Glutamate 90 at the Luminal Ion Gate of Sarcoplasmic Reticulum Ca2+-ATPase Is Critical for Ca2+ Binding on Both Sides of the Membrane*

The roles of Ser72, Glu90, and Lys297 at the luminal ends of transmembrane helices M1, M2, and M4 of sarcoplasmic reticulum Ca2+-ATPase were examined by transient and steady-state kinetic analysis of mutants. The dependence on the luminal Ca2+ concentration of phosphorylation by Pi (“Ca2+ gradient-dependent E2P formation”) showed a reduction of the apparent affinity for luminal Ca2+ in mutants with alanine or leucine replacement of Glu90, whereas arginine replacement of Glu90 or Ser72 allowed E2P formation from Pi even at luminal Ca2+ concentrations much too small to support phosphorylation in wild type. The latter mutants further displayed a blocked dephosphorylation of E2P and an increased rate of conversion of the ADP-sensitive E1P phosphoenzyme intermediate to ADP-insensitive E2P as well as insensitivity of the E2·BeF3− complex to luminal Ca2+. Altogether, these findings, supported by structural modeling, indicate that the E2P intermediate is stabilized in the mutants with arginine replacement of Glu90 or Ser72, because the positive charge of the arginine side chain mimics Ca2+ occupying a luminally exposed low affinity Ca2+ site of E2P, thus identifying an essential locus (a “leaving site”) on the luminal Ca2+ exit pathway. Mutants with alanine or leucine replacement of Glu90 further displayed a marked slowing of the Ca2+ binding transition as well as slowing of the dissociation of Ca2+ from Ca2E1 back toward the cytoplasm, thus demonstrating that Glu90 is also critical for the function of the cytoplasmically exposed Ca2+ sites on the opposite side of the membrane relative to where Glu90 is located.

The Ca 2ϩ -ATPase 2 of sarco(endo)plasmic reticulum (1)(2)(3) is an ion-translocating ATPase of P-type that mediates active transport of Ca 2ϩ from the cytoplasm to the endoplasmic reticulum lumen, thereby allowing rapid oscillation of Ca 2ϩ during cellular activation events. Ca 2ϩ pumping is ac-hieved by means of a reaction cycle (Scheme 1) involving the formation and decomposition of an aspartyl-phosphorylated intermediate coupled to protein conformational changes that facilitate the binding of two Ca 2ϩ ions at cytoplasmically facing high affinity binding sites and subsequent dissociation from luminally facing low affinity sites. Because of extensive efforts that in recent years have involved x-ray crystallography (4 -7), an increasingly detailed picture of the structural changes relating to Ca 2ϩ transport by the Ca 2ϩ pump is steadily emerging. The Ca 2ϩ -ATPase consists of 10 membrane-spanning helices (M1 through M10) connecting three major cytoplasmic domains named A (actuator), P (phosphorylation), and N (nucleotide binding) and some smaller luminal loops. Transmembrane helices M4-M6 and M8 contain the residues that coordinate the two Ca 2ϩ ions bound side-by-side in a binding pocket in the Ca 2 E1 and Ca 2 E1P states. A key element in the transport cycle is the action of the A domain, which is directly linked to M1-M3 via flexible linkers (see Fig. 1). Thus, occlusion of the Ca 2ϩ ions at the high affinity sites of Ca 2 E1 is achieved through an ATPinduced ϳ30°tilting of the A domain and accompanying partial unfolding and bending of the N-terminal part of M1 (3). After transfer of the ␥-phosphate of ATP to Asp 351 , a further ϳ90°r otation of the A domain during the Ca 2 E1P 3 Ca 2 E2P transition propagates to the transmembrane domain, disrupting the high affinity Ca 2ϩ sites and exposing the Ca 2ϩ ions to the lumen. The subsequent release of the Ca 2ϩ ions to the lumen is succeeded by dephosphorylation of the aspartyl phosphate in E2P, catalyzed by the 181 TGES phosphatase motif of the A domain (8) that has been brought into position in the catalytic site in E2P (5,6). The cycle is completed by reversal of the A domain rotation and restoration of the cytoplasmically facing high affinity Ca 2ϩ sites (E2 3 E1 in Scheme 1).
The molecular nature and properties of the Ca 2ϩ binding pocket in the membrane-spanning region of the Ca 2 E1 and Ca 2 E1P states are known in great detail. Little, however, is known about the exact entry and exit pathways to and from the Ca 2ϩ binding pocket and the structural elements involved in extrusion of the Ca 2ϩ ions to the endoplasmic reticulum lumen. The crystal structure of Ca 2ϩ -ATPase in the E2⅐BeF 3 Ϫ state (presumed E2P ground state analog) (7) provides some important clues. Thus, in the E2⅐BeF 3 Ϫ structure, part of the transmembrane domain near the lumen is in a considerably more open conformation compared with other crystal structures of the Ca 2ϩ -ATPase. The M4-residue Glu 309 , an essential residue for Ca 2ϩ coordination in the occluded binding pocket of the Ca 2 E1 and Ca 2 E1P states (4,7,9), is exposed to the lumen and associated with a Mg 2ϩ ion in the E2⅐BeF 3 Ϫ structure (Fig. 1). The E2⅐BeF 3 Ϫ crystals were formed in the presence of a high concentration of Mg 2ϩ (50 mM) and absence of Ca 2ϩ , and it may be speculated that the Mg 2ϩ ion is bound in place of Ca 2ϩ at a luminally facing low affinity Ca 2ϩ site. Ser 72 and Glu 90 at the luminal ends of transmembrane helices M1 and M2, respectively, are positioned very close to Glu 309 in E2⅐BeF 3 Ϫ , and the Glu 90 side-chain carboxylate seems to contribute to coordination of the luminal Mg 2ϩ ion. In the Ca 2 E1 and Ca 2 E1P states, on the other hand, Ser 72 and Glu 90 are 10 -20 Å apart from Glu 309 and the Ca 2ϩ sites due to translational movements of M1/M2 and M3/M4 relative to each other, and Ser 72 and Glu 90 instead interact with Lys 297 at the luminal end of M4 ( Fig. 1 and supplemental Table S1.
In the present study we investigated the significance of the above-described luminal interaction networks seen in the various crystal structures of the Ca 2ϩ -ATPase. Thus, mutants with alterations to Ser 72 , Glu 90 , and Lys 297 were compared with wild type Ca 2ϩ -ATPase by analyzing the partial reactions in transient and steady-state kinetic measurements.
We found that the mutations to Glu 90 affect the Ca 2ϩ binding properties profoundly both on the luminal side of the membrane, where Glu 90 is situated, as well as on the cytoplasmic side. Furthermore, single-substitution of either Ser 72 or Glu 90 with arginine gives rise to Ca 2ϩ pumps with a remarkably stable E2P state, compatible with the hypothesis that the positive charge of the arginine side chain interacts directly with the luminal Ca 2ϩ outlet, possibly mimicking Ca 2ϩ and thereby blocking further processing of the phosphoenzyme.

EXPERIMENTAL PROCEDURES
Site-directed mutagenesis of cDNA encoding the rabbit fast twitch muscle Ca 2ϩ -ATPase (SERCA1a isoform) inserted into the pMT2 vector (10) was carried out using the QuikChange site-directed mutagenesis kit (Stratagene), and the mutant cDNA was sequenced throughout. To express wild type or mutant cDNA, COS-1 cells were transfected using the calcium phosphate precipitation method (11). Microsomal vesicles containing either expressed wild type or mutant Ca 2ϩ -ATPase were isolated by differential centrifugation (12). The concentration of expressed Ca 2ϩ -ATPase was determined by an enzymelinked immunosorbent assay (13) and by determination of the maximum capacity for phosphorylation with ATP or P i ("active site concentration"; see Ref. 14). Transport of The respective Protein Data Bank accession codes corresponding to the structures shown are 1SU4 (4) and 3B9B (7). Amino acid side chains are shown for residues discussed in the text, and relevant distances are listed in supplemental Table S1. Carbon atoms are shown in gray, nitrogen in blue, oxygen in red, and Mg 2ϩ and Ca 2ϩ in green. The two arrows indicate the rotation of the actuator domain and the rearrangement of M1/M2 relative to M3/M4 during the conformational transition. The phosphorylation, nucleotide binding, and actuator domains are indicated by P, N, and A, respectively. SCHEME 1. Ca 2؉ -ATPase reaction cycle. Major conformational changes and substrate binding and dissociation steps are shown. 45 Ca 2ϩ into the microsomal vesicles and 45 Ca 2ϩ binding at 25°C were measured by filtration, and the ATPase activity was determined by following the liberation of P i (15) in the presence of 4 M calcium ionophore A23187 to prevent inhibition caused by rebinding of Ca 2ϩ to the luminally facing Ca 2ϩ sites (14). Measurements of phosphorylation and dephosphorylation were generally carried out by manual mixing at 0°C (14,16). Transient state kinetics at 25°C was analyzed using the Bio-Logic quench-flow module QFM-5 (Bio-Logic Science Instruments, Claix, France) with mixing protocols as previously described (17,18). The determination of the phosphorylation level by acid quenching followed by acid SDS-polyacrylamide gel electrophoresis and quantification of the radioactivity associated with the Ca 2ϩ -ATPase band was carried out using the previously established procedures (14,16).
The experiments were conducted at least twice, and average values are shown in Figs. 2-9. Generally, the complete data set (i.e. including all experimental points before averaging) was analyzed by nonlinear regression using the SigmaPlot program (SPSS, Inc.) with the equations described in the figure legends, giving the lines in the figures and the S.E. shown in the tables. To analyze the phosphorylation time courses in Fig. 2, the kinetic simulation software SimZyme was applied, allowing computation of the phosphorylation overshoot as detailed below and in previous publications (17,18). The best fit was in this case determined manually by trial and error, comparing the computed time courses with the experimental data points for various choices of rate constants. For any choice of reaction cycle 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 the reaction intermediates (17).

RESULTS
Expression and Assays of Overall Function-To study the potential roles of Glu 90 and Ser 72 for Ca 2ϩ interaction at the luminal ion gate in the E2P state as well as to address the significance of the Glu 90 -Lys 297 -Ser 72 interaction network seen in the Ca 2 E1 structures, we produced six point mutants with alterations to Ser 72 , Glu 90 , and Lys 297 of Ca 2ϩ -ATPase. Glu 90 was substituted with alanine, leucine, and arginine, Ser 72 was substituted with alanine and arginine, and Lys 297 was substituted with alanine. The five mutants with alterations to Ser 72 or Glu 90 could be expressed in COS-1 cells to levels comparable with that obtained for the wild type, whereas the expression level in COS-1 cells of mutant K297A was generally only ϳ30% that obtained with wild type (supplemental Table S2). A reduced expression level of mutants with alterations to Lys 297 has previously been reported by Chen et al. (19). Thus, K297G was not expressed (despite wild type-like levels of mRNA transcript in the cells), and K297F yielded low expression levels, whereas K297M, K297R, and K297E were expressed at wild type-like levels (19). Fortunately, the expression level of K297A was sufficiently high for us to carry out reliable functional measurements.
To assess the overall function of the mutants, the rate of ATP hydrolysis was measured at 37°C in the presence of 5 mM MgATP and 3 M free Ca 2ϩ . The catalytic turnover rates (ATP hydrolysis activity per enzyme molecule (14)) of S72A and K297A were wild type-like (80 and 87%, respectively, that obtained with wild type), whereas the remaining mutants displayed little or no ATPase activity (19% of the wild  type rate for E90A and Ͻ10% for S72R, E90L, and E90R, supplemental Table S2). The results of measuring the rate of 45 Ca 2ϩ transport into the microsomal vesicles at 37°C with 5 mM MgATP present were similar to those obtained in the ATPase activity assay, i.e. wild type-like Ca 2ϩ transport rates for S72A and K297A and much less for the remaining mutants (supplemental Table S2).
Transient State Kinetics of Phosphorylation of Ca 2 E1 by ATP-To determine the rate of phosphorylation of the Ca 2ϩbound enzyme from ATP and to obtain an initial overview of the effects of the mutations on the succeeding partial reactions of the pump cycle, we used rapid kinetic instrumentation at 25°C to examine the transient state kinetics of phosphorylation at 5 M [␥-32 P]MgATP at pH 7 of enzyme pre-equilibrated with Ca 2ϩ (Fig. 2). Under these conditions, the wild type and some of the mutants displayed overshoots of phosphorylation that could be reproduced by computation based on the simplified three-intermediate reaction cycle shown in the bottom right corner of Fig. 2 (for a detailed description of this approach as used previously in case of wild type and other mutants, see Refs. 17 and 18). We find that the best fit to the wild type data is obtained for the rate constants k A ϳ 50 s Ϫ1 for phosphorylation of Ca 2 E1, k B ϭ 6 s Ϫ1 for phosphoenzyme processing (i.e. Ca 2 E1P 3 Ca 2 E2P 3 E2P 3 E2), and k C ϭ 20 s Ϫ1 for the Ca 2ϩ binding transition of the dephosphoenzyme (i.e. E2 3 Ca 2 E1). As seen in Fig. 2, for all six mutants a good fit to the data could be obtained with a phosphorylation rate constant (k A ) similar to that of wild type. Mutants E90A and E90L displayed large overshoots of phosphorylation reflecting accumulation of a considerable amount of dephosphoenzyme at steady state. In such cases where the phosphorylation overshoot is large, fairly accurate values of the rate constants k B and k C can be extracted from the computational analysis. Thus, the increased levels of dephosphoenzyme accumulating at steady state with E90A and E90L were found to result from 2-fold increased rates of phosphoenzyme processing (k B ) combined with 10-and 20-fold reduced rates of the Ca 2ϩ binding transition of the dephosphoenzyme (k C ), respectively (Fig. 2). Direct measurements of the rates of the partial reactions involved in phosphoenzyme and dephosphoenzyme processing are presented below.
The Ca 2 E1P 3 Ca 2 E2P Conformational Transition-After the phosphorylation of Ca 2 E1 by ATP, the subsequent turnover of the phosphoenzyme occurs in at least three distinct steps comprising the Ca 2 E1P 3 Ca 2 E2P conformational transition, Ca 2ϩ dissociation from the Ca 2ϩ sites (now opening toward the luminal side and exhibiting low affinity), i.e. Ca 2 E2P 3 E2P, and the dephosphorylation of the Ca 2ϩ -free E2P intermediate (E2P 3 E2).
The time course of forward processing of the phosphoenzyme formed by phosphorylation of Ca 2ϩ -saturated enzyme with [␥-32 P]ATP was determined at 0°C by chasing the accumulated phosphoenzyme with an excess of the Ca 2ϩ chelator EGTA (terminating phosphorylation by removal of Ca 2ϩ from Ca 2 E1) followed by acid quenching at serial times (Fig. 3A, open symbols). Mutants S72A and K297A displayed phosphoenzyme turnover rates differing less than 2-fold from that of the wild type, whereas the phosphoenzyme turnover in mutants S72R and E90R was extremely slow (no dephosphorylation detected within the time frame of the experiment). Similar direct measurements of phosphoenzyme turnover could not be performed for E90A and E90L because of their low steady-state phosphoenzyme levels (cf. Fig. 2), but as noted above, reliable rates of phosphoenzyme processing could be extracted by computational analysis of the time course of phosphorylation from ATP measured by rapid quench instrumentation at 25°C (Fig. 2), showing that the phosphoenzyme processing was ϳ2-fold accelerated for E90A and E90L relative to wild type, in sharp contrast to the block of phosphoenzyme turnover observed for S72R and E90R.
The phosphoenzyme intermediates E1P and E2P can be distinguished experimentally by their differential sensitivities to Dephosphorylation was then studied at 0°C by the addition of excess EGTA (to remove Ca 2ϩ and, thus, terminate phosphorylation) followed by acid quenching at the indicated times (open symbols). The lines show the best fits of a monoexponential decay function EP ϭ EP max ⅐e Ϫkt (extracted rate constants are listed in Table 1 for the wild type, S72A, and K297A). The closed symbols represent experiments in which 1 mM ADP was added together with the EGTA chase medium. The symbol code is the same as for the corresponding open symbols. B, shown is the rate of loss of ADP sensitivity (i.e. Ca 2 E1P 3 Ca 2 E2P) for S72R and E90R. The enzyme was phosphorylated for the indicated time intervals at 0°C in 40 mM MOPS/Tris (pH 7.0), 80 mM KCl, 5 mM MgCl 2 , 50 M CaCl 2 , 2 M calcium ionophore A23187, and 5 M [␥-32 P]ATP followed by the addition of an equal volume of 40 mM MOPS/Tris (pH 7.0), 80 mM KCl, 5 mM MgCl 2 , 10 mM EGTA, and 2 mM ADP and acid quenching 4 s later. The lines show the best fits of a monoexponential rise to maximum function EP ϭ EP max ⅐(1 Ϫ e Ϫkt ), giving the rate constants listed in Table 1 for S72R and E90R. In each case the 100% value corresponds to the steady-state level of phosphoenzyme present 10 s after the initiation of phosphorylation without the addition of EGTA and ADP.  (17), giving the rate constants for Ca 2 E 1 3 Ca 2 E 1 P (k A ), Ca 2 E 1 P 3 E 2 (k B ), and E 2 3 Ca 2 E 1 (k C ) indicated in each panel as (k A , k B , k C ). The maximum phosphorylation level obtained in each case was taken as 100%.
ADP, E1P being able to transfer the phosphoryl group back to ADP forming ATP and E2P being insensitive to added ADP and dephosphorylating only in the forward direction by hydrolysis of the phosphoryl bond. The ADP sensitivity of the phosphoenzyme accumulated at steady state was determined by a 5-s chase with excess EGTA and 1 mM ADP (Fig. 3A, closed symbols). For the wild type, the ADP chase resulted in a complete disappearance of the phosphoenzyme, indicating that all the accumulated phosphoenzyme belonged to the ADP-sensitive E1P type, as expected because the Ca 2 E1P 3 Ca 2 E2P transition is known to be the rate-limiting step in the phosphoenzyme turnover of the wild type. The phosphoenzyme accumulated for mutant S72A was also exclusively E1P, and for mutant K297A ϳ90% of the phosphoenzyme was E1P (ϳ10% phosphoenzyme left after ADP chase, Fig. 3A, closed diamond). Thus, for the wild type, S72A, and K297A, the rate constant extracted from fits of a monoexponential function to the data shown in Fig. 3A, open symbols, reflects the rate of the Ca 2 E1P 3 Ca 2 E2P transition. In contrast, the phosphoenzyme accumulated for S72R and E90R was almost exclusively ADP-insensitive ( Fig. 3A, closed triangles), showing that the block of phosphoenzyme turnover in these two mutants results from a block of a step in the reaction sequence subsequent to the Ca 2 E1P 3 Ca 2 E2P conformational transition, i.e. a step in the sequence Ca 2 E2P 3 E2P 3 E2. For these two mutants it was feasible to determine the rate of Ca 2 E1P 3 Ca 2 E2P by following the formation of ADP-insensitive phosphoenzyme during incubation with [␥-32 P]ATP ( Fig.  3B), assuming that Ca 2 E1 3 Ca 2 E1P is as rapid as in the wild type (documented in Fig. 2, k A ) and, therefore, not rate-limiting for the reaction sequence Ca 2 E1 3 Ca 2 E1P 3 Ca 2 E2P. Hence, the data in Fig. 3B were obtained by incubating Ca 2ϩ -saturated enzyme with [␥-32 P]ATP for varying time intervals under conditions similar to those used in Fig. 3A followed by determination of the amount of ADP-insensitive phosphoenzyme accumulated (i.e. the sum of Ca 2 E2P and E2P) by adding 1 mM ADP for 4 s to remove any Ca 2 E1P present before acid quenching. The rate of the Ca 2 E1P 3 Ca 2 E2P transition obtained for S72R and E90R in this way was close to 10-fold faster than the rate of Ca 2 E1P 3 Ca 2 E2P obtained with the wild type as illustrated in Fig. 3A under the same buffer and temperature conditions. A summary of the relative Ca 2 E1P 3 Ca 2 E2P rate constants derived from the data of Figs. 2 and 3 is presented in Table 1.
The Stability of E2P and Ca 2 E2P-All six mutants were able to form E2P phosphoenzyme by phosphorylation of Ca 2ϩ -deprived enzyme from P i in the backward direction of normal turnover and displayed apparent affinities for P i at least as high as that of the wild type (supplemental Fig. S1). To measure the rate of E2P dephosphorylation (i.e. the forward reaction E2P 3 E2), E2P was formed in the presence of a saturating concentration of 32 P i and subsequently chased by dilution into a buffer containing excess nonradioactive P i followed by acid quenching at serial times (Fig. 4). Mutants S72A and K297A displayed dephosphorylation rates 1.4-and 3.4-fold lower than that of the wild type, respectively, whereas the dephosphorylation rates of mutants E90A and E90L were enhanced (2.8-and 1.8-fold, respectively) relative to wild type. Importantly, the dephosphorylation of E2P was completely blocked in mutants S72R and E90R, with no phosphoenzyme decay detected within the time frame of the experiment ( Fig. 4 and Table 1).
Such a block of dephosphorylation of E2P has previously been noted for mutations of some of the residues involved in Ca 2ϩ binding from the cytoplasmic side in the E1 conformation, which in E2P are candidates for luminal interaction sites for Ca 2ϩ or protons to be countertransported such as Glu 309 (9,20,21). For direct comparison with S72R and E90R, we also conducted a dephosphorylation experiment with mutant E309Q under the presently applied conditions (Fig. 4). In accordance with the previous result (20,21), the dephosphorylation of E2P was also markedly slowed in E309Q, although not quite as much as seen for S72R and E90R ( Fig. 4 and Table 1).
For S72R and E90R, the results summarized in Table 1 demonstrate that the first and the last reaction step in the sequence Ca 2 E1P 3 Ca 2 E2P 3 E2P 3 E2 is enhanced and inhibited, respectively. It is, however, unclear from these data how the intermediate reaction step Ca 2 E2P 3 E2P, i.e. the dissociation of Ca 2ϩ from the luminally exposed Ca 2ϩ sites, is affected by the mutations. To answer the question of whether the inserted arginine residue inhibits the luminal Ca 2ϩ dissociation such that Ca 2 E2P 3 E2P contributes to rate limitation of phosphoenzyme turnover, we examined whether Ca 2ϩ remained tightly bound ("occluded") in the accumulated phosphoenzyme (Fig. 5). A mutant previously studied by Daiho et al. (22), with four glycines inserted in the A/M1 linker between Gly 46 and Lys 47 (4Gi-46/47), was included as a positive control in which the luminal Ca 2ϩ dissociation step Ca 2 E2P 3 E2P is blocked. For reference we first demonstrated the ability of the mutants The values extracted by regression analysis or computation are shown relative to that of wild type. The S.E. is indicated, and the value n refers to the number of experimental data points on which the analysis is based.
a Rate constant of dephosphorylation of phosphoenzyme formed by phosphorylation with P i (experimental details are in the legend to Fig. 4)  to occlude Ca 2ϩ in the stable Ca 2 E1P-like Ca 2 E1⅐AlF 3 Ϫ ⅐ADP complex. The enzyme was incubated with 45 Ca 2ϩ in the presence of AlF 3 Ϫ and ADP followed by a chase with excess EGTA, removing free Ca 2ϩ from the medium (on both sides of the membrane due to the presence of calcium ionophore). At varying time intervals after the chase, the samples were rapidly filtered and washed to determine the amount of 45 Ca 2ϩ associated with the Ca 2ϩ -ATPase. As seen in Fig. 5, open symbols, S72R, E90R, and 4Gi-46/47 were all able to form a stable Ca 2ϩoccluded Ca 2 E1⅐AlF 3 Ϫ ⅐ADP complex similar to the wild type (23,24). Then in another set of experiments without AlF 3 Ϫ and ADP, the enzyme was phosphorylated by ATP in the presence of 45 Ca 2ϩ followed by the same EGTA chase and filtration procedure as described above (closed symbols in Fig. 5). In parallel experiments using [␥-32 P]ATP, the phosphoenzyme decay was followed under comparable conditions at 25°C (supplemental Fig. S2). In the wild type, the phosphoenzyme decayed too rapidly for any Ca 2ϩ occlusion in the phosphoenzyme to be detected by the manual filtration technique used here. However, in mutants 4Gi-46/47 and E90R, the phosphoenzyme was very stable even at 25°C (supplemental Fig. S2), and reliable measurements of Ca 2ϩ occlusion in the phosphoenzyme could easily be performed. For 4Gi-46/47, a high level of 45 Ca 2ϩ occlusion similar to that obtained in the presence of ADP⅐AlF 3 Ϫ was seen after the phosphorylation by ATP, in accordance with the previously described block of luminal Ca 2ϩ dissociation from the Ca 2 E2P state in this mutant (22). Notably, the E2P phosphoenzyme of E90R was not in the 45 Ca 2ϩ -occluded state, unlike that of 4Gi-46/47 (Fig. 5), thus indicating that in E90R the luminal Ca 2ϩ dissociation does not contribute to the rate limitation in the Ca 2 E2P 3 E2P 3 E2 reaction sequence. Likewise, no 45 Ca 2ϩ occlusion could be detected in S72R during the phosphorylation experiment, but in the latter case the conclusion is uncertain, because much of the phosphoenzyme may have disappeared during the measurement due to the higher dephosphorylation rate of S72R (compare Fig.  5 with supplemental Fig. S2).
Function of the Luminally Exposed Low Affinity Ca 2ϩ Sites-To examine the accessibility and affinity of the luminally exposed Ca 2ϩ sites in E2P, we applied a recently devised assay (25) that takes advantage of the well known Ca 2ϩ gradient-dependent formation of E2P from P i (26,27). Microsomal vesicles containing wild type or mutant Ca 2ϩ -ATPase were loaded passively with various Ca 2ϩ concentrations by overnight incubation and then diluted into a medium containing 32 P i and EGTA. The latter chelator was present to remove Ca 2ϩ in the medium outside the vesicles, thereby preventing Ca 2ϩ binding from the FIGURE 4. Rate of E2P dephosphorylation. Dephosphorylation of E2P accumulated by phosphorylation with P i is shown. The enzyme contained in microsomal vesicles was phosphorylated with 0.5 mM 32 P i for 10 min at 25°C in 100 mM MES/Tris (pH 6.0), 10 mM MgCl 2 , 2 mM EGTA, and 30% (v/v) dimethyl sulfoxide (the organic solvent ensuring maximal amount of phosphoenzyme). The phosphorylated enzyme was then chilled on ice, and dephosphorylation was studied at 0°C by a 19-fold dilution into ice-cold medium containing 40 mM MOPS/Tris (pH 7.0), 10 mM KCl, 2 mM MgCl 2 , 2 mM EGTA, and 0.5 mM nonradioactive P i followed by acid quenching at the indicated time intervals. The lines show the best fits of a monoexponential decay function EP ϭ EP max ⅐e Ϫkt , giving the rate constants listed in Table 1. outside (corresponding to the cytoplasmic side), which would inhibit phosphorylation by P i by shifting the E2-E1 equilibrium away from E2 (cf. Scheme 1). On the contrary, a high luminal Ca 2ϩ concentration facilitates the reaction of the wild type Ca 2ϩ -ATPase with P i , even under conditions (neutral pH, presence of K ϩ ) that in the absence of a Ca 2ϩ gradient favor dephosphorylation (27). A K 0.5 value of 7.5 mM for Ca 2ϩ activation from the luminal side was obtained for the wild type from the dependence of the phosphorylation by P i on the luminal Ca 2ϩ concentration (Fig. 6). For mutants S72A and K297A the apparent Ca 2ϩ affinity on the luminal side determined in this way was wild type-like, whereas E90A and E90L displayed markedly reduced apparent Ca 2ϩ affinities (K 0.5 increased ϳ3and Ն5-fold, respectively). Mutant 4Gi-46/47 displayed a 9-fold increased apparent Ca 2ϩ affinity (K 0.5 reduced) relative to wild type ( Fig. 6 and Table 2), which likely is associated with the high stability of the Ca 2 E2P state in this mutant (22). Remarkably, S72R and E90R were maximally phosphorylated even at 0.01 mM luminal Ca 2ϩ , where almost no phosphoenzyme is obtained with wild type, and remained maximally phosphorylated throughout the range of luminal Ca 2ϩ concentrations applied in the experiment (Fig. 6). As further seen in Fig. 6, E309Q likewise showed a rather constant phosphorylation level independent of the luminal Ca 2ϩ concentration. The phosphorylation level of E309Q was lower than that of S72R and E90R, consistent with the slightly higher rate of dephosphorylation of E2P in E309Q as compared with S72R and E90R (Fig. 4).
The phosphoenzyme that accumulates for the wild type upon phosphorylation of Ca 2ϩ -loaded vesicles with P i is ADP-sensitive Ca 2 E1P, due to the equilibrium E2P ϩ 2Ca lum 2ϩ 7 Ca 2 E2P 7 Ca 2 E1P (25). In contrast, as demonstrated by the experiment shown in the bottom panel of Fig. 6, the phosphoenzyme accumulated for mutants S72R, E90R, and 4Gi-46/47 remained ADP-insensitive even when the vesicles had been loaded with high amounts of Ca 2ϩ , implying either a block of Ca 2ϩ binding at the luminal Ca 2ϩ sites and/or of the Ca 2 E2P 3 Ca 2 E1P transition.
The Stability and Ca 2ϩ Sensitivity of E2P-like Analog States-We furthermore studied the properties of E2 in complex with the two inhibitory phosphoryl analogs BeF 3 Ϫ and vanadate. E2⅐BeF 3 Ϫ is believed to mimic the E2P ground state (7, 28), whereas vanadate is generally thought to capture the ATPase in a state similar to the transition state of E2P dephosphorylation (29). Both inhibitors bind to the Ca 2ϩ -deprived E2 state in a slow reaction that requires Mg 2ϩ . Panels A and B of Fig. 7 show the BeF 3 Ϫ and vanadate dependences, respectively, of the inhibition of the wild type and mutants S72R, E90R, and 4Gi-46/47 by reaction with the phosphoryl analog. S72R and 4Gi-46/47 deviated only marginally from the wild type with respect to the apparent affinities for BeF 3 Ϫ and vanadate. In contrast, E90R displayed markedly reduced apparent affinities for both BeF 3 Ϫ and vanadate (13-and 6-fold, respectively) relative to the wild type.
The E2⅐BeF 3 Ϫ complex is destabilized by luminal Ca 2ϩ binding at the transport sites in the luminally exposed configuration (28), unlike the complexes of the E2 state of Ca 2ϩ -ATPase with AlF 3 Ϫ and MgF 4 2Ϫ in which the transport sites are occluded inside the transmembrane domain and inaccessible to Ca 2ϩ present on either side of the membrane (5,6). E2⅐vanadate is also sensitive to Ca 2ϩ , although it seems that the dissociation of vanadate results from Ca 2ϩ binding at the cytoplasmically exposed Ca 2ϩ sites (30). Panels C and D of Fig. 7 show the rates of dissociation of E2⅐BeF 3 Ϫ (Fig. 7C) and E2⅐vanadate (Fig. 7D) for the wild type and mutants S72R, E90R, and 4Gi-46/47 after supplementation with 500 M excess Ca 2ϩ in the presence of calcium ionophore A23187 (to allow Ca 2ϩ access to the luminally exposed Ca 2ϩ sites). The E2⅐BeF 3 Ϫ complexes of S72R, E90R, and 4Gi-46/47 were, in contrast to that of the wild type, completely insensitive to Ca 2ϩ . The dissociation of E2⅐vanadate was also markedly slowed for S72R and E90R, whereas for  Table 2. The bottom panel shows 32 P i -phosphorylated Ca 2ϩ -ATPase bands on SDS-polyacrylamide separation gels in experiments assessing the ADP sensitivity of the phosphoenzyme accumulated with wild type and mutants S72R, E90R, and 4Gi-46/47 either under conditions of maximal E2P formation in the presence of dimethyl sulfoxide (max EP) or after phosphorylation of the Ca 2ϩ -loaded vesicles in the medium without dimethyl sulfoxide described above. The ADP sensitivity was studied by a 2-s incubation with 1 mM ADP at 25°C before the acid quenching (ADP ϩ). For comparison, the phosphorylation of samples that have not been incubated with ADP is also shown (ADP Ϫ). The samples indicated by bg represent background phosphorylation obtained in the presence of excess Ca 2ϩ (absence of EGTA) in the phosphorylation medium. mutant 4Gi-46/47 the dissociation of E2⅐vanadate was 2.4-fold accelerated relative to wild type.
Function of the Cytoplasmically Facing High Affinity Ca 2ϩ Sites-Phosphorylation of the Ca 2ϩ -ATPase by ATP depends on the binding of Ca 2ϩ at the high affinity cytoplasmically exposed sites of the E1 form. Fig. 8 shows the Ca 2ϩ concentration dependence of the steady-state level of phosphorylation from ATP. Mutants S72A, S72R, and K297A deviated less than 2-fold from the wild type with respect to the K 0.5 of Ca 2ϩ activation. The apparent Ca 2ϩ affinities of E90A, E90L, and E90R were more markedly affected, being 6-, 8-, and 4-fold reduced, respectively, relative to wild type (K 0.5 increased, see Table 2).
The left panels of Fig. 9 show the kinetics of the Ca 2ϩ binding transition measured by use of rapid kinetic instrumentation at 25°C. This transition comprises Ca 2ϩ binding to the dephosphoenzyme with accompanying enzyme conformational changes, i.e. reactions E2 3 E1 3 CaE1 3 Ca 2 E1 in Scheme 1. The assay takes advantage of the fact that only the Ca 2 E1 state is able to be phosphorylated by ATP (31). Ca 2ϩdeprived enzyme is incubated with Ca 2ϩ for varying time intervals (t in the mixing protocol at the top left of Fig. 9), and the amount of phosphorylatable Ca 2 E1 is determined for each time interval by a 34-ms incubation with [␥-32 P]ATP followed by acid quenching (32). The mutations S72R and E90R reduced the rate of the Ca 2ϩ binding transition 2-fold, whereas mutation K297A led to a 2-fold enhanced rate. Mutant S72A displayed a wild type-like rate of Ca 2ϩ binding. Importantly, the rate of the Ca 2ϩ binding transition was extremely slow in mutants E90A and E90L, being reduced by more than 2 orders of magnitude relative to wild type. This is in keeping with the large phosphorylation overshoots seen for E90A and E90L in Fig. 2 as a result of the accumulation of dephosphoenzyme at steady state and with the resulting small values for the derived rate constants k B . It should be emphasized that the rate constant of the Ca 2ϩ binding transition extracted from the data in Fig. 9 is independent of the rate of phosphorylation of Ca 2 E1 (32). The data in Fig. 2 indicate that the phosphorylation of Ca 2 E1, unlike the Ca 2ϩ binding transition, pro-ceeds with roughly the same rate constant (k A ϳ50 s Ϫ1 ) in mutants and wild type. To examine whether this is the case for E90A and E90L also under the conditions (pH 6.0) applied for Fig. 9, we repeated for these mutants the phosphorylation of Ca 2 E1 corresponding to Fig. 2 but this time at pH 6.0. The result was again that k A ϳ 50 s Ϫ1 for mutants as well as wild type (supplemental Fig. S3).
The dissociation of Ca 2ϩ from the high affinity sites of the Ca 2 E1 state back toward the cytosol was also examined by the rapid-quench technique (Fig. 9, right panels). Again, the assay used takes advantage of the dependence of the reaction with ATP on Ca 2ϩ occupancy of the Ca 2ϩ sites, as illustrated by the mixing protocol in the top right of Fig. 9. Thus, when an excess of EGTA is added to Ca 2ϩ -saturated enzyme, the ability to become phosphorylated by ATP will disappear at a rate corresponding to the rate of Ca 2ϩ dissociation from the enzyme. After incubation with EGTA for varying time intervals, the amount of enzyme still in the phosphorylatable Ca 2 E1 state is determined by a 34-ms incubation with [␥-32 P]ATP before acid quenching. Moderate deviations from wild type of the rate of Ca 2ϩ dissociation from the high affinity Ca 2ϩ sites of the Ca 2 E1 state were seen with mutants S72R, S72A, and E90R (1.5-and 2-fold reduced rates and 2.4-fold enhanced rate, respectively, Table 2). K297A displayed an ϳ3-fold slowing of Ca 2ϩ dissociation. Again the mutations E90A and E90L proved most detrimental, reducing the rates of Ca 2ϩ dissociation from Ca 2 E1 24-and 6-fold, respectively, relative to wild type. Thus, mutations E90A and E90L markedly inhibit the Ca 2ϩ binding transition as well as the dissociation of Ca 2ϩ from the high affinity cytoplasmically exposed Ca 2ϩ sites. Ca 2ϩ dissociation is also slowed by mutation K297A, but unlike E90A and E90L, K297A enhances the Ca 2ϩ binding transition slightly. These data on Ca 2ϩ interaction are summarized in Table 2. Caution is generally required in the interpretation of the K 0.5 values determined at steady state (Fig. 8) because they depend not only on the rate constants of Ca 2ϩ binding and Ca 2ϩ dissociation but also on the rate constants of all the other partial reaction steps in the Ca 2ϩ transport cycle, as previously discussed in detail

Summary of apparent affinities and rate constants relating to Ca 2؉ interaction
The values extracted by regression analysis are shown relative to that of wild type. The S.E. is indicated, and the value n refers to the number of experimental data points on which the analysis is based.
.5 values for Ca 2ϩ activation from the luminal side of the membrane of phosphorylation from P i , relative to that obtained with wild type (7.5 mM). For experimental details, see the legend to Fig. 6. b K 0.5 values for Ca 2ϩ activation from the cytoplasmic side of the membrane of phosphorylation from ATP, relative to that obtained with wild type (0.9 M). For experimental details, see the legend to Fig. 8. c Rate constants relative to that obtained with wild type (0.9 s Ϫ1 ). For experimental details, see the legend to Fig. 9, left panels. d Rate constants relative to that obtained with wild type (2.7 s Ϫ1 ). For experimental details, see the legend to Fig. 9, right panels. e Extraction of an apparent affinity for luminal Ca 2ϩ was not feasible for mutants S72R, E90R, and E309Q, because the phosphoenzyme levels of these three mutants were rather independent of the luminal Ca 2ϩ concentration, and for E90L the affinity was too low for an accurate determination (cf. Fig. 6). f ND, not determined in the present study. See Ref. 22 for a detailed functional study of mutant 4Gi-46/47. g ND, not determined, because E309Q does not undergo phosphorylation under the present conditions for reaction with ATP (21). (33). The shifts toward higher K 0.5 values for Ca 2ϩ activation observed for E90A and E90L are, however, in good accordance with the much larger inhibitory effects seen for the Ca 2ϩ binding transition as compared with Ca 2ϩ dissociation from Ca 2 E1 in these mutants. For E90R, the 2-fold slowing of Ca 2ϩ binding and 2-fold enhancement of Ca 2ϩ dissociation may both contribute to the increase of the K 0.5 for Ca 2ϩ activation of steadystate phosphorylation.

DISCUSSION
Here we have investigated the interaction network involving Ser 72 , Glu 90 , and Lys 297 at the luminally protruding ends of the transmembrane helices M1, M2, and M4 of the Ca 2ϩ -ATPase.
Our functional analysis of mutants with alterations to Glu 90 implicates this residue as a critical element at the luminal ion gate for extrusion of the Ca 2ϩ ions from the E2P state to the endoplasmic reticulum lumen. It is furthermore of note that some of the mutations to Glu 90 also had a pronounced impact on the function of the cytoplasmically exposed high affinity Ca 2ϩ sites of the E1 state, i.e. events that take place at the opposite side of the membrane relative to where Glu 90 is located.
Glu 90 and Glu 309 Provide Ligands for Luminal Ca 2ϩ Binding-The propagation of the movements of the A domain to the transmembrane region during the Ca 2 E1 3 Ca 2 E1P 3 Ca 2 E2P 3 E2P transitions leads to a considerable displacement of the transmembrane hairpins M1/M2 and M3/M4 relative to each other (cf. the arrows in Fig. 1). Consequently, in the presumed structural analog of the Ca 2ϩ -free E2P ground state, the E2⅐BeF 3 Ϫ crystal structure (7), Ser 72 and Glu 90 are positioned near Glu 309 , which is known as an essential Ca 2ϩ binding residue in Ca 2 E1 and Ca 2 E1P and probably plays a central role as well in the countertransport of protons (9,20,21). An issue of great importance for the understanding of the transport mechanism is where in the enzyme Ca 2ϩ is bound before it leaves from E2P in exchange for the protons to be countertransported. In E2⅐BeF 3 Ϫ the luminally protruding end of the transmembrane domain is in an open configuration compared with other crystal structures of the Ca 2ϩ -ATPase, with Glu 309 exposed to the lumen and associated with a Mg 2ϩ ion (7). The hypothesis that this Mg 2ϩ is bound in place of Ca 2ϩ at a luminally facing low affinity Ca 2ϩ site is consistent with the observation that under certain condi-   Table 2.
tions Mg 2ϩ in the lumen inhibits dephosphorylation of E2P, similar to the effect of Ca 2ϩ transported into the lumen (34,35). The close proximity of the side chains of Glu 90 and Ser 72 to the bound Mg 2ϩ ion (2.5 and 5.9 Å, respectively, see Fig. 1 and supplemental Table S1), therefore, led us to examine the functional roles of these two residues. The results in Fig. 6 suggest that the side chain of Glu 90 indeed does interact with Ca 2ϩ in the lumen, thus identifying an essential locus (a leaving site) on the hitherto unknown luminal Ca 2ϩ exit pathway. Hence, substitution of Glu 90 with alanine or leucine led to a marked reduction of the apparent affinity for luminal Ca 2ϩ , as indicated by the shift toward higher luminal Ca 2ϩ concentrations of the Ca 2ϩ activation curves for the reaction of E2 with P i (Fig. 6 and Table 2). The substitution of Ser 72 with alanine had no effect on the K 0.5 for Ca lum 2ϩ , suggesting a more peripheral position of Ser 72 relative to the luminal Ca 2ϩ site in accordance with the 5.9 Å distance to the Mg 2ϩ ion in the E2⅐BeF 3 Ϫ crystal structure. When, however, Ser 72 or Glu 90 was replaced by an arginine, possessing a longer and positively charged side chain, high amounts of phosphoenzyme were obtained even at very low luminal Ca 2ϩ concentrations, i.e. conditions where little or no phosphoenzyme was formed in the wild type or any of the other mutants examined here, except E309Q (Fig. 6). Like S72R and E90R, E309Q also displayed considerable activation of phosphorylation by P i at low luminal Ca 2ϩ concentrations and little further activation at higher luminal Ca 2ϩ concentrations within the range studied (Fig. 6). Thus, in S72R and E90R as well as E309Q, the E2P formation from P i appears to be constitutively activated without the need for Ca 2ϩ binding from the lumen. For S72R and E90R, this may be explained by occupation of a luminally facing Ca 2ϩ site by the guanidinium group of the arginine side chain, thus allowing the guanidinium group to mimic bound Ca 2ϩ to some extent due to its positive charge. For E309Q, the neutralization of the negative charge of the glutamate likewise provides a possible explanation in terms of mimicking Ca 2ϩ . To examine whether such ideas are realistic, we made structural models of the S72R and E90R mutants based on the E2⅐BeF 3 Ϫ crystal structure (Fig. 10). In E90R, the position of the guanidinium group of the arginine substituent overlaps with the position of the Mg 2ϩ ion for those rotamers of the arginine side chain with lowest energy, and after applying energy minimization to the arginine side chain as well as the side chain of Glu 309 , the resulting distance is consistent with bond formation between the guanidinium group and the side-chain carboxylate of Glu 309 (Fig. 10B), in accordance with the hypothesis that Glu 90 and Glu 309 provide oxygen ligands for a common luminally facing Ca 2ϩ site, which is occupied by the guanidinium group of the arginine side chain in E90R. It is also clear that such a site is disturbed by the guanidinium group in S72R (Fig. 10C). This model also explains the block of the E2P 3 E2 phosphoenzyme hydrolysis in S72R, E90R, and E309Q (Fig. 4), because dephos- The lines represent the best fits of a monoexponential rise to maximum function with an initial offset, EP ϭ EP min ϩ (EP max Ϫ EP min )⅐(1 Ϫ e Ϫkt ), giving the rate constants of E2 3 Ca 2 E1 listed in Table 2. In each case the EP max extracted from the fit was taken as 100%. Right panels, Ca 2ϩ dissociation from the high affinity Ca 2ϩ sites of Ca 2 E1 back toward the cytoplasm is shown. The enzyme, preincubated in 40 mM MES/Tris (pH 6.0), 80 mM KCl, 5 mM MgCl 2 , and 100 M CaCl 2 (to accumulate Ca 2 E1), was mixed with an equal volume of 40 mM MES/Tris (pH 6.0), 80 mM KCl, 5 mM MgCl 2 , and 4 mM EGTA to initiate Ca 2ϩ dissociation. At the indicated time intervals, the amount of phosphorylatable Ca 2 E1 remaining was determined by adding the double volume of 40 mM MES/Tris (pH 6.0), 80 mM KCl, 5 mM MgCl 2 , 2 mM EGTA, and 10 M [␥-32 P]ATP followed by acid quenching 34 ms later. To obtain the point corresponding to zero time, 4 mM EGTA was replaced by 100 M CaCl 2 . The lines show the best fits of a monoexponential decay function EP ϭ EP max ⅐e Ϫkt , giving the rate constants of Ca 2ϩ dissociation from Ca 2 E1 listed in Table 2. In each case, the phosphorylation level corresponding to zero time was taken as 100%.
phorylation of E2P during the normal Ca 2ϩ transport cycle does not take place until the translocated Ca 2ϩ has dissociated from the luminally exposed site(s).
It should be noted that E90A and E90L differ from E90R by not displaying a reduced rate of dephosphorylation of E2P (Fig.  4). The apparent affinity for luminal Ca 2ϩ was nevertheless found markedly reduced in these mutants relative to wild type (Fig. 6), thus indicating that the reduced luminal Ca 2ϩ affinity is not an indirect consequence of E2P stabilization. This supports the hypothesis that Glu 90 is directly involved in luminal Ca 2ϩ binding. Furthermore, because E2P is stabilized in E309Q, as in E90R, it is likely that neutralization of the negative charge of Glu 309 , similar to what is obtained by luminal Ca 2ϩ binding, is the reason for the E2P stabilization in E90R. This fits well with the hypothesis that in the normal Ca 2 E2P state Glu 90 and Glu 309 bind a Ca 2ϩ ion that through this interaction prevents dephosphorylation, probably through substantial rearrangements of the transmembrane domain, resulting in long-range effects of the altered positions of the transmembrane helices on the insertion of the 181 TGES phosphatase motif of the A domain into the catalytic site (5-7).
When phosphorylation was carried out with ATP in the forward-running mode of the Ca 2ϩ -ATPase cycle, the E2P phosphoenzyme accumulated for E90R (Fig. 3A) was Ca 2ϩ free (Fig.  5), which excludes the possibility that the stabilization of E2P seen under these conditions is due to hindrance of luminal Ca 2ϩ dissociation by the arginine side chain. This contrasts with the 4Gi-46/47 mutant that displayed a very stable E2P state with occluded Ca 2ϩ due to impaired Ca 2ϩ dissociation ( Fig. 5 and Ref. 22). Like 4Gi-46/47, the S72R and E90R mutants were unable to form ADP-sensitive Ca 2 E1P backward from the P i -phosphorylated enzyme (bottom panel of Fig. 6). In S72R and E90R the reason may be the arginine guanidinium group occupying a Ca 2ϩ site of E2P without being able to fulfill a similar role in E1P, because of the large distance between Ser 72 /Glu 90 and Glu 309 in this conformation (cf. Fig. 1). The hypothesis that E2P is stabilized by insertion of the arginine guanidinium group into a luminally exposed Ca 2ϩ site is also consistent with the highly accelerated loss of ADP sensitivity of the phosphoenzyme of E90R and S72R ( Fig. 3B and Table 1), because the measured rate of Ca 2 E1P 3 Ca 2 E2P is a net rate that depends both on the forward rate as well as the rate of the reverse reaction, which according to our hypothesis could be greatly reduced as a consequence of the guanidinium group interfering with luminal Ca 2ϩ binding in E2P, thereby enforcing Ca 2ϩ dissociation and depriving Ca 2 E2P. E90A and E90L likewise showed some acceleration of Ca 2 E1P 3 Ca 2 E2P, although not to the extent of S72R and E90R (Table 1), which may again be explained by a reduced rate of the reverse reaction due to enhanced Ca 2ϩ dissociation, as predicted from the reduced affinity of these mutants for luminal Ca 2ϩ ( Fig. 6 and Table 2).
The interaction of the arginine side chain with the luminal Ca 2ϩ site in the E2P state of S72R and E90R is further reflected by the results shown in Fig. 7 in which the Ca 2ϩ sensitivity of the E2P analog states E2⅐BeF 3 Ϫ and E2⅐vanadate were examined. Ca 2ϩ -induced dissociation of E2⅐BeF 3 Ϫ has been shown to result from Ca 2ϩ binding at a luminally exposed Ca 2ϩ site (28), in accordance with the idea that E2⅐BeF 3 Ϫ represents an E2P ground state-like conformation. The dissociation is most likely a three-step process in which Ca 2ϩ first binds at the luminally exposed Ca 2ϩ sites, forming Ca 2 E2⅐BeF 3 Ϫ (Ca 2 E2P analog) followed by a slow conformational transition to Ca 2 E1⅐BeF 3 Ϫ (Ca 2 E1P analog) and subsequent BeF 3 Ϫ release from Ca 2 E1⅐ BeF 3 Ϫ (36). E2⅐vanadate is generally believed to represent an E2P transition state-like conformation (29), and its dissociation is triggered by Ca 2ϩ binding at the high affinity Ca 2ϩ sites on the cytoplasmic side of the membrane allowed by the equilibrium of E2 with E1 (30). Accordingly, the Ca 2ϩ -induced dissociation of E2⅐BeF 3 Ϫ was very slow in mutant 4Gi-46/47 (Fig. 7C), reflecting the high stability of the occluded Ca 2 E2P state in this mutant (22), whereas the Ca 2ϩ -induced dissociation of the E2⅐vanadate complex was rapid (Fig. 7D), reflecting the more Ϫ state (same structure and view as in Fig. 1, right panel). B and C, procedures are the same as in panel A but with the luminal Mg 2ϩ ion removed and Glu 90 (B) or Ser 72 (C) substituted with arginine. The amino acid substitutions were carried out using the DeepView/Swiss-PdbViewer program (Swiss Institute of Bioinformatics). For E90R, the program provided a rotamer library of 28 possible conformations of the arginine side chain. The rotamer shown is one of three rotamers with the most favorable energy score, which all displayed arrangements where the arginine guanidinium group is held in place by the side chain carboxyl of Glu 309 , overlapping the region where the luminal Mg 2ϩ ion is positioned in the wild type structure. For S72R, the program provided 26 possible rotamers, of which the rotamer with the most favorable energy score is shown. In both mutant structures there was no steric hindrance to the arginine side chain. After the arginine substitution, energy minimization was also carried out for the Glu 309 side chain using the DeepView/Swiss-PdbViewer program and selecting the rotamer of the glutamate side chain with the most favorable energy score. Color codes as for Fig. 1. wild type-like characteristics of the E2P transition state of 4Gi-46/47, allowing formation of E2 from E2⅐vanadate at a normal rate and subsequent transition to E1 (22). In contrast, the dissociation of E2⅐BeF 3 Ϫ as well as that of E2⅐vanadate were very slow in both S72R and E90R (Fig. 7, C and D), thus reflecting the stabilization of a Ca 2ϩ -free E2P-like state with the arginine side chain located in the luminal Ca 2ϩ site.
Importance of Glu 90 for Cytoplasmic Ca 2ϩ Binding-The results of Fig. 9 and Table 2 show that Glu 90 is an important player not only at the luminal ion gate but is required for normal function of the cytoplasmic Ca 2ϩ sites as well.
The rate of Ca 2ϩ dissociation back toward the cytosol from Ca 2 E1 was markedly slowed by mutations E90A and E90L. Because Glu 90 is located at the luminal side of the membrane and, in the Ca 2 E1 state, some 15-20 Å below the Ca 2ϩ ions ( Fig.  1 and supplemental Table S1), any direct interaction with the cytoplasmic Ca 2ϩ sites seems to be excluded. In all the Ca 2 E1 crystal structures (with or without bound nucleotide or phosphoryl analogs), the side-chain carboxylate of Glu 90 is closely associated with the side-chain amino group of Lys 297 ( Fig. 1 and supplemental Table S1). This led us to speculate that a Glu 90 -Lys 297 ion bond might be involved in control of cytoplasmic Ca 2ϩ binding, and we, therefore, included mutation K297A in the study. The reduced rate of Ca 2ϩ dissociation seen for K297A supports the hypothesis that the presence of an ion bond between Glu 90 and Lys 297 influences the dissociation of Ca 2ϩ from Ca 2 E1. The finding that E90R displayed a 2-fold enhanced rate of Ca 2ϩ dissociation indicates that a surplus of positive charge in this area will destabilize the Ca 2 E1 state, thus pointing to a possible role for Glu 90 in neutralizing the positive charge of Lys 297 in Ca 2 E1. This influence on the cytoplasmically exposed Ca 2ϩ sites of E1 must be exerted by long-range effects involving repositioning of the transmembrane helices.
The Ca 2ϩ binding properties of E90A and E90L were further profoundly affected by a more than 2 orders of magnitudeslowing of the Ca 2ϩ binding transition E2 3 Ca 2 E1. Because K297A showed a 2-fold-enhanced rate of E2 3 Ca 2 E1 (Table 2), the slow rates of E2 3 Ca 2 E1 in E90A and E90L seem to be unrelated to the disruption of the ion bond between Glu 90 and Lys 297 in Ca 2 E1. Moreover, E90R displayed only a 2-fold reduction of E2 3 Ca 2 E1 relative to the wild type, implying that the very slow rate of this transition in E90A and E90L is related to the hydrophobic nature of the side chain in these mutants. The various structures of the Ca 2ϩ -ATPase crystallized in E2 state (37-41) provide a possible explanation. Thus, common to these structures is that the Glu 90 side chain is located inside a hydrophobic pocket consisting of Val 300 of M4 and Ile 788 , Pro 789 , and Val 790 of M6 (Fig. 11). It is, therefore, quite conceivable that substitution of Glu 90 with a hydrophobic residue such as alanine or leucine would strengthen the interaction between M2 and M4/M6, hindering the displacement of M1/M2 relative to M3/M4 during the E2 3 Ca 2 E1 transition, thus effectively locking the enzyme in the E2 conformation.
Importance of Lys 297 -In a previous mutational study predating the high resolution crystal structures of the Ca 2ϩ -ATPase, Chen et al. (19) carried out functional analysis of Ca 2ϩ pumps with other mutations to Lys 297 than the K297A mutation studied here and found that mutants K297M and K297F displayed slow E2P dephosphorylation, leading to the suggestion that Lys 297 seals the luminal gate of the Ca 2ϩ transport pathway (19). In the present study mutant K297A displayed a 3.4-fold reduction of the rate of E2P dephosphorylation ( Fig. 4 and Table 1), resembling the effects observed by Chen et al. (19) with K297M and K297F. However, the apparent affinity of the E2P state of K297A for Ca 2ϩ binding at the luminal sites was indistinguishable from that of the wild type ( Fig. 6 and Table 2), making it unlikely that Lys 297 is associated with the luminal ion gate of E2P. Rather, the effect of mutations to Lys 297 on E2P stability could be due to critical ion bonding between Lys 297 and the residue(s) in the loop connecting M1 and M2, one likely interaction partner candidate being Glu 79 (Fig. 1).
Conclusion and Perspective-In conjunction, all the data presented here seem to point to Glu 90 as a residue that participates together with Glu 309 in Ca 2ϩ binding at a luminally exposed site in the E2P state. This finding, together with previous functional (9) and structural (3, 7) evidence for an alternating exposure of Glu 309 at the two sides of the membrane during the pump cycle is consistent with a mechanism in which Glu 309 carries one of the two Ca 2ϩ ions along from the cytoplasmic side to a leaving site, where Ca 2ϩ is received by Glu 90 before the final exit to the lumen. Because the present data do not allow a distinction between the two Ca 2ϩ ions, it still remains to be clarified how the other Ca 2ϩ ion is translocated; that is, whether it also uses Glu 90 or takes a different exit pathway from the binding pocket?