Glycosyl phosphatidylinositol membrane anchoring of melanotransferrin (p97): apical compartmentalization in intestinal epithelial cells

p97, also known as melanotransferrin, is a membranebound sialo-glycoprotein with a striking homology to human transferrin and lactoferrin. p97 is also functionally related to these two proteins, since it binds iron (Brown et al., 1982), a reflection of the conserved disulfide bridges and residues involved in the iron-binding pocket of transferrin (Rose et al., 1986). Furthermore, the gene for p97 is localized on the same chromosomal region as the genes for transferrin and transferrin receptor (Seligman et al., 1986). These observations and studies on iron uptake in melanoma cells (Richardson and Baker, 1990, 1991) suggest that p97 plays a role in iron metabolism.

Although first identified as a melanoma-associated surface glycoprotein (Dippold et al., 1980; Woodbury et al., 1980), p97 was subsequently found on a wide range of cultured cell types, including renal and intestinal epithelial cell lines (Dippold et al., 1980; Woodbury et al., 1980; Real et al., 1988). Melanomas express very high levels of p97, whereas normal melanocytes have no detectable p97 (Dippold et al., 1980; Brown et al., 1981; Real et al., 1988), an observation that prompted its initial description as a transformation-related molecule. A further indication that p97 may be associated with cellular transformation came from the independent analysis of the immune response of a melanoma patient (FD) to autologous tumor cells. Serum antibodies from patient FD recognize an epitope which is exclusively expressed in the autologous tumor, but not in autologous normal cells or in allogeneic tumors (Real et al., 1984). This epitope is carried on the p97 molecule, as demonstrated using mouse monoclonal antibodies raised against the molecule bearing the FD epitope (Furukawa et al., 1989).

In contrast to its wide expression in proliferating cells in culture, p97 shows a highly restricted pattern of expression in normal human tissues. In normal tissues, p97 is expressed in human fetal intestine, umbilical cord, sweat gland ducts and liver sinusoidal lining cells (Brown et al., 1981; Real et al., 1988; Sciot et al., 1989). In the present study, we have examined the expression of p97 in intestinal epithelial cells. We have used colorectal cancer-derived cultured cell lines which have the ability to become polarized in the postconfluent state, and fresh intestine. In all cases, p97 was localized preferentially in the apical domain of polarized cells. Furthermore, we present evidence for membrane anchoring of p97 through a glycosyl phosphatidylinositol (GPI) moiety (Low and Saltiel, 1988), an anchoring mechanism for targeting proteins to the apical surface of epithelial cells (Lisanti et al., 1988, 1990a).

Caco-2, SK-CO-15 and SK-MEL-28 cells were obtained from the tumor cell bank of the Human Tumor Immunology Laboratory (Sloan-Kettering Institute, New York). HT-29 and HT-29 cells, selected by adaptation to culture in 10⁻⁶ M methotrexate (HT-29-MTX), were generously provided by A. Zweibaum (Institut Nationale de la Santé et la Recherche Médicale U178, Villejuif, France). PI-PLC purified from Bacillus thuringiensis was a gift from M. Low (Columbia University, New York). Hybridomas secreting monoclonal antibodies (mAbs) KF23, KF26 and KF104, were obtained as described previously (Real et al., 1988). mAb W6/32 detects a monomorphic epitope of MHC class I molecules and was obtained from the ATCC; mAb GRB1 detects a monomorphic epitope of MHC class II molecules and was kindly provided by F. Garrido (Granada, Spain); mAb SV63 detects placental and intestinal alkaline phosphatase and was kindly provided by W. Rettig (Sloan-Kettering Institute, New York). Reagents and isotopes were purchased as follows: ¹²⁵I-Na and [¹⁴C]ethanolamine, from ICN Biochemicals (Costa Mesa, CA); ¹²⁵I-streptavidin and [³²P]dCTP from Amersham (Buckinghamshire, England); Expre³⁵S³⁵S Protein Labeling Mix, from New England Nuclear (Boston, MA); sulfo-NHS-biotin from Pierce (Rockford, IL). Other reagents were commercial products of the highest grade available.

Cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (GIBCO, Grand Island, New York), supplemented with 10% heat-inactivated fetal bovine serum (FBS), penicillin (50 units/ml), streptomycin (50 μg/ml), glutamine (2 mM) and 1% nonessential amino acids. Caco-2 cells were seeded at 10⁴ cells/cm² and passaged every 5-6 d before reaching confluence. Cultures were regularly tested for mycoplasma contamination by a DNA hybridization method (Genprobe, San Diego, CA), and only mycoplasma-free cultures were used. Cells were used 10 days after reaching confluence.

Confluent cell monolayers were washed twice with methionine- and cysteine-free Minimal Essential Medium (MEM Select-Amine kit, Gibco), incubated in this medium for 30 min and labeled for 12 h with 100 μCi/ml of Expre³⁵S³⁵S Protein Labeling Mix in methionine- and cysteine-free MEM supplemented with 5% dia-lyzed FBS. Cells were washed 3 times with phosphate-buffered saline (PBS), lysed with low salt lysis buffer (10 mM Tris HCl, pH 7.4, 150 mM NaCl, 1 mM EDTA, 1% NP40, 2 mM phenyl-methylsulfonyl fluoride, 10 μg/ml aprotinin), and incubated for 45 min at 4°C with occasional vortexing. Lysates were cleared by centrifugation at 12,000 g for 15 min at 4°C, and supernatants were fractionated on concanavalin A-Sepharose (Pharmacia, Uppsala, Sweden), as described (Real et al., 1984). Fractions eluted with 0.2 M α-methyl-D-mannoside were pooled, pre-cleared with Pansorbin (1 mg/ml) (Calbiochem, La Jolla, CA) for 30 min at 4°C, and incubated with Protein G-agarose beads (Genex, Gaithes-burg, MD) pre-coated with the corresponding mouse mAb (5 μg purified antibody per sample) for 12 h at 4°C. After 5 washes with low salt lysis buffer and 3 washes with high salt lysis buffer (10 mM Tris-HCl, pH 7.2, 0.5 M NaCl, 1 mM EDTA, 0.5% NP40, 0.1 % SDS), beads were boiled for 3 min in sample buffer, and the released complexes were analyzed by SDS-PAGE. After electrophoresis, gels were soaked in 0.5 M sodium salicylate for 30 min (Chamberlain, 1979), dried and subjected to fluorography at −80°C.

For ethanolamine labeling experiments, SK-MEL-28 cells were incubated for 12 h with 100 μCi/ml [¹⁴C]ethanolamine (ICN Biochemicals) in D-MEM supplemented with 5% FBS. Preparation of lysates and immunoprecipitations were performed as above, except for the concanavalin A-affinity chromatography step.

Total cellular RNA was isolated as described (Chomczynski and Sacci, 1987). Samples containing 15 μg of total RNA were electrophoresed on 1% agarose-6% formaldehyde gels and transferred to nylon filters (Hybond-N, Amersham). The integrity and relative amounts of RNA were assessed by visualization of ribosomal RNA. DNA probes were labeled by the random prime method. The probe used to detect melanotransferrin transcripts was a 1.4 kb cDNA fragment from plasmid 1j1 (Rose et al., 1986). Filters were reprobed with a probe for glyceraldehyde-3-phosphate dehydrogenase to normalize for the amount of RNA loaded. Hybridizations were performed in hybridization solution (5×SSPE, 5×Denhardt’s solution, 0.5% SDS, 50% formamide, 100 μg/ml calf thymus DNA) at 42°C. Filters were washed and exposed to Hyperfilm-MP (Amersham) at −80°C.

Selective biotinylation of intestinal epithelial cell monolayers was performed as described by Lisanti et al. (1990). Cell monolayers grown on 24 mm TransWell filters (Costar) were rinsed 4 times with Dulbecco’s PBS (D-PBS) and incubated twice apically or basolaterally with sulfo-NHS-biotin (Pierce) (0.5 mg/ml in PBS) for 20 min at 4°C. After labeling, cells were rinsed with D-PBS and preincubated with D-MEM for 15 min. p97 was immunoprecipitated as described above, without the concanavalin A-affinity chromatography step. After SDS-PAGE, proteins were transferred to nitrocellulose filters, preincubated with skim milk (1% in PBS), and incubated with ¹²⁵I-streptavidin for 30 min at 4°C. Cell surface iodination of SK-MEL-28 cells was performed by the lactoperoxidase method (Marchalonis, 1969).

For treatment with PI-PLC, labeled cell monolayers (either biotinylated or iodinated) were washed three times with PBS, and incubated with PI-PLC (5 units/ml in D-MEM) for 45 min at room temperature. In control samples, PI-PLC was omitted. The resulting supernatant was collected and subjected to immunoprecipitation. Cell monolayers were washed three times with PBS, lysed and subjected to immunoprecipitation as above.

Ten days after confluence, monolayers were gently scraped, rolled, frozen directly in liquid N₂, and immediately cryopreserved in OCT embedding medium (Miles, Elkhart, IN). Fresh tissue was frozen in isopentane precooled at −115°C. 5 μm thin sections of cell pellets or fresh tissues were fixed with cold acetone for 10 min. After washing, sections were preincubated with 5% normal horse serum and incubated with mouse mAb for 1 h at 22°C. Sections were rinsed three times with PBS, incubated with biotinylated horse anti-mouse Ig (Vector) for 30 min, washed, and incubated with streptavidin-peroxidase (5 μg/ml) for 15 min. Reactions were developed with DAB/PBS/H₂O₂. Sections were counterstained with hematoxylin, dehydrated and mounted. For assays on fresh tissues, the same procedure was used and endogenous peroxidase activity was quenched by incubating with H₂O₂ for 15 min after the fixation step.

For PI-PLC treatment of immunohistochemical samples, 5 μm thick sections of cell pellets or fresh tissues were incubated with PI-PLC (5 units/ml in D-MEM) for 1 h at 37°C before fixation.

Cell pellets were fixed with 3% paraformaldehyde, 0.1-0.5% glutaraldehyde in 0.1 M phosphate buffer, pH 7.4, for 30 min at 22°C. Fixed cells were briefly rinsed with the same buffer and free aldehyde groups were quenched with 50 mM NH₄Cl in PBS for 1 h. Cells were embedded in Lowicryl K4M at −35°C as previously described (Carleman et al., 1982). Ultrathin sections (60-80 nm) were cut with glass knives, placed on parlodion/carbon-coated nickel grids and incubated with the corresponding monoclonal antibody for 12 h. Monodisperse colloidal gold particles with an average diameter of 14 nm were conjugated to Protein A and used to detect bound primary antibody as described previously (Frens, 1973; Roth, 1983).

In this study, cell lines derived from colorectal cancer displaying a range of differentiated features of intestinal cells were used. Caco-2 cells are polarized, form domes, and have a well developed brush border rich in hydrolases (Pinto et al., 1983); SK-CO-15 cells are polarized and form domes, but do not exhibit a brush border (Le Bivic et al., 1989); HT-29 cells grow as a multilayer of unpolarized cells and display a certain degree of heterogeneity, with fewer than 5% of cells showing a differentiated phenotype (Zweibaum et al., 1985; Lesuffleur et al., 1990); HT-29-MTX cells are a subpopulation of HT-29 cells selected for their ability to grow in 10⁻⁶ M methotrexate, and exhibit a polarized, mucin-producing phenotype (Lesuffleur et al., 1990). Since these epithelial cells only achieve the differentiated phenotype after they have reached confluence, the steady-state levels of p97 protein and RNA were analysed in 10-day post-confluent cultures. Immunoprecipitation of metabolically labeled cell extracts, northern blot analysis and immunocytochemistry were used to determine p97 expression in these cell lines. Caco-2, SK-CO-15, HT-29 and HT-29-MTX cells expressed p97 at variable levels (Fig. 1a,b). The melanoma cell line SK-MEL-28 showed high levels of p97, as reported (Dippold et al., 1980; Woodbury et al., 1980). Densitometric analysis of immunoprecipitation fluorograms revealed that steady state p97 levels were highest in SK-MEL-28 cells and lowest in Caco-2 cells. When p97 levels in the latter were given a reference value of 1, levels in HT-29-MTX, SK-CO-15 and SK-MEL-28 were 2, 3 and 8, respectively. Northern blot analysis revealed a 4 kb transcript in all these cell lines at levels that correlated with p97 protein levels (Fig. 1b).

To determine the subcellular localization of p97 in cultured intestinal epithelial cells, Caco-2 cells were selected because they are a prototype of enterocytic differentiation (Pinto et al., 1983). Confluent monolayers displaying differentiated phenotypes were used. Sections of frozen monolayers were analysed using the streptavidin-peroxidase technique. This method has been shown to be useful for the demonstration of compartmentalized expression of proteins in polarized cells (Le Bivic et al., 1989). Reactivity of mAb KF23 on Caco-2 monolayers was distributed mainly in the apical compartment, with a membrane pattern; reactivity with the basolateral compartment of the monolayer was either weak or absent (Fig. 2a). Two other independently derived mAbs, which also recognize p97 (KF26 and KF104), showed the same pattern of reactivity (data not shown).

Electron microscopic immunocytochemistry confirmed that p97 is preferentially localized in the apical domain of polarized cells. On ultrathin sections of Caco-2 monolayers, the majority of gold particles were distributed apically, in the microvilli and in a subapical vesicular compartment (Fig. 2b). There were no gold particles associated with the basolateral cytoplasmic region, basolateral membrane (Fig. 2c), nucleus or mitochondria.

To confirm the apical membrane distribution of p97, domain-selective biotinylation followed by immunoprecipitation was performed using polarized Caco-2 cells (Fig. 3). Approximately 80% of p97 was detected in the apical domain of Caco-2 cells. Similar results were observed in SK-CO-15 cells.

To examine whether p97 is also apically distributed in intestinal epithelium in vivo, immunohistochemical studies were performed on fresh specimens of normal human fetal and adult duodenum. p97 was not detected in normal adult duodenum (data not shown) but was strongly expressed in the apical membrane of epithelial cells in normal fetal duo-denum (22 weeks), with a brush border pattern. The immunohistochemical reaction in fetal duodenum showed a gradient, from undetectable levels in crypt cells to maximum intensity of signal at the top of the villus (Fig. 4a).

Many integral membrane proteins that show apical polarization in epithelial cells are anchored to the membrane via a GPI moiety covalently bound to the carboxy terminus (Lisanti et al., 1988). Upon examination of the published amino acid sequence for p97 deduced from its cDNA sequence (Rose et al., 1986), the characteristic carboxy-terminal hydrophilic cytoplasmic tail of class I transmembrane proteins (Singer, 1990) was not observed. Instead, the carboxy-terminal hydrophobic amino acid residues (from residue 711 to the C terminus) are preceded by a very short stretch of charged residues, with a hydrophilic profile (Fig. 5a). These features are reminiscent of the carboxy-terminal sequence of many membrane proteins attached through a GPI linkage (Low and Saltiel, 1988).

To determine if p97 is anchored to the membrane via GPI, SK-MEL-28 cells were used because they express higher levels of p97 than Caco-2 cells, making PI-PLC treatment and ethanolamine labeling assays more likely to yield conclusive results with regard to the mode of membrane attachment of p97. Cells were iodinated by the lactoperoxidase method and incubated with PI-PLC. Cell lysates and incubation media (supernatants) were then immunopreciptated with anti-p97 antibodies and analysed by SDS-PAGE. After PI-PLC treatment, most labeled p97 was released to the medium, with very small amounts remaining on the cells (Fig. 5b). Densitometric analysis indicated that, after PI-PLC treatment, 96% of p97 was released to the medium, whereas two transmembrane molecules (MHC class I and MHC class II) remained associated with the cell membrane. As a confirmation of its GPI membrane anchoring, we determined if p97 contained ethanolamine, a component of the GPI anchor. SK-MEL-28 cells were metabolically labeled with [¹⁴C]ethanolamine; cell lysates were then immunoprecipitated with anti-p97 mAb KF23 and radiolabeled p97 was revealed by autoradiography (Fig. 5c).

To identify whether p97 is also anchored to the cell membrane of normal intestine epithelial cells via a GPI linkage, sections of Caco-2 cells and sections of fetal duodenum were incubated with PI-PLC and processed for immunohistochemistry. Enzyme treatment abolished the reactivity of mAb KF23 with the apical membrane of epithelial cells, confirming that p97 is anchored to membranes through GPI in fresh tissue (Fig. 4b,c).

In the present work, we have analysed the expression of the iron-binding glycoprotein p97 in intestinal tissues and cultured cells. Furthermore, we have established that its mode of anchoring to the membrane is through a GPI moiety.

In tissues, two levels of regulation of p97 expression were observed: (1) p97 was detected in villus cells of the fetal small bowel, but not in crypt cells, suggesting that expression of p97 is regulated during the process of cell differentiation upon cell migration to the tip of the villus (Gordon, 1989); (2) p97 was not detected in any cells of normal adult small intestine, also suggesting that p97 expression is regulated during development.

The polarized expression of p97 in intestinal epithelial cell lines and fetal intestinal tissues is demonstrated here by several criteria. First, mAbs detecting p97 react exclusively with the apical compartment of cultured intestinal cells. Second, selective biotinylation of polarized Caco-2 cells followed by immunoprecipitation demonstrated that more than 80% of the biotinylated p97 is present in the apical compartment. Third, in fetal small intestine, mAbs detecting p97 react strongly with the brush border of villus cells.

Many membrane proteins that show apical polarization in epithelial cells are anchored to the membrane via a GPI moiety covalently bound to the carboxy terminus (Lisanti et al., 1988, 1990a). Although it has been proposed that p97 contains a transmembrane domain (Rose et al., 1986), a closer re-examination of the deduced amino acid sequence suggests that p97 may be GPI-anchored to the membrane because, (1) no charged cytoplasmic tail following the putative hydrophobic transmembrane domain can be observed and (2) the hydrophobic carboxy-terminal stretch is preceded by a very short stretch of charged residues. These are features commonly found at the carboxy terminus of many GPI-bound membrane proteins (Low and Saltiel, 1988). Indeed, the results presented here indicate that most p97 molecules are anchored to the membrane via GPI: it can be released from membranes by treatment of cells with PI-PLC and it contains ethanolamine, a characteristic component of the GPI moiety. PI-PLC releases more than 95% of the immunoprecipitable p97 present on SK-MEL-28 cells, a cell line with high levels of surface expression of this molecule (Dippold et al., 1980; Woodbury et al., 1980). These findings are extended to epithelial intestinal cells, as demonstrated by PI-PLC treatment of frozen sections of Caco-2 cells and fetal duodenum. Lisanti et al. (1990b) have described a GPI-linked, 95 kDa protein on Caco-2 and SK-CO-15 cells. Using the mAbs in our study, this protein appears to correspond to p97 (A. Le Bivic, personal communication).

The results presented here do not exclude the possibility that p97 or related proteins show other forms of association with membranes. The presence of even a very short charged tail at the carboxy end (Mitchell et al., 1991) or single amino acid substitutions (Waneck et al., 1988) of membrane proteins can convert them from GPI-anchored forms to transmembrane forms. Furthermore, multiple forms of certain membrane-bound proteins occur naturally: GPI-anchored, transmembrane or secreted (Fukuoka et al., 1991). These variations are generally presumed to be due to differential RNA processing (Hemperly et al., 1986), or alternative splicing (Caras et al., 1987; Stroynowski et al., 1987) of transcripts derived from a single gene. In immuno-precipitation experiments, we have not observed the release to the medium of forms other than PI-PLC-sensitive p97, arguing against the presence of secreted forms of the protein, at least under the conditions of these experiments. Overall, the evidence presented in this study indicates that most p97 molecules are anchored to the membrane via a PI-PLC-sensitive GPI moiety.

The apical localization of p97 in intestinal epithelial cells and its GPI-membrane anchoring are likely to be of major importance for the elucidation of its function. An iron-bind-ing membrane protein could theoretically be involved either in the uptake, secretion or transfer of iron. The slow internalization rates of GPI-anchored membrane proteins (Lisanti et al., 1990a) would not support a role in iron uptake for p97, unless it is associated with other transmembrane proteins. Precedents for such associations have been found in other GPI-linked membrane proteins, such as mouse Thy-1 (Kroczek et al., 1986) and Qa-2 (Robinson et al., 1989), rat Ly-6A (Su et al., 1991), or human CD59, Cd55, CD48, CD24 and CD14 (Stefanová et al., 1991), all of which can mediate signal transduction upon antibody-mediated cross-linking. Futhermore, other GPI-anchored proteins have been proposed to be involved in ligand internalization, such as the folate receptor (Rothberg et al., 1990) and a transferrin-binding protein of Trypanosoma brucei (Schell et al., 1991). Kinetic studies using SK-MEL-28 have also suggested a role in iron uptake for p97 (Richardson and Baker, 1990), although conflicting evidence is presented in more recent reports from the same laboratory (Richardson and Baker, 1991).

The tissue distribution of p97, which is expressed in several epithelia of secretory organs such as intestine, kidney, and sweat glands, would rather suggest a role in the secretion of iron. Indirect evidence is provided by the finding of p97 in the bile ductules of some patients with iron overload (Sciot et al., 1989). Iron can also be secreted by sweat gland epithelial cells (Falanga et al., 1984), one of the cell types in normal tissues which shows the highest levels of p97. The possibility of a role for p97 in the release of iron would imply the existence of a transfer mechanism by which absorbed iron, probably via the transferrin-transferrin receptor complex (located on the basolateral domain of polarized epithelial cells) would be released to p97 at an intracellular compartment before reaching the membrane. In our study, p97 was detected in a subapical vesicular compartment of polarized intestinal epithelial cells. These structures are reminiscent of caveolae involved in transcytosis or endocytosis of glycolipid-bound toxins or GPI-anchored membrane proteins such as the folate receptor (Rothberg et al., 1992). It remains to be determined which is the role of p97 in iron transport and whether these structures are indeed caveolae. Iron is essential for cellular proliferation; the upregulated expression of p97 in some cancers and in SV40-transformed fetal fibroblasts (F.X.R., unpublished observations) and its immunogenicity in the case of patient FD, suggest that p97 may be involved in the regulation of cellular proliferation. The studies presented here should help elucidate the physiological role of this iron-binding protein.

We thank Dr Alain Zweibaum for providing HT-29 and HT-29-MTX cells, Dr Martin Low for his generous gift of PI-PLC, Dr André Le Bivic for help with domain-selective biotinylation, and Drs Michael Davitz and Antonio García de Herreros for helpful suggestions and comments. This work was supported by a grant from the Fundación Knickerbokker of Barcelona, Spain, and grants from the Fondo de Investigaciones Sanitarias (to F.X.R.) and the Dirección General de Investigación Científica y Técnica (PM 89-0121) (to T.M.T.). R.A. and M.R.V. are recipients of a Beca para la Formación del Personal Investigador (Ministerio de Educación, Madrid). This study partially fulfils requirements towards the Ph.D. degree of R.A.