Comparative analysis of glutaredoxin domains from bacterial opportunistic pathogens

Glutaredoxin proteins (GLXRs) are essential components of the glutathione system that reductively detoxify substances such as arsenic and peroxides and are important in the synthesis of DNA via ribonucleotide reductases. NMR solution structures of glutaredoxin domains from two Gram-negative opportunistic pathogens, Brucella melitensis and Bartonella henselae, are presented. These domains lack the N-terminal helix that is frequently present in eukaryotic GLXRs. The conserved active-site cysteines adopt canonical proline/tyrosine-stabilized geometries. A difference in the angle of α-helix 2 relative to the β-­sheet surface and the presence of an extended loop in the human sequence suggests potential regulatory regions and/or protein–protein interaction motifs. This observation is consistent with mutations in this region that suppress defects in GLXR–ribonucleotide reductase interactions. These differences between the human and bacterial forms are adjacent to the dithiol active site and may permit species-selective drug design.

1. Introduction

Glutaredoxins (GLXRs) are redox enzymes that are important for the reduction of ribonucleotide reductase enzymes that synthesize deoxynucleotides from ribonucleotides (Uhlin & Eklund, 1994). Thus, they are required for efficient and sustainable synthesis of DNA. Additionally, GLXRs are important for detoxifying oxidizing agents such as reactive oxygen species (ROS), transition metals and metalloids, e.g. arsenic compounds (Fig. 1). Like other ROS defenses, i.e. glutathione peroxidases, this enzyme is connected to the glutathione pool: GLXRs catalyse the reaction of glutathione with peroxides and metals as shown in (1). Homeostatic levels of reduced glutathione are restored by the action of glutathione reductase (GSR) in (2) via reducing equivalents from the pentose phosphate shunt. Thus, the GLXR, glutathione peroxidase and glutaredoxin reductase enzymes are attractive targets for drug-mediated ROS amplification.

GLXRs have well conserved sequences within bacteria, but their sequences diverge between bacteria and humans. This distinctive difference in sequence should permit selective inhibition of bacterial GLXRs without perturbation of the host enzyme. This might kill bacteria by inhibition of DNA synthesis and/or through increases in ROS toxicity.

Figure 1 Diagram of the role that glutaredoxin and related enzymes play in DNA synthesis and ROS, metal and metalloid detoxification. Abbreviated pathways utilized by glutaredoxin and thioredoxin in DNA synthesis are highlighted in green; abbreviated ROS defense pathways are highlighted in blue.

Structures have been published for several forms of human (Sun et al., 1998; Yang et al., 1998), plant (Rouhier et al., 2007; Li et al., 2010), budding yeast (Gibson et al., 2008; Discola et al., 2009) and Escherichia coli GLXRs (Iwema et al., 2009; Fladvad et al., 2005; Xia et al., 1992, 2001; Bushweller et al., 1994; Sodano et al., 1991). However, it was unclear whether other bacterial GLXRs would adopt similar conformations. The aim of this study was to expand the existing knowledge base of GLXR structures and to find structural trends that might be exploited to design selective inhibitors of bacterial GLXR that leave host enzymes unperturbed. In particular, the GLXRs from the pathogens Brucella melitensis and Bartonella henselae were investigated as these organisms have significant relevance to medical and military biodefense. Here, we present the structures of GLXRs from Br. melitensis and Ba. henselae and compare these structures with the available structures from E. coli and human.

2. Methods

2.1. Protein expression and purification

GLXRs from Br. melitensis and Ba. henselae (NCBI YP_415222 and YP_033241.1; UniProt Q2YLN2 and Q6G5J5; Pfam ID PF00462; EC 1.20.4.1) were cloned into a pAVA vector (Choi et al., 2011) and expressed from RIL cells grown in 2 l of M9 medium supplemented with 4 g l^{−1 13}C glucose and 1 g l^{−1 15}N ammonium chloride. Protein expression was induced at an OD₆₀₀ of 0.6 with 0.5 mM IPTG and temperature reduction to 293 K for 12 h. Cell pellets were suspended in 50 ml buffer A (20 mM HEPES pH 7, 0.3 M NaCl, 5% glycerol, 2 mM DTT) supplemented with 2 µg lysozyme, freeze-fractured twice at 193 K and then lysed using a French press. The crude lysate was cleared by centrifugation at 15 000g and the soluble protein supernatant was filtered through a 0.22 µm GD/X membrane syringe filter (Whatman). Nickel IMAC on 5 ml HisTrap FF columns (GE Healthcare) was used to capture the proteins from this supernatant. Non­specific binding proteins were washed off the column with 10% buffer B (20 mM HEPES pH 7, 0.3 M NaCl, 5% glycerol, 2 mM DTT, 300 mM imidazole). The proteins were eluted with a 50 ml gradient from 10% to 100% buffer B and fractions containing purified protein were pooled, cleaved with 3C protease and rerun over the HisTrap column. The N-terminal tag introduced during cloning consisted of an MAHHHHHHMGTLEAQTQGPGS sequence appended to the native methionine; only the GPGS portion remained after protease cleavage. The HisTrap flowthroughs were collected, dialyzed against NMR buffer (20 mM phosphate, 120 mM NaCl pH 6) and purified by size-exclusion chromatography on a Superdex 75 column (GE Healthcare) equilibrated with NMR buffer. Fractions were pooled and concentrated via stirred cell (Amicon) to 0.5 mM for Ba. henselae GLXR and to 1.5 mM for Br. melitensis GLXR and placed in NMR microcells (Shigemi).

2.2. NMR data collection

For both proteins, the standard suite of NMR experiments were acquired (Sattler et al., 1999): ¹⁵N HSQC, ¹³C HSQC, 3D HNCO, 3D HNCA, 3D HN(CO)CA, 3D CBCA(CO)NH, 3D HNCACB, 3D HCCH-TOCSY, 3D ¹⁵N-TOCSY-HSQC (70 ms mixing), 3D ¹⁵N and ¹³C NOESY-HSQC (80 and 120 ms mixing times) and 2D ¹H/¹H D₂O-NOESY (100 ms mixing time). Two instruments were used for data collection: Bruker Avance 500 and 600 spectrometers equipped with cryoprobes. All data sets were collected in conventional, i.e. nonreduced dimensionality, formats with States–TPPI quadrature (States et al., 1982) in the indirect ¹³C and ¹H dimensions and Rance–Kay sensitivity enhancement (Kay et al., 1992; Cavanagh et al., 1991) for ¹⁵N dimensions. Proton carriers were set on water and the ¹⁵N carrier at 117 p.p.m. For α-carbon relevant spectra the ¹³C carrier was set to 52 p.p.m., while for CACB spectra it was set to 45 p.p.m. and for carbonyl spectra it was set to 176 p.p.m.. Spectra were referenced directly to DSS in proton dimensions and indirectly in ¹³C and ¹⁵N dimensions. NMR data sets were converted and processed with NMRpipe (Delaglio et al., 1995).

2.3. Assignments and structure calculations

Backbone assignments for both proteins were determined from pairs of triple-resonance spectra in the usual manner (Sattler et al., 1999; Lunde et al., 2010; Leeper et al., 2010). Backbone resonance correlations were compared and tabulated using CCPNMR (Vranken et al., 2005) using the manual assignment mode. Side chains were assigned from HCCH-TOCSY, ¹⁵N-TOCSY-HSQC and, in the case of aromatic residues, a 2D ¹H/¹H D₂O-NOESY. Distance constraints for structure calculations were obtained from 2D ¹H/¹H D₂O-NOESY and 3D ¹⁵N and ¹³C NOESY-HSQC spectra as unassigned peak lists. Peak intensities were exported directly from these spectra for use in CYANA structure calculations (Güntert, 2004), as were chemical shifts for TALOS-generated dihedral angle restraints (Shen et al., 2009). Hydrogen-bond constraints were determined for slowly D₂O-exchangeable backbone amides with acceptor-atom identities gleaned from preliminary structure calculations. Initially, the disulfide bond in the active site was left as a pair of thiols, but was ultimately restrained to be a disulfide based upon initial structure geometry and proximity. We decided to use the structure calculations to guide this decision since these residues are helical and the C^β shifts reside in the ambiguous border region between 30 and 33 p.p.m.: normal C^β shifts for reduced helical thiols range from 23.8 to 28.8 p.p.m., but oxidized helical disulfide C^β atoms range from 32.8 to 47.4 p.p.m. (Sharma & Rajarathnam, 2000). Note that no particular effort was made to maintain this pair of cysteines in the reduced state, so they are likely to have been oxidized spontaneously. We have not yet explored thorough pK_a calculations to determine whether these cysteines exist as a mixed thiol/thiolate state (Sun et al., 1998; Yang et al., 1998), but we may do so in future studies.

Seven rounds of automated NOE assignment and structure calculation using CYANA's CANDID tool (Herrmann et al., 2002) were used to calculate the structures, followed by one round of manual calculation of 100 structures. The final ensembles were selected as the 20 structures with the lowest CYANA target functions. These structures showed convergence via low r.m.s.d.s (Table 1) and excellent covalent geometry and clash scores (Table 2) as determined by MolProbity (Chen et al., 2010). Structure ensembles were analysed and rendered with PyMOL (DeLano & Lam, 2005).

Table 1
NMR restraints

	Ba. henselae (2klx )	Br. melitensis (2khp )
Distance restraints	1107	1747
Short range, |i − j| ≤ 1	621	886
Medium range, 1 < |i − j| < 5	207	374
Long range, |i − j| ≥ 5	279	487
Dihedral	202	194
Hydrogen bonds	40	22
CYANA target function (Å^2)	1.45	1.11
Dihedral r.m.s.d. (°)	0.76	0.94
Distance r.m.s.d. (Å)	0.012	0.006
Maximum NOE violation (Å)	0.24	0.34

Table 2
Ensemble statistics

Ensemble of 20 structures from 100 calculated structures.

	Ba. henselae (2klx )	Br. melitensis (2khp )
Backbone r.m.s.d. (mean) (Å)	0.52†	0.35‡
Heavy-atom r.m.s.d. (mean) (Å)	1.14†	0.82‡
Most favored (%)	83.8	90.8
Additionally allowed (%)	16.1	9.2
Generously allowed (%)	0.1	0.0
Disallowed (%)	0.0	0.0
MolProbity score (percentile)	3.57 (97th)	2.2 (99th)

†R.m.s.d. calculated over residues 3–84, excluding 1–2 and 85–89.
‡R.m.s.d. calculated over residues 6–92, excluding 1–5.

3. Results and discussion

3.1. Sequence conservation between domains

A BLAST search of Br. melitensis and Ba. henselae GLXR-domain sequences against the nonredundant protein database (Altschul et al., 1990) revealed that the E. coli GLXR3 domain was the closest known homolog (59% identity, E value = 3 × 10⁻²⁰ versus 2khp ). Upon inspection of closest homologs from available human sequences, the GLXR1 sequence was revealed to be most similar to the bacterial GLXR3 (38% identity, E value = 1 × 10⁻¹⁰ versus 2khp ). We assume that this represents a discrepancy in the annotation rather than a functional difference, as human GLXR3 is significantly less related (25% identity, E value = 2 × 10⁻³ versus 2khp ). A ClustalW alignment of the sequences using the BLOSUM matrix (Henikoff et al., 1999; Larkin et al., 2007) is shown in Fig. 2. From this comparison it is clear that for these sequences the region surrounding the redox active site is highly conserved (yellow box). There are very few overall differences between the new bacterial GLXR3s and the E. coli GLXR. However, when compared with the human GLXR1 sequence deviations are present in an N-terminal extension (α0), an inserted region in loop 1 between helix 1 and β-strand 2, and variations in the sequence of the loop between strand 2 and helix 2 and the N-terminal end of helix 2 are observed. As shown below, this last region is juxtaposed with the active site. As a result, we will refer to these latter two points of variation as the human-specific loop (HSL) and the sequence-specific helix (SSH), respectively.

Figure 2 Multiple sequence alignment of the Br. melitensis, Ba. henselae and E. coli glutaredoxin 3 domains and the Homo sapiens glutaredoxin 1 domain. The black-boxed region indicates the conserved active-site residues shared with all dithiol GLXRs, while the red-boxed region highlights the human-specific loop 2 region (HSL2) and adjacent sequence-specific helix (SSH) region. The blue-boxed region is the additional N-terminal helix found in the human protein.

3.2. Structures of glutaredoxin from Br. melitensis and Ba. henselae

NMR spectroscopy of the Br. melitensis and Ba. henselae GLXR domains revealed reasonably well resolved spectra that were amenable for structural study by NMR (Fig. 3). The Br. melitensis GLXR had a significantly larger number of unambiguously assignable NOEs than the Ba. henselae GLXR (Table 1). This is partially attributed to significantly stronger sample concentrations for the former (1.5 versus 0.5 mM), which are a result of a slight aggregation of the latter at higher concentrations as well as lower expression yields. Thus, the significantly larger numbers of medium and long-range constraints, which are also typically low signal-to-noise NOEs, for the Br. melitensis protein are a consequence of its higher concentration and improved spectral quality. Furthermore, Ba. henselae GLXR has ∼11 overlapped residues in the ¹⁵N HSQC, whereas Br. melitensis GLXR only has between two and six overlapped amides depending upon the field at which the spectra are collected (Fig. 3), thus further reducing the number of unambiguously assignable resonances.

Figure 3 ^15N HSQC spectra of the Ba. henselae (a) and Br. melitensis (b) GLXR domains with complete backbone and side-chain amide assignments labeled.

Structure calculations for both the Br. melitensis and Ba. henselae GLXR domains converged well (Table 1, Figs. 4a and 4b). Topo­logically, these domains adhere to the expected thioredoxin fold: βαβαββαα with a 2134 mixed parallel and antiparallel β-sheet with helices on both sides of the sheet. The active-site CPYC residues are in the expected location at the N-terminal end of helix α1. The N- and C-terminal tails of these full-length domains are somewhat short relative to many other proteins studied by NMR, resulting in a well defined backbone conformation over the entire domain (0.52 and 0.35 Å r.m.s.d. over all backbone atoms including the N- and C-­termini). The Ramachandran statistics and MolProbity scores are good (Table 2) and suggest a well refined structure in spite of the heavy reliance upon the CANDID automated NOE assignment.

Figure 4 (a, b) Stereopairs of ensemble superposition for the NMR solution structures of the Br. melitensis (a) and Ba. henselae (b) glutaredoxins. The conserved pair of active-site cysteine side chains are drawn in gray and yellow. (c) The lowest energy conformer from (a) rotated to show the angle of SSH relative to the β-sheet surface. (d) The lowest energy conformer from (b) with a more parallel SSH angle. (e) The human glutaredoxin 1 structure with the extra N-terminal helix (blue) and the HSL/SSH region (red trapezoid) indicated. (f) The E. coli glutaredoxin 3 structure. (g) Close-up view of the convergent superposition of the CPYC active-site region (left) juxtaposed with the divergent and non-overlapping structures for the HSL/SSH regions. Only the region near the active site and the HSL/SSH regions are colored according to the above figures, while the remainder of the proteins are drawn in white.

3.3. Comparison with other glutaredoxins: E. coli and human

The lowest energy structures for the Br. melitensis and Ba. henselae GLXR domains were compared with structures obtained for human GLXR1 (Fig. 4e) and E. coli GLXR3 (Fig. 4f). The most obvious difference is the presence of an extra N-terminal helix associated with the human domain (blue oval, Fig. 4e). On further inspection, slight deviations in the angle of the SSH region also become apparent. In the Ba. henselae GLXR the SSH helical angle relative to the vector perpendicular to the β-sheet is about 45° (Fig. 4c). This angle is similar to that of the human GLXR, which is sterically packed up against the C-terminal helical extension. In contrast, the Br. melitensis species-specific helix is more reminiscent of the E. coli structure, with an angle of about 20°. Thus, the SSH seems to vary among species at the levels of both primary sequence and three-dimensional structure.

4. Discussion and conclusion

We have determined the NMR structures of the GLXR domains from the pathogenic organisms Br. melitensis and Ba. henselae. These structures are typical examples of the thioredoxin fold present in many dithiol reductase enzymes. Furthermore, subtle differences in the ribonucleotide reductase binding platform on the SSH and the extension of the HSL suggest possible routes for rational species-selective drug design. For example, mutation of the SSH in E. coli GLXR3 allows it to thrive even in the inviable background con­taining gene knockouts for thioredoxin 1, thioredoxin 2 and GLXR1 (Ortenberg et al., 2004). This mutation of Met43 to valine, isoleucine or leucine in the SSH seems to exert the restoration of its viability via enhanced interactions with ribonucleotide reductase, consistent with studies on GLXR bound to model peptides that point to a direct interaction with the SSH (Berardi & Bushweller, 1999). E. coli GLXR residue Met43 is on the opposite side of the helix from the surface expected to directly interact with ribonucleotide reductase, which suggests that replacement by more hydrophobic residues may adjust the position of this helix relative to the adjacent β-­sheet. This result emphasizes that the manner in which the SSH lays across this GLXR β-sheet surface may be pertinent to interactions with ribonucleotide reductase, a detail that is also relevant to GLXR isoform and species substrate-specificity (Figs. 4c and 4d). Additionally, the expression levels of ribonucleotide reductase, thio­redoxin and GLXR are tightly regulated so as to maintain relative stoichiometries (Miranda-Vizuete et al., 1996). Thus, structural biology, biochemistry and epigenetics all point to the position of the sequence-specific helix (SSH) being important for recruitment of ribonucleotide reductase. Whether this is through direct interactions between ribonucleotide reductase and the SSH or whether the SSH acts as a displaceable cover for interactions mediated by the nearby β-sheet will require additional experiments to determine fully.

Either GLXR or thioredoxin is required for cellular viability (Russel & Holmgren, 1988). Unlike thioredoxin, GLXR requires no accessory enzymatic component to regenerate itself directly. Instead, it relies directly upon the state of the glutathione pool (typically at ∼99% GSH versus ∼1% GSSG; Higashi et al., 1985) and hence the availability of reducing equivalents in the form of NADPH. Therefore, as a simpler molecular system, it may be more difficult to develop resistance pathways beyond the inherent alternative pathway provided by thioredoxin. Indeed, small-molecule inhibitors of glutathione synthesis such as buthione sulfoximine (BSO) can reverse resistance to cellular toxins and stress (Griffith & Meister, 1979). For example, both tumor cells and Gram-negative facultative anaerobic bacteria are highly dependent on the glutathione pool for viability (Smirnova et al., 2005). It has been demonstrated that tumor cells that are resistant to radiation and chemotherapeutics can be sensitized via co-treatment with GSH synthesis inhibitors such as BSO. In a similar fashion, depletion of the glutathione pool using BSO-like compounds should amplify the effects of drugs targeting the GLXR in specific bacteria, although BSO itself has been shown to be only weakly effective against some strains of E. coli (Romero & Canada, 1991). Thioredoxin, on the other hand, senses the NADPH pool directly. Synthetic inhibition of thioredoxin and NADPH production might also be possible, since mutations in glucose-6-phosphate dehydro­genase, i.e. favism, are tolerated in the absence of ROS stress (Scriver, 2001). Thus, toxic side effects might be minimized for the host organism via direct inhibition of both thioredoxin and glucose-6-­phosphate dehydrogenase should that route be taken.

GLXR and thioredoxin are nonspecifically inhibited by cisplatin (Arnér et al., 2001) and cadmium (Chrestensen et al., 2000). Additionally, glutathione analogs are also potent but nonsequence-specific inhibitors of GLXR (Höög et al., 1982). Because these compounds are just as likely to disrupt host GLXRs as bacterial enzymes, they are not viable as drug candidates. Thus, the real challenge in finding dithiol active-site inhibitors lies in identifying compounds that disrupt or covalently react with the dithiol center but only after interrogating species-specific features. The relatively close proximity of the HSL region (Fig. 4g, trapezoid) to the conserved active site affords a promising option. Surface renderings of the proteins support this assertion and highlight a V-shaped indentation bordered on one side by the conserved dithiol and on the other by the HSL (Figs. 5a, 5b and 5c). This groove is much smaller within the surface of the human GLXR (Fig. 5d). Thus, it may be possible to rationally engineer bidentate drugs that anchor themselves into the region via the HSL by one epitope while attacking the adjacent dithiol with their other halves. In the case of GLXR, such drugs would be particularly useful if combined with the aforementioned BSO compound for perturbing the basal GSH and/or NADPH levels to enhance ROS-mediated cell death.

Figure 5 Surface renderings of the Br. melitensis (a), Ba. henselae (b), E. coli (c) and human (d) GLXR domains. The conserved disulfide active site is shown in yellow and the HSL loop regions are shown in cyan, magenta, green and gold as in Fig. 4. The V-shaped pocket amenable to drug design in the bacterial GLXRs is indicated by the red chevron. The protein poses are rotated 90° relative to Fig. 4(g).

Footnotes

‡Present address: Department of Chemistry, University of Akron, Akron, OH 44325, USA.

§Present address: Shanghai Hengrui Pharmaceutical Co. Ltd, Shanghai 201716, People's Republic of China.