The Holo-form of the Nucleotide Binding Domain of the KdpFABC Complex from Escherichia coli Reveals a New Binding Mode*

P-type ATPases are ubiquitously abundant enzymes involved in active transport of charged residues across biological membranes. The KdpB subunit of the prokaryotic Kdp-ATPase (KdpFABC complex) shares characteristic regions of homology with class II-IV P-type ATPases and has been shown previously to be misgrouped as a class IA P-type ATPase. Here, we present the NMR structure of the AMP-PNP-bound nucleotide binding domain KdpBN of the Escherichia coli Kdp-ATPase at high resolution. The aromatic moiety of the nucleotide is clipped into the binding pocket by Phe377 and Lys395 via a π-π stacking and a cation-π interaction, respectively. Charged residues at the outer rim of the binding pocket (Arg317, Arg382, Asp399, and Glu348) stabilize and direct the triphosphate group via electrostatic attraction and repulsion toward the phosphorylation domain. The nucleotide binding mode was corroborated by the replacement of critical residues. The conservative mutation F377Y produced a high residual nucleotide binding capacity, whereas replacement by alanine resulted in low nucleotide binding capacities and a considerable loss of ATPase activity. Similarly, mutation K395A resulted in loss of ATPase activity and nucleotide binding affinity, even though the protein was properly folded. We present a schematic model of the nucleotide binding mode that allows for both high selectivity and a low nucleotide binding constant, necessary for the fast and effective turnover rate realized in the reaction cycle of the Kdp-ATPase.

Ion transport is a vital prerequisite for cellular life, because ions play a crucial role in many biochemical processes, either as cofactors for enzyme function, as substrates for sym-and antiport with other substances, or maintenance of cellular pH or turgor. To maintain ion gradients across their membranes, cells employ highly sophisticated transport systems. Primary transport systems are driven by ATP to achieve transport against concentration gradients. A classic representative is the family of P-type ATPases, which play a fundamental role in the transport of heavy metal, alkali, and earth-alkali ions. Because of their enormous importance, eukaryotic P-type ATPases, such as the Ca 2ϩ -ATPase and the Na ϩ ,K ϩ -ATPase, have been studied intensively in recent years. A major breakthrough was achieved by Toyoshima et al. (1)(2)(3) with the elucidation of the Ca 2ϩ -ATPase structure in different reaction cycle intermediates. In addition, the structure of the nucleotide binding domain of the Na ϩ ,K ϩ -ATPase was solved by x-ray crystallography (4). However, distantly related members of the P-type ATPase group are less well examined, and in particular, the important question of the evolutionary relationship of the different P-type ATPases remains uncertain.
An unusual prokaryotic P-type ATPase, the KdpFABC complex, serves as a good model to address several questions for both mechanistic and evolutionary aspects. In Escherichia coli, the KdpFABC complex serves as a highly specific potassium transport system, which is only synthesized when the cells' need for K ϩ can no longer be satisfied by the constitutive K ϩ transport systems (5). This P-type ATPase differs from most others by the fact that catalytic activity (KdpB) and ion transport (KdpA) are associated with two different subunits (6 -8). For communication between KdpB and KdpA, charged residues within transmembrane helix 5 of KdpB are essential (9). The catalytic subunit KdpB is a P-type ATPase with a unique membrane-bound region (seven transmembrane helices) and a small nucleotide binding domain (17 kDa), probably reflecting an elementary nucleotide binding mechanism. Recently, we solved the solution structure of the apo-form of the nucleotide binding domain of KdpB (KdpBN; Asn 316 -Gly 451 ) by NMR spectroscopy (10). Based on nucleotide titration experiments and a small number of NOE 5 distance restraints, we proposed a preliminary model of the complex where the nucleotide was docked into apoKdpBN (10). It has been concluded that KdpBN does not show major conformational changes upon ATP binding, in contrast to previous discussions in the literature (11). Rather, it is suggested that KdpBN undergoes a rigid body movement upon ATP binding and that it is connected to the phosphorylation domain via a flexible linker. Here, we report for the first time the atomic resolution structure of the nucleotide bound state of KdpBN based on a complete NOE data set, refining and extending the proposed model. To substantiate the role of certain residues, amino acids Phe 377 and Lys 395 as well as Ser 384 and Asp 399 were substituted and their influence on ATP binding in the KdpFABC complex as well as in the isolated KdpBN domain was studied. In contrast to the previously formulated notion that KdpB belongs to type I P-type ATPases (12,13), this study corroborates that KdpB is more closely related to alkali-and earth-alkali-transporting P-type ATPases, sharing most similarities with the proton pump (type III).

MATERIALS AND METHODS
Growth Conditions, Media, and Supplements-E. coli cells were routinely grown in Luria Bertani medium (14) supplemented with the appropriate antibiotic. For the synthesis of recombinant KdpBN proteins, cells were grown in K115 minimal medium, as described previously (15). Strains carrying mutations in the kdp operon were grown in KML or K0 minimal medium with different potassium concentrations, according to Ref. 15. Supplements were added in the following concentrations: ampicillin, 100 g ml Ϫ1 ; thiamine, 1 g ml Ϫ1 ; isopropyl-␤-Dthiogalactopyranoside, 1 mM.
Generation of KdpB Mutants-The kdpFABC genes expressed from the pSMC10His vector (5) result in a KdpFABC complex, with KdpC carrying a C-terminal decahistidinyl epitope. The primer pairs KdpBK395A/BamHI_rev and KdpBD399A/BamHI_rev (see supplemental Table S1) were used to create constructs carrying the point mutations K395A and D399A, respectively. The resulting PCR products were ligated ClaI/BamHI into appropriately opened pSMC10His, in which a silent BamHI site was previously introduced. Using primer KdpBN2_for in combination with one of the mutagenesis primers KdpBF377A, KdpBF377Y, KdpBS384A, or KdpBS384T (see supplemental Table S1), PCR products were obtained carrying point mutations at positions Phe 377 and Ser 384 , respectively. These DNA fragments were ligated via ClaI/DraIII into appropriately opened pSMC10His. Modified KdpFABC complexes were synthesized using TKW3205 cells (5). This strain was routinely used for the expression of kdpFABC constructs, because it lacks the chromosomally encoded kdpFABC, atp, trkA, and trkD genes. Expression of the kdpFABC genes is under the control of the endogenous kdpD and kdpE gene products (6). Because not all kdpB mutants constructed enabled growth on low potassium concentrations, K ϩ concentrations in the millimolar range were used for induction, although the KdpD/E system triggers kdpFABC expression best below 1 mM KCl. Therefore, in these mutants, the expression of kdpFABC differed compared with wild type, which in consequence resulted in low yields of purified KdpFABC complexes. To generate modified KdpB nucleotide binding domains (KdpBN), the primer pair KdpBNfor and KdpBNrev was used to amplify the corresponding DNA fragments from the different pSMC10His derivatives carrying the point mutations described above. The resulting PCR fragments were ligated NdeI/XhoI into appropriately digested pET16b vector (Novagen). Modified KdpBN proteins were synthesized using BL21(DE3) cells carrying pLysS (Novagen). All constructs were checked by DNA sequencing (Department of Botany, University of Osnabrück, Osnabrück, Germany).
Sample Preparation-KdpFABC complexes were purified by affinity chromatography via a histidine tag at the C terminus of KdpC (9). 15 Nand 13 C-labeled KdpBN were synthesized and purified as described previously (10). Unlabeled KdpBN was purified as described by Bramkamp and Altendorf (16).
ATPase Assay-ATPase activity of purified KdpFABC complex was determined using the microtiter plate assay of Henkel et al. (17) following the modifications described previously (18).
NMR Spectroscopy of KdpBN-The AMP-PNP-bound NMR samples used for structure determination contained 1.0 mM protein to which 10% D 2 O was added. A 15-fold molar excess of AMP-PNP was used. All spectra were recorded at 300 K at 600, 750, or 900 MHz on Bruker DMX600, DMX750, and DMX900 spectrometers, respectively. Sequential and side chain assignment was carried out as reported previously (19). Distance data were derived from three-dimensional NOESY spectra, which were all measured with an 80-ms NOE mixing time. A three-dimensional 15 N-HSQC-NOESY was recorded on the 15 N-labeled sample, and a three-dimensional 13 C-HSQC-NOESY, a CCH-NOESY, and a CNH-NOESY (20) were recorded on the doublelabeled sample of the AMP-PNP-bound protein as well as an 2 -12 C, 14 N-filtered two-dimensional 13 C(H)-NOESY experiment to trace intermolecular constraints (21). Saturation transfer difference NMR experiments (22) and WaterLOGSY experiments (23) were performed on unlabeled samples of wild type and mutants F377A, F377Y, K395A, and D399A at protein concentrations of 0.5 mM and AMP-PNP concentrations of 5 mM. For the integration of the one-dimensional proton spectra of the saturation transfer difference and WaterLOGSY experiments, the exact relative sample concentration was determined by integrating over the ␦-methyl signal of Leu 355 at Ϫ0.2 parts/million. However, there is still a residual background from protein resonances visible in the WaterLOGSY spectra; therefore, the signal of H1Ј was chosen for evaluation, as it does not overlap with protein signals. In a reference WaterLOGSY experiment, a 5 mM solution of AMP-PNP only was recorded under similar buffer conditions. This represents the nonbinding case for the specific ligand signal and thus the integral over the H1Ј signal was set to Ϫ100%. All spectra were processed with X-WIN-NMR (Bruker Analytik GmbH, Karlsruhe, Germany) and evaluated with SPARKY software programs. 6 Integration of the one-dimensional spectra was performed with X-WINNMR.
Backbone Dynamics-For R 1 and R 2 experiments, typically 10 relaxation delays and three duplicate points for error estimation were recorded using experiments published earlier (24). In the 15 N dimension, 80 complex points were sampled with 48 transients and a recycle delay of 2 s between successive scans. For the heteronuclear NOE, two sets of experiments consisting of spectra with and without saturation of the amide protons were recorded. A relaxation delay between two transients of 5 s was followed by saturation of the amide protons for 3 s. Saturation was achieved by a train of 120°square pulses with the carrier centered in the middle of the amide region and the radio frequency amplitude set so as to achieve a null excitation at the water resonance. In the reference experiment, the pulse train was substituted by a delay of equal length. Relaxation dispersion data were recorded at B 0 fields of 14.1 and 17.6 tesla using a published sequence (25,26). For each profile, 14 experiments with different delays between the 180°pulses ranging from ϳ600 s to 20 ms and a total relaxation period of 80 ms were collected using 80 complex points in the 15 N dimension, 48 transients, and a relaxation delay of 2 s. In addition, a reference experiment lacking the relaxation period was recorded. Peak intensities were extracted using SPARKY. Relaxation rates were obtained from fitting the intensities to an exponential decay using CURVEFIT 7 and the perlscript SPARKY2RATE. 8 A minimum uncertainty of 2% for all rates was assumed. Initial guesses of the rotational diffusion tensor were obtained using QUADRIC_DIFFUSION. 7 Highly mobile residues or residues subject to chemical exchange were excluded from the estimation. Model-free analysis of the 15 N relaxation data were achieved using MODEL-FREE 7 interfaced with FastMODELFREE using a published protocol (27,28). Analysis of relaxation dispersion data was done using selfwritten scripts for SCILAB following a procedure described earlier (26,29,30).
Structure Calculations-Structures were calculated with X-PLOR version 2.9.3 using standard protocols (31). The evaluation of NOE cross-peaks in the three-dimensional spectra was carried out as described previously (10). Dihedral angle restraints for the backbone and angles were derived from C ␣ , C ␤ , and H ␣ chemical shifts using the program TALOS (32). Restraints were applied for the 85 high confidence predictions found by the program using the calculated range Ϯ10°.
Coordinates-The coordinates for the structure ensemble and a regularized average structure for the AMP-PNP bound form of KdpBN have been deposited in the Protein Data Bank (PDB) (accession codes 2A29 and 2A00, respectively).

RESULTS
Three-dimensional Structure of Nucleotide-bound KdpBN-We have previously published the NMR solution structure of the apo-form of the nucleotide binding domains of KdpB termed KdpBN, together with a model of the holo-form based on data for the apo-form, which suggested that no significant conformational changes occur within the enzyme upon nucleotide binding (10). Here, we present the NMR solution structure of AMP-PNP-bound KdpBN. For the structure calculations, 1642 intramolecular NOE distance restraints were evaluated (see Table 1). Another 244 dihedral restraints were derived from chemical shift data using the program TALOS (32). In addition, 24 intermolecular NOE contacts were assigned between the protein and the ligand. The resulting set of 19 final simulated annealing structures has a root mean square deviation of 0.24 Å for superimposition over backbone atoms of ordered residues (see Table 1). The enzyme consists of a six-stranded anti-parallel ␤-sheet flanked by two ␣-helices on each side, with a short 3 10 -helix between helix ␣2 and strand ␤2. The minimized average structure and the superimposition of the final set of 19 structures are shown in Fig. 1.
Structural Consequences of Nucleotide Binding-In general, the chemical shifts of apo-and holoKdpBN show only minor differences. Shifts characteristic of secondary structure elements, C ␣ , H ␣ , and C ␤ show changes that do not influence the classification of the residues according to the chemical shift index protocol (33). The secondary structure and the average estimated solvent accessibility of both sets of 19 structures, as determined by the program PROCHECK (34), are shown in Fig. 2, left panel. The only significant modification occurs to the C-terminal end of helix ␣1, which is extended by three residues upon nucleotide binding. The superimposition of the minimized averaged structures of the apo-and holoenzyme, shown in Fig. 2, right panel, exhibits a root mean square deviation of 1.26 Å over ordered backbone residues.
A more thorough examination of the chemical shift data helps in understanding the binding mechanism. In supplemental Fig. S1, the difference of the chemical shifts of holo-to apoKdpBN of selected nuclei is plotted against the sequence. C ␣ , H ␣ , C ␤ , H ␤ , H N , N H , and methyl groups were evaluated. Resonances of the backbone atoms changed most noticeably in the loop region between helices ␣1 and ␣2. Helix ␣1 had been extended by one turn, thus Ala 340 , which had previously been at the C-terminal end of helix ␣1, was then fully integrated. The side chain of Asp 344 , which had previously been involved in the hydrogen bonding to H of Lys 395 , was bent away to make way for the ligand and established a hydrogen bond to N6 of the nucleotide. The side chains of Glu 348 and Arg 317 were in close contact in the apoprotein (3.2 Å), but in the holoprotein, Glu 348 moved aside and Arg 317 then pointed toward the termini of the domain and thus toward the triphosphate group of AMP-PNP.
Another part that was affected considerably was the ␤-sheet region spanning the binding pocket (see supplemental Fig. S1). The turn connecting strands ␤2 and ␤3 closes down by ϳ15°on the ligand as the curved ␤-sheet stretches. Pro 376 , which shows the most intense C ␣ chemical shift deviation and almost identical / angles for apo-and holo-form, represents the pivotal point in strand ␤2. Ser 384 is the corre-    Table 1). This figure and Figs. 2, 3, and 5 were produced using PYMOL (45).
Chemical shifts of methyl groups are sensitive probes for structural alterations and electrostatic interactions caused by the ligand. Interestingly, the methyl carbons are hardly affected at all in contrast to the methyl protons close to aromatic systems. Significant changes in chemical shift were observed for residues in the loop region between helices ␣1 and ␣2 (see supplemental Fig. S1), as well as for Leu 431 situated on strand ␤5; its side chain is located in proximity to the purine ring. Both methyl groups come as close as 3-4 Å to the aromatic system, explaining the strong upfield shift.
Characterization of the Binding Pocket-The core of the binding pocket is dominated by neutral residues. Only one negatively charged residue, Asp 344 , is situated at the center and forms a hydrogen bond to the amino group of the nucleotide (N6). The only positively charged residue within the core is Lys 395 , forming a cation--stacking interaction to the six-membered part of the purine ring of AMP-PNP. Two positively charged residues, Arg 317 and Arg 382 , are situated diametrically opposed to each other on the outer rim of the binding pocket (Fig.  3a). Both form hydrogen bonds to the triphosphate group of the nucleotide and orient it toward the phosphorylation domain. Somewhat further away (6 -8 Å) are two negatively charged residues, Glu 348 and Asp 399 . They are also diametrically opposed to each other, and their axis forms an angle of ϳ45°with respect to the axis defined by Arg 317 and Arg 382 (Fig. 3b). These residues may provide stabilization by electrostatic repulsion within a certain range of flexibility. Furthermore, electrostatic interactions between the protein and the ligand are established by Ser 384 and N1 of AMP-PNP. Most residues in the binding pocket are, however, hydrophobic. The most important interaction is established by Phe 377 , which exhibits astacking with the nucleotide, whereas Leu 431 approaches the five-membered ring of AMP-PNP from the other side.
Backbone Dynamics-To further characterize structural changes upon binding of AMP-PNP, the backbone dynamics of apo-and holo-KdpBN were investigated on a fast (pico-to nanosecond) and a slow (millisecond) time scale, using the amide nitrogen as probes. Model-free analysis (35,36) of the R 1 , R 2 and heteronuclear NOE data show that internal motions on the fast time scale in the secondary structure elements of KdpBN are highly restricted, with averaged squared order parameters of 0.90 Ϯ 0.01 for the apoprotein and 0.91 Ϯ 0.01 for the holo-form, respectively. Regions with higher internal mobility are found in the loops connecting helix ␣2 and the 3 10 helix (Val 367 -Leu 370 ), between helices ␣3 and ␣4 (Asn 408 -Phe 412 ), and at the C terminus (Ile 448 -Gly 451 ); for the latter residues, virtually unrestricted motion is  observed. Ala 356 , although located in helix ␣2, stands out with an extremely low value for the squared order parameter S 2 (0.28 in the apoprotein and 0.24 in the holo-form).
As expected, the majority of residues showed no alteration in their dynamic behavior on a fast time scale upon ligand binding (see Fig. 4). Changes in the squared order parameter, defined as ⌬(S 2 ) ϭ S 2 (holo) Ϫ S 2 (apo), are considered to be significant, if the uncertainty is less than half of the corresponding parameter value. A dramatic decrease in S 2 upon ligand binding is observed for Ser 341 . This increase in mobility, also reflected in the relaxation parameters (R 2 is reduced by a factor of 2 in the complex), is somewhat surprising, because Ser 341 is located in a loop region in the apoprotein, whereas it is part of helix ␣1 in the complex. Phe 377 , located at the C-terminal end of strand ␤2, also experiences a higher degree of internal mobility in the holo-form, as indicated by the negative value of ⌬(S 2 ). In the apo-form, the amide proton of Phe 377 is hydrogen-bonded to A343CЈ, an interaction that is absent in the complex as a consequence of a different orientation of Ala 343 due to the elongation of helix ␣1. In contrast, Thr 378 features a positive ⌬(S 2 ) value. Thr 378 is involved in two hydrogen bonds with Met 383 (T378HN-M383CЈ and M383HN-T378CЈ); surprisingly, a negative ⌬(S 2 ) value is observed for Met 383 . The two regions of higher mobility in the apo-form appear more rigid in the complex.
On a slow (ms) time scale, 42 residues subject to chemical exchange, with an average exchange rate constant ͳk ex ʹ ϭ 881 Ϯ 40 s Ϫ1 , were identified for apoKdpBN. In the complex, 63 residues were found to be affected by an exchange process, with ͳk ex ʹ ϭ 414 Ϯ 36 s Ϫ1 . For residues subject to chemical exchange, the data were refitted assuming that all residues were affected by the same process, and the average values of k ex were used. For residues fitted to the general equation, averaged populations of the major populated site of 0.98 Ϯ 0.01 and 0.96 Ϯ 0.01 were obtained for the apo-and holo-form, respectively. In the absence of ligand, residues with exchange contributions cluster around the nucleotide binding site. In addition, exchange contributions are found for a number of residues in helix ␣4. For complexed KdpBN, chemical exchange affected a larger number of residues. In addition to residues in proximity to the binding site, the N-terminal winding of helix ␣2, virtually all residues in helix ␣4, and the C-terminal part of strand ␤5 showed significant exchange contributions. The differences in chemical shift for the apo-and holo-form, as obtained from the fitting procedure assuming a two-site exchange, suggest that the conformations adopted by KdpBN are not altered by ligand binding (see Fig. 5). A proposed mechanism by which ligand binding can be linked to dynamic changes distant from the binding site is presented in supplemental Fig. S2.
ATPase Activity of Different KdpB Mutants-To investigate the importance of individual residues for ATP hydrolysis, several KdpFABC mutants were constructed. The residues chosen for mutation were those of KdpBN, which showed strong alterations in their backbone 15 N chemical shift upon nucleotide addition (10). In each case, mutant Kdp-FABC complexes were detergent-solubilized from the membrane, affinity-purified via a histidine tag at the C terminus of KdpC, and analyzed for subunit composition by SDS-PAGE. Mutations in the N-domain of KdpB did not destabilize the KdpFABC complex as observed for mutations within the transmembrane region of KdpB (9). The ATPase activities of KdpFABC complexes are usually stimulated with 1 mM KCl by a factor of 3-5, and this was also observed for these mutant protein complexes. Because the K ϩ affinity of the wild-type KdpFABC complex is quite high (2 M), K ϩ impurities in the reaction buffer contribute to an already elevated basal activity of the protein, explaining the low stimulation factor for the wild type. In addition, the sensitivity of the ATPase activity to ortho-vanadate is comparable with wild type, indicating that all mutant protein complexes were capable of performing a full reaction cycle with E1-E2 transition. The results, summarized in Table 2, illustrate the important roles of amino acids Phe 377 and Lys 395 . Although changes in other amino acids located within the nucleotide binding    (23) is an ideal tool for studying weak binding ligands. The non-binding case (i.e. AMP-PNP in buffer without protein) provides the maximum negative signal, which can be standardized to Ϫ100% (Fig. 6a). Binding ligands show increasingly positive signals, reaching a maximum at concentrations around their binding constant. The ligand binds to the wild-type protein with an intensity of ϩ64%. The conservative mutant F377Y has the highest intensity of ϩ46%, which corresponds to 72% of wild-type binding capacity. F377A and K395A show only weak residual interaction with the ligand, i.e. Ϫ25% and Ϫ34%, respectively.
As strong binding ligands are indistinguishable from non-binding ligands in WaterLOGSY, saturation transfer difference experiments (22,37) were also performed. This experiment shows positive signals for binding ligands, with intensities proportional to the distance of individual groups from the protein surface. The most intense signal in the complex of AMP-PNP and wild-type enzyme was standardized to 100%. This was observed for the aromatic proton H2, which points directly into the binding pocket and which is involved in a close network of NOE contacts to the surrounding protein. The second observable aromatic proton is H8. It points toward the mouth of the binding pocket and is far from the protein surface. Consequently, the observed signal is weak (5%). H1Ј, the only observable proton from the sugar moiety has an intensity of 56%; it is closely anchored to the strands ␤3 and ␤4 via multiple NOE contacts (Fig. 6b). H2Ј resonates at the same frequency as water and is thus not detectable. The other protons, H3Ј, H4Ј, and H5Ј do not show a detectable signal, but they also resonate close to the water frequency and hence may be suppressed by the water suppression scheme. The sugar proton signals may also be below the detection threshold of ϳ5%. The conservative F377Y mutant shows 61% residual signal intensity for the H2 proton compared with the wild type, and 2.6 and 35% for H8 and H1Ј, respectively. The net signal intensity of the F377A mutant to the ligand is already below 5% compared with the wild type, whereas the K395A mutant no longer shows any signals. These results confirm that the mutated residues are responsible for ligand binding, as the mutants no longer have the ability to bind nucleotide, although NMR shows them to be natively folded and stable over several days.

DISCUSSION
A central question arising from the modular design of P-type ATPases has always been how nucleotide binding is realized in the clearly separated N-domain. In the very well studied case of P-loop ATPases, ATP is bound via a hydrophobicstacking to a conserved phenylalanine or tyrosine residue on the surface of the enzyme, and additionally, the triphosphate group of ATP caps the N terminus of an ␣-helix, leaving it embedded in a close network of hydrogen bonds. Evidently, this cannot be the case for P-type ATPases. It had been found earlier that a highly conserved lysine residue in the KGXX(D/E) motif and a phenylalanine 20 -25 amino acids upstream in the sequence are indispensable for nucleotide binding (11,38). However, both crystal structures of the Ca 2ϩ -ATPase with AMP-PCP bound (1,39) could not explain the necessity of the lysine residue. The modeling of the N-domain of KdpB with AMP-PNP bound, which was based on intermolecular distance restraints, suggested that the positively charged N end group is involved in a cation-interaction with the aromatic part of the nucleotide (10). The structure determination of the holo-form based on a completely new set of NMR data presented here confirms and extends this previous model. In the apo-form, the acidic side chain of Asp 344 is involved in H-bonding with Lys 395 . It also seems likely that its charge, positioned at the C-terminal end of helix ␣1, destabilizes the helix. In the holoenzyme, Asp 344 changes position to establish an H-bond to the amino group of the adenine, thereby retracting negative charge from the C-terminal end of helix ␣1, which is then elongated by one turn. Unlike the highly conserved lysine and phenylalanine residues, the aspartate has no conserved counterpart in other P-type ATPases, although Glu 442 seems to substitute for the aspartate in the calcium pump (1,39). Glu 442 is situated on the neighboring helix (␣2 in KdpBN numbering) and has its acidic side chain in the same position as Asp 344 in KdpBN, i.e. pointing toward the amino group of adenine. The nucleotide binding pocket of KdpBN (Fig. 7) is perfectly shaped by the curved ␤-sheet and helices ␣1 and ␣2, which allows a rapid exchange of nucleotide, a prerequisite for a functional reaction cycle (10,41). The preformed binding site is able to accept and release the ligand without the need for time-and energy-consuming conformational changes, consistent with a low bind- ing constant (1.4 mM for the isolated N-domain). The mobility on a slow time scale observed for the loop connecting helices ␣1 and ␣2 suggests that a motion of this stretch is necessary to provide access to the ligand binding site. Furthermore, our data show that this motion is present in both the apo-and holo-form of the protein; thus these slow motions likely represent an inherent property and do not require energy associated with the ligand binding event.
To compensate for this low affinity binding site described above, charged residues were placed strategically on the outer rim of the bind-ing pocket, which directed the triphosphate end group via electrostatic attraction and repulsion toward the solvent, thereby making it point toward the phosphorylation domain. The nucleotide itself is forced to adopt an almost linear form, accessorially facilitating the release of the ␥-phosphate to the neighboring P-domain. This peculiar binding mode where the nucleotide is "clipped" into the binding pocket by Phe 377 , Lys 395 , and Asp 344 in the core and stabilized and directed by Arg 317 and Arg 382 on the surface, is depicted schematically in Fig. 8.
The Ca 2ϩ -ATPase showed a similar arrangement of positive charges  around the outer rim of the nucleotide binding pocket; Arg 489 and Arg 560 both participate in the orientation of the ␤-phosphate (1). Mutation of Arg 560 in the Ca 2ϩ -ATPase reduced the apparent ATP affinity Ͼ100-fold (42) and showed a drastically reduced steady state level of phosphoenzyme production (5,43), obviously because of a hydrogen bond formed between Arg 560 and the P-domain (1). Despite these similarities, there is one major difference concerning the apparent nucleotide binding mode. The salt bridge formed by Lys 395 and Asp 344 in apoKdpBN is opened in holoKdpBN in favor of newly established contacts toward the nucleotide; Asp 344 makes a hydrogen bond to the amino group of AMP-PNP and Lys 395 is drawn toward the purine ring. This salt bridge finds its counterpart in the Ca 2ϩ -ATPase in residues Lys 515 and Glu 442 . However, this salt bridge remains intact in the crystal structure of the holoCa 2ϩ -ATPase. In the case of Ca 2ϩ -ATPase, the integrity of the enzyme seems to be dependent on the maintenance of this bridge (42), whereas for KdpBN, the salt bridge appears to be less important. A comparison of the chemical shifts of the two C ⑀ H 2 protons revealed that the lysine side chain is highly flexible in the apoenzyme versus an increased rigidity in the holoenzyme presumably because of the cation--interaction with the purine ring. In both cases, the positively charged NH 3 group cannot be observed in the 1 H, 15 N-HSQC spectra, as it is subject to fast proton exchange with water. Furthermore, the NMR spectra of the K395A mutant prove that the enzyme is properly folded and stable over several days without the stabilizing effect of a salt bridge, although it has completely lost the ability to bind nucleotides. The discrepancy in the binding mode may be due to the somewhat different structures of the two enzymes. In KdpBN, Asp 344 is located at the C-terminal end of helix ␣1 and changes position as this helix lengthens upon nucleotide binding. In the Ca 2ϩ -ATPase, Glu 442 is located at the N-terminal end of helix ␣2, and its position is unchanged in the apoand holo-forms. This difference is also reflected in the apparent K d values of the isolated N-domains, which is 10 -100 M for SERCA (44) and 1.4 mM for KdpBN (10).
The key role of Phe 377 and Lys 395 in the nucleotide binding process has been demonstrated clearly by mutational studies. Neither F377A nor K395A showed residual ATPase activity or a significant nucleotide affinity, substantiating the vital role of both. The nucleotide binding domain alone cannot discriminate between ATP, ADP, AMP, or AMP-PNP (10), as the ␤and ␥-phosphate groups are not involved in any contacts to the protein surface. Thus, the N-domain is likely to serve as a means of orienting the nucleotide in an appropriate way so that Asp 307 in the P-domain can be targeted at an optimal angle by the ␥-phosphate group. It is therefore necessary to enable on the one hand a rapid exchange of ADP by ATP to avoid a deceleration of the reaction cycle due to a lack of ATP, and on the other hand, the exchange rate should not exceed a certain speed limit, as this would again thwart the catalytic turnover by destabilization of the nucleotide-enzyme complex. This precisely balanced equilibrium of nucleotide uptake and release is a fundamental prerequisite for a functional and rate-optimized reaction cycle. It can be speculated that the different geometries of the nucleotide binding domains of the individual P-type ATPases are a result of finetuning the nucleotide exchange rate to the turnover rates of the individual reaction cycles.