Biosynthesis of Inositol Phosphoceramides and Remodeling of Glycosylphosphatidylinositol Anchors in Saccharomyces cerevisiae Are Mediated by Different Enzymes*

Metabolic labeling of cells with [3H]dihydrosphingosine ([3H]DHS) allows us to follow the incorporation of this tracer into ceramides (Cer), inositol phosphoceramides (IPCs), and mannosylated IPCs and at the same time to assess the remodeling of glycosylphosphatidylinositol proteins during which preexisting anchor lipid moieties are replaced by [3H]Cer-containing anchors. The results indicate that the remodelases in the endoplasmic reticulum and Golgi use as their substrate Cers that are not generated by the breakdown of IPCs but are newly synthesized. Aureobasidin A, an inhibitor of the IPC synthase Aur1p completely blocks IPC biosynthesis at 0.5 μg/ml but does not block remodeling of glycosylphosphatidylinositol anchors even at concentrations up to 10 μg/ml. In addition, a synthetic Cer analogue,N-hexanoyl-[3H]DHS, is used as a substrate by Aur1p but not by the remodelases. Thus, remodeling is not mediated by Aur1p although remodeling presumably proceeds by an analogous reaction. Studies with secretion mutants deficient in COPII or COPI coat proteins show that all COPII mutants are unable to introduce [3H]Cer by the Golgi remodelase at the restrictive temperature. This suggests that Cer has to be transported by a COPII-dependent way from the endoplasmic reticulum to Golgi for Golgi remodeling to occur. Golgi remodeling is also not operating in the erd2 mutant and is significantly reduced in COPI mutants, suggesting a dependence of Golgi remodeling on retrotransport.

The lipid moieties of mature GPI 1 anchors usually do not contain the DAG species found on the cells PI, although PI would appear as the natural starting point for the biosynthesis of the GPI (1,2). Studies in trypanosomes and in yeast strongly suggest that ordinary PIs indeed serve as the primary platform on which, by stepwise addition of sugars, GPIs are built but that the primary lipid moiety is remodeled later on, either shortly before transfer of the GPI to proteins (Trypanosoma brucei) or after this transfer (Saccharomyces cerevisiae and Trypanosoma cruzi) (3)(4)(5)(6)(7)(8). Curiously, it appears that after remodeling, lipids continue to be exchanged even on mature GPI proteins, although this further lipid exchange replaces lipids by lipids of the same structure (9,10).
The situation in S. cerevisiae is peculiar since two different types of lipids can be found on GPI anchors as follows: Cer and DAG, both containing C 26:0 or hydroxylated C 26:0 (11). Thus, all mature GPI proteins of yeast contain large lipid moieties with C 26 fatty acids.
The remodeling introducing Cer can be monitored by metabolic labeling experiments using tritiated DHS ([ 3 H]DHS) (9). When given to cells, this tracer is rapidly taken up and is incorporated into all sphingolipids as well as Cer-containing GPI proteins (Fig. 1). By labeling different secretion mutants with [ 3 H]DHS or by using preincubations with cycloheximide (Chx), we found that GPI remodelase 2 activity is present in both the ER and the Golgi (9). Curiously, in temperaturesensitive secretion mutants such as sec12 or sec18, the incorporation of [ 3 H]DHS into GPI proteins in the Golgi is completely blocked at 37°C, whereas [ 3 H]DHS continues to be incorporated into immature GPI proteins in the ER. This suggested that the substrate for the Golgi remodelase may not reach the Golgi, if the vesicular flow from the ER to the Golgi is interrupted.
To allow for a more definitive interpretation of this result, we set out to identify the lipid substrate of the remodeling reaction. It seemed likely that lipid remodeling proceeds by some transesterification or transglycosylation since activated forms of Cer such as CDP-Cer have not been found in yeast although they exist in certain bacteria (12). A priori we had to consider that remodeling may occur by introduction of IPC, Cer-P, or Cer in exchange for PI, phosphatidic acid, or DAG, respectively. Introduction of IPC could be excluded since metabolic labeling of cells with [ 3 H]Ins in the presence of the protein synthesis inhibitor Chx fails to label GPI proteins, whereas [ 3 H]DHS is strongly incorporated under this condition (9). We could, however, not exclude that remodelases use IPC as a substrate from which to take Cer-P or Cer for transesterification. This hypothesis even seemed attractive since, when the vesicular traffic from ER to Golgi is blocked in temperature-sensitive COPII mutants, the main IPC (IPC/C) continues to be made in the ER, but mannosylation of IPC/C to MIPC in the Golgi is abolished indicating that IPC is neither transported to the Golgi nor synthesized in the Golgi under these conditions (9,13,14). Thus, this lack of transport of IPC to the Golgi could explain the lack of GPI remodeling in the Golgi in COPII mutants.
Here we present evidence in favor of Cer rather than Cer-P as a remodelase substrate, and we establish that this Cer substrate is newly synthesized and is not taken from IPC or other more complex sphingolipids. By using aureobasidin A (AbA), a recently identified inhibitor of IPC biosynthesis (15), we can demonstrate that Golgi remodeling of GPI proteins is occurring normally even if IPC biosynthesis is blocked. Our data lend support to the hypothesis that it is newly synthesized Cer that is used for remodeling. We further present evidence against the possibility that the Aur1p and Ipt1p contain remodelase activity. Moreover, using secretion and retrieval mutants, we evaluate the necessity of COPI-and COPII-dependent vesicle flow for the transfer of Cer onto GPI proteins in the Golgi.

EXPERIMENTAL PROCEDURES
Yeast Strains, Media, and Materials-S. cerevisiae strains are indicated in Table I. Maintenance and growth conditions as well as the photometric determination of cell density have been described (16). "1 A 600 " of cells corresponds to 1-2 ϫ 10 7 cells depending on the strain. Reagents were obtained from the following sources: [ (18). Either 15 l of acetic anhydride or 15 l of caproic anhydride were added, and the mixture was left standing at room temperature for 2 h. The sample was dried and flashed with 100 l of dry methanol. This N-acylation procedure was repeated twice. The product was purified by preparative TLC using 0.2-mm thick silica gel plates (Merck, Darmstadt, Germany) and developed in solvent system 1. Labeled ceramides were eluted from the silica gel with methanol; methanol was evaporated and the remaining silica gel particles were eliminated by butanol/water partitioning (19).
Labeling of Cells H]DHS-C 6 exactly as described (9). Chx was added at 100 g/ml. Labelings were stopped and proteins were extracted and delipidated as described (9). For SDS-PAGE, proteins were resuspended and boiled during 5 min in reducing sample buffer (20), and 1/4 of the sample was loaded on a 7.5% SDS-PAGE gel. The gels were then processed for fluorography.
Lipid Extraction and Thin Layer Chromatography-When preparing cell extracts for protein analysis, the organic solvent extracts were saved, pooled, and dried. Where indicated, lipids were subjected to mild base treatment using NaOH (21). Lipids were desalted by partitioning between n-butyl alcohol and water followed by back extraction of the butanol phase with water (19). The desalted lipids were analyzed by ascending TLC using 0.2-mm thick silica gel plates with solvent system 1 (chloroform, methanol, 2 M NH 4 OH (40:10:1)), solvent system 2 (chloroform, methanol, 0.25% KCl (55:45:10)), or solvent system 3 (chloroform, methanol, 2 M acetic acid (18:10:2)). Radioactivity was quantified by radioscanning using a Berthold radioscanner (22). Results were calculated from the scans by summing up the radioactivity of all lipid peaks including [ 3 H]DHS and by dividing the radioactivity in any given lipid by this sum.
the TLC with methanol as described (17). The extract was pooled and dried. Lipids were desalted by partitioning between n-butyl alcohol and water followed by a back extraction of the butanol phase with water (19). Aliquots were then subjected to strong acid hydrolysis (23) or treated with PI-PLC for 16 h (24). After treatments, lipids were desalted with partitioning between n-butyl alcohol and water.
Purification of GPI Peptides and Preparation of Anchor-derived Cer and P-Cer Standards-GPI peptides were purified from 15 A 600 of cells and treated with PI-PLC (0.05 units) as described (9,17). GPI-PLD treatment was performed by resuspending the GPI peptides in 100 l of PI-PLD buffer II (50 mM Tris-HCl, pH 4.5, 10 mM NaCl, 2.6 mM CaCl 2 , 20% propanol) and adding 0.5 units of GPI-PLD (17). The samples were incubated overnight at 37°C. P-Cer obtained by GPI-PLD was dephosphorylated to Cer by alkaline phosphatase treatment (17).

Cells Do Not Contain Free Ceramide Phosphate-Previous
studies have revealed that the Cer moieties introduced into GPI anchors by remodeling are not representative of the cellfree Cers. In particular, if secretion is blocked in sec18, the prevalent PHS-C 26 OH is used to make IPC/C, whereas cells incorporate into GPI anchors a more hydrophobic, PHS-containing lipid, probably PHS-C 26 (9). Since both of these reactions occur in the ER, this discrepancy made us suspicious that the remodelases may draw on Cers present in form of Cer-P or else on IPCs.
To evaluate the possibility of Cer-P being the substrate for remodeling reactions, we set out to see if cells contain a pool of free Cer-P. Standards were prepared from [ 3 H]DHS-labeled GPI proteins using phospholipases. PI-PLC treatment of anchor peptides generated Cers ␣, ␤, and ␥ corresponding to DHS-C 26 , PHS-C 26 , and PHS-C 26 OH (Fig. 2, lane 2), the ␥Cer only being found on GPI proteins having reached the Golgi (6,9). GPI-PLD treatment of anchor peptides generated three more polar lipids representing phosphorylated forms of Cers ␣, ␤, and ␥ (Fig. 2, lane 7). They were resistant to mild alkaline hydrolysis and could be converted to Cers ␣, ␤, and ␥ by phosphatase treatment (not shown). Lipid extracts from wild type or heat-shocked sec18 cells did contain lipids migrating close to these Cer-P standards, but they were not resistant to alkaline hydrolysis and hence are not Cer-Ps (Fig. 2, lanes 3-6). In particular, after mild base hydrolysis, the region corresponding to ␤Cer-P, which is derived from the most abundant anchor species, was completely empty. Any isolated band containing Ͼ0.03% of the radioactivity spotted in each lane (5 Ci/1 cm) is detected by the radioscanner. In comparison, the free Cer (PHS-C 26 OH) comigrating with ␥Cer accounts for 7.8% (wt) and 16.8% (sec18) of the total lipid extract. Thus, Cer-P, if present at all, is at least 260-or 560-fold less abundant than the Cer PHS-C 26 OH. This makes it rather unlikely that free Cer-Ps would serve as substrates for GPI remodeling reactions. We also wanted to rule out the possibility that in a first step free [ 3 H]DHS or [ 3 H]PHS is introduced into GPI anchors, in exchange for diacylglycerol, and that the long chain base is N-acylated only subsequently. For this we used australifungin, a recently described antifungal drug that blocks the sphinganine N-acyltransferase thus blocking the conversion of DHS to  7) were used as standards. Lipids were resolved using solvent system 3. MIPC comigrates with a contaminating lipid X in this solvent system (see Fig. 4B).

H]DHS into Cer and IPCs as expected. At the same time it also completely abrogated incorporation of [ 3 H]DHS into GPI proteins (not shown). This argues in favor of the idea that [ 3 H]DHS has first to be converted to [ 3 H]Cer before being incorporated into GPI anchors. Aureobasidin A Dissociates GPI Anchor Remodeling from IPC Biosynthesis-Aureobasidin
A is a potent inhibitor of the IPC synthase Aur1p, an essential enzyme transferring Ins-P from PI to Cer, thus forming IPC (15). The main toxicity of AbA is due to its effect on Aur1p since point mutations in AUR1 render cells resistant to AbA (15,26). The subcellular localization of Aur1p has not been reported, but it is believed to reside in the ER since in temperature-sensitive COPII mutants the main IPC (IPC/C) continues to be made when the vesicular traffic from ER to Golgi is blocked, whereas mannosylation of IPC/C to MIPC in the Golgi is blocked indicating that IPC is neither transported to the Golgi nor synthesized in the Golgi under these conditions (9,13,14). To monitor the effect of AbA on the remodeling of GPI anchors, we labeled cells with [ 3 (Fig. 3) at a concentration at which incorporation into sphingolipids was completely blocked (not shown). We found that for a complete inhibition of IPC biosynthesis, it was necessary to have 0.2 g/ml (ϭ 0.17 M) AbA in the labeling medium, the same concentration as is required to achieve complete growth arrest. These experiments indicate that GPI remodeling can occur in the absence of IPC biosynthesis and argue against the possibility that [ 3 H]DHS is first incorporated into IPC in order to become integrated into a GPI anchor.

H]Ins and [ 3 H]DHS which both can be used to selectively label GPI proteins. AbA did not affect the incorporation of [ 3 H]Ins and [ 3 H]DHS into proteins
We wanted to know if both, ER and Golgi remodelase activities are independent of AUR1, and we therefore tested AbA in the previously established experimental settings in which only ER or only Golgi activity is assayed (9).
ER activity can be measured in the secretion mutant sec18, which at 37°C effectively blocks incorporation of [ 3 H]DHS into mature-sized GPI proteins in the Golgi. As can be seen in Fig.  4A, at a concentration of AbA which completely blocks the biosynthesis of IPCs and of M(IP) 2 C, the incorporation of [ 3 H]DHS into immature GPI proteins is somewhat reduced but so is the incorporation of [ 3 H]Ins, the latter being a reflection of the rates of GPI protein synthesis and turnover but not remodeling. The result suggests that AbA has no effect on the efficiency of remodeling of GPI proteins in the ER. The conditions used in Fig. 4A do not exclude the possibility that some of the remodeling activity is due to newly made Golgi remodelase in transit to the Golgi, which was trapped in the ER by the secretion block. To eliminate this possibility we used another preincubation protocol allowing us to specifically measure the autochthonous ER remodelase activity as detailed in Fig. 4B. Proteins were extracted, delipidated, and separated by SDS-PAGE. B, sec18 cells were preincubated at 24°C. myr, Chx, and AbA (1 g/ml final) were added at 100, 40, and 20 min before the addition of [ 3 H]DHS as indicated. Cells were shifted to 37°C 10 or 5 min before the addition of the label except for lane 1 where cells were labeled at 24°C. Cells were broken, and lipid extracts were resolved by TLC with solvent system 3, and proteins were separated by SDS-PAGE. The relative amounts of cpm found in IPC/B, IPC/C, IPC/D, and M(IP) 2 C obtained by scanning of TLCs are given in percent of total below each lane. MIPC is not indicated since it comigrates with an unrelated lipid that is resistant to strong alkaline and strong acid hydrolysis and hence is not a sphingolipid (data not shown). Two-dimensional TLC, however, showed that AbA also completely blocks the biosynthesis of MIPC (not shown). The rationale for the preincubations in B is as follows: inhibition of sphingolipid biosynthesis by myr results in the accumulation of GPI proteins in a non-remodeled state in the ER (67). myr additionally increases the efficiency of [ 3 H]DHS incorporation into lipids and proteins. Chx is added later during preincubation to arrest the de novo biosynthesis of Golgi remodelase and to allow the export of all remaining Golgi remodelase out of the ER. The use of sec18 cells at 37°C prevents labeling of mature proteins in the Golgi. and M(IP) 2 C is blocked even in the absence of AbA as a result of the secretion block, as mentioned above (9).
To evaluate the effect of AbA on the Golgi remodelase, cells were pretreated with Chx, allowing GPI proteins to be exported out of the ER before the label was added (9). As shown in Fig.  5, with this protocol, AbA diminishes the labeling of maturesized proteins with [ 3 H]DHS somewhat, whereas the biosynthesis of sphingolipids is completely blocked. Thus, both ER and Golgi remodelases are able to incorporate [ 3 H]DHS even if IPC biosynthesis is completely blocked.
Cers of GPI proteins seem to get exchanged for Cers (9), but this conservative lipid exchange does not seem to occur on sphingolipids since no labeling of IPC, MIPC, or M(IP) 2 C is observed in wt cells in the presence of AbA. This, however, does not rule out the possibility of such an exchange being mediated by Aur1p.
Cer has been found to be toxic to cells, and its toxicity was considered to be the reason for the lack of success in early attempts to generate mutants in IPC biosynthesis (15,27,28). Our results document, however, that AbA does not compromise the synthesis of GPI proteins nor their remodeling for the duration of the experiment (3 h), although it can be expected that the endogenously produced Cer reaches very high levels under these conditions (15). It should be noted that the accumulation of [ 3 H]PHS-C 26 OH does not reflect the increase in cold Cer. In the presence of AbA, [ 3 H]PHS-C 26 OH reaches only about 3-fold higher levels than in the absence of the inhibitor (Fig. 3). This elevation is moderate since even in the absence of AbA no more than 20 -25% of [ 3 H]DHS added to cells gets incorporated into sphingolipids, whereas the rest is degraded and incorporated into other types of phospholipids.
Aur1p and Ipt1p Have No Remodelase Activity-Collectively the results described so far support the idea that newly synthesized Cer is the substrate for the remodeling of GPI anchors in yeast and that the remodelases exchange DAG for Cer. This type of transesterification reaction is analogous to the biosynthesis of IPC by Aur1p, an enzyme replacing the DAG moiety of PI by a Cer. The following experiments were performed to see if Aur1p is the remodelase of GPI anchors. (i) As shown in Fig.  6, AbA does not interfere with GPI remodeling even at elevated concentrations (10 g/ml). (ii) When cells were incubated with chemically acylated [ 3 H]DHS (N-acetylated or N-hexanoy-lated), we did not find any incorporation of these short chain Cers into GPI proteins although they were efficiently incorporated into IPC and M(IP) 2 C (Fig. 7, A and B; not shown). Even when purifying large amounts of glycoproteins over concanavalin A-Sepharose and preparing anchor peptides in the same way as done in Fig. 9C, we could not find any traces of radioactivity in anchor peptides from DHS-C 2 -or DHS-C 6 -labeled cells. (iii) As mentioned, if secretion is blocked in sec18, Aur1p uses the prevalent PHS-C 26 OH to make IPC/C, whereas cells incorporate into GPI anchors a more hydrophobic, PHS-containing lipid, probably PHS-C 26 (9). These data suggest that the remodelase has a different substrate specificity than Aur1p. We also tested if the AUR1 homologue IPT1 is involved in Golgi remodeling. IPT1 has been shown to be necessary for the transfer of inositol phosphate from PI onto MIPC in order to make M(IP) 2 C, a reaction that is supposed to occur in the Golgi (13,29). We find that the deletion of IPT1 has no influence on Golgi remodelase (Fig. 7D) and also has no effect on the incorporation of [ 3 H]DHS into proteins when all remodelases are measured simultaneously (not shown). Recently it was found that exogenous [ 3 H]DHS is first phosphorylated and subsequently dephosphorylated again before being incorporated into sphingolipids (30 -32). Moreover, part of the phosphorylated DHS is broken down to [ 3 H]palmitaldehyde and phosphoethanolamine by Bst1p (33)(34)(35). Palmitaldehyde in turn can be used to make palmitoyl-CoA, and this explains the presence of labeled glycerophospholipids in the lipid extract of [ 3 H]DHS-labeled cells. Part of [ 3 H]DHS may also be transformed through other pathways since many of the lipid species seen in Fig. 7A are neither sphingolipids nor glycerophospholipids (9). However, multiple analysis of the protein-associated label after [ 3 H]DHS labeling showed that, irrespective of the labeling conditions, the bulk of label is in Cers, none is in diacylglycerols or in fatty acids, and only little is in a nonidentified lipid (9). The data in Fig. 7A show that in contrast to [ 3 H]DHS, exogenously added [ 3 H]DHS-C 6 is not degraded and is used exclusively to make sphingolipids. Hydrolysis of products indicated that [ 3 H]DHS-C 6 was transformed by cells into a slightly more hydrophilic, [ 3 H]DHS-containing compound which we tentatively designated as DHS-C 6 OH (Fig. 7A, lanes  6 -9, Fig. 7C, lanes 3 and 4). On the other hand, the main IPC (IPC 6 ) made by cells contained DHS-C 6 (Fig. 7C, lanes 5-7). (9), the incorporation of [ 3 H]DHS into mature high molecular mass (Ͼ150 kDa) GPI proteins in the Golgi seen in wt (Fig. 5, lane 5) is abrogated when the secretory pathway is blocked in sec12 (36) or sec18 mutants (37-39) (Fig. 4A, lane 3; Fig. 4B, lanes 3   FIG. 5. Golgi remodeling is not inhibited by aureobasidin A.  sec14 cells (lanes 1-4) and wt cells (lanes 5-8) were preincubated at 24°C. Addition of Chx, AbA (1 g/ml final), myr, and a temperature upshift to 37°C took place at 40, 30, 20, and 20 min, respectively, before adding 25 Ci of [ 3 H]DHS. Cells were broken, and proteins were separated by SDS-PAGE, and desalted lipids were resolved with solvent system 3 for quantification of sphingolipids. GPI proteins destined to become cell wall proteins lose the IPC moiety of the GPI anchor prior to being attached to the glucans of the cell wall (84). The use of sec14 allows us to retain GPI proteins in the Golgi and was chosen in an attempt to prevent this loss of the label.  1, 3, 5, 7, and 9) or during 2 h (lanes 2, 4, 6, 8, and 10). Proteins were extracted and separated on a 7.5% SDS-PAGE gel. Fig. 8A, the same phenomenon can also be observed in other COPII mutants such as sec23 (40,41) and sec13 (42) as well as in sec16 (Sec16p is thought of as a scaffold for COPII proteins (43)). These mutants are also unable to incorporate [ 3 H]DHS into IPC/D and M(IP) 2 C at 37°C (Fig.  8B). These data are consistent with the idea that, for Golgi remodeling to occur, Cer has to be transported to the Golgi by COPII-coated vesicles (42, 44 -46).

and 5). As shown in
When labeling COPI mutants (47) with [ 3 H]DHS, we found that some of them, similar to COPII mutants, accumulated immature forms of GPI proteins. At the same time the high molecular mass region (Ͼ150 kDa) showed weak labeling as compared with wt (Fig. 9A). This phenomenon was not unexpected for sec21-1, a mutant that was obtained by selecting for the inability to secrete soluble cargo proteins and that exhibits a relatively tight block of ER to Golgi transport at 37°C (48,49). On the other hand, the phenomenon was rather surprising for all other COPI mutants used in this study, since all of them secrete invertase and HSP150 (Pir2p) at a normal rate (50 -54) and hence are known to maintain the COPII-mediated pathway at least partially operational for the duration of our experiment. For all COPI mutants except ret1-1, the extent of the maturation deficit of GPI proteins was proportional to the inability to export CPY out of the ER as reported in the literature. Thus, the maturation deficit was strong in sec21-1, sec21-3, sec21-4, was less pronounced in ret2 and sec33-1, and was absent in sec27-1 and ret3-1 (Fig. 9A). When cells were preincubated at 37°C for 60 instead of 10 min, the maturation deficit became more pronounced in sec33-1, became also apparent in sec27-1, but could still not be observed in ret3-1 (not shown). The exception to this general rule is ret1-1 (Fig. 9A) which showed a strong maturation deficit of GPI proteins although it has been reported to transport CPY at a quite normal rate even after a preincubation of 90 min at 37°C (53). sec20, a mutant in a protein belonging to a t-SNARE complex involved in Golgi to ER retrograde transport (55), also showed a significant maturation deficit with a pronounced build up of immature GPI proteins. It is noteworthy that all these mutants affect essential genes and that all alleles used here are unable to grow at 37°C. In particular sec21-3 and sec21-4 have been shown to arrest the transport of CPY and ␣-factor precursor within 1 min after a temperature shift to 37°C but to continue to secrete HSP150 (Pir2p) and invertase at a normal rate (50).  6 OH, and IPC 6 , these lipids were scraped and purified from silica gel. All three lipids were subjected to strong acid hydrolysis (lanes 2, 4, and 7) or left untreated (lanes 1, 3, and 5), and IPC 6 was also treated with PI-PLC (lane 6). [ 3 H]DHS was used as standard, and lipids were resolved with solvent system 1. D, wt (lanes 1 and 2) and ⌬IPT1 cells (lanes 3 and 4) were preincubated at 24°C. Addition of Chx, AbA (1 g/ml final), myr, and a temperature upshift to 37°C took place at 40, 30, 20, and 20 min, respectively, before adding [ 3 H]DHS. Cells were broken, and proteins were separated by SDS-PAGE. Whereas the ER remodelase introduces ␣Cer and ␤Cer into GPI proteins, ␥Cer is only found on proteins that have reached the Golgi (6,9). To quantify the maturation deficit of sec21-3, we therefore quantified the relative amount of ␥Cer (PHS-C 26 OH) present in GPI anchors after 2 h of labeling. As can be seen in Fig. 9C, sec21-3 contained 6.5-fold less ␥Cer than wt cells.
In many COPI mutants we observed a very significant reduction in the relative amount of [ 3 H]DHS incorporation into IPC/D and M(IP) 2 C at the restrictive temperature (Fig. 9B), whereas in wt we observed an increase in IPC/D and no reduction of M(IP) 2 C (Fig. 8B).
Golgi Remodeling Is Reduced in sec21-3 and erd2-We wondered if sec21-3 also has a defect in Golgi remodeling as was observed with COPII mutants. We therefore investigated the amount of [ 3 H]DHS incorporation into mature GPI proteins after emptying the ER of GPI proteins by a prolonged preincubation in the presence of Chx, i.e. in the experimental settings in which only Golgi remodelase is assayed. As shown in Fig.  10A, lane 4, and Fig. 10D, lanes 4 -6, incorporation of [ 3 H]DHS is strongly reduced in COPII mutants but also is significantly diminished in sec21-3, a COPI mutant (Fig. 10A, lane 5). Golgi remodelase activity in sec21-3 is particularly weak if compared with the excessively high incorporation of [ 3 H]DHS into proteins in the absence of Chx (Fig. 10B). In wt cells the incorporation of [ 3 H]DHS into proteins in the absence of Chx and myr is of comparable magnitude as the incorporation in the presence of these two drugs (see Fig. 7D in Ref. 9). sec21-3 thus has a significant deficit in Golgi remodelase activity. COPI has been implicated in antero-as well as retrograde transport. We therefore asked whether mutants that selectively affect retrieval are also deficient in Golgi remodeling. We indeed found that erd2, a mutant defective in the retrieval of proteins with a C-terminal KDEL sequence (56,57), had strongly diminished Golgi remodeling activity (Fig. 10D, lane 3) although it incorporated [ 3 H]DHS normally in the absence of Chx (Fig. 10C,  lane 5). rer1, a mutant defective in the retrieval of several integral membrane proteins such as Sec12p, Sec63p, and Sec71p (58 -60) had normal Golgi remodeling activity (Fig.  10D, lane 2). An shr3 mutant defective in the packaging of Gap1p into COPII vesicles (61) also had normal Golgi remodeling activity (Fig. 10D, lane 1). DISCUSSION Remodeling of GPI proteins in S. cerevisiae has attracted our interest because it holds promise to reveal something about the importance of their lipid moiety for their forward movement and sorting in the secretory pathway (62,63). Since our attempts to reconstitute a remodelase activity in a microsomal system have been unsuccessful so far, we tried to characterize further the remodelase reactions by doing metabolic labeling experiments with [ 3 H]DHS in intact cells. The use of AbA allowed us to exclude IPCs as donors of the Cer moieties found in GPI anchors, since [ 3 H]DHS continued to be incorporated into GPI proteins under conditions in which its incorporation into all classes of IPCs was completely blocked. We further show that there are no Cer-Ps detectable with the methods used here, either in wt or in sec18 cells in which secretion is blocked. Also, based on the absence of remodeling in the presence of australifungin, there is no evidence for the attachment of free DHS or PHS to GPI proteins. These data suggest that free Cer is the substrate of the remodeling reaction. Although in [ 3 H]DHS-labeled wt cells we could not detect free [ 3 H]PHS-C 26 , i.e. the presumed Cer substrate for the principal remodeling reaction (9), PHS-C 26 has indeed been detected in scs7, a mutant that is deficient in the hydroxylation of very long chain fatty acids (64). In scs7, PHS-C 26 is converted by Aur1p into IPC/B, an IPC that is not very prominent in wt cells but becomes very prominent and is the only IPC in scs7 cells (64). If Cer is the substrate for remodelases that exchange the DAG of the primary GPI anchor for Cer, then the remodeling reaction becomes analogous to the IPC biosynthesis catalyzed by Aur1p. Since AbA does not block remodeling, it is unlikely that GPI remodeling is operated by Aur1p. Its homologue, Ipt1p, is not involved either, and we presently are testing other mutants for remodelase activity.
It appears that all GPI proteins of S. cerevisiae increase the size of the lipid moiety of their primary GPI anchor by way of lipid exchange (6,9,65,66). Lipid exchange introduces a very long chain fatty acid (C 26 or C 26 OH) either as part of a DAG or of a Cer. This phenomenon has been related to the fact that the export of GPI proteins out of the ER requires sphingolipid, i.e. that the inhibition of sphingolipid biosynthesis either by myr or in thermosensitive mutants (lcb1-100) rapidly slows the export of GPI proteins to the Golgi, whereas other membrane proteins or soluble cargo proteins continue to be transported at a normal rate (67)(68)(69). To explain the specific sphingolipid requirement for GPI protein transport, two different models have been proposed: the "remodeling hypothesis" and the "clustering hypothesis" (67). The remodeling hypothesis assumes that the incorporation of Cer into the anchors is a prerequisite for rapid transport. Indeed, the incorporation of Cer occurs in a majority of GPI proteins, but it does not represent an absolute requirement, since Gas1p contains C 26 -DAG (70) but no Cer and is efficiently transported. Also, the myriocin-resistant mutant myr1 can transport nonremodeled GPI proteins to the Golgi when sphingolipid biosynthesis is blocked (67). Although the remodeling hypothesis seems to be contradicted by these last findings, it remains a distinct possibility that transport to the Golgi requires the GPI anchor to contain a C 26 -containing lipid moiety of some sort, be it in the form of a Cer or a DAG. In fact we are presently unable to tell if under normal conditions Cer gets incorporated into GPI proteins of wt cells mainly in the ER or in the Golgi. The clustering hypothesis assumes some necessary interactions of GPI proteins with membrane microdomains having a high sphingolipid content. Recent in vitro experiments show that GPI proteins, as all other cargo molecules, bud off the ER in COPII-coated vesicles and that they continue to do so even when sphingolipid biosynthesis is blocked (68,71). This in vitro phenomenon led to the proposition that there are two classes of COPII vesicles, one carrying GPI proteins and another carrying other proteins, and that sphingolipids are not only enriched in the vesicles that carry GPI proteins but that they are required for their fusion with the cis-Golgi (68). In several experiments, AbA caused distinct molecular weight bands to accumulate (Fig. 3), but this only occurred after prolonged incubation with AbA, whereas the effect of AbA on IPC biosynthesis was almost immediate (not shown). On the other hand, myr affects GPI protein transport rapidly, i.e. after 10 min of preincubation (67). In view of the rapid effect of myriocin on GPI transport, it thus would appear that the transport of GPI proteins to the Golgi requires the continuous supply of Cer or a free long chain base, whereas IPC may be in sufficient quantity and only becomes limiting after a prolonged incubation with AbA.
Our data also represent confirmation and extension of the recently reported dependence of GPI protein transport on COPI (68). In particular, maturation of Gas1p and Yap3p was shown to be drastically slowed in ret1-1, whereas invertase is matured and secreted normally in this mutant. In other COPI mutants, the transport of Gas1p was retarded less drastically. Our data show that most COPI mutants with the exception of ret3-1 accumulate GPI proteins which are immature in size but which already are remodeled since they are labeled by [ 3 H]DHS.
In principle, the maturation deficit of GPI proteins in COPI mutants could be explained either by retention of immature GPI proteins in the ER or by a severe deficit in glycan maturation. Indeed, in several mutants affecting retrotransport to the ER (ret1-1, sec21-3, arf1, vti1, and erd1) as well as after treatment with brefeldin A, glycoproteins are secreted with smaller than normal glycans (50,68,(72)(73)(74). It has not been formally ruled out that the appearance of immature GPI proteins in COPI mutants is due to a deficit in Golgi glucan elongation, but this possibility appears unlikely for several reasons as follows. (i) The size of the GPI proteins accumulating in COPI mutants is the same as in COPII mutants, i.e. elongation is not reduced but absent (not shown). (ii) ␣1,3-Linked mannose is added onto partially elongated N-glycans of invertase in sec21-3 (50) but is rather absent from Gas1p accumulated in ret1-1 (68 In all experiments the cells were radiolabeled as indicated under "Experimental Procedures," and proteins were extracted and analyzed by SDS-PAGE and processed for fluorography. maturation cannot mature the glycans upon chase. Thus our data strongly argue that the maturation deficit in COPI mutants is due to specific retardation of the transport of GPI proteins to the Golgi. Since there is good evidence for the involvement of COPI in both antero-and retrograde transport between the ER and the Golgi, there are several possible interpretations for the retarded export of GPI proteins out of the ER (46,75). Their uptake into transport vesicles may require a special packaging molecule that has to be continuously retrieved from the Golgi to the ER by a COPI-dependent pathway. This kind of scenario has been proposed to explain why at 37°C sec21-3 and sec21-4 do export invertase and HSP150 but not CPY and ␣-factor precursor from the ER to the Golgi. Alternatively, as proposed by Riezman and colleagues (68), there may be two different types of COPII vesicles, only one of which is carrying GPI proteins. This subclass would require special lipids or proteins for fusion with the Golgi, whose enrichment in the vesicle depends on COPI function. Alternatively, COPII and COPI coats may act sequentially for the ER to Golgi transport of subclass of coated vesicles transporting GPI proteins as was proposed for mammalian cells (76,77).
Our data show that all COPII mutants are unable to incorporate [ 3 H]DHS into mature-sized proteins when secretion is blocked or after preincubations with Chx ( Fig. 10, A, lane 4, and D, lanes 4 -6), and there are several possible explanations for this phenomenon. (i) In the past we considered the possibility that Golgi remodeling is absent because the ER to Golgi transport of IPC, a potential remodelase substrate, is dependent on COPII function (9,13,14). The data with AbA now show that IPC is not the substrate for the Golgi remodelase, and we thus can exclude this possibility. (ii) Under "Results," we suggested that COPII mutants are unable to transport Cer to the Golgi. This would mean that Cer reaches the Golgi by COPIIdependent vesicular traffic, implying that in the absence of COPII-mediated transport, there is no COPI-mediated anterograde transport of Cer to the Golgi (50,78) and that Cer cannot be transported by lipid transfer proteins. Although we think this possibility is the most likely one, at present we cannot exclude an alternative explanation, namely (iii) that the Golgi remodelase is relocated to the ER in COPII mutants much in the same way as Emp47p is relocated to the ER is sec12 mutants (79). However, there is no labeling of mature-sized proteins also in sec18 where both antero-and retrograde transport are interrupted at 37°C so that relocation of remodelase to the ER is difficult to envisage. Thus, this latter possibility appears somewhat less likely.
The relative weakness of Golgi [ 3 H]DHS incorporation into mature sized GPI proteins in COPI mutants such as sec21-3 may similarly be caused by insufficient forward transport of Cer from the ER to the Golgi by COPII vesicles but may also be reflecting an inability of this mutant to retrieve GPI proteins from more distal locations to the Golgi for lipid exchange or an inability to retain the remodelase in the Golgi. A deficiency in forward transport of Cer is suggested by the reduced rate of MIPC and M(IP) 2 C synthesis in the Golgi (Table II) which is indicative of a reduction in the forward transport rate of IPC/C. However, we cannot discard the possibility that Cer is packaged into COPII vesicles with the help of molecules that have to be retrieved from the cis-Golgi in a COPI-dependent manner.
In all our experiments we measured the incorporation of [ 3 H]DHS into sphingolipids, particularly since the incorporation into IPC/D and M(IP) 2 C reflects the transport of IPC from the ER to the Golgi. The assessment of the ER to Golgi vesicular flow by measuring the synthesis of IPC/D and of M(IP) 2 C becomes a valuable tool in cases where the rate and extent of N-glycan elongation on glycoproteins cannot be used. Results obtained using the in vitro budding assay showed that the budding of vesicles from the ER is not reduced in microsomes made from Chx-treated cells (80). Our data show that extended preincubations with Chx reduced but never abolished the synthesis of IPC/D and M(IP) 2 C. As can be seen from Table II, sec21-3 when compared with wt, still made 45% of M(IP) 2 C and 93% of IPC/D in the presence of Chx. Thus, despite the toxic effect of Chx, the vesicular flux was maintained in the absence of cargo also in the intact cell. We also compared the synthesis of IPC/D and of M(IP) 2 C in COPI mutants and wt and observed that in most mutants it dropped by 0 -80% but was not completely abolished, although all COPI mutant alleles are lethal at 37°C. In particular, sec21-3 maintained a high level of M(IP) 2 C biosynthesis which was not diminished when the cells were preincubated for up to 60 min at 37°C (Table II).