Topological Analysis of Niemann-Pick C1 Protein Reveals That the Membrane Orientation of the Putative Sterol-sensing Domain Is Identical to Those of 3-Hydroxy-3-methylglutaryl-CoA Reductase and Sterol Regulatory Element Binding Protein Cleavage-activating Protein*

The Niemann-Pick C1 (NPC1) protein is predicted to be a polytopic glycoprotein, and it contains a region with extensive homology to the sterol-sensing domains (SSD) of 3-hydroxy-3-methylglutaryl-coenzyme A reductase (HMG-R) and sterol regulatory element binding protein cleavage-activating protein (SCAP). To aid the functional characterization of NPC1, a model of NPC1 topology was evaluated by expression of epitope-tagged NPC1 proteins and investigation of epitope accessibility in selectively permeabilized cells. These results were further confirmed by expression of NPC1 and identification of glycosylated domains that are located in the lumen of the endoplasmic reticulum. Our data indicate that this glycoprotein contains 13 transmembrane domains, 3 large and 4 small luminal loops, 6 small cytoplasmic loops, and a cytoplasmic tail. Furthermore, our data show that the putative SSD of NPC1 is oriented in the same manner as those of HMG-R and SCAP, providing strong evidence that this domain is functionally important.

Niemann-Pick type C (NPC) 1 is a rare autosomal recessive disorder that results in progressive neurodegeneration and hepatosplenomegaly, leading to death during early childhood (1). The most prominent biochemical feature is the accumulation of low density lipoprotein-derived unesterified cholesterol in lysosomes (2,3). In addition, cholesterol accumulates in the trans-Golgi network (TGN), and its relocation to and from the plasma membrane is delayed (1). In fibroblasts, the defect in cholesterol exit from lysosomes is accompanied by an attenuation in the down-regulation of two key components of choles-terol homeostasis, 3-hydroxy-3-methylglutaryl-coenzyme A reductase (HMG-R) and the low density lipoprotein receptor (4,5). Recently NPC1, the gene for the major linkage and complementation group, which maps to chromosome 18q11-12, was identified. It encodes an approximately 4.5-kilobase mRNA (6) that is predicted to produce a 1278-amino acid protein. At least 36 mutations have been described in NPC1 patients (6 -10), and although most are found throughout the gene and do not reveal any functionally critical protein domains, a small cluster are located in a the carboxyl-terminal third of the protein in a region that possesses conserved cysteine residues (9). A minor group of NPC patients does not map to the NPC1 locus, and complementation studies (11,12) have shown that these patients are defective for a second gene, NPC2, which is currently unknown.
The effects of reduced or absent NPC1 in both humans (4, 13) and animal models (1,3) have been well studied, establishing this gene as an essential component of intracellular cholesterol trafficking. In addition, several Chinese hamster ovary cell mutants that display the NPC1 phenotype have been characterized (14) and most have been shown to be defective for the NPC1 gene (15). However, the precise function(s) of the NPC1 protein is still unknown. NPC1 has been shown to be a membrane glycoprotein that localizes to Lamp-positive organelles, presumably endosomes and lysosomes (16 -18). Further studies have demonstrated that NPC1 resides primarily in late endosomes and only secondarily in lysosomes and the TGN (18). This is an important distinction in view of recent data suggesting that cholesterol accumulation in NPC cells occurs primarily in endosomes (19), which are sorting sites for various cellular components. In addition, NPC cells appear to be defective in their efflux of endocytosed sucrose and in the sorting of the mannose-6 phosphate receptor, suggesting that the retrograde movement of proteins and cargo from late endosomes to the TGN is perturbed (16,19).
Interestingly, although NPC1 is not closely related to any known proteins, a cluster of five potential membrane-spanning sequences, from residues 616 to 797, is homologous to the sterol-sensing domain (SSD) found in two key regulators of cholesterol homeostasis, HMG-R and sterol regulatory element binding protein cleavage-activating protein (SCAP). In SCAP, a single point mutation in its SSD is sufficient to render the protein sterol-insensitive, indicating the importance of this region (20,21), whereas in HMG-R the SSD mediates sterolregulated degradation of the protein (22,23). The transmembrane (TM) regions of NPC1, including the putative SSD, also share homology with regions of Patched (PTC), a receptor for the cholesterol-activated protein sonic hedgehog (24,25). Together these data support the view that the SSD is functionally significant.
In these studies we describe and validate a topological model for the structure of NPC1, showing that the orientation of the putative SSD is identical to those of HMG-R and SCAP. These data should support efforts in elucidating the precise function of this large membrane glycoprotein.

EXPERIMENTAL PROCEDURES
Materials-Dulbecco's modified Eagle's medium was purchased from Mediatech (Herndon, VA) and fetal bovine serum was from HyClone (Logan, UT). LipofectAMINE, L-glutamine, gentamicin, colcemid, and trypsin/EDTA were obtained from Life Technologies, Inc. Fluoromount-G was purchased from Southern Biotechnology Associates (Birmingham, AL). All enzymes were obtained from New England BioLabs (Beverly, MA), except for N-glycosidase F, which was from Roche Molecular Biochemicals (Indianapolis, IN). The goat anti-mouse IgG-fluorescein isothiocyanate (FITC) and goat anti-rabbit IgG-FITC were also from Roche Molecular Biochemicals, the anti-FLAG M2 monoclonal antibody was from Eastman Kodak Co. and the goat anti-mouse IgGrhodamine-X and anti-rabbit IgG-rhodamine-X were from Jackson Immunoresearch Laboratories Inc. (West Grove, PA). Filipin was purchased from Polysciences, Inc. (Warrington, PA). Oligonucleotides were synthesized using phosphoramidite chemistry on a 380B DNA synthesizer (Applied Biosystems, Foster City, CA). All other reagents were obtained from Sigma-Aldrich.
Creation of FLAG-tagged NPC1 Plasmids-All plasmids were constructed using the wild-type (wt) human NPC1 cDNA (6). For analysis of the full-length NPC1 protein, hydrophilic loops A to E were tagged with the FLAG peptide sequence DYKDDDDK by inserting an adapter oligonucleotide at a unique restriction endonuclease site in the wt NPC1 cDNA. For loops A to E, these sites were NdeI, ClaI, BsiWI, BclI, and Bsu36I and resulted in the insertion of the following sequences at the indicated amino acid positions: 36-FLAG-Y, 306-FLAG-I, 519-FLAG-VR, 554-FLAG-DQ, and 992-FLAG-DQ, respectively (Fig. 1B). A second loop E tag construct was created by placing the FLAG sequence at the SphI site resulting in the sequence 886-FLAG-HA. Each tag was verified by sequencing prior to cloning into the expression vector pAsc9.
For analysis of the glycosylation pattern in NPC1, a number of constructs were created. Truncated NPC1 cDNAs were generated by polymerase chain reaction using the wt NPC1 cDNA as a template and then cloned using standard procedures (26). Each cDNA extended from the start codon of NPC1 to the 3Ј-end of loops A to E (Fig. 2), followed by at least one putative membrane-spanning domain for correct protein anchorage and a BamHI site for cloning. All cDNAs were verified by sequencing on an automated ABI prism 377 DNA sequencer (Perkin-Elmer). Each truncated cDNA was cloned into the expression vector pAsc9 (18) at the BamHI site, adjacent to a 3Ј in-frame FLAG-tag linker sequence (DYKDDDDK-stop) for convenient detection. The plasmid constructs were named according to the last loop and TM domain in their cDNA sequences as follows: pLaTM1, pLbTM2, pLcTM3, pLdTM4, pLdTM6, and pLeTM13; and these encoded amino acids 1-296, 375, 549, 644, 727, and 1164, respectively (Fig. 2). An additional slightly truncated, FLAG-tagged protein was produced, as above, in which the carboxyl terminus tail after TM15 was completely replaced with a FLAG tag. This plasmid, pTM15tail, encoded NPC1 amino acids 1-1252.
Immunofluorescence Analysis of NPC1 and Filipin Staining-To assess the ability of the full-length FLAG-tagged NPC1 proteins to correct the NPC1 phenotype, human NPC1 fibroblasts (GM03123), obtained from Coriell cell repositories (Camden, NJ), were grown and transfected with the appropriate constructs using LipofectAMINE Plus reagent as described previously (18). At 24 h post-transfection, the medium was replaced and the cells were grown for a further 24 h in Dulbecco's modified Eagle's medium that contained 5% lipoprotein-deficient serum. The cells were fixed with 3% paraformaldehyde in phosphatebuffered saline (PBS) for 30 min, washed with PBS, and then quenched with blocking solution (3% bovine serum albumin (BSA), 50 g/ml filipin in PBS) containing 1.5 mg/ml glycine for 15 min. Cells were costained for cholesterol and the epitope-tagged NPC1 proteins by incubating with blocking solution that contained 2 g/ml M2 anti-FLAG antibody for 45 min, washing 3 ϫ 5 min with 50 g/ml Filipin in PBS, and then incubating with blocking solution that contained 2 g/ml anti-mouse IgG-rhodamine-X for 45 min. After further washing the cells were mounted and visualized as above.
Selective Permeabilization of Plasma Membranes and Immunofluorescence Microscopy-The expression and epitope accessibility of all FLAG-tagged proteins was assessed by transient transfection followed by selective permeabilization and immunofluorescence microscopy. COS-7 cells, obtained from the American Type Culture Collection (Manassas, VA) and grown as described (18), were treated with colcemid for 18 h prior to electroporation using a Bio-Rad gene pulser unit set at 500 microfarads and 250 V to transfect 10 g of plasmid DNA into approximately 10 6 cells. The cells were grown on glass coverslips that had been treated with 0.1 mg/ml poly-L-lysine to reduce cell loss during subsequent selective permeabilization steps. At 24 h post-transfection, the cells were washed twice with PBS. For complete permeabilization, cells were fixed in methanol for 6 min at 4°C and blocked with 1% BSA in PBS. Proteins were detected by incubation with 5.2 g/ml M2 anti-FLAG antibody and 1.4 g/ml rabbit anti-␣-galactosidase A (␣-Gal A) polyclonal antibody (27) in blocking solution for 45 min, followed by a wash with PBS, and then incubation with 2 g/ml anti-mouse IgG-FITC and 2 g/ml anti-rabbit IgG-rhodamine-X secondary antibodies in blocking solution. After further washing, the cells were mounted in Fluoromount G and visualized using a Nikon Eclipse fluorescence microscope equipped with a charge-coupled device (CCD) camera (Nikon, Melville, NY).
To selectively permeabilize the plasma membrane, cells were incubated at 4°C for 45 min in sucrose buffer (1% BSA, 0.3 M sucrose, 0.1 M KCl, 2.5 mM MgCl 2 , 1 mM EDTA, 10 mM HEPES, pH 7.4) containing 0, 1, 1.5, or 2 g/ml digitonin and with the same concentrations of M2 FLAG and ␣-Gal A antibodies used for complete permeabilization. The cells were then fixed with methanol and incubated in blocking solution for 30 min before addition of secondary antibodies and visualization as above.
Analysis of Truncated Glycosylated Proteins-For expression of various NPC1 constructs, COS-7 cells were grown in 10-cm diameter dishes and transfected using 4 g of DNA, 20 l of LipofectAMINE, and 30 l of Plus reagent, according to the manufacturer's recommendations. The transiently expressed proteins were analyzed by immunoprecipitation and SDS-polyacrylamide gel electrophoresis (PAGE), essentially as described previously (27). Briefly, at 36 h post-transfection, the cells were metabolically labeled with [ 35 S]methionine for 16 h. Cells were harvested using PBS, 2 mM EDTA and centrifuged at 800 ϫ g for 10 min. The cell pellet was resuspended in buffer (150 mM NaCl, 50 mM Tris, pH 8.0, 2 mM EDTA, 10 g/ml leupeptin, 5 g/ml pepstatin A, 25 g/ml N-acetyl-leucinal-leucinal-norleucinal, 1 M/ml phenylmethylsulfonyl fluoride, and 0.1 mM Pefabloc), disrupted by sonication using a Branson sonifier (Smith-Klein, Philadelphia, PA), and centrifuged at 22,000 ϫ g for 10 min. The resultant cell membrane pellet was resuspended in 0.5 ml of 0.5 M NaCl, 50 mM Tris, 2 mM EDTA, 1% Igepal, supplemented with the above proteinase inhibitors, using 10 passages through a 22-gauge needle, prior to rocking for 15 min and centrifugation at 22,000 ϫ g for 10 min. The solubilized membranes were incubated for 1 h with 2.8 g/ml of anti-FLAG M2, followed by a 45-min incubation with protein G-agarose beads (Roche Molecular Biochemicals). The immunoprecipitates were washed with buffer (10 mM Tris, pH 7.5, 150 mM NaCl, 0.1% Igepal, 1 mM EDTA, 0.25% gelatin, 0.2% NaN 3 ), resuspended in SDS-gel loading buffer, heated for 10 min at 65°C, and subjected to SDS-PAGE analysis (28). N-Glycosidase F digestions were carried out as described previously (27).

RESULTS
Predicted Membrane Topology of NPC1-The predicted NPC1 protein sequence was analyzed to identify potential transmembrane (TM) or hydrophilic domains. Using the TM-pred algorithm (29) at EMBnet-CH, the TM domains and topology of the hydrophilic loops of NPC1 were evaluated. The NPC1 protein is predicted to have 15 TM regions (Fig. 1A). The assignment of predicted TM regions 4 -8 is based on their relative hydrophobicity and their high degree of similarity to the sterol-sensing domains of SCAP and HMG-R (30, 31) (Fig.  1A, SSD). Based on these observations and computer analyses, the predicted model for NPC1 topology is shown in Fig. 1B and consists of 15 TM domains, 5 hydrophilic loops ranging in size from ϳ50 amino acids (loop B) to ϳ265 amino acids (loop A), and a 27-amino acid cytoplasmic tail containing a potential endosomal/lysosomal dileucine targeting motif. In addition, the model predicts that the amino terminus of NPC1 (loop A) remains in the lumen upon entering the endoplasmic reticulum (ER).
Characterization of Full-length Epitope-tagged NPC1 Pro- teins-To validate the locations of the major hydrophilic loops A to E in the predicted model (Fig. 1B), a series of epitopetagged, full-length NPC1 proteins was constructed and analyzed. These epitope-tagged NPC1 proteins were expressed in human NPC1 fibroblasts to determine whether the addition of the tags interfered with their correct processing, transport, and/or function. Transfected cells were costained with filipin and the anti-FLAG antibody to determine whether cells expressing the FLAG-tagged NPC1 were corrected for the NPC1 phenotype, thus producing filipin-negative cells. 50 -100 NPC1-positive cells were scored for correction. For all constructs ϳ60 -90% of positive cells appeared to be corrected, with the exception of the loop C construct for which about 22% of cells were corrected. Representative fields for each construct are shown in Fig. 3; human NPC1 cells positive for NPC1 expression (left panels; arrows) show a dramatic decrease in filipin staining (right panels; arrows) compared with neighboring cells, which are negative for NPC1 expression. Only one construct, which contained the tag in loop C (Fig. 3C), did not appear to complement the NPC1 phenotype. Upon careful inspection it is clear that NPC1 containing the loop C tag does not colocalize with the stored cholesterol and thus is unable to mobilize cholesterol, similar to results obtained by mutagenesis of the NPC1 tail (32). These results indicate that the FLAGtagged NPC1 proteins are functional and, with the exception of the loop C construct, were correctly targeted to the endosomal/ lysosomal system, the site of cholesterol accumulation in NPC1 cells. Thus, these epitope-tagged NPC1 proteins remained functionally active and could be used to determine the membrane topology of NPC1.
Epitope Mapping by Selective Permeabilization and Immunofluorescence-The full-length FLAG-tagged NPC1 constructs were transiently expressed in COS-7 cells following electroporation. Because COS-7 cells are easy to transfect and maintain, the majority of experiments were carried out using these cells. Initial studies showed no differences in the processing, distribution, and functional integrity of the expressed proteins, indicating that NPC1 is topologically identical whether expressed by human or simian cells (data not shown). To determine the positions of the hydrophilic domains, selective membrane permeabilization and immunofluorescence microscopy were used. At 24 h post-transfection, the cells were subjected to immunofluorescence analysis under one of three conditions: no permeabilization (absence of detergent); selective permeabilization of the plasma membrane using low concentrations of digitonin; or full permeabilization by methanol fixation. All NPC1 proteins were detected with the M2 FLAG antibody and a fluorescein-conjugated secondary antibody. As a control for permeabilization, the intralysosomal protein ␣-Gal A was detected using a specific polyclonal antibody and a rhodamine-conjugated secondary antibody. Pilot studies indi- Fig. 4. Immunolocalization of FLAG epitopes in NPC1 proteins of selectively permeabilized cells. The FLAG-tagged, full-length NPC1 proteins were transiently expressed in COS-7 cells. Proteins were detected by immunofluorescence analysis in nonpermeabilized (Non), completely permeabilized (Complete), and selectively permeabilized (Digitonin) cells to determine the accessibility of the FLAG antibody to the FLAG tags. NPC1 and the control protein were colocalized using the FLAG antibody and ␣-Gal A antibody, respectively. Proteins tagged at loops A, C, and D (panels A, C, and D, respectively) were undetected in selectively permeabilized cells, indicating a luminal location, whereas the loop B and tail protein was detected (panels B and G, respectively), showing that they were cytosolic. For loop E two different tagged NPC1 proteins were expressed (panels E and F), and both showed a luminal location. The first was observed at the plasma membrane surface of nonpermeabilized cells, whereas the second protein was undetectable in permeabilized cells. cated that concentrations of digitonin below 3 g/ml specifically permeabilized the plasma membrane but not intracellular membranes, as demonstrated by the inability of the ␣-Gal A antibody to detect its target protein within lysosomes (not shown). Under these conditions, cytosolic protein epitopes were detected while luminal ones, including the ␣-Gal A control protein, were negative. Nonpermeabilized cells were used as a negative control while transfection efficiency and locations of expressed proteins were analyzed in fully permeabilized cells. Fig. 4 shows the results of immunofluorescence detection of proteins from the FLAG-tagged constructs. Representative cells are shown for each condition. In fully permeabilized cells, a bright, vesicular staining pattern for the intraorganelle marker ␣-Gal A was observed, indicating complete antibody access. However, in digitonin-permeabilized cells there was no ␣-Gal A staining, confirming that permeabilization was selective for the plasma membrane. The six full-length NPC1 proteins, containing FLAG tags in domains A-E and the cytoplasmic tail, were analyzed (Fig. 4). Detection of the loop A-tagged NPC1 protein in the fully but not the partially permeabilized cells indicated that loop A was luminal. Similarly, detection of the loop B-tagged protein in digitonin-permeabilized cells, accompanied by a negative ␣-Gal A signal, indicated that this epitope, and thus loop B, was cytosolic (Fig. 4B). In contrast, the FLAG tags in loops C and D were both undetectable in digitonin-treated cells, suggesting their luminal location (Fig.  4, C and D). These results indicated that the orientation of loops D and E was reversed with respect to our predicted model, implying that the predicted TM3 domain does not span the membrane. This suggests that loops C and D comprise a single luminal loop and loop E is shifted from a cytoplasmic position to the luminal one. The results for loop E confirmed its luminal location, assuming elimination of TM3. However, its distribution was surprising. This protein was detected in large quantities at the plasma membrane of nonpermeabilized cells (Fig. 4E). Interestingly, the membrane staining appeared speckled and uneven, perhaps indicating that some form of aggregation occurred at the cell surface. However, this phenomenon does not apparently interfere with NPC1 function, because the protein successfully reversed the cholesterol storage observed in filipin-stained human fibroblasts (Fig. 3E). The mechanism by which a FLAG tag in this luminal loop causes mislocalization of NPC1 to the plasma membrane is currently being investigated.
To further confirm the location of loop E, two different approaches were used. First, an additional construct containing the FLAG tag at a different position within loop E was expressed and analyzed as above. The expressed protein did not mislocalize to the plasma membrane. As shown in Fig. 4F, failure to detect the FLAG-tagged protein in digitonin-treated cells confirmed the previous conclusion that loop E is luminal. Second, an antibody directed against loop E was used to detect its epitope in permeabilized cells that expressed wt NPC1, as above, lending further support to the luminal location of loop E (data not shown).
Notably, the above results indicate that elimination of TM3 from our topological model reverses the orientation of the SSD. In this orientation, the amino terminus of the SSD is luminal, in complete agreement with the topological models for the HMG-R and SCAP SSDs, further supporting the notion that the SSD plays an important role in cholesterol homeostasis in these proteins. Also, in the absence of TM3, the position of the NPC1 tail shifts from the cytoplasm to the lumen. Because the carboxyl-terminal dileucine (LLNF) motif normally functions in the cytoplasm to target proteins to the endosomal/lysosomal system (33), the precise number of TM domains and, therefore, FIG. 5. Localization of FLAG epitopes in truncated NPC1 proteins in semipermeabilized cells. The FLAG-tagged truncated NPC1 proteins were transiently expressed in COS-7 cells and detected by immunofluorescence, as described in Fig. 4. In A-F, expression of proteins from the pLaTM1, pLbTM2, pLcTM3, pLdTM4, pLdTM6, and pLeTM13 constructs (see Fig. 2), respectively, in nonpermeabilized (Non), completely permeabilized (Complete), and selectively permeabilized (Digitonin) cells is shown. The pLaTM1, pLdTM4, and pLdTM6 proteins were detected in selectively permeabilized cells indicating their cytosolic location. Proteins from pLbTM2 and pLcTM3 were detected in nonpermeabilized cells at the plasma membrane, and that of pLeTM13 was undetected in selectively permeabilized cells, confirming that their FLAG-tagged epitopes were luminal. the correct orientation of the tail needed to be established. Thus, a construct in which the tail was removed and replaced by a FLAG tag was used. As anticipated, the epitope tag was undetected in nonpermeabilized cells but was accessible following treatment with digitonin, indicating a cytosolic location for the tail (Fig. 4G). This is consistent with other proteins that contain the dileucine lysosomal targeting motif at their carboxyl terminus. However, for the model to correctly reflect the location of the tail, one of the proposed TM domains after TM13, most probably TM14, which has a lower hydropathy score than TM15 (Fig. 1A), would have to be eliminated as a potential transmembrane domain.
Expression and N-Glycosylation Analysis of Truncated NPC1 Proteins-To further confirm the NPC1 topology suggested by the above results, a number of truncated NPC1 constructs, whose sequences terminated after TM1, 2, 3, 4, 6, and 13 (Fig.  2), were expressed and analyzed as above. Based on the predicted model, constructs terminating after TM1, 4, and 6 were expected to have a FLAG tag on the cytosolic side of the membrane. pLaTM1 protein was undetected in nonpermeabilized cells but showed distinct plasma membrane staining in digitonin-treated cells, suggesting that its FLAG tag was cytosolic (Fig. 5A). pLbTM2 and pLcTM3 were also detected at the plasma membrane and because nonpermeabilized cells were stained, this indicated a luminal position for both, as expected in the absence of TM3 (Fig. 5, B and C). Interestingly, the pLdTM4 protein did not localize to the plasma membrane but showed a predominantly vesicular appearance, suggesting the presence of a targeting signal that either facilitated retrieval from the plasma membrane or caused intracellular sequestration (Fig. 5D). Because the tag was detected in digitonintreated cells, its cytosolic location was confirmed, further supporting the absence of TM3. pLdTM6 gave similar results, although its location was less vesicular and displayed a Golgi and/or ER pattern of distribution (Fig. 5E). pLeTM13 protein had a distribution similar to that of TM6 but was not detected in digitonin-treated cells, indicating a luminal tag position (Fig. 5F). These results confirmed previous observations that TM3 is not present and that loops A, C, D, and E are luminal.
Because the results obtained using the truncated NPC1 proteins were in complete agreement with the results obtained using the full-length FLAG-tagged NPC1 proteins, we extended our studies to investigate the level of glycosylation, if any, of each truncated protein. Loops that are glycosylated would have to enter the ER lumen, lending further support to their luminal location. The truncated proteins were expressed, immunoprecipitated, and then deglycosylated using N-glycosidase F, prior to SDS-PAGE analysis (Fig. 6). Comparison of the estimated molecular weights of the deglycosylated and untreated proteins allowed the extent of N-linked glycosylation to be determined (Table I). For pLaTM1 protein, there was a clear band shift after N-glycosidase F treatment (Fig. 6A), indicating that the amino terminus (loop A) was heavily glycosylated and therefore located in the ER lumen. Because the average molecular mass of one high mannose-type oligosaccharide is about 2 kDa, the shift of approximately 11 kDa probably represents glycosylation at all five potential sites. The protein expressed from pLbTM2 also showed an approximate shift of 11 kDa, and because there was no apparent increase in the extent of deglycosylation in comparison to the shift seen for pLaTM1, it was presumed that the glycosylation site in loop B was not utilized and that this loop was cytoplasmic, as predicted. The results for pLcTM3 expression are also in agreement with the proposed model. A band shift of about 16 kDa indicated that loop C was glycosylated, and it was therefore located in the lumen (Fig. 6A).
The results from the larger truncated proteins were less conclusive than those for loops A to C. The pLdTM6 construct was substituted for the pLdTM4 construct because of inefficient immunoprecipitation of the pLdTM4 protein (Fig. 6B),

TABLE I
Summary of results from the glycosylation analyses of the truncated NPC1 proteins The NPC1-truncated constructs were transfected into COS-7 cells, and the expressed proteins were radiolabeled and immunoprecipitated as described under "Experimental Procedures" and shown in Fig. 6. The molecular masses of each truncated protein were determined before and after deglycosylation using N-glycosidase F. The total number of Nglycosylation consensus sites contained in each construct and their predicted molecular masses without glycosylation are indicated. a Accurate determination of the molecular mass shifts for this construct was not possible (see text). even though immunofluorescent staining indicated that this protein was efficiently expressed in COS-7 cells. Following immunoprecipitation of the pLdTM6 protein, SDS-PAGE analysis revealed multiple bands, thus complicating estimations of the band shift (Fig. 6B). A similar problem was encountered with the pLeTM13 and full-length NPC1 proteins (not shown). However, following deglycosylation, both the pLeTM6 and pLeTM13 proteins showed a qualitatively greater shift than pLcTM3, indicating that loops D and E were glycosylated and thus located in the lumen (Fig. 6B and Table I). The additional bands may represent protein aggregation, or they may reflect additional forms of post-translational modification. Interestingly, there is a 5-to 8-kDa discrepancy between the predicted and observed molecular masses for the deglycosylated pLaTM1, pLbTM2, and pLcTM3 proteins (Table I). This may reflect additional post-translational modification of loop A, which contains six consensus sites for myristoylation. Further studies are necessary to determine whether NPC1 is subject to other types of post-translational processing.
Together, these data independently verify the positions of hydrophilic domains A-C and confirm the absence of TM3. Removal of TM14 to correctly orient the tail results in the topological model of NPC1 shown in Fig. 7. This model consists of 13 transmembrane domains, 3 large luminal and 4 small hydrophilic loops, 6 small cytoplasmic loops, and a cytoplasmic tail. Elimination of the proposed TM3 (Fig. 1) results in a reversal of the location of domains D and E, thus placing the putative SSD in the same orientation as those of the HMG-R and SCAP proteins. Because the carboxyl terminus tail is cytosolic, an additional membrane domain has been removed between TM13 and 15 in our modified model. DISCUSSION The data obtained in these studies are consistent with a topological model in which NPC1 has 13 TM domains, a luminal amino terminus and a cytosolic carboxyl-terminal tail. The results from the glycosylation analyses of truncated NPC1 proteins and the results obtained from investigating the antibody accessibility of FLAG-tagged NPC1 proteins in selectively permeabilized cells are also in agreement with this model. The prediction for TM3 has been determined to be incorrect. These studies have shown that loop D is luminal, not cytosolic as predicted, thus indicating that loops C and D are fused and that the proposed TM3 does not span the membrane. In addition, loop E, which is predicted to be cytosolic, is in fact luminal, suggesting that the orientation of protein domain from loop D to at least loop E is reversed. Thus, the orientations of loops D-E as well as transmembrane domains 4, 6, and 13, and presumably all intervening sequences, are reversed with respect to the predicted model (Fig. 1B). The loss of TM3 from the topological model is not surprising in view of the hydrophobicity profile in which TM3 has the lowest score compared with the other predicted membrane-spanning regions of NPC1 (Fig.  1A). Finally, immunolocalization of the NPC1 tail in cells with selectively permeabilized plasma membranes has confirmed its cytoplasmic location, which is expected, because the tail contains a dileucine targeting motif (33,34) that normally functions in the cytosol to target proteins to an endosomal/lysosomal compartment.
We therefore propose a new model for the topology of NPC1 (Fig. 7), which is in agreement with our extensive experimental data. In this model, the original TM3 does not span the mem- brane and the proposed loops C and D form one luminal domain. In addition, because our amended model reverses the orientation of the middle portion of the protein but maintains the cytosolic tail, the 14th predicted TM domain, which is very close to TM15, is presumed to be absent. Therefore, our data for NPC1 is consistent with a membrane glycoprotein that contains 13 transmembrane domains oriented as in Fig. 7, with a luminal amino terminus and cytosolic carboxyl-terminal tail. The three largest hydrophilic domains are luminal, and the majority of the remaining portions of the NPC1 protein is embedded within the membranes. Of interest, all three luminal hydrophilic domains are approximately the same length (ϳ250 amino acids), although the importance of this observation is currently unclear.
The results from our topological analyses have added to current speculation regarding the possible functional role of the domain from amino acids 615 to 797 (6,35). This region of NPC1 is homologous to the SSD of HMG-R and SCAP, two cholesterol-regulated proteins (22,31). In HMG-R, the putative SSD is required for regulation by enzyme degradation, which occurs in the presence of excess cholesterol (22,36). In SCAP, a D443N mutation in the SSD causes sterol insensitivity (20). The topologies of HMG-R and SCAP are known (30,31), and the current studies have shown that the topological orientation of the NPC1 SSD is identical to those of HMG-R and SCAP, supporting the notion that the SSD in NPC1 is functionally important. A recent study has confirmed the functional importance of the NPC1 SSD, using site-directed mutagenesis (32). However, as in the case of HMG-R and SCAP, the exact function of the SSD is not clear. For example, it is not known whether this domain directly interacts with sterols, such as cholesterol, or whether it interacts with another sterol-regulated protein. Alternatively, the SSD may interact with lipids other than cholesterol. Further studies are necessary to determine the function of this domain.
Even more extensive homology is seen between the membrane domains of NPC1 and the morphogen receptor, PTC, which not only share a similar putative SSD but have additional homologous TM regions. PTC has been implicated in embryogenesis, carcinoma formation, and cholesterol homeostasis (37), a role that is further strengthened by observations that its ligand, sonic hedgehog (24,25), is active when covalently bound to cholesterol (38). In light of the fact that the SSDs of HMG-R, SCAP, and NPC1 are in the same topological orientation, it is tempting to speculate that the SSD in PTC is oriented similarly, and this information would facilitate analysis of this receptor. Thus, knowledge of the topology of NPC1 may also aid in functional studies of the PTC protein.
In conclusion, we have provided extensive experimental evidence to support a model for the topology of the NPC1 membrane glycoprotein, indicating that its putative SSD is functionally significant, based on its identical orientation to homologous sequences. Knowledge of the topology of NPC1 and the functional data obtained in these studies will aid the complete elucidation of NPC1 function.