Determinants of Topogenesis and Glycosylation of Type II Membrane Proteins ANALYSIS OF Na,K-ATPase b 1 AND b 3 SUBUNITS BY GLYCOSYLATION MAPPING*

The structural and molecular determinants that gov-ern the correct membrane insertion and folding of membrane proteins are still ill-defined. By following the addition of sugar chains to engineered glycosylation sites (glycosylation mapping) in Na,K-ATPase b isoforms expressed in vitro and in Xenopus oocytes, in combination with biochemical techniques, we have defined the C-terminal end of the transmembrane domain of these type II proteins. N-terminal truncation and the removal of a single charged residue at the N-terminal start of the putative transmembrane domain influence the proper positioning of the transmembrane domain in the membrane as reflected by a repositioning of the transmembrane domain, the exposure of a putative cryptic signal peptidase cleavage site, and the production of protein species unable to insert into the membrane. Glycosylation mapping in vivo revealed that the degree of glycosylation at acceptor sites located close to the membrane increases with the time proteins spend in the endoplasmic reticulum. Furthermore, core sugars added to such acceptor sites cannot be processed to fully glycosylated species even when the protein is transported to the cell surface. Thus, the glycosylation mapping strategy applied in intact cells is a useful tool for the study of determinants for the correct membrane insertion of

The structural and molecular determinants that govern the correct membrane insertion and folding of membrane proteins are still ill-defined. By following the addition of sugar chains to engineered glycosylation sites (glycosylation mapping) in Na,K-ATPase ␤ isoforms expressed in vitro and in Xenopus oocytes, in combination with biochemical techniques, we have defined the Cterminal end of the transmembrane domain of these type II proteins. N-terminal truncation and the removal of a single charged residue at the N-terminal start of the putative transmembrane domain influence the proper positioning of the transmembrane domain in the membrane as reflected by a repositioning of the transmembrane domain, the exposure of a putative cryptic signal peptidase cleavage site, and the production of protein species unable to insert into the membrane. Glycosylation mapping in vivo revealed that the degree of glycosylation at acceptor sites located close to the membrane increases with the time proteins spend in the endoplasmic reticulum. Furthermore, core sugars added to such acceptor sites cannot be processed to fully glycosylated species even when the protein is transported to the cell surface. Thus, the glycosylation mapping strategy applied in intact cells is a useful tool for the study of determinants for the correct membrane insertion of type II and probably other membrane proteins, as well as for the processing of sugar chains in glycoproteins.
Subunit assembly of oligomeric membrane proteins often involves multiple but poorly understood interactions between the different subunits (for review see Ref. 1). Na,K-ATPase and H,K-ATPases are interesting model proteins for the study of functional roles of different subunit interaction sites, because these two enzymes are the only members of the cation-transporting P-type ATPase that are oligomeric and contain, in addition to the catalytic ␣ subunit, a second subunit, the ␤ subunit in the functionally active enzyme. Similar to most other P-type ATPases, the ␣ subunits of Na,K-and H,K-AT-Pases are polytopic membrane proteins with 10 transmembrane segments that carry the main functional properties. The ␤ subunits associated with Na,K-and H,K-ATPase ␣ subunits are type II membrane proteins with a short cytoplasmic N terminus, a single transmembrane domain, and a large glycosylated ectodomain. To date, three Na,K-ATPase and one gastric H,K-ATPase ␤ isoforms have been identified, which exhibit a similar domain structure but a low degree of sequence identity of 20 -35%. At present, we know that ␤ subunits have several functions that may be finely regulated by different isoforms. A primary role of the ␤ subunit is to support the maturation of the Na,K-and H,K-ATPase ␣ subunits, which, in contrast to other P-type ATPases, are stably expressed and become functionally active only when properly associated with a ␤ subunit (2). In addition to this chaperone function, the ␤ subunit has also been shown to influence the transport properties of the mature Na,K-ATPase, e.g. its apparent affinities for K ϩ and Na ϩ (3)(4)(5)(6)(7).
Interaction sites that have been identified in Na,K-and H,K-ATPase ␣ and ␤ subunits involve the extracellular, transmembrane, and cytoplasmic domains. By using the two-hybrid system, an extracellular ␤ domain adjacent to the transmembrane segment has been shown to interact with the extracellular loop between transmembrane segments M7 and M8 of the ␣ subunit of Na,K-ATPase (8) and H,K-ATPase (9). A ␤ sheet-like structure formed by the 10 most C-terminal amino acids most likely represents another ␣-interaction site in the ectodomain of the ␤ subunit (10). Interactions in the ectodomains of ␣ and ␤ subunits are important for the correct folding and the stabilization of the ␣ subunit of oligomeric P-type ATPases (11,12). Furthermore, studies performed on chimeras formed between Na,K-ATPase ␤ 1 and gastric H,K-ATPase ␤ subunits suggest that interactions with the ␤-ectodomain are responsible for the differences observed in the transport properties of the Na,K-ATPase associated with different ␤ isoforms (5,7,13). Evidence for interactions between transmembrane segments of the Na,K-ATPase ␣ and ␤ subunits has been obtained by crosslinking experiments (14,15), but the functional role of these interactions is not yet defined. Analysis of chimeric Na,K-ATPase/H,K-ATPase ␤ subunits suggests that only transmembrane interactions of Na,K-ATPase ␤ but not that of H,K-ATPase ␤ subunits permit the correct folding and ER 1 exit of the Na,K-ATPase ␣ subunit (7,13). Finally, the functional implications of subunit interactions in the cytoplasmic domains, which are supported by proteolysis protection assays (2,16), are the least well understood. Indeed, truncation of the N terminus of Na,K-ATPase ␤ 1 subunit does not impede ␣ interaction and stabilization but significantly decreases the appar-ent K ϩ and/or Na ϩ affinity of the Na,K-ATPase (2,6,7). However, results of a detailed mutational analysis indicated that the ␤ N terminus may not be directly involved in the functional effects observed after complete N-terminal truncation (7), but rather that N-terminal truncation could indirectly affect another domain of the ␤ subunit.
To better understand the structural and functional roles of ␣-interaction domains in Na,K-ATPase ␤ subunits, we aimed in this study to 1) define the membrane-spanning domain of the ␤ subunit by identifying the amino acids that actually make up the transmembrane ␣-helix of Na,K-ATPase ␤ subunits and 2) probe potential changes in the transmembrane domain after N-terminal truncation. To address these questions, we have applied a glycosylation mapping technique (17). This assay is based on the observation that an engineered consensus glycosylation acceptor site can be modified by oligosaccharyltransferase only if this site is placed at a precise "minimal glycosylation distance" from a transmembrane segment (17,18). Therefore, the active site of oligosaccharyltransferase can be used as a reference point against which the position of membrane helices can be determined (18). In this study, we apply for the first time the glycosylation mapping assay to proteins synthesized in intact cells, and our results show that the minimal glycosylation distance in intact cells is shorter than that of proteins synthesized in an in vitro translation system. Our studies also suggest that the C-terminal ends of the transmembrane helices of Na,K-ATPase ␤ 1 and ␤ 3 subunits are located near Leu 58 and Met 61 , respectively. N-terminal truncation of ␤ 1 and ␤ 3 subunits results in a repositioning of the transmembrane helices relative to the membrane.
Although these results do not resolve the question of the functional role of cytosolic ␣-␤ interactions, they clearly show that the N terminus of Na,K-ATPase ␤ subunits is crucial for a correct ␤ subunit topology that is compatible with proper assembly and, in consequence, the acquisition of the correct structural and functional properties of the Na,K-ATPase ␣ subunit. The results also support our hypothesis that structural changes in the ectodomain and/or the transmembrane domain are responsible for the K ϩ effect observed in Na,K-ATPase associated with N-terminally truncated ␤ subunits. Finally, our results validate the glycosylation assay applied in intact cells as a general tool to identify determinants of correct membrane insertion and, in addition, of the glycosylation processing of membrane proteins.

EXPERIMENTAL PROCEDURES
Site-directed Mutagenesis of Na,K-ATPase ␤ 1 and ␤ 3 Isoforms, Construction of Lep/␤ 1 Chimera, and cRNA Preparation-Truncation and point mutants of Xenopus Na,K-ATPase ␤ 1 and ␤ 3 isoforms contained in the pSD5 vector (19) were prepared by using the polymerase chain reaction method of Nelson and Long (20). ␤ 1 subunits lacking 33 amino acids after the initiator methionine (␤ 1 t34, see Fig. 1) were prepared as described previously (2). For the preparation of ␤ 3 t37, lacking 36 amino acids after the initiator methionine, a ␤ 3 cDNA fragment was amplified using an antisense oligonucleotide consisting of nucleotides 301-320 tailed by primer D of Nelson and Long (20) and a sense oligonucleotide comprising part of the noncoding sequence, the ATG coding for the first methionine, and the sequence coding for the amino acids Leu 38 to Tyr 43 of the ␤ 3 isoform. The amplified DNA fragment was then used as a primer to elongate the inverse DNA strand and was finally amplified using a sense oligonucleotide encoding part of the pSD5 vector and primer D of Nelson and Long. The mutated DNA fragment was introduced into the pSD5␤1 vector using NheI and StuI restriction sites.
N-Linked glycosylation acceptor sites were introduced in the ␤ 1 and ␤ 3 isoforms at various positions after the putative end of the transmembrane domain predicted by Kyte-Doolittle hydropathy analysis (21) (see Fig. 1). For this purpose, ␤ 1 fragments were amplified between an antisense oligonucleotide covering nucleotides 628 -648 and tailed by primer D of Nelson and Long and sense oligonucleotides containing codons for the glycosylation acceptor site Asn-Ser-Thr. In these replace-ment mutants, Asn was placed at amino acid positions 59, 63, 65, 67, 70, and 74 for ␤ 1 glycosylation mutants Ϫ5, Ϫ1, ϩ2, ϩ6, ϩ7, and ϩ11, respectively (see Fig. 1). The mutated DNA fragments were introduced into the ␤ 1 cDNA lacking codons for the three natural glycosylation sites (22) using NheI and BamHI restriction sites. The ␤ 1 glycosylation mutant ϩ2 served as a template for the mutant Q56Lglycϩ2 in which Gln 56 of ␤ 1 was replaced by a leucine residue. N-Linked glycosylation acceptor sites were introduced in the ␤ 3 isoform by first amplifying fragments of ␤ 3 cDNA using an antisense oligonucleotide consisting of nucleotides 301-320 and sense oligonucleotides containing codons for the glycosylation acceptor site Asn-Ser-Thr where Asn was placed at amino acid positions 65, 68, 70, 72, and 75 for ␤ 3 glycosylation mutants Ϫ1, ϩ3, ϩ5, ϩ7, and ϩ10, respectively (see Fig. 1). The mutated DNA fragments were introduced into the pSD5␤3 vector using NheI and StuI restriction sites.
For the preparation of ␤ 1 t34/K35L in which Lys 35 of ␤ 1 t34 was replaced by a leucine residue, a fragment of the ␤ 1 cDNA was amplified using ␤ 1 t34 cDNA as a template, a sense oligonucleotide containing the point mutation, and the same antisense oligonucleotide used for the preparation of ␤ 1 glycosylation mutants. For the preparation of ␤ 3 t37/ L38K in which Leu 38 of ␤ 3 t37 was replaced by a lysine residue, and of ␤ 3 t37/L62A/T64G and ␤ 3 t37/L62V/T64A, in which Leu 62 and Thr 64 were replaced by alanine and glycine residues, respectively, or by valine and alanine residues, respectively, a fragment of the ␤ 3 cDNA was amplified using ␤ 3 t37 cDNA as a template, a sense oligonucleotide containing the point mutation, and the same antisense oligonucleotide used for the preparation of ␤ 3 glycosylation mutants.
Chimera were constructed between the Xenopus Na,K-ATPase ␤ 1 subunit and Escherichia coli leader peptidase (Lep) (23) by replacing the second Lep transmembrane domain (H2) by amino acid residues Gly 27 to Thr 61 of ␤ 1 . This was performed by introducing BclI and NdeI restriction sites at nucleotides coding for amino acids 27 and 61 of ␤ 1 and amino acids 59 and 80 of Lep, which permitted the replacement of the BclI-NdeI Lep fragment with that of ␤ 1 subunits. The Lep cDNA contained N-linked glycosylation acceptor sites at positions ϩ3, ϩ16, and ϩ20 from the putative end of H2 (17).
The nucleotide sequences of all constructs were confirmed by dideoxy sequencing.
In Vitro Translations-In vitro translation was performed using the Promega TnT Quick Coupled Transcription/Translation System kit. Briefly, 60 ng of cRNA was added to 20 l of TnT Quick Master Mix, 2 l of [ 35 S]methionine (1000 Ci/mmol) at 10 mCi/ml (Amersham Pharmacia Biotech), and 0.3 l of canine pancreatic microsomal membranes to give a final volume of 25 l and was incubated at 30°C for 90 min. In some cases, the proteins were subjected to endoglycosidase H (EndoH, Calbiochem) treatment as described (13). Proteins were subjected to SDS-polyacrylamide gel electrophoresis (PAGE) and labeled proteins were detected by fluorography. Protein quantification was performed using a laser densitometer (KB Ultrascan 2202).
Expression in Xenopus Oocytes and Immunoprecipitations of ␣ and ␤ Subunits-Oocytes were obtained from Xenopus females as described (29). Routinely, 7 ng of Na,K-ATPase ␣ and/or 0.5-3 ng of ␤ cRNA were injected into oocytes, and in some cases ␤ and ␣ cRNA were co-injected with 4 Ci/oocyte of Easy Tag Express (NEN Life Science Products).
Oocytes not injected with [ 35 S]methionine were incubated at 19°C in modified Barth's medium containing 0.5 mCi/ml [ 35 S]methionine. After a pulse of 4, 6, or 24 h, oocytes were subjected to a chase period of 24, 48, or 72 h in the presence of 10 mM cold methionine. Microsomes and in some instances digitonin extracts were prepared as described (2), and the Na,K-ATPase ␣ and ␤ subunits were immunoprecipitated under denaturing or non-denaturing conditions as described (2) with a Xenopus ␤ 1 (30), ␤ 3 (26), or Bufo ␣ 1 subunit antibody (31). In some instances, the immunoprecipitates were subjected to EndoH treatment as described (13). The dissociated immune complexes were subjected to SDS-PAGE, and labeled proteins were detected by fluorography. Quantification of immunoprecipitated bands was performed with a laser densitometer (LKB Ultrascan 2202).
Saponin Permeabilization of Xenopus Oocytes-To identify the cytosolic localization of the 35-kDa protein species produced in oocytes expressing ␤ 3 t37 mutants, we followed its release into the medium after permeabilization of oocytes with saponin. For this purpose, oocytes were injected with wild type ␤ 3 or ␤ 3 t37 mutant cRNA and labeled for 20 h with [ 35 S]methionine. Twelve oocytes were incubated for 2 h at 19°C in 500 l of modified Barth's medium with or without 0.1% saponin before collection of the media and preparation of oocyte micro-somes by centrifugation of the yolk-depleted homogenate at 20,000 ϫ g for 30 min at 4°C. Denaturing immunoprecipitations were performed with a ␤ 3 antibody on the total volume of the collected media and on microsomes corresponding to six oocytes.
Protease Assays-To test the protease sensitivity of the 28-kDa protein species produced in ␤ 3 t37-expressing oocytes, we used two assays. In the first assay, oocytes were injected with wild type ␤ 3 or ␤ 3 t37 mutant cRNA together with 4 Ci/oocyte of Easy Tag Express (NEN Life Science Products). After a 4-h pulse period, oocytes were homogenized with a plastic pestle in an Eppendorf tube in a solution containing 50 mM Tris-HCl (pH 7.5), 0.25 M sucrose, 50 mM potassium acetate, 5 mM MgCl 2 , and 1 mM dithiothreitol. Aliquots were incubated with 10 mM CaCl 2 in the absence or presence of 0.2 mg/ml proteinase K (Merck) for 1 h at 4°C before addition of 1 mg/ml phenylmethylsulfonyl fluoride. Immunoprecipitations were performed under denaturing conditions with a ␤ 3 antibody. In the second protease assay, oocytes were injected with wild type ␤ 3 or ␤ 3 t37 mutant cRNA or with BiP cRNA and labeled for 24 h. Oocytes were then injected with 30 nl of H 2 O or 30 nl of a solution containing 25 mg/ml trypsin (TPCK-treated, Fluka). Oocytes were left for 1 h at 19°C before preparation of digitonin extracts and immunoprecipitation with a ␤ 3 or a BiP antibody (27).
Pump Current Measurements-Na,K-pump activity was measured as the K ϩ -induced outward current using the two-electrode voltage clamp method as described earlier (4). Current measurements were performed 3 days after injection of oocytes with Bufo ␣ cRNA together with different ␤ cRNAs. To determine the maximal Na,K-pump current (I max ), oocytes were loaded with Na ϩ in a nominally K ϩ -free solution containing 200 nM ouabain, a concentration that inhibits endogenous Na,K-pumps but not the moderately ouabain-resistant exogenous Bufo Na,K-pumps (32). The activation of the Na,K-pump current by K ϩ was determined in a Na ϩ -free solution (140 mM sucrose, 0.82 mM MgCl 2 , 0.41 mM CaCl 2 , 10 mM N-methyl-D-glucamine-HEPES, 5 mM BaCl 2 , 10 mM tetraethylammonium chloride, pH 7.4), and the current induced by increasing concentrations of K ϩ (0.02, 0.1, 0.5, and 5.0 mM K ϩ ) was measured at Ϫ50 mV. To determine I max values, the Hill equation was fitted to the data of the current (I) induced by various K ϩ concentrations ([K]) using a least square method, is the half-activation constant. According to previously published data (4), the Hill coefficient was set to a value of 1.0.

RESULTS
To define the transmembrane domains of Na,K-ATPase ␤ 1 and ␤ 3 isoforms and the possible changes in the transmembrane domain topology after N-terminal truncation, we have used a glycosylation mapping assay together with other biochemical techniques. For the glycosylation mapping assay, we have introduced Asn-Ser-Thr glycosylation acceptor sites at various positions around the predicted C-terminal ends of ␤ 1 and ␤ 3 subunit transmembrane domains and have used the concept of minimal glycosylation distance defined as the number of amino acids separating the C-terminal end of the transmembrane domain from the first Asn residue that is halfmaximally glycosylated (17).
Delineation of the Membrane-spanning Domain of Na,K-ATPase ␤ 1 and ␤ 3 Subunits by Glycosylation Mapping-Transmembrane domain predictions of ␤ 1 subunits by various computer programs are shown in Fig. 1A. All programs predict that ␤ 1 contains a transmembrane domain ␣-helix, but prediction of its N-and C-terminal ends vary between programs. Kyte-Doolittle hydropathy analysis predicts that the ␤ 1 transmembrane domain begins at Trp 33 and ends at Ser 63 . With the exception of the HMMTOP program, predictions of the N-terminal end of the ␤ 1 transmembrane domain was similar with all programs and was located downstream of that predicted by Kyte-Doolittle analysis. Predictions of the C-terminal end of the ␤ 1 transmembrane domain ranged between Gly 53 and Ser 63 .
To determine experimentally the C-terminal end of the ␤ 1 transmembrane domain, we first deleted the three natural glycosylation sites in the ␤ 1 lumenal domain and then introduced N-glycosylation acceptor sites (Asn-Ser-Thr) with Asn at positions Ϫ5, Ϫ1, ϩ2, ϩ4, ϩ7, and ϩ11 relative to Ser 63 , i.e. the C-terminal end of the ␤ 1 transmembrane domain predicted by Kyte-Doolittle hydropathy analysis (see Fig. 1A). Glycosylation of these proteins was first studied after in vitro translation in the presence of microsomes. About 80% of wild type Xenopus ␤ 1 subunits containing the three natural glycosylation sites were glycosylated and migrated on SDS-polyacrylamide gels as a higher molecular mass species ( Fig. 2A, lane 1) compared with ␤ subunits lacking the natural glycosylation sites (lane 2). Glycosylation was absent in ␤ 1 subunits containing a single engineered glycosylation site at position Ϫ5 (lane 3), but the proportion of glycosylated species gradually increased in ␤ 1 subunits containing single glycosylation sites at more distal positions and reached about 80% at position ϩ11 (lanes 4 -7, Fig. 2D). As shown for the ␤ 1 glycosylation mutant ϩ11 ( Fig.  2A, lane 8), all glycosylated species were sensitive to EndoH treatment, which specifically cleaves N-linked core sugars.
Because half-maximal glycosylation occurs at a distance of about 10 -11 amino acid residues away from the C-terminal end of natural, synthetic, or heterologous transmembrane domains introduced into the model membrane protein leader peptidase (Lep) (17,18,33), our results suggest that the ␤ 1 transmembrane domain is shorter than predicted by Kyte-Doolittle hydropathy analysis and ends around Leu 58 . To confirm this result, we also introduced the ␤ 1 transmembrane domain into the previously characterized Lep protein. For this purpose, we replaced the second transmembrane segment of Lep with the ␤ 1 transmembrane domain preceded by the last 6 cytoplasmic amino acids to ensure topological stability (Fig.  2B). Wild type Lep containing a glycosylation site at position ϩ20 from the end of the Lep transmembrane domain served as a control and was almost entirely glycosylated ( Fig. 2B, lanes 1 and 2). The Lep/␤ 1 chimera containing a glycosylation site at position ϩ3 (counting from the first Lep amino acid following the putative ␤ 1 transmembrane domain) was glycosylated by 30 -40% (lanes 3 and 4) and that containing a glycosylation site at position ϩ16 was entirely glycosylated (lanes 5 and 6). The minimal glycosylation distance for the Lep/␤ 1 chimera was thus similar to that of ␤ 1 subunits, which suggests that the C-terminal end of the wild type ␤ 1 transmembrane domain is indeed upstream of Ser 63 and most likely located near Leu 58 .
The glycosylation mapping assay was also used to delineate the C-terminal transmembrane domain end of the ␤ 3 isoform. Kyte-Doolittle hydropathy analysis predicts that the ␤ 3 transmembrane domain begins at Ser 35 and ends at Leu 65 (Fig. 1B). With the exception of the HMMTOP program, all programs predict that the N-terminal end of the ␤ 3 transmembrane domain is located downstream of that predicted by Kyte-Doolittle hydropathy analysis. Prediction of the C-terminal end of the ␤ 3 transmembrane domain varies among the different programs, ranging from Leu 56 to Leu 65 .
For the glycosylation mapping assay of ␤ 3 we left the four natural glycosylation sites intact and introduced new glycosylation acceptor sites at position Ϫ1, ϩ3, ϩ5, ϩ7, and ϩ10 with respect to Leu 65 , i.e. the C-terminal end of the transmembrane domain predicted by Kyte-Doolittle hydropathy analysis (see Fig. 1B). Glycosylation analysis of proteins synthesized in vitro showed that the wild type ␤ 3 subunit and all ␤ 3 glycosylation mutants were EndoH-sensitive and, according to the shift in the molecular mass, were glycosylated on the four natural glycosylation sites (Fig. 2C). Engineered glycosylation sites at positions Ϫ1, ϩ3, or ϩ5 (lanes 3, 5, and 7) were not significantly glycosylated, whereas those at positions ϩ7 and ϩ10 were nearly 100% glycosylated (lanes 9 and 11). Extrapolation of these data predicts that half-maximal glycosylation occurs at around position ϩ6 (Fig. 2D). Taking 10 amino acids as a reference for minimal glycosylation distance (see above), our data suggest that the ␤ 3 transmembrane domain, like the ␤ 1 transmembrane domain, is shorter than predicted by Kyte-Doolittle hydropathy and that it ends around Met 61 .
The reasons for the different shapes of the glycosylation curves observed with ␤ 1 and ␤ 3 isoforms (Fig. 2D) are not known. One possible explanation for the observed all-or-nothing glycosylation of ␤ 3 subunits compared with the nearly linear relationship between the percentage of glycosylation and the Asn position in ␤ 1 subunits may be that the presence of natural sugars decreases the flexibility of the protein and thus the probability of post-translational glycosylation of engineered glycosylation sites situated close to the membrane.
The Glycosylation Efficiency of Proteins Increases with the Time Spent in the ER-Because glycosylation in translation systems in vitro appears to be somewhat inefficient as reflected by the incomplete glycosylation of wild type ␤ 1 subunits ( Fig.  2A, lane 1), we compared glycosylation mapping results obtained by in vitro translations with results obtained after expression of glycosylation mutants in intact cells. Upon expression in Xenopus oocytes and labeling during a 6-h pulse, the total population of wild type ␤ 1 subunits (Fig. 3A, lanes 1 and  Fig. 3B). With the exception of the glycosylation mutant Ϫ5, which was never glycosylated (data not shown), the proportion of glycosylated species for all other glycosylation mutants further increased after a 24-or a 72-h chase period (Fig. 3A, lanes  5-8 and lanes 11-14; Fig. 3B). Xenopus ␤ 1 subunits expressed in oocytes in the absence of ␣ subunits are retained in the ER and are slowly degraded (2). Our results therefore suggest that the efficiency of glycosylation increases with the time proteins spend in the ER and that the actual minimal glycosylation distance is shorter in proteins synthesized in intact cells than in those translated in vitro, especially if the rate of exit from the ER is low.
Role of Gln 56 in Defining the C-terminal End of the ␤ 1 Transmembrane Domain-Because glycosylation mapping suggests that the C-terminal end of the ␤ 1 transmembrane domain is more proximal than that predicted by Kyte-Doolittle hydropathy analysis, we considered the possibility that Gln 56 plays a Amino acids with an attributed score superior to zero were considered as being part of the membrane helix. Kyte-Doolittle hydropathy prediction was performed by the ProtScale software package using a window size of nine residues (21) and is compared with six other programs, TMHMM (49), HMMTOP (50), Sosui (51), TopPred 2 (52), TMpred (53), and PHDhtm (54). Most of these programs can be found on the web. ␤1WT, ␤3WT ϭ wild type ␤ isoforms with indicated location of the natural sugar chains; ␤1WT non-glyc ϭ ␤ 1 subunits lacking the three natural glycosylation sites; ␤1t34 ϭ N-terminally truncated ␤ 1 subunits; ␤3t37 ϭ N-terminally truncated ␤ 3 subunits. Amino acids that were replaced by an asparagine residue in the glycosylation acceptor sequence Asn-Ser-Thr are shown in boldface and their position and the name of the corresponding mutant are indicated. role in defining the C-terminal end of the ␤ 1 transmembrane domain. Consistent with this idea, a ␤ 1 Q56L mutant containing a glycosylation site at ϩ2 (ϩ2 Q56L) was significantly less glycosylated than the glycosylation mutant ϩ2 (compare Fig.  3A, lanes 21-26 to lanes 9 -14). This suggests that introduction of a hydrophobic residue at this position "pulls" the C-terminal end of the transmembrane domain into the membrane.
Engineered Glycosylation Sites Adjacent to the ␤ 1 and ␤ 3 Transmembrane Domain Impede Efficient ␣-␤ Interactions and Correct Glycosylation Processing-Because an ␣-assembly domain is possibly located within the 68 amino acids succeeding the ␤ transmembrane domain, according to the two-hybrid assay (8), we wondered whether the glycosylation mapping assay could also provide information on the assembly process of ␤ with ␣ subunits. In other words, does introduction of glycosylation sites close to the ␤ transmembrane domain interfere with subunit oligomerization and maturation and/or does subunit oligomerization influence glycosylation at these sites?
After a 6-h pulse, wild type ␤ 1 subunits co-expressed with ␣ subunits in Xenopus oocytes co-immunoprecipitated and were thus assembled with ␣ subunits, mainly in their core-glycosylated ER form (Fig. 4A, lane 1). In contrast to unassembled ␣ subunits, which are rapidly degraded (lanes [22][23][24], formation of ␣-␤ complexes in the ER led to the stable expression of ␣ subunits and permitted ER exit of ␣-␤ complexes as reflected by the acquisition of complex sugars by the ␤ subunit (lanes 1-3) and the expression of functional Na,K-pumps at the cell surface as assessed by pump current measurements (Fig. 4B, lane 1). Significantly, non-glycosylated ␤ 1 subunits (Fig. 4A, lanes 4 -6) and in particular ␤-glycosylation mutants containing glycosylation sites at positions Ϫ1, ϩ2, and ϩ4 (lanes 7-15) co-immunoprecipitated efficiently, and thus associated with ␣ subunits, but only after a 24-and a 72-h chase and not after a 6-h pulse. The delayed assembly of the ␤-glycosylation mutants was reflected in the partial degradation of the ␣ subunits after the 24-h chase period. Oocytes co-expressing ␣ subunits with ␤ 1glycosylation mutants Ϫ1, ϩ2, or ϩ4 exhibited a significantly reduced pump current (Fig. 4B, lanes 3-5) compared with oocytes co-expressing ␣ subunits with wild type ␤ 1 subunits (lane 1). These results indicate that the sugar moieties and/or the mutations introduced at positions close to the transmembrane domain significantly impede correct ␣-␤ interactions and reduce cell surface expression. Despite the partial degradation of the ␣ subunit, non-glycosylated ␤ 1 subunits (Fig. 4B, lane 2) and the ␤ 1 -glycosylation mutants ϩ7 (lane 6) or ϩ11 (lane 7) produced a similar number of functional ␣-␤ complexes at the cell surface as wild type ␤ 1 subunits (lane 1) probably due to the regulated cell surface expression in Xenopus oocytes. Previous studies (22) have shown that Xenopus oocytes only express a limited number of exogenous Na,K-pumps at the cell surface, which does not exceed six to eight times the number of endogenous Na,K-pumps and which is therefore to a certain extent independent of the number of stable ␣-␤ complexes formed.
Interestingly, association of the ␤ 1 -glycosylation mutant Ϫ1 with the ␣ subunit reduced its glycosylation compared with that of the glycosylation mutant Ϫ1 expressed without ␣ subunits from 78 Ϯ 5% (n ϭ 3) to 47 Ϯ 4% (n ϭ 4) (p Ͻ 0.01) (compare Fig. 4A, lane 8 to Fig. 3A, lane 5). Furthermore, the extent of glycosylation remained stable over time in the ␣-associated ␤ 1 Ϫ1 mutant in contrast to the non-associated ␤ 1 Ϫ1 mutant in which glycosylation further increased after a 72-h chase (compare Fig. 4A, lanes 8 and 9 with Fig. 3A, lanes 5 and  7). These results may reflect that, after delayed assembly of the ␤ 1 Ϫ1 mutant, ␣-␤ complexes leave the ER in their non-glycosylated or core-glycosylated form. Finally, despite expression at the cell surface, all ␣-associated ␤ 1 subunits with engineered glycosylation sites up to position ϩ11 did not, in contrast to wild type ␤ 1 subunits (Fig. 4A, lanes 2 and 3), become fully glycosylated after 3 days of expression but remained in their core-glycosylated form (lanes 9, 12, 15, 18, and 21).
In contrast to ␤ 1 isoforms devoid of natural sugars or containing engineered glycosylation sites, ␤ 3 isoforms containing natural and engineered glycosylation sites associated efficiently with co-expressed ␣ subunits after a 6-h pulse (Fig. 4C,  lanes 4, 7, and 10) similar to wild type ␤ 3 subunits (lane 1). This is perhaps due to the presence of natural sugar chains, which may favor a more efficient folding of the ␤ 3 subunit compared with the non-glycosylated ␤ 1 subunit. Nevertheless, the presence of engineered glycosylation sites in ␤ 3 subunits prevented complete stabilization of the ␣ subunit and reduced the proc- essing of the ␤ 3 subunit to the fully glycosylated form (Fig. 4C,  compare lanes 4 -12 to lanes 1-3) further supporting the notion that the integrity of a domain close to the membrane is important for ␣-␤ interaction and maturation.
Effects of N-terminal Truncation on the Positioning of the ␤ 1 Transmembrane Domain in the Membrane-Previous observations (7) have led to the assumption that N-terminal truncation may indirectly affect the ectodomain and/or the transmembrane domain of the ␤ 1 subunit. To study this further, glycosylation acceptor sites were introduced into the N-terminally truncated ␤ 1 subunit (␤ 1 t34) at the same positions as in fulllength ␤ 1 subunits (see Fig. 1A). The glycosylation pattern of the glycosylation mutants of ␤ 1 and ␤ 1 t34 was compared after expression in Xenopus oocytes in the absence of ␣ subunits and labeled during a 24-h pulse period. Full-length (Fig. 5A, lanes 1  and 2) and N-terminally truncated ␤ 1 subunits (Fig. 5B, lanes  1 and 2) containing natural glycosylation sites were glycosylated, whereas the corresponding proteins lacking the natural glycosylation sites were not glycosylated (Fig. 5, A and B, lanes  3 and 4). On the other hand, the glycosylation pattern was different in ␤ 1 t34 mutants and full-length ␤ 1 subunits containing engineered glycosylation sites at the same positions (compare Fig. 5, A and B, lanes 5-12; Fig. 5C). Apart from the glycosylation site at position Ϫ5, which was glycosylated to a FIG. 4. Engineered glycosylation sites in ␤ subunits impede efficient ␣-␤ interactions, and the core sugars are not correctly processed. A, ␣-subunit assembly and glycosylation processing of ␤ 1 -glycosylation mutants. Oocytes were injected with 7 ng of Bufo ␣ 1 alone (lanes [22][23][24] or together with 0.5 ng of wild type (lanes 1-3), 1 ng of non-glycosylated ␤ 1 (lanes 4 -6), or 1 ng of ␤ 1 -glycosylation mutant cRNA (lanes 7-21). Digitonin extracts were prepared after a 6-h pulse, a 24-h chase, and a 72-h chase period, and immunoprecipitations were performed using an ␣-antibody under non-denaturing conditions that preserve ␣-␤ interactions. One out of four representative experiments is shown. B, Na,K-pump current measurements of cell surface expressed ␣-␤ complexes. Oocytes were injected with Bufo ␣ 1 and wild type (lane 1) or mutant (lanes 2-7) ␤ 1 cRNA. 3 days after injection, maximal pump currents (I max ) of ␣-␤ complexes were determined by extrapolation of K ϩ activation curves as described under "Experimental Procedures." Measurements were performed in the presence of 200 nM ouabain, which inhibits endogenous Xenopus Na,K-pumps but not the moderately ouabain-resistant Bufo Na,K-pumps. Shown are means Ϯ S.E. of data from 15-30 oocytes obtained from four different Xenopus females. *, p Ͻ 0.05; **, p Ͻ 0.01 compared with oocytes injected with wild type ␣-␤ complexes. C, ␣-subunit assembly and glycosylation processing of ␤ 3 -glycosylation mutants. Oocytes were injected with 7 ng of Bufo ␣ 1 and 1 ng of wild type (lanes 1-3) or ␤ 3 -glycosylation mutant (lanes 4 -12) cRNA. Digitonin extracts were prepared after a 6-h pulse, a 24-h chase, and a 72-h chase period, and immunoprecipitations were performed with an ␣-antibody under non-denaturing conditions. One out of two representative experiments is shown. ng, non-glycosylated; cg, core-glycosylated; fg, fully glycoylated ␤ subunits.
higher extent in ␤ 1 t34 than in full-length␤ 1 , glycosylation sites at more distal positions were less glycosylated in ␤ 1 t34 than in full-length ␤ 1 subunits.
These results clearly indicate that N-terminal truncation results in a repositioning of the ␤ 1 transmembrane domain in the membrane. Our results suggest that the C-terminal end of the ␤ 1 t34 transmembrane domain is located about 4 residues downstream of that of the ␤ 1 transmembrane domain, i.e. around Ile 62 . This would also be consistent with the increased glycosylation efficiency seen at position Ϫ5 in ␤ 1 t34. Although this Leu to Asn mutation lies right after the end of the ␤ 1 transmembrane domain (and hence should not affect the transmembrane domain's position in the membrane), it is 4 residues inside the ␤ 1 t34 transmembrane domain and could be expected to "push" the transmembrane domain out of the membrane, possibly far enough to allow some degree of glycosylation as is indeed observed.
We have previously shown that co-expression of ␣ subunits with ␤ 1 t34 containing the three natural glycosylation sites leads to the formation of stable and functional Na,K-pumps (7). ␤ 1 t34 glycosylation mutants devoid of natural but containing engineered glycosylation sites showed qualitatively similar defects in assembly and processing as full-length ␤ 1 glycosylation mutants (Fig. 4). However, the consequences of inefficient assembly on cell surface expression of functional Na,K-pumps were even more pronounced (data not shown).
In conclusion, the results obtained with the glycosylation mapping assay suggest that the position in the membrane of the ␤ 1 t34 transmembrane domain and the conformation of adjacent C-terminal domains differ from those of the wild type ␤ 1 subunit and that the cytoplasmic N terminus is therefore an important determinant for the correct membrane insertion and folding of this type II protein.
Effects of N-terminal Truncation on the Positioning of the ␤ 3 Transmembrane Domain in the Membrane-To confirm the general relevance of our observation we also investigated the effects of N-terminal truncations on ␤ 3 isoforms. For this purpose, we prepared a N-terminally truncated ␤ 3 t37 mutant, which placed the initiator methionine at the same position with respect to the putative N-terminal end of the transmembrane domain as in the ␤ 1 t34 mutant (see Fig. 1B) but which left the four natural glycosylation sites intact. Compared with fulllength ␤ 3 subunits, ␤ 3 t37 was less well expressed after a 24-h pulse and a 48-h chase both in the absence (data not shown) or in the presence of co-expressed ␣ subunits (Fig. 6A, compare  lanes 1 and 3 to lanes 5 and 7) indicating that ␤ 3 t37 is more susceptible to degradation. Also, N-terminally truncated ␤ 3 t37 did not become fully glycosylated, was not able to efficiently stabilize co-expressed ␣ subunits during a 48-h chase (Fig. 6C, compare lanes 1-4 to lanes [5][6][7][8], and produced low levels of functional ␣-␤ complexes at the plasma membrane (Fig. 6D,  lane 2). This contrasts with wild type ␤ 3 subunits (Fig. 6, A-C,  lanes 1-4, Fig. 6D, lane 1), ␤ 1 subunits (Fig. 4A, lanes 1-3, Fig.  6D, lane 4), or N-terminally truncated ␤ 1 t34 (Fig. 6, A-C, lanes 17-20, Fig. 6D, lane 5), which, after co-expression with ␣-subunits, could produce stable, functional Na,K-pumps at the cell surface and were themselves stably expressed and correctly processed to the fully glycosylated form.
Immunoprecipitations of ␤ 3 t37 revealed the presence of two major protein species in addition to the core-glycosylated protein both in the presence (Fig. 6A, lane 5) or absence (data not shown) of co-expressed ␣ subunits. These protein species were not apparent after expression of the wild type ␤ 3 subunit (lane 1) or the ␤ 1 t34 mutant (lane 17). The first of these products (asterisk) with a molecular mass of about 35 kDa migrated slightly faster than the core-glycosylated ␤ 3 t37 and was itself core-glycosylated as suggested by its cleavage by EndoH to a 24-kDa species (compare lanes 5 and 6). This product was also observed in microsomal preparations (Fig. 6B, lanes 5 and 6) that permit enrichment of membrane proteins as well as of soluble proteins contained within right-side-out ER vesicles (27). The 35-kDa protein species persisted after a 48-h chase period (lanes 7 and 8) and was able to transiently associate with the ␣ subunit (data not shown). In contrast to the wild type ␤ 3 subunit (Fig. 6, A and B, lanes 3 and 4), neither the authentic ␤ 3 t37 nor the glycosylated 35-kDa product became fully glycosylated during a 48-h chase period (Fig. 6, A and B,  lanes 7 and 8). The second protein species (Fig 6A, solid dot) of about 28 kDa, which was revealed in digitonin extracts (Fig.  6A, lane 5) but not in microsomal preparations (Fig. 6B, lane 5) after a 24-h pulse, migrated with the EndoH-treated ␤ 3 t37 (Fig.  6A, lane 6). This protein species disappeared after a 48-h chase period (lane 7) and was not able to associate with the ␣ subunit (data not shown). The identity of the 35-and 28-kDa products is discussed below.
Effects of N-terminal Charged Residues on the ␤ 1 and ␤ 3 Transmembrane Domains-One possibly important difference between ␤ 3 and ␤ 1 N-terminally truncated mutants is the presence, after the initiator methionine, of a positively charged lysine residue in ␤ 1 t34 as opposed to a leucine residue in ␤ 3 t37 (see Fig. 1). We tested whether the presence or absence of a single, positively charged residue could account for some of the differences observed between ␤ 1 t34 and ␤ 3 t37 expression by replacing the leucine by lysine in ␤ 3 t37 (␤ 3 t37L38K) and the lysine by leucine in ␤ 1 t34 (␤ 1 t34K38L). Replacing the leucine  (lanes 1 and 2), non-glycosylated ␤ 1 (lanes 3 and  4), ␤ 1 -glycosylation mutants (lanes 5-14). After a 24-h pulse, microsomes were prepared and immunoprecipitated with a ␤ 1 antibody before treatment or not with EndoH. B, glycosylation mapping of Nterminally truncated ␤ 1 subunits. Oocytes were injected with cRNA coding for ␤ 1 t34 containing (lanes 1 and 2) or not containing (lanes 3 and 4) the natural glycosylation sites or with ␤ 1 t34 glycosylation mutants ( lanes 5-14). Processing of the oocytes as in A. C, quantification of data shown in A and B (mean of two to three experiments). The percentage glycosylation of ␤ 1 (closed circles) and ␤ 1 t34 (open circles) subunits is shown as a function of the position of engineered asparagine (Asn) from the putative C-terminal transmembrane domain end. by a lysine residue in ␤ 3 t37 did indeed significantly improve its stability and led to the disappearance of the 28-kDa band and to a reduction in the 35-kDa protein species (Fig. 6A, lanes 9  and 10). Furthermore, ␤ 3 t37L38K could stabilize co-expressed ␣ subunits (Fig. 6C, lanes 9 -12), became fully glycosylated (Fig. 6, A and B, lanes 11 and 12), and increased, although only slightly, the pump current measured at the cell surface compared with that measured in oocytes expressing ␤ 3 t37 (Fig. 6D,  lanes 2 and 3). In contrast, replacing the lysine residue by a leucine residue in ␤ 1 t34 led to the appearance of a non-glycosylated product (35 kDa) that was mainly observed in digitonin extracts after a 24-h pulse period (Fig. 6A, lanes 13 and 14) analogous to the 28-kDa product of ␤ 3 t37 (lanes 5 and 6). ␤ 1 t34K38L was processed to a fully glycosylated, EndoH-resistant form (Fig. 6A, lanes 15 and 16) but was not able to stabilize the ␣ subunit as efficiently as ␤ 1 t34 (Fig. 6C, compare lanes  13-16 to lanes 17-20), which probably explains the significant decrease in the cell surface expression of functional pumps compared with oocytes expressing ␤ 1 t34 (Fig. 6D, lanes 5 and  6). These results confirm previous observations (34) that a single positive charge close to the N-terminal end of the transmembrane domain segment can indeed play an important role in the correct membrane integration of type II membrane proteins.
A final set of experiments were aimed at the identification of the nature of the additional protein species observed in ␤ 3 t37or ␤ 1 t34K38L-expressing oocytes to better understand the molecular basis of the perturbations in the biosynthesis and stability of these two proteins. The lack of glycosylation, the nearly complete absence in microsomal preparations, and the rapid degradation indicated that the 28-kDa protein species may be a cytosolic form of the ␤ 3 t37 protein. To test this possibility, we followed the appearance of the 28-kDa species in the medium of oocytes permeabilized with saponin. Saponin treatment produced some nonspecific detergent effects as reflected by the decrease in immunoprecipitable wild type ␤ 3 subunits from microsomal preparations (Fig. 7A, lanes 1 and 2) and the appearance of about 1% of core-glycosylated ␤ 3 subunits in the medium (lanes 3 and 4). However, under similar experimental conditions, saponin removed preferentially the contaminating 28-kDa protein species from microsomes (lanes 5 and 6) and, in particular, produced a much more significant release of the 28-kDa protein species than of the ␤ 3 t37 protein to the medium (lanes 7 and 8) indicating that the 28-kDa protein species might indeed be cytosolic. This result was further supported by the observation that proteinase K treatment of homogenates from oocytes expressing wild type ␤ 3 subunits or ␤ 3 t37 mutants only removed the cytoplasmic N-terminal tail of the wild type ␤ 3 subunit (Fig. 7B, lanes 1 and 2) and had no effect on the N-terminally truncated ␤ 3 t37, but completely digested the 28-kDa protein species as well as some additional protein species of lower molecular mass occasionally observed in ␤ 3 t37-expressing oocytes (lanes 3 and 4). Finally, a similar result was obtained when trypsin was injected into intact oocytes. Injected trypsin only removed the cytosolic tail of wild type ␤ 3 subunits (Fig. 7C, lanes 1 and 2) and did not digest ␤ 3 t37 or the glycosylated 35-kDa protein species but almost completely digested the 28-kDa and lower molecular mass species. Trypsin had no access to intralumenal ER proteins as reflected by the resistance to trypsinolysis of the molecular chaperone BiP (binding protein) (lanes 5 and 6). Altogether, these results support the hypothesis that the 28-kDa protein species is a cytosolic, nonglycosylated form of ␤ 3 t37, which, due to the severe topological effects of N-terminal truncation, cannot be integrated into the membrane.
In contrast to the 28-kDa species, the 35-kDa species of the ␤ 3 t37 mutant is glycosylated (Fig. 6A, lanes 5 and 6) and resistant to trypsinolysis performed on oocyte homogenates (Fig. 7C, lanes 3 and 4). Furthermore, the 35-kDa species shows a slightly lower molecular mass than the ␤ 3 t37 mutant.  (lanes 13-20), respectively. *, glycosylated 35-kDa protein species of ␤ 3 t37 mutants, which is sensitive to EndoH treatment and apparent in digitonin extracts (A) and microsomes (B); solid dot, non-glycosylated 28-kDa protein species of ␤ 3 t37 and ␤ 1 t34K35L mutants, which is EndoH insensitive and only apparent in digitonin extracts (A). C, denaturing immunoprecipitations from digitonin extracts of ␣ subunits with an ␣ antibody. D, Na,K-pump current measurements of cell surface-expressed ␣-␤ complexes. Oocytes were injected with Bufo ␣ 1 and wild type or mutant ␤ 1 or ␤ 3 cRNA. 3 days after injection, maximal pump currents (I max ) of ␣-␤ complexes were determined. Shown are means Ϯ S.E. of data from 16 oocytes obtained from two different Xenopus females. *, p Ͻ 0.01 compared with oocytes injected with wild type ␣-␤ complexes.
These characteristics suggest that this protein species has been translocated across the ER membrane, has become glycosylated, and has been cleaved by about 2 kDa. The difference in the molecular mass between the EndoH-treated ␤ 3 t37 mutant (28 kDa) and the EndoH-treated cleaved product (24 kDa) would be roughly compatible with the removal of the transmembrane domain, e.g. by signal peptidase cleavage. The computer software program SignalP (35) was used to evaluate the possibility that a signal peptidase cleavage site is present in the ␤ 3 transmembrane domain, which, due to ␣ helix perturbations of the transmembrane domain after N-terminal truncation of the ␤ 3 subunit, may become accessible. Signal peptidase cleavage sites characteristically contain small neutral residues at position Ϫ1 (1 amino acid residue upstream of the site of cleavage) and small neutral and uncharged residues at position Ϫ3 (for review see Ref. 36). The SignalP program predicted a probability of 50% for signal peptidase cleavage between amino acids Thr 64 and Leu 65 , which according to the glycosylation mapping assay are located adjacent to the Cterminal end of the ␤ 3 transmembrane domain. In an attempt to verify the hypothesis of the exposure of this putative signal peptidase cleavage site in the ␤ 3 t37 mutant, we replaced Leu 62 (position Ϫ4) by alanine and Thr 64 (position Ϫ1) by glycine residues (mutant ␤ 3 t37L62A/T64G) or by valine and alanine residues (mutant ␤ 3 t37L62V/T64A), respectively. These mutations should increase the cleavage probability by signal peptidase to 100%, according to the SignalP program. When expressed in Xenopus oocytes, the ␤ 3 t37L62A/T64G mutant no longer produced the 35-kDa protein species (Fig. 7D, lanes 3  and 4) contrary to the SignalP prediction, whereas the ␤ 3 t37L26V/T28A mutant (lanes 5 and 6) produced a higher proportion of 35-kDa protein species than ␤ 3 t37 (lanes 1 and 2).
In EndoH-treated samples, the 24-kDa product derived from the 35-kDa species represented 20% and 60% of the total ␤ population in oocytes expressing the ␤ 3 t37 (lane 2) and the ␤ 3 t37L26V/T28A (lane 6) mutants, respectively. These results suggest that in Na,K-ATPase ␤ 3 subunits, N-terminal truncation and consequent changes in transmembrane domain topology lead to the exposure of a cryptic signal peptidase cleavage site in a certain population of the newly synthesized proteins. DISCUSSION The glycosylation mapping and biochemical techniques used in this study have provided new information on the position of the C-terminal ends of the transmembrane domains of Na,K-ATPase ␤ subunits and on the role of the N terminus and of specific amino acids adjacent to the transmembrane domain in the definition of these domains. Furthermore, glycosylation mapping has allowed the identification of regions in the ␤ subunit that are important for ␣-interaction and have revealed novel features in the processing of glycoproteins in Xenopus oocytes.
Delineation of the Transmembrane Domains of Na,K-ATPase ␤ 1 and ␤ 3 Subunits and the Role of the Cytoplasmic N-terminal Tail and of Single Positive Charges for Correct Membrane Insertion-Our previous studies have suggested that the cytoplasmic N terminus of the Na,K-ATPase ␤ subunit interacts with the catalytic ␣ subunit (2), that its truncation changes the K ϩ and Na ϩ affinities of Na,K-ATPase (2, 7), but also that N-terminal interactions might not be directly involved in the observed functional effects (7). One of the aims of this study was to check whether N-terminal truncations may have consequences for the structural integrity of the ␤ subunit.
To assess this question, we have made use of a glycosylation FIG. 7. N-terminal truncation of ␤ 3 subunits produces soluble, cytosolic, and ER lumenal protein species. A, a 28-kDa protein species is preferentially released to the medium after cell permeabilization. Oocytes were injected with 3 ng of wild type ␤ 3 or ␤ 3 t37 cRNA, labeled during a 24-h pulse and incubated without or with 0.1% saponin for 2 h as described under "Experimental Procedures." Microsomes (M) prepared from the oocytes and the collected medium (Med) were immunoprecipitated with a ␤ 3 antibody. B and C, the 28-kDa but not the 35-kDa protein species is sensitive to trypsin digestion. B, oocytes were injected with 3 ng of wild type ␤ 3 or ␤ 3 t37 cRNA and 4 Ci/oocyte of [ 35 S]methionine. After a 4-h incubation, homogenates were prepared and subjected to trypsinolysis as described under "Experimental Procedures." Note the high and low proportion of the 28-and 35-kDa protein species, respectively, after this short pulse compared with that observed after a 24-h pulse (A and C) indicating that the 28-kDa species is an early synthesis product that is rapidly degraded and completely digested by trypsin (lane 4) and that the production of the 35-kDa species involves a slower process. C, oocytes were injected with 3 ng of wild type ␤ 3 , ␤ 3 t37, or BiP cRNA, labeled for 24 h and then injected or not with 0.75 g of trypsin per oocyte as described under "Experimental Procedures." After 1 h at 19°C, digitonin extracts were prepared and immunoprecipitated with a ␤ 3 antibody (lanes 1-4) or a BiP antibody (lanes 5 and 6). D, oocytes were injected with 3 ng of ␤ 3 t37, ␤ 3 t37L62A/T64G, or ␤ 3 t37L62V/T64A cRNA and subjected to a 24-h pulse before preparation of microsomes and immunoprecipitation with a ␤ 3 antibody. Samples were treated or not treated with EndoH. One out of three representative experiments is shown. *, glycosylated 35-kDa protein species; solid dot, non-glycosylated 28-kDa protein species.
mapping technique that has permitted us to delineate the C-terminal end of the transmembrane domain of wild type and mutant ␤ subunits. In previous studies using the glycosylation mapping technique in translation systems in vitro, the minimal glycosylation distance, e.g. the number of residues separating the end of a transmembrane domain and the active center of the oligosaccharyltransferase, was calibrated against transmembrane domains with known positions in the membrane (18). From these studies, a reference minimal glycosylation distance value of about 10 -11 residues was determined. Glycosylation mapping assays performed on Na,K-ATPase ␤ 1 and ␤ 3 subunits show that their transmembrane domains are significantly shorter than predicted by Kyte Doolittle hydropathy analysis and end around Leu 58 and Met 61 , respectively (Fig. 8), close to the C-terminal transmembrane domain ends predicted by several other prediction programs (see Fig. 1).
In ␤ 1 subunits, N-terminal truncation significantly decreased the minimal glycosylation distance compared with wild type ␤ 1 subunits (see Fig. 8), indicating that removal of the N-terminal cytoplasmic tail leads to a repositioning of the ␤ 1 transmembrane domain that may be transmitted to the ectodomain. This result lends support to our previously raised hypothesis that the effects on the cation affinities of Na,K-ATPase containing N-terminally truncated ␤ subunits are not due to the abolition of N-terminal ␤ interactions with the ␣ subunit but rather to structural changes in the extracytoplasmic domain and/or the transmembrane domain.
The observed importance of the ␤ N terminus for the correct positioning of the ␤ transmembrane domain in the membrane is likely to be of general relevance for type II proteins. Indeed, N-terminal truncations of other type II proteins such as invariant chain (37) and the asialoglycoprotein receptor (38) also lead to a shift in the position of the proteins in the membrane as reflected by the exposure of a cryptic signal peptidase cleavage site present in the transmembrane domain. We have observed a similar phenomenon after N-terminal truncation of Na,K-ATPase ␤ 3 subunits. The SignalP program predicts a higher probability for the existence of a signal peptidase cleavage site at the C-terminal end of the ␤ 3 transmembrane domain than of the ␤ 1 transmembrane domain. Accordingly, after N-terminal truncations of ␤ 3 , but not of ␤ 1 subunits, we observe a small population of glycosylated protein species with a molecular mass that is compatible with the removal of the transmembrane domain by signal peptidase acting at the predicted cryptic signal peptidase cleavage site. In addition, these protein species increase after certain mutations, which should increase the probability of signal peptidase cleavage, according to the SignalP program.
Beyond the effects on the membrane position of the transmembrane domain, N-terminally truncated ␤ 3 subunits produce a population of apparently cytosolic protein species, which are not glycosylated and are highly sensitive to cellular and tryptic degradation. This result suggests that in ␤ 3 subunits, in contrast to ␤ 1 subunits, N-terminal truncation changes the topology of the transmembrane ␣ helix to an extent that impedes stable membrane insertion of a detectable fraction of the molecules. ␤ 1 and ␤ 3 isoforms display about 39% and 62% sequence identity in the cytoplasmic N-terminal tail and the transmembrane domain, respectively. An apparently important sequence difference which may explain the less deleterious effects of the ␤ 1 than of the ␤ 3 N-terminal truncations, is the presence after the initiator methionine of a lysine residue in ␤ 1 subunits instead of a leucine residue in ␤ 3 subunits. Replacement of the lysine residue by a leucine residue in N-terminally truncated ␤ 1 subunits produces a population of non-glycosylated, unstable ␤ 1 species similar to those observed in N-terminally truncated ␤ 3 subunits. On the other hand, a lysine residue replacing the analogous leucine residue in N-terminally truncated ␤ 3 subunits reduces the proportion of these putatively non-membrane inserted protein species.
The predominance of positively charged residues at the cytoplasmic side of the transmembrane domain ("positive inside" rule (39); "charge difference" rule (40)) is a major determinant for the orientation of transmembrane domains in membrane Shown are linear sequence models of wild type ␤ 1 (␤1wt) and ␤ 3 (␤3wt) indicating the putative N-terminal start and the C-terminal ends of transmembrane domains (gray areas) defined by glycosylation mapping. Also shown are N-terminally truncated ␤ 1 (␤1t34) and ␤ 3 (␤3t37). For ␤ 1 subunits, the minimal glycosylation distances (MGD) determined in vitro (90-min pulse) and in intact cells (6-h pulse, 24-h chase, and 72-h chase) are shown. In ␤ 1 t34, the putative shift in positioning of the transmembrane domain C terminus in the membrane is indicated as a shaded area. In ␤ 3 t37, the positioning of the transmembrane domain is not clearly defined and is represented as a shaded area. The putative, exposed signal peptidase cleavage site (SP) is indicated by an arrow. Also indicated are the 38-kDa glycosylated protein species, which is produced after signal peptidase cleavage, and the 28-kDa non-glycosylated protein species, which cannot insert in the membrane. For detailed description of results see text.
proteins, although the length of the transmembrane domain and the presence or absence of hydrophilic N-terminal tails may also be important (41). In Na,K-ATPase ␤ subunits, the absence of positively charged residues at the cytoplasmic side in combination with the lack of the cytoplasmic N-terminal tail produces a population of non-membrane-inserted, cytoplasmic proteins. At present, it is not known whether topology changes in the transmembrane ␣ helix of these proteins diminish the interaction with the signal recognition particle responsible for the targeting of nascent membrane proteins to the ER membrane (for review see Ref. 42). Alternatively, it is also possible that part of the mutant proteins initially insert into the membrane in an unstable, inverted C cyt /N out orientation and, consequently, are eventually released into the cytoplasm.
The Glycosylation Mapping Assay Reveals New Features of Protein Glycosylation and Glycosylation Processing-The glycosylation mapping assay used in this study has also provided new information on protein glycosylation and glycosylation processing in Xenopus oocytes. For the first time, the glycosylation mapping assay has been carried out in intact cells where the time allowed for folding of newly synthesized proteins can be considerably increased compared with in vitro translation systems. Our results suggest that this time factor is an important determinant for the actual efficiency of glycosylation at sites close to the transmembrane domain. Indeed, translation of Na,K-ATPase ␤ 1 subunits in Xenopus oocytes rather than in translation systems in vitro significantly reduces the minimal glycosylation distance by 5-6 residues depending on the chase time (Fig. 8). Because the distance to the active site of the oligosaccharyltransferase is likely to be the same in in vivo and in vitro translation systems, this result suggests that the domain between the end of the transmembrane domain and the glycosylation acceptor sites is not a rigid structure and permits post-translational glycosylation of glycosylation acceptor sites while the protein is retained in the ER. Our results differ from observations made in HeLa cells that the major histocompatibility complex class II-associated invariant chain, a type II glycoprotein, cannot be glycosylated post-translationally on its natural glycosylation sites (43). Additional experiments are needed to decide whether the glycosylation process differs in Xenopus oocytes compared with mammalian cells; possibly, the slow synthesis, processing, and intracellular routing of proteins in oocytes caused by the low incubation temperature (19°C) may be one reason for the observed differences. Alternatively, the observed post-translational glycosylation may be characteristic for glycosylation sites close to the membrane. In any case, our results indicate that the minimal glycosylation distance is not a constant value, but is nevertheless, a valuable tool for defining the extracytoplasmic end of transmembrane domains of type II and probably other membrane proteins if assayed under controlled experimental conditions.
In addition to the observation that the glycosylation efficiency increases with the time proteins spend in the ER and thus that glycosylation can be post-translational, the in vivo glycosylation mapping assay also provides information on the requirements for glycosylation processing. Our results show that ␤ subunits, with certain engineered glycosylation sites close to the C-terminal end of the transmembrane domain, acquire core sugars and can associate with ␣ subunits and produce functional ␣-␤ complexes at the cell surface but never become fully glycosylated. This result is inconsistent with the dogma that, during intracellular transport in animal cells, high mannose core sugars are necessarily trimmed in the ER and proximal Golgi compartments and complex type sugars are added to glycoproteins in a distal Golgi compartment (for recent review see Ref. 44). A possible explanation of our results is that glycosylation processing enzymes have no access to the core sugars added to glycosylation sites situated close to the membrane. Because the core-glycosylated ␤ species, which escape ER retention, do not show any decrease in their molecular mass during long chases, it is possible that ER mannosidase I, which, as the oligosaccharyltransferase, is a type II membrane protein (45), is not able to remove the first five mannose residues that are necessary for the further trimming of glycoproteins. At present, the nature of the conformational constraints, which may impede mannose trimming of core sugars located as far as 16 amino acids from the C-terminal end of the transmembrane domain, is not known and the glycosylation mapping assay could become a valuable tool to determine the minimal distance from the membrane that permits mannose processing.
Introduction of Novel Glycosylation Sites Close to the ␤ Transmembrane Domain Impedes Efficient ␣-␤ Interaction and/or Intracellular Routing-Finally, our studies on the Na,K-ATPase ␤ subunits have also provided information on the location of putative sites of interaction with the partner ␣ subunit. According to the two-hybrid assay, an ␣-assembly domain is located within the 68 amino acids succeeding the ␤ transmembrane domain (8). The results of the present study show that introduction of new glycosylation sites within a region of about 10 amino acids from the C-terminal end of the ␤ transmembrane domain has a significant effect on the ␣-␤ interaction efficiency as reflected in delayed ␣-␤ assembly, decreased stability of the ␣ subunit, and reduced cell surface expression of functional Na,K-pumps. Similar defects have previously been observed in vesicular stomatitis virus G proteins (46,47) and influenza virus hemagglutinin (48) containing sugar chains at new glycosylation sites, and it was concluded from these studies that the primary role of natural sugars is to promote proper protein folding. However, neither in those nor in the present study, can it be definitively decided whether the alterations in the folding of the glycosylation mutants is due to the presence of a new sugar chain or to the amino acid substitutions introduced to create a new glycosylation site. ␤ 3 subunits, which contain a new, additional glycosylation site at apparent positions ϩ3 and ϩ5 from the C-terminal end of the transmembrane domain, are not or are only poorly glycosylated, but their ER exit is also significantly impeded as reflected by the absence of full glycosylation of the natural sugar chains. This result indicates that, at least in this case, the mutation itself rather than the presence of a sugar chain in this region is responsible for the folding defect of the protein. Further mutational analysis is needed to determine whether the decreased ␣-interaction efficiency of the ␤-glycosylation mutants reflects a discrete disruption of an assembly domain or is due to more general conformational perturbations introduced by the mutation and/or the sugar chain.