Membrane Topology of the Hepatitis C Virus NS2 Protein*

The hepatitis C virus (HCV) NS2 protein is a hydrophobic protein. Previous studies indicate that this protein is an integral membrane protein, which is targeted to the membrane of the endoplasmic reticulum (ER) by the signal sequence located in its preceding p7 protein. In this report, we demonstrate that the membrane association of NS2 is p7-independent and occurs co-translationally. Further deletion-mapping studies suggest the presence of two internal signal sequences in NS2. These two internal signal sequences, which are located within amino acids 839–883 and amino acids 928–960, could target the α-globin reporter, a cytosolic protein, to the membrane compartments in HuH7 hepatoma cells. The presence of multiple signal sequences for its membrane association suggests that NS2 has multiple transmembrane domains. The glycosylation studies indicate that both amino and carboxyl termini of NS2 are located in the endoplasmic reticulum lumen. Based on these results, a model for the NS2 membrane topology is presented.

The hepatitis C virus (HCV) NS2 protein is a hydrophobic protein. Previous studies indicate that this protein is an integral membrane protein, which is targeted to the membrane of the endoplasmic reticulum (ER) by the signal sequence located in its preceding p7 protein.
In this report, we demonstrate that the membrane association of NS2 is p7-independent and occurs co-translationally. Further deletion-mapping studies suggest the presence of two internal signal sequences in NS2. These two internal signal sequences, which are located within amino acids 839 -883 and amino acids 928 -960, could target the ␣-globin reporter, a cytosolic protein, to the membrane compartments in HuH7 hepatoma cells. The presence of multiple signal sequences for its membrane association suggests that NS2 has multiple transmembrane domains. The glycosylation studies indicate that both amino and carboxyl termini of NS2 are located in the endoplasmic reticulum lumen. Based on these results, a model for the NS2 membrane topology is presented.
Hepatitis C virus (HCV) 1 is an important human pathogen that can cause severe liver diseases, including cirrhosis and hepatocellular carcinoma (1)(2)(3)(4). This virus has a 9.6-kb positive-stranded RNA genome that codes for a polyprotein with a length of slightly more than 3000 amino acids. The HCV structural proteins, core, E1, and E2, are located at the amino terminus and are released from the polyprotein by the cellular signal peptidase located in the lumen of the endoplasmic reticulum (ER) (5)(6)(7). The non-structural proteins, NS3, NS4A, NS4B, NS5A, and NS5B, are located at the carboxyl terminus and are separated from each other by the protease activity residing within the NS3 protein (7)(8)(9).
Between E2 and NS3 there are two additional proteins named p7 and NS2. p7 may or may not be separated from the E2 protein (10) and has been shown to contain a signal sequence to direct the downstream NS2 protein to the membranes (11). NS2 is separated from its preceding p7 protein by the signal peptidase (11). The function of NS2 in the HCV life cycle is unclear. Its deletion did not abolish the replication of HCV RNA in cell cultures, indicating that it is not required for viral RNA replication (12,13). The separation of NS2 and its following NS3 protein requires most of the NS2 sequence and its adjacent NS3 protease domain (14).
The deduced amino acid sequence of NS2 indicates that it is a hydrophobic protein. Previous studies indicate that NS2 is a non-glycosylated integral membrane protein and mostly not exposed in the cytosol (15). However, the membrane topology of this protein remains unclear. In this report, we have studied the molecular mechanism that regulates the membrane translocation of NS2. Interestingly, our results indicate that the membrane-association of NS2 is p7-independent. Further studies indicate that NS2 contains at least two internal signal sequences for its membrane association and likely has multiple transmembrane domains.

EXPERIMENTAL PROCEDURES
Construction of DNA Plasmids-The HCV sequences used for these studies were derived from the HCV-1 strain (16). The isolation of the cDNA fragments was achieved by PCR, using oligonucleotide primers with the appropriate restriction sites. In the construction of DNA plasmids, the ATG start codon was inserted upstream of the HCV sequence for translation initiation. The plasmid pET(p7-NS2) was obtained by inserting the nucleotide (nt) sequence 2513-3428 into the NdeI-BamHI sites of the expression vector pET-3a (Novagen, Madison, WI). The upstream primer used for this construction was 5Ј-GTCGTTCATATGT-TCCTTCTGCTTG-3Ј, containing a NdeI site, denoted by the underline, and the downstream primer was 5Ј-CGTGGATCCCTACAGCAACCTC-CACCCCTT-3Ј, which contains a BamHI site. The expression of the plasmids containing the HCV cDNA fragment was under the control of the bacteriophage T7 promoter. The plasmid pET(NS2) was constructed in the same manner except that the nucleotide sequence 2778 to 3428 was used. pCMV(NS2) was similarly generated by PCR amplification of the sequence from nt 2778 to 3428. The upstream primer used was 5Ј-TA-CTAAGCTTCACCATGCTGGACACGGAGGTG-3Ј, containing a Hind-III restriction enzyme site, and the downstream primer used was 5Ј-A-GTATCTAGACTACAGCAACCTCCACCC-3Ј, containing an XbaI site. pCMV(FLAG-NS2) was generated by PCR amplification from nt 2778 to 3428 with a FLAG tag sequence at the 5Ј-end. The upstream primer used was 5Ј-TACTAAGCTTCACCATGGATTACAAGGATGACGACGA-TAAGCTGGACACGGAGGTG-3Ј, and the downstream primer used was the same as above. The HCV sequence was inserted into the polylinker cloning site of the pRc/CMV vector (Invitrogen, San Diego, CA) under the expression control of the cytomegalovirus (CMV) immediate early (IE) promoter and the T7 promoter.
The carboxyl-terminal truncations at amino acid (aa) 865 and 917 were achieved by digesting the plasmid pET(NS2) with BglII and StuI, respectively. The truncations at aa 839 and aa 883 were carried out by PCR. A stop codon was engineered at the end of these truncated sequences during PCR for translation termination.
The amino-terminal truncations at aa 884 and 928, initiating at nt 3000 and 3132, respectively, were created by PCR and cloned into the HindIII/XbaI sites of the expression vector pRc/CMV. An ATG codon was inserted at the 5Ј-end during PCR for translation initiation.
The PCR-based site-directed mutagenesis (17) was used to introduce the glycosylation sites. The sequences of the insertions and mutations are noted below by underlines. pCMV(p7NS2-812G) and pCMV(NS2-812G) were generated by inserting the sequence AATATTACGGCACT-GGAC, which codes for NITALD, between nt 2783 and 2784 of the NS2 sequence. pCMV(NS2-839G) was generated by inserting the sequence AATATT between nt 2867 and 2868 of the NS2 sequence to convert the sequence YISW to YINISW, creating the glycosylation site at aa 839. pCMV(NS2-869G) was generated by inserting the sequence AATAT-TACG, encoding NIT, between nt 2957 and 2958. pCMV(NS2-988G) was generated by mutating the amino acid from Ala to Asn at aa 988, changing the sequence from GADT to GNDT. pCMV(NS2-1018G) was generated by mutating the amino acid from Met to Asn, changing the sequence from GMVS to GNVS. All of the above constructs were cloned into the HindIII/XbaI sites of the pRc/CMV vector under the expression control of the CMV IE promoter and T7 promoter. Each construct also contained an HA tag at the 3Ј-end of the NS2 coding sequence prior to the stop codon. All DNA constructs were confirmed by DNA sequencing at the USC Microchemical Core Facility.
In Vitro Transcription and Translation-All plasmids used for in vitro RNA synthesis were linearized with the appropriate restriction enzyme and transcribed with T7 RNA polymerase, per the manufacturer's protocol (Promega, Madison, WI). In vitro translation reactions (12.5-13.5 l of mixture) were carried out at 30°C for 60 min using nuclease-treated rabbit reticulocyte lysates, in the presence [ 35 S]methionine (ICN, Irvine, CA), following the manufacturer's recommendations (Promega). If canine pancreatic microsomal membranes (MM) (Promega) were used, 1 l was added for each translation reaction. After translation, 1 l of the sample was mixed with 2ϫ Laemmli sample buffer and analyzed by SDS-PAGE, followed by autoradiography. If MM was added post-translationally, the translation reaction was first terminated by the addition of 5 mM CaCl 2 that activates the micrococcal nuclease. After the addition of MM, the sample was further incubated at 30°C for 5, 60, or 90 min.
Membrane Fractionation Experiments-Five microliters of the translational mixture was overlaid on a 650-l cushion of 13% sucrose in Tris-buffered saline (TBS, 10 mM Tris-HCl, pH 7, and 150 mM NaCl) and centrifuged at 45,000 rpm for 10 min at 4°C in a Beckman SW-55Ti rotor. For the high salt wash to remove the proteins peripherally associated with membranes, NaCl was added to the translational mixture to a final concentration of 1 M. After a further incubation on ice for 30 min, the mixture was then centrifuged as described above. For the alkaline carbonate extraction experiment, the translational mixture was mixed with 600 l of 100 mM Na 2 CO 3 , pH 11, prior to centrifugation. The proteins in the supernatants were precipitated with 2 volumes of saturated ammonium sulfate for 60 min at 4°C, washed with 5% trichloroacetic acid, and resuspended in 2ϫ Laemmli buffer. The membrane pellet was directly resuspended in 2ϫ Laemmli buffer and analyzed by SDS-PAGE. For membrane solubilization, the non-ionic detergent Nonidet P-40 was added to a final concentration of 0.3% prior to ultracentrifugation. All the experiments involving microsomal membranes such as the fractionation experiments were repeated at least three times to ensure the accuracy of the results.
Endoglycosidase H Treatment-After translation in the presence of MM, the endoglycosidase H (Endo H) digestion was performed per the manufacturer's recommendations (Roche Molecular Biochemicals). Briefly, 4 l of the mixture was added to 2 l of 10ϫ denaturing buffer (0.5% SDS, 1% ␤-mercaptoethanol) and 14 l of H 2 O, heated at 100°C for 5 min, then chilled on ice. The mixture was split evenly: in one aliquot, 1.5 l of 50 mM sodium citrate, pH 5.5, and 1 l of Endo H were added; in the other, 50 mM sodium citrate, pH 5.5, was added without the enzyme. Both samples were incubated at 37°C for 2 h. The protein mixtures were then analyzed by SDS-PAGE.
Cell Transfections-HuH7 cell monolayers were maintained in Dulbecco's modified Eagle's medium containing 10% fetal bovine serum and 1% penicillin/streptomycin. Cells were transfected with pCMV(839 -883-globin), pCMV(928 -960-globin), and pRc/CMV by calcium phosphate DNA precipitation (17,18). Forty-eight hours after transfection, cells were starved with methionine-free medium for 2 h then metabolically labeled with [ 35 S]methionine for 1-3 h. After labeling, cells were washed with Dulbecco's phosphate-buffered saline and then lysed with 0.1ϫ TBS with repeated passages through a 200-l pipette tip. Nuclei and cell debris were removed by brief centrifugation in the microcen-trifuge, and the cell lysates were divided evenly in two. To one aliquot, Nonidet P-40 was added to a final concentration of 0.3%. Fractionation experiments were performed by overlaying the samples on a 400-l cushion of 13% sucrose in TBS and centrifuged at 45,000 rpm for 30 min at 4°C in a Beckman SW-55Ti rotor. The supernatants were mixed with equal volumes of RIPA (10 mM Tris, pH 7, 150 mM NaCl, 1% Nonidet P-40, 1% sodium deoxycholate, 0.1% SDS), and the membrane pellets were resuspended in 1 ml of RIPA. The rabbit anti-human globin antibody (Dako, Santa Barbara, CA) was added for overnight incubation at 4°C. Subsequently, the immune complexes were precipitated by the addition of Pansorbin (Calbiochem, La Jolla, CA) at 4°C for 1 h, followed by four washes with RIPA. The immunoprecipitated proteins were analyzed on a 12.5% SDS-PAGE gel.
Immunofluorescence Staining-Huh7 cells transfected with the plasmid were fixed with 3.7% formaldehyde in phosphate-buffered saline for a 5-min period, 48 h after transfection. Cells were then stained with mouse anti-FLAG (Stratagene) and rabbit anti-calreticulin (Affinity Bioreagents, Inc.) primary antibodies and rhodamine-conjugated goat anti-mouse (Pierce) and fluorescein-conjugated goat anti-rabbit (Roche Diagnostics) secondary antibodies. For cells transfected with the ␣-globin reporter constructs, the primary antibodies used were goat antihuman globin (Bethyl Laboratories, Montgomery, TX) and rabbit anticalreticulin antibodies, and the secondary antibodies used were rhodamine-conjugated bovine anti-goat and fluorescein-conjugated bovine anti-rabbit antibodies (Santa Cruz Biotechnology, Santa Cruz, CA). The fluorescence staining images were visualized using a Zeiss fluorescence microscope and captured using the AxioVision image acquisition software.

RESULTS
The Membrane Association of NS2 Is p7-independent-Previous studies indicate that the p7 protein contains a signal sequence that targets the NS2 protein to the ER membrane, and the cleavage between p7 and NS2 is mediated by the signal peptidase located in the ER lumen. As shown in Fig. 1A, the in vitro translation of the NS2 sequence, which starts from codon 810 and ends at codon 1026 in the HCV-1 sequence, produced a 23-kDa protein product. The size of this protein was not affected by the microsomal membranes (MM). In contrast, the translation of the p7-NS2 sequence from codons 719 to 1026 yielded a 30-kDa protein. In the presence of MM, this protein was converted to 23-and 7-kDa proteins, which correspond to NS2 and p7 proteins, respectively. This result is consistent with the previous reports that p7 is separated from NS2 by the signal peptidase located within MM (11).
Interestingly, in the membrane fractionation experiment using ultracentrifugation, the majority (ϳ75%) of NS2 was found to be co-pelleted with MM even when it was synthesized in the absence of its preceding p7 sequence (Fig. 1B). This co-precipitation was specific and required MM, because most NS2 (96%) remained in the supernatant in the absence of MM or if MM was solubilized with the non-ionic detergent Nonidet P-40 (Fig.  1B). These results raise the possibility that NS2 may be able to become associated with the membranes in a p7-independent manner. Note that a small amount of NS2 was found in the pellet in the absence of MM. This may be due to protein aggregation caused by the hydrophobic sequence of NS2.
To investigate whether NS2 can indeed be associated with the membranes in the absence of its preceding p7 sequence in cells, we have performed the immunofluorescence staining experiment. In the absence of a good anti-NS2 antibody, we have fused a FLAG tag to the amino terminus of NS2 to facilitate the analysis. The DNA plasmid encoding this FLAG-tagged NS2 was transfected into Huh7 cells, a well-differentiated human hepatoma cell line. To ensure that the FLAG-tagged NS2 could be properly expressed in Huh7 cells, we first performed the immunoprecipitation experiment. As shown in Fig. 2A, the FLAG-tagged NS2 could be clearly detected in Huh7 cells. Next, we performed the immunofluorescence staining experiment using the anti-FLAG antibody. As shown in Fig. 2B, the FLAG-tagged NS2 protein displayed a punctate and perinuclear staining pattern typical of ER-associated proteins (panel a). Indeed, when the cells were also stained for calreticulin (panel b), an ER-associated protein, a significant fraction of the FLAG-tagged NS2 was found to colocalize with calreticulin in the cells (panel c, yellow color). This result is in support of the in vitro observation, which indicated that NS2 could be associated with the membranes independent of its preceding p7 sequence.
NS2 Is Associated with the Membranes Co-translationally-The association of NS2 with the membranes in the absence of p7 may occur post-translationally due to its hydrophobicity or co-translationally due to the presence of one or more internal signal sequences. To distinguish between these two possibilities, in vitro translation experiments were performed by adding MM either before or after the translation reaction. When the NS2 protein was synthesized with MM added at the onset of the translational reaction, the majority (ϳ73%) of the 23-kDa NS2 protein was found to be associated with the membrane pellet in the fractionation experiment (Fig. 3). However, when NS2 was synthesized with MM added after the termination of the translation reaction, the protein was recovered almost entirely (95%) in the supernatant fraction. The same result was obtained whether NS2 was incubated with MM for 5, 60, or 90 min after the translation reaction. These results indicate that the association of NS2 with the membrane is a co-translational event.
To examine whether NS2 synthesized in the absence of p7 was also an integral membrane protein, the membranes were washed with 1 M NaCl prior to fractionation. As shown in Fig.  4A, this high salt wash did not abolish the association of NS2 with the membranes, indicating that NS2 was not peripherally associated with the membranes. NS2 apparently also did not totally reside in the lumen of MM, because it could not be extracted from the membranes by the chaotropic alkaline carbonate buffer (Fig. 4B). Thus, NS2 appeared to be an integral membrane protein even when it was synthesized in the absence of its preceding p7 sequence.

NS2 Contains Two Internal Signal Sequences for Its Membrane Association-
The co-translational insertion of NS2 into the membrane would require the presence of one or more internal signal sequences in the NS2 sequence. To determine whether NS2 indeed contains internal signal sequences, the deletion-mapping analysis was performed. Three NS2 carboxyl-terminal deletion mutants, containing aa 810 to staggered C termini at aa 883, 917, and 965, were created. The proteins were synthesized in vitro either in the presence or absence of MM, followed by membrane fractionation using ultracentrifugation. As shown in Fig. 5A, the great majority of the NS2 deletion mutants were present in the supernatant in the absence of MM or if MM was solubilized with Nonidet P-40. In the presence of MM, significant fractions of the NS2 proteins truncated at aa 965, 917, and 883 were found to be associated with the membrane pellet. These results indicate a possible presence of a signal sequence located upstream of aa 883 to target NS2 to the membranes. Note that the fractions of the NS2 protein that were associated with the membranes varied from ϳ45% to 98%. These variations of the membrane association efficiency were presumably caused by the sequence truncations, which likely affects the conformation of the protein.
To further characterize this possible signal sequence, additional carboxyl-terminal truncation experiments were conducted. Due to the small size of the further truncated proteins, an ␣-globin reporter sequence was fused to the carboxyl termini of these NS2 proteins to facilitate the analysis. Approximately 75% of the fusion protein containing aa 810 -883 of the NS2 sequence was recovered in the membrane pellet, supporting the FIG. 3. Co-translational association of NS2 with the membranes. The NS2 RNA was translated in vitro using the rabbit reticulocyte lysates. The MM was added either at the initiation of the translation reaction (co-translational) (lanes 1-3) or after the termination of the translation reaction (post-translational) (lanes 4 -6). The translation reaction was terminated by the addition of CaCl 2 , which activates the micrococcal nuclease to degrade the RNA. The samples were fractionated by ultracentrifugation. M, the control translational mixture; S, the cytosolic supernatant; and P, the membrane pellet. Percentages indicate the fractions of the NS2 protein in the supernatant and in the pellet.

FIG. 4. Characterization of the membrane association of NS2.
NS2 was synthesized either in the absence or in the presence of the MM prior to fractionation by ultracentrifugation. A, the high salt wash experiment. In lanes 8 and 9, the MM in the translational mixture was treated with 1 M NaCl on ice for 30 min prior to fractionation. B, the alkaline carbonate extraction experiment. In lanes 6 and 7, the MM in the translational mixture was extracted with 100 mM Na 2 CO 3 , pH 11, prior to fractionation. M, the control NS2 protein without fractionation; S, the cytosolic supernatant; and P, the membrane pellet. The percentages indicate the relative amount of the NS2 protein in the cytosolic supernatant and in the membrane pellet. notion that this NS2 sequence indeed contains a signal for membrane association. Further truncation to aa 839 resulted in the detection of the fusion protein almost entirely in the supernatant fraction. This indicates that the NS2 sequence between aa 810 and 839 does not contain a signal sequence. Therefore, the signal sequence is likely located between aa 839 and 883. To examine this possibility, aa 839 -883 of the NS2 sequence was fused to the ␣-globin reporter. As shown in Fig.  5A, a significant portion of the fusion protein was detected in the membrane fraction, although the efficiency of membrane association was reduced to ϳ27%. The results shown in Fig. 5A were reproducible in at least three different experiments. These results indicate that NS2 likely contains an internal signal sequence, which is located within the region of aa 839 -883.
To determine whether additional signal sequences are present within the NS2 sequence, the amino-terminal deletionmapping experiments were also performed. As shown in Fig.  5B, ϳ60% of the NS2 protein truncated at aa 884 remained associated with the membrane fraction. A similar result (ϳ69% membrane association) was obtained when the NS2 protein was truncated at aa 928. These results indicate the possible presence of an additional signal sequence located downstream of aa 928. Further truncation analysis required the fusion of the small NS2 sequence to the ␣-globin reporter. As shown in Fig. 5B, further truncation to aa 961 abolished the membrane association of this NS2-globin fusion protein. This result suggests the presence of a signal sequence in a region between aa 928 and 961 or encompassing aa 961. For this reason, the sequence aa 928 -981 was again fused to the ␣-globin reporter, and the possible association of the fusion protein with the membranes was again analyzed. The fusion of aa 928 -981 to ␣-globin resulted in the association of this protein with the membrane fraction, albeit inefficiently at a percentage of ϳ23%. The results of the above deletion-mapping experiments were reproducible in at least three independent experiments. These results suggest the presence of an additional signal sequence within aa 928 -981 and that the efficiency of this signal sequence to direct membrane association can be enhanced by its downstream sequence. A summary of the deletion mapping experiments is illustrated in Fig. 6. These results indicate the presence of two possible signal sequences: one is located within aa 839 -883, and the other is within aa 928 -981.
To confirm that these two regions indeed contain signal sequences, the experiments were repeated in HuH7 cells, a well-differentiated human hepatoma cell line. The NS2 sequence encoding aa 839 -883, fused to the ␣-globin reporter, was placed under the expression control of the CMV IE promoter. The DNA plasmid or its control vector was then transfected into HuH7 cells, and the fusion proteins were metabolically labeled with [ 35 S]methionine. The cells were then lysed, fractionated by ultracentrifugation, and radioimmunoprecipitated with the anti-globin antibody. As shown in Fig. 7, 62% of the fusion protein was found in the membrane fraction. However, in the presence of Nonidet P-40 that solubilized the membranes, almost all of the protein was recovered in the supernatant. This NS2-globin fusion protein was not detected if the cells were transfected with the control pRc/CMV vector (data not shown). The same experiment was repeated with the NS2 sequence encoding either aa 928 -981 or aa 928 -960 and similar results were obtained (Fig. 7). The subcellular fractionation results were also confirmed by the immunofluorescence staining experiments. As shown in Fig. 8B, when the cytosolic human ␣-globin was expressed in Huh7 cells, it displayed a granular staining pattern with no resemblance to that of calreticulin, an ER-associated protein (Fig. 8A). However, when the globin sequence was fused to either aa 839 -883 or aa 928 -960 of the NS2 sequence and expressed in Huh7 cells, it displayed a reticular and perinuclear staining pattern similar to that of calreticulin. Thus, these results indicated that the sequences of aa 839 -883 and aa 928 -981 indeed contained signals for targeting NS2 to the membranes in HuH7 cells.
Moreover, the signal sequence located within aa 928 -981 could be further mapped to aa 928 -960.
Glycosylation Studies Indicate That NS2 Contains Multiple Domains in the ER Lumen-Multiple signal sequences for the membrane association of NS2 suggest that NS2 may have multiple transmembrane domains. To further characterize the membrane topology of NS2, glycosylation studies were performed. To determine whether the N terminus of NS2 is located within the ER lumen, a glycosylation site was introduced between aa 812 and 813 (812G mutant). As shown in Fig. 9A, the translation of the p7-NS2 sequence containing this glycosylation site resulted in the production of the full-length p7-NS2 protein in the absence of microsomal membranes. The p7 sequence was removed in the presence of MM to generate the mature NS2 protein. In addition, a protein slightly larger than NS2 was also detected. This larger protein was sensitive to endoglycosidase H (Endo H), indicating that it was the glycosylated form of the NS2 protein. This result is expected, because the cleavage of the p7-NS2 sequence is mediated by the signal peptidase located in the ER lumen, and therefore, the NS2 N terminus is expected to reside within the ER lumen. The glycosylation of the NS2 protein is inefficient. The reason for this is unclear. Similar studies in the past also generated inefficient glycosylation results (15,19). Surprisingly, even if the 812G mutant was synthesized in the absence of the p7 leader sequence, a glycosylated form that is sensitive to Endo H was also detected. This result indicates that the amino terminus of NS2 can be positioned in the lumen of the ER independent of the preceding p7 protein.

DISCUSSION
Previous studies indicate that the HCV NS2 protein is an integral membrane protein that is targeted to the ER membrane by a signal sequence located in its preceding p7 protein (11,15). In this report, we have further studied how NS2 associates with the membranes. In agreement with the earlier studies, we found that the NS2 protein was separated from the preceding p7 sequence only in the presence of membranes (Fig.  1A), indicative of separation by the signal peptidase. Interestingly, we also found that NS2, in the absence of p7, could become membrane-associated (Fig. 1B). This membrane association is supported by the immunofluorescence double-staining experiment using the ER-associated calreticulin as the ER marker (Fig. 2B). Because this membrane association could not be abolished by treating the membranes with 1 M NaCl or by extraction with the chaotropic alkaline carbonate buffer (Fig.  4), NS2 was apparently an integral membrane protein. The association of NS2 with the membranes only occurred co-translationally (Fig. 3), suggesting the presence of internal signal sequences in NS2. This possibility was tested by the deletionmapping experiments, which revealed two internal signal sequences located within aa 839 -883 and aa 928 -960 (Figs. 5-7). These two sequences could target the ␣-globin reporter, a cytosolic protein, to the membranes in HuH7 cells (Figs. 7 and 8).
The identification of the two internal signal sequences suggests that NS2 is a type III integral membrane protein with multiple transmembrane domains (20,21). To further study the membrane topology of NS2, we performed the glycosylation studies by introducing glycosylation sites into various regions of the NS2 sequence. Interestingly, the glycosylation site introduced near the amino terminus of NS2 was used independent of the p7 sequence (Fig. 9A). These results indicate that the amino terminus of NS2 can be localized to the ER lumen independent of p7. Because the presence of p7 did not significantly increase the glycosylation efficiency of NS2-812G (Fig.  9A), the role of the p7 signal sequence in the membrane translocation of NS2 is unclear. The p7 signal sequence may be needed only for the separation of NS2 from its preceding HCV sequences by signal peptidases.
The hydrophobicity plot of NS2 based on its deduced amino acid sequence is shown in Fig. 10A. The sequence aa 814 -835 near the amino terminus is highly hydrophobic. It is conceivable that this hydrophobic domain serves as the first transmembrane domain that will have a type I topology, because the amino terminus of NS2 is localized in the ER lumen. If this is correct, the hydrophilic sequence immediately following aa 835 is expected to be localized in the cytosol. The observation that the glycosylation site introduced at aa 839 was not used is consistent with this speculation (Fig. 10B). The first internal signal sequence was mapped between aa 839 and 883 (Figs. 5-7). This sequence likely contains the second transmembrane domain. As the glycosylation site introduced at aa 869 was used for glycosylation, the second transmembrane domain may be residing between aa 843 and 866, a sequence that is largely hydrophobic (Fig. 10A). This second transmembrane domain will have a type II topology with the amino terminus in the cytosol and the carboxyl terminus in the ER lumen.
The sequence from aa 872 to 919 is highly hydrophobic (Fig.  10A). This sequence may contain a third transmembrane domain. If this is correct, this transmembrane domain is expected to have a type I topology with the amino terminus in the ER lumen and the carboxyl terminus in the cytosol. The second signal sequence was identified between aa 928 and 960. This sequence likely contains the fourth transmembrane domain. The sequence from aa 928 to 956 is highly hydrophobic and may serve this function (Fig. 10A). If this second signal sequence indeed contains another transmembrane domain, it will likely have a type II topology with its carboxyl terminus localized to the ER lumen, because the glycosylation site introduced at aa 988 was used for glycosylation (Fig. 9B, also see Ref. 15). The carboxyl terminus of NS2 is predicted to localize in the ER lumen, because the glycosylation site introduced at aa 1018 near the carboxyl terminus was used for glycosylation (Fig.  9B). This result is consistent with the glycosylation studies of Santolini et al. (15), who also predicted that the carboxyl terminus of NS2 would be localized to the ER lumen. A model of the NS2 topology in the membrane is illustrated in Fig. 10B. This model predicts that the amino terminus and the carboxyl terminus are localized to the ER lumen and that NS2 has a total of four alternating type I and type II transmembrane domains.
In conclusion, our studies have allowed us to identify two internal signal sequences in the NS2 sequence, which allowed us to propose a model that NS2 has a type III membrane topology with multiple transmembrane domains. This model will now enable us to further explore the possible functions of NS2, which have remained largely elusive.