Epigenetic Regulation of Galectin-3 Expression by β1 Integrins Promotes Cell Adhesion and Migration*

Background: β1 integrins induce EMT-like phenomena. Results: β1 integrins epigenetically regulate the expression of galectin-3, which in turn enhances β1-mediated cell adhesion and migration. Conclusion: A functional feedback-loop exists between β1 integrins and galectin-3, to enhance integrin function. Significance: There is cross-talk between integrins and galectins at multiple levels, regulating events important for cancer biology. Introduction of the integrin β1- but not the β3-subunit in GE11 cells induces an epithelial-to-mesenchymal-transition (EMT)-like phenomenon that is characterized by the loss of cell-cell contacts, cell scattering, increased cell migration and RhoA activity, and fibronectin fibrillogenesis. Because galactose-binding lectins (galectins) have been implicated in these phenomena, we investigated whether galectins are involved in the β1-induced phenotype. We examined 9 galectins and, intriguingly, found that the expression of galectin-3 (Gal-3) is specifically induced by β1 but not by β3. Using β1-β3 chimeric integrins, we show that the induction of Gal-3 expression requires the hypervariable region in the extracellular domain of β1, but not its cytoplasmic tail. Furthermore, Gal-3 expression does not depend on RhoA signaling, serum factors, or any of the major signal transduction pathways involving protein kinase C (PKC), p38 mitogen-activated protein kinase (p38MAPK), extracellular signal-regulated kinase-1/-2 (ERK-1/2), phosphatidylinositol-3-OH kinase (PI3-K), or Src kinases. Instead, Gal-3 expression is controlled in an epigenetic manner. Whereas DNA methylation of the Lgals3 promoter maintains Gal-3 silencing in GE11 cells, expression of β1 causes its demethylation, leading to transcriptional activation of the Lgals3 gene. In turn, Gal-3 expression enhances β1 integrin-mediated cell adhesion to fibronectin (FN) and laminin (LN), as well as cell migration. Gal-3 also promotes β1-mediated cell adhesion to LN and Collagen-1 (Col)-1 in cells that endogenously express Gal-3 and β1 integrins. In conclusion, we identify a functional feedback-loop between β1 integrins and Gal-3 that involves the epigenetic induction of Gal-3 expression during integrin-induced EMT and cell scattering.

Introduction of the integrin ␤1but not the ␤3-subunit in GE11 cells induces an epithelial-to-mesenchymal-transition (EMT)-like phenomenon that is characterized by the loss of cellcell contacts, cell scattering, increased cell migration and RhoA activity, and fibronectin fibrillogenesis. Because galactosebinding lectins (galectins) have been implicated in these phenomena, we investigated whether galectins are involved in the ␤1-induced phenotype. We examined 9 galectins and, intriguingly, found that the expression of galectin-3 (Gal-3) is specifically induced by ␤1 but not by ␤3. Using ␤1-␤3 chimeric integrins, we show that the induction of Gal-3 expression requires the hypervariable region in the extracellular domain of ␤1, but not its cytoplasmic tail. Furthermore, Gal-3 expression does not depend on RhoA signaling, serum factors, or any of the major signal transduction pathways involving protein kinase C (PKC), p38 mitogen-activated protein kinase (p38MAPK), extracellular signal-regulated kinase-1/-2 (ERK-1/2), phosphatidylinositol-3-OH kinase (PI3-K), or Src kinases. Instead, Gal-3 expression is controlled in an epigenetic manner. Whereas DNA methylation of the Lgals3 promoter maintains Gal-3 silencing in GE11 cells, expression of ␤1 causes its demethylation, leading to transcriptional activation of the Lgals3 gene. In turn, Gal-3 expression enhances ␤1 integrin-mediated cell adhesion to fibronectin (FN) and laminin (LN), as well as cell migration. Gal-3 also promotes ␤1-mediated cell adhesion to LN and Collagen-1 (Col)-1 in cells that endogenously express Gal-3 and ␤1 integrins. In conclusion,

we identify a functional feedback-loop between ␤1 integrins and Gal-3 that involves the epigenetic induction of Gal-3 expression during integrin-induced EMT and cell scattering.
Integrins are transmembrane receptors composed of an ␣and a ␤-subunit that connect the extracellular matrix (ECM) 5 to the cytoskeleton, thus integrating the cell interior with its environment. In this way, integrins regulate cell adhesion and cell spreading, as well as cell migration, proliferation, differentiation, and ECM remodeling (1,2). Upon integrin-ligand binding, a variety of proteins associate with the cytoplasmic tail of the ␤-subunit, initiating cytoskeletal reorganization, assembly of macro-molecular adhesion complexes such as focal adhesions (FAs), and signaling through Rho GTPases and kinaseregulated pathways (3)(4)(5)(6). However, individual integrins can trigger very distinct cellular responses, even if they bind the same ligand.
We have previously characterized the cellular phenotypes induced by the ␤1 versus the ␤3 subunit in the ␤1-deficient murine epithelial cell line GE11 (7,8). GE11 cells grow in welldefined epithelial islands and express ␣v␤3 at relatively low levels. Introduction of the ␤1-subunit, which results primarily in cell-surface expression of the fibronectin (FN) receptor ␣5␤1, causes dramatic morphological changes reminiscent of an epithelial-to-mesenchymal transition (EMT), including the loss of cell-cell contacts, cell scattering, and a contractile, fibroblast-like phenotype with high cytoskeletal tension, large peripheral FAs, and multiple protrusions. The ␤1-induced phenotype is associated with high RhoA activity, fast but random cell migration, FN fibrillogenesis, and the assembly of fibrillar adhesions distributed along FN fibrils (7,8). In contrast, overexpression of the ␤3 subunit, leading to cell-surface expression of the alternative FN-receptor ␣v␤3, increases cell spreading but induces only a modest loss of cell-cell contacts and a pancake-like morphology with many small, randomly distributed FAs (8,9). Moreover, ␤3 stimulates Rac but not Rho activity and promotes directional rather than random migration. GE11 cells therefore constitute a unique model to study differential effects of integrins on Rho GTPase activation, cell migration, and other integrin-regulated processes.
Several of the ␤1-induced phenomena are associated with the actions of a family of matricellular proteins, the ␤-galactoside-binding lectins (galectins). The galectins represents 15 highly conserved proteins that bind through a carbohydraterecognition domain to N-glycosylated proteins at the cell surface and in the ECM, including integrins and several of their ligands such as laminin (LN) and FN (10 -14). Galectins can thus modulate integrin-mediated events, and this has been most extensively documented for galectin-1 (Gal-1), galectin-3 (Gal-3), and galectin-8 (Gal-8). For instance, Gal-1 stimulates integrin-mediated adhesion and signaling in platelets and smooth muscle cells, and Gal-8 modulates cell adhesion and cell spreading in a number of cell types including neutrophils, fibroblasts, and several tumor cell lines (15)(16)(17)(18)(19).
The best-characterized family member, Gal-3, has an important role in a wide variety of physiological and pathological phenomena, which is at least in part mediated through its interactions with integrins. Gal-3 regulates the adhesion of epithelial cells to collagens and laminins, and promotes keratinocyte migration over LN-332 and wound re-epithelialization in mice (14, 20 -23). Furthermore, Gal-3 stimulates neutrophil adhesion and migration, as well as eosinophil adhesion and rolling (24,25). Gal-3 also regulates adhesion and spreading of a number of cancer cell lines (26 -28). In addition, Gal-3 promotes FA turnover and ␣5␤1-dependent assembly of fibrillar adhesions and FN fibrillogenesis in tumor cells (29,30). Finally, Gal-3 increases the adhesion of disseminating cancer cells to endothelium, thereby protecting them from anoikis (12,31).
In this study, we investigated whether galectins are involved in the ␤1-induced phenotype in GE11 cells. Intriguingly, we found that ␤1 but not ␤3 specifically triggers transcriptional activation of Gal-3, through a mechanism that involves the demethylation of the Lgals3 promoter. In turn, Gal-3 promotes ␤1-mediated cell adhesion and cell migration. Thus, we identify a functional feedback-loop between ␤1 integrins and Gal-3.
Cells, Plasmids, and Transient Transfections-The pcDNA3-GFP-RhoA-Q63L construct was kindly provided by Dr. Sylvio Gutkind (NIH, Bethesda, MD). The ␤1-3 and ␤1-3-1 expression plasmids were a kind gift from Dr. Yoshikazu Takada (UC Davis, Sacramento, CA). GE11 cells are epithelial in origin and were obtained by injecting ␤1-null mouse embryonic stem cells into blastocysts, which were allowed to develop into whole chimeric embryos until day E10.5. Cells were then retrieved from these embryos, immortalized with SV-40, and selected with G418. Polarized cells which had formed small colonies were cloned (7). GE11 cells expressing human ␤1, human ␤3, or the chimeric human integrins ␤1-3 and ␤1-3-1 were generated by transfecting Phoenix packaging cells with retroviral constructs encoding the indicated integrins, to produce culture supernatants containing virus. GE11 cells were infected with the supernatants, selected with zeocin, and FACS-sorted for expression of the respective integrins at the cell surface (7,8,32). All GEderived cell lines were cultured in DMEM supplemented with 10% FCS and 100 units/ml penicillin/streptomycin (Invitrogen). Mouse keratinocytes MK␣3 were isolated as described previously (33). Briefly, skin was obtained from newborn mice and epidermis and dermis were separated with trypsin. Keratinocytes were retrieved by gentle shaking and centrifugation, and incubation in keratinocyte serum-free medium (Invitrogen) supplemented with 50 g/ml bovine pituitary extract, 5 ng/ml EGF, and penicillin/streptomycin. Spontaneously immortalized clones were obtained after several weeks. These were named MK␣3 and cultured routinely on Col-1 (3 g/ml). All cells were maintained at 37°C in a humidified atmosphere containing 5% CO 2 . Transient transfections were performed using the Amaxa nucleofector, according to the manufacturers' instructions.
Western Blotting-Cells were washed with ice-cold PBS and lysed in protein loading buffer (62.5 mM Tris-HCl, pH 6.8, 20% glycerol, 2% SDS, 5% mercaptoethanol) at 4°C. Proteins were separated on SDS-PAGE gels, transferred to polyvinylidene difluoride membranes (Millipore), and analyzed by Western blotting followed by ECL using the Super signal system (Pierce Chemical Co.).
Confocal Microscopy-Cells on glass coverslips were fixed in 2% paraformaldehyde, washed with PBS, permeabilized with 0.2% Triton X-100, and washed with PBS. Coverslips were subsequently blocked with 2% BSA in PBS for 1 h, incubated with primary antibodies for 1 h at room temperature, washed with PBS, and incubated with secondary antibodies for 1 h at room temperature. After washing with PBS, coverslips were mounted in Mowiol supplemented with DABCO (Calbiochem), and examined with a confocal Leica TCS NT microscope.
RNA Interference-For Gal-1 and RhoA knockdown, a SMARTpool from Dharmacon consisting of 4 siRNAs was used (catalogue number M-042642-00-0010 against Gal-1 and M-042634-01-005 against RhoA). For Gal-3 knockdown, a custom siRNA was ordered from Dharmacon with the target sequence GUAACACGAAGCAGGACAA. The standard siCONTROL siRNA from Dharmacon was used as negative control. Cells were transfected using the transfection reagent DharmaFECT according to the manufacturers' instructions. Experiments were performed 24 -48 h post-transfection and knockdown was verified in each experiment by Western blotting. Alternatively, short hairpins against Gal-3 (target sequences included GCAGTACAACCATCGGATGAA, CCCG-CTTCAATGAGAACAACA, CCGCATGCTGATCACAAT-CAT, CCTTTGAGAGTGGCAAACCAT, CCCAACGCAAA-CAGGATTGTT), cloned into pLKO.1, were obtained from the TRC shRNA Open Biosystems library and transfected into HEK 293FT cells together with the Virapower TM Packaging mix (Invitrogen), using Lipofectamine 2000 according to the manufacturers' instructions. Viral supernatant was harvested 48 h later, added to GE␤1 or MK␣3 cells, and positive cells were selected with puromycin.
Plates were then incubated overnight at 37°C, after which stop buffer (50 mM glycine, pH 10.4, 5 mM EDTA) was added, and the absorbance was measured at 405 nm on a Bio-Rad microplate reader. Alternatively, cells were fixed after washing and processed as described previously (33). Each condition was analyzed in triplicate, and graphs represent the average of three independent experiments.
Scratch Assays-Cells in 24-well plates were grown to confluency and subsequently serum-starved overnight. Proliferation was inhibited with 10 g/ml mitomycin C (Nycomed Inc.) 2 h before scratching the monolayers with a yellow pipette tip. Cells were then washed twice in serum-free medium to remove cell debris, stimulated with fresh medium, and incubated at 37°C. Scratched areas were photographed overnight at 10ϫ magnification. To determine the relative migration, the area that was not covered by cells at the end of the assay t(end) was calculated over the open area at the beginning of the assay t(0). The resulting ratio in untreated cells was set to 1. Experiments were performed three times in triplicate.
RhoA Assay-Cells were transfected with control or Gal-3 siRNAs, plated in tissue-culture plates, and RhoA activity was assessed 24 h post-transfection using the Rho G-LISA kit (Cytoskeleton), according to the manufacturers' instructions.
FN Fibrillogenesis Assay-Cells on coverslips were grown to confluency and then incubated overnight in DMEM containing FN-depleted serum, supplemented with 10 g/ml biotinylated FN. Thereafter, they were processed for immunofluorescence as described above. FN was visualized with FITC-conjugated strepavidin and nuclei were stained with DAPI. Several fields were photographed, and total fluorescence for each field was measured using ImageJ software and divided by the number of cells. Graphs represent the averages of three independent experiments.
Analysis of Promoter Methylation-For methylation analysis of the Lgals3 promoter, genomic DNA was isolated using the nucleospin tissue DNA isolation kit (Machery Nagel). The purified DNA was bisulfite-converted using the Epitect Bisulfite Kit (Qiagen) according to the manufacturers' instructions. Next, PCR for regions of the Lgals3 promoter was performed using the following mouse-and bisulfite-specific primers: 5Ј-TGT-TTTAGTTTATTGTGGGGAAGTT-3Ј (forward); 5Ј-AAAA-CCCAACCCTTTCTTAATACAC-3Ј (reverse), amplifying a 268-bp product which contained 21 CpGs located adjacent to the start of exon 1. PCR products were isolated from gel using a gel extracion kit (Qiagen), cloned into the pCRII-topo TA cloning vector (Invitrogen), and transformed into Top10 Escherichia coli competent cells (Invitrogen). Positive clones were selected by blue-white screening and sequenced using M-13 reverse primer (Invitrogen). At least 9 individual clones per cell line, derived from independent experiments, were analyzed. CpG island prediction and primer design were performed using MethPrimer (35).
For DNA methylation analysis using restriction enzyme digestion, genomic DNA was treated with HpaII or MspI, after which 30 cycles of PCR were performed using the following primers: 5Ј-TTGGGTGAGTCTGAGAGTCTG-3Ј (forward); 5Ј-CATCCCAAAGAGACTTAGTATG-3Ј (reverse), to amplify a 466-bp product. Reaction products were normalized for the amount of input DNA and analyzed on gel. All experiments were performed in triplicate.
Analysis of Global Genomic Methylation by HPLC-To assess global genomic methylation levels, the ratio of 5-methyldeoxycytosine (5-Me-dC) over total deoxycytosine (dC) was determined using HPLC (36). Cell lysates were depleted of RNA with 1 mg/ml RNase A (Sigma) for 5 min, after which DNA was extracted as described above. Next, 1 g of DNA was digested to single nucleotides using a combination of DNase I (Sigma), Nuclease P1 (Sigma) and alkaline phosphatase (New England Biolabs), and dC and 5-Me-dC content were quantified with an HPLC-UV system (Shimadzu) equipped with a 125 ϫ 4 mm Nucleosil 100 -10 SA column (Macherey-Nagel). The mobile phase consisted of 40 mM acetic acid in 15% acetonitrile (pH 4.8), and the flow rate was 0.6 ml/min. Global 5-Me-dC content was expressed as a percentage of the total dC content. Six replicates were analyzed per condition.

Gal-3 Expression Is Induced by the Integrin ␤1, but Not the ␤3
Subunit-To determine the expression of galectin family members in GE11 cells and GE11 cells stably expressing the ␤1-subunit (GE␤1), we performed a quantitative-PCR (Q-PCR) analysis of 9 galectins. The only galectin that was expressed at a detectable level in GE11 cells was Gal-1, and its expression was increased 4-fold in GE␤1 cells. Intriguingly, Gal-3 was hardly expressed in GE11 cells but its expression increased dramatically (28-fold) in GE␤1 cells (Fig. 1A). Next, we determined protein levels of Gal-1 and Gal-3 by Western blotting. To establish if the regulation of these galectins was specific for the ␤1 subunit, Gal-1 and Gal-3 expression was also analyzed in GE11 cells that overexpress ␤3 (GE␤3), which leads to elevated cellsurface expression of ␣v␤3 (8). Whereas Gal-1 protein was detected in GE11 cells and its expression was enhanced in both GE␤1 and GE␤3 cells, high expression of Gal-3 protein was only detected in GE␤1 but not in GE11 or GE␤3 cells (Fig. 1B). Thus, whereas both ␤1 and ␤3 enhance Gal-1 expression, high Gal-3 expression is induced by ␤1 but not by ␤3.
The Hypervariable Region of the I-like Domain of ␤1 Is Required for Gal-3 Expression-Because ␤1 but not ␤3 induces Gal-3 expression, and these integrin subunits trigger very different morphological responses, we investigated whether the induction of Gal-3 expression is correlated with cell morpho-logy. Furthermore, we addressed the importance of the ␤1-cytoplasmic tail in the regulation of Gal-3 expression. We therefore used the ␤1-3 chimeric integrin, which consists of the extracellular and transmembrane regions of ␤1, fused to the cytoplasmic domain of ␤3 ( Fig. 2A). Expression of this chimera in GE11 cells induces a phenotype similar to that induced by wild-type ␤1, including the loss of cell-cell contacts, cell scattering, and a fibroblast-like morphology with multiple protrusions and peripheral FAs ( Fig. 2A; Ref. 8). Thus, the extracellular but not the cytoplasmic domain of ␤1 is required for these effects, and the extracellular domain of ␤3 cannot substitute for it. The extracellular domains of integrin ␤-subunits vary widely in a region contained within the I-like domain, called the hypervariable region. This region affects ligand specificity and regulates the activity of Rho GTPases (37,38). Replacing the hypervariable region in ␤1 with that of ␤3 (creating ␤1-3-1) prevents RhoA-mediated contractility and FN fibrillogenesis, and induces a GE␤3-like morphology ( Fig. 2A; Ref. 32). Intriguingly, we detected high levels of Gal-3 expression in GE␤1 and GE␤1-3 cells, but not in GE␤3 or GE␤1-3-1 cells. In contrast, the level of Gal-1 expression was similar in GE␤1, GE␤1-3, and GE␤1-3-1 cells and was only marginally lower in GE␤3 cells (Fig. 2B). Thus, these findings show that Gal-3 expression is specifically correlated with the contractile, mesenchymal-like phenotype, and does not depend on the cytoplasmic tail of ␤1. Rather, the hypervariable region in the extracellular domain of   DECEMBER 28, 2012 • VOLUME 287 • NUMBER 53 JOURNAL OF BIOLOGICAL CHEMISTRY 44687 ␤1 is required for both the ␤1-induced cell morphology, and the induction of Gal-3 expression.

Epigenetic Regulation of Galectin-3 by ␤1 Integrins
␤1-induced Gal-3 Expression Is Not Regulated by RhoA Signaling or Kinase-regulated Signal Transduction Pathways-As the hypervariable region of ␤1 is required for both Gal-3 expression and RhoA activity, we reasoned that RhoA signaling might regulate Gal-3 expression downstream of ␤1. A major effector of RhoA is Rho kinase (ROCK). To test if ROCK is involved in Gal-3 expression, GE␤1 cells were incubated for 48 h with the ROCK-inhibitor Y-27632. This treatment efficiently inhibited ROCK activity, as judged by the dissolution of actin stress fibers and the induction of long, thin protrusions (supplemental Fig. S1A). However, Gal-3 expression was not affected by Y-27632 at any concentration tested (supplemented Fig. S1B). To assess whether RhoA regulates Gal-3 expression through another pathway, we depleted RhoA using RNAi. Whereas RhoA was completely absent 48 h after transfection, Gal-3 levels remained unchanged (supplemental Fig. S1C). We also investigated if active RhoA can induce Gal-3 expression independently of ␤1, by introducing a constitutively active RhoA mutant fused to GFP (GFP-RhoA-Q63L; Ref. 39) into GE␤3 cells. Transfected cells were FACS-sorted for GFP and subsequently grown overnight, after which Gal-3 protein levels were determined by Western blotting. Expression of GFP-RhoA-Q63L did not induce a GE␤1-like phenotype in GE␤3 cells, indicating that the activation of Rho signaling alone is not sufficient for the induction of EMT-like changes, and that ␤1 is absolutely required for this effect (data not shown). In line with the previous results, expression of GFP-RhoA-Q63L did not induce Gal-3 expression in the absence of ␤1 (supplemental Fig.  S1D). Taken together, these results suggest that RhoA signaling is not involved in the induction of Gal-3 expression by ␤1.
Besides Rho GTPases, ␤1-integrins activate a variety of signal transduction pathways (4,5). In an attempt to identify a ␤1-induced signaling pathway that regulates Gal-3 expression, we treated GE␤1 cells with standard concentrations of a variety of compounds to inhibit signal transduction pathways that are commonly activated by ␤1-integrins, including PD98059 and UO126 to inhibit the extracellular signal-regulated kinase-1/-2 (ERK-1/-2) cascade, SB203580 to inhibit p38 mitogen-activated protein kinase (p38MAPK), PI-103, and LY-294002 to inhibit phosphatidylinositol-3-OH kinase (PI3-K), PP1 to inhibit the Fyn/Src/Yes family of kinases, and Gö6983 to inhibit all isoforms of protein kinase-C (PKC). In addition, we used the broad-spectrum tyrosine kinase inhibitor genistein, and the broad-spectrum serine/threonine kinase inhibitor staurosporine. None of the used compounds prevented Gal-3 expression in GE␤1 cells although genistein did reduce Gal-3 expression to about 70% of control levels, which is possibly due to cytotoxic side-effects (supplemental Fig. S2A). Notably, whereas some inhibitors seemed to enhance Gal-3 expression in GE␤1 cells, they failed to do so in GE11 cells, further underlining that expression of Gal-3 cannot be activated in the absence of ␤1. We then analyzed whether Gal-3 expression in GE␤1 cells depends on soluble factors that can trigger signaling pathways at the cell-surface, by culturing GE11 and GE␤1 cells in the absence or the presence of serum for 24 to 48 h. Gal-3 expression was only slightly elevated by serum but was not prevented in its absence, suggesting that although serum-components may potentiate ␤1-induced Gal-3 expression, they are not absolutely required for it (supplemental Fig. S2B).

␤1 Integrins Induce Gal-3 Expression by Demethylation of the Lgals3 Promoter-The results reported in the previous sections indicate that Gal-3 expression in GE␤1 cells is not maintained by RhoA signaling, serum factors, or kinase-regulated signaling.
In addition, the ␤1-cytoplasmic tail, which contains docking sites for numerous signaling proteins, is dispensable for Gal-3 expression. It therefore seems unlikely that Gal-3 expression in GE␤1 cells depends on a classic integrin signaling event. The expression of Gal-3 and other galectins is often silenced epigenetically by DNA methylation, which can be reversed to induce de novo expression, for example during tumor progression (40 -42). We therefore investigated the DNA methylation status of the promoter region of Lgals3, the gene encoding Gal-3, by bisulfite sequencing. The murine Lgals3 gene contains a CpG island in the promoter region, which overlaps the start of exon 1 (Fig. 3A). Following bisulfite-conversion of genomic DNA, a 268 bp Lgals3 promoter fragment containing 21 CpG sites was amplified using a bisulfite treatment-specific primer set (Fig. 3A). The purified products were cloned into a TOPO vector, and sequencing of multiple individual clones (n Ͼ 9) revealed that virtually all the CpG sites were methylated in GE11 and GE␤3 cells, corresponding with the lack of Gal-3 expression (Fig. 3B). In contrast, the Lgals3 promoter was predominantly demethylated in GE␤1 cells (Fig. 3B).
We subsequently employed an alternative strategy to confirm the differences in methylation status of the Lgals3 promoter, making use of the selective digestive properties of the HpaII/MspI restriction enzyme pair. Whereas both enzymes recognize the CCGG sequence, digestion by HpaII but not MspI is prevented by DNA methylation. The Lgals3 promoter contains 3 HpaII/MspI sites. When fully methylated, digestion of genomic DNA with HpaII, but not MspI, should generate a PCR fragment of 466 bp with primers overlapping the restriction sites (Fig. 3C). Therefore, genomic DNA was treated with HpaII or MspI, and the 466 bp fragment was amplified by PCR and analyzed on gel. The expected fragment was generated from HpaII-digested genomic DNA isolated from GE11 and GE␤3, but not from GE␤1 cells (Fig. 3C). In contrast, digestion of genomic DNA with MspI never resulted in an amplified product, thus confirming that the Lgals3 promoter is indeed methylated in GE11 and GE␤3 cells, but not in GE␤1. These findings strongly suggest that the induction of Gal-3 expression in GE␤1 cells is caused by demethylation of the Lgals3 promoter. DNA methylation is maintained by DNA methyltransferases (Dnmts). To investigate whether the demethylation of the Lgals3 promoter in GE␤1 cells was the result of decreased overall expression or decreased activity of Dnmts, we tested Dnmt expression by Western blotting. Whereas Dnmt1 and Dnmt3a were not detected, the expression of Dnmt3b was not decreased in GE␤1 cells as compared with GE11 cells and perhaps even slightly elevated (Fig. 3D). As a readout for general Dnmt activity, we then tested the overall content of methylated cytosines in GE␤1 cells by HPLC (36). The ratio of methylated/ unmethylated cytosines was not decreased in GE␤1 cells, suggesting that demethylation of the Lgals3 promoter is a specific event, and is not the result of decreased overall Dnmt activity (Fig. 3D). To confirm that DNA methylation is the cause of Gal-3 silencing in GE11 cells, we incubated GE11 cells for 48 h with the Dnmt inhibitor decitabine, and performed digestion of genomic DNA with HpaII/MspI as described above. The 466 bp fragment was hardly amplified from GE11 cells after treatment with decitabine, indicating that the Lgals3 promoter was demethylated by this treatment (Fig. 3E). In addition, decitabine caused mRNA and protein expression of Gal-3 in GE11 cells in a dose-dependent manner, confirming that demethylation of the Lgals3 promoter leads to transcriptional activation of the Lgals3 gene (Fig. 3F). Finally, decitabine treatment also increased Gal-3 mRNA and protein in GE␤3 cells (Fig. 3G). These findings unambiguously demonstrate that 1) DNA methylation maintains Gal-3 silencing in GE11 and GE␤3 cells, and 2) demethylation is sufficient for activating Gal-3 transcription, even in the absence of ␤1.
Gal-3 Promotes ␤1 Integrin-mediated Cell Adhesion-We next investigated whether Gal-3 contributes to ␤1-mediated adhesion of GE␤1 cells. The ␤1-integrin repertoire in GE␤1 cells consists primarily of the FN-binding integrin ␣5␤1, whereas FN-binding ␣v␤1, as well as the LN-binding integrins ␣3␤1 and ␣6␤1 are also expressed at low levels (7,8). We first tested whether adhesion to FN was affected by the knockdown of Gal-3 expression with siRNA. Because ␤1 also enhanced Gal-1 levels, Gal-1 was depleted in parallel. Both siRNAs caused a strong and specific reduction of their respective targets within 24 h after transfection, while a scrambled siRNA-control had no effect (Fig. 4A). Interestingly, cell adhesion to FN was reduced to 50% of that of control cells by Gal-3 depletion, but was not affected by the loss of Gal-1 (Fig. 4, B and C). Thus, Gal-3 but not Gal-1 contributes to adhesion of GE␤1 cells to FN. Similar results were obtained using stable cell lines expressing shRNAs directed against Gal-3, and subsequently selected with puromycin (data not shown). Because the small amounts of ␣3␤1 and ␣6␤1 on GE␤1 cells allow weak adhesion to laminins, we next tested whether Gal-3 also promotes cell adhesion to LN-511, a ligand for both ␣3␤1 and ␣6␤1. In line with the results obtained on FN, Gal-3 depletion reduced cell adhesion to LN-511, albeit to a lesser extent (Fig. 4D). Thus, Gal-3 promotes ␤1-integrin-mediated cell adhesion to FN and LN-511 in GE␤1 cells.
We then determined whether Gal-3 affects ␤1-mediated adhesion in epithelial cells that endogenously express Gal-3 and ␤1-integrins. For this purpose we used a keratinocyte cell line designated MK␣3, which we have described previously (33). MK␣3 cells are non-transformed keratinocytes isolated from mouse epidermis, and their predominant ␤1-integrins are ␣2␤1 and ␣3␤1, whereas ␣5␤1 is not expressed (33). Knockdown of Gal-3 expression in MK␣3 cells reduced ␣2␤1-medi-  (n ϭ 3). D, expression of Dnmt3b in GE11 and GE␤1 cells was investigated by Western blotting (left). Content of 5-Me-dC, expressed as a percentage of total dC content, in GE11 and GE␤1 cells was determined by HPLC (right). E, GE11 cells were incubated for 48 h with 1 M decitabine, and the methylation status of the Lgals3 promoter was investigated using digestion with HpaII or MspI, followed by PCR. F, GE11 cells were incubated for 48 h with the indicated concentrations of decitabine, and expression of Gal-3 was investigated by Q-PCR (left) and Western blotting (right). G, GE␤3 cells were incubated for 48 h with the indicated concentrations of decitabine, and expression of Gal-3 was investigated by Q-PCR (left; expressed relative to GE11 without decitabine) and Western blotting (right). DECEMBER 28, 2012 • VOLUME 287 • NUMBER 53 ated cell adhesion to Col-1 (Fig. 4E). In addition, MK␣3 adhesion to LN-511, which is primarily mediated by ␣3␤1, was also decreased though less prominently than on Col-1 (Fig. 4F). Together, these results show that Gal-3 promotes ␤1-integrinmediated cell adhesion to FN and LN-511 in GE␤1 cells, and to Col-1 and LN-511 in MK␣3 cells.

Epigenetic Regulation of Galectin-3 by ␤1 Integrins
␤1-induced Gal-3 Expression Promotes Cell Migration, but Not FN Fibrillogenesis or RhoA-mediated Contractility-Next, we investigated whether Gal-3 is involved in other ␤1-induced phenomena in GE␤1 cells such as cell migration, FN fibrillogenesis, and RhoA activity. To study cell migration, GE␤1 cells were transfected with control or Gal-3 siRNAs and grown to confluence, and 24 h after transfection the monolayers were scratched with a pipette tip. Migration into the wound was then monitored by time-lapse microscopy, and migration speed was calculated from the movies. The knockdown of Gal-3 expression diminished migration speed by 30%, compared with that of cells transfected with control siRNAs (Fig. 5A). Thus, in addition to cell adhesion, Gal-3 expression in GE␤1 cells also promotes cell migration. We then investigated FN fibrillogenesis in cell monolayers 24 h after siRNA transfection by confocal microscopy. Whereas FN fibrils in GE11 cells were not detected, GE␤1 cells transfected with control or Gal-3 siRNAs both demonstrated abundant FN fibrillogenesis (Fig. 5B), and quantification revealed that the extent of fibril formation was not significantly decreased upon Gal-3 depletion (Fig. 5C). Fur-thermore, Gal-3 knockdown in GE␤1 cells did not affect RhoA activity, as determined by a RhoA G-LISA (Fig. 5D). Consistent with this finding, Gal-3 knockdown did also not disrupt the RhoA-mediated contractile phenotype of GE␤1 cells, characterized by stress fibers and FAs located in the periphery (Fig.  5E). Summarizing, these data indicate that Gal-3 expression in GE␤1 cells promotes ␤1-mediated cell adhesion and migration, but not RhoA-mediated contractility or FN fibrillogenesis.

DISCUSSION
In this study, we have identified a functional synergistic loop between ␤1-integrins and Gal-3, i.e. the epigenetic activation of Gal-3 expression during ␤1-induced EMT, which enhances ␤1-mediated cell adhesion and migration.
A number of reports demonstrate that galectins can regulate integrin-mediated cell adhesion. However, the exact mechanism underlying this phenomenon is poorly understood and galectins seem to have both promoting and inhibiting effects. For example, whereas overexpression of Gal-3 promotes adhesion and invasion of tumor cells, their adhesion is inhibited by addition of exogenous Gal-3 (27,28). Similarly, soluble Gal-8 has been shown to inhibit cell adhesion, whereas Gal-8-coated surfaces support cell adhesion, cell spreading, and integrin signaling in the same cells (15,16). It is conceivable that soluble galectins perform different functions than endogenously expressed galectins that are associated with membrane pro- teins, or immobilized galectins that are present in the ECM. However, additional complexity arises from the observation that galectins can have opposing effects across different cell types (which may have different integrin repertoires), and even on different integrins within the same cell type. An example of the former is that while soluble Gal-8 inhibits the adhesion of CHO, HeLa, and HaCaT cells, it stimulates adhesion of neutrophils (15,17). An example of the latter is that Gal-3 regulates ␣2␤1-mediated adhesion to collagens, but has no effect on FNor LN-binding integrins in MDCK cells (22).
We find that Gal-3 promotes ␤1-mediated cell adhesion to FN and LN in GE␤1 cells, and to Col-1 and LN in mouse keratinocytes. Earlier reports have documented that Gal-3 regulates surface expression of ␣6␤1 and ␣4␤7 on cancer cells (26,28,31), but we did not detect differences in ␤1 expression by flow cytometry upon knockdown of Gal-3 expression (data not shown). Moreover, we found that FN fibrillogenesis was not affected upon Gal-3 depletion, while a previous study has shown that soluble Gal-3 can promote this process in mammary tumor cells (30). The discrepancy between the results of the latter study and ours probably reflects differences in the used cell system, as well as differences in the effects of soluble Gal-3 versus endogenous Gal-3.
It is clear that the effects of galectins on integrin function are complex and depend on the integrin repertoire, the cell type, and the particular galectin involved. Accumulating evidence suggests that the pattern of glycosylation is of key importance to the interactions of galectins with their ligands, and thus to the effects of galectins on cell behavior. Galectins bind to N-acetyllactosamine (typically Gal␤1,4GlcNAc), which is a recurring sequence in branched N-glycans, and the affinity of galectins increases with the number of branches. Affinity is also regulated by adjacent saccharides, as different galectins show specificity for different oligosaccharides (11,43). Many galectins exist as dimers or oligomers, and Gal-3 can form up to pentamers, which are able to cross-link multiple ligands at the cell surface to form lattices of glycoproteins (44,45). In this way, galectins can cluster integrins but they can also form heterogeneous lattices consisting of integrins, non-integrin proteins that regulate adhesion such as MUC1, and other glycoproteins including cytokine-and growth factor receptors (46,47). Thus, it is increasingly recognized that the effects of galectins can only be understood completely with knowledge of the molecular composition of the glycoproteome at the cell surface, and its pattern and degree of glycosylation (43).
Irrespective of the mechanism by which Gal-3 enhances ␤1 function, our data reveal a new aspect of galectin-integrin biology, as we document the transcriptional activation of Lgals3 by an integrin, which clearly has significance for integrin function. Gal-3 was specifically induced by ␤1, whereas neither endogenous ␣v␤3 in GE11 cells, nor overexpression of the ␤3-subunit in GE␤3 cells was able to induce Gal-3 expression. Moreover, introduction of the ␤4-subunit in GE11 also failed to induce Gal-3 (data not shown). Neither of these integrins supports the EMT-like phenotypic changes induced by ␤1 integrins, and the correlation between this phenotype and Gal-3 expression is further strengthened by our observations using chimeric integrins. Surprisingly, the induction of Gal-3 expression was independent of the ␤1 cytoplasmic tail, RhoA signaling, or common integrin-induced signal transduction pathways. Rather, its expression in GE␤1 cells is regulated in an epigenetic fashion by the demethylation of DNA sequences around the transcriptioninitiation site of the Lgals3 gene, which are methylated in GE11 and GE␤3. Demethylation of the Lgals3 promoter is a specific FIGURE 5. Gal-3 promotes cell migration, but not FN fibrillogenesis or RhoA activity in GE␤1 cells. A, monolayers of GE␤1 cells transfected with control or Gal-3 siRNAs were wounded with a yellow pipette tip and then imaged overnight. The relative migration speed was determined as the ratio of the wound area after overnight migration t(end) over the wound area at the start of the experiment t(0), and expressed relative to the control-transfected cells. The graph represents the averages of three independent experiments (p Ͻ 0.05). Scalebar, 100 m. AU, arbitrary units. B, monolayers of GE11 cells and GE␤1 cells transfected with control or Gal-3 siRNAs were incubated in FN-depleted medium, supplemented with biotinylated FN. FN fibrils were visualized with FITCconjugated streptavidin (green), and nuclei were stained with DAPI (blue). Scalebar, 50 m. C, ratio between total FN fluorescence and the total number of cells was determined using ImageJ, and the averages of three independent experiments are depicted in the graph. AU, arbitrary units. D, GE␤1 cells were transfected with control or Gal-3 siRNAs and seeded on tissue-culture plates. After 24 h of incubation, RhoA activity was assessed using a G-LISA. AU, arbitrary units. E, GE␤1 cells were transfected with control or Gal-3 siRNAs, seeded on coverslips, and after overnight incubation they were processed for immunofluorescence. FAs were visualized using an antibody against paxillin (green), phalloidin was used to stain F-actin (red), and DAPI to stain nuclei (blue). Scalebar, 20 m. event, and is not the result of a loss or reduction of global methylation in GE␤1 cells. Our data emphasize the importance of DNA methylation in the regulation of the expression of various galectins, in line with previous reports (40 -42).
In the light of our data, two previous studies are particularly interesting. First, it has been shown that the cell-surface glycoprotein MUC1 controls Gal-3 expression in an epigenetic manner in cancer cells, be it through a miRNA-dependent mechanism rather than DNA-methylation. In turn, Gal-3 binds MUC1 at the cell surface and bridges it with the epidermal growth factor receptor, thereby establishing a functional feedback-loop (47). Second, integrin ␣6␤4 activates the expression of a number of target genes involved in invasion and metastasis, by demethylation of their promoters in MDA-MB-435 human mammary carcinoma cells (48). Thus, these reports provide additional evidence that integrins, as well as other cell surface receptors, can function in a regulatory feedback loop that involves the induction of the expression of accessory proteins via epigenetic mechanisms rather than through a conventional signal transduction cascade.
Future research should aim at the identification of factors that regulate the integrin ␤1-mediated demethylation of the Lgals3 promoter, and the link between Gal-3 expression and the acquirement of a mesenchymal phenotype. As Gal-3 is considered to be an important player in cancer progression and metastasis, and its expression is frequently regulated by DNA demethylation in tumor cells, it will be interesting to determine whether integrins are responsible for de novo expression of Gal-3 during tumor progression.