Ethanolamine Phosphate Linked to the First Mannose Residue of Glycosylphosphatidylinositol (GPI) Lipids Is a Major Feature of the GPI Structure That Is Recognized by Human GPI Transamidase*

Glycosylphosphatidylinositol (GPI) anchoring of proteins is catalyzed by GPI transamidase (GPIT), a multisubunit, endoplasmic reticulum (ER)-localized enzyme. GPIT recognizes ER-translocated proteins that have a GPI-directing C-terminal signal sequence and replaces this sequence with a preassembled GPI anchor. Although the GPI signal sequence has been extensively characterized, little is known about the structural features of the GPI lipid substrate that enable its recognition by GPIT. In a previous study we showed that mature GPIs could be co-immunoprecipitated with GPIT complexes containing functional subunits (Vainauskas, S., and Menon, A. K. (2004) J. Biol. Chem. 279, 6540–6545). We now use this approach, as well as a method that reconstitutes the interaction between GPIs and GPIT, to define the basis of the interaction between GPI and human GPIT. We report that (i) human GPIT can interact with GPI biosynthetic intermediates, not just mature GPIs competent for transfer to protein, (ii) the ethanolamine phosphate group on the third mannose residue of the GPI glycan is not critical for GPI recognition by GPIT, (iii) the ethanolamine phosphate residue linked to the first mannose of the GPI structure is a major feature of GPIs that is recognized by human GPIT, and (iv) the simplest GPI recognized by human GPIT is EtN-P-2Manα1–4GlcN-(acyl)-phosphatidyl-inositol. These studies define the molecular characteristics of GPI that are recognized by GPIT and open the way to identifying GPIT subunits that are involved in this process.

Glycosylphosphatidylinositol (GPI) 2 serves as a membrane anchor for many eukaryotic cell surface proteins that are involved in diverse biological functions such as immune recog-nition, signal transduction and cell adhesion (for reviews, see Refs. [1][2][3][4][5]. Attachment of a GPI anchor occurs rapidly upon completion of protein translation and translocation across the endoplasmic reticulum (ER) membrane, and involves replacement of the C-terminal GPI-directing signal sequence in the proprotein with a preformed GPI anchor.
More than 20 different ER membrane proteins are required for synthesizing the GPI lipid and attaching it to target proteins (3,4,6). GPI biosynthesis (Fig. 1A) is initiated on the cytoplasmic face of the ER by a multisubunit enzyme that transfers N-acetylglucosamine (GlcNAc) from UDP-GlcNAc to phosphatidylinositol (PI) (7)(8)(9)(10). GlcNAc-PI is then de-N-acetylated and, in mammalian cells and yeast, inositol acylated to yield GlcN-acyl-PI. GlcNacyl-PI is elaborated by addition of mannose and ethanolaminephosphate (EtN-P) residues that are contributed by the lipids dolichol-P-mannose and phosphatidylethanolamine, respectively (11)(12)(13)(14)(15). The mannose and EtN-P transfer reactions are presumed to occur in the lumenal leaflet of the ER.
The complete GPI structures that are generated by this pathway in mammals contain 3 (occasionally 4) mannose residues and 2-3 EtN-P residues. Modification of the 6-OH of the third mannose residue (Man3) with EtN-P, as seen in the GPI anchor precursors H7 and H8 (16,17) (Fig. 1), is essential in order for GPI to be attached to protein. Modification of the 2-OH of the first mannose residue (Man1) with EtN-P (catalyzed by PIG-N/ Mcd4p) is essential in yeast (18 -20) possibly because of a preference of the third mannosyltransferase (PIG-B/Gpi10p) for an EtN-P-modified GPI substrate. This EtN-P is also important in mammalian cells to generate a full complement of cell-surfaceexpressed GPI-anchored proteins (21).
GPI transamidase (GPIT), the enzyme that catalyzes the attachment of GPI to protein, is a penta-subunit complex consisting of the membrane proteins (in mammals) GPI8, GAA1, PIG-S, PIG-T, and PIG-U (22)(23)(24)(25)(26)(27)(28). In the first step of the GPITcatalyzed reaction, the GPI signal sequence is cleaved and the newly generated ␣-carbonyl group is attached via a thioester linkage to the GPI8 subunit of GPIT. Nucleophilic attack on the activated carbonyl by the amino group of the terminal EtN-P residue of GPI regenerates GPIT and yields a GPI-anchored protein (29 -31). While the GPI signal sequence has been extensively analyzed biochemically as well as by bioinformatics approaches (32)(33)(34)(35), very little is known about the structural features of the GPI lipid substrate that enable its recognition by GPIT.
In an effort to shed light on this problem, we considered the following three questions. (i) Is the terminal EtN-P required for GPI interaction with GPIT? (ii) Can GPI biosynthetic intermediates bind GPIT? (iii) What is the minimal GPI structure recognized by GPIT? To answer these questions we made use of our ability to co-precipitate GPIs (metabolically radiolabeled with [2-3 H]mannose in HeLa cells) with epitope-tagged human GPIT complexes (36). We used GPI biosynthesis inhibitors to generate a spectrum of radiolabeled GPI biosynthetic intermediates in situ and assayed whether these different lipids could be co-precipitated with human GPIT complexes. We also reconstituted the binding interaction between GPIs and GPIT, enabling us to test the ability of human GPIT to interact with non-mammalian GPIs that lack the side-chain modifications characteristic of GPIs seen in mammalian cells. Our results show that the EtN-P residue linked to Man1 is a major feature of the GPI structure that is recognized by human GPIT, while the EtN-P linked to Man3 and the trimannosyl core itself are relatively unimportant. Thus human GPIT is able to bind the minimal GPI structure EtN-P-2Man␣1-4GlcN-acyl-PI (lipid H5; Fig. 1). The implications of these data are discussed.

MATERIALS AND METHODS
Antibodies-Mouse monoclonal antibodies against the FLAG and hemagglutinin epitope tags were purchased from Sigma. Mouse monoclonal antibody against the V5 epitope tag was obtained from Invitrogen (San Diego, CA). Antibodies against GPI8 and GAA1 were obtained as described previously (37). Anti-PIG-T antibodies were a gift from Dr. Taroh Kinoshita (Osaka University). Horseradish peroxidase-conjugated anti-rabbit and anti-mouse IgGs were from Promega Corp. (Madison, WI).
Cell Culture and Transfection-HeLa cells were cultured in Dulbecco's modified Eagle's medium supplemented with 10% (v/v) fetal bovine serum. The cells were maintained in a humidified 5% (w/v) CO 2 atmosphere at 37°C. Transfection of cDNAs into the cells was done by electroporation as described previously (37).
Immunoprecipitations and Immunoblots-HeLa cells (1-2 ϫ 10 7 cells) were processed 48 h after being transfected with cDNAs encoding epitope-tagged GPIT subunits. The cells were scraped off the culture dish, washed once with phosphate-buffered saline, resuspended in 1 ml of MSB buffer (20 mM HEPES-KOH, pH 7.6, 200 mM NaCl, 1% (w/v) digitonin and 1ϫ protease inhibitor mixture set I (Calbiochem)) and incubated on ice for 60 min. The cell lysate was clarified by medium speed centrifugation (20 min, 10,000 ϫ g, 4°C), and the resulting supernatant was ultracentrifuged (45 min, 100,000 ϫ g (43,000 rpm, TLA-100.3 rotor), 4°C) to generate a clear detergent extract. Extracts were treated with 30 l of slurry of anti-FLAG-and/or anti-V5-coated agarose beads at 4°C for 4 h with gentle agitation. The beads were pelleted by centrifugation for 15 s at 10,000 ϫ g and washed with MSB buffer (4 ϫ 1.5 ml). Bound GPIT complexes were released from the beads by incubating with FLAG or V5 peptide (200 g/ml) in MSB buffer, then analyzed by SDS-PAGE, followed by immunoblotting using chemiluminescence reagents (Pierce).

Labeling HeLa Cells with [2-3 H]Mannose; Extraction and Analysis of Radiolabeled GPIs-HeLa cells were labeled with D-[2-
3 H]mannose, and radiolabeled GPIs were extracted with chloroform/methanol/water (10:10:3, by volume), desalted by partitioning between n-butyl alcohol and water, and analyzed as described previously (36). Radiolabeled GPIs were identified by their sensitivity to GPI-specific phospholipase D (GPI-PLD) and jack bean ␣-mannosidase (Jb␣m) and by co-migration with previously characterized GPIs on thin layer chromatograms. Chromatograms were visualized by scanning using a Berthold LB2842 TLC Linear Analyzer.
Enzyme treatments were done as follows. For treatment with Jb␣m, extracts containing [ 3 H]mannose-labeled GPIs were dried, dissolved in 0.1 M sodium acetate, pH 5.2, 0.5% (w/v) Nonidet P-40, and treated with 2 units of Jb␣m (Sigma) overnight at 37°C. For treatment with GPI-PLD, radiolabeled GPIs were dried, dissolved in 50 mM Tris-HCl, pH 7.4, 10 mM NaCl, 3 mM CaCl 2 , 0.1% (w/v) sodium deoxycholate, and treated with 10% (w/v) fetal bovine serum (a source of GPI-PLD) overnight at 37°C. At the end of the incubation with Jb␣m or GPI-PLD, samples were extracted with n-butyl alcohol. The n-butyl alcohol extracts were washed repeatedly with n-butyl alcohol-saturated water before being taken for TLC analysis on silica 60 thin layer plates (Merck, Darmstadt, Germany) using chloroform/methanol/water (10:10:3, by volume) as the solvent system.

Synthesis of [ 3 H]Mannose-labeled GPIs in a Cell-free System and Co-immunoprecipitation of GPIs with GPIT Complexes-
HeLa cells were transfected with cDNAs encoding different epitope-tagged GPIT subunits. Two days post-transfection cells (ϳ1 ϫ 10 7 ) were washed twice with phosphate-buffered saline, scraped off the culture dish, resuspended in 0.4 ml of labeling buffer (0.25 M sucrose, 25 mM HEPES-KOH, pH 7.6, 15 mM MgCl 2 , 15 mM MnCl 2 , 3 mM ATP, 1 mM UDP-GlcNAc) with 1.5 g/ml digitonin, 0.2 g/ml tunicamycin, and 10 Ci of GDP-[ 3 H]mannose (American Radiolabeled Chemicals, Inc., St. Louis, MO), and incubated for 45 min at 37°C. After labeling, digitonin was added to a final concentration of 1% (w/v), and the cells were solubilized on ice for 60 min. The cell lysate was clarified by centrifugation and subjected to immunoprecipitation using anti-FLAG beads as described above. GPI extraction and analysis was done as described above.

Synthesis of [ 3 H]GlcNAc-PI in a Cell-free System-Micro-
were incubated with 5 Ci of UDP-[ 3 H]GlcNAc for 45 min at 37°C to generate [ 3 H]GlcNAc-PI as described previously (38). After labeling, GPIs were extracted as described above.
Metabolic labeling of HeLa cells with D-[2-3 H]mannose was done as described above. Lipids were extracted with chloroform/methanol/water (10:10:3 by volume). The chloroform/ methanol/water extracts were dried and desalted by partitioning between n-butyl alcohol -saturated water and watersaturated n-butyl alcohol. The n-butyl alcohol phase containing [ 3 H]mannose-labeled GPIs was dried, and the residue was dissolved initially in MSB buffer containing 1% (w/v) Triton X-100, then diluted with detergent-free MSB to obtain a final concentration of 0.5% (w/v) Triton X-100. A similar procedure was followed to obtain a solution of in vitro synthesized [ 3 H]GlcNAc-PI in MSB containing 0.5% (w/v) Triton X-100.
Radiolabeled GPIs (ϳ30,000 cpm in 3 l of MSB containing 0.5% (w/v) Triton X-100) were added to 50 l of anti-FLAG bead slurry (in MSB buffer with 0.5% digitonin) with bound GPIT complexes. The mixture was incubated for 25 min at room temperature, then 0.5 ml of MSB with 0.5% digitonin was added, and the incubation was continued for an additional 10 min. The bead-bound immunoprecipitates were washed with MSB buffer containing 0.5% digitonin (5 ϫ 1 ml), then aliquoted for immunoblotting and GPI extraction.

RESULTS
The GPI structures discussed in this paper are illustrated schematically in Fig. 1B.
Specific Co-immunoprecipitation of GPIs with Human GPIT-We previously showed that the GPI anchor precursor H8 could be co-immunoprecipitated with detergent-solubilized human GPIT complexes (36). These results were obtained by transiently expressing functional, epitope-tagged subunits of human GPIT in HeLa cells, then metabolically radiolabeling the cells with [2-3 H]mannose in the presence of tunicamycin to generate radiolabeled GPIs in situ. After labeling, the cells were lysed in digitonin-containing buffer and GPIT complexes were immunoprecipitated. The immunoprecipitate (GPIT IP) was washed in digitonin-containing buffer before being extracted with organic solvents to solubilize co-precipitated lipids. Nonimmunoprecipitated material (post-IP supernatant) was extracted in parallel. The organic solvent extracts were desalted and analyzed by TLC to identify [ 3 H]mannose-labeled GPIs. TLC profiles from a typical experiment (Fig. 2) show that of the many [ 3 H]mannose-labeled GPIs evident in the cell extract (upper panel; post-IP supernatant), only H8 is significantly coprecipitated with GPIT (lower panel; GPIT IP). M4-H8, 3 a minor GPI species corresponding to H8 with an additional ␣-linked mannose residue (Fig. 1), is also recovered in the GPIT IP but at a lower level than H8. In contrast, the [ 3 H]mannoselabeled GPIs H4 and H3 ( Fig. 1) are not detected in the GPIT IP. These structures are completely de-mannosylated on treatment with the exoglycosidase Jb␣m, consistent with the absence of any EtN-P (or other) substitutions of the mannose core. Thus mature GPI structures (H8, and to a lesser extent, M4-H8) are "pulled down" with GPIT in the immunoprecipitation assay, whereas GPIs lacking EtN-P altogether are not.
Many GPI-anchors found on mammalian proteins resemble H7 (which has two EtN-P groups, one linked to the first mannose and another to the third mannose of the GPI glycan (Fig.  1B)) rather than H8. Thus we were interested to see whether H7 interacts with GPIT. In HeLa cells, H8 is the main GPI species generated during metabolic radiolabeling with [ 3 H]mannose ( Fig. 2A), and levels of radiolabeled H7 are typically too low to test in GPIT co-precipitation experiments. We were able to generate readily detectable levels of H7 by growing HeLa cells in the presence of the aminoglycoside G418 (42). G418 treatment alters the profile of metabolically radiolabeled GPIs, increasing the levels of H7 without significantly decreasing the level of H8. HeLa cells were grown in medium containing G418 (0.75 FIGURE 1. Outline of the GPI biosynthetic pathway and structures of GPI anchor precursors and biosynthetic intermediates analyzed in this study. A, a brief outline of the GPI biosynthetic pathway is shown. Biosynthesis is initiated by the synthesis of GlcNAc-PI from UDP-GlcNAc and phosphatidylinositol. This lipid is de-N-acetylated and inositol acylated before receiving mannose and EtN-P additions. The first significant mannosylated intermediate is H5, which is elaborated in a series of steps to H6, H7 and H8. In the figure, H8 is shown attached to protein (depicted as a squiggle linked to the capping EtN-P of H8), although H7 (illustrated in B) can also be attached. GPI attachment to protein is catalyzed by GPIT. B, EtN-P-containing mammalian GPIs that interact with GPIT are shown in the top box. "MN" represents the major GPI intermediate that accumulates in cells treated with ManN. GPI intermediates that do not bind GPIT as well as trypanosome GPIs used in this study are shown below. With the exception of the symbols for inositol and EtN-P, symbols used are as described on the web site of the Consortium for Functional Glycomics. mg/ml) for 5 days. The cells were transfected with cDNAs encoding epitope-tagged GPIT subunits, metabolically labeled with [2-3 H]-mannose, and processed as described above to immunoprecipitate GPIT. Lipid extraction and TLC analysis of the GPIT IP showed that both H7 and H8 were present in the immunoprecipitated complexes (data not shown) indicating that both H7 and H8 are able to interact with GPIT.
Although the pulldown assay provides only a semiquantitative assessment of GPI interaction with GPIT, clear distinctions can nevertheless be made between the ability of different GPIs to be recognized by the enzyme (Fig. 2). Thus M4-H8 is less well recognized than H8 (recovery of M4-H8 in the IP versus post-IP extracts is ϳ2-fold lower than that of H8 (Fig. 2)), and H4 and H3 do not bind GPIT at all. The pulldown assay is also highly specific. We previously showed that GPIT complexes containing a non-functional GAA1 subunit do not pull down H8 (36). This observation pointed to a role for GAA1 in GPI recognition by GPIT and also served as a specificity control for the pulldown assay. The results described above further document the specificity of the assay since not all GPIs are pulled down with wild-type GPIT.
The Terminal EtN-P Residue Is Not Required for GPI Recognition by Human GPIT-GPIT transfers "complete" GPIs, like H8 and H7, to ER-translocated proteins that display a signal sequence for GPI anchor attachment. These complete GPIs are characterized by the presence of a capping EtN-P residue linked to the 6-OH of Man3. The amino group of the EtN-P cap ultimately participates in the amide bond that links the GPI structure to the amino acid of the target protein. Because of its direct role in GPI anchoring we speculated that the EtN-P cap would be a critical feature of GPIs recognized by GPIT. This speculation is consistent with the data presented in Fig. 2 that indicate that trimannosylated GPIs (H3 and H4) lacking EtN-P altogether are not recognized by GPIT. We wanted to refine our approach to this question by asking whether a fully mannosylated GPI lacking the EtN-P cap but possessing EtN-P residues elsewhere could be recognized by GPIT in our pulldown assay. To do this we took advantage of the fact that 1,10-phenanthroline (PNT), a commonly used membrane-permeant inhibitor of metalloproteases, is an effective inhibitor of GPI EtN-P transferases in mammalian cells (43). As reported previously, inclusion of 250 M PNT in the culture medium, while metabolically labeling HeLa cells with [2-3 H]mannose, blocked production of radiolabeled H8 and caused the accumulation of [ 3 H]mannoselabeled H6, a trimannosylated GPI intermediate with only one EtN-P residue linked to Man1 (Fig. 3A; compare with the lipid profile shown in Fig. 2A). A species corresponding to M4-H6 was also labeled, as were the Jb␣m-sensitive lipids H4, H3, and H2. The identity of H6 and M4-H6 was confirmed by diagnostic enzymatic treatments. Both H6 and M4-H6 were completely FIGURE 2. Co-immunoprecipitation of H8, but not H3 or H4, with GPIT. HeLa cells were transfected with cDNAs corresponding to FLAG-and V5-tagged versions of the five GPIT subunits; 48 h later, the cells were metabolically radiolabeled with [2-3 H]mannose for 2 h. The cells were lysed in digitonin-containing buffer, and the clarified digitonin extract was treated with anti-FLAG and anti-V5 beads to immunoprecipitate GPIT complexes. The GPIT IP and post-IP supernatants were subjected to lipid extraction and TLC analysis; lipid extracts corresponding to ϳ95% of the immunoprecipitated material (B) and ϳ1.5% of the post-IP supernatant (A) were analyzed by TLC. The chromatograms were visualized with a radioactivity scanner. The migration positions of characterized GPI intermediates are indicated. The structure of H8 is shown in the lower panel (symbols are the same as in Fig. 1).

FIGURE 3. EtN-P on the third mannose is not critical for GPI recognition by GPIT.
HeLa cells expressing epitope-tagged GPIT subunits were metabolically labeled with [2-3 H]mannose for 2 h. Cells were pretreated for 30 min with 250 M PNT or 10 mM ManN before labeling, and the same concentrations of PNT or ManN were used during labeling. Cell lysates were prepared, and GPIT complexes were immunoprecipitated. GPI extracts corresponding to ϳ95% of the immunoprecipitated material and ϳ1.5% of the non-immunoprecipitated material (post-IP supernatant) were subjected to lipid extraction and TLC analysis. Chromatograms were visualized with a radioactivity scanner. A and B, samples from PNT-treated cells; C and D, samples from ManN-treated cells. E, extracts from ManN-treated cells were incubated with GPI-PLD or Jb␣m and analyzed by TLC. Both MN and MN1 were completely susceptible to GPI-PLD; MN1 was partially susceptible under the reaction conditions used, giving rise to a lipid product that co-migrated with H5 (asterisk).
hydrolyzed by GPI-PLD, and after treatment with Jb␣m, both lipids were converted to a radiolabeled species that co-migrated with H5 by TLC analysis (data not shown).
HeLa cells expressing epitope-tagged variants of all five GPIT subunits were labeled with [2-3 H]mannose in the presence of 250 M PNT. After labeling, the cells were lysed in digitonincontaining buffer, and GPIT complexes were recovered from the digitonin extract by immunoprecipitation. TLC analysis of the organic solvent extract of the GPIT IP shows that both H6 and M4-H6 are co-immuoprecipitated with GPIT complexes while H4, H3, and H2 are not. This result clearly shows that the EtN-P cap of complete GPI anchor precursors is not required for GPI recognition by GPIT.
We previously described a point mutant of human GAA1 (GAA1(P609L)) that interacts with other GPIT subunits to form a non-functional GPIT complex (36). GAA1(P609L)-containing GPIT complexes bind proproteins but do not bind H8. These non-functional complexes similarly do not interact with H6 (data not shown), suggesting that both H6 and H8 interact with GPIT in the same way.
EtN-P-containing GPIs with an Incomplete Glycan Core Are Recognized by GPIT-To test whether EtN-P-containing GPI intermediates with an incomplete glycan core structure can be recognized by human GPIT, we used mannosamine (ManN) to inhibit GPI biosynthesis. ManN inhibits in vivo GPI anchoring of proteins in a variety of cell types by preventing addition of the third mannose to the core glycan of GPI (44 -48). The mechanism by which ManN inhibits GPI biosynthesis is not entirely clear. It has been proposed that it acts as a chain terminator once it is incorporated into the GPI structure; alternatively, ManN has been proposed to be an inhibitor of the third GPI-mannosyltransferase.
HeLa cells expressing epitope-tagged GPIT subunits were preincubated with 10 mM ManN and then metabolically labeled with [2-3 H]mannose in the continued presence of 10 mM ManN. Cells were lysed in digitonin-containing buffer, and GPIT IP and post-IP supernatant fractions were generated and analyzed. TLC analysis of the post-IP supernatant revealed a major radiolabeled species (termed MN) that migrated faster than H6 (R f (MN) ϳ0.65 versus R f (H6) ϳ0.6) (Fig. 3C) and another, slower moving lipid species termed MN1. Both MN and MN1 could be quantitatively hydrolyzed by GPI-PLD indicating that they are GPIs (Fig. 3E). MN1 was resistant to Jb␣m, whereas MN was partially degraded to a product that co-migrated with H5 by TLC (Fig. 3E). These data suggest that MN has the structure ManN-(EtN-P)Man-GlcN-acyl-PI, (Fig. 1B) consistent with previous proposals (45,48). The exact structure of MN1 remains to be determined. The properties of MN1 (mobility on TLC comparable with that of H7 and resistance to Jb␣m) suggest that it could have the structure ManN-Man-Man-GlcN-acyl-PI with EtN-P residues on the terminal ManN and the first mannose. Both MN and MN1 were present in the GPIT IP (Fig. 3D), indicating that they are recognized by GPIT. We conclude that EtN-P-containing GPIs with an incomplete or altered glycan core are recognized by GPIT.
A Minimal EtN-P-containing GPI Is Recognized by Human GPIT-Since GPIT can bind incomplete EtN-P-containing GPIs such as H6 and MN, but not other incomplete structures such as H4, H3, and H2 that lack EtN-P (Figs. 2 and 3, A and B), we wondered if GPIT could interact with H5, the minimal EtN-P-containing GPI synthesized in mammalian cells. Since it is difficult to detect H5 in lipid extracts of [ 3 H]mannose-labeled HeLa cells, we used in vitro labeling with GDP-[ 3 H]mannose instead. Membranes prepared from FLAG-GAA1-expressing HeLa cells were incubated with GDP-[ 3 H]mannose for 45 min. GPIT IP and post-IP supernatants were generated from a digitonin extract of the labeled membranes. TLC analysis of lipids in the post-IP supernatant indicated H8, as well as a major peak of radiolabeled lipid that migrated with the R f expected for H5 (Fig. 4A). This lipid was confirmed to be H5 by demonstrating its sensitivity to GPI-PLD and resistance to Jb␣m (Fig. 4D). Both H5 and H8 were pulled down with GPIT (Fig. 4B). Fig. 4C shows an immunoblot of the GPIT IP, indicating the presence of all three subunits that were probed with available antibodies. The efficiency of co-precipitation with GPIT was ϳ3-fold greater for H8 than for H5 (note the ratio of H8/H5 peaks in the post-IP supernatant versus the GPIT IP), indicating that even though the minimal H5 structure is recognized by GPIT, the enzyme may prefer the complete GPI structure. For purposes of comparison, we note that the efficiency of co-precipitation is ϳ2-fold greater for H8 than for H6 in extracts where both lipids are present (data not shown). We conclude that H5, a minimal EtN-Pcontaining GPI, is recognized by human GPIT.

Reconstitution of the Interaction between GPIs and GPIT Points to a Critical Role for the EtN-P Residue Linked to Man1-
We considered the possibility that EtN-P linked to any of the mannose residues of the GPI core glycan, not just Man1, would enhance recognition of the GPI structure by GPIT. None of the data presented thus far address this possibility since the only GPIs that we were able to generate and test through metabolic or in vitro radiolabeling lack EtN-P altogether (these structures do not bind GPIT) or have EtN-P residue linked to Man1. To circumvent this problem, we developed methods to reconstitute the interaction of GPIT with GPIs to test GPIs of suitable structure. GPIs synthesized by Trypanosoma brucei lack EtN-P on Man1 and Man2. We were interested in testing the interaction between human GPIT and the trypanosome GPIs PP1 and PP3 (Fig. 1B) (39,40) that have a single EtN-P linked to Man3.
Human GPIT complexes were purified from Nonidet P-40 extracts of HeLa cells expressing FLAG-GAA1 (or FLAG-GAA1(TM1-3), a GAA1 mutant lacking the last 4 transmembrane (TM) domains of the protein). Nonidet P-40 was used since it strips GPIs from GPIT (36). The bead-bound GPIT complexes were transferred to a digitonin-containing buffer (permissive for GPI binding) and incubated with detergent-solubilized radiolabeled GPIs. After washing, bound GPIs were extracted and analyzed. Fig. 5B shows immunoblot analyses of GPIT complexes used for this experiment. In addition to wild-type GAA1 or mutant FLAG-GAA1(TM1-3) as indicated, the anti-FLAG immunoprecipitates contained GPI8 and PIG-T as expected (PIG-S and PIG-U are also likely to be present in these complexes but the non-availability of antibodies against these two subunits precluded confirmation of this point). We previously showed that FLAG-GAA1(TM1-3) interacts with other subunits of GPIT but that the resulting GPIT complex is non-functional (37).
We first tested the interaction between H8 and GPIT. A lipid extract containing [ 3 H]mannose-labeled H8 (Fig. 5A, panel a) was incubated with bead-bound GPIT complexes. Fig. 5A, panels d and g, show that H8 can be co-precipitated with GPIT complexes containing wild-type GAA1 but not with complexes containing GAA1(TM1-3). Thus the reconstitution protocol recapitulates the result seen in pulldown assays with metabolically labeled cells (Fig. 2). The inability of complexes containing GAA1(TM1-3) to interact with H8 reaffirms the importance of GAA1 for GPI binding that we reported previously (36). We also tested whether GlcNAc-PI, the first GPI biosynthetic intermediate, could bind GPIT. An extract containing [ 3 H]GlcNAc-PI (Fig. 5A, panel c) was incubated with beadbound GPIT complexes under the same conditions used for the experiment with H8 described above. No binding was observed (Fig. 5A,  panel f ). This result is consistent with data on H4, H3, and H2 described above and also provides a specificity control for the pulldown assay.
The trypanosome GPI preparation (Fig. 5A, panel b, shows the TLC profile of radiolabeled lipids tested) showed weak but significant binding; low levels of PP1 and PP3 were recovered with functional GPIT complexes (Fig. 5A, panel e), but neither lipid was pulled down with GPIT complexes containing GAA1(TM1-3) (Fig. 5A, panel h). These data suggest that although a complete GPI structure lacking EtN-P on Man1 can be recognized, albeit weakly, by human GPIT, binding is greatly enhanced when the GPI structure contains this critical EtN-P residue.

DISCUSSION
The results presented in this paper define the structural features of the GPI substrate that are important for its interaction with human GPIT. The main findings of this study are that the terminal EtN-P is not required for GPI binding to GPIT and that partially mannosylated, EtN-P-modified GPI biosynthetic intermediates can be recognized by GPIT. Examination of the ability of GPIT to pull down a variety of in situ synthesized or exogenously supplied GPIs suggests that the GPI lipid must have at least one EtN-P residue to be recognized and that recognition is by far the strongest when that residue is located on the first mannose of the GPI glycan.
Our data are consistent with in vivo studies of PIG-N/ Mcd4p, the transferase that attaches EtN-P to the first mannose in mammalian cells and yeast (18 -21). Hong et al. (21) reported the characterization of a mouse embryonal carcinoma cell line in which the Pig-N gene had been disrupted. The Pig-N-knockout cells synthesized the entire spectrum of GPIs seen in wildtype cells, except that the GPIs lacked EtN-P on Man1. The Pig-N-knock-out cells also continued to express the GPI-anchored protein Thy-1; however, the expression level of Thy-1 per cell, and the number of Thy-1-expressing cells, were both reduced by an order of magnitude. 4 This suggests that complete 4 The authors claim that Thy-1 expression was only partially affected in Pig-Nknock-out cells. While this conclusion is technically correct, closer examination of their data indicate that the effect is substantial: roughly 10-fold fewer Pig-N-knock-out cells express Thy-1, and for the cells that do, surface expression levels of the protein are ϳ10-fold lower than wild type. GPI intermediates lacking the initial EtN-P modification are inefficiently transferred to proteins in mammalian cells, consistent with the biochemical analyses that we report here. It was recently reported that a yeast strain (Mcd4⌬/ TbGPI10) that synthesizes complete GPIs lacking EtN-P on Man1 accumulates these GPI anchor precursors and displays lower levels of [ 3 H]inositol incorporation into proteins (19). This result indicates a reduction in efficiency of GPI anchoring caused by the absence of EtN-P on the first mannose. We suggest that yeast GPIT, like human GPIT, processes GPIs more efficiently if they display EtN-P on the first mannose.
Our results suggest that human and trypanosome GPIT complexes are likely to recognize somewhat different aspects of GPI since trypanosomatids do not modify the first mannose of their GPI structures with EtN-P. Trypanosome GPIT is a pentasubunit complex consisting of three subunits that resemble those found in mammalian systems (TbGPI8, TbPIG-T/TbGPI16, TbGAA1) and two subunits (TTA1 and TTA2) that differ (49). Using the approaches described here it should be possible to identify GPI features that are recognized by trypanosome GPIT and to exploit the difference in subunit architecture between human and trypanosome GPIT complexes to determine the role of individual subunits in GPI recognition.
Where does GPI bind within the human GPIT complex? We previously showed that GAA1 is involved in GPI recognition by GPIT, since human GPIT complexes containing a truncated form of GAA1 are able to interact with pro-proteins but unable to interact with GPIs (36). It is possible that GAA1, perhaps in conjunction with PIG-U (also implicated in GPI recognition (28)), binds GPI to channel or present the lipid to GPI8. It can be envisaged that binding of GPI to GAA1/PIG-U via the innermost EtN-P side chain would leave the terminal EtN-P free to carry out its role in the second step of the transamidation reaction. Our ability to reconstitute the GPI-GPIT interaction provides an opportunity to identify GPI-binding subunits within GPIT by using synthetic GPIs bearing a photocross-linkable prosthetic group.