The Crystal Structure of a Novel, Inactive, Lysine 49 PLA2 from Agkistrodon acutus Venom

The crystal structure of acutohaemolysin, a lysine 49 phospholipase A2 protein with 1010 non-hydrogen protein atoms and 232 water molecules, has been determined ab initio using the program SnB at an ultrahigh resolution of 0.8 Å. The lack of catalytic activity appears to be related to the presence of Phe102, which prevents the access of substrate to the active site. The substitution of tryptophan for leucine at residue 10 interferes with dimer formation and may be responsible for the additional loss of hemolytic activity. The ultrahigh resolution of the experimental diffraction data permits alternative conformations to be modeled for disordered residues, many hydrogen atoms to be located, the protonation of the Nϵ2 atom in the catalytic residue His48 to be observed experimentally, and the density of the bonding electrons to be analyzed in detail.

The crystal structure of acutohaemolysin, a lysine 49 phospholipase A2 protein with 1010 non-hydrogen protein atoms and 232 water molecules, has been determined ab initio using the program SnB at an ultrahigh resolution of 0.8 Å. The lack of catalytic activity appears to be related to the presence of Phe 102 , which prevents the access of substrate to the active site. The substitution of tryptophan for leucine at residue 10 interferes with dimer formation and may be responsible for the additional loss of hemolytic activity. The ultrahigh resolution of the experimental diffraction data permits alternative conformations to be modeled for disordered residues, many hydrogen atoms to be located, the protonation of the N⑀2 atom in the catalytic residue His 48 to be observed experimentally, and the density of the bonding electrons to be analyzed in detail.
Phospholipase A2 (PLA2, 1 EC 3.1.1.4) is an enzyme that hydrolyzes the sn-2 fatty aryl bond of phospholipids to produce free fatty acid and lysophospholipids (1). Although PLA2s from different species share similar structures, they show diverse pharmacological and toxicological functions, including myotoxicity, anticoagulant activity, hemolyticity, and lipolysticity (2,3). Accurate atomic resolution crystal structures are required to answer questions concerning the possible structural basis for these biological functions. According to conventional wisdom, the catalytic activity of PLA2s depends on the binding of a calcium ion and on the presence of residue Asp 49 (4). (The conventional numbering system of Renetseder et al. (5) is used throughout this report.) The substitution of other residues for Asp 49 eliminates the binding of calcium ion and is thought to abolish catalytic activity (6). However, the side-chain N atom of lysine residues can play a role similar to that of a calcium ion, thereby enabling Lys 49 PLA2s to display noticeable catalytic activity (7,8). Lys 49 PLA2s are abundant in snake venoms from various species, and they share highly homologous sequences and three-dimensional structures as evidenced by four crystal structures determined at a resolution of 2.0 Å or lower (2, 9 -11).
Acutohaemolysin is a Lys 49 PLA2, isolated from the venom of Agkistrodon acutus, that lacks both catalytic and hemolytic activity. The crystal structure determination of this protein was undertaken to investigate the structural basis of its inactivity, and the results indicate that specific substitutions at residues 10 and 102 may be responsible for this loss. Because the acutohaemolysin crystals diffracted extremely well, it was possible to measure x-ray data at the ultrahigh resolution of 0.80 Å and to solve the structure using the ab initio dual-space direct method that is known as Shake-and-Bake (12) and implemented in the computer program SnB (13,14). This structure, which contains ϳ1250 non-hydrogen atoms, is one of the largest ever determined by direct methods. The atomic coordinates for acutohaemolysin have been deposited in the Protein Data Bank with accession code 1MC2.
In recent years, several methodological improvements, including cryogenic crystal mounting techniques, bright synchrotron x-ray sources, and highly sensitive area detectors, have made the measurement of ultrahigh resolution diffraction data more feasible. This is important because high resolution studies provide much more detailed and accurate information than that obtainable from structures determined in the range of 2-3 Å. For example, in the acutohaemolysin case, each atom can be described by nine parameters, including three coordinates and six thermal factors, that express the anisotropic nature of the thermal motion, and this permits the solvent molecules to be divided into two hydration shells. In addition, atomic resolution difference electron density maps allow the positions of a large number of hydrogen atoms, including those involved in interactions among active-site residues, to be located (15). Moreover, alternative or multiple conformations of some disordered residues can be resolved and modeled.
When ultrahigh resolution (e.g. better than 0.9 Å) crystallographic diffraction data are available, the deformation of the atomic electron density due to chemical bonding can be observed (16). Following a refinement using a spherical atom model, residual maps may reveal additional information about chemical bonding and electron transfer between atoms. At subatomic resolutions (d ϳ 0.5 Å), the experimental molecular charge density can be precisely refined using a multipolar electron density model (17). Multipolar density can be decomposed into three contributions: the core spherical density, the valence spherical density, and the valence non-spherical density. Such information is important for deriving the overall electrostatic potential distribution in protein structures (18 -22), for elucidating the mechanisms of biological processes involving electron transfers between catalytic atoms, and for determining the oxidation states of active metal ions in metalloproteins (23).
It is often asserted that charge density analyses are restricted to diffraction data from crystals of small molecules (24). Recently, however, it has been demonstrated that charge density refinement for biological macromolecules can also be carried out if the diffraction resolution is high enough and if the atomic thermal motion is sufficiently low. For example, the application of charge density refinement to the structure of a scorpion protein at 0.96-Å resolution showed visible bonding density between some main-chain atoms (21). The best example of charge density analysis applied to biological macromolecules, however, is provided by exhaustive refinement, at the ultrahigh resolution of 0.54 Å, of the valence electron distribution in crambin, a small protein with 46 residues (22). Herein we report the results of a charge density analysis applied to acutohaemolysin.

EXPERIMENTAL PROCEDURES
Crystallization and Diffraction Data Collection-Acutohaemolysin was purified and crystallized using previously reported procedures (25). Crystals with a size of ϳ0.8 ϫ 0.8 ϫ 1.0 mm were soaked in a 0.05 M sodium citrate solution (pH 5.6) containing 18%(v/v) isopropanol, 18% (w/v) polyethylene glycol 4000, and 5% (v/v) glycerol and then subjected to cryogenic flash freezing. Diffraction data were collected from cryoprotected crystals, under a nitrogen stream at a temperature of 100 K, at beamline 19-ID of the Advanced Photon Source (Argonne). The wavelength of the incident x-rays was set to 0.92 Å to meet the requirements for ultrahigh resolution data collection. A total of 600 frames was recorded from one crystal with different exposure time to cover all of the reciprocal space at both low and high resolution. Data processing was performed using the program HKL2000 (26), and the statistics of data collection and processing are summarized in Table I. The cryo-protected crystals of acutohaemolysin belong to space group C2 with unit cell dimensions a ϭ 44.743 Å, b ϭ 59.100 Å, c ϭ 45.319 Å, and ␤ ϭ 117.421°. These dimensions are slightly shorter than those observed under room temperature conditions. The agreement factor R merge and the data completeness were 5.5% and 76.7%, respectively, for all resolution shells, and the corresponding values were 27% and 34.7% for the highest resolution shell (0.83-0.80 Å). An average isotropic thermal factor (B) of 5.6 Å 2 , a relatively small value for a protein data set, was estimated from a Wilson plot using data in the range 4.0 -0.8 Å.
Phasing by Direct Methods-Initial phases were obtained using the direct methods program SnB version 2.1 (14). First, normalized structure-factor magnitudes ( E values) in the resolution range 8 -0.8 Å were calculated using the LEVY/EVAL programs in the SnB package (27), and then 107,100 triplet structure invariants were generated using the 10,710 strongest E values. Next, trial structures consisting of 300 randomly positioned atoms (S17, C283) were created, and each of them was subjected to 1,071 cycles of dual-space Shake-and-Bake refinement that included parameter shift optimization of the minimal function (R min ) and selection of 300 new peaks in each cycle (12). Three figures of merit (i.e. R min , a crystallographic R factor, and a correlation coefficient between E obs and E calc (28)) were calculated for each trial, and a histogram of the R min values examined following completion of refinement for the 16th trial suggested that a solution had been found. After 140 cycles, a trace of the R min values versus SnB cycle number showed a sudden drop (characteristic of a solution) resulting in a decrease of R min from an initial value of 0.48 to a stable value of 0.439. Trial 16 was refined for 107 more dual space cycles while selecting the 964 strongest peaks in each cycle, and a final 107 cycles of Fourier refinement alone were performed while slowly increasing the number of reflections phased from 10,710 to 43,457 (i.e. all reflections with E were Ͼ0.75). The final R min , crystallographic R-factor, and correlation coefficient values were 0.443, 0.32, and 0.66, respectively.
Structural Model Building and Refinement-To avoid lower ranking peaks that might be spurious, only the strongest 500 peaks were then input to unconstrained free-atom refinement using SHELXL (29). The highest 17 peaks were assigned to be sulfur atoms, and the remaining peaks were assigned as oxygen atoms. Refinement continued until there was no further decrease in the crystallographic R factor, which converged to a value of 0.25 after 100 cycles. Next, the main chain was built automatically by the program ARP/wARP (30) using diffraction data between 20 and 1.0 Å. The Connectivity Index increased steadily from 0.80 to 0.97 after 50 cycles of model building, indicating that a successful main-chain tracing had been found. Except for a few hydrophilic residues at the molecular surface, it was very easy to recognize and build almost all side chains based on both 2F obs Ϫ F calc and F obs Ϫ F calc maps. Hydrophilic residues with poor electron density were assigned according to a cDNA clone from the same species (31). The crystallographic R factor of the model built in this way dropped to a value of 22% without including any solvent molecules.
Further refinement with the SHELX package was performed with cross-validation using 5% of the observed reflections that were selected at random. Using data in the resolution range 10 -1.2 Å, the model was first refined to convergence with an R factor of 22.0% and an R free factor of 24.0%. Then, water molecules were added gradually. Using the instruction STIR 1.3 0.02, the upper resolution limit was gradually expanded to 1.0 Å. Both the electron density map and the model were inspected carefully, and the model was adjusted to fit the density using the program O (32). Thereafter, all reflections were used, and, after addition of more water, the resolution was extended to 0.8 Å yielding R and R free values of 18.91% and 20.08%, respectively.
Subsequent refinements were carried out by introducing anisotropic thermal parameters for all atoms except for the water molecules. This adjustment gave a sharp decrease in both R and R free (13.98% and 14.96%, respectively). At this point, alternative conformations could be modeled for disordered residues using 2F obs Ϫ F calc maps as well as clues in the diagnostic table output by SHELXL. For these modeled conformations, the initial B factors of all side-chain atoms were reset to an isotropic value of 10 Å 2 , the occupancies of the main conformers were fixed at 0.66, and atoms with isotropic and anisotropic B factors were refined alternately. Altogether, alternative conformations for nine residues were modeled. At this stage, the R value was 10.72%, and the R free value was 12.63%. Distance restraints were applied during all refinements; the allowed standard deviations from the distance targets was of 0.02 Å for covalent bonds and 0.04 Å for atoms linked by two bonds. Near the end of refinement, the addition of riding hydrogen atoms based on stereo-chemical requirements further reduced R and R free to 9.21% and 12.1%, respectively.
Experimental Charge Density Refinement-Charge density refinement for acutohaemolysin was carried out using the program MOPRO (33), an extension of the program MOLLY (16) for biological macromolecules. This program is able to refine large structures with the use of conjugate gradient iterations, and the sparse matrix principle was applied to the least squares method. A common difficulty in charge density refinement is the separation of the anisotropic mean square displacements of the atoms from the static molecular electron distribution. The proper experimental deconvolution requires very accurate diffraction data to ultrahigh resolution (e.g. 0.4 -0.6 Å). The contribution of the bonding electron density to the diffraction does not go beyond 0.8 Å (23), but scattering by the core electrons, located close to the atomic nuclei, extends to 0.5 Å or better depending on the size of the atomic thermal factors. Therefore, in small molecules, charge density analyses as well as accurate atomic positional and thermal parameters are obtained by a high order refinement where only reflections in the upper resolution ranges are taken into account.
In the case of acutohaemolysin, the starting structure for charge density studies was the final model obtained from the SHELXL refinement. Only the atoms with a B eq factor lower than 9 Å 2 were allowed to vary in the high order MOPRO refinement, and the stereo-chemical restraints that were applied were similar to those used in SHELXL. MOPRO enables the application of reflection weights that depend on the experimental intensity I H , on its uncertainty I , and on the resolution d. The W H weighting scheme results in reflections at high resolution that are overweighted, as in Equation 1.
Using this weighting scheme, the charge density could be partially retrieved for regions of the protein with low thermal motion. When a spherical atom model was used, a residual map revealed the non-modeled deformation of the electron density after high order refinement.
Coordinates and anisotropic displacement parameters were not refined simultaneously but, rather, in successive refinement cycles. The Scale factor and solvent correction parameters were regularly refined using a different weighting scheme, which was independent of the resolution. Refinement using all diffraction data and both weighting schemes (until convergence of the positional and thermal parameters was achieved) resulted in a final Rfactor of 15%. This relatively high value was due to overweighting of the upper resolution reflections during the refinement.

RESULTS AND DISCUSSION
Model Quality and Stereochemical Parameters-The structure of acutohaemolysin contains 1010 non-hydrogen protein atoms, 2 isopropanol molecules, and 232 water molecules, and it was refined to a crystallographic R factor of 9.21% (for data with F Ͼ 4 F ) or 10.52% (for all data). Fourier maps clearly revealed the atomicity of the electron density, and non-hydrogen protein atoms (i.e. C, N, and O) could be easily differenti-ated, because the heavier atoms have a larger spherical radius of density. Moreover, peaks accounting for 428 hydrogen atoms were observed on difference Fourier maps, thereby indicating that the refined model is of high quality.
The stereochemical parameters of the model were analyzed by the use of programs PROCHECK (34), WHATCHECK (35), and PROVE (36), and the results are shown in Table I. The main-chain dihedral angles of all non-glycine and non-proline residues lie in the preferred and allowed regions (91.6% and 8.4%, respectively) of the Ramachandran diagram (37), and the standard deviations of the bond lengths and bond angles (0.018 Å and 2.3°, respectively) are reasonable. The overall average G factor of 0.03 computed by PROCHECK is very close to its theoretic value of zero, and this is another indication of the excellent overall quality of the model. All root-mean-square (r.m.s.) Z scores are close to 1.0 except for those of the bond angles and side-chain planarity. The poor r.m.s. Z scores for these two quantities might be the result of the relatively low quality of the standard parameters used by WHATCHECK.
In addition, the standard deviations of the stereochemical parameters were also estimated directly from the experimental data by inversion of the full normal matrix (38), and the pa-  2 Ϫ F calc (hkl) 2 ) 2 /(n Ϫ p)] 1/2 , where n is the number of reflections, p is the number of refined parameters, F obs and F calc are the observed and calculated amplitudes, respectively, of the structure factors for reflection hkl, and w is a weighting scheme. e The G-factor and the Ramachandran plot were generated by program PROCHECK (34). f All r.m.s. Z scores are from WHATCHECK (35) with the exception of the volume, which was calculated by PROVE (36).
rameter uncertainties determined in this way are listed in Table I as well. Matrix inversion and the statistical program WHATCHECK give almost the same values for the standard deviations of the bond lengths, but they give quite different values for the standard deviations of the bond angles and the main-chain dihedral angles. The differences are 1.08°and 4.77°, respectively, indicating that the stereochemical parameters currently used in structural validation are not appropriate, especially for ultrahigh resolution data, and that the angles in protein structures may possess a wider range than the usually accepted values. The structure of acutohaemolysin refined with ultrahigh resolution diffraction data could be a candidate for deriving better structural parameters for validation.
Overall Structure-The amino acid residues of acutohaemolysin were sequenced from the crystallographic electron density, and the identities of the 27 N-terminal residues were determined by Edman degradation as well (25). These results were found to be in agreement with the sequence of a cDNA clone of unknown functions from the same species (31). The excellent model-to-density fitting of residue Lys 49 clearly shows that acutohaemolysin is a member of the Lys 49 PLA2 family instead of the Asp 49 or Ser 49 PLA2 families. A comparison of the primary structures of acutohaemolysin and four other Lys 49 PLA2 structures shows a high degree of homology (Fig. 1).
In common with other Lys 49 PLA2 structures, acutohaemolysin consists of three ␣ helices, two anti-parallel ␤ strands, and a few connecting loops (Fig. 2). Two long ␣ helices interlocked by two pairs of disulfide bonds (Cys 44 -Cys 105 and Cys 51 -Cys 98 ) constitute a rigid platform that stabilizes the overall structure. Other elements of secondary structure are attached directly or indirectly to this platform by five additional disulfide bonds. Compared with the other Lys 49 PLA2 structures (Fig. 3), the two regions of acutohaemolysin with the greatest conformational variation are the C-terminal loop where a serine residue is inserted at position 131 and the so-called calcium-binding loop containing residues 25-35. Fig. 4, is a highly conserved feature of the PLA2 enzymes, including the Lys 49 variants that do not have a calcium ion present in this region. The existence of a consensus sequence Tyr 25 -Gly-Cys-Asn-Cys-Gly-X-X-X-Arg-Gly 35 implies that this region has a similar function in all members of the family. Compared with the neighboring region containing Cys 27 -Asn-Cys-Gly 30 , the electron density of the peptide segment Val 31 -Gly-Gly 33 appears to be quite poor, indicating that these residues have relatively high thermal motion. In fact, the average temperature factor for Val 31 -Gly-Gly 33 is 16 Å 2 , double the value for the segment Cys 27 -Asn-Cys-Gly 30 . Fig. 4 also shows that a strong electron density peak marks the position of a water molecule in the calcium-binding loop. This structurally well ordered water molecule (named WAT2010) forms two hydrogen bond and four electrostatic interactions in the range 2.76 -3.38 Å with the carbonyl oxygen atoms of four residues in the loop as well as the N atoms of Lys 49 and Lys 122 (see Table  II for the geometric parameters of hydrogen bonds). Therefore, these six atoms, including the carbonyl oxygen atoms of Val 31 and Gly 33 , stabilize the water molecule, and it is reasonable to infer that the segment Val 31 -Gly-Gly 33 has an impact on the behavior of the water molecule. The electron density accounting for the two hydrogen atoms of this water molecule could be clearly seen on the maps.

Calcium-binding Loop-The calcium-binding loop, illustrated in
The calcium-binding loop is one of most conservative regions in PLA2s family. For the enzymatic activity of Asp 49 PLA2s, Scott et al. (41) proposed that the backbone amide N-H of residue Gly 30 plays a role in stabilizing the oxyanion of the substrate tetrahedral intermediate (41). On the other hand, for the enzymatic activity of Lys 49 PLA2s, Lee et al. (9) demonstrated that the hyperpolarization of the peptide bond between residues 29 and 30, induced by the hydrogen interaction between the main-chain carbonyl of Cys 29 and Lys 122 via a water molecule, may increase the affinity for the enzyme to the reaction product (fatty acid). In acutohaemolysin, the poor density appearance and high average temperature factor of peptide segment Val 31 -Gly-Gly 33 indicate relatively high dynamics. The hydrogen interactions between carbonyl oxygen atoms of Val 31 and Gly 33 to the structurally well ordered water molecule, which has electrostatic interactions with the N atoms of Lys 49 and Lys 122 , suggest coordination between Val 31 -Gly-Gly 33 , the water molecule (WAT2000), and the peptide bond of Cys 29 -Gly 30 . Therefore, it could be postulated that this loop may be of importance in stabilizing the substrate tetrahedral intermediate and in releasing the reaction products.
Catalytic Site-PLA2s use the conserved catalytic triad consisting of His 48 , Tyr 52 , and Asp 99 (see Figs. 2 and 5) to hydrolyze substrates into free fatty acids and lysophospholipids (1). It is believed that His 48 and Asp 99 , together with a nearby structurally conserved water molecule, are involved in the nucleophilic attack of the water molecule at the sn-2 position of the phospholipid substrate. The success of this process requires the stabilization of a tetrahedral transition-state intermediate (41). For Asp 49 PLA2s containing calcium ion, the essential factors stabilizing the tetrahedral intermediate are the interactions of calcium ion with the catalytic water molecule, with the carboxyl oxygen atom of Asp 49 , and with the main-chain carbonyl oxygen atoms of the calcium-binding loop. Due to the substitution of lysine for Asp 49 , Lys 49 PLA2s are unable to bind calcium, but some of them have, nevertheless, been demonstrated to have at least limited catalytic activity (7). Furthermore, the crystal structures of two Lys 49 PLA2s complexed with their substrates reveal that these proteins have the ability to bind substrates to their active sites (8,42). It has been proposed that Lys 49 PLA2s possess interrupted catalytic activity, because the residue Lys 122 hyperpolarizes the peptide bond between residues Cys 29 and Gly 30 , resulting in failure to release the reaction product (9).
Although all of the required catalytic apparatus (including His 48 , Tyr 52 , Tyr 73 , and Asp 99 , and the catalytic water molecule WAT2010) is conserved in the acutohaemolysin structure, no measurable activity is exhibited in standard assays (25). When the catalytic triad of acutohaemolysin is superimposed with the corresponding residues of four structurally homologous Lys 49 PLA2s, at least one of which possesses catalytic activity (9), it is apparent that all the triads have nearly identical conformations and are anchored tightly to the scaffold composed of two ␣ helices (Fig. 5). Therefore, it is clear that the local conformation of the triad in acutohaemolysin is not responsible for the lack of catalytic ability, and the structural basis for this functional deficiency needs to be investigated further.
By sequence alignment (see Fig. 1), Phe 102 is detected as a unique residue that exists only in acutohaemolysin, whereas other Lys 49 PLA2s possess a short side-chain residue (e.g. Val) at this site. As shown in Fig. 5, by comparison of the local conformation near the catalytic triad of acutohaemolysin with that of piratoxin II, a catalytically active Lys 49 PLA2 complexed with a 13-carbon fatty acid molecule from Bothrops pirajai venom (9), it is apparent that the benzene ring of Phe 102 may interfere with the binding of potential substrates. If this fatty acid molecule were placed at the same position in the acutohaemolysin structure, there would be an impossibly short contact distance (about 1.3 Å) between it and the benzene ring of Phe 102 . Therefore, the docking of a similar or larger fatty acid molecule into the hydrophobic channel of acutohaemolysin seems to be prohibited. Because all other structural components in or near the active site are the same, and because the catalytic process requires a space to allow substrate entry and product release, it can be postulated that Phe 102 is responsible for the loss of catalytic activity. Clearly, this hypothesis should be validated experimentally by the site-directed mutation of F102V in acutohaemolysin.
The acutohaemolysin structure provides a high resolution look at the hydrogen-bond network in the vicinity of the active site. Peaks of electron density corresponding to the hydrogen atoms riding on the C␦ 2 , C⑀ 1 , and N⑀ 2 atoms of His 48 as well as the O h atom of Tyr 52 have been readily observed (Fig. 6). The two hydrogen-bond distances involving the O␦ atom of Asp 99 are 2.59 and 2.77 Å, indicating the strength of these interactions. Because the dissociation pK a value of the N⑀ 2 H ϩ group in a free histidine residue is about 6.0, the proton could easily bind to or release from the N⑀ 2 atom under neutral conditions. Acutohaemolysin was crystallized at pH 5.6 (a little lower than the dissociation pK a value) so that the environment was appropriate for the protonation of both the N␦ 1 and the N⑀ 2 atoms. However, it should be remembered that the pK a values of amino acid residues located at the surface of proteins are known to be highly dependent on the local chemical environment (43). The protonation of the N⑀ 2 atom is favored by its role as hydrogen-bond donor to the negatively charged carboxylate of Asp 99 . On the other hand, N␦ 1 seems to be non-protonated and is the receptor in the hydrogen bond involving WAT2009.
Conformational Flexibility and Dimerization-Biological macromolecules are much more flexible than small organic molecules, and conformational disorders in proteins can be either dynamic or static. Dynamic disorder can be due to thermal vibrations or to rotations around single bonds, whereas static disorder can result from slight displacements of parts of protein molecules in the crystal lattice or local differences in solvation. Relatively low resolution x-ray diffraction experiments give an average positional distribution for disordered atoms or residues. At ultrahigh resolution, however, it is possible to decipher the disorder and to view alternative conformations. In the case of acutohaemolysin, such conformations have been modeled for several residues, including Trp 10 , Val 18 , Leu 24 , Arg 43 , Met 92 , Arg 107 , Glu 108 , Asp 111 , and Ser 116 . All of these residues are located at the molecular surface, and some of these cases of disorder may have possible functional significance.
In some Lys 49 PLA2s, the N-terminal ␣ helix and an antiparallel ␤ sheet in the region of residues 74 -85 (noted as the ␤ wing in Fig. 2) play a role in the formation of a molecular dimer that, in turn, has been considered essential for the presence of hemolytic activity (23,42). On the other hand, acutohaemolysin exists as a monomer in solution, and it has no hemolytic activity. Whereas other Lys 49 PLA2s typically possess Leu or Phe residues at position 10 at the end of the N-terminal ␣ helix, this residue is a tryptophan in acutohaemolysin (Fig. 1). The large indole group of Trp 10 shows two preferred conformations differing by a rotation of about 180°around the C␤-C␥ bond (Fig.   FIG. 4. A stereo view of the calciumbinding loop of acutohaemolysin. The electron density is contoured at 4.0 for the 2F obs Ϫ F calc map (dark green) and at 2.4 for the F obs Ϫ F calc map (red). The peptide segment Val 31 -Gly-Gly 33 shows very poor electron density relative to that for Asn 28 -Cys-Gly 30 . The structurally ordered (and highly conserved) water molecule 2010 has multiple interactions with neighboring atoms, and peaks on the F obs Ϫ F calc map indicate that this water is the hydrogen-bond donor for the Gly 31 and Gly 33 carbonyl oxygen atoms and receives hydrogens from Lys 49 and Lys 122 . The carbonyls of Cys 27 and Asn 28 also have electrostatic interactions with the water molecule.   7), and a series of other conformers with varying torsion angles about this bond may, in fact, exist. The electron density in this region is poor due to the disorder as well as the large atomic thermal motion (B eq ϭ 20 -40 Å 2 ). Nevertheless, it is clear that the Trp 10 side chain, which occupies a volume of about 300 Å 3 , extends into the surrounding solvent, and its flexibility may prevent molecular aggregation and dimer formation from occurring. Thus, the presence of Trp 10 may be responsible for the absence of hemolytic activity in acutohaemolysin.
Although acutohaemolysin exists as a monomer in solution, it packs in the crystal as a dimer composed of two molecules related by a crystallographic 2-fold axis parallel to the b axis of the unit cell. This dimer is distinctly different from the hemolytically active dimer observed in other Lys 49 PLA2s. One of the interactions that appears to stabilize the crystallographic dimer is a pair of salt bridges between the disordered Arg 107 and Glu 108 residues. The two positively charged N atoms of Arg 107 in one monomer are hydrogen-bonded to the two negatively charged O⑀ atoms of Glu 108 in the other (symmetryrelated) half of the dimer with the shortest contact distances being 2.75 and 2.98 Å. In addition, the Arg-Glu salt-bridge pairs are further stabilized by longer range electrostatic interactions with N . . . O distances shorter than 4 Å (44). The definite contribution of salt bridges and longer range electrostatic interactions in the crystal packing formation could be supported by the analysis of several electrostatic variants of cutinase surface residues in different crystal forms (45).
Solvent Molecules-X-ray diffraction experiments performed at liquid nitrogen temperatures permit more solvent molecules to be located than would be possible using data measured at higher temperature. In all, 210 fully occupied water molecules, 22 half-occupied water molecules, and 2 isopropanol molecules were located on the acutohaemolysin difference electron-density maps. All of these solvent molecules were visible as electron density peaks contoured at 4, and they possess equivalent isotropic thermal parameters (B eq ) lower than 40 Å 2 . Each water molecule forms at least one hydrogen bond with adjacent protein atoms or other solvent atoms.
The thermal anisotropy of both protein and solvent atoms in acutohaemolysin was analyzed statistically (see Table I) using PARVATI, a program that has been implemented to study this property for structures determined at a resolution higher than 1.4 Å (46). Anisotropy is defined as the square axial ratio of an ellipsoid and can be determined from the anisotropic displacement parameters. For the protein atoms, the distribution of anisotropy values has a typical Gaussian shape, and the average anisotropy of 0.443 is almost the same as the expected value of 0.45 statistically computed from 67 crystal structures with a resolution higher than 1.4 Å. On the other hand, the anisotropy distribution for the solvent molecules has two peaks. One of these peaks (at ϳ0.44) matches the results for the protein atoms, but the other (at ϳ0.25) indicates a larger anisotropy. Such a bimodal distribution implies that the water molecules in the refined structure of acutohaemolysin can be divided into two hydration shells. Water molecules in the inner shell (anisotropy peak at 0.44) have thermal parameters similar to those of the protein atoms and could be considered to be an integral part of the protein molecule. Inner-shell waters exhibit an average hydrogen-bond length of 2.70 Å, whereas the corresponding value for waters in the outer shell is 3.13 Å. These results are similar to observations made for the hydration shells around other protein molecules (47).
Hydrogen Atoms-Ultrahigh resolution diffraction data are of good quality if a high percentage of the hydrogen atoms can be recognized on difference electron density maps. In the case of acutohaemolysin, a total of 428 hydrogens, including most main-chain and many side-chain atoms (about 38% of the possible 1132 hydrogen atoms in the structure), could be discerned on F obs Ϫ F calc difference maps contoured at 2.1 (1 ϭ 0.09e/ Å 3 ). All of these peaks had a distance of 1.0 Ϯ 0.3 Å to the nearest non-hydrogen atom. There were 89 and 88 peaks (72% of all possible) observed for the electron density of the hydrogen  6. A stereo view of the hydrogen-bond network for the catalytic triad of acutohaemolysin illustrates the probable protonation. The electron density is contoured at 2.5 for the 2F obs Ϫ F calc map (dark green) and at 2.3 for the F obs Ϫ F calc map (red). Clearly, the N⑀ 2 atom of His 48 is protonated, but the N␦ 1 atom is not. The O-O distance of 2.59 Å is evidence of a strong hydrogen bond between Tyr 52 and Asp 99 . atoms riding on the main-chain C ␣ and N atoms, respectively. Although a relatively low percentage (28%) of the side-chain hydrogen atoms was observed, the peaks of electron density for the hydrogen atoms of some residues in hydrophobic regions could be found easily. The geometry of the hydrogen bonds listed in Table II gives an indication of the good quality of the hydrogen positions.
Experimental Deformation Density Analysis-The ultrahigh resolution (0.8 Å) and high quality (R merge ϭ 5.5%) of the acutohaemolysin diffraction data as well as the low experimental temperature factors estimated from a Wilson plot (5.6 Å 2 ) and observed as the average for the refined model parameters (6.5 Å 2 ) support the application of charge density analysis. For some hydrophobic residues in the core region of the structure, the atomic temperature factors are even below 4.0 Å 2 , low enough to allow partial differentiation of the deformation density features from the thermal movement. Following MOPRO refinement of the coordinates and thermal parameters of the protein atoms with B eq Ͻ 9 Å 2 , a residual map displayed unambiguously the covalent bonding densities in the regions with low B-factors. These regions include the catalytic residue His 48 (Fig. 8A) and residue Tyr 64 (Fig. 8C). Planes through several peptide bonds show residual peaks corresponding to valence electrons (e.g. peptide bond Phe 92 -Ala 93 shown in Fig.   8B). Some bonding densities are even visible for some residues in the difference map after the refinement with the SHELXL program Disulfide bonds are formed by the oxidation of two thiol groups of cysteine residues. The formation of such intra-molecular or inter-molecular bonds is important to fold and stabilize functional structures. Besides, disulfide bonds are sensitive to temperature or reducing environments and may easily dissociate or reform by increasing or decreasing the temperature. S-S bonds are weak covalent interactions, and the corresponding bonding density peaks appear lower compared with other bonds (C-S, C-C, and C-N, etc.) in a charge density analysis of L-cystine performed at 0.45-Å resolution (44). After high order refinement, the map also shows residual electron density between the two sulfur atoms that form the Cys 43 -Cys 95 disulfide bond (Fig. 8D).
The presence of seven disulfide bridges in the protein (Fig. 2) and the relatively low amount of bulk solvent volume in the unit cell contribute to the overall low thermal motion in the PLA2 structure. Because a significant number of the protein atoms (39%) have low B factors in the range of 4 -6 Å 2 , the crystals reported here seem to have the potential to diffract to resolutions even higher than 0.8 Å (22), if the crystallization conditions and the data collection techniques are optimized. A FIG. 8. Residual maps that were obtained after high order refinement and contoured at ؎0.05 e/Å 3 reveal peaks of electron density between the bonded atoms. The solid blue lines show the electron density accumulation, and the dashed red lines show the electron density depletion (i.e. blue, positive density; red, negative density). These residual maps illustrate the catalytic residue His 48 (A), the dipeptide Phe 92 -Ala 93 (B), residue Tyr 64 (C), and the disulfide bond between Cys 43 and Cys 95 (D). charge density analysis performed using such data would provide additional information about the protonation and electronic states of the residues in acutohaemolysin.