Bdellovibrio bacteriovorus phosphoglucose isomerase structures reveal novel rigidity in the active site of a selected subset of enzymes upon substrate binding

Glycolysis and gluconeogenesis are central pathways of metabolism across all domains of life. A prominent enzyme in these pathways is phosphoglucose isomerase (PGI), which mediates the interconversion of glucose-6-phosphate and fructose-6-phosphate. The predatory bacterium Bdellovibrio bacteriovorus leads a complex life cycle, switching between intraperiplasmic replicative and extracellular 'hunter' attack-phase stages. Passage through this complex life cycle involves different metabolic states. Here we present the unliganded and substrate-bound structures of the B. bacteriovorus PGI, solved to 1.74 Å and 1.67 Å, respectively. These structures reveal that an induced-fit conformational change within the active site is not a prerequisite for the binding of substrates in some PGIs. Crucially, we suggest a phenylalanine residue, conserved across most PGI enzymes but substituted for glycine in B. bacteriovorus and other select organisms, is central to the induced-fit mode of substrate recognition for PGIs. This enzyme also represents the smallest conventional PGI characterized to date and probably represents the minimal requirements for a functional PGI.


Introduction
Bdellovibrio bacteriovorus is a small predatory deltaproteobacterium that preys on other Gram-negative bacteria. Bdellovibrio bacteriovorus leads a complex biphasic life cycle switching between extracellular predatory and intraperiplasmic replicative phenotypes; using swimming and/or gliding motility, B. bacteriovorus seeks out prey and, upon encounter, attaches to and invades the periplasm of the prey cell. Within the periplasm, B. bacteriovorus remodels the prey cell into a transient structure called the bdelloplast, wherein it progressively metabolizes the cytoplasm and periplasm of its host and uses the nutrients it harvests to grow and divide into new progeny cells. Upon exhaustion of nutrients, B. bacteriovorus progeny lyse their host cell and exit in search of new prey cells to repeat the life cycle. With the sharp rise of antibiotic resistance, B. bacteriovorus has shown promise and is currently being investigated as a 'living antibiotic' option for treating multidrug-resistant Gram-negative bacterial pathogens [1].
Prior studies have indicated that B. bacteriovorus is unlikely to use polysaccharides as the primary substrates for energy metabolism and instead respires amino acids (aa) to provide the energy needed for intraperiplasmic growth [2]. In agreement, high activities of citric acid cycle enzymes and low activities of glycolytic enzymes are observed in B. bacteriovorus cell extracts, with the notable exceptions of phosphoglucose isomerase (PGI) and glyceraldehyde 3-phosphate dehydrogenase [3]. Sequencing of the B. bacteriovorus genome has revealed a full complement of enzymes required for the production of ATP through glycolysis [4]. Why high B. bacteriovorus PGI (BbPGI) activity is detected in B. bacteriovorus cell extracts, when glycolysis is not the primary route for energy production, is unknown [3].
PGI mediates reversible isomerization of glucose-6phosphate (G6P) to fructose-6-phosphate (F6P) in glycolysis and gluconeogenesis. The catalytic mechanism of PGI mediated isomerization has been well documented using mammalian PGIs as model systems [5][6][7][8][9]. The process begins with the PGI opening the substrate ring before mediating a proton transfer, using a cis-enediol mechanism, between the C1 and C2 positions before ring closure. The mechanism of bacterial PGIs is assumed to be identical, owing to bacteria having homologous structures with similar domain arrangements. However, while many bacterial PGIs have been structurally characterized, relatively few have been solved in complex with a substrate. In addition, no structure is currently available from a deltaproteobacterial source.
In addition to the classical role of isomerization, PGIs often function as 'moonlighting' proteins. In humans, PGI moonlights as autocrine motility factor, neuroleukin, and differentiation and maturation mediator [10][11][12]. Conversely, in Lactobacillius crispatus, PGI moonlights as a pH-dependent adhesion protein [13]. Thus, PGI enzymes can have diverse functions in addition to their roles in central metabolism, and there are likely additional roles for this family yet to be discovered. Bdellovibrio bacteriovorus PGI is part of a novel grouping of smaller PGIs which are typically approximately 100 aa smaller than other conventional PGIs (for example, Escherichia coli PGI is 549 aa compared to only 408 aa for BbPGI). All structurally characterized PGIs reside in one of two groupings, those above 540 aa and those below 450 aa in size. To gain valuable insights into the role of BbPGI, we have obtained two crystal forms of BbPGI which we have solved to 1.74 Å and 1.84 Å. Using ligand-soaking experiments, we have obtained a crystal structure of BbPGI bound to its substrate. Through these crystal structures, we determine that the BbPGI active site, in contrast to those of previously characterized PGIs, does not undergo significant conformational change upon substrate binding. This lack of active site mobility can be partially attributed to the substitution of a Phe residue, conserved in most PGIs, for a Gly residue in BbPGI. We also note a novel feature in the smaller PGIs wherein the hook region is in a different position to that of the larger PGI structures.

Overall structure of BbPGI
To determine the BbPGI structure we used full-length recombinant BbPGI protein in crystallography experiments. Crystallization trials yielded two crystal forms of BbPGI which were processed in P12 1 1 and P3 1 21 spacegroups (1.74 Å and 1.84 Å, respectively). Data from the P3 1 21 form were solved by molecular replacement using PDB 1B0Z from Geobacillus stearothermophilus as a search model [14]. Data processing and refinement statistics are displayed in table 1. The P3 1 21 form structure was used as a search model for the molecular replacement solution of the P12 1 1 dataset. The P3 1 21 form had a single copy of BbPGI comprising the asymmetric unit, whereas the P12 1 1 dataset asymmetric unit is composed of eight copies. The eight copies in the P12 1 1 dataset form four largely identical dimers (root mean square deviations of 0.107-0.190 Å between dimers). A dimeric assembly also exists in the P3 1 21 form but is generated via crystallographic symmetry axes. Dimer partners are related to one another by a two-fold rotational symmetry. The polypeptide backbone for residues 3-406 can be fully traced into the electron density map, with only the first and last two aa of the polypeptide sequence and the His-tag being disordered and absent from the final refined models.
The core structure of the BbPGI is similar to other characterized PGI structures but lacks much of the loop decoration characteristic of larger PGIs (figure 1). BbPGI is composed of three domains: the large domain, consisting of residues 1-31 and 213-384 which forms a mixed 5-strand β-sheet sandwiched by α-helices; the small domain, consisting of residues 44-195 which forms a parallel 5-strand β-sheet flanked by α-helices and the C-terminal arm subdomain, a single α-helix composed of residues 392-408 (figure 1a). The small and large domains are linked to each other by residues 32-43 and 196-212, with residues 199-211 forming an α-helix. A small loop (residues 385-391) connects the large domain to the C-terminal arm subdomain. A hook region (residues 306-348) extends from the large domain and packs against the small domain of the dimer partner providing dimer-stabilizing hydrogen bonding and hydrophobic interactions (figure 1b). We note that positioning of the hook region varies in PGIs, with the interaction loop extending from either the N-terminus or C-terminus of α20 (BbPGI helix numbering) to interact with either the small domain or large domain of the dimer partner, respectively (electronic supplementary material, figure S1). In PGIs of size >540 aa the loop extends from the C-terminal position to contact the large domain (determined using 15 unique representative structures in the PDB). Conversely, in BbPGI and PGIs of less than 450 aa, an alternative conformation is adopted with an extension of the loop from the N-terminus of α20 (six unique representative structures). This is an intuitive observation since α20 tilts away from the dimer partner, with the N-terminus being much closer for contact to the dimer partner than the C-terminus, thus requiring fewer residues to make dimer contact. The C-terminal arm subdomain in BbPGI sandwiches the hook region between itself and the small domain, whereas in larger PGIs the hinge region wraps around the outside of the C-terminal arm subdomain.
All three domains contribute to the proposed active site, with the small domain positioned to coordinate the phosphoryl group of the incoming substrate. Active site architecture is completed by recruitment of the large domain of a dimer partner. This is necessary for catalysis since the dimer partner donates the conserved residue H285 required for sugar ring-opening/closing. BbPGI is one of the smallest conventional PGI family enzymes reported to date, with distance matrix alignment analysis revealing strong structural similarity to PGIs from the thermophiles Thermotoga maritima (PDB 2Q8N, Z-score: 48.6) and Thermus thermophilus HB8 (PDB 1ZZG, Z-score: 48.2) [15]. The BbPGI dimeric assembly is supported by a PISA calculated interface area of 4811 Å 2 (25% of total monomer surface area) between dimer partners, suggesting the dimer to be physiologically relevant [16]. However, the T. thermophilus HB8 PGI exists in solution as a dynamic equilibrium between monomer and dimer and, royalsocietypublishing.org/journal/rsob Open Biol. 11: 210098 owing to the high structural similarity, BbPGI may also exist in a similar equilibrium state [15].

The BbPGI active site and ligand accommodation/ coordination
Although the conformation of the active site is similar to larger PGIs we decided to validate substrate binding in this smaller grouping of compact PGIs. To do so, we performed a ligand-soaking experiment with G6P using the P3 1 21 crystal form of BbPGI. Strong F o -F c difference density was observed in the active site and could be unambiguously modelled as an extended, open ring phosphosugar (figure 2). Since G6P can be interconverted to F6P we queried which had bound to the active site; unfortunately, electron density is most diffuse at the C1 position and it is difficult to conclusively state whether a tetrahedral (G6P) or trigonal planar (F6P) geometry is present at the C2 position. However, the F6P O2 can be more confidently modelled into the electron density than the G6P O2, thus we have tentatively modelled F6P into the active site and will refer to it as such hereafter. It is possible this ambiguity is the result of a mixed population of open-ring G6P and F6P being bound.
In the active site, the phosphoryl group of F6P is extensively coordinated by the small domain through the hydroxyl-containing sidechains of S72, S118, S120 and T123, the backbone nitrogen of K119 and S120, and three ordered waters (figure 2). F6P is also coordinated through O1 interacting with the backbone nitrogen of R180 (small domain) and ordered water. F6P interactions are further provided by: O3 bonding with the backbone nitrogen of H285, O4 binding the backbone nitrogen of G71 and ordered water, and O5 interacting with the sidechain of H285 (figure 2). Active site residues known to be involved in mediating isomerization (R180, E256, H285 and K398) are positioned and orientated similarly to other PGIs (figure 3).

Conformational changes at the BbPGI active site
Upon ligand binding in other PGIs, relatively mobile active site residues facilitate hydrogen bonding with the phosphoryl group of G6P or F6P. This is evident in the unliganded, F6P and 5-phospho-D-arabinonate bound structures of rabbit  Interestingly, all PGI structures less than 450 aa use a Gly, while in PGI structures of more than 540 aa this residue is a Phe. A second conformational change is observed in rPGI upon F6P binding, wherein residues 385-389 are also recruited to the active site repositioning H388 for catalysis (figure 3a) [7]. Our structure undergoes no conformational changes in the equivalent histidine residue (H285) upon ligand-binding. Previous work using cis-enediol(ate) intermediate mimics in rPGI noted a conformational change in residues 513-521 which brought K518 towards the bound ligand. Again, we observe no conformational changes in the equivalent residues (393-401) upon F6P binding, agreeing with previous G6P complex structures [9,17]. Only cis-enediol(ate) intermediates may drive this particular conformational change. We believe the 'closed state' observed in the unliganded structure is not an artefact of crystal packing since identical structures are observed in two different crystal forms, which pack against this region differently. In addition, although they make some minor interactions with symmetry-related chains, residues that contribute to the 'closed state' are not subject to close crystal packing effects in either crystal form.

Discussion
In solution, G6P and F6P exist primarily in cyclic hemiacetal and hemiketal forms [18]. Our structure presents F6P bound in an open ring conformation, strongly suggesting the enzyme has facilitated ring opening. In addition, the placement of active site residues is consistent with other active PGIs. It is interesting that BbPGI undergoes no conformational changes upon ligand-binding unlike other characterized PGIs. Certainly, it is apparent that conformational movement in the polypeptide backbone is not a prerequisite for isomerase activity, which may also be the case in other, similar PGIs. The Leishmania mexicana eukaryotic parasite PGI has also been demonstrated to undergo no ligand driven conformational changes; however, unlike our structure, it is held in the 'open' conformation rather than the 'closed' state (the authors suggest the loop would still be required to move approximately 5 Å during isomerization in order to be within a suitable distance of the catalytic glutamate residue) [19]. The positioning of a Gly (G121) instead of a Phe residue in the active site may also be indicative of PGIs which do not reposition loops for binding of the phosphoryl group. The structure of BbPGI (408 aa) is significantly smaller than other bacterial/ mammalian PGIs (normally more than 500 aa) and matches PGIs from thermophiles. Certainly, this structure is likely to approximate the minimum structural requirements for an active PGI. It would be interesting if the compact nature of BbPGI was an adaptation particular to the predatory lifestyle of B. bacteriovorus. The reasoning as to why other PGIs have not evolved to be more compact and still have ligand-induced conformational changes is unknown and may relate to their active sites/states being involved in other moonlighting functions. From our structures, it is not obvious why PGI activity is noticeably higher than other glycolytic processes in B. bacteriovorus cell extracts.
We have observed that the open reading frame of Bbpgi is separated by only 38 nucleotides from the diguanylate cyclase dgcb, although they are unlikely to share an operon  royalsocietypublishing.org/journal/rsob Open Biol. 11: 210098 ( personal communication with Luke Ray in R. E. Sockett lab). Deletion of dgcb abolishes the ability of B. bacteriovorus to initiate host cell invasion [20]. We have previously characterized DgcB as a diguanylate cyclase probably activated by a phosphorylation event, which drives the production of a localized pool of cyclic-di-GMP that signals to downstream proteins to initiate invasion [21]. DgcB is composed of a GGDEF domain appended to a forkhead-associated (FHA) domain [21]. While GGDEF-domain-containing proteins are common in bacteria, the GGDEF-FHA domain arrangement has limited distribution. In closely related species to Bdellovibrio which do possess this arrangement, such as Myxococcus xanthus DK 1622, GGDEF-FHA domain genes (mxan4029, mxan1525 and mxan5199) are distant from the gene encoding M. xanthus PGI (mxan6908), revealing the close genetic distance between BbPGI and dgcB to be Bdellovibrio specific. Remarkably, enzymes from metabolic pathways have been isolated from B. bacteriovorus through a c-di-GMP capturing compound [22,23]. Additionally, c-di-AMP has also been implicated in binding metabolic enzymes, and G6P has recently been directly correlated to regulating c-di-GMP levels [24,25]. Considering the propensity of PGI enzymes to take on moonlighting roles we wondered whether BbPGI links c-di-GMP signalling to metabolism by directly interacting with DgcB. We carried out pull-down assays to test this hypothesis, however results indicated that they did not interact under the conditions assayed (electronic supplementary material, figure S2). We cannot, however, exclude the possibility that the wild-type variants of DgcB (c-di-GMP bound dimer) and/or BbPGI exist in a state non-permissive for interaction or require an additional third protein complexed to facilitate interaction, or that G6P/F6P regulate c-di-GMP signalling through an unknown mechanism.
In summary, BbPGI is representative of the minimal PGI unit and highlights that, through substitution of the Phe residue found in most PGIs (rPGI F212) to a Gly, a less drastic or absent conformational change in the active site is likely to occur in enzymes with this variant. Whether this lack of mobility impacts upon PGI catalysis and its ability to interact with other ligands and protein partners remains an open question.

Cloning and protein production
The full open reading frame for Bbpgi (UniProt [26] entry Q6MPU9, gene bd0741, aa 1-408) was amplified from B. bacteriovorus HD100 genomic DNA using primers detailed in electronic supplementary material, table S1. Amplified construct DNA was inserted into a modified pET41c plasmid (Novagen, GST-tag removed and thrombin cleavable 8xHistag introduced at C-terminus) by restriction-free cloning [27]. Correct plasmid construct was confirmed by sequencing. Plasmid was transformed into E. coli expression strain BL21λDE3 (New England Biolabs) for expression. Cells were grown in a shaking incubator (180 r.p.m.) at 37°C in 2× LB supplemented with 100 µg ml −1 Kanamycin until an OD 600 of 0.6-1 was reached. Expression was induced with 0.5 mM IPTG before 15°C overnight shaking incubation. Cells were harvested by centrifugation at 6675g for 6 min (4°C) before being frozen at −20°C.

Protein purification
Cell pellets were defrosted and resuspended in Buffer A (20 mm Imidazole pH 7.0, 400 mM NaCl and 0.05% Tween 20) with lysozyme. Cells were lysed on ice by sonication and lysate was clarified by centrifugation at 48 400 g for 1 h (4°C). Supernatant was loaded onto buffer A equilibrated HisTrap FF nickel columns (GE Healthcare). Columns were washed with 12 CVs of buffer A before two 20 ml elution steps at 8% buffer B (400 mm Imidazole pH 7.0, 400 mM NaCl and 0.05% Tween 20) in buffer A and 100% buffer B. The purity of samples was confirmed by sodium dodecyl sulfate-polyacrylamide gel electrophoresis. Eluted protein was dialysed overnight at 4°C against 20 mM HEPES pH 7.0 and 200 mM NaCl. Protein was concentrated to 40 mg ml −1 using Vivaspin spin-concentrators (Sartorius) and snap-frozen.

Crystallization and structure determination
Protein at 40 mg ml −1 was screened for conditions to promote crystallization (1 : 1 ratio of protein to reservoir solution, 4 µl total drop size). Diffraction quality crystals were obtained in 0.2 M ammonium acetate, 0.1 m sodium acetate pH 5.0 and 20% PEG 4000 (crystal form 1), and 0.2 M ammonium acetate, 0.1 M sodium acetate pH 4.0 and 15% PEG 4000 (crystal form 2). Crystals were cryoprotected in mother liquor supplemented with 20% ethylene glycol (v/v) before being flash cooled in liquid nitrogen. To obtain the F6P bound BbPGI structure, crystal form 1 was soaked in mother liquor supplemented with 3 mM G6P and 20% ethylene glycol for 5 min before being flash cooled in liquid nitrogen. Diffraction data were collected at Diamond Light Source in Oxford, UK. Data reduction and processing were performed through the xia2 suite [28][29][30]. Crystal form 1 (P3 1 21 spacegroup) was successfully phased by molecular replacement in MRage using PDB accession 1B0Z as a search model [31]. Phenix auto build was used to partially rebuild incorrect sections of the model [32]. Crystal form 2 (P12 1 1 spacegroup) and the G6P soaked crystal form 1 data were solved by MRage molecular replacement using crystal form 1 as a search model. Models were finalized through iterative manually building in Coot and refinement in REFMAC [33,34]. Structure representations were produced in Chimera and ChimeraX [35,36].

Sequence analysis
Protein sequences were aligned through Clustal Omega and figures produced through ESPript3 [37,38]. Data accessibility. All data have been provided to the PDB under accession codes 7NSS, 7O0A and 7NTG.
The data are provided in electronic supplementary material [39].