An N-Linked Glycan Modulates the Interaction between the CD1d Heavy Chain and β2-Microglobulin*

Human CD1d molecules consist of a transmembrane CD1 (cluster of differentiation 1) heavy chain in association with β2-microglobulin (β2m). Assembly occurs in the endoplasmic reticulum (ER) and involves the initial glycan-dependent association of the free heavy chain with calreticulin and calnexin and the thiol oxidoreductase ERp57. Folding and disulfide bond formation within the heavy chain occurs prior to β2m binding. There are four N-linked glycans on the CD1d heavy chain, and we mutated them individually to ascertain their importance for the assembly and function of CD1d-β2m heterodimers. None of the four were indispensable for assembly or the ability to bind α-galactosyl ceramide and to present it to human NKT cells. Nor were any required for the CD1d molecule to bind and present α-galactosyl ceramide after lysosomal processing of a precursor lipid, galactosyl-(α1-2)-galactosyl ceramide. However, one glycan, glycan 2 at Asn-42, proved to be of particular importance for the stability of the CD1d-β2m heterodimer. A mutant CD1d heavy chain lacking glycan 2 assembled with β2m and transported from the ER more rapidly than wild-type CD1d and dissociated more readily from β2m upon exposure to detergents. A mutant expressing only glycan 1 dissociated completely from β2m upon exposure to the detergent Triton X-100, whereas a mutant expressing only glycan 2 at Asn-42 was more stable. In addition, glycan 2 was not processed efficiently to the complex form in mature wild-type CD1d molecules. Modeling the glycans on the published structure indicated that glycan 2 interacts significantly with both the CD1d heavy chain and β2m, which may explain these unusual properties.

The human CD1 5 (cluster of differentiation 1) family consists of five transmembrane glycoproteins encoded by linked genes (1). They are divided into two groups based on amino acid sequence homology; group 1 includes CD1a, -b, and -c, and group 2 consists of CD1d, the only isoform present in mice and rats. The fifth member of the family, CD1e, has an amino acid sequence intermediate between the two groups. CD1 heavy chains are structurally similar to MHC class I molecules and possess a short C-terminal cytosolic tail, a hydrophobic transmembrane region, and an extracellular region that interacts non-covalently with ␤ 2 -microglobulin (␤ 2 m). The role of CD1 molecules is to bind lipid antigens and present them to T cells, and the ␣1 and ␣2 domains of the extracellular region fold in a similar manner to the analogous domains in MHC class I molecules to generate the lipid binding site. CD1 heavy chain folding and association with ␤ 2 m occurs in the endoplasmic reticulum (ER). After exiting the ER, the assembled CD1 molecules pass through the secretory pathway and reach the plasma membrane. From there, with the exception of CD1a, they enter the endocytic system by adaptor protein (AP)-dependent internalization using tyrosine-based endocytic motifs (YXX⌽, X ϭ any amino acid and ⌽ ϭ bulky hydrophobic amino acid). Similar to MHC class II molecules, antigen loading occurs mainly in the endocytic system (1). The affinity of CD1 molecules for lipid antigens is explained by the architecture of their antigen binding sites, which are more hydrophobic and deeper than those of MHC class I molecules. They accommodate the alkyl chains of the lipids, leaving the more polar regions exposed for recognition by T cell receptors (2)(3)(4).
CD1d molecules are expressed by many cell types, in particular by dendritic cells, macrophages, B cells, and thymocytes (1). They present lipid antigens to a special subset of T cells that express the NK1.1 marker. The majority of these NKT cells are called invariant NKT cells because they express fixed rear-ranged T cell receptor ␣ chains (V␣14-J␣18 in mice and V␣24-J␣18 in humans) that are associated with different V␤ chains. After activation, they rapidly secrete Th1 and Th2 cytokines, suggestive of an immunoregulatory function (5). An exceptionally potent antigen presented by CD1d to invariant NKT cells is ␣-galactosyl ceramide (␣-GalCer), a lipid from a marine sponge (6,7). Recent data have shown that both endogenous lipids and exogenous lipid antigens, e.g. from Sphingomonas and Ehrlichia species, can be presented to NKT cells by CD1d molecules (8 -15). Loading of exogenous lipids occurs mainly in the endocytic system and is catalyzed by saposins or the GM2 activator protein, which are small lipid transfer proteins (16,17). Initial endogenous lipid loading occurs in the ER (8) and is facilitated by the action of microsomal lipid transfer protein (18 -20).
CD1d molecules have four occupied N-glycosylation sites (five in the case of mouse CD1d) and six Cys residues, four of which are involved in the formation of intrachain disulfide bridges. Correct folding of the heavy chain depends on glucose trimming (21) and involves the chaperones calnexin (CNX) and calreticulin (CRT) and the oxidoreductase ERp57. After release from the CNX/CRT cycle, the heavy chain binds non-covalently to ␤ 2 m before trafficking to the plasma membrane (21). In addition to gaining access to the endocytic pathway by AP2and AP3-dependent endocytosis, CD1d molecules can also be directed there by association with the invariant chain, normally responsible for the endocytic localization of MHC class II molecules, or by interaction with MHC class II invariant chain complexes (22,23). CD1d undergoes multiple rounds of recycling between the endocytic system and the plasma membrane.
The four N-linked glycans of human CD1d are localized in the ␣1 and ␣2 domains, effectively surrounding the antigen binding site, and in this study, we have focused on the role of these glycans in the assembly and stability of CD1d molecules. We show that in mature human CD1d molecules, one of the glycans (at position Asn-42) exists mainly in a form that is sensitive to the enzyme endoglycosidase H (Endo H), likely the unprocessed high mannose form, and that its deletion affects the stability of the heavy chain-␤ 2 m heterodimer. We propose a role for this glycan in mediating interactions between the heavy chain and ␤ 2 m.
NKT Cell Clone-The human V␣24/V␤11 CD4ϩ NKT clone 6F5 was isolated by staining peripheral blood mononuclear cells from a normal human subject with mAbs specific for CD4 (OKT4; hybridoma obtained from ATCC, Manassas, VA) and the invariant T cell receptor ␣-chain expressed by CD1d-re-stricted NKT cells (6B11, Pharmingen). Cells staining positive for both mAbs were sorted as individual cells directly into wells of sterile 96-well plates using a MoFlo high speed fluorescence activated cell sorter with autoclone module (Dako-Cytomation). CD4ϩ NKT cell clones, including clone 6F5, were expanded using irradiated (50 grays) allogeneic peripheral blood mononuclear cell as feeders with PHA and recombinant interleukin-2 as described previously (27).
Generation of N-Glycosylation Mutants of Human CD1d-Mutant constructs were generated using a human CD1d cDNA expression plasmid, pSR␣-neo.CD1d. CD1d cDNAs containing single N-glycosylation site mutations were generated by the PCR method using primers that mutated each of the codons encoding Asn to generate Ala substitutions. CD1d mutants containing two glycosylation site mutations were generated using single mutants as templates for second round PCR-based mutagenesis and so forth for mutants at three and eventually all four glycosylation sites.
Generation of Stable Transfectant C1R Cells-C1R cells were transfected with the vector pSR␣-neo containing the gene of interest by electroporation at 230 mV/960 microfarads and selected for neomycin resistance at 1.8 mg/ml G418 (Invitrogen). The cells were assayed for CD1d surface expression by flow cytometry. Cells were maintained in Iscove's modified Dulbecco's medium (Invitrogen) supplemented with 10% fetal bovine serum (Hyclone), penicillin/streptomycin (Invitrogen), and G418 at 37°C in 5% CO 2 atmosphere.
NKT Cell Stimulation Assay-C1R cells transfected with wild-type and mutant CD1d constructs were pulsed for 6 h with ␣-GalCer or GalGalCer (100 ng/ml in Me 2 SO) or Me 2 SO as control. After three washes with phosphate-buffered saline, the cells were fixed in 0.05% glutaraldehyde. Cells were washed twice in Iscove's modified Dulbecco's medium and once in NKT cell medium before use. Triplicate samples of 75,000 cells were co-cultured with 25,000 6F5 NKT cells in 96 U-bottom well plates for 22 h. The medium was supplemented with phorbol 12-myristate 13-acetate (1 ng/ml) (Sigma) and interleukin-2 (1 nM) (Preprotech). Secreted human IFN-␥ was detected by sandwich ELISA.
Metabolic Labeling and Immunoprecipitation-Labeling with [ 35 S]methionine/cysteine (ICN, PerkinElmer Life Sciences) and immunoprecipitations were performed as described (21). Reimmunoprecipitation was performed by heating the washed immunoprecipitates in 1% SDS, 5 mM dithiothreitol for 5 min at 100°C before diluting 10-fold in 1% Triton X-100 in 0.15 M NaCl, 0.01 M Tris, pH 7.4 (TBS), 1.5 mM iodoacetamide. The supernatants were then used for reimmunoprecipitation with D5 antibody. The samples were boiled in SDS sample buffer and separated by SDS-PAGE (12% acrylamide) before autoradiography. When Endo H digestion was performed, the samples were incubated overnight with 50 units of Endo H according to the manufacturer's protocol.
Purification of CD1d from C1R.CD1d Cells-5 ϫ 10 9 C1R.CD1d cells were lysed in 2% C 12 E 9 in TBS. The lysate was loaded onto a CD1d51 immunoaffinity column in tandem with a precolumn coupled with mouse IgG to remove nonspecific contaminating proteins (21). After washing with 0.1% C 12 E 9 / TBS, the proteins were eluted in acidic conditions (0.1% C 12 E 9 , 0.1 M NaCl, 0.05 M sodium acetate, pH 3.5). The fractions were neutralized with 1 M Tris, and the proteins precipitated with ethanol. The proteins were then solubilized in 0.2% SDS.
Western Blotting-The proteins were resolved by SDS-PAGE and electrophoretically transferred to Immobilon-P membranes (Millipore). Blots were incubated with D5 mAb (1:5000) or rat anti-human Grp94 mAb (1:10,000) followed by horseradish peroxidase-conjugated goat anti-mouse IgG (1:5000) antibodies or alkaline phosphatase-conjugated goat anti-mouse or goat anti-rat IgG (1:5000) for quantitative analysis. The proteins were detected by enhanced chemiluminescent Western blotting (SuperSignal West Pico, Pierce) in the first case or imaged using a Fluorimager with ECF substrate (Amersham Biosciences) in the second. For quantitative immunoblots, the protein content of the postnuclear supernatants was quantitated by the bicinchoninic acid method (BCA protein assay, Pierce). 3-fold serial dilutions were prepared, and the samples were resolved by reducing 12% SDS-PAGE.
Peptide Analysis-After overnight digestion of the purified CD1d-␤ 2 m complexes with 100 units of Endo H, heavy chains were separated by SDS-PAGE. The proteins were stained with Coomassie Blue, and the bands were excised from the gel. The gel samples were washed in 50% acetonitrile, 10 -50 mM NH 4 HCO 3 , dried, and incubated for 24 h at 37°C with 0.1 g of trypsin/15 mm 3 gel in 10 mM NH 4 HCO 3 . The peptides were extracted in 0.1% trifluoroacetic acid, 60% acetonitrile, dried, and resuspended in 0.05% trifluoroacetic acid, 5% acetonitrile. Fractionation of the samples was performed on a Hewlett Packard 1090 HPLC system equipped with an Isco Model 2150 peak separator and a 1 mm ϫ 25 cm Vydac C-18 (5-m particle size, 300-m pore size) reverse phase column equilibrated with 98% buffer A (0.06% trifluoroacetic acid) and 2% buffer B (0.052% trifluoroacetic acid, 80% acetonitrile). Peptides were eluted at 50 l/min (gradient program: 0 -60 min (2-37% buffer B), 60 -90 min (37-75% buffer B), and 90 -105 min (75-98% buffer B) and detected by their absorbance at 210 nm. Fractions were collected, and the peaks of interest were subjected to Edman sequencing using an ABI Precise 494 cLC instrument with an on-line HPLC for detection of the phenylthiohydantoin amino acids.
Molecular Modeling of Glycosylated CD1d-The model of glycosylated CD1d was generated starting from the crystal structure of human CD1d complexed with the ␣-GalCer, Protein Data Bank code 1zt4 (3), in which the four glycan structures are not resolved in the electron density map. The glycans attached by in silico modeling to the heavy chain core of CD1d were of the GlcNAc 2 Man 9 type in a configuration consistent with the most populated conformer derived from a data base of experimental glycan structures (28). Glycan attachment, conformational search, and clash analysis were performed with Glyco-Pack, an in-house software package designed for glycoprotein structural analysis. Refined modeling was performed using the Insight II software from Accelrys using the Discover module with the cvff force field for the simulated annealing and the energy minimization. The solvent accessibility was calculated with Naccess 2.1.1 with a 1.5 Å radius probe. 6

None of the Four N-Linked Glycans Are Essential for CD1d
Assembly and Transport-To investigate the roles of the CD1d glycans, we used mutagenesis to change the Asn residues in each of the four N-glycosylation sequons (Fig. 1A) to Ala and generated transfectants of the human C1R B cell line stably expressing the wild-type or mutant forms. Quantitative Western blotting revealed that each of the CD1d mutants was present at somewhat less than 50% of the wild-type level, with the glycan 4 mutant the most poorly expressed (Fig. 1B). To investigate plasma membrane expression, the cells were stained with a PE-conjugated mAb CD1d42 specific for CD1d-␤ 2 m heterodimers and analyzed by flow cytometry. All the glycosylation mutants were expressed at the cell surface (Fig. 1C). This suggests that assembly and transport through the secretory pathway is not dependent on the presence of any single N-linked glycan.
None of the Four Glycans of CD1d Are Essential for Recognition by NKT Cells-We then wished to determine whether the absence of any of the N-linked glycans interferes with lipid loading and presentation to CD1d-restricted NKT cells. ␣-Gal-Cer can bind to cellular CD1d by direct exchange at the plasma membrane or in the endocytic system (6,7). A derivative, Gal-GalCer, can bind to CD1d but cannot stimulate NKT cells unless the terminal galactose residue is removed by lysosomal ␣-galactosidase to generate ␣-GalCer (30). C1R cells expressing wild-type or mutant CD1d molecules were incubated for 6 h at 37°C with vehicle control (Me 2 SO) or lipid (100 ng/ml ␣-Gal-Cer or GalGalCer), fixed, and then co-cultured with the 6F5 NKT cell line for 22 h. The culture supernatants were assayed for secreted IFN-␥ by a sandwich ELISA. All the CD1d mutants were able to present both ␣-GalCer and GalGalCer to the NKT cell line to about the same extent as wild-type CD1d (Fig. 2). Moreover, even in the absence of exogenously added antigen, IFN-␥ release was readily detected, indicating that endogenous lipid loading by the mutants was normal. It was also evident from these experiments that the reduced surface expression of the mutant forms, seen in Fig. 2C, did not result in proportionate reductions in NKT cell recognition. Thus the absence of any single glycan had no discernable effect on the lipid binding and T cell recognition characteristics of CD1d.
Intracellular Transport of the CD1d Single N-Glycosylation Mutants-N-glycans can facilitate glycoprotein folding and trafficking through the secretory pathway (31). The expression of the CD1d mutants on the cell surface showed that they were transported but did not address whether the absence of any of the individual glycans affects the efficiency of folding or the rate of transport. To examine this question, the C1R transfectants were labeled with [ 35 S]methionine/cysteine for 15 min, chased up to 8 h, and lysed in 1% Triton X-100 in TBS. The proteins were immunoprecipitated with the monoclonal antibodies CD1d51 or D5, which recognize CD1d-␤ 2 m heterodimers or free heavy chains, respectively (25). Transport was monitored by acquisition of resistance to Endo H treatment. The transport properties of the mutants lacking glycan 1, 3, or 4 were similar to that of wild-type CD1d. The glycans of the CD1d-␤ 2 m dimers began to acquire Endo H resistance at 2 h of chase, and by 8 h, the majority was in the mature, complex form (Fig. 3). The mutant lacking glycan 2 was transported more rapidly, with Endo H-resistant mature glycans detectable at 1 h of chase. Control immunoprecipitation using the anti-HLA-A2 monoclonal antibody BB7.2 gave no bands upon SDS-PAGE (data not shown).
Wild-type CD1d molecules with mature glycans were predominantly associated with ␤ 2 m, measured by efficient immunoprecipitation with the CD1d51 mAb as compared with the D5 mAb. This is best seen at the 4-and 8-h time points, looking at the Endo H-treated lanes of Fig. 3. The mutants lacking gly-can 3 or 4 resembled wild-type CD1d. In contrast, more of the mature forms of the mutants lacking glycan 1 or 2 reacted with the D5 mAb, consistent with a lack of associated ␤ 2 m. The effect was most pronounced for the mutant lacking glycan 2. CD1d heavy chains associate with ␤ 2 m in the ER, as suggested by Fig. 3 and as described previously (21), and surface CD1d molecules are predominantly ␤ 2 m-associated. Therefore these observations suggested either that CD1d heavy chains lacking glycan 1 or 2 can be transported from the ER efficiently without ␤ 2 m association or that the heterodimers they form dissociate more readily upon detergent solubilization.
CD1-␤ 2 m Dimers Lacking Glycan 2 Dissociate Readily in Detergents-Different detergents have variable effects on protein-protein interactions. For example, the interaction of transporter associated with antigen processing (TAP) with the transmembrane glycoprotein tapasin is preserved in digitonin but disrupted in Triton X-100 (32). We therefore compared the stability of wild-type and mutant CD1d-␤ 2 m dimers in digitonin and Triton X-100. Transfected C1R cells were labeled for 15 min with [ 35 S]methionine/cysteine and chased for 4 h to ensure that the majority of the molecules had exited the ER. Aliquots of the cells were extracted in either 1% Triton X-100 or 1% digitonin in TBS, and immunoprecipitations were performed with the D5 mAb or a rabbit anti-human ␤ 2 m antibody to compare the amounts of free heavy chain and CD1d-␤ 2 m dimers. Control experiments using untransfected C1R cells showed little background in immunoprecipitations using these antibodies (Fig.   4A). In digitonin, wild-type CD1d and the mutants lacking glycan 1, 3, or 4 were predominantly ␤ 2 m-associated (Fig.  4A). In contrast, in Triton X-100, substantially more free heavy chains were observed, indicating that digitonin preserves the dimers better than Triton X-100. In the case of the glycan 1 mutant, a sharp band likely corresponding to unprocessed free heavy chains was observed in both digitonin and Triton X-100. This is consistent with the data presented in Fig. 3. For the mutant lacking glycan 2, there was considerable free mature heavy chain even in digitonin, suggesting that this glycan has a critical role in maintaining the stability of the CD1d-␤ 2 m heterodimer upon detergent extraction.
To confirm the results in Fig. 4A and to rule out a role for differences in the extraction capacity of the two detergents, cells were similarly labeled and extracted in 1% digitonin, and the CD1d-␤ 2 m heterodimers were immunoisolated with anti-␤ 2 m antibody and protein A-Sepharose beads. The beads were then incubated for 4 h either in TBS or in TBS containing 1% digitonin or 1% Triton X-100 followed by centrifugation. The extent to which free heavy chains dissociated from ␤ 2 m upon incubation in the various conditions was assessed by immunoprecipitation of the supernatants with the D5 antibody. As shown in Fig. 4B, for all the species, Triton X-100 induced substantial heavy chain dissociation from ␤ 2 m as compared with incubation in the absence of detergent or in the continued presence of digitonin. As expected from the data in Fig. 4A, the most pronounced effect was observed for CD1d-␤ 2 m dimers lacking glycan 2.
To investigate the effect of the presence of glycan 1 or 2 on CD1d interactions with ␤ 2 m, we generated triple N-glycosylation mutants of human CD1d and expressed them in C1R cells. Mutants expressing only glycan 1 or only glycan 2 were readily detected at the plasma membrane in association with ␤ 2 m, although at lower levels than wild-type CD1d (Fig.  4C). Mutants expressing only glycan 3 or 4 were expressed at even lower levels, and it was very difficult to isolate significant quantities of radiolabeled CD1d-␤ 2 m heterodimers from these cells. Similarly, a mutant CD1d lacking all four glycans was not detectably surface-expressed at all. Therefore these mutants were not further analyzed (data not shown). For the mutants expressing glycan 1 or 2, we used the same experimental approach as in Fig. 4B to examine the stability of the singly glycosylated CD1d-␤ 2 m dimers. The dimers were immunoisolated using the anti-␤ 2 m antibody from digitonin extracts. After incubation without detergent or in digitonin or   DECEMBER 29, 2006 • VOLUME 281 • NUMBER 52

JOURNAL OF BIOLOGICAL CHEMISTRY 40373
Triton X-100, the free heavy chains released were quantitated. The results, shown as the percentage of the total amount of heavy chain that remains ␤ 2 m-associated, are presented in Fig.  4D. In TBS or digitonin, very little dissociation occurs for any of the three forms. In Triton X-100, ϳ55% of the wild-type CD1d heavy chains remained ␤ 2 m-associated. If only glycan 2 was present, the amount was ϳ20%. However, if glycan 2 was absent and only glycan 1 was expressed, virtually none of the CD1d heavy chain remained ␤ 2 m-associated. The data suggest that glycan 2 is essential for the stability of the CD1d-␤ 2 m complex in the presence of Triton X-100. Glycan 2 Is Not Processed to Complex Forms in Wild-type CD1d-Endo H digestion of mature wild-type CD1d heavy chains reproducibly results in a slight increase in electro- In A, the cells were extracted in 1% digitonin or 1% Triton X-100 (Triton X), and free heavy chains and CD1d-␤ 2 m dimers were isolated using the D5 mAb or rabbit anti-human ␤ 2 m serum, respectively. After stripping in SDS, the CD1d heavy chains were reimmunoprecipitated with the D5 mAb and detected by SDS-PAGE and autoradiography. In B, the cells were extracted in 1% digitonin, and ␤ 2 m-associated heavy chains were isolated with anti-␤ 2 m antibody. The beads bearing the antigen-antibody complexes were incubated in TBS (control), 1% digitonin, or 1% Triton X-100 for 4 h. After pelleting, the beads (B) were then heated in 1% SDS to release heavy chains still associated with ␤ 2 m, and they and the free heavy chains released into the supernatants by detergent treatment (S) were immunoprecipitated using the D5 mAb and detected by SDS-PAGE and autoradiography. C, C1R cells expressing wild-type CD1d or the triple mutants expressing glycan 1 or 2 only were stained with PE-conjugated CD1d42 anti-CD1d (solid line) or PE-conjugated isotype control IgG 1 (dotted line) and analyzed by flow cytometry. D, C1R cells expressing wild-type CD1d or glycan 1(⌬2, ⌬3, and ⌬4) or 2 (⌬1, ⌬3, and ⌬4) only were labeled, digitonin-extracted, and immunoprecipitated with rabbit anti-␤ 2 m. The heavy chains released by incubation in TBS, digitonin, or Triton X-100 and the residual ␤ 2 m-associated heavy chains were isolated, separated, and identified as in panel B. The graphs represent the percentage of the total CD1d heavy chain still associated with ␤ 2 m under the three conditions. phoretic mobility on SDS-PAGE (Fig. 3). The mobility shift corresponds to a reduction in the apparent molecular mass of ϳ2-3 kDa, expected when a single N-linked glycan is removed. To identify the Endo H-sensitive glycan, CD1d-␤ 2 m dimers were purified from 5 ϫ 10 9 C1R.CD1d cells using a CD1d51 immunoaffinity column. The cells were lysed in the detergent C 12 E 9 , which, similar to digitonin, preserves the CD1d-␤ 2 m interaction (data not shown). The dimers were acid-eluted, ethanol-precipitated, and solubilized in 0.2% SDS. The yield was ϳ200 g. After overnight digestion with Endo H or PNGase F, aliquots were separated by SDS-PAGE and transferred electrophoretically to an Immobilon P membrane, and the CD1d heavy chains were detected by probing with the D5 mAb. The purified CD1d heavy chains were predominantly Endo H-resistant as expected, but the slight increase in mobility previously observed upon Endo H digestion was apparent (Fig. 5,  upper panel, inset). Deglycosylation was complete when PNGase F was used.
To identify the Endo H-sensitive glycan, 4 g of untreated and Endo H-digested purified CD1d heavy chains were separated by SDS-PAGE, and "in-gel" tryptic digestions were performed. The peptides were extracted from the gel and sepa-rated by reverse phase HPLC using a C18 column. As shown in Fig. 5, comparison of the two samples revealed only a single peak with an altered retention time. The peak from the Endo H-treated sample had a longer retention time (84 min) than the one from the control sample (82 min). The peptides eluting in these peaks were N-terminally sequenced by Edman degradation. The peak from the control sample contained a mixture of two peptides with the N-terminal sequences TDGLAXLGELQTH and LXYPX-ELQVSAGXEVKP, where X is an undetermined amino acid. The first sequence was dominant and corresponds to the peptide containing glycan 2 (residues 28 -48, with the anticipated glycosylated Asn residue at position 42). The second corresponds to the peptide containing glycan 3 (residues 90 -121, with the anticipated glycosylated Asn residue at position 108). The specific peak from the Endo H-treated sample contained a single peptide with the N-terminal sequence TDG-LAXLGELQTH, clearly corresponding to the peptide derived from the glycopeptide containing glycan 2. The residual small peaks from the Endo H-treated sample eluting at 82-83 min were also subjected to Edman sequencing. This revealed a low level mixture of the two peptides found in the untreated sample (data not shown). The data clearly show that glycan 2 in the CD1d heavy chain remains predominantly Endo H-sensitive, suggesting that it may be relatively inaccessible to the Golgi glycosyl transferases responsible for the formation of complex glycans. A likely explanation is that the glycan is obscured in the transport-competent folded form of the glycoprotein.
Glycan 2 Interacts with the Surface of ␤ 2 m-To gain insight into the possible orientation of glycan 2 in the assembled CD1d-␤ 2 m dimer, we performed molecular modeling using the published structure of the CD1d-␤ 2 m-␣-GalCer complex (3). For simplicity, the glycans were all modeled in the GlcNAc 2 Man 9 form. The analysis revealed that glycan 2, positioned at Asn-42, exhibited potentially significant glycan-protein clashes, whereas the three remaining sites could accommodate glycans without any stereochemical conflicts. Further extensive conformational searches showed that to avoid clashes with either the heavy chain or ␤ 2 m, the allowed configuration of the Asn-42-glycan linkage was severely restricted (Fig. 6A). This is due to the very low accessibility of this site, which restricts the glycan in relation to heavy chain, and also its loca-FIGURE 5. Glycan 2 is predominantly Endo H sensitive in wild-type (WT) CD1d-␤ 2 m dimers. CD1d-␤ 2 m dimers were isolated from C1R.WT.CD1d cells solubilized in 2% C 12 E 9 in TBS, pH 7.4, by affinity chromatography using a CD1d51 immunoaffinity column. They were incubated overnight with or without Endo H, and the proteins were separated by SDS-PAGE. Endo H-treated or untreated heavy chains were in-gel digested with trypsin, and the extracted peptides were separated by reverse phase-HPLC. The peptides from the starred peaks were sequenced by Edman degradation. In the upper panel, the inset shows a Western blot of the purified CD1d heavy chains, untreated or digested with Endo-H or PNGase F, detected by the D5 mAb.
tion with respect to the associated ␤ 2 m. Finally, as a starting point for modeling, an Asn-42-glycan linkage configuration was chosen that was the best compromise between a minimal deviation from the experimental Asn-42 side chain conformation, a minimal deviation from the closest experimentally documented Asn-glycan linkage conformation, and constraints imposed by the nature of clashes and the reduction of their number: for example, the need to avoid superposition of backbone atoms and side chain ␤-carbons.
The starting model with glycans attached in the experimental configuration for Asn-20 (glycan 1), Asn-108 (glycan 3), and Asn-163 (glycan 4) and the optimal configuration at Asn-42 (glycan 2) was subjected to repeated rounds of local simulated annealing followed by energy minimization, with constraints imposed on ␣-carbons and further constraint-free energy minimization. Modeling the structure around site Asn-42 needed careful attention because of the very limited allowed conformational freedom. To avoid steric clashes with the CD1d heavy chain, the glycan had to be persistently oriented toward ␤ 2 m, and only a narrow space available at the surface of ␤ 2 m avoided clashes with both heavy chain and ␤ 2 m. In contrast, glycan 1 at Asn-20 is relatively distant from ␤ 2 m and thus forms fewer contacts with it. The location of the remaining two glycans in the structure makes contacts with ␤ 2 m impossible. A structural model with glycan 2 added is presented in Fig. 6B.
Differential solvent accessibility analysis of the model also reveals extensive contacts between glycan 2 and the protein core of CD1d, including both heavy chain and ␤ 2 m. As described above, this glycan practically sits on the surface of ␤ 2 m (Fig. 6B), but more importantly, it shelters the interface between heavy chain and ␤ 2 m. For example, four residues located at this interface, heavy chain residues Arg-25 and Asp-27 and ␤ 2 m residues Asp-35 and Leu-55, show a direct and consistent reduction of their solvent accessibility when the glycan is attached. These observations suggest a possible explanation for the experimental data showing that this glycan retains sensitivity to Endo H and for the complex behavior in detergents of CD1d-␤ 2 m dimers in the mutants lacking glycan 2.

DISCUSSION
The addition of N-linked glycans to newly synthesized glycoproteins is thought to occur during translocation into the ER, shortly after the three-residue glycosylation site (NXT or NXS, where X is any amino acid except proline) enters the ER lumen (33). At this stage, conformational restraints are limited, and consequently, almost all available glycosylation sites are occupied. In their monoglucosylated form, the glycans can serve as substrates for CNX or CRT binding, facilitating glycoprotein folding by the CNX/CRT/ERp57-dependent folding cycle in which the enzymes glucosidases I and II and UDP-glucose glycoprotein transferase work together to ensure the correct conformation and disulfide bonding pattern (31). Once folding is complete, the glycoproteins typically leave the ER and transit the Golgi apparatus on their way to the plasma membrane. During transport through the Golgi, multiple additional modifications and sugar additions take place, resulting in the generation of complex-type glycans terminating in sialic acid residues. Characteristically, N-linked glycans are surface-exposed, and because of this, the majority of them mature to the complex type. Occasionally, a particular N-linked glycan in a glycoprotein remains in the immature form. Steric constraints that are imposed after folding and consequent shielding of the glycan from the trimming enzymes and glycosyl transferases present in the Golgi that are responsible for the later stages of glycan maturation are a plausible explanation for this course of events (34).
The initial stages of human CD1d folding and disulfide bond formation are glucose trimming-dependent and involve CNX, CRT, and ERp57 (21). Notably, disulfide bond formation is complete before the association of the heavy chain with ␤ 2 m, and the chaperones do not interact with CD1d-␤ 2 m dimers. Therefore a likely scenario based on what we have observed is that glycan 2, which remains almost exclusively Endo H-sensitive after CD1d transport to the cell surface, becomes inaccessible after ␤ 2 m association. The structural model presented in Fig. 6B is entirely consistent with this hypothesis; glycan 2 substantially interacts with both the heavy chain and ␤ 2 m in the folded structure. Glycans 3 and 4 are not predicted to interact with the folded protein, and only minor interactions are possible for glycan 1. It is also interesting to note that the CD1d mutant lacking glycan 2 is transported more rapidly from the ER, based on the rate at which it acquires Endo H resistance (Fig. 3). Based on the model structure in Fig. 6B, it is likely that the significant structural constraints imposed on the CD1d heavy chain-␤ 2 m interaction by glycan 2 slow down the rate of dimer formation and subsequent exit from the ER. The prolonged residence in the ER provoked by glycan 2 may be important in facilitating lipid association with CD1d molecules in the ER, which is mediated by microsomal lipid transfer protein (18).
The proximity of glycan 2 to the region of the CD1d-␤ 2 m interaction presumably in some way also explains the ready detergent-induced dissociation observed in the absence of this glycan (Fig. 4). Mutants lacking glycan 2 that express only glycan 1 or glycans 1, 3, and 4 dissociate more readily in Triton X-100 than wild-type CD1d-␤ 2 m dimers or dimers incorporating mutants that express glycan 2. The interactions of glycan 2 with both the heavy chain and ␤ 2 m, as predicted by the model, could play a role in stabilizing the dimer. There are two possible mechanisms that could explain this. First, detergent monomers could interact with the lipid binding site. The first structure obtained for a CD1 molecule with a bound ligand, that of CD1b, had a detergent molecule and a lipid molecule present simultaneously in the binding groove (35), and we have observed detergent-dependent dissociation of CD1d-bound lipids in the laboratory (data not shown). Conceivably, eliminating associated lipid could reduce the stability of the CD1d-␤ 2 m dimer, similar to the way that lack of peptide association reduces the stability of MHC class I-␤ 2 m dimers (36), and eliminating the interactions of glycan 2 with the protein components could exacerbate this. Alternatively, monomeric detergent molecules could insert more readily into the site of interaction of the CD1d heavy chain and ␤ 2 m when glycan 2 is absent. The solvent accessibility constraints imposed by the presence of glycan 2 that we observed during the modeling procedure would be consistent with this idea. The hydrophobic domain of digitonin is a rigid, steroid-based structure with a covalently bound pentameric sugar forming the hydrophilic portion of the detergent. On the other hand, Triton X-100 is relatively flexible with an octylphenol hydrophobic component and a polyethoxylate hydrophilic component. The flexibility of Triton X-100 may allow it to insert more readily than digitonin into the CD1d-␤ 2 m interaction site, accounting for the differential abilities of the two detergents to induce dissociation (Fig. 4). Alternatively, and for similar reasons, Triton X-100 may be better able than digitonin to induce lipid dissociation from the binding site. Currently, no data are available that would allow us to distinguish between these two hypotheses.
The proposed roles of N-linked glycans are varied, ranging from protein folding (31), intracellular transport (37), and protection from degradation (31,38). However, there are few examples in the literature where defined interactions between glycans and the proteins that bear them have demonstrated biological significance. One interesting example is in the proform of the lysosomal enzyme cathepsin Z, where an interaction between a high mannose oligosaccharide and a ␤-hairpin loop in the protein appears to maintain the glycan in a conformation better able to interact with the cargoreceptor ERGIC-53 (39). Elimination of the glycan or the loop by mutation reduces the rate of export of the proenzyme from the ER. Interestingly, maintenance of the loop is required to prevent the maturation of the glycan into the complex form. This is similar to the suggestion from our own work that its interaction with the CD1d-␤ 2 m interface prevents the maturation of glycan 2.
CD1d molecules use the glycan-dependent CNX/CRT cycle during folding, although assembly can still proceed in the absence of any single glycan, and indeed, it can occur to some extent if only glycan 1 or 2 is present (Fig. 4C). Assembly is seriously impeded, however, if only glycan 3 or 4 is present, and in the complete absence of glycans, no assembly or surface expression of heavy chain-␤ 2 m dimers occurs (data not shown). This is likely to reflect a requirement for the glycan-dependent CNX/CRT folding cycle. Glycan 1 is the only glycan absolutely conserved among all the CD1d family members and therefore may be of particular importance. However, CD1d molecules clearly fold to some extent and they maintain function in its absence. Glycan 2 is not present in CD1b or CD1c, but a similarly placed glycan is present in CD1a. Whether this glycan interacts with ␤ 2 m in CD1a remains unknown.
CD1d molecules recycle continuously through late endocytic compartments and are therefore exposed to proteases for prolonged periods. Presumably because of this, they have evolved to be highly protease-resistant (29,40), allowing their survival in this environment. The presence of four N-linked glycans flanking the lipid binding site may play a role in this stability. However, the pulse-chase data presented in Fig. 3 do not indicate that any of the mutants lacking a single glycan are more rapidly turned over than the wild-type molecule. Thus the reason why CD1d molecules have evolved to express four glycans in the particular positions they occupy is not clear. However, the data we have presented argue that glycan 2 in particular is important for maintaining the stability of the CD1d molecules and that interactions between this glycan and both components of the heterodimer are likely to play a critical structural role.