Structures of the glycosyl-phosphatidylinositol anchors of porcine and human renal membrane dipeptidase. Comprehensive structural studies on the porcine anchor and interspecies comparison of the glycan core structures.

The glycan core structures of the glycosyl-phosphatidylinositol (GPI) anchors on porcine and human renal membrane dipeptidase (EC 3.4.13.19) were determined following deamination and reduction by a combination of liquid chromatography, exoglycosidase digestions, and methylation analysis. The glycan core was found to exhibit microheterogeneity with three structures observed for the porcine GPI anchor: Man alpha 1-2Man alpha 1-6Man alpha 1-4GlcN (29% of the total population), Man alpha 1-2Man alpha 1-6(GalNAc beta 1-4)Man alpha 1-4GlcN (33%), and Man alpha 1-2Man alpha 1-6(Gal beta 1-3GalNAc beta 1-4)Man alpha 1-4GlcN (38%). The same glycan core structures were also found in the human anchor but in slightly different proportions (25, 52, and 17%, respectively). Additionally, a small amount (6%) of the second structure with an extra mannose alpha (1-2)-linked to the non-reducing terminal mannose was also observed in the human membrane dipeptidase GPI anchor. A small proportion (maximally 9%) of the porcine GPI anchor structures was found to contain sialic acid, probably linked to the GalNAc residue. The porcine GPI anchor was found to contain 2.5 mol of ethanolamine/mol of anchor. Negative-ion electrospray-mass spectrometry revealed the presence of exclusively diacyl-phosphatidylinositol (predominantly distearoyl-phosphatidylinositol with a minor amount of stearoyl-palmitoyl-phosphatidylinositol) in the porcine membrane dipeptidase anchor. Porcine membrane dipeptidase was digested with trypsin and the C-terminal peptide attached to the GPI anchor isolated by removal of the other tryptic peptides on anhydrotrypsin-Sepharose. The sequence of this peptide was determined as Thr-Asn-Tyr-Gly-Tyr-Ser, thereby identifying the site of attachment of the GPI anchor as Ser368. This work represents a comprehensive study of the GPI anchor structure of porcine membrane dipeptidase and the first interspecies comparison of mammalian GPI anchor structures on the same protein.

Glycosyl-phosphatidylinositol (GPI) 1 membrane anchors are present in organisms at most stages of eukaryotic evolution, including protozoa, yeast, slime molds, invertebrates, and vertebrates, and are found on a diverse range of proteins. They are primarily responsible for the anchoring of cell-surface proteins in the plasma membrane and may be considered as an alternative to the hydrophobic transmembrane polypeptide anchor of integral membrane proteins. Many other functions have been proposed for GPI anchors, including roles in intracellular sorting, transmembrane signaling, and the novel endocytic process of potocytosis. A GPI anchor might also allow the protein to associate in membrane microdomains or be selectively released from the cell-surface by phospholipases. These functions, as well as the structure, biosynthesis, and distribution of GPI anchors have been extensively reviewed Cross, 1990;Thomas et al., 1990;Ferguson, 1991;Ferguson 1992a;Englund, 1993;McConville and Ferguson, 1993).
From these studies it is apparent that the GPI anchor consists of a highly conserved core structure of ethanolamine-PO 4 -6Man␣1-2Man␣1-6Man␣1-4GlcNH 2 ␣1-6myo-inositol-1-PO 4 -lipid. Often attached to this conserved core are variable side chains which may be protein-, tissue-, and/or speciesspecific . Examples include an ␣-galactose branch on some T. brucei VSG molecules and additional ␣-mannose residue(s) on a number of structures (including yeast, slime mold and mammalian glycoproteins, and T. cruzi 1G7 antigen). A single sialic acid (N-acetylneuraminic acid, NANA) residue is present on a proportion (30%) of the scrapie prion protein anchors (Stahl et al., 1992) while an average of five sialic acid residues are found on the T. brucei PARP anchor . In contrast to lower eukaryotic GPI anchor structures, all metazoan GPI structures studied to date contain at least one additional ethanolamine phosphate residue, although the exact position(s) and linkages have only been determined in rat brain Thy-1 and human erythrocyte acetylcholinesterase Roberts et al., 1988a;Deeg et al., 1992). The function, if any, of side chain modifications in general remains obscure, although the ␣-galactose branch in VSG has been suggested to be involved in the dense packing of the protective surface coat of the trypanosome (Homans et al., 1989).
From the lipid moieties that have so far been determined it appears that this part of the anchor structure can be quite variable. They range from ceramide in most yeast glycoproteins (Fankhauser et al., 1993) and slime mold pre-spore antigen (Haynes et al., 1993), to sn-1-alkyl-2-acylglycerols in L. major promastigote surface protease (Schneider et al., 1990), T. cruzi 1G7 antigen (Heise et al., 1995), human erythrocyte acetylcholinesterase (Roberts et al., 1988a(Roberts et al., , 1988b, human placental alkaline phosphatase (Redman et al., 1994), and several other mammalian GPI-anchored proteins (reviewed in McConville and Ferguson, 1993), sn-1,2-diacylglycerols in T. brucei VSG (Ferguson et al., 1985), Torpedo acetylcholinesterase (Butikofer et al., 1990) and human CD52 (Treumann et al., 1995), and an unusual sn-1-acyl-2-lyso-glycerol in T. brucei PARP (Field et al., 1991). In addition, several GPI anchors contain an addi-FIG. 1. Structures of known mammalian GPI anchors. The conserved core region contains an extra ethanolamine phosphate (EtNP) in all mammalian structures. Various side chain modifications of carbohydrate, additional EtNP, and/or palmitate (R 1 -R 4 ) are found as indicated (see text for references except human and porcine membrane dipeptidase; this study). In some proteins certain residues may only be present in a proportion of GPI anchors, and this is indicated by Ϯ. OH indicates that no modification is thought to be present. R 5 ϭ lipid moiety present. n.d. ϭ not determined. tional fatty acid (palmitate) in ester linkage to the myo-inositol ring. Examples include human erythrocyte acetylcholinesterase (Roberts et al., 1988b), T. brucei PARP (Field et al., 1991;Ferguson, 1992c) and one form of human CD52 (Treumann et al., 1995). The latter study showed that the modification is at the 2-position of the myo-inositol ring.
The mRNAs of known GPI-anchored proteins encode an Nterminal signal sequence, to direct the protein to the endoplasmic reticulum, and a C-terminal GPI attachment signal sequence. This sequence is believed to hold the nascent protein in the membrane prior to anchor addition and is cleaved posttranslationally with the concomitant addition of the GPI anchor. The anchor attachment site () may be one of six amino acid residues, all of which have small side chains (Ala, Asn, Asp, Cys, Gly, or Ser) (Moran et al., 1991;Gerber et al., 1992). In addition, the residue at the ϩ 2 position is restricted to Ala, Gly, or Ser (Gerber et al., 1992).
Membrane dipeptidase (MDP, dehydropeptidase-I, renal dipeptidase, microsomal dipeptidase, EC 3.4.13.19) is a zincmetalloenzyme found predominantly in the brush-border membrane of the kidney and also in the lung (for review see Keynan et al., 1995). The enzyme is capable of hydrolyzing a wide variety of dipeptides (Campbell, 1970;Armstrong et al., 1974) may have a role in the metabolism of glutathione and leukotriene D 4 (Kozak and Tate, 1982;Campbell et al., 1990) and is the only known mammalian enzyme to exhibit ␤-lactamase activity (Kropp et al., 1982). Porcine MDP was the first mammalian peptidase that was found to be anchored via GPI (Hooper et al., 1987) and subsequently human MDP was also shown to be GPI-anchored (Hooper and Turner, 1988). The GPI anchor on both porcine and human MDP has been extensively characterized in terms of its hydrolysis by bacterial phosphatidylinositol-specific phospholipase C (PI-PLC), and a serum phospholipase D (Littlewood et al., 1989;Hooper and Turner, 1989;Hooper et al., 1990;Brewis et al., 1994) and polyclonal antisera have been generated against the cross-reacting determinant of both proteins (Hooper et al., 1991;Broomfield and Hooper, 1993).
In the present study we have continued our molecular characterization of the GPI anchors on porcine and human MDP. The complete structure of the porcine MDP GPI anchor has been determined, including the nature of the lipid species present, the extent of sialylation, and the site of GPI anchor attachment in the polypeptide chain. In addition, we have compared the GPI anchor glycan core structures of porcine and human MDP.
Purification of Membrane Dipeptidase-MDP was purified from porcine or human kidney cortex, following solubilization with either PI-PLC (hydrophilic form) or n-octyl-␤-D-glucopyranoside (amphipathic form), by affinity chromatography on cilastatin-Sepharose (Littlewood et al., 1989;Hooper and Turner, 1989). Protein was determined using the bicinchoninic acid method of Smith et al. (1985) modified for use in 96-well microtiter plates (Hooper, 1993) with bovine serum albumin as standard.
Isolation of the Radiolabeled Neutral Glycan Fraction from Hydrophilic Membrane Dipeptidase-Except for the desialylation procedure this method is as described in Ferguson (1992b). Purified hydrophilic MDP (0.47 mg, 10 nmol) was freeze-dried and desialylated with 40 mM trifluoroacetic acid (100 l) at 80°C for 1 h and freeze-dried again. The sample was then subjected to mild base hydrolysis (to remove any acyl groups and improve the solubility of the products), aqueous HF dephosphorylation, trichloroacetic acid precipitation, and desalting to yield the glycan core fraction. This material was deaminated with nitrous acid and 20% was reduced with tritiated sodium borohydride (NaB 3 H 4 ), followed by sodium borodeutoride (NaBD 4 ). The remaining 80% was reduced with NaBD 4 alone. In the case of the NaB 3 H 4 -reduced material, the resulting neutral glycan fraction (containing 1[ 3 H]-2,5-anhydromannitol, AHM) was extensively repurified by downward paper chromatography followed by high voltage electrophoresis (HVE) to remove radioactive contaminants. The tritium-labeled neutral glycans were eluted from paper with water, recombined with the NaBD 4 -reduced material, and stored at Ϫ20°C.
Liquid Chromatography-For resolution of the GPI neutral glycans of porcine and human MDP, two methods of liquid chromatography were used. Samples (5% of the total neutral glycan fraction) were first chromatographed by Dionex HPLC using a Carbopac PA-1 column, and each peak was subsequently desalted and rechromatographed by gelfiltration chromatography on Bio-Gel P4 (Ferguson, 1992b). The radiolabeled neutral glycans were detected using a Raytest Ramona on-line radioactivity monitor (Raytest Instruments, Sheffield, U.K.), and their elution positions were defined in relation to those of ␤-glucose oligomer internal standards. Each neutral glycan was assigned two elution positions, one in "Dionex units" (Du, Dionex HPLC) and the other in "glucose units" (Gu, Bio-Gel P4 gel-filtration) by linear interpolation of their elution positions between the ␤-glucose oligomer internal standards. The relative abundance of each neutral glycan species was calculated as a percentage of the total radioactivity.
Analysis of Neutral Glycans Using Thin Layer Chromatography-Exoglycosidase digestions performed on the neutral glycans were assessed by thin layer chromatography as described in Schneider et al. (1993). Samples before and after digestion were applied to aluminumbacked silica gel HPTLC plates and run alongside tritium-labeled neutral glycan standards from rat brain Thy-1  and tritium-labeled reduced ␤-glucose oligomer standards prepared as described in Schneider et al. (1993). The plates were developed in 1-propanol/acetone/water (9:6:5, v/v) followed by 1-propanol/acetone/water (5:4:1, v/v). Radioactive glycans were detected by spraying the plates with En 3 Hance prior to fluorography with an intensifying screen at Ϫ

70°C.
Isolation of the N-Acetylated Glycan Core Fraction of Porcine Membrane Dipeptidase-The N-acetylated glycan core fraction was prepared by aqueous HF dephosphorylation and N-acetylation of porcine hydrophilic MDP (8.2 mg, 175 nmol) as described in Ferguson (1992b) except that the time of aqueous HF dephosphorylation was reduced to 40 h. The glycan core was separated from the protein using a Centricon-10 concentrator (Amicon, MA), with the protein remaining in the retentate and the glycan core of the GPI anchor passing through the membrane in the filtrate. The filtrate was then applied to an AG50X12(H ϩ ) column (5 ml), eluted with 25 ml of water, freeze-dried, and redissolved in 0.5 ml of water. The sample was then applied to a small Bio-Gel P4 gel-filtration column (1.5 ϫ 30 cm) and eluted with water at 0.2 ml/min. The N-acetylated glycan core fraction was located by analyzing groups of fractions for carbohydrate composition.
Carbohydrate Compositional Analysis-Compositional analysis was performed as described in Ferguson (1992b) using methanolysis followed by re-N-acetylation and trimethylsilyl (TMS) derivatization. The resulting neutral sugar, hexosamine, and sialic acid derivatives were analyzed by gas chromatography-mass spectrometry (GC-MS), and scyllo-inositol (1 nmol) was used as an internal standard.
Methylation Analysis-Methylation analysis was carried out on the porcine neutral glycan fraction (9.5 nmol), prior to fractionation by liquid chromatography, and on the N-acetylated glycan core fraction (150 nmol). Methylation analysis was performed using a modification of the method of Ciucanu and Kerek (1984) as described in Ferguson (1992b). Analysis of the resulting partially methylated alditol acetates (PMAAs) was by GC-MS. GC-MS analyses were performed using both SE-54 and SP2380-bonded phase columns to allow detection of hexosamine PMAAs and the resolution of mannose and glucose PMAAs, respectively.
Isolation of Sialylated Glycans from the GPI Anchor of Porcine Membrane Dipeptidase-Hydrophilic porcine MDP was subjected to deamination, reduction, dephosphorylation, N-acetylation, and desalting essentially as described in Schneider and Ferguson (1995). Briefly, MDP (8.2 mg, 175 nmol) was dissolved in 0.3 M sodium acetate, pH 4.0 (1 ml) and deaminated and reduced as described, except that only 5% of the sample (8.5 nmol) was reduced with NaB 3 H 4 , and the remainder was reduced with NaBD 4 (166.5 nmol). Following dialysis against water, both samples were freeze-dried and dephosphorylated with 50% aqueous HF (0°C, 40 h), N-acetylated, and desalted. The NaB 3 H 4 -reduced sample was subjected to paper chromatography followed by HVE to remove radiochemical contaminants. Regions of the HVE electrophoretogram containing neutral and acidic glycans were eluted with water and combined. An aliquot of the 3 H-labeled glycans (2 ϫ 10 5 cpm, 1 nmol) was added to the majority of the NaBD 4 -reduced material (140 nmol) to act as a tracer. The GPI glycan sample was adjusted to 10 mM ammonium acetate, pH 6.0, and then subjected to ion-exchange chromatography on a column of QAE-Sephadex A25 (10 ϫ 1.5 cm), eluted with a linear gradient of 10 -500 mM ammonium acetate, pH 6.0, over 100 min at 0.5 ml/min. Fractions of 1 ml were collected and aliquots of these fractions were taken for scintillation counting.
Ethanolamine and Inositol Analyses-Porcine hydrophilic MDP (0.5 nmol) was subjected to ethanolamine analysis as described in Ferguson (1992b) using a Waters Pico-Tag amino acid analyzer and nor-leucine (1 nmol) as an internal standard. Inositol analysis was also performed on the same MDP sample (100 pmol) using GC-MS with scyllo-inositol (20 pmol) as the internal standard (Ferguson, 1992b).
Analysis of the Phosphatidylinositol Moieties of the Membrane Form of Porcine Membrane Dipeptidase-The amphipathic membrane form of porcine MDP (2 nmol) was dissolved in 50 l of water and extracted three times with 50 l of butan-1-ol saturated with water, to remove any contaminating phospholipids. The sample was subsequently freezedried and redissolved in 15 l of 0.3 M sodium acetate buffer, pH 4.0. Deamination was performed by adding sodium nitrite, 7.5 l of a fresh 1 M solution, and incubating for 1 h at room temperature, followed by 2 h at 56°C with a further 15 l of buffer and 7.5 l of sodium nitrite solution. The released PI moieties were recovered by extraction once with 100 l, and twice with 50 l, of butan-1-ol saturated with water. The combined butan-1-ol phases were washed twice with 200 l of water saturated with butan-1-ol and dried in a stream of N 2 . The sample was redissolved in 100 l chloroform/methanol (2:3, v/v), containing 1 mM NH 3 , and introduced in 20-l aliquots into the electrospray source of a VG-Biotech Quattro triple-quadruple mass spectrometer (Fisons Instruments, U.K.) at 10 l/min via a Michrom UMA microbore HPLC system (Michrom Associates, CA). Negative ion mass spectra were recorded after optimizing the source conditions for maximal PI pseudomolecular ion response with standards of soybean PI and bovine liver PI. Typical conditions were: capillary voltage, 3.0 kV; high voltage lens, 0.3 kV; focus lens, 40 V; skimmer lens, 45 V. Experiments employing collision-induced dissociation tandem mass spectrometry (parent ion and daughter ion spectra) used argon in the collision cell (cell pressure 2.5 ϫ 10 Ϫ3 mbar). Ions were accelerated into the collision cell through a potential difference of 60 V. All data were processed using MassLynx software (Fisons Instruments, U.K.).
Determination of the Site of GPI Anchor Attachment to Porcine Membrane Dipeptidase-Porcine hydrophilic MDP (0.94 mg, 20 nmol) was reduced with 20 mM dithiothreitol (1 h at 37°C), S-carboxymethylated with 50 mM iodoacetic acid (1 h at 37°C in the dark) and trypsindigested for 16 h at 37°C using a protein/TPCK (L-1-tosylamido-2phenylethyl chloromethyl ketone)-treated trypsin ratio of 50:1 (w/w) as described in Adachi et al. (1990b). The sample was carefully adjusted to pH 5.0 with 5% (v/v) acetic acid, and an aliquot (equivalent to 1.0 nmol) was freeze-dried and redissolved in 50 mM sodium acetate-HCl, 20 mM CaCl 2 , pH 5.0 (equilibration buffer, 0.1 ml final volume). This was applied to an anhydrotrypsin (AHT)-Sepharose column prepared as described in Ishii et al. (1983) (10 mg AHT, 3.5-ml bed volume, 1-ml void volume and pre-equilibrated in equilibration buffer (100 ml)). The column was sealed and mixed overnight, the AHT-Sepharose was allowed to settle in the column, and the first 15 ml of the run-through fraction was collected while equilibration buffer was applied at 10 ml/h. The run-through fraction was freeze-dried and redissolved in 0.1 ml of water. This sample was subjected to reverse-phase HPLC using a Bondapak C 18 steel column (3.9 ϫ 300 mm) with a UV (214 nm) detector at a flow rate of 2.0 ml/min and a 30-min gradient of 0 -70% (v/v) acetonitrile in 0.08% H 3 PO 4 at pH 2.5 followed by 5 min elution at the final conditions. The fractions corresponding to the major peaks were pooled, freeze-dried, and subjected to solid-phase peptide sequencing as described in Findlay et al. (1989) and Hooper et al. (1990).

RESULTS
Purification of Membrane Dipeptidase-The hydrophilic forms of both porcine and human MDP were purified from kidney cortex following PI-PLC solubilization to apparent homogeneity as assessed by SDS-polyacrylamide gel electrophoresis (data not shown). The apparent molecular masses were 47 and 59 kDa for the porcine and human forms, respectively, in agreement with Littlewood et al. (1987) and . The amphipathic form of porcine MDP was purified following solubilization with n-octyl-␤-D-glucopyranoside to apparent homogeneity with a molecular mass of 45 kDa, in agreement with Hooper and Turner (1989).
Sequencing of the GPI Glycan Cores-Neutral glycans terminating with AHM were isolated from both porcine and human MDP. An aliquot of each of these samples (1 ϫ 10 6 cpm, 4 -5% of total radioactivity) was subjected to Dionex HPLC and each resolved peak then further chromatographed using Bio-Gel P4 gel-filtration. The results obtained for the porcine neutral glycan fraction are illustrated and show that two peaks were resolved on Dionex HPLC (Fig. 2a). One of these peaks chromatographed as a single species on Bio-Gel P4 whereas the other was further separated into two distinct species (Fig. 2, b  and c). The elution positions of the three resolved neutral glycan structures (B, C, and D) on both Dionex HPLC and Bio-Gel P4 gel-filtration, and their relative abundance, are summarized in Table I. Two of the structures (C and D) were assigned the provisional structures Man␣1-2Man␣1-6(GalN-Ac␤1-4)Man␣1-4AHM and Man␣1-2Man␣1-6Man␣1-4AHM based on the known chromatographic properties of other GPI neutral glycans (Ferguson, 1992b) and in the knowledge that, to date, no two neutral glycans have been found to co-elute on both Dionex HPLC and Bio-Gel P4 gel-filtration. It was not possible to assign a provisional structure to neutral glycan structure B as a GPI glycan with these chromatographic properties has not been previously reported. The results of the human neutral glycan liquid chromatography (data not shown) were similar to those for the porcine neutral glycans, except that an additional neutral glycan structure (A), that eluted at 3.5 Du on Dionex HPLC and 6.5 Gu on Bio-Gel P4 gel-filtration, was resolved. Based on these elution positions and comparison with previously determined GPI neutral glycans this structure was provisionally assigned as Man␣1-2Man␣1-2Man␣1-6-(GalNAc␤1-4)Man␣1-4AHM (Table I).
A preliminary set of exoglycosidase digestions on the unknown porcine neutral glycan (structure B) were performed and assessed by Dionex HPLC and/or Bio-Gel P4 gel-filtration (data not shown). Digestions with jack bean ␤-hexosaminidase, jack bean ␤-galactosidase, and coffee bean ␣-galactosidase were all negative. However, bovine testes ␤-galoactosidase successfully digested all of structure B to structure C. Resistance to jack bean ␤-galactosidase and sensitivity to bovine testes ␤-galoactosidase is consistent with the presence of a non-reducing terminal ␤Gal residue linked 1-3 to a subterminal residue (Li et al., 1975;Distler and Jourdain, 1978). Thus neutral glycan structure B appears to differ from structure C by the presence of a non-reducing terminal 1-3-linked ␤Gal residue. The provisional assignment of the neutral glycan structures from their chromatographic properties greatly assisted in designing the minimum number of exoglycosidase digestions needed to sequence them. An authentic standard of Man␣1-2Man␣1-6Man(GalNAc␤1-4)Man␣1-4AHM isolated from rat brain Thy-1  was used as a control throughout. Two series of exoglycosidase digestions were performed on the porcine neutral glycans, and the products were analyzed by HPTLC (Fig. 3, a and b). Identical results were obtained for the human neutral glycan structures B, C, and D (data not shown), and the results for the human neutral glycan structure A are shown in Fig. 3c. The digestions shown in Fig. 3a define the position of the Gal␤1-3GalNAc␤1 and GalNAc␤1 branches of structures B and C to the ␣Man residue adjacent to the AHM.
All of the digestions were successful except that the first set of jack bean ␣-mannosidase digestions for each of the structures in Fig. 3a achieved only partial (about 60 -80%) digestion. The known standard was also only partially digested, and therefore it was concluded that each of the samples was homogeneous and that the jack bean ␣-mannosidase digestion in this step was incomplete. The incomplete jack bean ␣-mannosidase digestions were most likely due to steric hindrance from the ␤GalNAc residue. Each of the samples were completely digested in the second jack bean ␣-mannosidase digestion which would not have been the case if there were other structures present. The use of Man␣1-2Man-specific A. saitoi ␣-mannosidase in Fig. 3, b and c, was important to show that the nonreducing terminal Man residue in structures B, C, and D was linked ␣1-2 to the subterminal Man residue and that both the non-reducing terminal ␣Man residue and the subterminal ␣Man residue of neutral glycan structure A were similarly 1-2-linked.
In addition to the assessment of the exoglycosidase digestions by HPTLC, all of the digestions performed on porcine structure D were also assessed by Dionex HPLC and Bio-Gel P4 gel-filtration chromatography. The APAM digestion product (Fig. 3b) eluted at 2.1 Du and 3.2 Gu and the jack bean ␣-mannosidase digestion products (Fig. 3, a and b) eluted at 1.0 Du and 1.7 Gu. These chromatographic properties are consistent with Man␣1-6Man␣1-4AHM and AHM, respectively (Ferguson, 1992b). All of the exoglycosidase digestions performed in Fig. 3b on the porcine structures B and C, to the point where the structure was the same as porcine structure D, were also assessed by Dionex HPLC and Bio-Gel P4 gel-filtration chromatography, and these results were also consistent with the intermediate structures presented. Identical results were also obtained for the human structures B, C, and D. The exoglycosidase digestions of human structure A presented in Fig. 3c were also assessed by Dionex HPLC and Bio-Gel P4 gel-filtration chromatography. The jack bean ␤-hexosaminidase digestion product eluted at 3.0 Du and 5.1 Gu, the A. saitoi ␣-mannosidase digestion product eluted at 2.1 Du and 3.2 Gu, and the jack bean ␣-mannosidase digestion product eluted at 1.0 Du and 1.7 Gu. These results are consistent with Man␣1-2Man␣1-2Man␣1-6Man␣1-4AHM, Man␣1-6Man␣1-4AHM, and AHM, respectively (Ferguson, 1992b).
Methylation analysis of the total porcine neutral glycan fraction (Table II) is consistent with the exoglycosidase results described above. The presence of 4,6-di-O-substituted Man defines the linkage position of the GalNAc residue to the 4-position of the ␣Man residue adjacent to the AHM. The presence of terminal-GalNAc is consistent with the presence of structure C and the presence of terminal-Gal and 3-O-substituted GalNAc confirms the presence of the Gal␤1-3GalNAc branch proposed for structure B. Methylation analysis of the N-acetylated glycan core sample (Table II) showed, in addition, the presence of 4-O-substituted GlcNAc (in place of the 4-O-substituted AHM) and 6-O-substituted myo-inositol. These results further define the original GPI structures as containing GlcN1-6-myo-inositol termini.
Isolation of Sialylated Glycans from the GPI Anchor of Porcine Membrane Dipeptidase-GPI glycans were prepared from porcine MDP by deamination, reduction, dephosphorylation, and N-acetylation, as described under "Experimental Procedures." Following radiochemical decontamination the glycans (containing a radioactive tracer, 2 ϫ 10 5 cpm) were applied to a QAE-Sephadex A25 (OH Ϫ ) column and eluted with a gradient of ammonium acetate. Three radioactive peaks were resolved ( Fig. 4), freeze-dried, and each fraction was analyzed for monosaccharide composition by GC-MS. Peak 1, eluting in the column void volume, contained Man, GalNAc, and Gal (in a molar ratio of 3.0:0.9:0.4) but no sialic acid. This monosaccharide ratio is consistent with the mixture of neutral glycans B, C, and D indicated in Table I. Peak 2 was considered to be irrelevant residual radiochemical contamination since it contained only a trace of Man. Peak 3 contained Man, GalNAc, Gal, and sialic acid (in a molar ratio of 3.0:1.7:0.4:1.7). The equimolar GalNAc and sialic acid contents, and the low Gal content, suggest that the major sialylated glycan is Man␣1-2Man␣1-6(GalNAc␤1-4)Man␣1-4AHM (i.e. glycan C in Table II). The complete absence of fucose, a component of the complex sialylated N-linked oligosaccharides of MDP, 2 suggested that peaks 1 and 3 are composed solely of GPI glycan material. The total radioactivity recovered in peaks 1-3 was 1.92 ϫ 10 5 cpm which represented a recovery of 96% of the total radioactivity applied to the QAE-Sephadex column. The Man contents of peaks 1 and 3 were 22.6 and 10.5 nmol, respectively. From this the sialylated glycans (peak 3) were calculated to represent 4.5% of the recovered glycans. The extent of desialylation due to aqueous HF treatment was calculated to be 31 and 52% for ␣2-3and ␣2-6-linked sialic acid, respectively. Hence, the maximal proportion of sialylated GPI anchor species (if they were all ␣2-6-linked and not polysialylated) was determined as 9% of the total glycan population.
Ethanolamine Analysis-To assess the ethanolamine phosphate content of the porcine MDP GPI anchor ethanolamine and inositol analyses were performed. This revealed that porcine MDP contained 2.5 mol of ethanolamine/mol of myo-inositol. Hence, as each GPI anchor contains one myo-inositol residue, the porcine GPI anchor probably contains two ethanolamine phosphates in all of the glycan structures and three ethanolamine phosphate moieties in 50% of the glycans. The locations of these ethanolamine phosphate groups, as indicated in Fig. 1, are made by analogy with the known structures of rat brain Thy-1 and human erythrocyte acetylcholinesterase Deeg et al., 1992).

Identification of the PI Moieties of the Membrane Form of Porcine MDP-High resolution negative ion ES-MS analysis of
the PI-containing fraction, isolated by nitrous acid deamination and butan-1-ol extraction, revealed major pseudomolecular ions at m/z 815, m/z 837, and m/z 865 (Fig. 5a). The low resolution parent ion scan (for parents of the PI-specific daughter ion at m/z 241, corresponding to the inositol-1,2-cyclic phosphate fragment ion (Sherman et al., 1985) revealed that the m/z 865 parent ion was due to a PI species, whereas the m/z 815 ion was not (Fig. 5b). The nature of the m/z 837 ion was equivocal by this analysis. The mass of the major ion at m/z 865 is consistent with that predicted for the [M-H] Ϫ pseudomolecular ion of distearoyl-PI (or any other isometric PI species). The daughter ion spectrum of m/z 865 (Fig. 5c) shows the presence of the PI-specific m/z 241 fragment ion (i.e. [inositol-1,2-cyclic phosphate] Ϫ ), together with another intense daughter ion at m/z 283. The latter ion is the carboxylate ion of stearic acid (i.e. [CH 3 (CH 2 ) 16 COO] Ϫ ). The absence of any other signficant daughter ions shows that the parent ion contains only this fatty acid. Thus, the major PI species released from porcine MDP by nitrous acid deamination must be distearoyl-PI. From this unequivocal identification of the m/z 865 ion, the minor ion at m/z 837 (28 mass units lower) is likely to be due to a stearoyl-palmitoyl-PI species. Significantly, no m/z 885 ion, corresponding to the [M-1] Ϫ ion of the major PI species (1-stearoyl-2-arachidonoyl-PI) of mammalian cells (Michell, 1975;Kerwin et al., 1994) was observed. This latter finding strongly suggests that there was no significant PI-phospholipid contamination of the sample.
Determination of the Site of GPI Anchor Attachment to Porcine Membrane Dipeptidase-Porcine hydrophilic MDP was reduced, S-carboxymethylated, and trypsin-digested. SDS-polyacrylamide gel electrophoresis performed under non-reducing conditions on aliquots after each treatment confirmed that the enzyme had been successfully reduced and trypsin-digested (results not shown). Trypsin digestion selectively cleaves proteins on the C-terminal side of the basic amino acid residues Arg and Lys and the trypsin digestion of MDP therefore contained 28 peptides (determined using the cDNA-derived amino acid sequence (Rached et al., 1990)). All these peptides, except for the C-terminal peptide, have a C-terminal Arg or Lys, and this fact was exploited to isolate the C-terminal peptide by affinity chromatography using AHT-Sepharose. AHT is a catalytically inert derivative of trypsin but still retains a strong 3.0 6.7 38 17 binding affinity toward the products of trypsin digestion. The AHT-Sepharose should therefore bind all the peptides except for the C-terminal peptide (Ishii et al., 1983). Following AHT-Sepharose chromatography, the run-through fraction was sub-jected to reverse-phase HPLC to assess the previous step, and this revealed that the AHT-Sepharose chromatography appeared to have bound all but three of the peptides in the trypsin-digested sample as three significant peaks were pres- The standard used as a control was authentic Man␣1-2Man␣1-6(GalNAc␤1-4)Man␣1-4AHM isolated from rat brain Thy-1. Each lane contains the equivalent of 5000 cpm of material. Tritium-labeled reduced ␤-glucose oligomer standards (Dex, 10,000 cpm) were also chromatographed on the HPTLC plate. Fluorography was performed at Ϫ70°C for 1 week. ASAM, A. saitoi ␣-mannosidase; BTBG, bovine testes ␤-galactosidase; JBAM, jack bean ␣-mannosidase; JBBH, jack bean ␤-hexosaminidase; ent (results not shown). Each of these peak fractions was subjected to peptide sequencing, and this revealed the presence of a single peptide in each fraction. The relative abundance of each peptide was estimated from the amount of the first residue sequenced. The most abundant peptide had the sequence TNYGYS and corresponded to positions 363-368 near the C terminus of the cDNA-derived amino acid sequence (Fig. 6). No amino acid residue was identified in the seventh cycle of the sequencer indicating that this was the complete sequence of this peptide. The two less abundant peptides (63 and 54% as compared with the hexapeptide TNYGYS) corresponded to the tryptic peptides at positions 262-275 and 10 -54, respectively, of the amino acid sequence predicted from the cDNA sequence. These two peptides are relatively large and each contain one of the only two N-linked glycosylation sites present in porcine MDP. On N-terminal sequencing of the whole protein Asn 41 failed to sequence consistent with it being modified by an N-linked sugar (Rached et al., 1990). In the present study Asn 263 also failed to sequence, again consistent with it being modified by an N-linked sugar. Thus the large size and presence of N-linked sugars on these peptides may well account for their failure to be bound effectively by the AHT-Sepharose. DISCUSSION A combination of Dionex HPLC and Bio-Gel P4 gel-filtration (Fig. 2), exoglycosidase digestions (Fig. 3), and methylation analyses (Table II) produced a complete qualitative and quantitative analysis of the glycan cores of MDP isolated from both porcine and human kidney ( Table I). All of the structures identified contained the conserved GPI core sequence found throughout eukaryotic evolution of Man␣1-2Man␣1-6Man␣1-4GlcN (McConville and Ferguson, 1993). The carbohydrate side chains found in both MDP samples included GalNAc and Gal␤1-3GalNAc, both linked ␤1-4 to the conserved core (see Fig. 1). In human MDP one additional minor structure, containing an extra ␣Man residue, as well as a GalNAc side chain, was also observed. This is the first trans-species comparison of the GPI anchor structures of the same protein expressed in the same tissue. It is noteworthy that the porcine and human anchor preparations display identical structures in quite similar molar ratios, thus suggesting that side chain modifications to GPI anchors are not species-specific but rather proteinand/or tissue-specific. Interestingly, the side chain modifications (an extra ␣Man residue and a GalNAc) to the GPI anchor of human urine CD59 (Nakano et al., 1994), which is probably derived from the epithelial cells lining the kidney tubules, are similar to those found on human MDP (see Fig. 1).
Microheterogeneity of glycan core structures has also been commonly observed with other mammalian GPI structures although none is found either in human erythrocyte acetylcholinesterase (Roberts et al., 1988a) or human placental alkaline phosphatase (Redman et al., 1994). The presence of GalNAclinked ␤1-4 to the conserved GPI core has been previously described in rat brain Thy-1 , Torpedo acetylcholinesterase (Mehlert et al., 1993), mouse skeletal muscle NCAM (Mukasa et al., 1995) and human urine CD59 (Nakano et al., 1994), and HexNAc has been observed in hamster brain scrapie prion protein (Stahl et al., 1992) and bovine liver 5Ј-nucleotidase (Taguchi et al., 1994). The presence of the Gal␤1-3GalNAc side chain has not been previously noted in GPI anchor structures. However, it is reminiscent of the partial  FIG. 4. Ion-exchange chromatography of an N-acetylated glycan core fraction of porcine membrane dipeptidase. GPI glycans were prepared from porcine MDP by deamination, reduction, dephosphorylation, and N-acetylation. The glycans (containing a radioactive tracer) were then subjected to ion-exchange chromatography on a QAE-Sephadex A25 (OH Ϫ ) column. Three radioactive peaks (labeled 1-3, solid line) were resolved, eluted, and analyzed for monosaccharide composition by GC-MS. Also presented is the ammonium acetate gradient applied to the column (dashed line). See text for details. structure described for the hamster brain scrapie prion protein GPI anchor of Hex-HexNAc and NANA-Hex-HexNAc side chains (Stahl et al., 1992). The Gal␤1-3GalNAc motif is also common in glycoprotein and mucin O-linked oligosaccharides (Carlstedt et al., 1985;Van Halbeek et al., 1988;Hokke et al., 1994), and it is possible that the same ␤1-3 galactosyltransferase is used to modify both O-linked GalNAc residues and GPI anchor GalNAc residues.
The possibility that some of the GPI glycans were originally sialylated was investigated by releasing the GPI anchor of porcine MDP by aqueous HF dephosphorylation under conditions that retain half to two-thirds of the sialic acid residues, depending on the linkage. The low yield of sialylated GPI glycans obtained after anion-exchange chromatography suggested that a maximum of 9% of the original GPI anchors contained sialic acid (presuming ␣2-6 linkage and no polysialylation). The low abundance of sialylated species, and their inherent lability to the aqueous HF conditions necessary to prepare them, prevented a detailed structural characterization. However, the composition of the sialylated fraction suggests that the presence of GalNAc is a prerequisite for sialylation whereas only a small fraction of the sialylated species contained Gal. This leads us to tentatively suggest that in the case of the porcine MDP GPI anchor most, if not all, of the sialic acid is attached to the side chain GalNAc residue. The presence of sialic acid in GPI anchors has been reported only twice. In one case, that of T. brucei PARP, an average of 5 sialic acid residues/anchor were found linked ␣2-3 to the terminal ␤Gal residues of branched polylactosamine side chain structures . In the other example, that of hamster brain scrapie prion protein, about 30% of the anchors were found to be monosialylated with the sialic acid attached to a Hex-HexNAc side chain, predominantly in the form NANA- FIG. 6. Comparison of the deduced C-terminal amino acid sequence of porcine membrane dipeptidase with the cDNAderived amino acid sequences of porcine and human membrane dipeptidase. a, determined C-terminal peptide sequence of porcine dipeptidase. b, porcine dipeptidase cDNA-derived protein sequence (Rached et al., 1990). c, human dipeptidase cDNA-derived protein sequence (Adachi et al., 1990a). S ϭ site of GPI anchor attachment. 2 ϭ site of cleavage by trypsin. The porcine and human cDNA sequences are aligned to give maximum homology between them, hence the two gaps (-) in the porcine sequence. Numbering refers only to the porcine sequence.
Hex-HexNAc (Stahl et al., 1992). Thus, the possibility exists that, like O-linked oligosaccharides and ganglioside glycans, sialic acid can be linked to both Gal and GalNAc residues in GPI anchor side chains.
The porcine MDP was shown to contain 2.5 mol of ethanolamine/mol of GPI anchor indicating that, in addition to the ethanolamine phosphate bridge, there is an extra ethanolamine phosphate moiety in the GPI anchor and, probably in half the structures, a third ethanolamine phosphate residue is present. The presence of at least one extra ethanolamine phosphate in the GPI anchors of metazoan (but significantly not unicellular) eukaryotes is well documented . Deeg et al. (1992) first precisely analyzed the presence of two additional ethanolamine phosphates in human erythrocyte acetylcholinesterase in 10 -20% of the structures. Two additional ethanolamine phosphates are also thought to occur in a proportion of the GPI anchors of bovine liver 5Ј-nucleotidase (only 3.5% of total population) (Taguchi et al., 1994), Torpedo acetylcholinesterase (up to 30% by compositional analysis) (Mehlert et al., 1993), human placental alkaline phosphatase (up to 40% by compositional analysis) (Redman et al., 1994), and human CD52 (40% of the total population) (Treumann et al., 1995). The presence of two additional ethanolamine phosphate residues was not found in rat brain Thy-1 nor hamster brain scrapie prion protein suggesting that this is not a ubiquitous feature of all metazoan GPI structures Stahl et al., 1992). The presence of this third ethanolamine phosphate in about half of the porcine MDP anchors would cause microheterogeneity of charge, in addition to the microheterogeneity of the carbohydrate residues.
In general, the GPI PI moieties are substantially different from the cellular pool of PI phospholipids . For example, several of the mammalian GPI anchors contain exclusively alkylacyl-PIs (Roberts et al., 1988b;Walter et al., 1990;Redman et al., 1994) (see Fig. 1) as opposed to sn-1-stearoyl-2-arachidonoyl-PI that is the predominant cellular PI species in these organisms (Michell, 1975;Kerwin et al., 1994). In the case of porcine MDP, the predominant PI moiety is a diacyl-PI (distearoyl-PI). Although diacyl-PI moieties are quite common in non-mammalian GPI anchors (e.g. in Torpedo acetylcholinesterase (Bü tikofer et al., 1990), T. brucei VSG (Ferguson et al., 1985), Saccharomyces cerevisiae gp125 (Fankhauser et al., 1993), and human CD52 (Truemann et al., 1995)), the presence of predominantly distearoyl-PI has only been reported previously for one of the two forms of human CD52 (Truemann et al., 1995). The only other example of a GPI anchor diacyl-PI that contains predominantly one type of acyl chain is that of T. brucei VSG. The dimyristoyl-PI moiety of the VSG anchor is produced by a process of fatty acid remodeling (Masterson et al., 1990), whereby the original heterogeneity in the PI moiety (Doering et al., 1994) is removed by sequentially replacing the sn-2 and sn-1 fatty acids with myristate. While most mammalian GPI intermediates and precursors contain alkylacyl-PIs, some contain diacyl-PIs (Puoti and Conzelmann, 1993). Thus, it is possible that some kind of analogous fatty acid remodeling may occur on the diacyl-PI containing GPI intermediates in the cells producing CD52 and porcine MDP. Alternatively, these cells may simply select for distearoyl-PI as the precursor for the GPI intermediates that will be transferred to these proteins.
The results of the trypsin digestion of porcine MDP and isolation of the C-terminal peptide on AHT-Sepharose identified the site of GPI anchor attachment as Ser 368 . Thus the mature protein lacks the predominantly hydrophobic C-terminal sequence of 25 residues predicted from the cDNA (Rached et al., 1990). The Ser anchor attachment site () has a pair of Ala residues immediately C-terminal to it. This is consistent with the proposals of Gerber et al. (1992), who have found that the next but one residue C-terminal to the site of anchor attachment ( ϩ 2 position) is generally restricted just to Ala, Gly, or Ser. Also displayed in Fig. 6 is the cDNA-derived protein sequence for human MDP and the previously determined GPI anchor attachment site of Ser 369 (Adachi et al., 1990b). The human and porcine proteins are 80% homologous and, from sequence alignment studies, it appears that in the porcine MDP either Ser 368 or Ser 369 has been mutationally deleted. However, it is impossible to predict from the cDNA codon information which Ser is absent. Whatever the case both the enzymes utilize Ser in the same part of the protein sequence for GPI anchor attachment and have Ala at the ϩ 2 position.
In conclusion, we have determined the glycan core structures of the GPI anchors on porcine and human MDP, representing the first interspecies comparison of mammalian GPI anchor structures on the same protein. The glycan core structures were remarkably similar, with three major structures observed in approximately equal amounts: Man␣1-2Man␣1-6Man␣1-4GlcN, Man␣1-2Man␣1-6(GalNAc␤1-4)Man␣1-4GlcN, and Man␣1-2Man␣1-6(Gal␤1-3GalNAc␤1-4)Man␣1-4GlcN. Also the GPI anchor on porcine MDP was found to be anchored to the protein at Ser 368 , to contain 2.5 mol of ethanolamine/mole of anchor, and almost exclusively distearoyl-PI. In addition, a small proportion of the porcine GPI structures were shown to contain sialic acid residues, probably attached to GalNAc.