The glycosylation of the complement regulatory protein, human erythrocyte CD59.

Human erythrocyte CD59 contains N- and O-glycans and a glycosylphosphatidylinositol (GPI) anchor, all of which have been analyzed in this study. The anchor consists principally of the minimum core glycan sequence Manα1-2Manα1-6Manα1-4GlcN-linked to a phosphatidylinositol moiety with the structure sn-1-O-alkyl(C18:0 and C18:1)-2-O-acyl(C20:4)glycerol-3-phospho-1-(2-O-palmitoyl(C16:0))myo-inositol. This structure is essentially identical to that of human erythrocyte cholinesterase (Deeg, M. A., Humphrey, D. R., Yang, S. H., Ferguson, T. R., Reinhold, V. N., and Rosenberry, T. L. (1992) J. Biol. Chem. 267, 18573-18580). This first comparison of GPI anchors from different proteins expressed in the same tissue suggests that human reticulocytes produce only one type of anchor structure. The N- and O-glycans were sequenced using a novel approach involving digestion of the total glycan pool with multiple enzyme arrays. The N-glycan pool contained families of bi-antennary complex-type structures with and without lactosamine extensions and outer arm fucose residues. The predominant O-glycans were NeuNAcα2-3Galβ1-3GalNAc and Galβ1-3[NeuNAcα2-3]GalNAc. Inspection of a molecular model of CD59, based on the NMR solution structure of the extracellular domain and the structural data from this study, suggested several roles for the glycans, including spacing and orienting CD59 on the cell surface and protecting the molecule from proteases. This work completes the initial structural analysis of CD59, providing the most complete view of any cell surface glycoprotein studied to date.

been proposed that CD59 mediates T cell adhesive interactions by synergizing with CD58 via direct interactions with CD2 (4,5) although this is controversial (6,7). It has also been proposed that CD59 participates in T cell activation pathways and platelet secretory responses (1).
CD59 is normally attached to the cell surface via a glycosylphosphatidylinositol (GPI) 1 anchor (8) although a number of soluble forms have been found, for example in saliva, amniotic fluid, milk, and urine (1). The complementary DNA sequence of the CD59 gene (8) has shown that the translated precursor contains 128 amino acids, including a 25-amino acid N-terminal signal peptide and a C-terminal sequence that contains a cluster of 18 mainly hydrophobic amino acids preceded by eight polar amino acids. This sequence is typical of signal peptides for GPI anchored proteins (9). The cleavage site for the anchor attachment is between Asn 77 and Gly 78 (10). The molecule contains eight potential O-glycosylation sites at Thr 10 , Thr 15 , Thr 29 , Thr 51 , Thr 52 , and Thr 60 , and at Ser 20 and Ser 21 . There is one fully occupied N-glycosylation site (Asn 18 -Cys-Ser) (8) which is completely conserved in all known CD59 sequences except rat, which is glycosylated at the adjacent residue, Asn 16 (11). The role of the glycans at this site in CD59 is controversial. Removal or modification of the N-glycans has been shown to reduce the co-stimulation of proliferation by CD59 and to eliminate its complement inhibitory properties (12,13), whereas other studies have shown that this enhances or has no effect on CD59 function (14 -16).
NMR analyses of recombinant (17) and urine-derived (18) glycosylated forms of soluble human CD59 revealed that the extra-cellular region consists of a single disk-like domain, formed by single two-and three-stranded beta sheets and a * 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.
short alpha helix, that is attached to the GPI anchor by a short 7-residue stalk. A recent mutational analysis 2 has shown that the active site is centered on Trp 40 of the exposed glycosylated face of CD59 and that this is equivalent to the "concave" face used by neurotoxins to bind to target nicotinic and muscarinic acetylcholine receptors and acetylcholinesterases.
GPI membrane anchors are used by a wide variety of cell surface glycoproteins (reviewed in Refs. 1, 19 -22). All GPI anchors characterized to date contain a conserved core of ethanolamine-HPO 4 -6Man␣1-2Man␣1-6Man␣1-4GlcN␣1-6myo-inositol-1HPO 4 -lipid where the lipid can be a diacylglycerol, an alkylacylglycerol, or a ceramide (22). In some cases, the inositol is substituted with a fatty acid (palmitate), rendering the structure resistant to the action of bacterial PI-specific phospholipase C (PIPLC) (23,24). The core glycan can be substituted with other sugars, and all GPI anchors from metazoan sources contain one or two additional ethanolamine phosphate groups attached to the conserved trimannosyl backbone (22).
In this study, the glycan component of the GPI anchor of human erythrocyte CD59 was characterized, enabling its structure to be compared with that of the glycan anchor attached to another human erythrocyte GPI-anchored glycoprotein, acetylcholinesterase (25,26). Further, the N-and O-linked glycans covalently attached to human erythrocyte CD59 were characterized, and the full HPLC glycosylation profile compared with that of human platelet CD59 to explore the extent of cellspecific glycosylation. Molecular modeling, based on the solution structure of the protein (18) was used to explore the spatial relationships between the lipid bilayer, the protein, and the N-and O-linked glycans. Several possible roles for the heterogeneous array of glycans attached to CD59 were proposed, including orientation of the active site, protection from protease cleavage, and limiting interactions with other CD59 molecules and with the cell surface. This work provides the most complete view of any cell surface glycoprotein studied to date.

EXPERIMENTAL PROCEDURES
Purification of Human Erythrocyte and Platelet CD59 -CD59 was purified from human erythrocytes and platelets by immunoaffinity chromatography of butanol-extracted erythrocyte membranes essentially as described previously (27,28). The purity of the proteins was confirmed by analysis on SDS-PAGE (15% gels) and by N-terminal amino acid sequencing. Protein concentration was measured using the Pierce protein assay reagent.
Nitrous Acid Deamination and Sodium Borohydride Reduction-CD59 (200 g) was dissolved in 10 l of pH 4.0 buffer (0.1 M sodium acetate adjusted to pH 4 with acetic acid). A freshly prepared solution of NaNO 2 (10 l, 0.5 M) was added, and the mixture was incubated for 2.5 h at room temperature after which boric acid (5 l, 0.8 M) was added. The amount of NaOH required to adjust the pH to about 10.5 (determined using buffer controls and typically 8 -10 l) was added to the reaction mixture. NaB 3 H 4 (5 l, 10 -15 Ci/mmol, 36 mM in 0.1 M NaOH) was added immediately, followed 2 h later by freshly prepared NaBH 4 (1 M aqueous) to give a final concentration of 0.2 M NaBD 4 . After a further 2 h, the reduction mixture was acidified with acetic acid (1 M). The reaction was carried out in the fume hood as tritium gas was evolved. The sample was dialyzed against water using a Spectra/Por membrane (Orme Scientific) with a 2,000 Da cut off and was freeze dried.
Aqueous HF Dephosphorylation and Re-N-acetylation-Deaminated and reduced CD59 (40 g), prepared as described above, was placed in a screw top Eppendorf tube and incubated with 50 l of 48% aqueous HF at 0°C for 48 h. The aqueous HF was neutralized with saturated LiOH and the products re-N-acetylated (29).
Dionex HPLC-The system was fitted with a Dionex CarboPac PA-1 column. A calibration standard of mixed glucose oligomers was coinjected with the sample and detected by pulsed amperometric detection. The elution positions of the neutral glycans were defined in Dionex units (Du) by linear interpolation of their elution positions between two adjacent internal standards. These Du have no specific meaning but are a reliable chromatographic property that correlate with particular glycan structures. The radiolabeled glycans were detected on line by a Raytest Romana radioactive monitor and by scintillation counting of collected fractions.
Bio-Gel P4 Gel Filtration-Filtration was performed using the Oxford GlycoSystems GlycoMap TM Instrument, which is based on a BioGel P4 gel permeation chromatography system (GPC).
Electrospray Mass Spectrometry (ES-MS)-A sample of CD59 (200 g) in 0.5 ml of phosphate-buffered saline, 0.05% CHAPS, 0.02% sodium azide was extracted four times with an equal volume of butan-1-ol saturated with water to remove the CHAPS detergent and any noncovalently associated lipids. The aqueous phase was freeze dried and redissolved in 100 l of 0.3 M sodium acetate buffer, pH 4.0). The sample was deaminated by addition of 100 l of freshly prepared 0.5 M NaNO 2 , sonication for 15 min, and incubation for 3 h at 37°C. The PI fraction, released by the deamination procedure, was recovered by two extractions with 200 l and one extraction with 100 l of butan-1-ol saturated with water. The combined butan-1-ol phases were washed with 300 l of water, dried under a stream of N 2 and redissolved in 100 l of chloroform/methanol (2:3) containing 1 mM NH 3 . Aliquots (20 l) were introduced into the electrospray source of a Micromass (Altringham, Cheshire, UK) mass spectrometer in a stream of the same solvent at 10 l/min. Negative ion mass spectra were recorded over the mass range m/z 700 -1500 and processed using MassLynx software. The source conditions were optimized using soybean PI from Sigma. For tandem mass spectrometry, mass resolution in mass spectrometer 1 was reduced to allow the simultaneous transmission of m/z 1135 and m/z 1137 into a hexapole collision cell containing argon (2.5 ϫ 10 Ϫ3 Torr) under an accelerating voltage of 60 V. Daughter ions were collected over the mass range m/z 200 -450.
Release and Re-N-acetylation of Glycans-0.6 g (31.6 nmol) of CD59 was dialyzed against 0.1% trifluoroacetic acid and lyophilized. Glycans were released from CD59 by hydrazine at 95°C and re-N-acetylated using a GlycoPrep 1000 (Oxford GlycoSystems, Ltd.) optimized for maximum recovery (approximately 85%) of both N-and O-linked sugars. The recovery was assessed by desialylating and degalactosylating bi-antennary complex sugars and then incubating with UDP-[ 3 H]Gal and galactosyl transferase. The free glycans associated with the anchor were also released under these conditions. However, since these glycans terminate in inositol, they cannot be reduced or 2-aminobenzamidelabeled and are therefore not detected in the oligosaccharide analyses.
Matrix-assisted Laser Desorption Ionization Mass Spectrometry (MALDI MS)-MALDI data were acquired on a Micromass (Altringham, Cheshire, UK) AutoSpec-FPDQ magnetic sector instrument fitted with a pulsed nitrogen laser (337 nm) and an array detector (33). Samples were prepared by adding the oligosaccharide sample (1 l) in water to the matrix solution (3 l of a saturated solution of 2,5-dihydroxybenzoic acid in acetonitrile) on the mass spectrometer target and allowing it to dry at room temperature. The mixture was then recrystallized from 1 l of ethanol (34). The array detector was set to the high resolution position, and the mass range was set to be appropriate to the sample being examined. For data acquisition, the laser was operated at full power, and the laser beam was moved manually over the sample in order to compensate for sample depletion under the laser beam.
N-terminal Sequence Analysis-Purified CD59 (10 g) was run under reducing conditions on a 15% Tris/glycine gel-PAGE (35) and blotted onto Problott TM membrane (Applied Biosystems, Warrington, UK). The blot was stained with Coomassie Brilliant Blue, the CD59 band was excised, and the protein was reduced and alkylated with dithiothreitol and iodoacetamide. For enzymatic digestion 60 g of CD59 was precipitated with ice-cold acetone, resuspended in phosphate-buffered saline, 2 mM EDTA, 2 mM dithiothreitol, and boiled for 5 min. 10 mM iodoacetamide and 3 g of endoproteinase Lys C (Promega, Southampton, UK) were added. Digestion was carried out for 16 h at 37°C. Fragments were separated under reducing conditions on a 15% Tris/Tricine gel (36) and blotted onto Problott TM membrane. The blot was stained with Amido black, and bands were excised. N-terminal sequence analysis of the excised bands was performed using an Applied Biosystems 476A protein sequencer and 610A data analysis software.
Molecular Modeling-Molecular modeling was performed on a Silicon Graphics 4D/35 workstation using the program Insight II and Discover (Biosym Tech. Inc.). Coordinates for the CD59 solution structure were retrieved from the Protein Data Bank, Brookhaven National Laboratory (37).

RESULTS
CD59 was purified from butanol-extracted erythrocyte membranes by affinity chromatography and gel-filtration as described previously (27,28). On SDS-PAGE, the purified protein migrated as a broad band in the apparent molecular mass range 20 -25 kDa. No other bands were detected even on overloaded gels. N-terminal sequencing indicated that no N-terminal residues other than that of CD59 (leucine) were detectable (data not shown). Together these data indicate that the preparation was at least 95% pure.
The strategy for the subsequent analysis of the GPI anchor and the N-and O-glycans attached to CD59 is outlined in Fig.  1A, and the results of the study are correspondingly organized into sections A-H below. Fig. 1B shows a sialylated bi-anten-nary complex glycan with lactosamine extensions and some of the cleavage positions of the enzymes used in this study.

GC-MS Composition Analysis of All N-and O-linked
CD59 Sugars (Fig. 1A) The monosaccharide composition of the total CD59 oligosaccharides, following methanolysis, re-N-acetylation, and trimethylsilylation, was determined by GC-MS (29) (Table I)  mannose (1.6 nmol) suggested that CD59 was associated with a high percentage of multi-antennary-and polylactosamine-type oligosaccharides. Third, the proportion of fucose (2.5 nmol) suggested that many structures contained outer arm as well as core fucose. Finally, the detection of 3 nmol of sialic acid residues indicated that many of the glycans were charged.

Characterization of the PI Moiety of the GPI Anchor
( Fig. 1C) The PI moiety released by nitrous acid deamination of CD59 was recovered by extraction into butan-1-ol and analyzed by FIG. 6. MALDI-MS analyses of glycans from HuE CD59. a, asialo glycans; b, asialo glycans following digestion with C. lampas fucosidase; c, asialo glycans following digestion with C. lampas fucosidase and endo ␤-galactosidase; and d, asialo glycans following digestion with C. lampas fucosidase and endo ␤-galactosidase and bovine testes galactosidase. The data are analyzed in Table II.   . 3a). Simultaneous collision induced dissociation of the two ions resulted in a daughter ion spectrum (Fig. 3b)    The glycan and PI data described here are summarized in Fig. 4, which is a schematic representation of the GPI anchor of CD59. The inclusion of the ethanolamine groups in this model is solely by analogy with other mammalian GPI anchors such as rat brain Thy-1 antigen (40), human erythrocyte acetylcholinesterase (26), and human leukocyte CD52 (24).

Comparison of the Glycosylation Profiles of Human
Erythrocyte and Human Platelet CD59 (Fig. 1D) The 2AB-labeled sialylated glycan pools released from human erythrocyte and human platelet CD59 were resolved by normal phase HPLC ( Fig. 5a and b). The differences in these profiles, which compare the overall glycosylation of CD59 expressed in two different cell types, confirm previous findings that glycosylation is cell type-specific (41). There are more than 40 peaks in each of these profiles indicating that platelet and erythrocyte CD59 are both expressed as a population of many different glycoforms. The bar chart below the figure indicates the elution positions of standard 2AB-labeled oligosaccharides. Using this as a calibration, it can be predicted that both platelet and erythrocyte CD59 contain bi-and multi-antennary structures and glycans with lactosamine extensions. Both contain sugars, for example peaks A and B, in the region of the FIG. 7. a-f, HPLC profiles of the N-glycan population of HuE CD59 simultaneously digested with a series of enzyme arrays. The HPLC analysis of the total glycan pool (a) and the products resulting from the digestion of five aliquots of the total CD59 glycan pool with a series of enzyme arrays (b-f). The shaded areas define the peaks that contain glycans which were subsequently digested by the additional enzyme present in the next array. The glucose unit value of each peak was calculated (Table III)  Characterization of the N-Glycans Attached to HuE CD59 (Fig. 1E) MALDI-MS Analysis of CD59 N-linked Asialo Glycan Pool (Fig. 1E.1)-MALDI-MS of the desialylated N-glycan pool (Fig.  6a, Table II) indicated that asialo CD59 contained at least 123 glycan structures associated with the one N-glycosylation site. These included families with compositions consistent with those of bi-, tri-, and tetra-antennary sugars with and without polylactosamine extensions and outer arm fucose. Since MALDI-MS determines only molecular masses, this type of analysis does not give information with respect to isomers which arise from differences in linkage and arm specificity. These additional variations, as well as diversity arising from the addition of sialic acid residues to many of the neutral oligosaccharides, may be expected to increase the number of structures even further. The maximum size of the sugars could not be determined because their relative abundance decreased with increasing molecular weight and eventually fell below the detection limit of the MALDI experiment. The largest structures detected contained approximately 24 monosaccharide residues.
Normal Phase HPLC Analysis of CD59 N-linked Glycan Pool (Fig. 1E.2)-The sialylated N-glycan pool from HuE CD59 was resolved into at least 36 peaks by normal phase HPLC (Fig. 7a). Each peak was numbered and assigned an HPLC glucose unit

) and HPLC analysis of the products of a series of five simultaneous exoglycosidase digestions of aliquots of the total pool of CD59 sugars (columns 4 -8)
Colume 4-8 refer to the products of the enzyme assay digests shown in Fig. 7, b-f respectively. The numbering of the peaks is shown in Fig. 7 a, denotes a shoulder on the main peak. value by comparison with the elution positions of a standard 2AB-labeled dextran hydrolysate mixture (shown at the top of the figure). Individual peaks were then assigned structures from their glucose unit values using the elution positions of standard glycans and pre-determined incremental values for monosaccharide residues (32) ( Table III). The data indicated that human erythrocyte CD59 contained families of bi-, tri-, or tetra-antennary complex glycans, which were both neutral and sialylated; neither hybrid nor oligomannnose structures were detected.
In view of the wide heterogeneity revealed by both the MALDI-MS and HPLC analyses of the CD59 N-glycans, no attempts were made to purify and define every individual oligosaccharide. Classically, enzyme arrays and P4 GPC have been used to sequence individual sugars purified to Ͼ80% homogeneity from the glycan pool. In the case of CD59, this would have required an extremely lengthy and difficult separations procedure. Therefore, a strategy was developed which involved the simultaneous sequencing of aliquots of the total glycan pool with enzyme arrays followed by HPLC analysis (Fig. 7). The glycan structures assigned by this novel strategy were confirmed as follows: (i) the sialylation status of the sugars was confirmed by WAX chromatography followed by normal phase HPLC; (ii) further studies with exoglycosidase enzymes confirmed the structures of the bi-antennary nonextended complex glycans and the oligosaccharides containing core and/or outer arm fucose; and (iii) glycans with lac-tosamine extensions were further characterized using endo-␤galactosidase.
Simultaneous Sequencing of the Total Glycan Pool Using Enzyme Arrays (Fig. 1E.3)-A preliminary assignment of structures to the N-glycans from HuE CD59 was made from the HPLC gu values (Fig. 7a and Table III). These assignments were confirmed by digesting aliquots of the total pool of CD59 glycans with five different enzyme arrays. The digestion mix- tures were resolved by normal phase HPLC (Fig. 7, b-f) ( Table  III, columns [4][5][6][7][8]. The assignment of each of the main peaks in the CD59 glycan pool was checked by following its predicted elution position through each of the enzyme digests. Predicted elution positions were based on predetermined incremental values for monosaccharide additions to standard glycan cores (32) and the known specificity of the enzymes (Fig. 1b).
The naturally neutral N-glycans (pool N) and de-sialylated N-glycans prepared from pools A1, A2, and A3 were analyzed by normal phase HPLC (Fig. 9). The monosialylated (A1) Nglycan pool contained bi-, tri-, and tetra-antennary sugars; the disialylated (A2) N-glycans were mainly of the bi-antennary type, while the trisialylated (A3) N-glycan fraction contained tri-and tetra-antennary glycans with and without lactosamine extensions. Table IV indicates the charge states which were associated with each glycan. N-glycans (Fig. 1E.5)-As a result of the above analyses, peaks 1-14, 16, and 17 in Fig. 7a were assigned to a series of neutral complex bi-antennary glycans. The major neutral glycan, A2G2FB, was isolated and its structure further confirmed by the reagent array analysis method (42) (data not shown). A comparison of the native and asialo HPLC profiles (Fig. 7, a and b) indicated that A2G2FB eluted in peak 14, while the mono-and di-sialylated forms of this sugar eluted in peaks 19 and 22, respectively. This was consistent with the WAX chromatograhy data (Table IV).

Analysis of the Major Glycan Population of Bi-antennary Non-extended Complex
Further Characterization of Oligosaccharides Containing Fucose (Fig. 1E.6)-The 2AB-labeled asialo glycan pool was incubated with C. lampas fucosidase, which removes both core fucose (GlcNAc␤1-4(Fuc␣1,6)GlcNAc) and outer arm fucose residues linked Fuc␣1-2Gal and Gal␤1-3(Fuc␣1-4 GlcNAc)3 Gal␤1-4(Fuc␣1-3)GlcNAc (43). MALDI-MS analysis of the digestion products indicated that most fucose residues were removed by the enzyme (Fig. 6b and Table 2, footnote c), but a number of structures, in particular the mono, di, and tri-fucosylated forms of H7N6, were resistant. These glycans contained fucose residues linked ␣1,3 to N-acetyl glucosamine in the outer arm and which were susceptible to almond meal fucosidase that is specific for this linkage (data not shown). These data were consistent with the results of the enzyme digests (Fig. 7, b and c).
Further Characterization of Multi-antennary Glycans and Glycans with Polylactosamine Extensions (Fig. 1E.7)-In the normal phase HPLC separation of the total glycan pool (Fig.  7a), peaks 18 and 20 -36 were assigned to multi-antennary and polylactosamine structures. The largest structure resolved by this system was a bi-antennary extended glycan with the composition formula H7N7 (peak 36). However, MS identified 75 additional structures which were larger than H7N7. Most of these contained more than one fucose residue, and 57 were present at 0.3% or less (Table 2). Interestingly, some of the compositions were consistent with those of multi-antennary glycans with outer arm branching such as have been detected on the lactosamine sugars released from erythrocyte band 3 protein (44).
The polylactosamine-type structures were investigated by digesting the asialo glycan pool with EBG after defucosylation with C. lampas fucosidase. The digestion mixture was analyzed using both MALDI-MS ( Fig. 6d and Table 2) and HPLC (Fig.  10b). In both cases, the major products of the EBG digestion were A2G0B (H3N5), A2G1B or A3G1 (H4N5), A4G2B (H5N7), A4G3B (H6N7), and A4G4B (H7N7). These structures are drawn in Fig. 10c. The glycan with the composition H7N7 has been shown with the fourth galactose attached to the bisecting GlcNAc since the alternative oligosaccharide with this composition is the fully galactosylated tetra-antennary structure, which is not a product of an EBG digest. There is no further evidence for the proposal that this structure contains galactose linked to the bisecting GlcNAc, but, interestingly, it has been noted (44) in the glycans attached to erythrocyte band 3 protein.
These data indicate that CD59 contains (i) bi-antennary glycans with lactosamine extensions on both arms, (ii) tetraantennary glycans with two and three lactosamine extensions, and (iii) many larger polylactosamine structures present at low levels which were digested by EBG to A2G1(1,3)B and/or A3G1 (H4N5). These latter structures are derived from tri-antennary glycans with extensions on two arms and/or bi-antennary glycans with bisecting GlcNAc and lactosamine extensions on one arm. Since no A2G1(␣1,6) glycans were detected by HPLC analysis after the EBG digestion, A2G1(1,3)B was derived from bi-antennary glycans extended on the ␣1,6 arm and containing only one (terminal) galactose on the ␣1,3 arm. This tendency for the 1,6 arm to be extended in preference to the 1,3 arm has been noted by Fukuda et al. (44) in band 3 protein from erythrocytes.
A2G1(1,3)B and A3G1 co-elute on this HPLC system and cannot be resolved by MS since they have the same composition (H4N5). The basic core structures were therefore determined by incubating the EBG-treated glycan pool with bovine testes galactosidase and analyzing the products by MS (Fig. 6d and Table 2). The major core structures were A2G0 (30%), A2G0B/ A3G0 (60%), and A3G0B (5%).

Characterization of the O-Glycans Attached to Human
Erythrocyte CD59 (Fig. 1F) The presence of O-linked sugars, not previously reported on CD59, was initially suggested by GC-MS composition analysis of the glycan pool (Table I). Consistent with this finding, the HPLC profile of the pool of sugars released from CD59 contained peaks eluting with glucose unit values corresponding to O-linked glycans (Fig. 5a). The two major peaks (A and B) were analyzed by enzyme digests coupled with normal phase HPLC (data not shown). After incubation with A. ureafaciens sialidase, both A and B were digested to one peak. This was assigned by comparison with known standards to Gal␤1, 3GalNAc. After incubation with an enzyme array containing both sialidase and bovine testes galactosidase, both A and B were digested to GalNAc. The elution positions of A and B and the products of the digests were consistent with the presence of two species of glycans, both containing the disaccharide Gal␤1,3GalNAc but with sialic acid-linked ␣2,3 to the Gal in one case (A) and ␣2,6 to the GalNAc in the other (B). FIG. 10. a-c, analysis of polylactosamine structures attached to  CD59. a, HPLC profile of the total pool of CD59 glycans after digestion with A. ureafaciens sialidase and C. lampas fucosidase. Glycans with lactosamine extensions elute from approximately 9 -12 gu. 20% of the glycans associated with CD59 are larger than this and elute later in a broad peak which is not resolved using this HPLC gradient. b, HPLC profile of CD59 glycans after incubation with A. ureafaciens sialidase, C. lampas fucosidase, and EBG. The major products of the digestion are in peaks 1-5. c, a schematic diagram showing the structures of the major digestion products. The structure with the composition H7N7 has been assigned tentatively to a tetra-antennary structure (A4G4B*) containing a galactose linked to the bisecting GlcNAc. (Fig. 1G) Since O-glycans attached to CD59 have not been reported previously, attempts were made to locate the glycosylation sites in the protein. Edman degradation of intact CD59 was carried out for 30 cycles. This covered all the Thr (10,13,29) and Ser (20,21) residues, which are potential O-glycosylation sites, in the first half of the CD59 molecule. The data that were obtained were identical to the published amino acid sequence (45,46). The repetitive yields for the five residues at potential sites for O-glycosylation were close to the predicted value (data not included). The previously published amino acid sequence of the second half of the molecule was derived solely from the cDNA sequence. No amino acid analysis had been carried out on this section of the molecule, with the exception of the last six amino acids which contain the GPI-anchoring site (47). Here protein enzymatic digestion was carried out with endoproteinase Lys-C, which specifically digests at the C terminus of lysine. As predicted, this approach yielded a 3-kDa fragment from amino acid 42 to amino acid 64 of the mature protein. This fragment was separated from partial digests and smaller fragments on a Tris/tricine gel, blotted, and sequenced. The amino acid sequence was identical to the cDNA derived sequence. All potential O-glycosylation sites were detected, and the repetitive yields for residues Thr 51 and Thr 52 were reduced, although not entirely absent (data not included). The level of the reduction was consistent with the finding that approximately 30% of the glycan pool consists of O-linked sugars. These findings, while not definitive, suggest that O-linked sugars may be variably attached to residues Thr 51 and Thr 52 . Further studies are being undertaken to confirm the presence of the O-linked gly-cans, and to determine the linkage positions since these structures have not been identified in earlier studies of CD59.

Molecular Modeling Studies of CD59 (Fig. 1H)
Molecular modeling studies using the x-ray crystallographic co-ordinates (48) and the enzyme binding site residues (49) indicate that PNGase F will only cleave N-glycosidic linkages when the sugar is in a conformation in which there are no interactions of the outer arm residues and the target protein.
In this study, CD59 was incubated with PNGase F to probe for possible interactions of the outer arm residues with the protein. Analysis of CD59 by SDS-PAGE showed that the protein migrated as a broad band with an apparent molecular mass of 20 -25 kDa (data not shown), consistent with the presence of a range of glycoforms. After incubation of native, undenatured CD59 with PNGase F, the protein migrated as a narrow band with an apparent molecular mass of approximately 15 kDa. This indicated that the glycoform population had been digested by the enzyme and that the Asn 18 -sugar amide linkage is fully accessible to PNGase F. These data are consistent with NMR analyses of glycosylated forms of CD59, which provided little evidence for protein-sugar interactions beyond the first N-acetylglucosamine residue of the N-linked glycan (17,18). Moreover, incorporation of model oligosaccharides in structure calculations (18) suggested that the N-glycan projects away from the protein in the plane of the disc-like extracellular domain.

DISCUSSION
The GPI-anchored glycoprotein, CD59, consists of a heterogeneous mixture of more than 120 glycoforms. The full signif- To illustrate the probable relationship between CD59 and the erythrocyte cell surface, the plasma membrane is depicted by two parallel red lines; the distance between these lines reflects the maximum dimensions of the bilayer based on the work of Hauser et al. (61). Views c and d, which are slightly enlarged with respect to views a and b, are orthogonal to the plane of the membrane which, for clarity, is not depicted. In view c, all residues are colored as in views a and b, whereas in view d, additional potential (exposed) O-glycosylation sites, rather than active site residues, are shown in purple and are labeled.
icance of glycosylation for the structure and function of CD59 can only be explored properly when its glycan population is viewed in the context of the complete structure of the molecule. In previous work, the three-dimensional structure of the protein component of CD59 (17,18) and the location of the active site of the molecule 2 have been determined. The aim of the present study was to determine the structure of both the GPI anchor and the N-and O-linked glycans of CD59 in order to complete the initial structural analysis and to provide a context for considering the function of the glycosylation of the molecule. With respect to its structure, CD59 is now the best defined cell surface molecule studied to date.
A Comparison of the GPI Anchors Attached to CD59 and Acetylcholinesterase Suggests That Structural Features of GPI Anchors May Be Cell-type Specific-This study has enabled the first comparison of GPI anchors expressed on two different glycoproteins in the same tissue, human erythrocytes. The majority (91 mol %) of the GPI anchors of CD59 were identical, in terms of PI and glycan structure, to that described for acetylcholinesterase (23,26). The identical PI moieties in particular, which contain alkyl chains and C22:4 fatty acid components that are not common in mammalian PI pools (50), suggest that, in reticulocytes, the two proteins receive the same GPI precursor in exchange for their different COOH-terminal GPI signal peptides. The small proportion of CD59 anchors subsequently modified by the addition of ␤-GalNAc suggests that at least some human reticulocytes possess the ␤-GalNAc transferase required for this relatively common GPI modification.

Analysis of the GPI Anchor and the N-and O-Glycans of Human Erythrocyte CD59 Indicates That Soluble Urinary CD59 (U-CD59) Does Not Originate From Erythrocytes-While
CD59 is normally attached to cell surfaces, a number of soluble forms have been found including urine (U-) CD59. Two studies (51,52) have shown that soluble U-CD59 contains a GPI anchor without any lipid moieties, consistent with a PIPLD cleavage product of PIPLC-sensitive material. In addition, there was extensive modification of the tri-mannosyl core with GalNAc (45 mol %). This is in contrast to the present finding that only 9 mol % of the trimannosyl core of the anchor of membrane bound erythrocyte CD59 was modified. Analysis of the N-linked glycans of U-CD59 also revealed significant differ-ences between urine and erythrocyte CD59 and, in addition, no O-glycans have been reported on U-CD59. Thus, an examination of the structures of the GPI anchor and the glycans suggests that little or no urinary CD59 originates from erythrocytes. An explanation for the apparent sequestration of erythrocyte CD59 from urine can be proposed on the basis of the new data. GPI-anchored proteins which appear in a soluble form in the urine may be the result of PIPLC-or PIPLD-type cleavages. The PIPLC resistance of HuE CD59 may prevent its appearance in urine even if it undergoes transfer between plasma membranes as has been suggested by several studies (53,54).
Human Erythrocyte CD59 Consists of at Least 120 Different Glycoforms-While the structure of the GPI anchor glycan was relatively homogeneous, the N-glycan population was extremely heterogeneous. However, despite the apparent heterogeneity of the N-glycans, the major population of glycoforms consisted of a family of structures based on bisected, corefucosylated, bi-antennary glycans carrying varying numbers of lactosamine extensions. A high proportion of the larger sugars were bisected bi-antennary glycans with polylactosamine extensions on one arm containing one or more fucose residues ␣1-3 linked to a GlcNAc residue.
Molecular Modeling Suggests Several Roles for the Glycans Attached to CD59 -The N-linked oligosaccharides (size range 3-6 nm in length) attached to the disc-like extracellular region of CD59 (diameter of approximately 3 nm) project away from the protein domain in the plane of the active face and adjacent to the membrane surface (Fig. 11). The glycans do not appear to restrict access to proposed active site residues of human CD59 (Asp 24 , Trp 40 , Arg 53 , and Glu 56 ) located on the membrane distal surface of the extracellular domain 2 . However, the glycans would be expected to restrict the rotational freedom of the extracellular domain around axes parallel to the membrane that may, in turn, stabilize an exposed location for the active face (55). Removal of the conserved N-linked glycan might therefore reduce the affinity of CD59 for the membrane attack complex without eliminating it completely. The effects of removing the N-linked glycan might therefore be expected to depend on the density of expression of the glycoprotein at the cell surface, and this may explain the observed variation in the Three different glycoforms are shown. Both the O-and N-linked oligosaccharides (size range 2-6 nM) attached to CD59 (diameter of approximately 3 nM) restrict the conformational space available to the protein and limit its interaction with the lipid bilayer. The sugars may therefore orient the active site of CD59 toward the C5b-9 complex, which is also inserted into the cell membrane. In addition, the heterogeneity of the sugars suggests that the glycans influence the geometry of the packing, and it is unlikely that CD59 molecules will form a regular array on the cell surface. The active site residues are highlighted in CPK. activities of unglycosylated CD59 2, 3 (13,45). A discrepancy between the measured two-dimensional affinity of the cell surface recognition molecules CD2 and LFA-3 and of the twodimensional affinity predicted on the basis of the measured three dimensional affinity (6) suggests that the mobility of the CD2 ligand binding site is restricted perpendicular to the cell surface (56). It seems likely that N-glycans attached to the highly conserved, membrane proximal glycosylation site at the base of CD2 may contribute to this effect. This suggests that restricting the conformational space available to cell surface glycoproteins may be a key function of N-glycosylation.
Molecular modeling of the GPI-anchored glycoprotein Thy-1 (57) suggested that the glycoprotein sits directly on the membrane and that there are extensive interactions between the protein and the GPI glycan. In contrast to Thy-1, where the anchor is attached to a cysteine residue involved in an intramolecular disulfide bond (40), CD59 contains a hydrophilic C-terminal sequence linking the protein to the GPI anchor. The protein component of CD59 would therefore have considerable dynamic freedom relative to the membrane, and it seems likely that there would be very few non-covalent interactions between the protein and the anchor. This view is supported by NMR data from the analysis of the soluble form of U-CD59, which shows that the last few hydrophilic residues have a poorly defined structure. The bulky, hydrophilic glycans in the anchored form of CD59 would be expected to limit interactions with the lipid bilayer and in this way facilitate diffusion of the glycosylated protein in the membrane. Fig. 12 shows three differently glycosylated CD59 molecules inserted into the cell membrane. The heterogeneity of the sugars suggests that the glycans influence the geometry of the packing, and it is likely that they will also prevent the aggregation of CD59 molecules on the cell surface. By limiting such protein-protein interactions, the glycans could influence the distribution of CD59 molecules at the cell surface where GPI anchored proteins may associate in micro-domains in dynamic equilibrium with isolated individual molecules. The large Nglycans may also be important in preventing proteolysis of the extracellular domain since N-glycosylation has been shown to increase the dynamic stability of a protein while different glycoforms variably increase its resistance to protease digestion (55).
Human Erythrocyte CD59 Contains Potential O-Glycosylation Sites-A population of sialylated O-glycans was recovered from human erythrocyte CD59; the major species that were identified were two mono-sialylated forms of the disaccharide Gal␤1,3GalNAc. Interestingly, asialo human erythrocyte CD59 failed to bind peanut agglutinin (13), a lectin which is specific for Gal␤1,3GalNAc. This suggests that the O-linked Gal␤1, 3GalNAc glycans detected on CD59 may not be accessible to the lectin. In view of this result (13), and the fact that O-glycans have not been identified previously on CD59, further studies are being undertaken to confirm the presence of the O-linked sugars and to determine their attachment sites. In the present study, attempts to locate the O-glycosylation sites on the basis of signal supression during protein sequencing suggested that the O-glycans may partially occupy Thr 51 and Thr 52 .
In support of this, inspection of the molecular model suggested that several potential sites are accessible. Two structures are available, both derived from NMR solution studies (17,18). Rotation and inspection of both structures indicates that Ser 20 , which forms part of the N-glycosylation sequon, may not be readily accessible once the N-linked sugar is attached at Asn 18 . However, the side chain of the adjacent resi-due, Ser 21 , presents a feasible site for O-glycosylation. Thr 15 is also close to Asn 18 , but is accessible, as are the side chains of Thr 52 and Thr 60 . Thr 29 is inaccessible. In the structure determined by Fletcher et al. (18), the side chain of Thr 51 is also accessible (Fig. 11d). Interestingly, in the structure determined by Kieffer et al. (17), a small difference in the conformation of the helix to which Thr 51 is attached causes the side chain to be differently oriented so that, in this case, it is inaccessible. Thr 10 is partially hindered in both models. As a result, it may be concluded that potential sites at Ser 20 , Thr 10 , and Thr 29 are unlikely to be O-glycosylated, while both Thr 51 and Thr 52 are situated on a helix and sufficiently well exposed to be O-glycosylated. These residues are also adjacent to active site residues (Arg 53 and Glu 56 ) identified by site-directed mutagenesis 2 (Fig.  11). This raises the possibility that the O-glycans may form part of the active site of CD59. Soluble, recombinant CD59 expressed in insect cells is active in complement inhibition assays (47) but the extent of O-glycosylation of this form of the molecule has not been determined. Conversely, Chinese hamster ovary cell-derived soluble CD59 (17) is not O-glycosylated, 4 but it is not clear if protein from this source is active.