Holo- and Apo-bound Structures of Bacterial Periplasmic Heme-binding Proteins*

An essential component of heme transport in Gram-negative bacterial pathogens is the periplasmic protein that shuttles heme between outer and inner membranes. We have solved the first crystal structures of two such proteins, ShuT from Shigella dysenteriae and PhuT from Pseudomonas aeruginosa. Both share a common architecture typical of Class III periplasmic binding proteins. The heme binds in a narrow cleft between the N- and C-terminal binding domains and is coordinated by a Tyr residue. A comparison of the heme-free (apo) and -bound (holo) structures indicates little change in structure other than minor alterations in the heme pocket and movement of the Tyr heme ligand from an “in” position where it can coordinate the heme iron to an “out” orientation where it points away from the heme pocket. The detailed architecture of the heme pocket is quite different in ShuT and PhuT. Although Arg228 in PhuT H-bonds with a heme propionate, in ShuT a peptide loop partially takes up the space occupied by Arg228, and there is no Lys or Arg H-bonding with the heme propionates. A comparison of PhuT/ShuT with the vitamin B12-binding protein BtuF and the hydroxamic-type siderophore-binding protein FhuD, the only two other structurally characterized Class III periplasmic binding proteins, demonstrates that PhuT/ShuT more closely resembles BtuF, which reflects the closer similarity in ligands, heme and B12, compared with ligands for FhuD, a peptide siderophore.

Successful pathogenic bacteria must acquire nutrients from the host and one such essential nutrient is iron. Numerous Gram-negative bacterial pathogens have developed a sophisticated mechanism for recruiting host heme iron (1,2). The first step involves a TonB-dependent cell surface receptor that acquires heme. Heme transport across the periplasmic space involves an active transport system, composed of a soluble periplasmic heme-binding protein, an inner membrane heme permease, and an ATPase. Upon crossing the inner membrane, the heme is broken down by heme oxygenase thus releasing iron in the cytosol.
Two of the better characterized heme transport systems in Gram-negative bacteria are from Pseudomonas aeruginosa and Shigella dysenteriae. Shigella dysenteriae is a Gram-negative, non-spore forming bacteria that causes deadly epidemics in many developing regions and nations. This type of bacteria typically resides in the human gastrointestinal tracts and causes an inflammatory disorder of the lower gastrointestinal tract (3). Pseudomonas aeruginosa is widespread in nature-inhabiting soil, water, plants, animals, and humans and is an important cause of diseases, including pneumonia and urinary tract infections, especially in patients with compromised host defense mechanisms (4). Given that the heme transport proteins are unique to these bacteria, heme transport proteins provide potential targets for anti-bacterial agents.
Both Pseudomonas and Shigella encode a periplasmic binding protein (PBP) 2 required to shuttle heme from the outer membrane to the inner membrane. In Pseudomonas PhuT encodes a 33-kDa protein, whereas the Shigella ShuT protein at 28.5 kDa is somewhat smaller. Both PhuT and ShuT are members of a broad family of PBPs that are important in maintaining selectivity and specificity of substrate transport (1,5). PBPs consist of N-and C-terminal domains with a ligand binding cleft between domains. PBPs can be divided into three main classes. Class I and II are defined by having two or three interdomain connections, respectively. The maltose-binding protein belongs to Class I, and this group of PBPs undergoes a fairly large interdomain open/close motion required for substrate binding and release (6). Our crystal structures revealed that ShuT and PhuT belong to a third class that contains a single connecting helix between the two globular domains. The vitamin B 12 -binding BtuF (5, 7) and FhuD (8), which binds ironhydroxamic acid siderophores, are the only Class III PBPs whose crystal structures are known. Here we report the crystal structures of heme-free ShuT, partially heme-bound ShuT, and heme-bound PhuT. These structures provide further insights into how the ligand binding architecture adapts to the requirements of substrate specificity.

MATERIALS AND METHODS
ShuT Crystallization-ShuT and PhuT bacterial transformations, sub-cloning, protein expression, and purification were carried out as previously described (1,2). Crystals of ShuT belonging to the triclinic space group P1 with unit cell dimensions a ϭ 54.82 Å, b ϭ 73.28 Å, c ϭ 72.38 Å, ␣ ϭ 71.44°, ␤ ϭ 77.8°, and ␥ ϭ 89.9°were grown by hanging-drop vapor diffusion at room temperature over a reservoir of 0.1 M sodium citrate, pH 4.0, 0.05 M ammonium bromide, and 15% polyethylene glycol 4000. Hanging drops consisted of 2 l of 15 mg/ml ShuT mixed with an equal volume of reservoir solution. Platelike crystals with dimensions 600 -400 m by 20-40 m thick grew within 5-10 days. These crystals tend to grow as large clusters. Although faint brownish in color, they were found to have no observable heme bound. To introduce heme for phasing, a second batch of ShuT crystals was grown in the presence of 1-5 mM hemin chloride (Aldrich), by hanging-drop vapor diffusion at room temperature. These crystals also belong to the triclinic P1 space group with unit cell dimensions a ϭ 54.7 Å, b ϭ 73.31 Å, c ϭ 73.32 Å, ␣ ϭ 70.25°, ␤ ϭ 79.02°, and ␥ ϭ 90.22°. Crystals with the addition of hemin chloride tend to grow larger and thicker with small satellites within 2 weeks or even as a single crystal, both appearing much darker in color. Based on the estimation of Matthew's coefficient (9), there are four molecules per asymmetric unit. Cryoprotectant soaking for the first batch of ShuT crystals consisted of a six-step transfer to artificial precipitant solution (0.1 M sodium citrate, pH 4.0, 0.05 M ammonium bromide, and 15% polyethylene glycol 4000) with increasing concentrations of glycerol from 5% to 30%. For the second batch of ShuT crystals, the same precipitant plus increasing concentrations of glycerol from 5% to 30% were used.
Data Collection for and Phasing Attempts for ShuT-All data were collected using crystals flash cooled in liquid nitrogen. A high resolution data set (50 -2.05 Å) for the first batch of ShuT crystals and a high resolution data set (50 -2.0 Å) for the second batch of ShuT crystals were collected on Stanford Synchrotron Radiation Laboratory beamline 9-2 with a MAR 325 chargecoupled device detector and on Advanced Light Source beamline 5.0.2 with an ADSC Quantum 315 charge-coupled device detector, respectively. The lack of fully occupied heme sites even for crystals grown in the presence of hemin precluded the use of iron multiwavelength anomalous dispersion (MAD) phasing. The most likely reason for poor heme incorporation is the low pH (pH 4.0) required for crystallization. At such low pH the heme propionates may be partially protonated, and deprotonation of the Tyr heme ligand is more difficult. The attempted molecular replacement using the vitamin B 12 periplasmic binding protein, BtuF (Protein Data Bank (PDB) code 1N2Z (7)), failed to provide a solution. This is not too surprising because the sequence identity was only 20.8%. Attempts to obtain crystals of selenomethionine ShuT were unsuccessful, and data obtained from 15 different heavy atom soaks failed to provide useful phasing information. It became clear that we needed to adopt a new strategy. We therefore turned to PhuT in hopes that this might crystallize in the hemebound form. This would enable iron MAD phasing to be used to solve the PhuT structure, which then could be used to solve the ShuT structure by molecular replacement.
PhuT Crystallization-Crystals of PhuT belonging to the hexagonal space group P6 3 22 with unit cell dimensions a ϭ b ϭ 132.9 Å, c ϭ 72.9 Å, and ␥ ϭ 120°were grown by hanging-drop vapor diffusion at room temperature over a reservoir of 0.1 M sodium cacodylate, pH 6.5, 0.25 M sodium chloride, 1.8 M ammonium sulfate (optimized Magic 96 condition #58, University of Washington). Vapor diffusion drops consisted of 2 l of 10 mg/ml PhuT mixed with an equal volume of reservoir solution. Rod crystals with length of 200 -300 m and 50 -75 m thickness grew within 1 month. These crystals tend to grow as large clusters, with a brownish color indicating good binding of heme. Based on the Matthew's coefficient calculation, there is only one molecule per asymmetric unit. Cryoprotectant soaking for PhuT consisted of a four-step transfer to artificial precipitant solutions (0.1 M sodium cacodylate, pH 6.5, 0.25 M sodium chloride, 1.8 M ammonium sulfate) with increasing concentrations of glycerol from 5% to 20%.
Data Collection for PhuT-All data were collected using crystals flash cooled in liquid nitrogen. A high resolution data set (50 -2.4 Å) was collected on Stanford Synchrotron Radiation Laboratory beamline 1-5 with a Q4 charge-coupled device detector. A MAD data set was collected at the Fe-K absorption edge, and a fluorescence scan was taken at the Fe-K absorption edge before data collection. Owing to the small size of PhuT crystals, only two x-ray wavelengths, inflection and remote, were used to prevent crystal decay from long x-ray exposures. Optimization of data collection was guided by the STRATEGY function of MOSFLM (10). Two sets of data were collected for two crystals. All data were reduced using DENZO and SCALEPACK, and rejections were performed with SCALEPACK. All data collection and process statistics are shown in Table 1.
Phase Calculations, Model Building, and Refinement for PhuT-Two MAD data sets at 3.14-and 2.8-Å resolution were collected using two wavelengths (inflection and remote) at the Fe-K edge of 1.74 Å. Phase information from both MAD data sets was calculated with HKL2MAP (11) and SOLVE (12). HKL2MAP gave a dЉ/ value of 1.43 (0.77) for MAD data set 1 and a dЉ/ value of 1.26 (0.96) for MAD data set 2. Values in parenthesis correspond to highest resolution shell. Both SHELXD and SOLVE found the same one iron site for two MAD data sets, which was in agreement with the predicted one molecule per asymmetric unit. The Z-score reported by SOLVE for the first MAD data set is 7.07 and for the second MAD data set is 5.31.
Electron density maps produced from either MAD data set alone with density modification using DM were quite noisy with no secondary structure elements clearly visible. SIGMAA (13,14) was used to combine phases from the two MAD data sets. The combined phases were further improved by density modification using DM (13,15) with solvent flattening or using SOLOMON (13,16) with solvent flipping. The electron density maps showed clear features for some major ␣-helices and ␤-strands. However, the electron density map from SOLOMON was better showing clearer density for side chains. A run of ARP/wARP Helix Build (13,17) was able to fit some helical fragments nicely into the electron density map produced from SOLOMON. Phases from ARP/wARP and the two MAD data sets then were combined and further improved by density modification using SOLOMON followed by stepwise phase extension from 3.2 to 2.40 Å. The final electron density map was of high quality, showing very clear density for both the main chain and the side chains.
Even though PhuT has only ϳ16.4% sequence identity to BtuF, BtuF proved useful as a guide in building PhuT structure. Map fitting was carried out with the graphic modeling package O (18). Regions of the map showing good connectivities were modeled first as polyalanines. Once enough main-chain atoms were built in, side-chain atoms were included in regions where the sequence was clear. Then the partial model was submitted to the simulated annealing refinement with CNS (19). The free R-factor was calculated with 5% of total reflections that were set aside throughout the refinement. Once the R-factor and free R-factor were Ͻ30%, simulated annealing was replaced by minimization followed by the isotropic B-factor refinement. When the R-factor reached ϳ25%, water molecules were automatically picked at a level of 3.5 in a F o Ϫ F c difference map using CNS, followed by visual inspection and manual adjustment. The final model consists of all the residues in the PhuT molecule except for the last 4 residues and has 11 extra residues at the C terminus introduced by cloning. The extra 11 residues is part of the commercial vector pET 101/D-TOPO (Invitrogen), which serves as a spacer between the natural C terminus and His tag used for purification. 158 water molecules, 2 sulfate molecules, and 2 glycerol molecules were also included in the final model with a crystallographic R-factor of 19.3% and the free R-factor of 24.7%. Refinement statistics are listed in Table 1.
ShuT Structure Solution-A high resolution data set of ShuT was collected at a resolution of 2.0 Å with an x-ray wavelength of 0.999 Å. Molecular replacement in the P1 space group was also carried out with PHASER (20) using PhuT as a search model. Good solutions were found when the full-length polyalanine PhuT was used as the search model. A log likelihood gain value of 3871 and a Z-score of 46.9 were obtained in PHASER for four solutions with minor packing clashes. The model phases were improved by running DM, including solvent flattening, histogram matching, and 4-fold non-crystallographic symmetry averaging. The final density map showed clear densities for side chains that were omitted in the search model.
Electron density map fitting beginning from the polyalanine PhuT model was carried out with the graphic modeling package O (18). Multiple rounds of model building and refinement with simulated annealing at 3000 K in CNS were carried out. Model building was on one chain only, and the models for the other three chains were generated by the non-crystallographic symmetry. Non-crystallographic symmetry restraints were imposed on main-chain atoms until the final stages of refinement. The starting R-factor was 0.420 (R-free ϭ 0.477) and dropped to 0.356 (R-free ϭ 0.414) after one round of anneal and temperature factor refinement. Model building and refinement continued until the final R-factor reduced to 0.21 (R-free ϭ 0.255). The final structure of ShuT consists of 499 water molecules. Final data collection and refinement statistics are summarized in Table 2.
is the sum over all reflections, and ⌺ i is the sum over all I measurements of reflection h. e Figure of merit (FOM) is the mean of the cosine of error in the phase angle: FOM ϭ ͉F(hkl) best ͉/͉F(hkl)͉, where F(hkl) best ϭ ⌺ ␣ P(␣)F hkl (␣)/⌺ ␣ P(␣). f R free ϭ R-factor calculated using 5% of the reflection data chosen randomly and set aside throughout refinement. a Native data set collected at Stanford Synchrotron Radiation Laboratory beamline 9-2 (first batch of crystals). b Native data set collected at Advanced Light Source beamline 5.0.2 (second batch of crystals). c The values in parentheses were obtained from the outermost resolution shell.
is the sum over all reflections, and ⌺ i is the sum over all I measurements of reflection h. e R free ϭ R-factor calculated using 5% of the reflection data chosen randomly and set aside throughout refinement.

RESULTS AND DISCUSSION
Overall Structures of PhuT and ShuT-The overall fold of PhuT and ShuT share common structural features, having two topologically similar globular domains and a long, rigid ␣-helix as an interdomain linker (Fig. 1). These structures most closely resemble that of the Class III periplasmic binding protein, BtuF, a vitamin B 12 -transporting protein (5,21), and FhuD, an Escherichia coli periplasmic protein that binds hydroxamate-type siderophores (8). Each domain consists of a 5-stranded ␤-sheet flanked by helices. Although almost all of the secondary structure elements described for the BtuF (5) are conserved in PhuT and ShuT, we do observe a few more small extra hel-   DECEMBER 7, 2007 • VOLUME 282 • NUMBER 49 ices in the two new periplasmic binding protein structures. By adopting the nomenclature in BtuF (5), there are additional small helices, ␣-1Ј-(47-50) in the N-terminal domain and ␣-9Ј-(245-248) and ␣-9Љ-(250 -254) in the C-terminal domain of PhuT. The r.m.s. deviation between BtuF and PhuT is 1.95 Å for 199 common C␣ atoms, and the r.m.s. deviation between BtuF and ShuT is 1.8 Å for 197 C␣ atoms, whereas the r.m.s. deviation between ShuT and PhuT for 232 common C␣ atoms is 1.4 Å. FhuD is also a Class III PBP, and the r.m.s. deviation between FhuD and PhuT is 2.1 Å for 174 common C␣ atoms, whereas the r.m.s. deviation between FhuD and ShuT is 1.9 Å for 165 common C␣ atoms.

Structures of Periplasmic Heme-binding Proteins
Heme Binding Cleft in PhuT and ShuT-Spectroscopic studies indicate that the heme is high spin pentacoordinate with the side-chain oxygen of a Tyr residue providing the one protein ligand (1,2). The spectral properties of ShuT are very similar to catalases, which also use a tyrosinate ligand (22). The crystal structure is fully consistent with the spectral results. The heme is embedded in a cleft between the N-and C-terminal domains (Figs. 1 and 2) and is pentacoordinate with Tyr 71 in PhuT from the C-terminal end of ␤-3 strand in the N-terminal domain providing the heme ligand. A characteristic feature of catalases shared by PhuT is an arginine in the proximal pocket, which H-bonds with the tyrosinate ligand. Arg 73 serves this function in PhuT (Fig. 2). One difference is that in PhuT, but not in catalases, Arg 73 also H-bonds to the heme propionate. In ShuT, the homolog to the PhuT Arg 73 is Lys 69 . Lys 69 is not near the ShuT heme ligand, Tyr 67 , so if Lys 69 H-bonds to Tyr 67 to mimic the Tyr 71 -Arg 73 interaction in PhuT, there must be a reorientation of the Lys 69 side chain. If, as it appears, ShuT lacks a good H-bonding partner to the Tyr ligand, this could help to explain why ShuT gives a higher Fe-O vibrational frequency obtained from resonance Raman studies (1). The O-Fe bond also is longer in PhuT, 2.3 Å compared with to 2.03 Å in beef liver catalase (2.5-Å resolution, pdb ID: 7CAT (22)) and 1.8 Å in Helicobacter pylori catalase (1.6 Å resolution, pdn ID: 1QWL (23)). The O-Fe-N bond angle ranges from 93 to 98°in PhuT and the two catalases, which is not very significant, but the longer bond distance is important. If ShuT is similar to PhuT, then the longer O-Fe bond is another reason for a higher vibrational frequency in ShuT. This also suggests that the heme is not as tightly held in place in ShuT/PhuT compared with catalases, which is functionally important because the role of ShuT/PhuT is to bind and release heme.
In PhuT the heme propionate groups are pointing outward toward the surface of the protein, and each is bent to the opposite sides of heme plane, respectively (Fig. 2). One propionate group is directly hydrogen bonded to Arg 228 from the C-terminal domain. Note that Arg 228 is stacked against the distal surface of heme.
There are four molecules present in each unit cell of ShuT. For ShuT crystals that were grown in the presence of hemin chloride, the electron density in molecule A shows partial density for heme (Fig. 2B), whereas molecules B, C, and D show no density for heme. Possible differences between the apo-and heme-bound structures can be derived by comparing the three molecules in the structure that do not have heme bound. In addition, we also solved the structure of ShuT that did not have hemin included in crystallization solutions and here, there are four molecules in the unit cell that are heme-free. The most obvious difference between the heme-free molecules and partially heme-bound molecule is in the orientation of the Tyr 67 heme ligand. Tyr 67 either points "in" toward the heme pocket where it is in position to coordinate the heme iron, or it flips out  toward the surface (Fig. 3). For the partially heme-bound crystals, Tyr 67 points inward in molecule A, because the heme is partially bound, but also points inward in molecule D, even though there is no heme density. In molecules B and C Tyr 67 points "out." For the apo crystal, Tyr 67 points inward in molecules A and B, but outward in C and D. Thus there is considerable flexibility in the orientation of Tyr 67 in the absence of heme. There also is a slight movement of the loop centered on Gln 88 toward the heme in the heme-bound molecule. There are, however, no noteworthy changes on the distal side of the heme. A similar Tyr reorientation has been observed in a periplasmic ferric binding protein and has been proposed to play a role in guiding iron into the site (24).
There are no major differences between ShuT and PhuT on the proximal ligand binding side of the heme with one exception. Val 266 (Fig. 4) in PhuT extends from a loop that is one residue shorter in ShuT. This provides a somewhat more com-pact environment around the Tyr heme ligand. However, there are substantial differences on the opposite distal side of the heme pocket. Fig. 4 highlights two key differences. First, a hairpin loop after ␤-6 strand centered on Gly 169 in ShuT and Gly 177 in PhuT is pointing downward into the heme binding pocket in ShuT, but upward in PhuT. Although the loop is made up of two Gly residues at the turn, the down position found in ShuT still occupies a large space on the distal side of the heme, making potential Van der Waals contacts with heme. Second, a hairpin loop in PhuT dips down into the heme pocket, which allows Arg 228 to lie over the face of the heme and interact with one of the heme propionates. The corresponding region in ShuT consists of a small one-turn helix that is folded away from the heme and is centered on Gln 222 (Fig. 4). ShuT has no homolog that can interact with the heme propionates as can Arg 228 in PhuT.
Comparison with BtuF and FhuD-Class one PBPs such as the maltose-binding protein undergo a large change in struc-  ture upon ligand binding as the domains close down around the ligand. The ligand-free and -bound structures of BtuF show that the change upon vitamin B 12 binding is modest, resulting in at most a 1-Å shrinking of the ligand binding pocket (5). Because the ShuT apo and partially heme-bound structures are in the same crystal form, it is possible that the crystal lattice prevents important structural changes required for heme binding. Even so, because ShuT/PhuT are structurally so similar to BtuF, it is doubtful that heme binding leads to major changes in structure other than a reorientation of the Tyr heme ligand and a slight movement of the loop centered on Gln 88 toward heme. However, the conformational changes in the heme binding pocket, such as the reorientation of the Tyr heme ligand, may play a role in the heme binding and/or release processes. By analogy with BtuF heme binding probably has no large effect on structure, then differences in the heme pocket between PhuT and ShuT are not the result of heme binding but rather a major difference in heme pocket architecture. Fig. 5 provides a comparison of the binding cleft in PhuT, BtuF, and FhuD. Both PhuT and BtuF form well defined deep clefts, whereas FhuD forms a depression on the molecular surface. The less well defined pocket in FhuD may reflect a lack of selectivity in FhuD compared with PhuT and BtuF. In addition, the ligands for PhuT and BtuF are relatively flat aromatic groups with a large surface area that requires a deep cleft for binding. Another similarity is the positioning of the B 12 or heme relative to the "proximal" ligand. In BtuF B 12 binds in the "baseon" conformation with the N3B nitrogen of the dimethylbenzimidazole coordinated to the cobalt. In PhuT the heme ligand, Tyr 71 , occupies approximately the same position. Therefore, the BtuF pocket is somewhat larger because, unlike heme where the protein provides a metal ligand, the B 12 substrate provides its own dimethylbenzimidazole ligand. A more detailed comparison of the pockets shows that the B 12 ring is rotated ϳ45°r elative to the heme in PhuT (Fig. 6). Even so, the major elements of secondary structure on the proximal ligand binding side of the substrate are conserved. For example, the loop centered on Trp 66 in BtuF corresponds to the loop containing the PhuT heme ligand, Tyr 71 (Fig. 6). The main difference is that this loop in BtuF is positioned closer to the ligand thus allowing Trp 66 to interact with the vitamin B 12 ring. However, like the comparison between ShuT and PhuT, the architecture on the opposite distal surface of the substrate is quite different. The main difference is centered on Arg 228 in PhuT, which approximates to Gly 216 in BtuF. PhuT has four extra residues in this loop that enables Arg 228 to dip into the distal cavity where it can interact with the heme propionate.
In summary, the ShuT and PhuT structures illustrate a very similar architecture on the proximal Tyr ligand binding side of the heme. This was not unexpected given the stereochemical restrictions in forming a Tyr-Fe coordination bond. The con-servation of secondary structural elements on the ligand binding side extends to BtuF as well, although the proximal side elements of secondary structure must reposition relative to PhuT/ShuT to accommodate the B 12 dimethylbenzimidazole ligand. However, the substantial differences on the opposite distal side of the heme between ShuT and PhuT were unexpected. Such diversity could not be tolerated in heme enzymes where the distal pocket often provides essential catalytic groups that are evolutionarily conserved. However, the role of PhuT and ShuT is to bind and deliver heme and not to catalyze critical reactions, and thus, diversity in the heme pocket can be tolerated.