Deletion of GPI7, a Yeast Gene Required for Addition of a Side Chain to the Glycosylphosphatidylinositol (GPI) Core Structure, Affects GPI Protein Transport, Remodeling, and Cell Wall Integrity*

Gpi7 was isolated by screening for mutants defective in the surface expression of glycosylphosphatidylinositol (GPI) proteins. Gpi7 mutants are deficient in YJL062w, herein named GPI7. GPI7 is not essential, but its deletion renders cells hypersensitive to Calcofluor White, indicating cell wall fragility. Several aspects of GPI biosynthesis are disturbed in Δgpi7. The extent of anchor remodeling, i.e. replacement of the primary lipid moiety of GPI anchors by ceramide, is significantly reduced, and the transport of GPI proteins to the Golgi is delayed. Gpi7p is a highly glycosylated integral membrane protein with 9–11 predicted transmembrane domains in the C-terminal part and a large, hydrophilic N-terminal ectodomain. The bulk of Gpi7p is located at the plasma membrane, but a small amount is found in the endoplasmic reticulum. GPI7 has homologues inSaccharomyces cerevisiae, Caenorhabditis elegans, and man, but the precise biochemical function of this protein family is unknown. Based on the analysis of M4, an abnormal GPI lipid accumulating in gpi7, we propose that Gpi7p adds a side chain onto the GPI core structure. Indeed, when compared with complete GPI lipids, M4 lacks a previously unrecognized phosphodiester-linked side chain, possibly an ethanolamine phosphate. Gpi7p contains significant homology with phosphodiesterases suggesting that Gpi7p itself is the transferase adding a side chain to the α1,6-linked mannose of the GPI core structure.

resent a subclass of surface proteins found in virtually all eukaryotic organisms (1). The genome of Saccharomyces cerevisiae contains more than 70 open reading frames (ORFs) encoding for proteins that, as judged from the deduced primary sequence, can be predicted to be modified by the attachment of a GPI anchor (2,3). In about 25 of them, the presence of an anchor has been confirmed biochemically. A majority of them lose part of the anchor and become covalently attached to the ␤1,6-glucans of the cell wall (4 -6). A minority of GPI proteins retain the GPI anchor in an intact form and stay at the plasma membrane (PM).
For the biosynthesis of GPI anchors, phosphatidylinositol (PI) is modified by the stepwise addition of sugars and ethanolamine phosphate (EtN-P), thus forming a complete precursor lipid (CP) which subsequently is transferred en bloc by a transamidase onto newly synthesized proteins in the ER (7,8). The identification of genes involved in the biosynthesis of the CP and its subsequent attachment to proteins has been possible through the complementation of mammalian and yeast gpi Ϫ mutants, i.e. mutants being deficient in GPI anchoring of membrane proteins (7, 9 -20). In our laboratory, a series of recessive gpi Ϫ mutants (gpi4 to gpi10) has been obtained by screening for yeast mutants that are unable to display the GPI-anchored ␣-agglutinin (Sag1p) at the outer surface of the cell wall, although the synthesis and secretion of soluble proteins is normal (21,22).
Materials were obtained from the sources described recently (22). Cysteamine was from Sigma; [ 3 H]dihydrosphingosine was synthesized as described (24) Cloning, Partial Sequencing, and Disruption of GPI7-The GPI7 gene was cloned by complementation of the ts growth phenotype of the gpi7-1/gpi8-1 double mutant as described (19). The three plasmids complementing gpi7-1 contained a 4.3-kb common DNA restriction fragment that was partially sequenced by the dideoxy sequencing method (25). The complementing insert SphI/SspI of 3.6 kb was cloned into the SphI/SmaI-digested YEp352 multicopy vector (26) or YCplac33 single copy vector (27) to generate pBF41 (Fig. 3C) and pBF43, respectively.
One step disruption of GPI7 was done as described (28). Briefly, the 1.5-kb long KanMX4 module was PCR-amplified by using pFA6a-KanMX4 as template and the following two adapter primers: GPI7forwards (5Ј-CTTCACCAAGTTAGCAAGATGAACTTGAAGCAGTTC-ACGTGCCtcgatgaattcgagctc-3Ј) with 17 nucleotides (nt) of homology to the pFA6a-KanMX4 multiple cloning site (in lowercase) and 43 nt of homology to GPI7 (in uppercase) starting 18 nt upstream of the start codon (bold); GPI7-backwards (5Ј-ATCAAGAGCGCAAAGGAGGGCC-AATTCAGGTAACCAGCCATTCAcgtacgctgcaggtcgac-3Ј) with 18 nt of homology to the pFA6a-KanMX4 multiple cloning site (lowercase) and 44 nt of homology to the ORF of GPI7 in the region immediately upstream of the stop codon. This PCR DNA fragment was used to transform the diploid strain FBY118, homozygous for GPI7, and FBY40, a heterozygous gpi7-1/GPI7 strain. The correct targeting of the PCRmade KanMX4 module into the GPI7 locus in geneticin-resistant clones was verified by PCR with whole yeast cells using primers GPI7-plus (5Ј-GTTCATCTACCACGCAC-3Ј) starting 36 nt upstream of the ATG, GPI7-minus (5Ј-GACCCAAGTAATGCAGG-3Ј) starting 631 base pairs downstream of the ATG and the K2 primer of the KanMX4 module (5Ј-GTATTGATGTTGGACG-3Ј).
Purification of Recombinant His-tagged Gpi7p and Antibody Production-Plasmid pBF41 (Fig. 3C) was digested with BstYI and EcoRV to generate a 633-base pair fragment of GPI7. This fragment was inserted into the multiple cloning site of the bacterial expression vector pQE-30 (Qiagen) digested with BamHI/SmaI thus generating the plasmid pBF402. This plasmid was used to transform the E. coli strain M15[pREP4]. Expression of the recombinant protein was induced with isopropyl-1-thio-␤-D-galactopyranoside and purified on a nickel-nitrilotriacetic acid-agarose column (Qiagen) under denaturing conditions according to the manufacturer's instructions. A polyclonal antiserum was raised against this Gpi7p fragment by repeated intramuscular injections of 100 g of recombinant protein into a rabbit. Ten mg of the recombinant protein were coupled to CNBr-activated Sepharose 4B (Amersham Pharmacia Biotech) according to the manufacturer's instructions, and antiserum was affinity purified as described (29).
Membrane Association, Protease Sensitivity, and Cellular Localization of Gpi7p-The nature of the association of Gpi7p with the membrane was determined following a previously described protocol (30) except that the EDTA concentration in buffer G was increased from 2 to 20 mM. Gpi7p protease sensitivity was examined by proteinase K digestion of microsomes essentially as described (31). Briefly, 100 A 600 of washed W303 cells were resuspended in 1 ml of lysis buffer (20 mM HEPES, pH 7.5, 500 mM sucrose, 3 mM magnesium acetate, 20 mM EDTA, 1 mM dithiothreitol) and were lysed by agitation with glass beads at 4°C. The homogenate was centrifuged for 5 min at 600 ϫ g to remove unbroken cells, and the supernatant was centrifuged for 15 min at 13,000 ϫ g. The membrane pellet was resuspended in 240 l of the same lysis buffer and split into 6 aliquots of equal size. Aliquots of microsomes were incubated for 20 min on ice with or without 0.5% Triton X-100 and proteinase K. Digestions were stopped by addition of phenylmethylsulfonyl fluoride (final concentration 4 mM, added from a 200 mM stock in ethanol) and kept on ice for an additional 10 min before being boiled in sample buffer (32). The subcellular localization of Gpi7p was determined essentially as described (33). Briefly, 500 A 600 units of mid-log phase W303-1B cells were broken by agitation with glass beads in 200 mM sorbitol, 25 mM PIPES, pH 6.8, 50 mM KCl, 5 mM NaCl, 10 mM EDTA, 10 mM NaN 3 /NaF, 1 mM phenylmethylsulfonyl fluoride, leupeptin, pepstatin, and antipain, each at 30 g/ml). After removal of the unbroken cells the homogenate was centrifuged for 10 min at 8,000 ϫ g at 4°C to generate pellet P8 and supernatant S8. S8 was divided in two and either precipitated by the addition of trichloroacetic acid to 10% or centrifuged at 100,000 ϫ g for 1 h to generate pellet P100. For zymolyase treatment the cells were washed and resuspended at 50 A 600 units/ml in zymolyase buffer (1.2 M sorbitol, 50 mM K 2 HPO 4 pH 7.5, 40 mM 2-mercaptoethanol, 20 mM EDTA, 10 mM NaN 3 , 10 mM NaF) containing zymolyase 20T. After a 40-min incubation at 30°C, cells were placed on a cushion of 1.5 M sorbitol, 50 mM K 2 HPO 4 , pH 7.5, 20 mM EDTA, 10 mM NaN 3 , 10 mM NaF and were centrifuged. For Western blotting, all the samples were denatured during 5 min at 95°C in reducing sample buffer and run on a 6, 7.5, or 10% SDS-PAGE for detection of antigens, respectively (32). Western blotting was carried out with anti-Wbp1p, anti-alkaline phosphatase, anti-Gas1p, or anti-Kex2p antisera or with affinity purified anti-Gpi7p or anti-Och1p antibodies, always using the chemiluminescence ECL kit from Amersham Pharmacia Biotech, Buckinghamshire, UK.

H]DHS, or [2-3 H]Ins
Analytical Methods-Lipid extracts were deacylated with NaOH (34) and treated with JBAM (35) as described. For GPI-PLD treatment lipid extracts were dissolved in 20 mM Tris-HCl, pH 7.4, 0.1 mM CaCl 2 , 20% 1-propanol. Incubations were for 12 h at 37°C. All treated lipid extracts were desalted by partitioning between n-butyl alcohol and an aqueous solution of 0.1 mM EDTA, 5 mM Tris-HCl, pH 7.5, and back extraction of the butanol phase with water before TLC (23). Lipids were treated with methanolic NH 3 to remove the acyl group on Ins (36) and cleaved using nitrous acid (37) as described. Lipids were purified by preparative TLC on 0.2-mm thick Silica Gel 60 plates (Merck, Germany) in solvent 2. Radioactive spots were localized by radioscanning, scraped, and eluted with solvent 2. A second run on TLC was done to obtain radiochemically pure M0, M4, and CP2.
Limiting HF Treatment of Head Groups-For limiting HF treatment, aliquots of radiolabeled head groups derived from CP2 and M4 and prepared as above were dephosphorylated with 50 l of 48% aqueous HF at 0°C as described (38) for 0 -28 h. After neutralization with saturated LiOH, samples were desalted by gel filtration through an 8-ml Sephadex G-10 (Amersham Pharmacia Biotech) column. Samples were then dried in the Speed-Vac and treated with JBAM prior to complete HF dephosphorylation (60 h, 0°C). Samples were neutralized again with LiOH and N-acetylated. Aliquots were dried and then directly applied to Whatman paper No. 1M and analyzed by descending chromatography as described above.

RESULTS
Cloning of GPI7-As reported before (19) and shown in Fig.  2, wild type (wt) cells do not contain polar GPIs (lane 1), gpi8-1 accumulates CP2 as the most polar GPI lipid (lane 8), and gpi7-1 and the gpi7-1/gpi8-1 double mutant accumulate M4 (lanes 4 and 6), thus demonstrating that gpi7-1 is epistatic to gpi8-1 and suggesting that, during GPI biosynthesis, Gpi7p may act before Gpi8p. Although the original gpi7 mutants and the unrelated gpi8-1 mutant were not significantly temperature-sensitive (ts) for growth, the growth of the gpi7-1/gpi8-1 double mutant was strongly temperature-dependent. Transfection of a genomic library into this double mutant allowed the isolation of clones containing complementing plasmids (19).
These clones were labeled with myo-[ 3 H]inositol ([ 3 H]Ins) at 37°C, and the lipids were extracted and analyzed by TLC. Upon transfection some gpi7-1/gpi8-1 indeed had regained the ability to make CP2 (Fig. 2, lanes 6 and 7) and showed the same lipid profile as gpi8-1 (Fig. 2, lane 8). All these clones harbored plasmids containing YJL062w as the only complete ORF. Transfection of a multicopy vector containing YJL062w under its own promoter (pBF41, Fig. 3C) into gpi7-1 almost completely cured the accumulation of M4 (Fig. 2, lane 5). As expected, the accumulation of CP2 by gpi8-1 was not abolished by the overexpression of YJL062w (Fig. 2, lanes 8 and 9). YJL062w predicts an 830-amino acid membrane protein with an N-terminal signal sequence for insertion into the ER, 5 potential N-glycosylation sites, and about 9 -11 putative transmembrane domains (Fig. 3, A and B). YJL062w was deleted and replaced by the selectable marker KanMX4. On rich medium the deletants grew about as rapidly as wt cells at all temperatures. Thus, YJL062w is not an essential gene. We were unable to sporulate ⌬YJL062/⌬YJL062 diploids indicating that YJL062 is required for sporulation. However, ⌬YJL062/YJL062 heterozygotes sporulated readily and ⌬YJL062 spores germinated normally. In accordance with previous results on gpi7 mutants (21), growth of ⌬YJL062 (ϭ ⌬gpi7, see below) on plates at 37°C was severely inhibited by 0.5 mg/ml Calcofluor White. ⌬YJL062 accumulated M4 at even higher levels than gpi7-1, and this accumulation was almost completely suppressed by the transfection of pBF41 (Fig. 2,  lanes 2 and 3). Residual accumulation of M4 may be due to 2 I. Flury, unpublished observations. some cells that lost the complementing plasmid. Transfection of YJL062w under its own promoter on a single copy vector (plasmid pBF43) was sufficient to suppress the accumulation of M4 in a homozygous ⌬YJL062/⌬YJL062 diploid (Fig. 2, lanes 11 and 12). As can be seen in Fig. 2, gpi7-1, ⌬YJL062, ⌬YJL062/ ⌬YJL062, and gpi8-1 mutants also show minor amounts of the GlcN␣1,6(acyl3)Ins-P-DAG GPI intermediate M0, the accumulation of which is believed to reflect a build up of GPI intermediates throughout the biosynthetic pathway (Fig. 2, lanes 2, 4, 6, 8, and 11). (It should be noted that some intermediates of intermediate size are obscured on TLC by PI and inositol phosphoceramide (41).) As expected, expression of YJL062w abolishes the accumulation of M0 in gpi7-1 and ⌬YJL062 (Fig. 2, lanes 3, 5, and 12) but not in gpi7-1/gpi8-1 nor gpi8-1 (lanes 7 and 9), since in the latter the GPI biosynthesis remains blocked. To evaluate if the mutation in gpi7-1 is genetically linked to YJL062w, YJL062w was disrupted in a heterozygous gpi7-1/GPI7 diploid. Correct replacement of one YJL062w locus was verified by PCR in two independent geneticin-resistant transformants. The verified deletants were sporulated, and a total of 26 complete tetrads was labeled with [ 3 H]Ins to analyze the accumulation of M4. In all 26 tetrads only two of the four segregants showed accumulation of M4, whereas the other two showed the lipid profile of wt cells. Geneticin resistance also segregated 2:2 and cosegregated with M4 accumulation in all cases. This demonstrates that the mutation of gpi7-1 is tightly linked to YJL062w which we henceforth call GPI7. Since a construct containing only 348 nucleotides 5Ј of the initiation codon of GPI7 still retained significant complementing activity, we also can dismiss the possibility that the complementing activity of pBF41 is due to one of the two small ORFs located on the opposite strand in the 5Ј upstream region of GPI7 and starting at Ϫ409 and Ϫ503 with regard to the start codon of GPI7.
Characterization of the GPI Intermediate M4 -We found that M4, contrary to our initial expectation, contained an HFsensitive group on Man3 (Fig. 1). Indeed, treatment of the lipid extracts of gpi7-1 with jack bean ␣-mannosidase (JBAM, an exomannosidase) shifted M4 to a slightly less hydrophilic position on TLC (Fig. 4A, lanes 1 and 2) but not to the position of M0. It seemed conceivable that JBAM did not remove more than one Man from M4 because it was sterically hindered by the detergent micelle in which M4 was embedded. To circumvent this problem, M4 was purified by preparative TLC, and its hydrophilic head group was liberated by GPI-PLD, O-deacylated by NH 3 , and then subjected to several treatments as indicated at the top of Fig. 5, A-D. The N-acetylated fragment comigrated with the Man 4 -GlcNAc-Ins standard (Fig. 5A). When treated with JBAM before HF, the resulting N-acetylated fragment comigrated with the Man 3 -GlcNAc-Ins standard, clearly indicating the presence of a blocking group on Man3 (Fig. 5B). The blocking group on Man3 was HF-sensitive, since JBAM done after HF produced a fragment comigrating with GlcNAc-Ins (Fig. 5C). Aspergillus satoi ␣-mannosidase (ASAM), a linkage-specific ␣1,2-exomannosidase, when used after HF treatment, produced Man 2 -GlcNAc-Ins (Fig. 5D). The migration of the fragments shown in Fig. 5, A and B, was much slower when N-acetylation was omitted (not shown). This partial characterization of M4 is consistent with the presence of a classical Man␣1,2(EtN-P3)Man␣1,2Man␣1,6Man␣1,4-Glc-N␣1,6Ins core structure.
Having recently discovered an additional EtN-P on Man1 of CP2 (22), we considered the possibility that M4 may be lacking EtN-P on Man1. We thus proceeded to compare the non-dephosphorylated head groups of M4 and CP2 by Dionex HPLC using a system in which the presence of negatively charged phosphodiesters greatly retards the elution of oligosaccharides (42). The non-dephosphorylated head groups of M4 and CP2 eluted as sharp peaks at fractions 22 and 31, respectively (not shown). This wide separation suggested that the head group of M4 contains less negative charge than the one of CP2. To assay directly for a side chain on Man1 of M4, the head group of [ 3 H]Ins-labeled M4 was first cleaved by acetolysis, a procedure which, under mild conditions, selectively cleaves ␣1,6-glycosidic bonds (Fig. 1). Here this procedure is expected to produce the labeled fragment (X-P3?)Man␣1,4-GlcN␣1,6-[ 3 H]Ins with X-P-being the substituent in question. The fragment was then either treated with JBAM or control incubated and finally dephosphorylated by HF, N-acetylated, and analyzed by paper chromatography. As can be seen in Fig. 6, A and B, the (X-P3?)Man␣1,4GlcN␣1,6Ins fragment of M4 is JBAM-resistant, since successive treatment by acetolysis, JBAM, and then HF generates Man␣1,4GlcN␣1,6Ins. The same had previously been found for CP2 (22). Thus, the difference between the head groups of M4 and CP2 cannot be explained by the presence or absence of an HF-sensitive substituent on Man1: both lipids have the same classical Man 4 -GlcN-Ins carbohydrate core structure, they both contain HF sensitive groups on Man1 and Man3 (Fig. 1), but they migrate differently on TLC, and their non-dephosphorylated head groups elute differently on Dionex . Lipid extracts were treated (ϩ) or control incubated (Ϫ) with either JBAM or GPI-PLD or were deacylated by mild base treatment (NaOH). Desalted products were analyzed by TLC (solvent 2) and fluorography. B, sec53 and ⌬gpi7 were preincubated and labeled with [ 3 H]Ins at 37°C (5 Ci/A 600 ), and lipid extracts were treated with PI-specific phospholipase C to get rid of labeled PI, and the GPI intermediates M0 and M4 were purified from sec53 and ⌬gpi7, respectively, by two rounds of preparative TLC. Anchor peptides were prepared from [ 3 H]Ins-labeled W303 in the experiment described in Table I. M4 and anchor peptides were treated with HNO 2 to liberate the acyl-[ 3 H]Ins-P-lipid and [ 3 H]Ins-P-lipid moieties, respectively (Fig. 1). Samples were incubated with (ϩ) or without (Ϫ) methanolic NH 3 to remove the acyl from the Ins, desalted and separated by TLC (solvent 1), and processed for fluorography. pG1, protein-derived glycerophospholipid 1; pC1 and pC2, protein-derived ceramides 1 and 2, see Sipos et al. (41). Other results of this same experiment were described before (Ref. 41 HPLC (data presented above and in Refs. 22 and 23). We thus hypothesized that CP2 may contain either additional HFsensitive groups on GlcN, Man1, Man2, or Man3 or may contain additional groups linked through the amino group of the EtN-P on M1. Of these several theoretical possibilities, the only ones that have been documented in other organisms are the Man1-P group on the GlcN in Paramecium aurelia (43) and the EtN-P group on Man2 in several mammalian GPI proteins, e.g. human erythrocyte acetylcholinesterase (44), CD52-II (45), and bovine liver 5Ј-nucleotidase (46). We used limiting HF treatment to test specifically if CP2 contains an HF-sensitive group on Man2. If we assume that during HF treatment the EtN-Ps are hydrolyzed in a random order, we may expect to find some reaction intermediates lacking the HF-sensitive group on Man3 while retaining EtN-P on Man2 or Man1. When such intermediates subsequently are treated with JBAM and then are dephosphorylated to completeness with HF, they should yield Man 2 -GlcN-Ins and Man 1 -GlcN-Ins fragments, respectively. For preliminary tests, CP2 head groups were first treated with HF for 0.5, 1, 3, 9, 12, 18, or 28 h, then with JBAM, and finally with HF for 60 h. These experiments showed that both Man 2 -GlcN-Ins and Man 1 -GlcN-Ins fragments became visible after 1 h of limiting HF treatment, peaked at 12 h, and remained detectable at all time points up to 28 h. In quantitative terms it appeared that Man 3 -GlcN-Ins Ͼ Man 2 -GlcN-Ins Ն Man 1 -GlcN-Ins at all time points. Importantly, treatment of the head group of CP2 with HF for 12 h yielded substantial amounts of Man 2 -GlcN-Ins and Man 1 -GlcN-Ins (Fig. 6C), whereas the identical treatment performed with the head group of M4 only yielded Man 1 -GlcN-Ins but no Man 2 -GlcN-Ins (Fig. 6D). This result is compatible with the idea that ⌬gpi7 cells are unable to add an HF-sensitive group onto Man2 of the GPI core (Fig. 1). It also confirms the presence of an HF-sensitive group on Man1 of both M4 and CP2.
The Lipid Moieties of GPI Intermediates in ⌬gpi7 Are Normal-We looked for additional differences between M4 and CP2 by analyzing the lipid moiety of M4. M4 is sensitive to GPI-specific phospholipase D (GPI-PLD) and mild base treatment (Fig. 4A, lanes 3-6), suggesting that its lipid moiety consists of Ins-P-DAG. We previously reported that M4 is resistant to PI-specific phospholipase C (21). This finding, together with the GPI-PLD sensitivity, can be taken as an indication for the presence of an acyl moiety attached to the Ins of M4. We further released the (acyl3)Ins-P-DAG moiety of M4 with HNO 2 as described recently (41). As shown in Fig. 4B, the treatment of purified M4 by HNO 2 produced a very hydrophobic species, which migrates very closely to M0, i.e. the GlcN(acyl3)Ins-P-DAG accumulating in sec53 (Fig. 4B, lanes 7 and 9) (41). (As reported previously, the presence of GlcN on these early precursors does not significantly influence their migration in TLC, for discussion see Sipos et al. (41).) Partial deacylation of the M4-derived lipid moiety by NH 3 produced PI and lyso-PI (Fig. 4B, lane 10). This PI was compared with pG1, the PI species obtained by HNO 2 treatment of protein-bound GPI anchors from the corresponding wt strain (Fig. 4B, lane  11). The comparison shows that M4 contains a PI moiety that migrates clearly less than pG1, whereas a lyso-PI of M4 migrates slightly more than the lyso-PI species generated by methanolic NH 3 treatment of anchor peptides (Fig. 4B, lanes 10  and 12). Very similar results had been obtained previously when comparing protein-derived PI moieties with the PI moieties of M0 from sec53 and of CP2 from gpi8-1 (41). In addition we isolated from ⌬gpi7 the recently identified GPI intermediates that are obscured in TLC by PI and inositol phosphoceramides (Ref. 41, therein Fig. 6A), and we found that they are exactly the same as the corresponding intermediates from wt cells by all criteria (not shown). Thus, it seems that M4 and other GPI intermediates of ⌬gpi7 contain the same PI moiety as early and late GPI intermediates accumulating in other mutants or in wt cells, and we therefore conclude that the difference between CP2 and M4 is solely due the difference in their head groups.
Lack of Gpi7p Affects the in Vitro Biosynthesis of GPI Precursor Lipids-When yeast microsomes are incubated in the presence of UDP-[ 3 H]GlcNAc, ATP, coenzyme A, GDP-Man, and tunicamycin, they generate labeled GPI intermediates as the only kind of labeled lipids (22,47). Wild type microsomes make GPI intermediates up to CP2. Although a large array of incomplete intermediates is also generated, the pattern of labeled intermediates is fairly reproducible. When we used ⌬gpi7 microsomes, they reproducibly made all the normal intermediates down to a band comigrating with M4 of [ 3 H]Ins-labeled ⌬gpi7 cells, but they consistently failed to make CP2 (Fig. 7). When the labeled lipid extract was treated with JBAM, most of the band comigrating with M4 was shifted to a less hydrophilic position, much in the same way as seen for [ 3 H]Ins-labeled M4 (Fig. 2, lanes 1 and 2). Thus, the M4 accumulation of ⌬gpi7 can be reproduced in vitro. This result implies that the Gpi7p present in wt microsomes is functional in vitro.
Characterization of Gpi7p-Gpi7p was characterized using affinity purified rabbit antibody made against the N-terminal, hydrophilic part of GPI7 (Fig. 3B). As shown in Fig. 8A, the antibody recognized a heterogeneously glycosylated 208-kDa protein, the estimated molecular mass of various glycoforms ranging, after heavy exposure, from about 130 to 230 kDa (Fig.  8A, lane 2). The predicted mass of the protein before and after removal of the signal sequence is 94,832 and 92,207 Da, respectively. In glycosylation mutants ⌬och1/mnn1 or mnn9 which are totally or partially deficient in the elongation of N-glycans in the Golgi, Gpi7p has an estimated mass of 108 and 115 kDa, respectively (Fig. 8A, lanes 1 and 7). pmi40 has a ts deficiency in Man biosynthesis that is partial at 24°C (48). In pmi40 grown at 24°C the average mass of Gpi7p is around 150 kDa (Fig. 8A, lane 6). (This suggests that full elongation of Nglycans is not necessary for Gpi7p function since, when shifted from 24 to 37°C, pmi40 cells are able to make CP2 (23).) Tunicamycin treatment of wt or pmi40 cells resulted in the appearance of a single, relatively sharp band of an apparent molecular mass of about 83 kDa (Fig. 8A, lanes 3 and 5). The protein could also be deglycosylated to an apparent molecular mass of 86 kDa by treatment with endoglycosidase H (not shown). All these data concurrently indicate that Gpi7p contains several N-glycans that are heavily elongated in the Golgi but contains no or only few O-glycans. In the cell lysate Gpi7p was rapidly degraded by an endogenous protease which, how-ever, could be inhibited by 10 mM EDTA. Gpi7p is associated with membranes since it could be sedimented by ultracentrifugation of lysates at 100,000 ϫ g for 60 min (Fig. 8B, lanes 1  and 2). Gpi7p was neither dissociated from membranes by NaCl nor sodium carbonate at pH 11, nor urea, but it was efficiently solubilized by Triton X-100 or SDS (Fig. 8B). The presence of an N-terminal signal sequence and the large amount of N-glycans suggested that the hydrophilic N-terminal part of the protein, which contains 3 of the 5 potential N-glycosylation sites, would reside on the lumenal or ectocytoplasmic side of the membrane. We tried to confirm this orientation by protease protection assays on microsomes. As shown in Fig. 8C, neither Gpi7p nor Wbp1p, which was used as a control, were degraded by proteinase K unless microsomes were permeabilized with Triton X-100. Wbp1p has been demonstrated to be a lumenal ER protein (49), and our result thus shows that the N-terminal hydrophilic part of Gpi7p is not accessible to protease in these microsomes. It is noteworthy that proteinase K did not reduce the molecular mass of Gpi7p, thus indicating that the cytosolic loops between the predicted transmembrane domains of the C-terminal part of Gpi7p are not accessible to proteinase K in native microsomes.
The cellular localization of Gpi7p was investigated by subcellular fractionation as shown in Fig. 8D. Differential centrifugation at 8,000 and 100,000 ϫ g for 10 and 60 min, respectively, achieved satisfactory separation of the Golgi markers Kex2p and Och1p from the ER marker Wbp1p and from the vacuolar alkaline phosphatase. We were concerned that the relative amounts of these proteins in the 100,000 ϫ g pellet FIG. 7. Microsomes of ⌬gpi7 synthesize M4 and not CP2. Microsomes of W303 wild type or ⌬gpi7 were incubated with 6 Ci of UDP-[ 3 H]GlcNAc, GDP-Man, tunicamycin, and ATP for 1 h at 37°C as described (22). The glycolipid products were extracted and then run on TLC with solvent 2. The extract in lane 3 was first treated with JBAM. On the basis of the preceding analysis the band denoted with an asterisk can be presumed to be an M4 derivative in which Man4 has been removed.
FIG. 8. Membrane association, orientation, and localization of Gpi7p. A, exponentially growing cells were broken with glass beads in TEPI buffer as described (19); lysates were centrifuged at 10,000 ϫ g for 15 min at 4°C, and microsomal pellets were processed for SDS-PAGE. Cells in lanes 3 and 5 had been grown in 20 g/ml tunicamycin (Tm) for 90 min. B-D, exponentially growing W303 cells were broken with glass beads using the buffers indicated under "Experimental Procedures," and cell wall debris was removed by centrifugation at 600 ϫ g. B, aliquots of cell lysate were incubated for 30 min at 0°C with 0.5 M NaCl, 0.8 M urea, 1% Triton X-100 (TX-100), 0.1 M Na 2 CO 3 , pH 11, or 1% SDS. Subsequently membranes were sedimented by ultracentrifugation to get supernatant (S) and pellet (P) fractions. C, cell lysate was sedimented at 13,000 ϫ g for 15 min, and the membrane pellet was thoroughly resuspended and digested with 10 or 25 g/ml proteinase K (prot K) at 0°C for 20 min in the presence or absence of 0.5% Triton X-100. D, cell lysates were subjected to differential centrifugations at 8,000 and 100,000 ϫ g. These centrifugations generated pellet P8 containing ER, PM, and vacuolar membranes and pellet P100 which contains Golgi membranes. The 8,000 ϫ g supernatant was also precipitated with trichloroacetic acid (TCA). In all panels the lanes contain material derived from 1 A 600 of cells except for lanes 1 and 7 of A which contain 0.3 A 600 . may be underestimated due to ongoing proteolytic degradation during the 100,000 ϫ g spin. Therefore the supernatant of the 8,000 ϫ g spin was split whereby proteins were immediately precipitated with trichloroacetic acid in one half, and the other half was pelleted at 100,000 ϫ g. Gpi7p was exclusively found in the 8,000 ϫ g pellet (P8) and thus is associated with either the ER, the vacuole, or the PM but not with the Golgi. As shown in Fig. 9A low amounts of proteinase K added to intact cells hydrolyzed all of the mature Gpi7p. In contrast, Wbp1p and the 105-kDa ER form of Gas1p were completely resistant to this treatment unless membranes were permeabilized with Triton X-100 (Fig. 9A, lane 6). The mature 125-kDa form of Gas1p was found to be partially resistant to proteinase K, indicating either a tighter interaction of Gas1p with some cell wall components or the existence of an internal pool of Gas1p as proposed earlier (50). Crude zymolyase treatment of intact cells also removed all of the mature form of Gpi7p (Fig. 9B), whereas recombinant zymolyase left Gpi7p intact (not shown). Longer exposures showed the presence of several minor bands of smaller size all of which were also present in ⌬gpi7 cells except for a 108-kDa form of Gpi7p (Fig. 9C, lanes 7 and 8). This material seems to be an ER form in transit to the surface since it was no more detectable if cells were preincubated with either cycloheximide or tunicamycin (Fig. 9C, lanes 1-6). These data also indicate that mature Gpi7p is relatively rapidly degraded or becomes resistant to extraction with SDS. Globally these data indicate that the bulk of Gpi7p is exposed at the cell surface but that a small amount of core-glycosylated material is found in the ER in transit to the cell surface. For the moment it is unclear why Gpi7p was completely protected in microsomes, since it has been claimed that PM does not form closed vesicles upon homogenization (51). It is conceivable that Gpi7p resides in special PM subdomains that form closed vesicles upon homogenization or that centrifugation of microsomes generated proteaseresistant membrane aggregates (51).
Deletion of GPI7 Alters GPI Protein Transport and Remodeling-We previously reported on the accumulation of the immature 105-kDa ER form of Gas1p in gpi7 mutants (21). We therefore investigated GPI protein transport in ⌬gpi7. Indeed, by pulse-chase experiments we found that the maturation of GPI proteins Sag1p and Gas1p was slowed 2-3-fold as compared with wt cells, whereas the maturation of carboxypeptidase Y proceeded with normal kinetics (not shown). This indicates that the transport of GPI proteins in ⌬gpi7 is specifically retarded. Nevertheless, in rich media ⌬gpi7 cells grow at roughly the same rate as wt cells. They also incorporate [ 3 H]Ins with the same efficiency as wt cells.
The lipid remodeling of GPI anchors is significantly altered in ⌬gpi7. As seen in Table I, the proportion of ceramides (pC1 and pC2) in anchor peptides from ⌬gpi7 is drastically decreased, whereas the fraction of DAG-containing lipids ((pG1) is correspondingly increased. It should be noted that at the time of analysis, i.e. 75 min after addition of [ 3 H]Ins, the relative amounts of mild base-sensitive and mild base-resistant anchors are no longer changing and represent the steady state proportion of these two anchor types (41,52). It is important to realize that pG1 also represents a remodeled form of the anchor lipid in which a long chain fatty acid has replaced the original fatty acid present in sn-2 of the glycerol of the CP (Fig. 1). It thus appears that the relative decrease of ceramide remodeling goes along with a compensatory increase in DAG remodeling. A relative reduction in ceramide remodeling was also observed when we compared the efficiency of [ 3 H]Ins and [ 3 H]dihydrosphingosine ([ 3 H]DHS) incorporation into GPI proteins. As can be seen in Fig. 10, the ratio of [ 3 H]DHS/[ 3 H]Ins incorporation into proteins is much higher in wt than in ⌬gpi7 (Fig. 10,  lanes 1-4). The lack of incorporation of [ 3 H]DHS in ⌬gpi7 cannot be explained by an increase of the endogenous production of DHS in ⌬gpi7, since the difference between wt and ⌬gpi7 persists, even when all endogenous DHS biosynthesis is blocked by myriocin (Fig. 10, lanes 5 and 6). The defect in FIG. 9. Gpi7p is localized at the cell surface. Cells in the early exponential phase growing at 30°C in YPD were used. A, intact W303 cells were treated with cysteamine chloride and then treated with the indicated concentrations of proteinase K (prot K) exactly as described (65) except that the EDTA concentration was raised to 20 mM. In lane 6, Triton X-100 was added to 1%. B, W303 cells were treated with zymolyase 20T at the indicated concentrations. C, W303 or ⌬gpi7 cells were either lysed directly or after having been incubated for 30 or 60 min at 10 A 600 /ml in the presence of cycloheximide (200 g/ml) or tunicamycin (Tm, 20 g/ml). Cells were lysed by boiling in sample buffer and processed for SDS-PAGE and Western blotting with antibodies against Gpi7p, Gas1p, or Wbp1p. H]Ins into lipids ranged from 43 to 49% of the radioactivity added to cells. Proteins were extensively delipidated, and anchor peptides were prepared as described (41). Anchor peptides were treated with HNO 2 to liberate the [ 3 H]Ins-P-lipid moieties; the products were desalted by partitioning between butanol and water, and the labeled lipid products were separated by TLC and quantitated by radioscanning. Results of the quantitation of anchor lipids pG1, pC1, and pC2 are given as percentage of the total radioactivity of anchor peptides. The remaining 7-11% that are not accounted for stayed at the origin and represent residual anchor peptides that had not been cleaved by HNO 2 treatment. An aliquot of W303 anchor lipids is shown in Fig.  4B remodeling seems to be affecting mostly the maturation processes in the Golgi and/or PM (Golgi/PM remodeling) since, as shown in Fig. 10, lanes 7-10, the ratio of [ 3 H]DHS/[ 3 H]Ins incorporation into proteins in the ⌬gpi7/sec18 double mutant was the same as in sec18. Also, when using stringent conditions under which one observes only ER or only Golgi/PM remodelase (24), remodeling in the ER appeared relatively normal, whereas remodeling in the Golgi/PM was reduced (Fig. 10,  lanes 11-14, 11Ј, and 12Ј). The relatively low amount of pC2 in anchor lipids of ⌬gpi7 (Table I) may be a consequence of this relative deficiency of Golgi/PM remodeling, since pC2 type anchors are only generated by the Golgi/PM but not the ER remodelase (24,41). The relationship between the specific retardation of GPI protein transport, reduced Golgi/PM remodeling, and increased remodeling toward pG1 is for the moment unclear.

DISCUSSION
Yeast and mammals contain the same GPI carbohydrate core structure. This suggests that the GPI anchoring pathway has been established early in evolution and has rigorously been conserved in widely diverging organisms. On the other hand, the side chains added to this core as well as the lipid moieties of the anchor tend to vary a lot between different species (1). The GPI anchors of S. cerevisiae contain two types of side chains as follows: one or two mannoses are linked to Man3 (53) and an EtN-P side chain is linked to Man1. 3 Both side chains are already present on the precursor lipid CP2 (22,23). These two side chains are also found in some vertebrates, including mammals, and possibly in Dictyostelium discoideum (1), suggesting that not only the GPI core structure but also certain kinds of side chains have been invented in and conserved since early times of evolution. Here we present evidence for yet a further, possibly conserved HF-sensitive substituent on CP2 which is attached to Man2. So far, the only side chain attached to Man2 reported in the literature is EtN-P. EtN-P was found by mass spectrometry on 15% of anchors of human erythrocyte cholinesterase and 3% of bovine liver 5Ј-nucleotidase (44,46). Partial acid hydrolysis has also indicated an HF-sensitive substituent on Man2 in 40% of CD52-II (45). Analysis of the ethanolamine/Ins ratio in GPI anchors of porcine renal membrane dipeptidase and of human placental alkaline phosphatase yielded values of 2.5 and 2.4, suggesting the presence of EtN-P on Man2 in 50 and 40% of their anchors, respectively (54,55). However, no such side chain was detected in other mammalian GPI proteins such as rat brain Thy-1 glycoprotein (56) or hamster scrapie prion (57). It may be that in many studies part or all of EtN-P side chains on Man2 and Man1 were hydrolyzed by an unspecific phosphodiesterase during the purification of the respective GPI proteins and preparation of anchor peptides using Pronase. Such a phosphodiesterase activity may explain why we failed to detect any EtN-P side chain on GPI-anchored yeast proteins in the past (53), although we now have firm evidence for the presence of an HF-sensitive substituent on Man1. 3 The chemical nature of the side chain on Man2 of CP2 remains to be determined. The presence of an HF-sensitive side chain on Man2 of CP2 has its parallel in human cells. Indeed, there is evidence for an HF-sensitive group on Man2 of H8, the most polar GPI lipid of HeLa cells (58). The EtN-P side chains on Man1 and Man3 being conserved between mammalian organisms and yeast, it appears reasonable to speculate at this point that the analogy between mammalian and yeast anchors may extend to the substituent on Man2, i.e. that also the side chain on Man2 of yeast GPI structures may consist of an EtN-P and that this EtN-P may be present on some mature GPI proteins of S. cerevisiae.
⌬gpi7 cells are hypersensitive to Calcofluor White and hence have some difficulty in constructing their cell walls. Several reasons can be envisaged. (i) The side chain on Man2 may be important for the interaction of CPs with the transamidase complex and for their efficient transfer onto proteins. Recent data show that a small reduction of Gpi8p renders transamidase activity rate-limiting. 3 The synthetic effect of gpi7 mutations with gpi8 mutations suggests that deletion of GPI7 may have a similar effect. A decreased transamidase activity may particularly affect the anchoring of certain GPI proteins that have a low affinity for the transamidase even though the global rate of GPI biosynthesis and [ 3 H]Ins incorporation into proteins of ⌬gpi7 is not grossly reduced. Thus it is conceivable that some GPI proteins important for cell wall architecture are lacking in ⌬gpi7. (ii) The side chain on Man2 may serve as an attachment point for the covalent linkage of ␤1,6-glucans to the anchor moiety of cell wall proteins although a recent analysis of the linkage region between the GPI anchor remnant and ␤1,6glucans rather showed a direct glycosidic linkage between Man1 and the ␤1,6-glucan (6). (iii) The side chain may serve as a recognition signal for enzymes or proteins that facilitate the packaging of GPI proteins into vesicles, for remodelases that  (24). Lanes 1-6, precultures and labelings were at 30°C. Cells were preincubated with (ϩ) or without (Ϫ) 40 g/ml of myriocin (myr) for 20 min before addition of the radiotracers. Lanes 7-10, sec18 or sec18 ⌬gpi7 double mutants were precultured at 24°C and preincubated for 10 min at 37°C before addition of the tracers. exchange their lipid moieties, or for hydrolases or transglycosidases that remove parts of the GPI anchor of cell wall proteins and hook the GPI remnant onto ␤1,6-glucans (4,5).
Our data further show that ceramide remodeling in the Golgi/PM is significantly reduced in ⌬gpi7, whereas remodeling toward pG1 is increased whereby it is not clear if pG1 remodeling is increased because ceramide remodeling is decreased or if ceramide remodeling is decreased because pG1 remodeling is increased. Moreover, the relationship of the alteration of GPI remodeling with the other phenotypic changes of ⌬gpi7can be explained in several ways. (i) Previous studies showed that remodeling toward pG1 occurs in the ER and that retention of GPI proteins in the ER in secretion mutants maintains a high pG1/pC1 ratio on these proteins (41). Thus, if the substituent on Man2 of GPI anchors is important for efficient packaging of GPI proteins into transport vesicles, the delay in export of GPI proteins out of the ER may give the ER remodelase generating pG1 prolonged access to the GPI proteins and may thus cause a relative increase of pG1. (ii) The side chain on Man2 may serve as a recognition signal for Golgi/PM remodelase. (iii) We also considered the possibility that Gpi7p itself may be a Golgi/PM remodelase. This latter hypothesis would not directly explain why ⌬gpi7 cells cannot attach the HF-sensitive substituent onto Man2 and would imply that the addition of this side chain somehow is directed by the prior attachment of a ceramide moiety. This, however, is clearly not the case, since CP2 also contains the HF-sensitive side chain on Man2, although its lipid moiety consists of DAG (23). Thus we believe that the reduced Golgi/PM remodeling of GPI proteins in ⌬gpi7 is secondary to the lack of a substituent on Man2.
Our previous data suggested that CP2 can be transferred to proteins (23), and our working hypothesis until recently assumed that CP2 represents the GPI lipid used for GPI anchoring also by normal cells that do not accumulate this lipid ("CP2 hypothesis"). By consequence we would have predicted that all the enzymes required for the elaboration of CP2 are localized in the ER. Paradoxically, the subcellular fractionation experiments and protease treatment of intact spheroplasts strongly suggest that the bulk of Gpi7p resides at the cell surface (Fig.  9, A and B). Moreover, although we recently succeeded in demonstrating the presence of an HF-sensitive group on Man1 of immature ER forms of GPI proteins, we presently lack the tools to look for such a group on Man2. Thus, the so far available data raise a doubt whether it is CP2 which is added to GPI proteins in the ER, and we therefore are presently considering the possibility that other GPI lipids than CP2 are the physiological substrate of the ER transamidase. In fact, neither CP2 nor M4 can be detected in wt cells. It therefore seems possible that under physiological conditions cells add M4 to GPI pro-teins ("M4 hypothesis") and that CP2 is elaborated only in mutants in which M4 cannot be transferred to proteins, spills out of the ER, and reaches the PM. It is noteworthy that ⌬gpi7 incorporates [ 3 H]Ins at a normal rate into proteins suggesting that the transamidase is perfectly able to transfer M4. Thus, the side chain on Man2 may normally not be added to GPI proteins or only be added after GPI proteins arrive at the cell surface. The M4 hypothesis, however, does not explain why M0 and M4 accumulate in ⌬gpi7, whereas M0, M4, and CP2 remain undetectable in wt cells (Fig. 2, lanes 1 and 2) or why gpi8-1, deficient in the transfer of GPIs onto proteins, accumulates CP2 (19,21). It also fails to explain the delayed maturation of GPI proteins and the reduced rate of GPI remodeling observed in ⌬gpi7. To save the M4 hypothesis, the accumulation of GPI intermediates in ⌬gpi7 could be rationalized by assuming that the substituent on Man2 serves to mark supernumerary GPIs for degradation, but also this assumption does not explain the observed accumulation of CP2 in gpi8. Thus, although our results raised the possibility that M4 is the physiological GPI lipid for GPI anchoring, this M4 hypothesis leaves many results unexplained and the data are more easily explained by our original CP2 hypothesis. For one, the synchronous accumulation of M4 and CP2 in all our gpi8 mutants argues that M4 is not a better substrate for the transamidase than CP2. CP2 may physiologically be produced by the small amount of Gpi7p in the ER (Fig. 9C). Alternatively, it is conceivable that M4 is transported from the ER to the PM, is converted there to CP2, and is then transported back to the ER by some not yet elucidated mechanism. In this context it is noteworthy that the biosynthesis of GPIs by wt microsomes in vitro produces CP2 in good yield, i.e. the in vitro system adds the substituent on Man2. This in vitro system does not contain cytosol nor GTP and hence should not support vesicular transport from ER-to Golgi-derived microsomes (59). It is possible, however, that GPI lipids are transported between microsomes or membrane fragments by means of lipid transfer proteins or through direct contact between membranes. It also can be envisaged that juxtaposition of membranes allows enzymes present in one membrane to work on lipids in another membrane. The same mechanisms may also operate in intact cells. Clearly the identity of the physiological GPI lipid substrate of the transamidase will have to be established by further experiments.
Homology searches show that two ORFs of S. cerevisiae are related to GPI7, MCD4 (ϭ YKL165c), and YLL031c. They belong to a novel gene family comprising for the moment the nine members shown in Table II which, based on the many predicted transmembrane domains, were previously classified as putative permeases (60). Pairwise alignment allows us to group them into three subfamilies of more closely related ORFs. All nine ORFs predict membrane proteins of about 100 kDa having an N-terminal signal sequence, a hydrophilic N-terminal part, and multiple transmembrane domains in their C-terminal half. mcd4 mutants were obtained in a screen for cells deficient in the cell cycle controlled polarization of growth, a phenotype also generated by mutations in the exocyst or in N-glycosylation (61). The subfamilies typified by GPI7 and YLL031c are more closely related to each other than to the MCD4 subfamily. All nine family members contain two conserved motifs at about the same position in the hydrophilic N-terminal domain, namely HXLGXDXXGH and DHGMXXXGXHG. These motifs are also found in two EST clones from human cDNA that have high homology to MCD4 (NCBI PID 1779747 and 1765215, ClustalW alignments giving aligned scores of 46 and 35). Very interestingly, by a reiterated Psi Blast search at the National Center of Biotechnology Information (NCBI) (62) one can find a highly significant homology of all three subfamilies with a large family of phosphodiesterases. The large majority of these homologous sequences encode mammalian cell-surface proteins classified as alkaline phosphodiesterase I, nucleotide pyrophosphatase, or alkaline phosphatase. The homology comprises a region of about 220 -240 amino acids in the N-terminal hydrophilic part of GPI7, YLL031c, and MDC4. The homology of GPI7 in this region with mammalian and plant phosphodiesterases amounts to 17-18% identity and 30 -34% similarity and comprises a motif PTXTX 8 TGX 2 P which is common to bacterial, viral, plant, and mammalian phosphodiesterases. This homology may suggest that Gpi7p itself is the transferase adding the phosphodiester-linked substituent on Man2. In this context it is interesting to note that the EtN-P on Man3 of the GPI anchor has been shown to be transferred by transesterification using phosphatidylethanolamine as donor of EtN-P (63,64). Mutants in YLL031c also accumulate abnormal GPI intermediates which on TLC have about the same mobility as M4 suggesting that YLL031c is similarly involved in adding EtN-P. 2 Thus, it is conceivable that not only the GPI7 subfamily but also other subfamilies are involved in the transfer of EtN-P onto the GPI core structure. However, the transesterification activity of Gpi7p will have to be shown directly before one can exclude that the primary function of Gpi7p is to generate some signal from the cell wall which regulates GPI protein transport and remodeling as well as side chain addition to GPI structures. It is interesting that subfamily members belonging to different species are more closely related to each other than family members belonging to a single species. This can be seen when comparing the pairwise alignment scores among the three ORFs of S. cerevisiae or the three ORFs of Schizosaccharomyces pombe with the scores among subfamily members (Table II). In evolutionary terms this suggests that the divergence of these three subfamilies occurred earlier than the separation of the lineages leading to S. cerevisiae, S. pombe, and Caenorhabditis elegans. This implies that the HF-sensitive group on Man2 is of very ancient origin. GPI7 bears no resemblance with PIG-F, a mammalian gene encoding for a highly hydrophobic membrane protein involved in the addition of EtN-P to Man3 (10). The exact role of PIG-F has not yet been elucidated.
It will be interesting to find the human homologues of GPI7. It may be that this gene, as in S. cerevisiae, plays a more dispensable role in GPI anchoring than the enzymes involved in the elaboration of the carbohydrate core structure such as for instance PIG-A/GPI3/CWH6/SPT14 (7). Thus, although deficiencies in PIG-A are only acquired by somatic cells, deficiencies in the human GPI7 homologue may be transmittable through the germ line as well.