ATP Hydrolysis in Eg5 Kinesin Involves a Catalytic Two-water Mechanism*♦

Motor proteins couple steps in ATP binding and hydrolysis to conformational switching both in and remote from the active site. In our kinesin·AMPPPNP crystal structure, closure of the active site results in structural transformations appropriate for microtubule binding and organizes an orthosteric two-water cluster. We conclude that a proton is shared between the lytic water, positioned for γ-phosphate attack, and a second water that serves as a general base. To our knowledge, this is the first experimental detection of the catalytic base for any ATPase. Deprotonation of the second water by switch residues likely triggers subsequent large scale structural rearrangements. Therefore, the catalytic base is responsible for initiating nucleophilic attack of ATP and for relaying the positive charge over long distances to initiate mechanotransduction. Coordination of switch movements via sequential proton transfer along paired water clusters may be universal for nucleotide triphosphatases with conserved active sites, such as myosins and G-proteins.

Motor proteins couple steps in ATP binding and hydrolysis to conformational switching both in and remote from the active site. In our kinesin⅐AMPPPNP crystal structure, closure of the active site results in structural transformations appropriate for microtubule binding and organizes an orthosteric two-water cluster. We conclude that a proton is shared between the lytic water, positioned for ␥-phosphate attack, and a second water that serves as a general base. To our knowledge, this is the first experimental detection of the catalytic base for any ATPase. Deprotonation of the second water by switch residues likely triggers subsequent large scale structural rearrangements. Therefore, the catalytic base is responsible for initiating nucleophilic attack of ATP and for relaying the positive charge over long distances to initiate mechanotransduction. Coordination of switch movements via sequential proton transfer along paired water clusters may be universal for nucleotide triphosphatases with conserved active sites, such as myosins and G-proteins.
In a wide variety of protein machines, the binding and hydrolysis of ATP are chemically coupled to large scale conformational changes that occur within and distant from the nucleotide site, resulting in force generation and movement. Bimolecular motors, like kinesins (1,2), and proteins not conventionally considered motors, such as DNA-modifying enzymes, use this principle of mechanotransduction to perform their cellular roles. However, direct cause-and-effect linkage between the catalytic steps and powered structural changes in these proteins has not been established, principally as the deprotonation partner in the initial chemical step of proteindriven ATP hydrolysis is not known.
Within kinesins, as well as myosins and G-proteins, common sequence motifs, such as switch I, switch II, and the P-loop, organize into an active site that can accelerate NTP hydrolysis. It is well accepted that the nucleophile for catalysis is a water molecule and that protonation and deprotonation events are at the core of this acid-base reaction. Upon substrate binding, the initial chemical step is that a catalytic base abstracts a proton from the lytic water. The resulting hydroxide ion attacks the substrate at the ␥-phosphate and alters the partial charge distribution between the ␥and ␤-phosphate groups. The ␤-␥phosphoanhydride bond is cleaved upon the shift in electron charge within the substrate, and the reaction products, a nucleotide diphosphate leaving group and an inorganic phosphate (P i ), are formed.
Elusive for kinesins specifically and other ATPases in general is experimental identification of the catalytic base. The chemical nature of the base defines how catalysis in motor proteins can be accelerated, controlled, or reversed. The archetypal model of protein-mediated ATP hydrolysis predicts that the proton acceptor should be a conserved amino acid side chain and that the proton-donating and proton-accepting groups are in close proximity. However, protein crystallographic models (3) have not identified an obvious proton acceptor within 5 Å of the phosphate ester linkage of ATP. Thus, two alternate proposals have been put forth: substrate-assisted catalysis, in which a proton of the attacking water molecule is transferred to an oxygen of the ␥-phosphate (4), and a proton relay mechanism in which a water network spanning the distance between conserved residues and substrate provides a proton shuttle (5). Implicit within the latter proposal is that the distance between the acid-base pair can be increased by water molecules. Discrimination between these proton transfer models is necessary for not only understanding the mechanism of ATP hydrolysis proper but also for describing how events at the active site are transmitted to distal parts of the protein.
Here we have determined the crystal structure of the wildtype motor domain of human Eg5 kinesin in a prehydrolytic state, trapped by the non-hydrolyzable analogue AMPPNP. 2 This study provides the first experimental observation of the catalytic base for any nucleotide triphosphatase (NTPase) and the first atomic details regarding the prehydrolysis state for any motor protein. Several observations in our data support these assignments. The electron density around the bound nucleotide (see Fig. 1a) cradles three phosphate groups in the active site, and the ␥-phosphate is in a tetrahedral geometry. Second, both switch regions are in a closed conformation (see Fig. 1b) and linked by a salt bridge, revealing an ordered water network spanning the distance from an interswitch salt bridge to the nucleotide ␥-phosphate. Lastly, structural elements required for microtubule (MT) binding have distinct conformational organizations and molecular interactions not previously observed. This initial set of structural and chemical isomerizations necessary to prime ATP hydrolysis and mechanotransduction in kinesins is likely conserved in myosin, another major superfamily of motor proteins. Our data also provide a mechanistic model for how transmission of a proton across a water network disrupts an interswitch salt bridge, thereby promoting a second major conformational transition.
Crystallization, Data Collection, and Data Processing-Crystals obtained with Hampton Crystal Screen Cryo TM solution 42 were used to streak seed drops containing a 1:1 ratio of protein to well solution (40 mM KH 2 PO 4 , 20% glycerol w/v, 15% polyethylene glycol 8000 w/v). Crystals were grown by vapor diffusion at 4°C in hanging drops. Rectangular crystals appeared overnight and were allowed to grow for several days before flash-freezing in LN 2 . Diffraction data (Table 1 and supplemental Table S1) were collected at 100 K using a Bruker MICROSTAR x-ray generator equipped with HELIOS optics and a PROTEUM 4K PLATINUM 135 CCD camera and were integrated and scaled with PROTEUM2 software (Bruker AXS). Crystals belong to space group P2(1) with cell parameters a ϭ 60.50 Å, b ϭ 71.44 Å, c ϭ 94.85 Å, and ␤ ϭ 98.86°. Volume considerations indicated that two Eg5 molecules existed in the crystal asymmetric unit.
Structure Solution and Refinement-The Eg5⅐AMPPNP complex structure was determined using the program AMoRe (7) and molecular replacement. The search model was molecule A of the Eg5⅐ADP⅐monastrol complex (Protein Data Bank (PDB) ID 1X88) without associated waters, ADP, Mg 2ϩ , and monastrol. These coordinates are also missing switch II cluster residues 272-286. Using data in the range 8 -3.5 Å and a sphere of 25 Å, the rotation function yielded two clear solutions. The translation function verified that the top two rotation function orientations were correct. Rigid-body fitting in AMoRe using data in the range 8 -3.5 Å yielded an R cryst of 48.2. At this point, 15 steps of CNS (8) rigid-body refinement using data from 25 to 2.42 Å yielded R cryst ϭ 49.5, R free ϭ 48.8. Next, a 5,000 K slow cool-simulated annealing using torsion angle dynamics produced coordinates with R cryst ϭ 40.0, R free ϭ 45.6 for 25-2.42 Å data. Inspection of initial 2F o Ϫ F c and F o Ϫ F c maps indicated significant differences primarily in the two switch regions. These portions of the model were removed, and data were extended to 2.19 Å. Successive rounds of manual model building with O (9) and automated refinement were performed. When the model (not including waters, nucleotide, or Mg 2ϩ ) was essentially complete, the CCP4 (10) program ACT was used to guide the application of non-crystallographic symmetry restraints. Non-crystallographic symmetry was applied to 34 residues in each of the two molecules. At a very late stage in model building, the coordinates and topology/parameter files for AMPPNP were acquired from Hic-Up (11), and the nucleotide and Mg 2ϩ were positioned into the active sites.
The final model contains residues 17-366 for each molecule of the asymmetric unit, 281 waters, two AMPPNP molecules, two Mg 2ϩ ions, and one inorganic phosphate (Table 1 and supplemental Table S1). To determine whether we had inappropriately applied non-crystallographic symmetry, a 5,000 K slow cool-simulated annealing refinement using torsion angle dynamics was performed on the final model without the use of non-crystallographic symmetry restraints. The root mean square difference between the two Eg5 molecules after this where F oh and F ch are the observed and calculated structure factor amplitudes for the reflection h. Working set includes 36,987 reflections. b R-factor calculated for 4,140 reflections that were chosen randomly and not included in the refinement. c r.m.s., root mean square. refinement was 0.64 Å for all C␣ atoms and 0.10 and 0.28 Å between C␣ atoms from regions that include switch I (residues 217-238) and switch II (264 -281) of each Eg5 molecule. Table S2 were made using PyMOL (12). C␣ atoms of residues 102-111 of Eg5 and the equivalent residues from Eg5⅐ADP (PDB ID 1II6), Kif1A⅐AMPPNP (PDB ID 1VFV), and myosin II⅐ADP⅐vanadate (PDB ID 1VOM) were used for aligning coordinates for Figs. 1 and 2 and also coordinates used to produce supplemental Table  S2. The AMPPNP-bound-to-ADP-bound movies (supplemental Movies S1 and S2) were made using the RigiMOL add-on in PyMOL, as well as the eMovie (13) plug-in. The start state included the Eg5⅐AMPPNP structure (PDB ID 3HQD), and the end state was Eg5⅐ADP (PDB ID 1II6). For clarity, the missing portion of the Eg5⅐ADP switch II loop was assembled in Swiss-Pdb Viewer and energy-minimized.

RESULTS
Closed Active Site-To our knowledge, there is no prior published crystallographic observation of the fully closed conformation of an active site or of a ␥-phosphate-sensing mechanism for kinesin motor proteins. Kinesins, in common with myosins, are thought to undergo an ATP-induced conformational transition in which the switch regions flanking the active site shift from "open" to "closed" states (14,15) with the open state resembling the ADP-bound conformation. Indeed, numerous NTPases employ a ␥-phosphate-sensing mechanism that alters their molecular conformations dependent upon the presence or absence of the ␥-phosphate. The closed state of NTPases, in which critical determinants have been correctly positioned, is competent for supporting nucleotide hydrolysis, whereas the open state is not.
Accordingly, the conformations of switch regions in the closed state of kinesins should be structurally distinct from either ADP-bound protein or the initial ATP⅐kinesin collision complex. Although the overall architecture of the Eg5⅐AMPPNP complex is akin to that of other microtubule motors (Fig. 1b), there are differences among the loops and helices surrounding the core ␤-sheet; these include L5, L9 (containing the 229 NAYSSR 234 switch I sequence (16)), ␣3, L11 (containing the 265 DLAGSE 270 switch II sequence (16)), ␣4, L12, and ␣6 (Fig.  1b). These regions are dissimilar from ADP-bound Eg5, exhibit two novel structural features that distinguish them from the structures of other kinesins bound to ATP analogues or transition state mimics (supplemental Table S2), and are consistent with expectations of a closed, catalytically competent state.
First, the switch I conformation in Eg5⅐AMPPNP (Fig. 2a) adopts a closed conformation, unique for kinesins but previously seen in myosins (17). In the Eg5⅐ADP structure (18), switch I residues 231-234 adopt a helical conformation that is 4 Å removed from the location of the (absent) ␥-phosphate (Fig. 2a). The C terminus of ␣3 (residues 221-226; Fig. 2a), immediately preceding switch I, is unwound in the Eg5⅐AMPPNP structure in comparison with the Eg5⅐ADP structure. Unwinding of the ␣3 C terminus increases the length of the switch I loop, fosters interactions of the NAYSSR motif observed at the active site, and rotates the ␣3 helix slightly (ϳ11°, Fig. 2a) relative to the core ␤-sheet. As the switch I region of a kinesin had never been observed in a conformation analogous to that of a closed myosin switch I, it has been hypothesized that kinesin switch I functions differently from that of myosin (15). However, our data confirm that the switch I of kinesins and myosin operates via closely related mechanisms, with each adopting a similar switch I conformation (Fig.  2b) that is required to fully enclose the nucleotide-binding site.
Second, the switch II region in the Eg5⅐AMPPNP complex assumes a completely closed conformation (Fig. 2c), and the switch II loop interacts with switch I and the nucleotide ␥-phosphate. Comparison of the ADP-bound, ATP analogue-bound, and transition state mimic-bound forms of kinesins (supplemental Table S2) does not show large changes in the secondary structure of switch II, and no large secondary structure changes are noted in this region of the Eg5⅐AMPPNP complex. However, for the first time in any wild-type kinesin crystal structure, the Eg5⅐AMPPNP structure features a completely ordered switch II loop (L11). The ordered switch II loop (Fig. 2c, red), which includes a short helical element formed by residues 272-275, is extended toward the active site. Residues in the switch II loop participate in H-bonding interactions with residues of switch I, the P-loop, ␣6, active site water molecules, and the nucleotide proper (see below). Because the switch II loop in Eg5  FEBRUARY 19, 2010 • VOLUME 285 • NUMBER 8 ADP-bound structures is disordered (supplemental Table S2), it appears that interactions provided by the ␥-phosphate of AMPPNP have a stabilizing effect on these residues.

Proton Transfer Drives Eg5 Mechanotransduction
Propagated Changes at the Microtubule-binding Site-Interaction with microtubules is crucial to kinesin function, and the ATP-bound form of motor proteins is accepted as the conformational state in which microtubules are stably bound (19,20) and in which forward motion is produced in dimeric forms of kinesin. Therefore, substrate binding to the active site should elicit distal structural changes at the MT-binding surface, and these microscopic changes have been probed using crystallography and cryo-EM techniques. The "ATP-like" or "helix-up" structural orientation of the switch II relay helix (␣4) in kinesins (19), also observed in our Eg5⅐AMPPNP structure (Fig. 1c and supplemental Movie S1), has been correlated with the chemical state of the nucleotide bound to the motor domain. This ATPlike position differs from the "ADP-like" state as the N-terminal portion of ␣4 is rotated toward the nucleotide-binding site and the helix as a whole is shifted away from the central ␤-sheet. In the ATP-like position, ␣4 is shifted out of the plane of the ␣6 helix, allowing for neck linker docking against the core ␤-sheet.
Cryo-EM studies revealed that the ␣4 relay helix is also lengthened in kinesin⅐microtubule complexes (14,21,22). This lengthening has not uniformly been observed in the kinesin crystal structures in a presumed ATP-like state, prompting the suggestion that MT interactions are necessary for stabilizing a longer ␣4 helix (22,23). Challenging this idea, the ␣4 helix in our crystal structure, found at the MT-binding face, agrees well in both length and orientation with cryo-EM densities of Drosophila Kinesin-5⅐AMPPNP⅐ MT complexes (21). Comparison of and structural agreement between our data and this recent cryo-EM study (21) reveal that not only the orientation but also the length of the ␣4 helix is important for achieving a tight, AMPPNP-MT-bound state and is directly correlated with the stability of the switch II loop.
In addition, the fully ordered switch II region, only observed in the AMPPNP-form of Eg5, appears to organize additional residues into the ␣4 helix and alter the orientation of the relay helix. Coincident with these conformational changes is the observation of a novel network of H-bonding among residues of the switch II helix, the P-loop, and the ␣6 helix. Asn 342 of ␣6, Arg 274 of the switch II loop, and Gln 106 of the P-loop form an H-bonding triad (supplemental Fig. S1). This interaction appears to stabilize the closed conformation of the switch II loop and may also serve as a mode of communication from the active site to the MT-binding surface.
Atomic Interactions in the Closed Active Site-The closing of the nucleotide-binding pocket produces changes critical for enzyme-catalyzed ATP hydrolysis. Three mechanistic outcomes result from the conformations adopted by switch regions and their adjacent secondary structures and drive toward a conclusion regarding the biochemical role of the switch elements. First, there is formation of an anticipated interswitch salt bridge (15) between the side chains of switch I Arg 234 and switch II Glu 270 that closes the active site (Fig. 3a). This Eg5 salt bridge is equivalent to a salt bridge between analogous residues in myosin switches, shown to be essential for closing the active site and ATP hydrolysis (24 -26). The salt-  (a and b) and of the relay helix and switch II loops (c and d) from representative kinesin and myosin structures are shown. a, differences in ␣3 and L9 between Eg5⅐AMPPNP (red) with the equivalent sequences of Eg5⅐ADP (cyan) and Kif1A⅐AMPPNP (yellow). Disordered residues are shown as dashed lines. Note that L9 of Eg5⅐AMPPNP approaches the nucleotide in this closed conformation. b, Eg5⅐AMPPNP (red) and myosin⅐ADP⅐vanadate (gray) have similar switch I loop conformations. Conserved switch I arginines are shown. c, relay helices and switch II loops of Eg5⅐AMPPNP (red), Eg5⅐ADP (cyan), and Kif1A⅐AMPPNP (yellow). L11 of Eg5⅐AMPPNP is ordered, and the relay helix is in an ATP-like position. d, Eg5⅐AMPPNP (red) and myosin⅐ADP⅐vanadate (gray) have similar switch II loop conformations. Conserved switch II glutamates are shown. bridge geometry seen in the myosin⅐ADP⅐vanadate (17) and our Eg5⅐AMPPNP substrate pockets has not been observed in other wild-type kinesin structures in which the salt bridge is not formed at all or which feature a salt bridge with different geometries and distances (23,27,28). These kinesin structures may represent intermediate states that precede an ATP-induced, fully closed conformation.
A second mechanistic outcome resulting from the closed conformations adopted by Eg5⅐AMPPNP switch elements is the observation of not one but two water molecules (Figs. 1a and 3, W1 and W2) in the active site, one of which (W1) is coordinated by conserved switch residues and lies close to the scissile phosphate of AMPPNP. These two waters are present in each of the two Eg5⅐AMPPNP complexes in the crystal asymmetric unit with nearly identical orientations and with W1-W2 separation distances in each complex of 2.41 and 2.43 Å. Furthermore, omit electron density maps show that the densities for all four of the water molecules are well formed and equal in size (data not shown), and the temperature factor values for these water molecules have been refined to below the average for waters within the structure.
The first water, W1, is oriented 3.3 Å from the ␥-phosphorus atom in a nearly perfect "in-line" position for a nucleophilic role in catalysis (Fig. 3). This is the first observation of a water molecule proximal to the ␥-phosphate of the kinesin substrate, and thus the lytic water, in these motor proteins. W1 is coordinated by the amide carbonyl group of switch I Ser 233 , the backbone amide group of switch II Gly 268 , and a second water molecule, W2. Given the chemistry of the available moieties surrounding W1, we reach the unexpected conclusion that W2 serves as the proton acceptor for the nucleophilic W1 (see below). The second water is secured in the nucleotide pocket via the backbone oxygen of Gly 268 and the side chain oxygen of switch II Glu 270 .
Thus, this active site water pair is positioned by a tight network of hydrogen bonds and bridges the distance between the ␥-phosphate and the interswitch salt bridge. The advantage of such a hydrogen-bond network has been calculated for myosins, G-proteins, and F 1 -ATPase (29 -32) but never experimentally observed.
A third important feature of the closed active site is the extensive protein-mediated electrostatic stabilization and H-bonding of the bound nucleotide that are central features in enzyme-catalyzed nucleotide hydrolysis. These H-bonding interactions with the nucleotide, complex in nature and also large in number, can have two effects: polarization of the ␥-phosphate P-O bonds and reduced Coulombic repulsion in preparation for an associative nucleophilic attack by the lytic water (below). Overall, there are also two active site residues that have three H-bonds. Switch I Ser 233 and switch II Gly 268 each have hydrogen bonds to the ␥-phosphate of AMPPNP and to W1. Interactions with Mg 2ϩ and W2 form the third coordinations for Ser 233 and Gly 268 , respectively (Fig. 3a). There are also two functional groups that form bidentate ligations in the active site; Mg 2ϩ tethers oxygens of both the ␥-phosphates and the ␤-phosphates of the adenine nucleotide and Lys 111 in the P-loop H-bonds with both the ␥-phosphate and the ␤-phosphate.
In detail, the Eg5 switch elements form extensive H-bonding interactions with the ␤-␥ bridging oxygen and ␥-phosphoryl oxygens of AMPPNP and with the Mg 2ϩ ion. The nitrogen of AMPPNP, a chemical substitute for the ␤-␥ bridging oxygen atom in the native substrate, is 2.9 and 2.6 Å from the switch I Asn 229 N␦ atom and the amide nitrogen of the P-loop Gly 108 , respectively. The elongated conformation of the switch I loop, described above, allows Asn 229 to approach the atomic ␤-␥ bridging substitute. In the Eg5⅐ADP structure, the compara-  FEBRUARY 19, 2010 • VOLUME 285 • NUMBER 8 tively shorter switch I loop alters the location of Asn 229 and prohibits interaction with the nucleotide. Thus, in the ATPbound motor, closure of switch I would allow for H-bonding between Asn 229 and the ␤-␥ bridging oxygen, which would stabilize developing negative charge on this atom during catalysis.

Proton Transfer Drives Eg5 Mechanotransduction
The three oxygen atoms of the ␥-phosphate are securely coordinated in the prehydrolysis state of this kinesin; there are at least two chemical groups within interaction distance of each ␥-phosphoryl oxygen. The O1 atom of the ␥-phosphate is coordinated by the N atom of the P-loop Lys 111 and the amide nitrogen of switch II Gly 268 (Fig. 3b). The O2 atom of the ␥-phosphate is within hydrogen-bonding distance of the hydroxyl group of the P-loop Thr 107 and the side chain of switch I Ser 232 . Analogous to the Kif22⅐AMPPNP structure (23), the presence of the ␥-phosphate elicits an alternate rotamer conformation of Thr 107 that facilitates interaction with the ␥-phosphate. In this position, Thr 107 comprises an element of the ␥-phosphate sensor, which reorients following phosphate release to form an H-bond to Glu 270 in switch II (supplemental Movie S2). In addition, the O3 atom of the ␥-phosphate is within hydrogen-bonding distance of the amide nitrogen of Ser 233 in switch I and the Mg 2ϩ ion.
The Mg 2ϩ ion, required for charge stabilization in ATP hydrolysis (33), coordinates six ligands in an octahedral geometry (Fig. 4) in the Eg5⅐AMPPNP complex. These ligands are ␥and ␤-phosphoryl oxygens of the substrate mimic, the O␥ atom of switch II Ser 233 , O␥ of the P-loop Thr 112 , and water molecules (W3 and W4). We propose that the Mg 2ϩ ion secures the inline nucleophilic attack configuration, stabilizes alterations in charge distribution between the ␤and ␥-phosphate moieties of the substrate, and potentially facilitates protonation of the ADP leaving group oxyanion. The latter point is supported by examination of the open nucleotide-binding site in the Eg5⅐ADP complex (18); Mg 2ϩ becomes coordinated with two additional water molecules that displace bonds to Ser 233 and to the ␥-phosphate in the AMPPNP state.
Notably, the atomic interactions described above lead to the conclusion that the conventional function attributed to switch regions in NTPases in general is not a complete description. Switches I and II are traditionally portrayed as ␥-phosphate sensors (16). Our data show that these conserved sequence elements do interact with the substrate as anticipated but are equally critical for transiently organizing both water molecules in the active site. The majority of atomic interactions involving switch I and switch II residues (Fig. 5a) provide charge stabilization of the nucleophilic water and W2, the catalytic base. The P-loop residues bind the ␣and ␤-phosphate groups of the nucleotide and the scissile phosphate moiety (Fig. 5a).

DISCUSSION
This structural study provides direct experimental detection of the general base that initiates catalysis in motor proteins, a first among kinesins as well as a first among any group of protein machines that use ATP hydrolysis to drive function. The major surprise of this work on kinesin proteins is the strong correspondence of the mechanistic observations in our experimentally derived structure to hypothetical and calculated models of NTP hydrolysis in other classes of proteins. Multiple waters have been postulated to be favorable in the F 1 -ATPase through mutational analysis (34) and computational efforts (35) and in myosin via remarkably consistent theoretical studies (36). Multiple waters have previously been experimentally observed in p21 ras (31), although in quite a different spatial organization from this study. Coupled with our data, this detection of waters in the G-protein active site opens the door for the argument that the chemical nature of the catalytic base is conserved in other P-loop NTPases.
Our data provide new information beyond the theoretical studies published and serve as a powerful aid in building a description of the initial reaction steps that are likely to occur within the kinesin active site. Questions we will address are (i) the general mechanism of nucleotide hydrolysis for motor proteins, (ii) the catalytic role of the waters present in the active site, and (iii) the cause-and-effect link by which active site reactions are transmitted to distal sites. ATP hydrolysis is proposed to proceed through either of two general mechanisms (37)(38)(39)(40): dissociative or associative. A dissociative mechanism occurs when the ␤-␥ bond of ATP is cleaved prior to binding of the nucleophilic water to the ␥-phosphorus. This mechanism would involve development of a negative charge on the bridging oxygen and an HPO 3 Ϫ intermediate. In contrast, an associative mechanism features bond formation between the nucleophilic water and the ␥-phosphorus, creating a pentacoordinate transition state that destabilizes the ␤-␥ bond, leading to bond cleavage. In this case, negative charge would develop on the ␥-phosphate.
These hydrolytic mechanisms can be distinguished by examining distances between the ␥-phosphorus atom and the attacking and leaving group oxygens, the ␥-phosphate geometry, and charge stabilization of the ␥versus ␤-phosphate. An associative mechanism requires the close approach of the attacking nucleophile (41), and distances as short as 2.7 Å between the nucleophilic water and the ␥-phosphorus are indicative of this reaction pathway (17). In contrast, distances as long as 5.3 Å (42) between these two atoms are reflective of a dissociative reaction. In our structure, W1 is 3.3 Å from the ␥-phosphorus atom and 2.9 Å from O2 of the ␥-phosphate, distances supporting an associative mechanism (Fig. 3a). Additionally, W1 is in a near equidistant position from all three ␥-phosphate oxygens, a geometry favoring formation of a pentacoordinated intermediate and in-line nucleophilic attack. Finally, as discussed previously, there is extensive charge stabilization of the ␥-phosphate of the AMPPNP. In contrast, contacts with the bridging oxygen, limited to Asn 229 and the backbone carbonyl of Gly 108 , are not likely to foster accumulation of negative charge in preparation for a dissociative mechanism. Based on these observations, our data favor the conclusion that the prehydrolysis state of Eg5 kinesin is primarily associative in character.
The second major question we address is the catalytic role of the two waters present in the active site of Eg5⅐AMPPNP. Foremost, these water molecules are recruited as critical chemical reagents. They are the key players in the hydrolysis reaction: the nucleophile (W1) that transfers its negative charge to the substrate and the catalytic base (W2) that accepts the proton (Fig.  5a). This transient water bridge can shuttle protons rapidly across the solvent H-bonds and delocalize their charge. The observed 2.4 Å distance between the oxygens of W1 and W2 (Fig. 3), uncommonly small for any structure despite allowance for noise in the diffraction experiment, is diagnostic for the protonation state of these solvent molecules. Although x-ray diffraction methods are not traditionally associated with observing protons, proton transfer is exquisitely dependent on structure; as the distance a proton can tunnel is less than a covalent bond length, atoms participating in the acid-base catalysis must be in quite close proximity.
In other protein structures (43), representative distances between oxygens of adjacent water molecules are significantly larger at nearly 2.8 Å, suggesting that W1-W2 in the Eg5 orthosteric site are not in the ground state. Reported O-O distances that approximate our W1-W2 pair are 2.55, 2.45, and 2.4 Å, measured between a water molecule and a protonated hydronium ion (44), a hydroxide ion (45), and a shared proton with a second water molecule (46), respectively. We deduce that a proton is localized and shared between the water pair in this crystal structure of the Eg5 prehydrolysis state and that proton abstraction from the nucleophile is facilitated via a Grotthuss-type mechanism (47). Real-time monitoring of Eg5 kinesin catalysis 3 corroborates that a delocalized proton within a water cluster is observed. As the abstracted proton on W2 creates a hydronium ion (Fig. 5b), the switch II Glu 270 carboxylate can provide the required charge stabilization for the pos-  itively charged hydronium ion via complex salt-bridge formation (Fig. 5b).
The W1-W2 proton relay network is the central cause-andeffect link between chemistry and conformational changes distal from the orthosteric site and critical for motor protein function. Unlike other catalytic reaction mechanisms, the second protonation event in the Eg5 active site is equally important as the first. A second proton transfer from the W2 hydronium ion to Glu 270 will destabilize the interswitch Glu 270 -Arg 234 salt bridge. Breaking the H-bond between this residue pair will result in conformational changes in the switch I and switch II regions that lead to opening of the active site (Fig. 5d). Thus, the scission of the interswitch Arg 234 -Glu 270 salt bridge is the basis for local chemical modifications amplifying into structural rearrangements. Subsequently, the open nucleotide-binding cavity may allow P i release and permit reorientation of the Thr 107 side chain to move within 2.4 Å of Glu 270 (Fig. 5d), reestablishing an interaction seen in the Eg5⅐ADP structure (18). Environmentally sensitive salt-bridge structures are known to play an important role in biological systems and act as controllable switches participating in various gating mechanisms. The hydrogen-bonded ion pair is required for efficient and synchronized motor function (15), and this simple salt bridge can stabilize the closed conformation in Eg5 kinesin (18). Our data provide a mechanistic rationale for kinesins and perhaps ATPases in general.
Usage of protonated water clusters in protein machines capable of NTP hydrolysis is equivalent in importance to the role of transition state stabilization through electrostatics or H-bonding. Biochemically, this W1-W2 cluster forms an H-bonding network that lifts the constraint of an acid-base pair being in close proximity; a shuttled proton along the water cluster can communicate chemical modification of the scissile phosphate to regions of the protein distant from the substrate. Functionally, an H-bonded water network permits communication with a larger number of switch regions in the protein and can trigger a greater range of conformational changes distant from the active site and required for force production and movement. These mechanistic outcomes would be unachievable with a single proton transfer event or one active site water.
Important for its cellular role, control of the direction and reversibility of proton transfer can be realized with this orthosteric water cluster in tandem with a terminal salt bridge at the edge of the motor domain active site. Metamorphosis from a simple to a W2-complexed salt bridge results in minor, stepwise changes in geometry as the Arg 234 -Glu 270 interaction in kinesin is already constrained. Moreover, the organization and chemistry of the active site players permit reversibility in kinesin catalysis. The W1 and W2 are equivalently amphiprotic, can both accept and donate protons, and can allow the same Grotthuss-type mechanism (47) to proceed in either direction. Similar bidirectionality of transfer can be constructed with other, adjacent amphiprotic species, such as the neighboring Ser 232 and Ser 233 residues in switch I. We speculate that motor proteins may have repeated the use of this theme in their catalytic sites at the level of both protein and solvent to ensure control of catalytic commitment and coordination with events at the cytoskeletal-binding site.