Segregation of Glycosylphosphatidylinositol Biosynthetic Reactions in a Subcompartment of the Endoplasmic Reticulum*

Glycosylphosphatidylinositols (GPIs) are synthesized in the endoplasmic reticulum (ER) via the sequential addition of monosaccharides, fatty acid, and phosphoethanolamine(s) to phosphatidylinositol (PI). While attempting to establish a mammalian cell-free system for GPI biosynthesis, we found that the assembly of mannosylated GPI species was impaired when purified ER preparations were substituted for unfractionated cell lysates as the enzyme source. To explore this problem we analyzed the distribution of the various GPI biosynthetic reactions in subcellular fractions prepared from homogenates of mammalian cells. The results indicate the following: (i) the initial reaction of GPI assembly, i.e. the transfer of GlcNAc to PI to form GlcNAc-PI, is uniformly distributed in the ER; (ii) the second step of the pathway, i.e.de-N-acetylation of GlcNAc-PI to yield GlcN-PI, is largely confined to a subcompartment of the ER that appears to be associated with mitochondria; (iii) the mitochondria-associated ER subcompartment is enriched in enzymatic activities involved in the conversion of GlcN-PI to H5 (a singly mannosylated GPI structure containing one phosphoethanolamine side chain; and (iv) the mitochondria-associated ER subcompartment, unlike bulk ER, is capable of the de novosynthesis of H5 from UDP-GlcNAc and PI. The confinement of these GPI biosynthetic reactions to a domain of the ER provides another example of the compositional and functional heterogeneity of the ER. The implications of these findings for GPI assembly are discussed.

Glycosylphosphatidylinositols (GPIs) 1 were originally discovered covalently linked to eukaryotic cell-surface glyco-proteins and were recognized to be an important alternate mechanism for anchoring proteins to cell membranes (1,2). GPI-anchored proteins have the general structure protein-(ethanolamine-P-6Man␣1-2Man␣1-6Man␣1-4GlcN␣1-6myo-inositol-1-P-lipid), where the structure within parentheses represents a minimal GPI moiety, and the carboxyl terminus of the protein is amide-linked to the GPI ethanolamine residue. GPIs can also be found in cells as free glycolipids representing protein anchor precursors or biosynthetic intermediates. Cellular pools of non-protein-linked GPIs vary depending on their structure as well as cell type, and their physiological significance remains to be determined.
Many aspects of the GPI biosynthetic pathway can be reproduced in cell-free systems using cell lysates or subcellular membrane fractions supplemented with appropriate soluble metabolites (e.g. UDP-GlcNAc and GDP-mannose). Initial work with membrane fractions prepared from African trypanosomes showed that it was possible to achieve the synthesis of a range of GPI structures from GlcNAc-PI to the phosphoethanolamine-containing protein anchor precursors (23,24). While attempting to establish a comparable cell-free system for the assembly of mammalian GPI structures, we found that the biosynthesis of mature GPI species was significantly impaired when purified ER preparations were substituted for unfractionated or coarsely fractionated cell lysates as the enzyme source (8,25). Since the ER was previously identified as the site of initiation of GPI biosynthesis (10), these preliminary data suggested that factors required for one or several post-initiation reactions might have been lost from the ER preparation during the fractionation procedure. Alternatively, cell homogenization may have resulted in fragmentation of the ER resulting in the separation of biosynthetically active ER subfractions from "bulk" ER; these subfractions may sediment differently from conventionally isolated ER during the fractionation procedure, resulting in an apparent loss of activity. Growing evidence of the lateral compositional heterogeneity of the ER (10) and the recent isolation of functionally specialized subregions of the ER (27,28) lend credence to the latter possibility. In order to explore these issues we undertook an analysis of the distribution of the various GPI biosynthetic reactions in subcellular fractions prepared from homogenates of mammalian cells. The results, described in this paper, indicate that postinitiation GPI biosynthetic reactions are most active in a subcompartment of the ER. This "GPI-active" subcompartment of the ER co-fractionates with mitochondria and probably corresponds to a previously described ER fraction shown to be active in phospholipid synthesis (27), and possibly also involved in phospholipid trafficking between the ER and mitochondria (29). . Protease inhibitor mixture and tunicamycin were obtained from Calbiochem. TLCK, leupeptin, and Staphylococcus aureus nuclease were from Roche Molecular Biochemicals. Rabbit reticulocyte lysate was obtained from Promega (Madison, WI). Plasmid pGEM-4Z/miniPLAPSer was a gift from Dr. Sidney Udenfriend (formerly at the Roche Institute of Molecular Biology, Nutley, NJ). GPI-PLD was a gift from Dr. Peter Bü tikofer (University of Bern, Bern, Switzerland). Anti-ribophorin I antibodies were provided by Dr. Christopher Nicchitta (Duke University Medical Center, Durham, NC). Antibodies against the yeast mitochondrial proteins Hsp60 and Hsp70 were provided by Dr. Elisabeth Craig (University of Wisconsin, Madison, WI). Rabbit polyclonal anti-GRP78 (BiP) antibodies were purchased from Affinity Bioreagents, Inc. (Golden, CO). Horseradish peroxidase-conjugated anti-rabbit IgG was from Santa Cruz Biotechnology (Santa Cruz, CA). The ECL Western blotting detection system for chemiluminescent detection of Western blots was obtained from Amersham Pharmacia Biotech (Buckinghamshire, UK). Glass-backed silica 60 thin layer plates were from Merck (Darmstadt, Germany). GlcN[ 14 C]Ac-PI(diC16) and GlcN-PI(diC16) were gifts from Dr. Michael Ferguson (University of Dundee, Dundee, Scotland).
Cell Culture-The Thy-1 ϩ mouse thymoma cell line BW5147.3 was maintained in suspension culture in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum, 100 units/ml penicillin, and 100 g/ml streptomycin in an atmosphere of 5% CO 2 at 37°C. The CHO-K1 cell lines G9PLAP (stably expressing human placental alkaline phosphatase (PLAP)) and its derivative G9PLAP.85 (a mutant defective in the second step of GPI anchor assembly) were provided by Dr. Victoria Stevens (Emory University School of Medicine, Atlanta, GA) and described previously (30). CHO-K1 cells were cultured in 5-dm dishes in Ham's F-12 medium supplemented with 10% fetal calf serum.
Preparation of Subcellular Fractions-Subcellular fractions were prepared from CHO cells or thymoma cells (see Fig. 2 for an outline of the fractionation procedure). For typical fractionation experiments, CHO-K1 G9PLAP or G9PLAP.85 cells were scraped from 30 dishes into 5 ml/dish of phosphate-buffered saline. BW5147.3 thymoma cells (1 ϫ 10 9 ) were collected by centrifugation (1,000 ϫ g for 5 min). The cells were washed twice with phosphate-buffered saline and resuspended in 10 ml of buffer A (0.25 M sucrose, 10 mM Hepes/NaOH, pH 7.5, 1 mM DTT) supplemented with protease inhibitor mixture. Cells were transferred to a 15-ml nitrogen cavitation bomb (Kontes Glass Co., Vineland, NJ) and held at 400 pounds/square inch for 30 min. The cavitated cells were treated with S. aureus nuclease (30 units/ml) in the presence of 1 mM CaCl 2 for 20 min on ice. The treatment was stopped by adding 1 mM EGTA. Cells were further disrupted by three passages through a 21gauge needle. The lysate was clarified by centrifugation at 1,000 ϫ g for 10 min at 4°C to remove large debris and nuclei. The resulting supernatant was centrifuged at 10,000 ϫ g for 10 min at 4°C. The pellet (P10) was washed twice with buffer A and was resuspended in buffer A for enzyme activities measurement or in isolation buffer (250 mM mannitol, 25 mM Hepes/NaOH, pH 7.5, 0.5 mM EGTA) for further purification at a protein concentration of 10 -15 mg/ml. The supernatant from the 10,000 ϫ g spin (S10) was centrifuged at 100,000 ϫ g for 1 h (Ti 50.2 rotor (Beckman Instruments Inc., Fullerton, CA)). The crude microsome pellet (P100) so obtained was resuspended in 2 ml of buffer A at a concentration of 5-10 mg/ml protein. In some experiments the supernatant was layered on top of a discontinuous sucrose gradient (2 ml of 30% and 4 ml of 38% sucrose prepared in 10 mM Hepes/NaOH, pH 7.5) and fractionated by centrifugation at 100,000 ϫ g for 2 h. The pellet and 38% sucrose step including the interface were combined, diluted at least 3-fold with buffer A, and collected by centrifugation at 100,000 ϫ g for 2 h. The resulting pellet, containing relatively pure endoplasmic reticulum membranes (10), was resuspended in buffer A. Subcellular fractions were snap-frozen in liquid nitrogen and stored at Ϫ70°C.
Preparation of the Mitochondria-associated Membrane (MAM) Fraction and Mitochondria-MAM and mitochondrial fractions were isolated following procedures described for rat liver (27,31) and CHO cells (29) with minor modifications. Briefly, the P10 pellet (obtained as described above) was resuspended in 2 ml of isolation buffer at ϳ10 -15 mg/ml and layered on top of a 20-ml solution of 30% (v/v) Percoll (prepared in 225 mM mannitol, 25 mM Hepes/NaOH, pH 7.5, 1 mM EGTA) in a centrifuge tube. Following 30 min centrifugation at 95,000 ϫ g a light upper band (running approximately half-way down the tube) corresponding to mitochondria-associated membranes (MAM) and a denser lower band corresponding to mitochondria were recovered. The two fractions were collected separately and diluted at least 2-fold with isolation buffer. The mitochondrial fraction was subjected to sonication with a 3-mm probe at 60-V output (sonicator model W185F, Heat Systems-Ultrasonics Inc., Plainview, NY) for 10 s on ice. Both fractions (MAM and mitochondria) were then centrifuged (10,000 ϫ g, 15 min) yielding pellets (MAM p and M, respectively) and supernatants. The pellets were washed 3 times by repeated centrifugation at 10,000 ϫ g for 15 min to remove Percoll and then suspended in buffer A. Since the supernatants from both fractions showed similar marker enzyme activities, they were combined and collected by centrifugation at 100,000 ϫ g for 1 h. The resulting pellet (MAM s ) was washed twice at 75,000 rpm for 30 min in a Beckman TLA 100.3 rotor and resuspended in 2 ml of buffer A.
Characterization of Subcellular Fractions-The protein content of each fraction was measured with the Micro BCA Protein Assay Reagent (Pierce). Subcellular fractions were characterized by measuring organelle-specific marker enzyme activities or by determining the amount of marker proteins via Western blot analysis. The following marker enzyme activities were assayed using previously described procedures. NADPH-cytochrome c reductase activity (ER marker) was assayed as described previously (29) using 10 -30 g of protein for P100 fraction and 30 -100 g of protein for the other fractions. The determination of succinate dehydrogenase (mitochondrial marker) activity was carried out as previously reported (32). The ER proteins ribophorin I, protein disulfide isomerase, and BiP (heavy chain binding protein (GRP78)), and the mitochondrial proteins Hsp60 and Hsp70 were quantitated using standard Western blotting procedures. After probing with primary antibodies (chicken anti-protein disulfide isomerase or rabbit anti-BiP, anti-Hsp60, and anti-Hsp70), blots were developed using peroxidase-conjugated second antibodies and the ECL Western blotting detection system following the protocol provided by the manufacturer. Band intensities were determined by densitometry (at least two different amounts of protein (typically 1 and 2.5 g) were analyzed and several different film exposures were obtained to verify linearity of the film response for densitometric analysis). Table I and Fig. 2, mitochondrial markers (Hsp60 and succinate dehydrogenase) were recovered mainly in the P10 fraction. However, the ER markers NADPH cytochrome c reductase, ribophorin I, BiP, and protein disulfide isomerase were present in both the P10 and P100 fractions (Table I; data for protein disulfide isomerase are not shown), indicating that the P10 fraction contains a substantial amount of ER. Ribophorin I, BiP, and protein disulfide isomerase were only slightly (1.1-1.5-fold) enriched in P100 versus P10, whereas NADPH cytochrome c reductase activity (specific activity as well as total) was about 4 times higher in the P100 fraction than in the P10 fraction. A similar trend was also seen in ER fractions obtained from the P10 and P100 (see below). The results point to compositional heterogeneities within the ER, consistent with a variety of other reports (Ref. 10; see Ref. 27 for specific data on NADPH cytochrome c reductase).

Summary of Characteristics of Subcellular Fractions-As shown in
Approximately 45% of the total ER marker activities measured in the P10 fraction were recovered in MAM s ( Table I). The MAM p and Mito fractions contained ϳ20 and ϳ7%, respectively (figures represent averages of the data obtained with the markers listed in Table I), of the total ER marker activities measured in the P10 fraction. Taking into account that recovery of the ER markers was ϳ70% after the fractionation procedure, this result indicates that the majority of the ER membranes present in P10 are recovered in the MAM fractions (MAM s ϩ MAM p ), specifically MAM s . The mitochondrial content of each of the three fractions was determined by measuring levels of succinate dehydrogenase, Hsp60, and Hsp70. Succinate dehydrogenase was recovered mainly in the MAM p and Mito fractions (ϳ30 and ϳ36% of the P10 total level, respectively), although a significant amount (ϳ22% of the total P10 amount) was also recovered in MAM s . Broadly similar results were obtained with the Hsp70 and Hsp60 markers. The recovery of these proteins in Mito and MAM p was 39 and 20% of the amount in P10, respectively, with only 13% being recovered in MAM s . It is possible that the slightly higher recoveries of succinate dehydrogenase compared with Hsp60/Hsp70 in all 3 fractions are due to the fact that succinate dehydrogenase is a mitochondrial membrane protein, whereas the Hsps are soluble mitochondrial matrix proteins that would most likely be lost from damaged mitochondria.
In Vitro Biosynthesis of GlcNAc-PI and GlcN-PI-Procedures for in vitro biosynthesis of GlcNAc-PI and GlcN-PI and product verification were essentially as described previously (10,33,34). Briefly, isolated fractions (typically 50 g of protein) were incubated with UDP-[ 3 H] GlcNAc (1 Ci) in buffer A in the presence of 5 mM EDTA, in a total volume of 50 l at 37°C. In some experiments 1 g/ml PI (dissolved in 0.1% Nonidet P-40 (the final concentration of Nonidet P-40 in the reaction mixture was 0.01%)) was also added. The addition of exogenous PI equally stimulated the formation of GlcNAc-PI (about 1.5-2-fold) in all examined fractions, but no difference in the ratios of products was observed. Incorporation of [ 3 H]GlcNAc into products increased linearly for up to 30 min. The reaction was stopped by placing samples on ice, and a single phase lipid extract was obtained by adding 150 l of water and 1.5 ml of ice-cold chloroform/methanol (1:2, v/v). A two-phase mixture was induced by adding 0.5 ml of chloroform and 0.5 ml of water. The lipid containing chloroform-rich lower phase was washed several times with 0.5 ml of mock upper phase, dried, and dissolved in watersaturated n-butyl alcohol. The products were analyzed by thin layer chromatography (silica gel 60 thin layer plates, using chloroform/methanol, 1 M ammonium hydroxide 10:10:3, v/v/v, as the solvent) and detected with a TLC radioscanner (Berthold Analytical Instruments, Inc., Nashua, NH). Incorporation of radioactivity into the individual species was determined using the integration software supplied with the scanner.
De-N-acetylase Assay Using Synthetic GlcN[ 14 C]Ac-PI Substrate-Each subcellular fraction to be assayed was adjusted to contain 150 g of protein suspended in 250 mM sucrose, 50 mM Hepes, pH 7.5, 10 mM EDTA in a reaction volume of 100 l. All measurements were carried out in duplicate. The de-N-acetylation reaction was started by the addition of 5 l of a 0.1% Nonidet P-40 solution of GlcN[ 14 C]Ac-PI(diC16) (containing 2500 cpm, ϳ80 pmol), and samples were incubated at 37°C for 4 h. The reactions were terminated by removing 100 l of the incubation mixture at the appropriate time points and adding it to 50 l of 100% propan-1-ol, vortexing, and snap-freezing. The samples were thawed, diluted with 100 l of 1 M ammonium acetate and 750 l of water (resulting in approximately 1 ml of 5% propan-1-ol, 100 mM ammonium acetate per reaction), and passed through a pre-equilibrated 100-mg C8 Sep-Pak cartridge (Waters Corp., Milford, MA). The cartridge was then washed with a further 1 ml of 5% propan-1-ol, 100 mM ammonium acetate. The combined washes (ϳ2 ml, containing released [ 14 C]acetate) were taken for liquid scintillation counting. Control incubations carried out with boiled membrane preparations were processed identically and used as assay background. Release of [ 14 C]acetate was confirmed by strong anion exchange high pressure liquid chromatography analyses using a 5-500 mM ammonium acetate gradient: the released radioactivity co-chromatographed with a [ 3 H]acetate standard while other possible GlcN[ 14 C]Ac-PI cleavage products (e.g. GlcNAc, GlcNAc-inositol, GlcNAc-inositol-P, GlcNAc-inositol-P-glycerol) eluted in the void volume or bound so strongly to the column that they could only be eluted with a follow-up potassium phosphate gradient (66).
Translation-Translocation Assays-Prepromini-PLAP mRNA was prepared from plasmid pGEM-4Z/miniPLAPSer (35,36) by HindIII linearization and SP6 RNA polymerase-mediated transcription. In vitro translation-translocation reactions were performed in nuclease-treated rabbit reticulocyte lysate as described previously (35,37). To initiate the translocation reaction, 1 l (5 g) of membranes was added to 11 l of translation mixture. The reaction was allowed to proceed for 60 min at 25°C, after which the products were either directly analyzed by SDS-PAGE (12.5% gels) and fluorography or first subjected to proteinase K treatment prior to SDS-PAGE. For proteinase K treatment, the translation-translocation reaction samples were placed on ice, diluted to 50 l with 150 mM KOAc, 50 mM Hepes/NaOH, pH 7.5, 2.5 mM Mg(OAc) 2 , 1 mM DTT, and proteinase K was added to a final concentration of 100 g/ml. After 30 min incubation on ice, proteolysis was stopped by adding 500 l of buffer A supplemented with 2 mM phenylmethylsulfonyl fluoride (prepared freshly from a 150 mM stock in ethanol). The membranes were collected by centrifugation (Beckman TLA 100.2 rotor, 75,000 rpm, 45 min, 4°C) through a 100-l sucrose cushion (0.5 M sucrose in 50 mM Hepes/NaOH, pH 7.5). The pellet was dissolved in SDS-PAGE sample buffer and analyzed as described above.

Synthesis of GlcNAc-PI and GlcN-PI in Subcellular
Fractions-We investigated the synthesis of early GPI biosynthetic intermediates (GlcNAc-PI and GlcN-PI; Fig. 1) in subcellular fractions obtained from thymoma cells. Thymoma cell lysates were prepared by nitrogen cavitation and, following the removal of nuclei and unbroken cell debris, were subjected to differential centrifugation to obtain fractions sedimenting at 10,000 ϫ g (P10) and at 100,000 ϫ g (P100) (Fig. 2). The P10 fraction is typically characterized as a crude mitochondrial preparation but has been shown previously to contain also a subfraction of ER tightly associated with the mitochondrial membrane (27,38,39). The P100 fraction consists primarily of bulk ER membranes but also contains other membranes, including those derived from the Golgi and plasma membrane. Both the P10 and P100 fractions were able to synthesize GlcNAc-PI and GlcN-PI; the biosynthetic activity for the combined synthesis of both lipids on a per mg of protein basis was slightly higher (ϳ1.3-fold) in the P10 fraction compared with P100 (Fig. 2). In order to examine this result more closely, we isolated ER membranes from the S10 (Fig. 2) by sucrose gradient centrifugation, and subfractionated the P10 material on Percoll gradients (Fig. 2).
The S10-derived ER fraction was generally similar to the P100 in terms of marker enzyme distribution, except that all markers (ER markers and residual mitochondrial markers) were about 1.5-2-fold enriched relative to P100 ( Fig. 2 and Table I). Subfractionation of P10 yielded three distinct fractions (mitochondria (Mito), MAM s , and MAM p ; see Fig. 2; the acronym MAM stands for mitochondria-associated membranes (27)) which we characterized in terms of ER and mitochondrial marker enzyme activities before testing for GPI biosynthetic capability. The overall analysis of marker enzyme distribution (see Fig. 2, Table I, and "Experimental Procedures" for details) is consistent with the following assignments for the P10-derived fractions. 1) The Mito fraction consists primarily of mitochondria with very little ER contamination. 2) MAM s contains most (Ն70%) of the ER membranes that are found in the P10, along with some mitochondrial contamination (in contrast ER membranes isolated from the S10 contain relatively little mitochondrial contamination; see Fig. 2 and Table I). Mitochondrial markers recovered in MAM s may be carried by mitochondrial membrane fragments or damaged mitochondria since MAM s represents a relatively light membrane mixture that pellets only during a 100,000 ϫ g spin. 3) MAM p , which has marker enzyme characteristics intermediate between MAM s and Mito, and quite similar to that of P10, may consist of intact and/or damaged mitochondrial structures in association with ER. Since MAM p is recovered as a pellet from a 10,000 ϫ g spin, it represents a relatively dense membrane fraction.
The four fractions (S10-derived ER, and P10-derived Mito, MAM s , and MAM p ) were tested for their ability to synthesize GlcNAc-PI and GlcN-PI from UDP-[ 3 H]GlcNAc and PI. Since it is difficult to uncouple the synthesis of GlcN-PI from GlcNAc-PI in normal cells, the combined synthesis of both lipids (referred to as GlcNAc/GlcN-PI) was measured. As shown in Fig. 2, all fractions except the Mito fraction were able to synthesize GlcNAc/GlcN-PI. Although the ability to synthesize GlcNAc/GlcN-PI correlated with the presence of ER marker enzymes in a particular fraction, the MAM fractions were significantly more active than the P100 and ER fractions. For example, the specific activity of GlcNAc/GlcN-PI synthesis in MAM s was roughly 4-fold higher in comparison with P100 and 2.5 times higher in comparison with relatively pure ER (Fig. 2). Since the inclusion of exogenous PI had a similar stimulatory effect on all fractions (under our experimental conditions GlcNAc/GlcN-PI synthesis was enhanced ϳ2-fold in samples with added PI relative to samples without added PI), the observed differences in the specific activity of GlcNAc/GlcN-PI synthesis between the fractions cannot be explained by postulating that the fractions contain different amounts of one of the reaction substrates (PI). Also, differential enzyme lability in the various fractions was discounted as an explanation for FIG. 1. GPI biosynthetic pathway in mammalian cells. GPI biosynthesis is initiated by GlcNAc transfer from UDP-GlcNAc to PI yielding GlcNAc-PI. GlcNAc-PI is then de-N-acetylated, inositol-acylated, and further modified with mannose and phosphoethanolamine residues. CHO mutant cells (G9PLAP.85) defective in the PIG-L gene are unable to de-N-acetylate GlcNAc-PI. The singly mannosylated GPI intermediates that are detectable in the cell-free GPI synthesis assays shown in this paper are H2 and H5. H5 is decorated by a phosphoethanolamine residue linked to the 2-position of mannose. The mature, triply mannosylated GPIs (H7 and H8) bear two or three phosphoethanolamine side chains as shown. Biosynthetic intermediates between H5 and H7 are not shown.

FIG. 2. Subcellular fractionation and intracellular distribution of GlcNAc/GlcN-PI synthesis.
The subcellular fractionation procedure is described in detail under "Experimental Procedures." Briefly, cells were disrupted by nitrogen cavitation, and the mixture was clarified by low speed centrifugation to remove large debris and nuclei. The resulting supernatant (homogenate) was centrifuged at 10,000 ϫ g to obtain P10 (pellet) and S10 (supernatant) fractions. S10 was processed either by direct centrifugation at 100,000 ϫ g to yield P100 (pellet) or by sucrose gradient centrifugation to yield an ER fraction (referred to as S10-derived ER). P10 was resolved into two fractions (lower band and upper band) by centrifugation through a Percoll gradient. The bands were collected and respun at 10,000 ϫ g. The pellets so obtained were resuspended resulting in mitochondrial (Mito) and MAM p fractions. The supernatants from the 10,000 ϫ g spin were combined and respun at 100,000 ϫ g to obtain MAM s (pellet). Fractions were characterized by measuring marker enzyme activities. The panels in the lower part of the figure show the specific activities of early GPI biosynthetic enzymes (the data show the combined synthesis of GlcNAc-PI and GlcN-PI; cpm ϫ 10 Ϫ3 /min/mg protein; black bars), ribophorin I (ER marker; arbitrary units/mg protein; white bars), and Hsp60 (mitochondrial marker (M); arbitrary units/mg protein; gray bars) in subcellular fractions from thymoma cells. Data are presented as mean Ϯ S.E. differences in the specific activity of GlcNAc/GlcN-PI synthesis since recovery of activity in the subfractionation steps was excellent (see Fig. 3B, for example); ϳ93% of the total GlcNAc/ GlcN-PI synthetic activity of P100 was recovered in the ER fraction and ϳ102% of the activity of P10 was recovered in the combined MAM fractions. These data suggest that ER mem-branes (MAM s , and MAM p ) recovered from P10 are not only capable of initiating GPI biosynthesis but possess a somewhat higher specific activity for the formation of GlcNAc/GlcN-PI in comparison with ER isolated more conventionally from the S10 (Fig. 2).
Further Analysis of GlcNAc/GlcN-PI Synthesis in Subcellular Fractions-In the experiments described above, GPI biosynthesis in the various subcellular fractions was assessed as the combined synthesis of GlcNAc-PI and GlcN-PI. In order to determine the extent of synthesis of each of the two lipids, lipid extracts were resolved by thin layer chromatography (TLC), and the incorporation of radioactivity into GlcNAc-PI and GlcN-PI was quantitated. The results are shown in Fig. 3. Although the P10 and P100 fractions were able to synthesize comparable amounts of combined products (gray bars, Fig. 3B, P10, 1080 cpm ϫ 10 Ϫ3 /min; P100, 870 cpm ϫ 10 Ϫ3 /min; specific activity data are shown in A), the relative proportions of the two products differed significantly (hatched bars, GlcNAc-PI; black bars, GlcN-PI; Fig. 3). The yield of GlcN-PI was much higher in P10 versus P100, corresponding to 62 and 32% of the total lipid synthesized, respectively (Fig. 3B). The ER and MAM fractions appeared much like the parental P100 and P10 fractions, respectively, in terms of the yield of GlcN-PI relative to GlcNAc-PI. The yield of GlcN-PI (relative to GlcNAc-PI ϩ GlcN-PI) was 24, 45, and 72% in the ER, MAM s , and MAM p fractions, respectively (Fig. 3B). The apparent specific activity for GlcN-PI synthesis (cpm ϫ 10 Ϫ3 /min/mg protein) was 15, 20.5, 43.4, 80, and 90.6 for P100, ER, P10, MAM s , and MAM p , respectively. These data suggest that the GlcNAc-PI de-Nacetylase activity responsible for GlcN-PI synthesis is more strongly represented in P10 than in P100 and concentrated in the P10-derived MAM fractions.
Distribution of GlcNAc-PI De-N-acetylase Activity in Subcellular Fractions-In order to test the hypothesis that the GlcNAc-PI de-N-acetylase activity is enriched in the P10 and MAM fractions relative to P100 and ER, we assayed the de-Nacetylase directly by incubating the fractions with chemically synthesized GlcN[ 14 C]Ac-PI(diC16) and measuring the release of radioactive acetate (Fig. 4). Very low levels of de-N-acetylation were seen in the crude homogenate (H), P100, and mitochondrial (Mito) fractions. Although it is certain that the P100 possesses some de-N-acetylase activity (see Fig. 3A, for exam-

TABLE I Distribution of ER and mitochondrial markers in subcellular fractions from thymoma cells
The table presents the protein content and specific marker enzyme activities (mean Ϯ S.E.) of subcellular fractions from thymoma cells. Fractions were prepared from 10 9 thymoma cells as described under "Experimental Procedures" and outlined in Fig. 2. Ribophorin, BiP, and NADPH:cytochrome c reductase are ER markers. Hsp60 and succinate dehydrogenase are mitochondrial markers. Ribophorin, BiP, and Hsp60 were quantitated by Western blotting and densitometry, and the data are presented in arbitrary units. The units of NADPH:cytochrome c reductase activity and succinate dehydrogenase activity are nmol of cytochrome c reduced per min and nanomoles of p-nitrotetrazolium violet reduced per min/mg of protein, respectively. Total activities are presented in parentheses.  ple, for GlcN-PI synthesis by P100), the activity is low and appears to be at the limit of detection of the GlcN-[ 14 C]Ac-PIbased assay. The P10 fraction was active, as were the P10-and S10-derived ER fractions. The P10-derived ER fractions (MAM s and MAM p ) had 10.8-and 7.3-fold higher specific de-N-acetylase activity than the S10-derived ER (Fig. 4). Moreover, the total GlcNAc-PI de-N-acetylase activity recovered in these fractions was 4.8-and 2.2-fold higher for MAM s and MAM p , respectively, than in the S10-derived ER. Thus, the total GlcNAc-PI de-N-acetylase activity in the combined MAM fractions is ϳ7-fold higher than in the S10-derived ER. These data indicate that the GlcNAc-PI de-N-acetylase is greatly concentrated in the MAM fractions relative to the S10-derived ER.

Synthesis of GlcNAc-PI in Subcellular Fractions Derived from Wild-type CHO Cells and CHO Cells Defective in Glc-
NAc-PI De-N-acetylation-The results described above are consistent with the hypothesis that the enzymes responsible for the synthesis and de-N-acetylation of GlcNAc-PI are differently distributed in subcellular fractions. In order to explore this idea further we repeated the analysis with CHO cells to test whether the subcellular fractionation pattern of activities that we had obtained with thymoma cells also applied to CHO cells. The experiments showed that the pattern of GlcNAc/GlcN-PI synthesis in CHO cell fractions was identical to that seen in thymoma cell fractions (Figs. 2 and 3), and as with the thymoma cell fractions, the relative synthesis of GlcN-PI was higher in the P10 and MAM fractions (data not shown).
A mutant strain of Chinese hamster ovary (CHO) cells (G9PLAP.85) has been recently isolated and shown to be defective in GlcNAc-PI de-N-acetylase activity (30). With these cells it is possible to uncouple the GlcNAc-PI synthesis reaction from the de-N-acetylation step and to measure the specific activity of GlcNAc-PI synthesis explicitly in various subcellular fractions. Subcellular fractions were prepared from the CHO G9PLAP.85 cells and assayed for GlcNAc-PI synthesis. The results, shown in Fig. 5, indicate that all the fractions were generally efficient at synthesizing GlcNAc-PI. In the particular experiment shown, the P10 fraction was slightly more active in synthesizing GlcNAc-PI (ϳ1.4-fold) than the P100, and the P10-derived ER (MAM s ) was only marginally more active (ϳ1.2-fold) than the S10-derived ER. The results indicate that, unlike GlcNAc-PI de-N-acetylase activity, GlcNAc-PI synthetic activity is similarly distributed among the ER subfractions isolated from P10 and P100.
In Vitro Biosynthesis of Mannosylated GPI Species from UDP-GlcNAc-Elaboration of GlcN-acyl-PI in mammalian cells occurs via the addition of mannose and phosphoethanolamine residues (Fig. 1). These biosynthetic steps have not been studied individually thus far since the various GPI reaction substrates (GlcN-acyl-PI and partially mannosylated GPI structures) are not generally available, and cell-free assay conditions remain to be worked out. Efficient delivery of lipid substrates is perhaps the greatest hurdle to be overcome in assaying these later biosynthetic steps in vitro since the reactions are known to be detergent-labile. Nevertheless, it is possible to derive information on the elaboration of GlcN-acyl-PI by incubating subcellular fractions with more distant substrates, such as radioactive sugar nucleotides, and tracking the incorporation of radioactivity into mannosylated GPI structures. In order to extend our analyses of the subcellular distribution of GPI biosynthesis to include these "later" biosynthetic reactions, we therefore incubated the various subcellular fractions ( Fig. 2) with GDP-[ 3 H]Man, in the presence or absence of nonradioactive UDP-GlcNAc, and we analyzed the radiolabeled lipids formed by organic solvent extraction and TLC (Fig. 6). Related experiments in which fractions were incubated with GDP-[ 3 H]Man in the presence or absence of nonradioactive GlcN-PI are described in the legend to Fig. 7. Fig. 6, A and B, shows representative chromatograms of radiolabeled lipids synthesized from GDP-[ 3 H]Man in the MAM p fraction. The major radiolabeled lipid synthesized in the absence of UDP-GlcNAc was dolichol-P-mannose (dol-P-Man) (Fig. 6A). Minor but detectable amounts of singly and triply mannosylated GPIs (H5, H7, and H8, see Fig. 1) were also synthesized. The extent of incorporation of radiolabel into GPI species (H8) under these conditions (no added UDP-GlcNAc) was roughly similar in all P100-and P10-derived ER fractions examined, i.e. ER, MAM s , and MAM p .
When the reaction was supplemented with nonradioactive UDP-GlcNAc, biosynthesis of the GPI intermediate H5 (EtN-P-2Man␣1-4GlcN-acylPI) increased 3-4-fold in the MAM s and MAM p fractions (Fig. 6, C and D; see Fig. 6, A and B, for TLC scans showing H5 synthesis in MAM p ). In contrast, no increase in the formation of mature GPI species (H7 and H8) was observed in any of the fractions (Fig. 6, C and D) indicating that the synthesis of mature [ 3 H]mannose-labeled GPI species occurs predominantly via [ 3 H]mannose transfer to endogenous acceptors (GlcN-acyl-PI and partially mannosylated GPIs) rather than to de novo synthesized GlcN-acyl-PI species. Al- The data are presented as specific activities (cpm/min/mg; mean Ϯ S.E.). Total recovery of activity (cpm/min; mean values) was 38, 160, 180, 84, and 6 for ER, P10, MAM s , MAM p , and Mito, respectively. The total de-N-acetylase activity in MAM s ϩ MAM p (264 cpm/min) is ϳ1.7fold greater than that in P10 (160 cpm/min). One explanation for this observation is that the de-N-acetylase assay (unlike other assays used in this paper) uses an exogenously added lipid substrate, GlcN-[ 14 C]Ac-PI. Deposition of this lipid in biosynthetically inactive membranes such as the Mito component of the P10 would result in removing it from the assay pool. Since MAM s ϩ MAM p have fewer "GPI-inactive" membranes than P10, the total activity recovered in these fractions appears higher.
FIG. 5. Distribution of GlcNAc-PI synthase in subcellular fractions from CHO cells. GlcNAc-PI biosynthesis in fractions from CHO mutants defective in GlcNAc-PI de-N-acetylase activity. Subcellular fractions (50 g of protein) prepared from CHO G9PLAP.85 mutant cells (5 ϫ 10 8 cell equivalents) were incubated with UDP-[ 3 H]GlcNAc as described under "Experimental Procedures." Lipids were extracted, analyzed, and quantitated as described in Fig. 3. Homogenate and MAM p were not determined (ND). The data are presented as specific activities (cpm ϫ 10 Ϫ3 /min/mg; mean Ϯ S.E.). Total recovery of activity (cpm ϫ 10 Ϫ3 /min; mean values) was 183, 181, 215, 100, and 6 for P100, ER, P10, MAM s , and Mito, respectively. MAM s accounts for ϳ47% of the Glc-NAc-PI synthase activity in the P10; the remainder is probably contained in MAMp that was not analyzed in these experiments. though all fractions tested were able to synthesize mannosylated GPI species (H5, H7, and H8) to some extent, the UDP-GlcNAc-dependent stimulation of H5 biosynthesis was only seen in the MAM fractions. Variations in synthesis of dolichol-P-mannose, the mannose donor, are unlikely to account for this observation since dolichol-P-mannose synthesis differed by no more than ϳ1.7-fold between the various fractions and was lower in the MAM fractions relative to ER (see Fig. 6 legend).
The results presented above indicate the following: (i) that the MAM fractions, unlike the S10-derived ER, are capable of the de novo synthesis of H5, and (ii) de novo synthesis of more elaborate GPI structures (e.g. H7 and H8) does not occur in either the S10-derived ER or the MAM fractions. Possible reasons for the latter result are detailed under "Discussion." The de novo (UDP-GlcNAc-dependent) synthesis of H5 in the MAM fractions may be a result of the concentration of GlcNAc-PI de-N-acetylase activity in these membranes (Fig. 4B) rather than a consequence of correspondingly high levels of the rele-vant inositol acyl-, mannosyl-, and phosphoethanolaminetransferases. However, results presented below indicate that the MAM is also enriched in these latter enzymes.

Post-GlcNAc-PI De-N-acetylation Reactions Leading to the Biosynthesis of H5 Are Concentrated in the MAM Fractions-
The synthesis of H5 from GlcNAc-PI involves four reactions yielding the biosynthetic intermediates GlcN-PI, GlcN-acylPI, and H2; phosphoethanolamine transfer to the mannose residue in H2 yields H5 (Fig. 1). We were interested in discovering whether the ability of the MAMs to synthesize H5 de novo is due solely to their enhanced ability to synthesize GlcN-PI or whether all reactions involved in the conversion of GlcNAc-PI to H5 are enriched in the MAMs. To test these possibilities, we assayed the conversion of exogenously added GlcN-PI to [ 3 H]Man-labeled H5. This approach allowed us to bypass the GlcNAc-PI de-N-acetylase and directly assess the cumulative activity of the following "H5 biosynthetic enzymes": GlcN-PI inositol acyltransferase, mannosyltransferase, and phosphoethanolaminetransferase (Fig. 1).
When subcellular fractions were incubated with GDP-[ 3 H]Man in the presence or absence of nonradioactive GlcN-PI(diC16), [ 3 H]Man-labeled H5(diC16) was synthesized. Fig. 7 shows that P10, MAM s , and MAM p were considerably more effective at converting GlcN-PI(diC16) to [ 3 H]Man-labeled H5 (white arrowheads in Fig. 7A) than P100 and ER. Quantitation of the data (Fig. 7B) showed that the apparent specific activity of H5(diC16) biosynthesis was ϳ6.6-fold higher in P10 relative The distribution of radioactivity in the chromatograms was visualized using a TLC scanner and quantitated using the integration software supplied with the scanner in conjunction with liquid scintillation counting. A and B, thin layer chromatograms of GPIs synthesized from GDP-[ 3 H]mannose (ϩ/Ϫ UDP-GlcNAc) in the MAM p fraction. C and D, incorporation of radioactivity into the mannosylated GPIs H5, H7, and H8 in S10-derived ER, MAM s , and MAM p . Results are presented as specific activities (cpm ϫ 10 Ϫ3 /min/mg). Dol-P-Man synthesis (cpm ϫ 10-3 /min/mg) was 22.5, 19.6, and 16.8 for ER, MAM s , and MAM p , respectively, in the absence of UDP-GlcNAc (C), and 20.4, 15.2, and 12.1 for ER, MAM s , and MAM p , respectively, in the presence of UDP-GlcNAc. The data indicate that dol-P-Man is synthesized similarly in all fractions tested. to P100 and ϳ5.2-fold higher in the MAM fractions relative to ER. The relative enrichment of these biosynthetic activities in the P10 and MAMs is likely to be higher than indicated since the assay readout was not linear over the time periods required to obtain detectable levels of [ 3 H]Man-labeled H5. The lack of linearity is likely due to the complexity of the incubation mixtures, suboptimal assay conditions, and the fact that the assay measures the cumulative output of three enzymes. Fig. 7B also shows the incorporation of radioactivity into dolichol-P-mannose, the donor of the mannose residue in H5; synthesis of dolichol-P-mannose was comparable in all fractions but higher in P100 and ER relative to P10 and the MAM fractions, respectively. Thus reduced levels of dolichol-P-mannose synthesis cannot account for low H5 synthesis in P100 and ER. These results show clearly that the inositol acyltransferase, mannosyltransferase, and phosphoethanolaminetransferase required to convert GlcN-PI to H5 are, in combination, enriched in the P10-derived ER membranes relative to S10-derived ER. The relative enrichment of these activities in the MAM versus ER (ϳ6-fold) is less than that obtained for the GlcNAc-PI de-Nacetylase (ϳ9-fold, Fig. 4), but as pointed out above, this figure is likely to be an underestimate.
Translocation and Processing of Prepromini-PLAP in Isolated Fractions-The final step in the assembly of GPI-anchored proteins is the transamidase-catalyzed replacement of a carboxyl-terminal signal sequence by a pre-formed GPI anchor, following protein translocation across the ER membrane. A model protein, prepromini-PLAP (derived from placental alkaline phosphatase; PLAP), has been extensively used to study conversion of proproteins to the mature GPI-anchored form in cell-free assays (7). Upon membrane translocation and aminoterminal signal sequence cleavage, prepromini-PLAP is converted to a lower molecular weight, protease-protected form, promini-PLAP. Promini-PLAP is subsequently converted to GPI-anchored mini-PLAP which also resists protease treatment of intact vesicles. We tested the ability of P10, P100, ER, and MAM s to translocate and convert prepromini-PLAP into GPI-anchored form (mini-PLAP). Although it is impossible to deduce specific activities for the translocation and GPI-anchoring steps using this assay, all four fractions were able to translocate and GPI anchor prepromini-PLAP (data not shown). Protease-protected products corresponding to promini-PLAP and mini-PLAP were generated in every case at approximately similar levels, and the amount of mini-PLAP synthesized relative to the total amount of protease-protected material was also similar in all fractions (data not shown).

DISCUSSION
The results described in this paper point to an unusual segregation of GPI biosynthetic reaction(s) in a subcompartment of the ER (Fig. 8). The cell-free assembly of GPI intermediates subsequent to the initial GlcNAc transfer reaction and up to the formation of the singly mannosylated species H5 is substantially enhanced in this ER subcompartment compared with conventionally isolated ER. This ER subcompartment appears to be similar to the one previously identified by a number of investigators as a region of ER tightly associated with mitochondria (27, 28, 38 -42). The association between mitochondria and regions of the ER was demonstrated morphologically and biochemically in these earlier studies and shown to be a general feature of many cell types (e.g. rat hepatocytes, CHO cells, S. cerevisiae). Cell homogenization apparently disrupts the connection between this mitochondria-associated ER subcompartment (termed mitochondria associated membrane or MAM (27)) and the rest of the ER resulting in its isolation, together with mitochondria, in a 10,000 ϫ g pellet (P10, Fig. 2).
We fractionated the P10 material to generate two MAM fractions, MAM s and MAM p (Fig. 2), which probably represent operational variants of the same membrane structure. Most of the ER content and all of the GlcNAc/GlcN-PI biosynthetic activity of the P10 were recovered in the MAM fractions (Table  I and Fig. 2). The MAM fractions contain mitochondrial marker proteins, a likely consequence of the close association between the MAM and mitochondria that is readily apparent in electron microscopic images of intact cells and isolated MAM fractions (31, 38 -40). The mitochondrial inner membrane marker succinate dehydrogenase was found to be preferentially enriched in the MAM relative to the mitochondrial matrix marker, Hsp60 (Table I), indicating that the MAM fractions are probably contaminated with mitochondrial membrane fragments (possibly contact site membranes (31)) or damaged mitochondria. However, the absence of GlcNAc/GlcN-PI biosynthetic activity in mitochondria (Figs. 2 and 4) indicates that the GPI biosynthetic activity detected in the MAM can be directly attributed to ER membranes rather than to mitochondria.

Subcellular Distribution of GlcNAc-PI De-N-acetylase and Enzymes Required for the Conversion of GlcN-PI to H5-The
MAMs represent hot spots of GlcNAc-PI de-N-acetylase and H5 biosynthetic activity in the ER as judged by our subcellular fractionation results. The specific activity of GlcNAc-PI de-Nacetylase was 9.4-fold higher in the combined MAM fractions (MAM s ϩ MAM p ) relative to the S10-derived ER, and the total activity recovered in the MAM fractions was 7-fold higher than that recovered in the ER. However, immunofluorescence visu-  Fig. 2). The MAM and bulk ER are separated upon cell homogenization and are recovered separately from the P10 and S10 fractions, respectively. GlcNAc-PI is synthesized equally efficiently in the ER and MAM fractions, but de-N-acetylation of GlcNAc-PI occurs preferentially in the MAM (the specific activity of the de-N-acetylase in MAM is ϳ9-fold higher than in the ER and the total activity recovered in the MAM (MAM s ϩ MAM p ) is ϳ7-fold higher than in the ER). It is unclear whether GlcNAc-PI can diffuse between the two membrane domains and whether GlcNAc-PI synthesized in the bulk ER can access the de-N-acetylase in the MAM fraction (dashed line). De novo biosynthesis of the singly mannosylated GPI intermediate H5 occurs in the MAM but not in the ER. The location(s) at which H5 is elaborated to the mature structures H7 and H8 is unknown. A potential complication is the topological distribution of the mannosylation steps in the ER and/or MAM; trans-bilayer movement (double-headed dashed arrow) of lipid intermediates (H5, dolichol-P-mannose) may be required to allow synthesis of mature GPIs, but it is unknown whether such movement occurs in the MAM. Dolichol-P-mannose, the mannose donor for GPI mannosylation reactions, is synthesized equally well in the ER and MAM fractions and is known to undergo trans-bilayer movement in the ER. For further details, see under "Discussion." alization of the distribution of the PIG-L protein, a candidate GlcNAc-PI de-N-acetylase, showed a homogeneous reticular staining pattern very similar to that of the ER resident protein disulfide isomerase (43). The absence of heterogeneity in the immunofluorescence distribution of PIG-L (an epitope-tagged version of PIG-L was used for these studies) may be attributed to the relatively low resolution of the technique or to the possibility that the localization of PIG-L may not provide an accurate representation of the membrane distribution of the de-Nacetylase activity; for example, the active form of the de-Nacetylase may be a complex between PIG-L and another, hitherto unidentified, heterogeneously distributed (MAM-localized) protein (9).
The enzymes responsible for inositol acylation, mannosylation, and phosphoethanolamine modification of GlcN-PI are also concentrated in the MAM. Owing to difficulties in assaying these activities explicitly, we measured these enzymes in combination by monitoring the conversion of exogenously added GlcN-PI to H5. The combined activity was enriched (ϳ6-fold) in the P10 and MAM fractions relative to P100 and ER.
Our data show that the MAM fractions are uniquely able to carry out the de novo synthesis of H5 (Fig. 6, C and D). It is clear from the results summarized in this section and shown in Figs. 3, 4, and 7 that this unique capability is due to the enrichment of GlcNAc-PI de-N-acetylase as well as the H5 biosynthetic enzymes in the MAM.
Biosynthesis of Mature GPIs-The de novo synthesis of GPIs proceeds up to the formation of the intermediate H5 in the MAM fraction. Further mannosylation and phosphoethanolamine addition reactions appear to be so severely attenuated that it is impossible to detect de novo synthesized mature GPIs (H7 and H8). The MAM fraction undoubtedly has the ability to synthesize these structures since low levels of radiolabeled H7 and H8 can be detected, but synthesis appears to occur primarily via mannose transfer to endogenous acceptors rather than via de novo synthesis from UDP-GlcNAc and PI (Fig. 5).
The inability of de novo synthesized H5 to progress further along the GPI assembly pathway may be explained in a number of ways. We enumerate a few possibilities as follows. (i) Dolichol-P-[ 3 H]mannose synthesized in the MAM is present only in radiochemical amounts and is possibly insufficient to supply the second and third GPI mannosyltransferases at reasonable rates leading to detectable H7 and H8. However, in vitro biosynthesis of mature [ 3 H]mannose-labeled GPIs proceeds readily, albeit not at optimal rates, in other cell-free systems primed with GDP-[ 3 H]mannose (23,24,51), suggesting that other factors must also contribute to the elaboration of H5. (ii) The second mannosyltransferase (GPIMT-II) is highly unstable and consequently non-functional in the subcellular fractions. (iii) The efficient synthesis of H5 is a result of substrate channeling through a metabolon (44) composed of the de-N-acetylase, inositol acyltransferase, and the first GPI mannosyltransferase; discharge of H5 from the metabolon and consequent dilution into the membrane milieu would result then in the inefficient, essentially undetectable synthesis of more elaborate GPIs. Substrate channeling in this segment of the GPI biosynthetic pathway has been suggested previously on the basis of results obtained with trypanosomal and Leishmania cell-free systems for GPI biosynthesis (45) although other data (46) suggest that this may not occur in mammalian systems. (iv) GPIMT-II and its substrates are topologically separated (Fig. 7). The substrates for GPIMT-II are H5 and dolichol-P-mannose (the mannose donor in the GPIMT-II catalyzed mannosyl transfer reaction in trypanosomes and presumably also in mammalian cells (47)), both of which are synthesized in the MAM (Fig. 5 and legend). The prevailing evidence (6,8,48) suggests that the two lipids are synthesized on the cytoplasmic face of the endoplasmic reticulum membrane and, in the absence of lipid translocators (flippases), may well be confined to this membrane leaflet. Although GPIMT-II has yet to be identified, its active site may face the ER lumen, similar to the PIG-B gene product that probably encodes the third GPI mannosyltransferase (49). Trans-bilayer movement of H5 and dolichol-P-mannose would overcome such a topological barrier. Protein-mediated flipping of dolichol-P-mannose has been described in "microsomal" ER preparations (50), and it remains to be seen whether such an activity also exists in the MAM. There is no information on trans-bilayer movement of GPIs except that such movement is likely to occur (6,8,51,52). At this point, none of the possibilities discussed above can be ruled out, and further investigations are clearly necessary in order to understand the membrane localization and regulation of the later steps of GPI biosynthesis.
ER Heterogeneity and MAM Function-The confinement of GPI biosynthetic reactions to a domain of the ER provides yet another example of the compositional and functional heterogeneity of the ER. The functional diversity of ER membranes and the existence of functionally specialized and morphologically distinct subregions of the ER have been the subjects of several recent reviews (26,(53)(54)(55). The existence and functional uniqueness of a mitochondria-associated ER domain have been documented in a number of studies using biochemical and ultrastructural methods. For example, Wassler and Fries (28) showed that a protease involved in post-translocational processing of the secretory protein haptoglobin is localized to ER membranes that sediment similarly to mitochondria after cell homogenization. Connections between this membrane fraction and the remainder of the ER could be re-established through GTP-dependent membrane fusion, resulting in the effective processing of haptoglobin. It is not clear why the protease is confined to a mitochondria-associated ER domain and whether mitochondria contribute in any way to the control of the proteolytic event. In another example, the ER proteins calreticulin and the inositol 1,4,5-triphosphate receptor were shown to be clustered in regions of the ER that are closely associated with mitochondria and that appear to define sites of enhanced Ca 2ϩ release from ER stores (56 -58). Here the functional significance of a mitochondria-associated ER subcompartment is somewhat clearer since mitochondrial activity was shown to be required to preserve cytosolic Ca 2ϩ responses (59).
Mitochondria-associated ER domains have a clearly defined function in the transfer of lipids between the ER and mitochondria (29,60,61). Phosphatidylserine (PS) which is synthesized in the ER must be transported to the mitochondrial inner membrane in order to be decarboxylated to phosphatidylethanolamine (PE) (62). PE, in turn, must be exported from the mitochondria to participate in ER-localized N-methylation reactions (63) involved in the synthesis of phosphatidylcholine. In rat liver, the PE N-methyltransferase, PEMT-2, is uniquely localized to the MAM (63), and the activity of PS synthase in liver and CHO cells is about 2.5-fold higher in the MAM than in microsomal ER fractions (29,64).
At this point it is difficult to explain the precise physiological significance of our finding that post-initiation GPI biosynthetic reactions are localized to the MAM. It is possible that the phosphoethanolamine residues in the GPI structure are preferentially derived from PE synthesized via the mitochondrial PS decarboxylation pathway rather than PE generated via the ER-localized CDP-ethanolamine pathway (65). Phosphoethanolamine transfer reactions in a MAM-localized GPI biosynthetic pathway would thus be adequately primed with PE since PE is exported from mitochondria to the MAM (29). Alterna-tively, the differential distribution of GlcNAc-PI de-N-acetylase (restricted to the MAM) versus GlcNAc-PI synthase (uniformly distributed in the ER) may serve to generate two functionally distinct pools of GlcNAc-PI. Although it is conceivable that GlcNAc-PI synthesized in other regions of the ER in intact cells diffuses into the MAM to be de-N-acetylated and subsequently elaborated, it is tempting to speculate that GlcNAc-PI serves an independent function in the cell and that non-MAM pools of GlcNAc-PI are preserved for this purpose by restricting their access to the de-N-acetylase.