Biosynthesis, processing, and intracellular transport of GM2 activator protein in human epidermal keratinocytes. The lysosomal targeting of the GM2 activator is independent of a mannose-6-phosphate signal.

The processing, intracellular transport, and endocytosis of the GM2 activator protein (GM2AP), an essential cofactor of β-hexosaminidase A for the degradation of ganglioside GM2, was investigated in human epidermal keratinocytes. The GM2AP precursor is synthesized as an 18-kDa peptide, which is singly glycosylated, resulting in 22-kDa high mannose and 24-27-kDa complex glycoforms. A small portion of the 22-kDa form bears phosphomannosyl residues. About 30% of the GM2AP precursor is secreted during 12 h after synthesis, consisting almost exclusively of complex glycoforms. In a post-Golgi compartment, the intracellular remainder is converted to a 20-kDa mature form within 24 h, bearing a heavily trimmed N-glycan on a 17-kDa backbone. Interestingly, even nonglycosylated GM2AP is delivered to the lysosome, as shown by tunicamycin treatment and subcellular fractionation. Also, its endocytosis is independent of carbohydrate-linked signals and is even more effective for nonglycosylated GM2AP. We conclude that a mannose-6-phosphate-independent pathway for the lysosomal delivery of GM2AP exists in cultured human keratinocytes.

The lysosomal degradation of cellular glycosphingolipids with short oligosaccharide chains requires the presence of nonenzymatic cofactors, the glycosphingolipid activator proteins (reviewed in Refs. 1 and 2). Five different activator proteins have been discovered to date. Four of them (sphingolipid activator proteins or saposins A-D) are proteolytically generated from a single precursor (3)(4)(5). The fifth activator, the G M2 1 activator protein (G M2 AP), is the product of a separate gene (6). It is an essential cofactor to ␤-hexosaminidase A in the degradation of G M2 to G M3 (1). Its physiological significance became apparent when functional defects in G M2 AP were found to be the underlying cause for the AB variant of G M2 gangliosidosis (7). This group of lysosomal storage diseases is characterized by a pronounced accumulation of undegraded G M2 and related glycolipids in the neuronal tissue of affected patients (reviewed in Ref. 8). G M2 AP has been fully sequenced on the protein level; its cDNA and genomic structure have been characterized; its human gene has been localized to chromosome 5, and a processed pseudogene has been localized to chromosome 3 (reviewed in Ref. 1); and its mode of action is a subject of current debate (2,9).
On its biosynthesis and transport, limited information was available from an earlier study in human fibroblasts (10), but many aspects of its intracellular transport remained unclear, since the expression level of G M2 AP is extremely low in this cell type. Although the targeting of soluble lysosomal proteins usually involves the mannose-6-phosphate receptor system (reviewed in Ref. 11), it was found that recombinant, nonglycosylated G M2 AP is effectively endocytosed by AB variant fibroblasts (12). Thus, at least the endocytosis of G M2 AP can occur independently of M6P residues and this raises the question, how far its intracellular transport would depend on the M6P pathway. The existence of M6P-independent targeting mechanisms has been known for a long time. Lysosomal enzymes are hypersecreted by many cell types from I-cell disease patients, since a defect in phosphotransferase activity prevents proper phosphorylation of mannose residues in this disease. However, almost normal levels of lysosomal enzymes were found in some tissues from these patients (13), indicating compensation of the phosphotransferase defect by an alternative targeting mechanism in some cell types. M6P-independent targeting has also been described for cathepsin D in HepG2 cells (14), breast cancer cells (15), and I-cell disease lymphoblasts (16), as well as for ␤-aspartylglucosaminidase (17) and ␣-glucosidase (18) in transfected COS1 cells. For cathepsin D, this type of targeting seems to involve a transient, M6P-independent association of the precursor with intracellular membranes (15,19). Cathepsin D has been found to be associated with sphingolipid activator protein precursor during early stages of transport, and it was proposed that such an intermolecular association of lysosomal enzyme precursors might be one of the essential features of M6P-independent transport (20). It has also been suggested that the targeting signal of cathepsin D might reside in a peptide determinant in its carboxyl-terminal region (16).
To define the intracellular transport of G M2 AP with particular attention to its M6P dependence, we were in need of a physiological cell culture system allowing pulse-chase analysis with reasonable amounts of radioactive label in sufficiently short pulse times. Although no data were yet available on the biosynthesis and processing of lysosomal proteins in human epidermal keratinocytes (hEKs), we decided to choose this cell type for our study, since in hEKs, G M2 AP biosynthesis is 5-fold enhanced over human fibroblasts. 2 With the present study, we established the biosynthesis and processing of G M2 AP in hEKs and we will show that an M6Pindependent targeting pathway exists for G M2 AP in this cell type, which cannot be used by the precursors of cathepsins D and L. In this way, we also give first examples of lysosomal enzyme processing in hEK. We will also show, that G M2 AP endocytosis is independent of known signals for N-glycan receptor-mediated endocytosis in hEK.

EXPERIMENTAL PROCEDURES
Reagents-The following reagents were purchased from commercial sources: MCDB 153 medium base, keratinocyte growth medium supplements, Dulbecco's modified Eagle's medium, N-acetylalanylglutamine (Biochrom, Berlin, Germany), cysteine-and methionine-free MCDB 153 medium base (Cytogen, Lohmar, Germany), bovine pituitary extract (Beckton Dickinson, Bedford, MA), porcine insulin, epidermal growth factor, brefeldin A, tunicamycin, yeast mannan, and protein G-Sepharose fast flow (Sigma), [ 35  Cell Culture-Epidermal keratinocytes from human foreskin were obtained according to the method of Rheinwald and Green (21). They were cultured at 0.1 mM Ca 2ϩ in MCDB 153 supplemented as described (22), with N-acetyl-alanyl-glutamine substituting for glutamine. At approximately 70% confluency, the cells were split in a 1:5 ratio, resulting in seeding densities of 2 ϫ 10 4 cells/cm 2 . Passaging was repeated every 4 -6 days. In the third or fourth passage, cells from a single donor were used for experiments.
Antibodies-A goat antiserum raised against recombinant G M2 AP (23) was used for all experiments. Sheep anticathepsins L or D were commercially available (BioAss, Schwalbach, Germany); goat anti-␤hexosaminidase ␤-chain was a kind gift of Dr. R. Proia (NIDDK, National Institutes of Health, Bethesda, MD).
Metabolic Labeling-Epidermal keratinocytes grown in 60-mm dishes were kept in 1.5 ml of medium lacking methionine and cysteine for 2 h. When inhibitors were applied, they were already present during this starvation period. All inhibitors were added from 1000-fold-concentrated stock solutions in ME 2 SO. Half of the medium was then removed, and the cells were labeled with [ 35 S]cysteine (5.55 MBq; specific activity, Ͼ37 TBq/mmol) for 60 min. After the pulse period, the cells were washed twice with phosphate-buffered saline, and the chase was initiated by addition of 1.8 ml of complete medium. The media were saved, and the cells were extracted. Phosphate labeling was performed analogously, except that the labeling medium was phosphate deficient. 18.5 MBq of [ 33 P]orthophosphate (specific activity, Ͼ111 TBq/mmol) were used in a pulse of 5 h.
Endocytosis Experiments-Epidermal keratinocytes grown in 75-cm 2 flasks were starved for 2 h in the presence of 10 mM NH 4 Cl and labeled with 22.2 MBq of [ 35 S]cysteine in 4 ml of medium containing 10 mM NH 4 Cl for 8 h. Nonglycosylated G M2 AP was obtained by labeling of hEKs in the simultaneous presence of 10 mM NH 4 Cl and 5 g/ml tunicamycin. The secretions were dialyzed overnight against two 500-ml changes of MCDB 153 medium base in dialysis tubing with a 10-kDa cutoff to remove the inhibitors and were added to three 60-mm dishes of unlabeled cells in a total volume of 2 ml/dish. Endocytosis was allowed to proceed for 24 h. Afterward, the cells were extracted and immunoprecipitated as described below. Of the endocytosis media, 200 l were filled up to 700 l with cell extraction buffer (see below) and immunoprecipitated.
Preparation of Cell Extracts and Immunoprecipitation-After the pulse-chase experiment, the cells were washed twice with ice-cold phosphate-buffered saline and scraped off with 0.7 ml of ice-cold lysis buffer (phosphate-buffered saline containing 5 mM EDTA, 0.5% bovine serum albumin, 1% Nonidet P-40, 1 mM phenylmethylsulfonyl fluoride, and 1 g/ml each leupeptin and pepstatin A). After the removal of insoluble matter by a 20-min centrifugation at 17,500 ϫ g and 4°C, 10 l of the extracts were saved for acid precipitation. Immunoprecipitation was carried out as described (24), except that 3.5 l of goat anti-G M2 AP serum/ml of sample and 10 l of protein G-Sepharose/l of serum were used. Finally, the immune complexes were solubilized by boiling in 0.5% SDS/1% ␤-mercaptoethanol.
Deglycosylation of the precipitates with endo H or PNGase F was carried out according to the manufacturer's instructions. The samples were subjected to reducing SDS gel electrophoresis in the Tris/Tricine buffer system of Schä gger and von Jagow (25) using 12.5% polyacrylamide in the separating gel. Afterward, the gels were soaked in Amplify (Amersham), dried, exposed overnight to a Fuji BAS 1000 imaging system screen, and recorded with a Fuji BAS 1000 system. For a permanent record, the gels were exposed to Kodak BioMax MR x-ray film.
Magnetic Isolation of Lysosomes-The preparation of ferromagnetic dextran and the magnetic isolation of lysosomes were carried out essentially as described by Rodriguez-Paris et al. (26) with the following modifications. 3 The cells were loaded with the iron/dextran probe for 24 h and chased for at least 10 h. They were harvested by trypsinization and lysed in homogenization buffer (27) supplied with 0.5 mM 4-aminoethylbenzylsulfonyl fluoride (Calbiochem) and 1 g/ml each leupeptin and pepstatin A by 20 passages through a 24-gauge needle. Nuclei and insoluble components were removed by centrifugation (750 ϫ g, 5 min, 4°C). The supernatant was passed over a steel wool column placed in a magnetic cell-sorting device (Miltenyi Biotec, Bergisch-Gladbach, Germany). The flow-through was collected. After washing with homogenization buffer (5 ϫ 1 ml), the column was removed from the magnet for elution of lysosomes. Nondestructive elution was achieved with 5 ϫ 1 ml of homogenization buffer. Destructive elution was performed with 5 ϫ 1 ml of water. Thereafter, each fraction was assayed for the following marker enzymes as described (28): alkaline phosphatase (plasma membrane), lactate dehydrogenase (cytoplasm), NADPH-dependent cytochrome-C reductase (endoplasmic reticulum), galactosyl-transferase (Golgi apparatus), and succinate dehydrogenase (mitochondria). ␤-Hexosaminidase (lysosomes) was assayed as described (29) using (4-nitrophenyl)-␤-D-N-acetylglucosaminide as a substrate. For hEKs, it was found that 60 -80% of the preloaded lysosomes were lost into the 750 ϫ g pellet. However, losses of 40 -60% were also observed with unlabeled cells. More than 90% of total protein and all nonlysosomal marker enzyme activities were recovered from the flow-through. Only 6 Ϯ 2% of total hexosaminidase activity could be recovered from the nondestructive eluate and were contaminated by mitochondria and some plasma membrane, but sufficiently enriched lysosomal material was eluted destructively. 36 Ϯ 10% of hexosaminidase was found in this fraction, being 65 Ϯ 9-fold enriched (mean of six experiments in all cases) and being contaminated only by trace amounts of mitochondrial and plasma membrane marker activities.

RESULTS
Biosynthesis and Processing of G M2 AP in hEKs-hEKs were pulse-labeled with [ 35 S]cysteine for 1 h and chased for 2, 6, 24, and 48 h. G M2 AP was immunoprecipitated from the media (Fig.  1A, left panel) and from cell extracts (Fig. 1A, right panel). G M2 AP was not yet detectable in the medium after the pulse period, whereas an intense single band of 22 kDa was precipitated from the cell extract. After a 2-h chase, an additional band of 24 kDa appeared in the medium as well as in the cells. In the medium, the intensity of this band increased within the next 10 h but then remained constant during a chase time of 48 h. This secreted form of G M2 AP was accompanied by at least three additional weaker bands of 25, 26, and 27 kDa and by traces of the intracellular 22-kDa form and an 18-kDa form. In the cells, however, the initial band doublet of 22 and 24 kDa was replaced during 24 h by a single, broad band centered around 20 kDa after having passed through several intermediate stages of less clearly defined molecular mass. This form persisted for at least 48 h without being further processed.
Removal of the G M2 AP N-glycan by endo H or PNGase F (Fig.  1B) revealed that most of the apparent molecular mass difference between individual G M2 AP forms was due to carbohydrate processing events. The early 22-kDa band and the 24 -27-kDa forms shared the same 18-kDa peptide backbone, but the 22-kDa G M2 AP bore an endo H-sensitive carbohydrate, whereas the N-glycan of the 24 -27-kDa G M2 AP was endo H-resistant. Experiments with more closely spaced chase times (not shown) demonstrated that about 70% of the 22-kDa G M2 AP was converted to the 24 -27-kDa forms during the first 6 h of chase. At 6 h of chase, conversion of the 18-kDa peptide to a 17-kDa product was already detectable, indicating the onset of proteolytic processing on the peptide chain of G M2 AP. Processing was complete after 24 h, and no endo H-sensitive glycoforms could be detected at this time any more.
For comparison, cathepsin L was subsequently immunoprecipitated from the cells and media (not shown). It was chosen as a representative of soluble lysosomal enzymes not involved into glycosphingolipid metabolism and because it has a unique Nglycosylation site (30), just like G M2 AP. After 24 h of chase, a 42-kDa precursor was found in the medium, and two mature forms of 29 and 24.5 kDa (31) were detected intracellularly. The oligosaccharide side chains of all forms were entirely endo H-sensitive. Fig. 2 summarizes the results for G M2 AP schematically. It anticipates some of the observations given in detail later, but it is intended as a guide to the large number of G M2 AP forms and their intracellular localization. In hEKs, G M2 AP is synthesized as a precursor with a backbone size of 18 kDa after translocation into the ER and removal of the signal peptide. The precursor is cotranslationally glycosylated at its unique N-glycosylation site, yielding a 22-kDa protein with a high mannose type carbohydrate chain (precursor, high mannose (P HM )). Passing through the Golgi apparatus, a significant amount of this early precursor is converted to an endo H-resistant glycoform of 24 kDa, which most likely bears a complex type oligosaccharide (precursor, complex (P C )) and is partially secreted. Glycoforms of 25-27 kDa are generated from a minor fraction of the 24-kDa precursor immediately before its secretion. They presumably bear multiantennary glycan structures (precursor, multiantennary (P MA )), and only trace amounts of them are found inside the cell. A small portion (Յ5%) of the secretions consists of 22-kDa G M2 AP. The secretion ceases between 6 and 12 h after synthesis, when about one-third of the total amount of precursor present after a 1-h pulse is found outside the cell. Simultaneously, late processing events occur on the protein backbone as well as on the oligosaccharide of the intracellular forms, suggesting that the bulk of G M2 AP is then entering the lysosome. The intermediates of this late processing cover the entire molecular mass range between 20 and 24 kDa. After 24 h, processing has yielded a 20-kDa mature form (M) with a 17-kDa protein backbone. Its carbohydrate side chain is completely endo H-resistant and probably heavily trimmed, since the mass shift between the precursor glycoforms and M is 2 or 4 kDa, but proteolytic processing accounts for only 1 kDa of this apparent loss (compare with Fig. 1).
The half-life of mature G M2 AP must be well beyond the chosen chase times, since no significant decrease in overall signal intensity was observed even after 72 h of chase (not shown). Varying amounts (2-4%) of the total secreted material consisted of 18-kDa dP, suggesting that a part of G M2 AP had escaped N-glycosylation intracellularly, since a trace of dP was also observed within the cells.
Carbohydrate Phosphorylation of G M2 AP-hEKs were labeled with [ 33 P]phosphate to assess which glycoform of G M2 AP  The localization of individual glyco and protein forms of G M2 AP during the course of transport and processing as determined by metabolic labeling and immunoprecipitation is shown. Thick arrows, major pathways; thin arrows, minor pathways. Approximate transit or conversion times are indicated for each step. Blocking agents are given for transitions that could be inhibited. BafA1, bafilomycin A1. bears phosphomannosyl residues (Fig. 3). In the media, only a faint band of P HM appeared after the pulse period of 6 h, the label of which could completely be removed by endo H. In the cells, a single, broader band of 22-23 kDa was observed. In this case, 90% of the label was removed by endo H and the remainder by PNGase F, demonstrating that the carbohydrate rather than the backbone of G M2 AP is phosphorylated, and that this marker resides predominantly on P HM . Since the labeling effectivity seemed to be rather low, we excluded the possibility of an inefficient label incorporation by sequential immunoprecipitation of cathepsins L and D and hexosaminidase ␤-chain from the same samples. In all cases, the precursors and mature forms were easily detectable, exemplified for cathepsin L in Fig. 3, right panel. The phosphate-labeling efficiency of cathepsin L was 2.3-fold enhanced over that of G M2 AP. Since the biosynthesis level of G M2 AP was found to be 5.0 Ϯ 0.6-fold enhanced over that of cathepsin L by [ 35 S]cysteine labeling (data not shown), one may estimate that less than 10% of the G M2 AP precursor is tagged with a M6P recognition marker.
[ 35 S]Sulfate labeling of hEK revealed that a tiny fraction of P C bears oligosaccharide-linked sulfate residues (not shown), but the signal intensity was close to the detection limit even after 14 h of pulse.
Processing of G M2 AP in the Presence of Inhibitors of Early Stages of Intracellular Transport-Various inhibitors of intracellular transport were used to verify the precursor-product relationship for G M2 AP and to assign individual G M2 AP modifications to particular cellular compartments more closely. First, the G M2 AP precursor was arrested in stages of early processing by a temperature shift and brefeldin A (BFA) treatment.
After a 1-h pulse at 37°C, hEKs were chased at 14°C for 24 h, which should trap early G M2 AP forms by freezing vesicle transport between the ER and the Golgi apparatus (32) (Fig. 4,  lanes 1-4). Only a trace of P C appeared in the medium (Fig. 4, lane 1). In the cells, more than 80% of the total signal still corresponded to P HM (Fig. 4, lanes 2-4), and no mature form appeared after this chase time. This is consistent with the view, that conversion of P HM to P C is a carbohydrate processing event caused by Golgi type modifications in the oligosaccharide chain of G M2 AP. The small amounts of P C present at this point of time had not been processed to a mature form, which gave a clear indication that G M2 AP matures in a compartment distal to the Golgi apparatus. Two additional hitherto unobserved products appeared after the temperature shift, one of 19 kDa and one of 16.5 kDa (Fig. 4, lane 2). Their origin remained obscure, and they disappeared after glycosidase treatment (Fig. 4, lanes 3 and 4), thus revealing that they did not represent M and dM. Release of the temperature block after 24 h fully restored the usual processing pattern (not shown).
An unexpected result was obtained when the cells were exposed to BFA (5 g/ml), a fungal macrolide antibiotic, which leads to an immediate vesiculation of the Golgi apparatus. It blocks anterograde vesicular transport from the ER but does not interfere with retrograde membrane flow between Golgi and the ER (33). In this respect, BFA would have been predicted to exert essentially the same effects on G M2 AP transport and maturation as the 14°C shift. Fig. 4, lanes 5-14, shows that this was not quite the case. If the cells were treated with BFA during the starvation period, a 1-h pulse, and a 6-h chase, the secretion of G M2 AP was completely prevented (Fig. 4, lane  5). Within the cells, however, a band triplet of 22, 23, and 24 kDa was found (Fig. 4, lane 8) instead of the expected P HM . These forms proved to be largely endo H-resistant (Fig. 4, lane  9). After 24 h of chase in the presence of BFA, the triplet had been converted to P C (Fig. 4, lanes 10 and 11) with the usual dP backbone (Fig. 4, lane 12). This result also supported the view that the proteolytic processing events leading to the generation of the 17-kDa mature peptide must take place in a post-Golgi compartment, presumably in endosomes and/or lysosomes. No alteration of the N-glycan was observed for cathepsin L, the precursor still being in its high mannose form even after 24 h of chase in the presence of BFA (Fig. 4, lanes 17 and 18).
To assess whether the modified glycosylation pattern had any effect on G M2 AP maturation, the cells were chased in the presence of BFA for 24 h and for another 24 h in the absence of the drug. In this case, secretion was reconstituted but rose to 50% (Fig. 4, lane 7). In the cells, P C was processed to mature G M2 AP (Fig. 4, lanes 13 and 14), indicating that the complex glycoform of the G M2 AP precursor is also competent for lysosomal transport. Checking for the phosphorylation state of the BFA-induced glycoforms, we treated hEKs with BFA during a 6-h pulse with [ 33 P]phosphate and a 24-h chase. Only P HM was  5-14, 17, and 18). For lanes 7, 13, and 14 3 and 4). recovered from the cells (Fig. 4, lanes 15 and 16). Remarkably, P HM had completely escaped detection by [ 35 S]cysteine labeling under these conditions (compare with Fig. 4, lanes 10 and 11), again arguing for the view that only a minor subfraction of G M2 AP bears phosphomannosyl residues.
Processing of G M2 AP in the Presence of Inhibitors of Late Stages of Intracellular Transport and Processing-Late stages of G M2 AP transport and processing were disturbed by treating hEKs with ammonium chloride or protease inhibitors. Ammonium chloride induces hypersecretion of lysosomal enzyme precursors by increasing the intraorganellar pH of acidic compartments (34). The elevated pH probably prevents dissociation of the M6P recognition markers from M6P receptors. The occupied receptors are depleted from the Golgi apparatus, whereby newly synthesized ligands are diverted to the secretory pathway (35,36). Leupeptin was chosen as an inhibitor of thiol proteases, and pepstatin A was chosen as an inhibitor of aspartyl proteases (37).
Maturation of G M2 AP precursor was suppressed by ammonium chloride treatment of hEKs (Fig. 5, left panel). High amounts of P C were secreted into the chase media. After 6 h of chase, both P HM and P C were immunoprecipitated from the cells. Only 10% of total G M2 AP remained in the cells after 24 h of chase, which were in the P C form and not processed to mature G M2 AP. Similar results were obtained (not shown) when hEKs were exposed to bafilomycin A1, an inhibitor of the vesicular proton pump (38), or to monensin, a Na ϩ and K ϩ ionophore, which disturbs transport at a late Golgi stage (39).
In contrast, 60% of the cathepsin L precursor had been secreted after 24 h of chase in the presence of NH 4 Cl, but the remaining 40% had been converted to the 29-kDa form intracellularly (Fig. 5, right panel). The 24.5-kDa form was not generated, possibly because the autoprocessing of cathepsin L (40) is inhibited at elevated pH.
Apparently, the proper targeting of G M2 AP in hEKs is strongly dependent on acidification of the compartments responsible for late sorting steps, whereas the targeting of cathepsin L is less restricted in this respect.
Treatment of hEKs with leupeptin (0.1 mM) completely prevented maturation of the G M2 AP peptide backbone (not shown). Instead of 20-kDa mature G M2 AP, a 21-kDa protein was precipitated from the cells after 24 h, from which dP (18 kDa) was liberated by PNGase F treatment. Thus, a thiol protease is involved in the maturation of the G M2 AP precursor, and the late processing of its N-glycan may occur independently from the proteolytic step. Treating the cells with pepstatin A (0.1 mM) had no effect on G M2 AP maturation.
Transport of Nonglycosylated G M2 AP-The significance of the carbohydrate for lysosomal targeting of G M2 AP was probed by desoxymanno-nojirimycin and swainsonine treatment, respectively (not shown). These compounds are potent inhibitors of Golgi mannosidases prematurely terminating the trimming of N-glycans at high mannose stages (reviewed in Ref. 41). The maturation of G M2 AP was not affected by these drugs, but both inhibitors led to the generation of entirely endo H-sensitive glycoforms, including even secreted and mature G M2 AP. In conjunction with the results of the BFA treatment presented in Fig. 4, it therefore seemed questionable whether the carbohydrate of G M2 AP has any significance for its lysosomal delivery at all. Consequently, we suppressed the cotranslational attachment of its N-glycan by tunicamycin treatment of hEKs. Tunicamycin interferes with the biosynthesis of the dolicholpyrophosphate precursor of N-linked carbohydrate side chains, thereby globally preventing N-glycosylation (41,42).
hEKs were treated with tunicamycin (5 g/ml) during the starvation period and a 1-h pulse with [ 35 S]cysteine. They were chased for 0, 6, and 24 h in the absence of the drug and immunoprecipitated for G M2 AP (Fig. 6). Qualitatively, the same G M2 AP band pattern was obtained as for deglycosylated samples from pulse-chase experiments conducted without tunicamycin (compare with Fig. 1B). dP was precipitated from the media, the intensity of which increased over 24 h of chase. In the cells, dP appeared after the pulse, showing first signs of proteolytic processing after 6 h of chase and having been completely converted to dM after 24 h. Tunicamycin treatment seemed to have no adverse effect on G M2 AP synthesis and stability. Overall protein synthesis dropped to 70% as judged by trichloroacetic acid precipitation of the lysates, but the immunoprecipitable proportion of G M2 AP remained constant under these conditions. The combined intensity of dP and dM did not significantly decrease over the chase period. The kinetics of G M2 AP transport also seemed to be unaffected compared with the normal situation. However, two additional bands (16 and 14 kDa) appeared, which had not been observed in the absence of the inhibitor and which were not subjected to further proteolytic processing. Again, the secretion of the G M2 AP precursor rose to 50% instead of the normal 30%, which had also been observed after release of the BFA block. Quite another situation was met with cathepsin L (Fig. 6,  right four lanes). Here, a nonglycosylated precursor of 38 kDa was synthesized and completely excreted into the medium. Similar results were obtained for cathepsin D (not shown).
This experiment indicated that an alternative pathway for lysosomal targeting exists in hEKs, which allows effective delivery of G M2 AP to the lysosome even when no N-glycan is attached to its protein backbone. This mechanism obviously does not operate on the precursors of cathepsins L and D.
In an attempt to determine whether the transport of G M2 AP or cathepsin L is a membrane-associated process, we tried to recover their precursors from the cellular scaffold of hEKs differentially permeabilized with saponin according to the method of Rijnboutt et al. (19). However, G M2 AP as well as cathepsin L could only be detected in the saponin supernatant, not in the membrane pellet (not shown).
Lysosomal Localization of Mature G M2 AP by Magnetic Isolation of hEK Lysosomes-To prove the lysosomal localization of processed G M2 AP, hEK lysosomes were isolated by magnetic fractionation.
For cold experiments, hEKs were preloaded for 24 h with a probe of colloidal iron coupled to dextran, which was ingested by nonspecific endocytosis. After a chase of at least 10 h, the cells were lysed, a 750 ϫ g supernatant was subjected to magnetic separation of lysosomes, and the fractions were assayed for the distribution of subcellular marker enzyme activities. The bulk of activity of the marker enzymes for cytoplasm, plasma membrane, mitochondrial membrane, ER, and Golgi was recovered from the flow-through. About 54% of total hexosaminidase activity was also eluted here, most probably having been liberated from disrupted lysosomes. Hexosaminidase was enriched 65 Ϯ 9-fold in the fractions eluted destructively and depleted 2.5 Ϯ 0.5-fold from the flow-through.
For the localization of nonglycosylated G M2 AP, hEKs were pulse-labeled with [ 35 S]cysteine for 5 h in the presence or absence of tunicamycin. They were chased for 14 h in the presence of the magnetic probe and for another 10 h in its absence. Fig. 7 shows that mature G M2 AP could be immunoprecipitated from the flow-through and, more significantly, from the lysosomal fraction of the magnetic separation, regardless of whether it was glycosylated or not. G M2 AP was 2-fold depleted from the flow-through and about 45-fold enriched in the destructive eluates from these separations, which essentially parallels the hexosaminidase enrichment given above.
In this manner, we showed that in hEKs, G M2 AP is properly targeted to the lysosome even if no N-glycan is cotranslationally attached to its unique N-glycosylation site.
Endocytosis of G M2 AP-We examined whether the carbohydrate would be of any relevance for the uptake of exogenously added G M2 AP by offering the [ 35 S]cysteine-labeled secretions of ammonium chloride-treated hEK to unlabeled cells for a 24-h period in the presence of M6P or glucose-6-phosphate (10 mM each) or yeast mannan or asialo-orosomucoid (1 mg/ml each) or a combination of M6P, mannan, and asialo-orosomucoid at the above concentrations, respectively. These inhibitors were chosen to interfere with endocytosis by the receptors for M6P, exposed mannoses, or asialoglycoproteins (43). The cells and 10% of their media were sequentially immunoprecipitated for G M2 AP, cathepsin L, and hexosaminidase ␤-chain (Fig. 8).
Three additional G M2 AP bands appeared in the medium besides P C and P MA , which were assigned to P HM , dP, and an unknown component according to their molecular masses. The endocytosis of G M2 AP was not significantly affected by any of the inhibitors (Fig. 8, left panel). Under M6P, glucose-6-phosphate, mannan, and asialo-orosomucoid alone, the uptake was 95-120% of the untreated control. It dropped to 75% in the presence of the inhibitor mixture, which we believe to be a nonspecific effect due to overloading of the serum-free medium with unusual additives, since none of the inhibitors had an adverse effect when applied alone. The endocytosis of cathepsin L was only slightly affected by M6P, whereas the uptake of hexosaminidase ␤-chain was almost completely inhibited in the presence of M6P (Fig. 8, right panel). The relative molar concentrations offered in the media were calculated to be 4.4:1:1.1 for G M2 AP, cathepsin L, and hexosaminidase ␤-chain, respec- tively. After 24 h of endocytosis in the absence of any inhibitor, 4.1 Ϯ 0.5% of the added material were recovered from the cells for G M2 AP, 5.0 Ϯ 0.5% for cathepsin L, and 12.5 Ϯ 1.5% for hexosaminidase ␤-chain, respectively (mean of two experiments). These variations may reflect either the individual rates of uptake and/or different rates of degradation.
A second endocytosis experiment was conducted as above, except that the secretions were collected from cells that had been labeled in the additional presence of tunicamycin and that no inhibitor was present during the endocytosis phase (not shown). Under these conditions, G M2 AP was the only one of the four referenced antigens (G M2 AP, cathepsins L and D, and hexosaminidase ␤-chain) that was endocytosed and processed effectively. Surprisingly, even 8.5 Ϯ 1% (mean of two experiments) of the offered G M2 AP was recovered from the cells after 24 h in this case.
Both findings indicate that endocytosis of G M2 AP is largely independent of the known N-glycan-linked signals for receptormediated endocytosis in hEKs.

DISCUSSION
In this article, we have clearly established that in hEKs, G M2 AP is synthesized as a 22-kDa precursor (P HM ) bearing a single high mannose N-linked oligosaccharide chain on a peptide backbone of 18 kDa. In the Golgi apparatus, at least 70% of P HM is converted to 24-kDa P C by remodeling of the N-glycan to a complex type oligosaccharide. About one-third of the G M2 AP precursor is secreted, and more than 90% of these secretory forms consist of P C . The intracellular remainder is segregated from the secretory pathway and processed in a post-Golgi compartment to yield a 20-kDa mature form (M), which bears an endo H-resistant N-glycan on a 17-kDa peptide (summarized in Fig. 2).
Thus, the early and late processing of G M2 AP shows many features typical of soluble lysosomal enzymes (reviewed in Ref. 44). It is synthesized as a precursor protein, the N-glycan of which receives Golgi type modifications. It is partially secreted into the medium, and it becomes subject to proteolytic and glycolytic processing events in acidic compartments after segregation. The molecular mass of the major secreted form (P C , 24 kDa) was in agreement with earlier findings (10). Regarding the increased half-width of the mature G M2 AP band, one may assume that M is microheterogeneous in its N-glycan and/or its protein backbone. It should be noted in this context that two protein species were also present in M purified from human kidney: one major form with Phe 34 and a minor one with Ser 32 as the amino terminus (45). The molecular mass found for dM (17 kDa by SDS-PAGE) is close to its calculated molecular mass from cDNA and protein sequencing data (17,369 Da starting from Phe 34 ). Since dP has an apparent molecular mass of 18 kDa, it is very likely that the signal peptide of G M2 AP is removed on entry into the ER. The algorithm of von Heijne (46) predicts signal peptidase cleavage sites at Gln 22 or His 24 , leading to precursors of 18,561 or 18,362 Da, respectively.
Very recently, a report by Wu et al. (47) proposed that an alternatively spliced form of G M2 AP (Met 1 -Glu 142 plus Val-Ser-Thr, termed G M2 AP(A) below) might exist in human placenta and fibroblasts, which in vitro has an activating effect on clostridial sialidase but not on hexosaminidase. Taking into account the expected decrease in molecular mass of such a splice variant (Ϫ5 kDa for all forms), we rescreened our fluorographs for the presence of G M2 AP(A). At least for hEKs, we were not able to detect G M2 AP-related products in the relevant molecular mass window. Shortened G M2 AP forms appeared only when inhibitors of early transport (see Fig. 4, lane 2) or tunicamycin (see Fig. 6) were used. However, they were not processed in the same manner as full-size G M2 AP, and we assume that they represent prematurely and aberrantly processed forms arising from the presence of the inhibitors.
P C first appears in the medium 60 -90 min after synthesis, a value that is typical for other lysosomal enzymes too (44). In hEKs, however, its secretion does not cease before 12 h after synthesis, and its final maturation takes more than 12 h. Similar kinetics were observed for cathepsin L. It seems that segregation and transport of at least these two lysosomal proteins is somewhat delayed between late Golgi and lysosomal compartments. To our knowledge, this is the first study dealing with the processing of lysosomal proteins in hEKs. Since the maturation of lysosomal protein precursors occurs in ranges from minutes to days depending on cell type and species (44), we attribute these slowed-down kinetics to a yet uncharacterized effect specific to hEK.
Furthermore, we have shown that G M2 AP is targeted to the lysosomes by a M6P-independent pathway in hEKs. After Klima et al. (12) had found that recombinant, nonglycosylated G M2 AP is effectively endocytosed by human fibroblasts, the question arose whether its intracellular transport would also be at least partially independent of M6P or other carbohydratelinked signals. Several different lines of evidence demonstrated that such a pathway must exist in hEKs.
First, we have shown that only a minor fraction of G M2 AP, possibly 10%, bears phosphomannosyl residues, and that this tag is confined to P HM exclusively. Since at least 70% of P HM s are converted to P C s and since 70% of G M2 AP precursors are retained intracellularly, a significant proportion of M should be derived from P C . The endo H resistance of M also suggests that it may be derived mainly from a precursor with complex type oligosaccharides. It has to be kept in mind, however, that extensive trimming of mannose-rich N-glycans may abolish recognition by endo H (44). In addition, we could trap P C exclusively by BFA treatment and observed that it is correctly processed to M on detoxification of the cells (Fig. 4, lanes 13 and  14). Under BFA, the oligosaccharide of P HM was remodeled almost instantaneously by the action of Golgi glycosidases and glycosyltransferases, which retain at least part of their activity in the fused ER-Golgi compartment (49). It may well be that the underphosphorylated N-glycan of G M2 AP is a preferred substrate for such modifications, since M6P serves as a protecting group against Golgi glycosidase trimming (44,48,50). Indeed, the efficiently phosphorylated carbohydrate of cathepsin L was under no circumstances converted to a complex form in hEKs.
Second, only P C is hypersecreted on NH 4 Cl treatment. For enzymes sorted via M6P receptors, NH 4 Cl interferes with sorting in a post-Golgi compartment, and the secretions are highly enriched in phosphorylated precursor glycoforms (35). Therefore, it is likely that P C passes a similar base-sensitive compartment, and the high selectivity by which NH 4 Cl induces its hypersecretion and not that of P HM suggests that it is mainly P C that moves through this compartment. P C would then be the dominant precursor to M, implying that the M6P receptor system possibly plays only a minor role in the segregation of G M2 AP, since P C does not bear phosphomannosyl residues.
Finally, and most strikingly, we found that G M2 AP was correctly targeted at an apparently normal rate even when its glycosylation had been suppressed by tunicamycin treatment (Figs. 6 and 7), suggesting that the lysosomal targeting signal resides on its protein backbone rather than on its carbohydrate. Interestingly, the secretion of the G M2 AP precursor rose from 30 to 50% under tunicamycin. This difference is close to the 10% value estimated to be the mannose-6-phosphorylated fraction of G M2 AP. Usually, tunicamycin treatment strongly enhances the secretion of lysosomal enzymes transported in re-sponse to the M6P system (51)(52)(53), which then are shed from the cells as unprocessed precursors (52,53). In this respect, the nonglycosylated precursors of cathepsins L and D behaved as expected.
Thus, the intracellular transport of G M2 AP adds to an increasing number of reports about correct targeting of lysosomal enzymes independent of phosphomannosyl residues. Such phenomena have been described for cathepsin D in HepG2 cells (14,19), in breast cancer cells (15), and in ICD lymphoblasts (16), as well as for ␤-aspartylglucosaminidase (17) and ␣-glucosidase (18) in transfected COS1 cells. In the cases of cathepsin D (14,15,19,54), ␣-glucosidase (18), glucocerebrosidase and sulfatide activator protein (19), and cathepsin L (55), a M6P-independent membrane association of the precursors was observed. We could not demonstrate a membrane association of the G M2 AP precursor by saponin permeabilization, but we cannot exclude a weak interaction, which may be disrupted by this treatment. For ␣-glucosidase, the membrane association is mediated by an uncleaved signal peptide (18). However, our data indicate that the signal peptide of G M2 AP is instantaneously removed on entry into the ER.
Since the function of G M2 AP is to serve as a lipid-binding protein (1), and since G M2 AP is able to transfer gangliosides from donor to acceptor liposomes in vitro (56), an intriguing option would be that it is transported and segregated in direct association to the lipid bilayer. Such an interaction has also been proposed for glucocerebrosidase, which does not bear M6P residues (57) and is possibly bound to acidic phospholipids (58).
Comparing the transport of G M2 AP and cathepsin L in hEKs, it seems that two different targeting mechanisms operate on these proteins. The efficient segregation of cathepsin L is crucially dependent on the presence of an N-glycan, which is not the case for G M2 AP. On the other hand, acidification of the transporting machinery is essential for the delivery of G M2 AP, whereas it is less substantial to the targeting of cathepsin L. In this respect, each of both proteins displays one feature typical of M6P-dependent and one typical of M6P-independent transport. Obviously, the unknown recognition system responsible for G M2 AP segregation shows a similar pH dependence as the M6P-receptor system. The transport of cathepsin L might proceed in a manner analogous to myelocytes and macrophages, in which base-sensitive and base-insensitive packaging mechanisms are known to exist in parallel (reviewed in Ref. 44).
The endocytosis of G M2 AP is also independent of the known signals for carbohydrate-mediated internalization (Fig. 8). We had expected that M6P would not affect its uptake, since more than 90% of the offered secretions are made up of P C and the P MA , which are not phosphorylated. However, its endocytosis remained equally unaffected by inhibitors of mannose and asialoglycoprotein receptors. Nonglycosylated G M2 AP is also internalized, as had already been shown earlier for human fibroblasts (12), but we were surprised to find that its uptake seemed almost twice as effective as for the glycosylated form. A similar observation has been made for the sphingolipid activator protein precursor in fibroblasts (24), and it has been suggested that the carbohydrate might partially mask an epitope involved in recognition between this protein and a yet unknown receptor system on the cell surface. It may again be speculated that G M2 AP reaches the interior of the cell by binding to lipid components of the cell surface.
In the granular layer of the epidermis, keratinocytes start to form acidic lamellar granules, which contain precursors of specialized stratum corneum lipids, lysosomal hydrolases, and a vesicular ATP-dependent proton pump (reviewed in Refs. 59 and 60). The contents of these storage bodies is eventually shed into the uppermost, cornified layer of the epithelium. The pres-ent work was performed with basal keratinocytes, in which the basal characteristics were maintained by culture at low calcium concentrations (0.1 mM), suppressing the onset of terminal differentiation. Further experiments, however, may address the question of whether G M2 AP is also targeted to the lamellar granules in differentiated keratinocytes and whether this targeting involves the same lysosomal type delivery mechanism as observed in basal keratinocytes, or whether a separate routing strategy is used. Additionally, future work may reveal the nature of the mechanisms responsible for the M6Pindependent intracellular targeting and uptake of G M2 AP and whether both of these processes are mediated by the same or different targeting systems.