O- Glycan Sialylation and the Structure of the Stalk-like Region of the T Cell Co-receptor CD8*

Studies of mucins suggest that the structural effects of O- glycans are restricted to steric interactions between peptide-linked GalNAc residues and adjacent polypeptide residues. It has been proposed, however, that differential O- glycan sialylation alters the structure of the stalk-like region of the T cell co-receptor, CD8, and that this, in turn, modulates ligand binding (Daniels, M. A., L., and Cell 107, 501–512). We characterize the glycosylation of soluble, chimeric forms of the (cid:1)(cid:1) - and (cid:1)(cid:2) -isoforms of murine CD8 containing the O-glycosylated stalk of rat CD8 (cid:1)(cid:1) , and we show that the stalk O- glycans are differentially sialylated in CHO K1 versus Lec3.2.8.1 cells (82 versus (cid:1) 6%, respectively). Sedimentation analysis indicates that the Perrin functions, P exp , which reflect overall molecular shape, are very similar (1.61 versus 1.54), whereas the sedimentation coefficients ( s ) of the CHO K1- and Lec3.2.8.1-derived proteins differ considerably (3.73 versus 3.13 S). The hydrodynamic properties of molecular models also strongly imply that the sialylated and non-sialylated forms of the chimera have parallel, equally highly extended stalks ( (cid:1) 2.6 Å/residue). Our analysis indicates that, as in the case of mucins, the overall structure of O- glycosylated stalk-like peptides is sialylation-independent and that the functional effects of differential CD8 O- glycan sialylation need careful interpretation.

The early responses of T lymphocytes are determined by interactions of both the T cell receptor (TCR) 1 and CD8 or CD4 molecules with major histocompatibility complex (MHC) molecules on antigen-presenting cells. CD8 is required by mature T cells restricted to recognizing foreign antigenic determinants (peptides) complexed with class I MHC antigens (MHCp), whereas CD4 is involved in MHC class II-restricted T cell responses (1). Because CD8 and CD4 are believed to interact with the same MHCp complex as the TCR (2), these proteins are commonly referred to as T cell co-receptors.
CD8 is a glycoprotein consisting of disulfide-linked subunits, ␣ and ␤, which are encoded by two closely linked genes near the immunoglobulin locus (3). Despite sharing relatively low sequence similarity (ϳ20%), the CD8 subunits are structurally related and predicted to have identical topologies (4). Each chain consists of an extracellular immunoglobulin superfamily (IgSF) V-set domain attached to hydrophobic transmembrane sequences and short cytoplasmic tails via extended, disulfidelinked stalk-like peptides of 48 -51 (␣-chain) or 37-42 (␤-chain) residues (reviewed by Gao et al. (5)). The IgSF domains in CD8 exhibit highly variable N-linked glycosylation; in humans, only a single site on the ␤-chain is glycosylated, whereas in mice the ␣and ␤-chains have three and one sites, respectively. The ratio seen in the mouse is reversed for rat CD8. In all species, the stalk-like region of each chain is rich in proline, serine, and threonine residues and is O-glycosylated. Amino acid sequencing of rat CD8␣ has indicated that the four threonine residues 122, 126, 132, and 134 clustered at the membrane-distal end of the stalk are occupied with O-glycans (6). The cytoplasmic domain of CD8␣ is attached to the tyrosine kinase p56 lck (reviewed in Ref. 7). Most T cells express CD8 as an ␣␤ heterodimer, although a homodimeric form (CD8␣␣) is found on subsets of intraepithelial T lymphocytes of the gut, ␥␦ T cells, NK cells, and lymphoid-related dendritic cells (reviewed by Zamoyska (8)). CD8␣␣ and CD8␣␤ T cells undergo different pathways of selection, perhaps reflecting an underlying functional specialization within the CD8 T cell population (9).
The interaction between CD8␣␣ and class I MHCp ligands was initially investigated using cell adhesion assays (10). More * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
b Both authors contributed equally to this work. e Recipient of a Royal Society traveling fellowship award. g Supported by the Wellcome Trust. i Present address: Avidex Ltd., 57c Milton Park, Abingdon, OX14 4RX, UK.
j To whom correspondence may be addressed. E-mail: pmr@glycob. ox.ac.uk.
k To whom correspondence may be addressed. E-mail: sdavis@ molbiol.ox.ac.uk. 1 The abbreviations used are: TCR, T cell receptor; CHO, Chinese hamster ovary; MS, mass spectrometry; IgSF, immunoglobulin super-family; NP-HPLC, normal phase-high performance liquid chromatography; MALDI-TOF MS, matrix-assisted laser desorption/ionization time-of-flight mass spectrometry; MHC, major histocompatibility complex; MHCp, MHC-peptide complex; Q-TOF, quadrupole time-of-flight mass spectrometry; sCD8, soluble form of CD8; sCD8␣␣K1, murine sCD8␣␣ expressed in CHO K1 cells; sCD8␣␣Lec, murine sCD8␣␣ expressed in Lec3.2.8.1 cells; sCD8␣␣E, human CD8␣␣ expressed in recently, the affinities and kinetics of the interaction of soluble forms of human and mouse CD8␣␣ with human and mouse class I MHC, respectively (11,12), and of human CD4 and murine MHC class II (13), have been determined using surface plasmon resonance-based methods. In both cases, the interactions were shown to have extraordinarily low affinities and extremely rapid dissociation kinetics. Crystallographic analyses of the complexes of CD8␣␣/class I MHCp from humans and mice (14,15) indicate that CD8 binds the membrane-proximal ␣2 and ␣3 domains of MHCp and makes additional contacts with ␤ 2 -microglobulin. An analogous interaction was revealed by the crystal structure of the complex of human CD4 domains 1 and 2, and murine MHC class II (16), insofar as CD4 binds a membrane-proximal cavity formed by residues from both the ␣2 and ␤2 domains of class II. It has been proposed that the weak interactions between the TCR and MHCp may be enhanced by simultaneous interactions involving CD8 (17). However, CD8 interactions are generally at least 10-fold weaker than those involving the TCR, and CD4 interactions are possibly even weaker. Therefore, although they will contribute somewhat to the overall interaction, it seems more likely that the key function of the co-receptors is to recruit sufficient p56 lck to pre-formed TCR⅐MHCp complexes to consolidate the early signaling response (11).
An additional level of complexity regarding these interactions arises from the recent suggestion that ligand binding by CD8 is regulated in vivo by changes in its O-glycosylation. Specifically, Moody et al. (18) and Daniels et al. (19) each report that the binding of tetrameric forms of MHCp in a CD8-dependent, non-cognate manner to double-positive thymocytes diminishes as O-glycan sialylation increases and the thymocytes progress through positive selection. To account for this effect, sialylation-induced changes in the structural properties of the stalk, either involving electrostatic repulsion between the two chains (18) or chain extension effects (19), were proposed to affect the ligand-binding site even though this is located ϳ30 Å from the stalk. As has already been noted (20), thermodynamic considerations make an electrostatic repulsion-based mechanism for these effects extremely unlikely. In the present study, we test the second proposed mechanism, i.e. that O-glycan sialylation modulates the extension of the stalklike region of CD8. Our findings suggest that sialylation has little, if any, effect on the overall structure of CD8.

EXPERIMENTAL PROCEDURES
Chemicals and Reagents-Anhydrous hydrazine was prepared by distillation from reagent-grade hydrazine (Pierce, catalog number 21515-1) mixed with calcium oxide and toluene. Reagents for automated hydrazinolysis were from Oxford GlycoSciences (Abingdon, UK). Chromatography (grade 3MM) paper was from Whatman. Acetonitrile and methanol were from Riedel-de-Haën (Sigma). All other chemicals were AnalaR-grade from Sigma.
Enzymes-Sequencing-grade exoglycosidase and peptide N-glycosidase F (PNGaseF, EC 3.5.1.52) enzymes were obtained from Glyko Ltd. (Upper Heyford, UK) unless specified. Incubation conditions for single enzyme digests were as described by Rudd et al. (21), using the buffers provided by the enzyme manufacturers. When enzyme arrays were used, the buffer was 50 mM sodium acetate, pH 5.5. Exoglycosidases were used at the following concentrations. Preparation of sCD8␣␣ and sCD8ab-Four recombinant soluble forms of CD8 were prepared as described previously (22). Briefly, sCD8␣␤ was produced by co-expressing sCD8␣ with sCD8␤ constructs (Fig. 1) in both Chinese hamster ovary (CHO) K1 cells (referred to as sCD8␣␣K1 and sCD8␣␤K1) and the mutant CHO cell line Lec3.2.8.1 (referred to as sCD8␣␣Lec and sCD8␣␤Lec). The sCD8␣␤ was purified to homogeneity by OX-8 affinity and/or ion-exchange chromatography followed by gel filtration and the sCD8␣␣ by immunoaffinity chromatography on an antibody (OX-8) column. For ultracentrifugation-based structural comparisons with these mammalian cell-expressed forms of murine CD8, Escherichia coli-expressed unglycosylated human sCD8␣␣ consisting of residues 1-120 of the mature protein, referred to here as sCD8␣␣E, was prepared as described in detail elsewhere (23).
Hydrazine Release, Re-N-acetylation, and Labeling of N-and O-Glycans-Approximately 50 g of each protein was dialyzed against 0.1% trifluoroacetic acid and then lyophilized. N-Glycans were released by hydrazine at 95°C (N* mode) and purified using a GlycoPrep 1000 (Oxford GlycoSciences Ltd., Abingdon, UK). These hydrazinolysis conditions represent a compromise between achieving non-selective release, maximizing yield, and minimizing released sugar degradation (24,25). Manual hydrazinolysis for selective release of O-glycans was performed as described previously (24,26). For in-gel release of the glycans, the ␣and ␤-chains of CHO K1 sCD8 were separated on a one-dimensional SDS-PAGE gel that was stained with Coomassie Blue. The N-linked glycans were released from gel slices by incubation with PNGaseF as described previously (27).
Labeling and High Performance Liquid Chromatography (HPLC)-Fluorescent labeling with 2-aminobenzamide (2-AB) was performed as described by Bigge et al. (28) using the kit provided by Glyko Ltd. HPLC was performed essentially as described by Guile et al. (29). The procedures for analyzing N-glycans were as described by Rudd et al. (21) and for O-glycans as described by Royle et al. (30). Glucose units (29) were calculated from retention time by reference to a standard consisting of a partially hydrolyzed dextran ladder using a prototype version of PeakTime, a program developed by Dr. E. Hart at the Oxford Glycobiology Institute.
Simultaneous Exoglycosidase Sequencing of the Released Glycan Pool-Exogylcosidase sequencing was performed as described previously (31,32). Normal phase HPLC separations (NP-HPLC) were performed on a GlycoSep-N chromatography column (Glyko Ltd.) using low salt conditions, and structures were assigned as described previously (29).
Trypsin Digestion-50 g of each sample (sCD8␣␣K1 and sCD8␣␣Lec) was dissolved in 200 l of 6 M guanidine HCl, 0.5 M Tris, pH 8.0, with 3 l of 0.5 M dithiothreitol and reduced by incubation at 37°C for 2 h. The sample was then alkylated by addition of 3 l of 4-vinylpyridine and incubated for a further 1.5 h in the dark at 37°C and then dialyzed against 50 mM NH 4 HCO 3 . Samples were dried, then resuspended in 40 l of trypsin solution (1.25 g/ml in 25 mM NH 4 HCO 3 10% acetonitrile), and incubated overnight at 37°C. Peptides were extracted with a C18 ZipTip (Millipore Corp., Bedford, MA) and analyzed directly by MALDI mass spectrometry. A fifth of each sample was desialylated with A. ureafaciens sialidase, and the glycopeptides were analyzed by MALDI after extraction with a C18 ZipTip.
MALDI Mass Spectrometry of Glycopeptides-Samples were purified using C18 ZipTips (Millipore, Bedford, MA) prior to analysis by MALDI-TOF mass spectrometry (33). Positive ion MALDI-TOF mass spectra were recorded with a Micromass TOFSpec 2E reflectron-TOF mass spectrometer (Micromass Ltd., Atlas Park, Simonsway, Manchester, UK) fitted with delayed extraction and a nitrogen laser (337 nm). The extraction voltage was 20 kV; the pulse voltage was 3200 V, and the delay was 500 ns. Samples were prepared by mixing 0.3 l of the solution of tryptic peptides and glycopeptides on the MALDI target together with the matrix solution (0.3 l from a solution of ␣-cyano-4hydroxycinnamic acid (10 mg/ml) in 7:3 (v/v) acetonitrile, 0.1% trifluoroacetic acid) and allowing it to dry at room temperature. The resulting molecular weights were used to search the sequence data base maintained at the European Molecular Biology Laboratory (EMBL) in Heidelberg, Germany.
MALDI Mass Spectrometry of Released Glycans-Underivatized, released glycans were purified using a Nafion 117 membrane (Aldrich) and examined by MALDI mass spectrometry with the TOFSpec instrument in positive ion reflectron mode. Samples (0.3 l in water) were mixed with a saturated solution of 2,5-dihydroxybenzoic acid on the MALDI target and allowed to dry at room temperature. The sample was then recrystallized from ethanol. Operating conditions for the mass spectrometer are as follows: extraction voltage, 20 kV; pulse voltage, 3200 V. The instrument was calibrated with dextran oligomers. Monoisotopic masses of the [M ϩ Na] ϩ ions are listed in Tables I and II; all masses were within 0.1 mass unit of the calculated values.
Q-TOF Mass Spectrometry of Glycopeptides-Electrospray MS and MS/MS spectra were recorded with a Micromass Q-TOF mass spectrometer fitted with a Z-spray ion source and a nanoflow injection system. The ion source temperature was 150°C; the needle voltage was 1668 V, and the cone voltage was 80 V. Samples were infused in 1:1 double-distilled water/acetonitrile containing 0.1% formic acid at about 20 nl/min from borosilicate capillary needles. For MS/MS, the collision energy was 35-40 V for the doubly charged [M ϩ 2H] 2ϩ ions from the C-terminal glycopeptide ions and 100 -110 V for the singly charged ions. The mass window for parent ion selection was about 4 Da. Argon at 20 pounds/square inch was used as the collision gas. Sample acquisition and processing were performed with the Micromass MassLynx data system.
Analytical Ultracentrifugation-Analytical ultracentrifugation experiments were performed using a Beckman Optima XL(I) analytical ultracentrifuge, which is equipped with absorbance and interference optics. Samples of sCD8␣␣E, sCD8␣␣K1, and sCD8␣␣Lec were spun at 40,000 rpm at 20°C. Sample distributions were recorded, using the interference or absorbance systems, at 2-min intervals. The data were analyzed using the Sedfit software (34,35). The partial specific volumes and masses of sCD8␣␣⌲1, sCD8␣␣Lec, and sCD8␣␣E were calculated using the known amino acid composition of CD8 and the glycosylation analysis of sCD8␣␣⌲1 and sCD8␣␣Lec reported herein. The calculation was accomplished using a derivative program of AtoB (36). The sedimentation coefficients of sCD8 calculated over a concentration range were extrapolated to infinite dilution and corrected for buffer viscosity and density compared with water as a standard solvent. The program Sedfit was used to calculate the frictional ratio of each sample f/f 0 , this being the ratio between the experimental frictional coefficient and the frictional coefficient of a spherical species of the same mass and partial specific volume. The frictional and sedimentation coefficients obtained were corrected for hydration effects for a series of different levels of hydration as previously described (37), yielding "dry" Perrin functions P exp and sedimentation coefficients s 20,w,␦ 0 (0) (S) (38): where is the partial specific volume of the protein (ml/g), 0 the buffer density (g/ml); and ␦ app is the hydration of the protein (g/g). The bead molecules used to predict hydrodynamic properties were calculated using the program AtoB (36). The solution properties of the bead models were calculated using SOLPRO (39).

RESULTS
The structure/function analysis of protein glycosylation can be problematic because preventing glycosylation, either genetically or with inhibitors, can lead to protein misfolding and/or non-secretion (for example see Ref. 40). In the case of O-glycans, post-expression enzymatic deglycosylation is difficult because commercially available endo-␤-N-acetylgalactosaminidases do not cleave all types of O-glycan chains (41). The O-glycosylation of proteins secreted by CHO cell-derived Lec3.2.8.1 cells has not been characterized previously in detail, but these cells are known to be deficient in enzymes required for sialylation (Lec2 mutations; Refs. 42 and 43) and galactosylation (Lec8 mutations; Ref. 44). It was therefore anticipated that these cells would produce soluble forms of sCD8 ␣␣ and ␣␤ with substantially modified O-glycans. We confirm this proposal here and examine its consequences for the structure of sCD8 expressed in Lec3.2.8.1 versus non-mutant CHO K1 cells using velocity sedimentation analysis, which is one of the few techniques that presently yield structural data for native glycoproteins with the properties of CD8. To allow reliable hydrodynamic modeling we also characterize the N-glycosylation of the soluble forms of CD8.
CD8 Constructs-For the present study, four recombinant soluble forms of glycosylated CD8 were prepared as described previously (22). Briefly, sCD8␣␣ consists of the mouse CD8␣ chain (residues 1-130 of the mature polypeptide) expressed as a soluble, homodimeric fusion protein with 17 residues of the rat CD8␣ stalk-like peptide (residues 122-138) containing the O-linked sugars and the OX-8 anti-rat CD8 monoclonal anti-body epitope (Fig. 1a). Overall, the stalk-like region of this protein, measured from the first residue beyond ␤-strand G of the murine ␣-chain V-set IgSF domain, consists of 26 residues. sCD8␣␤ was generated by co-transfecting the sCD8␣ construct ( Fig. 1a) together with a construct consisting of residues 1-116 of the mouse ␤-chain sequence attached to the 23 residues forming the C terminus of the sCD8␣ construct (Fig. 1b). The sCD8 isoforms were expressed in CHO K1 and Lec3.2.8.1 cells and are referred to as the K1 and Lec derivatives, respectively. For structural comparison with the glycosylated murine proteins, unglycosylated human sCD8␣␣ consisting of residues 1-120 of the mature polypeptide, i.e. completely lacking the stalk-like region of CD8, was expressed in E. coli as described in detail elsewhere (23). This protein, for which a crystal structure exists (14), is referred to as sCD8␣␣E.
Glycosylation Analysis, N-Linked Glycans-The N-glycans were released by hydrazinolysis from purified sCD8␣␣K1, sCD8␣␤K1, and sCD8␤␤K1 expressed in CHO K1 cells, and sCD8␣␣Lec and sCD8␣␤Lec expressed in Lec3.2.8.1 cells. The glycan pools were analyzed directly by MALDI-TOF mass spectrometry to obtain the composition of the constituent isobaric monosaccharides (Tables I-III) or labeled with 2-AB for analysis by NP-HPLC ( Fig. 2 and Tables I-III). The HPLC procedure separates glycans on the basis of their hydrophilicity which, in practice, corresponds very closely to molecular weight (29). HPLC chromatograms were calibrated against a standard dextran hydrolysate, enabling sample retention times to be expressed as glucose units. Preliminary structural assignments were made by comparison with glucose unit values in a data base of standard N-glycans (21). The glycosylation profiles of the ␣␣ and ␣␤ constructs, and of a non-physiological form of sCD8 consisting of the ␤␤ homodimer, were all very similar (Fig. 2a, top 3 panels), indicating that there is little difference in the glycan processing of the ␣and ␤-chains and that the N-glycosylation sites are equally accessible to the processing enzymes.
The identity of the N-glycans was confirmed by exoglycosidase digestion of the pool of labeled glycans and re-analysis by NP-HPLC (representative data for sCD8␣␣ is shown in Fig.  2b). The relative proportions of each glycan, derived from integration of the fluorescence intensities of the peaks (Fig. 2a), are FIG. 1. Structures of sCD8 constructs. a, the mouse CD8␣ chain (residues 1-130 of the mature polypeptide, boxed) was fused to 17 residues of the rat CD8␣ stalk peptide (residues 122-138, italics) containing the O-glycosylation sites (arrowed). Overall, the stalk-like region of this protein, measured from the first residue beyond ␤-strand G of the murine ␣-chain V-set IgSF domain (i.e. Val-122), consists of 26 residues. b, the mouse CD8␤ chain (residues 1-116, boxed) was fused to the 23 residues forming the C terminus of the sCD8␣ construct (residues 125-147 of construct a).
shown in Table I. About a third of the glycans (34%) were of the oligomannose types Man 5,6,7,8 GlcNAc 2 . The remaining glycans were all bi-antennary complex-type glycans. Of these, 57% were sialylated, mostly as the mono-sialylated form. Thirty nine percent of the bi-antennary glycans were core-fucosylated (Table I).
The doublet was not due to the existence of distinct sCD8␣ glycoforms because it remained after digestion with PNGaseF (data not shown). Importantly, identical heterogeneity was also apparent in Lec3.2.8.1-derived protein (data not shown), but its  (Fig. 3a), were subjected to exoglycosidase sequencing as described under "Experimental Procedures" to derive the structures shown. The relative abundance was calculated from the fluorescence intensities of the peaks from the NP-HPLC analysis (Fig. 3b). The scheme for glycan representation is as follows: s, N-acetylglucosamine; {, galactose, diamond with dot, fucose; E, mannose; *, N-acetylneuraminic acid; dashed line, ␣-linkage; full line, ␤-linkage.

TABLE I Analysis of N-glycans of sCD8␣␣ expressed in CHO K1 cells
The N-glycan pool from sCD8␣␣K1 was subjected to exoglycosidase sequencing and MALDI-TOF mass spectrometry, as described under "Experimental Procedures," to derive the structures shown. The relative abundance was calculated from the fluorescence intensities of the peaks from the NP-HPLC analysis (Fig. 2a, top panel).
# N.D., not detected by NP-HPLC. * N.D., H 5 N 4 was not detected by NP-HPLC but was produced by desialylation of H 5 N 4 S 1 during the MALDI ionization process.
§ A n refers to antennae number (n); G n is the number of galactose residues; F n is the number of fucoses; and S n is the number of sialic acids; Man n refers to the number of mannose residues. The scheme for glycan representation is as follows: s ϭ N-acetylglucosamine; {, ϭ galactose; diamond with dot ϭ fucose; ⅜ ϭ mannose; ✭ ϭ N-acetylneuraminic acid; dashed line, ␣-linkage; full line, ␤-linkage.

cells
The N-glycan pool from sCD8␣␣Lec was subjected to exoglycosidase sequencing and MALDI-TOF mass spectrometry as described under "Experimental Procedures" to derive the structures shown. The relative abundance was calculated from the fluorescence intensities of the peaks from the NP-HPLC analysis (Fig. 2a, 4th panel). The scheme for glycan representation is as follows: s, N-acetylglucosamine; E, mannose; dashed line, ␣-linkage; full line, ␤-linkage. Definitions of compositions and trivial names are given in the footnote to Table I. source remains unresolved. The profiles of the glycans determined by NP-HPLC revealed differences in the proportions of glycans present in the ␣and ␤-chains; the most abundant glycan on the ␣-chain was the mono-sialylated bi-antennary glycan, N4, whereas for the ␤-chain, the fucosylated nonsialyated glycan, N3, was most abundant and another structure, N5, was absent ( Fig. 3b; Table III).
The NP-HPLC profile of sCD8␣␣Lec O-glycans (Fig. 4a) was characterized by the complete absence of the di-sialylated O3glycan structure and a substantial reduction in the monosialylated structure (Fig. 4c and Table IV). Therefore, the key difference between the two forms of sCD8␣␣ is that although 86% of the core 1 O-glycans are mono-and di-sialylated in CHO K1 cells, 82% of the Lec3.2.8.1-derived core 1 O-glycans are non-sialylated. This glycan composition is consistent with the known defects in Lec3.2.8.1 cells (42)(43)(44), which are predicted to result in the production of single GalNAc residues. However, although hydrazinolysis cleaves single GalNAc residues from proteins, some monosaccharides are removed along with peptides at a subsequent paper chromatography step. Therefore, in principle, the NP-HPLC analysis could overestimate the degree of sialylation of sCD8␣␣Lec. In order to determine whether or not single GalNAc residues occupy any of the sites in sCD8␣␣Lec, mass spectrometric analysis of the tryptic fragments of sCD8␣␣ was undertaken.
Electrospray mass spectrometry and subsequent MS/MS  Tables I and II. FIG. 3. SDS-PAGE-based separation of the sCD8 ␣and ␤-chains expressed in CHO K1 cells. a, sCD8␣␣ and -␣␤ were separated into their constituent chains using SDS-PAGE. The two isoforms of the ␣-chain are labeled ␣1 and ␣2 and were analyzed separately. N-Glycans were released using PNGaseF digestion of the Coomassie Blue-stained gel bands. b, the N-glycan pools from the gel bands were labeled with 2-AB and analyzed by NP-HPLC. GU, glucose units. The structures identified are given in Table III.

FIG. 4. NP-HPLC and sequence analysis of the O-glycans from sCD8 constructs expressed in CHO K1 cells and in Lec 3.2.8.1 cells. a, NP-HPLC analysis of total
O-glycan pools from sCD8␣␣K1, sCD8␣␤K1, and sCD8␤␤K1 and from sCD8␣␣Lec and sCD8␣␤Lec. Digestion with bovine testes ␤-galactosidase (not shown) confirmed the identity of peak O1 as Gal␤1-3GalNAc and the blank peak (labeled B) as lactose (Gal␤1-4Glc, an environmental contaminant). The other unlabeled peaks were not identified. GU, glucose units. b, exoglycosidase analysis of the total O-glycan pool from sCD8␣␣K1 (top panel in a) using A. ureafaciens sialidase (ABS). c, analysis of the total O-glycan pool from sCD8␣␣Lec (4th panel in a) using A. ureafaciens sialidase. Glycans were released from the proteins by automated hydrazinolysis. The digests were analyzed by NP-HPLC, and the structures derived are given in Table IV. fragmentation of the [M ϩ H] ϩ ions of tryptic peptides derived from sCD8␣␣Lec and sCD8␣␣K1 produced a cleavage profile consistent with that of a 15-amino acid C-terminal ␣-chain glycopeptide with the sequence APTPVPPPTGTPRPL (examples are given in Fig. 5). This peptide contains three of the four threonine residues in the sCD8 stalk likely to be sites of Oglycan attachment (6). The fragmentation pattern indicated that, when present, the hexoses are attached to the HexNAc residues.
MALDI mass spectrometry (Fig. 6a) showed that in excess of 92% of the tryptic glycopeptides of sCD8␣␣Lec had all three sites occupied (m/z ϭ 2107.1, 2269.1, 2431.3, and 2593.3); 7% had two sites occupied (m/z 1903.2, 2066.1, and 2227.2), and 1% had one site occupied (m/z 1701.6 and 1863.6). The relative amounts of the fully occupied, sCD8␣␣Lec glycopeptide glycoforms are also apparent in the MALDI-MS profile of the whole tryptic digest. Relative to the amount of the glycoform consisting of single GalNAc residues at each of the three glycosylation sites (m/z ϭ 2107.1), glycoforms with one (m/z ϭ 2269.1), two (m/z ϭ 2431.3), and three (m/z ϭ 2593.3) additional hexose sugars were present at 107, 62, and 13%, respectively. By taking these ratios into account, and the results of the NP-HPLC analysis (Fig. 4, a and c; Table IV), ϳ70% of the Oglycans present on Lec3.2.8.1-derived CD8 consisted of single GalNAc residues; the remainder were extended with ␤1,3-galactose to form type 1 core Gal-␤1,3GalNAc, and only 18% of these (ϳ6% of the O-glycans overall) were sialylated.
Similar MALDI-MS analysis was performed on sCD8␣␣K1 (Fig. 6b). This showed that 100% of the available sites were occupied with GalNAc residues and that 96% of the glycopep-tides (m/z 2593.2) contained an additional hexose (galactose) residue at each site. From the HPLC analysis of the glycans (Table IV), 55% of the O-glycans containing galactose were monosialylated and 30% were disialylated. Thus, at least 82% of all the O-glycans present in the CHO K1-derived CD8␣␣ were sialylated (i.e. 85% of the 96% core 1 O-glycans detected by MALDI-MS).
Analytical Ultracentrifugation-sCD8␣␣K1, sCD8␣␣Lec, and sCD8␣␣E have apparent sedimentation coefficients at 293 K (s 20,w 0 ) that vary inversely with concentration, presumably due to the limiting effects of macromolecular crowding on diffusion (Fig. 7) (Fig. 4a, top panel) and sCD8␣␣Lec (Fig. 4c, top panel) were subjected to exoglycosidase sequencing as described under "Experimental Procedures" to derive the structures shown. The relative abundance was calculated from the fluorescence intensities of the peaks from the NP-HPLC analyses. The scheme for glycan representation is as follows: }, N-acetylgalactosamine; {, galactose; *, N-acetylneuraminic acid; dashed line, ␣-linkage; full line, ␤-linkage. To aid in the interpretation of these effects, we calculated the hydrodynamic properties of low resolution molecular models of each protein. The calculated s and P values for sCD8␣␣E, i.e. 2.61 and 1.14, respectively, modeled explicitly on the crystal structure of this protein (14) for ␦ ϭ 0.3 g/g, are very similar to the experimental values (i.e. 2.65 and 1.12) and indicate that in solution, as in crystals, this form of sCD8 lacking the stalk-like region is very compact (Table V; Fig. 8a). The sCD8␣␣Lec and sCD8␣␣K1 models were based on the crystal structure of the ligand-binding domain of murine CD8 (15), the sequence of the stalk-like regions, and the foregoing glycosylation analysis. The hydrodynamic properties of a model of sCD8␣␣K1 with a highly extended parallel stalk (i.e. ϳ2.6Å per residue, the likely upper limit for mucin-like polypeptides) are also in good agreement with the experimental values (i.e. s and P values of 4.05 and 1.53 at ␦ ϭ 0.3 g/g, versus 3.73 and 1.61, respectively; Fig.  8c). Crucially, a model of sCD8␣␣Lec with the same highly extended, parallel stalk also gives calculated s and P values that closely match the experimental values (3.23 and 1.53 at ␦ ϭ 0.3 g/g, versus 3.13 and 1.54, respectively; Fig. 8b).
To confirm that hydrodynamic modeling is sensitive to the overall structure of CD8, the predicted hydrodynamic properties of models of sCD8␣␣Lec, in which the stalk is absent to different degrees or is not fully extended, were determined (Fig.  8d). In combination, models 4 -6 in Fig. 8d approximate the hydrodynamic properties of an essentially unconstrained, flexible stalk. Along with the comparisons between the models of the unglycosylated and glycosylated forms of sCD8␣␣ (Fig. 8, a-c) this shows that, in combination, s and P are capable of discriminating between models of CD8 whose stalk-like regions have distinct conformational properties. DISCUSSION We have expressed proteins in glycosylation mutant and wild-type Chinese hamster ovary cells in order to circumvent the pitfalls inherent in analyzing the biological effects of glycosylation indirectly with inhibitors or by mutagenesis, or after incomplete enzymatic deglycosylation. We characterize the glycosylation of the two isoforms of recombinant sCD8 and the O-glycosylating capacity of the Lec3.2.8.1 cell line. We show that Lec3.2.8.1 cells are considerably restricted in forming the type 1 core disaccharide and profoundly deficient in O-glycan sialylation. We use these observations to characterize the effects of O-glycan sialylation on the structure of the stalk-like region of CD8. Our results suggest that relatively few of the shortest possible O-glycans profoundly affect the extension of the stalk-like region of this cell surface molecule and that sialylation has little or no additional effect.
The presence of the ␤-chain in sCD8 ␣␤, or its absence in ␣␣, might have been expected to give rise to different N-glycosylation patterns, particularly given the very low level of protein sequence conservation. Although differing slightly in proportion, the N-glycans of CHO K1 cell-derived sCD8 ␣␣ and ␣␤ were essentially of the same type, suggesting that the glycosylation sites are equally well exposed on both chains and accessible to the relevant processing enzymes. The N-glycan structures are restricted and consist of oligomannose and biantennary complex-type glycans. This glycan profile is presumably a consequence of the three-dimensional structure of the protein, as CHO K1 cells are known to be capable of processing tri-and tetra-antennary sialylated glycans on other glycoproteins, such as recombinant tissue plasminogen activator (50) and erythropoietin (51). Overall, however, our results argue strongly against the possibility that structure-based Nglycosylation differences are responsible for any functional differentiation of the ␣␣and ␣␤-isoforms of CD8.
The O-glycosylation in CHO cells is also very restricted and probably much simpler than for CD8 expressed on lymphocytes. In CHO K1 cells, the O-glycans consist of mono-and di-sialylated core 1 structures. Differences in O-glycosylation in the Lec3.2.8.1 cell line are likely due to the Lec2 and Lec8 mutations, which result in defective CMP-sialic acid and UDPgalactose translocation into the Golgi, respectively (43,44). Lectin-binding studies of the O-glycosylation of recombinant glycophorin A expressed in Lec8 cells (52) indicated that the glycans are truncated and non-sialylated, and this is confirmed here for sCD8 expressed in Lec3.2.8.1 cells. Most of the Olinked glycans were predicted to be restricted to single GalNAc residues, although leakiness of the Lec2 and Lec8 phenotypes may generate O-linked structures with galactose and/or sialic acid residues added to the GalNAc. 2 A 15-amino acid fragment, identified as the glycopeptide from the C terminus (position 133-147), was present as three distinct glycoforms, with the dominant species bearing a single GalNAc sugar at each of the three threonine residues. However, this glycopeptide is also present as glycoforms that contain an additional hexose residue at one or more of these sites, and mono-sialylated core 1 disaccharide was detected by sialidase digestion and HPLC analysis (Fig. 4c), confirming that the Lec3.2.8.1 mutant is indeed leaky.
Numerous studies have shown that, in leukocytes, O-glycosylation is very complex and varies in an activation-dependent and tissue-specific manner. For example, marked changes in core 2 branching and 6-GlcNAc transferase activities have been associated with T cell maturation, and a thymus-specific core 2-6GlcNAc-transferase has now been identified (53,54). Of most relevance to the present study, the sialylation of core 1 O-linked glycans is known to be up-regulated during thymocyte maturation (55,56). Changes in O-glycan processing of this 2 P. Stanley, personal communication.  (19) showed that CD8 binds MHC class I tetramers less avidly and that it becomes less effective as an adhesion molecule as sialylation increases during thymocyte maturation. They argue that changes in the flexibility or extension of the stalk-like region of CD8 may be critical for optimal ligand binding.
This proposal raises the question of whether or not the structural properties of CD8 can in fact be altered by sialylation. Previously, the structural effects of O-glycans had been thought to depend only on steric interactions between the peptide-linked GalNAc and the adjacent amino acids of the polypeptide (58 -60). This conclusion was based on the analysis of mucins but was not confirmed for cell surface proteins with much shorter stalk-like polypeptides, such as CD8. Comparison of the hydrodynamic data for sCD8␣␣K1 and sCD8␣␣Lec clearly indicates that O-glycans consisting of a single GalNAc are as effective as sialylated core 1 glycans in extending the stalk region of CD8. Therefore, our data generalize the concept that steric interactions between the peptide-linked GalNAc and the adjacent amino acids of the polypeptide account for the major structural effects of O-glycans, regardless of the length of the polypeptide.
The hydrodynamic modeling suggests that the degree of extension may be as great as 2.6 Å per residue. This is greater than for leukosialin (2 Å per residue (61)), comparable with that for bovine and porcine submaxillary gland mucins (2.5 Å per residue (59, 62)) but much less than the theoretical maximum (3.4 Å per residue). Remarkably, this apparent degree of extension of the CD8 stalk is achieved at half the O-glycan density of leukosialin (63), i.e. approximately one glycan for every six amino acids. However, it is necessary to acknowledge the weaknesses of hydrodynamic modeling, wherein a static model is substituted for a dynamic molecule and the imperfectly understood hydration of proteins has to be arbitrarily fixed. Whereas our data indicate that the degree of extension is essentially indistinguishable for sCD8␣␣K1 and sCD8␣␣Lec, systematic errors could confound a more quantitative analysis.
How  a M w is molar mass calculated from the known amino acid composition of sCD8␣␣ dimer and the glycosylation analysis reported herein. is the partial specific volume, and ␦ is the hydration in g of water/g of protein. These were calculated using the same data as for the M w . The s 0 20,w (0) is the sedimentation coefficient extrapolated to infinite dilution, and the ␦ in the second instance of this term refers to the correction to s 0 20,w (0) for the calculated hydration level of the protein. s(c) is the dependence of the sedimentation coefficient on concentration. P exp is the Perrin function, the frictional ratio corrected for hydration (in this case for the theoretical ␦ values listed here). The units of s are Svedbergs (S).
FIG. 8. Molecular models of sCD8␣␣ used to simulate hydrodynamic parameters, and the results of the simulations. a, the model of sCD8␣␣E is based on the crystal structure of human sCD8␣␣ from the sCD8␣␣Ϫclass I MHCp complex (14). The sCD8␣␣Lec (b) and sCD8␣␣K1 (c) structures are based on the crystal structure of murine sCD8␣␣ from the sCD8␣␣Ϫclass I MHCp complex (15), the amino acid sequence of the stalks, and the glycan analysis described herein. An extension of 2.6 Å per residue was used to model the stalk region. s and P are the calculated sedimentation coefficients and Perrin functions derived for the models (see text for details of the calculation), whereas s exp and P exp are the experimentally determined sedimentation coefficients and Perrin functions corrected for hydration using a value of ␦ ϭ 0.3 g/g (see Table V). In a, b, and c, the protein (purple) and N-and O-glycans (small red and dark blue spheres) are shown as "bead" models (assemblages of larger, transparent spheres). d, to demonstrate the extent to which the modeling is sensitive to the structure of CD8, the hydrodynamic properties are shown for models of sCD8␣␣Lec (shown schematically) in which the stalk is fully extended as in b (model 1), absent to different degrees (i.e. by 50 and 80% in models 2 and 3) or not fully extended (models 4 -6).
for MHC class I molecules is extremely low (11,15) and much weaker than for the TCR⅐MHCp. Interactions that are this weak are very likely to be sensitive to avidity effects (i.e. density-dependent binding effects), which can occur in the absence of structural changes at the level of individual molecules. These effects are distinct from affinity changes, which are structure-dependent. Given that neither group directly tested the effect of CD8 sialylation on the monovalent binding affinity, in the light of our data, we suggest that the binding changes observed by Moody et al. (18) and Daniels et al. (19) are more likely the result of avidity changes. The simplest explanation is that de-or unsialylated CD8 tends to aggregate, increasing the likelihood of observable tetramer binding, for example. The differential sialylation of CD8 during thymopoiesis may simply occur coincidentally along with that of other glycoproteins for which sialylated O-glycans are of greater functional significance. The dominant thymocyte sialoglycoprotein is not CD8 but CD43 (64) after all. It could be argued that our results are not representative of the behavior of CD8 in vivo, where the ␣␤-isoform predominates, given that we have characterized the solution properties of the ␣␣-isoform. However, in all species, the analogous region of the ␤-chain of CD8 is also rich in proline and threonine residues and is likely to have a very similar structure to that of the ␣-chain. Therefore, although we cannot rule out the possibility that, in contrast to the ␣-chain, the structure of the ␤-chain is sensitive to sialylation, there is at present no obvious structural basis for suspecting that this is the case. More generally, as has been noted (65), because many proteins are affected by manipulations of the type employed by Moody et al. (18) and Daniels et al. (19) and glycosylation can affect proteins in several ways, it will often be difficult to establish that the glycosylation of a particular protein is important.
Finally, we note that the O-glycosylated region of the ␤-chain is significantly shorter than that of the ␣-chain (by nine residues in mouse CD8, measured to the inter-chain disulfide). Given the same degree of extension implied by the present hydrodynamic data, the ␤-chain is likely to cause the ␣-chain to arch, favoring a docking interaction with class I MHCp parallel with the cell surface, comparable with that seen in the crystal structures (14,15). It is now accepted that the co-receptors and the TCR each have to bind the same MHCp molecule (2). If this is initiated by TCR binding to MHCp, as seems likely given the higher affinity of the TCR interaction (66), both co-receptors will be required to dock precisely to a binding site fixed ϳ150 Å from the surface of the T cell. The question remains as to why CD4 and CD8 have evolved such different solutions to this docking problem. It seems relevant that cytotoxic CD8 ϩ T cells are required to recognize their targets largely in the absence of interactions involving other similarly sized adhesive and costimulatory molecules, such as CD2 and CD28, that facilitate CD4 ϩ T cell contacts with their targets. The orientating effects of the stalk, coupled with its inherent flexibility, may overcome this limitation and facilitate CD8-class I MHCp interactions during the very earliest stages of CD8 ϩ T cell activation and immunological synapse formation. In this, O-linked sugars would appear to be key.