Chondroitin sulfate–mediated N-cadherin/β-catenin signaling is associated with basal-like breast cancer cell invasion

Tumor metastasis involves cancer cell invasion across basement membranes and interstitial tissues. The initial invasion step consists of adherence of the tumor cell to the extracellular matrix (ECM), and this binding transduces a variety of signals from the ECM to the tumor cell. Accordingly, it is critical to establish the mechanisms by which extracellular cues influence the intracellular activities that regulate tumor cell invasion. Here, we found that invasion of the basal-like breast cancer cell line BT-549 is enhanced by the ECM component chondroitin sulfates (CSs). CSs interacted with and induced proteolytic cleavage of N-cadherin in the BT-549 cells, yielding a C-terminal intracellular N-cadherin fragment that formed a complex with β-catenin. Of note, the cleavage of N-cadherin increased cytoplasmic and nuclear β-catenin levels; induced the matrix metalloproteinase 9 (MMP9) gene, a target of β-catenin nuclear signaling; and augmented the invasion potential of the cells. We also found that CS-induced N-cadherin proteolysis requires caveolae-mediated endocytosis. An inhibitor of that process, nystatin, blocked both the endocytosis and proteolytic cleavage of N-cadherin induced by CS and also suppressed BT-549 cell invasion. Knock-out of chondroitin 4-O-sulfotransferase-1 (C4ST-1), a key CS biosynthetic enzyme, suppressed activation of the N-cadherin/β-catenin pathway through N-cadherin endocytosis and significantly decreased BT-549 cell invasion. These results suggest that CSs produced by C4ST-1 might be useful therapeutic targets in the management of basal-like breast cancers.

Breast cancer comprises heterogeneous tumors with different clinical characteristics, disease courses, and responses to specific treatments. Tumor-intrinsic features, including classical histological and immunopathological classifications, separate breast tumors into multiple groups. The presence of specific markers in breast cancer has long been recognized to define subtypes with differential overall prognosis and to identify tumors susceptible to targeted treatments. The chief markers assessed are estrogen receptor (ER), 2 progesterone receptor (PR), and human epidermal receptor 2 (HER2). Combinations of these markers allow for the assignment of individual cases to specific categories, namely ER ϩ (ER ϩ /HER2 Ϫ ), HER2 ϩ (ER Ϫ / HER2 ϩ ), triple negative (ER Ϫ /PR Ϫ /HER2 Ϫ ), and triple positive (ER ϩ /PR ϩ /HER2 ϩ ). Basal or basal-like subtype broadly corresponds to the ER Ϫ /PR Ϫ /HER2 Ϫ cohort (1,2), and these subtypes are linked to the worst prognosis among the subtypes (3). Thus, a better understanding of the specific nature of the basallike subtype, especially its highly invasive property, is required.
N-cadherin is a cell-cell adhesion molecule that contributes to the invasive/metastatic phenotype and is also known as an epithelial-mesenchymal transition marker. Up-regulation of N-cadherin is more likely to be observed in breast tumors with a basal-like phenotype (4). The expression of N-cadherin correlates with both invasion and cell motility, and the forced expression of N-cadherin in non-invasive cells produces invasive cells (5). It has been reported that N-cadherin increases cell motility and invasive potential mediated by FGF receptor signaling (5). FGF receptor signaling dramatically up-regulates the expression of the matrix metalloproteinase MMP-9, which may endow cancer cells with a greater ability to degrade matrix protein barriers (6). Another report showed that the cleavage of N-cadherin by the disintegrin metalloproteinase ADAM10 regulates ␤-catenin signaling and plays a significant role during tumor invasion (7). These studies indicate that N-cadherin is responsible for invasive/metastatic properties and have examined N-cadherin-dependent intracellular events. However, which extracellular cues control tumor cell properties via N-cadherin has remained unclear.
Chondroitin sulfate (CS), a type of glycosaminoglycan, is present on the cell surface and in the extracellular matrix (ECM). There is ample evidence for a pro-tumorigenic role of CS in enhancement of cell proliferation, cell motility, and metastasis (8 -11). CSs in the ECM act on tumor cells to pro-mote tumor progression. For example, CS interacts with tumor cells via the multifunctional transmembrane protein CD44 (12,13). The interaction of CS with CD44 triggers CD44 cleavage and thereby may be involved in tumor progression (14). Furthermore, it has been reported that the tumor-associated glycocalyx that includes CSs plays a key role in promoting and regulating breast cancer progression and metastasis (15).

Results
The most common sites of breast cancer metastasis are the bone, liver, and lung. In particular, the basal-like subtype of breast cancer preferentially metastasizes to the liver and brain (21). We speculated that the structural differences in CSs could affect the invasive potential of basal-like breast cancer cells and permit site-specific metastasis. Of the six breast cancer cell lines shown in Fig. 1A, panel a, HCC1954 (HER2 subtype), MDA-MB-231, and BT-549 cells (basal-like subtype) showed higher invasive potential than the other subtypes (Fig. 1A, panels a, open bars, and b). Interestingly, the invasive activity of the basal-like subtypes (MDA-MB-231 and BT-549 cells) was elevated by treatment with chondroitin sulfate E (CS-E), whereas invasion of the HER2 subtype (HCC1954 cells) was not affected by CS-E (Fig. 1A, panels a, closed bars, and panel b). CS consists of repeating disaccharide units of N-acetylgalactosamine (GalNAc) and glucuronic acid, and it has various structural modifications resulting from sulfation of different positions in the sugar residues. There are at least five different types of CSs (CS-A, CS-B, CS-C, CS-D, and CS-E), as defined by the sulfation pattern of the repeating disaccharide moieties. Major component disaccharide units of CS-A, CS-C, and CS-E are the A-unit, the C-unit, and the E-unit, respectively. Although CS-E enhanced the invasive activity of BT-549 cells, CS-A and CS-C did not (Fig. 1B). These results indicated that CS-E specifically activated the invasion potential of BT-549 cells. In addition, cell proliferation activities were not affected by CS-E (Fig. 1C).
We first focused on the previously described link between canonical Wnt signaling activation and mammary gland tumors that was shown in the mouse (22) to investigate the mechanism underlying CS-E-dependent up-regulation of the invasion potential of BT-549 cells. In addition, frequent alterations in expression of ␤-catenin have been reported in breast cancer cell lines (23). We therefore examined whether CS-E could activate the canonical Wnt-signaling pathway, in which ␤-catenin functions as a central player. ␤-Catenin functions as a transcription cofactor in a complex with T cell factor/lymphoid enhancer factor (TCF/LEF) to regulate target gene expression. Canonical Wnt signaling can be measured using the reporter vector, pTcf7wtluc, which carries seven repeats of the TCF-binding consensus sequence upstream of the IFN-␤ basal promoter followed by the luciferase gene (24). TCF reporter activities in BT-549 cells were increased by treatment with CS-E in a dose-dependent manner (Fig. 1D, panel a). The effect of CS-E on TCF reporter activities was not canceled by overexpression of LEF1 (Fig. 1D, panel b). These results suggest that CS-E controls the expression level of ␤-catenin in these cells. We next examined the mRNA expression levels of genes related to the Wnt-signaling pathway in the six breast cancer cell lines using quantitative real-time PCR. The mRNAs for the Wnt ligands, Wnt-1, Wnt-2, and Wnt-3a, were very weakly expressed, and the mRNAs for the receptors for the canonical Wnt-signaling pathway, LRP5 and LRP6, were expressed at low levels in all six human breast cancer cell lines (Fig. 1E). The mRNA for the Dickkopf canonical WNT signaling pathway inhibitor 1 (Dkk1), which antagonizes canonical Wnt signaling by inhibiting LRP5/6 interaction with Wnt ligands, was expressed at various levels in the six cell lines (Fig. 1E), and Dkk1 proteins were highly expressed in BT-549 and MDA-MB-231 cells (Fig. 1F). We examined the possibility that Dkk1 might be a target of CS-E in BT-549 cells, because surface plasmon resonance (SPR) analysis demonstrated that Dkk1 specifically binds to CS-E (Fig. 1G). Neutralization of Dkk1 by anti-Dkk1 antibodies did not affect TCF reporter activities in BT-549 cells (Fig. 1H, panel a). In addition, CS-E could activate TCF reporter activities even when Dkk1 was blocked by its antibodies (Fig. Wnt-1  Wnt-2  Wnt-3a  LRP5  LRP6  DKK1   MCF7  T47D  BT-474  HCC1954  MDA-MB-231  BT-549  MCF7  T47D  BT-474  HCC1954  MDA-MB-231  BT-549  MCF7  T47D  BT-474  HCC1954  MDA-MB-231  BT-549  MCF7  T47D  BT-474  HCC1954  MDA-MB-231  BT-549  MCF7  T47D  BT-474  HCC1954  MDA-MB- Chondroitin sulfate drives invasion of breast cancer cells 1H, panel a). Furthermore, anti-Dkk1 antibodies had no effects on the invasion activities of BT-549 cells, and did not abrogate the increased invasion induced by CS-E (Fig. 1H, panel b).

Chondroitin sulfate drives invasion of breast cancer cells
These results suggested that CS-E up-regulated ␤-catenin-dependent transcriptional activities in BT-549 cells, but that the effect of CS-E was not mediated via the canonical Wnt-signaling pathway. ␤-Catenin is not only a key transcription factor in canonical Wnt signaling but also a structural adaptor protein linking cadherins to the actin cytoskeleton in cell-cell adhesion (25). The cadherin-bound ␤-catenin can be released and made available for signaling. We therefore examined whether CS-E elevates the invasion potential of BT-549 cells by acting on cadherins, thereby increasing ␤-catenin-mediated signaling. It was previously reported that BT-549 cells highly express N-cadherin (5) and that CS-E binds to N-cadherin with high affinity (26). Consistent with these reports, adhesion of BT-549 cells to immobilized CS-E was significantly inhibited by anti-N-cadherin antibodies ( Fig. 2A). In addition, blocking of the binding of CS-E to N-cadherin by anti-N-cadherin antibodies suppressed ␤-catenin-mediated transcriptional activities (Fig. 2B). These results suggested that CS-E mainly utilizes N-cadherin as a receptor and controls ␤-catenin signaling via N-cadherin in BT-594 cells. Next, we investigated whether up-regulation of ␤-catenin signaling by CS-E via N-cadherin could enhance cell migration and invasion. A scratch assay indicated that anti-Ncadherin antibodies completely abrogated the increased migration that was induced by CS-E at 24 h (Fig. 2C). Cells treated solely with anti-N-cadherin antibodies migrated slightly faster at 12 h compared with untreated cells, probably because cellcell adhesion was reduced by dissociation of homophilic interactions between N-cadherin molecules. In addition, the invasion of BT-549 cells into Matrigel that was enhanced by CS-E was completely blocked by treatment with anti-N-cadherin antibodies (Fig. 2D). These combined data suggested that the binding of CS-E to N-cadherin up-regulates the migration and invasion of BT-549 cells. Moreover, as shown in Fig. 2E, the activities of CS-E were unaffected by heat treatment of CS-E at 94°C for 20 min, excluding the possibility that up-regulation of the invasion was caused by protein components contaminated in CS-E preparations. Furthermore, heparin did not enhance the invasive activity of BT-549 cells, although highly sulfated CS like CS-E likely behaves as a heparin mimetic (Fig. 2F). In addition, removal of endogenous CSs from BT-549 cells by pre-treatment with chondroitinase ABC remarkably diminished the invasive potential of BT-549 cells (Fig. 2F).
We further confirmed that N-cadherin could function as a receptor of CS-E to elevate the invasive activities. A previous report showed that non-invasive MCF7 cells acquired an invasive phenotype by overexpression of N-cadherin (6). As shown in Fig. 2G, N-cadherin-expressing MCF7 cells showed an invasive property, and their invasive activities were enhanced by CS-E. In contrast, although HCC1954 cells showed an invasive phenotype, their invasive activities were not enhanced by CS-E (see Fig. 1A). Remarkably, HCC1954 cells expressed no N-cadherin (Fig. 2H, panel a), and ␤-catenin transcriptional activities were not affected in the presence of CS-E (Fig. 2H, panel b). These results suggest that the cell-surface expression levels of N-cadherin control the CS-E-induced invasive properties of breast cancer cells.
Next, we examined the cellular localization of N-cadherin and ␤-catenin in BT-549 cells in the presence or absence of CS-E. In the absence of CS-E, N-cadherin was mainly expressed at the plasma membrane, but small amounts of intracellular vesicular staining could also be seen scattered throughout the periphery of the cells (Fig. 3A, panels a and e). Treatment of the cells with CS-E increased the intracellular distribution of N-cadherin (Fig. 3A, panels b and f), and vesicular N-cadherin staining was observed in a perinuclear position. In parallel with the altered localization of N-cadherin from the plasma membrane to the cytoplasm, the intracellularly localized pool of ␤-catenin also increased with CS-E treatment (Fig. 3A, panels c-h). Previous studies have indicated that matrix metalloproteinase-and ␥-secretase-mediated N-cadherin cleavage leads to the generation of intracellular C-terminal fragments and modulates signal transduction through influencing the cytoplasmic ␤-catenin pool (7). We therefore investigated whether N-cadherin was cleaved in the presence of CS-E using immunoblotting. In the absence of CS-E, N-cadherin was detected mostly as the full-length form, whereas a proteolytic fragment appeared in the presence of CS-E, albeit at a low level (Fig. 3B). These results suggested that N-cadherin is cleaved in a CS-Edependent manner. The intracellular C-terminal fragments of N-cadherin are considered to be involved in signal transduction (7,27). Therefore, to investigate the function of the C-terminal cytoplasmic domain released by proteolytic cleavage of Ncadherin, an EGFP-fused cytoplasmic domain of N-cadherin (N-Cad(C)-EGFP) was transiently expressed in BT-549 cells Three independent experiments were carried out, and box-and-whisker graph was plotted using data combined from these experiments. Statistical significance was assessed using a Student's t test. Panel b, invaded cells on the lower surface of the membrane in the invasion assay were stained with Giemsa stain, photographed under a light microscope, and counted. B, invasion activity of BT-549 cells treated with CS-A, CS-C, or CS-E was examined (n ϭ 3). C, effect of CS-E on cell proliferation of six human breast cancer cell lines was investigated. Cells were treated with or without CS-E, and the cell number was measured by 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide assay (n ϭ 3). D, panel a, canonical Wnt signaling activity in BT-549 cells was measured using the pTCF7wt-luc reporter vector. The relative luciferase activity in the presence of different concentrations of CS-E was measured in duplicate, and the data from two independent experiments are presented with the S.D. (error bars). Panel b, pTCF7wt-luc reporter vector was co-transfected with the LEF expression vector (catalog no. SC114370, ORIGENE) into BT-549 cells, and the cells were treated with or without CS-E as indicated. The relative luciferase activity was measured (n ϭ 6). E, mRNA expression levels of Wnt signal pathway-related genes in six breast cancer cell lines were measured using quantitative real-time PCR (n ϭ 3). F, expression of Dkk-1 in six human breast cancer cell lines was examined by immunoblotting using anti-Dkk-1 antibody. G, panel a, concentration-dependent binding of Dkk-1 to CS-E was investigated by SPR. Wnt-3a is reported as a CS-E binding protein.  Chondroitin sulfate drives invasion of breast cancer cells (Fig. 3C). When EGFP alone was expressed, ␤-catenin was distributed not in the nucleus but in the plasma membrane and cytoplasm (Fig. 3C, panels a-c, the cells indicated by white arrows in panel a and b). In contrast, the forced expression of N-Cad(C)-EGFP resulted in the translocation of ␤-catenin from the plasma membrane into the nucleus (Fig. 3C, panels d-f, the cells indicated by white arrowheads in d and e). Immunoprecipitation followed by immunoblotting analysis revealed a direct interaction between N-Cad(C)-EGFP and ␤-catenin (Fig. 3D). Immunoprecipitates of N-Cad(C)-EGFP obtained using either anti-EGFP antibody or anti-N-Cad(C) antibody contained ␤-catenin, whereas N-Cad(C)-EGFP was co-immunoprecipitated by an anti-␤-catenin antibody (Fig. 3D). The combined data suggested that N-Cad(C) remains complexed with ␤-catenin and translocates into the nucleus. We next examined whether the expression of N-Cad(C) could enhance ␤-catenin-dependent transcriptional activities. As shown in In contrast, knockdown of MMP2 decreased the invasion of EGFP-expressing BT-549 cells but did not affect the invasive activities of N-Cad(C)-EGFP-expressing BT-549 cells (Fig. 4F, panel b). These results suggested that the C-terminal fragments of N-cadherin led to the transcriptional induction of MMP9 in cooperation with ␤-catenin and thereby to enhancement of cell invasion. However, it is considered that ␤-catenin-targeted genes other than MMP9 are also involved in the invasion activities accelerated by the expression of N-Cad(C)-EGFP, because the inhibitor of MMPs, GM-6001, and the knockdown of MMP9 could not completely decrease the invasion activities of N-Cad(C)-EGFPexpressing cells to the same extent as the control cells expressing EGFP (Fig. 4, D and F). Next, we aimed to visualize the movement of transfected full-length N-cadherin (N-Cad(FL)-EGFP) within cells using GFP fluorescence before and after the addition of CS-E. First, we expressed N-Cad(FL)-EGFP in BT-549 cells; however, overexpression of N-Cad(FL)-EGFP caused CS-E-independent intracellular relocation of N-Cad(FL)-EGFP. This phenomenon was probably due to the fact that endogenous N-cadherin is expressed at high levels in BT-549 cells (5). When N-Cad(FL)-EGFP was expressed in HeLa cells, which express N-cadherin at lower levels compared with BT-549 cells, we could observe CS-E-dependent intracellular relocation of N-Cad(FL)-EGFP. We therefore used HeLa cells to visualize N-Cad(FL)-EGFP endocytosis induced by CS-E. When N-Cad(FL)-EGFP was expressed in confluent cells that form stable cell-cell contacts, it was mainly localized on the plasma membrane (Fig. 5A, panels a and b). In contrast, treatment of the cells with CS-E enhanced the internalization of N-Cad(FL)-EGFP, and consequently N-Cad(FL)-EGFP disappeared from the cell surface ( Fig. 5A, panels c and d). These effects of CS-E on the endocytosis of N-cadherin were specific, and other differently structured chondroitin sulfates, CS-A and CS-C, could barely induce internalization of N-Cad(FL)-EGFP (Fig. 5B). Because it was previously shown that N-cadherin specifically binds to CS-E with high affinity (26), it was considered that the interaction between N-cadherin and CS-E stimulates internalization of N-cadherin. Immunofluorescence analysis indicated that intracellular N-Cad(FL)-EGFP was partially co-localized with GM130, a marker of cis-Golgi (Fig. 5C, panels u and uЈ), but showed little merging with LAMP-1 or EEA-1, markers of lysosomes and early endosomes, respectively (Fig. 5C, panels o, oЈ, r, and rЈ). In addition, the intracellular localization of N-Cad(FL)-EGFP accorded well with that of caveolin-1 (Fig. 5C, panels x and xЈ). These results suggested that CS-E stimulates the internalization of N-cadherin, which is mediated via caveolin-positive vesicles.

Chondroitin sulfate drives invasion of breast cancer cells
Based on the data shown in Figs. 3 and 4, it was considered that proteolytic release of the cytoplasmic domain of N-cadherin is important for ␤-catenin-dependent transcriptional induction. We therefore investigated where in the cell the proteolysis of N-cadherin that is triggered by CS-E occurs. It is generally accepted that N-cadherin is cleaved on the cell surface by multiple metalloproteases containing a disintegrin domain (ADAMs), MMPs, or other transmembrane proteases (7, 28 -30). We analyzed the effect of the broad-range MMP inhibitor GM-6001, which is a potent, reversible inhibitor of zinc-containing proteases, including MMPs and ADAMs, on N-Cad(FL)-EGFP cleavage induced by CS-E. For this purpose, antibodies against the N-terminal domain of N-cadherin (clone GC-4) that were detected with Alexa594-labeled second antibodies in immunofluorescence were used in conjunction with EGFP fluorescence for detection of the cleavage of N-cadherin (Fig. 6A, right panel). Thus, if N-Cad(FL)-EGFP is cleaved and its C terminus is released into the cytosol, then the signals of the EGFP that are fused to the C terminus of N-cadherin will not co-localize or merge with the signals of GC-4 staining. In the absence of CS-E, N-Cad(FL)-EGFP was localized in the plasma membrane, where the signals derived from EGFP (green signal) and GC-4 (red signal) merged (Fig. 6A, panels a and aЈ). Following treatment with CS-E, some internalized N-Cad(FL)-EGFP molecules were labeled with both EGFP and GC-4, although others were detected as either a red or a green signal only (Fig.  6A, panels b and bЈ). When cells were treated with CS-E in the presence of GM-6001, internalized N-Cad(FL)-EGFP molecules were detected as a full-length N-cadherin protein that was labeled with both EGFP and GC-4 (Fig. 6A, panels c and cЈ). These results raise the possibility that N-cadherin is cleaved by zinc-containing proteases, including MMPs and ADAMs, after internalization into the cells, probably in their journey from the plasma membrane to a caveolin-positive intracellular compartment. However, further studies are needed to clarify the underlying mechanism.
We next examined whether endocytosis is required for the proteolysis of N-cadherin induced by CS-E. Because N-Cad(FL)-EGFP molecules incorporated within the cells were mostly co-localized with caveolin-1 (Fig. 5D, panels x and xЈ), we inhibited endocytosis of N-cadherin using nystatin, an inhibitor of the caveolae-mediated endocytosis pathway (31,32). Pre-treatment of cells with nystatin inhibited the internalization of N-Cad(FL)-EGFP that was induced by CS-E, and consequently N-Cad(FL)-EGFP was retained on the cell surface even in the presence of CS-E (Fig. 6B, panel d). In addition, N-Cad(FL)-EGFP that was internalized following CS-E treatment was co-localized with caveolin, a component of the caveolae membranes (Fig. 6C, panels b, bЈ, and bЉ). Interestingly, although nystatin blocked the transport of N-Cad(FL)-EGFP into the cells (Fig. 6C, panels c, cЈ, and cЉ), Pitstop2, a potent inhibitor of clathrin-mediated endocytosis (33), had little effect on the endocytosis of N-Cad(FL)-EGFP (Fig. 6C, panels d, dЈ, and dЉ). We next tried to detect the C-terminal cleavage fragments of N-cadherin in the presence of CS-E, but we failed to do so. Therefore, to detect these C-terminal cleavage fragments, we incubated the cells in the presence or absence of MG132 (benzyloxycarbonyl-Leu-Leu-Leu-al), an inhibitor of protein degradation by the proteasome, because it has been shown that some nuclear transcription factors can be detected by blocking their rapid proteasome-mediated degradation. Pre-treatment of BT-549 cells with MG132 allowed the C-terminal fragments of N-cadherin (N-Cad(C)) to be visualized by Western blotting (Fig. 6D), suggesting that cleavage of N-cadherin occurred in a cell-autonomous manner in BT-549 cells without the addition of exogenous CS-E (probably by endogenous CS-E). This phenomenon is considered to contribute to the high invasive activities of BT-549 cells. In addition, exogenously added CS-E enhanced the cleavage of N-cadherin (Fig. 6D). We next investigated whether the cleavage of N-cadherin could be affected by blocking the endocytosis of N-cadherin. As shown in Fig. 6E, nystatin inhibited the production of N-Cad(C), whereas Pit-stop2 could not (Fig. 6E), suggesting that caveolae-mediated endocytosis is required for proteolysis of N-cadherin. Furthermore, the invasive activities of BT-549 cells were not increased even in the presence of CS-E when the caveolae-mediated endocytosis pathway was suppressed by nystatin (Fig. 6F).
As shown in Fig. 2F, digestion of endogenous CSs of BT-549 cells with chondroitinase ABC significantly reduced the invasive activities of BT-549 cells. In addition, a small but significant proportion of N-cadherin was constitutively cleaved in BT-549 cells (Fig. 6D). It is considered that endocytosis and subsequent proteolysis of N-cadherin occur in a cell-autonomous manner, because BT-549 cells endogenously express E-units, albeit at low levels (Fig. 7A, panel a, and Table 1). We therefore hypothesized that BT-549 cells that were deficient for E-units would display a non-invasive phenotype. As shown in Fig. 7A, panel b, the E-unit is biosynthesized by the successive action of C4ST-1 and GalNAc4S-6ST using the O-unit as a precursor. Therefore, to generate BT-549 cells that were deficient for E-units, we established C4ST-1-knock-out BT-549 cells (C4ST-1KO) using the CRISPR/Cas9 system. The sequences of the guide

Chondroitin sulfate drives invasion of breast cancer cells
RNA (gRNA) target sites were analyzed, and the sites cleaved by the Cas9 nuclease were confirmed (Fig. 7B). As shown in Fig.  7A, panel a, E-units were completely depleted, A-units were decreased, and the C-and D-units were increased by the loss of C4ST-1. C4ST-1 is dominantly involved in the synthesis of E-units, as reported previously (34). Thus, C4ST-1KO cells are defective for the biosynthesis of E-units. Residual A-units in C4ST-1KO cells are synthesized by other sulfotransferases involved in the 4-O-sulfation of GalNAc residues in CS/derma-tan sulfate, C4ST-2, and D4ST-1. In addition, the amounts of CS-disaccharides were decreased in C4ST-1KO cells (Fig. 7A,  panel a), consistent with previous reports that C4ST-1 regulates CS content (34,35).
As shown in Fig. 6, the cell surface-expressed N-cadherin is transported into the cells in response to the binding of CS-E and is cleaved to release its cytoplasmic domain, which forms a complex with ␤-catenin. We therefore hypothesized that N-cadherin might be retained on the cell surface in the C4ST-

Chondroitin sulfate drives invasion of breast cancer cells
1KO cells that were defective for CS-E expression. As expected, a cell-surface biotinylation assay demonstrated that the cellsurface expression levels of N-cadherin were increased in C4ST-1KO cells compared with the parental BT-549 cells (Fig.  7C, panels a-c). Consistent with this result, flow cytometric analysis also showed an increased expression level of N-cadherin on the cell surface in C4ST-1KO cells (Fig. 7C, panel d).
In addition, we checked the total expression level of N-cadherin and ␤-catenin by Western blotting. Both N-cadherin and ␤-catenin were more highly expressed in C4ST-1KO cells than in BT-549 cells (Fig. 7D, panels a and c), probably because N-cadherin is degraded together with ␤-catenin in BT-549 cells. The binding of ␤-catenin to N-cadherin was not altered in the absence of CS-E (Fig. 7D, panel b and d). Moreover, the gene expression levels of N-cadherin and ␤-catenin were not significantly different between BT-549 cells and C4ST-1KO cells (Fig. 7D, panel e). We further compared the cellular localization of N-cadherin between BT-549 cells and C4ST-1KO cells using immunofluorescence. In non-permeabilized cells, the cell-surface distribution of N-cadherin was visualized using GC4, a monoclonal anti-N-cadherin antibody that reacts with the N-terminal half of the extracellular domain of N-cadherin. In this experiment, more N-cadherin molecules were distributed on the cell surface in C4ST-1KO cells, compared with BT-549 cells (Fig. 7E, panel a). The plasma membrane and intracellular localization of N-cadherin was examined in permeabilized cells using C32, a monoclonal anti-N-cadherin antibody that reacts with the C-terminal half of the intracellular domain of N-cadherin (see Fig. 6). In BT-549 cells, N-cadherin was localized in the cytoplasm and nucleus as well as in the plasma membrane, and in C4ST-1KO cells a large fraction of N-cadherin resided on the cell surface (Fig. 7E, panel b). We next examined which CS-PGs interact with N-cadherin. Syndecan (SDC) family is well-known as cell-surface PGs modified with HS and CS. Thus, we checked the gene expression levels of four members of the SDC family. As shown in Fig. 8A, the gene expression levels of SDC1 and SDC3 were significantly higher in C4ST-1KO cells than parental BT-549 cells. Consistent with this result, the expression of SDC1 core proteins increased in C4ST-1KO cells more than BT-549 cells (Fig. 8B). Interestingly, SDC1 core proteins were hardly modified with CS chains in C4ST-1KO cells (Fig. 8B, lanes 6 -8). We then investigated the cellular localization of N-cadherin and SDC1 in BT-549 and C4ST-1KO cells (Fig. 8C). N-cadherin was mostly co-localized with SDC1 at the plasma membrane of BT-549 cells (Fig. 8C, panels c and d). In contrast, the localiza-tion of N-cadherin and SDC1 did not merge well on the plasma membrane of C4ST-1KO cells (Fig. 8C, panels g and h). Furthermore, the interaction between SDC1 and N-cadherin was investigated. As shown in Fig. 8D, N-cadherin was co-immunoprecipitated with SDC1 in BT-549 cells, albeit at low efficiency, whereas no bands of N-cadherin interacting with SDC1 could be detected in C4ST-1KO cells. These results suggest the possibility that CS chains on the SDC1 could interact with N-cadherin.
To confirm that a difference in endocytosis gave rise to this difference in cellular localization of N-cadherin, the surface expression levels of N-cadherin were analyzed using both cellsurface biotinylation and FACS analysis in the presence of endocytosis inhibitors. As expected, the surface expression of N-cadherin in parental BT-549 cells was increased to the level of that in C4ST-1KO cells by blocking endocytosis of N-cadherin using nystatin (Fig. 9, A and B). Furthermore, a cycloheximide chase analysis demonstrated that the half-life of the fulllength N-cadherin expressed in BT-549 cells was ϳ2 h (Fig. 9C, open circle). In contrast, the level of N-cadherin in BT-549 cells remained relatively constant over the 2-h chase when endocytosis was blocked using NH 4 Cl (Fig. 9C, closed circle). N-cadherin expressed in C4ST-1KO cells was relatively stable both in the absence and presence of NH 4 Cl (Fig. 9C). Finally, we compared the production of the C-terminal fragment of N-cadherin by BT-549 and C4ST-1KO cells using Western blotting. As shown in Fig. 9D, the amount of the C-terminal fragment of N-cadherin was significantly decreased in C4ST-1KO cells compared with BT-549 cells. As shown in Fig. 9E, ␤-catenindependent transcriptional activities, which were measured using the TCF reporter vector, were significantly suppressed in C4ST-1KO cells (Fig. 9E, panel a), and thus the expression level of the ␤-catenin-target gene MMP9 was reduced (Fig. 9E, panel  b). These results indicated that the N-cadherin/␤-catenin pathway was depressed in C4ST-1KO cells due to a lack of CS-E. As expected, C4ST-1KO cells showed lower invasive activities compared with the parental BT-549 cells (Fig. 9F, panel a). In addition, exogenously added CS-E restored the invasive activities of C4ST-1KO cells to the same level as those of BT-549 cells stimulated by CS-E (Fig. 9F, panel b), suggesting that CS-E directly promotes tumor invasion. Furthermore, these data suggested that molecules involved in tumor invasion, except for C4ST-1, are fully functional in C4ST-1KO cells. Based on the combined data, it is concluded that tumor cells drive invasive activity by taking advantage of both tumor-and tissue-derived CSs.  3). E, BT-549 cells were treated with or without Pitstop2 or nystatin for 2 h in the presence or absence of MG132 and were then subjected to immunoblotting analysis using an anti-N-cadherin antibody (n ϭ 3). D and E, the band intensity of N-Cad(C) was normalized to that of N-Cad(FL). F, invasion potential of BT-549 cells treated with or without nystatin in the presence or absence of CS-E was measured using an in vitro invasion assay (n ϭ 3).

Discussion
Extensive evidence supports the importance of tumor-associated CSs in promoting aggressive and metastatic behavior of malignant cells (9,36). In addition, CSs present in the ECM of normal tissue affect tumor cells (14). A previous report showed that CD44, an ECM-binding molecule of tumor cells, is proteolytically cleaved by interaction with certain ECM components such as hyaluronan and CSs and that CD44-dependent motility is thereby up-regulated (14). This finding indicates that tumor cells utilize the glycocalyx containing CSs to invade into normal tissue. We found here that CSs could significantly increase the invasion potential of two basal-like cancer cell lines, BT-549 and MDA-MB-231 (Fig. 1A, panel a). However, the molecular mechanism by which CSs contribute to tumor cell invasion has remained unclear. We previously found that CS is a receptor for N-cadherin in osteoblasts (26). In addition, N-cadherin is expressed in highly invasive tumor cell lines, including BT-549 cells, and contributes to the invasive phenotype (5,6). These findings raised the possibility that CSs transduce signals into breast cancer cells through N-cadherin to up-regulate the tumor invasion potential. As expected, the invasion activity of basal-like subtype BT-549 cells that was elevated by stimulation with CSs was inhibited by pre-treatment with an anti-N-cadherin antibody (Fig. 2D).
What happens to N-cadherin after the binding of CSs to N-cadherin? Classical cadherins are proteolytically cleaved by multiple proteases in a context-dependent manner (30). It is generally accepted that proteolysis of cadherins proceeds in two steps. This proteolysis process is well-known to be a regulated intramembrane proteolysis, by which transmembrane proteins are cleaved to release cytoplasmic domains that often enter the nucleus to regulate gene transcription (37). In the first step, the extracellular region of N-cadherin is cleaved by multiple proteases, including ADAMs, MMPs, and other transmembrane proteases. Subsequently, a second cleavage occurs within the transmembrane segment by another protease to liberate the cytoplasmic C-terminal fragment (CTF). The CTFs of classical cadherins can also influence intracellular signaling events by interacting with molecules involved in multiple pathways. For example, the N-cadherin CTF can interact with ␤-catenin. This complex enters the nucleus, leads to ␤-catenindependent transcriptional activation, and can promote cell migration and invasion (7). Thus, proteolysis of N-cadherin is considered to be a regulated step in N-cadherin/␤-catenin nuclear signaling; however, what regulates the proteolysis of N-cadherin has remained unclear. We have shown here that binding of CSs with N-cadherin triggers endocytosis-associated proteolysis of N-cadherin to activate ␤-catenin nuclear signaling. To date, it has been reported that N-cadherin is cleaved by ADAM10, MMP-7, MT5-MMP, and presenilin 1 (7,28,29,38,39) in intracellular compartments as well as at the cell surface. In addition, ADAM10 is predominantly localized in the Golgi, and L1 adhesion molecule ectodomain cleavage occurs in the Golgi as well as at the plasma membrane (40). Therefore, endocytosis is probably necessary to deliver N-cadherin cargo to a subcellular compartment where proteases are active. Consistent with this notion, previous studies showed that proteolytic processing of protocadherin (Pcdh) is tightly linked to subcellular trafficking that is mediated by the endosomal sorting complex required for transport (ESCRT complex)-0, -I, -II, and -III (41). In early brain development or in undifferentiated cell lines, rapid endocytosis of Pcdh facilitates its proteolytic processing to generate an intracellular fragment that may enter the nucleus and regulate gene expression. As development progresses or cells differentiate, Pcdh intracellular fragments are no longer produced probably because of inhibition of endocytosis due to homophilic interactions between Pcdh molecules. In addition, the amount and the structure of CSs are tightly  Fig. 6A.

Chondroitin sulfate drives invasion of breast cancer cells
regulated during brain development (42). These findings suggest that sugar signals control cadherin homophilic interactions, endocytosis of cadherin, and its subsequent cleavage. Furthermore, as described above, the regulation of a regulated intramembrane proteolysis is thought to occur at the level of the first proteolysis step. The first ectodomain shedding of N-cadherin is mediated by MMPs. MMP7 is reported to be one of the proteases involved in the first cleavage of N-cadherin (28) and is activated by CS-E (43). Thus, CS-E might also enhance N-cadherin proteolysis through activation of the proteases involved in N-cadherin cleavage.
Dynamic behavior of cadherin that is regulated by endocytic pathways in cell migration in physiological and pathological conditions has been reported, focusing particularly on neural development and cancer metastasis (44). N-cadherin endocytosis at neural synapses is regulated by ␤-catenin, an N-cadherin binding partner (45). Activation of NMDA receptors decreases tyrosine phosphorylation of ␤-catenin, increases  2nd to 4th, 6th to 8th, 10th to 12th, and 14th to 16th lanes), which was followed by a pulldown assay using StreptAvidin-agarose beads. The

Chondroitin sulfate drives invasion of breast cancer cells
binding of ␤-catenin to N-cadherin, and thus the rate of N-cadherin endocytosis is reduced (45). Thus, tyrosine phosphorylation of ␤-catenin regulates NMDA-dependent N-cadherin endocytosis. In addition, proteoglycans control the phosphorylation of ␤-catenin through activation of the PTP1B-like phosphatase (46). Furthermore, the Slit/Robo pathway, which requires proteoglycans bearing CS modifications for its activation (47), activates its downstream tyrosine kinase Abl that phosphorylates ␤-catenin, resulting in a decrease in the affinity of ␤-catenin for N-cadherin (48). Therefore, CSs might indirectly control N-cadherin endocytosis through regulation of ␤-catenin phosphorylation in neuronal systems. In cancer cells, cadherin endocytosis plays an important role in maintaining a rounded cell morphology to adopt mitosis-related cell rounding and to decrease cell-cell adhesion for control of cancer cell migration. Thus, cadherin internalization is a key requirement for cancer metastasis. In addition, cadherin endocytosis is also regulated by cancer-associated cell signaling, such as receptor tyrosine kinase signaling. Lu et al. (49) reported that EGF signaling enhanced internalization of E-cadherin to induce epithelial-mesenchymal transition in a breast carcinoma cell line. Furthermore, it has been reported that EGF signaling and CSs are reciprocally regulated. EGF signaling increases the production of CSs in astrocytes (50), heparin-binding-EGF molecules secreted into the tumor microenvironment bind to CS-E with high affinity (51), and CSs controlled EGF signaling (52). Therefore, we cannot exclude the possibility that CSs control cadherin endocytosis in an indirect manner through tyrosine kinase signaling, although we did examine whether N-cadherin endocytosis occurred via direct interactions between N-cadherin and CS-E in this study. Why does the interaction of CSs with N-cadherin lead to endocytosis of N-cadherin? During the biogenesis of cell-cell contacts, N-cadherin shuttles between an intracellular and a plasma membrane pool, and the formation of the N-cadherindependent cell-cell contacts results from the recruitment of the intracellular pool of N-cadherin to the plasma membrane (53). It is considered that exogenously added CSs interact with cadherin and loosen the homophilic interaction between cadherin molecules. X-ray crystallography analysis has shown that homophilic interaction of cadherin is mediated by interaction between extracellular cadherin (EC)-1 (EC1) domains. These domains are present in an N-terminal cell-surface-exposed region of cadherin that is composed of tandemly repeated EC domains, EC1-EC5. The interfaces between EC1 domains have clusters of positively charged amino acids that can electrostatically interact with negatively charged CSs. Thus, binding of CSs to an EC1 domain might destabilize homophilic interactions between N-cadherins, leading to enhanced endocytosis of N-cadherin.
What is the pathophysiological significance of CS-E-induced proteolysis of N-cadherin? Breast cancer mainly metastasizes to the bony skeleton, the lungs, liver, and brain via the circulation. Liver metastasis in particular is a frequent occurrence in patients with breast cancer (54). If breast cancer liver metastasis is left untreated, the survival time is only 4 -8 months. It is therefore necessary to understand the molecular mechanism that underlies liver metastasis in breast cancer patients.
Although liver metastasis consists of multiple steps, liver-specific homing of breast cancer cells is an initial important step of metastasis, which means that it is necessary to understand the interaction between breast cancer cells and the liver microenvironment. Liver and bone express CSs and have a high content of CS-E. As shown in this study, breast cancer cells might bind through their N-cadherin to CS-E in the liver microenvironment to elevate their invasive potential via activation of the N-cadherin/␤-catenin pathway (Fig. 10). In addition, CS-E that is expressed in cancer cells might form a positive feedback loop, thereby amplifying N-cadherin/␤-catenin signaling. This notion is consistent with our finding that loss of CS-E led to a suppression of invasive potential in C4ST-1 knock-out cancer cells (Fig. 9F). Indeed, the genes encoding CS-E biosynthetic enzymes, C4ST-1 and GalNAc4S-6ST, are transcribed to a higher degree in breast carcinoma samples than in healthy tissue (55). Therefore, cell-autonomous activation of N-cadherin/ ␤-catenin signaling by CS-E might contribute to breast cancer malignancy. Further studies are needed to elucidate whether cancer cell-derived CS-E promotes tumor growth and metastasis in vivo.

Cell culture and stable transfection
The human breast cancer cell lines MCF7 (ATCC HTB-22 TM ), T47D (ATCC HTB-133 TM ), HCC1954 (ATCC CRL-2338 TM ), and BT-549 (ATCC HTB-122 TM ) and the human cervical cancer cell line HeLa (ATCC CCL-2 TM ) were obtained from the American Type Culture Collection (ATCC). The MDA-MB-231 cell line (no. 92020424) was purchased from the European Collection of Cell Culture (Salisbury, UK). RPMI 1640 medium supplemented with 10% heat-inactivated fetal bovine serum (FBS), 100 units/ml penicillin, 100 g/ml streptomycin, and 1% L-glutamine was used as the culture medium for the human breast cancer cell lines. HeLa cells were grown in DMEM supplemented with 10% heat-inactivated FBS, 100 units/ml penicillin, 100 g/ml streptomycin, and 1% L-glutamine. All cell lines were maintained at 37°C and 5% CO 2 .
The expression plasmids (pEGFP-N1-N-Cad(C) or pEGFP-N1) were transfected into BT-549 cells using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. Transfectants were cultured in the presence of 25 g/ml G418. Colonies surviving in the presence of 25 g/ml G418 were picked up and purified using a fluorescence-activated cell sorting (FACS)-based method (FAC-SAria TM III, BD Biosciences).

Plasmid construction
A human N-cadherin ORF clone, pF1KB4700-human Ncadherin, was obtained from the Kazusa DNA Research Institute (Chiba, Japan). For expression of the EGFP-fused fulllength or C-terminal region of N-cadherin, a polymerase chain reaction using the following primers and pF1KB4700-human N-cadherin as a template was performed: forward primer 1 for amplification of N-Cad(FL), 5Ј-CTAGCTAGCGCCATGT-GCCGGATAGC-3Ј (underline, NheI site; bold, start codon); forward primer 2 for amplification of N-Cad(C)), 5Ј-CTA-GCTAGCGCCATGAAACGCCGGGATAAAG-3Ј (underline,

Real-time PCR
Total RNA was isolated from cells using the Maxwell 16 LEV simplyRNA purification kit (Promega). For reverse transcription, 1 g of total RNA was treated with Moloney murine leukemia virus reverse transcriptase (Invitrogen) using random primers (nonadeoxyribonucleotide mixture; pd(N) 9 ) (Takara Bio Inc., Shiga, Japan). Quantitative real-time PCR was conducted using FastStart DNA master plus SYBR Green I in a LightCycler 1.5 (Roche Applied Science) according to the manufacturer's protocols. The housekeeping gene, GAPDH, was used as an internal control for quantification. The primers used for real-time PCR are shown in Table 2.

Invasion assay
The invasion assay was carried out using Corning BioCoat TM Matrigel invasion chambers according to the manufacturer's instructions. Cells (5 ϫ 10 4 cells/ml) were incubated in the presence or absence of 50 g/ml CS isoforms (CS-A, CS-C, and CS-E were derived from whale cartilage, shark cartilage, and squid car-tilage, respectively) (Seikagaku Corp., Tokyo, Japan) for 20 min at room temperature and were then added to the upper chamber and allowed to invade for 22 h at 37°C in a CO 2 incubator. In the case of treatment with MMP inhibitor, cells were pre-incubated for 20 min in the presence of 100 M of the MMPs inhibitor, GM-6001 (BML-E1300, Enzo Life Science), before adding CS to the cells. Cells were transfected with the following siRNA corresponding to MMP2 and MMP9. Silencer Select siMMP2 (catalog no. 4427038; Assay ID s8852) and MMP9 siRNA (catalog no. sc-29400) were purchased from Thermo Fisher Scientific and Santa Cruz Biotechnology, Inc., respectively. Twenty four hours after transfection, cells (5 ϫ 10 4 cells/ml) were seeded on a Matrigel-coated membrane and cultured for 22 h at 37°C. MCF7 cells were transfected with either pEGFP-N1 empty vector or pEGFP-N1-N-cadherin(full-length) vector and cultured for 12 h. GFPpositive cells (0.83 ϫ 10 4 cells/ml) were inoculated on a Matrigelcoated membrane.
After the non-invading cells and Matrigel were removed from the upper surface of the membrane by scrubbing with a cotton swab, invading cells (those adhering to the bottom surface of the membrane) were stained with Giemsa (WAKO Pure Chemicals, Osaka, Japan). Alternatively, cells expressing EGFP or EGFP-N-Cad(C) were counted taking advantage of the fluorescence of EGFP. For this purpose, all invaded cells on a mem-β-catenin First, CS chains expressed in a CS-E-rich target organ or in BT-549 cells themselves bind to N-cadherin. The binding of CSs to N-cadherin triggers endocytosis of N-cadherin, followed by cleavage of N-cadherin. As a result, the C-terminal domain of N-cadherin that is associated with ␤-catenin is released and is translocated into the nucleus, and transcriptional induction of MMP9 by ␤-catenin is increased, which enhances cancer cell invasion. CS chains expressed in the BT-549 cells themselves are considered to be utilized for cell-autonomous enhancement of invasive potential, whereas BT-549 cells might take advantage of CS chains expressed in a CS-E-rich N-cadherin target organ for metastasis.

Chondroitin sulfate drives invasion of breast cancer cells
brane were photographed using a Keyence BZ-8000 fluorescence microscope and counted.
It was checked whether the Matrigel-coated membrane might contain proteases using Amplite TM universal fluorimetric protease activity assay kit (AAT Bioquest Inc.). No activities of chymotrypsin, trypsin, thermolysin, protease K, protease XIV, and human leukocyte elastase were detected. In addition, the amounts of HS and CS in a Matrigel-coated membrane were measured. The membrane contained only HS (64.9 Ϯ 10.5 pmol of HS-disaccharides).

Luciferase reporter assay
One day before transfection, cells were seeded on a 24-well culture plate. The reporter plasmid, pTCF7wt-luc (1 g) (24), and the reference plasmid, pRL-TK (harboring the thymidine kinase promoter just upstream of Renilla luciferase; Promega) (0.1 g), were transfected into cells using Lipofectamine 2000 (Invitrogen). After 12 h, the cells were washed once with PBS; fresh growth medium was added, and the cells were cultured for an additional 12 h with or without either 50 g/ml CS-E or 200 ng/ml anti-Dkk1 antibody (H-120) (sc-25516, Santa Cruz Biotechnology). The cells were washed with PBS and lysed in 100 l of Passive Lysis Buffer (Promega). Firefly luciferase and Renilla luciferase activities were measured using 5 l of cell lysate with the Dual-Luciferase Reporter Assay System (Promega). The Lumat LB9507 luminometer (EG&G Berthold, Germany) was used for measurement in the linear range. "Relative activity" was defined as the ratio of firefly luciferase activity to Renilla luciferase activity.

Chondroitin sulfate drives invasion of breast cancer cells Biacore analysis
The interaction between Dkk-1 and CSs was analyzed using BiacoreJ (GE Healthcare) as described previously (18).

Scratch assay
Cells were seeded in 6-well plates and cultured until confluent. Using a pipette tip, a straight scratch was made in the center, simulating a wound. After one wash of 1 ml/well of PBS, cells were cultured in 1 ml/well of RPMI 1640 medium supplemented with 10% heat-inactivated fetal calf serum (FCS), 100 units/ml penicillin, 100 g/ml streptomycin, and 1% L-glutamine in the presence or absence of 5 g/ml anti-N-cadherin antibodies (Clone GC-4) for 30 min at 37°C, and then 2.5 l/well of 20 g/ml CS-E were added. After 0, 6, 12, and 24 h, the cells were imaged at ϫ10 magnification, and three fields of view were imaged along the wound edge. The number of cells migrating from the leading edge of the monolayer were counted.

Binding assay
Biotinylated CS-E was immobilized on a Nunc Immobilizer TM streptavidin 96-well plate (Nalge Nunc International, Rochester, NY) as described previously (56,57). Briefly, CS-E was dissolved in 100 mM MES-NaOH, pH 5.5, at a concentration of 1 mg/ml. To this solution were added 2.5 g of biotin-LC-hydrazide freshly dissolved in dimethyl sulfoxide and 0.25 l of 1ethyl-3-(3-dimethylaminopropyl)carbodiimide hydrochloride (Pierce). The mixture was incubated overnight at room temperature with continuous shaking. Excess biotinylating reagents were removed by centrifugal filtration using the Ultrafree-MC (5000 NMWL filter unit) (Millipore, Billerica, MA) against several changes of phosphate-buffered saline. After the Nunc Immobilizer TM streptavidin was washed with PBS containing 0.1% Tween 20 (PBST), 40 g/ml biotinylated CS-E was added to each well, and incubated for 1 h at room temperature. After washing with PBST, each well was blocked with PBS containing 2% BSA for 30 min at room temperature. After three washes with 100 l of PBS, 50,000 cells/well in RPMI 1640 medium containing 10 mM HEPES were added to each well and were then incubated for 30 min at 37°C. After the cells were washed with PBS containing 0.03% BSA, cellular lactate dehydrogenase was released from the cells by adding 1 ml of 1% Triton X-100. The enzyme activity of cellular lactate dehydrogenase was measured as an indicator of cell viability using the Cyto Tox-ONE homogeneous membrane integrity assay (Promega).

Generation of C4ST-1 knock-out BT-549 cells
CRISPR/Cas9 plasmids targeting the human C4ST-1 gene were constructed using a GeneArt CRISPR nuclease vector with orange fluorescent protein reporter kit (catalog no. A21174) (Thermo Fisher Scientific). Two single-stranded oligonucleotides specific for the C4ST-1 gene were annealed as follows: top oligonucleotide, 5Ј-GCGGAGGAACCCCTTCG-GTGGTTTT-3Ј, and bottom oligonucleotide, 5Ј-CACCGAA-GGGGTTCCTCCGCCGGTG-3Ј (underlined sequences were complementary to the overhang sequence in the linearized CRISPR nuclease vector). The resulting double-stranded oligonucleotides encoding CRISPR RNA specific for the C4ST-1 gene were inserted into a GeneArt CRISPR nuclease vector. The constructed CRISPR/Cas9 plasmids targeting the human C4ST-1 gene were transfected into BT-549 cells using Lipofectamine 2000 (Thermo Fisher Scientific). Forty eight hours after transfection, orange fluorescent protein-positive cells were isolated using the FACSAria TM III cell sorter. These cells were cloned by limiting dilution analysis and were then analyzed by genome sequencing to check the accurate and efficient gene editing of the C4ST-1 gene using the following primers: forward sequencing primer for C4ST-1, 5Ј-TTTCTTCCTC-CCTCCCAGAT-3Ј, and reverse sequencing primer for C4ST-1, CAGGAAAGGGCA.

Disaccharide analysis of GAGs from BT-549 cells
CSs isolated and purified from human breast cancer BT-549 cells were analyzed as described previously (18,19,58).

Cell-surface protein biotinylation assay
After washing with pre-chilled PBS containing 1 mM MgCl 2 and 0.1 mM CaCl 2 (PBS-CM), cells were incubated with PBS-CM containing 0.5 mg/ml Ez-Link Sulfo-NHS-Biotin (Thermo Fisher Scientific) for 30 min at 4°C with occasional shaking. After removing this reagent, PBS-CM containing 0.5 mg/ml Ez-Link Sulfo-NHS-Biotin was added to the cells again and incubated for 30 min at 4°C with occasional shaking. The cells were washed with PBS-CM and then incubated with quenching solution (4 mM glycine) for 10 min at 4°C to neutralize uncross-linked reagents.

Flow cytometry
Cells (1 ϫ 10 5 cells) were fixed with PBS containing 4% paraformaldehyde on ice for 30 min. After washing with PBS, the cells were incubated with PBS containing 2% BSA on ice for 30 min and were then stained with anti-N-cadherin antibody (dilution ratio 1:100) (Clone GC-4, catalog no. C3865, Sigma) on ice. After 1 h, the cells were washed and were then incubated with mouse IgG(H&L) antibody conjugated with DyLight TM 488 and pre-adsorbed (dilution ratio 1:200) (Rockland Immunochemicals Inc., Limerick, PA) on ice for 1 h. The cells were analyzed using the BD Accuri TM C6 flow cytometer (BD Biosciences).

Statistical analysis
Data are expressed as the mean Ϯ S.D. of the mean. Statistical significance was determined using Student's t test.
Author contributions-S. N. and H. Kinouchi performed the research and analyzed the data. S. N. and H. Kitagawa conceived the idea, designed the research, and wrote the manuscript. H. Kitagawa coordinated the study. All authors reviewed the results and approved the final version of the manuscript.