INTRODUCTION

TOP ABSTRACT INTRODUCTION RESULTS DISCUSSION REFERENCES

In all cells, ionic homeostasis is maintained by members of the large family of P-type ion pumps (Axelsen and Palmgren, 1998). Ca²⁺-ATPase and Na⁺/K⁺-ATPase are the two best-studied members, and they have been characterized in great detail with regard to reaction kinetics, chemical modification, and site-directed mutagenesis (Andersen, 1995; Moller et al., 1996; MacLennan et al., 1997). Structural studies by electron microscopy have recently culminated in structures at 8-Å resolution for both Ca²⁺-ATPase (Zhang et al., 1998) and H⁺-ATPase (Auer et al., 1998). These structures revealed similar arrangements of 10 transmembrane helices, and, in the case of Ca²⁺-ATPase, connections were identified between four of these transmembrane helices and the corresponding stalk helices. In addition, a particular pathway for calcium through the transmembrane domain was proposed and the sequences of stalk and transmembrane helices were assigned, using a variety of constraints. Similar assignments were not possible for the cytoplasmic domains because they contain mainly -strands and short helices, which are not easily identified at 8-Å resolution.

Nevertheless, based on secondary structure predictions and on functional considerations, a sophisticated model has evolved for the cytoplasmic portion of Ca²⁺-ATPase, which includes the delineation of several domains (MacLennan et al., 1985; Taylor and Green, 1989; Green and Stokes, 1992). According to this model, the main cytoplasmic loop between transmembrane helices M4 and M5 begins with a phosphorylation domain that is separated from a nucleotide-binding domain by the dominant tryptic cleavage site (at R⁵⁰⁵) and ends with a 60-residue "hinge" domain, which was postulated to fold back onto the phosphorylation domain to produce the tertiary fold. An alternative to these three sequential domains is a model in which the nucleotide-binding domain forms an insert relative to a combined phosphorylation/hinge domain. Although the sequential domain model was never entirely satisfactory, the alternative with an inserted domain was unusual and, at that time, was not supported by evidence for an insertion site.

However, Aravind et al. (1998) recently proposed the existence of a superfamily of HAD hydrolases that includes the P-type pumps and that specifically supports a model with an inserted nucleotide-binding domain. This superfamily is based on homology between the phosphorylation domains of the P-type ATPases and the catalytic domains of serine/threonine phosphatases, phosphoglycomutases, and haloacid dehalogenases (HADs). Besides the sequence homology, these families also share the use of an aspartyl ester intermediate (Collet et al., 1998), the dynamics of which, in the pumps, are closely linked to the occlusion and transport of cations across the membrane. Here we show how the proposed homology leads to a new domain structure for the P-type ATPases, which we correlate with domains identified in the cytoplasmic portion of the 8-Å density map for Ca²⁺-ATPase. By fitting the crystal structure for a haloacid dehalogenase to this map, we propose a model not only of the phosphorylation domain, but also of the structural components coupling this domain with the ion transport sites within the membrane. This model provides insight into the effects of site-directed mutation and chemical modification; it supports the existence of two overlapping nucleotide-binding sites, and it suggests a role for Mg²⁺ in initiating the conformational changes required for Ca²⁺ translocation.

	RESULTS

TOP ABSTRACT INTRODUCTION RESULTS DISCUSSION REFERENCES

Sequence alignment and domain structure of dehalogenases and P-type pumps

A homology between the L-2 haloacid dehalogenase from Pseudomonas sp. YL (1JUD) and P-type pumps was proposed by Aravind et al. (1998), based on four sequence motifs that contribute to the active sites of these divergent enzymes. We begin our comparison by reviewing evidence from a wide variety of sources that support this homology. The crystal structure of 1JUD reveals the existence of two distinct domains (Fig. 1) (Hisano et al., 1996), the core domain being an / sandwich with the topology of a Rossmann fold (-strand order 3-2-1-4-5-6). The catalytic aspartate is at the C-terminus of 1 and is followed first by a five-residue coil (cyan in Fig. 1) and then by the inserted subdomain, which consists of a distorted four-helix bundle (gray in Fig. 1). The chain then returns to the first helix (5) of the core domain and completes the Rossman fold. The sequence motifs involved in the proposed homology are in four of the loops at the top of the -sheet (red residues in Fig. 1 b, including the catalytic aspartate and the cyan coil). According to mutagenesis studies (Kurihara et al., 1995) and the crystal structures of dehalogenases in the presence of a substrate analog (Li et al., 1998; Ridder et al., 1997), most of these conserved residues contribute directly to the active site. R⁴¹ is the only residue from the -helical subdomain that is directly involved in dehalogenation and serves to stabilize and to abstract the halide (Ridder et al., 1997). This role suggests that the subdomain is primarily involved in substrate specificity and binding, which would explain its variability across the proposed superfamily.

View larger version (28K): [in this window] [in a new window]   FIGURE 1   Structure of chloroacetate dehalogenase (1JUD). (a) This ribbon model was generated by ICM (Abagyan et al., 1994), using a pdb file of 1JUD, in which important functional residues were mutated by Quanta (MSI, San Diego, CA) to the corresponding residues of the sequence of the Ca2+-ATPase (shown in red in b). The backbone of the Rossmann fold is color-coded from blue at the N-terminus to red at the C-terminus. The inserted four-helix subdomain is gray. Catalytic residues are colored as follows: Asp (351, 601, 703, 707) are yellow, Lys (352, 684) are green, Arg (678) is purple, the 1-loop (KTGTL356) is cyan, and VO3 is white. (b) The topology diagram shows the positions of three major inserts by Ca2+-ATPase into the sequence of 1JUD (dotted lines, with the number of inserted residues in brackets). The largest insert replaces 4 in the A-domain and includes sites of TNP-ATP and FITC labeling (K492 and K515). The two short strands that link the domains form a hinge and have been redesignated as h1 and h2, and the strands of the -sheet have been renumbered from 1 to 6. Residue numbers correspond to CaATPase.

We have expanded the original alignment of these four sequence motifs to encompass the entire sequence between M4 and M6. Fig. 2 compares the sequence from 1JUD with four representative P-type ATPases from three main subfamilies (Axelsen and Palmgren, 1998): the type I heavy metal ATPases (CadA), the type III plant and fungal H⁺-ATPases (H-ATP), and the type II animal Na⁺/K⁺- and Ca²⁺-ATPases (NaATP, CaATP). The sequences homologous to the dehalogenase core domain are referred to as the phosphorylation or P-domain and the inserted subdomain (140-240 residues in the ATPases) as the adenosine-binding or A-domain. The presence of this inserted subdomain in Ca²⁺-ATPase is strongly supported by the recent isolation and characterization of a proteolytic fragment with termini (T³⁵⁷-T⁶⁰⁸) close to those of the putative A-domain (Champeil et al., 1998). Furthermore, this proteolytic fragment was shown to retain tertiary structure, to bind nucleotides, and to react specifically with fluorescein isothiocyanate (FITC), supporting its proposed functional role in binding nucleotide. The independently predicted secondary structures of the P-domains are consistent both with each other and with the crystal structure of 1JUD. The only significant disparities are a variable insert following 6 and some differences in the loop between 5 and 6 (which is the site of a two-helix insert in the dehalogenase from Xanthobacter autrophicus; Ridder et al., 1997). In view of the low overall sequence identity (8-14%), further support was obtained from the structure-based GenTHREADER program, which picked 1JUD from the structure database as the only certain match (p = 1) to any of the ATPases in Fig. 2 (N. M. Green and J. Saldanha, unpublished observations). All together, these results provide firm ground for fitting the structure of the 1JUD core domain into the 8-Å density map of Ca²⁺-ATPase, which we describe below.

View larger version (66K): [in this window] [in a new window]   FIGURE 2   Alignment of the phosphorylation (P) and adenosine binding (A) domains of the ATPases with 1JUD. Four sequences of ATPases (SWISSPROT codes cada_staau, pma1_neucr, atn1_sheep, atcb_rabit) are aligned together with secondary structures predicted independently for three families by PHD (Rost and Sander, 1993). The predicted structure was used to supplement the sequence motifs and to guide the alignment to 1JUD. The A-domain has been divided at a conserved intron site (dashed line) into an N-terminal A1 region, with some similarity to 1JUD, and a C-terminal A2 region containing adenosine-binding sites, but with no resemblance to 1JUD. The intron sites for CadA and H-ATP were deduced from homologous positions of the Menkes Cu2+-ATPase (Tumer et al., 1995) and the Nicotiana H+-ATPase (Perez et al., 1992) genes. Sites of chemical labeling (1-9), mutation (a-j), proteolysis () and Ca2+ binding (++) are shown for the Ca2+-ATPase. Many of these sites are also relevant to Na+/K+-ATPase, but three unique sitesfor Na+/K+-ATPase are designated with primes (4', 5', and 6'). These sites are characterized in more detail in Tables 1 and 2. Finally, residues directly involved in the catalysis of 1JUD are indicated by x's below their sequence.

Given the dissimilarity of their substrates, it is not surprising that the subdomain of 1JUD differs from the A-domains of the pumps, which are all much larger. Nevertheless, near the N-terminus of the A-domain in Ca²⁺-ATPase there are three predicted helices that could parallel those of 1JUD, one of which (3) includes a notable sequence similarity (EATETAL/QATEDAL). In 1JUD these three helices are reconnected to the core domain by a fourth helix, whereas in the pumps, this connection expands to form a region of mixed secondary structure, including sites labeled by FITC and by azido-adenosine nucleotides (Figs. 1 b and 2). These differences in homology and secondary structure as well as the existence of a conserved intron site (dashed line in Fig 2; McIntosh, 1998) suggest that the A-domain may be subdivided into two parts (A1 and A2).

The proposed functional distinction between the P- and A-domains is supported by the effects of site-directed mutagenesis and chemical modification of Ca²⁺-ATPase and Na⁺/K⁺-ATPase (Tables 1 and 2). ATP analogs with reactive groups on the ring invariably label residues in the A-domain (sites 4, 5, 5', 6); as a result, these analogs block ATP-activated phosphorylation and transport, but not hydrolysis of acetyl phosphate or phosphorylation by P_i. In contrast, reactive groups near the -phosphate label residues in the P-domain (sites 6', j, k); the modified ATPases are completely inactive and phosphorylation either by ATP or by P_i is blocked. Similarly, mutations of residues in the A-domain (sites 5, 5', and 6) have much smaller effects on activity than do those in the P-domain (sites a-k), and the mutations at site 5 are the only ones with specific effects on ATP binding (McIntosh et al., 1996).

Domain boundaries in the density map

These considerations suggest that the cytoplasmic part of Ca²⁺-ATPase is made up of three domains: the A- and P-domains for the loop between M4 and M5 and the "-domain" for the loop between M2 and M3, which is predicted to be predominantly -strand (Green and Stokes, 1992). We have therefore looked for these three domains in the cytoplasmic portion of our Ca²⁺-ATPase density map (Fig. 3). The transmembrane and stalk regions had previously been fitted with -helices (Zhang et al., 1998), and we therefore considered only the densities between the top of the stalk and the top of the molecule. In particular, we examined the map for low-density regions that might represent boundaries between domains, both in contour plots of map sections (Fig. 3, d-f) and in surface representations of the molecule (Fig. 3, a-c). The molecular surface depends on the density level that is chosen, and the figure shows the cytoplasmic region at a relatively high density threshold, with three domains (colored yellow, orange, and blue). The boundaries of the yellow domain were the best defined and delineate a block-like domain that sits on top of the stalk. On the other hand, the boundary between orange and blue domains was unclear, and we have left the corresponding part of the surface uncolored in this region. Contour plots correspond to cross sections through a dimer, with domains outlined on one of the two protomers.

View larger version (65K): [in this window] [in a new window]   FIGURE 3   Domains in the cytoplasmic region of Ca2+-ATPase. (a-c) Surface representation of the density map in which the three domains composing the cytoplasmic head have been contoured at a higher density and colored. Yellow corresponds to the P-domain, orange to the A-domain, blue to the -domain, and purple to the putative decavanadate site; the green line indicates a possible delineation between the two parts of the A-domain. Domain boundaries were difficult to discern at the top of the molecule, which has therefore been left uncolored. The molecule is shown in three orientations related by ~90° rotations about the vertical (z) axis. The three zones indicated at the right margins correspond to the sections shown in d-f. The white arrows on the left correspond to the levels shown in Fig. 4, a and c. (d-f) Sections through two molecules related by a twofold axis, showing the low density boundaries between the three cytoplasmic domains. The colored contours represent a single molecule masked from the surrounding crystal lattice and correspond to four superimposed 2-Å-thick sections. Green, dashed contours are lowest, black are intermediate, and magenta are the highest density; thus, domain boundaries are likely to exist in regions filled with green contour lines. Twofold related molecules are represented by a single 2-Å section with only black contours, and the domains have been outlined in colors corresponding to those in a-c. The single arrow in d and # in a show the cavity occupied by Cr-ATP (Yonekura et al., 1997). The double arrowhead in e indicates a deep, curved cavity that is postulated as a potential second nucleotide binding site. Densities corresponding to two proposed decavanadate binding sites are indicated in f by the asterisks and by the triple arrowhead; the latter is colored magenta in a and c.

According to the homology with dehalogenases, the P-domain composing the Rossman fold begins and ends only a few residues from the ends of stalk helices S4 and S5, respectively, so it was logical to assign it to the yellow domain in Fig. 3. Indeed, the volume of this yellow domain corresponds closely to the expected mass of the P-domain (Table 3), leading us to fit the atomic coordinates for 1JUD (see next section). The nucleotide is expected to bind at the interface between the P-domain and the A-domain, which is consistent with the previously defined site of CrATP binding (Yonekura et al., 1997). Although the 14-Å resolution of the CrATP difference map was insufficient to define boundaries, the CrATP difference density lies on the boundary between the orange and yellow domains (# in Fig. 3 a and arrow in Fig. 3 d). This suggests that the orange domain corresponds to the A-domain and that the blue domain corresponds to the -domain

Two very high densities are apparent in Fig. 3 f, which we believe are good candidates for the decavanadate used to induce crystallization. One of these is present on the dyad axis (* in Fig. 3 f), and, given its position on a symmetry axis, it is unlikely to represent protein. The peak is ~10 Å in diameter and is therefore the right size and shape for a decavanadate ion (Day et al., 1987). Accordingly, this peak was omitted from the molecule in previous publications (Zhang et al., 1998; Toyoshima et al., 1993). Further examination of Fig. 3 f reveals a second peak of high density within the A-domain, which is much higher than that from an -helix (magenta in Fig. 3, a and c; triple arrowhead in Fig. 3 f). Again, this peak is ~10 Å in diameter and may represent a second, internal site for decavanadate. If so, it would imply a total of 1.5 moles of decavanadate per mole of ATPase. Previous measurement of decavanadate binding (Csermely et al., 1985) determined a stoichiometry of 1.5-2 moles of decavanadate per mole of Ca²⁺-ATPase, which is consistent with our model.

The assignment of this high-density peak to decavanadate leaves only a narrow neck of density (green line, Fig. 3, a, c, f) between the upper and lower parts of the A-domain. These could correspond to the division of the corresponding sequence into A1 and A2 subdomains (see above), the upper region (A1) adjoining the hinge region of the P-domain and the lower region (A2) lining the cavities that bind the nucleotide. Indeed, A2 contains all of the functional sites in the A-domain, and its calculated molecular weight is consistent with the size of this part of the map (Table 3). It is also noteworthy that the internal decavanadate site contacts both parts of the A-domain as well as the P- and -domains and so is in an ideal position to stabilize their interactions. This stabilization could account for the compact structure of this E₂ form of the pump compared to the E₁Ca₂ form (Stokes et al., 1999).

Docking of dehalogenase core domain into Ca²⁺-ATPase

Our next step was to dock the atomic structure of the dehalogenase core domain into the yellow density assigned to the Ca²⁺-ATPase P-domain (Figs. 4 and 5). We tried two different automated methods for docking, but neither produced convincing or consistent results. Docking was therefore done manually, and the general orientation was chosen to be consistent with short connections between the stalk helices S4 and S5 and the ends of the dehalogenase fold, so that the -strands run away from the membrane. The pattern of density within the P-domain was then used for a more precise positioning of the dehalogenase fold. Generally speaking, -helices generate relatively strong density peaks at 8 Å resolution, whereas -strands remain undefined. Thus we positioned 5 and 9 in a plausible region of high density along the back of the P-domain, which also matched 6 to a high-density rod that protrudes from the side of the P-domain. In addition, a low-density region in the P-domain was matched with a low-density region in maps generated from the dehalogenase coordinates between 5, 9, and the -sheet (Figs. 1 a and 6). There were, however, several discrepancies remaining between the two structures, which are not surprising given the large differences in sequence and the probability that extensions of stalk helices S2 and S3 run through this region and thus contribute extra density. In particular, two regions of 1JUD project beyond the yellow envelope of the P-domain. One is the L₅₆ loop (Fig. 4 a), which is variable among dehalogenases and is predicted to be slightly shorter and -helical for Ca²⁺-ATPase (we have remodeled this loop in Fig. 6 to minimize this protrusion). The other is the loop between 3 and 7 (Fig. 4 c), which is actually a short 3₁₀ helix in the dehalogenase and is two residues shorter in Ca²⁺-ATPase. On the other hand, there are unassigned densities in the P-domain, the largest being at the bottom of Fig. 4 c. This particular density is next to the loop between 6 and 3, which consists of a short 3₁₀ helix in the dehalogenase but has an extra 28 residues in Ca²⁺-ATPase; these extra residues could plausibly fill this unassigned space. We can find no well-resolved connecting density between the -domain and stalk helices, although chains could be accommodated by adjustments of 8 and L₅₆, which occupy a relatively large, high-density region at the interface with the A-domain (Fig. 3 d). Alternatively, the -domain terminates as a small column of density that resembles a helix (Fig. 3 d), and it is possible that it is connected to the stalk by a disordered, and therefore invisible, loop.

View larger version (28K): [in this window] [in a new window]   FIGURE 4   Fitting the dehalogenase core domain to the P-domain density from Ca2+-ATPase. The density envelopes from Ca2+-ATPase and dehalogenase are shown in yellow and green, respectively. a is from the top of the P-domain (topmost arrow in Fig. 3 a), whereas c is from the bottom (lower arrow in Fig. 3 a). The ribbon diagram for the entire dehalogenase core domain is shown in b, with the same coloring scheme as in Fig. 1. Axes are the same as in Fig. 3.

View larger version (98K): [in this window] [in a new window]   FIGURE 5   New model for the architecture of Ca2+-ATPase, showing transmembrane and stalk helices as well as the P-domain. Transmembrane and stalk domains are as fitted by Zhang et al. (1998) with hypothetical connecting loops. The hinge between the P-domain and A-domain is indicated by h, and the putative decavanadate peak is indicated by V. The N- and C-termini of the P-domain are indicated in b, as is the location of the catalytic D351 (star). Axes are the same as in Figs. 3 and 4.

View larger version (39K): [in this window] [in a new window]   FIGURE 6   Catalytic region of Ca2+-ATPase superimposed on a section from the density map. The section is at the same orientation and position as those of Fig. 3 d. For this figure, the pdb file of 1JUD was mutated in Quanta, replacing functional residues in the loops at the ends of 1, 2, 3, and 4 by the corresponding residues of the Ca2+-ATPase (as for Fig. 1). In addition, the L56 loop was shortened and remodeled as two turns of helix, based on the predicted secondary structure of the corresponding Ca2+-ATPase sequence. This remodeling would reduce the projection of this loop outside the density envelope of the ATPase as seen in Fig. 4 a. ADP-Mg-VO3 was imported from the structure of myosin (1VOM; Smith and Rayment, 1996) and oriented between D351 and site a of the density map. Energy minimization of the whole structure was performed using the CHARMm option of Quanta. Only the remodeling of L56 required changes in backbone configuration; other changes were restricted to the side chains. The functionally important residues are shown in full, as is R678, which can be cross-linked by glutaraldehyde to the A-domain. The figure was prepared using the program MOLSCRIPT (Kraulis, 1991).

	DISCUSSION

TOP ABSTRACT INTRODUCTION RESULTS DISCUSSION REFERENCES

Modeling of the catalytic site

To evaluate the consistency of this model with the results of mutagenesis and chemical modification, a particular binding configuration for ATP must be specified. Although the active site for nucleophilic attack on the -phosphate is defined by the dehalogenase structure, the binding of the rest of the ATP molecule is not, both because it is substantially different from dehalogenase ligands and because the contribution of the subdomain to this binding pocket is currently undefined. Nevertheless, according to our model, the binding pocket for ATP should occur at the interface between the A-domain and the P-domain, and two low-density cavities are seen at this interface (labeled a and b in Fig. 6, single and double arrowheads in Fig. 3, d and e). These two cavities ultimately merge and provide plausible binding pockets for either one or possibly two adenosine ring systems. They are both close enough to the catalytic aspartate (D³⁵¹) to permit interaction with the -phosphate; however, the location of the difference density produced by CrATP is consistent with the position of cavity a, even though the two cavities could not be resolved in the corresponding maps because of their lower resolution (Yonekura et al., 1997). Cavity b might contribute to a second ATP binding site, for which there is considerable evidence in the literature (see below).

The -phosphate itself, which is modeled as a transition state VO₃, occupies an active site at the topological switch point between -strands 1 and 4; such a switch point is generally favorable for an active site in /-type structures (Branden, 1980) and specifically matches the active site of the dehalogenases (Hisano et al., 1996; Ridder et al., 1997). The side chains shown in Fig. 6 correspond to mutation-sensitive residues in Table 2, which have been modeled into the dehalogenase fold. The location of all of these residues at the ligand binding surface, mostly in close juxtaposition with the -phosphate, is strongly supportive of our model. In particular, the -phosphate is surrounded by four mutationally sensitive residues (D³⁵¹, D⁷⁰³, D⁷⁰⁷, K⁶⁸⁴), whose homologs in the dehalogenase are directly involved in nucleophilic attack on the substrate. The signature sequence for P-type pumps (DKTGTL³⁵⁶) is represented by the cyan loop, following 1_, which links D³⁵¹ to the hinge and which is stabilized by a continuous network of H-bonds in the dehalogenase structure. The loop following 2 contains D⁶²⁷, which is more distant, but could indirectly affect the disposition of the preceding several residues (T⁶²⁵ and G⁶²⁶), which also contribute ligands to the hydrogen-bonding system at the active site of dehalogenases; indeed, these preceding residues constitute the fourth sequence motif that was used to define the superfamily (Aravind et al., 1998). The final residue shown is R⁶⁷⁸ at the C-terminus of 3, whose orientation in Fig. 6 was determined by energy minimization. This residue is located on the surface of the P-domain, pointing toward the A-domain, consistent with its observed cross-linking by glutaraldehyde to K⁴⁹² in the A-domain. The ability of decavanadate to block this cross-link is evidence for the effect of this reagent on domain interactions.

Chemical modification and evidence for two nucleotide binding sites

The binding of nucleotides and nucleotide analogs has been the subject of many types of experiment, and the complexity of the results has given rise to a variety of different explanations. The basic observation is that the high-affinity catalytic site binds ATP with a K_d of ~5 µM to give a phosphoenzyme whose subsequent transformations are further activated by ATP binding at sites of medium (50 µM) and low (1 mM) affinity (e.g., Moller et al., 1980). In other experiments the binding constants for inhibitors or for protection by ATP against irreversible inhibitors have been measured. The simplest explanation (Champeil et al., 1988; Moczydlowski and Fortes, 1981) postulates a single nucleotide site that changes in affinity after phosphorylation of D³⁵¹, giving rise to low-affinity modulations. Others assume separate catalytic and regulatory sites, which may be on the same peptide chain (Ward and Cavieres, 1998a) or on opposing chains of a dimeric unit (Linnertz et al., 1998).

Our observation of two cavities at the domain interface is consistent with the existence of two nucleotide sites within a single molecule (a and b in Fig. 6). Because --CrATP occupies site a and has the same stereochemistry as MgATP (as found in the active sites of myosin and F₁ ATPase), we assume that this is the catalytic ATP site. Site b is near K⁶⁷⁸ and extends upward out of the section in Fig. 6. As mentioned, K⁶⁷⁸ can be cross-linked to K⁴⁹² by glutaraldehyde, which in turn can be cross-linked to K⁵¹⁵ by 4,4'-diisothiocyanostilbene-2,2'-disulfonic acid, suggesting that these three residues are close together. Association with a nucleotide binding site is suggested by the fact that K⁴⁹² is modified by a number of nucleotide analogs and that FITC labeling of K⁵¹⁵ blocks ATP binding (Table 1). Thus we propose that the well-studied residues K⁴⁹² and K⁵¹⁵ are associated with site b and form a secondary nucleotide binding site.

In fact, there is ample evidence from various ATP analogs that suggest the existence of two nucleotide binding sites that can be simultaneously occupied. For example, although labeling of K⁴⁹² or K⁵¹⁵ generally blocks phosphorylation by ATP, D³⁵¹ can still react with smaller substrates such as P_i, acetyl phosphate, or p-nitrophenyl phosphate (and even, in the Na⁺/K⁺-ATPase, with O-methyl fluorescein phosphate). In addition, nucleotides and their analogs are still able to bind to the labeled enzyme with moderate affinity and inhibit its reactivity with these small substrates, supporting the existence of a site distinct from K⁴⁹² and K⁵¹⁵ (Champeil et al., 1988; Mignaco et al., 1996; Scheiner-Bobis et al., 1993; Ward and Cavieres, 1998b). One explanation for the complexity of these effects is that reagents bound at the catalytic site a produce a conformational change that closes down the secondary site b, whereas reagents bound at site b have mainly local effects. This explanation is supported by the general observation that the reactivity of K⁴⁹² and K⁵¹⁵ is blocked when the catalytic site a is occupied. Furthermore, the reactivity of 11 of 16 cytoplasmic thiol groups is reduced after ATP binding to the catalytic site (Thorley-Lawson and Green, 1977), but reaction of FITC at the secondary site does not protect thiol groups to any significant extent (N. M. Green, unpublished observations) and does not prevent phosphorylation of D³⁵¹ with P_i or with p-nitrophenyl phosphate. Assuming that ATP binds preferentially to the catalytic site, this could explain why direct binding measurements have failed to support the existence of two sites. An inconsistency exists in the large effects of mutations at F⁴⁸⁷ and K⁴⁹² on binding of ATP to the catalytic site (McIntosh et al., 1996), but such inconsistencies could be reconciled by postulating ligand-induced domain movements, an idea that is supported by preliminary observations of the Ca²⁺-bound, E₁ state, discussed below. Moreover, it should be remembered that we do not yet have a 3D structure for the E₁ form and that our structure for the E₂ form includes both decavanadate and thapsigargin, which may influence the relations between the domains.

The multiplicity of binding sites is further extended by our observation of two sites for decavanadate distinct from the CrATP site, both of which remain occupied in the helical arrays containing CrATP (Yonekura et al., 1997). At the same time there is evidence for competition between decavanadate and ATP (Coan et al., 1986), although the data did not allow quantitative conclusions about the dissociation constants. The competition was not caused by orthovanadate, because the latter does not prevent firm binding of ATP (Andersen and Moller, 1985). This implies allosteric interaction between the decavanadate sites and an ATP site, which is also suggested by a decreased decavanadate binding by the FITC derivative of the ATPase (Csermely et al., 1985).

Mechanistic implications of the P-domain structure

This model also provides potential insights into the coupling between sites of phosphorylation and Ca²⁺ transport. Some earlier proposals have emphasized effects transmitted by the direct link between the Ca²⁺ binding site in M4 and the phosphorylation site, while others have considered transport to be the result of more global changes (reviewed by McIntosh, 1998). In our proposed structure, the phosphorylated D³⁵¹ is closely linked to the hinge segment h1 (TTN³⁵⁹) by the short, fully conserved sequence KTGTL³⁵⁶. In the dehalogenase, this connection forms a five-residue loop, which is stabilized by two hydrogen bonds between i and i + 5 positions. Also according to this analogy, another mutation-sensitive sequence, DPP⁶⁰³, composes the return half of the hinge (segment h2) and is linked to TTN³⁵⁹ by two H-bonds. We suggest that this close linking between the KTGTL³⁵⁶ loop and the hinge allows changes at the catalytic site to initiate domain movements.

In particular, there is evidence that the bonding of Mg²⁺ changes markedly during the reaction cycle. Previous studies have measured a fall in the rate constant for Mg²⁺ release from 80 s¹ in the E·Mg·ATP complex (Reinstein and Jencks, 1993) to less than 0.5 s¹ in E₁~P·Mg(Ca²⁺)₂ (Wakabayashi and Shigekawa, 1984). This dramatic increase in Mg²⁺ affinity after phosphate transfer and dissociation of ADP suggests that loss of phosphate coordination has been compensated for by new protein ligands with Mg²⁺. Similar large changes in the Mg²⁺ off-rate accompany phosphorylation from P_i (Ogurusu et al., 1991), and analogous behavior is observed in Na⁺/K⁺-ATPase, using Co²⁺ as a substitute for Mg²⁺ (Richards, 1988). So far these results have not resulted in any detailed mechanistic proposals, because of lack of a good structural model. However, the H-bond network between the MgATP site and the hinge in our model provides a basis for Mg²⁺-bonding to initiate movements of the P- and A-domains. Such movements could ultimately be transmitted via the stalk to the Ca²⁺ sites within the membrane.

There are several plausible candidates for Mg²⁺ ligands among the mutation-sensitive sites of Ca²⁺-ATPase, including T³⁵³ and T³⁵⁵ of the catalytic loop, D⁶²⁷, D⁷⁰³, and D⁷⁰⁷. A recent structural comparison (Ridder and Dijkstra, 1999) between the catalytic sites of CheY and 1JUD, has suggested D³⁵¹, D⁷⁰³, and D⁷⁰⁷ as Mg²⁺ ligands. In the related FixJ, the conformational change that accompanies formation of the aspartyl phosphate involved reorientation of T⁶²⁵ to bind phosphate together with a 6 Å movement of the histidine corresponding to D⁶²⁷ towards the KTGTL³⁵⁶ loop (Birck et al., 1999). In Ca²⁺-ATPase, analogous changes are bound to influence Mg²⁺ bindings as well as the configuration of the hinge region.

The structure of Ca²⁺-ATPase used for our fitting is likely to represent the E₂ conformation of the enzyme, because crystallization is prevented by Ca²⁺ and promoted by thapsigargin (Stokes and Lacapere, 1994). The binding of Ca²⁺ to the transport site initiates a major structural transition from the E₂ to the E₁ conformation, which activates the catalytic ATP site and leads to nucleophilic attack of D³⁵¹ on ATP. Comparison of the fitted 3D structure with projection images of Ca²⁺-ATPase in the presence of saturating calcium concentrations (i.e., E₁·Ca₂; Cheong et al., 1996; Ogawa et al., 1998) and with the structure of H⁺-ATPase in the E₁ conformation (Kühlbrandt et al., 1998; Stokes et al., 1999) reveals a substantial rearrangement of the cytoplasmic nose. According to our model, this rearrangement can be explained by a movement of the A-domain relative to the P-domain. In particular, the interface that contains the putative ATP-binding site appears to open up and produce a gap between the domains, perhaps promoting the binding of nucleotide. The two hinge segments between P- and A-domains could provide the necessary flexibility for such a rearrangement. Indeed, analogous domain movements have been shown to modulate the accessibility of the ligand binding site of various structurally related enzymes, such as the family of phosphoribosyl transferases (Smith, 1999). Clearly, revealing the nature of this conformational change is an important step in understanding the structural basis for active transport.