Mammalian lectins bind to a variety of glycans and mediate important biological functions. Many lectin–glycan interactions involve anionic glycans (Kornfeld 1990; Varki 1994; Springer 1995; Crocker et al. 1998; Nakayama et al. 1998; Bernfield et al. 1999; Fukuda et al. 1999). We previously identified non-selectin ligand carboxylated endothelial cell glycans that mediate leukocyte–endothelium interactions during inflammation (Norgard-Sumnicht et al. 1995; Srikrishna et al. 2001b). Structural analysis of these glycans is underway. Meanwhile, we searched for lectins that mediate the glycan-based interactions with the endothelium. Affinity columns containing immobilized carboxylated glycans bound three proteins from solubilized bovine lung in a carboxylate-dependent manner (Srikrishna et al. 2001a). We previously identified two of these as inflammation-related proteins annexin I and S100A8/A9 complex (Srikrishna et al. 2001a). Here we identify the third lectin as amphoterin, a protein linked to neuronal development and invasive cell migration (Rauvala and Pihlaskari 1987; Hori et al. 1995; Rauvala et al. 2000), differentiation of erythroleukaemia cells (Melloni et al. 1995b), endotoxaemia (Wang et al. 1999; Abraham et al. 2000), and tumor growth and metastasis (Taguchi et al. 2000). We provide evidence that these carboxylated glycans are present on RAGE, a well-documented signal-transducing receptor for amphoterin (Hori et al. 1995; Huttunen et al. 1999; Taguchi et al. 2000). Furthermore, we show that the carboxylated glycans mediate RAGE–amphoterin interactions.
In this study we show that embryonic neurite growth-promoting protein amphoterin binds to carboxylated N-glycans previously identified on mammalian endothelial cells. Since amphoterin is a ligand for the receptor for advanced glycation end products (RAGE), and the ligand-binding V-domain of the receptor contains two potential N-glycosylation sites, we hypothesized that N-glycans on RAGE may mediate its interactions with amphoterin. In support of this, anti-carboxylate antibody mAbGB3.1 immunoprecipitates bovine RAGE, and PNGase F treatment reduces its molecular mass by 4.5 kDa, suggesting that the native receptor is a glycoprotein. The binding potential of amphoterin to RAGE decreases significantly in presence of soluble carboxylated glycans or when the receptor is deglycosylated. Oligosaccharide analysis shows that RAGE contains complex type anionic N-glycans with non-sialic acid carboxylate groups, but not the HNK-1 (3-sulfoglucuronyl β1–3 galactoside) epitope. Consistent with the functional localization of RAGE and amphoterin at the leading edges of developing neurons, mAbGB3.1 stains axons and growth cones of mouse embryonic cortical neurons, and inhibits neurite outgrowth on amphoterin matrix. The carboxylated glycans themselves promote neurite outgrowth in embryonic neurons and RAGE-transfected neuroblastoma cells. This outgrowth requires full-length, signalling-competent RAGE, as cells expressing cytoplasmic domain-deleted RAGE are unresponsive. These results indicate that carboxylated N-glycans on RAGE play an important functional role in amphoterin-RAGE-mediated signalling.
CHO Lec2, Chinese hamster ovary cell glycosylation mutants defective in CMP-Sia transporter
and CONHMe Gps, bovine lung glycopeptides charge fractionated on a DEAE column and desialylated and carboxylate-enriched (COO−) or methylamidated (CONHMe)
fluorescence-activated cell sorting
high mobility group box 1
epitope, 3-sulfoglucuronyl β1–3 galactoside, originally identified in human natural killer cells
c-jun N-terminal kinase
receptor for advanced glycation end products.
Materials and methods
Chemicals and biologicals
Chemicals were from sources indicated before (Srikrishna et al. 2001a, 2001b) and as follows: expression vector pQE-32, Escherichia coli strain M15p(REP4), Ni-NTA-resin and anti-penta-His mAb, Qiagen (Hilden, Germany); vector pIZ/V5-His, Invitrogen (Groningen, Netherlands). RAGE cDNA fragment and sRAGE were generous gifts from Dr Angelika Bierhaus, University of Tubingen and Dr Ann-Marie Schmidt, Columbia University, respectively. N18 mouse neuroblastoma cells stably transfected with full-length or cytoplasmic domain-deleted RAGE were generated as described (Huttunen et al. 1999). Preparation of GAG-free, carboxylate-enriched bovine lung glycopeptides, and generation of anti-carboxylate monoclonal antibody mAbGB3.1 were as described (Srikrishna et al. 2001a, 2001b). Briefly, the glycopeptides were first desialylated by mild acid treatment (10 mm HCl, 30 min at 100°C) to remove sialic acids, split into two equal aliquots and lyophilized. These are referred to as COO– glycopeptides. The carboxylate groups on one aliquot were modified by two cycles of methylamidation as described earlier (Norgard-Sumnicht et al. 1995). This treatment neutralizes about 80% of the carboxylate groups. These are referred to as CONHMe glycopeptides. The glycopeptides were coupled to bovine serum albumin (BSA) using glutaraldehyde. Purification of baculovirus expressed rat amphoterin and generation of affinity-purified rabbit antibodies to amphoterin were also described earlier (Parkkinen et al. 1993). Anti-HNK-1 antibody and culture supernatants containing anti-HNK-1-reactive proteins from Lec2 cells transfected with glucuronyl transferase and HNK-1 sulfotransferase (Ong et al. 1998) were kindly provided by Drs Junya Mitoma and Minoru Fukuda (Burnham Institute, La Jolla, CA, USA).
Cell lines and tissues
Human tumor cell lines were generously provided by Dr William Stallcup (Burnham Institute). Fresh bovine lung was obtained from Mory's Meats (Escondido, CA, USA).
Generation of affinity columns containing immobilized COO– glycans, and purification and analysis of bound lectins were described earlier (Srikrishna et al. 2001a). Briefly, BSA neoglycoproteins generated as above were coupled to Affigel-10. Fresh bovine lung homogenates prepared as described earlier (Srikrishna et al. 2001a), were first pre-cleared by passing over a 2-mL BSA–Affigel column, and equal volumes were then loaded on BSA–COO––glycopeptide affinity column or the corresponding CONHMe–glycopeptide column run in phosphate-buffered saline (PBS; 10 mm phosphate, 150 mm NaCl) pH 7.5. Unbound proteins were washed out with 10 column volumes of starting buffer, and bound proteins were eluted with 50 mm sodium citrate in 10 mm phosphate buffer (no change in pH or net ionic strength over loading/wash buffer). The eluates were then dialysed, lyophilized, reconstituted in PBS and analysed on sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS–PAGE)/western blots. Amphoterin was detected in the affinity-purified bovine lung proteins using rabbit anti-amphoterin (Parkkinen et al. 1993).
Amphoterin binding to COO– glycopeptide-coated plates was assayed as described earlier for annexin I and S100 proteins (Srikrishna et al. 2001a), and bound protein detected using anti-amphoterin. mAbGB3.1 binding to sRAGE was done by first␣coating sRAGE on microtiter wells for 4 h at 37°C. After␣blocking with 3% BSA in PBS overnight, wells were incubated with mAbGB3.1 in the presence and absence of COO– or CONHMe-glycopeptides and developed as described (Srikrishna et al. 2001b).
Overexpression and purification of recombinant RAGE fragment
The internal BamHI sites of human RAGE cDNA were used to subclone the fragment coding for amino acids 90–347 into expression vector pQE-32. Overexpression of the recombinant RAGE with an N-terminal His-tag was performed in E. coli M15p (REP4), as recommended by the manufacturer. RAGE was purified from cell lysates on Ni-NTA-resin and purity assessed by SDS–PAGE and western blotting using mouse anti-penta-His mAb.
Generation and characterization of anti-RAGE monoclonal antibody
Recombinant RAGE fragment was used to immunize BALB/c mice and hybridomas were generated according to standard procedures. They were screened against recombinant RAGE fragment, and positive colonies were cloned. One IgG2a (mAbA11) recognizing recombinant RAGE fragment, bovine RAGE, and High Five insect cells stably transfected with human full-length RAGE cDNA (cloned into vector pIZ/V5-His) was used here as culture supernatant.
Immunoprecipitation and deglycosylation of RAGE
Bovine lung homogenates were prepared as described earlier (Srikrishna et al. 2001a). A 50% ammonium sulfate precipitation was carried out for enrichment of RAGE. For immunoprecipitations using mAbGB3.1, enriched homogenenates were pre-cleared with normal mouse IgG immobilized on Affigel 10 beads, followed by incubation with mAbGB3.1 immobilized beads. After overnight incubation at 4°C, the individual pellets were washed several times to remove unbound material. RAGE was released from the immunoprecipitate by 0.1 m triethanolamine, pH 11.5 or 0.1 m glycine, pH 2.5, and neutralized using 1 m Tris–HCl, pH 7.5. The protein was digested using PNGaseF (New England Biolabs, Beverly, MA, USA) according to manufacturer's protocol, but where necessary, effective deglycosylation of RAGE was achieved without protein denaturation.
Analysis of tumor cells
Tumor cells were grown in Dulbecco's modified Eagle medium (DMEM; high glucose, Gibco-BRL, Gaithersburg, MD, USA) containing 10% fetal bovine serum, 2 mm l-glutamine,␣100 units/mL penicillin, 100 µg/mL streptomycin and 100 µm non-essential amino acids. Cells were detached by incubation with PBS and 10 mm EDTA, washed thrice with ice-cold Hank's balanced salt solution (HBSS) containing 1% BSA (staining buffer) and incubated with mAbGB3.1. Bound antibody was detected with fluorescein isothiocyanate (FITC)-conjugated goat anti-mouse Ig secondary antibody. Flow cytometry was performed on a Becton Dickinson (Mountain View, CA, USA) FACscan equipped with CellQuest software. Proteins from whole cell lysates or membrane preparations were also analysed by western blotting using anti-RAGE or mAbGB3.1.
Metabolic labelling of RAGE and analysis of labelled oligosaccharides
SKNSH human neuroblastoma cells were grown as above. Metabolic labelling of cells using [2-3H]mannose and immunoprecipitation of labelled RAGE were done as described (Srikrishna et al. 2001a). N-linked carbohydrate chains were released from the protein using PNGaseF. Analysis of oligosaccharides by Concanavalin A lectin affinity chromatography, desialylation, methylesterification to␣neutralize carboxylates and to remove sulfate esters, and QAE-Sephadex fractionations were performed as described earlier (Etchison et al. 1995; Norgard-Sumnicht et al. 1995). Saponification (0.1 N NaOH, 2 h, 50°C) was used to regenerate the free carboxylate from the methylesters followed by neutralization with 1 m HCl.
Anti-HNK-1 reactivity of RAGE
HNK-1 epitope content of RAGE was estimated by comparing anti-HNK-1 binding to an equivalent amount of HNK-1-reactive proteins secreted by CHO Lec2 cells co-transfected with glucuronyl and HNK-1 sulfotransferase (Ong et al. 1998).
Binding assays using [125I]amphoterin
Amphoterin was radioiodinated using Na125I and Iodo-Beads Iodinating Reagent (Pierce, Rockford, IL, USA) to a specific activity of 5 × 103cpm/ng protein. Saturation binding experiments were done by a slight modification of the method of Hori et al. (1995). Briefly, purified RAGE was immobilized on microtiter wells and the wells blocked with 3% BSA. Wells were incubated at 4°C overnight with increasing concentrations of [125I]amphoterin in HBSS containing 0.1% BSA in the presence or absence of desialylated carboxylate-enriched (COO–) or CONHMe bovine lung glycopeptides, or 20 µg/mL mABGB3.1 or an unrelated anti-carbohydrate monoclonal antibody, or 100-fold excess of unlabelled amphoterin. Binding was also tested on RAGE deglycosylated by PNGaseF under non-denaturing conditions. Wells were washed, bound amphoterin was eluted with 2 m NaCl, and radioactivity was measured using a gamma counter. Non-linear regression analysis was done using the Prism program (GraphPad Software Inc., San Diego, CA, USA).
Neurite outgrowth assays and immunohistochemistry
Chamber slides were coated with recombinant amphoterin (20 µg/mL) for 16 h at 37°C. Cortical neurons were isolated from the cerebral hemispheres of day 15 mouse embryos as described (Miura et al. 2001), plated on amphoterin-coated slides, and incubated for 24 h in neurobasal medium containing B27 serum substituent (both from Gibco-BRL) in the presence and absence of modifiers. Cells were washed, fixed with 4% buffered paraformaldehyde, and stained with lipophilic neuronal tracers DiO or DiI (Molecular Probes, Eugene, OR, USA). In separate experiments, cells were also grown on BSA-conjugated COO– or CONHMe glycopeptides (20 µg/mL, generated as described in Materials and methods) coated on polylysine matrices. About 70% of the added conjugates bound to the matrices based on protein estimation before and after coating (not shown). Morphometric analysis of neurite lengths was made on captured images of fixed, DiI-stained cultures. Neurite-bearing cells were defined as cells having neurites longer than one cell body diameter. Neurite outgrowth assays using N18 mouse neuroblastoma cells stably transfected with full-length or cytoplasmic domain-deleted RAGE were performed as described (Huttunen et al. 1999, 2000). For immunochemical localization of mAbGB3.1 glycotope, embryonic cortical neurons were grown on amphoterin substrate and fixed as above. They were blocked with 3% BSA in PBS, incubated with mAbGB3.1 or non-specific control IgG, followed by FITC-conjugated secondary antibody and examined by standard immunofluorescence microscopy.
Amphoterin binds carboxylated N-glycans
We previously showed that a non-selectin ligand-based pathway functions during acute inflammation (Srikrishna et al. 2001b). The critical feature of this mechanism is a previously unknown carboxylate modification on endothelial cell glycoproteins (Norgard-Sumnicht et al. 1995; Srikrishna et al. 2001b). We searched for lectins that might mediate these endothelial–leucocyte interactions. A very small fraction of the proteins detergent-solubilized from whole bovine lung bound to affinity columns with immobilized carboxylated glycans (Srikrishna et al. 2001a). This included three proteins of masses 35, 37 and 30 kDa which did not bind to columns on which the carboxylate groups were converted to methylamides (Srikrishna et al. 2001a). The 35- and 37-kDa bands were S100A8/A9 and annexin I, respectively (Srikrishna et al. 2001a). In this study we report that the 30-kDa protein is amphoterin, based on its mass and anti-amphoterin reactivity (Fig. 1). Native amphoterin from human erythroleukaemia cell lysates, and purified recombinant amphoterin (Parkkinen et al. 1993) also bind to immobilized glycans in a carboxylate-dependent manner (data not shown).
Amphoterin binds to chondroitin-sulfate, heparin and heparan sulfate proteoglycans (Rauvala and Pihlaskari 1987; Salmivirta et al. 1992; Milev et al. 1998). However, desialylated carboxylate-enriched glycopeptides that we use in the glycan affinity column and in binding assays are moderately anionic and composed of sugars commonly found in N- and O-glycans (Srikrishna et al. 2001a). Their compositional analysis also shows no evidence for the presence of any uronic acids typical of glycosaminoglycan (GAG) chains. This rules out amphoterin binding to any potential GAG chains in our preparations.
Evidence for funtional N-glycans on the amphoterin receptor, RAGE
RAGE is a physiologically important binding protein for amphoterin (Hori et al. 1995; Huttunen et al. 1999, 2000). It is an immunoglobulin superfamily member consisting of an N-terminal V-type domain and two C-type domains. It also has a single transmembrane spanning domain that anchors the protein, and a cytosolic tail which is essential for signalling (Neeper et al. 1992). The extracellular V-type domain is believed to be the principal ligand-binding site that interacts with a diverse group of ligands, including advanced glycation end products, β-amyloid peptides and S100A12, in addition to amphoterin (Kislinger et al. 1999; Schmidt et al. 2000b). How does RAGE bind so many structurally unrelated ligands and exert its location-specific effects? What are the structural determinants involved? One possibility is that post-translational modifications, such as its N-glycans, affect RAGE binding and signalling, possibly by altering its association with various cell surface molecules. The V-type ligand domain of RAGE has two potential N-glycosylation sites (Neeper et al. 1992), but it is not known if they are utilized. Since amphoterin binds to chondroitin-sulfate, HNK-1 on sulfoglycolipids, heparin and heparan sulfate proteoglycans (Rauvala and Pihlaskari 1987; Salmivirta et al. 1992; Nair and Jungalwala 1997; Milev et al. 1998), we hypothesized that N-glycans on RAGE may mediate amphoterin–RAGE interactions.
mAbGB3.1 is an anti-carbohydrate antibody that was generated against carboxylate-enriched desialylated bovine lung N-glycans (Srikrishna et al. 2001b). Antibody reactivity is abrogated by carboxylate-methylamidation of bovine lung glycopeptides, showing that it recognizes the non-traditional carboxylated epitopes on bovine lung N-linked oligosaccharides (Srikrishna et al. 2001b). The antibody therefore provides a tool to identify carboxylated N-glycans on candidate proteins. Consistent with our predictions, we found that soluble RAGE (sRAGE), consisting of only the extracellular domain, binds to mAbGB3.1. Soluble COO− glycopeptides, but not CONHMe-species, blocked binding at␣low concentrations (Fig. 2a). Purified oligosaccharides released from the glycopeptides by hydrazinolysis inhibited equally well (not shown), showing that this effect did not depend on the peptides from the proteins.
Next, we immunoprecipitated native RAGE from bovine lung with mAbGB3.1. Bovine lung was used as the source as it has the highest level of RAGE expression, and was also the original source of the carboxylated glycans (Brett et al. 1993; Norgard-Sumnicht et al. 1995). Low or high pH buffers release bound RAGE from the antibody, as do 200 µm COO− glycans alone at neutral pH (not shown), showing that the␣antibody–RAGE interaction is carbohydrate-based. Anti-peptide RAGE and mAbGB3.1 both recognize immunoprecipitated RAGE. PNGaseF digestion eliminates mAbGB3.1 binding and decreases the protein mass by about 4500 Da, consistent with the loss of two N-linked oligosaccharide chains (Fig. 2b).
RAGE is expressed in early neuronal development, down-regulated in adulthood, and re-expressed in many types of␣tumors (Schmidt et al. 2000a). We found that carboxylated glycans are expressed on the surface of many tumor␣cells,␣especially glioblastomas and neuroblastomas (Fig. 3a).␣In fact, RAGE was by far the most prominent mAbGB3.1-reactive band in these and several other tumor cells that we examined (Fig. 3b and not shown). mAbGB3.1 immunoprecipitates 65% of RAGE from bovine lung, whereas more than 90% of RAGE from neuroblastoma cells is immunoprecipitable (not shown). The extent of precipitation may depend on the precise structure of the sugar chains, or number or density of epitopes. It may also be cell type-specific.
Analysis of RAGE oligosaccharides
Next we metabolically labelled SKNSH human neuroblastoma cells with [2-3H]mannose to tag RAGE oligosaccharides. More than 80% of the 3H label on mAbGB3.1-immunoprecipitated proteins from these cells was associated with RAGE (not shown). Concanavalin A affinity chromatography of the PNGaseF-released sugar chains showed that about 25% of the label occurred in tri- and tetra-antennary chains, 10% in bi-antennary and 15% in hybrid-type chains, with 50% in high mannose-type species (not shown). Based on the 2–3-fold higher mannose content in high mannose-type oligosaccharides, each RAGE molecule is therefore likely to have at least one complex-type N-linked glycan chain. Consistent with this, one of the oligosaccharides is sensitive to Endoglycosidase H digestion, indicating that it is a high-mannose or hybrid-type chain. We used QAE-Sephadex chromatography and selected chemical modifications to identify the anionic substituents. In replicate experiments, approximately 42% of the radiolabel was anionic without any treatment, and 15–17% remained charged after desialylation with mild acid that removes sialic acid (Fig. 3c). The remaining charges were nearly all neutralized by methanolysis, which cleaves sulfate esters and converts carboxylates to neutral methylesters. Base hydrolysis of the desialylated, methanolysed material, almost completely regenerates anionic species seen on desialylated oligosaccharides, consistent with reconversion of the methyl ester into the carboxylate. This provides further chemical evidence that RAGE oligosaccharides contain these unusual carboxylate groups.
Amphoterin is identical to a previously described sulfoglycolipid binding protein-1 (SBP-1), which recognizes 3-sulfoglucuronyl β1–3 galactoside or HNK-1 (Nair and Jungalwala 1997; Chou et al. 2001). It is therefore possible that amphoterin–RAGE binding simply involves HNK-1 glycans on RAGE. However, bovine lung RAGE did not react with HNK-1 antibody under conditions that can easily detect 0.1 pmole HNK-1/nmole of protein (not shown). Therefore, we believe that RAGE does not contain this epitope. As we reported previously, mAbGB3.1 does not recognize HNK-1 carbohydrates (Srikrishna et al. 2001b) and the carboxylate-enriched bovine lung fractions used to prepare the affinity columns do not contain HNK-1 epitope or GAG chains either (not shown).
RAGE-amphoterin binding is glycan-dependent
Purified RAGE specifically binds [125I]amphoterin (Fig. 4a) with a Kd of approximately 10.7 ± 1.9 nm and a Bmax of approximately 52.7 ± 4.3 fmole/well (binding potential Bmax/Kd of 4.9, Fig. 4) consistent with previous observations using recombinant rat amphoterin and bovine lung RAGE (Hori et al. 1995). Specific binding is defined as the total binding minus non-specific binding measured in presence of 100-fold excess of unlabelled amphoterin. The non-specific binding was 3–4% of total binding. We then examined the possibility that amphoterin binding is dependent upon the N-glycans on RAGE. Non-linear transformations showed that the binding potential of amphoterin-RAGE (Bmax/Kd) is significantly reduced in presence of soluble COO− glycopeptides (Kd of approximately 12.6 ± 2.7 nm and a Bmax of approximately 34.7 ± 3.7 fmole/well; Bmax/Kd = 2.8), while it is unaffected by neutral glycopeptides (not shown). We could effectively deglycosylate RAGE using PNGaseF without prior protein denaturation (not shown), suggesting that the glycans are exposed and readily accessible in the native protein. Deglycosylation again significantly reduces the binding potential (Kd of approximately 18.2 ± 5.3 nm and a Bmax of approximately 47.9 ± 7.9 fmole/well; Bmax/Kd = 2.6). We also tested the binding of a single concentration of labelled ligand in presence of various concentrations of COO− or neutral glycopeptides. Again, while the COO− glycopeptides progressively inhibit binding, neutral species are without effect (Fig. 4b). Binding is also reduced in presence of mAbGB3.1, while a non-relevant antibody is ineffective (Fig. 4c). The inability of COO− glycopeptides, mAbGB3.1 or of RAGE deglycosylation to completely block binding suggests that the interaction between amphoterin and RAGE occurs at more than one site. However, these results strongly imply that glycans on RAGE are important in defining conformational epitopes on amphoterin-binding V-domain of the native receptor.
The carboxylated glycans are involved in neurite outgrowth
RAGE and amphoterin co-localize at the leading edges of various motile cells, including embryonic neurons and tumor cells. Their interaction is crucial for embryonic neurite outgrowth and tumor cell invasion (Huttunen et al. 1999, 2000; Rauvala et al. 2000; Taguchi et al. 2000). To determine if the carboxylated glycans mediated this interaction in vitro, we first sought to establish whether embryonic cortical neurons expressed mAbGB3.1 glycotope. mAbGB3.1 stains axons and growth cones of embryonic cortical neurons grown on amphoterin (Figs 5b and c). Cells incubated with a control IgG are negative (Fig. 5a). We also found that soluble mAbGB3.1 inhibits neurite outgrowth (Fig. 5f), but a control antibody does not (Fig. 5e). In the absence of a modifier, the percentage of total cells bearing neurites is 38.7 ± 10.9 (mean ± SD) as determined by examination of 4–6 different fields in two replicate analyses. In presence of mAbGB3.1, the neurites are shorter, and the percentage of cells with neurites is significantly reduced (14.9 ± 6.9, p < 0.005).
Migrating cells are known to secrete amphoterin (Parkkinen et al. 1993). Cell surface localization of amphoterin has also been previously demonstrated (Rauvala et al. 1988). We reasoned that COO− glycopeptides coated on the culture dish could provide an excellent amphoterin-binding surface and thus promote extension of processes. Also, immunoglobulin superfamily members quite often interact homophilically. Though it is not known if RAGE ectodomain is able to bind to itself, it is likely that coated glycopeptides could promote neurite outgrowth by interacting with RAGE itself, or by forming a RAGE–amphoterin-immobilized glycan complex. To check this, we first tested the growth of cortical neurons␣on␣BSA conjugated glycopeptides immobilized on polylysine matrices. We found that COO− glycopeptides enhance neurite outgrowth, but CONHMe glycopeptides do not (Fig. 6). Both the number of neurite-bearing cells, and the length of the neurites decrease on CONHMe–glycopeptide substrate.
To confirm that the effects of glycopeptides and mAbGB3.1 on neurite outgrowth are RAGE-dependent, we tested them in N18 mouse neuroblastoma cells transfected with either full-length or cytoplasmic domain-deleted RAGE. This domain is crucial for RAGE signalling (Huttunen et al. 1999; Taguchi et al. 2000). Cells that express the tail-deleted form cannot extend neurites on surfaces coated with RAGE ligands such as amphoterin, whereas the cells expressing the full-length RAGE display a clear neurite morphology (Huttunen et al. 1999, 2000). Parental neuroblastoma cells show very low mAbGB3.1 reactivity (not shown), while amphoterin is expressed and secreted by N18 cells (Merenmies et al. 1991). As seen in Fig. 7, mAbGB3.1 significantly inhibits amphoterin-induced outgrowth of N18 neuroblastoma cell neurites. The neurites are shorter and have a distorted morphology. Soluble COO− glycopeptides at 200 µm do not block amphoterin-induced neurite outgrowth; however, as seen above with cortical neurons, when the glycans were used as immobilized matrices, they were fully capable of inducing neurite outgrowth. These neurite outgrowth-promoting effects of glycopeptides are RAGE-dependent, as cells expressing the cytoplasmic domain-deleted RAGE do not display neurite morphology.
Collectively our results indicate that: (i) amphoterin binds carboxylated N-glycans; (ii) RAGE is a glycoprotein containing these glycans; and (iii) the glycans play an important role in RAGE-amphoterin binding and signalling in neurite growth.
Earlier we found that bovine lung COO− glycopeptides bound to annexin I and S100A8/A9 and to an unknown 30-kDa protein (Srikrishna et al. 2001a). While this work was in progress, Hofmann et al. (1999) reported that EN-RAGE or S100A12, a close homologue of S100A9, bound to RAGE. The other well-documented ligand for RAGE is the 30-kDa heparin-binding protein amphoterin (Schmidt et al. 2000b). We therefore suspected that our unknown 30-kDa protein was amphoterin, and confirmed this by its immunoreactivity.
Amphoterin has a highly dipolar charge distribution, and is a developmentally regulated protein that is abundant in embryonic brain and in transformed cell lines (Rauvala and Pihlaskari 1987; Parkkinen et al. 1993). It has extensive sequence similarity to HMGB1-type DNA-binding proteins (Merenmies et al. 1991), but its functional role in the nucleus, if any, remains unclear. Amphoterin is also a cytosolic protein that localizes to growth cones of embryonic neuronal cells and leading edges of tumor cells when extension of cytoplasmic processes are stimulated on appropriate matrices (Merenmies et al. 1991; Parkkinen et al. 1993). RAGE is a major cellular binding site for amphoterin at the leading edges of invasive cells. Amphoterin is also a late mediator of endotoxin lethality and acute lung inflammation in mice (Wang et al. 1999; Abraham et al. 2000).
Elegant studies from the Schmidt and Stern laboratories show that RAGE engages a series of structurally unrelated ligands (for recent reviews see Schmidt and Stern 2000; Schmidt et al. 2000a, 2000b), and that RAGE–amphoterin interaction is a key checkpoint in tumor growth, invasion and metastasis (Taguchi et al. 2000). A ternary complex between amphoterin, plasminogen and plasminogen activator at the leading edge of neurites and tumor cells activates metalloproteinases (MMP-2 and MMP-9) that degrade extracellular matrix molecules (Rauvala et al. 2000; Taguchi et al. 2000). While overexpression of RAGE enhances tumor growth and metastasis, anti-RAGE, expression of dominant-negative RAGE lacking a cytoplasmic tail, or addition of soluble RAGE, prevent tumor growth and metastasis in mice (Taguchi et al. 2000). These effects are brought about by three coexisting MAPK modules: p38, JNK and p42/p44 (Taguchi et al. 2000).
Full-length RAGE has two potential N-linked glycosylation sites, and its alternatively spliced and secreted isoform from human lung and brain has an additional site (Neeper et al. 1992; Malherbe et al. 1999). The two N-linked sites on mature RAGE occur in the principal ligand-binding V-domain (Kislinger et al. 1999). There are no previous studies either showing the presence of oligosaccharides on RAGE or their structure. Our studies here show that mature RAGE is glycosylated and that the N-linked glycans have the carboxylated epitope. The abundance of mAbGB3.1 reactivity in bovine lung, endothelial cells, macrophages, tumor and embryonal cells (Srikrishna et al. 2001b and present findings) parallels the reported tissue expression of RAGE (Brett et al. 1993). The function of the alternatively spliced and secreted isoform of RAGE in human brain and lung with three potential N-glycosylation sites (Malherbe et al. 1999) is unknown. Incidentally, the additional N-glycosylation site in the alternatively spliced form is not present on the V-domain of the secreted protein. Malherbe et al. (1999) suggest that the soluble form could act as a physiological antagonist, analogous to the addition of sRAGE in experimental systems, where it competes with cell surface RAGE for amphoterin binding (Hori et al. 1995; Taguchi et al. 2000; Yan et al. 2000).
RAGE also binds to a group of seemingly unrelated ligands including advanced glycation end products, β-amyloid proteins and S100A12. The structural basis for the multiligand binding properties of RAGE is not understood, but our findings suggest that carboxylated modifications on RAGE and/or other associated glycoproteins (Schmidt et al. 1994) could be important modulators. Incidentally, S100A12, the other established ligand for RAGE, is most homologous to S100A9, which also binds the carboxylated glycans (Robinson and Hogg 2000; Srikrishna et al. 2001a).
We found that carboxylated glycans themselves promote neurite outgrowth in embryonic neurons and RAGE-transfected neuroblastoma cells. This outgrowth requires full-length, signalling-competent RAGE, as cells expressing cytoplasmic domain-deleted RAGE are unresponsive. These results suggest that the glycans could induce or stabilize a multivalent protein complex through homophilic association (Hakomori et al. 1998), or by forming a RAGE–amphoterin-immobilized glycan multivalent complex. A variable aggregation threshold could be crucial for differential signalling in␣vivo. In fact, a novel paradigm for supermolecular assembly and signal transduction based on cross-linking of multivalent carbohydrates with multivalent lectins has been recently proposed, based on studies of receptor clustering involving endogenous galectin-1 and its counter receptors on human T cells (Sacchettini et al. 2001). N-Glycans are known to modulate signalling (Ellies et al. 1998; Hennet et al. 1998; Priatel et al. 2000). A GlcNAc transferase differentially modulates Notch-1 binding to its ligands, Delta and Serrate (Moloney et al. 2000). Deficiency of β1,6 GlcNAc transferase (Mgat-5) lowers T-cell activation threshold by enhancing T-cell receptor clustering (Demetriou et al. 2001), and multivalent galectin-Mgat-5 modified glycoprotein lattices limit agonist-mediated clustering. These studies join the growing list of proven effects of glycans on apoptosis, immunomodulation and immune response.
Amphoterin localizes in the cytoplasm of resting cells and lacks a classic secretion signal, and yet it is secreted upon activation (Melloni et al. 1995a; Wang et al. 1999; Fages et al. 2000; Rouhiainen et al. 2000). Annexin-I, and the S100A8/A9 complex, the other two carboxylated glycan-binding lectins, are also cytosolic and secreted by a non-classical pathway (Perretti 1997; Rammes et al. 1997), as are the galectins (Hughes 1999). The three carboxylate-binding lectins share other properties: all three bind to signal-transducing cell surface receptors (Hori et al. 1995; Newton and Hogg 1998; Walther et al. 2000) and all have been linked to inflammation (Perretti 1997; Kerkhoff et al. 1998; Hofmann et al. 1999). Identifying annexin I, S100A8/A9 and amphoterin as a new family of lectins now offers a new dimension to their roles as effectors in signalling pathways.
We are grateful to Dr Ajit Varki, University of California, San Diego, for his insight, advice and invaluable contributions to our long-term collaborative study on bovine lung oligosaccharides, and for his critical review of this manuscript. We thank Drs Fumitoshi Irie and Ryu Miura for help with the embryonic neurite outgrowth assays, and␣Violet Abraham and Antje Oehmichen for their excellent␣technical␣assistance. This work was supported by grants from NIGMS, PO1-CA71932 (to HF), RO1-NS32717 (to YY), and the Finnish Academy, the Technical Research Center of Finland (Program of Molecular Neurobiology) and the Sigrid Juselius Foundation (to HR). LJ is supported by a research fellowship from Wenner-Gren␣Foundations, Stockholm, and the work of BW was funded by Bundesministerium fuer Bildung und Forschung (BMBF- 01ZZ9604).