Sialylation of β1 Integrins Blocks Cell Adhesion to Galectin-3 and Protects Cells against Galectin-3-induced Apoptosis*

In previous studies, we determined that β1 integrins from human colon tumors have elevated levels of α2-6 sialylation, a modification added by β-galactosamide α-2,6-sialyltranferase I (ST6Gal-I). Intriguingly, the β1 integrin is thought to be a ligand for galectin-3 (gal-3), a tumor-associated lectin. The effects of gal-3 are complex; intracellular forms typically protect cells against apoptosis through carbohydrate-independent mechanisms, whereas secreted forms bind to cell surface oligosaccharides and induce apoptosis. In the current study, we tested whether α2-6 sialylation of the β1 integrin modulates binding to extracellular gal-3. Herein we report that SW48 colonocytes lacking α2-6 sialylation exhibit β1 integrin-dependent binding to gal-3-coated tissue culture plates; however, binding is attenuated upon forced expression of ST6Gal-I. Removal of α2-6 sialic acids from ST6Gal-I expressors by neuraminidase treatment restores gal-3 binding. Additionally, using a blot overlay approach, we determined that gal-3 binds directly and preferentially to unsialylated, as compared with α2-6-sialylated, β1 integrins. To understand the physiologic consequences of gal-3 binding, cells were treated with gal-3 and monitored for apoptosis. Galectin-3 was found to induce apoptosis in parental SW48 colonocytes (unsialylated), whereas ST6Gal-I expressors were protected. Importantly, gal-3-induced apoptosis was inhibited by function blocking antibodies against the β1 subunit, suggesting that β1 integrins are critical transducers of gal-3-mediated effects on cell survival. Collectively, our results suggest that the coordinate up-regulation of gal-3 and ST6Gal-I, a feature that is characteristic of colon carcinoma, may confer tumor cells with a selective advantage by providing a mechanism for blockade of the pro-apoptotic effects of secreted gal-3.

yltransferase, which adds ␣2-6-linked sialic acids to glycoproteins (4,6), is up-regulated in a number of tumors including colon adenocarcinoma, and its expression positively correlates with tumor metastasis and poor patient survival (4,5,7). Moreover, both in vitro and animal studies have implicated ST6Gal-I in regulating tumor cell invasiveness and metastasis (7)(8)(9)(10). However, the mechanisms linking elevated ␣2-6 sialylation to tumor progression are still poorly understood. Previously, our group identified the ␤1 integrin subunit as a target for oncogenic Ras-induced ST6Gal-I activity (11) and further determined that ␣2-6 sialylation of ␤1 integrins stimulated cell attachment and migration on collagen I (12). We also reported that ␤1 integrins in colon tumor tissues carry elevated levels of ␣2-6-linked sialic acid (12). Collectively, these results suggest that hypersialylation of the ␤1 integrin may contribute to the invasive tumor cell phenotype by modulating cell-matrix interactions.
There is a vast literature directed at understanding integrin association with traditional extracellular matrix ligands such as collagen, fibronectin, etc. However, accumulating data suggest that integrins also bind to galectins, a family of lectins that can associate with the matrix through interactions with laminin and fibronectin (13,14). Galectins are ␤-galactoside-binding proteins that bind to target molecules through conserved carbohydrate recognition domains. So far, at least 15 mammalian galectins have been identified, and these are known to regulate numerous biological processes including cell adhesion/migration, apoptosis, angiogenesis, and immune responses (13,15). One unique member of the galectin family is galectin-3 (gal-3), which has only one carbohydrate recognition domain, in combination with an extended N-terminal domain that is thought to promote oligomerization (16). Galectin-3 is expressed intracellularly but can also be secreted through a nonclassical pathway (17). Intracellular gal-3 generally mediates anti-apoptotic effects through carbohydrate-independent processes, whereas extracellular gal-3 binds to cell surface oligosaccharides and thereby induces apoptosis and also modulates cell-matrix interactions (15,16,18). Given the diverse locales and functions of gal-3, the ultimate effect of this lectin on tumor cell behavior, and particularly survival, likely depends upon the balance of signals arising from intracellular versus extracellular gal-3.
Like ST6Gal-I, gal-3 appears to play important roles in neoplastic cell metastasis. Transfection of gal-3 into low metastatic colon cancer cells enabled the cells to become more metastatic after inoculation into spleen or cecum of nude mice. In con-trast, transfection of antisense gal-3 into a highly metastatic colon cancer cell line reduced metastatic capability (19). Galectin-3 is reported to be highly expressed in several tumor types including colon carcinoma, and up-regulated gal-3 expression is associated with tumor aggressiveness (15,16).
Several studies have suggested that galectin binding to ␤-galactosides may be sensitive to terminal sialylation. For example, Hirabayashi's group (20) used frontal affinity chromatography to show that the binding of gal-3 to synthetic oligosaccharides (as opposed to cellular glycoproteins) could occur in the presence of ␣2-3, but not ␣2-6, sialylation of ␤-galactose. Studies of immune cells support this general concept; cell surface ␣2-6 sialylation was reported to block apoptosis induced by galectin-1 (gal-1) (21,22). However, the potential roles of gal-3 and ST6Gal-I in regulating epithelial cell survival have received little attention despite the fact that both gal-3 and ST6Gal-I are implicated in tumor progression. Accordingly, we compared the effects of gal-3 on SW48 cells that either lack endogenous ST6Gal-I (parental cells) or have forced expression of ST6Gal-I. These studies showed that gal-3 preferentially binds to cells with unsialylated ␤1 integrins. Moreover, ␣2-6 sialylation of the ␤1 integrin was found to protect cells against gal-3-mediated cell apoptosis, implicating a role for ST6Gal-I in regulating tumor cell survival.

EXPERIMENTAL PROCEDURES
Cell Lines-The human SW48 colon epithelial cell line was purchased from the ATCC (Manassas, VA). These cells are derived from a stage IV colon adenocarcinoma, and subcutaneous inoculation of SW48 cells into nude mice reportedly results in tumor formation 100% of the time (23). SW48 cells have no detectable ␣2-6 or ␣2-3 sialyltransferase activity (24). SW48 cells stably expressing either empty vector (EV) or ST6Gal-I (ST6) were established as described previously using a lentiviral vector (12), and pooled populations of lentiviral-infected clones were utilized for all studies. The cells were grown in Leibovitz's L-15 medium with 2 mmol/liter L-glutamine (Mediatech, Herndon, VA) supplemented with 10% fetal bovine serum (Hyglone, Logan, UT). EV and ST6 cells were maintained with 0.5 g/ml puromycin (Sigma) in complete culture medium (the puromycin was removed from the medium at least 1 day in advance of all experiments). All cells were cultured at 37°C in a CO 2 -free incubator.
Galectin-1/3 Cell Adhesion Assay-InnoCyte ECM adhesion assay gal-1/gal-3 kits were purchased from Calbiochem. According to the vendor, the wells were coated with 100 l/well of either 5 g/ml gal-1 or 5 g/ml gal-3 in PBS followed by blocking with 2% BSA in PBS. BSA-and poly-L-lysine-coated wells were used as negative and positive controls, respectively. The wells were never exposed to serum-containing medium during commercial processing. Cells were harvested by cell stripper (Mediatech) and washed extensively in PBS. Then the cells (at subconfluent density) were seeded onto each well in serum-free medium and incubated for 2 h at 37°C. The 2-h attachment interval was selected because epithelial cells typically bind maximally to most substrates within 2 h. As well, allowing cells to adhere for longer time points increases the possibility that cells will secrete matrix molecules such as fibronectin or collagen, which could compromise the interpretation of results. After a 2-h incubation, cells were washed with PBS twice and labeled with calcein AM for 1 h at 37°C. The number of calcein AM-labeled cells was measured by Versa-Fluor fluorometer (Bio-Rad) at 520 nm. For some experiments, cells were incubated in the presence or absence of either lactose (Sigma) or sucrose (Bio-Rad). For the ␤1 integrin blocking study, 35 g/ml function blocking antibodies against the ␤1 integrin (catalog number MAB1965, Chemicon International, Temecula, CA) or a mouse IgG1 isotype control (R&D Systems) were preincubated with SW48 cells for 45 min at 37°C prior to seeding cells onto the galectin-coated plates. All data were analyzed by a Student's paired t test.
Flow Cytometry-Cells were harvested by cell stripper (Mediatech) and were adjusted to 1 ϫ 10 6 cells/tube with 100 l of 0.2% BSA in PBS. 10 l of FITC-conjugated anti-␤1 integrin antibody (Chemicon International), FITC-conjugated mouse IgG1 (Invitrogen), or PBS only were added to each tube and incubated on ice for 60 min. The cells were washed three times with 0.2% BSA in PBS and resuspended in 500 l of 0.2% BSA in PBS for flow cytometric analysis on FACScan (BD Biosciences).
Neuraminidase Treatment of Cell Lysates-90 g of cell lysates were treated in 50 mM sodium citrate buffer (pH 6.0) in the presence or absence of different concentrations of neuraminidase from Clostridium perfringens (New England Biolabs, Ipswich, MA) for 1 h at 37°C. The cell lysates were subsequently boiled in 2ϫ SDS-PAGE sample buffer, resolved by SDS-PAGE, and immunoblotted for the ␤1 integrin (BD Biosciences).
Neuraminidase Treatment of Cells-4 ϫ 10 5 cells were treated in the presence or absence of 50 units of neuraminidase from C. perfringens (New England Biolabs) in PBS (pH 7.0) for 1 h at 37°C. Then the cells were washed three times with PBS and resuspended in serum-free culture medium for gal-3 cell adhesion assay.
Immunofluorescent Staining-Round German glass coverslips (Electron Microscopy Science, Hatfield, PA) were coated with 20 g/ml collagen I (Inamed Biomaterials, Fremont, CA) overnight. The cells were plated on the collagen-coated coverslips for 3 h at 37°C and then fixed with 3.7% formaldehyde at room temperature for 15 min. The cells were treated with 0.2% Triton X-100 in PBS for 2 min and blocked with 5% donkey serum. The cells were subsequently incubated with anti-gal-3 antibody (R&D Systems), no primary antibody, or mouse IgG2b isotype control (R&D Systems) for 1 h at room temperature followed by Alexa Fluor 488-conjugated anti-mouse IgG (Invitrogen) for 45 min at room temperature. The coverslips were mounted with Aqua-Poly/Mount medium (Polysciences Inc., Warrington, PA), and cells were viewed under ϫ60 oil objective of a fluorescent microscope (Nikon, Tokyo, Japan). Images were taken by a Nikon CoolSNAP camera.
Galectin-3 Overlay-Cell lysates (500 g of total protein/ sample) were incubated overnight at 4°C with a glycosylationinsensitive anti-␤1 integrin antibody, MAB 2000 (Chemicon International). Protein A/G agarose beads (Santa Cruz Biotechnology Inc.) were then added, and samples were incubated for 2 additional hours at 4°C with rotation. The agarose beads were washed with lysis buffer and then boiled in 2ϫ SDS-PAGE sample loading buffer. Precipitated proteins were resolved by on 7% SDS-PAGE and transferred to PVDF membranes. Membranes were exposed to alkaline phosphatase (AP)-conjugated gal-3 and developed with nitro blue tetrazolium/5-bromo-4-chloro-3-indolyl phosphate (25).
Galectin-3-induced Apoptosis-Cells were seeded in 4-well chamber slides (BD Biosciences) or 6-well plates in complete medium for 24 h. The adherent cells (40 -60% confluency) were then washed with PBS twice and incubated in serum-free medium in the presence or absence of gal-3 (R&D Systems) or gal-3 plus ␤-lactose (Sigma) for 24 h. Direct visualization of the cells following the gal-3 incubation did not reveal any significant increases in the number of detached cells, suggesting that the addition of soluble gal-3 had no effect on cell adhesion to tissue culture plastic. After gal-3 treatment, the cells were either fixed for Hoechst 33258 staining or solubilized in protein lysis buffer for cleaved caspase-3 Western blots. The 24-h time point for analyzing apoptotic responses to gal-3 was selected because this is a common time interval for evaluating apoptosis of epithelial cells and has been used previously to study gal-1and gal-3-induced apoptosis of lymphocytes. We recognize that soluble gal-3 was used in this assay, rather than the immobilized form of gal-3 used for binding assays. We considered examining apoptosis in response to immobilized gal-3 but concluded that such experiments might be difficult to interpret. Specifically, ST6 cells do not adhere well on immobilized gal-3 (as shown in Fig. 1), and extended incubations would likely result in anoikis. Accordingly, it might be difficult to discriminate between anoikis-related apoptosis versus gal-3-induced apoptosis.
To evaluate integrin involvement in gal-3-induced apoptosis, cells were allowed to adhere to tissue culture plates for 24 h as before, and then anti-␤1 integrin function-blocking antibodies were added to the cells for 45 min followed by the addition of the gal-3-containing medium. Four different antibodies were tested (at 35 g/ml): two distinct anti-␤1 function-blocking antibodies (catalog numbers MAB1965 and MAB2253, Chemicon International) and two distinct IgG1 isotype controls (Chemicon International, catalog number PP100; and R&D Systems, catalog number MAB002). Neither the anti-␤1 nor the control antibodies had any apparent effect on cell attachment or morphology (data not shown).
Hoechst 33258 Staining-Cells were fixed with 3.7% formaldehyde in PBS and treated with 0.2% Triton X-100 in PBS for 2 min. The cells were then washed with PBS and stained with 1 g/ml Hoechst 33258 (Invitrogen) for 30 min at room temperature. The slides were mounted with Aqua-Poly/Mount medium (Polysciences Inc.), and the percentage of apoptotic cells was calculated from 10 different randomly selected fields at ϫ40 objective. Apoptotic nuclei were visualized by nuclear condensation and fragmentation. All data were analyzed by a Student's paired t test.
Staurosporine-induced Apoptosis-Cells were seeded in 6-well plates in complete medium for 24 h before treatment. The cells were washed with PBS twice and incubated in 1 M staurosporine (Cell Signaling Technology) for 4 h. The cells were lysed in protein lysis buffer for cleaved caspase-3 Western blots.

RESULTS
␣2-6 Sialylation Inhibits the Binding of SW48 Cells to Galectin-3-Parental (Par), EV-transduced, or ST6Gal-I-expressing (ST6) SW48 cells were evaluated for cell adhesion to either gal-1 or gal-3. Briefly, cells were allowed to adhere for 2 h to tissue culture wells precoated with either gal-1 or gal-3. BSAand poly-L-lysine-coated wells were used as negative and positive controls, respectively. As shown in Fig. 1A, none of the cell lines exhibited any specific binding to gal-1; therefore this lectin was not studied further. In contrast, Par and EV cells demonstrated substantial binding to gal-3, whereas the binding of ST6Gal-I-expressing cells was limited. Cell attachment to gal-3 was significantly inhibited by ␤-lactose but not by an equivalent concentration of sucrose, indicating that gal-3 binds to specific carbohydrate structures (Fig. 1B). Taken together, these data suggest that ␣2-6 sialic acids, added by the ST6Gal-I sialyltransferase, block the binding of gal-3 to cell surface glycan targets.
Cell Adhesion to Galectin-3 Is Dependent upon ␤1 Integrins-The dual findings that ␤1 integrins can be differentially sialylated (5,11,12) and are binding partners for galectins (26) prompted us to test whether gal-3 binding was mediated by the ␤1 subunit. Of note, Par and ST6 cells express equivalent amounts of cell surface ␤1 integrins (Fig. 1C); therefore diminished attachment of ST6 cells is not due to a reduction in integrin expression. To assess the role of the ␤1 integrin in gal-3 binding, cell attachment was evaluated in the presence of anti-␤1 integrin function blocking antibodies. As shown in Fig.  1D, anti-␤1 blocking antibodies markedly inhibited the binding of parental cells to gal-3, and interestingly, also attenuated the binding of ST6Gal I expressors. These results suggest that ␤1 integrins play a major role in regulating cell adhesion to this lectin.
Desialylation of ST6 Cells Increases Cell Adhesion to Galectin-3-To verify that ␣2-6 sialylation inhibits gal-3 binding, cells were treated with neuraminidase to remove sialic acids and then evaluated for binding to gal-3-coated plates. We first tested neuraminidase efficacy by monitoring the electrophoretic mobility of the ␤1 integrin following neuraminidase treatment. As shown in Fig. 2A, the mature form of the ␤1 integrin from ST6 cells has a reduced electrophoretic mobility as compared with integrins from Par cells; this altered mobility was previously confirmed to result from increased ␣2-6 sialylation (12). Neuraminidase treatment of integrins from ST6 cells caused a dose-dependent increase in the electrophoretic mobility of the mature ␤1 isoform, consistent with the enzymatic removal of sialic acids. In contrast, the electrophoretic mobility of mature ␤1 integrins from Par cells was not altered by neuraminidase treatment, consistent with the fact that parental SW48 cells do not express either ␣2-3 or ␣2-6 sialyltransferases (24). Neuraminidase treatment also had no effect on the mobility of the precursor ␤1 isoform from either Par or ST6 cells, which was expected given that the precursor is thought to be localized to the endoplasmic reticulum (and is therefore not a substrate for ST6Gal-I).
As shown in Fig. 2B, ST6 cells treated with neuraminidase demonstrated a 3-fold increase in gal-3 binding as compared with untreated ST6 cells. In contrast, neuraminidase treatment had no effect on the binding of parental cells to gal-3. Collectively, these results strongly suggest that ␣2-6 sialylation inhibits the binding of SW48 cells to gal-3.
␣2-6 Sialylation of ␤1 Integrins Abolishes Galectin-3 Binding-To determine whether gal-3 binds directly to ␤1 integrins, we used a gal-3 overlay technique. Briefly, ␤1 integrins were immunoprecipitated from either parental or ST6Gal-Iexpressing cells, resolved by SDS-PAGE, and transferred to PVDF membrane. The membrane was then overlaid with an AP-conjugated gal-3. As shown in Fig. 3, AP-conjugated gal-3 can directly bind to the mature form of ␤1 integrins immunoprecipitated from parental, but not ST6Gal-I-expressing, cells. These data suggest that ␣2-6 sialylation of ␤1 integrins blocks gal-3 binding. Interestingly, a higher molecular weight band was also detected in the gal-3 overlay of ␤1 immunoprecipitates, suggesting that another gal-3 ligand associates with, and is co-precipitated by, ␤1. The identity of this glycoprotein is currently unknown. To verify that equal amounts of ␤1 were precipitated, duplicate immunoprecipitated samples were immunoblotted for ␤1 (Fig. 3). Notably, gal-3 does not appear to bind to the precursor ␤1 isoform, which is consistent with the hypothesis that this endoplasmic reticulum-resident ␤1 isoform is modified with high mannose-type glycans that are not substrates for galectins.

␣2-6 Sialylation of Cell Surface Proteins Does Not Alter the Expression Level or Subcellular Localization of Endogenous Galectin-3-A
number of studies have reported that gal-3 is highly expressed in tumor cells, including colon cancer cells (27); therefore we tested whether SW48 cells produce endogenous gal-3. To evaluate the levels of endogenous gal-3, lysates from parental, EV, and ST6Gal-I-expressing SW48 cells were immunoblotted. Results from these experiments showed that Par, EV, and ST6 cells express similar levels of endogenous gal-3 but no detectable gal-1. Promonocytic U937 cells were included as a positive control for gal-1 (Fig. 4A). In addition to assessing gal-3 protein levels, gal-3 subcellular distribution was monitored by immunofluorescent microscopy. As shown in Fig. 4, B and C, the staining of endogenous gal-3 was mainly cytoplasmic in all three SW48 populations, which is noteworthy in light of studies suggesting that a cytoplasmic localization of gal-3 is pro-tumorigenic (15,16). Thus, our data suggest that Par, EV, and ST6 cells all produce robust amounts of endogenous, cytoplasmic gal-3, and forced expression of ST6Gal-I does not alter either gal-3 levels or subcellular distribution.
ST6Gal-I Protects Cells from Galectin-3-induced Apoptosis-Given our results suggesting that ␣2-6 sialylation blocks gal-3 binding, we hypothesized that ST6Gal-I expression would attenuate gal-3-induced apoptosis in SW48 cells. To test this hypothesis, Par and ST6 cells were seeded onto chamber slides, treated with different concentrations of human recombinant gal-3 for 24 h, and then evaluated for nuclear morphology. Hoechst 33258 staining was used to detect nuclear condensa-FIGURE 1. ␣2-6 sialylation inhibits the attachment of SW48 cells to gal-3 and cell adhesion to gal-3 is dependent upon ␤1 integrins. Par, EV, or ST6Gal-I-expressing (ST6) SW48 cells were seeded onto cell culture plates coated with either gal-3 or gal-1, and binding was quantified by measuring the fluorescence of calcein AM-labeled cells. BSA-and poly-L-lysine (PL)-coated wells were used as negative and positive controls respectively. A, expression of ST6Gal-I inhibits cell adhesion to gal-3 (* denotes significant difference relative to parental cells on gal-3, p Ͻ 0.001). However, none of the cell lines tested adhered to gal-1. B, cell binding to gal-3 is inhibited by 20 mM ␤-lactose but not 20 mM sucrose, indicating that gal-3 binding is carbohydrate-dependent (* denotes significant difference from parental cells on gal-3; # denotes significant inhibition by ␤-lactose, p Ͻ 0.001). C, Par and ST6 cells were stained with FITC-conjugated anti-␤1 integrin antibody (dotted line) or FITC-conjugated mouse IgG1 (solid line) and analyzed by flow cytometry for cell surface staining of ␤1 integrins. Par and ST6 cells have similar ␤1 integrin expression on the cell surface. Mean fluorescent intensity (MFI) Ϯ S.D. from three independent experiments are given in each plot. D, parental and ST6 cells were preincubated with either an anti-␤1 integrin function blocking antibody or a mouse IgG1 isotype control for 45 min at 37°C. The cells were subsequently seeded onto gal-3-coated wells, and adhesion was quantified as described previously. As shown, gal-3 binding was significantly inhibited by the anti-␤1 integrin function blocking antibody (* denotes significant difference from parental cells; # denotes significant inhibition from ST6 cells, p Ͻ 0.001). All values from adhesion experiments represent the means and S.E. for three independent experiments performed with at least three samples per cell line. tion and fragmentation. As shown in Fig. 5A, the addition of exogenous gal-3 induced apoptosis in parental cells in a dosedependent manner, whereas the ST6Gal-I-expressing cells were protected against apoptosis. To confirm that the effects of exogenously added gal-3 were due to the binding of gal-3 to cell surface oligosaccharides, gal-3-induced apoptosis was examined in the presence and absence of ␤-lactose. These studies showed that ␤-lactose significantly inhibited gal-3-induced apoptosis (Fig. 5B). To further confirm that ␣2-6 sialylation inhibits gal-3-mediated apoptosis, we measured levels of activated caspase-3 in the parental, EV, and ST6Gal-I-expressing populations. As shown in Fig. 5C, exogenously added gal-3 stimulated caspase-3 activation (as measured by immunoblotting for the cleaved form of caspase-3) in Par-and EVtransduced cells but not in ST6Gal-I-expressing cells. As well, gal-3-induced caspase-3 activation was blocked by ␤-lactose, confirming that the apoptotic effects of gal-3 are mediated through gal-3 binding to oligosaccharide determinants.
Galectin-3-induced Apoptosis Is Dependent upon ␤1 Integrins-As shown in Figs. 1 and 3, cell adhesion to gal-3 is dependent upon ␤1 integrins, and gal-3 binds directly to the unsialylated ␤1 glycoform. Therefore, we questioned whether gal-3induced apoptosis was mediated by ␤1 integrins. To test this hypothesis, gal-3 was added to adherent parental cells in the presence or absence of anti-␤1 integrin function blocking antibodies. The cells were then lysed and monitored for caspase-3 activation. As shown in Fig. 6A, two distinct anti-␤1 integrin blocking antibodies had a markedly inhibitory effect on gal-3-induced caspase activation, whereas control antibodies had a negligible effect. To further evaluate whether differential integrin sialylation plays a specific role in regulating gal-3-induced apoptosis, we examined apoptosis in response to staurosporine. Staurosporine targets the mitochondria, and therefore the efficacy of this agent should not be influenced by changes in cell surface sialylation. As shown in Fig. 6B, staurosporine induced equivalent levels of caspase-3 activation in parental and ST6Gal-I expressors, consistent with the hypothesis that cell surface ␣2-6 sialylation functions as a specific block for gal-3 binding and activity.

DISCUSSION
Numerous reports suggest that ␣2-6 sialic acid and ST6Gal-I are elevated in multiple types of human tumors, particularly in highly invasive ones (4,5). However, the mechanisms underlying the role of ST6Gal-I in tumor cell growth and metastasis remain to be elucidated. Here, we provide new insight into how ST6Gal-I overexpression might promote tumor progression. Specifically, we find that ␣2-6 sialylation of the ␤1 integrin receptor protects epithelial tumor cells from gal-3-induced apoptosis. An expanding literature supports the idea that extracellular galectins, including gal-3 and gal-1, promote cell apoptosis. However, the vast majority of studies have focused on cells of the immune system. For example, gal-1 is thought to stimulate cell death in T lymphoma cells (28) and human immunodeficiency virus, type 1-infected T cells (29) were resuspended in serum-free medium and subsequently seeded onto gal-3-coated wells, and adhesion was quantified as described previously. As shown, the adhesion of ST6 cells to gal-3 was significantly increased by neuraminidase treatment (* denotes significant difference from parental cells; # denotes significant difference between ST6 cells with and without neuraminidase treatment, p Ͻ 0.001). PL, poly-L-lysine. integrins were immunoprecipitated (IP) from parental or ST6Gal-I-expressing cells, resolved by 7% SDS-PAGE, and transferred to PVDF membrane. The membrane was overlaid with an AP-conjugated gal-3. As shown, the AP-conjugated gal-3 bound the mature ␤1 integrin from parental cells but not ST6Gal-I expressors. Duplicate immunoprecipitated samples were immunoblotted with an anti-␤1 integrin antibody to verify equivalent levels of ␤1 in the immunoprecipitates. through binding to core 2 O-linked glycans on the cell surface. In the latter study, gal-1 binding to infected T cells was enhanced upon loss of sialic acid. Likewise, extracellular gal-3 stimulates apoptosis in human leukemia T cell lines (26), human peripheral blood mononuclear cells (26), activated mouse T cells (26), human B cell lymphoma (30), mouse thymocytes (31), mast cells (32), and neutrophils (33). Our studies now indicate that, as with immune cells, gal-3 plays a role in regulating epithelial cell survival.
Our study is also noteworthy because it addresses an apparent paradox. Why would a tumor cell simultaneously up-regulate both gal-3 and ST6Gal-I when ST6Gal-I blocks the binding of gal-3 to the cell surface? We hypothesize that increased levels of intracellular forms of gal-3 (which function largely through carbohydrate-independent mechanisms) promote tumor cell survival, whereas extracellular gal-3 is pro-apoptotic. This concept is supported by a substantial literature. For example, intracellular gal-3 protects cells against apoptotic stimuli including cisplatin, genistein, and nitric oxide (34). Furthermore, the anti-apoptotic activity of intracellular gal-3 is thought to be mediated through direct binding to molecules such as Bcl-2 family members (35), synexin (36), CD95 (37), and activated K-Ras (38). The binding of gal-3 to K-Ras prolongs Ras-induced signaling (39), which is interesting in light of studies showing that Ras activation increases ST6Gal-I expression (11,40). The subcellular distribution of intracellular gal-3 is also thought to be important. It was reported that overexpression of gal-3 in the cytoplasm increased its anti-apoptotic activity, whereas overexpression of gal-3 in the nucleus decreased its proliferative activity in prostate cancer cells (41). In our study, both parental and ST6Gal-I-expressing SW48 cells expressed high levels of intracellular gal-3 that localized primarily to the cytoplasm, suggesting that this endogenously expressed gal-3 might have anti-apoptotic activity. In contrast to the effects of intracellular gal-3, prior studies of immune cells, coupled with our current results from epithelial cells, suggest that extracellular gal-3 functions largely as an apoptosis inducer. When taken together, these results suggest that upregulating ST6Gal-I, in conjunction with gal-3 overexpression, would attenuate the pro-apoptotic effects of secreted gal-3, while leaving intact the anti-apoptotic effects of intracellular gal-3, thus providing a selective advantage for the tumor cell.
The up-regulated synthesis of gal-3 in tumor cells results in increases in both intracellular and secreted forms of gal-3, as exemplified by the observation that patients with metastatic  Values represent means and S.E. for two independent experiments (* denotes significant difference from parental cells, p Ͻ 0.001). B, Par, EV, and ST6 cells were incubated in serum-free medium in the presence or absence of 20 g/ml gal-3 or 20 g/ml gal-3 ϩ 20 mM ␤-lactose for 24 h. The percentage of apoptotic cells stained by Hoechst 33258 was counted as above. Values represent means and S.E. for two independent experiments (* denotes significant difference from parental cells, p Ͻ 0.001; # denotes significant inhibition by ␤-lactose, p Ͻ 0.001). C, cells were treated as described in panel B and then lysed and Western blotted for cleaved (activated) caspase-3 and ␤-actin.
disease have higher levels of gal-3 in sera (42). Within the tumor microenvironment, secreted gal-3 is thought to promote angiogenesis and also attenuate anti-tumor immune responses by inducing T cell apoptosis (15). As with gal-3, the synthesis of gal-1 by LNCaP prostate cancer cells caused the death of bound T cells (43). Thus, we hypothesize that the pro-apoptotic functions of galectins secreted from tumor cells will contribute to tumor evasion from the immune system by acting on T cells but will have no effect on tumor cells that have increased cell surface ␣2-6 sialylation.
In the current study, we found that expression of ST6Gal-I in colon epithelial cells markedly inhibited cell attachment to gal-3-coated plates, and attachment could be restored by pretreating ST6 cells with neuraminidase. In addition, we used a blot overlay approach to show that gal-3 binds directly and preferentially to unsialylated ␤1 integrins, as compared with sialylated integrins. Interestingly, ST6Gal-I-expressing cells did retain some binding to gal-3-coated plates, and much of this binding was ␤1 integrin-dependent. The molecular mechanisms underlying ␤1-dependent attachment of ST6 cells to gal-3 are not currently understood. It is possible that cells with ␣2-6 sialylation maintain some level of affinity for gal-3, perhaps through the lateral binding of gal-3 to polylactosamine chains, as suggested by Cummings' group (48). Clearly, additional studies will be needed to elucidate the molecular basis for ST6 cell attachment to immobilized gal-3; nonetheless, our results unequivocally demonstrate a strong inhibitory effect of ␣2-6-linked sialic acids on cell binding to gal-3, as well as gal-3-directed apoptosis.
Information concerning which cell surface receptors are responsible for transducing galectin-dependent signals is still very limited; however, the literature suggests that there is some specificity in this regard. Suzuki et al. (32) showed that gal-3 binding to RAGE (receptor for advanced glycation end products) receptors on mast cells induced oxidative stress, leading to caspase-mediated cell death. Baum's group (49) demonstrated that gal-1 could bind to ␤1 integrins, CD45, CD71, and CD43 in T cells; however, in this system, CD45 appeared to be the major receptor regulating gal-1-induced apoptosis. In contrast to Baum's study, Fukumori et al. (26) suggested that the ␤1 integrin was the crucial receptor for gal-3-mediated T cell apoptosis. The ␤1 family of integrins appears to be a particularly prevalent ligand for galectins, as evidenced by the multiple galectin/ integrin associations that have been reported: for example, gal-1 binding to ␣7␤1, gal-3 binding to ␣1␤1, and gal-8 binding to either ␣3␤1 or ␣6␤1 (15). Our results suggest that the ␤1 integrin is one of the predominant receptors mediating colon epithelial cell association with, and apoptotic responses induced by, extracellular gal-3. Importantly, we show that sialylation of ␤1 by ST6Gal-I blocks the pro-apoptotic effects of extracellular gal-3. Thus, differential sialylation of ␤1 integrins not only modulates cell responses to traditional extracellular matrix proteins like collagen and fibronectin, as we have previously reported (11,12,50,51), but also serves as a potential mechanism for regulating cell interactions with matrix-tethered galectins. In the aggregate, these studies suggest that variant integrin glycoforms play a multifactorial role in regulating tumor cell interaction with matrix. Importantly, the expression of hypersialylated integrin glycoforms occurs in response to biologically relevant events, such as oncogenic Ras-induced cell transformation (11). Moreover ␣2-6 sialylation of ␤1 integrins is elevated in human colon tumors, as compared with pairmatched controls (12).
Interestingly, anti-␤1 integrin function blocking antibodies did not eliminate all of the cell adhesion to gal-3, suggesting that other molecular interactions may be important. In accordance with these results, the gal-3 overlay experiments (Fig. 3) revealed that another galectin-binding protein was present in ␤1 integrin immunoprecipitates; the identity of this glycoprotein is currently unknown. Several other cell-associated ligands for gal-3 have been reported, including the epidermal growth factor receptor (52), transforming growth factor-␤ receptor (52,53), T cell receptor (53), Mac-2-binding protein (54), RAGE (32), Lamp-1,2, and carcinoembryonic antigen (55). Interestingly, carcinoembryonic antigen is highly expressed in colon cancer cells, and like the ␤1 integrin, is known to be hypersialylated with ␣2-6 sialic acid (56). Additional experiments will be needed to identify other galectin-binding proteins and also to dissect the specific galectin-dependent functions of these respective molecules.
In sum, the important finding of this study is that ␣2-6 sialylation of the ␤1 integrin attenuates gal-3-induced epithelial cell apoptosis, which in turn offers a molecular explanation for why the simultaneous up-regulation of ST6Gal-I and gal-3 in tumor cells might provide a selective advantage. Finally, our results have translational implications in that recombinant galectins are currently being considered for use in cancer treatment (57). It is highly likely that such a strategy would be less effective with cancers (e.g. colon carcinoma), which typically express high levels of ␣2-6 sialylation.