GPI1 Stabilizes an Enzyme Essential in the First Step of Glycosylphosphatidylinositol Biosynthesis*

Attachment of glycosylphosphatidylinositol (GPI) is essential for the surface expression of many proteins. Biosynthesis of glycosylphosphatidylinositol is initiated by the transfer ofN-acetylglucosamine from UDP-N-acetylglucosamine to phosphatidylinositol. In mammalian cells, this reaction is mediated by a complex of PIG-A, PIG-H, PIG-C, and GPI1. This complexity may be relevant for regulation and for usage of a particular phosphatidylinositol. However, the functions of the respective components have been unclear. Here we cloned the mouse GPI1 gene and disrupted it in F9 embryonal carcinoma cells. Disruption of the GPI1 gene caused a severe but not complete defect in the generation of glycosylphosphatidylinositol-anchored proteins, indicating some residual biosynthetic activity. A complex of PIG-A, PIG-H, and PIG-C decreased to a nearly undetectable level, whereas a complex of PIG-A and PIG-H was easily detected. A lack of GPI1 also caused partial decreases of PIG-C and PIG-H. Therefore, GPI1 stabilizes the enzyme by tying up PIG-C with a complex of PIG-A and PIG-H.

The synthesis of GPI is initiated by the transfer of N-acetylglucosamine (GlcNAc) from UDP-GlcNAc to phosphatidylinositol (PI) to yield GlcNAc-PI (1,11). Genes involved in this step have been cloned by means of complementation of mutants derived from mammalian cells and yeast Saccharomyces cerevisiae: PIG-A, PIG-H, and PIG-C from mammalian cells (12)(13)(14), and GPI1, GPI2 and GPI3/SPT14/CWH6 from yeast (5,(15)(16)(17). PIG-A and PIG-C are homologues of GPI3 and GPI2, respectively, whereas PIG-H and GPI1 are not similar to each other (12,16). A mammalian GPI1 homologue has also been cloned, based on the sequence homology (18). A PIG-H homologue is not found in the S. cerevisiae genome. It was shown that PIG-A, PIG-H, PIG-C, and mammalian GPI1 proteins form a complex in the endoplasmic reticulum that has UDP-GlcNAc:PI GlcNAc transferase (GPI-GnT) activity in vitro (18). Because PIG-A has homology to a bacterial GlcNAc transferase and other glycosyltransferases, this protein is thought to be a catalytic subunit (17,19,20). However, functions of three other proteins cannot be predicted from their sequences. Class A, C, and H mutant cells, corresponding to PIG-A, PIG-H, and PIG-C mutations, do not express GPI-anchored proteins on the cell surface, and their membranes do not have GPI-GnT activity, indicating that these three proteins are essential for the enzyme (21,22). Similarly, S. cerevisiae gpi2 and gpi3 mutants are defective in GPI anchoring, indicating essential roles of GPI2 and GPI3 in GPI synthesis (5). In contrast, S. cerevisiae gpi1 is not essential for growth, and the gpi1-disrupted mutant showed temperature-sensitive growth. Moreover, the gpi1-disrupted cells incorporated a significant amount of inositol to proteins and generated mature Gas1p, a major GPI-anchored protein. Therefore, GPI1 may have a regulatory function rather than a direct role in the enzymatic function (15).
As described above, human GPI1 (hGPI1) is contained in the GPI-GnT complex (18). It has also been shown that hGPI1 physically associates with three other subunits, suggesting a role for hGPI1 in complex formation (18). More recently, it was reported that human and mouse GPI1 cDNAs complemented S. cerevisiae gpi1 and Schizosaccharomyces pombe gpi1 mutants (23), indicating that mammalian GPI1 is functionally homologous to yeast Gpi1p (24). To clarify the function of GPI1 in GPI-GnT, in the present study, we generated GPI1 knockout mouse cells and found that GPI1 is necessary for stable formation of the GPI-GnT complex and stable expression of the PIG-C and PIG-H proteins.
Plasmids-To express mouse GPI, its cDNA was cloned to a pMEEB expression vector. To express GST-tagged PIG-A, GST-tagged-PIG-H, and FLAG-tagged-PIG-C at the same time, each fragment was cloned between the PGK promoter and the PGK poly(A) signal in pBS-PGKtk (7). Three constructs were tandemly blunt-ligated in pBS-PGKzeo to introduce the zeocin resistance gene. This plasmid restored the surface expression of GPI-anchored proteins on human PIG-A-deficient JY5 cells, Pig-c-deficient T1M1 cells (Thy-1 Ϫ c), and Pig-h-deficient S49 cells (Thy-1 Ϫ h). To express FLAG-tagged hGPI1, it was cloned between the PGK promoter and the PGK poly(A) signal in pBS-PGKtk and then ligated with pBS-PGKzeo to generate pBS-FLAG-hGPI1/zeo. All of the other constructs used in this study were described previously (25).
Preparation of cDNA and Genomic Clones of mGPI1-We designed degenerate primers on the basis of hGPI1 and yeast Gpi1p amino acid sequences to clone mGPI1 cDNA (15,18). Two primers, forward primer HF1 (5Ј-GTGCTCCCGCNGGNYTNAARATGAA-3Ј) (N; A, T, G, and C: Y; C and T: R; A and G) and reverse primer HR1 (5Ј-GGAYGGTGAA-GAGNARNGTNCCDAT-3Ј) (D; G, A, and T), and template DNA derived from a mouse testis cDNA library were used for 35 cycles of amplification reaction under the conditions of 94°C (1 min), 55°C (30 s), and 68°C (1.5 min) for denaturation, annealing, and extension, respectively. A sequence of the PCR product was confirmed to be homologous to hGPI1. To obtain the 5Ј and 3Ј regions, we screened 96 subpools of a cDNA library, each of which contained 2 ϫ 10 4 independent clones, with PCR using the same primers. We selected four subpools showing a positive PCR band. Using DNAs of these pools, forward vector primer, and HR1 primer in PCR, fragments containing the 5Ј region were cloned from the subpools. The longest clone was named pGH5c7. Using HF1 primer and reverse vector primer, fragments containing the 3Ј region were cloned. All subpools gave the same fragments. The longest 5Ј and 3Ј fragments were ligated in the EcoRV site to generate full-length cDNA. An analysis by 5Ј rapid amplification of cDNA ends with the Marathon RACE kit (CLONTECH) using RNA from F9 cells confirmed that this cDNA contained the longest 5Ј sequence.
To obtain genomic clones of mGPI1, we screened 1 ϫ 10 6 plaqueforming units of lambda FIXII from a mouse 129/SvJ liver genomic library (Stratagene) using a 1.4-kb pGH5c7 fragment, which contained the 5Ј region of mGPI1 cDNA, as a probe. We obtained eight positive clones.
Disruption of the mGPI1 Gene in F9 Embryonal Carcinoma Cells-A 9.4-kb HindIII genomic fragment of mGPI1 was blunted and transferred to the SmaI site of pBS, the EcoRI-XbaI fragment was then transferred to pPNT (26), and the plasmid was named pPNT-18HdelF. A 2.1-kb XbaI-EcoRV fragment of mGPI1 containing exons 9 and 10 was blunted and transferred to pBS. Its XhoI-NotI fragment and a 1.8-kb XbaI-XhoI fragment derived from PGKpuro were ligated in tandem in the XbaI-NotI site in pPNT-18HdelF. This plasmid was then used as the first targeting vector. A 3.0-kb EcoRV fragment containing a region extending from the middle of exon 6 to exon 10 was transferred to the SmaI site of pBS. Its XhoI-XbaI fragment and a 1.9-kb XbaI-XhoI fragment derived from PGKneo were ligated in the XbaI-NotI site of pPNT-18HdelF. The generated plasmid was used as the second targeting vector.
Northern Blot Analysis-For Northern blot analysis of mouse Pig-c, poly(A) RNA (10 g) prepared using Trizol (Life Technologies, Inc.) and the Poly(A)Tract mRNA isolation system (Promega) was separated on a 1% formaldehyde-agarose gel and then transferred to a nylon membrane. The membrane was hybridized with a Pig-c cDNA probe and rehybridized with a EF1␣ cDNA probe (a gift from Dr. S. Nagata). The Pig-c cDNA probe was amplified with forward primer 5Ј-TGGGCTGT-GGTATTTGAGTCCAG-3Ј and reverse primer 5Ј-TGGCTCCCACAG-CACTGGAC-3Ј from a mouse cDNA library.
Fluorescence-activated Cell-sorting Analysis-To examine the sur-face expression of GPI-anchored proteins, cells were stained with biotinylated anti-Thy1.2 or anti-CD59 antibody and phycoerythrin-conjugated streptavidin (Biomedica) and then subjected to analysis in a FACScan (Becton Dickinson) (25).
Immunoprecipitation and Western Blotting-Transfectant cells (2-4 ϫ 10 8 ) were hypotonically lysed in water containing 0.1 mM TLCK and 1 g/ml leupeptin, mixed with an equal volume of a lysate buffer (100 mM HEPES/NaOH (pH 7.4), 50 mM KCl, 10 mM MgCl 2 , 20% glycerol, 0.1 mM TLCK, and 1 g/ml leupeptin) and stored at Ϫ80°C until used. After thawing on ice, the cell lysates were further treated using a tight pestle Dounce homogenizer. Membranes and cytosol fractions were divided by centrifugation at 100,000 ϫ g for 1 h. The collected membranes were solubilized in 1% digitonin in 50 mM HEPES/NaOH (pH 7.4), 25 mM KCl, 5 mM MgCl 2 , 0.1 mM TLCK, and 1 g/ml leupeptin or 1% Nonidet P-40 in 10 mM Tris⅐HCl (pH 7.7), 150 mM NaCl, 1 mM EDTA, 0.1 mM TLCK, and 1 g/ml leupeptin. Insoluble materials were removed by centrifugation at 100,000 ϫ g for 1 h, and supernatants were used for immunoprecipitation. For the cytosol fractions, Nonidet P-40 was added at 1% and used for immunoprecipitation. Anti-FLAG M2 affinity gel beads (Sigma) and glutathione-Sepharose beads (Amersham Pharmacia Biotech) were used to collect FLAG-and GST-tagged proteins, respectively. Proteins were separated by SDS-polyacrylamide gel electrophoresis and analyzed by Western blotting. Western blotting was carried out with biotinylated anti-FLAG monoclonal antibody M2 plus horseradish peroxidase-conjugated streptavidin (Amersham Pharmacia Biotech) or with anti-GST antibody (Amersham Pharmarcia Biotech) plus horseradish peroxidase-conjugated anti-goat IgG antibody (Organon Teknika) and visualized with chemiluminescence (Renaissance; DuPont). The band intensities in Western blotting were quantitated by measuring chemiluminescence with a Fuji Image analyzer LAS1000.
Fluorescence in Situ Hybridization-Fluorescence in situ hybridization was carried out with a biotinylated genomic DNA probe for mGPI1 and a biotinylated cDNA probe for hGPI1 prepared as described previously (10).

RESULTS
Cloning of mGPI1-The full-length mGPI1 cDNA contained 3062 nucleotides except a poly(A) tract, encoding a 581-amino acid protein that had an 89% amino acid identity to hGPI1 (GenBank accession number for mGPI1 cDNA is AB008895.). Using a mGPI1 cDNA probe, we obtained eight genomic clones of mGPI1, mapped them, and sequenced around the exons (GenBank accession numbers for the genomic sequences are AB008915 to AB008921). Exon sequences were exactly the same as that of the cDNA. The mGPI1 gene had 11 exons, containing a start codon in exon 2 and a stop codon in exon 11 (Fig. 1A). This structure was similar to a partial structure of hGPI1 determined by analyzing a sequence obtained from Gen-Bank TM (Fig. 1B). Their corresponding exons had the same lengths.
The amino acid sequence of mGPI1 was very different from that published recently as mouse GPI1 (24) (GenBank accession number AF030178), having only a 93% amino acid identity. We think that the published sequence is not that of mouse GPI1 because: 1) our cDNA sequence and genomic exon sequence were exactly the same, 2) many mouse expressed sequence tag (EST) nucleotide sequences in the current databases matched exactly with our sequence, 3) the mGPI1 gene was successfully disrupted using our mGPI1 DNA fragments, and 4) their sequence was simply that of an EST clone and was not confirmed in mouse cDNA or DNA.
Disruption of mGPI1 in the F9 Embryonal Carcinoma Cell Line-We used the F9 cell line to generate mGPI1-disrupted cells because it has a relatively high frequency of homologous recombination and is easier to culture than embryonic stem cells. We designed two targeting vectors: the vector for the first targeting had the puromycin resistance gene in place of exon 2 to exon 8, and the vector for the second targeting had the neomycin resistance gene in place of exon 2 to a part of exon 6 ( Fig. 2A). Colonies grown after antibiotic selection were screened with PCR ( Fig. 2B) and confirmed with genomic Southern blotting (Fig. 2C). Bands corresponding to the wild type allele decreased after the first targeting (lanes 2) and disappeared after the second targeting with generation of the predicted targeted bands (lanes 3). After the second targeting, no 3.1-kb RNA band of mGPI1 was seen in Northern blotting (Fig. 2D), indicating that both alleles were disrupted.
Characterization of mGPI1 Knockout Mutants-We first assessed the surface expression of Thy-1, a GPI-anchored protein on mGPI1-knockout F9 cells, and found that the expression decreased greatly but remained at a low level (Fig. 3A). The surface Thy-1 expression was restored upon transfection of mGPI1 cDNA but not upon transfection of an empty vector (Fig. 3B). To confirm that mGPI1 knockout cells have some ability to generate GPI-anchored proteins, we transfected a cDNA of CD59. CD59 was more efficiently expressed on the cell surface, whereas the expression of endogeneous Thy-1 stayed very low (Fig. 3C). These results showed that mGPI1 knockout cells still have some GPI-GnT activity.
We next measured GPI-GnT activity in vitro using cell lysates and microsomes of mGPI1 knockout cells (Fig. 4). In contrast to the significant surface expression of GPI-anchored proteins on mGPI1 knockout cells, their lysates and microsomal membranes did not generate GlcNAc-PI upon incubation with UDP-GlcNAc (lanes 2 in Fig. 4A and B), whereas they had comparable activities of dolichol phosphate glucose (Dol-P-Glc) synthase. This defect was restored by transfection of mGPI1 cDNA (lanes 3) but not by transfection of an empty vector (lanes 4). The lysates of mGPI1 knockout cells gave several radioactive spots, one of which had the same mobility as GlcN-PI (lane 2), the second intermediate derived from Glc-NAc-PI by deacetylation. To determine whether it is GlcN-PI or a non-relevant material, we first tested for the presence of PI by treatment with PI-PLC (Fig. 5A). The spot was resistant to PI-PLC (lanes 3 and 4), whereas the GlcN-PI generated by the wild type cells was sensitive, as shown by a great decrease of the intensity (lanes 1 and 2), indicating that the spot did not contain PI. We next tested acetylation that would convert GlcN-PI to GlcNAc-PI (Fig. 5B). GlcN-PI generated by the wild type cells was acetylated to GlcNAc-PI, resulting in a decrease of GlcN-PI and a slight increase of GlcNAc-PI (lanes 1 and 2); however, the spot generated by mGPI1 knockout cells did not shift to the GlcNAc-PI position (lanes 3 and 4). As a final test, we treated the spot with mild alkali (Fig. 5C). If it contained diacylglycerol, then the acyl chains would be removed, causing a mobility shift or disappearance from the lipid fraction. The spot was resistant (lanes 3 and 4), whereas GlcN-PI and Glc-NAc-PI were sensitive (lanes 1 and 2). All of these results indicated that the spot was not GlcN-PI. Other spots seen above and below the area corresponding to GlcNAc-PI and GlcN-PI should also be non-relevant. They are not related to N-glycans because a 10ϫ higher concentration of tunicamycin did not affect them (data not shown). Therefore, the membranes of mGPI1 knockout cells did not have a detectable level of GPI-GnT activity in vitro.
We thought that the weak expression of Thy-1 on mGPI knockout cells (Fig. 3A) might be due to inefficient recognition of GPI attachment signals of Thy-1 by GPI transamidase in the presence of a limited amount of GPI. To test this, we replaced the GPI attachment signal sequence of CD59 with that of Thy-1 FIG. 1. Genomic structures of mGPI1 and hGPI1. The genomic structure of mGPI1 (A) was determined from overlapped restriction maps of genomic lambda clones and sequences of regions containing exons. The genomic structure of hGPI1 (B) was determined by analyzing nucleotide sequences from the GenBank TM . GenBank TM accession numbers of genomic sequences of hGPI1 are Z98881 and Z98883. f and Ⅺ, noncoding and coding exon regions, respectively. Exon numbers are indicated. Restriction enzyme sites are as follows: Bm, BamHI; E, EcoRI; H, HindIII; RV, EcoRV; X, XbaI; and Sp, SpeI.

FIG. 2. Disruption of mGPI1 in F9 cells.
A, targeting vectors for homologous recombination. The first and second targeting vectors were designed to disrupt exon 2 to exon 8 and exon 2 to the middle of exon 6, respectively, and to contain the puromycin (puro) and neomycin (neo) resistance genes, respectively. The HSV thymidine kinase gene (tk) was included at the 5Ј end to select against random integration. All selection marker genes were driven by the PGK promoter and had a PGK poly(A) signal. Restriction enzyme site: Xh, XhoI. B, structures of disrupted mGPI1 after homologous recombination. To screen homologous recombinants, PCR primers were used as indicated (arrows). Probes A and B, which were used for Southern blotting, are in bold. C, Southern blotting of targeted mutants. Samples of genomic DNA (5 g) were cut with XbaI (left panel) and SpeI (right panel) and probed with probes A and B (see B), respectively. Lanes 1, F9 (wild type); lanes 2, single knockout mutant; lanes 3, double knockout mutant. D, Northern blotting of mGPI1 knockout mutant. Samples of total RNA (30 g) were separated in a 1% formaldehyde-agarose gel, transferred to a nylon membrane, and probed with radiolabeled mGPI1 cDNA. The membrane was reprobed with EF-1␣ cDNA as a control. Lane 1, F9; lane 2, mGPI1 knockout mutant. and compared it with wild-type CD59. As shown in Fig. 6, CD59 bearing the GPI attachment signal of Thy-1 was processed much less efficiently than CD59 bearing its own signal.

GPI1 Is Required for the Stable Formation of GPI-GnT Complexes and for the Full Expressions of PIG-C and PIG-H but not PIG-A-
The above results demonstrated that a very low level of GPI-GnT activity was expressed in the absence of mGPI1. A possible explanation for this is that the GPI-GnT complex cannot be formed efficiently in the absence of mGPI1. Another possible explanation is that a protein complex of PIG-A, PIG-C, and PIG-H is formed even in the absence of mGPI1 but has a very weak activity. To test the former possibility, we stably transfected mGPI1 knockout cells with a vector bearing cDNAs for FLAG-tagged PIG-C and GST-tagged PIG-A and PIG-H. A vector for free GST was also cotransfected to the same cells. A clonal cell population was isolated by limiting dilution and further transfected with either mGPI1 cDNA or an empty vector. Therefore, two cells would potentially express tagged PIG-C, PIG-A, and PIG-H and free GST at the same levels, with the sole difference being the presence or absence of mGPI1. From digitonin extracts of these cells, we precipitated FLAG-tagged PIG-C with anti-FLAG beads, and coprecipitations of GST-tagged PIG-A and PIG-H were assessed by West-ern blotting with anti-GST (Fig. 7A) 2 and 4) or a reaction buffer (lanes 1 and 3), and reextracted lipids were analyzed by TLC. B, the same samples as in A were treated with acetic anhydride (lanes 2 and 4) or a buffer (lanes 1  and 3). C, the same samples as in A and B were treated with mild alkali (lanes 2 and 4) or a neutral buffer (lanes 1 and 3). cytoplasm (bottom panel). Therefore, only a trace amount, if any, of the protein complex was formed in the absence of mGPI1.
To evaluate the amounts of expressed FLAG-tagged PIG-C, we analyzed the immunoprecipitates with anti-FLAG beads by Western blotting against anti-FLAG antibody (Fig. 7A, lanes 3  and 4). The amount of FLAG-PIG-C was significantly lower in the absence (lane 4) than in the presence (lane 3) of mGPI1, whereas a nonspecific band (asterisk) and a light chain of the antibody (L) showed comparable intensities. We also evaluated the amounts of expressed GST-tagged PIG-A and PIG-H by collecting them from Nonidet P-40 extracts with glutathione-Sepharose beads followed by Western blotting against anti-GST antibody (lanes 5 and 6). The amount of GST-tagged PIG-H was significantly lower in the absence (lane 6) than in the presence (lane 5) of mGPI1, whereas the amounts of GSTtagged PIG-A were comparable. The intensities of these bands and bands of free GST collected from the cytoplasm were quantitated by an image analyzer. Expression levels of PIG-C and PIG-H decreased to one-third of the wild type levels in the absence of mGPI1, whereas PIG-A and a control free GST were expressed at similar levels.
These results were from cloned cells. To eliminate the chance that an unusual event might have occurred due to the use of cloned cells, we used cells without limiting dilution. For this we transiently cotransfected mGPI1 knockout cells with a vector bearing cDNAs for FLAG-tagged PIG-C and GST-tagged PIG-A and PIG-H in combination with mGPI1 cDNA or an empty vector and assessed the complex formation and expressions of three tagged proteins. We obtained results very similar to those with the cloned cells (data not shown).
To test the association between PIG-A and PIG-H in the absence of GPI1, we cotransfected GST-tagged PIG-H and FLAG-tagged PIG-A into mGPI knockout cells (Fig. 7B, lane 1). As a positive control, GST-tagged GPI1 and FLAG-tagged PIG-A were cotransfected (lane 2). As negative controls, GSTtagged ALDH (aldehyde dehydrogenase, a control endoplasmic reticulum membrane protein (29)) and FLAG-tagged PIG-A (lane 3) and GST-tagged PIG-H and FLAG-tagged ALDH (lane 4) were cotransfected, respectively. FLAG-tagged PIG-A was efficiently coprecipitated with GST-tagged PIG-H (lane 1), but not with GST-tagged ALDH (lane 3). FLAG-tagged ALDH was not coprecipitated with GST-tagged PIG-H (lane 4). Therefore, PIG-A and PIG-H associated specifically and efficiently in the absence of GPI1. This indicated that the efficient association of PIG-C with a complex of PIG-A and PIG-H is dependent upon GPI1.
Because GP11 is an endoplasmic reticulum membrane protein (18), it is unlikely that the GPI1 protein affects transcriptions of the PIG-C and PIG-H genes. In fact, the expression levels of endogenous mouse Pig-c mRNA assessed by Northern blotting did not change between wild type and mGPI1 knockout cells (data not shown). It is likely that GPI1 affects the stability of the PIG-C and PIG-H proteins. However, the decreased levels of PIG-C and PIG-H were still 30% of the wild type levels, which does not fully explain a nearly complete lack of complexes of PIG-A, PIG-C, and PIG-H. Therefore, GPI1 must function in the maintenance of the GPI-GnT complex itself.
To see whether the expression of GPI1 is conversely influenced by PIG-C and PIG-H, we cotransfected FLAG-tagged hGPI1 and free GST, as a control, into Pig-c-deficient TIM1 cells (Thy-1 Ϫ c) and Pig-h-deficient S49 cells (Thy-1 Ϫ h), obtained the cloned cells, and transfected them with PIG-C and PIG-H cDNA, respectively, or with an empty vector. An analysis by fluorescence-activated cell sorting confirmed that PIG-C-transfected T1M1 (Thy-1 Ϫ c) cells and PIG-H-transfected S49 cells (Thy-1 Ϫ h) but not vector transfectants restored the surface expression of Thy-1 proteins (data not shown). However, the amounts of expressed FLAG-tagged GPI1 did not change significantly in the presence (Fig. 8A, lane 1) and absence (lane 2) of PIG-C or in the presence (Fig. 8B, lane 1) and absence (lane 2) of PIG-H. Therefore, it seems that GPI1 influences the stability of PIG-C and PIG-H, but not PIG-A, but that the stability of GPI1 is not influenced by PIG-C and PIG-H.
Chromosomal Localization of mGPI1 and hGPI1-We investigated the chromosomal localization of mGPI1 and hGPI1 using fluorescence in situ hybridization. With a 15.2-kb genomic fragment from a lamda clone as a probe, we localized mGPI1 to mouse chromosome 17B (Fig. 9A). Using a 2.9-kb hGPI1 cDNA probe, we localized hGPI1 to human chromosome 16p13.3 (Fig. 9B). These mouse and human chromosome regions are syntenic (30). The autosomal location of mammalian GPI1 is consistent with the notion that only PIG-A is X-linked among all GPI-anchor synthesis genes (1).

DISCUSSION
In this report, we present direct evidence that GPI1 is necessary for the stable formation of GPI-GnT. We disrupted mouse GPI1 in F9 embryonal carcinoma cells and found that the generation of GPI-anchored proteins was severely affected in the absence of GPI1. This defect was accounted for by a decrease of complexes of PIG-C with PIG-A and PIG-H to a nearly undetectable level, whereas an association between PIG-A and PIG-H was still seen. In the absence of GPI1, the levels of PIG-C and PIG-H were only one-third of the wild type levels, indicating that GPI1 is needed to maintain normal levels of these proteins. The level of PIG-A was not affected by the absence of GPI1. These partial losses of PIG-C and PIG-H alone cannot fully account for the nearly complete lack of the complex of these components. Therefore, it is indicated that GPI1 is required for the efficient association of PIG-C with a complex of PIG-A and PIG-H. This is consistent with our previous report that GPI1 directly associates with each of the three other proteins (18).
Among the four proteins participating in GPI-GnT, PIG-A, PIG-C, and PIG-H are essential for activity because their mutant cells are completely defective in GPI-GnT (21,22), whereas GPI1 is not essential because GPI1 knockout cells still had some ability to generate GPI-anchored proteins. This phenotype of mGPI1 knockout F9 cells is similar to that of gpi1disrupted S. cerevisiae that was lethal at a higher temperature but was viable and incorporated some inositol to proteins at 25°C (15). Expression of mammalian GPI1 in gpi1-disrupted S. cerevisiae restored growth at a higher temperature and synthesis of GlcNAc-PI in vitro (24), indicating that mammalian and yeast GPI1 proteins have conserved sites for association with other components.
In the absence of GPI1, proteins that are normally GPIanchored would compete for a limited amount of GPI. Proteins bearing different carboxyl-terminal GPI attachment signals may have different affinities for GPI transamidase that recognizes the GPI attachment signal and replaces it with GPI (2). The mGPI1 knockout F9 cells expressed a very low level of Thy-1 on the surface but expressed CD59 quite efficiently, indicating that the GPI attachment signal of CD59 functioned more efficiently than that of Thy-1. A possible reason for this is that the signal sequence of CD59 contains asparagine at the site (31) to which GPI is linked, whereas the signal of Thy-1 has cysteine at the site (32). There are reports that the asparagine site functions more efficiently than the cysteine site in the GPI attachment reaction (33)(34)(35).
We localized the mouse GPI1 gene to chromosome 17B and localized human GPI1 to chromosome 16p13.3. This autosomal location is consistent with the fact that two homologous recombinations were required to eliminate mGPI1. This is also consistent with the observation that most, if not all, patients with paroxysmal nocturnal hemoglobinuria lost the ability for biosynthesis of GlcNAc-PI in their affected blood cells due to somatic mutation of the PIG-A gene in the hematopoietic stem cell (36). Because PIG-A is X-linked, a single somatic mutation should cause the GPI-GnT deficiency, whereas two somatic mutations of GPI1 should occur in the same cell to cause the mutant phenotype. The latter probability would be extremely low. It is difficult to predict the outcome of a hereditary lack of GPI1. Although a complete lack of GPI synthesis causes embryonic lethality (7,8), no information is available about the effects of a partial GPI deficiency.