|Home | About | Journals | Submit | Contact Us | Français|
Many eukaryotic cell surface proteins are anchored to the membrane via glycosylphosphatidylinositol (GPI). The GPI is attached to proteins that have a GPI attachment signal peptide at the carboxyl terminus. The GPI attachment signal peptide is replaced by a preassembled GPI in the endoplasmic reticulum by a transamidation reaction through the formation of a carbonyl intermediate. GPI transamidase is a key enzyme of this posttranslational modification. Here we report that Gaa1p and Gpi8p are components of a GPI transamidase. To determine a role of Gaa1p we disrupted a GAA1/GPAA1 gene in mouse F9 cells by homologous recombination. GAA1 knockout cells were defective in the formation of carbonyl intermediates between precursor proteins and transamidase as determined by an in vitro GPI-anchoring assay. We also show that cysteine and histidine residues of Gpi8p, which are conserved in members of a cysteine protease family, are essential for generation of a carbonyl intermediate. This result suggests that Gpi8p is a catalytic component that cleaves the GPI attachment signal peptide. Moreover, Gaa1p and Gpi8p are associated with each other. Therefore, Gaa1p and Gpi8p constitute a GPI transamidase and cooperate in generating a carbonyl intermediate, a prerequisite for GPI attachment.
Numerous eukaryotic cell surface proteins are anchored to the membrane via a covalently attached glycosylphosphatidylinositol (GPI). Posttranslational attachment of GPI is essential for expression of those proteins on the cell surface. This type of membrane anchoring is widely used in all eukaryotic organisms.
In mammalian cells, more than 100 cell surface proteins with various sizes and functions are GPI-anchored (Low, 1989 ; Kinoshita et al., 1995 ). At the cellular level, GPI-anchoring is not essential, and many GPI-deficient mutant cell lines have been established (Takeda and Kinoshita, 1995 ), indicating roles of GPI-anchored proteins in cell-to-cell interactions rather than cell growth itself. At the levels of tissues and the whole body, GPI-anchoring is critical. Keratinocyte-specific disruption of one of the GPI biosynthesis genes, PIG-A, demonstrated that GPI is essential for normal development of skin (Tarutani et al., 1997 ). Disruption of PIG-A gene in the whole body resulted in embryonic lethality (Kawagoe et al., 1996 ; Nozaki et al., 1999 ). A human disease paroxysmal nocturnal hemoglobinuria is caused by somatic mutation of the PIG-A gene occurring in the multipotential hematopoietic stem cell (Takeda et al., 1993 ).
In Saccharomyces cerevisiae, GPI is essential for growth (Leidich et al., 1995 ). Analysis of the S. cerevisiae genome demonstrated that of ~6200 ORFs, ~60 encode GPI-anchored proteins (Caro et al., 1997 ). Many of these are cell wall proteins. They are first synthesized and transported to the plasma membrane in the GPI-anchored form and then are incorporated into cell wall glucan after cleavage of the GPI portion (Lu et al., 1995 ; Kollar et al., 1997 ).
GPI-anchored proteins are formed in the endoplasmic reticulum (ER) from a preformed GPI and a protein precursor (Kinoshita et al., 1995 ; Udenfriend and Kodukula, 1995 ). Proteins that are to be GPI-anchored have two signal peptides (Udenfriend and Kodukula, 1995 ). One is an amino-terminal signal peptide that directs translocation across the ER membrane. The other is a C-terminal signal peptide that directs attachment of the GPI anchor. Shortly after translation, the C-terminal GPI attachment signal peptide is recognized by a GPI transamidase that cleaves the signal and replaces it with GPI.
The amino acid to which GPI is attached is termed the ω site (Gerber et al., 1992 ), and it must have a small side chain (Micanovic et al., 1990 ; Moran et al., 1991 ; Nuoffer et al., 1993 ). The second residue carboxyl terminal to the ω site (ω+2) must also be a small amino acid, whereas the ω+1 site can be any amino acid except proline and tryptophan (Gerber et al., 1992 ). The ω+2 site is followed by a stretch of hydrophilic amino acids, usually 5–7 residues, and a hydrophobic segment of 12–20 amino acids (Furukawa et al., 1997 ). These are characteristics of the GPI attachment signal peptide, but there is no consensus sequence. The GPI transamidase is proposed to bind to the GPI attachment signal peptide and attack the carbonyl group of ω site amino acid with its catalytic site to release the signal peptide and generate a carbonyl intermediate between a precursor protein and the enzyme. GPI is then presented to this intermediate, whose amino group in the terminal ethanolamine would attack the intermediate to complete the transamidation reaction (Udenfriend and Kodukula, 1995 ; Sharma et al., 1999 ).
The GPI transamidase that mediates GPI attachment has not been clearly characterized. Two S. cerevisiae mutants, gaa1 (Hamburger et al., 1995 ) and gpi8 (Benghezal et al., 1996 ), are defective in the attachment of GPI to proteins. GPI8 encodes a protein with homology to members of a family of cysteine proteases (Benghezal et al., 1996 ), one of which, a jack bean asparaginyl endopeptidase, showed transamidase activity in vitro (Abe et al., 1993 ). A human mutant cell line termed class K that is defective in attachment of GPI (Mohney et al., 1994 ; Chen et al., 1996 ) is due to a defect in the human GPI8 gene (Yu et al., 1997 ). Microsomal membranes of class K cells did not have GPI transamidase activity (Chen et al., 1996 ; Yu et al., 1997 ). It was therefore suggested that Gpi8p is a component of the GPI transamidase (Benghezal et al., 1996 ; Yu et al., 1997 ). On the other hand, Gaa1p has no homology to other proteins in the databases, so it is not possible to predict its function. In the present investigation, we demonstrate that Gaa1p and Gpi8p form a protein complex, that Gaa1p is required for a precursor protein to form a carbonyl intermediate with the GPI transamidase, and that a conserved cysteine residue of Gpi8p is involved in cleavage of the signal peptide.
Mouse F9 and human K562 cells were obtained from the American Type Culture Collection. The class K mutant was a gift from Dr. M. E. Medof (Case Western Reserve University). Mouse EL4 and its class F GPI-anchor–deficient mutant line Thy-1−f were provided by Dr. R. Hyman (Salk Institute). They were cultured in high glucose DMEM supplemented with 10% FCS. Chinese hamster ovary (CHO) cells were cultured in Ham's F-12 medium containing 10% FCS.
All expression plasmids were constructed on pMEPyori in which the expression of the cloned insert is driven by SRα promoter (Ohishi et al., 1996 ). We cloned the human GAA1 cDNA into pMEPyori (pMEPyori–hGAA1) (Inoue et al., 1999 ). A full-length human GPI8 cDNA was obtained by ligation of the following two fragments at a unique MscI site: an EST I.M.A.G.E. Consortium clone 33372 containing most of the ORF and a 5′ RACE product containing the missing region, which was amplified from placental mRNA. The resulting full-length human GPI8 cDNA was cloned into pMEPyori (pMEPyori–hGPI8). The yeast GPI8 ORF was amplified by PCR and cloned into pMEPyori (pMEPyori–yGPI8). Amino-terminally FLAG-tagged human GAA1 and carboxyl-terminally GST-tagged human and yeast GPI8s were constructed as follows. To obtain pMEPyori-FLAG–hGAA1, a PIG-A–coding fragment of pMEPyori FLAG–PIG-A (Maeda et al., 1998 ) was replaced with a PCR-amplified SalI–EcoRV fragment containing the 5′ portion of human GAA1 in which an initiation codon was substituted for a SalI site and an EcoRV–NotI fragment containing the 3′ portion of human GAA1 excised from pMEPyori–hGAA1. pMEPyori-HA–hGAA1 was obtained in a similar way. We obtained pMEPyori-hGPI8–GST by assembling three fragments on pMEPyori, an EcoRI–NheI fragment containing the 5′ portion of human GPI8 from pMEPyori–hGPI8, a PCR-amplified NheI–MluI fragment containing the 3′ portion of human GPI8 in which a MluI site was substituted for a stop codon, and a MluI–XbaI fragment from pMEEB-PIG-A–GST (Watanabe et al., 1998 ) encoding GST. pMEPyori-yGPI8–GST was obtained in a similar way by assembling an EcoRI–Bsp119I fragment from pMEPyori–yGPI8, a PCR-amplified Bsp119I–MluI fragment and the GST fragment derived from pMEEB-PIG-A–GST. Amino-terminally FLAG-tagged human GPI8 (pMEPyori-FLAG–hGPI8) was constructed by insertion of an HA epitope between Ile30 and Glu31 of human GPI8 by means of oligonucleotide-directed mutagenesis. FLAG- and GST-tagged microsomal aldehyde dehydrogenase (msALDH) were described previously (Maeda et al., 1998 ). Tagged versions of human Gaa1p and Gpi8p were functional because their cDNAs complemented GAA1-knockout cells and class K cells, respectively, to the same extent as cDNAs for nontagged counterparts. pMEpuro was constructed by cloning PGKpuro cassettes into the HindIII site of pME vector and used for the establishment of class K cells stably expressing GST-tagged Gpi8ps. Sequences of primers used are available on request.
Targeting vectors were constructed as follows. A 7-kilobase (kb) BamHI and blunt-ended XbaI fragment of mouse GAA1 was cloned into BamHI and blunt-ended EcoRI sites of pPNT (a gift from Dr. R. Mulligan, Harvard Medical School) (Tybulewicz et al., 1991 ). A NotI–XhoI fragment containing a 2-kb SacII–BamHI genomic fragment was excised from pBluescript (pBS) bearing the 2-kb fragment at the SmaI site. pPGKBSD was obtained by replacing the puromycin-resistance gene in pPGKPuro (a gift from Dr. T. Yagi, National Institute for Physiological Sciences) (Watanabe et al., 1995 ) with a blasticidin resistance gene from pMAM2-BSD (Kimura et al., 1994 ). PGKneo, PGKpuro, PGKhyg, and PGKbsd cassettes from pPNT, pPGKPuro, pPGK-Hygro (a gift from Dr. A. Berns, The Netherlands Cancer Institute), and pPGKBSD, respectively, were cloned into blunt-ended HindIII sites of pBS and excised as XhoI–BamHI fragments. Each of these fragments containing the PGK-driven drug resistance genes and the NotI–XhoI fragment of mouse GAA1 described above were cloned into NotI–BamHI sites of pPNT bearing the 7-kb fragment of mouse GAA1 (see Figure Figure1A).1A). F9 cells were electroporated with NotI-linearized targeting plasmids and selected 1 d later with appropriate drugs. Concentrations of G418, puromycin (Sigma, St. Louis, MO), hygromycin, and blasticidin were 380, 2, 500, and 4 μg/ml, respectively. Recombinants were screened by PCR with common 3′ primer and drug cassette-specific 5′ primers and were confirmed by Southern blotting using 1-kb EcoRV–EcoRI and 0.6-kb BamHI–BamHI genomic fragments as 5′ and 3′ probes, respectively (see Figure Figure11A).
Cells were stained with biotinylated anti–Thy-1 G7 or anti-CD59 5H8 followed by phycoerythrin-conjugated streptavidin (Biomeda, Foster City, CA) and analyzed in a FACScan (Becton Dickinson, San Jose, CA) (Maeda et al., 1998 ).
GPI intermediates were metabolically radiolabeled with [3H]mannose (American Radiolabeled Chemicals, St. Louis, MO) in the presence of tunicamycin, extracted, and analyzed by TLC (Hirose et al., 1992 ).
The amino acid sequence of mini-placental alkaline phosphatase (mini-PLAP) with Ser at the ω site, designed based on the reported sequence (Millan, 1986 ), was basically identical to that used by Udenfriend's group (Kodukula et al., 1991 ), except for the presence of an additional five residues (Met-Leu-Gly-Pro-Cys) at the amino terminus. A coding region of this mini-PLAP was divided into three regions and amplified from human placental mRNA by RT-PCR and assembled in pSPUTK, an in vitro transcription vector (Stratagene, La Jolla, CA). Microsomal membranes were isolated basically according to a reported method (Maxwell et al., 1995b ; Chen et al., 1996 ). The microsomal membranes were suspended in a suspension buffer containing 50 mM triethanolamine/250 mM sucrose, pH 7.5, at a determined concentration (50 OD units/ml at 280 nm in 1% SDS), frozen in liquid nitrogen, and stored at −80°C. Membranes of mutant and corresponding wild-type cells were prepared at the same time.
One microliter-capped mini-PLAP RNA (1 μg/μl) was translated at 30°C for 90 min in the following reaction mixture: 12.5 μl nuclease-treated rabbit reticulocyte lysate, 1 μl methionine-free amino acid mixture, 0.5 μl RNasin (all from Promega [Madison, WI]), 2 μl Redivue l-[35S]methionine (Amersham, Arlington Heights, IL), 2.5 μl buffer composed of 100 mM potassium acetate, 4 mM magnesium acetate, 20 μg/ml each of antipain, aprotinin, bestatin, chymostatin, leupeptin, and pepstatin, 1.5 μl water, 4 μl microsomal membranes, and 1 μl water or 260 mM hydrazine. Reaction mixtures were diluted in 1 ml of a precipitation buffer consisting of 1% NP-40, 50 mM Tris, 150 mM NaCl, 0.025% sodium azide, and a complete protease inhibitor mixture tablet at the recommended concentration (Boehringer Mannheim, Mannheim, Germany), pH 7.8. Mini-PLAP proteins were precipitated with rabbit anti-PLAP antibody (Biomeda) and protein A-Sepharose, fractionated on a 15% SDS-PAGE gel, and visualized by BAS image analyzer (Fuji Photo Film, Tokyo, Japan).
To generate chimeric GPI8s, we divided human and yeast GPI8 coding regions into four segments encoding an amino-terminal signal sequence (amino acids 1–42 of hGpi8p, 1–35 of yGpi8p), a highly conserved region (amino acids 43–304 of hGpi8p, 36–297 of yGpi8p), a nonconserved juxtatransmembrane region (amino acids 305–370 of hGpi8p, 298–384 of yGpi8p), and transmembrane cytoplasmic domains (amino acids 371–395 of hGpi8p, 385–411 of yGpi8p). Unique restriction enzyme sites were designed at boundaries of the regions. We amplified these segments by PCR and confirmed the sequences. All nucleotide sequences introduced to make restriction enzyme sites did not change amino acids. These segments were assembled in all possible combinations in pMEPyori. Chimeric GPI8 cDNAs (2 μg) were lipofected into class K cells with DMRIE-C in Opti-MEM (Life Technologies, Gaithersburg, MD), and transfectants were analyzed for CD59 surface expression 2 d later.
An EcoRI–NheI fragment of human GPI8 and an EcoRI–Bsp119I fragment of yeast GPI8 encoding a highly conserved region were excised from pMEPyori–hGPI8 and pMEPyori–yGPI8 and subcloned into pBS. We replaced codons of interest with alanine by means of oligonucleotide-directed mutagenesis. Mutated fragments were cloned back into original plasmids. Plasmids were transfected into class K cells by electroporation or by DMRIE-C. Restoration of CD59 expression was measured by flow cytometry. Deletion mutants of human GPI8 were constructed by replacement of a NheI–NotI fragment of pMEPyori–hGPI8 with PCR-amplified shortened fragments.
CHO cells (4 × 106) were electroporated with 15 μg each of plasmids at 960 μF and 250 V. Cells were grown in medium for 2 d to allow protein expression and then solubilized in 1 ml precipitation buffer (containing 1% NP-40) on ice for 1 h. We centrifuged the cell lysates at 18,000 × g for 20 min to remove cell debris and nuclei and then centrifuged the supernatants at 100,000 × g for 1 h to remove insoluble materials. The resulting cleared lysates were subjected to immunoprecipitation with anti-FLAG M2 beads (Eastman Kodak, Rochester, NY) or anti-HA (Roche, Mannheim, Germany) plus protein G beads (Pharmacia, Piscataway, NJ). After the first precipitation, the remaining supernatants were subjected to precipitation with glutathione beads (Pharmacia) or anti-FLAG M2 beads. These precipitates were washed in 1 ml precipitation buffer five times and analyzed by Western blotting as reported (Watanabe et al., 1998 ).
Human and mouse Gaa1ps have only 25% amino acid identity with S. cerevisiae Gaa1p (Hiroi et al., 1998 ; Inoue et al., 1999 ). To demonstrate that mammalian Gaa1p is involved in attachment of the GPI anchor, we disrupted mouse GAA1 genes in F9 embryonal carcinoma cells by means of homologous recombination (Figure (Figure1A).1A). Perhaps because of an unexpected amplification of the GAA1 gene during the disruption procedures, we needed to perform four homologous recombinations to eliminate all wild-type alleles of GAA1 (Figure (Figure1B).1B). The GAA1-knockout F9 cells lost the surface expression of GPI-anchored proteins Thy-1 (Figure (Figure2A,2A, a and b), stem cell antigen-1, and heat stable antigen (our unpublished results). Their expression was restored after transfection of human GAA1 cDNA (Figure (Figure2A,2A, c and d) and mouse GAA1 cDNA (our unpublished results), indicating that mammalian GAA1 is necessary for the cell surface expression of GPI-anchored proteins.
To confirm that biosynthesis of GPI is normal after disruption of GAA1, we metabolically labeled GAA1-knockout F9 cells with [3H]mannose in the presence of tunicamycin and analyzed mannolipids by TLC. Mature forms of GPI termed H7 and H8 (Hirose et al., 1992 ) were synthesized and accumulated in the GAA1-knockout cells (Figure (Figure2B,2B, lane 5) compared with the wild-type F9 cells (lane 6). This phenotype of GAA1-knockout cells was similar to that of GPI8-defective class K mutant cells that accumulate large amounts of H8 and H7 (lanes 3 and 4). These results together indicate that mammalian Gaa1p is not required for biosynthesis of GPI but is essential for attachment of GPI to proteins.
It was reported previously that a temporary carbonyl intermediate between the precursor protein and the GPI transamidase is formed during GPI-anchoring with a release of a cleaved signal peptide (Maxwell et al., 1995a ). We tested whether microsomal membranes of GAA1-knockout F9 cells can generate the carbonyl intermediate. We used a cell-free system (Kodukula et al., 1991 ) in which a radiolabeled precursor protein, mini-PLAP, generated by in vitro transcription and translation is processed by the microsomal membranes bearing GPI transamidase. Microsomal membranes from wild-type CHO, EL4, K562, and F9 cells completed processing, generating the GPI-anchored form of mini-PLAP (Figure (Figure3A,3A, lanes 2, 4, 8, and 12) as well as pro–mini-PLAP (with a cleaved amino-terminal signal sequence). Membranes from CHO and F9 cells generated small amounts of free mini-PLAP (which lost the amino-terminal signal sequence as well as the GPI attachment signal peptide because of hydrolysis; lanes 2 and 12) (Maxwell et al., 1995b ), and membranes from K562 cells generated relatively more free mini-PLAP (lane 8). In the presence of hydrazine, generation of the GPI-anchored form was almost completely inhibited because of competition between GPI and excess hydrazine, resulting in generation of the hydrazide of free mini-PLAP (lanes 3, 5, 9, and 13). As reported previously, membranes from class F GPI synthesis mutant cells did not generate the GPI-anchored form because of a lack of mature GPI (lane 6) but formed the enzyme-substrate carbonyl intermediate, which was sensitive to hydrazine (lane 7) (Chen et al., 1996 ). The membranes of class K cells did not generate the GPI-anchored form (lane 10) or the carbonyl intermediate (lane 11), as reported (Chen et al., 1996 ). Similarly to class K cells, membranes from GAA1-knockout cells processed the amino-terminal signal peptide generating pro–mini-PLAP but did not generate the GPI-anchored form (lane 14) or the carbonyl intermediate (lane 15). Generation of the GPI-anchored form was restored by transfection of human GAA1 cDNA into GAA1-knockout cells (Figure (Figure3B,3B, lanes 6 and 7). Transfection of human GPI8 cDNA had no effect (Figure (Figure3B,3B, lanes 8 and 9). These results indicate that Gaa1p acts before or during formation of the carbonyl intermediate.
It has been suggested, based on sequence homology to cysteine proteases, that Gpi8p is the catalytic component of GPI transamidase that cleaves off the GPI attachment signal peptide (Benghezal et al., 1996 ). To obtain experimental evidence for this, we mutagenized Cys206 and His164 in human Gpi8p, residues that are conserved in yeast and nematode Gpi8ps and members of a cysteine protease family and are likely to be involved in catalytic reaction (Figure (Figure4A).4A). We also mutagenized Cys92, which is conserved in Gpi8ps but not in other members of a cysteine protease family, and Ser67, which was suggested to be a possible active site (Benghezal et al., 1996 ). As shown in Figure Figure4B,4B, Ala substitutions of Cys206 and His164 resulted in complete loss of complementation of class K cells, whereas substitution of Cys92 only partially decreased the activity. Ser67 was not important. Expression of both Cys206Ala and His164Ala mutants was efficient as shown by Western blot analysis (Figure (Figure4C).4C).
To demonstrate that a lack of complementation of class K cells was due to a defect in cleavage of GPI attachment signals, we assayed transamidase activity using microsomal membranes of class K cells expressing these mutant Gpi8ps. The microsomes bearing these mutants produced no hydrazide form of mini-PLAP in the presence of hydrazine (Figure (Figure4D,4D, lanes 6 and 8). These results demonstrate that Cys206 and His164 are essential to form carbonyl intermediates.
We next tested whether corresponding cysteine and histidine in yeast Gpi8p (Cys199 and His157) are essential. Yeast Gpi8p had no activity when transfected into class K cells, maybe because of incompatibility with another mammalian component or components of GPI transamidase (our unpublished results). Therefore, we divided Gpi8ps into four regions, R1–R4 (Figure (Figure5A),5A), and constructed chimeras of yeast and human Gpi8ps. Replacement of R3, the least conserved intralumenal juxtamembrane region, of yeast Gpi8p with that of human origin rendered the chimeric protein functional in class K cells (our unpublished results), indicating that yeast R3 caused incompatibility and that other regions of yeast origin are interchangeable. To determine roles of Cys199 and His157 in yeast Gpi8p, we prepared respective Ala mutants using a chimera in which only R2 was of yeast origin. Substitution of His157 or Cys199 with Ala showed complete loss of complementation activity (Figure (Figure5B,5B, b and c). Therefore, these conserved Cys and His are essential in both human and yeast Gpi8p for transamidase activity.
The above results indicated that both Gaa1p and Gpi8p are components of the GPI transamidase. We next tested whether the two components form a complex. For this, we tagged Gaa1p and a control ER membrane protein ALDH (Masaki et al., 1994 ) with the FLAG epitope, and Gpi8p and ALDH with GST. The tagged proteins were expressed in various combinations in CHO cells (Figure (Figure6).6). FLAG-tagged Gaa1p or FLAG-tagged ALDH were immunoprecipitated with anti-FLAG beads from detergent extracts of the cells. The precipitates were analyzed by Western blotting with anti-GST (middle panel) and anti-FLAG (top panel) to assess coprecipitation. GST-tagged Gpi8p was coprecipitated with FLAG-tagged Gaa1p (lane 5), but not with FLAG-tagged ALDH (lane 4). GST-tagged ALDH was not coprecipitated with FLAG-tagged Gaa1p (lane 6), indicating a specific interaction between Gaa1p and Gpi8p. Analysis of the supernatant after immunoprecipitation with anti-FLAG beads by means of glutathione beads (bottom panel) demonstrated that more than half of GST-tagged Gpi8p was associated with FLAG-tagged Gaa1p (lanes 7 and 8).
Nematode Gpi8p contains only 322 amino acids and lacks a transmembrane domain. To determine the functional importance of a region of human Gpi8p including a transmembrane domain, we constructed mutants in which various lengths of the carboxyl-terminal portion were deleted (Figure (Figure7A).7A). Gpi8p mutant 321del bearing 321 amino acids and lacking the transmembrane domain retained its activity to complement class K mutant cells, indicating that the transmembrane domain is not necessary (Figure (Figure7B,7B, b). A mutant 310del bearing 310 amino acids did not have any activity (a). Therefore, a region from amino acids 311–321 is critical (see Figure Figure5A5A for location of this region and the transmembrane domain).
We next tested whether the loss of function of 310del was due to a lack of formation of a complex with Gaa1p. We analyzed complex formation between amino-terminally FLAG-tagged mutant Gpi8ps and HA-tagged Gaa1p by a coprecipitation assay (Figure (Figure7C).7C). Precipitates by anti-HA antibody were blotted with anti-HA (top panel) and anti-FLAG (middle panel) antibodies. FLAG-tagged wild-type Gpi8p was coprecipitated with HA-tagged Gaa1p (lane 12) but not with HA-tagged control protein ALDH (lane 7). FLAG-tagged ALDH was not coprecipitated with HA-tagged Gaa1p (lane 8), confirming the specificity of this assay. Deletion mutants 321del and 332del with class K complementation activities formed complexes with Gaa1p (lanes 10 and 11). The nonfunctional mutant 310del also formed a complex with Gaa1p (lane 9), indicating that the transmembrane domain and amino acids 311–321 are not necessary for association with Gaa1p. Therefore, amino acids 311–321 must have another function that is essential for attachment of GPI.
A remarkable feature of the protein modification by GPI is that the carboxyl-terminal GPI attachment signal peptides do not have any consensus sequence but have only several rather nonstrict characteristics (Udenfriend and Kodukula, 1995 ). Nevertheless, they direct GPI-anchoring specifically. Characterization of the GPI transamidase that mediates this reaction is important for understanding the molecular mechanisms of GPI attachment.
In the present study, we disrupted GAA1 in murine F9 cells and found that Gaa1p is essential for GPI-anchoring of precursor proteins but not for GPI synthesis. In the absence of Gaa1p, a carbonyl intermediate between the precursor protein and the GPI transamidase was not formed. These characteristics of GAA1-disrupted cells are very similar to those of class K mutant cells that are defective in GPI8 (Yu et al., 1997 ). We also found that Gaa1p and Gpi8p formed a complex. Therefore, the two proteins are necessary for generation of the carbonyl intermediate.
Two steps are involved in generation of the carbonyl intermediate. First, the transamidase recognizes a GPI attachment signal peptide located at the carboxyl terminus of the precursor protein and presents it to the catalytic site. Second, the signal peptide should be cleaved by the catalytic site, resulting in formation of a carbonyl intermediate. It is known that GPI is not required for the transamidase to generate a carbonyl intermediate (Maxwell et al., 1995a ). Gpi8p should function in the second step because it has sequence homology to cysteine proteases and because its cysteine, which is conserved among members of the cysteine protease family, is essential for its function (Figures (Figures44 and and5).5). It is not possible to predict functions of Gaa1p from its sequence because it has no significant homology to other proteins of known functions. Gaa1p could recognize the GPI attachment signal peptide. This possibility is supported by previous experiments that showed that overexpression of yeast GAA1 could partially suppress the processing defect seen in the GPI signal peptide mutants of Gas1p (Hamburger et al., 1995 ), or Gaa1p could act during formation of the carbonyl intermediate together with Gpi8p.
The GPI attachment signal peptide contains a carboxyl-terminal moderately hydrophobic region. It is thought to act as a temporary membrane anchorage of a precursor protein until it is recognized by the transamidase (Udenfriend and Kodukula, 1995 ). It was reported that substitution of valine for aspartic acid located within a hydrophobic region of GPI attachment signal of Qa-2 abolished GPI attachment and resulted in the expression as an integral transmembrane protein (Waneck et al., 1988 ). It was also reported that the hydrophobic region is highly sensitive to substitution with charged residues (Nuoffer et al., 1991 ; Yan et al., 1998 ). These results indicate that moderate hydrophobicity is required for the GPI attachment signal and suggest that a hydrophobic domain of the transamidase would recognize it. Deletion mutants of human Gpi8p lacking a transmembrane region retained full activity to complement class K cells (Figure (Figure7B),7B), indicating that another component or components should be responsible. Both yeast and mammalian Gaa1ps have several hydrophobic regions (Hamburger et al., 1995 ; Inoue et al., 1999 ). Whether they are involved in this recognition is not clear at the moment.
We found that Cys206 and His164 of human Gpi8p and Cys199 and His157 of yeast Gpi8p are conserved in nematode Gpi8p, jack bean asparaginyl endopeptidase with transamidase activity and other members of a cysteine protease family (Figure (Figure4A).4A). Those residues were in fact essential for the function of human and yeast Gpi8ps (Figures (Figures4,4, B and D, and 5B). Consistent with these results, this conserved His was predicted to be an important residue for a catalytic site of cysteine proteases (Alonso and Granell, 1995 ). The conserved Cys was not discussed in the same article, but instead another Cys (at the position of Leu66 of human Gpi8p) that is conserved in members of a cysteine protease family was speculated to be important (Alonso and Granell, 1995 ). The latter Cys, however, is not conserved in human and yeast Gpi8ps, and the same position is Leu in both Gpi8ps (Figure (Figure4A).4A). Based on this, Ser at the next position was alternatively predicted to be catalytic (Benghezal et al., 1996 ); however, this Ser is not conserved in nematode Gpi8p, and indeed, Ser67 of human Gpi8p was not important (Figure (Figure4B).4B). Therefore, we conclude that cysteines homologous to Cys206 of human Gpi8p are essential for catalytic activities of members of this cysteine protease family.
The final step of GPI attachment would be a nucleophilic attack of the carbonyl intermediate between the precursor protein and Gpi8p by the terminal amino group of the preassembled GPI. Therefore, the GPI transamidase may contain a component that binds GPI. At the moment there is no information about this putative component. Gaa1p contains a large hydrophilic amino-terminal domain that would reside in the lumen of the ER (Hamburger et al., 1995 ). This domain probably contains a binding site for Gpi8p but in addition could recognize GPI. Another possibility is the existence of a third component. Amino acids 311–321 of Gpi8p are essential for GPI attachment. This region is not necessary for association of Gpi8p with Gaa1p. It is unlikely that this region is involved in the catalytic reaction because there is no consensus amino acid in this region. Therefore, a likely role of this region is to associate with a third component. A genetic approach in yeast, isolating and characterizing mutants that are synthetic lethal with gaa1–1, may help to identify a putative third component. A biochemical approach, involving purification of the protein complex containing Gaa1p and Gpi8p, could also lead to identification of additional components.
We thank Reika Watanabe for discussion and Keiko Kinoshita for technical assistance. This work was supported by grants from the Human Frontier Science Program and the Ministry of Education, Science, Sports and Culture of Japan.