Conformational changes in four regions of the Escherichia coli ArsA ATPase link ATP hydrolysis to ion translocation.

Structures of ArsA with ATP, AMP-PNP, or ADP.AlF(3) bound at the A2 nucleotide binding site were determined. Binding of different nucleotides modifies the coordination sphere of Mg(2+). In particular, the changes elicited by ADP.AlF(3) provide insights into the mechanism of ATP hydrolysis. In-line attack by water onto the gamma-phosphate of ATP would be followed first by formation of a trigonal intermediate and then by breaking of the scissile bond between the beta- and gamma-phosphates. Motions of amino acid side chains at the A2 nucleotide binding site during ATP binding and hydrolysis propagate at a distance, producing conformational changes in four different regions of the protein corresponding to helices H4-H5, helices H9-H10, helices H13-H15, and to the S1-H2-S2 region. These elements are extensions of, respectively, the Switch I and Switch II regions, the A-loop (a small loop near the nucleotide adenine moiety), and the P-loop. Based on the observed conformational changes, it is proposed that ArsA functions as a reciprocating engine that hydrolyzes 2 mol of ATP per each cycle of ion translocation across the membrane.

In Escherichia coli resistance to the metalloids arsenic and antimony is conferred by the ars operon of plasmid R773 (1). The arsA and arsB genes of the operon encode, respectively, the catalytic subunit ArsA 1 (ATPase) and the membrane subunit ArsB of a pump that extrudes arsenite (As(III)) and antimonite (Sb(III)) ions from the cytosol (2).
Arsenic efflux in bacteria is catalyzed by either ArsB alone, functioning as a secondary transporter, or by the ArsAB complex, functioning as a transport ATPase (3). E. coli can utilize either mode physiologically; however, the ATP-coupled pump is more efficient, capable of producing concentration gradients as high as 10 6 , equivalent to a concentration of 1 nM intracellular arsenite at 1 mM external arsenite.
Although bound to ArsB in vivo, ArsA can be expressed and purified as a soluble protein (4) whose ATPase activity is stimulated by As(III) or Sb(III) (5). ArsA is composed of two homologous domains, designated A1 and A2, connected by a linker of 23 amino acids; each domain contains a consensus sequence for a nucleotide binding site (NBS) (see Fig. 1A).
We have recently determined the crystal structure of the enzyme in complex with Mg⅐ADP (6). The A1 and A2 halves of the protein are related by a pseudo-2-fold axis of symmetry. The two NBSs are located at the interface between A1 and A2, in close proximity of each other. Both NBSs are formed by residues from both A1 and A2. However, one NBS is contributed mostly by A1 residues and is thereby named A1 NBS; the other NBS is contributed mostly by A2 residues and is named A2 NBS.
Also at the interface between A1 and A2, but at the opposite end of the molecule with respect to the NBSs, is a site in which three distinct As(III) or Sb(III) ions bind (6). Three cysteines (Cys 113 , Cys 172 , Cys 422 ), two histidines (His 148 , His 453 ), and one serine (Ser 420 ) ligate the ions. Each As/Sb(III) is coordinated by one residue from A1 and one residue from A2. Thus, binding of each of the three metalloids tightens the interaction between A1 and A2, possibly triggering ATP hydrolysis.
One of the major unanswered questions in the ArsA mechanism is how the two NBSs work together to provide the energy necessary for ion transfer. It is clear that both NBSs bind nucleotides (7). Furthermore, pre-steady-state analyses of ATP hydrolysis by ArsA in the absence or presence of arsenite or antimonite show that both NBSs are catalytic, although not equivalent (8,9). These kinetic data are consistent with binding studies using the ATP analog 5Ј-p-fluorosulfonylbenzoyladenosine, which suggest that nucleotides can be easily exchanged at the A2 NBS but not at the A1 NBS (10). A possible basis for these observations is found in the structure of ArsA in complex with Mg⅐ADP, which shows the A1 NBS in a "closed" conformation, while the A2 NBS is in an "open" conformation (6).
ArsA is structurally similar to NifH, the Fe-protein of bacterial nitrogenases (11; see also sequence alignment in Fig. 1A below). Like ArsA, NifH has two NBSs facing each other and its iron-sulfur center is almost coincident with the As/Sb(III) cluster of ArsA. Because NifH has long been recognized as a relative of G-proteins (12)(13)(14), it has become customary to identify specific regions of ArsA with definitions borrowed from Gprotein terminology. For example, in G-proteins, as well as in ArsA, Mg 2ϩ is coordinated (directly or via a water molecule) by an aspartic acid located in a strand-loop-helix structure that is referred to as the Switch I region (15). Likewise, the DTAPTGH signature sequence of ArsA (6,16,17) has an exact counterpart in NifH (Fig. 1A) and is believed to correspond to the Switch II region of G-proteins (14). The terms Switch I and Switch II are commonly used in reference to these regions throughout this report.
To elucidate the structural basis of ATP hydrolysis at the NBSs of ArsA, we have determined the crystal structure of the enzyme in complex with ATP, the non-hydrolyzable ATP analog AMP-PNP, and the transition state analog of ATP hydrolysis, ADP⅐AlF 3 . The results presented here in conjunction with previous studies of the pre-steady-state kinetics of ATP binding and hydrolysis (8,9) suggest that the enzyme may function as a reciprocating engine.

MATERIALS AND METHODS
Crystals of ArsA in complex with Mg⅐ADP were prepared by vapor diffusion in hanging drops at 30°C (18). Drops were prepared by mixing equal amounts of a solution containing 20 mg/ml ArsA protein, 2 mM ADP, 1 mM MgCl 2 , 2 mM NaAsO 2 , 1 mM CdCl 2 , 10 mM Bis-Tris-propane (pH 8.0) with the reservoir solution containing 4% (w/v) polyethylene glycol 3000, 100 mM Bis-Tris-Propane (pH 8.0). Crystals were harvested from the crystallization tray in a holding solution consisting of 20% (w/v) polyethylene glycol 3000, 5 mM ADP, 2.5 mM MgCl 2 , 4 mM NaAsO 2 , 2 mM CdCl 2 , 100 mM Bis-Tris-propane (pH 8.0), and 15% glycerol (v/v) as cryoprotectant. Crystals of ArsA in complex with Mg⅐ATP were initially obtained by soaking crystals of the Mg⅐ADP enzyme in the presence of 5 mM ATP and 2.5 mM MgCl 2 . Under these conditions the crystals undergo a space group change from I222 (a Х 73 The new unit cell is only apparently different from the original as shown by the observation that the choice of the unconventional space group P22 1 2 1 instead of P2 1 2 1 2 would be associated with essentially unchanged unit cell dimensions (a Х 74 Å, b Х 77 Å, c Х 222 Å) with respect to the original I222 space group. The asymmetric unit of P2 1 2 1 2 crystals contains two molecules related by a 2-fold axis of local symmetry. De novo crystallization of the enzyme, using ATP instead of ADP, also yielded crystals of space group P2 1 2 1 2, which were indistinguishable from those obtained by soaking I222 crystals in the presence of MgCl 2 and ATP. Crystals of the enzyme with AMP-PNP also belong to space group P2 1 2 1 2 and could be obtained by incubating crystals of the ADP complex with AMP-PNP but not by de novo crystallization in the presence of this ATP analog. Crystals of the ADP⅐AlF 3 enzyme were obtained using the same conditions that yield ADP crystals, except for the addition of 8 mM NaF and 2 mM AlCl 3 . Crystals of the ADP⅐AlF 4 enzyme were obtained by incubating crystals of the ADP⅐AlF 3 enzyme in a mother liquor at pH 6.2 containing MES as buffer instead of Bis-Tris-propane.
Data sets were collected at 100 K with a R axis IV image plate detector (Table I) and processed with HKL (19). The structure of ArsA in complex with Mg⅐ATP was determined by molecular replacement using the enzyme from space group I222 (6) as the starting model. All the steps of the molecular replacement procedure were carried out using the CNS v. 1.0 suite of crystallographic programs (20). Selfrotation and cross-rotation function analyses clearly identified the presence of two molecules in the asymmetric unit, related by a 2-fold axis of local symmetry. Correct placement of the molecules was obtained with two consecutive translation searches. Model refinement was also carried out with CNS v. 1.0 using cross-validated maximum likelihood as the target function (21). Solvent molecules were added during the final stages of refinement after the protein model had stabilized. Because crystals of the ADP⅐AlF 3 enzyme are isomorphous to crystals of the Mg⅐ADP enzyme, the structure of the former was obtained by simple crystallographic refinement.
Difference distance matrices were computed with CNS v. 1.0 and displayed with Mathematica (Wolfram Research). Least-square fits were carried out with LSQMAN written by Gerard Kleywegt.

Structure of the Enzyme in Complex with ATP-When
ArsA is crystallized in the presence of ATP, or when crystals containing Mg⅐ADP are soaked with ATP, the resulting crystals belong to space group P2 1 2 1 2 and contain two molecules in the asymmetric unit (Table I). Several choices of a dimer are consistent with crystal symmetry: The dimer with the largest buried surface is shown in Fig. 2. It should be emphasized that, although there is some suggestion that ArsA may be a dimer in vivo (22), the crystallographic dimer shown in Fig. 2 may not reflect the actual structure of the enzyme when it is part of the ArsAB complex.
In both monomers, the A1 NBS is occupied by Mg⅐ADP, while the A2 NBS is occupied by Mg⅐ATP. The local 2-fold axis that relates the monomers corresponds to a crystallographic 2-fold axis in the crystal of the I222 space group described previously (6), which contains ADP at both the A1 and A2 NBSs.
A superposition of the A2 NBS with bound ADP (I222 crystals) and of the same site with bound ATP (P2 1 2 1 2 crystals, Molecule B) is shown in Fig. 3. The most striking difference is observed in the region corresponding to residues 567-573. Throughout this report we adopt the convention that corresponding structural elements of A1 and A2 are referred to by a single denomination as helix H# or strand S# of A1 or A2, respectively ( Fig. 1). Thus, for example, residues 567-573 form a small loop between strand S8 and helix H16 of A2, which wraps around the adenine moiety of the nucleotide. The equivalent loop between strand S8 and helix H16 of A1 encompasses residues 277-283 (see sequence alignment of Fig. 1A). We refer FIG. 1. ArsA amino acid sequence and secondary structure. A, structure-based sequence alignment of ArsA A1 and A2 and of NifH, the iron protein of Azotobacter vinelandii nitrogenase. A1 and A2 sequences are in aquamarine; NifH sequence is in blue. Secondary structure and A1 numbering are above the alignment, A2 numbering is below. Strands, helices, and loops are drawn as arrows, quadrilaterals, and thick lines, respectively. Disordered regions are represented as dashed lines. P-loop, green; Switch I, pink; Switch II, cyan; A-loop, red; ligands of As/Sb(III) in ArsA or of the [2Fe-2S] cluster in NifH, yellow; A1-A2 linker, gray. B, ribbon drawing of A1. Strands, purple; P-loop, green; Switch I, pink; Switch II, cyan; A-loop, red; region between H9 and H10, disordered when ADP is bound at the A2 NBS, gold. Images were generated with MOLSCRIPT (45) and RASTER3D (46). H, helix; S, strand. to this loop as the "adenine loop" or "A-loop." The A-loop was shown to form a photo-adduct with [␣-32 P]ATP (23). In the presence of ATP the A-loop is displaced producing a wider opening of the A2 NBS cavity (Fig. 3). The view presented in Fig. 3 clearly shows how Mg 2ϩ is located at the crossroad of the P-loop (green trace), the Switch I region (pink trace), and the Switch II region (cyan trace). When ADP is bound at the A2 NBS, Mg 2ϩ displays a distorted octahedral geometry with six ligands, two axial and four equatorial (Fig. 4A). The two axial ligands are a water molecule hydrogen bonded to a ␤-phosphate oxygen and the hydroxyl of Thr 341 . The four equatorial ligands are a ␤-phosphate oxygen, a water molecule that is also hydrogen-bonded to an ␣-phosphate oxygen and to the carboxylate of Asp 364 (Switch I), a water molecule that is also hydrogenbonded to the carboxylate of Asp 447 (Switch II) and to the hydroxyl of Ser 363 (Switch I), and a water molecule that is also hydrogen-bonded to the carbonyl of Asp 447 . By virtue of these interactions, the P-loop, the Switch I, and the Switch II elements are held tightly together around the Mg 2ϩ ion. When ATP binds at the A2 NBS, one of the ␥-phosphate oxygens replaces water as an axial ligand to Mg 2ϩ , such that the metal now bridges the ␤and ␥-phosphates (Fig. 4, B and C). How-ever, there are differences between the two ArsA monomers in the equatorial coordination of Mg 2ϩ . In one monomer (Molecule B) one of the water molecules that coordinates Mg 2ϩ is hydrogen-bonded to another more distant solvent, which in turn is hydrogen-bonded to Ser 363 and Asp 447 (Fig. 4B). In the other monomer (Molecule A) the equatorial coordination of Mg 2ϩ resembles more closely that observed in the presence of ADP, with a water ligand forming a direct hydrogen bond with Ser 363 and Asp 447 (Fig. 4C). In molecule B, Mg 2ϩ displays a perfect octahedral coordination (Fig. 4B), while in molecule A the coordination geometry is slightly distorted (Fig. 4C). In these various configurations of the A2 NBS with ADP or ATP bound, Mg 2ϩ appears to move back and forth among the nucleotide, the Switch I, and the Switch II elements. This motion is reflected by changes in the distance between Mg 2ϩ and nearby residues of the P-loop, the Switch I, and the Switch II region (Table II).
The Switch II region contains the signature sequence 447 DTAPTGH 453 , whose N-terminal end, Asp 447 , is involved in Mg 2ϩ coordination, and whose C-terminal end, His 453 , is one of the As/Sb(III) ligands (Fig. 1A). It has been proposed that this sequence may act as a signal transduction element connecting   2. Dimeric ArsA. A dimer of ArsA in the asymmetric unit of P2 1 2 1 2 crystals (red trace) is shown together with a leastsquare superposition of the equivalent dimer in I222 crystals (ivory trace). Least square fitting was carried out to maximize the superposition of only one of the molecules. The two ArsA molecules in I222 crystals are derived from application of a 2-fold axis of crystallographic symmetry and are therefore identical. In contrast, the two ArsA molecules in P2 1 2 1 2 crystals (here labeled as Molecule A and Molecule B) are similar (root mean square deviation of 0.63 Å for 546 C␣ values corresponding to the ordered regions of the protein) but not identical, which explains the space group difference. events at the NBS with changes in metal binding (6,17). Surprisingly, no significant changes in the conformation of the DTAPTGH element are observed between the ADP-and the ATP-complexed enzymes. The geometry of As/Sb(III) ligation at the metal binding site is also essentially unchanged in the two structures. In contrast, a large change is observed in the region of A1 that provides one of the cysteine ligands (Cys 172 ), represented by helices H9 -H10 and by the intervening loop. This loop is disordered in the ADP enzyme but becomes well ordered when ATP binds at the A2 NBS (Fig. 1B, gold trace).
Although all the crystals analyzed in this report were obtained by incubating the enzyme in the presence of As(III), difference Fourier maps show residual positive density at the positions of the bound metal, suggesting it should be modeled as an element heavier than arsenic. One possible interpretation of this observation is that Cd(II) ions, which may assume a coordination similar to that of As(III), and whose presence in the crystallization medium is necessary for crystal formation (see "Materials and Methods"), slowly replace As(III) at the metal binding site.
A structure essentially indistinguishable from that of ArsA in complex with ATP was observed when crystals obtained in the presence of ADP were incubated with the non-hydrolyzable ATP analog AMP-PNP. In this experiment AMP-PNP exchanged with ADP at the A2 NBS (data not shown). However, if AMP-PNP was added prior to crystallization, no crystals were obtained.
Structure of the Enzyme in Complex with ADP⅐AlF 3 -The structure of the enzyme in complex with ADP⅐AlF 3 is much more similar to that of the ADP complex than to that of the ATP complex. This is also evident in the fact that crystals of both the ADP and ADP⅐AlF 3 complex belong to the same space group I222 (Table I). However, as was the case for ATP and AMP-PNP, ADP⅐AlF 3 binds only at the A2 NBS. At this site the aluminate is clearly recognizable as a planar trigonal molecule occupying a position equivalent to that of the ␥-phosphate of ATP (Fig. 4D). Consequently, the hydroxyl moiety of the ␤-phosphate of ADP corresponding to the bridging oxygen between the ␤and ␥-phosphate of ATP is only 2.3 Å from aluminum. Mg 2ϩ is 1.9 Å from one of the fluorine atoms and bridges the latter to one of the ␤-phosphate hydroxyls. On the other side of the aluminum atom, a water molecule occupies a position almost symmetrical with respect to the ␤-phosphate hydroxyl (Fig. 4D). This water molecule is stabilized by a hydrogen bond to another solvent molecule, which in turn is hydrogen-bonded to the hydroxyl of Ser 363 (Switch I). There is no water near the ␥-phosphate in the structures of the enzyme with ATP bound at the A2 NBS.
As previously mentioned, the structure of the enzyme in complex with ADP⅐AlF 3 at the A2 NBS is very similar to that of the ADP complex. In particular, the position of the A-loop (residues 567-573) is similar to that observed in the ADP complex, and the loop between helices H9 and H10, which is disordered in the ADP enzyme but visible when ATP binds at the A2 NBS, is also disordered in the ADP⅐AlF 3 complex.
It is worth emphasizing that the ADP⅐AlFx complex described above is of the ADP⅐AlF 3 form rather than of the ADP⅐AlF 4 form observed in similar experiments with other enzymes (24 -26). Because the coordination number of the AlF x in transition state analogs of phosphoryl transfer is pH-dependent and the probability of the aluminate to be in the AlF 4 form increases with decreasing pH (27), we have sought to determine whether lowering the pH of the mother liquor of ArsA crystals would affect the aluminum coordination. In fact, when crystals of the ADP⅐AlF x complex are incubated at pH 6.2, four fluoride ions are present as square-plane ligands of the aluminum ion (not shown), suggesting that the aluminum coordination number has changed from 5 to 6 (including water and the ␤-phosphate hydroxyl as axial ligands). It is widely accepted that hydrolysis of ATP proceeds by attack of a water molecule on the ␥-phosphorus (28 -30) with formation of a trigonal bipyramidal transition state (31,32). Thus, the ADP⅐AlF 3 complex (trigonal coordination) observed at pH 8.0 represents a better structural analog of the transition state of ATP hydrolysis than the ADP⅐AlF 4 complex (square-planar coordination) observed at pH 6.2.
Global Conformational Changes Associated with the Binding of Different Nucleotides at the A2 NBS-A quantitative analysis of the differences between structures can be achieved through the use of difference distance matrices (33). In a distance matrix, the differences between the coordinates of each C␣ of a protein and every other C␣ of the same protein are utilized to build a density plot in which residues that are close in space are represented as a point of high density. In a difference distance matrix the distance matrix of a first structure is subtracted from the distance matrix of a second structure. If the distance A n Ϫ A m between two residues A n and A m of the A structure changes in the B structure, then the distance B n Ϫ B m between the same two residues in the B structure will be either Ͼ A n Ϫ A m or Ͻ A n Ϫ A m . In the first case a positive value will be associated with those two residues in the difference distance matrix between B and A; in the second case a negative value will be obtained. Thus, to evaluate whether changes have occurred between two structures, it is informative to look at both positive and negative values of the differ- Altogether, there are only a few regions of ArsA that undergo the same or very similar conformational changes in all pairwise comparisons in response to the binding of different nucleotides at the A2 NBS. The smallest differences are observed between the structures of ArsA in complex with ADP⅐AlF 3 versus ADP (Fig. 5, row 1), whereas the largest values of both positive and negative differences occur between the two molecules of ArsA with ATP bound (Fig. 5, row 4). It is important to notice that these differences cannot be interpreted as trivial consequence of the variation in crystal contacts between the two molecules of ArsA in the asymmetric unit. On the contrary, it is the addition of ATP to crystals of space group I222 that produces the change to space group P2 1 2 1 2 by forcing the establishment of new crystal contacts to accommodate the conformational changes in the enzyme. These conformational changes involve primarily (a) helices H4 -H5 (pink rhombus; residues 75-105 of A1 or 385-415 of A2), (b) helices H9 -H10 (cyan circle; residues 155-180 of A1), (c) helices H13-H15 and the A-loop (red triangle; residues 235-280 of A1 or 535-570 of A2), (d) a loop and strand located at the end of the Switch I region of A1 (yellow square; residues 55-70 of A1), (e) the region S1-H2-S2 encompassing the P-loop (green star; residues 320 -360 of A2). Overall, the largest conformational changes are observed in A1 (residues 1-300, left half of each panel) despite the fact that different nucleotides are exchanged only at the A2 NBS.

DISCUSSION
The findings reported in this report bear relevance to two fundamental aspects of the ArsA mechanism: how ATP is hydrolyzed, and how As/Sb(III) ions are translocated across the membrane. When crystals of ArsA are either formed or incubated in the presence of ATP, this nucleotide is found at the A2 NBS, while ADP is at the A1 NBS. One possible explanation for this observation is that the A1 NBS is catalytic and the A2 NBS is not. However, if ATP is rapidly hydrolyzed at the A1 NBS, such that only ADP is observed at this site, then, when crystals of the enzyme are incubated with the non-hydrolyzable analog AMP-PNP, this compound should be found at the A1 NBS. Instead, AMP-PNP is found only at the A2 NBS. Moreover, if de novo crystallization is attempted in the presence of AMP-PNP, no crystals are obtained, suggesting that crystals can be formed only if the A1 NBS contains ADP. This might occur because conformational changes associated with ATP binding and hydrolysis at the A1 NBS are not allowed in the crystal. Constraints imposed by the crystal lattice on catalysis are also likely to be the reason why ATP and not ADP is found at the A2 NBS, despite overwhelming evidence from pre-steady-state kinetic data showing that the A2 site is also hydrolytic (8,9). With regard to this point, it may be worth noting that, when a number of intact crystals are incubated in a holding solution containing a regenerating assay mixture to monitor ATP hydrolysis, only minimal enzymatic activity is detected (data not  shown), which is likely to originate from the small amount of free protein in equilibrium with crystalline ArsA. However, when ArsA crystals are dissolved, full enzymatic activity is restored. Thus, it can be argued that ArsA molecules are catalytically competent inside the crystals, but they cannot turn over because the conformational changes that occur in association with ATP hydrolysis are denied within the rigid frame of the crystal lattice.
Instead of being a drawback, this situation can be an advantage; it allows visualization of the initial changes that occur in the active site during the first turnover. In particular, the structures presented here draw attention to the role played in catalysis by residues from both the Switch I and Switch II region of the enzyme and by water molecules in the coordination sphere of Mg 2ϩ (Fig. 4). When ATP binds at the A2 NBS, in one molecule of the crystallographic dimer Asp 447 (Switch II) and Ser 363 (Switch I) form direct hydrogen bonds with a water ligand of Mg 2ϩ , while in the other molecule an additional solvent molecule is intercalated between these residues and the ion. Thus, at least two different conformations of the A2 hydrolytic site are possible in the presence of ATP. When ADP and trifluoroaluminate bind together at the A2 NBS, they mimic the transition state of the hydrolysis reaction (Fig. 4D). This conclusion, which is in line with similar observations made with other enzymes that hydrolyze triphosphonucleosides (24 -26, 34 -43), is supported by the observation that a water molecule, which in this complex is located in proximity of the aluminum atom, mimics an in-line nucleophilic attack onto the ␥-phosphate of ATP. Altogether, the four panels of Fig. 4 can be viewed (going clockwise from B to A through C and D) as consecutive movie frames displaying the scene of ATP hydrolysis. In this movie, the walls of the active site (P-loop, Switch I, and Switch II regions) are pulsing around the Mg 2ϩ ion, producing subtle distortions of its coordination geometry that may be essential for the breaking of the scissile bond.
The vibrations involving residues around the nucleotide and Mg 2ϩ appear to propagate to distant parts of the protein and are likely to provide the physical basis for the events that propel As/Sb(III) into ArsB. A comparison of the structure of the enzyme with ATP bound at the A2 NBS versus the structures in which this site is filled with ADP or ADP⅐AlF 3 (Fig. 5) identified conformational changes involving primarily helices H4 -H5, H9 -H10, H13-H15, a loop located at the end of the Switch I region, and a region encompassing strands S1, helix H2, and strand S2. Surprisingly, the largest positional shifts take place in A1 (for example, see right panel of row 4 in Fig. 5) despite the fact that nucleotides are being exchanged only at the A2 NBS. With regard to this point, it may be of relevance that the cavity of the A2 NBS is completed by residues at the end of strand S6 and the beginning of helix H12 of A1 (e.g. Gln 208 of A1 is hydrogen-bonded to the 3Ј-hydroxyl of the nucleotide ribose at the A2 NBS), two structural elements that follow immediately helices H4 -H5 and H9 -H10 of A1 (Fig. 6). Also in contact with the A-loop of the A2 NBS are helices H14 -H15 of A1 (not shown). Conformational changes in these helices may carry information on the nucleotide type and occupancy from the A2 NBS to the A1 NBS via the A-loop of the latter (see below). Clearly, there is extensive cross-talk between A1 and A2, which is likely to be essential for catalysis.
Interestingly, helices H4 -H5 are also connected to the Switch I region; helices H9 -H10 are connected to the Switch II region; helices H13-H15 are connected to the A-loop; and the region S1-H2-S2 contains the P-loop. Thus, the structure of each half of ArsA can be dissected in four different components, each of which acts as a transducer of events occurring at the NBS (Fig. 7); all the ligands of As/Sb(III) are connected to residues of the hydrolytic site via either the Switch I pathway (Cys 113 is connected to Asp 45 in A1; Ser 420 and Cys 422 are connected to Asp 364 in A2) or the Switch II pathway (His 148 and Cys 172 are connected to Asp 142 in A1; His 453 is connected to Asp 447 in A2). In our study the largest conformational changes were observed between molecule A and molecule B of ArsA in complex with ATP (Fig. 5, row 4), and between either of these molecules and the enzyme in complex with ADP⅐AlF 3 (Fig. 5,   FIG. 5. Difference distance matrices between structures of ArsA in complex with various nucleotides. For each pairwise comparison positive and negative values of the matrix are shown in the left and right panels, respectively. In each panel conformational changes that involve equivalent structural elements in A1 and A2 are marked with symbols of the same shape and color: pink rhombus, helices H4 -H5; cyan circle, helices H9 -H10; red triangle, helices H13-H15 and A-loop; yellow square, a loop and a strand located at the end of the Switch I region of A1; green star, region S1-H2-S2 encompassing the P-loop of A2. Completely clear areas of the density plots correspond to disordered regions of the structures. Visible pixels correspond to positional shifts between 1.0 and 3.5 Å. Residue numbers are labeled on both the X and Y axes of each panel. rows 2-3). In contrast, no significant differences were observed between the structure with ADP⅐AlF 3 and that with ADP, suggesting that a large conformational change is associated with ATP hydrolysis at the A2 NBS but not with phosphate release. Finally, we have not been able to obtain a crystal structure of the enzyme with the A2 NBS unoccupied, even after prolonged incubation of the crystals with periodic changes of a holding solution devoid of any nucleotides. This fact suggests that release of ADP from the A2 NBS is also associated with a conformational change not allowed by the crystal lattice.
These observations are of considerable importance for understanding the mechanism of energy coupling to ion translocation. Although ArsB alone can form a channel for the diffusion of As(III) or Sb(III) ions across the membrane (3), by itself ArsA FIG. 6. Cross-talk between A1 and A2. Stereoview showing the relationship between the A2 NBS and structural elements of A1. The A1 C␣ trace is shown starting with elements H4 -H7 (pink) and continuing with elements S5-H12 (cyan). At the end of this region, residues in the loop between S6 and H12 and at the beginning of H12 complete the A2 NBS cavity. The H9 -H10 region of A1 (cyan) intercalates between H4 and H5 of A1 (pink) and the equivalent region of A2 (here appearing as an upward directed protuberance of the A2 surface), providing a ceiling for the metal binding site. A2 surface, ivory; As/Sb(III), CPK in purpleblue; ATP, bonds colored according to atom types: oxygen, red; nitrogen, blue; carbon, yellow; phosphorus, orange.

FIG. 7. ArsA functional domains.
Center panel, the molecular surface of A1 is dissected into four regions whose conformational changes are under control of the P-loop (green), the Switch I region (pink), the Switch II region (cyan), and the A-Loop (red). ADP bound at the A1 NBS is shown as bonds colored according to atom types; Mg 2ϩ and As/Sb(III) are shown as CPK in magenta and purpleblue, respectively. Corner panels, C␣ traces of the domains whose surface is shown in the center panel. Mg 2ϩ -ADP is shown for reference next to each trace. Side chains are shown with yellow bonds; Mg 2ϩ and As/Sb(III) as CPK in magenta and purple-blue, respectively.
FIG. 8. ArsA catalytic cycle. Helices H9 -H10 of A1 (red) and A2 (cyan) are the arms of a gate alternating in the "open" and "closed" positions. An As(III) ion is shown as a blue sphere. For each cycle of ion translocation, one ATP is used at the A2 NBS in the transfer step, and one at the A1 NBS in the re-isomerization step. Although the scheme depicts a hypothetical situation in which only one As(III) ion is translocated per catalytic cycle, the actual stoichiometry of ions translocated per ATP hydrolyzed is not known.