Biosynthesis of Glycosyl Phosphatidylinositol Membrane Anchors*

One kind of lipid anchor, is a glycosyl phosphatidylinositol (GPI). The anchor of the Trypanosoma brucei variant surface glycoprotein exemplifies the GPI solution to membrane attachment. Its structure consists of phosphatidylinositol linked glycosidically to a linear tetrasaccharide composed of one glucosaminyl and three mannosyl residues. The terminal mannose, at the nonreducing end of the glycan, is linked to phosphoethanolamine, and the ethanolamine is in amide linkage to the α-carboxyl group of the protein's C-terminal amino acid residue

Faced with the problem of membrane attachment, proteins resort to several solutions. For example, many bond to biological membranes via some type of lipid modification. One kind of lipid anchor, described only in the last few years, is a glycosyl phosphatidylinositol (GPI)' (see Refs. l-3 for reviews). GPIs anchor an extensive array of cell surface hydrolases, surface antigens, and adhesion molecules to plasma membranes; they are found in organisms ranging from protozoans, yeast, and slime mold to Drosophila and man. None have been reported in prokaryotes or in plants. Recently, detailed structural analyses of several GPI anchors have been performed (see below), and the GPI biosynthetic pathway in one organism has been elucidated (see below).

GPI Structure
The anchor of the Trypunosoma brucei variant surface glycoprotein (VSG) exemplifies the GPI solution to membrane attachment. Its structure ( Fig. 1) consists of phosphatidylinositol linked glycosidically to a linear tetrasaccharide composed of one glucosaminyl and three mannosyl residues. The terminal mannose, at the nonreducing end of the glycan, is linked to phosphoethanolamine, and the ethanolamine is in amide linkage to the cY-carboxyl group of the protein's Cterminal amino acid residue (4). This structure contains several novel linkages, as well as a non-acetylated glucosamine; the latter compound is rare in eukaryotic oligosaccharides, although common to all GPI structures described (3).
Remarkably, the core portion of the VSG anchor (consisting of phosphatidylinositol, the linear tetrasaccharide, and phosphoethanolamine) is also present in the GPI of rat brain Thy-1 (5) and in the GPI of human erythrocyte acetylcholinesterase (6,7). Where linkages have been determined they are also conserved. Despite these similarities, there are many colorful variations on the shared GPI motif (Fig. 2). These include several modifications of core mannosyl residues, such as the attachment of variously sized galactosyl side chains (in trypanosome VSG (4, 8)), extra phosphoethanolamines (in rat brain Thy-l (5) and in human erythrocyte acetylcholinester- ase (6)), GalNAc, or additional mannosyl residues (rat brain Thy-l (5)). The inositol may be modified by an extra fatty acid (human erythrocyte acetylcholinesterase (7)). There is also considerable variation in the fatty acyl or fatty alkyl groups linked to glycerol; the VSG anchor is unusual in containing exclusively myristate (see below). Because of the apparent strict conservation of the phosphatidylinositol-glucosamine-mannosea-phosphoethanolamine core, it is likely that the basic biosynthetic pathway for GPIs is conserved across a spectrum of organisms. Although this review concentrates on the membrane anchors of proteins, the biosynthetic processes described may also apply to GPI anchors of carbohydrate antigens (9, lo), or to free GPI structures, like the one implicated as an insulin second messenger (11).

How Do GPIs Attach
to Proteins?
cDNA sequences for GPI-anchored proteins exhibit a characteristic pattern. They encode a typical N-terminal signal sequence, which directs the protein to the endoplasmic reticulum, and a C-terminal sequence, which predicts 20-30 predominantly hydrophobic amino acids. The predicted C-terminal sequences are absent in mature (GPI-anchored) polypeptides (for examples, see Refs. 12 and 13); presumably they are removed during glycolipid addition.
In the case of the trypanosome VSG, replacement of the Cterminal peptide with an anchor occurs within 1 min of polypeptide synthesis (14, 15). This rapid processing suggested that the GPI may be preconstructed, as shown in the model in Fig. 3. According to this model, the C-terminal peptide serves transiently to tether the newly synthesized polypeptide 60 the endoplasmic reticulum membrane. The ethanolamine amino group of the GPI precursor may make a nucleophilic attack on the appropriate peptide bond of the protein to be anchored, resulting in transfer of the protein to the GPI.
Several experiments indicate that the transfer of protein to a preformed GPI occurs in the endoplasmic reticulum. These include the rapidity of GPI addition after protein synthesis (mentioned above), as well as study of a yeast mutant (secl8). In secl8 cells vesicular transport of glycoproteins from the rough endoplasmic reticulum to the Golgi apparatus is blocked at 37 "C, but GPI anchors are still added correctly to a 125-kDa plasma membrane glycoprotein; this finding provides direct evidence that GPI addition occurs in the former compartment (16).
GPI addition to a protein has not yet been demonstrated in uitro. However, in the case of placental alkaline phosphatase, cleavage of the C terminus of the primary translation product occurs in the presence of a microsomal extract (17). This cleavage could be due to an abortive reaction catalyzed by the transferase which usually, in the presence of the appropriate GPI precursor, attaches the protein to its anchor.

Signals for GPI Addition
One focus of recent study has been the identification of the sequence determinants, within a given polypeptide, that direct addition of GPI anchors. Such signals probably reside in or near the C-terminal sequences removed during anchor addition. However, these sequences for various GPI-anchored proteins share few features other than the presence of a hydrophobic region in the portion displaced by the anchor. These hydrophobic regions differ from typical transmembrane domains, as they often include several charged or hydrophilic residues. Identification of sequences responsible for GPI addition has usually involved transfection of mammalian cells with an expression vector carrying a cDNA encoding a GPI-anchored protein; the cDNA may be altered in the region of interest by various mutagenic procedures. Localization of the protein product by immunofluorescence, demonstration of its labeling with ['Hlethanolamine, and determination of its sensitivity to bacterial phosphatidylinositol-specific phospholipase C have all been used to determine whether it is GPI-anchored to the plasma membrane. Several studies of this type involving decay-accelerating factor (a complement regulatory protein), Qa-2 (a glycoprotein found on some hematopoietic cells), and placental alkaline phosphatase prove that signals for GPI addition must reside near the C terminus (for examples, see . In the case of decay-accelerating factor, transfection with a cDNA encoding glycoprotein D (a herpesvirusencoded protein that is normally secreted) linked to the 37residue C-terminal sequence of decay-accelerating factor yielded glycoprotein D GPI-anchored to the plasma membrane (18). Additional studies showed that a hydrophobic sequence of 17 residues, at the extreme C terminus of the decay-accelerating factor precursor, can be replaced by other hydrophobic sequences, or even by a randomized version of the wild type sequence, without perturbing GPI addition (21). These data indicate that extensive sequence variation is permitted in this hydrophobic domain. Other mutational studies indicate that some residues in a hydrophilic sequence just upstream of the hydrophobic 17 residues are also essential for GPI addition (22); the residue to which the GPI is added (not yet identified) may reside in that upstream sequence. Additional mutational analysis of this hydrophilic region of the decay-accelerating factor precursor, as well as similar analyses of other proteins, may help to clarify the rules for GPI addition. It does appear, however, that C-terminal hydrophobic regions, as well as hydrophilic regions near anchor attachment sites, are involved in proper anchor addition.

The GPI Biosynthetic Precursor: Glycolipid A
The model for GPI addition in Fig. 3 postulates a preformed glycolipid anchor to which protein is transferred. T. brucei contains a glycolipid with many properties consistent with a role as an anchor precursor (23,24). This glycolipid, termed glycolipid A, can be radiolabeled with glucosamine, mannose, ethanolamine, phosphate, and myristate (as predicted from the known anchor structure); the kinetics of labeling are those expected of a precursor for the VSG anchor. Glycolipid A shares with the VSG anchor susceptibility to several GPIspecific phospholipases and specific chemical cleavage reactions. The two molecules are also antigenically related. Like the VSG anchor, the only fatty acid in glycolipid A is myristic acid. Unlike the VSG anchor, it contains no galactose (see below). So, although the structure of glycolipid A has not been rigorously determined, all of its known properties (23, 24) are consistent with the structure shown schematically in Fig. 2.

Biosynthesis of Glycolipid A
To address the biosynthesis of GPIs in T. brucei, a cell-free system for the biosynthesis of glycolipid A was developed. This system, consisting of washed trypanosome membranes, allows incorporation of radiolabeled sugars or fatty acids into glycolipid A, and into a series of less polar biosynthetic intermediates (25).* Structural characterization of the intermediates, and examination of their precursor-product relationships through pulse-chase experiments, led to the proposed pathway of glycolipid A biosynthesis (shown in Fig. 4 (25, 26)).
The first step in the biosynthesis of glycolipid A is the transfer of GlcNAc from UDP-GlcNAc to phosphatidylinositol (PI; Fig. 4, a and b). Whether this glycosylation step is specific for a certain class of PIs is not known. The resulting N-acetylglucosaminyl PI (GlcNAc-PI) is then deacetylated to form glucosaminyl PI (GlcN-PI; Fig. 4c (26)). The two activities involved in GlcN-PI formation, the sugar transfer from UDP-GlcNAc to endogenous PI and the deacetylation of GlcNAc-PI, have been demonstrated in the cell-free system (26), and both have been detergent-solubilized." Once GlcN-PI has been formed, it is sequentially mannosylated in the presence of GDP-mannose (Fig. 4,). The immediate sugar donor of these steps has not been determined. GDP-mannose may be the direct donor of one or more residues, and a species related to mannosyl phosphoryldolichol appears to be involved. Evidence for the latter includes failure of a cell line deficient in mannosyl phosphoryldolichol synthase activity to form GPI structures (27,28). Also, inhibitor studies indicate the involvement of mannosyl phosphoryldolichol in the biosynthesis of trypanosome GPI species (29).
The triply mannosylated GPI intermediate is next modified by the incorporation of phosphoethanolamine from an as yet unknown donor (Fig. 4g (25)). Based on studies with the cellfree system, this phosphoethanolamine donor, present in trypanosome membranes, is probably not CDP-ethanolamine (25); it may be an ethanolamine containing lipid, perhaps phosphatidylethanolamine (a phosphoethanolamine donor to oligosaccharides in other systems (30)). Phosphoethanolamine incorporation yields a compound designated glycolipid A' (Fig. 4g)  It is now clear that glycolipid A' and all earlier GPI intermediates contain fatty acids more hydrophobic than myristate (25).' These fatty acids are replaced with myristate late in biosynthesis through a series of remodeling steps (Fig. 4, g-j). Studies in the cell-free system indicate that the first steps are the removal of a fatty acid from glycerol and its replacement, at position sn-2, with myristate from a myristoyl-coenzyme A donor. Subsequently the other fatty acid is replaced by myristate to form glycolipid A, but the details of this reaction are not yet worked out.'

Modifications of the Glycan Core
Up to fatty acid remodeling, the biosynthetic pathway ( Fig.  4) yields the core structure common to the several GPIs (from different species) that have been studied in detail. Therefore, these steps could be common to GPI biosynthetic pathways in all cells. Little information is available about when noncore components of GPI structures (see Fig. 2) are added. In the trypanosome case, galactosylation of the core is a late modification of the completed anchor; these residues are added roughly 15 min after the anchor is attached to protein (31, 32), possibly in the Golgi apparatus. In general, anchor modifications may be governed by the structure of the anchored protein, the cell type, or the metabolic state of the cell.

Defective
GPI Biosynthesis Is Implicated in a Human Disease Patients with paroxysmal nocturnal hemoglobinuria experience periodic hemolysis due to increased sensitivity of blood cells to autologous complement-mediated lysis (for review see Refs. 33,34). This is due to the absence of two proteins, decay-accelerating factor and C8 binding protein, from the cell surface of erythrocytes, platelets, and leukocytes. Both are GPI-anchored proteins (35-37). On normal cells, these molecules act to prevent such incidental host cell lysis during complement activation.
Since the defective cells also lack acetylcholinesterase, alkaline phosphatase, and other GPIanchored proteins, it is likely that the molecular basis of