Sphingolipids Are Required for the Stable Membrane Association of Glycosylphosphatidylinositol-anchored Proteins in Yeast*

Ongoing sphingolipid synthesis is specifically required in vivo for the endoplasmic reticulum (ER) to Golgi transport of glycosylphosphatidylinositol (GPI)-anchored proteins. However, the sphingolipid intermediates that are required for transport nor their role(s) have been identified. Using stereoisomers of dihydrosphingosine, together with specific inhibitors and a mutant defective for sphingolipid synthesis, we now show that ceramides and/or inositol sphingolipids are indispensable for GPI-anchored protein transport. Furthermore, in the absence of sphingolipid synthesis, a significant fraction of GPI-anchored proteins is no longer associated tightly with the ER membrane. The loose membrane association is neither because of the lack of a GPI-anchor nor because of prolonged ER retention of GPI-anchored proteins. These results indicate that ceramides and/or inositol sphingolipids are required to stabilize the association of GPI-anchored proteins with membranes. They could act either by direct involvement as membrane components or as substrates for the remodeling of GPI lipid moieties.

Sphingolipids are a major lipid component in eucaryotic cells. They are not only membrane components, but they also have important functions in regulation of cellular events (1). In Saccharomyces cerevisiae, sphingolipids are involved in stress responses, calcium homeostasis, regulation of cell growth, cell cycle control, and membrane traffic (1).
Glycosylphosphatidylinositol (GPI) 1 -anchored proteins are a group of proteins that are expressed on the surface of eucaryotic cells. They are modified post-translationally with a glycolipid anchor in the ER, which provides an alternative mechanism to a transmembrane domain for membrane insertion. In yeast, the ER to Golgi transport of GPI-anchored proteins has requirements distinct from that of other secretory proteins (2)(3)(4)(5). One of these is ongoing sphingolipid synthesis. This requirement has been identified by the observation of impaired trafficking of these proteins in several mutants defective at different steps of sphingolipid biosynthesis. These mutants include lcb1-100, which have a temperature-sensitive serine palmitoyltransferase, the first enzyme in sphingolipid biosynthesis (3,4), lag1 lac1 double mutant cells, which are defective in acyl-CoA-dependent ceramide synthesis (6 -8), and fen1 (elo2) and sur4 (elo3) mutants, which are defective in the synthesis of C26 fatty acyl-CoA, a precursor of ceramides and sphingolipids in yeast (9,10). These results suggest strongly that certain sphingolipids are required specifically for the efficient transport of GPI-anchored proteins, but so far the relevant sphingolipids have not been identified.
GPI-anchored proteins also are transported from the ER to the Golgi apparatus in distinct vesicles from other secretory proteins (11). The sorting of GPI-anchored proteins is dependent on the proper functions of the Rab GTPase Ypt1p and the tethering factors Uso1p and COG complex (Sec34/35p) (12). In addition to these protein factors, sphingolipid/sterol-enriched detergent-insoluble microdomains, also known as lipid rafts, may also function in the sorting of GPI-anchored proteins. Rafts may function in several cellular events such as signal transduction, protein sorting, and lipid traffic (13). In yeast, raft-like structures have been proposed to play several roles in protein sorting (14,15). It was shown that GPI-anchored proteins are detergent-insoluble in the ER, and thus, rafts have been proposed to have a sorting function there (14). Defining the precise function of sphingolipids in GPI-anchored protein transport in yeast may unravel the sorting mechanism of GPIanchored proteins in the ER and the significance of raft structure in their transport.
In this study, we show that ceramide and/or inositol sphingolipids are indispensable for GPI-anchored protein transport to the Golgi compartment. Interestingly, we found that in the absence of ceramide, a significant fraction of GPI-anchored proteins is no longer associated tightly with the ER membrane but behaves like peripheral membrane protein. This loose membrane association was neither because of the lack of a GPI-anchor attachment nor to the retention of the GPI-anchored proteins in the ER. The lack of tight membrane association is likely to be one of the reasons for the inefficient transport of GPI-anchored proteins to the Golgi.

MATERIALS AND METHODS
Strains, Plasmids, Media, and Reagents-The strains used in this study are listed in Table I. RH5465 was obtained by crossing RH2043 with RH3804. Yeast cells were grown overnight in semi-synthetic medium, SDYE (2). To express wild-type and mutant Gas1p (L526R) in gas1 knock-out strains, we used plasmids YEplac195-GAS1 and PC-NYG-(L526R) (16), respectively. To construct YEplac195-GAS1, a 2.4kbp fragment of the GAS1 gene including the promoter region was amplified by PCR using the primers 5Ј-GCTCGGGCATGCGAATTCA-CAGGCCAGCCCTGGCTA and 5Ј-GCTCGGGAGCTCCAGCTGATAT-TATGGAGAAAGTACATA. The fragment was subcloned into YE-plac195 (2 m; URA3) using SphI and SacI sites in the fragment and the vector. To induce Gap1p expression we precultured strains in SUD (1.6 g/liter yeast nitrogen base without ammonium sulfate and amino acids, 20 g/liter glucose, and 1.0 g/liter urea plus required supplements) overnight before starting cultures in SDYE. Stereoisomers of dihydrosphingosine (DHS) and phytosphingosine (PHS) were purchased from Matreya (Pleasant Gap, PA) and Sigma, respectively, and were dissolved in 100% ethanol at 10 mM as stock solutions.
Sphingolipid Analysis-In vivo [ 3 H]myo-inositol labeling was performed as described previously (17) with slight modifications. Briefly, cells were grown overnight in SDYE at 24°C to 0.5-2 ϫ 10 7 /ml, harvested, and suspended in S.D. medium without inositol. 2 ϫ 10 8 cells were used for each sample and were preincubated at 37°C for 15 min in the absence or presence of 50 M DHS stereoisomers and then labeled with 25 Ci of [ 3 H]myo-inositol (10 -25 Ci/mmol in H 2 O; PerkinElmer Life Sciences) for 60 min. The reactions were terminated by addition of chloroform:methanol (1:1, v/v). After removal of the insoluble fraction by centrifugation, the supernatant was dried. Half of the dried sample was treated by mild alkaline hydrolysis with 0.6 M NaOH in methanol to degrade glycerolipids, and lipids were re-extracted and desalted by n-butanol partitioning. The lipids were separated by TLC using Silica gel 60 plates (5553; Merck) in solvent system I (chloroform:methanol: 0.25% KCl (55:45:10) per vol), and radiolabeled lipids were visualized using a Cyclone Storage Phosphor System (Packard). Ceramide synthesis was assayed in vitro using stereoisomers of [ 3 H]DHS as described previously (18). The D-erythro and L-erythro stereoisomers of [4, H]DHS were synthesized as described (19). Microsomal membranes and cytosol from wild-type cells were incubated with [4, H]DHS (2,000,000 cpm) at 24°C for 2 h in the presence or absence of 100 M fumonisin B1, a specific inhibitor of acyl-CoA-dependent ceramide synthesis (20). Lipids were extracted as described above and analyzed by TLC in solvent system II (chloroform:acetic acid (9:1) (v/v)).
Pulse-Chase Experiments for Gas1p Maturation and GPI Anchor Attachment-Radiolabeling and immunoprecipitation to measure Gas1p maturation were performed as described previously (3). Briefly, cells were grown in SDYE at 24°C to 0.5-2 ϫ 10 7 /ml, harvested, and resuspended in SD medium without methionine and cysteine. 5 ϫ 10 7 cells were used for each time point and preincubated at 37°C for 15 min in the absence or presence of the individual stereoisomer of DHS or PHS or the specific inhibitors and labeled with 100 Ci of EasyTAG TM express protein labeling mix, [ 35 S] (1175 Ci/mmol in aqueous solution; PerkinElmer Life Sciences) for 6 min. The chase was initiated by addition of cold methionine and cysteine in 0.3 M (NH 4 ) 2 SO 4 to 1% final concentration, respectively. The reactions were terminated by adding both NaN 3 and NaF to 10 mM final concentration. The radiolabeled cells were suspended in TEPI (100 mM Tris-HCl, pH 7.5, 10 mM EDTA, protease inhibitors) and lysed with glass beads. The lysates were boiled in the presence of 1% SDS for 5 min and centrifuged to remove insoluble material. The supernatant was diluted four times with TNET (100 mM Tris-HCl, pH 8, 100 mM NaCl, 5 mM EDTA, 1% Triton X-100) and used for immunoprecipitation of Gas1p with anti-Gas1p rabbit antiserum and protein A-Sepharose (Amersham Biosciences). The samples were separated by SDS-PAGE and analyzed and quantified using the Cyclone Storage Phosphor System (Packard). The percentage of mature Gas1p was determined by calculating the ratio of the 125-kDa mature form to the total signal (125-and 105-kDa ER form) and multiplying by 100. GPI anchor attachment was examined as described (3) with slight modifications. In brief, crude cell extracts from radiolabeled cells were solubilized with Triton X-114 at a final concentration of 1%. After two rounds of partitioning into detergent and aqueous phases, the detergent phase was diluted 10 times with TEPI and divided into two fractions, which were incubated in the presence or absence of 0.1 unit of phosphatidylinositol-specific phospholipase C (PI-PLC) (Roche Molecular Biochemicals) at 30°C for 3 h. Triton X-114 phase separation was repeated, and proteins were precipitated from the final detergent and aqueous phases using 10% trichloroacetic acid. The precipitates were resuspended and boiled in 1% SDS for 5 min, and samples were processed for Gas1p immunoprecipitation. Unanchored Gas1p segregated into the primary aqueous phase. The anchored Gas1p partitioned into the primary detergent phase and shifted into the aqueous phase after PI-PLC treatment. The measurement of Gas1p released into the medium was done as described previously. Pulse-chase experiments were done in the presence of 0.125 mg/ml bovine serum albumin and ovalbumin in the medium (16).
Membrane Association Assay-To follow the membrane association of transport intermediates, we pulse-labeled cells for 6 min and chased for 15 min as described above. The radiolabeled cells were suspended in TEPI and lysed with glass beads. After removal of insoluble material by centrifugation at 2,500 ϫ g for 5 min, the crude cell extracts were divided into three fractions. One was mixed with an equal amount of TEPI, one was mixed with 0.2 M Na 2 CO 3, pH 12, in TEPI (to give final pH of 11), and one was mixed with 2% (w/v) Triton X-100 in TEPI and incubated on ice for 30 min. Soluble and pellet fractions were prepared by centrifugation at 150,000 ϫ g for 30 min at 4°C. The soluble fractions from 0.1 M Na 2 CO 3 treatment were precipitated with 10% trichloroacetic acid, and other fractions were solubilized directly at 55°C for 5 min by addition of SDS to 1%. All were processed for immunoprecipitation using the corresponding rabbit antiserum and protein A-Sepharose.

RESULTS
Substrate Specificities of Ceramide Synthase and IPC Synthase in Yeast-To better define the role of sphingolipids in GPI-anchored protein transport from the ER to the Golgi compartment, we first investigated the structural requirements for sphingolipid synthesis using stereoisomers of the ceramide precursor, DHS, whose structures are shown in Fig. 1A. D-erythro-DHS is the naturally occurring form. To analyze sphingolipid synthesis, we preincubated wild-type and lcb1-100 mutant cells at 37°C for 15 min in the presence or absence of each DHS isomer and then labeled sphingolipid by adding [ 3 H]myo-inositol for 60 min. As described previously (17), lcb1-100 has an extremely low level of sphingolipid synthesis ( Next, to determine the stereospecificity of ceramide synthesis, we measured ceramide synthase activity using microsomal membranes in the presence of D-erythro and L-erythro forms of 3 [H] DHS, as described previously (18). We found that dihydroceramide was synthesized from the D-erythro but not the Lerythro form of [ 3 H] DHS (Fig. 1C), suggesting that ceramide synthesis is stereospecific. We observed that the synthesis of dihydroceramide was fumonisin B1-sensitive, confirming we are measuring acyl-CoA-dependent ceramide synthesis activity (20). Identification of Indispensable Sphingolipids for GPI-anchored Protein Transport from the ER to the Golgi Compartment-Next, we studied GPI-anchored protein transport ( Fig.  2A) under the same conditions we used to assay sphingolipid biosynthesis. After a 15-min preincubation of lcb1-100 mutant cells at 37°C in the presence or absence of individual DHS stereoisomers, we labeled cells with 35 [S]-labeling mix for 6 min and chased for the indicated time. We analyzed maturation of Gas1p, whose ER form migrates at 105 kDa and whose mature form, produced after arrival at the Golgi, migrates at 125 kDa by SDS-PAGE. Mutant lcb1-100 cells show a severe delay of Gas1p maturation ( Fig. 2A, EtOH) (3). Only the two DHS stereoisomers, D-erythro and L-threo, that can be incorporated into ceramide and sphingolipids could restore ER to Golgi GPIanchored protein transport completely ( Fig. 2A). These results suggest that ceramide and/or inositol sphingolipids are required for transport. To confirm this, we studied the effects of inhibitors specific for syntheses of ceramide and inositolphosphorylceramide (IPC). In the presence of the ceramide synthesis inhibitor australifungin (21), the restoration by D-erythro-DHS was reduced significantly, suggesting that ceramide synthesis is necessary for GPI-anchored protein transport. The IPC synthesis inhibitor aureobasidin A (22) also affected the restoration efficiency somewhat, suggesting that inositol sphingolipids may play a role in transport. The incomplete inhibition by both inhibitors could be because of incomplete inhibitions of ceramide and IPC synthases under these conditions (high DHS concentration) or could be because of a partial restoration by DHS. To provide additional evidence for a requirement for ceramide synthesis, we tested transport efficiency under the condition when ceramide synthesis is eliminated almost completely by other means. The mutant lcb3, ysr3 double knock-out mutants, which are defective for dephosphorylation of phosphorylated sphingoid base, combined with lcb1-100, produce very little ceramide and inositol sphingolipid from exogenous DHS (17,23). We compared the concentration of DHS that efficiently restored transport defect in these cells versus that needed for the lcb1-100 mutant (Fig. 2B). For lcb1-100 cells, 10 M Derythro-DHS restored GPI-anchored protein transport considerably. In contrast, in lcb1/3, ysr3 cells, transport was not restored significantly even at 50 M D-erythro-DHS. From these results, we conclude that ceramide and/or inositol sphingolipid synthesis is necessary for GPI-anchored protein transport.
To address the possibility that other DHS derivatives, specifically phosphorylated sphingoid bases or PHS, might be required for GPI-anchored protein transport, we used two other mutant strains. Mutant lcb4, lcb5 double knock-out cells also carrying the lcb1-100 allele cannot produce DHS-1P from exogenous D-erythro-DHS (17,24), and sur2-disrupted cells, combined with the lcb1-100 allele, are defective for hydroxylation of exogenous DHS to form PHS (17,25). GPI-anchored protein transport was clearly restored by exogenous D-erythro-DHS in lcb1/4/5 mutant cells and by both D-erythro form of DHS or PHS in lcb1/sur2 mutant cells (Fig. 2C). These results show that phosphorylation of sphingoid bases and their interconversion between DHS and PHS are not required for GPI-anchored protein transport.
Loose Membrane Association of GPI-anchored Proteins in lcb1-100 Mutant Cells-To understand the role of sphingolipids in GPI-anchored protein transport, we attempted to use an in vitro assay that measures the efficiency of packaging of proteins into ER-derived vesicles, sorting from other secretory proteins, and the efficiency of delivery of cargo proteins to the Golgi structure (11). During the course of this study, we noticed that a significant fraction of GPI-anchored proteins was released specifically from lcb1-100 mutant membranes to the soluble fraction. This release was cytosol-independent (data not shown) and was enhanced at elevated temperature (30°C) and by addition of an energy source (ATP and GTP; data not shown). Therefore, we investigated the membrane association of GPI-anchored proteins in lcb1-100 cells using conventional fractionation methods. We preincubated mutant or wild-type cells for 15 min at 37°C, pulse-labeled for 6 min, and chased for 15 min. Then, we prepared crude cell extracts and assessed membrane association of two GPI-anchored proteins under three different conditions, buffer alone, high pH, and detergent (Fig. 3A). In lcb1-100 mutant cells, but not in wild-type cells, a significant fraction of Gas1p was no longer associated with membranes even under the buffer conditions (39% versus 7.7%). More Gas1p was released from the membrane by Na 2 CO 3, pH 11, treatment (54% versus 7.2%). Gas1p has been proposed to associate with detergent-insoluble structures in the ER (26). In accordance with these findings, Gas1p was solubilized poorly by 1% Triton X-100 from wild-type cells. However, Gas1p was solubilized almost completely from lcb1-100 mutant cells (91% versus 53%). We also observed a similar weak membrane association of another GPI-anchored protein, Yps1p, in lcb1-100 cells. In contrast to wild-type cells, most of Yps1p was in the ER form after a 15-min chase in lcb1-100 cells, confirming that the general maturation delay of GPIanchored proteins in lcb1-100 cells. We also noticed that the small amount of mature forms of Gas1p and Yps1p seen in lcb1-100 cells were associated more tightly with membranes than were the ER forms. As controls for these experiments, we examined membrane association of ␣ϪCOP (peripheral protein) and Gap1p (integral membrane protein). Both proteins behaved as expected in these cell types. From these results, we conclude that the binding of GPI-anchored proteins to the ER membrane is weakened severely in lcb1-100 mutant cells.
To rule out the possibility that prolonged ER retention is the cause of the weakened membrane association of GPI-anchored proteins in lcb1-100 mutant cells, we assayed membrane association in sec18 mutant cells, which shows a block in all ER to Golgi protein transport. The ER forms of Gas1p and Yps1p accumulated in sec18 mutant cells were not released like those in the lcb1-100 mutant. The results confirmed that the weak membrane association in lcb1-100 cells was not because of its retention in the ER.
Next, to investigate the possibility that loose membrane association might be because of inefficient GPI-anchor assembly or addition in lcb1-100 cells, we examined the partitioning behavior of Gas1p using Triton X-114 (16). We labeled cells as in Fig. 3A, prepared cell extracts, solubilized extracts in Triton X-114 at 4°C, and subjected them to phase partitioning at 32°C. Fig. 3B shows that Gas1p from wild-type and lcb1-100 cells partitions into the Triton X-114 detergent phase with the same efficiency (left columns). In addition, the Gas1p found in the detergent phase could be shifted completely to the aqueous phase by treatment with phosphatidylinositol-specific phospholipase C to remove the diacylglycerol moiety of GPI-anchor

FIG. 3. Membrane association of GPI-anchored proteins is severely weakened in lcb1-100 mutant cells.
A, membrane association of GPI-anchored proteins was studied in wild-type (RH2874), lcb1-100 (RH3804), and sec18-20 (RH5465) cells. The cells were preincubated for 15 min at 37°C, labeled for 6 min, and chased for 15 min. Crude extracts were incubated under three conditions (buffer, TEPI; CO 3 2Ϫ , 0.1 M Na 2 CO 3 , pH 11, in TEPI; TX-100, 1% Triton X-100 in TEPI). After separation by ultracentrifugation, the soluble (S) and the pellet (P) fractions were processed for immunoprecipitation using individual rabbit antiserum. B, GPI-anchor attachment was not affected in lcb1-100 cells. The same crude extracts as for A were solubilized with Triton X-114 at 1% final concentration. After partitioning into detergent and aqueous phases, the detergent phase was incubated in the presence or absence of PI-PLC. Phases were re-extracted and processed for Gas1p immunoprecipitation. Unanchored Gas1p segregated into the primary aqueous phase, whereas anchored Gas1p partitioned into the primary detergent phase and shifted into the aqueous phase after phospholipase C treatment. The total amount of Gas1p quantified in each partition was set to 100. A1 and D1, first aqueous and detergent phases, respectively (left panels); A2 and D2, aqueous and detergent phases from D1 after mock (middle panels) and PI-PLC (right panel) treatment, respectively. C, membrane association of the prenylated protein Ypt1p was studied in wild-type (RH1638) and lcb1-100 (RH5241) cells as A.
(right columns). Thus we confirmed that GPI-anchoring efficiency was not affected in lcb1-100 cells, in agreement with previous observations (3,4).
Finally to test whether the weakened membrane association of GPI-anchored proteins reflects a general defect in membrane association of lipid-anchored proteins, we assessed membrane association of the prenylated protein, Ypt1p (27). Fig. 3C shows that membrane association of Ypt1p is similar in wild-type and lcb1-100 cells, suggesting that the weak membrane association is specific for GPI-anchored proteins.
The Soluble Form of Gas1p Shows Slow Maturation in Wildtype Cells-As exogenous DHS restores GPI-anchored protein transport we checked whether it also restored membrane association of GPI-anchored proteins. D-erythro-DHS clearly restored membrane association, as well as maturation, to a wildtype levels. On the other hand, L-erythro-DHS, which was not incorporated into ceramide, did not restore membrane association of Gas1p (data not shown). These results lead to the conclusion that ceramide and/or inositol sphingolipid are necessary for stable membrane association of GPI-anchored proteins. These results also suggest that the weak membrane association could be the reason for the transport defect in lcb1-100 cells.
To test whether weak membrane association of Gas1p could cause the transport delay of Gas1p in wild-type cells, we used a mutant Gas1p containing a point mutation (L526R) in the hydrophobic stretch of the carboxy-terminal sequence that is required for GPI-anchor addition. This mutant Gas1p (L526R) is no longer GPI-anchored and is released into the medium as a soluble mature form (16). A large fraction of the mutant protein behaves as a soluble lumenal protein, and the remaining membrane-associated protein can be extracted completely from the membrane by carbonate treatment (Fig. 4A). The behavior of this mutant Gas1p is similar in wild-type and lcb1-100 cells. In a pulse-chase experiment (Fig. 4B), we observed the ER form of mutant Gas1p (L526R) in wild-type cells even after 60 min of chase, although a significant proportion of the protein was lost (lower left). Wild-type Gas1p was matured almost fully in wild-type cells by this time point (upper left). This result demonstrates that a soluble form of Gas1p is matured inefficiently even in wild-type cells. The same result was observed in lcb1-100 cells (lower right). These results suggest that proper membrane association may be necessary for efficient maturation of Gas1p.
Because the mutant Gas1p (L526R) is secreted into the medium as a mature form (16), it is possible that the released Gas1p in lcb1-100 cells may also be secreted into the medium as a mature form and escaped our detection. Therefore, we assayed for the secretion Gas1p into the medium after a long period of chase. Wild-type Gas1p in lcb1-100 cells was neither matured nor released into medium after 1 and 2 h (Fig. 4C,  upper right). In contrast, a significant amount of Gas1p (L526R) was released as a mature form into the medium, consistent with the decrease of intracellular Gas1p (lower panels). The secreted product from both strains showed a normal apparent molecular weight, demonstrating that there is no general defect in Golgi modifications of Gas1p in lcb1-100 cells. Therefore, most of the Gas1p in lcb1-100 cells seems unable to reach the Golgi compartment and instead was degraded gradually after a long chase time.
The GPI-anchored Protein Transport Defect Is Partially Reversible in lcb1 Mutant Cells-As mentioned above, exogenous D-erythro-DHS restored GPI-anchored protein transport fully in lcb1-100 mutant cells if added from the beginning of preincubation. More than 95% of Gas1p was matured after 90 min (Fig. 5A, lane 4). In contrast, if we added DHS after a 15-min chase, we observed a clear but only partial restoration of transport (lane 6, 57%). The simplest explanation for the partial restoration is that once Gas1p molecules are released from the FIG. 4. A soluble Gas1p mutant shows slow maturation in wildtype cells. A, membrane association of mutant Gas1 protein (L526R) was studied in wild-type (RH1638) and lcb1-100 (RH5241) cells as in Fig 3. B, maturation of the wild-type (wt) and mutant Gas1 proteins was examined by pulse-chase experiments. C, after the different chase periods, cells were separated, and the protein in the media was precipitated by addition of trichloroacetic acid to 10%. Gas1p in both fractions were analyzed by immunoprecipitation. For the estimation of the amount of Gas1p inside of the cells and the medium, the amount of Gas1p recovered without chase was set to 100. In this experiment to suppress synthesis of endogenous sphingoid base completely, serine palmitoyltransferase inhibitor, myriocin (5 g/ml), was added from the beginning of the preincubation. ER membrane, they can no longer be associated efficiently with membrane nor transported to the Golgi for maturation. To test this, we studied the membrane association of Gas1p before and after adding DHS (same conditions as for lanes 5 and 6 in Fig.  5A). Unexpectedly, the amount of released ER form of Gas1p decreased by chasing for another 75 min after adding DHS 2 (Fig. 5B, right column). This result shows that membrane association, as well as the Gas1p maturation defect, is reversible and can be restored. We also checked for membrane association of Gas1p after a 90-min chase in the absence of DHS (left column). Only a small fraction of Gas1p was matured after 90 min. As a result we found that all of the Gas1p that was transported to the Golgi compartment in lcb1-100 mutant was membrane-associated, whereas the ER form was always in the equilibrium between membrane-associated and -soluble forms.

DISCUSSION
The major finding of this study is that ceramide and/or inositol sphingolipid synthesis is required for ER to Golgi GPIanchored protein transport in yeast. We also observed that in lcb1-100 mutant cells, GPI-anchored proteins do not show stable membrane association but rather behave like peripheral membrane proteins. The stable membrane association is dependent on ceramide and/or inositol sphingolipid synthesis. The lack of tight membrane association in the absence of sphingolipid synthesis could be one of the reasons why GPI-anchored proteins are not transported efficiently to the Golgi apparatus.
To identify which sphingolipids are required for ER to Golgi transport of GPI-anchored proteins in yeast, we first determined the substrate specificity for ceramide and sphingolipid synthesis. Our study showed that two stereoisomers of DHS, D-erythro and L-threo, but not L-erythro, are incorporated into ceramide in yeast. The stereospecificity for ceramide synthesis in yeast is similar to that found recently for mammalian cells (19). Both stereoisomers of ceramide that could be made were used for inositol sphingolipid synthesis. We found that only the DHS stereoisomers that could be incorporated into ceramide and inositol sphingolipids restored GPI-anchored protein transport in lcb1-100 cells at 37°C. The endocytic defect in lcb1-100 mutant cells at 37°C was restored significantly by all four stereoisomers of DHS, 3 suggesting that all stereoisomers can be taken up and are active. In addition, both ceramide and IPC synthesis inhibitors reduced the restoration of GPI-anchored protein transport by D-erythro-DHS in lcb1-100 cells significantly. Finally, GPI-anchored protein transport was not restored in mutant cells that synthesize almost no ceramide even in the presence of DHS. Taken together, these results demonstrate that ceramide and inositol sphingolipids are essential for GPI-anchored protein transport.
This conclusion is consistent with results of studies showing that GPI-anchored protein transport is affected in mutant cells defective at several steps of sphingolipid metabolism. It was shown previously that a lag1 lac1 double knock-out mutant (6 -8), and fen1 and sur4 mutant cells (9,10) are also defective for GPI-anchored protein transport to the Golgi compartment. These studies suggested that ceramide and sphingolipids might be important for GPI-anchored protein transport. However, all these mutants accumulate sphingolipid intermediates such as PHS and synthesize unknown lipids (6 -8), which could confer negative effects on GPI-anchored protein transport. In addition, the effect of these mutations on GPI-anchored protein transport was much weaker than that found for the lcb1-100 cells (6,10). Because exogenous DHS could restore the conditional and severe defect in lcb1-100 cells completely, it was an ideal system to identify required sphingolipids. DHS itself also seems to be important for membrane trafficking steps, like the internalization step of endocytosis, where it is important in regulating protein kinases (17,28,29). Our results show clearly that DHS alone is not sufficient for GPI-anchored protein transport.
One of the significant consequences of the lack of ceramide and sphingolipid synthesis is the incomplete membrane association of the ER form of GPI-anchored proteins. This is not because of a lack of GPI anchor attachment. It was reported recently that ceramide synthase, mediated by Lag1p and Lac1p, is localized in ER membranes (6 -8) and that IPC synthase, mediated by Aur1p, is localized to the Golgi structure (30). This makes it likely that ceramide, rather than inositol sphingolipids, is responsible for the ER membrane association of GPI-anchored proteins, although we cannot rule out that inositol sphingolipids can be transported from the Golgi apparatus back to the ER.
Based on the observation that GPI-anchored proteins behave like peripheral membrane proteins in lcb1-100 cells, we studied the maturation efficiency of a soluble Gas1p mutant (L526R) in wild-type cells to test whether membrane association is important for the efficient transport. The soluble form of Gas1p showed delayed maturation in wild-type cells clearly, suggesting that membrane association is an important factor for transport. However, it could also be that the transport delay is because of the lack of a GPI anchor on this mutant Gas1p, because the anchor may possibly function as an exit signal from the ER (31). We also found that a small amount of the mutant Gas1p was matured and secreted into medium, in contrast to the loosely associated GPI-anchored Gas1p in lcb1-100 mutant cells. These differences suggest that there are additional reasons for the maturation delay of GPI-anchored proteins in the absence of tight membrane association in lcb1-100 cells. One possibility is that the hydrophobic part of the GPI anchor of the released proteins may be recognized by the quality control apparatus and serve to retain these proteins in the ER. Alternatively, the hydrophobic anchor could cause formation of micelle-like structures that would be excluded from ER-derived vesicles because of their size.
We found that whenever Gas1p was found in the mature form in lcb1-100 cells, it was associated tightly with the membrane. In contrast, the immature form of Gas1p in lcb1-100 mutant always showed a certain equilibrium between membrane-associated and -soluble forms. These observations also suggest that the transport is coupled to membrane association. There are two possible models to explain the different behavior of ER and mature Gas1p in lcb1-100 cells. One model is that only the small amount of ER form of Gas1p in lcb1-100 cells that is tightly membrane-associated (for example, by remodeling of GPI-lipid moiety) would be delivered to the Golgi structure and matured. Alternatively, all ER forms of Gas1p could be transported and matured very inefficiently. After arrival in the Golgi compartment, all Gas1p would become tightly membrane-associated because of the different membrane composition. Perhaps sphingolipids are not depleted from the Golgi compartment as severely as from the ER in lcb1-100 mutant cells. So far, we cannot differentiate these possibilities experimentally, but in both cases the weak membrane association could be one of the reasons for the delayed transport.
Ceramides and sphingolipids may have an additional role in GPI-anchored protein transport. GPI-anchored proteins are transported by a different vesicle population than are other 2 The maturation efficiency in this experiment was higher than 57%. Especially under these conditions, maturation efficiency varied among experiments; however, we obtained reproducible correlations between membrane association and maturation. 3 B. Zanolari, unpublished result. secretory proteins such as pro-␣-factor and the general amino acid permease, Gap1p (11). Ceramides and/or inositol sphingolipids may be required specifically for this GPI-anchored protein-specific pathway. Alternatively, in the absence of sphingolipids, sorting of GPI-anchored proteins from other secretory proteins in the ER could be impaired. This could result in inefficient transport in lcb1-100 cells. Previously, it was reported that lcb1-100 mutant membranes did not show any defect at the step of ER exit of Gas1p in an in vitro budding assay (3). This in vitro budding assay was carried out at 20°C using wild-type cytosol. Sphingolipid synthesis is temperature-sensitive in lcb1-100 mutant cells (17), and GPIanchored protein transport is less defective at lower temperature (3). This could explain the lack of defect seen in the previous study. As shown here, GPI-anchored proteins in lcb1-100 mutant cells at 37°C behave like peripheral proteins, and a large fraction is released from membranes easily during the assay. Therefore, we could not use the in vitro budding assay to examine ER exit and sorting efficiency in lcb1-100 mutant cells under non-permissive conditions.
There are several possible, not mutually exclusive, ways that ceramide could be required for stable association of GPI-anchored proteins with the ER membrane. Ceramide may be required as a membrane component for the stable association of the GPI-anchor with the membrane. In yeast, both ceramide and GPI-lipid moieties contain C26 fatty acids (32,33). The hydrophobic interaction between ceramide and the GPI-lipid moiety may be critical for stable membrane association. Another possibility is that ceramide may participate in the formation of lipid raft structures in the ER. Ceramide has been shown to promote membrane microdomain formation in vitro (34). Our results could be interpreted in favor of the hypothesis that microdomain structure is important for GPI-anchored proteins to associate stably with membranes. However, transport of Gas1p from ER to Golgi was not affected significantly in ergosterol biosynthesis mutant cells (erg3/6). These mutants showed a complete loss of mature Gas1p insolubility in Triton X-100 at 0°C (35) and therefore, supposedly had greatly perturbed lipid raft formation.
Another obvious possibility is that ceramide may function as a substrate for remodeling of the GPI-lipid moiety. In yeast, the GPI-lipid moieties are remodeled from more hydrophilic PI to ceramide or diacylglycerol, both carrying C26 fatty acid, after transfer to protein (36). This remodeling occurs mainly in the ER and partially in the Golgi structure (33). The majority of GPI-anchored proteins in yeast have ceramide moieties. For these proteins, it is natural to imagine a function of ceramides as remodeling substrates. However, mature Gas1p has a glycerol backbone with a C26 fatty acid (37). The introduction of a long chain fatty acid in the sn-2 position of diacylglycerol moiety of GPI-anchored proteins occurs in the ER (33). Because a C26 fatty acid moiety is found mainly in phytoceramide and sphingolipids in normal cells (32), it is possible that ceramide also functions as a donor of C26 fatty acid for the PI-lipid moiety of GPI-anchored proteins with a glycerol backbone. If so, the lack of ceramide would cause a general defect in GPIanchor remodeling. The significance of remodeling is not yet known, but both remodeling reactions introduce long chain fatty acids onto GPI-anchored proteins (33). This may be nec-essary for their stable association with the ER membrane. Analysis of the lipid moieties of Gas1p and other GPI-anchored proteins in lcb1-100 cells will help to address this possibility.