Interactions between Na,K-ATPase (cid:1) -Subunit ATP-binding Domains*

The reaction mechanism of the Na,K-ATPase is thought to involve a number of ligand-induced conformational changes. The specific amino acid residues re-sponsible for binding many of the important ligands have been identified; however, details of the specific conformational changes produced by ligand binding are largely undescribed. The experiments described in this paper begin to identify interactions between domains of the Na,K-ATPase (cid:1) -subunit that depend on the presence of particular ligands. The major cytoplasmic loop (be-tween TM4 and TM5), which we have previously shown contains the ATP-binding domain, was overexpressed in bacteria either with a His 6 tag or as a fusion protein with glutathione S -transferase. We have observed that these polypeptides associate in the presence of MgATP. Incubation with [ (cid:2) - 32 P]ATP under conditions that result in phosphorylation of the full-length Na,K-ATPase did not result in 32 P incorporation into either the His 6 tag or glutathione S -transferase fusion proteins. The MgATP-induced association was strongly inhibited by prior modification of the fusion proteins with fluorescein iso-thiocyanate or by simultaneous incubation with 10 (cid:3) M eosin, indicating that the effect of MgATP is due to interactions within the nucleotide-binding domain. These data are consistent with Na,K-ATPase The GST-ABD incubated (cid:4) g of purified Na,K-ATPase solubilized C E . A , SDS-PAGE. The incubation medium contained m M (pH 7.4) alone ( lanes 1 and 2 ) or the of one of following: lane 3 and The were probed with either anti-KETYY (gift or (Affinity There is clearly strong between the full-length Na,K-ATPase in the presence of MgATP lane 7 ) and weaker in the of MgADP ( lane 6 ). , There to be no difference in the ability of the untreated or the C 12 E 8 -treated Na,K-ATPase to hydrolyze suggesting that the was not denatured.

The Na,K-ATPase is an integral membrane protein that plays a central role in ionic homeostasis in animals by mediating the translocation of Na ϩ and K ϩ ions against their electrochemical gradients across the plasma membrane (for a review, see Ref. 1). The Na,K-ATPase functions as a heterodimer composed of a 105-kDa ␣-subunit that spans the plasma membrane 10 times (2) and a ϳ55-kDa glycosylated ␤-subunit that has a short cytoplasmic N-terminal domain, a single transmembrane domain, and a large extracellular domain. At present, whether the ␤-subunit plays a part in the transport process remains unclear, but evidence is accumulating that indicates the impor-tance of ␤ in targeting the enzyme complex to the plasma membrane (3,4). The Na,K-ATPase belongs to a large family of enzymes known as the P 2 -type ATPases (5). Members of this important protein class couple the hydrolysis of ATP to the transmembrane translocation of cations and have been identified in every taxonomic phylum.
Recently, the three-dimensional structures of several P 2 -type ATPases including the H-ATPase (6), the Na,K-ATPase (7), and the sarcoplasmic reticulum Ca-ATPase (8,9) at 8-, 11-, and 2.6-Å resolution, respectively, have been solved. As one might predict, the high resolution sarcoplasmic reticulum Ca-ATPase (SERCA) 1 pump structure shows that the 10-transmembrane segments are largely ␣-helical (8). Indeed, the SERCA crystal structure confirmed several previously proposed structural suggestions based on functional studies, including the suggestion that the fifth and sixth transmembrane domains were surrounded by the other helices as opposed to directly contacting the membrane lipid (10,11). In addition, the structure of the cytoplasmic loops results in the formation of three distinct domains: 1) the N domain, containing the nucleotide-binding site; 2) the P domain, containing the phosphorylation site; and 3) the A domain, referred to as the "activator" domain by Toyoshima et al. (8,9), consisting of the N terminus and the M2M3 loop.
The Na,K-ATPase and the gastric H,K-ATPase are the only members of the P 2 -type ATPase family that possess two obligatory subunits, ␣ and ␤. Consequently, these enzymes have a quaternary structure; whether this structure is simply an ␣␤ protomer or contains higher oligomers remains a central issue of scientific investigation (see Ref. 1). Identification of the subunit domains involved in assembly and trafficking of Na,K-ATPase has been approached by immune precipitation experiments with truncated ␤-subunits (12,13) and chimeras between the Na,K-ATPase and gastric H,K-ATPase ␣-subunits (14,15).
The oligomeric state of the Na,K-ATPase remains a controversial issue. Indeed, it seems hard to dispute the findings that monomeric (i.e. detergent-solubilized) ␣␤ protomers of the sodium pump are sufficient to perform Na,K-ATPase activity (16,17). However, it has been convincingly demonstrated that sodium pump protomer-protomer interactions do indeed take place and that adjacent pumps can be tethered together via chemical cross-linking (18). This apparent paradox has been explained by suggesting that close packing of sodium pumps within high density preparations leads to incidental contact, which can be captured via cross-linking (19). Although some functional measurements appear more easily resolved by a functional diprotomer (20 -22), more complex models for a single functioning protomer can accommodate many of these findings (23). Nevertheless, it is clear that sodium pump molecules are in close proximity to each other within some cell membrane preparations. Although this association may not be necessary for sodium pump action, it may play a role in cell function by bringing other proteins that interact with the sodium pump into close proximity. For example, the Na,K-ATPase has been shown to be a membrane anchor for phosphoinositide-3 kinase in opossum kidney cells (24).
In this paper, we demonstrate that the isolated nucleotide binding domain can directly associate with full-length Na,K-ATPase purified from dog kidney. In addition, this interaction was significantly enhanced by the binding of magnesium and ATP. To further investigate the sites of interaction, we determined whether the isolated ATP-binding domain (ABD) could interact with itself in the absence of other pump domains. In the presence of MgATP, we found that a GST-tagged ABD associated with a His 6 -tagged ABD. Consistent with the facilitation of interaction by nucleotide binding was the observation that both FITC and eosin significantly decreased the degree of association. Taken together, these data imply that the Na,K-ATPase is capable of self-association and that this quaternary structure is stabilized upon MgATP binding. Preliminary accounts of this work have been previously reported (25).

Reagents and Media
Glutathione-Sepharose 4B and Rainbow protein molecular weight markers were from Amersham Biosciences. NaCl, KCl, MgCl 2, Na 2 HPO 4 , NaH 2 PO 4 , glutathione, Tris-base, Coomassie Brilliant Blue R-250, phenylmethylsulfonyl fluoride, antipain, leupeptin, pepstatin A, FITC, eosin, and imidazole were purchased from Sigma. Recombinant streptavidin-tagged superoxide dismutase was a generous gift of John Eisses (Oregon Health and Science University). The bacterial lysis reagent, BPER II, was purchased from Pierce. The primary antibody, mouse anti-penta-His, was obtained from Qiagen, and the secondary antibody, horseradish peroxidase-conjugated goat anti-mouse, was from Sigma.

Protein Expression and Purification
Escherichia coli was used to overexpress the M4M5 cytoplasmic loop from ␣-subunit of the rat Na,K-ATPase. The M4M5 loop includes the amino acid residues between Lys 354 and Lys 774 of the ␣ 1 -subunit from rat. Two fusion protein versions of this peptide were constructed for this study: 1) a glutathione S-transferase (GST)-tagged M4M5 loop and 2) a His 6 -tagged M4M5 loop.

GST-M4M5 Loop Purification
The GST construct was produced via PCR amplification of a 1260-bp section encoding the M4M5 loop from a pGEM-rat ␣ 1 cDNA construct (generous gift from Dr. Robert Mercer, Washington University, St. Louis, MO). An EcoRI restriction endonuclease site was engineered into both primers, allowing a single digestion and ligation step into the multiple cloning site of pGEX-1T vector (Amersham Biosciences). Positively transformed DH5␣ cells were selected for ampicillin resistance conferred by pGEX-1T, and correctly oriented clones were determined via restriction endonuclease mapping and subsequent DNA sequencing.
A single colony was grown overnight and used to inoculate 500 ml of LB amp (100 g/ml) containing 2% ethanol. (We previously observed that the presence of ethanol dramatically increased the amount of fusion protein appearing in the soluble fraction after bacterial cell lysis (26).) After an A 600 of 0.8 was attained, protein synthesis was induced with 0.1 mg/ml isopropyl ␤-D-thiogalactoside, and the cells were grown for an additional 12 h at room temperature. Bacterial cells were collected by centrifugation (7500 ϫ g for 25 min), resuspended in 10 ml of BPER II (Pierce), and lysed via gentle Dounce homogenization. Lysis was performed in the presence of a protease inhibitor mixture containing 80 M phenylmethylsulfonyl fluoride and 10 g/ml each of leupeptin, antipain, and pepstatin A. After lysis, soluble proteins were separated from cellular debris by centrifugation (13,000 ϫ g, 45 min). The GST fusion protein was purified from the supernatant via a glutathione-Sepharose 4B (Amersham Biosciences) affinity column, and then the bound protein was washed with ϳ50 column volumes of buffered saline (140 mM NaCl, 2.7 mM KCl, 10 mM Na 2 HPO 4 , 1.8 mM KH 2 PO 4 , pH 7.3) to remove nonspecifically bound E. coli proteins. The protein was used in subsequent procedures either bound to the support or eluted off the affinity column with 10 mM reduced glutathione. The eluent was dialyzed overnight against 50 mM Tris (pH 7.4) to remove the glutathione and phosphate for interaction experiments. When support-bound GST-protein was used, the column was washed with two volumes of 50 mM Tris (pH 7.4) to remove the phosphate. The size of the GST-M4M5 construct is ϳ63 kDa, estimated by SDS-PAGE according to the method of Laemmli (27). Protein content was determined by the method of Bradford (28). The yield was typically 5-8 mg of GST-M4M5 loop/liter of cell culture.

His 6 -M4M5 Loop Purification
The His 6 -tagged M4M5 loop was produced and purified as described previously (26). Briefly, the identical region of the rat ␣ 1 -subunit (i.e. Lys 354 -Lys 774 ) used for the GST construct was cloned into the sixhistidine fusion protein expression vector, pET-28b (Novagen), between the EcoRI and NdeI sites. The E. coli transformants were selected by the kanamycin resistance (30 g/ml) conferred by pET-28. A single colony was grown overnight, and this culture was used to inoculate 500 ml of LB kan (2% ethanol). After reaching an A 600 of 0.8, protein expression was induced by the addition of a final concentration of 1 mM isopropyl ␤-D-thiogalactoside, and the cells were grown at room temperature for an additional 12 h. Bacteria were collected and lysed, and soluble proteins were separated from cellular debris via centrifugation, as described above for the GST-M4M5 loop. The His 6 -tagged ATPbinding domain was purified by passing the lysate over a Ni 2ϩ affinity column, His-Bind resin (nitrilotriacetic acid-agarose, Novagen). The column was washed with ϳ50 column volumes of binding buffer containing 5 mM imidazole, 500 mM NaCl, and 20 mM Tris-HCl (pH 7.9). The His 6 -tagged protein was eluted from the resin with elution buffer containing 400 mM imidazole, 500 mM NaCl, 20 mM Tris-HCl (pH 7.9). The eluent was dialyzed overnight against 50 mM Tris (pH 7.4) to remove the imidazole for interaction experiments. The size of the His 6 construct is ϳ46 kDa, estimated by SDS-PAGE according to the method of Laemmli (27). Protein content was determined by the method of Bradford (28).

Binding of C 12 E 8 -solubilized Dog Kidney Na,K-ATPase to Immobilized Fusion Protein ATP-binding Domains
Dog kidney Na,K-ATPase was purified by the method of Jorgensen (29) with the modifications described previously (30). In order to assay for binding between the immobilized GST-ABD and dog kidney ATPase, 50 g of dog kidney enzyme in 200 l of 50 mM Tris (pH 7.4) and 0.1% C 12 E 8 was combined with a 200-l slurry of GST-ABD conjugated to glutathione-Sepharose in the presence or absence of enzyme substrates (as shown in the figure legends). Control assays were performed using slurries of glutathione-Sepharose 4B that was not bound to GST-ABD. Microcentrifuge tubes containing the 400-l reaction slurries were rotated at room temperature for 60 min, and the unbound enzyme was removed by centrifugation (1000 rpm in a tabletop microcentrifuge for 5 min). The supernatant was discarded, and the Sepharose pellet was resuspended in 1 ml of 50 mM Tris (pH 7.4, with corresponding substrate where appropriate) and rotated at room temperature for 5 min. The supernatant was removed, and the Sepharose was washed two additional times in the same manner.
Proteins were removed from the Sepharose by the addition of 100 l of Laemmli sample buffer (i.e. 1:1:1 (v/v/v) of 8 M urea, 10% SDS, and 125 mM Tris-HCl, pH 6.8, and 5% ␤-mercaptoethanol), and a 50-l aliquot was resolved on a 7.5% SDS-PAGE gel according to the method of Laemmli (27). After electrophoresis, proteins were transferred onto PVDF membranes by electroblotting in 10 mM CAPS, 10% MeOH, pH 11.0, for 2 h at 180-mA constant current (31). The PVDF membrane was blocked with 10% dry milk protein solution in phosphate-buffered saline for 1 h. The membrane was then incubated with an antibody against the Na,K-ATPase ␣or ␤-subunit for 1 h at room temperature. The primary antibody was removed, and the membrane was washed three times with phosphate-buffered saline plus 0.1% Tween 20 and then incubated for 1 h with the appropriate horseradish peroxidaseconjugated secondary anti-IgG at room temperature. The membrane was then washed five times with phosphate-buffered saline plus 0.1% Tween 20, and the proteins were visualized by chemiluminescent de-tection of peroxidase activity using the SuperSignal substrate kit (Pierce).

Interaction between Expressed ATP-binding Domains via "Pull-down" Assays
The GST-ABD fusion protein was bound to the glutathione-Sepharose and subsequently washed with TBS (50 mM Tris, 120 mM NaCl, pH 7.4) to remove unbound fusion protein. Domain-domain interactions were initiated by adding 50 g of the His 6 -ABD to a slurry of the conjugated GST-ABD in a final volume of 200 -400 l of TBS alone or containing various ligands (see figure legends for details). The fusion proteins were rotated at 4°C for 1-3 h. After the interaction period, the glutathione-Sepharose was pelleted via centrifugation (1000 rpm in a tabletop microcentrifuge for 5 min) and washed three times with a 20-volume quantity of 50 mM TBS, containing any additives that were present during the interaction period. Finally, the GST loop and any His 6 loop bound to it were eluted from the Sepharose with 100 l of 10 mM reduced glutathione or 100 l of Laemmli sample buffer. Proteins were separated via SDS-PAGE and electrotransferred to PVDF as described above. Evidence for His 6 -ABD interaction with GST-ABD was demonstrated via immunostaining with mouse anti-penta-His antibody (Qiagen) and horseradish peroxidase-conjugated goat anti-mouse IgG secondary antibody (Sigma).

Treatment of Fusion Protein ABDs with ATP Site Probes
Eosin-Eosin, tetrabromofluorescein, has been shown to be a potent inhibitor of the Na,K-ATPase by competing with ATP (32). Thus, we performed the domain-domain interaction experiments described above in the presence and absence of 10 M eosin.
FITC-FITC is a potent irreversible inhibitor of the Na,K-ATPase and has been shown to specifically label Lys 501 , which resides in the nucleotide binding site (33,34). The GST loop was modified by incubating the protein, bound to glutathione-Sepharose 4B, with 20 M FITC in 50 mM Tris buffer (pH 9.0, 30 min, 25°C). The unbound FITC was removed by washing the Sepharose twice with a 10-volume quantity of TBS. FITC labeling was confirmed by UV illumination of the labeled protein on a gel (see Fig. 7B, middle panel). The His 6 loop was labeled similarly, except the protein was not bound to the Ni 2ϩ -nitrilotriacetic acid; thus, excess FITC was removed via dialysis (12-kDa cut-off) overnight against a 1000-volume quantity of 50 mM Tris (pH 8.0). In order to determine whether FITC modification was necessary at either or both ABD partners, both FITC-labeled and unlabeled GST-ABD were incubated in the presence of 5 mM MgATP and 50 g of either FITClabeled or unlabeled His 6 -ABD.
Na,K-ATPase Activity Measurements-Na,K-ATPase activity was determined in a standard assay medium containing 1 mM EGTA, 130 mM NaCl, 20 mM KCl, 3 mM MgCl 2 , 3 mM Na 2 ATP, 50 mM imidazole, pH 7.2, and 0.5 g of purified dog kidney enzyme (or enzyme solubilized with C 12 E 8 ). The mixture was incubated at 37°C for 15 min, and the amount of inorganic phosphate released through ouabain-sensitive ATP hydrolysis was measured as described previously (35).
Phosphorylation with [ 32 P]ATP-The phosphorylation measurements were carried out essentially as described previously (36) in 50 l of medium containing 100 mM NaCl (or 100 mM KCl), 5 mM MgCl 2 , 50 mM Tris-HCl, pH 7.2, and 50 g of protein. The reaction was initiated by the addition of ATP ([␥-32 P]ATP (PerkinElmer Life Sciences) and 7.3 M Tris-ATP) and incubated in an ice bath for 60 s. The phosphorylation was stopped with 750 l of "ice-cold" 5% (v/v) perchloric acid containing 0.5 mM Tris-ATP and 1.5 mM Tris-phosphate. The samples were filtered through Millipore filters (pore size 0.45 m), washed three times with 3 ml of stopping buffer, and counted in a scintillation counter. Specific phosphorylation was calculated from the difference between 32 P incorporation in native protein preparations and those where perchloric acid was added before ATP to denature the proteins.

RESULTS
Interaction between Purified Dog Kidney Na,K-ATPase with a Recombinant ATP-binding Domain-For these studies, a GST fusion protein of the large cytoplasmic loop between the fourth and fifth transmembrane segments was constructed. We reported previously that this GST-M4M5 loop was able to bind ATP as determined via protection against FITC labeling (37). More recently, the corresponding domain in the full-length rabbit fast twitch SERCA has been shown to coordinate TNP-AMP binding by x-ray crystallographic analysis (8).
To determine whether the GST-M4M5 loop (GST-ABD) as-sociated with purified full-length Na,K-ATPase, C 12 E 8 -solubilized dog kidney Na,K-ATPase (50 g) was incubated at 25°C with a slurry of the GST loop bound to a glutathione-Sepharose affinity resin. The incubation was performed both in the absence and presence of pump substrates (Fig. 1A). There was a clear interaction between the GST-ABD and the intact Na,K-ATPase in the presence of MgATP (Fig. 1A, lane 6). Occasion- After the incubation period, the Sepharose was washed with the corresponding media, and bound proteins were removed by adding Laemmli sample buffer. Aliquots were run on a 7.5% Laemmli gel, and proteins were electrotransferred to a PVDF membrane. The membranes were probed with either anti-KETYY antibody (upper gel) (gift from Dr. Jack Kyte, University of California, San Diego) or anti-␤ antibody (lower gel) (Affinity Bioreagents). There is clearly a strong association between the GST-ABD and the full-length Na,K-ATPase in the presence of MgATP (lane 7) and a somewhat weaker association in the presence of MgADP (lane 6). B, ATPase activity. There appears to be no difference in the ability of the untreated or the C 12 E 8 -treated Na,K-ATPase to hydrolyze ATP, suggesting that the enzyme was not denatured. ally, we observed that there was a less prominent band in the presence of MgADP as well (Fig. 1A, lane 5), suggesting that MgADP may also promote an interacting conformation. Specifically, we observed a MgADP-facilitated interaction in three of five experiments where the PVDF membrane was probed with an anti-␣ antibody and in one experiment where membrane was probed with anti-␤ (Fig. 1B). In contrast, we never observed GST-ABD/Na,K-ATPase interactions in the absence of substrates or in the presence of sodium, potassium, or magnesium, alone. Additionally, when experiments were performed with dog kidney enzyme that had not been treated with C 12 E 8 (see "Experimental Procedures"), no interactions were observed in the presence or absence of any substrates (data not shown). The observed interactions were not due to C 12 E 8 denaturation of the enzyme, since the detergent-solubilized enzyme has been shown to retain function (16,38), and we confirmed these findings for our experimental conditions (Fig. 1C).
We next determined the concentration of MgATP required to stabilize the interaction between the GST-ABD and C 12 E 8solubilized sodium pump. These experiments are possible because the affinity for MgATP binding to the full-length Na,K-ATPase is more than 3 orders of magnitude greater than to the bacterially produced constructs (see "Discussion") (26). Fig. 2 shows that near millimolar MgATP concentrations are necessary to stabilize the heterodimer between GST-ABD and the full-length pump.
Evidence Suggesting That Interactions Take Place within the Nucleotide-binding Domain-The Na,K-ATPase nucleotidebinding domain comprises nearly 40% of the mass of the catalytic ␣-subunit. Furthermore, it has been shown that this domain undergoes dramatic conformational changes in response to substrate binding (39,40). Consequently, we decided to test whether the interactions observed between the intact Na,K-ATPase and the GST-ABD were mediated via contacts within the ATP-binding domain. These experiments were performed using either the GST-ABD or a His 6 -tagged fusion protein with the same polypeptide as the GST-ABD. The nucleotide-binding properties of the His 6 -ABD were described previously (26); in addition, we found that the His 6 -ABD, like the GST-ABD, associates with the intact Na,K-ATPase in a MgATP-dependent manner (Fig. 3). Therefore, the observed interactions between native Na,K-ATPase and the fusion proteins were mediated by enzyme contacts with the ATP-binding domain and not by either of the fusion protein affinity tags.
However, the segment (or segments) of the full-length Na,K-ATPase that interacted with the bacterially produced ATPbinding domains remained to be determined. Thus, we initially designed experiments to determine whether the interactions occurred between the two ATP-binding domains (i.e. the M4M5 loop of the intact enzyme with the equivalent heterologously expressed isolated domain). We tested this hypothesis by measuring interactions between the two purified cytoplasmic loops themselves. The soluble His 6 -ABD was incubated with GST-ABD bound to glutathione-Sepharose resin in the presence of varying substrates. We observed that the two constructs did associate and that this association depended upon the simultaneous presence of both magnesium and ATP (Fig. 4A). In other words, neither ATP alone nor magnesium alone was an effective promoter of the interaction. MgADP and MgAMP were unable to facilitate interaction between the two nucleotide-binding domains (Fig. 4B). The inability of MgADP to promote interactions between the GST-ABD and the His 6 -ABD, was somewhat surprising, considering that it did facilitate interactions between the GST-ABD and intact Na,K-ATPase (Fig. 1A). Thus, this finding may suggest that MgADP binding to the full-length ␣-subunit elicits a conformation slightly different than when it is bound to the isolated M4M5 loop alone. Indeed, the nucleotide-bound crystal structure of SERCA indicates that both the N terminus and the M2M3 cytoplasmic loop (i.e. the "A" domain) are in close proximity to the M4M5 loop (8,9); the A domain is obviously absent from our bacterial constructs.
It was important to ensure that MgATP does not facilitate nonspecific protein interactions with GST-ABD or glutathione-Sepharose. Therefore, we incubated a bacterially produced superoxide dismutase (SOD) fusion protein with the GST-ABD under the same conditions that promote interaction with the His 6 -ABD. Fig. 5 shows that there was no interaction between the GST-ABD and SOD in the presence or absence of MgATP. The Western blot was probed with an anti-SOD antibody, and a positive control lane shows that the immunostaining was successful (Fig. 5) and that SOD would have been detectable had it interacted with the GST-ABD.
Specificity of the Interactions between ATP-binding Domains-To more directly examine the MgATP-dependent asso- . The GST-ABD bound to glutathione-Sepharose was incubated with 50 g of purified Na,K-ATPase solubilized with 0.1% C 12 E 8 . The incubation medium contained 50 mM Tris (pH 7.4) with the indicated concentrations of magnesium and ATP. After the incubation period, the Sepharose was washed with the corresponding media, and bound proteins were removed by adding Laemmli sample buffer. Aliquots were run on a 7.5% Laemmli gel, and proteins were electrotransferred to a PVDF membrane. The membranes were probed with anti-␣ antibody (Affinity Bioreagents). The lack of protein-protein interactions at low MgATP suggests that the GST-ABD must be in the nucleotide-bound state to interact with the ␣-subunit.

FIG. 3. His 6 -ABD interaction with C 12 E 8 -solubilized dog kidney
Na,K-ATPase. The His 6 -ABD bound to Ni 2ϩ -nitrilotriacetic acid was incubated with 50 g of purified Na,K-ATPase solubilized with 0.1% C 12 E 8 . The incubation medium contained 50 mM Tris (pH 7.4) with or without 3 mM MgATP. After the incubation period, the Ni 2ϩ -nitrilotriacetic acid was washed with the corresponding media, and bound proteins were removed by adding Laemmli sample buffer. Aliquots were run on a 7.5% Laemmli gel, and proteins were electrotransferred to a PVDF membrane. The membranes were probed with anti-KETYY antibody. It is clear that the presence of MgATP dramatically increased the amount of Na,K-ATPase bound to the His 6 -ABD. ciation between these ATP-binding domains, we determined whether known inhibitors of ATP binding to the Na,K-ATPase could block the interaction. For example, eosin is a reversible inhibitor of the Na,K-ATPase, and the binding of eosin and ATP are mutually exclusive (32). Thus, eosin should block GST-ABD and His 6 -ABD interactions, by preventing MgATP binding. Consistent with this prediction, the presence of 10 M eosin greatly reduced the binding of His 6 -tagged loop to GSTtagged loop in the presence of 1 mM MgATP (Fig. 6). Both the Coomassie Blue-stained gel (Fig. 6A) and the anti-penta-His stained Western blot (Fig. 6B) indicate that a much smaller amount of the His 6 -tagged loop bound to the GST fusion protein in the presence of eosin.
FITC is a fluorescent amine-reactive molecule that labels Lys 501 in the purified Na,K-ATPase; this reaction is prevented by the simultaneous presence of ATP (33,34). Similarly, ATP has been shown to protect both the His 6 -ABD (26) and the GST-ABD (37) against FITC labeling. When both the GST-ABD and His 6 -ABD were labeled with FITC, domain-domain interactions were substantially reduced (Fig. 7B, right panel). Fig.  7B (middle panel) shows FITC incorporation into the GST-ABD, whereas FITC-labeled His 6 -ABD is not observed, since it did not associate with the GST-ABD and thus was lost during the washing steps. (The FITC labeling protocol used for His 6 -ABD modification was identical to those published previously (26).) Interestingly, FITC labeling of only one of the interacting fusion proteins did not inhibit the interaction; clearly, FITC modification of the His 6 -ABD (Fig. 7A, middle panel) did not significantly reduce its ability to associate with the nonmodified GST-ABD (Fig. 7A, right panel).
Phosphorylation Cannot Explain Domain-Domain Interactions-It is clear that MgATP, and not ATP alone, promotes the observed ABD interactions. An obvious possibility might be that the proteins are undergoing magnesium-dependent phosphorylation. However, there was no difference between the K ϩ -containing control with the intact Na,K-ATPase and either the His 6 -ABD or the GST-ABD in the presence of Na ϩ or K ϩ (Fig. 8). This may not be surprising, since the isolated ABDs are devoid of cation-binding sites. Nonetheless, the phosphorylation levels observed for the fusion proteins (Fig. 8), compared with the level of 32 P i captured on the filter with perchloric acid-denatured protein (i.e. background), was minimal. Moreover, considering that the fusion proteins are less than half the molecular mass of the Na,K-ATPase and constitute a purer protein preparation, the difference in phosphoprotein production between the intact enzyme and the fusion proteins is a very conservative estimate (Fig. 8). Also, considering that FIG. 4. Binding between the GST-ABD and His 6 -ABD is elicited by MgATP. GST-ABD tethered to glutathione-Sepharose was incubated with a 50-g quantity of His 6 -ABD as described under "Experimental Procedures." A, separately, incubations contained 5 mM ATP, AMP, P i , or TBS in the absence of Mg 2ϩ ions (left) or combined with 5 mM Mg 2ϩ (right). B, a separate experiment is shown that also included 5 mM MgADP. Both blots were probed with anti-penta-His antibody (Qiagen). Clearly, both magnesium and ATP together produce the most stable interaction between the two constructs.

FIG. 5. Specificity of the GST-ABD for binding the His-ABD.
In order to determine whether there were significant nonspecific interactions between bacterially purified fusion proteins and the GST-ABD and/or the glutathione resin, we used an unrelated protein, streptavidin-tagged SOD, in identical interaction experiments. Clearly, there was no interaction between SOD and the GST-ABD in the presence or absence of ATP. As a positive control for immunodetection, we ran 1 g of purified SOD in an adjacent lane (left). neither the His 6 -ABD nor the GST-ABD can hydrolyze ATP (data not shown), the inability to isolate a phosphorylated ABD (Fig. 8) cannot be due to a more labile acyl-phosphate intermediate. Rather, it seems likely that the ABD constructs cannot "close" sufficiently to bring the N and P domains together, a requirement of phosphoenzyme formation (8,9). This inability to close is probably due to the lack of both the membrane domains and the A domain in the isolated ABDs. Thus, our data suggest that the protein conformation producing the strongest domain-domain interactions is produced simply by the binding of both magnesium and ATP and not phosphoenzyme formation.

DISCUSSION
The high resolution structures of the Ca-ATPase (8,9) clearly show that, as predicted previously (41), the cation and nucleotide binding domains of P 2 -type ATPases are not only functionally separate but are spatially separate as well. In other words, all of the residues involved in ATP-binding and hydrolysis are located in the cytoplasmic loop between M4 and M5, forming the two separate N and P subdomains. Conversely, all of the residues suggested to be involved with cation coordination in the occluded state are located in the transmembrane-spanning regions of the enzyme (1). Since occupation of the cation-binding site dramatically alters nucleotide affinity, it is obvious that communication exists between the cationbinding membrane domains and the cytoplasmic ATP-binding domain. Indeed, the conformational changes observed between the E 1 and E 2 SERCA structures (see Refs. 8 and 9, respectively) show dramatic movements of transmembrane helices and changes in their structure, which push and pull on large cytoplasmic domains. Instability of helices associated with cation binding has been observed previously in both the Na,K-ATPase (10) and the gastric H,K-ATPase (11).
In the current work, we provide evidence that tight proteinprotein interactions occur between two differentially tagged constructs of the large cytoplasmic loop between M4 and M5 of the Na,K-ATPase. This loop has been shown to contain all of the residues that compose the nucleotide binding domain in the sodium pump (26) as well as other members of the P 2 -type ATPase family (8,42). The ABD-ABD interaction reported here was dependent upon the presence of both magnesium ions and ATP, yet neither fusion protein was phosphorylated (Fig. 8); nor did they possess significant ATPase activity (26). These experiments were designed to determine whether FITC modification of both the GST-and His 6 -ABDs was required to prevent the MgATP-dependent association. For FITC modification, the fusion proteins were treated with 20 M FITC for 30 min at room temperature (50 mM Tris, pH 9.0). Unreacted FITC was removed via dialysis (see "Experimental Procedures"). A, Unlabeled GST-ABD, tethered to glutathione-Sepharose, was incubated with a 50-g quantity of either unlabeled His 6 -ABD or FITC-labeled His 6 -ABD. Interactions were measured in the presence of 1 mM MgATP. Equal aliquots from the respective interactions were run in separate lanes on a 12% Laemmli gel. The gel was cut in half, and one section was used for Western analysis with anti-penta-HIS antibody (right panel), whereas the other half was first photographed under UV illumination (middle panel) and then stained with Coomassie Brilliant Blue (left panel). Lane 1 in each panel shows interactions in the complete absence of FITC labeling. Lane 2 in each panel shows the interaction between unlabeled GST-ABD and FITC-labeled His 6 -ABD. Clearly, FITC-modified His 6 -ABD was still able to associate with unlabeled GST-ABD in the presence of 1 mM MgATP (right panel, lane 2). B, FITC-labeled or unlabeled GST-ABD, tethered to glutathione-Sepharose, was incubated with a 50-g quantity of either unlabeled His 6 -ABD or FITC-labeled His 6 -ABD. Interactions were measured in the presence of 1 mM MgATP. Equal aliquots from the respective interactions were run in separate lanes on a 12% Laemmli gel. The gel was cut in half, and one section was used for Western analysis with anti-penta-His antibody (right panel), whereas the other half was first photographed under UV illumination (middle panel) and then stained with Coomassie Brilliant Blue (left panel). Lane 1 in each panel shows interactions when both fusion proteins were labeled with FITC. Lane 2 is the control, showing normal association in the complete absence of FITC modification. Clearly, FITC modification of the fusion proteins dramatically reduced their ability to associate with one another (right panel, lane 1).

FIG. 8. Phosphorylation of the Na,K-ATPase and bacterially expressed domains by [ 32 P]ATP.
Purified His 6 -ABD, dog kidney Na,K-ATPase, and GST-ABD were incubated with [ 32 P]ATP on ice as described under "Experimental Procedures." 32 P i associated with protein denatured a priori with perchloric acid was subtracted from nondenatured protein, and the difference is shown in the figure as specific 32 P i incorporation. There appears to be no covalent transfer of phosphate to either fusion protein, whereas native Na,K-ATPase was readily phosphorylated. Data represent means, and bars represent the S.E. from three experiments with triplicate determinations.
One Versus Two ATP Molecules Bound-Whether the MgATPdependent interaction between the M4M5 loops required both partners to have MgATP bound remained a question. Thus, we determined the MgATP concentration dependence for the interaction between GST-ABD and intact (C 12 E 8 -solubilized) Na,K-ATPase. It was shown previously that the isolated ATPbinding domain has a K d(ATP) of ϳ500 M, consistent with the ABD existing in an E 2 -like conformation (26). In contrast, the binding affinity for ATP of intact Na,K-ATPase is less than 1 M (43). Thus, we measured interactions between the GST-ABD and intact enzyme at varying MgATP concentrations (Fig.  2). At low MgATP concentrations (e.g. Ͻ50 M), where intact enzyme is saturated with ATP and GST-ABD is less than 10% occupied by ATP, no protein-protein interactions were detected (Fig. 2). However, at MgATP concentrations greater than 500 M, when both the GST-ABD and intact enzyme are largely in the ATP-bound state, the GST-ABD was able to pull down the intact Na,K-ATPase ␣-subunit (Fig. 2). These observations are consistent with the necessity of ATP binding to both interacting proteins to facilitate association, or alternatively, they may reflect a necessary MgATP-dependent stabilization or decrease in flexibility of the soluble GST-ABD to achieve this stable protein-protein interaction.
If ATP-binding domain interactions only occur when each protein is in the nucleotide-bound form, then modification of one of the fusion proteins with FITC should eliminate ABD-ABD interactions, because the FITC-labeled partner would be unable to bind MgATP. However, we found that labeling either the GST-ABD or the His 6 -ABD could not disrupt the MgATPdependent association between the two (Fig. 7A). Rather, FITC modification of both fusion proteins was required to block the GST-ABD/His 6 -ABD interaction (Fig. 7B, right panel). Consequently, it appears that MgATP binding is only necessary to one of the constructs to facilitate protein-protein interactions, or that a single ATP molecule is binding in part to both fusion proteins and forming a bridge between the two ABDs. In other words, FITC modifies Lys 501 that forms part of the binding site for the adenosine moiety in the N domain of the Na,K-ATPase ␣-subunit. Since FITC binding to both partners is required to prevent association, either partner can supply this segment. The other partner may provide the terminal phosphate binding segment (close to Asp 369 ) in the P domain. In this way, the ATP molecule bridges two ABD polypeptide loops via N and P domains from each member of the dimer (see below).
Intermolecular Interactions between Adjacent Sodium Pumps-Whether P-type ATPases exist as monomers, dimers (diprotomers for the H,K-and Na,K-ATPases), or higher oligomers is a subject of debate. There have been several reports demonstrating oligomeric forms of various P-type ATPases (e.g. the sarcoplasmic reticulum Ca-ATPase (44), the Na,K-ATPase (45), and the H,K-ATPase (46)). In particular, there have been several studies that suggest the Na,K-ATPase exists as an ␣/␤ multimer (i.e. (␣␤) n ). Initially, (␣␤) 2 models were proposed to explain the biphasic kinetics of ATP on enzyme activity (47) and later expanded to (␣␤) 4 models (48). Structural support for oligomeric models comes from a number of diverse studies, including cross-linking (18), fluorescence resonance energy transfer between the ATP-site probes FITC and ErITC (22), and more directly by co-immunoprecipitation experiments (49).
It has been suggested that the findings of the cross-linking and energy transfer studies could be a result of the high density of Na,K-ATPase in kidney membrane preparations used in these studies (19). Indeed, a recent study reported that thermal denaturation of the Na,K-ATPase resulted in the formation of ␣-␣ dimer and tetramer aggregates (50). Interestingly, these oligomers were devoid of ␤and ␥-sub-units, demonstrating that the important contacts were within the ␣-subunit (50). Blanco et al. (49) demonstrated associations between different ␣-subunit isoforms (i.e. ␣ 1 and ␣ 3 ) heterologously expressed in insect cells, via co-immunoprecipitation experiments with isoform-specific antibodies. In these experiments, the expressed Na,K-ATPase is less than 10% of the total membrane protein and most likely closer to 1-2% (3). Clearly, the pump density in insect cell preparations is sufficiently low that ␣-␣ interactions are unlikely to have been caused by overcrowding. In the present study, the ability of both GST-and His 6 -ABD to pull down intact ␣-subunit from a C 12 E 8 -solubilized kidney preparation of Na,K-ATPase (Figs. 1 and 3) in the presence of defined ligands is more consistent with specific interactions than incidental contact. Indeed, the lack of interaction between the ABDs and intact Na,K-ATPase ␣-subunit, in the absence of solubilization, suggests that endogenous pump-pump associations must be broken before the exogenous ABDs can bind. Moreover, the demonstration that the two tagged ABDs associate with the same substrate dependence as with the intact Na,K-ATPase indicates that pump-pump interactions may be solely through interactions between the large cytoplasmic loops. Indeed, Koster et al. (45) reached similar conclusions based upon ␣-␣ associations measured between different Na,Kand H,K-ATPase chimeric enzymes expressed in insect cells. Specifically, the sodium pump ␣ 1 subunit selectively associated with chimeras containing the Na,K-M4M5 loop but not with constructs containing the H,K-ATPase large cytoplasmic loop (45). In fact, these authors observed that wild-type sodium pump ␣ 1 associated with a chimera containing only the sodium pump M4M5 loop inserted within the H,K-ATPase (45).
Significance of ␣-␣ Interactions-It seems clear that the catalytic properties of Na,K-ATPase can be mediated by a functional monomeric form (i.e. ␣␤) of the enzyme (16,19,23,38). Consequently, the relevance of the higher oligomeric states sometimes proposed for this enzyme remains a puzzle. There are several reports measuring various kinetic properties of Na,K-ATPase function that can be easily explained by ␣-␣ interactions. For example, fluorescence resonance energy transfer between several Na,K-ATPase labeling probes reveals distances great enough to indicate that the probes reside on different protomers (22). These observations are directly in contrast with FRET measurements from another laboratory indicating that FITC and Co(NH 3 ) 4 ATP are close enough to reside on a single ␣␤ protomer (52). In addition, the ability of TNP-ADP to inhibit the residual p-nitrophenyl phosphatase activity of FITC-labeled ␣␤ protomers, solubilized with C 12 E 8 , makes it difficult to justify invoking the involvement of additional protomers (21). Indeed, Martin and Sachs (23) convincingly demonstrate that the appearance of the "low affinity" nucleotide site, which allows TNP-ADP to bind and inhibit p-nitrophenyl phosphatase activity, is actually a result of FITC modification and does not appear on native enzyme. Thus, there seems to be no compelling functional reason to suggest that the Na,K-ATPase exists as a diprotomer (or higher oligomer). Nevertheless, the current observations (Figs. 1-3) as well as previous reports from other laboratories (18, 45, 48 -50) strongly suggest that under certain conditions the Na,K-ATPase does self-associate.
A Cell Biological Role of Dimerization?-Recently, Caplan and colleagues (53) initially identified, using the yeast twohybrid system, several cellular proteins that interacted with the Na,K-ATPase. Following careful characterization, specific candidates were identified: 1) polycystin-1, involved in polycystic kidney disease; 2) SNAPAP, a protein involved in vesicular targeting; and 3) the catalytic subunit of protein phosphatase 2a (53). Interestingly, the site of interaction of the Na,K-ATPase with these proteins was suggested to be the M4M5 loop by measuring direct protein-protein interactions with a bacterially purified GST-M4M5 loop, a preparation equivalent to the construct used in our studies. 2 Furthermore, the interactions of these foreign proteins with the GST-M4M5 loop were significantly enhanced by the presence of MgATP. 2 Considering the observations of Caplan and co-workers in light of our MgATP-facilitated ABD-ABD dimerization, it is possible that the role of sodium pump dimerization may be related to proper trafficking or regulation (or both) via proteinprotein interactions. In other words, the association of other cellular regulatory proteins (e.g. protein phosphatase 2a or SNAPAP) may depend upon the pump first self-associating and then forming a binding site for the regulatory partner.
ABD-ABD Dimerization Mimicking Intramolecular Interactions-The E 1 Ca SERCA structure (8) shows that the M4M5 loop is separated into two distinct structures comprising the nucleotide-binding domain (N domain) and the phosphorylation domain (P domain). These two domains are separated by ϳ50 Å in E 1 ; thus, it was predicted that a significant conformational closure must occur for the transfer of phosphate from the nucleotide (in the N domain) to the catalytic aspartate residue (in the P domain). This closure was observed recently with the thapsigargin-stabilized high resolution E 2 Tg SERCA structure (9). Although these two structures have been invaluable for understanding the structure-function relationship of the P-type ATPases, the steps that occur between E 1 Ca and E 2 Tg remain unclear. It appears that the binding of MgATP to the N domain results in a conformational change that is more receptive for interacting with the P domain. In addition, significant concomitant changes within the cation-binding domain in the membrane along with the A domain also occur between E 1 Ca and E 2 Tg (9). However, what remains unresolved is whether initial movements between the N and P domains produce changes in the cation-binding and A domains, or vice versa.
In the intact Na,K-ATPase, closure of the N and P domains occurs in a single ␣-subunit. However, an intermolecular interaction might account for our observed GST-ABD/His 6 -ABD association. Our observations that FITC modification was required of both fusion proteins to disrupt the interaction suggest that eliminating MgATP binding to one of the partners alone was without effect. In the native protein, it seems likely that following MgATP binding, closure of P and N domains occurs, facilitated by changes in the intramembrane helices and the A domain. Since the GST-ABD and His 6 -ABD are devoid of both the transmembrane segments and A domains, they would be unable to close upon themselves (i.e. an intramolecular N and P interaction cannot occur within either fusion protein). Rather, MgATP binding induces a conformational change in the N domain within one fusion protein, but the P domain for the interaction is supplied from another molecule that interacts. Thus, FITC modification of one partner would not eliminate interactions, since that partner would supply the P domain for the interaction (see above).
An alternative explanation is that MgATP forms a chemical bridge between the GST-ABD and the His 6 -ABD. In this case, the adenine moiety could bind to a portion of the N domain on one construct, and the phosphate chain would occupy the appropriate contacts on the adjacent construct. In the experiments where only one construct was FITC-modified, that fusion protein would provide the phosphate chain contacts, whereas the unmodified construct could coordinate the adenine moiety. In this case, MgADP may be too short to effectively bridge the two constructs together.
Conclusions-We have shown that two separate constructs of Na,K-ATPase ATP-binding domain can interact with each other or with C 12 E 8 -solubilized intact enzyme. Moreover, this interaction is greatly enhanced by MgATP binding. Although our data cannot rule out the possibility of higher oligomeric structures, the simplest explanation for these interactions is that the sodium pump can exist as a dimer under certain circumstances. Indeed, ATP-dependent dimerization of the two ATP-binding domains within MJ0796 (an ABC transporter), has been demonstrated recently (54). The ATP-mediated ABD dimerization in the ABC transporters has been suggested to be critical for coupling ATP hydrolysis to substrate translocation (55).
Alternatively, such dimerization between Na,K-ATPase ABDs may perform a necessary cellular function as a membrane anchor for other signaling proteins (24,53). Experiments are under way to confirm whether the associations presented here are dimers or higher oligomers and to further understand the role that these sodium pump interactions may play in cellular physiology.