Crystal structure of activated CheY. Comparison with other activated receiver domains.

The crystal structure of BeF(3)(-)-activated CheY, with manganese in the magnesium binding site, was determined at 2.4-A resolution. BeF(3)(-) bonds to Asp(57), the normal site of phosphorylation, forming a hydrogen bond and salt bridge with Thr(87) and Lys(109), respectively. The six coordination sites for manganese are satisfied by a fluorine of BeF(3)(-), the side chain oxygens of Asp(13) and Asp(57), the carbonyl oxygen of Asn(59), and two water molecules. All of the active site interactions seen for BeF(3)(-)-CheY are also observed in P-Spo0A(r). Thus, BeF(3)(-) activates CheY as well as other receiver domains by mimicking both the tetrahedral geometry and electrostatic potential of a phosphoryl group. The aromatic ring of Tyr(106) is found buried within a hydrophobic pocket formed by beta-strand beta4 and helix H4. The tyrosine side chain is stabilized in this conformation by a hydrogen bond between the hydroxyl group and the backbone carbonyl oxygen of Glu(89). This hydrogen bond appears to stabilize the active conformation of the beta4/H4 loop. Comparison of the backbone coordinates for the active and inactive states of CheY reveals that only modest changes occur upon activation, except in the loops, with the largest changes occurring in the beta4/H4 loop. This region is known to be conformationally flexible in inactive CheY and is part of the surface used by activated CheY for binding its target, FliM. The pattern of activation-induced backbone coordinate changes is similar to that seen in FixJ(r). A common feature in the active sites of BeF(3)(-)-CheY, P-Spo0A(r), P-FixJ(r), and phosphono-CheY is a salt bridge between Lys(109) Nzeta and the phosphate or its equivalent, beryllofluoride. This suggests that, in addition to the concerted movements of Thr(87) and Tyr(106) (Thr-Tyr coupling), formation of the Lys(109)-PO(3)(-) salt bridge is directly involved in the activation of receiver domains generally.

Two-component signal transduction systems control a variety of cellular processes, including chemotaxis and expression of some genes in bacteria and lower eukaryotes (1)(2)(3)(4). Signal transduction is mediated by phosphotransfer from a histidine kinase to a conserved aspartyl residue of a response regulator. To date, more than 300 response regulators have been identified based on homology in a domain of ϳ120 residues, commonly referred to as a receiver or regulatory domain (5). The structures of several receiver domains have been solved (4). They all have a similar (␤/␣) 5 fold with an active site comprised of five highly conserved residues, including three aspartates, a lysine, and either threonine or serine (6).
CheY, the response regulator of bacterial chemotaxis, has served as the model for understanding phosphorylation-induced activation of response regulators (7,8). Early biochemical, genetic, and structural studies on CheY indicate that phosphorylation induces a structural change from an inactive to an active conformation (for a review, see Ref. 6). The five conserved active site residues were shown to be important for phosphorylation and/or conformational changes subsequent to phosphorylation. Asp 57 was established as the site of phosphorylation (9) and, along with Asp 12 and Asp 13 , is required for magnesium binding (10 -12). While Thr 87 and Lys 109 were implicated in post-phosphorylation events (10,(13)(14)(15), it is only recently that their roles have been defined (see below). Finally, the rotameric position of Tyr 106 , another conserved residue adjacent to the active site, was thought to be correlated with the signaling state of CheY (16). Although the importance of these residues has been noted in numerous studies of inactive and mutant forms of CheY as well as other response regulators (17,18), a detailed understanding of the mechanism of the phosphorylation-induced transformation from an inactive to active conformation could not be reached because of the short half-life of the aspartyl-phosphate linkage (half-life of a few seconds to hours).
We have shown through biochemical and structural studies that BeF 3 Ϫ forms persistent complexes with receiver domains, mimicking the phosphorylation-activated states (19). For example, similar to phosphorylated CheY (P-CheY), BeF 3 Ϫ -CheY shows enhanced binding to the N-terminal 16 residues of its target, FliM, enhanced affinity for CheZ, and decreased affinity for CheA (19,20). Our recent NMR structure of BeF 3 Ϫ -CheY (20) revealed many aspects of the structural changes induced upon activation. The hydroxyl group of Thr 87 forms a hydrogen bond with an active site acceptor, presumably BeF 3 Ϫ -Asp 57 , and the side chain of Tyr 106 is restrained in a buried conformation. Unfortunately, the positions of the BeF 3 Ϫ moiety, magnesium cation, and the side chain conformation of Lys 109 could not be defined by the NMR data.
Recently, the crystal structures of the phosphorylated forms of two other response regulators, Spo0A r (21) and FixJ r (22), have been reported (superscript r denotes receiver domain). Spo0A r was unknowingly crystallized in the phosphorylated state with calcium as the divalent metal rather than magnesium, and FixJ r was crystallized in the absence of a divalent metal ion to circumvent problems associated with hydrolysis of the phospho-aspartate. It is known that removal of magnesium from phosphorylated CheY does not alter its enhanced affinity for FliM (23). This suggests that, while magnesium is important in the chemistry of phosphorylation of receiver domains, it is probably not required for stabilizing the active conformations. Thus, the structures of P-Spo0A r and P-FixJ r do likely represent the phosphorylation-activated states, although the activities of these two proteins in the conditions used for crystallization cannot be directly assessed. Importantly, the residues homologous to Thr 87 and Tyr 106 in both structures adopt similar conformations to those seen the NMR structure of BeF 3 Ϫ -CheY. We have recently determined the crystal structures of BeF 3 Ϫ -CheY and BeF 3 Ϫ -CheY complexed with a 16-residue peptide derived from the N terminus of FliM. The binding interactions of the CheY-peptide complex have been discussed (24). Herein we report the crystal structure of BeF 3 Ϫ -CheY complexed with the divalent cation manganese solved at 2.4-Å resolution. A detailed comparison of the active sites of BeF 3 Ϫ -CheY and P-Spo0A r (21) clearly shows that BeF 3 Ϫ -aspartate activates receiver domains by reproducing the geometry and electrostatic potential of a phospho-aspartate. Indeed, all of the active site interactions in BeF 3 Ϫ -CheY are identical to those in P-Spo0A r , indicating that the structural changes induced by BeF 3 Ϫ activation of response regulators are the same as those induced by phosphorylation. We also show, through a comparison of backbone coordinates of BeF 3 Ϫ -CheY with inactive magnesiumbound CheY, that activation results in only relatively small structural differences, except in loops, and that these differences are similar in magnitude to those observed between inactive and phosphorylated FixJ r (22).

MATERIALS AND METHODS
Escherichia coli-CheY was overexpressed and purified as described previously (20,25) and was prepared as a solution containing 2 mM CheY, 8 mM BeCl 2 , 50 mM NaF, and 4 mM MnCl 2 at pH of 8.4. Crystals of the complex were obtained at room temperature using the hangingdrop vapor diffusion method using a well solution containing 1.8 M ammonium sulfate, 5-10% glycerol, and 100 mM Tris (pH 8.4). The crystallization droplets contained the CheY solution mixed with an equal volume of the well solution. Crystals appeared after 1 day and grew to ϳ0.7 ϫ 0.4 ϫ 0.4 mm after 3 days. The concentration of glycerol in the well solution was increased (5% at each step) every 2 days to a final concentration of 25 volume %. The glycerol was added as a cryoprotectant to allow freezing for data acquisition. The protein crystallized in space group P2 1 2 1 2 1 with unit cell dimensions a ϭ 53.5 Å, b ϭ 53.8 Å, and c ϭ 161.3 Å with two molecules in the asymmetric unit.
The diffraction data were collected at 100 K on the Mar345 detector at the Stanford Synchrotron Radiation Laboratory using beamline 7-1 (wavelength of 1.08 Å). A crystal detector distance of 180 mm was used to collect data to 2.37-Å resolution. The data set was integrated and scaled to 2.37-Å resolution, using DENZO and SCALEPACK (26).
Cystal Structure of BeF 3 Ϫ -CheY-Initial phases were determined by molecular replacement (MR) using AMoRe (27). The structure of BeF 3 Ϫ -CheY from a BeF 3 Ϫ -CheY-N16-FliM complex (Protein Data Bank accession code 1F4V) was used as the search model with Tyr 106 replaced by alanine. Two solutions were easily found with a correlation coefficient of 67% and an R-factor of 34% (20 -3.5 Å). The MR 1 model was refined by several rounds of simulated annealing and B-group refinements using CNS (28) to an R-factor and R free of 29.5% and 31.4%, respectively. Structure factors (F c and ⌽ c ) were calculated from this partially refined MR model using SFALL of the CCP4 package (29). These phases were improved by 20 rounds of 2-fold noncrystallographic symmetry averag-ing and solvent flattening at 2.4 Å using the RAVE package (30), with a mask made from the MR solution and operators from the MR solution refined by the IMP program using the 2F o Ϫ F c MR map. In each round the map was calculated using coefficients 2F o Ϫ F c , with F c and ⌽ c calculated from the density-modified map of the previous cycle.
An unbiased (2F o Ϫ F c ) map calculated with phases and F c from the final symmetry-averaged, solvent-flattened map was displayed using the graphics program O (31) and used as a guide in modeling Tyr 106 , positioning Mn 2ϩ ions, and manually modifying the model in places where it did not fit the electron density. Refinement was performed using CNS (28). Anisotropic B-factor and bulk solvent corrections as well as the cross-validation method (32) were applied throughout the refinement. 15.0 to 2.37 Å data were included in the refinement with tight noncrystallographic symmetry restraints (300 kcal mol Ϫ1 Å 2 ). Water picking was performed after the R-factor/R free dropped to 24%/ 27% using CNS.
The electron density for beryllofluoride on Asp 57 was clearly seen (10 ) in the resulting CNS F o Ϫ F c SIGMAA weighted map. This moiety was modeled on both protomers and refinement continued giving a final R-factor and R free of 21.0/24.0%. Geometric parameters for the structure were monitored using PROCHECK (33) and WHAT_CHECK (34).

BeF 3
Ϫ -CheY crystals were grown in the presence of manganese (Mn 2ϩ ). Although in vivo CheY is complexed with magnesium (Mg 2ϩ ), NMR studies have shown that the active site readily accommodates larger divalent cations (12,35). Given that Mn 2ϩ has the same coordination geometry as Mg 2ϩ and supports phospho-transfer from CheA-P to CheY (11), we expect that it does not perturb the active structure significantly. BeF 3 Ϫ -CheY complexed with Mn 2ϩ crystallized in the space group P2 1 2 1 2 1 and diffracted to 2.4 Å. The two molecules in the asymmetric unit form a noncrystallographic symmetric dimer ( Fig. 1) similar to that seen for P-FixJ r (22), with helix H4 of one molecule packing against the H4-␤5-H5 face of the second molecule. For CheY, dimer formation in the crystal must be due to lattice packing forces and is not biologically relevant, because in solution BeF 3 Ϫ -CheY remains monomeric even at 3 mM protein concentrations. Refinement statistics are summarized in Table I.
The overall crystal structure of BeF 3 Ϫ -CheY retains the (␤/␣) 5 fold of receiver domains (6) (Fig. 1) and is very similar to the NMR structure of BeF 3 Ϫ -CheY (20) as well as the crystal structure of inactive Mg 2ϩ -bound CheY (36). Superposition of C␣ coordinates (residues 6 -125) of the BeF 3 Ϫ -CheY x-ray tructure with the mean BeF 3 Ϫ -CheY NMR structure and the Mg 2ϩbound CheY x-ray structure yielded root mean square deviations of only 1.2 and 0.8 Å, respectively. Differences in C␣ coordinates between BeF 3 Ϫ -CheY and the crystal structure of inactive Mg 2ϩ -bound CheY, based on a superposition of the residues least affected by activation, are shown in Fig. 2a. The biggest changes are observed in the ␤4/H4 loop, the ␤5/H5 loop, and the N terminus of helix H5. The significance of the changes in the ␤4/H4 loop are particularly hard to interpret, because it adopts different conformations in the various crystal structures 1 The abbreviation used is: MR, molecular replacement.

Crystal Structure of BeF 3
Ϫ -CheY 16426 of inactive CheY (6). Indeed, dynamics studies of Mg 2ϩ -bound CheY showed that this region is flexible in solution (37), and a superposition of the x-ray structure of BeF 3 Ϫ -CheY with the NMR structures of Mg 2ϩ -bound CheY shows that the ␤4/H4 loop of the active (x-ray) structure falls on the edge of the bundle formed by the inactive (NMR, Mg 2ϩ -bound) structures. Rather than a conformational change, we prefer to view the activation-induced changes in the ␤4/H4 loop as a stabilization of the active conformation that may be sampled by the inactive protein. Unfortunately, it is hard to make similar conclusions for the ␤5/H5 loop, because the relaxation data for residues in this loop could not be reliably interpreted due to complications caused by chemical exchange of Mg 2ϩ in the active site (37).
Active Site of BeF 3 Ϫ -CheY-From the NMR structure of BeF 3 Ϫ -CheY we determined that the switch from an inactive to an active conformation involves hydrogen bond formation between the hydroxyl group of Thr 87 and an active site residue, presumably BeF 3 Ϫ -Asp 57 . As a consequence of, or in conjunction with, formation of this hydrogen bond, ␤-strand ␤4 (along with Thr 87 ) is displaced, and the aromatic ring of Tyr 106 becomes buried in a hydrophobic pocket between helix H4 and ␤5. However, the NMR data did not define the positions of either the BeF 3 Ϫ moiety or the divalent cation. In addition, the NMR data for Lys 109 , a residue known to be critical for switching to the active conformation (10), were insufficient to define the position of the side chain in the active site accurately. The BeF 3 Ϫ -CheY crystal structure verifies the previous conclusions and extends the detail in the active site (Fig. 3). The hydroxyl group of Thr 87 does hydrogen-bond with one of the fluorine atoms of the BeF 3 Ϫ moiety that is bonded to Asp 57 O␦ (O␦-Be distance 1.5 Å) in a tetrahedral configuration. The aromatic ring of Tyr 106 is seen exclusively in the buried position, stabilized in this rotameric conformation by a hydrogen bond that was not previously identified in the NMR studies between the tyrosine hydroxyl group and the backbone carbonyl oxygen of Glu 89 as well as hydrophobic interactions. The divalent cation (Mn 2ϩ ) is located adjacent to Asp 57 -BeF 3 Ϫ in the crystal structure and is coordinated by Asp 13 O␦, Asp 57 O␦, backbone carbonyl oxygen of Asn 59 , a fluorine atom, and two water molecules. Finally, the side chain of Lys 109 forms a salt bridge with BeF 3 Ϫ and Asp 12 O␦.
Active Site Comparisons with P-Spo0A r , P-FixJ r , and Phosphono-CheY-Comparison of the BeF 3 Ϫ -activated CheY active site with those of phosphorylated receiver domains determined to high resolution, including P-Spo0A r (21), P-FixJ r (22), and phosphono-CheY (38), provides a structural basis by which BeF 3 Ϫ mimics the phosphoryl group (Fig. 3). Of these, BeF 3 Ϫ -CheY is best compared with P-Spo0A r because both contain a divalent cation, hexavalent Mn 2ϩ , and heptavalent Ca 2ϩ , respectively, in the active site. Similar to the phospho-aspartate in P-Spo0A r , BeF 3 Ϫ -aspartate acts as a ligand to the divalent metal atom, forms a salt bridge with Lys 109 N, and hydrogen bonds with Thr 87 O␥ and the backbone amides of Trp 58 , Asn 59 , and Ala 88 . The measured distances for the common interactions in the structural models of BeF 3 Ϫ -CheY and P-Spo0A r are within coordinate uncertainties (ϳ0.3 Å) (Table II). It appears that, although calcium has an extra coordination site relative to manganese (and magnesium), which is occupied by a water molecule in P-Spo0A r , the extra ligand is accommodated without requiring a significantly different active site geometry.
Although lacking a divalent cation, the distances for the analogous interactions in P-FixJ r are also similar to those seen for BeF 3 Ϫ -CheY and P-Spo0A r ( Table II). The only exception is the large distance (4.1 Å) between Lys 109 N and Asp 12 O␦ in P-FixJ r , indicating that this salt bridge is broken in the ab-

FIG. 2. Activation-induced C␣ coordinate changes (active-inactive) for CheY (a), FixJ r (b), and phosphono-CheY (c). For
CheY, a ␦-distance plot comparing crystal structures of Mg 2ϩ and BeF 3 Ϫ -CheY showed that residues 5-55 and 65-84 were the least affected by activation. These residues were used to superimpose the Mg 2ϩ and BeF 3 Ϫ -CheY structures from which changes in C␣ positions were calculated. For FixJ r , residues least influenced by phosphorylation (residues 1-8, 16 -52, and 106 -122) were used to superimpose Mn 2ϩ -bound and phosphorylated (no metal) structures from which changes in C␣ positions were calculated. For phosphono-CheY, residues least influenced by the phosphono group (residues 5-50) were used to superimpose Mg 2ϩ -CheY and phosphono-CheY (no metal) structures. The horizontal line in each plot denotes the overall backbone root mean square deviation for each pair of structures.

Crystal Structure of BeF 3
Ϫ -CheY sence of metal. It is interesting to note that, although a divalent cation is necessary for the chemistry of phosphorylation and dephosphorylation of CheY, removal of the metal after phosphorylation apparently does not alter the affinity of P-CheY for FliM (23). Similarly, the fact that P-FixJ r purifies as a dimer in the absence of metal, consistent with its activated state, suggests that the metal is not required for inducing the active conformation of FixJ r . This may be a general feature of receiver domains.
In phosphono-CheY, except for the salt bridge between Lys 109 N and a phosphonate oxygen, the distances measured for the analogous hydrogen bonds are outside of the acceptable range (2.5-3.1 Å) ( Table II). The absence of these interactions leads to much smaller changes in the ␤4/H4 and ␤5/H5 loops (Fig. 2c). The modest structural differences relative to inactive CheY appear to be consistent with the partial activity of phosphono-CheY, which shows an 8-fold increase in affinity for N16-FliM (38), whereas BeF 3 Ϫ -and phosphorylation-activated CheY show a 25-fold increase in affinity (19,39). Considering that the S␥-C␦ bond in phosphono-cysteine is only 0.5 Å longer than the C␥-O␦ bond in phospho-aspartate, it is surprising that the phosphonate analog does not better activate CheY. Since the salt bridge formed by Lys 109 N and an active site partner (BeF 3 Ϫ , PO 3 Ϫ , phosphonate) is the only common interaction in P-Spo0A r , P-FixJ r , BeF 3 Ϫ -CheY, and phosphono-CheY, it appears to be an important part of the active site interactions that together induce the fully active conformation. Based on crystal structures of mutant forms of CheY, it was previously  Table II. c, stereo view of active site residues for BeF 3 Ϫ -CheY(Mn 2ϩ ) (blue), phosphorylated FixJ r (no metal) (lime), and phosphorylated Spo0A r (Ca 2ϩ ) (copper). Mn 2ϩ and Ca 2ϩ are shown as red and green balls, respectively. Residue numbers are based on E. coli CheY. For clarity, phosphono-CheY was not included.

Crystal Structure of BeF 3
Ϫ -CheY suggested that Lys 109 plays a role in positioning the ␤5/H5 loop (40,41). Activation-induced Conformational Changes-CheY and FixJ r are the only receiver domains that have been solved with sufficient resolution in both active (22) and inactive (36,42) states to allow a detailed comparison of activation-induced structural changes. The largest activation-induced C␣ coordinate changes for both proteins occur in loop regions, particularly the ␤4/H4 loop. In addition, the ␤5/H5 loop shows significant displacement in CheY, potentially due to the Lys 109 N-BeF 3 Ϫ salt bridge, but the analogous conformational change is not seen in FixJ r . As stated previously, for inactive CheY the ␤4/H4 loop is conformationally flexible according to solution NMR studies (37). Activation results in the formation of a new hydrogen bond between the hydroxyl of Tyr 106 and the backbone carbonyl of Glu 89 . This likely helps to stabilize the active conformation of this loop. Thus, a comparison of just the active and inactive CheY crystal structures could lead one to conclude that activation induced dramatic conformational changes in the ␤4/H4 loop. However, in light of the NMR data, the exact magnitude of this change is hard to quantify. In FixJ r the residue homologous to Tyr 106 is a phenylalanine, which cannot stabilize the ␤4/H4 loop through a side chain-backbone hydrogen bond. It would be interesting to determine whether this loop in FixJ r is also conformationally flexible in the inactive state and becomes stabilized upon activation.
Even though the loops show significant activation-induced changes, activation of CheY and FixJ r does not result in any major structural rearrangements. Whereas some ␤-strands and ␣-helices are slightly displaced, the actual residues that define these elements of secondary structure remain unchanged in both proteins. In both BeF 3 Ϫ -CheY and P-FixJ r the N terminus of H4 moves slightly upon activation, and in CheY there is also a small displacement of the N terminus of H5. Indeed, even when compared as a group (Fig. 4), including P-Spo0A r , there are no dramatic structural differences among either the active or inactive forms of the receiver domains.
Although there are small differences in the tilt and inclination of the helices, these differences do not give rise to changes in atomic coordinates of more than a few Ångstroms.
The structures of Spo0A r and NtrC r have also been determined in both the active (21, 43) and inactive states (44,45). It was difficult to analyze the activation-induced structural changes for Spo0A r , because the inactive form crystallized as a domain-swapped dimer, the biological relevance of which is unclear. In contrast to CheY and FixJ r , the low resolution NMR structures of active and inactive NtrC r show major structural differences, especially for residues that define helix H4. A higher resolution structure of BeF 3 Ϫ -activated NtrC r will more clearly define these changes.
Conclusions-The comparable interactions in the active sites of BeF 3 Ϫ -CheY and P-Spo0A r indicates that BeF 3 Ϫ -aspartate is almost a perfect structural mimic of phospho-aspartate. In conjunction with our previous biochemical data that show functional activation of receiver domains with BeF 3 Ϫ (19,20), it appears that beryllium fluoride is a convenient tool that can be applied to biochemical as well as structural studies of a host of response regulators.
Given the high sequence conservation, it is perhaps not surprising that the structures of P-Spo0A r , P-FixJ r , and BeF 3 Ϫ -CheY all show similar interactions in the active site. Based on a comparison of these structures we can begin to define a general mechanism of activation. As predicted by previous biochemical and genetic studies on CheY, the hydroxyl group of Thr 87 (Thr 84 Spo0A r , Thr 82 FixJ r ) (13) and the side chain of Lys 109 (Lys 106 Spo0A r , Lys 104 FixJ r ) (10) form what appears to be critical active site interactions with the phosphoryl group (or BeF 3 Ϫ ). As a consequence of, or in conjunction with, these interactions, Tyr 106 (Phe 103 Spo0A r , Phe 101 FixJ r ) adopts a buried conformation. Just how general these three events are in the transition from an inactive to an active conformation and how they affect the overall structures of receiver domains generally remains to be determined.
Comparison of just crystal structures suggests that there is a coupling between phosphorylation of CheY and FixJ with structural changes, especially in the ␤4/H4 loop. The extent to which phosphorylation induces an actual conformational change versus a stabilization of the active state from a preexisting equilibrium between the active and inactive conformations in solution is not clear. Positive evidence for the idea of stabilization comes from NMR studies of inactive, constitutively active mutant forms (46), and phosphorylated NtrC (43), which indicate that phosphorylation stabilizes the active conformation (47). Additional NMR and x-ray studies of active and inactive forms of response regulators will help to clarify this issue and will help define how phosphorylation-induced conformational changes ultimately regulate the diverse processes controlled by two-component signal transduction.