethanol 0.05% and bromophenol blue 0.05%) and separated by 5-15% SDS-PAGE. After migration, the gels were dried and autoradiographed for 4 days at -80°C. Then, the bands were cut and their radioactivity was counted in 12 ml scintillation liquid (Ready Safe™, Beckman, Palo Alto, CA). Background levels of radioactivity were obtained from controls without exogenous acceptor.

Cells seeded on glass lamellae were fixed in 2% formaldehyde in culture medium for 20 minutes. After washing in saline buffer (PBS containing 1 mM CaCl₂ and MgCl₂) for 5 minutes (3×), cells were permeabilized with 0.05% Triton X-100 and 0.05% Tween 20 in saline buffer for 3 minutes at room temperature. After three washes in saline buffer, nonspecific binding sites were blocked by incubation with 1% BSA in saline buffer for 30 minutes. The cell monolayer was then incubated with the anti-STn TKH2 diluted 1/2 in saline buffer 1% BSA for 1 hour, washed in saline buffer (3×) before incubation with an anti-mouse IgG FITC-labeled antibody for 30 minutes, washed again in saline buffer (3×) and mounted in Mowiol. Negative controls were incubated with the secondary antibody alone. Rhodamine-labeled phalloidin, used to label polymeric actin, was incubated for 2 minutes at 2 μg/ml. The slides were observed on a Leica TCS SP (Heidelberg, Germany) confocal fluorescence microscope.

For morphometric parameters acquisition, cells were seeded at low density (6 to 7 cells per mm²). Samples were viewed using a 20× phase contrast objective and projected onto a video screen. Outlines of 100 cells of each experimental group were acquired manually, quantitative measurements of area (A) and perimeter (P) being performed using the Gold software (Clara Vision, Orsay, France). Cell circularity (shape factor: f) was calculated according to the following formula: f=2πA P^-2 (f=1 corresponds to an ideal circle and decreasing values correspond to appearance of cell extensions).

Five thousand cells were plated in 96-well culture microtiter plates (Falcon) and incubated at 37°C in 5% CO₂. Every 24 hours one plate was quantified for cell viability: cells were loaded for 3 hours with neutral red at a final concentration of 50 μg/ml in culture medium. This weakly cationic dye penetrates cell membranes by nonionic diffusion and binds intracellularly to anionic carboxylic and/or phosphate groups of the lysosomal matrix. Thereafter, the medium was removed, cells were fixed for 5 minutes with a mixture of 1% formaldehyde/1% CaCl₂ and the dye extracted with 0.2 ml of 1% acetic acid in 50% ethanol. Plates were left overnight at 4°C and transferred to a microplate reader. Absorbance was recorded at 535 nm (Fluorolite 1000, excitation wavelength: 535 nm, emission wavelength: 600 nm) and corrected with a blank. Experiments were performed in triplicate, eight wells per cell line being used. For cell adhesion assays, wells were coated with fibronectin at different concentrations (1 to 20 μg/ml) in PBS containing 1 mM CaCl₂ for 2 hours at 37°C. After flicking and drying, cells were plated at 100 000 cells per well, medium was removed by flicking the plates and processed as described above.

We used a wound healing assay in which cells' abilities to repopulate a wound made in confluent cultures were compared. Linear wounds were done in just confluent cell monolayers plated in 6-well culture microtiter plates (Falcon) by denuding a lane (three lanes per well) with a plastic pipette tip and rinsing to remove debris. Wound breadth was measured all along the denuded monolayers at different time points using a Clara Vision image analysis system coupled to a CCD video camera and a Nikon TMS inverted light microscope equipped with a 10× objective. Results are the mean of three independent experiments.

Migration of individual cells was evaluated in 24-well microtiter plates (Falcon) using a modified Albrecht-Buehler G (1977) assay. Briefly, wells were coated with 1% bovine serum albumin, rinsed with ethanol and coated with a colloidal gold particle suspension (Sigma). After removing the nonimmobilized gold salts, wells were coated with various extracellular matrix components diluted in PBS containing 1 mM CaCl₂ for 2-3 hours. Wells were then gently washed with culture medium and seeded with 7000 cells/well. We first controlled that all cells used in this study had the same size in cell suspension as tracks are proportional to cell diameter. Plates were incubated overnight at 37°C in a 5% carbon dioxide humidified incubator. Migration was arrested by addition of 2% formaldehyde and quantified by measuring track areas of at least 100 cells (about ten fields, in triplicate) using the Gold software (Clara Vision Orsay). Control wells were coated with bovine serum albumin (0.1% in PBS). For inhibition or activation assays, cells were incubated for 30 minutes at 37°C before seeding with reagents, mAbs at concentrations ranging from 1 to 25 μg/ml.

A total of 5000 to 10,000 cells were plated in the upper chamber of 8 μm pore (24 wells) transwells (Corning, NY) coated on their lower side with either BSA (0,1%) or fibronectin (10 μg/ml) for 2-3 hours at 37°C. After 16 hours, cells were fixed with 100% cold methanol and cells localized at the top of the membranes were removed with cotton swabs. Cells remaining on the lower surface were stained with toluidin blue and the intensity of staining was quantified using the Gold software for image analysis. Experiments were performed three times in duplicate.

For western blotting, total proteins were solubilized as cells reached confluency by a 30 minute incubation in PBS pH 7.4 containing 5 mM EDTA, 1% Triton X-100. The preparations were centrifuged at 10,000 g for 15 minutes. Protein concentration of supernatants was measured using bicinchoninic acid. Thirty micrograms of proteins were subjected to electrophoresis in 10% SDS-polyacrylamide gels. Separated proteins were electrophoretically transferred to nylon filter in 25 mM Tris, 192 mM glycine, 20% methanol at 200 mA for 1 hour. After transfer, membrane strips were incubated in PBS containing 3% skim milk for 2 hours at 37°C. Membranes were then incubated with the mAbs TKH2 or B72.3 diluted in PBS containing 1% skim milk overnight at 4°C. After washing twice in PBS/Tween 0.05% and once in PBS, membranes were incubated for 1 hour with peroxidase-labeled anti-mouse IgG (Sigma) diluted in PBS containing 1% skim milk. After washes, the bound antibodies were revealed by chemiluminescence using the ECL Kit from Amersham (Little Chalfont, Great Britain). For immunoprecipitation, protein G-coated agarose beads (Sigma) were incubated with PBS containing 5% defatted milk overnight at 4°C and washed with lysate buffer. They were then incubated with 150 μg proteins from cell lysates and mAb TKH2 for 2 hours under rotation. Beads were washed three times with lysate buffer and bound proteins were eluted by boiling in Laemmli buffer for 10 minutes. Eluates were subjected to SDS-PAGE and transferred to nylon membranes. Membranes were then probed with the anti-β1 integrin or with the anti-STn TKH2 as described above.

Analysis of differences between STn+ and STn-cells was performed by Student's t test.

Clone B10 isolated from the TS/A cell line (STn-) was transfected with the ST6GalNAc I cDNA. Stable transfectants expressing STn epitopes were obtained. However, after several passages in the presence of the selecting antibiotic, antigenic expression was lost. To obtain stably STn expressing cells (STn+), positive clones were isolated by flow cytometry 1 week after transfection and amplified on a feeder of irradiated rat fibroblasts that do not express the antigen. Several clones were thus obtained that strongly expressed STn as detected with either the TKH2 or the B72.3 mAbs (Fig. 1A). As the ST6GalNAc I transferase competes with other glycosyltransferases that contribute to the synthesis of O-glycans for the GalNAc-peptide acceptor, we tested whether the expression of other cell surface carbohydrate markers would be altered. No evidence for a difference in staining with a set of lectins including GS1-B4, ECA, LEA, STA, DSA, VVA, MAA, SNA, WGA, ConA, L-PHA and SBA was found between STn- and STn+ cells (data not shown). However, a strong decrease of staining with the PNA (peanut agglutinin) and JAC (Jacalin) lectins was observed. These two lectins recognize preferentially the Galβ1,3GalNAc disaccharide, which constitutes the core 1 of O-glycans. The addition of a sialic acid on the GalNAc residue by the ST6GalNAc I enzyme precludes the addition of a galactose, therefore explaining the drop in PNA and JAC reactivity. To confirm that the STn expression concomitant with the loss of PNA and JAC reactivity is due to the presence of a strong ST6GalNAc I enzyme activity, the transfer of sialic acid was tested using fetuin as an acceptor. As shown in Fig. 1B, a much stronger activity was obtained using extracts from STn+ cells than from the STn-parental and mock-transfected cells. The localization of the STn epitopes was then determined by confocal microscopy using mAb TKH2. Reactivity was clearly localized at the cell surface, with almost no labeling detectable intracellularly, indicating that the proteins carrying the STn epitopes are cell surface glycoproteins (Fig. 1C).

Observation of the STn+ clones revealed clear changes in cellular morphology (Fig. 2A), whereas sensitivity to apoptosis was not significantly modified (data not shown). Parental and control mock-transfected cells grew as islands with strong cell-cell contacts, exhibiting the typical paving stone appearance of epithelial cells. By contrast, transfected cells were more refractive, larger (Fig. 2B) and elongated (Fig. 2C) and were less able to grow as colonies of cohesive cells. F-actin staining using rhodamine-labeled phalloidin revealed a gross perturbation of the actin cytoskeleton with partial disappearance of stress fibers at the basal membrane, cortical actin bundles being less dramatically disorganized (Fig. 2D). Moreover, the formation of focal contacts was impaired as visualized using an anti-vinculin antibody (Fig. 2E). As microfilament organization is crucial for cell migration, the ability of STn+ and STn-cells to repair a wound was compared. It appeared that STn+ cells had an impaired ability to repair a scratch wound (Fig. 3A,B). Because we found that wound closure efficiency was largely influenced by cell density (data not shown), we quantified STn+ and STn-cell growth, and it appeared that STn+ cells had a decreased proliferation rate (Fig. 3C). Therefore, STn+ cells were seeded at a higher cell density than control cells to reach the same density at the time of wounding. Using such experimental conditions, the mean speed of healing was 13.3±1.1 μm/hour for parental or mock-transfected cells and 6.28±1.2 μm/hour for STn+ cells (n=3, P<0.001). Therefore, the presence of the sialylated epitote STn on membrane glycoproteins reduces cell growth and cell saturation density and modulates cell phenotype. STn+ cells grow more dispersedly compared with wild type, they are more fusiform and larger but less flattened, therefore contacting the extracellular matrix at fewer sites.

To determine the effect of the neo-expression of ST6GalNAc I on undirected cell motility on various insoluble bound extracellular matrix components (haptotaxis), phagokinetic track assays were performed. This assay allows the co-investigation of various experimental conditions and the quantification of a great number of cell displacements visualized by the displacement of coated gold particles. As the data obtained with the B10 and F4 STn-clones were never significantly different and that the STn+ clones also behaved similarly, the results of the two STn-clones on the one hand and of the two STn+ clones on the other were pooled in all the experiments presented below.

TS/A cells significantly moved on BSA and were variously stimulated according to the matrix component tested (Fig. 4A). Laminin inhibited cell movement irrespective of the ST6GalNAc I expression and matrigel gave very variable results (not shown). No significant difference was observed between STn- and STn+ cells migrating on collagen I, collagen IV, collagen VI (1 to 50 μg/ml) (Fig. 4A). STn-cell motility was strongly stimulated by fibronectin, 10 μg/ml being the more efficient coating concentration (P<0.001), and to a lesser extent by hyaluronic acid at 5 μg/ml (P<0.05). In the representative experiment shown, cell migration on fibronectin and hyaluronic acid was nearly 40% and 30% over that on BSA, respectively (Fig. 4A). By contrast, STn+ cell motility was weakly stimulated by fibronectin and hyaluronic acid at all concentrations tested (1 to 50 μg/ml). Motility on fibronectin and hyaluronic acid between STn- and STn+ was significantly different with P<0.001 and P<0.01, respectively. The distinct motility on fibronectin between STn- and STn+ cells in the phagokinetic track assay is easily visualized in Fig. 4B. The specificity of the effect was verified by inhibiting STn-cells' motility on immobilized fibronectin with soluble fibronectin (P<0.001). The motility of STn+ cells was unaffected, confirming their lack of sensitivity to this matrix protein (Fig. 4C). The difference in motility on fibronectin between STn- and STn+ cells was more precisely quantified through ten independent phagokinetic track experiments involving two STn- and STn+ clones (Fig. 4D). STn- and STn+ cells' migration on fibronectin was stimulated 49% and 15%, respectively, this difference being highly significant (P<0.001). This was confirmed using a transwell assay (Fig. 4E). It is important to note that fibronectin did not modify adhesion of either STn- or STn+ cells over a range of concentration (1 to 20 μg/ml) and that both cell types adhered similarly (P<0.001) (Fig. 4F).

To determine which glycoproteins could have their O-glycosylation altered by transfection of ST6GalNAc I, total protein extracts from STn+ and STn-cells were submitted to western blotting revealed with the TKH2 and B72.3 STn specific mAbs (Fig. 5). No bands were detected on protein extracts from STn-cells. By contrast, various bands ranging from over 200 kDa to less than 50 kDa were detected with both mAbs TKH2 and B72.3. A major band at about 130 kDa was recognized by both antibodies (Fig. 5A). As the B72.3 mAb binding requires a repetition of the STn motif (Ogata et al., 1998; Zhang et al., 1995), this observation suggests that the 130 kDa band carries STn epitopes in cluster. The main cell surface receptor for fibronectin is the α5β1 integrin. Analysis of the mouse β1 integrin subunit sequence using the NetOglyc program (Hansen et al., 1998) revealed the presence of a cluster of three potential O-glycosylation sites at Ser 177, Thr 178 and Thr 179. We therefore tested whether the 130 kDa band could correspond to the β1 chain of the integrin. Cell extracts were immunoprecipitated with the anti-STn mAb TKH2 and the reactivity of immunoprecipitates with an anti-β1 and with TKH2 was detected by western blotting. A band at 130 kDa reacting with the anti-β1 was specifically precipitated from extracts of STn+ cells, indicating that indeed the STn+ glycoprotein at 130 kDa corresponds to the integrin β1 subunit (Fig. 5B). Western blots performed on total cell extracts showed that the amount of β1 chain was not significantly different between STn- and STn+ cells (Fig. 5A).

The presence of STn epitopes on the β1 chain induces a less motile phenotype on fibronectin. The fact that cell adhesion on this extracellular matrix was not modified indicates that substrate recognition is unchanged and therefore suggests that subsequent signalling required for motility stimulation could be altered by the presence of the STn glycans. Some anti-integrin mAbs have the property to activate integrin function through induction of conformational changes that promote integrin clustering and further recruitment of a variety of intracytoplasmic proteins (Liddington and Ginsberg, 2002; Mould et al., 1998). Most of the anti-β1 subunit activating antibodies recognize a region of the molecule located in close proximity to the fibronectin binding sites (Takada and Puzon, 1993) or in the cysteine-rich stalk region (Faull et al., 1996). As the putative O-glycosylation sites of the β1 chain are also located in the ligand binding region, we tested whether anti-STn antibodies could affect cell motility on fibronectin. Yet, as integrins act synergistically with various growth factor receptors, all following experiments were performed using cells cultured in the presence of 10% FCS as in the initial experiments. As shown in Fig. 6A, the anti-STn mAb TKH2 increased migration of STn+ cells by nearly 30% (P<0.001), whereas an isotype matched control immunoglobulin had no effect. The anti-STn mAbs had no significant effect on STn-cells. Because motility on fibronectin of STn+ cells is 34% higher than that of parental STn-cells (Fig. 4), it indicates that the anti-STn mAb almost completely restored migration on fibronectin of the STn+ cells at a concentration of 1 μg/ml. A similar result was obtained with the anti-STn mAb B72.3 at 2 μg/ml (data not shown). Both anti-α5 and anti-β1 antibodies inhibited STn-cell migration on fibronectin, confirming involvement of α5β1 in this process. Strikingly, treatment with the anti-β1 subunit at 5 μg/ml fully restored STn+ cells' motility on fibronectin (P<0.01) like the anti-STn mAb did. A slight nonsignificant stimulatory effect on STn+ cells' motility on fibronectin was observed using the anti-α5 subunit at a higher concentration (25 μg/ml) (Fig. 6B). Treatment with the anti-STn or anti-integrin subunits did not affect motility on BSA. Fibronectin is a high molecular weight multidomain glycoprotein present in plasma and in insoluble multimeric forms in the extracellular matrix. Loss of this matrix frequently accompanies transformation (Ruoslahti, 1994) and in fact no fibronectin could be detected in TS/A cells by indirect immunofluorescence (data not shown).

Fibronectin binding to the β1 integrin subunit induces activation of the small G protein RhoA, also known as a major regulator of actin stress fiber formation and organization (Danen et al., 2002; Hall, 1998). The changes in actin stress fiber organization and motility in STn+ cells could thus involve a defect in β1-integrin-mediated RhoA activation. To assess this hypothesis, we used HIV Tat-mediated delivery of a constitutively active RhoA protein (TAT-RhoA^Val-14) into TS/A cells. Transduction of TATRhoA^Val-14 protein (6 μg/ml) activated motility of the STn+ cells on fibronectin by about 25%, without a significant effect on STn-cells (Fig. 7A). TAT-RhoA^Val-14 transduction had no effect on STn+ or STn-cells' motility on hyaluronic acid, underscoring the specificity of its effect on the fibronectin transduction pathway. As transfection of ST6GalNAc I induced morphological changes on TS/A cells, the effect of TATRhoA^Val-14 on cell shape was examined. It restored an epithelial cell shape to STn+ cells without noticeably affecting the morphology of STn-cells (Fig. 7B). Similarly, staining of the transduced cells by rhodamine-phalloidin showed that TATRhoA^Val-14 protein restored actin stress fibers in STn+ cells, whereas an inactive TAT-RhoA (N19) did not show any effect (Fig. 7C). Collectively, these results show that while cells' ability to recognize fibronectin as an extracellular substrate is unchanged, downstream recruitment and reorganization of structural and signalling elements at the cytoplasmic face are impaired in the presence of the STn glycans.

The major findings of the present study show that O-glycosylation of the β1-integrin in mouse mammary carcinoma cells impairs the integrin-mediated signalling, leading to major morphological and cell behavior alterations.

Mucin-type-O-linked glycosylation is initiated in the cis-Golgi where the first sugar (GalNAc) is added to the hydroxyl group of serine or threonine residues. It then proceeds by sequential addition of monosaccharides, each reaction being catalyzed by a specific enzyme. Depending on the cell type and their tissue of origin, a large variety of O-glycans, including very large structures, can be synthesized (Brockhausen et al., 2001). A common feature of cancer cells is the frequent occurrence of short O-glycans that correspond to tumor-associated antigens. Thus, simple O-glycans comprising only the first GalNAc residue correspond to the Tn antigen. The addition of a galactose in β1,3 linkage to this structure gives the core 1 of O-glycans or T antigen (Galβ3GalNAc), which can be sialylated into sialyl-T antigen (NeuAcα2,3Galβ3GalNAc). The ST6GalNAc I transferase can use each of these molecules as substrate to add a sialic acid residue in α2,6 position. The addition of this residue signals a termination of the elongation of the carbohydrate chain (Saitoh et al., 1991). It is thus expected that the premature addition of a sialic acid residue in α2,6 position on the Tn antigen should generate STn epitopes (NeuAcα2,6GalNAc) and block the elongation of O-glycan chains. The results presented here show indeed that increasing the expression of the ST6GalNAc I enzyme activity in the murine TS/A cells generates cell surface STn epitopes. This probably results from a competition with the endogenous β3-galactosyltransferase responsible for the synthesis of core 1 as manifest by the major decrease of staining with PNA and JAC, which bind to the T and/or sialyl-T antigens, respectively.

Increasing the expression of the ST6GalNAc I enzyme also led to major changes in cellular behavior. The STn+ cells presented a distinct morphology with disorganization of actin stress fibers and a low proliferation rate, and their migratory phenotypes were quite different from those of parental or mock-transfected cells. These changes were not artefactual as they were never observed on a large series of mock transfected clones or on ST6GalNAc I transfected clones that did not express the STn antigen. By contrast, all STn+ clones presented the same altered morphology and decreased proliferation, as well as decreased motility on fibronectin and hyaluronic acid. The main receptor for hyaluronic acid is CD44, which is a highly glycosylated adhesion molecule, and it has been previously shown that cellular adhesion and motility on hyaluronic acid is modulated in a complex manner by both the N- and O-glycans of CD44 (English et al., 1998; Katoh et al., 1995; Sheng et al., 1997; Skelton et al., 1998). It is possible that CD44 variants are carriers of the STn O-glycan chains of ST6GalNAc I transfectants as western blotting with anti-STn revealed glycoproteins that may correspond to CD44 high molecular weight variants and since anti-CD44 inhibited the migration of STn-cells but not of STn+ cells on hyaluronic acid (data not shown). However, this remains to be shown.

Integrins are important receptors for cell adhesion to extracellular matrix proteins. They also make connections to the cytoskeleton at sites of focal adhesion and regulate various intracellular signalling pathways. For this reason they have been involved not only in the regulation of cell adhesion and movement but also of cell growth, survival and differentiation (Hynes, 2002). Among the 24 integrins described, the main receptor for fibronectin is α5β1, and a large number of previous studies have shown that its N-glycans modulate adhesion and motility on fibronectin. Thus, N-glycosylation of the β1 chain is required for adhesion to fibronectin via this integrin (Zheng et al., 1994); incomplete maturation of the β1 chain N-glycans decreases interaction with fibronectin (Ringeard et al., 1996) reduces α5β1 clustering (Guo et al., 2002) and the presence of sialic acid residues on mature N-glycans of the β1 chain decreases binding to fibronectin (Kawano et al., 1993; Semel et al., 2002). As the ST6GalNAc I transfectants presented a major alteration of their ability to migrate on a fibronectin matrix, we determined whether the α5β1 integrin could carry STn O-glycans. Although it has never been shown previously that the α5β1 integrin can be O-glycosylated, this was plausible given that analysis of the sequence of the two chains revealed an STT motif with a high glycosylation potential on the β1 chain. Western blotting and immunoprecipitation experiments showed that indeed the β1 chain carried STn epitopes. We controlled by western blotting that its level of expression was not affected.

The STT motif for O-glycosylation is localized within the I-like domain and present on both mouse and human β1 chains. Yet, it is not found on all integrin β chains such as β3, for which much structural information is available. It is the I-like domain of the β1 chain, which binds to the RGD peptide and which in conjunction with the β-propeller domain on the α subunit participates in the binding of fibronectin (Mould et al., 1997). This domain is also proposed to transmit conformational change upon ligand binding (Hynes, 2002). Here, we have observed that the STn+ cells' adhesion to fibronectin was not altered; only their motility was affected. Signal transduction mediated by integrins requires their conformational change and/or clustering at the cell surface on ligand binding (Bazzoni and Hemler, 1998; Hogg et al., 2002; Springer, 2002). Motility on fibronectin of STn+ cells was restored by both a mAb anti-β1 and a mAb anti-STn, suggesting that the blocking of elongation of the O-glycan chain on STn+ cells led to a conformational change in the β1 chain impairing the transfer of signal from the extracellular matrix to the cytoplasmic domain of the integrin. Addition of these antibodies could either facilitate the tethering of α5β1 or restore the spatial conformation required for efficient outside-inside signalling, thus inducing motility. The change in β1 chain O-glycosylation might also modify its susceptibility to protease action and its turnover as previously described for LAMP-1 (Fukuda, 1991).

The small G proteins of the Rho family are guanine nucleotide binding proteins that cycle between an active GTP-bound and an inactive GDP-bound state. They are involved in the reorganization of the actin cytoskeleton that takes place on cell adhesion and migration (Malliri and Collard, 2003). The small G protein RhoA is involved in the formation of actin stress fibers in response to α5β1 integrin-mediated adhesion on a fibronectin matrix (Danen et al., 2002). We observed that STn+ cells had disorganized F-actin fibers and decreased levels of vinculin-positive cell focal adhesions. The restoration of a normal morphological and migrating phenotype in STn+ cells treated with a constitutively active form of RhoA, and the loss of stress fibers in the wild-type cells with its dominant negative form, supports the hypothesis of a defecting coupling between α5β1 and RhoA. However, we could not completely rule out that the effect of the active form of RhoA was due to the activation of a parallel pathway.

The results presented here are at odds with various reports indicating that the presence of STn epitopes on tumor cells is associated with bad prognosis (David et al., 1996; Ghazizadeh et al., 1997; Imada et al., 1999; Itzkowitz et al., 1990; Kakeji et al., 1995; Kinney et al., 1997; Victorzon et al., 1996; Werther et al., 1996). Indeed, tumor cells with a low proliferation rate and a low ability to migrate on either fibronectin or hyaluronic acid are expected to present a poorly aggressive phenotype. In line with this, when STn+ cells where injected subcutaneously into syngeneic mice, all resulting tumors had almost completely lost expression of the STn epitope, indicating that STn+ cells had been counterselected in vivo (data not shown). The mechanisms that led to overexpression of the STn structure on tumor cells are not related to overexpression of ST6GalNAc transferase as in the ST6GalNAc I transfectants. For exemple, the normal human colonic epithelium expresses STn, but it is not detected by most anti-STn mAbs as it is present in an acetylated form not recognized by these reagents. The loss of an acetyltransferase allows unmasking of the STn epitope in colon tumors (Jass et al., 1995; Jass et al., 1994; Ogata et al., 1995). It has also been shown that the expression of the glycosyltransferases that elongate the O-glycan chains is decreased in tumors or cell lines that express STn epitopes (Brockhausen et al., 1998). In this situation, no increase in expression of ST6GalNAc transferase activity is required to synthesize the STn disaccharide given that the competition with the elongating enzymes is released. Moreover, in tumors, the STn epitopes are mainly expressed on mucin-type glycoproteins either secreted or membrane bound (Itzkowitz et al., 1992). These mucin-type glycoproteins present a large number of O-glycosylation sites that should outcompete a glycoprotein such as β1 as substrates of the ST6GalNAc transferase. It is therefore unlikely that in these conditions the integrin will carry significant levels of STn O-glycans. Indeed, we found that the clonal cells LSC derived from the LS174T human colon carcinoma cell line, the only cell line described as spontaneously expressing the STn antigen (Ogata et al., 1994), expressed STn epitopes on a very high molecular weight glycoprotein but not on the β1 integrin chain (data not shown). Transfection of the ST6GalNAc I transferase cDNA into the MDA-MB-231 human breast carcinoma cell line has been recently reported (Julien et al., 2001). This led to expression of STn epitopes and was accompanied by a decreased proliferation and by morphological alterations similar to those described here. The glycoproteins that carried the STn motif were not characterized but these results and ours suggest that the synthesis of STn through overexpression of the ST6GalNAc I enzyme alters cellular properties in a manner distinct from that of the STn synthesis occurring spontaneously in tumors.

In conclusion, the data presented here indicate that in nonmucin-producing cells, overexpression of the ST6GalNAc I sialyltransferase generates STn O-glycans mainly carried by the β1 chain of the α5β1 integrin. This leads to major alterations in the cellular phenotype that are mediated at least in part by impairment of signalling events downstream of the α5β1 integrin. These results address a novel issue in the involvement of glycosylation in the regulatory loop of integrins for cytoskeletal rearrangement-linked signalling events.

The authors thank Dr Narimatsu for the gift of the pCXN2 ST6GalNAc I plasmid, Dr H. Clausen for the generous gift of the anti-STn TKH2 and Dr S. Dowdy for the gift of the TAT-RhoA^Val-14 and TAT-RhoA^Asp-19 recombinant proteins. This work was supported by the Association pour la Recherche contre le Cancer (ARC) and by the Institut National de la Recherche Médicale (INSERM).