The Rapid-onset Dystonia Parkinsonism Mutation D923N of the Na+,K+-ATPase α3 Isoform Disrupts Na+ Interaction at the Third Na+ Site*

Rapid-onset dystonia parkinsonism (RDP), a rare neurological disorder, is caused by mutation of the neuron-specific α3-isoform of Na+,K+-ATPase. Here, we present the functional consequences of RDP mutation D923N. Relative to the wild type, the mutant exhibits a remarkable ∼200-fold reduction of Na+ affinity for activation of phosphorylation from ATP, reflecting a defective interaction of the E1 form with intracellular Na+. This is the largest effect on Na+ affinity reported so far for any Na+,K+-ATPase mutant. D923N also affects the interaction with extracellular Na+ normally driving the E1P to E2P conformational transition backward. However, no impairment of K+ binding was observed for D923N, leading to the conclusion that Asp923 is specifically associated with the third Na+ site that is selective toward Na+. The crystal structure of the Na+,K+-ATPase in E2 form shows that Asp923 is located in the cytoplasmic half of transmembrane helix M8 inside a putative transport channel, which is lined by residues from the transmembrane helices M5, M7, M8, and M10 and capped by the C terminus, recently found involved in recognition of the third Na+ ion. Structural modeling of the E1 form of Na+,K+-ATPase based on the Ca2+-ATPase crystal structure is consistent with the hypothesis that Asp923 contributes to a site binding the third Na+ ion. These results in conjunction with our previous findings with other RDP mutants suggest that a selective defect in the handling of Na+ may be a general feature of the RDP disorder.

dient of involvement and lack of response to dopaminergic medications. The disorder typically presents in the teens to twenties, triggered by stressful events. The pathophysiological mechanism underlying RDP is poorly understood.
Na ϩ ,K ϩ -ATPase uses the energy liberated by ATP hydrolysis to actively transport three Na ϩ out of the cell in exchange for two K ϩ during each enzyme cycle, thereby creating transmembrane Na ϩ and K ϩ gradients fundamental to many physiological functions, including electrical excitability and neurotransmitter uptake in the nervous system. The catalytic ␣-subunit of the Na ϩ ,K ϩ -ATPase ␣␤␥ complex exists in four different isoforms (␣1-␣4) encoded by different genes whose expression is developmentally regulated and tissue-specific (3,4). In the brain, ␣2 and ␣3 are expressed preferentially in glial cells and neurons, respectively (5).
Recently the three-dimensional structure of Na ϩ ,K ϩ -ATPase with two K ϩ /Rb ϩ bound at membranous transport sites has been determined by x-ray crystallography (6,7). It is believed that two Na ϩ ions can bind at sites that overlap substantially with the K ϩ sites, whereas the third Na ϩ ion is bound at a different, yet unknown site that in some way interacts functionally with the C terminus (6,8).
Nine different RDP mutations have thus far been identified in the ␣3 gene (1, 2, 9 -12). Here, we report on the functional consequences of an RDP mutation replacing Asp 923 by an asparagine (D923N). 4 This was identified as a de novo mutation in two unrelated patients; that is, a child with onset of bulbar symptoms and gait disturbance already at age 4 years after a mild head injury and later development into somewhat exceptional symptoms consisting of periods of flaccidity and stiffness (11) and an adult with onset of typical symptoms at age 20 years after an exhaustive 5-day march while serving in the army (12). We have characterized the D923N mutant functionally and show that D923N interferes profoundly with Na ϩ interaction, probably disturbing the third Na ϩ site of Na ϩ ,K ϩ -ATPase. D923N is one of several RDP mutations found to interfere selectively with Na ϩ interaction without effect on K ϩ interaction, and our findings reveal new information about the location of the third Na ϩ site of Na ϩ ,K ϩ -ATPase.

EXPERIMENTAL PROCEDURES
Mutagenesis and Expression-Using the QuikChange sitedirected mutagenesis kit (Stratagene), mutation D923N was introduced directly into full-length cDNA encoding the human ␣3-isoform of Na ϩ ,K ϩ -ATPase previously made ouabain-resistant by mutations Q108R and N119D (13). Recombinant expression in COS-1 cells was carried out applying the ouabain selection methodology (13,14), which relies on the more than a 100-fold difference in ouabain sensitivity between the endogenous COS-1 cell Na ϩ ,K ϩ -ATPase and the exogenous Na ϩ ,K ϩ -ATPase, allowing stable cell lines expressing the exogenous enzyme to be isolated in the presence of ouabain, due to preferential inhibition of the endogenous Na ϩ ,K ϩ -ATPase (13)(14)(15)(16). The stable integration of the D923N mutation into the genome of the cells was verified by sequencing of the genomic DNA (17).
Functional Analysis-Plasma membrane vesicles isolated by differential centrifugation (14) were made leaky by treatment with alamethicin or sodium deoxycholate before functional analysis. Na ϩ ,K ϩ -ATPase activity was measured at 37°C by following the liberation of P i by the Baginski method (14,18). For kinetic measurements of steady-state and transient reactions, a manual mixing technique (at 0°C) or a quench-flow module (QFM-5, Bio-Logic Science Instruments, Claix, France) (at 25°C) was applied (17, 19 -22). Ouabain (10 M) was included in all experiments to make the contribution from the endogenous Na ϩ ,K ϩ -ATPase negligible.
In all phosphorylation experiments the acid-precipitated 32 P-labeled phosphoenzyme was washed by centrifugation and subjected to SDS-polyacrylamide gel electrophoresis at pH 6.0, and the radioactivity associated with the separated Na ϩ ,K ϩ -ATPase band was quantified by "imaging" using a Packard Cyclone TM Storage Phosphor System (15,20).
Data Analysis and Statistics-All experiments were performed at least three times on different enzyme preparations, and average values are shown in the figures, with error bars indicating standard errors (seen only when larger than the size of the symbol). Data were processed using SigmaPlot (SPSS Inc.) for nonlinear regression analysis with the equations described in the legends to Fig. 1 and Fig. 2 (16,17,20). Ligand concentration dependences with both an activating and an inhibition phase were fitted by a function consisting of two Hill components as previously (16).
To analyze the dephosphorylation time courses, the kinetic simulation software SimZyme (23) was applied. For any choice of reaction path and rate constants, SimZyme solves the relevant differential equations using the fourth order Runge-Kutta numerical method and provides a graphical representation of the time dependence of the concentration of each of the reaction intermediates. The rate constants and initial concentrations of reaction intermediates providing the best fit to the experimental data points were determined by comparing computed time courses with the data points (see examples in supplemental Fig. S1).
Structural Modeling-The structural model was created by substituting the relevant amino acid side chains of the Ca 2ϩ -ATPase in appropriate conformation (see the legend to Fig. 6) with those of the Na ϩ ,K ϩ -ATPase followed by energy minimization and selection of appropriate rotamers of the inserted residues using the DeepView/Swiss-PdbViewer program (Swiss Institute of Bioinformatics). The structure-based sequence alignment of the transmembrane region shown in supplemental Fig. S2 was used in the construction of the model. Structural figures were prepared using PyMol. This software was also used to make the structural alignment.

RDP Mutant D923N Is Well
Expressed-Like the wild type enzyme, mutant D923N was capable of conferring ouabain resistance to the COS-1 cells, as evidenced by sustained cell viability in the presence of ouabain, thus indicating that D923N does transport Na ϩ and K ϩ across the membrane at a certain rate. Measurement of the maximal amount of acid-stable phosphoenzyme intermediate formed from [␥-32 P]ATP ("active-site concentration" (19)) on the isolated plasma membrane fraction showed that D923N is well expressed at a level above wild type (ratio between mutant and wild type active-site concentrations 1.7 Ϯ 0.23, n ϭ 13). The catalytic turnover rate (Na ϩ ,K ϩ -ATPase activity per active site) determined at 130 mM Na ϩ , 20 mM K ϩ , and 3 mM MgATP was, on the other hand, reduced by no less than 90% in D923N (843 Ϯ 27 min Ϫ1 (n ϭ 10) versus 8199 Ϯ 209 min Ϫ1 in wild type (n ϭ 10)). Taken together, these results show that the RDP mutation does not prevent the biogenesis and plasma membrane targeting of Na ϩ ,K ϩ -ATPase but, rather, interferes with its function.
D923N Reduces the Affinity for Intracellular Na ϩ Conspicuously-It is generally believed that Na ϩ ,K ϩ -ATPase exchanges three intracellular Na ϩ ions for two extracellular K ϩ ions in a consecutive mechanism, where Na ϩ triggers phosphorylation from ATP by binding at internally facing sites, and K ϩ activates dephosphorylation by binding at externally facing sites, cf. Scheme 1 (24 -26). At the internal sites K ϩ inhibits by binding in competition with Na ϩ , whereas Na ϩ inhibits at the external sites by binding in competition with K ϩ . Examination of the Na ϩ dependence of ATPase activity in the presence of K ϩ (Na ϩ ,K ϩ -ATPase activity) showed a significant reduction of the apparent affinity for Na ϩ activation in the D923N mutant as compared with wild type (Fig. 1A). The mutation furthermore SCHEME 1. Post-Albers model for the reaction cycle of the Na ؉ ,K ؉ -ATPase. Cytoplasmic and extracellular ions are indicated by subscript c and e, respectively. Occluded ions are shown in brackets. E 1 and E 2 are the Na ϩ -and K ϩ -selective conformations, respectively. E 1 P and E 2 P are phosphoenzyme intermediates sensitive to ADP and K ϩ , respectively. The binding of three Na ϩ ions from the cytoplasmic side to E 1 activates phosphorylation from ATP. The binding of two K ϩ ions from the external side to E 2 P activates dephosphorylation by hydrolysis (24 -26). Two Na ϩ ions may replace K ϩ here, resulting in a dephosphorylation, which is slower than with K ϩ . caused a shift toward higher Na ϩ concentrations of the inhibition of ATPase activity resulting from Na ϩ interaction with externally facing low affinity sites. By fitting a function consisting of two Hill components, one corresponding to the rising phase and the other corresponding to the inhibition phase, a ϳ25-fold reduction of the Na ϩ affinity for activation of ATPase activity and a ϳ1.7-fold reduction of the Na ϩ affinity for inhibition was revealed.
To determine the Na ϩ affinity of the internal sites without interference from K ϩ , the Na ϩ dependence of phosphorylation from ATP was examined in the absence of K ϩ and presence of oligomycin to stabilize the phosphoenzyme (25)(26)(27). As seen in Fig. 1B, D923N exhibited a most conspicuous 216-fold reduc-tion of Na ϩ affinity relative to wild type in the phosphorylation assay.
To examine whether the reduction of apparent Na ϩ affinity at the internal activating sites is caused by a shift of the E 1 -E 2 conformational equilibrium away from the Na ϩ binding E 1 form toward the K ϩ binding E 2 form, we determined the apparent affinity for vanadate, which can be used for probing the E 1 -E 2 distribution because it binds to E 2 and not to E 1 . As evident from Fig. 1C, the apparent affinity for vanadate inhibition was reduced (K 0.5 increased) by around 9-fold in D923N relative to wild type, showing that the RDP mutation actually favors the Na ϩ -binding E 1 state. Altogether, these data indicate a direct effect of mutation D923N on the true Na ϩ affinity of the internal Na ϩ sites of the E 1 form.
Oligomycin allows the phosphoenzyme to build up to a maximum level due to inhibition of the conversion of E 1 P to E 2 P and the promotion of Na ϩ occlusion in E 1 P, cf. Scheme 1 (25)(26)(27)(28)(29). Fig. 1D shows the steady-state level of phosphoenzyme reached in the absence of oligomycin (EP) relative to that reached in its presence (EP max ). For the wild type, the EP/EP max ratio was 84%, whereas it was reduced to 56% in D923N. The lower EP/EP max ratio in the mutant may indicate that Na ϩ interaction is even more defective in the absence of oligomycin than in its presence. Moreover, an increased phosphoenzyme turnover rate in the absence of oligomycin (see further below) may contribute to lower the EP/EP max ratio in the mutant.
D923N Does Not Reduce the Affinity for External K ϩ - Fig. 2A displays the K ϩ dependence of the ATPase activity at Na ϩ concentrations of 40 and 400 mM. K ϩ exerts two different effects; (i) binding from the external side with high affinity to E 2 P, K ϩ triggers rapid dephosphorylation, and (ii) binding from the internal side with low affinity to E 1 in competition with Na ϩ , K ϩ inhibits phosphorylation and, thus, ATP hydrolysis. At 40 mM Na ϩ , both the activation phase and the inhibition phase could be distinguished for the wild type as well as the mutant; however, the inhibition at high K ϩ concentrations was much more pronounced in the mutant, consistent with the marked reduction of the affinity for Na ϩ at the internal sites described above. As the Na ϩ concentration was raised 10-fold to 400 mM, the inhibition phase was reduced markedly, reflecting the increased Na ϩ /K ϩ ratio at the internal sites. At both Na ϩ concentrations, the apparent affinity for K ϩ activation at the external sites, determined from the rising phase of the curves, was slightly higher in the mutant than in the wild type (2-3-fold). Because this could be an indirect consequence of a change in a reaction step preceding or subsequent to the binding step, the K ϩ binding properties of the external sites on E 2 P were assessed more directly by determining the rate of E 2 P dephosphorylation at a non-saturating and a saturating K ϩ concentration (Fig. 2B). The ratio between the dephosphorylation rates at 1 and 20 mM K ϩ did not differ significantly between mutant and wild type (26 versus 31%), showing that their intrinsic affinities for external K ϩ are quite similar.
D923N Affects the Interaction with Extracellular Na ϩ -The decay of the phosphoenzyme was furthermore examined at 20 and 600 mM Na ϩ in the absence of K ϩ after phosphorylation at 150 mM Na ϩ (Fig. 3). At the lower Na ϩ concentration (open triangles), the spontaneous dephosphorylation was rather slow FIGURE 1. Na ؉ interaction. A, Na ϩ dependence of Na ϩ ,K ϩ -ATPase activity is shown. The ATPase activity was measured at 37°C in 30 mM histidine (pH 7.4), 20 mM KCl, 3 mM ATP, 3 mM MgCl 2 , 1 mM EGTA, 10 M ouabain, and NaCl to obtain the indicated concentrations of Na ϩ . Each line represents the best fit of the equation (16)  for both mutant and wild type. The slightly higher rate of the mutant may be explained by a higher amount of E 2 P accumulated during the phosphorylation (see below and Table 1). In the absence of K ϩ , Na ϩ at a high concentration has dual effects on dephosphorylation in the wild type; (i) Na ϩ , bound with low affinity at the external sites of E 2 P, is able to mimic K ϩ , thus activating dephosphorylation to some extent and being transported toward the cytoplasmic side at a stoichiometry of two Na ϩ per ATP hydrolyzed (30,31), and (ii) Na ϩ bound at an external site(s) reverses the E 1 P 3 E 2 P conformational transition, promoting accumulation of E 1 P (32,33). In the wild type the net result of these two opposite effects is that the rate of dephosphorylation is higher at 600 mM Na ϩ compared with 20 mM Na ϩ . However, the acceleration of dephosphorylation by Na ϩ at high concentrations was found considerably more pro-nounced for the D923N mutant than for the wild type (Fig. 3, open circles). To explore which partial reaction step(s) is responsible for this difference, we examined the effect of ADP on the rate of phosphoenzyme decay in the presence of 600 mM FIGURE 2. K ؉ interaction. A, K ϩ dependence of Na ϩ ,K ϩ -ATPase activity at 40 and 400 mM Na ϩ is shown. The ATPase activity was measured at 37°C in 30 mM histidine (pH 7.4), 3 mM ATP, 3 mM MgCl 2 , 1 mM EGTA, 10 M ouabain, with KCl and NaCl added to obtain the indicated concentrations of K ϩ and Na ϩ . Each line represents the best fit of the equation , the latter term being omitted for wild type at 400 mM Na ϩ because no inhibition phase was apparent. The K 0. 5 1 values (K ϩ activation) are 0.56 mM (wild type, closed circles) and 0.17 mM (D923N, open circles) at 40 mM Na ϩ and 3.2 mM (wild type, closed circles) and 1.6 mM (D923N, open circles) at 400 mM Na ϩ . The K 0. 5 2 values (K ϩ inhibition) are 65 mM (wild type) and 8 mM (D923N) at 40 mM Na ϩ and 18 mM (D923N) at 400 mM Na ϩ . Hill coefficients are n1 ϭ 1.5, n2 ϭ 2.1 (wild type) and n1 ϭ 1.8, n2 ϭ 1.5 (D923N) at 40 mM Na ϩ and are n1 ϭ 1.6 (wild type) and n1 ϭ 1.2, n2 ϭ 0.8 (D923N) at 400 mM Na ϩ . The V max and V 0 values in % are V max ϭ 102, V 0 ϭ 4.1 (wild type) and V max ϭ 105, V 0 ϭ 29 (D923N) at 40 mM Na ϩ and are V max ϭ 100, V 0 ϭ 10 (wild type) and V max ϭ 144, V 0 ϭ 46 (D923N) at 400 mM Na ϩ . Turnover rates (min Ϫ1 ) corresponding to the ordinate of 100% are for wild type 8323 (40 mM Na ϩ ) and 4940 (400 mM Na ϩ ) and for D923N 1009 (40 mM Na ϩ ) and 877 (400 mM Na ϩ ). B, K ϩ dependence of E 2 P dephosphorylation is shown. Using the quench-flow module QFM-5 (Bio-Logic Science Instruments, Claix, France) at 25°C (17,21), the decay of phosphoenzyme formed from ATP was followed in 20 mM Tris (pH 7.5), 20 mM NaCl, 3 mM MgCl 2 , 1 mM EGTA, 10 M ouabain, 1 mM ATP, and either 1 mM KCl with 130 mM choline chloride (closed circles) or 20 mM KCl with 110 mM choline chloride (closed squares). Each line represents the best fit of a monoexponential decay function giving rate constants for dephosphorylation of 56 s Ϫ1 (wild type) and 40 s Ϫ1 (D923N) at 1 mM K ϩ and 178 s Ϫ1 (wild type) and 157 s Ϫ1 (D923N) at 20 mM K ϩ . or 600 mM Na ϩ and 1 mM ATP plus 2.5 mM ADP (closed circles). The lines were generated by computation using the SimZyme software with the choices of rate constants and initial concentrations of E 1 P and E 2 P indicated in Table 1.

TABLE 1
Kinetic constants extracted from the data in Fig. 3

by computation
The SimZyme software (23) was used to compute the time courses of phosphoenzyme decay with and without ADP. The table indicates the choices of rate constants and initial concentrations of E 1 P and E 2 P providing the best fit to the experimental data points (shown as lines in Fig. 3) as determined by comparing the computed time courses with the data points (see examples in supplemental Fig. S1). The computation was based on the reaction scheme shown below. A first order rate constant, k 12 , describes the reaction E 1 P 3 E 2 P, whereas the reverse reaction is described by a pseudo first order rate constant, k 21 , which depends on the extracellular Na ϩ concentration, as only E 2 P with bound Na ϩ is able to reverse to E 1 P (Scheme 1). The value of k 21 determined is for each Na ϩ concentration higher for the wild type than for the D923N mutant, consistent with a lower affinity of an external Na ϩ site(s) in the mutant. At 150 mM Na ϩ , the condition of phosphorylation, the weaker binding of extracellular Na ϩ to the mutant also seems to play a role in determining the initial (steady state) concentrations of E 1 P and E 2 P, resulting in a higher amount of Initial E 2 P in the mutant compared with wild type. The value of k 3 depends on extracellular Na ϩ binding to E 2 P in place of K ϩ to activate dephosphorylation and is for each Na ϩ concentration identical in mutant and wild type. This can be explained by assuming that the third Na ϩ site, being critical for k 21 and not for k 3 , is defective in the mutant. Na ϩ . ADP dephosphorylates E 1 P, forming ATP, whereas E 2 P dephosphorylates only by hydrolysis activated by K ϩ or high Na ϩ . It is seen in Fig. 3 (filled circles) that the ADP-induced acceleration of dephosphorylation was less pronounced in the mutant as compared with the wild type. Hence, the difference between the phosphoenzyme decay rates with and without ADP was much smaller for D923N than for the wild type. This observation suggested that 600 mM Na ϩ is less capable of driving the E 1 P 3 E 2 P transition backwards in the mutant. To further substantiate the latter hypothesis, we used the simulation software SimZyme (23) to perform a series of computations of predicted dephosphorylation time courses for various assumptions regarding the rate constants of interconversion of E 1 P and E 2 P and dephosphorylation of E 2 P (see supplemental Fig. S1 for illustration of some of the computations). These computer simulations revealed that all the data in Fig. 3 are compatible with the model described in Table 1 (generating the lines in Fig. 3) in which the only difference between the dephosphorylation characteristics of mutant and wild type is a reduced ability of the mutant to interact with Na ϩ at the external low affinity site(s) responsible for driving the E 1 P 3 E 2 P transition backward. Hence, the dephosphorylation time courses observed in the presence of 600 mM Na ϩ with and without ADP are compatible with a 14-fold reduced rate constant of the backward conversion of E 2 P to E 1 P in the mutant relative to wild type. Furthermore, the analysis indicates that during the initial incubation with ATP at 150 mM Na ϩ to form the phosphoenzyme, more E 2 P accumulates in the D923N mutant as compared with wild type (85% versus 50%, Table 1) as a consequence of a reduced rate constant of the backward conversion of E 2 P to E 1 P in the mutant. The higher amount of E 2 P initially present in the mutant also explains the slight difference between the phosphoenzyme decay rates of mutant and wild type seen when the enzyme phosphorylated at 150 mM Na ϩ is diluted, allowing dephosphorylation to take place at 20 mM Na ϩ (Fig. 3, open  triangles). The reduced ability of Na ϩ to drive the E 1 P 3 E 2 P transition backward in the mutant can be explained by assuming that mutation D923N not only interferes with the binding of Na ϩ from the intracellular side of the membrane but also to some extent with the binding of extracellular Na ϩ . Besides it is worth noting that the model fitting the data does not imply any difference between mutant and wild type with respect to the rate constant of E 2 P dephosphorylation. The latter increases with increasing Na ϩ concentration from 0.07 s Ϫ1 at 20 mM Na ϩ to 0.4 s Ϫ1 at 600 mM Na ϩ , thus reflecting that Na ϩ at a high concentration binds in place of K ϩ and activates E 2 P dephosphorylation to the same extent in mutant and wild type. Hence, it is only the ability of Na ϩ to drive the E 1 P 3 E 2 P transition backward that differs between D923N and wild type.
Na ϩ -ATPase Activity-From the above-described finding of an increased phosphoenzyme turnover of D923N relative to wild type in the absence of K ϩ , caused by reduced backward conversion of E 2 P to E 1 P, it might have been expected that the ATPase activity in the absence of K ϩ (so-called Na ϩ -ATPase activity) was enhanced in the mutant. The Na ϩ -ATPase turnover rate (Na ϩ -ATPase activity per active site) determined at 600 mM Na ϩ , i.e. corresponding to that in Fig. 3, was, however, very similar for wild type and mutant (516 Ϯ 12 min Ϫ1 (n ϭ 13) and 474 Ϯ 13 min Ϫ1 (n ϭ 13), respectively), leading to the conclusion that the phosphoenzyme turnover is not rate-limiting for the Na ϩ -ATPase activity in the mutant.

DISCUSSION
The functional analysis of RDP mutant D923N disclosed a striking ϳ200-fold reduction of the Na ϩ affinity for activation of phosphorylation in the absence of K ϩ and presence of oligomycin (Fig. 1B), the largest effect on Na ϩ affinity reported so far for any Na ϩ ,K ϩ -ATPase mutant. This effect cannot be attributed to a displacement of the conformational equilibrium E 1 -E 2 away from the Na ϩ binding E 1 form because E 1 , in fact, is favored in the mutant (Fig. 1C). From these data we infer that D923N interferes profoundly with the interaction of E 1 with Na ϩ that binds from the intracellular side of the membrane. This interpretation is supported by the observation of a marked inhibition of the ATPase activity by high K ϩ concentrations in D923N ( Fig. 2A), which may be explained by the reduced Na ϩ affinity allowing K ϩ to compete more efficiently with Na ϩ at the internal sites.
Our measurements of the K ϩ activation of ATPase activity showed a slight increase of the apparent K ϩ affinity of D923N relative to the wild type enzyme ( Fig. 2A, rising phase). This result in conjunction with the more direct rapid kinetic studies of the rate of K ϩ -induced E 2 P dephosphorylation (Fig. 2B) led us to conclude that mutation D923N causes a selective defect in Na ϩ handling without disrupting the external K ϩ sites. Furthermore, K ϩ binding at the internal E 1 sites appears intact in the mutant, as evidenced by the strong inhibition of the ATPase activity by K ϩ at high concentrations ( Fig. 2A).
D923N not only disturbs the interaction with intracellular Na ϩ , but also the interaction with extracellular Na ϩ seems to be affected by the mutation, as indicated by the dephosphorylation data in Fig. 3, which could be fitted by a model where the only difference between the dephosphorylation characteristics of mutant and wild type is a reduced rate constant of the Na ϩdependent backward conversion of E 2 P to E 1 P in the mutant ( Table 1). The reduced backward conversion of E 2 P to E 1 P leads to accumulation of more E 2 P and less E 1 P and, therefore, to reduced reactivity toward ADP in the D923N mutant. In the absence of ADP the consequence is an enhanced dephosphorylation by hydrolysis of E 2 P without any increase of the rate constant of the latter reaction. The requirement for a single model to fit all the three data sets in Fig. 3, obtained at 20 and 600 mM Na ϩ with and without ADP, places severe restrictions on the rate constants, thus reducing the latitude (cf. supplemental Fig. S1). Hence, the results generated by the computation can be considered rather accurate. Na ϩ is released at the extracellular side of the membrane in connection with the E 1 P 3 E 2 P conformational transition (Scheme 1), and the binding of extracellular Na ϩ reverses this transition in the wild type (32,33), thus promoting accumulation of E 1 P. Hence, the dephosphorylation data are consistent with the hypothesis that one or more external Na ϩ site(s) is affected. Because the Na ϩ activation of E 2 P dephosphorylation was not disturbed by the mutation, the affected external site is probably not one of those binding K ϩ during the Na ϩ ,K ϩ -ATPase cycle, which is in good accordance with the conclusion drawn above that K ϩ interaction is normal in the mutant.
The results allow some conjectures regarding the rate-limiting step of the enzyme cycle responsible for the conspicuous 90% reduction of the catalytic turnover rate in the mutant, determined in the presence of 130 mM Na ϩ and 20 mM K ϩ . Given that the net rate of the conversion of E 1 P to E 2 P is enhanced in D923N (Fig. 3), the E 2 P dephosphorylation is wild type-like (Fig. 2), and the net rate of the conversion of E 2 to E 1 is enhanced (Fig. 1C), it appears that the reduced turnover rate in D923N must be accounted for by slowing of the Na ϩ -dependent E 1 3 E 1 P reaction. Furthermore, a wild type-like Na ϩ -ATPase activity was determined for the mutant in the absence of K ϩ and presence of 600 mM Na ϩ . Under these conditions phosphoenzyme turnover is ratelimiting in the wild type, and the fact that the Na ϩ -ATPase turnover rate was normal in the mutant, despite an increased phosphoenzyme turnover, indicates that another partial reaction step has become rate-limiting. Again this could be the Na ϩ -dependent E 1 3 E 1 P reaction.
During the transport cycle the bound Na ϩ ions become occluded, i.e. unable to access the medium on either side of the membrane. Na ϩ is occluded in the E 1 P form, cf. Scheme 1 (26), and there is furthermore evidence suggesting that Na ϩ occlusion actually precedes phosphorylation of E 1 and is required for transfer of the ␥-phosphoryl group of ATP to the enzyme (28,29,34). Because oligomycin stabilizes the Na ϩ -occluded E 1 conformers (28,29), the reduced ratio between the phosphorylation levels in the absence and presence of oligomycin (EP/EP max ratio) observed for D923N (Fig. 1D) could reflect a defective Na ϩ occlusion, which is rescued by oligomycin, i.e. the Na ϩoccluding conformational change that follows Na ϩ binding within E 1 conformers is affected by the mutation. This could mean that the rate of phosphorylation, which depends on Na ϩ occlusion, and, thus, the apparent Na ϩ affinity, is even lower in the absence of oligomycin than in its presence. Another likely reason for the reduced EP/EP max ratio in the mutant is the enhanced phosphoenzyme turnover relative to wild type in the absence of oligomycin (Fig. 3), which may also reflect destabilization of the Na ϩ occluded E 1 P form.
It emerges that all the kinetic differences between the D923N mutant and wild type can be ascribed to a defective handling of Na ϩ on both sides of the membrane. Hence, a site that binds and occludes one of the three Na ϩ ions during translocation by opening alternately toward either side of the membrane may be defective. An intriguing question is which of the three Na ϩ sites is affected by D923N. In the crystal structures of the K ϩ /Rb ϩbound Na ϩ ,K ϩ -ATPase in the E 2 form, the two membranous K ϩ binding sites are well defined (6,7). No high resolution structure is yet available for the Na ϩ -bound E 1 form of the enzyme, which could reveal the location of the three Na ϩ sites, but biochemical (35) and mutagenesis experiments (15, 36 -38) in conjunction with homology modeling based on the structure In all panels the ␣-subunit is gray except for the 11 most C-terminal residues, which are colored wheat. The ␤-subunit is violet, and the FXYD protein is blue. Selected residues including the two C-terminal tyrosines (Y1012 and Y1013) and Asp 923 (D923) are shown as sticks, and Asp 923 is highlighted by an orange color. Blue spheres represent bound K ϩ ions. A, shown is an overview of the structure, cytoplasmic side up; MgF in the catalytic site is green. B, shown is a view of the transmembrane region from the cytoplasmic side showing the location of Asp 923 in a putative channel between the transmembrane helices M5, M7, M8, and M10. C, shown is a side view of the same transmembrane region as shown in B, illustrating the channel and the short distance (5.5 Å) of Asp 923 to the side-chain hydroxyl group of the most C-terminal tyrosine. Also highlighted as sticks are residues previously suggested to make up the third Na ϩ site (Y768, T804, Q920, and E951) (6, 39 -41). D, a surface representation of the channel seen from the cytoplasmic side of the membrane shows that the channel is capped by the two C-terminal tyrosines. The 11 most C-terminal residues, the P-domain, and the ␤-subunit are colored wheat, green, and violet, respectively. In the left panel the two tyrosines are shown as sticks, and Asp 923 (orange surface representation) can be spotted below, whereas in the right panel the tyrosines are shown as wheat surface representation, completely covering the entrance to the channel, such that the orange Asp 923 below is hidden. The structure shown is the K ϩ -bound E 2 form with Protein Data Bank code 2ZXE (7). Residues are numbered according to the human ␣3-isoform involved in the RDP disorder.
of the closely related Ca 2ϩ -ATPase in the E 1 form (39) have indicated that the residues making up two of the three Na ϩ sites are almost identical to those binding K ϩ in the E 2 form. The location of the third Na ϩ site has remained elusive (39 -41). The third Na ϩ site is selective toward Na ϩ and does not bind K ϩ (42), and its occupancy triggers a rearrangement in the catalytic site that activates phosphorylation by ATP (43). The large decrease in affinity for Na ϩ activation of phosphorylation observed for D923N together with the normal behavior of the K ϩ sites, both in their externally facing configuration in E 2 P, where they display wild type-like interaction with K ϩ as well as Na ϩ (Figs. 2 and 3, and Table 1), and in the internally facing configuration in E 1 , where K ϩ apparently competes with two of the three Na ϩ ions, led us to conclude that Asp 923 is specifically associated with the third Na ϩ ion. In this connection, it is interesting to note that Asp 923 is highly conserved among the various isoforms of Na ϩ ,K ϩ -ATPase, whereas the sarco(endo)plasmic reticulum Ca 2ϩ -ATPases, which transport only two ions, possess an asparagine at the homologous position like the RDP mutant studied here (see e.g. the alignment in supplemental Fig. S2A).
In the Na ϩ ,K ϩ -ATPase crystal structures, Asp 923 is located in the cytoplasmic half of transmembrane helix M8 inside a channel-like structure, the lining wall of which is made up of residues from the transmembrane helices M5, M7, M8, and M10 ( Fig. 4 and supplemental Fig. S2B). The channel is capped at the cytoplasmic membrane surface by the two C-terminal tyrosines (Fig. 4D). The hydroxyl group of the most C-terminal tyrosine is only 5-6 Å away from the carboxyl group of Asp 923 (Fig. 4C), and one might, therefore, imagine that the C terminus controls the access to the channel from the intracellular side. In this respect it is interesting that we recently found that the C terminus has a hitherto unknown critical role in the binding of Na ϩ , an effect that could be ascribed to the third Na ϩ site (6,8). Hence, it is possible that the channel between M5, M7, M8, and M10 functions as part of a Na ϩ transport pathway for the third Na ϩ ion and that Asp 923 contributes to binding/occlusion of this Na ϩ ion. Furthermore, the two C-terminal tyrosines might be involved in guiding the Na ϩ ion to its binding site, thus, explaining their influence on the apparent Na ϩ affinity (8). Because most of the residues believed to contribute ligands to binding of the other two Na ϩ ions are located in M4, M5, and M6, the corollary is that the third Na ϩ ion takes a path different from the two other Na ϩ ions. In the Na ϩ ,K ϩ -ATPase crystal structures, Asp 923 also seems to have access to the extracellular surface, as Asp 923 can be seen from the extracellular side when the ␤-subunit and surface loops have been removed (Fig. 5).
To examine the hypothesis that Asp 923 is part of a Na ϩ binding site in the E 1 form, we have made a homology model based on the structure of the Ca 2ϩ -ATPase in E 1 (44) with bound Ca 2ϩ ions (Fig. 6). This approach seems justified by the similarity of the overall structures of the Ca 2ϩ -ATPase and the Na ϩ ,K ϩ -ATPase ␣-subunit in E 2 form (6,7,45); see the alignments in supplemental Fig. S2. Importantly, the channel-like structure between M5, M7, M8, and M10 is preserved in the E 1 homology model, in accordance with M7-M10 being rather static during the E 2 -E 1 transition (46). In both conformations the channel is most narrow around Asp 923 , and it widens toward the extracellular side as well as the intracellular side. In fact, Asp 923 can be seen from both sides of the membrane in the E 1 model, when flexible surface loops have been removed ( Fig.  6C and 6D). Hence, it is not too far fetched to imagine that by minor rearrangements Na ϩ could have alternate access to Asp 923 from the two sides of the membrane, i.e. from the cytoplasmic side in E 1 conformation and from the extracellular side in E 2 P conformation. Due to movement of M5, Thr 771 and Ser 772 have approached Asp 923 in the E 1 model structure such that the distances between the side-chain oxygens of Asp 923 , Thr 771 , and Ser 772 are less than 3 Å (Fig. 6B). These side-chain oxygens might, therefore, contribute to a binding site for the third Na ϩ ion in the E 1 conformation. In addition, the sidechain hydroxyl group of Tyr 768 is in the vicinity of Asp 923 (ϳ6 Å) and might also contribute to this binding site, possibly indirectly through a water molecule. The exact geometry of the site would furthermore depend on the modifying influence of the ␤-subunit on the transmembrane helices through its interaction with, particularly, M7 and M10 (6), which cannot be taken appropriately into account in modeling based on the Ca 2ϩ -ATPase structure.
In support of our hypothesis, mutation of Thr 771 , Ser 772 , and Tyr 768 has led to quite significant (10-to 20-fold) reductions of the apparent Na ϩ affinity (38,41,47,48). Importantly the effect was selective for Na ϩ in mutants T771A (38,41,48), Y768L (47), and Y768F (49), consistent with an involvement in the binding of the third Na ϩ ion. For mutant S772A, both Na ϩ and K ϩ affinities have been reported significantly reduced relative to wild type (48,50,51). This is in accordance with our hypothesis, because a comparison of the Na ϩ ,K ϩ -ATPase E 2 structure with the homology model shows that the E 2 -E 1 transition brings Ser 772 in position for binding of the third Na ϩ ion FIGURE 5. Extracellular view of Asp 923 in the Na ؉ ,K ؉ -ATPase E 2 crystal structure. The structure shown is the ␣-subunit of the K ϩ -bound E 2 form with Protein Data Bank code 2ZXE (7), i.e. the same molecule as in Fig. 4, viewed from the extracellular side as surface representation with surface loops and ␤-and ␥-subunits removed. Asp 923 is indicated in orange. M5, M6, M7, and M8 are colored light pink, green, dark gray, and blue, respectively. Other parts of the protein are colored light gray (due to shadow effects, they appear darker in some areas).
together with Asp 923 and Thr 771 , whereas Ser 772 contributes to K ϩ binding in the E 2 form (compare Figs. 4 and 6) (6, 7). Gln 920 might also contribute with a backbone carbonyl oxygen to the binding of the third Na ϩ ion together with Asp 923 (5 Å distance to Asp 923 ; see Fig. 6B), whereas the side chain of Gln 920 likely interacts with one of the two other Na ϩ ions, because its homolog in Ca 2ϩ -ATPase, Glu 908 , contributes to Ca 2ϩ binding (Fig. 6B). Accordingly, Gln 920 is an essential residue in Na ϩ ,K ϩ -ATPase, mutations of Gln 920 being incompatible with cell viability (41,52).
Previous modeling and mutagenesis studies have pinpointed Tyr 768 , Thr 804 , and Glu 951 (39,40) or Thr 771 , Ser 772 , and Gln 920 (41) as putative liganding residues at the third Na ϩ site. With respect to Thr 804 , its side chain, like that of Gln 920 , appears from the homology with the Ca 2ϩ -ATPase to interact with the first Na ϩ ion rather than the third (Fig. 6, A and  B). As seen, some of the previously proposed liganding residues are the same as those of our model, but none of the previous homology models implicated Asp 923 , which is shown here for the first time to have more impact on Na ϩ affinity than any of the other residues and without effect on K ϩ interaction.
A study of the homologous mutation D925N in the ␣2-isoform concluded from Na ϩ ,K ϩ -ATPase activity measurements that the apparent Na ϩ affinity is reduced only 3-5-fold in the mutant relative to wild type ␣2 (36), whereas the present Na ϩ ,K ϩ -ATPase activity data obtained under similar conditions were consistent with a 25-fold difference between mutant and wild type (Fig. 1A). One might, therefore, ask whether the mutated residue plays a less critical role in ␣2 than in ␣3. We do not believe this to be the case; rather, the explanation of the discrepancy should be found in the uncertainty of estimation of the apparent affinity in the previous report (36), where the Na ϩ concentrations studied were much too low (Ͻ 50 mM) to reach saturation in the mutant and too low for revealing the inhibition phase at high Na ϩ (cf. Fig. 1A), which influences the results obtained by curve-fitting. In fact, below 50 mM the Na ϩ dependence of ATPase activity in Ref. 36 follows much the same course as in our Fig. 1A. Moreover, in support of the notion that there is no real difference between the isoforms with respect to mutational effects, in Ref. 36 the K ϩ dependence of the ATPase activity (at 30 mM Na ϩ ) shows the same inhibition phase, starting above 5 mM K ϩ , as seen in our Fig. 2A, left panel, which is the result of a marked reduction of Na ϩ affinity.
In support of the existence of a channel-like conduction pathway in relation to the third Na ϩ site, it has previously been demonstrated that the function of the third Na ϩ site is associated with a passive inward leak current of protons (53). Furthermore truncation of the C terminus, which as mentioned above, is close to Asp 923 , potentially plugging the channel from the cytoplasmic side, was recently found to activate the inwardly Ϫ with Protein Data Bank code 2ZBD (44). To model the putative third Na ϩ site, Ser 766 , Glu 908 , and Asn 911 of Ca 2ϩ -ATPase were replaced with Thr, Gln, and Asp, respectively. Rotamers free from steric clash and with minimum energy were chosen. Residues discussed under "Discussion" are highlighted as sticks numbered according to the human ␣3-isoform. Yellow spheres with numbers represent the bound Ca 2ϩ ions in Ca 2ϩ -ATPase that correspond to Na ϩ ions 1 and 2 in Na ϩ ,K ϩ -ATPase. A, shown is a side view of the transmembrane segments M4-M10 with indication of residues proposed to contribute to the binding of the third Na ϩ ion (D923, Y768, T771, S772, and Q920) as well as Thr 804 (T804), cytoplasmic side up. B, shown is a close-up view of the same transmembrane region illustrating the short distances of Asp 923 to Tyr 768 , Thr 771 , Ser 772 , and the backbone carbonyl of Gln 920 . C, surface representation of the model structure is seen from the intracellular side of the membrane (surface loops removed). No attempt was made to model the C terminus, which is not present in the Ca 2ϩ -ATPase. D, surface representation of the model structure is seen from the extracellular side of the membrane (surface loops removed). In C and D, Asp 923 is indicated by orange surface representation. Note that Asp 923 can be seen from both sides of the membrane, although the direct access to Asp 923 from the cytoplasmic side is partially hindered because M5 has moved relative to the structure in Fig. 4, possibly reflecting Na ϩ occlusion. M5, M6, M7, M8, and M10 are colored light pink, green, dark gray, blue, and dark blue, respectively. Other parts of the protein are colored light gray (due to shadow effects, they appear darker in some areas).
directed H ϩ current and possibly also allows an inwardly directed Na ϩ current (54,55). These experiments were carried out with the ␣1and ␣2-isoforms, thus providing further support for the existence of this channel in all Na ϩ ,K ϩ -ATPase isoforms (see also Note Added in Proof).
Electrophysiological measurements on HeLa cells expressing the ␣2 mutant D925L have demonstrated that the electrogenicity of the pump is retained with this mutation; hence, the 3Na ϩ / 2K ϩ stoichiometry is probably also intact (56). It is noteworthy in this connection that the D923N mutant supported cell viability in our experiments, thus being able to transport Na ϩ and K ϩ and probably to generate a membrane potential. Taken together these findings lead to a picture in which the mutant pump carries out transport with a normal stoichiometry but with altered kinetics due to reduced Na ϩ affinity.
The closely related H ϩ ,K ϩ -ATPase is electroneutral, meaning that the transport stoichiometry is either 2H ϩ /2K ϩ or 1H ϩ /1K ϩ , i.e. the third site is not present or is inactive. One may, therefore, ask how our proposal fits with the fact that in all H ϩ ,K ϩ -ATPases an aspartate is present at the position corresponding to Asp 923 of the Na ϩ ,K ϩ -ATPase. The answer seems to be provided by the fact that in the H ϩ ,K ϩ -ATPase a lysine replaces the serine in transmembrane segment M5 (Ser 772 ) (see the alignment in supplemental Fig. S2A), which in our model contributes to the third Na ϩ site. Hence, our hypothesis would predict that in H ϩ ,K ϩ -ATPase the third site is occupied by the positively charged side-chain amino group of the lysine, probably bonding to the aspartate. These ideas are supported by mutagenesis studies showing that the electrogenicity of Na ϩ ,K ϩ -ATPase is lost upon replacement of Ser 772 with an arginine (lysine disrupts transport completely) and, conversely, that the H ϩ ,K ϩ -ATPase gains electrogenicity upon introducing a serine in place of the lysine (57).
Nine different RDP mutations have thus far been identified in the Na ϩ ,K ϩ -ATPase ␣3 gene (1, 2, 9 -12), all of which, except T613M in the P-domain, are located in the cation binding transmembrane domain or in closely associated regions. Four of these mutants, including T613M, have previously been characterized functionally in detail, all exhibiting a reduced affinity for intracellular Na ϩ (10,21,22), as much as 41-fold for a mutant (Y1013dup) in which an extra tyrosine is added to the C terminus (10), probably interfering with the channel inlet. For F780L (22), located in transmembrane segment M5, there is evidence similar to the present for D923N, indicating that the third Na ϩ site is directly or indirectly affected with no effect on K ϩ interaction. The present results in conjunction with these previous studies of other RDP mutations suggest that a selective defect in the handling of Na ϩ may be a general feature of the RDP disorder.