Structural Biology and Crystallization Communications

BrabA.11339.a: anomalous diffraction and ligand binding guide towards the elucidation of the function of a 'putative b-lactamase-like protein' from Brucella melitensis The crystal structure of a-lactamase-like protein from Brucella melitensis was initially solved by SAD phasing from an in-house data set collected on a crystal soaked with iodide. A high-resolution data set was collected at a synchroton at the Se edge wavelength, which also provided an independent source of phasing using a small anomalous signal from metal ions in the active site. Comparisons of anomalous peak heights at various wavelengths allowed the identification of the active-site metal ions as manganese. In the native data set a partially occupied GMP could be identified. When co-crystallized with AMPPNP or GMPPNP, clear density for the hydrolyzed analogs was observed, providing hints to the function of the protein.


Introduction
The mission of the Seattle Structural Genomics Center for Infectious Disease (SSGCID) is to provide the scientific community with a library of crystal structures of proteins from infectious organisms that are potential drug targets. This collection of structures is hoped to promote structure-guided drug design against National Institutes for Allergy and Infectious Disease (NIAID) class A, B and C infectious disease organisms. The target organisms are divided between SSGCID and its sister organization the Center for Structural Genomics of Infectious Diseases (CSGID). Both centers provide plasmids (BEI repository), protein samples (SSGCID), coordinates and structure factors (Protein Data Bank), and raw diffraction images (http://www.csgid.org/csgid/pages/diffraction_images) to the community immediately after completion of a target.
Brucella melitensis is a Gram-negative bacterial pathogen and one of the several Brucella species identified to cause brucellosis (Moreno & Moriyon, 2002). Brucellosis affects mostly goats, sheep and cattle, causing reproductive diseases such as abortions and stillbirths, hence the popular name 'contagious abortion'. The pathogen can be transmitted to humans via contact with infected animals or the consumption of unpasteurized milk or cheese (Young, 1995). In humans the disease was initially called 'Malta fever' or 'undulating fever' and is multi-systemic, with nonspecific symptoms such as fever, chills, dementia, fatigue, headaches, nausea, vomiting, muscular and bodily pain. Regions with significant impact of brucellosis are parts of the Mediterranean, Asia, Africa and Latin America (Corbel, 1997). Infections can be successfully treated with antimicrobial agents. The pathobiology of Brucella is unique owing to the absence of classical virulence factors (Moreno & Moriyon, 2002).
Here, we report the high-resolution structure of BrabA.11339.a, a structural genomics target that entered the pipeline as a '-lactamaselike protein'. -Lactamases break down -lactam antibiotics such as penicillin and lead to the resistance of pathogens to antibiotics. There are four classes of -lactamase: classes A-D. While classes A, C and D have an active site-serine, class B metallo--lactamases have a binuclear metal center in the active site that activates a water. With high-resolution data including anomalous information at various wavelengths, we could obtain a detailed description of the active site of BrabA.11339.a and an identification of the active-site metals.
Ligand-bound structures lead us towards an understanding of the function of this protein.
The protein was expressed in Escherichia coli using BL21 (DE3) R3 Rosetta cells and autoinduction medium (Studier, 2005) in a LEX Bioreactor. Starter cultures of LB broth with appropriate antibiotics were grown for $18 h at 310 K. ZYP-5052 auto-induction medium was freshly prepared as per Studier's published protocol (Studier, 2005). Antibiotics were added to 2 l bottles of sterile auto-induction medium. The bottles were inoculated with all of the overnight culture. Inoculated bottles were then placed into a LEX bioreactor. Cultures were grown for $24 h at 298 K; the temperature was reduced to 288 K and the cultures were grown for a further $60 h. To harvest, the medium was centrifuged at 4000g for 20 min at 277 K. The cell paste was flash-frozen in liquid nitrogen and stored at 193 K.
The frozen cells were resuspended in lysis buffer (25 mM HEPES pH 7.0, 500 mM NaCl, 5% glycerol, 30 mM imidazole, 0.025% sodium azide, 0.5% CHAPS, 10 mM MgCl 2 , 1 mM TCEP, 250 ng ml À1 AEBSF and 0.05 mg ml À1 lysozyme). The resuspended cell pellet was disrupted on ice for 30 min with a Virtis sonicator (408912) set to 100 W power with alternating cycles of 15 s pulse-on and 15 s pulseoff. The cell debris was incubated with 20 ml Benzonase nuclease (25 units ml À1 ) at room temperature for 45 min and clarified by centrifugation using a Sorvall SLA-1500 centrifuge at 14 000 rev min À1 for 75 min at 277 K. The protein was purified from the clarified cell lysate by immobilized metal-affinity chromatography using a 5 ml HisTrap FF column (GE Healthcare) equilibrated with binding buffer (25 mM HEPES pH 7.0, 500 mM NaCl, 5% glycerol, 30 mM imidazole, 0.025% sodium azide, 1 mM TCEP). The recombinant protein was eluted with 250 mM imidazole. The hexahistidine tag was removed by overnight incubation with 3C protease. The sample was then passed over HisTrap resin in a gravity-flow column. The tagless protein was collected in the flowthrough. It was further purified by size-exclusion chromatography (SEC) using a HiLoad 26/60 Superdex 75 column (GE Healthcare) equilibrated with SEC buffer (25 mM HEPES pH 7.0, 500 mM NaCl, 2 mM DTT, 0.025% sodium azide and 5% glycerol). The protein eluted as a single peak. The peak fractions were pooled and concentrated to 19.6 mg ml À1 using an Amicon Ultra centrifugal filter with a 10 kDa cutoff (Millipore). The concentrated sample was flash-frozen in liquid nitrogen and stored at 193 K.

Crystallization and X-ray data collection
For the target BrabA.11339.a at a protein concentration of 19.6 mg ml À1 , two sparse-matrix screens were set up following the strategy of Newman et al. (2005): JCSG+ (Emerald BioSystems) and PACT (Molecular Dimensions). 0.4 ml protein solution was mixed with 0.4 ml well solution and equilibrated against 100 ml reservoir using 96-well Compact Jr crystallization plates (Emerald Bio-Systems). Crystals suitable for diffraction studies were obtained under two very similar conditions from the PACT screen: G9 (200 mM sodium/potassium tartrate, 100 mM Bis-Tris propane pH 7.5, 20% PEG 3350) and H9 (200 mM sodium/potassium tartrate, 100 mM Bis-Tris propane pH 8.5, 20% PEG 3350). Crystals were cryoprotected in a buffer consisting of well solution to which 20% ethylene glycol was added.
A native data set was collected in-house at Cu K wavelength using a Rigaku SuperBright FR-E + rotating-anode X-ray generator equipped with Osmic VariMax HF optics and a Saturn 944+ CCD detector (Table 1, native Cu K). The diffraction data were reduced with XDS/XSCALE (Kabsch, 2010) and scaled to 1.85 Å resolution in space group C222 1 .
In order to obtain experimental phase information, a crystal from the same well was incubated for 5 min in an iodide-containing buffer consisting of 200 mM sodium/potassium tartrate, 100 mM HEPES pH 7.0, 25% PEG 3350, 250 mM KI. A data set with high multiplicity was collected in-house using the same set up as described above. 720 images were collected with a width per frame of 0.5 . The diffraction data were reduced with XDS/XSCALE (Kabsch, 2010) and scaled to 2.0 Å resolution. Data analysis in XSCALE (Anomalous Correlation and SigAno) indicated a strong anomalous signal out to 2.0 Å resolution.
A high-resolution data set was collected on beamline 5.0.1 of the Berkeley Center for Structural Biology in the context of the Collaborative Crystallography program. The wavelength of this beamline is optimized for an Se-SAD experiment: 0.9774 Å /12.4 keV (Morton et al., 2007). The beamline is equipped with a 3 Â 3 tiled ADSC Q315r detector. The diffraction data were reduced with XDS/XSCALE (Kabsch, 2010)  In order to validate ligand binding, the protein was incubated with 1 mM AMPPNP or GMPPNP and 2 mM MgCl 2 . A small optimization screen was set up: 15-20% PEG 3350, 200 mM sodium/potassium tartrate, 100 mM Bis-Tris pH 7.0-8.0. After crystals had grown to full size within a few days, they were incubated for 1 min in a cryoprotection buffer consisting of 1 mM AMPPNP or GMPPNP, 2 mM MgCl 2 , 15% ethylene glycol, 21% PEG 3350, 165 mM sodium/ potassium tartrate, 83 mM Bis-Tris pH 7.5. For each of the co-crystals, a data set with high multiplicity was collected at the home source as described for the iodide soak above. Each data set was scaled to 1.7 Å resolution. For another AMPPNP cocrystal from the same optimization screen, a data set with high multiplicity was collected at a wavelength of 1.3476 Å on beamline 5.0.2 of the Berkeley Center for Structural Biology in the context of the Collaborative Crystallography program. The wavelength was selected to be at lower energy than the Zn edge (1.284 Å /9.659 keV) but at significantly higher energy than the Mn edge (1.896 Å /6.359 keV) The ALS beamline is equipped with a 3 Â 3 tiled ADSC Q315r detector. The data were scaled to 1.6 Å resolution.

Structure solution and refinement
The packing density of the crystals (Matthews, 1968), 2.3.2. Native synchrotron data set. In a separate approach, the native high-resolution synchrotron data set was used to phase the BrabA.11339.a structure. The XSCALE analysis had indicated a moderate anomalous signal to $2.0 Å resolution. phenix.hyss (Grosse-Kunstleve & Adams, 2003) was able to locate two strong anomalous scatterers when the diffraction data were limited to 3 Å resolution. Phaser (McCoy et al., 2007) refined and extended the anomalous substructure to three atoms and yielded an FOM of 0.360 using data to 1.8 Å resolution. After density improvement with Parrot (Cowtan, 2010), ARP/wARP (Langer et al., 2008) could build a good initial model with 267 residues and a R work of 0.200 with data to 1.7 Å resolution.
The model for the high-resolution structure of BrabA.11339.a contains one protein molecule spanning residues Ser3-Val272, two manganese ions, a potassium ion, a sodium ion, a guanosine 5-monophosphate (GMP), a tartrate molecule and 348 water molecules. The final structure was validated with Coot (Cowtan, 2010) and MolProbity (Chen et al., 2010). The structure was deposited in the PDB with the identifier 3md7 (see Table 2).
The three structures cocrystallized with AMPPNP or GMPPNP were solved by direct refinement of the high-resolution structure from which all nonprotein atoms had been removed. Data and refinement statistics are summarized in Tables 1 and 2.

Results and discussion
The target BrabA.11339.a entered the SSGCID pipeline as a '-lactamase-like' protein from B. melitensis. A search of sequence databases for homologies yielded a variety of annotations, primarily other '-lactamase-like' proteins, metal-dependent hydrolases, uncharacterized proteins, PhnP (a phosphodiesterase of the carbonphosphorus lyase pathway for phosphonate degradation) proteins and tRNase Z proteins. However, the sequence is highly conserved between various species, with many proteins sharing more than 50% sequence identity. A search of the Protein Data Base yielded a striking lack of structures with high sequence homology. The phosphodiesterase of the carbon-phosphorus lyase pathway (PhnP; PDB entry 3g1p) from E. coli is the closest homologue; PhnP shares 23% sequence identity with BrabA.11339.a, with a large number of inserts and deletions.  Table 2 Refinement and model statistics.
The models for the iodide and native Cu K data sets have only been partially refined. Preliminary refinement statistics are shown in italics. Values in parentheses are for the highest of 20 resolution shells. jF obs j À jF calc j = P hkl jF obs j. The free R factor was calculated with an equivalent equation using 5% of the reflections that were omitted from the refinement.
BrabA.11339.a readily crystallized in a well diffracting C-centered orthorhombic crystal form. The structure could not be solved by molecular replacement. SSGCID's first approach to obtaining experimental phases is quick soaks with iodide (Abendroth et al., 2010(Abendroth et al., , 2011, following the ideas of Nagem and Dauter (Nagem et al., 2001). The principles behind this approach are (i) the strong anomalous signal of iodide at Cu K X-ray wavelength (f 00 = 6.8; Bricogne et al., 2003), (ii) the compatibility of iodide with weak binding sites of different natures, both hydrophobic and basic, and (iii) the high concentrations of iodide that a crystal lattice can tolerate. In the SSGCID context, this experimental phasing strategy has been extremely successful (Abendroth et al., 2010(Abendroth et al., , 2011. For BrabA.11339.a the phasing information was not only obtained from iodide ions: two metal ions in the active site and a potassium ion contributed in addition to the iodide ions. In fact, the high-resolution native data set was collected on a beamline which was optimized for data collection at the Se edge (0.9774 Å ). The minor anomalous signal from these three anomalous scatterers was strong enough to phase the structure independently. The structure solution of BrabA.11339.a underlines the power of anomalous phasing by both intrinsic anomalous scatterers such as metals and externally added scatterers such as iodide.
BrabA.11339.a crystallizes with one monomer per asymmetric unit. A dimer is formed by crystallographic symmetry (Fig. 1). The dimer buries 1444 Å 2 accessible surface per protomer in the interface as calculated by PISA (Krissinel & Henrick, 1997). BrabA.11339.a assumes a metallo--lactamase-like fold with two central -sheets (Fig. 2). A search for structural homology using SSM (Krissinel & Henrick, 2004) Fig. 2). The structural homology is focused mostly around the central -sheets and the helices 4 and 5, while there is more structural variety between -strands 5 and 6.
As a very typical feature of -lactamase-like proteins and the phosphodiesterase protein family, a two-metal center is present in the active site of BrabA.11339.a (Fig. 3). The two metals are 3.40 Å apart, bridged by water 330 and liganded by several histidine residues (His86, His88, His170 and His241) with bond distances between 2.22 and 2.24 Å , aspartate residues (Asp90 and Asp188) with bond distances between 2.28 and 2.31 Å and the phosphate group of GMP. Fold comparison. Comparison of the fold of BrabA.11339.a with its structural homologues. 3g1p is the PhnP protein from E. coli, 1zkp is a putative ribonuclease from B. anthracis and 1y44 is ribonuclease Z from B. subtilis. The structural homology is focused around the two -sheets and helices 4 and 5. Color scheme: Mn, magenta; Zn, cyan; Na, green; K, blue Structure of BrabA.11339.a dimer. In the ribbon representation of the crystallographic dimer the -strands for each protomer are colored slightly differently. The crystallographic dyad runs vertically. Various ions are shown as spheres: Mn 2+ in magenta, K + in blue and Na + in green. Five strong iodide sites are shown as orange spheres for one protomer. GMP and tartrate are shown as stick models.
The bond distances around the metal center are summarized in Table 3. They fall within the distances expected for the coordination of manganese ions (Zheng et al., 2008). Podzelinska et al. (2009) have performed a detailed functional analysis of the metal-dependency of PhnP; they report a strong preference for Mn 2+ and Ni 2+ (Podzelinska et al., 2009) and model the metals as manganese ions in their structure. In the 1zkp structure of a putative ribonuclease (Midwest Center for Structural Genomics, unpublsihed work) the metals are modeled as Zn 2+ ions, likely because of the presence of zinc chloride in the crystallization buffer. In the structure of RNAse Z (PDB entry 1y44) the metals are modeled as Zn 2+ ions because of functional studies, even though this is not further discussed in the structure paper and the B factors of the metal ions are higher than for the protein environment (de la Sierra-Gallay et al., 2005).
The iodide data set already indicated significant anomalous density for the two active-site metals and the potassium site. In order to analyze the metal center in more detail, diffraction data sets with high multiplicity were collected at long X-ray wavelengths from the ligand co-crystals: in-house at = 1.5418 Å and at the synchrotron at = 1.3476 Å . Anomalous Fourier peak heights for the metal sites are summarized in Table 4(a) and are compared with the theoretical f 00 coefficients for various atoms in Table 4(b). Potassium was part of the crystallization, soaking and cryobuffer at high concentrations that were consistent between the data sets. The occupancy of the potassium ion can therefore assumed to be constant and can be considered as a reference. Interestingly, the potassium-ion site is not conserved between the structural homologs. For both long-wavelength data sets and the Se-edge data set strong anomalous Fourier peaks can be observed for the two active-site metals. Each of the peaks is about four times as strong as the signal for potassium. Manganese is the only metal in question that has a significant anomalous signal at all three wavelengths. The anomalous signal for Zn 2+ is strong at the Se edge, but is very weak at either of the longer wavelengths. The anomalous signal for Ni 2+ is strong at the Se edge and at 1.3476 Å ; however, it is low at 1.5418 Å . The bond distances between the metals are in the range expected for Mn 2+ ions (Zheng et al., 2008). Therefore, the dinuclear center of BrabA.11339.a can be assigned as two manganese ions with a high degree of confidence.
In the native structure, a partially occupied GMP molecule can be observed in the immediate proximity of the binuclear Mn 2+ center. At no stage during the purification was GMP added to the protein; thus, it must have been copurified from the expression host. In order to confirm the binding of nucleotides, crystals were grown in the presence of GMPPNP and AMPPNP. In both cases, however, very solid electron density could be observed for GMP and AMP, respectively (see Fig. 4). The -phosphate group is very well defined, while there is no density for the -phosphate or -phosphate groups. The -phosphate groups of GMPPNP or AMPPNP, which had been added during crystal growth and had been supplemented for a short soak in the cryobuffer, are nonhydrolyzable; the -phosphate groups, however, can be hydrolyzed. It can be assumed that BrabA.11339.a is able to hydrolyze AMPPNP and GMPPNP between the -phosphate and the -phosphate to AMP or GMP, respectively.
The GMP or AMP molecules are bound in the dimer interface and make multiple interactions. The phosphate group is involved in hydrogen bonds to both manganese ions, the bridging water molecule 330 (called water 1 in the ligand-bound structures), histidine residues His88 and His241, aspartate residue Asp90 and further water molecules (315, 454 and 511 in the high-resolution structure). The ribose group makes two interactions with the carboxyl group of Asp96 (O2* and O3*) and two interactions with the guanidine group of Arg99 (O2*), each from the other protomer. The guanine group stacks with Tyr127 from the other protomer. The guanine moiety engages in Dinuclear manganese center. The active site of BrabA.11339.a consists of two manganese ions (magenta), which are bridged by a water molecule (red). The phosphate group of GMP is in close proximity to this activated water molecule. The A -weighted 2F o À F c electron density for the high-resolution data set is contoured at 1 (blue) and the corresponding F o À F c electron density is contoured at AE3 (green/red). The anomalous Fourier electron density is contoured at 10 (orange). The strong anomalous Fourier density at all three wavelengths indentifies the active-site metal as manganese. For bond distances refer to Table 3; for anomalous peak heights refer to Table 4. Table 3 Bond distances of the dinuclear Mn 2+ center for the high-resolution data set.   (Bricogne et al., 2003). Manganese is the only metal in question that has a stronger signal than potassium at both wavelengths. hydrogen bonds to the amide N atom of Ala89 (N3) and to water molecules. Neither of these interactions should be able to provide specificity for either AMP or GMP.
In the PhnP structure (PDB entry 3g1p), a malonate is modelled in the space that is occupied by the phosphate and the ribose. No ligand was modelled in the putative ribonuclease (PDB entry 1zkp). In the structure of ribonuclease Z from B. subtilis (PDB entry 1y44, r.m.s.d. 2.3-2.4 Å , 20% sequence identity) a phosphate ion is modelled in an identical location to the phosphate group of GMP in BrabA.11339.a, even though the B factors for the phosphate ion are significantly higher than those for the environment (de la Sierra-Gallay et al., 2005). None of these three homologous structures are able to accommodate a GMP or an AMP molecule in the conformation observed in BrabA.11339.a without conformational changes.
The discovery of a purine monophosphate in the vicinity of the dinuclear manganese center allows speculation regarding the function of BrabA.11339.a. For various enzymes with a dinuclear center it is known that the bridging water is activated by the two metal ions. The water reacts as a hydroxide, starting a hydrolytic reaction. The phosphate group of the GMP or AMP is located in close proximity to this deprotonated water molecule 303 (2.2 Å ). There is enough space in the active site to fit two more phosphate groups. It therefore seems likely that BrabA.11339.a is a GTP/ATP or GDP/ADP hydrolase.
The structure of BrabA.11339.a and its binuclear manganese center were determined with high accuracy using high-resolution data and anomalous diffraction. The initially serendipitous discovery of GMP in the active site and the subsequent high-resolution co-crystal structures with in situ hydrolyzed AMP or GMP shed light on the function of this protein, which entered the SSGCID pipeline as a '-lactamase-like protein'. Further analysis will be needed to validate the biological function of this protein and its suitability for structureguided drug design against the pathogen B. melitensis.