Lewis type 1 antigen synthase (beta3Gal-T5) is transcriptionally regulated by homeoproteins.

The type 1 carbohydrate chain, Galbeta1-3GlcNAc, is synthesized by UDP-galactose:beta-N-acetylglucosamine beta1,3-galactosyltransferase (beta3Gal-T). Among six beta3Gal-Ts cloned to date, beta3Gal-T5 is an essential enzyme for the synthesis of type 1 chain in epithelium of digestive tracts or pancreatic tissue. It forms the type 1 structure on glycoproteins produced from such tissues. In the present study, we found that the transcriptional regulation of the beta3Gal-T5 gene is controlled by homeoproteins, i.e. members of caudal-related homeobox protein (Cdx) and hepatocyte nuclear factor (HNF) families. We found an important region (-151 to -121 from the transcription initiation site), named the beta3Gal-T5 control element (GCE), for the promoter activity. GCE contained the consensus sequences for members of the Cdx and HNF families. Mutations introduced into this sequence abolished the transcriptional activity. Four factors, Cdx1, Cdx2, HNF1alpha, and HNF1beta, could bind to GCE and transcriptionally activate the beta3Gal-T5 gene. Transcriptional regulation of the beta3Gal-T5 gene was consistent with that of members of the Cdx and HNF1 families in two in vivo systems. 1) During in vitro differentiation of Caco-2 cells, transcriptional up-regulation of beta3Gal-T5 was observed in correlation with the increase in transcripts for Cdx2 and HNF1alpha. 2) Both transcript and protein levels of beta3Gal-T5 were determined to be significantly reduced in colon cancer. This down-regulation was correlated with the decrease of Cdx1 and HNF1beta expression in cancer tissue. This is the first finding that a glycosyltransferase gene is transcriptionally regulated under the control of homeoproteins in a tissue-specific manner. beta3Gal-T5, controlled by the intestinal homeoproteins, may play an important role in the specific function of intestinal cells by modifying the carbohydrate structure of glycoproteins.

Carbohydrate antigens are involved in many biological events. Among the antigens, Lewis antigens, which have Gal␤1-3/4GlcNAc␤-as a backbone structure, are the major ones expressed in human tissue. Whereas the type 2 (Gal␤1-4GlcNAc␤-) structure is distributed ubiquitously, the type 1 (Gal␤1-3GlcNAc␤-) structure is synthesized only in certain tissues, such as epithelium of the digestive tracts or pancreatic tissue. This restricted distribution of Lewis type 1 antigens has been considered to be the result of the limited expression of UDP-Gal:GlcNAc ␤1,3-galactosyltransferases (␤3Gal-Ts), 1 which synthesize the type 1 structure. Sialyl Lewis a (sLe a ), known as CA19-9 antigen in the clinical field, belongs to the Lewis type 1 antigens. CA19-9 is one of the most useful tumor markers and shown to be frequently accumulated in sera of colonic and pancreatic cancer patients (1).
We reported previously (2) the molecular cloning of ␤3Gal-T5 and presented evidence that this enzyme is responsible for the synthesis of Lewis type 1 antigens in colonic mucosa, pancreatic tissue, and cell lines derived from them. Moreover, we showed that the amount of type 1 antigen, such as CA19-9, on glycoproteins was well correlated with the ␤3Gal-T5 transcript level in these cell lines, suggesting that the transcriptional control of ␤3Gal-T5 might be the major regulatory factor for the synthesis of the Lewis type 1 structure. This enzyme efficiently acts on glycolipids as well as N-linked glycoproteins such as carcinoembryonic antigen, a well known tumor marker, and arranges its carbohydrate composition (3). In addition to the activity for synthesizing the type 1 structure, ␤3Gal-T5 can transfer a galactose in the ␤1,3-linkage to the mucin core-3 structure (GlcNAc␤1-3GalNAc␣-) which is abundantly expressed in colonic epithelium (4).
It has been reported that alterations of glycosyltransferase activity give rise to a dramatic modification of carbohydrate structures in several biological processes, such as differentiation and carcinogenesis. Caco-2 cells, derived from colon adenocarcinoma, have been frequently used as a model for intestinal differentiation because of their unique character, i.e. spontaneous differentiation into enterocyte-like cells as defined morphologically and enzymologically. In this process, ␤3Gal-T * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The tissue-specific expression and unique acceptor substrate specificity of ␤3Gal-T5 and the alteration in ␤3Gal-T activity during carcinogenesis and intestinal cell differentiation suggested its biological importance, which prompted us to examine the molecular basis for the transcriptional regulation of the ␤3Gal-T5. In a previous study, we determined the organization of ␤3Gal-T5 gene and the transcription initiation site by the 5Ј-rapid amplification of cDNA ends method, and we found several consensus sequences for known transcription factors within about 200 bp upstream of the transcription initiation site by computer analysis (2). Thus, we carried out the present study to identify essential transcription factors for the ␤3Gal-T5 gene, and we found that the intestine-specific transcription factors, Cdx and HNF1, were functional in transcriptional activation for ␤3Gal-T5. We also examined the alteration of ␤3Gal-T5 expression upon intestinal differentiation and carcinogenesis both in vitro and in vivo.

EXPERIMENTAL PROCEDURES
Cell Cultures-All cell lines were obtained from American Type Culture Collection (Manassas, VA) and maintained in Dulbecco's minimal essential medium supplemented with 10% fetal calf serum at 37°C in a 5% CO 2 , 95% air atmosphere. For differentiation experiments using Caco-2 cells, 2 mM sodium butyrate (Sigma) was added to the medium after confluence. Cells were harvested just before confluence and 2 or 4 weeks after confluence and stored at Ϫ80°C until total RNA isolation.
Cloning of cDNA for Homeoproteins and Expression in COS-1 Cells-Complementary DNAs for homeoproteins Cdx1, Cdx2, HNF1␣, and HNF1␤, were obtained with PCR using Pfx DNA polymerase (Invitrogen) and subcloned into a eukaryotic expression vector, pcDNA3.1 (Invitrogen). Their entire sequences were verified by sequencing. Sequences of primers are listed in Table I. These plasmids were also used as standard DNA for quantitative RT-PCR analysis.
For the expression of homeoproteins in COS-1 cells, ϳ1 ϫ 10 6 cells were transfected with LipofectAMINE 2000 Reagent (Invitrogen) and 1 g of the expression plasmid for each well of a 6-well plate, according to the manufacturer's directions. Cells were harvested after 24 h and used for electrophoretic mobility shift assay (EMSA) experiments.
Tissue Samples-Twenty nine pairs of colon cancer tissue and adjacent normal mucosa were obtained from surgically resected specimens. Informed consent was obtained from all patients. The specimens were clinically and pathologically examined and diagnosed as adenocarcinoma arising from the colon. The tissue samples were immediately frozen in liquid nitrogen and stored at Ϫ80°C until total RNA isolation.
Preparation of cDNA and Quantification of Transcript-Total RNA was isolated from tissue specimens and cell lines with the acid/guanidine/phenol/chloroform extraction method. Complementary DNAs were synthesized with an oligo(dT) primer from 5 g of total RNA in a 20-l (total volume) reaction mixture using a Superscript Preamplification System for First Strand cDNA Synthesis (Invitrogen). After cDNA synthesis, the reaction mixture was diluted 50-fold with water and then stored at Ϫ80°C until use.
The quantitative RT-PCR using the LightCycler real time PCR instrument (Roche Applied Science) was performed according to the distributor's manual. Standard plasmids of Cdx1, Cdx2, HNF1␣, and HNF1␤ were prepared with PCR cloning as described above. The sequences of primers and fluorescent (fluorescein isothiocyanate and LC-Red 640) probes for quantitative RT-PCR are listed in Table I. Primers were designed to be located in different exons and not to detect contaminated genomic DNA. After 5 min of preheating, 50 cycles of PCR were performed. Each cycle consisted of 0 s of denaturing at 95°C, 5 s of annealing at 55°C, and 10 s of extension at 72°C.
Reporter Plasmid Constructs-The cosmid human genomic library, kindly provided by Dr. H. Inoko (Tokai University, Kanagawa, Japan), was screened by hybridization with a full-length ␤3Gal-T5 cDNA as a probe. Two clones out of three positives were found to contain the ␤3Gal-T5 exon 1 by restriction mapping and Southern blotting. A 3.8kbp XbaI fragment containing exon 1 was subcloned into pBluescript SK(Ϫ) (Stratagene). As the 3Ј-end of this fragment was located within the first intron, a SalI/EcoRI fragment, containing the 2.9-kbp 5Јflanking region and 5Ј-half of the first exon, was subcloned into pBluescript and subsequently subcloned into the KpnI and SacI sites of the pGL3 basic vector (Promega) for promoter assay. Several deletion mutants were prepared by appropriate restriction enzyme digestion and self-ligation. Smaller mutants (Ͻ200 bp) were cloned by PCR, and entire sequences were verified by sequencing. A site-directed mutation in the homeoprotein consensus sequence was introduced into a 200-bp HincII/EcoRI construct (containing from Ϫ151 to ϩ49 of the promoter region) with a PCR-based mutagenesis kit (Takara).
Reporter Assay-Approximately 24 h before transfection, cells were removed with trypsin, and ϳ5 ϫ 10 5 cells were seeded in a 24-well plate. Cells were transfected using the lipofection method with Lipo-fectAMINE 2000 Reagent (Invitrogen), 0.5 g of reporter plasmid (pGL3), and 10 ng of internal reference plasmid pRL-CMV, according to the manufacturer's protocol. For co-transfection experiments, 0.5 g of the expression vector was transfected together with reporter vectors. Fluorescence-labelled probe sequences used for quantitative PCR ATGGCTTTCCCGAAGATGAGAT GGTTCCTCTTACCCAGCAAAAA Twenty four hours later, reporter activity was quantified with a Dual luciferase assay kit (Promega). The luciferase activity was normalized to an internal reference activity for each well, and the average for at least three experiments was shown as relative light units. Nuclear Protein Extraction and Electrophoretic Mobility Shift Assay (EMSA)-Nuclear extracts were prepared as described previously (8). Radiolabeling of probe oligonucleotides and protein-DNA binding reactions were performed using a BandShift kit (Amersham Biosciences) according to the manufacturer's directions. Briefly, protein-DNA binding reactions were performed in the reaction buffer (10% glycerol, 10 mM Tris-HCl (pH 7.5), 50 mM NaCl, 0.05% Nonidet P-40, and 0.5 mM dithiothreitol), including 1 g of poly(dI-dC)-poly(dI-dC) and 50,000 cpm of probe. Reactions were initiated by the addition of nuclear extract and run for 30 min at room temperature to form protein-nucleotide complexes that were separated on a 5% nondenaturing polyacrylamide gel in 1ϫ TAE running buffer (40 mmol/liter Tris acetate, 1 mmol/liter EDTA (pH 8.3)). Gels were dried before the visualization of radiolabeled complexes with a BAS2000 image analyzer. For competition assays, unlabeled competitor was added to the reaction mixtures at a 100-fold molar excess of the concentration of labeled probe before the addition of the nuclear extract. Supershift assays were performed by adding 1 l of antibody (anti-HNF1␣; sc-6547, anti-HNF1␤; sc-7411, purchased from Santa Cruz Biotechnology) after an initial incubation period of 30 min. The incubation was then continued for an additional 30 min.
Preparation of Anti ␤3Gal-T5 Monoclonal Antibody-A partial cDNA encoding a truncated ␤3Gal-T5, which lacks a cytoplasmic tail and a transmembrane domain, was amplified by RT-PCR and subcloned in a pBluescript vector. The insert was subsequently subcloned in pET14b bacterial expression vector (Promega). The recombinant protein expressed in Escherichia coli accumulated in the insoluble fraction. The inclusion body was solubilized and dialyzed; rats were immunized with dialysate, and the production and screening of hybridomas were performed as described previously. Finally, a monoclonal antibody reacting to the recombinant ␤3Gal-T5 protein was established. Immunological specificity was confirmed by immunostaining and Western blotting using Namalwa cells which express ␤3Gal-T1 to -T5 (2).
Immunohistochemical Staining-Small intestine, stomach, and nor-FIG. 1. Analysis of the promoter region of the human ␤3Gal-T5 gene using deletion constructs. Reporter constructs containing sequentially deleted 5Ј-flanking fragments were prepared and transfected into SW1116 and HCT-15 cells as described under "Experimental Procedures." Results are presented as the mean relative luciferase activity of three different experiments carried out in triplicate. The mean value for the cells, which were transfected with promoterless pGL3 vector, is presented as 1. Values for SW1116 and HCT-15 were indicated by solid and hatched bars, respectively. Two Cdx consensus sequences (shown in Fig. 2), named Cdx(A) and Cdx(B), are represented by dark and light gray boxes, respectively. Site-directed mutations that destroy these two Cdx consensus sequences were separately introduced (shown with X) and designated Ϫ151m1 and Ϫ151m2 constructs. mal and cancerous colon tissues were fixed in 10% formaldehyde and embedded in paraffin. Deparaffinized 4-m sections of the tissues were washed in PBS three times. For antigen retrieval, they were heated in 10 mM citrate buffer (pH 7.0) by autoclave at 121°C for 10 min. They were treated with 0.3% (v/v) H 2 O 2 in methanol for 30 min to block endogenous peroxidase and then washed in PBS three times. To block nonspecific staining, they were incubated for 20 min with 1.5% normal rabbit serum in PBS at room temperature. Antigen detection was carried out by applying anti-␤3Gal-T5 monoclonal antibody (rat IgG) at 2 g/ml for 16 h at 4°C, followed by the streptavidin-biotin complex method (Elite ABC Kit, Vector Laboratories). The specimens were washed with PBS between each step, and 0.1 mg/ml 3,3Ј-diaminobenzidene (Wako Pure Chemical Industries, Ltd., Osaka, Japan) in 50 mM Tris-HCl (pH 7.6) was used for the peroxidase reaction. Nuclei were counterstained with Mayer's hematoxylin.

RESULTS
Identification of the cis-Element Required for Activation of ␤3Gal-T5-To determine the region essential for the transcriptional regulation of ␤3Gal-T5, we isolated a genomic clone encompassing the ␤3Gal-T5 promoter region (about 2.9 kbp in length) for a reporter assay. We constructed several deletion plasmids containing various lengths of the 5Ј-flanking region of the ␤3Gal-T5 gene and examined the reporter activity with the two types of cells expressing or non-expressing ␤3Gal-T5, SW1116, or HCT-15, both of which were derived from colon cancer. As shown in Fig. 1, reporter plasmids containing Ϫ2951 to Ϫ151 bp of the 5Ј-flanking region induced a 35-90-fold increase above the promoterless reporter plasmid in relative luciferase activity, whereas plasmids containing Ϫ121 to Ϫ31 achieved only a 3-8-fold increase when assayed with SW1116 cells (␤3Gal-T5 expressing cells). This result revealed that the 30-bp region, from Ϫ151 to Ϫ121, is important for the promoter activity; therefore we designated this region the "␤3Gal-T5 control element" (GCE). By analysis using TRANSFAC, a data base of transcription factor binding sites (www.transfac. gbf.de/), we found that the promoter region, from Ϫ151 to Ϫ1, contains several consensus binding sites for known transcription factors, such as Cdx, HNF1, C/EBP, AP-1, and GATA etc. (Fig. 2). Two Cdx consensus sequences (ATAAA and ATTAT in Fig. 2), named as Cdx(A) and Cdx(B), were found in this region. The Cdx(A) site, overlapped by a putative HNF1-binding site (ATTAATAAATACC in Fig. 2), was located within the GCE, whereas the Cdx(B) site was downstream of GCE. A site-directed mutation was separately introduced in each Cdx-binding site to destroy it, and the mutant constructs were designated "Ϫ151m1" and "Ϫ151m2," respectively. When a 4-bp mutation was introduced to the Cdx(A) site (Ϫ151m1), the promoter activity decreased in 20% of wild-type construct ("Ϫ151"), whereas the mutation in the Cdx(B) site (Ϫ151m2) essentially did not affect the activity (Fig. 1). When assayed with HCT-15 cells that did not express ␤3Gal-T5, deletion ("Ϫ121") or mutation (Ϫ151m1) of GCE did not exert for the promoter activity. This significant difference between two types of cells indicated that GCE is the most important ciselement for promoter activity of ␤3Gal-T5.
Identification of Transcription Factors Essential for Regulation of ␤3Gal-T5-For further characterization of transcription factors that bind to GCE, EMSA was performed. When the radiolabeled double-stranded oligonucleotide encoding the GCE sequence was reacted with the nuclear extract of SW1116 cells, two bands were observed (Fig. 3). The specific binding to GCE was confirmed by the disappearance of the bands when an excess of non-radiolabeled GCE probe was added as a compet- itor. The lower band disappeared when an excess of SIF1, a Cdx-binding oligonucleotide from the sucrase-isomaltase promoter (9), was added as an unlabeled competitor, although the upper band did not disappear. Addition of anti-HNF1␣ antibody to the EMSA reaction mixture resulted in a supershift of the upper band but not the lower band. On the other hand, anti-HNF1␤ antibody did not shift any bands. Thus, the upper band in this EMSA experiment using the SW1116 nuclear extracts was identified as a GCE-HNF1␣ complex, and the lower band was indicated to be a GCE-Cdx1 and/or Cdx2 complex.
Four homeoproteins, Cdx1, Cdx2, HNF1␣, and HNF1␤, were cloned in an expression vector in this study, and each of them was transiently expressed in COS-1 cells. The nuclear extracts of COS-1 cells expressing Cdx1 or Cdx2 formed a complex with the GCE probe, which appeared as a band almost at the same migration length as the lower band in the previous SW1116 experiment (Fig. 4). These results further indicated that the lower band in the SW1116 EMSA experiment is a GCE-Cdx1 and/or Cdx2 complex. Specific protein-DNA complexes were also observed when reactions were performed with nuclear extract of COS-1 cells expressing HNF1␣ or HNF1␤ protein, and the complex of GCE-HNF1␣ appeared at the same position as the upper band in the SW1116 experiment (Fig. 4). The complex of GCE-HNF1␤ appeared at a position slightly lower than that of GCE-HNF1␣. As shown with asterisks in Fig. 4, nonspecific bands were observed, which also appeared when the reaction was carried out with nuclear extract of COS-1 cells transfected with a null expression vector as a negative control. Thus, four homeoproteins, Cdx1, Cdx2, HNF1␣, and HNF1␤, were demonstrated to bind to GCE.
␤3Gal-T5 Was Up-regulated during Enterocytic Differentiation of Caco-2 Cells Correlating to Up-regulation of HNF1␣ and Cdx2-Changes of transcript levels for ␤3Gal-T5 and homeoproteins were examined during differentiation of Caco-2 cells, which have been used as an in vitro model for intestinal differentiation. As shown in Fig. 6, when the enterocytic differentiation of Caco-2 cells was induced by sodium butyrate, the transcript levels for ␤3Gal-T5, Cdx2 and HNF1␣ were elevated in 2 or 4 weeks after confluence, although the Cdx1 transcript was not detected and the HNF1␤ transcript level did not change. Whereas the transcript levels for Cdx2 and HNF1␣ peaked 2 weeks after confluence, the ␤3Gal-T5 transcripts gradually increased in 4 weeks after confluence. These results indicated that the ␤3Gal-T5 gene is transcriptionally regulated by Cdx2 and/or HNF1␣ during the differentiation of Caco-2 cells.
Axial Gradient Expression of ␤3Gal-T5 and HNF1␤ in Colon and Significant Decrease of ␤3Gal-T5 and Cdx1 in Colon Cancer-To examine the alteration of ␤3Gal-T5 expression in cancer in vivo, we quantified the amounts of transcripts for ␤3Gal-T5 and four homeoproteins in 29 pairs of tissue samples from colon cancer and the surrounding normal mucosa using real time RT-PCR. The samples were divided into two groups based on the location of the lesion, i.e. "left colon," descending to the sigmoid colon and rectum, and "right colon," including the cecum, ascending, and transverse colon. No significant difference was observed between the two groups in sex, age, clinical stage, nodal metastasis, depth of invasion, and pathological features such as histological type or vascular/lymphatic invasion (data not shown). As shown in Fig. 7, the ␤3Gal-T5 transcript level of both groups decreased in cancer compared with normal mucosa with significance (p Ͻ 0.05), and interestingly, the ␤3Gal-T5 transcript level was significantly higher in the left colon than the right colon. Regarding the homeoproteins, the transcript levels for Cdx1 and HNF1␤ were also significantly lower in the left colon than the right colon. The amounts of Cdx1 and HNF1␤ transcript decreased significantly in cancer (p Ͻ 0.05). In contrast, those of Cdx2 and HNF1␣ presented no significant difference between the left colon and right colon. In some cases, the serum CA19-9 level was determined; however, no significant relation was found between the CA19-9 level and ␤3Gal-T5 transcript (data not shown).
Detection of ␤3Gal-T5 in Normal, Metaplastic, and Cancerous Tissue by Immunohistochemical Analysis-We newly raised an anti-␤3Gal-T5 monoclonal antibody and examined normal and cancerous tissues by immunohistochemistry. The specificity of the antibody was confirmed by Western blot analysis using the lysate of Namalwa cells transfected with each of the ␤3Gal-T1 to -T5 genes (Fig. 8). This antibody specifically reacted to the lysate of ␤3Gal-T5 Namalwa cells alone and not to the other cell lysates. It detected the specific band of ␤3Gal-T5, ϳ42 kDa in size (Fig. 8). As shown in Fig. 9, perinuclear staining was observed in columnar cells of both small intestine (in Fig. 9, A and E) and colon (B and F). In colon, strong signals were observed especially at the top of villi rather than crypt, whereas the signal strength was almost equal through the top to the bottom of villi in small intestine. In colon cancer tissue, the signals were apparently weakened as compared with those of normal tissue (Fig. 9, C and G). However, in several cancer samples, which were immunohistochemically confirmed to express CA19-9 antigen, the signals for ␤3Gal-T5 were also weakened as compared with those of the corresponding normal tissue (data not shown). Whereas ␤3Gal-T5 was barely expressed in normal gastric foveolar epithelium and proper glands, strong signals were detected in absorptive cells of intestinal metaplasia in the stomach (Fig. 9, D and H). DISCUSSION More than 120 genes for glycosyltransferases have been cloned to date, whereas only a few genes have been analyzed for their transcriptional mechanism. In this study, we analyzed the molecular mechanism for transcriptional regulation of the ␤3Gal-T5 gene. The members of two independent homeoprotein families, the Cdx and HNF1 families, were demonstrated to form a specific complex with the ␤3Gal-T5 promoter region and activate its gene transcription. This is the first report to demonstrate that homeoproteins are involved in the regulation of glycosyltransferases at the molecular level. In fact, our results indicated that four factors were differentially expressed depending on the system. Cdx1, Cdx2, and HNF1␣ were expressed in SW1116 cells and probably functioned as transcriptional factors. Cdx2 and HNF1␣ expression were correlated with ␤3Gal-T5 expression in the Caco-2 system. Finally, Cdx1 and HNF1␤ expression were correlated with ␤3Gal-T5 expression in actual colon cancer tissue. To explain this discrepancy, we speculate as follows. SW1116 and Caco-2 cells are cell lines  7. Regulation of ␤3Gal-T5, Cdx, and HNF1 transcripts in human normal and cancerous colon tissue. Quantitative real time RT-PCR was performed using 29-pair tissues surgically resected from colon cancer patients. The difference between normal and tumor, or left colon and right colon, was compared with the t test, and a statistically significant difference (p Ͻ 0.05) is indicated with asterisks. that expand clonally in vitro. On the other hand, the transcriptional level in colon cancer tissue reflects the mixture of such clonal cells. As demonstrated in the transfection experiment in Fig. 5, each of the four factors is active for the ␤3Gal-T5 gene expression. The expression of each factor may be regulated clonally dependent on the cell type or differentiation. Very recently, frequent mutations in the HNF1 gene were found to occur in colon cancers (10) and hepatic adenomas (11). Another speculation is that, in SW1116 and Caco-2 cells, some genes involved in the regulation of homeobox genes and/or homeobox gene themselves may be lost or mutated.
The luciferase reporter assay using the ␤3Gal-T5 promoter revealed that a 30-bp region, which is located about 150 bp upstream of the transcription initiation site, is an essential cis-element for transcriptional regulation of this gene, and it was named GCE. The EMSA experiment in Fig. 4 demonstrated that four homeoproteins, Cdx1, Cdx2, HNF1␣, and HNF1␤, are able to form a specific complex with the GCE oligonucleotide, and the co-transfection assay in Fig. 5 demonstrated that they are also able to transactivate the ␤3Gal-T5 promoter.
We determined the transcript levels for the four homeoproteins in SW1116 cells by the real time RT-PCR method. The transcripts for Cdx1, Cdx2, and HNF1␣ were found to be abundantly expressed in SW1116 cells, but the transcript for HNF1␤ could not be detected (data not shown). This is consistent with the EMSA experiment (Fig. 4), in which we detected the GCE-HNF1␣ complex as an upper band and probably the GCE-Cdx1 and/or 2 complex as a lower band. The lower band was strongly indicated to be the complex of GCE-Cdx, because 1) it appeared at the same position as that of the complex of GCE-Cdx protein expressed in COS-1 cells, and 2) it was competed by an excess of unlabeled SIF1 that specifically binds to the Cdx protein. These results suggested the direct involvement of Cdx and HNF1 homeobox family proteins in the transcriptional regulation of ␤3Gal-T5.
Cdx homeoproteins, human orthologs of Drosophila caudal, are transcription factors that are expressed exclusively in the small intestine and colon (12). Many intestine-specific genes, such as those for sucrase-isomaltase (SI) (13,14), lactase-phlorizin hydrolase (LPH) (13,(15)(16)(17), intestine phospholipase A/lysophospholipase (18), and claudin-2 (19) have been demonstrated to be transcriptionally regulated by Cdx proteins. It has also been reported that the Cdx family is involved in intestinal carcinogenesis, differentiation, and morphogenesis using several experimental approaches (20 -23). These results suggest, in general, that Cdx family proteins induce intestinal differentiation, and their down-regulation gives rise to cancer. Downregulation of Cdx1 expression in colon cancer tissue has been demonstrated with Northern blotting (12) and immunostaining (24), being compatible with our results using the quantitative RT-PCR method in this study. Mallo et al. (12) also documented a significant decrease of Cdx2 transcripts in human colon cancer, whereas we did not observe a decrease of the Cdx2 transcripts in cancer tissue (Fig. 7). This discrepancy might be due to a difference of methodology.
The HNF1 family was first cloned as liver-specific transcription factors but recently shown to be involved in the expression of several genes in the intestine, such as the SI (14,17,25,26), LPH (16,17,27), claudin-2 (19), and dipeptidyl peptidase IV genes (28). The involvement of HNF1 in carcinogenesis is not clearly understood; however, recent genetic analysis revealed that frequent mutations in the HNF1 gene occur in colon cancers (10) and hepatic adenomas (11). In the present study, we found a relative decrease of HNF1␤ transcription in colon cancer, which implies that inactivation or down-regulation of HNF1 might affect carcinogenesis.
Interestingly, the transcription for the ␤3Gal-T5 gene is under the cooperative control of Cdx and HNF1 as the transcription for SI (14), LPH (16), and claudin-2 (19), which are well known proteins specifically expressed in intestine, is controlled in the same manner. The intestine phospholipase A/lysophospholipase promoter also has binding sites for Cdx and HNF1 (18), although it has yet to be reported whether the HNF1 site is functional or not. The promoter regions of these genes share certain features, namely HNF1-, Cdx-, and GATAbinding sites within 200 bp upstream of their transcription initiation site (Fig. 10). Two closely located Cdx-binding sites were contained in the claudin-2 and SI promoters. The ␤3Gal-T5 promoter also has two putative Cdx-binding sites, but only the first site Cdx(A), overlapped by the HNF1-binding site, was demonstrated to be functional in this study. It was demonstrated that both HNF1 and Cdx proteins can bind to this site, although the question of whether Cdx and HNF1 act synergistically or exclusively remains to be answered.
The activity of several glycosyltransferases changed during the enterocytic differentiation of Caco-2 cells (5,29), resulting in a wide variety of alterations in carbohydrate structure such as a decrease in polylactosaminoglycans (6,30) and increase at the type 1 antigen (6). Once the type 1 structure is formed on the nonreducing end of the polylactosamine chain through the activity of ␤3Gal-T5, it probably works as a terminating signal for elongation of the chain, resulting in the shortening of the chain. Amano et al. (5) reported that the increase of type 1 antigen is well correlated with up-regulation of the ␤3Gal-T activity. In this study, we demonstrated a gradual elevation of ␤3Gal-T5 transcript levels upon the differentiation of Caco-2 cells, indicating that ␤3Gal-T5 directs the major part of ␤3Gal-T activity in Caco-2 cells. This study also demonstrated that transcription for Cdx2 and HNF1␣ was augmented, whereas the transcript for Cdx1 or HNF1␤ was absent or did not change, respectively, during this process. These results matched well the previous reports that documented up-regulation of Cdx2 (31) and HNF1␣ (32) expression during the differentiation of Caco-2 cells. These results strongly suggested that the up-regulation of ␤3Gal-T activity during the differentiation of Caco-2 cells is controlled by Cdx2 and/or HNF1␤.
Immunohistochemical study using anti-␤3Gal-T5 monoclonal antibody showed strong signals in epithelial cells of the small intestine and colon, whereas signals in colon adenocarcinoma cells were weakened as compared with those of normal mucosa. This finding is consistent with the transcript levels in this study and the ␤3Gal-T activity level in the previous study (7). In normal colon, the ␤3Gal-T5 protein was detected in both goblet cells and absorptive cells, especially in highly differentiated epithelial cells in the top of villi. This observation is compatible with the use of ␤3Gal-T5 as a marker for intestinal differentiation. In stomach, ␤3Gal-T5 is markedly expressed in intestinal metaplasia, comparable with that in small intestine but barely expressed in normal gastric mucosa. We have reported the strong expression of Le a antigen in intestinal metaplasia, correlating with a marked augmentation of the expression of ␣1,3-fucosyltransferase III (FUT3), which is involved in the synthesis of Le a (33). The results of the present study suggested that ␤3Gal-T5 was also involved in this phenomenon. It is noteworthy that both Cdx1 and Cdx2 are barely detectable in normal gastric mucosa but strongly expressed in intestinal metaplasia (24,34,35). Recently, Cdx2 transgenic mice were shown to develop intestinal metaplasia in the stomach, which indicated Cdx2 was essential for metaplastic formation (36). Taken together, we assume that intestinal metaplasia arising in the gastric mucosa is induced by the upregulation of Cdx1 and Cdx2 expression, resulting in transcriptional activation of ␤3Gal-T5, followed by the augmented expression of Le a antigen.
It is of interest what kind of modification in the carbohydrate structure and population is induced by the alteration of ␤3Gal-T5 during several biological processes such as differentiation, metaplastic formation, and carcinogenesis. Salvani et al. (3) reported that transfection of ␤3Gal-T5 in CHO-FT3 cells gave rise to the expression of Lewis type 1 antigens, reduction of Lewis type 2 antigens, and shortening of the polylactosamine chain. In our previous study, we introduced ␤3Gal-T5 into HCT-15 colon cancer cells, and we established HCT-3GT5H cells that strongly express Lewis type 1 antigens, such as Le a , Le b , and sLe a . We also observed the complete disappearance of Lewis type 2 antigens such as Le x , Le y , and sLe x on these cells by flow cytometric analysis (data not shown). Thus, it is speculated that down-regulation of ␤3Gal-T5 expression in cancer cells might cause the alteration of carbohydrate structures, for example, an extension of the polylactosamine chain or conversion from type 1 to type 2. In fact, the N-acetyllactosamine structure was significantly increased in colon cancer (37,38). Although Ichikawa et al. (38) concluded that an increase of ␤4Gal-TI in cancer cells was responsible for such alterations, the ␤4Gal-T activity in cancerous tissue was reported to be essentially unchanged (7).
In our previous study, we reported that ␤3Gal-T5 is essential for the expression of sLe a antigen in cancer cells, which is known as CA19-9 antigen in the clinical field (2). The amount of ␤3Gal-T5 transcript was well correlated with the amount of CA19-9 antigen expressed in many cancer cells derived from the digestive tracts (2). However, we found in the present study that ␤3Gal-T5 does not determine the amount of CA19-9 antigen in the tissues of cancer patients. We explain this unex-pected finding as follows. ␤3Gal-T5 is expressed at a saturated level in normal tissue, and enough for formation of the CA19-9 epitope exists in cancer tissue even after the down-regulation. Some other glycosyltransferase(s) or unknown factors, which contribute to the formation of the CA19-9 epitope, may be the key in determining the amount of CA19-9 antigen in cancer tissue.
It is quite interesting that homeoproteins essential for the cell differentiation in the intestine transcriptionally regulate the ␤3Gal-T5 gene, leading to the specific formation of the type 1 carbohydrate structure. ␤3Gal-T5 may play a crucial role in the maintenance of the intestinal phenotype and function through the modification of carbohydrate antigens.