Crystal Structure of Monomeric Actin in the ATP State

A nucleotide-dependent conformational change regulates actin filament dynamics. Yet, the structural basis of this mechanism remains controversial. The x-ray crystal structure of tetramethylrhodamine-5-maleimide-actin with bound AMPPNP, a non-hydrolyzable ATP analog, was determined to 1.85-Å resolution. A comparison of this structure to that of tetramethylrhodamine-5-maleimide-actin with bound ADP, determined previously under similar conditions, reveals how the release of the nucleotide γ-phosphate sets in motion a sequence of events leading to a conformational change in subdomain 2. The side chain of Ser-14 in the catalytic site rotates upon Pi release, triggering the rearrangement of the loop containing the methylated His-73, referred to as the sensor loop. This in turn causes a transition in the DNase I-binding loop in subdomain 2 from a disordered loop in ATP-actin to an ordered α-helix in ADP-actin. Despite this conformational change, the nucleotide cleft remains closed in ADP-actin, similar to ATP-actin. An analysis of the existing structures of members of the actin superfamily suggests that the cleft is open in the nucleotide-free state.

guish" between ATP-and ADP-actin suggests that these two states are structurally different. Consistent with this view, biochemical (3)(4)(5), spectroscopic (6 -8), and electron microscopic (9) evidence has suggested that a conformational change in actin subdomain 2 accompanies the hydrolysis of ATP and the release of inorganic phosphate.
Visualization of the structural details of such a conformational change has come from a comparison of crystal structures of the actin monomer (G-actin) in the ATP and ADP states. ATP-actin structures have been determined from complexes with actin-binding proteins that keep actin in a monomeric state: DNase I (10), profilin (11), gelsolin (12,13), and vitamin D-binding protein (14,15). ADP-actin, on the other hand, was crystallized in a monomeric state after binding tetramethylrhodamine-5-maleimide (TMR) 1 to Cys-374, which blocks polymerization (16). A comparison of the structures in the two states reveals how the release of the nucleotide ␥-phosphate triggers a sequence of events that propagate into a loop to helix transition in the DNase I-binding loop in subdomain 2. However, a proper comparison of the ATP-and ADP-bound states of actin would require for the two structures to be determined under similar conditions.
Meanwhile, an analysis of the existing structural data has led to conflicting interpretations. Recent reports have questioned the validity of the structure of TMR-modified actin with bound ADP as truly representative of the ADP state (17)(18)(19). Some (18,20,21) think that in the "real" ADP state the cleft that separates the two major domains of actin must be open, as observed in the so-called "open" state structure of the actinprofilin complex determined with bound ATP (22). It has also been suggested that the TMR probe bound to Cys-374 prevents the two domains of actin from opening apart so that actin can assume its "true" ADP conformation (18). The TMR probe, rather than the change in the nucleotide, has also been held responsible for the conformational change observed in actin subdomain 2 (17,19). According to this hypothesis, long range allosteric effects due to the binding of the TMR probe to the C terminus of actin could have induced the conformational change observed at the opposite end of the molecule in subdomain 2. Another explanation for the loop to helix transition in the DNase I-binding loop, which disregards any role of the nucleotide state, is that it is triggered by crystal packing contacts (18). A structure of monomeric TMR-actin in the ATP state would provide an excellent way to resolve this controversy (23,24).
In this report, we describe the structure of TMR-modified actin with bound AMPPNP, a non-hydrolyzable ATP analog (25), at 1.85-Å resolution. The symmetry and unit cell param-eters of the crystals of monomeric AMPPNP-actin are identical to those of the ADP-actin structure reported previously (16). As compared with the ADP-actin structure, the presence of the nucleotide ␥-phosphate in the new structure forces the rotation of the side chain of Ser-14, in line with all of the existing structures of ATP-actin (10 -15). The rotation of Ser-14 brings about a change in the conformation of a neighboring loop, which contains the methylated His-73. Because of the central role that this loop plays in transmitting nucleotide-dependent conformational changes from the catalytic site to subdomain 2, we refer to it as the "sensor loop." In the current ATP-actin structure, the movement of the sensor loop is propagated into subdomain 2, resulting in a small rotation and increased thermal disorder of this subdomain and the melting of the ␣-helix in the DNase I-binding loop, this helix being a characteristic feature of the structure of ADP-actin (16). These structural features are common to all of the existing ATP-actin structures (10 -15). Therefore, it now can be asserted that the differences observed between the structures of monomeric actin in the ATP and ADP states (16) are due to the release of the nucleotide ␥-phosphate and not because of crystal packing contacts or the binding of the TMR probe.

EXPERIMENTAL PROCEDURES
Protein Preparation-Actin modified with TMR (Molecular Probes) was prepared with minor modification of the protocol described previously (16). All of the steps were carried out at 4°C. Actin was prepared as described previously (26). A F-actin pellet was homogenized in Gbuffer (2 mM MOPS, 0.2 mM CaCl 2 , 0.2 mM ATP, 0.01% NaN 3 , pH 7.5) with the addition of 10 mM dithiothreitol to reduce actin fully. After 1 h, actin was dialyzed exhaustively against G-buffer to remove the dithiothreitol and then centrifuged for 30 min at 100,000 ϫ g to pellet any F-actin that did not depolymerize and any denatured actin. The resulting G-actin at 5-8 mg/ml was reacted with a 2.2-molar excess of TMR (dissolved as a 30 mM solution in dimethylformamide) and stirred overnight, followed by the addition of 5 mM dithiothreitol to react with any excess TMR. The solution was then brought into F-buffer conditions (40 mM NaCl, 2 mM MgCl 2 , 2 mM MOPS, 0.2 mM CaCl 2 , 0.2 mM ATP, 0.01% NaN 3 , pH 7.5). After 1 h, the solution was centrifuged for 30 min at 100,000 ϫ g to pellet any unreacted F-actin and insoluble TMR. The supernatant was exhaustively dialyzed against G-buffer and then against G-buffer in which ATP was replaced with AMPPNP (Sigma). Note that, for two main reasons, in this procedure special effort was made to minimize the time that TMR-actin spent in F-actin buffer. First, nucleotide hydrolysis in F-buffer is significant but practically zero in G-buffer (27). Second, unpublished evidence from analytical ultracentrifugation 2 and electron microscopy 3 indicates that TMR-actin kept in F-buffer aggregates slowly into a non-filamentous form that can no longer be crystallized.
Crystallization-Crystals of TMR-modified AMPPNP-actin were obtained at 20°C using the hanging drop vapor diffusion method under conditions somewhat different from those used for ADP-actin (16). In a typical experiment, 2 l of AMPPNP-actin at 9 -12 mg/ml (concentrated using a Centricon device) were mixed with an equal volume of a reservoir solution containing 18% polyethylene glycol 3350, 9 mM ACES, pH 7.0, 365 mM CaCl 2 , and 9% dimethyl sulfoxide. The crystals, which typically formed within 2-3 days, were harvested immediately to minimize the possibility for nucleotide hydrolysis to take place. The crystals were flash-cooled in propane directly from their crystallization buffer.
Data Collection and Structure Determination-An x-ray diffraction dataset was collected from a crystal of AMPPNP-actin at the BioCARS beamline 14-ID-B (Advanced Photon Source, Argonne, IL) ( Table I). The data was indexed and scaled with programs DENZO and SCALEPACK (28). Because the symmetry and unit cell parameters of the crystal of AMPPNP-actin were identical to those of ADP-actin (16), refinement was started from this structure with program CNS (29) and using all of the data between 24.0-and 1.85-Å resolution (no -cutoff was applied). Graphic inspection of the electron density maps (2F o Ϫ F c ) did not reveal any density for the region encompassing amino acids His-40 to Asp-51, which had been included during the first round of refinement.
These amino acids were removed from the model, and some minor changes were made to the structure. During the final stages of the refinement, alternative conformations were assigned to 12 amino acids for which the electron density map revealed more than one rotamer orientation. The important residue, Ser-14, located in the catalytic site, and its neighbor, Ser-33, which marks the beginning of subdomain 2, are among these 12 amino acids. In the case of Ser-14, the second rotamer corresponds to that in the ADP-actin structure. However, Ser-14 spends only 10 -15% of the time in that position, whereas the remaining 85-90% of the time, it is oriented as in the other ATP-actin structures (10 -15). The relative distribution of the two orientations of Ser-14 was determined independently from refinement of the occupancies of the two side chain rotamers and from inspection of difference Fourier electron density maps. A positive peak in the difference Fourier map, corresponding to the second rotamer of Ser-14, appears when its occupancy is estimated to be lower than 0.1 but is substituted by a negative peak for occupancies higher than 0.15. The final model resulted in R-factor and R-free values of 18.7 and 22.6%, respectively. Although the overall quality of the electron density was very good, subdomain 2 was less well ordered and the DNase I-binding loop, in particular, was fully disordered and could not be traced. The final structure (PDB entry 1NWK) includes amino acids Thr-6 to Arg-39, Ser-52 to His-371, the TMR probe, a molecule of AMPPNP, four Ca 2ϩ ions, and 256 water molecules.

RESULTS AND DISCUSSION
General Description of the Structure of TMR-Actin in the ATP State-Crystals of TMR-actin with bound ATP can be obtained under conditions similar to those for ADP-actin (16). However, nucleotide hydrolysis takes place during the time needed to grow these crystals, which is probably the result of the high salt concentration in the crystallization buffer. A way to circumvent this problem was to crystallize TMR-actin with AMP-PNP, a commonly used non-hydrolyzable ATP analog (25). The crystallization conditions for AMPPNP-actin, although similar to those for ADP-actin (16), had to be modified slightly (see "Experimental Procedures"). An analysis of the diffraction pattern from these crystals revealed unit cell parameters and symmetry identical with those of ADP-actin. However, a characteristic feature of the crystals of AMPPNP-actin is that they do not diffract the x-rays as strongly as those of ADP-actin. This is probably due to increased disorder in actin subdomain 2 (see below). After screening a large number of crystals at the BioCARS beamline 14-ID-B, a complete x-ray diffraction dataset was collected from one of the crystals that diffracted to the resolution of 1.85 Å ( Table I). Refinement of the structure converged to a R-factor value of 18.7% and a R-free value of 22.6%.
The position and geometry of AMPPNP ( Fig. 1) is almost indistinguishable from that of ATP in the ATP-bound actin structures (10 -15). Indeed, a superimposition of AMPPNPactin (henceforth referred to as ATP-actin) with all of the existing actin structures places the nucleotide and associated divalent cation in nearly identical positions ( Fig. 2) with the exception of one of the structures, that of the open state of the actin-profilin complex that contains ATP bound in an unusual manner (22) (discussed below). Moreover, a comparison of the ATP and ADP bound structures of monomeric TMR-actin reveals how the ADP moiety of the nucleotide and associated Ca 2ϩ ion remain bound after hydrolysis in precisely the same position, linked by a network of conserved interactions that are often mediated by water molecules (Fig. 3).
The structures of TMR-actin in the ADP (16) and ATP states are similar overall (Fig. 4A). However, important differences occur around the nucleotide ␥-phosphate site, Ser-14, the loop containing the methylated His-73, and subdomain 2 (Figs. 1, 3, and 4). Most remarkably, the DNase I-binding loop, which formed a well ordered ␣-helix in the ADP-actin structure (16), is now fully disordered in the ATP structure (Fig. 5). Although the crystals corresponding to the two nucleotide states are identical, the ATP-actin structure is more flexible overall but significantly more so in subdomain 2. The higher thermal mobility of the ATP structure is reflected by higher temperature factor values, which in this case can be compared directly because of the identity of the crystals (30). The conformational change upon P i release can be described as a sequence of events that originate at the nucleotide-binding site and the Ser-14 ␤-hairpin loop, propagating through the His-73 loop into subdomain 2.
The Ser-14 ␤-Hairpin Loop-In the structure of ATP-actin, the presence of the nucleotide ␥-phosphate forces the rotation of the side chain of Ser-14 into a common rotamer with that of all of the other ATP-actin structures (10 -15) but different from that of ADP-actin (Figs. 1B and 3) (16). However, Ser-14 is one of 12 amino acids in the structure that adopted two alternative side chain orientations. The second rotamer of Ser-14, which is only occupied ϳ10% of the time (see "Experimental Procedures"), corresponds to that observed in the ADP-actin structure, i.e. directed toward the ␤-phosphate of the nucleotide. The remaining ϳ90% of the time Ser-14 is oriented as in all of the other ATP-actin structures where it is hydrogen-bonded to an oxygen atom from the nucleotide ␥-phosphate and to the carbonyl oxygen of Gly-74 from the loop containing the methylated His-73 (Fig. 3A). It is unclear whether the second lower occupancy rotamer of Ser-14 results from the use of AMPPNP (in place of ATP) or whether this is a genuine feature of the ATP state. However, it is important to note that there is no evidence from inspection of the difference Fourier electron density maps that the second rotamer of Ser-14 results from partial occupancy of the ␥-phosphate moiety of the nucleotide, i.e. it is not the consequence of partial hydrolysis or nucleotide impurities.
The Sensor Loop-When the ATP-and ADP-bound structures of TMR-actin are superimposed, it becomes apparent that the two different orientations of Ser-14 result in two different conformations of the loop containing the methylated His-73 ( Fig. 4). Because of steric hindrance, the conformation of this loop in the ADP-actin structure, where it had moved toward the ␤-hairpin loop containing Ser-14, would have been inconsistent with the presence of the intact nucleotide and the resulting orientation of Ser-14 in the ATP structure. Therefore, there exists a direct correlation among the state of the nucleotide (ATP or ADP), the orientation of the side chain of Ser-14, and the conformation of this loop. Thus, we refer to this loop as the sensor loop. The sensor loop, Pro-70 to Asn-78, constitutes an insert between actin subdomains 2 and 1 and appears to function as a switch, linking changes in the nucleotide site to structural transitions in subdomain 2 (Fig. 4). Similar inserts exist in the actin-related proteins 2 and 3 (Arp2 and Arp3) (31) and in the bacterial actin homologue MreB (32). The conformation of the sensor loop in ATP-actin appears to be less stable than that in the ADP-actin structure, which is reflected by temperature factor values that are well above the average value for the rest of the structure. In contrast, in the ADP-actin structure, the loop displays temperature factors similar to those of the rest of the structure. One element that probably contributes to the stability of the loop in the ADP-actin structure (but which is missing in the ATP structure) is a stacking interaction between the side chains of Glu-72 and the methylated side chain of His-73 (Fig. 3). When the constraints imposed upon the loop in the ATP structure (because of the presence of the ␥-phosphate of the nucleotide and resulting orientation of the side chain of Ser-14) are removed upon P i release, the sensor loop moves toward the ␥-phosphate site. Identical changes in Ser-14 and the sensor loop were observed previously between the structures of ATP-and ADP-actin complexed with DNase I (10). However, because the ADP structure in this case was obtained following the hydrolysis of ATP within the crystals in which DNase I is tightly bound to actin subdomains 2 and 4, the changes in the sensor loop could not propagate into a conformational transition in subdomain 2. Such constraints do not exist in the structures of monomeric TMR-actin.
Subdomain 2 and the DNase I-binding Loop-The change in conformation of the sensor loop in the ATP structure is accompanied by a small ϳ4°rotation of subdomain 2. Note, however, that this rotation does not bring subdomain 2 entirely back to the orientation observed in the other ATP-actin structures (10 -15), which would have required a rotation of ϳ10°. A number of factors could help explain this small difference, including the use of AMPPNP in place of ATP, the binding of TMR to Cys-374, or the fact that this is the first ATP-actin structure to have been determined without any other protein bound to it. More significant, however, is the increased disorder in subdomain 2 in the ATP state. Indeed, as evidenced by the electron density maps and temperature factor values, subdomain 2 is far more mobile in the ATP-actin structure than in the ADPactin structure. For instance, witness the striking difference between the 2F o Ϫ F c electron density maps contoured around subdomain 2 (Fig. 5A) and around a more typical region of the structure such as the catalytic site (Fig. 1A). As previously observed in other ATP-actin structures (12)(13)(14)(15), the DNase I-binding loop within subdomain 2 becomes fully disordered with 12 amino acids (His-40 to Asp-51) undetermined in the final structure (Fig. 5A). The average temperature factor for the 25 amino acids (of 37) of subdomain 2 that remain visible in the structure is 42.2 Å 2 as compared with 32.5 Å 2 for the rest of the structure. In contrast, in the ADP state, subdomain 2 is well ordered and the DNase I-binding loop forms a stable ␣-helix (Fig. 5B). Therefore, one reason for the melting of the ␣-helix in the ATP structure appears to be the increased flex- d Free R-factor, R-factor calculated for a subset of the reflections (5%), which were omitted during the refinement and used to monitor its convergence.
ibility of subdomain 2, which in turn emanates from increased flexibility in the sensor loop. Another reason could be the disruption of a hydrophobic cluster that stabilizes the inner face of the ␣-helix in the structure of ADP-actin. In the ADP-actin structure, amino acids Val-43, Met-44, and Met-47 from the inner face of the ␣-helix form part of a hydrophobic cluster with Pro-38, Tyr-53, Ile-64, and Leu-65 in subdomain 2. In the ATP structure, the main chain carbonyl oxygen of Ile-64, which is connected to the sensor loop by a short ␤-strand, is rotated toward the core of this hydrophobic cluster. Such an orientation of the carbonyl of Ile-64 would be inconsistent with the formation of a ␣-helix in the DNase I-binding loop in the ATP structure. However, a comparison of the structures alone does not allow determining whether the rotation of the carbonyl at Ile-64 is a consequence of the melting of the ␣-helix in the DNase I-binding loop or whether, on the contrary, it causes the melting of the ␣-helix in response to the movement of the sensor loop.
The DNase I-binding loop can adopt a broad range of conformational states. In F-actin, this loop is thought to participate in intermonomer interactions (33,34). In the structure of the actin-DNase I complex, the loop is attached as an additional ␤-strand to a ␤-sheet in DNase I (10). In a number of ATP-actin structures, including the current structure of TMR-actin, the loop is disordered (12)(13)(14)(15). In contrast, in the structure of ADP-actin, the DNase I-binding loop is very stable, forming a ␣-helix (16). Such an array of conformational states is made possible by specific features of the amino acid sequence. For instance, three Gly residues at positions 42, 46, and 48 of the loop may account for its flexibility. The loop contains four hydrophobic amino acids, two valines (Val-43 and Val-45), and two methionines (Met-44 and Met-47). The presence of a sulfur atom within the unbranched side chain of methionine imparts this amino acid with uniquely high flexibility and polarizability (35). Thus, methionines are thought to play an important part in a number of protein-protein contacts by creating highly adaptable surfaces (36,37). The occurrence of two methionine residues, a relatively rare amino acid (2.3/100), in the loop may reflect the need for conformational adaptability. The DNase I-binding loop is also unique in that it is among the most exposed regions of the actin structure, yet together with the rest of subdomain 2 and the sensor loop, it is one of the most FIG. 1. Nucleotide binding site. A, all-atom-stereodiagram of the AMPPNP nucleotide analog bound in the catalytic site of actin. The 2F o Ϫ F c electron density map, contoured at 1.2 , is also shown. B, stereodiagram of a superimposition of the catalytic site region in the structures of AMPPNP-actin (cyan) and ADP-actin (red). The AMPPNP analog is colored according to atom type. Note how the change in nucleotide state leads to two different orientations of Ser-14 and the subsequent rearrangement of the loop containing the methylated His-73. Also shown are Ser-33, which marks the beginning of subdomain 2, and Arg-183 from subdomain 4, which changes conformation between the two nucleotide states so that it interacts with the His-73 loop in the ADP state but not in the ATP state.
conserved regions of the actin sequence (1).
It has been proposed that the stable ␣-helical conformation of the DNase I-binding loop in the ADP-actin structure is the result of crystal packing contacts (18,19). The current structure of TMR-actin in the ATP state demonstrates that this is not the case. The crystal parameters of TMR-actin in the ATP and ADP states are identical, yet their conformations differ, ruling out crystal packing contacts as a plausible explanation for the ␣-helix formation in the DNase I-binding loop. It has been also suggested that the ␣-helix in the DNase I-binding loop results from an allosteric effect because of the TMR modification (17,19). The fact that the two structures of monomeric actin contain the TMR modification also excludes this possibility.
The Structures of ATP-and ADP-bound Monomeric Actin Are in Agreement with the Biochemical Data-Subtilisin cleaves the DNase I-binding loop between Met-47 and Gly-48 (38). Substitution of ATP by ADP substantially decreases the rate of this reaction, which has been interpreted as evidence of a nucleotide-dependent conformational change in actin involving the DNase I-binding loop (3)(4)(5). Subdomain 2 of actin is also susceptible to hydrolysis by trypsin at Arg-62 and Lys-68. Although these two cleavage sites are outside the DNase Ibinding loop, their accessibility is also affected by the state of the nucleotide, becoming more accessible in the ADP state (3,4). Fluorescence spectroscopy also reveals the existence of a conformational change affecting the DNase I-binding loop upon P i release. Replacing ATP with ADP produces large changes in the emission properties of fluorescence probes attached to Gln-41 of the loop (6 -8).
Additional support for the existence of interdependence between the state of the nucleotide and the conformation of Ser-14, the sensor loop, and subdomain 2 comes from site-directed mutagenesis studies. The mutation of Ser-14 to Ala in yeast actin leads to decreased thermal stability, decreased affinity for ATP, decreased ATPase activity, and altered protease susceptibility in the DNase I-binding loop (39,40). In full agreement with the structural results described here, the protease digestion pattern of the Ser-14 to Ala mutant with bound ATP becomes similar to that of the ADP-bound state, whereas in wild-type actin, the two nucleotide states are characterized by different protease susceptibilities (39). In other words, the removal of the hydroxyl group of Ser-14 seems to break the coupling mechanism by which cleavage of the DNase I-binding loop becomes susceptible to the state of the nucleotide. Interestingly, the mutation of the residue equivalent to actin Ser-14 in heat shock cognate (Hsc) 70 (41) and BiP (42,43) also leads to decoupling of ATP binding from a conformational change. Mutation of Ser-14 to Cys in ␤-actin also results in decreased thermal stability and lower binding affinity for DNase I, which has been interpreted as evidence of a change in the interdomain relationship of actin, i.e. open versus closed (19,44). However, a different interpretation would be that this mutation affects the conformation of the DNase I-binding loop, thereby changing the binding affinity for DNase I (just as ADP-actin exhibits a lower affinity for DNase I than does ATP-actin).
His-73 occupies a special position within the sensor loop. This amino acid is absolutely conserved in actin, and most commonly it is also methylated, a rare posttranslational modification. However, the function of this modification remains unclear. His-73 mutagenesis studies attest to the importance of the sensor loop in determining the conformation of the DNase I-binding loop. Substitutions of His-73 by basic amino acids (Arg and Lys) have a general stabilizing effect on actin, whereas anionic or neutral substitutions are destabilizing (19,45). As revealed by limited proteolysis using trypsin, subtilisin, and ␣-chymotrypsin, mutants of His-73 are characterized by changes in the conformation of subdomain 2 and, in particular, the DNase I-binding loop (45). The effect of a His-73 to Ala mutation has been also investigated using an indirect assay that monitors the DNase I-inhibitory activity of actin (19). Based on the activity of actin, the affinity of this mutant for DNase I was estimated to be less than that of wild-type actin, which was interpreted as evidence that a change had occurred in the degree of opening of the cleft that separates the two major domains of actin. However, as for the Ser-14 to Cys mutant of ␤-actin described above (44), such an interpretation may be unwarranted since a change in the activity of actin in the presence of DNase I could be interpreted in a number of different ways, including the possibility for a conformational change limited to the DNase I-binding loop. Although these two studies on the mutants of His-73 (19,45) confirm the existence of a relationship between the sensor loop and the structure of subdomain 2, they do not specifically investigate the role of the nucleotide. However, since interactions of the loop (with Ser-14 and Arg-183 in the ATP and ADP structures, respectively) involve main chain atoms (Fig. 3) and not any specific side chain, mutants of this loop may not always be sensitive to the state of the nucleotide.
Nucleotide-dependent Conformational States of Actin-Similar to G proteins (46) and myosin (47), members of the actin superfamily (48) appear to undergo at least two major nucleotide-dependent conformational changes, one upon release of the nucleotide ␥-phosphate and a second one upon ADP release. Members of this superfamily include the Hsp70 molecular chaperones, hexokinase, the sugar kinases, the Arps (31), and the prokaryotic actin-like proteins MreB (32) and ParM (49). Crystal structures corresponding to their different nucleotide states are starting to reveal a general pattern. Structures with bound nucleotide (either ATP or ADP) are characterized by a  (12), vitamin D-binding protein (14), and the original actin-profilin complex (11). In actin, as in most phosphoryltransferases, the ADP moiety of the nucleotide and divalent cation remain bound after hydrolysis in a position that is nearly identical to that occupied by the ATP parent molecule. Shown in red is the so-called open-state structure of the actin-profilin complex (22) for which the location and conformation of the nucleotide differ from all of the other structures. Observe, in particular, how the divalent cation in this structure occupies the position of the ␤-phosphate in all of the other structures, breaking the entire network of interactions that characterizes the coordination of the nucleotide and divalent cation in the other actin structures (see also Fig. 3). As described in the "Results and Discussion" section, such a shift of the nucleotide has the effect of weakening the connection between the two major actin domains, allowing them to open apart (see also Fig. 6). The opening of the domains is reflected by a separation of the two homologous ␤-hairpin motifs on each side of the nucleotide to a distance of ϳ8 Å as compared with ϳ5 Å in all of the other structures.
closed conformation of the main interdomain cleft, whereas nucleotide-free structures are generally distinguished by an open cleft (Fig. 6). The reason for this is that the nucleotide and associated divalent cation bind at the base of the deep interdomain cleft, forming an elaborate network of hydrogen-bonding interactions (many of which are mediated by solvent molecules). This network of interactions helps to hold the two major domains together. Two homologous ␤-hairpin loops (actin residues 11 DNGSGLVK 18 and 154 DSGDGVTH 161 ), one from each of the two major domains on each side of the nucleotide, account for the majority of the interactions with the nucleotide and divalent cation (Figs. 2 and 3). Although some of these interactions are lost upon release of the nucleotide ␥-phosphate, the majority remains because of the presence of the ADP  4. The sensor loop. A, superimposition of the ATP-and ADP-bound structures of TMR-actin (cyan and red, respectively). Subdomain 2 rotates slightly between the two nucleotide states, although most of that rotation is in a direction perpendicular to that of the plane of the figure. The DNase I-binding loop within subdomain 2 is disordered in ATP-actin but becomes an ordered ␣-helix in the ADP-bound structure upon release of the nucleotide ␥-phosphate. B, close-up view of the area contoured by the black rectangle in part A. The loop containing the methylated His-73, which spans from residue Pro-70 to Asn-78, constitutes an insert between actin subdomains 2 and 1. Upon P i release, Ser-14 (partially covered by the sensor loop in this view) rotates and the polypeptide linkage between amino acids Glu-72 and His-73 of the loop flips around (see also Figs. 1B and 3), resulting in two different conformations of this loop. The two conformations of the loop translate into two different conformations of subdomain 2, more notably, the reorganization of the DNase I-binding loop from a disordered structure in the ATP state into an ordered ␣-helix in the ADP state. The nucleotide is only shown for one of the structures, that of AMPPNP-bound actin.   Ile-136 to Gly-146, green), which cannot be considered as belonging to either one of the two domains but which during opening of the cleft moves together with domain I. The only other connection between the two domains occurs in a loop centered at residue Lys-336. A, actin conformation in the ATP state as derived from the structure of AMPPNP-actin described here. For completeness, the missing DNase I-binding loop was modeled from the original structure of the actin-profilin complex (11) where the loop was seen in the absence of interactions with other proteins. B, actin conformation in the ADP state as observed in the structure of monomeric TMR-actin (16). The major changes upon P i release occur in the three regions shown in yellow (the Ser-14 ␤-hairpin loop, the sensor loop containing the methylated His-73, and the DNase I-binding loop). There is also a small rotation of subdomain 2 in a direction perpendicular to that of the plane of the figure (shown by an arrow). Note, however, that these changes do not lead to an open cleft, as interactions mediated by the ADP moiety of the nucleotide and divalent cation hold the two domains together. C, model of nucleotide-free actin built by homology with Arp3 (31). Indeed, although there is no structure of nucleotide-free actin available as yet, this state may be represented by the structures of nucleotide-free Arp2 and Arp3 (31), which share significant sequence similarity with actin. According to these two structures, as well as that of nucleotide free ParM (49), opening of the cleft can be described as a combination of two perpendicular rotations of ϳ12°each (indicated by the arrows). The ␣-helix between amino acids Ile-136 and Gly-146 shown in green serves as a hinge for the first of these two rotations (to the right) while the loop at the end of this helix rearranges slightly to accommodate the second rotation. moiety of the nucleotide and the divalent cation, which remain bound in nearly the same position (Fig. 3). However, once ADP dissociates, very little remains to keep the two major domains together, which then could (but not necessarily have to) open apart (Fig. 6C). As a result, ATP-actin is generally more stable than ADP-actin (50) and nucleotide-free actin denatures rapidly and irreversibly (51). What is the evidence from the crystal structures in support of two conformational transitions?
A well studied member of the actin superfamily is Hsc70 (a representative of the Hsp70 family). The structures of Hsc70 in the ATP and ADP states are both in a closed conformation (52). However, the structure of nucleotide-free DnaK (a bacterial homologue of Hsc70) bound to the nucleotide exchange factor GrpE reveals an open cleft (53). The binding of ATP to the ATPase domain of DnaK dissociates GrpE, presumably because of a conformational change that brings about the closure of the interdomain cleft. Thus, GrpE helps stabilize an open conformation that cannot occur when a nucleotide (either ATP or ADP) is bound to the catalytic site of DnaK. Although the crystal structures of ATP-and ADP-bound Hsc70 do not reveal any major conformational change (52), these two states are characterized, respectively, by low and high substrate affinity, suggesting that they are structurally different. In agreement with this finding, the existence of a conformational change upon release of the nucleotide ␥-phosphate has been demonstrated by solution x-ray-scattering experiments (54).
Another example is provided by the recently determined structures of the prokaryotic actin-like protein ParM (49). Two different structures of ParM were determined corresponding to the ADP-bound and nucleotide-free states. Although the nucleotide-free structure is characterized by an open cleft, that of ADP-ParM is in the closed conformation, very similar to ADPactin (16).
Even more closely related to actin are Arp2 and Arp3. The nucleotide-free structures of Arp2 and Arp3 both reveal an open cleft (31). Binding of nucleotide to Arp2 and Arp3 is predicted to lead to the closure of their respective clefts and the concomitant activation of the Arp2/3 complex (31).
An apparent exception to the rule described here is provided by the structure of nucleotide-free MreB, which is in the closed conformation (32). Therefore, although opening of the cleft is more likely to take place in the absence of a nucleotide, under certain conditions even a nucleotide-free cleft could remain closed. Interestingly, in MreB, a salt bridge between Lys-49 from subdomain 2 and Glu-204 from subdomain 4 helps stabilize the closed conformation. Two additional positively charged amino acids from subdomain 2 of MreB (Arg-63 and Arg-66) face negatively charged amino acids in subdomain 4 (Asp-180 and Glu-200). Such a charge balance between subdomains 2 and 4 does not exist in actin. Another feature that distinguishes MreB from actin is that the structural alignment of their respective polypeptide chains breaks apart toward the C terminus so that a ␣-helix in MreB (residues Lys-325 to Leu-332) occupies a position right in between subdomains 1 and 3. The presence of this ␣-helix adds to the number of contacts between the two major domains of MreB as compared with actin. Together, these structural features may help explain why nucleotide-free MreB is quite stable in solution while nucleotide-free actin is not (51).
The Open State Structure of the ATP-Actin-Profilin Complex Cannot Be Equated with That of ADP-Actin-As discussed above, by analogy with other members of the actin superfamily, nucleotide-free actin would be expected to be open. However, because of the general instability of nucleotide-free actin (51), no crystal structure of this state has been determined as yet that would confirm this view. Nevertheless, there exists one structure of actin with an open cleft, namely that of the socalled open state of the actin-profilin complex (22). Although this structure contains ATP bound, it has been interpreted by some as the true ADP-actin structure (18,21). To understand the meaning of this structure, one must first analyze how it was obtained. Similar to all of the other ATP-actin structures, the original actin-profilin structure in which the crystals had been stabilized in an ammonium sulfate-containing solution was characterized by a closed nucleotide cleft (11). But when the crystals that produced that closed cleft structure are stabilized in a different solution, containing 1.8 M potassium phosphate, their unit cell parameters change, giving rise to the open-state structure of the actin-profilin complex (22). Therefore, this is not a nucleotide-dependent but rather a solutiondependent conformational change. Moreover, ATP, which in the original closed-state structure occupied a position similar to that in all of the other ATP-actin structures (11), is shifted upwards from this position in the open-state structure (Fig. 2). Such a displacement of the nucleotide has not been observed for any other member of the actin superfamily. Possibly, the nucleotide in the open actin-profilin structure is being competed out of its site by the high concentration of free phosphate ions in the solution. Moreover, as currently modeled, the stereochemistry of the nucleotide and associated divalent cation in this structure (PDB code 1HLU) does not fall within the range of accepted values, suggesting poor definition in the electron density map. The displacement of the nucleotide and divalent cation from their genuine binding sites has the effect of breaking the linkage that they would normally exert between the two major actin domains, thereby allowing the cleft to open (Fig. 2), within the limits allowed by the crystal contacts.
It has been suggested that the two domains are not allowed to open apart in the structure of TMR-modified ADP-actin because of steric hindrance between the TMR probe and actin subdomains 1 and 3 (18). However, that ADP-actin is in a closed conformation is in full agreement with the fact that the nucleotide cleft is also closed in the ADP-bound structures of other members of the actin superfamily including ParM (49) and Hsc70 (52). Moreover, considering that TMR does not make any direct contact with subdomain 3, it is unlikely that it would interfere with the opening of the cleft. Profilin, on the other hand, makes extensive contacts with both subdomains 1 and 3 yet does not prevent the cleft from opening apart when the crystals are transferred into 1.8 M potassium phosphate (22). As it stands, the open-state structure of the actin-profilin complex has far more in common with a nucleotide-free structure than with either an ATP-or ADP-bound actin structure (Fig. 6). CONCLUSIONS This work and previous evidence (16) reveal a conformational change in actin upon P i release, culminating in a disordered loop to ordered ␣-helix transition in the DNase I-binding loop in subdomain 2. Such a conformational change would affect the monomer-monomer interface in F-actin (33,34), which may explain why ADP-actin dissociates more readily from the filament than ATP-or ADP-P i -actin. This change could also provide the basis for why certain actin-binding proteins, such as actin-depolymerizing factor/cofilin, specifically target ADP-actin for accelerated dissociation from the filament, thereby regulating treadmilling in the cell (2). It is also concluded that the conformational change upon P i release does not lead to an open nucleotide cleft in G-actin. Neither does the cleft appear to be open in the actin filament. Indeed, Holmes et al. 4 have recently used the structure of ADP-actin (16) to refine a 6-Å resolution model of the actin filament obtained from fiber diffraction. This is the highest resolution structure of F-actin so far available, and it does not reveal an open actin cleft. Because the structures of nucleotide-free Arp2, Arp3 (31), ParM (49), and DnaK (53), as well as nucleotide-free and glucose-free hexokinase (55), all display an open cleft, it is proposed here that nucleotide-free actin will be open as well. Although in vitro nucleotide-free actin is unstable, in the cell this open-cleft state may be stabilized by actin-binding proteins, such as profilin, that promote nucleotide exchange during treadmilling.