Only the first and the last hydrophobic segments in the COOH-terminal third of Na,K-ATPase alpha subunit initiate and halt, respectively, membrane translocation of the newly synthesized polypeptide. Implications for the membrane topology.

We studied the topogenic properties of five hydrophobic segments (H5-H9) in the COOH-terminal third of Na,K-ATPase α subunit using in vitro insertion of fusion proteins into endoplasmic reticulum membranes. These fusion proteins consisted of several different lengths of truncated α subunit starting at Met729 and a reporter protein, chloramphenicol acetyltransferase, that was linked in frame after each hydrophobic segment. We found that membrane insertion of the newly synthesized COOH-terminal third was initiated by H5 and terminated by H9, indicating that here only H5 and H9 have topogenic function. The other three, H6-H8, did not have topogenic function in the native context and were translocated into the endoplasmic reticulum lumen. These results were in striking contrast to the previous models in which four or six hydrophobic segments were proposed to cross the membrane. Furthermore, the findings suggest a novel mechanism for achieving the final membrane topology of the COOH-terminal third of the α subunit.

Na,K-ATPase consists of two non-covalently linked subunits: a larger non-glycosylated ␣ subunit (about 100 kDa) and a glycosylated ␤ subunit (approximately 55 kDa) (Brotherus et al., 1983), both of which are transmembrane proteins. The complete amino acid sequence of the ␣ subunit from different sources has been cloned (see review by Mercer (1993)). Hydropathy analysis of the sequence has shown nine hydrophobic segments (H1-H9). Four segments (H1-H4) are in the NH 2terminal third and the rest (H5-H9) are in the COOH-terminal third of the subunit. However, H5 is often proposed to cross the membrane twice as H5 and H6, so that the remaining segments are often named H7-H10.
The four hydrophobic segments in the NH 2 -terminal third have been demonstrated to be transmembrane spans by immunochemical studies (Kano et al., 1990;Arystarkhova et al., 1992;Ning et al., 1993;Mohraz et al., 1994;Yoon and Guidotti, 1994). The sidedness of the NH 2 terminus has been determined to be cytoplasmic by immunochemical methods (Felsenfeld and Sweadner, 1988;Antolovic et al., 1991;Ning et al., 1993;Canfield and Levenson, 1993;Yoon and Guidotti, 1994). The orientation of these four hydrophobic segments in the lipid bilayer has been determined by the in vitro ER 1 membrane insertion experiments (Xie and Morimoto, 1995). In contrast to the NH 2 -terminal third, the membrane topology of the COOH-terminal third of the ␣ subunit has been quite controversial. Since this region is involved in several important functions of the Na,K-ATPase such as cation binding and occlusion (Karlish et al., 1990), ouabain binding, and conformational changes during the E1 to E2 transitions that occur during the catalytic cycle (see review by Vasilets and Schwartz (1993) and Lingrel and Kuntzweiler (1994)), structural information on this particular region is crucial for understanding the structure function relationship of the enzyme.
In this paper, we describe studies on the transmembrane disposition of the COOH-terminal third of the ␣ subunit using an experimental strategy based on the topogenic property of hydrophobic segments. The fundamental principle of this approach is that the transmembrane disposition of a polytopic protein is achieved cotranslationally by the action of a series of alternating insertion signals that initiate translocation of downstream portions of the polypeptide, and of halt transfer sequences that block the translocation of the downstream sequences (Blobel, 1980;Sabatini et al., l982). Since the hydrophobic segments in the COOH-terminal third occur after a long cytoplasmic stretch, this principle implies that the first transmembrane segment in this region must be an insertion signal that has also anchoring function. Fusion proteins were made consisting of several different lengths of truncated ␣ subunit, linked in frame at their COOH termini to a reporter protein, chloramphenicol acetyltransferase (CAT), that has a consensus N-linked glycosylation site (Gorman et al., 1982). Fusion proteins were designed such that the topogenic properties of all the hydrophobic segments could be examined individually and sequentially in their native context. Occurrence of N-linked glycosylation and sensitivity to protease digestion of the reporter protein were used as markers for luminal and cytoplasmic disposition, respectively. The results provide both structural (orientation with respect to the membrane) and functional (an insertion signal, a halt transfer signal, or an inactive element as signals) information on the hydrophobic segment in the context tested. Furthermore, these findings shed light on the membrane insertion mechanism of polytopic proteins.

Plasmids and Construct of Mutants
Plasmids pGEM-3Z and pCAT basic were obtained from Promega Corp., and pT7/T3 19R was obtained from Life Technologies, Inc. Okayama-Berg vector containing entire coding region of rat brain Na,K-ATPase ␣-1 subunit (pRNKA1) was supplied by Dr. Hara (Tokyo Dental Medical School, Tokyo, Japan).
cDNA , and pC5B-6, which encode the fusion proteins of the truncated Na,K-ATPase ␣ subunit and CAT as a reporter protein, are derived from plasmids pF1 and pCAT (Xie and Morimoto, 1995).
pC5-A 3.2-kbp fragment containing an entire coding region of the ␣ subunit including the modified 5Ј end noncoding region was obtained by digestion of pF1 with PstI and SacI. An aliquot of 3.2-kbp fragment was digested with KpnI to generate a 2.1-kbp fragment with KpnI/SacI ends. Another aliquot was successively digested with StuI and HaeIII, and the generated 0.48-kbp fragment was ligated to 8-mer SalI linker, followed by digestion with KpnI and SalI. The resultant 0.48-kbp fragment was inserted, together with 2.1-kbp PstI/KpnI fragment, into the plasmid pCAT that had been cut by PstI and SalI (this construct was defined as pN5). After plasmid pN5 was partially digested with NcoI, the isolated NcoI-cut plasmid pN5 that encodes the fusion protein C-5 was self-ligated to construct plasmid pC5.
pC6 -A 3.2-kbp fragment (see "pC5") was successively digested with StuI and Xmn I to generate a 0.64-kbp fragment. After ligation of the fragment and 8-mer SalI linkers, the 0.64-kbp fragment was digested with KpnI and SalI. The generated 0.6-kbp fragment and a 2.1-kbp PstI/KpnI-cut fragment from pC5 were inserted together into the pCAT vector that had been cut with PstI and SalI (defined as pN6). This construct contains an additional consensus N-linked glycosylation site at the ligation site between the truncated ␣ subunit and CAT protein. Plasmid pC6 was constructed from pN6 in the same way as described for pC5 construction.
pC7-After digestion of the 3.2-kbp PstI/SacI fragment (see "pC5") with EcoRI, a generated 2.8-kbp fragment was subcloned into a PstI-EcoRI-cut plasmid pSP73, which was then digested with PstI and EcoRV. The PstI-EcoRV fragment was ligated to the pCAT vector that had been cut with PstI and HincII (defined as pN7). Plasmid pC7 was constructed from pN7 in the same way as described for pC5 construction. This construct contains an additional consensus N-linked glycosylation sequence like pC6.
pC8 -After digestion of pF1 with Bsu36I, a 2.95-kbp fragment was generated and its protruding ends were filled in with Klenow enzyme. The blunt ended fragment was then ligated to 8-mer SalI linker, digested with SalI and PstI, and subcloned into the plasmid pCAT that had been cut with PstI and SalI (defined as pN8). Plasmid pC8 was constructed from pN8 in the same way as described for pC5 construction.
pC9 -Plasmid pF1 was partially digested with HindIII and BglI, and a generated 3.5-kbp fragment was blunt-ended with T4 DNA polymerase, followed by digestion with PstI. The resulting fragment was then subcloned into the pCAT that had been cut with PstI and HincII (defined as pN9). Plasmid pC9 was constructed from pN9 in the same way as described for pC5 construction.
pCB5-6 -A 490-base pair DNA segment generated by PCR using a linearized plasmid pC6 as the template and two primers (upstream primer: 45-mer oligonucleotide described in the construction of plasmid pF1 (Xie and Morimoto, 1995) and downstream primer: 5Ј-TAGCTCCT-GAAAATCTCGCC-3Ј) was digested with DpnI. After ligation of the isolated 389-base pair fragment to 8-mer SalI linker, the ligated fragment was digested with SalI and PstI. The resulting 372-base pair fragment was inserted into the plasmid pCAT that had been cut with SalI and PstI to prepare plasmid pCB5-6.
pMT2-N5-A eukaryotic expression vector pMT2 was cut with EcoRI, and the resulting protruding ends were filled in with Klenow enzyme. The blunt-ended vector was further digested with PstI, and the isolated PstI/EcoRI-cut pMT2 was ligated with a fragment generated by digesting pN5 (see "pC5") with PstI and SmaI to construct plasmid pMT2-N5.
pMT2-C5-This plasmid was constructed from plasmids pMT2 and pC5 in the same way as described for pMT2-N5 construction.
pX5, -6, -7, -8, and -9 -These plasmids were prepared from plasmids pSG5rGH and pC5, -6, -7, -8, and -9. About 800-base pair DNA segments containing one of the five hydrophobic segments and CAT were generated by PCR using two primers described below and a linearized plasmid that contains the hydrophobic segment interested as the template. The NH 2 -terminal side of the primers that contain PmlI site are given below.

Oligo C: GAATTCGAGCTCGGTACCCGGGGATTC
These PCR products, which had been digested with PmlI and BamHI, were ligated to pSG5rGH, which had been digested with Eco47III and BamHI.
pXc and pX4 -These two plasmids were prepared as a negative and a positive controls. Plasmid pXc was constructed by ligating an entire coding region of CAT cDNA to the pSG5rGH that had been cut with Eco47III and BamHI in frame. For construction of plasmid pX4, DNA fragment that encodes a fusion peptide consisting of hydrophobic segment H4 and CAT protein was made by PCR using a linearized plasmid pN4 (Xie and Morimoto, 1995) as the template and two primers (Oligo C and Oligo 4: TGATGAGCTCGCACGTGCT GTCATCTTCCTCAT-TGG). The PCR product was inserted to plasmid pSG5rGH as described for pX5 construction.
All plasmids encoding a fusion protein have been examined by DNA sequencing, the size of their primary translation products and immune reactivity to anti-CAT antibody in order to confirm the in frame ligation. These fusion proteins are shown schematically in Fig. 2.

Site-directed Mutagenesis
Replacement of Pro 789 by Leu was carried out basically according to the procedure developed by Deng and Nickoloff (l992) using the following two primers. TZ-12 contains the desired single point mutation, 5Ј-CAAGAAGAGGGTGATTTCCGG-3Ј, which is complementary to the (Ϫ) strand of ␣ subunit coding region 2340 -2358 except for a single nucleotide mismatch (C 2351 and T 2351 ). TZ-14, 5Ј-CAGGCATGCTAGCT-TGAGT-3Ј, is complementary to the (Ϫ) strand of the multiple cloning sites of pGEM-3Z vector except for a single nucleotide mismatch (A 56 and T 56 of the pGEM-3Z sequence). This change converted a unique HindIII site to a unique NheI site.

Cell Culture, Transfection, Metabolic Labeling, and Sample Preparation for Immunoprecipitation Analyses
COS-1 cells were grown in six-well culture plates containing 10% calf serum Dulbecco's modified Eagle's medium. At mid-log phase cells were transfected with plasmids containing cDNAs according to the procedure provided by Life Technologies, Inc. using Lipofectamine. At 44 -48 h after the start of transfection, cells were metabolically labeled with [ 35 S]methionine (50 Ci/well) for 60 min and chased for 3 h. For immunoprecipitation analyses, at the end of chasing, the culture medium was saved and half was used for examining the amount of the secreted products, if any, and cells were washed three times with cold phosphate-buffered saline. After draining the residual buffer completely, cells were solubilized in 0.5 ml/well of 10 mM Tris-HCl, pH 7.4, 1% Nonidet P-40, 0.1% SDS, 0.15 M NaCl, 1 mM EDTA, 10 g/ml aprotinin (Lysis buffer). Culture medium was added SDS to be a final concentration of 1%, and incubated in boiling water for 5 min. Upon being cooled down to room temperature, the medium was mixed with an equal volume of 2 ϫ Lysis buffer. Both cell lysate and medium were subjected to immunoprecipitation as described previously (Xie and Morimoto, 1995).
For preparation of the total membrane fraction, cells from four wells were used. After transfection, followed by pulse-chasing as described above, cells were washed three times with cold phosphate-buffered saline, and scraped in 0.3 ml/well of 10 mM Tris-HCl, pH 7.4, 10 mM KCl, 0.5 mM MgCl 2 , 1 mM phenylmethylsulfonyl fluoride, 1 mM dithiothreitol, and 2 g/ml aprotinin and immediately homogenized in a tight fitting Dounce homogenizer with 15 strokes. The homogenate was centrifuged at 40,000 rpm for 30 min at 4°C in a Beckman Ti-50 rotor. The pellet was suspended in 200 l of 0.25 M sucrose, 10 mM Tris-HCl, pH 7.4, 3 mM Tetracaine, incubated at room temperature for 10 min, and then subjected to various treatments.
In vitro transcription, in vitro translation, isolation of microsomes from the translation mixture, removal of globin from the translation reaction mixture, Endo H treatment, immunoprecipitation with anti-CAT antibody, protease digestion, and SDS-PAGE were carried out as described by Xie and Morimoto (1995).

RESULTS
Membrane Insertion of the COOH-terminal Third of the ␣ Subunit Was Initiated by the Hydrophobic Segment H5 (Ile 782 -Ile 805 )-We used fusion proteins that contained different lengths of truncated ␣ subunit starting at Met 729 , instead of Met Ϫ5 , which is the initiator methionine in the wild type ␣ subunit (see Figs. 1 and 2). Because a molecular mass of the fusion proteins starting at Met Ϫ5 becomes larger and unglycosylated and glycosylated forms, which differ by only 3 kDa, becomes indistinguishable in SDS-PAGE. In addition, we observed that translocation efficiency of hydrophobic segments became progressively lower as the peptide length increased. Met 729 was chosen from the following reasons. (a) Since all hydrophobic segments in the COOH-terminal third occurred after a long cytoplasmic stretch, the first hydrophobic segment H5, which is located about 50 amino acid residues downstream of Met 729 , was considered as a signal/anchor type II (Sabatini and Adesnik, 1994), because the segment may function as an insertion signal that has an anchoring function and the NH 2terminal flanking region is predicted to be in the cytosol after insertion. (b) The H5 in this truncated form was considered to have the same topogenic properties as H5 in the wild type ␣ subunit. (c) ␣ subunit cDNA had two NcoI sites at Met Ϫ5 and Met 729 in frame.
To examine whether the H5 in the truncated subunits initiates translocation of the downstream peptides in living cells, we constructed two eukaryotic expression vectors: (a) pMT2-N5, which encoded a fusion protein (N-5) of NH 2 -terminal 981 amino acid residues (containing five hydrophobic segments, H1, H2, H3, H4, and H5 in the native context) and the CAT protein (219 amino acid residues); and (b) pMT2-C5, which encoded a fusion protein (C-5) consisting of a truncated sequence (89 amino acid residues including H5) starting at Met 729 and CAT (see Fig. 2). These plasmids were expressed FIG. 1. Amino acid sequence deduced from the nucleotide sequence of rat brain Na,K-ATPase ␣-1 cDNA (Hara et al., 1987) and its hydropathy plot. Amino acids are numbered beginning at Gly 1 of the mature protein and preceded by 5 amino acid residues present in the primary translation product. Hydrophobic segments H1-H9 are in boldface and are also indicated by a dashed line. The site at which truncated ␣ subunits are linked in frame with COOH-terminal reporter protein CAT at cDNA level is indicated by an upward arrow. Hydropathy plot was based on the Kyte and Doolittle algorithm (1982). Hydrophic regions are above and hydrophilic regions are below the x axis. Underline and closed circle indicate respectively core hydrophobic segments and the initiator methionine of the fusion proteins C-5, C*-5, C*-6, C*-7, C*-8, and C*-9.

FIG. 2. Schematic drawing of the constructs used in the study.
H, T7, SP6, and GH indicate hydrophobic segment, T7 RNA polymerase, SP6 RNA polymerase, and signal sequence of rat growth hormone, respectively. transiently in COS-1 cells.
As shown in Fig. 3, immunoprecipitation of the total membrane fractions by anti-CAT antibody yielded a single band with a molecular mass of about 110 kDa from pMT2-N5-transfected cells and two bands with molecular masses of 40 and 37 kDa from pMT2-C5-transfected cells. Upon treatment with Endo H, the 110-kDa protein band did not shift, but the 40-kDa protein band distinctly shifted to 37 kDa. The relative amount of the 40 kDa protein was about 50% of the total immunoprecipitates. When the membrane fractions were digested with proteases, followed by immunoprecipitation with anti-CAT antibody, two bands at 33 and 30 kDa were obtained from both transfected cells. The 33-kDa band shifted to a band at 30 kDa by Endo H treatment, but no shift was observed in the 30-kDa band by the same treatment. This membrane-protected 30-kDa protein corresponded to the expected size of the unglycosylated fusion protein C-5 (37 kDa) from which the NH 2 -terminal flanking region was cleaved. Since the consensus N-linked glycosylation site was only one in these fusion proteins and located within the CAT molecule downstream of H5, the results clearly demonstrated that H5 initiated translocation of both fusion proteins, and that the CAT protein was translocated across the membrane into the lumen where it was glycosylated. Since the membrane-protected 30-kDa protein was recovered by anti-CAT antibody as a major product from both pMT2-N5and pMT2-C5-transfected cells, the 37-kDa protein from pMT2-N5-transfected cells was apparently translocated but failed N-linked glycosylation for unknown reason. Therefore, the extent of glycosylation was low, but the majority of the fusion proteins were considered to be translocated.
H5 Translocated the Downstream Peptide in Vitro-Since the previous experiment showed that, in vivo, the H5 in the fusion protein C-5 translocated the downstream peptide as efficiently as the H5 in the fusion protein N-5, we examined whether the H5 segment in C-5 functioned as efficiently in a rabbit reticulocyte lysate cell free system. We used the plasmid pC5 that encoded the same fusion protein as encoded by pMT2C5 (see Fig. 2). As a reference for the efficiency of in vitro translocation, we used the plasmid pN1 that encoded a fusion protein consisting of the NH 2 -terminal 130 amino acid residues of the ␣ subunit and the CAT protein. The truncated ␣ subunit in this plasmid contained the hydrophobic segment H1, which was shown to initiate translocation very efficiently (Xie and Morimoto, 1995).
Translation of the pC5-derived mRNA in the absence of microsomes and subsequent immunoprecipitation by anti-CAT antibody yielded: 1) a major protein with a molecular mass of about 36 kDa (Fig. 4, C-5, lane 1), which was the predicted size of the primary translation product (defined as the product that has not been modified co-and/or post-translationally, and generally referred to as the translation product obtained in the absence of microsomes) of the fusion protein C-5, and 2) a few minor proteins that could be polypeptides initiated by internal methionine residues, or degradation products. When the same mRNA was translated in the presence of microsomes, an additional protein with a molecular mass of about 39 kDa was obtained (Fig. 4, C-5, lane 2). This protein shifted its molecular weight down to about 36 kDa by Endo H treatment (lane 3). From the density of the band, the glycosylated form was estimated as about 20% of the total immunoprecipitate. This was much smaller than the glycosylated form obtained in the pN1derived mRNA translation (about 80%) (Fig. 4, N-1, lanes 2  and 3).
When the translation product shown in lane 2 was digested with proteases in the absence of Triton X-100, and membraneprotected fragments were recovered by anti-CAT immunoprecipitation, three major bands with molecular masses of 39, 36, and 30 kDa were observed. From the size of the NH 2 -terminal flanking region of H5 in the fusion protein C-5 (53 amino acid residues) and the shift in molecular mass by N-linked glycosylation (about 3 kDa), the 30-kDa protein corresponded to the expected size of the unglycosylated form from which the NH 2terminal flanking region was cleaved. The 36-kDa form was probably an undigested, glycosylated form which was initiated by the internal Met 743 , although the size appeared to be smaller than the size calculated from the position of the internal methionine residue, or the undigested, nonglycosylated form. The largest molecular mass, 39 kDa, corresponded to the expected size of the undigested, glycosylated form. No fragments were recovered by immunoprecipitation upon digestion in the presence of Triton X-100 (C-5, lane 6). However, the possibility that the peptides in lane 5 were produced by incomplete digestion could not be ruled out. That these proteins were inserted into microsomal membranes was confirmed by their resistance to alkaline (pH 11.5) extraction according to Fujiki et al. (1982), using membrane insertion of the fusion protein N-1 and X-c as positive and negative controls, respectively (data not shown here).
Mutation of One of the Proline Residues in H5 Greatly Increased Its Translocation Efficiency-The translocation efficiency of H5 in the fusion protein C-5 was much lower than that of H1 in the fusion protein N-1 (Fig. 4, C-5 and N-1). The NH 2 -terminal flanking region of these two fusion proteins appeared to be similar, and the hydrophobicities of H1 and H5 were also very similar, except that H5 had 3 proline residues FIG. 3. Hydrophobic segment H5 initiates membrane translocation of the newly synthesized COOH-terminal third of the ␣ subunit. Top, total membrane fractions were prepared from cells that had been transiently expressed with pMT2N5 or pMT2C5, followed by metabolic labeling with [ 35 S]methionine as described under "Experimental Procedures." Each membrane fraction was divided into two aliquots. One aliquot was treated with proteases and the other was used as a control. Both aliquots were subjected to immunoprecipitation. Each immunoprecipitate was dissolved in 30 l of 1% SDS, 50 mM dithiothreitol, incubated in boiling water for 3 min, and then mixed with an equal volume of 0.2 M citrate buffer, pH 5.5. Each aliquot was further divided into two. One was treated with Endo H, and the other was the control. After incubation at 37°C overnight, samples were analyzed in SDS-PAGE as described under "Experimental Procedures." Bottom, schematic drawing shows the orientation of the inserted chimera before (ϪP) and after (ϩP) protease digestion. Solid hearts, N, and C indicate glycosylated forms, NH 2 terminus, and COOH terminus, respectively. while H1 did not. Proline is known as an ␣ helix breaker and could account for the low translocation efficiency of H5. If this were the case, change of 1 or more proline residues in H5 should increase its translocation efficiency. In fact, in parallel to this work, Homareda et al. (1992) observed improvement of the translocation efficiency of H5 by changing proline residues to hydrophobic amino acid residues. It was desirable to have an H5 with higher translocation efficiency for examination of the topogenic properties of the hydrophobic segments H6, H7, H8, and H9 that followed it in the native context. We mutated 1 proline residue at position 784 to leucine and examined the translocation efficiency. Among 3 proline residues, proline 784 was chosen because it was located very close to the beginning of the hydrophobic segment H5 and therefore this would cause the least effect on the structure of H5 in the lipid bilayer. (Constructs and fusion proteins that contain proline-mutated H5 are denoted by "*" like pC*5 and C*-5, respectively.) The immunoprecipitates from the translation of the pC*5 transcript in the presence of microsomes contained two major proteins with molecular masses of 36 and 39 kDa at a mass ratio of about 1: 1 (Fig. 4, C*-5, lane 2). Endo H treatment resulted in the shift of 39 kDa band to 36 kDa (lane 3), indicating that the relative amount of the glycosylated form was about 50%. This value was much higher than the value obtained in the pC5 translation. Interestingly, the Endo H treatment also shifted the molecular mass of the 36-kDa protein to 33 kDa. This shift was presumably due to deglycosylation of a fusion protein that was initiated by an internal Met. These results clearly demonstrated that the mutation of Pro 784 improved the translocation efficiency of H5. Thus, the plasmids that encoded fusion proteins containing "Pro 784 -mutated H5" (referred to as H*5) were used for testing the topogenic prop-erties of the downstream hydrophobic segments H6 -H9.
No Transmembrane Domains Existed between H5 and H6 -The H5-H6 loop is a relatively long stretch, in which a residue (Arg 832 ) was reported to be present in the cytosolic side (Karlish et al., 1993), which suggests there may be a transmembrane span between the ligation site of the reporter protein and H6. To examine this possibility, the plasmid pCB*5-6 was constructed. This plasmid encoded a fusion protein that was similar to C*5 but CAT was ligated at residue Ile 845 instead of Ala 818 .
Two proteins (39 and 42 kDa) were immunoprecipitated by anti-CAT antibody from translation of the pCB*5-6 transcript in the presence of microsomes (Fig. 4, B*5-6, lane 2). The 42-kDa protein was converted by Endo-H treatment to 39 kDa, which was the expected size of the primary translation product of the fusion protein (lane 3). From the comparison of the translocation efficiency between fusion proteins C*-5 and B*5-6, the H5-H6 loop was considered to be translocated into the ER lumen. Therefore, we concluded that there were no hydrophobic segments within the loop that functioned as a halt transfer signal. However, it cannot be ruled out completely the possibility that some regions in the loop may be inserted posttranslationally to microsomal membranes together with H6, H7, and/or H8.
Among Hydrophobic Segments H6 (Leu 844 -Ala 869 ), H7 (Phe 911 -Ile 931 ), H8 (Ile 948 -Leu 973 ), and H9 (Thr 981 -Tyr 996 ), Only H9 Functioned as a Halt Transfer Signal-The primary translation products of the pC*6, pC*7, pC*8, and pC*9 transcripts had molecular masses of 43, 49, 53, and 57 kDa, respectively (Fig. 5, lane 1). Translation in the presence of microsomes (lane 2) produced two additional proteins in C*-6 and C*-7 and one additional protein in C*-8, but no additional FIG. 4. Mutation of 1 proline residue in the hydrophobic segment H5 greatly increases its translocation ability in vitro. Left column, transcripts of pN-1, pC-5, pC*-5, and pCB*5-6 were translated in the absence (lane 1) and presence (lanes 2-6) of microsomes. The latter products were divided into two portions. One portion was used for Endo H treatment (lanes 2 and 3), and the other portion was further divided into three (lanes 4, 5, and 6) and used for protease digestion. After treatment, the samples were subjected to immunoprecipitation, followed by SDS-PAGE as described under "Experimental Procedures." Open circles indicate the position of glycosylated forms. Right column, schematic drawing presents the orientation of the inserted fusion proteins of lane 2. Solid hearts, N, and C indicate glycosylated form, NH 2 terminus, and COOH terminus, respectively.
protein in C*-9. These additional proteins were converted by Endo H treatment to the size of their respective primary translation products (lane 3). Majority of the glycosylated forms of C*-6, C*-7, and C*-8 were not digested when treated with proteases in the absence of Triton X-100, but C*-9 was digested into fragments that could not immunoprecipitate with anti-CAT antibodies under the same condition (lane 5). The glycosylated forms were, however, digested in the presence of Triton X-100 (lane 6). These results demonstrated that H6, H7, and H8 were not capable of halting the translocation and were translocated into the ER lumen where the downstream CAT peptide was glycosylated. Thus, the translocation initiated by H*5 was halted only by H9.
Topogenic Properties of the Hydrophobic Segments Observed in the Native Context Seemed to Be Intrinsic-To examine whether the topogenic properties of H5, H6, H7, H8, and H9 observed in Figs. 4 and 5 were the same when examined in a different context, we constructed plasmids that encoded a fusion protein consisting of 66 amino acid residues (including the cleavable signal sequence) of the NH 2 -terminal rat growth hormone and one of the five hydrophobic segments, which was linked to the CAT protein. These plasmids were defined as pX5, pX6, pX7, pX8, and pX9 according to the location of the hydrophobic segment in the ␣ subunit (Fig. 2). Two additional plasmids were constructed: pXc, which encoded a fusion protein of the NH 2 -terminal 66 amino acid residues of rat growth hormone and CAT protein; and pX4, which encoded a fusion protein of the NH 2 -terminal 66 amino acid residues of rat growth hormone and the hydrophobic segment H4, linked to CAT. Since the H4 in the NH 2 -terminal third of Na,K-ATPase ␣ subunit has been confirmed to function as a halt transfer signal (Xie and Morimoto, 1995), pX4 served as a positive control and pXc was used as a negative control.
As shown in Fig. 6, a single protein with the same molecular weight as the primary translation product (33 kDa) was obtained from the translation of pXc transcript in the presence of microsomes (X-c, lane 2). This protein was converted to a 30-kDa protein by Endo H treatment (X-c, lane 3). Protease treatment in the absence of Triton X-100 did not cause any change in the molecular mass of the protein, but the same treatment in FIG. 5. Membrane insertion of newly synthesized COOH-terminal third of the ␣ subunit that was initiated by hydrophobic segment H5 was halted by hydrophobic segment H9. Left column, transcripts of pC*-6, pC*7, pC*8, and pC*9 were translated in the absence (lane 1) and presence (lanes 2-6) of microsomes. The latter products were divided into two parts and analyzed as described in the legend to Fig.  4. Open triangles indicate the position of the glycosylated forms. Right column, schematic drawing presents the orientation of the inserted fusion proteins of lane 2. Solid hearts and diamond indicate glysocylated and unglycosylated forms, respectively. N and C represent NH 2 and COOH termini, respectively. the presence of detergent digested the protein (lanes 4 and 5). Since the signal sequence of rat growth hormone has 26 amino acid residues and the shift of molecular mass in SDS-PAGE due to one N-linked glycosylation was about 3 kDa, the translation product in the presence of microsomes would be almost the same size as the primary translation product if the translocation of a fusion protein initiated by the signal sequence was not halted by the downstream segment, and this product would be completely protected from digestion by exogenously added proteases. However, if the translocation was halted by the downstream segment, the translation product would be smaller by about 3 kDa than the primary translation product and digested by exogenously added proteases into fragments that could not immunoprecipitate with anti-CAT antibody. Therefore, the results showed that the translocation of the fusion protein X-c was initiated by the cleavable signal sequence and the following CAT protein was translocated into the ER lumen where it was glycosylated.
The primary translation products of pX7, pX8, and pX9 transcripts showed a single protein with a molecular mass of 36 kDa (Fig. 6, lane 1). The immunoprecipitates from the pX5 and pX6 translation products showed two major proteins and one minor protein (lane 1). The largest ones corresponded to the expected size of the respective fusion proteins; the other two could have been the products that were initiated by internal methionines or degradation products. Translation of the pX5 and pX8 transcripts in the presence of microsomes yielded, respectively, the same size of the protein as obtained in the absence of microsomes (lane 2). Translation of the pX6 and pX7 transcripts yielded additional proteins (39 kDa in fusion proteins X-6 and X-7, 33 kDa in fusion protein X-7) and the translation of the pX9 transcript yielded a single protein with smaller molecular mass (lane 2). Endo H treatment resulted in the decrease by about 3 or 6 kDa of the molecular mass of each fusion protein, except the pX9 translation products, whose molecular mass was not changed (lane 3). Protease treatment did not cause any change in the molecular mass of any of the fusion protein except the smaller protein of the pX5 translation products (X-5, lanes 2 and 4) and the pX9 translation products (X-9, lane 4), both of which were digested into the fragments that did not immunoprecipitate by anti-CAT antibody. These data have confirmed the previous observation that among hydrophobic segments H6 -H9, only H9 has a halt transfer signal function. It should be pointed out here that H5 functions as an insertion signal like H1, but, as opposed to H1, H5 has a weak halt transfer function as seen in Fig. 6 (X-5, lanes 2 and 4).
Hydrophobic Segments H6, H7, and H8 That Were Translocated into the ER Lumen Seemed to Remain in the ER without Being Secreted-To examine whether H5, H6, H7, H8, and H9 have the same topogenic properties in living cells, plasmids FIG. 6. Topogenic nature of the hydrophobic segments observed in the native context are intrinsic. Top, transcripts of the plasmids pX-c, pX-4, pX-5, pX-6, pX-7, pX-8, and pX-9 were translated in the absence (lane 1) and presence (lanes 2-5) of microsomes. The products translated in the presence of microsomes were divided into two portions. One portion was used for Endo H treatment (lanes 2 and 3); the other portion was used for protease digestion (lanes 4 and 5). After treatment, the samples were subjected to immunoprecipitation, followed by SDS-PAGE as described under "Experimental Procedures." Open circle (X-9) indicates the position of the inserted fusion protein from which the signal sequence was cleaved. Bottom, schematic drawing shows the orientation of the inserted fusion proteins before (no mark) or after (ϩP) protease digestion. Solid hearts and diamonds represent glycosylated and unglycosylated forms, respectively. N, C, and GH represent NH 2 terminus, COOH termini, and signal sequence of rat growth hormone, respectively. pXc, pX5, pX6, pX7, pX8, and pX9 were expressed transiently in COS cells. The immunoprecipitates from the pXc-expressing cells showed a single protein with a molecular mass of 33 kDa, which corresponded to the expected size of the unglycosylated form of the fusion protein X-c (Fig. 7, lane C Ϫ ). More than 50% of the 33-kDa protein was converted by Endo H treatment to 30-kDa protein (lane C ϩ ). The immunoprecipitates from the culture medium showed also a single protein (33 kDa), but Endo H treatment did not cause any shift in the molecular mass (lanes M Ϫ and M ϩ ). These results demonstrated that the fusion protein X-c was translocated into the ER lumen where the reporter protein was glycosylated and then secreted into culture medium like a regular secretory proteins.
The immunoprecipitates from cells expressing plasmids pX5, pX8, and pX9 showed a single protein with molecular masses of 37, 36, and 36 kDa, respectively, and those from cells expressed with pX6 and pX7 showed three proteins with molecular masses of 33, 36, and 39 kDa (lane C Ϫ ). Endo H treatment did cause a decrease by 3 or 6 kDa in the molecular mass of each of the products except for the pX9 translation product, which remained unchanged (lane C ϩ ). No proteins were recovered from their culture media by the immunoprecipitation (lane M Ϫ ), even after a 5-h chase (data not shown). These results have confirmed that all hydrophobic segments show the same topogenic properties both in the native context and in the different context. Interestingly, the fusion protein containing either H5, H6, H7, or H8 was totally translocated into the ER lumen, but no products were secreted into the culture medium, while the pXc products were secreted like regular secretory proteins. The only difference between the pXc translation prod-uct and pX5, -6, -7, and -8 translation products was that the latter group of the fusion proteins contained a hydrophobic segment (about 20 amino acid residues) at the ligation site between the truncated growth hormone peptide and the CAT protein. Since the glycosylated form was sensitive to Endo H treatment, the products may remain in the ER lumen by unknown mechanisms.

DISCUSSION
Unexpected Topogenic Properties of the Hydrophobic Segments in the COOH-terminal Third of ␣ Subunit-Our data demonstrated the following. (a) Membrane insertion of the COOH-terminal third of the ␣ subunit was initiated by H5 and terminated by H9 (Figs. 4 and 5). (b) H6, H7 and H8, located between H5 and H9, lacked halt transfer function and were translocated across the membrane into the ER lumen (Fig. 5).
(c) H5 functioned as an insertion signal, but its halt transfer signal activity was weak (Fig. 6). This was quite a different property from H1 that initiated membrane insertion of the NH 2 -terminal third of the ␣ subunit as shown by Xie and Morimoto (1995). (d) When H6, H7, and H8 were individually translocated into the ER lumen, they appeared to remain there without being secreted. These results clearly showed that the hydrophobic segments in the COOH-terminal third had very different membrane topology from those predicted by hydropathy analysis and other studies.
Why H6, H7, and H8 did not function as a halt transfer signal can not be explained by factors that are usually considered to be determinants of topogenic properties of hydrophobic segments. For example, the average hydrophobicity indices of the hydrophobic segments calculated using the residue hydrophobicity index taken from the data by Kyte and Doolittle (1982), were 2.19 (H1), 1.87 (H2), 2.42 (H3), 2.00 (H4), 2.07 (H5), 1.48 (H6), 1.69 (H7), 1.43 (H8), and 1.17 (H9). The values for H6, H7, and H8, which had no topogenic function, were significantly lower than those for H1, H2, H3, H4, and H5, which had topogenic function. However, the values for H6, H7, and H8 were higher than the value for H9, which had topogenic function. Thus, the topogenic properties of hydrophobic segments could not be defined by their average hydrophobic indices. Distribution of charged amino acids in the NH 2 -and COOH-terminal flanking regions of each hydrophobic segment is also considered to be an important factor that affects the topogenic properties of the segment (review by von Heijne (1994)). However, as seen in Fig. 1, there were no distinct differences in the distribution of charged amino acids of the flanking regions among the segments that had topogenic function and those who did not. These observations suggest that there must be other factors that determine the topogenic properties of the hydrophobic segments H6, H7 and H8. Similar examples have been reported with CHIP28 water channel protein (Skach et al., 1994), human P-glycoprotein (Skach et al., 1993) and H ϩ ,K ϩ -ATPase (Bamberg and Sachs, 1994).
Another hydrophobic segment that displayed unexpected topogenic properties was H5. This segment could initiate translocation of COOH-terminal third of ␣ subunit, but the efficiency of translocation was much lower than that of H1 (Fig. 4), and the halt transfer function was much weaker than that of H1 (Figs. 4 and 6). When these two segments were compared, their hydrophobicity indices were similar, but H5 contained 3 proline residues while H1 did not contain any. Since proline is known as an ␣ helix breaker, the unusual topogenic properties of H5 may be to some extent due to the proline residues. This possibility was supported by the present observation (Fig. 4) that mutation of Pro 784 to leucine resulted in a significant increase in the translocation efficiency (Fig. 4). Homareda et al. (1992Homareda et al. ( , 1993 observed that increase in translocation efficiency was FIG. 7. Hydrophobic segments H6, H7, and H8 that were translocated in the ER lumen seem to remain there. COS-1 cells were transfected with plasmids pX-c, pX-5, pX-6, pX-7, pX-8, and pX-9. After metabolic labeling with [ 35 S]methionine, followed by chasing, cell lysates were subjected to immunoprecipitation with anti-CAT antibody and half of the immunoprecipitates was treated with Endo H. At the same time the culture media were analyzed in the same way to examine whether the translocated peptides may be secreted. All the samples were analyzed in SDS-PAGE as described under "Experimental Procedures." C and M represent cell lysate and medium, respectively.
proportional to the number of proline residues mutated.
When the 10 residues flanking the NH 2 termini of H1 and H5 were compared, H1 had 2 positively and no negatively charged amino acid residues, but H5 had no positively and 1 negatively charged residue (Fig. 1). Since H5 initiates translocation of the COOH-terminal third of the ␣ subunit after a long cytoplasmic stretch, it is reasonably considered to have the same topogenic function as H1 in membrane insertion of its downstream peptide. Therefore, the positively charged amino acid residues in the NH 2 -terminal flanking region may play an important role in membrane insertion as indicated by several investigators (review by von Heijne (1994)). However, the NH 2 -terminal flanking region of H5 had only negatively charged residue. Such a distinct difference from H1 might be at least partially accounted for the unexpected topogenic properties of H5.
Sidedness of the Loop between Two Adjacent Hydrophobic Segments-Contradictory results have been reported on the membrane location of the sequence between H5 and H6. Our data (Fig. 4) demonstrated that the H5-H6 loop was located on the luminal side of the membrane, which supports the results of Mohraz et al. (1994). These authors demonstrated using immunoelectron microscopy with site specific antibody VG2 that the domain Val 812 -Ile 826 was located on the extracellular side. However, several other investigators reported that the loop was located in the cytosolic side of the membrane. Those include the following. (a) Protease digestion of Na,K-ATPase in the right side out membrane vesicles in the presence or absence of detergent showed that Arg 832 was located on the cytosolic side of the membrane (Karlish et al., 1993). (b) Immunochemical localization of the oligopeptide specific antibody on the two dimensional crystal of the enzyme showed that the domain Leu 815 -Arg 830 was located in the cytosolic side of the membrane (Ning et al., 1993). (c) Yoon and Guidotti (1994) demonstrated that the domain Leu 815 -Arg 830 was located in the cytoplasmic side of the membrane using "epitope insertion" also used by Canfield and Levenson (1993). In this method, the nucleotide sequence that encoded an immunoreactive peptide was inserted into various region within the ␣ subunit cDNA, and the location of the peptide was determined by immunofluorescence method using cells that had been transiently expressed with the mutated cDNA.
In addition to the present data, five independent studies showed that the loop between H6 and H7 was located in the extracellular side. These studies are as follows. (a) Mohraz et al. (1994) showed that the domain Trp 889 -Phe 911 was located on the extracellular side by immunogold method using site specific antibody IIC9 and large membrane fragments rich in Na,K-ATPase. (b) Ning et al. (1993) localized the domain Asn 889 -Gln 903 on the extracellular side by immunoelectron microscopy using an epitope-specific antibody. (c) Yoon and Guidotti (1994) demonstrated extracellular location of Gln 900 . (d) Schultheis et al. (1993) showed that Arg 880 was a portion of ouabain binding site in the COOH-terminal third. (e)  demonstrated that a stretch of 26 amino acid residues (Asn 831 -Ala 916 ), which included the beginning of H7, were involved in the assembly with ␤ subunit.
Contradictory results have been also reported on the sidedness of the H7-H8 loop. Our data (Fig. 5) that showed the extracellular localization of the loop agreed with the results by Canfield and Levenson (1993). These authors found that using an epitope insertion a peptide inserted at Gly 943 was positive to the peptide-specific antibody in the absence of detergent when the mutated cDNA was expressed transiently in cells. However, Yoon and Guidotti (1994) demonstrated that using the same "epitope insertion" the residues Val 939 and Phe 940 were located in the cytoplasmic side. Furthermore, Fisone et al. (1994) found that Ser 936 was phosphorylated by protein kinase A, which indicates the cytoplasmic location of this amino acid residue.
Concerning the sidedness of the H8 -H9 loop, our data indicated that it was in the extracellular side, but Canfield and Levenson (1993) showed by using "epitope insertion" method that Leu 978 was located on the cytoplasmic side. The sidedness of the COOH-terminal flanking region of H9 was shown to be in the cytosolic side of the membrane by immunochemical method (Antolovic et al., 1991), epitope insertion method (Canfield and Levenson, 1993), and the present topogenic analysis of the hydrophobic segments.
Most of contradictory results stem from studies on the localization of antigenic sites by immunofluorescence microscopy of intact and permeabilized cells, and studies on the protease digestion of Na,K-ATPase-rich vesicles in the presence or absence of detergent. Positive results obtained by these methods in the absence of detergent can be safely interpreted as extracellular location. However, those obtained only in the presence of detergent can not always be interpreted as intracellular location. For example, if an antigenic site on the extracellular side is folded in such a way that it is covered by other domains of the peptide, it will not react with antibodies in the absence of detergent. However, when cells are permeabilized, conformation of membrane spanning helices may induce unfolding of the peptide which covers the antigenic site. Even the antigenic sites that are partially embedded in the lipid bilayer from the surface but not exposed to the cytosolic side will become exposed.
A Model for the Biogenesis of Na,K-ATPase ␣ Subunit-In a previous study, we reported membrane topology of the NH 2terminal third of Na,K-ATPase ␣ subunit (Xie and Morimoto, 1995). Based on the present and the previous studies, we propose a model for biogenesis of the ␣ subunit in which the first hydrophobic segment H1 targets the nascent chain ribosome complex to the ER membrane and initiates translocation of the COOH-terminal flanking region. This translocation is terminated by the second hydrophobic segment H2. The second cycle of translocation is initiated by the third hydrophobic segment H3 and is terminated by the fourth hydrophobic segment H4. After a long cytoplasmic stretch of the middle third of the subunit, the third cycle of translocation is initiated by the fifth hydrophobic segment H5. This translocation continues until it is terminated by the ninth hydrophobic segment H9 that is located near the COOH terminus of the subunit. Hydrophobic segments H6, H7, and H8, which were located between H5 and H9, are translocated into the ER lumen. Thus, six membranespanning helices are established. H6, H7, and H8 may associate on the luminal surface of ER or may be embedded from the lumen into the lipid bilayer of the ER membrane during translocation or post-translationally shortly after the translocation is terminated.
The unique feature of this model is that the membrane insertion of the ␣ subunit is achieved by combination of three cycles of alternate initiation and termination of translocation to establish transmembrane spans and co-or post-translocational interaction of H6, H7, and H8 within the luminal surface of the ER to establish partially embedded membrane spans (Fig. 8).
Recently Bamberg and Sachs (1994) reported a study of the topological analysis of H ϩ ,K ϩ -ATPase ␣ subunit using a method similar to the one used in the present study to identify transmembrane segments of the subunit. 10 hydrophobic segments (M1-M10) were chosen from either biochemical analyses or hydropathy plots, and their topogenic properties were examined. The results showed that four segments M1-M4 in the NH 2 -terminal portion were inserted into microsomal mem-branes as in the case of four NH 2 -terminal hydrophobic segments (H1-H4) of Na,K-ATPase ␣ subunit. Unexpectedly, M5, M6, and M7 did not have topogenic function like H6, H7, and H8 of Na,K-ATPase. M8 and M10 had a halt transfer function but not an insertion signal function, and M9 functioned as both an insertion and a halt transfer signals. Based on these results, the authors have proposed a model that H ϩ ,K ϩ -ATPase is inserted co-translationally (M1-M4 and M9 -M10) by three cycles of alternate initiation and termination of translocation and post-translationally (M5-M8) by a different mechanism. This indicates that both H ϩ ,K ϩ -ATPase ␣ subunit and Na,K-ATPase ␣ subunit seem to be inserted into microsomal membranes by similar mechanisms except that the insertion signal of the third cycle of translocation is M9 for H ϩ, K ϩ -ATPase and H5 for Na,K-ATPase and the sidedness of the post-translational insertion is cytosol for H ϩ ,K ϩ -ATPase and endoplasmic reticular lumen for Na,K-ATPase. An interesting question is whether such mechanisms of membrane insertion are common to P-type ATPase. To answer to this question, topological studies of the other P-type ATPases using the method similar to the one used by Bamberg and Sachs (1994) and us will be required. In fact, there was a recent report on the membrane topology of the Neurospora plasma membrane H ϩ -ATPase by Lin and Addison (1995). These authors described that the ATPase had 10 putative transmembrane segments M1-M10 and that fusion proteins with various combinations of transmembrane segments were all integrated co-translationally into microsomal membranes. However, as opposed to their conclusion, the results were not supportive primarily because of the lack of combination of transmembrane segments for constructing fusion proteins. Interestingly, the results are rather suggestive of the possibility that the enzyme may also be inserted into microsomal membranes by the mechanisms similar to those of H ϩ ,K ϩ -ATPase and Na,K-ATPase.
Implication for the Membrane Topology of the ␣ Subunit-Recently several studies have attempted to resolve the membrane topology of the COOH-terminal third of the ␣ subunit. These include the following. (a) 19-kDa fragment of the COOHterminal portion may be involved in occlusion and cation transport (Karlish et al., 1990). (b) Some hydrophobic segments in the ␣ subunit may be partially or totally exposed extracellularly (Modyanov et al., 1991). (c) A channel-like structure may be formed . (d) Glu 781 , which is located just before the hydrophobic segment H5, and Asp 928 , which is located near the COOH terminus of H7, are considered to be cation binding sites (see review by Lingrel and Kuntzweiler (1994)). (e) COOH-terminal third of the ␣ subunit popped out of the membrane when membranes associated enzyme was heated to 50°C while the NH 2 -terminal third remained in the membrane associated (Arystarkhova et al., 1995). (f) The protein mass of the ␣ subunit in the lipid is about 40%, which is close to the relative area occupied by 10 transmembrane spans (see review by Lingrel and Kuntzweiler (1994)). These observations suggest that hydrophobic segments H6, H7, and H8 are not simply attached on the luminal surface of the ER, but are at least partially embedded in the lipid bilayer and form a complex structure with the other hydrophobic segments.
What membrane topology of the COOH-terminal third can be predicted from the present studies and the primary structure of the ␣ subunit shown in Fig. 1? Hydrophobic segment H6 has 2 consecutive glycine residues in the middle, and both flanking regions of the segment are definitely located on the extracellular side (Fig. 4). Since glycine is an ␣-helix breaker, the segment may be embedded in the lipid bilayer from the lumen forming a hairpin structure in which the two glycine residues make a turn. The hydrophobic segment H7 has 1 Asp near the COOH terminus and H8 has 2 Glu residues in the middle of the segment. The H7-H8 loop has 16 amino acid residues, of which 5 are positively charged and 2 are prolines, but no negatively charged amino acid residues are present. The H8 -H9 loop has 7 amino acid residues, of which 2 are positively charged and 2 are proline residues, but no negatively charged amino acid residues are present. Since H9 was established as the last transmembrane segment whose COOH-terminal flanking region was on the cytoplasmic side, the H8 -H9 loop must be on the luminal side. Therefore, if H7 and H8 are embedded in such a way that the NH 2 terminus of H7 and COOH terminus of H8 face the lumen, the H7-H8 loop, which is hydrophilic, would be folded between H7 and H8, possibly by interaction between negatively charged amino acids in the segments and positively charged amino acids in the loop. The embedded H7 and H8 may form a complex with the other hydrophobic segments in the COOH-terminal third as well as hydrophobic segments in the NH 2 -terminal third.
Among hydrophobic segments that have topogenic function, only H5 has 3 proline residues. Because of this, the segment had weak halt signal function. As the COOH-terminal third of ␣ subunit may be involved in occlusion and cation transport (Karlish et al., 1990), and this portion pops out when incubated at 50°C (Arystarkhova et al., 1995), it is reasonable to consider that the complex formed by H6, H7, and H8 is movable within the lipid bilayer and the nature of the hydrophobic segment H5 may facilitate the movement of the complex. Thus, the unique nature of H5 found in this study may contribute not only to the folding process of the COOH-terminal third of the peptide during the enzyme biogenesis but also to the movement of the folded peptide complex within the membrane, coupling it to the gating activity of the enzyme. Such complexity in the structurefunction relationship of the COOH-terminal third of the ␣ subunit might have given rise, at least partially, to the contradictory results on the sidedness of the loops. Further studies on the role of the 3 proline residues in the topogenic function of H5, the role of hydrophobic segments H5, H6, H7, H8, and H9 in the folding process during enzyme biogenesis, and the involvement of ␤ subunit in these processes, which are currently in progress, will provide crucial information on the conformational changes of the folded peptide complex that accompanies the ion transport function of the enzyme. Acknowledgment-We thank Dr. David D. Sabatini for encouragement and support throughout this work. We also thank Dr. Y. Hala for the rat brain Na,K-ATPase ␣1 subunit cDNA, J. Culkin and F. Forcino   FIG. 8. A model for the membrane topology of the Na,K-ATPase ␣ subunit. Hatched structures with numbers 1-9 represent hydrophobic segments H1-H9 (see Fig. 1). N and C indicate NH 2 and COOH termini, respectively. Lengths of NH 2 -and COOH-terminal flanking regions and those of the loops between two adjacent hydrophobic segments are approximately proportional to the number of the constituent amino acid residues.
for their excellent photographic work, and H. Plesken for her excellent artwork. We are very grateful to Dr. Manijeh Mohraz for critical reading of the manuscript and helpful discussion and Dr. J. Shafland for advice during preparation of the manuscript.