The membranes of various intracellular compartments of all eukaryotic cells have distinct lipid compositions (van Meer, 1998; Schneiter et al., 1999). The molecular mechanisms that establish and maintain these different lipid compositions are largely unresolved. These processes must be highly regulated to ensure correct membrane assembly for normal cellular function (Futerman et al., 1998). Lipid localization can be achieved by the dynamic interplay between local synthesis, modification, transport, and sorting of various lipids. The first step in the biosynthesis of most membrane lipids occurs in the ER, and then lipids travel to their final destination. This movement must occur selectively by specific mechanisms for each lipid (McIntyre and Sleight, 1994; Trotter and Voelker, 1994; Liscum and Munn, 1999; van Meer and Holthuis, 2000; Voelker, 2000). It is assumed that transport of lipids between intracellular compartments occurs via either vesicular or nonvesicular transport or both. Nonvesicular transport could occur by transport of lipid monomers through the cytosol or by transfer of lipids at regions of close apposition between membranes. Although several lipid transfer proteins and mutants defective in lipid transport have been identified and isolated (Nishijima et al., 1997; Emoto et al., 1999; Liscum and Munn, 1999; Raggers et al., 2000), the precise roles of proteins in net transport and sorting of lipids remain unclear.

In mammalian cells, ceramide is formed by decarboxylation and condensation of L-serine with palmitoyl-CoA, followed by reduction of 3-ketodihydrosphingosine to generate dihydrosphingosine (DHS),* N-acylation of DHS with a fatty acid, and finally desaturation of dihydroceramide (DHS-Cer) (Merrill and Jones, 1990). The biosynthesis of ceramide seems to occur at the cytosolic surface of the ER in mammalian cells (Mandon et al., 1992; Hirschberg et al., 1993), although the ceramide synthase has not been identified. Ceramide, a common precursor for all complex sphingolipids, is then transported to the Golgi apparatus for conversion to both sphingomyelin and glucosylceramide by yet-unknown mechanisms (van Meer and Holthuis, 2000). Recently, it has been shown by in vivo analysis of a mutant resistant to lysenin, an SM-directed cytolysin, that the transport of ceramide to the sites of sphingomyelin and glucosylceramide synthesis occurs via a mechanism that differs from the ER-to-Golgi vesicular transport of glycoproteins (Fukasawa et al., 1999).

In yeast cells, the steps in ceramide biosynthesis are similar to those in mammalian cells except that DHS-Cer is hydroxylated by Sur2p to give phytoceramide (Haak et al., 1997; Grilley et al., 1998), but the exact localization and topology of the reactions of ceramide biosynthesis have not been studied. In contrast to the situation in mammalian cells, it was assumed that ceramide is converted to inositolphosphorylceramide (IPC), one of three major classes of yeast sphingolipids, by addition of inositolphosphate from phosphatidylinositol (PI) in the ER. Subsequently, IPC was thought to be delivered to the Golgi apparatus where IPC is mannosylated to form mannosyl-IPC followed by mannosyl-diinositolphosphorylceramide (Dickson and Lester, 1999; Schneiter, 1999). This idea was based on the fact that IPC synthesis continued at nonpermissive temperature in sec mutants that block ER-to-Golgi vesicular transport of proteins (Puoti et al., 1991). Recently, Aur1p, a protein required for IPC synthesis, has been localized to the Golgi apparatus (Levine et al., 2000), implying that ceramide is delivered to the Golgi apparatus. This suggests that a nonvesicular transport mechanism exists for ceramide transport to the Golgi apparatus.

In vitro systems that reproduce transport events occurring in intact cells are powerful tools to understand molecular mechanisms (Wuestehube and Schekman, 1992; Moreau et al., 1993; Barlowe et al., 1994; Funakoshi et al., 2000). Therefore, we have developed a cell-free system with isolated donor and acceptor membranes that allows us to study the mechanisms of ceramide transport and ceramide synthesis. Here, we show that ceramide is transported from the ER to the Golgi apparatus via two distinct pathways, one vesicle dependent and the other vesicle and ATP independent. We provide biochemical evidence that ER-Golgi membrane contacts play an important role in nonvesicular ceramide transport. A cytosolic factor is required for transport after formation of ER-Golgi contacts.

Figure 1. Incorporation of [3H]DHS into sphingolipids in sec mutants. (A) Wild-type (RH448) and sec18 mutant (RH1737) cells were preincubated with Chx for 20 min at 24°C, then incubated for 20 min at 24 or 37°C, and labeled with [3H]DHS for 2 h (more ...)

Figure 2. Analysis of lipids labeled with [3H]DHS in vitro. (A) In vitro [3H]DHS labeling with microsomal membranes and cytosol from RH981 was performed at 10°C for 15 min and then at 4 or 24°C for 2 h as described in Materials and methods. Aliquots (more ...)

Figure 3. Requirements for ceramide and IPC synthesis in vitro. (A and B) In vitro [3H]DHS labeling under different conditions as indicated and NaOH treatment of the labeled lipids were performed as described in the legend to Fig. 2 A. To the samples containing (more ...)

The identity of ceramide was established by its chemical characteristics and by using specific inhibitors of ceramide synthesis. Base-resistant IPCs and presumed ceramides synthesized in vitro were purified from TLC plate by scraping followed by PI-PLC and/or HCl treatments (Fig. 2, B–D). The presumed ceramides (Fig. 2 B, lane 5) comigrated with ceramides derived from base-resistant IPCs treated with PI-PLC (Fig. 2 B, lane 2) and yielded the sphingoid bases DHS and PHS and an unidentified lipid X on strong HCl hydrolysis (Fig. 2 B, lanes 3, 4, 7, and 8) in solvent system I. When solvent system III was used, two different ceramides could be resolved from PI-PLC–treated IPCs (Fig. 2 C, lane 2). From the presumed ceramide fractions, we observed one extra lipid Z, which migrates further than the two ceramides and was hydrolyzed by HCl (Fig. 2 C, lanes 5 and 6). We have not further investigated the identity of this lipid. Two ceramides, *I and *II (Fig. 2 C, lane 6) were purified and hydrolyzed individually with HCl. This treatment produced DHS from ceramide *I. PHS and X were produced from ceramide *II (Fig. 2 D). These results show that ceramide-1 (*I) is DHS-Cer and ceramide-2 (*II) is a complex of phytoceramide and a ceramide having X. Ceramide-1 and -2 formation was blocked by specific inhibitors of ceramide synthesis, fumonisin B1 (FuB) (Fig. 3 B) (Merrill et al., 1993) and australifungin (unpublished data) (Mandala et al., 1995). The inhibition by FuB reveals that ceramide synthesis is based on the fatty acyl-CoA–dependent reaction, not on the reversal of ceramidases (Mao et al., 2000a,b).

This in vitro system using microsomal membranes showed that ceramide synthesis is temperature dependent and requires ATP, cytosol, and a membrane fraction (Fig. 3 B). When the liposomes containing hexacosanoic acid (C26) and PI were omitted from the reaction mixture, a small reduction (18%) in the synthesis of IPCs was found. We also found that N-ethylmaleimide (NEM) completely inhibited ceramide synthesis, suggesting the requirement of an NEM-sensitive protein or compound (unpublished data).

Next, we investigated whether ceramide and IPC synthesis could be reconstituted using purified ER- and Golgi-enriched membranes. ER-enriched membranes were prepared as described previously (Wuestehube and Schekman, 1992), and Golgi-enriched membranes were isolated by differential centrifugation (see Materials and methods). Western blotting analysis showed that Wbp1p was found exclusively in P27 fraction and P45 fraction, resulting from further fractionation of S27, whereas Emp47p, a Golgi marker (Schröder et al., 1995) and Aur1p-HA were found mainly in the final supernatant (S45) (Fig. 4 A). Therefore, we used the S45 fraction as source of Golgi-enriched membranes even though a small amount of ER membranes remained in this fraction. Fractionation of the P27 fraction on a sucrose gradient resulted in further purification of ER membranes (at the 40/50% sucrose interface) as shown in Fig. 4 B. Fig. 4 C shows an experiment in which ceramide and IPC synthase activities were examined with the purified ER- and Golgi-enriched membranes. As expected, efficient synthesis and subsequent conversion of ceramide to IPC required the addition of both ER- and Golgi-enriched membranes. When only ER membranes were incubated, the amount of ceramide synthesized remained high (87% of complete incubation), but the level of IPC synthesis was low. This confirms that IPC synthase activity is localized in Golgi membranes. In contrast, addition of only Golgi membranes resulted in significant reductions of ceramide and IPC synthesis by 41 and 27%, respectively. Since a detectable amount of Wbp1p was found in the Golgi membrane fraction, the lack of complete loss of ceramide synthase activity is most likely due to ER contamination in the Golgi fraction. The above results are consistent with the localization of ceramide synthesis to the ER. Ceramide would then be transported to the Golgi apparatus for conversion to IPC.

Figure 4. Reconstitution of ceramide and subsequent IPC synthesis with ER- and Golgi-enriched membranes. (A and B) Purification of ER- and Golgi-enriched membranes was performed as described in Materials and methods. (A) Microsomal membranes from RH5256 were subjected (more ...)

Figure 5. Cytosol- but not ATP-dependent transport of ceramide in vitro. (A) Wild-type (WT, RH5247) and AbA-resistant (RH5255) cells were preincubated with ethanol or AbA (20 μg/ml) for 20 min at 24°C and then labeled with [3H]DHS for 30 min at (more ...)

Figure 7. Ceramide transport occurs through ER-Golgi membrane contacts. In vitro ceramide transport assay was performed under ATP depletion conditions as described in the legend to Fig. 5 C. (A) For dilution experiments, the reaction mixtures for transport assay (more ...)

Since cytosol was required for conversion of ceramide to IPC, we next tested whether the cytosolic component was specific and proteinaceous (Table I). The cytosol requirement was specific because BSA could not substitute for cytosol. Furthermore, addition of Triton X-100 resulted in an almost complete loss of the cytosol requirement, suggesting that cytosol is required for ceramide transport but not IPC synthase activity. Cytosol treated with trypsin or treated at 95°C for 15 min did not support the efficient conversion to IPC. These results indicate that a heat-labile and trypsin-sensitive cytosol protein(s) is required for ceramide transport. In contrast, treatment of donor, acceptor, or both membranes with trypsin had little inhibitory effect on conversion of ceramide to IPC, although under the same conditions trypsin treatment resulted in complete removal of the HA epitope tag from the COOH terminus of Aur1p (unpublished data).

Table I. Cytosolic protein is required for ATP-independent transport of ceramide

The next set of experiments was performed to determine whether ceramide transport in our assay requires ER-to-Golgi vesicular transport. Several studies using in vitro cell-free assays have shown that NEM and nonhydrolyzable analogs of GTP such as GTPγS inhibit ER-to-Golgi-vesicular transport (Baker et al., 1990; Barlowe et al., 1994). As seen in Fig. 6 A, neither GTPγS nor NEM had a strong inhibitory effect on ceramide transport. The nonvesicular nature of the transport reaction was confirmed by using cytosol and microsomal membranes from sec18 mutant cells, which are defective for delivery of ER-derived vesicles to the Golgi at restrictive temperature in vitro (Muñiz et al., 2001). Formation of IPC was monitored by a one step assay at permissive (24°C) and restrictive (30 and 37°C) temperatures. Although high temperature reduced the extent of IPC synthesis in both wild-type and sec18 extracts, no specific defect inthe sec18 mutant was detected (Fig. 6 B). Therefore, our results demonstrate that transport of ceramide from ER to the Golgi compartment in our assay occurs by nonvesicular transport.

Figure 6. Ceramide is transported via a nonvesicular pathway. (A) In vitro ceramide transport assay were performed in the presence of cytosol and ATP as described in the legend to Fig. 5 B. GTPγS (0.4 mM), GTP (0.4 mM), GTP (0.4 mM) plus Sar1p (90 μg/ml), (more ...)

Two mechanisms for nonvesicular transport between membranes have been postulated. One is transfer of lipid through the cytosol and the other is transfer of lipid through direct membrane contacts (McIntyre and Sleight, 1994; Trotter and Voelker, 1994). The former should be strongly dependent on ER and Golgi membrane concentrations throughout the reaction. However, preliminary experiments suggested that the reaction was dilution independent at several time points of the reaction at 24°C. Therefore, we examined the effect of dilution on ceramide transport after a preincubation of donor and acceptor membranes together at 4°C where transport does not occur. Fig. 7 A shows the kinetics of ceramide transport with different preincubation conditions. IPC synthesis was not affected by dilution after premixing of membranes at 4°C for 5 min. Control experiments showed that when membranes were diluted independently before the mixing (0 min preincubation), no time-dependent conversion of ceramide to IPC was observed (Fig. 7 B). Therefore, once the two membranes have been preincubated at 4°C, synthesis of IPC is a zero order reaction with respect to membranes. This strongly suggests that ceramide transport occurs through ER-Golgi membrane contacts. In addition, we found that when donor and acceptor membranes were preincubated at 4°C with cytosol and then diluted without cytosol, no efficient conversion of ceramide to IPC was observed (Fig. 7 C), demonstrating that cytosol is required after ER-Golgi membrane contacts. In contrast, preincubation at 4°C without cytosol and subsequent dilution with cytosol did not affect the conversion of ceramide to IPC. Therefore, cytosol is not required under our conditions to establish ER-Golgi membrane contacts.

Figure 8. SEC-independent IPC biosynthesis does not require ATP in intact cells. Mutant sec18 (RH1737) cells labeled with [3H]myo-inositol were subjected to ATP depletion procedure and then chased at restrictive temperature as described in Materials and methods. (more ...)

Transport of lipids between subcellular compartments must be achieved by one or more of the following possible mechanisms: vesicular transport with budding of vesicles from donor membranes and targeting/fusion with the acceptor membranes, transfer of lipid monomers through the cytosol, or transfer of lipid through membrane contacts (McIntyre and Sleight, 1994; Trotter and Voelker, 1994; Liscum and Munn, 1999; van Meer and Holthuis, 2000; Voelker, 2000).

Our study provides evidence that ceramide is transported from the ER to the Golgi apparatus by both vesicular and nonvesicular means in yeast. From in vivo experiments, IPC synthesis from radiolabeled DHS was reduced but still observed when ER-to-Golgi vesicular transport was blocked in temperature-sensitive secretion mutants even under the conditions where protein synthesis was also blocked in sec18 mutant cells. This SEC18-independent IPC synthesis was also not affected by ATP depletion. Consistent with this, IPC synthesis was inhibited partially by ATP depletion in wild-type cells (unpublished data). These results suggest that ceramide is transported from the ER to the Golgi apparatus via two distinct pathways in intact cells, one vesicle dependent and the other nonvesicular. Analysis using an in vitro assay demonstrated that transport of ceramide from the ER to the Golgi compartment was not sensitive to ATP depletion and was SEC18 independent, indicating a nonvesicular transport mechanism. The correlation between in vivo and in vitro experiments provides evidence that the in vitro assay for ceramide transport measures a physiologically relevant transport process.

Some previous studies have suggested that nonvesicular transport of ceramide in vitro does not require cytosol (Moreau et al., 1993; Kok et al., 1998). Recently a cytosolic protein requirement has been demonstrated using semiintact cells to measure ceramide transport from the ER to the site of synthesis of sphingomyelin in the Golgi compartment (Funakoshi et al., 2000). In vitro analysis with the LY-A mutant, displaying a specific defect in ER-to-Golgi trafficking of ceramide to the site of sphingomyelin synthesis, showed that LY-A mutant cells are defective in a cytosolic protein involved in the ceramide transport. Since ER-to-Golgi transport of GPI-anchored or transmembrane proteins in LY-A cells appeared to be normal, it was suggested that ceramide is delivered to the Golgi apparatus by a mechanism that differs from ER-to-Golgi vesicular transport of proteins. Although the delivery of ceramide from the ER to the Golgi apparatus is mediated by two pathways, ATP-dependent and ATP-independent (nonvesicular), the defect in LY-A cells is specific to ATP-dependent trafficking of ceramide (Fukasawa et al., 1999). It is not clear whether the ATP-dependent delivery of ceramide is via one of multiple vesicular pathways that is specific for ceramide or nonvesicular pathways. In this respect, evidence for the existence of two distinct vesicle populations upon budding from the ER was provided recently (Muñiz et al., 2001).

The dilution independence of ceramide transport after ER and Golgi membranes have been incubated together at 4°C, where transport does not occur, strongly suggests that stable membrane contacts are formed and that these are relevant to the nonvesicular transport process. The ER-to-Golgi membrane contacts are likely to be saturable and specific for ceramide transport, since microsomes from wild-type cells inhibited ceramide transport in vitro, whereas addition of purified mitochondria did not. Our data provide important biochemical evidence for a role of membrane–membrane contacts in intracellular lipid transport and provide a system to begin to characterize this process.

Experimental evidence for an important function of membrane contacts has been found for ER-to-mitochondria trafficking of phosphatidylserine (Voelker, 1993; Achleitner et al., 1999). In addition, a close apposition of trans-ER with trans-Golgi cisternae has been reported by a high voltage electron microscope tomographic study (Ladinsky et al., 1999). Since the trans-ER lacks buds, it is most likely that this close apposition is associated with transport of lipids via nonvesicular rather than with vesicular transport of proteins.

ER-to-Golgi transport of ceramide for glucosylceramide synthesis was almost independent of ATP and cytosol (Funakoshi et al., 2000). In contrast, our experiments have shown that ATP-independent nonvesicular transport of ceramide requires a heat-labile and trypsin-sensitive cytosol protein(s). To our knowledge, this is the first clear evidence for a cytosolic protein requirement for nonvesicular transport of ceramide. What could be the function of this cytosolic factor?

Our dilution experiments have shown that the cytosol requirement is involved in the late step of ceramide transport after establishment of ER-Golgi membrane contacts. Thus, the cytosolic protein(s) seems to be required for transfer of ceramide from ER to Golgi membranes. It is unlikely that a cytosolic factor would be required for the transbilayer movement of ceramide after it has reached the Golgi compartment. Therefore, the late cytosolic requirement also makes it unlikely that ceramide is transferred from ER to Golgi during our short incubation at 4°C. The processes of membrane–membrane contact and transfer could be obligatorily coupled because cytosol could not efficiently extract ceramide out of the ER without acceptor membranes (unpublished data).

Analysis using our in vitro transport assay also revealed that ER-to-Golgi transport of ceramide continues for >60 min with a lag period from 5 to 15 min. The rather long time requirement could be due to ceramide transport to the Golgi complex and/or reflect the need to translocate ceramide from the cytosolic to the lumenal side of Golgi apparatus. Based on the predicted topology of Aur1p and localization of its putative active site in the lumen of the Golgi compartment, this translocation step would be required in our assay (Levine et al., 2000). A lumenal active site is consistent with our findings that trypsin treatment of membranes did not destroy conversion of ceramide to IPC. The half times for spontaneous translocation of short acyl chain ceramide (C5-DMB-ceramide) across lipid bilayers has been reported to be 22 min (Bai and Pagano, 1997), but measurements for natural ceramides are not available. Since short acyl chain lipids appear to undergo rapid transbilayer movement, on the basis of the correlation between the hydrophobicity of acyl chain and the rate of transbilayer movement (Bai and Pagano, 1997), it would follow that the spontaneous translocation of natural ceramide would occur in the order of hours and that specific proteins must be involved in the movement of ceramide. This translocation should be ATP independent because nonvesicular transport of ceramide does not require ATP. Such specific lipid translocators or flippases exist in the ER, Golgi apparatus, and plasma membrane (Raggers et al., 2000). Most of them require energy (e.g., ATP), but some proteins such as scramblase are involved in ATP-independent transbilayer lipid movement. In addition, ATP-independent protein-mediated lipid movement has been postulated in the translocation of glucosylceramide from the cytosolic to the lumenal leaflet of the Golgi membranes.

Our in vitro transport assay seems to reconstitute only the nonvesicular ceramide transport pathway. The reasons for the lack of reconstitution of the vesicular pathway are not clear, but in yeast the Golgi apparatus is made up of apparently unconnected compartments. Aur1p is located primarily in the medial-Golgi but not cis-Golgi compartments (Levine et al., 2000). Therefore, it seems plausible that ceramide transported to the cis-Golgi compartment via vesicular mechanism could be not delivered to the IPC synthase compartment efficiently in our assay because it involves too many transport steps. This may help explain why GTP plus Sar1p inhibits ceramide transport in our assay. GTP and Sar1p stimulate vesicle budding (Barlowe et al., 1994), and ceramide could be sequestered in these vesicles. In any event, it seems likely that nonvesicular pathway is used directly to generate IPC. To show that the direct fusion of ER with Golgi membranes did not occur under our assay conditions, we assayed for the acquisition of α-1,6 mannose modifications onto the ER form of the GPI-anchored protein, Gas1p (Muñiz et al., 2001). Less than 5% (background) of the Gas1p received this modification. Therefore, this cannot account for the >20% transport of ceramide that we see in our assay (unpublished data).

We postulate that the two different pathways of ceramide transport may have different functions. Vesicular-dependent transport would function to establish and maintain the lipid composition of ER and Golgi membranes during ER to early Golgi trafficking associated with possible functions of ceramide in regulation of cellular processes (Futerman et al., 1998; Hannun and Luberto, 2000) such as protein transport and signal transduction. For example, the sorting of GPI-anchored proteins from the ER is regulated by sphingoid base or ceramide synthesis (Horvath et al., 1994; Sütterlin et al., 1997). Nonvesicular transport would function to provide ceramide directly for de novo biosynthesis of sphingolipids for their important roles. These hypotheses will be addressed by further studies of the molecular mechanisms underlying the nonvesicular transport of ceramide.

For the energy deprivation experiments, sec18 cells were preincubated for 20 min at 1–3°C and labeled with [³H]myo-inositol at 1–3°C. After 45 min, myo-inositol (100 μg/ml) and NaF/NaN₃ (each 10 mM) were added and incubated for 20 min at 1–3°C. The cells were then chased for 30 min at 37°C after the addition of 4 vol of SD medium containing myo-inositol.

After the incubations, cells were placed on ice and washed, and cell pellets were subjected to lipid extraction (Zanolari et al., 2000). If necessary, the extracted lipids were submitted to mild alkaline hydrolysis with NaOH. After neutralization with acetic acid, the lipids were desalted by partitioning between N-butanol and water. The pooled organic phases were dried under nitrogen. The lipids were dissolved in chloroform/methanol/water (CMW; 10/10/3 [vol/vol/vol]) for TLC on Kieselgel 60 plates (20 × 20; Merck) and developed in solvent I (chloroform/methanol/4.2 N NH₄OH, 9/7/2 [vol/vol/vol]) or solvent II (chloroform/methanol/0.25% KCl, 55/45/10 [vol/vol/vol]). Radiolabeled lipids were visualized and quantified on a Cyclone Storage Phosphor System using a tritium-sensitive screen (Packard).

We thank R. Schekman for strains and antibodies, S. Munro for plasmids, J. Holenstein and T. Aust for technical assistance, and members of the Riezman lab for comments on this article.

This work was supported by a grant from the Swiss National Science Foundation and a European Community grant (HPRN-2000-00077) funded through the Bundesamt für Bildung und Wåssenschatt (to H. Riezman).