Journal list menu

Volume 282, Issue 16 p. 3030-3042
Original Article
Free Access

Structure and proposed mechanism of l-α-glycerophosphate oxidase from Mycoplasma pneumoniae

Callia K. Elkhal

Callia K. Elkhal

Department of Biochemistry and Biophysics, Oregon State University, Corvallis, OR, USA

These authors contributed equally to this workSearch for more papers by this author
Kelsey M. Kean

Kelsey M. Kean

Department of Biochemistry and Biophysics, Oregon State University, Corvallis, OR, USA

These authors contributed equally to this workSearch for more papers by this author
Derek Parsonage

Derek Parsonage

Center for Structural Biology, Wake Forest School of Medicine, Winston-Salem, NC, USA

Search for more papers by this author
Somchart Maenpuen

Somchart Maenpuen

Department of Biochemistry, Faculty of Science, Burapha University, Chonburi, Thailand

Search for more papers by this author
Pimchai Chaiyen

Pimchai Chaiyen

Department of Biochemistry and Center of Excellence in Protein Structure and Function, Faculty of Science, Mahidol University, Bangkok, Thailand

Search for more papers by this author
Al Claiborne

Al Claiborne

Center for Structural Biology, Wake Forest School of Medicine, Winston-Salem, NC, USA

Search for more papers by this author
P. Andrew Karplus

Corresponding Author

P. Andrew Karplus

Department of Biochemistry and Biophysics, Oregon State University, Corvallis, OR, USA


P. Andrew Karplus, 2011 ALS Bldg, Department of Biochemistry and Biophysics, Oregon State University, Corvallis, OR 97331, USA

Fax: +1 (0)541 737 0481

Tel: +1 (0)541 737 3200

E-mail: [email protected]

Search for more papers by this author
First published: 16 February 2015
Citations: 19


The formation of H2O2 by the FAD-dependent l-α-glycerophosphate oxidase (GlpO) is important for the pathogenesis of Streptococcus pneumoniae and Mycoplasma pneumoniae. The structurally known GlpO from Streptococcus sp. (SspGlpO) is similar to the pneumococcal protein (SpGlpO) and provides a guide for drug design against that target. However, M. pneumoniae GlpO (MpGlpO), having < 20% sequence identity with structurally known GlpOs, appears to represent a second type of GlpO that we designate as type II GlpOs. In the present study, the recombinant His-tagged MpGlpO structure is described at an approximate resolution of 2.5 Å, solved by molecular replacement using, as a search model, the Bordetella pertussis protein 3253 (Bp3253), comprising a protein of unknown function solved by structural genomics efforts. Recombinant MpGlpO is an active oxidase with a turnover number of approximately 580 min−1, whereas Bp3253 showed no GlpO activity. No substantial differences exist between the oxidized and dithionite-reduced MpGlpO structures. Although, no liganded structures were determined, a comparison with the tartrate-bound Bp3253 structure and consideration of residue conservation patterns guided the construction of a model for l-α-glycerophosphate (Glp) recognition and turnover by MpGlpO. The predicted binding mode also appears relevant for the type I GlpOs (such as SspGlpO) despite differences in substrate recognition residues, and it implicates a histidine conserved in type I and II Glp oxidases and dehydrogenases as the catalytic acid/base. The present study provides a solid foundation for guiding further studies of the mitochondrial Glp dehydrogenases, as well as for continued studies of M. pneumoniae and S. pneumoniae glycerol metabolism and the development of novel therapeutics targeting MpGlpO and SpGlpO.


Structural data have been deposited in the Protein Data Bank under accession numbers 4X9M (oxidized) and 4X9N (reduced).


  • AML
  • artificial mother liquor
  • Bp
  • Bordetella pertussis
  • DAAO
  • d-amino acid oxidase
  • DH
  • l-α-glycerophosphate dehydrogenase
  • DHAP
  • dihydroxyacetone phosphate
  • Glp
  • l-α-glycerophosphate
  • GlpO
  • l-α-glycerophosphate oxidase
  • Mp
  • Mycoplasma pneumoniae
  • Sp
  • Streptococcus pneumoniae
  • Ssp
  • Streptococcus sp
  • Introduction

    Mycoplasma pneumoniae is a human respiratory tract pathogen that causes 40% or more of community-acquired pneumonias [1], with the typical syndrome being tracheobronchitis in children. This pathogen initiates the colonization of the host airway mucosal epithelium via a special attachment organelle [2], which also provides an important gliding function. Among other distinctive features of M. pneumoniae are a small genome, the lack of a rigid cell wall and limited metabolic capabilities. As in the Gram-positive streptococci from which the mycoplasmas diverged approximately 600 million years ago [1], the tricarboxylic acid cycle, an electron transport chain and respiratory cytochromes are absent. However, glycerol metabolism appears to be an important pathogenicity factor for M. pneumoniae [3-6] and thus the enzymes involved provide potential drug targets for combating respiratory infectious diseases.

    Of particular interest in this context is the M. pneumoniae glpD gene (MPN051), which is annotated as encoding a glycerol 3-phosphate dehydrogenase [4] and, based on its sequence, can be identified as a member of the d-amino acid oxidase (DAAO) superfamily [7] of FAD-dependent enzymes combining a ‘glutathione-reductase-2’ type FAD-binding domain and an antiparallel β-sheet based substrate-binding domain. Despite the annotation as a dehydrogenase, the encoded enzyme has been shown to be a constitutively expressed cytosolic FAD-dependent l-α-glycerophosphate oxidase (GlpO) (EC [4], using O2 as the final electron acceptor and producing dihydroxyacetone phosphate (DHAP) and H2O2 (Scheme 1).

    Details are in the caption following the image
    Reaction catalyzed by GlpO.

    We therefore refer to the encoded protein as M. pneumoniae GlpO (MpGlpO), rather than GlpD. The gene is tightly linked with that for glycerol kinase (glpK) and, together, these enzymes can catalyze the ATP-dependent conversion of glycerol to the glycolytic intermediate dihydroxyacetone phosphate [4]. The peroxide produced by MpGlpO has been shown to be crucial for pathogenicity, and the ortholog from the animal pathogen, Mycoplasma mycoides subsp. mycoides SC, has also been implicated as a primary virulence factor [3]. Similarly, Mahdi et al. [8] have shown that Streptococcus pneumoniae GlpO (SpGlpO) is responsible for H2O2-mediated cytotoxicity against human brain microvascular endothelial cells, as well as for promoting pneumococcal meningitis.

    Given the clinical significance of M. pneumoniae and S. pneumoniae and the importance of MpGlpO and SpGlpO in cytotoxicity and virulence, structural studies of these enzymes would provide a valuable foundation for drug design. A good understanding of the SpGlpO structure is in hand because it is quite similar (approximately 62% sequence identity) to the structurally known Streptococcus sp. GlpO (SspGlpO) [9]. However, insufficient information exists for modeling the 43-kDa MpGlpO enzyme, because it is remarkably divergent from the higher molecular weight (~ 65 kDa) streptococcal enzymes: it is not only missing a C-terminal α-helical domain, but also it has only approximately 20% sequence identity with SspGlpO and is more sequence similar to other DAAO superfamily enzymes, such as glycine oxidase, than it is to SspGlpO. In the present study, we report its crystal structure, aiming to provide a foundation for understanding catalysis and guiding drug design against MpGlpO. As part of this work, we also carried out functional characterization of the protein encoded by Bordetella pertussis gene 3253 (Bp3253), which is a related protein of unknown function solved by the NorthEast Structural Genomics group [Protein Data Bank (PDB) entry 3DME; deposited 2008]. It became of interest because it is the structurally known protein most similar to MpGlpO (27% sequence identity).

    Results and Discussion

    Expression and biochemical properties of MpGlpO and the B. pertussis protein Bp3253

    Previously published studies of MpGlpO were carried out with an N-terminal His-tagged protein derived from M. pneumoniae M129 [4]. In the present study, the pET28a vector introduces a 34-residue His-tag in-frame with the codon-optimized synthetic gene, and this tag was present in all studies. The characterization of B. pertussis Bp3253 was carried out employing the same N-terminally His-tagged construct used in the crystal structure determination of Protein Data Bank (PDB) entry 3DME. Both recombinant MpGlpO and Bp3253 were expressed and purified (yielding ~ 5 and ~ 60 mg·L−1 culture, respectively), and the visible absorption spectra for the purified proteins are characteristic of properly folded flavoenzymes (Fig. 1). The native molecular weight for His-tagged MpGlpO, as determined by gel filtration to be 41.7 kDa, is reasonably close to the value of 46.3 kDa calculated for His-tagged MpGlpO, indicating that the recombinant protein is a monomer in solution.

    Details are in the caption following the image
    Sulfite titrations of MpGlpO and Bp3253. (A) MpGlpO (35.4 μm in 0.8 mL of 50 mm potassium phosphate, 0.5 mm EDTA, pH 7.0) was titrated aerobically with a 1 m sodium sulfite solution. The spectra shown correspond to the addition of 0 mm (black), 1.43 mm (blue), 4.27 mm (green), 11.3 mm (orange) and 47.6 mm (red) total sulfite. Inset: A450 as a function of added sulfite. (B) The titration of Bp3253 (36.2 μm in 1 mL) was carried out as described in (A), with 25 and 100 mm solutions of sodium sulfite. The spectra shown are with 0 mm (black), 0.15 mm (blue), 0.50 mm (green), 0.80 mm (orange) and 2.0 mm (red) total sulfite. The inset shows the changes in A449 as sulfite is added.

    Although the oxidase activity of MpGlpO was described previously [4], we assayed the recombinant His-tagged MpGlpO purified in the present study and found a specific activity of 12.8 U·mg−1 at 25 °C with 200 mm d/l-Glp and approximately 260 μm O2 (the saturating concentration in water at 25 °C). This is equivalent to a kcat of 9.7 s−1 (i.e. 580 min−1). This level of activity is reasonable for an oxidase, even though it is somewhat lower than that seen for native and recombinant GlpOs from E. casseliflavus (EcassGlpO) [10] and Streptococcus sp. (SspGlpO) [11] which gave kcat values of 70–90 s−1 at 25 °C. When Bp3253 was tested using the same assay, it showed no ability to catalyze the Glp-dependent formation of H2O2.

    Many flavoprotein oxidases show a pronounced reactivity with sulfite to form a reversible adduct at the flavin N5-position [12]. The EcassGlpO and SspGlpO give similar values of Kd = 0.82 mm and 1.5 mm [10, 11] and, in the present study, we performed similar sulfite titrations with freshly prepared MpGlpO and Bp3253 (Fig. 1A). For MpGlpO, the measured Kd was 3.0 mm, although flavin absorbance was only partially bleached with approximately half of the MpGlpO flavin being refractory to sulfite adduct formation, even after an incubation of 30 min at 48 mm sulfite. A parallel experiment with the Bp3253 protein (Fig. 1B) shows a monophasic sulfite titration giving essentially complete bleaching of the flavin absorbance spectrum with a Kd of 0.42 mm, being approximately seven-fold more favorable than that for MpGlpO. At present, we do not have any good proposals explaining why only approximately half of the MpGlpO reacts with sulfite.

    Structure determination of oxidized and reduced MpGlpO

    Crystals of recombinant His-tagged MpGlpO yielded diffraction data to an approximate resolution of 2.5 Å. Attempts to solve the structure by molecular replacement with a structure of SspGlpOΔ (PDB code: 2RGH) as the search model were unsuccessful, which was not unexpected given the low, approximately 20%, sequence identity between MpGlpO and SspGlpO. A PDB database query for better search models led to the identification of the B. pertussis protein Bp3253 (PDB code: 3DME; 27% identity) as the known structure with the highest sequence similarity. Using this as a search model, molecular replacement was successful and led to structures of an approximate resolution of 2.5 Å with acceptable statistics (Table 1) for both oxidized and dithionite-reduced MpGlpO.

    Table 1. Data collection and refinement statistics for MpGlpO structures. Numbers in parentheses represent data for the high-resolution shell
    Oxidized Reduced
    PDB code 4X9M 4X9N
    Data quality statistics
    Wavelength (Å) 0.9765 0.9765
    Space group P23 P23
    Cell dimensions, c (Å) 111.59 111.61
    Resolution range (Å) 50–2.4 (2.53–2.4) 60–2.5 (2.64–2.5)
    Reflections 430 576 836 296
    Unique reflections 18 401 16 368
    Completeness (%) 100 (100) 100 (100)
    Multiplicity 23.4 (23.6) 51.1 (28.6)
    R pim 0.022 (0.19) 0.072 (0.28)
    R meas 0.109 (0.90) 0.533 (1.51)
    I 22.7 (4.4) 19.9 (3.0)
    Refinement statistics
    Rwork (%) 14.8 (21.5) 14.6 (23.8)
    Rfree (%) 20.4 (24.6) 21.0 (34.9)
    Number of amino acid residues 384 384
    Number of solvent atoms 201 201
    Number of non-hydrogen atoms 3301 3309
    <B> protein (Å2) 35 42

    The refined models contain a single MpGlpO chain in the asymmetric unit including residues 1–384, the FAD cofactor, ordered water sites and a Ni2+ atom. The 2Fo – Fc map for the FAD cofactor and some of its environment illustrates the quality of electron density in well-ordered regions of the protein (Fig. 2). Missing from the models and presumed to be mobile are all 34 residues of the N-terminal His-tag and two side-chains; also, a few residues are modeled with alternate conformations (see Experimental procedures). The Ni2+ ion, identified using an X-ray fluorescence scan and presumably introduced during Ni-affinity chromatography, is situated at a crystallographic three-fold packing interface where it is coordinated by three His59 side chains. Because the gel filtration result described above shows that the recombinant His-tagged MpGlpO is a monomer, this three-fold interface and the two-fold crystal packing interaction involving strand β15 and burying approximately 1600 Å2 of surface area are not physiologically relevant. Thus, the nickel, possibly picked up during purification, can be considered a fortuitous crystallization aid.

    Details are in the caption following the image
    Electron density quality for the flavin and nearby side chains. Stereoview of the final 2Fo – Fc electron density map (cyan; contoured at 2.0ρrms) for the MpGlpO flavin (yellow carbons) and the adjacent protein atoms (off-white carbons) and one water (red sphere). Hydrogen bonds (dashed lines) to the flavin and the water are indicated. Residues are labeled.

    Overall structure

    As expected for a DAAO superfamily member, the MpGlpO chain is organized into an FAD-binding domain with a predominantly parallel, six-stranded β-sheet and a substrate-binding domain with a core antiparallel, eight-stranded β-sheet (Fig. 3). The two domains are discontinuous, with the FAD-binding domain including residues 1–87, 149–219 and 330–364, and the substrate-binding domain formed by residues 86–148 and 227–323.

    Details are in the caption following the image
    Tertiary structure of the MpGlpO monomer. Stereo ribbon diagram of the MpGlpO monomer showing the FAD (sticks with yellow carbons) and labeling the secondary structural elements. The FAD-binding domain is at the bottom and the substrate-binding domain is at the top.

    A search using the DALI server [13] to identify structural homologs of MpGlpO confirms that there are no highly similar known structures. The most similar structure is that of the molecular replacement search model Bp3253 (PDB code: 3DME; Z-score = 44; 26% sequence identity and 2.0 Å rmsd over 352 residues) and slightly less similar is a glycine oxidase from Bacillus subtilis (PDB code: 1NG4; Z-score = 42; 20% sequence identity and 1.9 Å rmsd over 340 residues). An overlay with these proteins shows the high similarity extends throughout the protein chains (Fig. 4A). Remarkably, the enzymes with more similarity in function, SspGlpO (PDB code: 2RGO) and Escherichia coli GlpD (PDB code: 2QCU), are structurally less similar, giving Z-scores of only 28 and 36, respectively (with higher rmsd values of 2.9 and 3.7 Å), and showing only approximately 17% sequence identity over the roughly 350 residues that are in the two domains common to the proteins. Visually, the overlay of MpGlpO with SspGlpO and E. coli GlpD shows their greater divergence (Fig. 4B). A structure-based sequence alignment of these proteins (Fig. 4C) provides further details of the comparisons, and also is useful for tracking the similarities and differences in the active site residues, as is discussed further below.

    Details are in the caption following the image
    Comparisons of the MpGlpO structure and sequence with select homologs. (A) The flavin region in an overlay of MpGlpO (off-white protein and yellow FAD) on Bp3253 (violet; PDB code: 3DME) and glycine oxidase (salmon; PDB code: 1RYI) [17]. For this view, the molecule in Fig. 3 was rotated 180° around a vertical axis (i.e. this view is from the back of that image). Note the similar paths in front of the flavin of the loop that, in MpGlpO contains Ser 348. (B) As in (A) but overlaying MpGlpO (colored as in A) with a form of SspGlpO missing a 50-residue segment (blue; PDB code: 2RGH) [9] and EcGlpD (cyan; PDB code: 2QCU) [14]. Note the different paths of the loop in front of the flavin of the type I enzymes compared to MpGlpO. (C) Structure-based sequence alignment of the five enzymes shown in (A) and (B). Conserved residues (*) and residues involved in β-strands (yellow), α-helices (cyan) and 310-helices (blue) are indicated. Also highlighted are residues noted as being important in flavin binding (green boxes; and including the cis-Pro in MpGlpO and Bp3253) and substrate binding (red boxes). Dots above the MpGlpO sequence mark every tenth residue and, at the end of each line, a residue number for each of the sequences is shown. Residues in lower case letters are disordered in the structure.

    That MpGlpO is more similar to other DAAO superfamily enzymes than it is to SspGlpO led us to consider whether GlpO activity may have independently evolved twice in the DAAO superfamily. To explore this possibility, we generated a relatedness tree of structurally known proteins similar to MpGlpO (Fig. 5). In this tree, despite the DALI scores, the MpGlpO and SspGlpO and E. coli GlpD are shown to be more closely-related to each other than to other functionally characterized DAAO superfamily members. The SspGlpO-like enzymes that cluster together include both GlpOs and the mitochondrial/bacterial GlpD dehydrogenases [14]. They represent a very widely distributed group that we designate as type I GlpO/DH enzymes. By contrast, the MpGlpO-like enzymes, which we designate type II GlpO/DH enzymes, are more narrowly distributed, being found only in bacteria of the class mollicutes (including M. pneumoniae) and the closely related low G+C Gram-positive anaerobic bacteria of the class Erysipelotrichia [15].

    Details are in the caption following the image
    Relatedness tree of structurally known DAAO superfamily members most similar to MpGlpO. A dali [13] search in November 2014 using the PDB90 database option and oxidized MpGlpO as the search model provided a gap-removed alignment for hits with Z-scores higher than 20. These were used to generate a tree with phyml [30]. Branches are labeled with individual PDB code names and known enzyme types are indicated.

    Interestingly, the type II GlpO/DHs from anaerobic bacteria appear to all have an additional approximately 85 residue C-terminal domain that is not closely related to any known structure but includes two segments with conserved pairs of Cys residues (CxCE and CQxGFC) and has some similarity (e-value 4 × 10–10) with the Pfam family of bacterioferritin-associated ferredoxin-like [2Fe-2S] domains (Pfam04324). The one biochemically studied type II GlpO/DH containing such a C-terminal domain is the E. coli glpA gene product that is expressed under anaerobic conditions (replacing GlpD); consistent with the assignment of the additional domain having an iron-sulfur center is a study reporting that E. coli GlpA binds both FAD and nonheme iron [16]. Our proposal based on these observations is that these type II GlpO/DHs with the additional C-terminal domain are all dehydrogenases for which the type II GlpO/DH module converts Glp to DHAP, and the C-terminal domain serves as a conduit to receive electrons from the flavin and pass them on to a further (anaerobic) acceptor.

    Flavin binding and active site

    The general features of FAD binding to DAAO superfamily enzymes have been well-described [7] and so we do not detail those here but, instead, focus on the flavin and the substrate-binding site. In MpGlpO, the bound flavin is slightly twisted with a 310-helix involving residues Thr42-Asn46 closely covering its si face and Ser47-Val49 interacting with the N5, O4 and N3 atoms of the flavin (Fig. 2). Optimizing the hydrogen bond between flavin-N3 and Val49-O appears to be a cause of the flavin twist. Above the flavin (on the si side), there is open space above and in front of atom N5, creating a cavity lined by His51, Arg320 and Ser348 (Fig. 2). Interestingly, we do not see any substantive differences between the active sites of oxidized and reduced MpGlpO structures at this resolution, even though the pale color of the reduced MpGlpO crystals provided visual evidence that they truly were reduced. Given the lack of differences, in the remaining discussion of the active site features, we focus solely on the somewhat higher resolution oxidized structure.

    Unfortunately, attempts to obtain an MpGlpO structure with a substrate or substrate analog bound were unsuccessful (see Experimental procedures), as was also true for our earlier structural work on SspGlpO [9]. However, the protein structure that is the most similar to MpGlpO, that of Bp3253, fortuitously has a ligand, l-tartrate, bound in its active site. Although the function of Bp3253 is not yet known, it binds tartrate similarly to how glycine oxidase [17, 18] and DAAO [19] bind their substrate analogs (overlays not shown), implying that the Bp3253–tartrate complex is an informative one. Specifically, one tartrate α-carbon (i.e. α to one of the carboxylates) is placed just 3.7 Å from flavin N5 and with excellent geometry for hydride transfer (Fig. 6).

    Details are in the caption following the image
    Comparison of the Bp3253 tartrate binding pocket and MpGlpO. Stereoview shown of the Bp3253–tartrate complex (violet carbons) overlaid on the equivalent region of MpGlpO (semi-transparent off-white carbons for protein and yellow for flavin). Also indicated are hydrogen bonds (grey dashed lines for MpGlpO, violet dashed lines for Bp3253–tartrate complex) and the close approach of the tartrate C3 atom to the flavin N5 (green thick dotted line). Select residues in the Bp3253–tartrate complex (violet) and MpGlpO (grey) are identified. H259 in the Bp3253–tartrate complex and I261 in MpGlpO are equivalent residues, although their labels are placed near their respective side chains.

    An overlay of MpGlpO with Bp3253 (Fig. 6) shows that the flavins and most nearby peptide backbone segments align well. In the Bp3253 complex, there are just six side chains (and no backbone atoms) making van der Waals or hydrogen-bond contacts with the tartrate. Starting with the key Arg sitting above the flavin C7 and C8 methyl groups, the Bp3253 residues, with their MpGlpO equivalents in parentheses, are Arg316(Arg320), Pro272(Pro274), His259(Ile261), Tyr248(Phe250), His52(His51) and Ser350(Ser348). Hydrogen bonds connect Arg316 with the C1-carboxylate, His259 via a water molecule with the C3 hydroxyl, and Tyr248, His52 and Ser350 with the C4-carboxylate (Fig. 6). Four of these positions are perfectly conserved in MpGlpO, both in identity and potential placement, recognizing that Arg320 in MpGlpO could easily shift to match the position seen for the equivalent Bp3253 residue. The conserved placement of Ser350(Ser348) projecting over the flavin depends on a conserved cis-peptide bond between the Ser and Pro351(Pro349). Notable differences in the active sites are the Tyr248 to Phe replacement in MpGlpO and the His259 to Ile261 replacement that is compounded by a large difference in the path of the β12-β13 loop, with the backbone of MpGlpO Gly259 occupying the space filled by the His259 side chain in Bp3253.

    Regarding this β12-β13 loop difference, a key question is why the path of MpGlpO residues 256–261 is different. Are they adopting an arbitrary conformation in this crystal form or a conformation that reliably represents the chain path of the Type II GlpO/DHs? There is evidence to suggest it is representative rather than arbitrary: first, the MpGlpO path is similar to the paths seen in broader superfamily members such as glycine oxidase, meaning that it is the Bp3253 path that is unusual; second, the distinct Bp3253 path appears to be directly related to it having a proline (Pro249) at the end of strand β12 that disrupts the normal β-sheet hydrogen bonding (Fig. 6) and is not present in MpGlpO (Fig. 4C); and, third, the loop is well ordered and not involved in crystal contacts.

    Modeling Glp binding to type II GlpO/DHs

    Given the reliability of the β12-β13 loop conformation, we can now consider how MpGlpO binds substrate (Fig. 7A). Compared with Bp3253, the presence of Phe250 and the Gly259 methylene make the pocket above the pyrimidyl portion of the flavin less polar, leading us to speculate that this region would recognize the C1-end of Glp and the region near Arg320 would accommodate the negatively-charged phosphoryl group. Regarding what could provide additional recognition of the phosphoryl, the side chains of Arg230, Lys258 and Lys347 are nearby and appear to be adjustable (B-factors at the side-chain tips being ~ 20 Å2 higher than at the backbone). These side chains are also well conserved among the 70 sequences having > 40% sequence identity with MpGlpO (i.e. BLAST e-value < 10−80): Lys258 is fully conserved; Lys347 is conserved as either a Lys or a Gln; and Arg230 is Arg or Lys in all but four sequences, although in these it is a Leu, implying a lesser importance. Given these insights, we created a model for docked Glp guided by the constraints that the C2-hydrogen be oriented for transfer to the flavin-N5, that either the C1- or C2-hydroxyl replaces the water binding to His51 and Ser348 (Fig. 2), and that the phosphate interacts with Arg320. The binding mode obtained (Fig. 7A) fortuitously places His51 ideally for serving as a catalytic base that could deprotonate the C2-hydroxyl in the forward reaction.

    Details are in the caption following the image
    A predicted MpGlpO-Glp complex and its comparison with EcGlpD. (A) Stereoview of the MpGlpO active site with a roughly positioned Glp-bound (off-white carbons for protein, yellow for flavin, and green for Glp). Hydrogen bonds (grey dashed lines) and the close approach of the modeled Glp C2 atom to the flavin N5 (green thick dotted line) are also shown. Criteria used for placing the Glp are described in the text. (B) Stereoview of the MpGlpO (colored as in A, although semitransparent and without hydrogen bonds shown) overlaid on the equivalent region of EcGlpD (cyan carbons) with its bound inorganic phosphate (cyan phosphorus) and hydrogen bonds (cyan dashed lines). Select residues in EcGlpD (cyan) and MpGlpO (grey) are identified.

    Extrapolation of Glp binding to type I GlpO/DHs

    Interestingly, this proposed mode of substrate binding differs from those proposed previously for the type I GlpO/DHs SspGlpO [9] and EcGlpD [14], neither of which were very satisfactory. For SspGlpO, we had proposed that the substrate would be oriented the other way, with the 3′-phosphoryl moiety near the equivalent of MpGlpO His51 and the C1-hydroxyl interacting with the equivalent of MpGlpO Arg320 [9]. We had also noted that the equivalent of His51 appeared to be the only potential base, although how this could be achieved was not clear. For EcGlpD [14], a phosphate bound in the native crystal structure interacted with Arg317 (equivalent of MpGlpO Arg320), Arg54 and Tyr55, although ligand soaks were taken to indicate binding modes that placed the substrate C2-atom far from flavin-N5.

    Given the unsatisfactory nature of those proposals, we explored whether the mode of binding that we predict for MpGlpO could also work for the type I enzymes. An overlay of EcGlpD onto MpGlpO with its docked Glp shows a remarkable compatibility, with the predicted position of the phosphoryl of Glp matching closely with the experimentally observed phosphate of EcGlpD (Fig. 7B). The ligand fits reasonably into the EcGlpD active site, and it can be seen that the functionality of key MpGlpO residues not conserved in EcGlpD appears to be fulfilled by substitutions of residues often from different parts of the chain: with Lys354 replacing Ser348 in interacting with the C2-hydroxyl; Arg54 and Tyr55 replacing Lys258 and Lys347 in interacting with the phosphoryl; and Phe257 replacing Phe250 in providing a nonpolar environment for the C1-methylene (Fig. 7B). These functional substitutions are most easily seen in schematic drawings of the interactions (Fig. 8). We emphasize that, although the general features associated with this rough placement of Glp are plausible and have explanatory power, the details are not reliably defined because the side chain and backbone positions of protein groups are expected to shift during ligand binding from the positions they adopt in the unliganded MpGlpO structure used to guide the modeling.

    Details are in the caption following the image
    Schematic drawings of residues involved in substrate binding and catalysis in MpGlpO and EcGlpD. (A) The MpGlpO active site atoms approximately in the plane of the flavin are shown smaller and with thinner bonds and hydrogen bonds (dashed); atoms in front of the flavin are shown with thicker bonds and hydrogen bonds (dashed). Residues shown interacting with the flavin and the Glp are highlighted in red and green boxes in Fig. 4C, respectively. Curved arrows indicate the proposed flow of electrons during the reductive half-reaction. (B) Similar diagram as in panel A, showing the EcGlpD active site as a representative of a type I GlpO.

    Catalytic mechanism and outlook

    Given the predicted mode of Glp binding to the MpGlpO (Fig. 8A) and the SSpGlpO (Fig. 8B) active sites, how the electrons flow during catalysis and how the forward reaction is enhanced by base catalysis become readily apparent. As indicated in Fig. 8A (arrows), in MpGlpO, the substrate is well-aligned for His51 to deprotonate the C2-hydroxyl, promoting the formation of a ketone and facilitating hydride transfer from the C2-atom to the flavin N5-atom; the ensuing shifting of flavin electrons would lead to the formation of the reduced flavin N1-anionic form, with the negative charge at the N1/O2 locus stabilized by three hydrogen bonds donated by the Leu351 and Thr352 backbone amides and the Thr352 hydroxyl. The geometries of the interactions are also stereoelectronically reasonable, in that the His51 interaction with the C2-hydroxyl has an approximately ‘anti’ orientation with respect to the hydride leaving group. The Ser47 hydroxyl, supported by its interactions with Tyr234, is well-placed to help stabilize the protonated N5 of the reduced flavin through minor shifts in its position in association with flavin reduction. A set of equivalent interactions exist in the SSpGlpO active site (Fig. 8B).

    A full kinetic analysis of MpGlpO has recently been completed [20] and our structural results allow us to propose explanations for some of the observations made in that study. One such area is the redox potential difference between the MpGlpO flavin, which, at –167 mV, is much lower than those seen for type I GlpOs (e.g. Enterococcus casseliflavus GlpO at –118 mV) [10]. Although many subtle factors can influence redox potential, we note that one very clear difference in the flavin electrostatic environment consistent with this shift is that the type I GlpO/DH enzymes conserve an active site lysine (equivalent to Lys354 in EcGlpD) (Figs 4C, 7B and 8B) having its amino group just at van der Waals distance above the flavin N1/O2 locus, whereas the type II enzymes have a conserved neutral serine side chain in that place (equivalent to Ser348 in MpGlpO). The additional local positive charge in the type I enzymes would make them more easily reduced. As noted in Maenpuen et al. [20], the difference in redox potential may be the reason that MpGlpO can catalyze the reverse reaction (i.e. DHAP oxidation of the reduced flavin), whereas those type I enzymes tested cannot.

    Another very interesting result of the kinetics study was the observation of two enzyme populations, partitioning as 70 : 30, with 70% of the oxidized enzyme reacting rapidly with Glp and 70% of the reduced enzyme reacting slowly with DHAP in the reverse reaction but reacting rapidly with O2. Based on the structure, a plausible single explanation for all of these observations is that the two populations are defined by the protonation state of the catalytic acid/base His51, with 70% being deprotonated and 30% protonated at pH 7 (i.e. at which the studies were conducted). For the reaction with Glp, the deprotonated form of His51 is required for abstracting the C2-hydroxyl proton, and so 70% of the enzyme would react rapidly. The slower population would reasonably be limited by the rate of deprotonation of His51 in the ligand bound form. For the reaction of the reduced flavin with DHAP, the deprotonated 70% of the enzyme reacts poorly (in bimolecular limited fashion at 56 m−1·s−1), whereas the protonated 30% reacts sufficiently rapidly so that the koff of the Glp produced (at 6 s−1) is rate limiting. In terms of the O2 reactivity of the reduced enzyme, it is reasonable that the His51-protonated/charged and the His51-deprotonated/neutral active sites would react differently and our proposal implies that the deprotonated 70% reacts rapidly with oxygen (at ~ 600 s−1), whereas the protonated 30% reacts more slowly (at ~ 100 s−1). This explanation predicts that the enzyme will be more active above pH 7, and a set of assays investigating the pH dependence of the enzyme at a single high substrate concentration shows the optimum pH is near 8. More extensive studies of kcat and KM as a function of pH are now planned to test these ideas and better define the enzyme mechanism.

    The proposed binding mode of Glp also implies the importance of the negatively-charged phosphoryl group for substrate recognition because both the type I and type II enzymes have three or four positively-charged side chains involved in its recognition (Fig. 8). The importance of the phosphoryl is consistent with the binding studies reported by Maenpuen et al. [20] showing that MpGlpO effectively binds the negatively-charged glyceraldehyde-3-phosphate, lactate and malate, although not the neutral glycerol. Although much remains to be clarified, we see the most influential contributions of these analyses are two-fold: the first is the recognition of the type I and type II GlpO/DH enzymes as distinct variants with striking active site differences that were not predicted from sequence alignments and the second is the plausible concrete predicted mode for substrate recognition and catalysis that is relevant not only for bacterial GlpO enzymes, but also for the widespread mitochondrial GlpD dehydrogenases, which will guide mutational studies that test the proposed mechanism and dissect the roles of specific active site residues. Also, the distinction between the type I and type II GlpO/DH active sites raises the encouraging possibility that it will be possible to design inhibitors selectively blocking the activity of bacterial type II enzymes such as MpGlpO at the same time as not inhibiting the mitochondrial GlpD dehydrogenase of the host.

    Experimental procedures

    Expression and purification of MpGlpO and Bp3253

    The codon-optimized MPN051 gene encoding MpGlpO was synthesized by GenScript (Piscataway, NJ, USA) and subcloned into the expression plasmid pET28a (Novagen, Darmstadt, Germany). MpGlpO was expressed with an N-terminal His-tag in E. coli B834(DE3) cells using autoinduction medium at 28 °C. All steps of purification were conducted at 4 °C. Harvested cells were resuspended in 50 mm potassium phosphate (pH 7.0) containing 200 mm NaCl, 20 mm imidazole, 0.5 mm 4-(2-aminoethyl)benzenesulfonyl fluoride protease inhibitor and 10% glycerol. Cells were disrupted using an Avestin EmulsiFlex-C5 homogenizer (Avestin Inc., Ottawa, ON, Canada) and centrifuged (27 000 g for 60 min). The clarified extract was loaded onto a 25-mL Co-Sepharose High Performance column (GE Healthcare, Piscataway, NJ, USA), and MpGlpO protein was eluted with 0.5 m imidazole after washing with 20 mm imidazole. The pooled yellow fractions were dialyzed overnight against 50 mm potassium phosphate (pH 7.0) containing 0.5 mm EDTA. FAD (0.25 mm) was added to the enzyme and incubated for 45 min on ice before loading on a 75-mL SP-Sepharose HP column (GE Healthcare). The column was washed with 50 mm potassium phosphate (pH 7.0), 0.5 mm EDTA and 100 mm NaCl before MpGlpO was eluted with a 100 mm to 1 m NaCl gradient. Fractions were analyzed by SDS/PAGE, pooled, buffer-exchanged into 50 mm potassium phosphate (pH 7.0) and 0.5 mm EDTA and concentrated to 10 mg·mL−1 before freezing in aliquots at –80 °C.

    The pET21_NESG clone for expression of His-tagged Bp3253 (corresponding to PDB code: 3DME) was purchased from the DNASU Plasmid Repository (Tempe, AZ, USA), and its expression and purification were performed in accordance with the protocol described above for MpGlpO, with certain modifications. Recombinant E. coli B834(DE3) cells were grown in TYP medium, prior to induction with 0.5 mm isopropyl thio-β-d-galactoside and overnight protein expression at 16 °C. The cells were broken in a solution containing 50 mm Tris-Cl (pH 8.0) (4 °C). Nucleic acids were removed by adding 2% (w/v) streptomycin sulfate and centrifugation. Crude extract was loaded to the Co column in 50 mm sodium phosphate (pH 8.0), 300 mm NaCl and 20 mm imidazole, the column was washed with more buffer, and pure Bp3253 was eluted with 500 mm imidazole. Bp3253 protein was concentrated to 10 mg·mL−1 in an ultrafiltration cell with a YM30 membrane (Amicon Corp., Danvers, MA, USA) and, for freezing, was buffer-exchanged into 50 mm potassium phosphate (pH 7.0) containing 0.5 mm EDTA.

    Biochemical characterization

    Extinction coefficients for His-tagged MpGlpO and Bp3253 were determined by standard methods, using a model 8453 diode-array spectrophotometer (Agilent Technologies Inc., Santa Clara, CA, USA). The specific activities of the two proteins were measured, as described previously for SspGlpO [11], using the standard spectrophotometric assay with a Cary 50 spectrophotometer (Varian Inc., Palo Alto, CA, USA) thermostatted at 25 °C. The native Mr for recombinant His-tagged MpGlpO was determined by gel filtration at 25 °C, in a phosphate buffer (pH 7) with 150 mm NaCl, using a Superdex 200 10/300 GL column (GE Healthcare) calibrated with six standard proteins covering the range 15 600–440 000. Titrations of both MpGlpO and Bp3253 proteins with sulfite followed established protocols [10, 11].

    Crystallization and data collection

    Thawed aliquots of wild-type His-tagged MpGlpO were subjected to a variety of crystal screens at 4 °C and the most promising crystal leads grew using a reservoir of 2.68 m NaCl, 3.35% v/v isopropanol and 0.1 m Hepes (pH 7.5) (Wizard Precipitant Synergy Screen Block 1 condition B7; Rigaku Reagents, Inc., Bainbridge Island, WA, USA). Optimization at 4 °C using hanging-drops led to yellow, trigonal-pyramidal crystals measuring approximately 0.3 × 0.3 × 0.3 mm3 growing within 1 week using drops made from 1 μL protein at 5 mg·mL−1 plus 2 μL of a reservoir solution as above but containing 2% (v/v) isopropanol. For data collection, crystals in an artificial mother liquor (AML) of 3 m NaCl in 0.1 m Bis-Tris (pH 7.0) were placed for 3 min in AML with 15% glycerol as a cryoprotectant before being flash-frozen by plunging into liquid nitrogen. Data sets, first at a resolution of 2.50 Å and then at a resolution of 2.40 Å, were collected at beamline 5.0.3 at the Advanced Light Source synchrotron (Berkeley, CA, USA). For analyses of additional forms of MpGlpO, crystals were soaked in either 10 mm dithionite, 10 mm l-Glp, 10 mm l-tartrate, 10 mm 2-phosphoglycerate or 10 mm phosphoenolpyruvate in AML for 1 h prior to freezing and data collection. Two experiments with dithionite-soaked crystals gave a complete merged data set to 2.50 Å. Data sets at a resolution of between 2.5 and 3.0 Å were collected for the ligand soaks.

    MpGlpO phasing and structure refinement

    Diffraction data for oxidized MpGlpO were processed with imosflm [21] and with scala in the ccp4 suite [22, 23]. MpGlpO crystallizes in space group P23 and there is one molecule per asymmetric unit. Attempts to solve the structure by molecular replacement using amore [24] with the SspGlpO coordinates were not successful. However, using chain A of PDB code: 3DME, annotated as the ‘conserved exported protein (BP3253) from Bordetella pertussis’, as a search model gave a solution using the 2.50-Å data set and automr [25]. Manual rebuilding of the initial model was carried out in coot [26], and this model was refined with phenix [27]. Six rounds of simulated annealing and minimization refinement gave a partially refined model with R/Rfree of 32%/45%, respectively. At this point, the improved 2.40-Å resolution MpGlpO data were used, and six rounds of refinement with buster [28] led to an Rfree of 28%. Water molecules were added, as indicated by both electron density peak height and hydrogen-bonding interactions, and refinement continued with refmac [29]. A nickel ion (see Results) was introduced late in the process and was confirmed by X-ray fluorescence, and further manual modeling and refinement led to the final oxidized MpGlpO structure with R/Rfree 14.8%/20.4%. The model does not include the Lys79 and Lys298 side chains beyond -CB because these surface residues have little or no side chain density. Alternate side chain and/or backbone conformations are included for Gln40, His244 and the Trp375-Asn376-Gly377 backbone.

    The 2.50-Å refinement of the dithionite-reduced GlpO structure began from the oxidized structure. Rounds of manual modeling and refmac refinement gave a model with only minor changes, and with R/Rfree values of 14.6%/21.0%. Of the soaks with Glp, 2-phosphoglycerate, phosphoenolpyruvate and tartrate, no substantive interpretable density differences were observed in the vicinity of the isoalloxazine and so these structural analyses were not pursued further.


    We thank Dale Tronrud for helpful guidance with the crystallographic work. This work was supported by the North Carolina Biotechnology Center (Grant 2011-MRG-1116 to A.C.), the Thailand Research Fund (MRG5580066 to S.M. and RTA5680001 to P.C.) and the Howard Hughes Medical Institute (grant 52005883 to C.K.E.). Synchrotron data were collected at the Advanced Light Source, supported by contract DE-AC02-98CH10886 from the Office of Basic Energy Sciences of the US Department of Energy.

      Author contributions

      CKE, DP, SM, PC, AC and PAK planned the experiments. CKE, DP, SM and PC performed the experiments. CKE, KMK, DP, SM, PC, AC and PAK analyzed the data. DP contributed reagents. KMK, AC and PAK wrote the paper.