Functional Organization of the Golgi Apparatus in Glycosphingolipid Biosynthesis LACTOSYLCERAMIDE AND SUBSEQUENT GLYCOSPHINGOLIPIDS ARE FORMED IN THE LUMEN OF THE LATE GOLGI *

Biosynthesis of plasma membrane sphingolipids involves the coordinate action of enzymes localized to individual compartments of the biosynthetic secretory pathway of proteins. These stations include the endoplasmic reticulum and the Golgi apparatus. Although a precise localization of all the enzymes that synthesize glycosphingolipids has not been achieved to date, it is assumed that the sequence of events in glycosphingolipid biosynthesis resembles that in glycoprotein biosynthesis, i.e. that early reactions occur in early stations (endoplasmic reticulum and cis/medial Golgi) of the pathway, and late reactions occur in late stations (trans Golgi/trans Golgi network). Using truncated analogues of ceramide and glucosylceramide that allow measurement of enzyme activities in intact membrane fractions, we have reinvestigated the localization of individual enzymes involved in glycosphingolipid biosynthesis and for the first time studied the localization of lactosylceramide synthase after partial separation of Golgi membranes as previously described (Trinchera, M., and Ghidoni, R. (1989) J. Biol. Chem. 264, 15766–15769). Here, we show that the reactions involved in higher glycosphingolipid biosynthesis, including lactosylceramide synthesis, all reside in the lumen of the late Golgi compartments from rat liver.

Biosynthesis of plasma membrane sphingolipids involves the coordinate action of enzymes localized to individual compartments of the biosynthetic secretory pathway of proteins. These stations include the endoplasmic reticulum and the Golgi apparatus. Although a precise localization of all the enzymes that synthesize glycosphingolipids has not been achieved to date, it is assumed that the sequence of events in glycosphingolipid biosynthesis resembles that in glycoprotein biosynthesis, i.e. that early reactions occur in early stations (endoplasmic reticulum and cis/medial Golgi) of the pathway, and late reactions occur in late stations (trans Golgi/trans Golgi network).
Using truncated analogues of ceramide and glucosylceramide that allow measurement of enzyme activities in intact membrane fractions, we have reinvestigated the localization of individual enzymes involved in glycosphingolipid biosynthesis and for the first time studied the localization of lactosylceramide synthase after partial separation of Golgi membranes as previously described (Trinchera, M., and Ghidoni, R. (1989) J. Biol. Chem. 264, 15766 -15769). Here, we show that the reactions involved in higher glycosphingolipid biosynthesis, including lactosylceramide synthesis, all reside in the lumen of the late Golgi compartments from rat liver.
Sphingomyelin (SM) 1 and glycosphingolipids (GSLs) are major constituents of the outer leaflet of the plasma membrane in eukaryotic cells. Their involvement is discussed in neuronal growth (1), in signal transduction (2)(3)(4), cell growth (5), cell-cell recognition (2,6), and the receptor-mediated endocytosis of bacterial toxins and viruses (6,7). Although a variety of reports exist on the subcellular localization of individual enzymes involved in their biosynthesis and on their biosynthetic transport to the plasma membrane (reviewed in Ref. 8), our picture of the functional organization of the endomembrane systems involved in GSL biosynthesis is not complete.
The lipid moiety of sphingolipids, ceramide, is produced on the endoplasmic reticulum (9,10) and, within the lumen of the ER, part of it in some cell types is converted to form galactosylceramide (11,12). For further sphingolipid synthesis, ceramide is then carried to the Golgi apparatus in either a vesicular (13,14) or a nonvesicular transport step (15,16). The Golgi apparatus comprises stacks of various cisternae, the cis Golgi network, cis, medial, and trans cisternae, and the trans Golgi network (TGN) (17). Ceramide is then converted to sphingomyelin in the lumen of the early (cis/medial) Golgi (18,19) by transfer of the phosphorylcholine head group of phosphatidylcholine to ceramide (20). In some cell types, a minor part of SM seems to be generated in the outer leaflet of the plasma membrane (21).
In glycoprotein biosynthesis, functions of the various Golgi subcompartments are well established: early reactions in glycoprotein trimming and processing take place in the lumen of the early (cis/medial) Golgi, and late reactions take place in the trans Golgi/trans Golgi network (summarized in Refs. 17, 22, and 23). Biosynthesis of complex glycosphingolipids, like that of glycoproteins, involves sequential glycosyltransferase reactions, starting with the formation of glucosylceramide, and it was assumed that the various transferases are functionally organized within the Golgi in a way similar to protein glycosyltransferases (24). This view results from experiments including partial separation of Golgi subcompartments by sucrose density gradient centrifugation and subsequent determination of the various enzymatic activities in the presence of detergent (25)(26)(27) or the use of inhibitors that block various steps in vesicular transport, like the fungal macrolide brefeldin A (BFA) (28 -31) and the antibiotic monensin (32,33). However, differences exist in the topology of these reactions. Rather than lumenal, GlcCer is synthesized on the cytoplasmic face of the Golgi (34 -36), not only in the early but also in the late subcompartment of this organelle (36,37). GlcCer is then converted to LacCer by LacCer synthase (38 -40), a lumenal Golgi enzyme (12,39), the localization of which to Golgi subcompartments has not been described to date. This has prompted us to investigate the localization within the Golgi of this enzyme and to reinvestigate the topography of various additional glycosyltransferases involved in GSL biosynthesis using sucrose density gradient centrifugation. To this end, the conditions for assaying enzyme activities in partially separated Golgi subfractions (25,26,41) were optimized. Using a truncated analogue of ceramide with only eight carbon atoms in both its sphingosine and fatty acid moieties (C 8 C 8 -ceramide), transferase activities can be measured in the absence of detergent, because the truncated substrates readily permeate mem-branes (39,42). Assaying intact membranes has the advantage that overall higher transferase activities are obtained and the enzymes are characterized functionally, because in intact membranes their activities depend on functionally intact translocators for their respective substrates, such as, e.g. CMP-NeuAc or UDP-Gal (43,44). Moreover, care was taken to inhibit ␤-glucosidase, which is present at different levels of activity in the various subfractions examined, because degradation of the substrate added or the product formed in an enzyme assay may profoundly change the result. Here we show that, in contrast to the functional organization of the Golgi apparatus in glycoprotein biosynthesis, all lumenal activities investigated for the formation of more complex GSLs reside in the late Golgi, including LacCer synthesis.
Isolation of Golgi Membranes-For enzyme activity determinations without further partial separation into subcompartments, intact Golgi membranes from livers of fasted rats (Wistar male rats, 180 -200 g) were isolated according to the method described in Ref. 45 with some modifications. All sucrose solutions were in 10 mM Tris-maleate buffer, pH 7.4. The homogenization buffer contained 0.5 M sucrose, 5 mM EDTA, and 5 mM dithiothreitol.
Characterizations with marker enzyme activities were performed as described earlier (36,39). Protein concentrations were determined using the BCA method (Pierce).
Treatment of Golgi with Pronase E-Pronase E treatment of membranes was as described (36,39), with the following modifications: 100 g membrane protein in 10 mM Tris-maleate, pH 7.4, containing 50 mM NaCl and 250 mM sucrose was incubated with Pronase E (protein/ protease ϭ 5:1) in a total volume of 100 l at 37°C for various time intervals. After incubation, the samples were diluted 3-fold with icecold buffer, loaded on 1 ml of 10 mM Tris-maleate, pH 7.4, containing 500 mM sucrose and 2 mg of bovine serum albumin/ml, and the membranes were pelleted by centrifugation (30 min at 170,000 ϫ g and 4°C). The pellets were resuspended in the above buffer, and aliquots were used for the determination of the following enzyme-activities: for GlcCer and SM synthase, [ 3 H]t-Cer was used as a substrate, and for ␤-glucosidase, [ 3 H]t-GlcCer was used.
Isolation and Subfractionation of Golgi Membranes for Topographical Studies-To obtain partially purified Golgi subcompartments, a protocol was followed as described in Ref. 25, and the following markers were used to characterize the subfractions obtained: esterase and galactosyltransferase (Gal-T) as described (46) and N-acetylglucosaminylphosphotransferase (GlcNAc-P-T) according to the method described in Ref. 47. Of each 1-ml fraction, the following aliquots were used: 10 l for esterase, 20 l for Gal-T, and 20 l for GlcNAc-P-T. In addition, various types of membranes were analyzed immunologically with antibodies directed against dipeptidylpeptidase IV (DPP IV) (rat liver plasma membranes (48,49)), rab5 (early endosomes (50, 51)), and TGN38 (trans Golgi network (52)). Several independent gradients were run with nearly the same results. A partial separation of early and late Golgi membranes was always found, with the maxima of marker activities separated by three fractions. Fractionation of the gradients allowed a reproducibility of the maxima range of Ϯ1 fraction without affecting their distance. Three gradients that showed highly reproducible patterns are shown under "Results"; they were used for the immunological and activity assays presented. Immunological analyses were performed twice in independent gradients and identical results were obtained (see Fig. 2).
Assays of Sphingolipid Biosynthesis with Truncated Substrates-Activities of SM and GlcCer synthase were determined as described (36). A standard assay to follow t-GSL synthesis in Golgi membranes contained in a total volume of 10 l, 1-5 g of protein, 50 M [ 3 H]t-GlcCer (specific activity, 0.2 Ci/nmol), 1 mM conduritol B epoxide (CBE), 100 mM KOAc, 1 mM Mg(OAc) 2 , 1 mM MnCl 2 , 5 mM NaCl, and 2 mM UDP-Gal in 50 mM Hepes/KOH, pH 6.5, to measure LacCer synthase. Further addition of 2 mM CMP-NeuAc leads to the synthesis of t-GM3 and t-GD3. These conditions were found to be optimal after variations of buffer, pH, ion concentrations, and concentration of substrates. Enzyme activities were found to be linear for up to 2 h. For standard assays, samples were incubated for 40 min at 37°C. To measure these enzyme activities across the gradients, 4 l of each fraction was used. To determine the activities in the presence of detergent, a final concentration of 0.05% Triton CF54 was used. The reaction was stopped by the addition of an equal volume of 2-propanol, and then the samples were centrifuged for 5 min at 10,000 ϫ g, and an aliquot (10 l) of the supernatants was analyzed by TLC.
Identification of t-GM3 and t-GD3-t-GM3 was isolated by scraping the band from a preparative TLC and hydrolysis with sialidase from V. cholerae (Sigma). Only t-LacCer was found as the product. When the isolated t-GM3 was used as a substrate, Golgi membranes in the presence of detergent converted this ganglioside into t-GD3 as evidenced by an additional spot on the TLC that appeared only after the addition of CMP-NeuAc.
Assay of ␤-Glucosidase Activity-Activity of ␤-glucosidase was determined using [ 3 H]t-GlcCer (50 M) as substrate, under conditions described above, in the absence of CBE and detergent. A total volume of 10 l contained 4-l gradient fractions. The hydrolysing activity was calculated from the sum of [ 3

H]t-SM and [ 3 H]t-Cer formed.
Assays of Ganglioside Biosynthesis with Long Chain Substrates and Detergent-treated Membranes-GM3 synthase (SAT-I) was determined with minor modifications according to the method described in Ref. 41. In a total volume of 20 l, 200 M long chain LacCer, 0.05% Triton CF 54, 50 mM Hepes/KOH, pH 6.5, 100 mM KOAc, 5 mM Mg(OAc) 2 , 10 mM MnCl 2 , 1 mM [ 3 H]CMP-NeuAc (4000 -6000 cpm/nmol), and 10 l of gradient-fractions were incubated for 30 min at 37°C. The reaction was stopped by the addition of 20 l of chloroform/methanol (2:1), and an aliquot (10 l) of the organic phase was analyzed by TLC. Identification of GM3 was performed by comparison with an authentic standard substance.
TLC and Evaluation of Radioactivity-Whatman silica gel plates (LK 6) were developed with a mixture of chloroform, methanol, and 0.22% CaCl 2 in water (65:35:8) as a solvent. Identification of the various glycosphingolipids was performed either by incorporation of 3 H-labeled sugars from their nucleotide-activated precursors (long chain GSLs) or by hydrolysis with the appropriate specific glycosidases (t-GSLs). For fluorography, the chromatograms were prepared according to the method described in Ref. 54. For determination of radioactivity, the chromatograms were evaluated on an automatic TLC two-dimensional analyzer (digital autoradiograph) (Berthold, Wildbad, Germany). The scanner counts the radioactivity of 3 H-labeled spots on the TLC plate with a yield of 0.88% compared with liquid counting. When the spots counted for 1 h (cph) in the scanner (e.g. 2700 cph) are measured by liquid scintillation counting, they yield 113-fold of the radioactivity registered (in our example, 5085 cpm). In this system a significant signal is obtained down to 100 cph, representing 188 cpm in the scintillation system.

SDS-Polyacrylamide Gel Electrophoresis and Immunoblot
Analysis-Proteins were separated on SDS-10% polyacrylamide gels under reducing conditions (55). Western blot analysis was performed according to the method described in Ref. 56 with antibodies directed against DPP IV, rab5, and TGN38. Immunoreactions were visualized by chemiluminescence (ECL, Amersham Corp.).
Electron Microscopy-For morphological evaluations, aliquots of fractions 15 and 18 were diluted 10-fold with 10 mM Tris/maleate, pH 7.4, and the membranes were sedimented by centrifugation (60 min at 4°C and 100,000 ϫ g). The pellets were fixed with 2.5% glutaraldehyde in 0.1 M sodium-cacodylate buffer, pH 7.6, containing 4% polyvinylpyrrolidone (M r 30,000) and 0.05% calcium chloride. After washing with 0.1 M cacodylate buffer, pH 7.6, three times, the pellets were embedded in agar (extra pure) and cut into thin slices. Postfixation with osmium tetroxide and tannic acid treatment were according to the method described in Ref. 57. After washing in cacodylate buffer, the slices were dehydrated in a gradient of ethanol and embedded in Epon 812. Ultrathin sections were stained with alkaline lead citrate and analyzed using a Zeiss EM 10 electron microscope.

␤-Glucosidase Activity in Partially Separated
Golgi Subfractions-To obtain valid activity quantifications, GSL-degrading activities need to be excluded in the corresponding assays. The observation of ␤-glucosidase activity in intact Golgi membranes (35,39) prompted us to characterize the topography and topology of this activity in partially separated Golgi subcompartments isolated by sucrose density gradient centrifugation according to the method described in Ref. 25. Addition of [ 3 H]t-GlcCer and UDP-Gal to nonfractionated Golgi membranes led to the formation of t-LacCer (Fig. 1). In addition, free [ 3 H]t-Cer was formed (indicative for the presence of ␤-glucosidase), part of which was converted into [ 3 H]t-SM (Fig. 1, lane 2). In the presence of CBE hydrolysis of [ 3 H]t-GlcCer was efficiently inhibited (Fig. 1, lane 3) (58, 59). These conditions allowed the determination of the ␤-glucosidase (without added CBE) or LacCer synthase (in presence of CBE). ␤-Glucosidase activity was found by protease latency experiments to be lumenal and did not follow typical Golgi peak activities after subfractionation (data not shown). Part of this activity may be present in the lumen of Golgi subcompartments en route to the lysosomes.
Lactosylceramide Is Synthesized in the Lumen of the Late Golgi-The activity profile of LacCer synthase after partial separation of Golgi subcompartments is shown in Fig. 2A with a peak in fraction 18. The following marker proteins were quantitated immunologically to determine membrane contaminants within the fractions across the gradient: for liver plas-mamembranes DPP IV (48,49) (Fig. 2B), for early endosomal membranes rab5 (50, 51) (Fig. 2C), and for the trans Golgi network TGN38 (52) (Fig. 2D). As an ER marker, the activity of esterase was determined (46) (Fig. 2E). Golgi subcompartments were characterized by GlcNAc-P-T (cis Golgi (60)) and by Gal-T (trans Golgi (61)), as shown in Fig. 2E, and confirm their distribution across the gradient as described earlier (18,25,26,36). In Fig. 2F, the sucrose density and protein profiles of the gradients investigated are given. The activity for t-LacCer synthesis is restricted to late Golgi subfractions, and the peak activity coincides with the marker activity for trans Golgi, protein-Gal-T (compare Fig. 2, A and E). Clearly, contaminating plasma membranes, as well as early endosomes, peak differently in the gradient (see Fig. 2, B and C) excluding these membranes as residences for LacCer synthase.
It is of note that determination of sphingolipid synthesizing activities with either t-Cer or t-GlcCer as a substrate can be performed in the absence of detergent and therefore reflects the enzyme activities of the functional Golgi compartments. Generally, GSL biosynthesis beyond GlcCer is the result of both translocation and enzyme catalyzed processes. t-LacCer synthesis, for example, requires the translocation of t-GlcCer and UDP-Gal into the lumen of the Golgi and the conversion to t-LacCer by the lumenal LacCer synthase (39,44). To investigate whether enzyme activity in additional subcompartments of the Golgi is restricted by the absence of translocation activities for the substrates needed, we determined GSL synthesis in the presence and absence of detergent. To this end, the activities for [ 3

H]t-LacCer formation from [ 3 H]t-GlcCer and UDP-Gal and for [ 3 H]t-GM3 synthesis from [ 3 H]t-GlcCer
, UDP-Gal, and CMP-NeuAc were compared in Golgi membranes in the presence and absence of Triton CF 54 at a concentration of 0.05%, optimal for GM3 synthase (SAT-I) when assessed by use of long chain, physiological LacCer as substrate (41). In the presence of detergent, the synthesis of [ 3 H]t-LacCer and [ 3 H]t-GM3 was decreased by a factor of four. This has still allowed us to determine, with [ 3 H]t-GlcCer and UDP-Gal, the activity of LacCer synthase in the presence of detergent. The resulting profile is shown in Fig. 2A, with a peak of LacCer synthase activity coincident with that obtained in the absence of detergent.
Interestingly, not only is a functional partial separation of membranes observed across the sucrose gradients, but partial separation is also reflected morphologically: electron microscopy of the early Golgi peak fraction (Fig. 2E, 15) shows a high content of individual Golgi cisternae and some stacks (Fig. 3A), whereas the late Golgi peak fraction (Fig. 2E, 18) contains only a few individual cisternae and consists mainly of a more homogeneous population of vesicular structures (Fig. 3B).
GM3, GD3, and GM2 Synthase Activities Reside in the Lumen of the Late Golgi-Thus, surprisingly, LacCer synthase, the first lumenal activity in the synthesis of gangliosides, is a late Golgi enzyme. In light of reports that SAT-I (catalyzing the biosynthetic step after LacCer) resides in the early Golgi (25)(26)(27), an exclusive late Golgi activity for LacCer synthesis would imply a retrograde transport step for LacCer. This has prompted us to reinvestigate the localization of the enzyme activities for steps in ganglioside biosynthesis subsequent to LacCer.
To this end, we analyzed the activity profiles for GM3 and GD3 synthesis in the partially separated Golgi subfractions. This was performed with [ 3 H]t-GlcCer as a substrate in the presence of both UDP-Gal and CMP-NeuAc. As a result, t-GM3 formation was observed in the late Golgi with a peak activity coincident with the peak of t-LacCer synthesis (not shown; see observed, indicating that t-LacCer formation represents the rate-limiting step for t-GM3 synthesis. This result shows GM3 synthase activity in the late Golgi but does not exclude the presence of additional GM3 biosynthesis steps in earlier subcompartments of the organelle, because t-GM3 formation under the conditions used for the assay strictly depends on Lac-Cer synthase activity, which in turn is restricted to the late Golgi. Therefore, GM3 synthase activity was followed using an assay system with long chain LacCer and [ 3 H]CMP-NeuAc as substrates, in the presence of detergent (41). Under these conditions, part of the resulting GM3 was further sialylated to yield GD3 (not shown), and therefore the formation of both GM3 and GD3 reflects the overall activity of GM3 synthase. Those activities show a profile (Fig. 4A) nearly identical with that of LacCer synthase (depicted in Fig. 2A), irrespective of the experimental conditions. Thus, both LacCer and GM3 synthases peak in the late Golgi. Likewise, the activity profile for GM2 synthase in the presence of detergent shows localization of this enzyme to the late subcompartments of the organelle (Fig. 4B). In addition, all of these activity profiles peak with the established marker activity of the trans Golgi, protein-Gal-T (61) (Fig. 2E).

Characterization of Early and Late Golgi-
We have characterized the partial separation of Golgi subcompartments obtained by sucrose density gradient centrifugation according to the method described in Ref. 25 by use of the following markers: the activity profile of GlcNAc-P-T (60, 62, 63) peaks in fraction 15 representing the early Golgi, and the late Golgi is characterized by a peak in fraction 18 of protein Gal-T, a marker for the trans Golgi (61, 64 -66). Additional markers for membranes known to contaminate the Golgi fractions show that neither ER nor early endosomal or plasma membranes peak with early or late Golgi markers. TGN38, a marker for the TGN (52,67) shows its strongest signal coincident with the peak of Gal-T.
Localization of Lumenal Enzymes Involved in GSL Biosynthesis-The activity profiles for synthesis of t-LacCer and GM3 colocalize both with each other and with the late Golgi markers Gal-T and TGN38. The activity profile for GM2 synthesis peaks with the same marker enzymes. GM2 synthase was found also to catalyze the formation of GA2 and GD2 (53). These results were confirmed by Lutz et al. (68), who characterized the cDNA for GalNAc-T and showed that these activities are catalyzed by the same enzyme. Thus, the activities for formation of GA2, GD2, and GM2 are likely to colocalize with the late Golgi markers.
LacCer synthase has been shown to represent the first activity in the lumen of the Golgi during step by step formation of cph (3955 cpm, as described under "Experimental Procedures")) and in the presence of 0.05% Triton CF54 (E) (peak activity (100%) corresponds to 600 cph (1130 cpm)) using [ 3 H]t-GlcCer as substrate. t-LacCer was characterized and quantified as described (39). Immunological quantification of DPP IV as a marker for the apical domains of liver plasma membranes (B) (48,49), rab5 as a marker for early endosomes (C) (50,51), and TGN38 as a marker for TGN (D) (52). Relative immunoactivity in % in B-D equals percentage of pixels in each individual lane relative to the pixels summarized from all lanes. E, esterase activity as a marker for ER (46), GlcNAc-P-T for cis Golgi (60), and Gal-T for trans Golgi (61) (peak activities (100%) correspond to 0.16 mol/min for esterase, 8.400 cpm for Gal-T, and 3700 cpm for GlcNAc-P-T). F, sucrose density and protein profiles. Three independent gradients were analyzed and showed highly reproducible patterns. Immunological and enzyme activity assays were performed in the same gradient. For details, see "Experimental Procedures."  (11489 and 6592 cpm, respectively, as described under "Experimental Procedures"). more complex GSLs (12,39), but it had not previously been localized to an individual Golgi subsite. The localization of LacCer synthase and SAT-I to the late Golgi came as a surprise, because SAT-I, the enzyme needed to convert LacCer into GM3, had been described as an early Golgi resident in rat liver (25,26). In addition, in primary cultured cerebellar neurons of 6-day-old mice, SAT-I has also been attributed to an early Golgi site (27). In the same system, however, SM synthase, an early Golgi enzyme (18,19), had been localized to the late Golgi (29). These alterations may be due to a maturation process within the organelles of the secretory pathway in the developing brain, similar to that described in chick embryo retina cells (69).
Two factors may account for the discrepancy between earlier results (25,26) and those described here: (i) in the earlier reports, detergent concentrations of 0.2% Triton CF54 were used for determination of SAT-I, too high to detect significant enzyme activity, as shown in a subsequent publication (41). These authors describe 0.05% of the detergent (Triton CF54) as optimal for SAT-I activity and show complete inactivation already at 0.15%. In addition, another sialyltransferase, SAT-IV, localized to the late Golgi (25,26), was reported to be less sensitive to detergent and to have a broad substrate specificity, including conversion of LacCer to GM3 (41). This raises the question why SAT-I activity (catalyzed by SAT-IV) was not found in both the early and the late Golgi in earlier studies (25,26). (ii) The use of detergent may allow access of substrates to SAT-I present in the early Golgi in transit to the late Golgi. However, the latter possibility is unlikely, because in this study, early Golgi SAT-I activity was not detectable in assays both with and without detergent. This issue cannot be resolved at present.
A variety of reports deal with the intra-Golgi localization of individual enzymes involved in GSL biosynthesis, analyzed by use of inhibitors of intracellular vesicular transport, like the antibiotics monensin (32,33) and BFA (28 -31), and by investigation of mitotic cells (16). These reports describe blocks of GSL biosynthesis in steps after GM3 and/or GD3 formation and localize to the early Golgi the enzymes acting before this block. In chick embryo retina cells, increased amounts of GM3, GD3, and GT3 were found in the presence of BFA (31). In the presence of monensin, an inhibition of GT3 formation was observed, whereas the amounts of GM3 and GD3 remained unchanged (33). The authors concluded that synthesis of GM3 and GD3 takes place in the cis/medial Golgi and that synthesis of GT3 takes place in the trans Golgi, and they postulated the existence of individual enzymes for these sialylation reactions. However, Nakayama et al. (70) found that an expression-cloned cDNA coding for GT3 synthase was identical with the cDNA for GD3 synthase and concluded that both products are formed by the same enzyme. Thus, the different results obtained with monensin and BFA may need to be explained by some direct or indirect effect on GD3/GT3 synthase of monensin, such as, for example, a shift of pH, as discussed in Ref. 33.
A Revised Model for the Functional Organization of the Golgi Apparatus in Sphingolipid Biosynthesis-In the above mentioned studies, localization of the biosynthetic activities for LacCer, GM3, GD3, and GT3 was assigned dependent on SAT-I as a reference enzyme, localized to the cis Golgi earlier (25,26). With our finding that all lumenal activities in GSL biosynthesis reside in the late Golgi, the above results would easily and consistently fit into a model as depicted in Scheme I: GlcCer is synthesized at the cytosolic sides of both the early and the late Golgi. If it is translocated into the lumen of the early Golgi, GlcCer may be carried to the late Golgi by vesicular transport, where it can serve as a substrate for LacCer synthesis. GlcCer expressed at the cell surface possibly stems from the cytosolic side of the Golgi and represents molecules that have not been translocated into the lumen but transported monomolecularly to the inner leaflet of the plasma membrane. A very recent report (71) indicates that ATP binding cassette proteins in the plasma membrane are able to translocate various GlcCer-species to the outer leaflet, thus expressing these monohexosyl lipids at the cell surface.
The blocks observed with BFA of GSL biosynthesis after LacCer, GM3, and GD3 and before GA2, GD2, and GM2 (28,29) indicate that LacCer-, GM3-, and GD3-forming activities reside in the trans Golgi part of the late Golgi, because BFA mediates fusion of Golgi membranes with the ER, including the trans Golgi (65,(72)(73)(74). In contrast, the trans Golgi network is insensitive to BFA (75,76). These data are only in conflict with a recent report (77) that ascribes GalNAc-T, an enzyme for the formation of GA2, GM2, and GD2, to various cisternae of the Golgi. For these ultrastructural investigations, however, an overexpressed myc-tagged form of the enzyme was used, and the immunopositive cisternae were not defined by corresponding markers. Thus, the variable location of the tagged enzyme might well be due either to a dose effect (overexpression was 4 -8-fold of the physiological level), and/or the myc tag might interfere with the proper localization of the enzyme. Thus, taken together, the evidence implies that GA2, GM2, and GD2 biosynthesis is likely to reside in the trans Golgi network. Furthermore, because of the broad specificity of GalNAc-T (53,68) and galactosyltransferase III acting on GA2, GM2, and GD2 (78), all steps leading to the complex gangliosides of the 0, a, b, and c series would occur in this Golgi compartment. This would also agree with data on GSL biosynthesis in mitotic cells, SCHEME. I. A revised model for the organization of sphingolipid biosynthesis within organelles of the secretory pathway in rat liver. Note that the results described in this work were obtained with rat liver Golgi and that the above model includes results obtained from a variety of mammalian cell types. The differentiation of activities between trans Golgi and TGN is based on results using BFA (28,29) and mitotic cells (16). For details, see text.
where the Golgi is dissociated into vesicles functionally identical to the stacks from which they have formed (16): analysis of the GSLs in interphase HeLa cells after incubation with [ 3 H]serine results in labeled Cer, GlcCer, LacCer, and GA2, whereas in mitotic HeLa cells, a block beyond LacCer synthesis was observed, indicating that LacCer has to move to another compartment to be converted to GA2 (Scheme I). Furthermore, localization of LacCer synthase to the late Golgi is in line with earlier results that have localized to this Golgi subsite another activity necessary for LacCer synthesis, explained as follows: the UDP-Gal translocator functions as an antiport system, in which a molecule of UMP is transferred to the cytosol for each molecule of UDP-Gal brought into the lumen (79). Lumenal conversion to UMP of the UDP resulting from the transfer of the galactose residue to an acceptor is performed by UDPpyrophosphatase, and this enzyme was localized to the trans Golgi and the TGN by electron microscopy (61,80). Absence of this hydrolase in the early Golgi again is in line with the presence of SAT-I in the late part of the organelle, because UDP is known to inhibit the activity of this sialyltransferase (81).
Thus, the starting compounds for the 0, a, b, and c series of gangliosides most likely are formed in the trans Golgi, and the following steps leading to more complex gangliosides are localized to the TGN (Scheme I).
In light of this contention, an alternative or additional vesicular step for expression of GlcCer on the cell surface cannot be excluded: GlcCer could translocate into the TGN, where it cannot be converted to higher species. A single vesicular transport step would then be sufficient for its expression.
A GSL transport assay has been described that makes use of donor Golgi membranes from a Chinese hamster ovary mutant cell line defective in CMP-NeuAc transport. Thus, LacCer accumulates in these membranes and served as a probe for intra-Golgi transport, when an acceptor Golgi membrane was used from a mutant that cannot form LacCer within its lumen because it is devoid of UDP-Gal translocator activity. Upon fusion of donor vesicles with acceptor membranes, LacCer from the donor fraction is then converted into GM3 (24). From the data outlined above, it becomes likely that this assay reflects homotypic fusion of trans Golgi membranes.
In summary, unlike in glycoprotein biosynthesis, the lumenal activities in GSL biosynthesis seem to be organized in such a way as to minimize vesicular transport steps during GSL formation. This organization might help to minimize unselective vesicular retrograde transport back to the early Golgi and to the ER of these ubiquitous plasma membrane constituents.