Intersubunit Ionic Interactions Stabilize the Nucleoside Diphosphate Kinase of Mycobacterium tuberculosis

Most nucleoside diphosphate kinases (NDPKs) are hexamers. The C-terminal tail interacting with the neighboring subunits is crucial for hexamer stability. In the NDPK from Mycobacterium tuberculosis (Mt) this tail is missing. The quaternary structure of Mt-NDPK is essential for full enzymatic activity and for protein stability to thermal and chemical denaturation. We identified the intersubunit salt bridge Arg80-Asp93 as essential for hexamer stability, compensating for the decreased intersubunit contact area. Breaking the salt bridge by the mutation D93N dramatically decreased protein thermal stability. The mutation also decreased stability to denaturation by urea and guanidinium. The D93N mutant was still hexameric and retained full activity. When exposed to low concentrations of urea it dissociated into folded monomers followed by unfolding while dissociation and unfolding of the wild type simultaneously occur at higher urea concentrations. The dissociation step was not observed in guanidine hydrochloride, suggesting that low concentration of salt may stabilize the hexamer. Indeed, guanidinium and many other salts stabilized the hexamer with a half maximum effect of about 0.1 M, increasing protein thermostability. The crystal structure of the D93N mutant has been solved.


Introduction
Nucleoside diphosphate kinases (NDPKs) catalyze the reversible transfer of the phosphoryl c of nucleoside triphosphates to nucleoside diphosphates [1,2]. The two-step reaction proceeds via a covalent intermediate, the enzyme being transiently phosphorylated on a conserved histidine residue, His 117 in Mycobacterium tuberculosis NDPK (Mt-NDPK) [3]. In addition to their catalytic function, eukaryotic NDPKs are involved in complex regulatory functions, some of which unrelated to kinase activity. Drosophila melanogaster NDPK (Dm-NDPK, product of the awd gene) is essential for larvae development [4]. The isoform A of the human NDPK (NDPK-A or Nm23-H1) is an anti-metastatic protein [5,6]. The isoform B of the human NDPK, also called Nm23-H2, is a transcription factor of the proto-oncogene c-myc [7] and possesses nuclease activity [8].
The gene coding for NDPK has been identified in Mycobacterium tuberculosis (Mt) by genome sequencing. Mt-NDPK has been shown to be active and to have secondary functions besides the kinase activity. It cleaves single strand DNA within the human c-myc promoter [9], acts as a GTPase-activating protein for Rho-GTPases [10] and damages the nuclear DNA when present in the nuclei of HeLa and COS-1 cells [11]. Importantly, it is cytotoxic for mammalian cells when secreted [12]. The toxicity mechanism is unknown, but may be related to tuberculosis pathology. In transfected human cells Mt-NDPK localizes to the nucleus [11], whereas human NDPKs localize both to the cytoplasm and the nucleus. The interesting biology of the Mt-NDPK prompted us to study its solution properties and stability to denaturation.
Crystal structure of the Mt-NDPK has been solved [13]. It is a hexamer with a tertiary and quaternary structure very similar to other hexameric NDPKs [14]. It has been shown that several prokaryotic NDPKs are tetramers [15,16]. Whatever the quaternary architecture, NDPKs share a common-dimer unit. Recently, such dimer unit has been found in solution for the NDPK from moderately halophilic bacteria [17,18]. As the subunit assembly is very different in hexameric and tetrameric NDPKs, the role of the quaternary structure for protein varies between the two types of NDPKs [19]. Our study focuses on the hexameric type of NDPK enzyme. The Mt-NDPK protein sequence of 135 amino acids long is very similar to that of other hexameric NDPKs (.50% identity without gaps or insertions), except for a missing 15 amino acids Cterminal segment (Figure 1). In ''long'' NDPKs, this segment is extended, without secondary structure, and interacts with the neighboring subunits over about 300 Å 2 . As this interaction is repeated six-fold in the long hexamer it has a very large contribution to the overall hexamer stability. Indeed, the deletion of a few residues in the C-terminus of NDPK (cytosolic isoform) from Dictyostelium discoideum (Dd-NDPK) dramatically decreased hexamer stability [20]. The puzzling issue is that the missing interactions do not affect the Mt-NDPK, which is hexameric and quite thermostable, having a temperature of half denaturation (T m ) of 73uC [13]. The molecular bases of the high stability of Mt-NDPK have never been understood.
One important point was to identify interaction(s) compensating the missing free energy due to the absence of the C-terminal segment. Several publications demonstrated that quaternary structure is crucial for NDPK stability and activity. We therefore focused on the analysis of intersubunit interaction. A detailed analysis failed to identify any significant differences of interfaces, between Mt-NDPK and other hexameric NDPKs. Dd-NDPK structure and properties in solution have been extensively studied. Mt-NDPK and Dd-NDPK overlap with a root mean square deviation (rmsd) rmsd of 1.0 Å distributed over the common sequence. Dd-NDPK having a larger subunit interface is nevertheless less thermostable than Mt-NDPK.
Solvent exposed salt bridges are common determinants for the thermostability of proteins [21]. The importance of specific steric and electrostatic interactions in the dimer-dimer assembly of the NDPK from moderately halophilic bacteria has been established [18,22]. One interaction present in Mt-NDPK but missing in other NDPKs, is the intersubunit salt bridge Arg 80 -Asp 93 located on the protein surface. Here, to elucidate the crucial role of that salt bridge for hexamer and overall protein stability, a mutant having Figure 1. Sequence alignment of NDPKs whose structure has been solved. Sequence alignment was performed using ClustalW and mapped onto the secondary structure elements of Mt-NDPK, which derived from the crystal structure (PDB id 1k44) [13], by ESPript (http://espript.ibcp.fr/ ESPript/ESPript/). The Kpn loop was named after the killer of prune (Kpn) mutation of Drosophila. Among the fully conserved residues indicated on red background, the activesite residues are denoted with a blue star. Triangles indicate Arg80 and Asp93 which form the salt bridge discussed in this paper. The quaternary structure and the pdb code are indicated at the end of the sequences. The enzymes of the first group from M. tuberculosis to B. halodenitrificans are hexameric, while the second tetrameric or dimeric (H. sp. 593). doi:10.1371/journal.pone.0057867.g001 the salt bridge abolished (D93N) has been prepared. The aspartate was changed to asparagine since many NDPKs have asparagine in that position. The activity, the stability and the crystal structure of the wild-type Mt-NDPK and the D93N mutant were compared.

Mutagenesis and Protein Purification
D93N gene mutation was introduced using the Transformer TM site-directed mutagenesis kit (Clonetech). The recombinant proteins Mt-NDPK wt and D93N mutant were expressed using a pET24 vector (Novagen) in the BL21-derived host strain BL21-CodonPlusH(DE3)-RIL (Stratagene). The mutation was confirmed by nucleotide sequencing and the molecular weight of proteins checked by mass spectrometry. The culture medium contained 16 g/L bacto tryptone, 10 g/L bacto yeast extract, 5 g/L sodium chloride, in the presence of 80 mg/mL of kanamicyn; expression was induced with 1 mM IPTG for 6 hours at 37uC, once the optical density reached 0.5-0.7 units. The purification steps were carried out at 4uC. After harvesting, the E. coli cells were sonicated and centrifuged in order to recuperate the soluble fraction containing Mt-NDPKs. The DNase-treated bacterial extract was applied to a Q-Sepharose column equilibrated in 100 mM Tris-HCl, pH 7.4. The enzyme was eluted at 0.5-0.6 M sodium chloride, in a linear gradient of 0-0.8 M sodium chloride in the same buffer. Active fractions were precipitated with 80% saturated ammonium sulfate and further purified by salting-out chromatography on a unmodified sepharose 6B column equilibrated with 80% ammonium sulfate, 100 mM Tris-HCl, pH 7.4. The protein was eluted by a linear gradient from 80% to 20% ammonium sulfate in the same buffer. The active fractions were pooled, dialyzed against 100 mM Tris-HCl, pH 7.4, and further purified on a Source 15Q column, under the conditions described for the Q Sepharose chromatography. The enzymes were precipitated by dialysis against a saturated solution of ammonium sulfate, recovered by centrifugation and further purified by size-exclusion chromatography on a Sephacryl S-200 column equilibrated with 0.2 M sodium phosphate buffer, pH 7.0 (Buffer A). This step allowed to cleanup the sample, by eliminating aggregated and dissociated protein.
The enzymes were essentially pure as ascertained by polyacrylamide gel electrophoresis in the presence of SDS. The concentrations of WT and mutant Mt-NDPKs were determined from the optical density at 280 nm using an extinction coefficient of 0.48 for 1 mg/mL, which was calculated from the amino acid composition.

Calorimetry
Heat capacity versus temperature profiles were recorded with a VP-DSC differential scanning microcalorimeter (MicroCal Inc., Northampton, MA) at a scan rate of 1uC/min. Protein samples were diluted to 0.2-0.4 mg/mL concentration, dialyzed against buffer A and degassed before the calorimetric experiment. The reference cell was filled with degassed buffer A. Both cells were kept under an excess pressure of 30 psi to avoid bubbling during the scan. At the end of each run, the solutions were cooled and subjected to a second heating cycle under the same conditions to determine the reversibility of the transitions. Thermograms were corrected by subtracting the instrumental baseline, obtained with both cells filled with buffer A, and normalized for protein concentration. The T m (temperature at which excess heat capacity reaches a maximum) and the denaturation enthalpy (DH) were determined with the ORIGIN software provided by MicroCal, after subtraction of a progress baseline connecting the pre-and post-transition traces. Errors are estimated to be 60.05uC for the T m .

Stability and Enzymatic Activity Measurements
For the unfolding/refolding curves, native or unfolded Mt-NDPK was diluted at the final protein concentration of 10 mg/mL in 0-8 M urea or 0-5 M guanidinium hydrochloride (GuHCl), and 20 mM phosphate buffer, pH 7.0 at 25uC and incubated for 16 hours. Fluorescence intensities of the single tryptophan residue Trp132 were measured at 335 nm (bandwidth of 5 nm) with an excitation at 295 nm (bandwidth of 5 nm). Data were normalized after linear fitting correction of the pre-and post-transition. Enzymatic activities were measured with the spectrophotometric assay, containing 1 mM ATP and 0.2 mM 8-bromoinosine-59diphosphate as substrates [23]. The errors associated with the kinetic parameters are less than 20%.

Size-exclusion Chromatography
Size-exclusion chromatography was performed using a Superdex 75 HR 10/30 or a Superose 12 HR 10/30 column (Pharmacia, Uppsala) equilibrated with a buffer solution of 50 mM Hepes pH 7.4 containing 150 mM sodium chloride, and eluted at a flowrate of 0.4 mL/min. The column was calibrated with a set of molecular weight markers (BioRad Markers). Protein was detected by absorbance or by fluorescence intensity at 340 nm with excitation at 280 nm (excitation and emission bandwidths of 10 nm) using a flow cell on the LS50B spectrofluorimeter (Perkin-Elmer).

Circular Dichroism
CD ellipticity at 222 nm was recorded on a Jasco J 810 spectropolarimeter between 25 and 80uC at 1uC/min heating rate using a 1 mm quartz cuvette.

Crystallization of the D93N Mutant
The protein solution was dialysed against 20 mM Tris-HCl, pH 7.5 buffer containing 20 mM MgCl 2 and concentrated to 11 mg/mL. Crystallization screening was carried out using a Honeybee 961 robot (Cartesian Technology) mixing 200 nL of protein solution with 200 nL of reservoir solution (Crystal Screen, Hampton Research and The Classics Screen, Nextal). Crystals grew at 20uC in a few hours. Two different crystal forms were obtained: (i) hexagonal plates with 2.0 M ammonium sulfate, 0.1 M Tris-HCl, pH 8.5, (ii) rods with 2.0 M ammonium sulfate, 2% (v/v) PEG400, 0.1 M Hepes, pH 7.5. Crystals were cryoprotected in mother liquor supplemented with 20% glycerol (v/v) and flash-frozen in liquid nitrogen.

X-Ray Diffraction Data Collection
Complete data sets were collected at 107 K on the ID23-2 beamline (ESRF, Grenoble), scaled with SCALA from CCP4 suite and processed with MOSFLM [24]. The structures were solved by molecular replacement with MOLREP using the coordinates of the wild type Mt-NDPK (PDB id: 1k44) [13] as search model. Refinement was done using phenix.refine [25] alternated with manual model building using COOT [26]. Data collection and refinement statistics are gathered in Table 1. The surface areas and hydrogen bonds were calculated using PISA [27]. The crystal structure was drawn using PYMOL [28].

Miscellaneous
All experiments were repeated three times. The experiments were performed at 25uC in 20 mM sodium phosphate buffer (pH 7.0) unless otherwise stated. The pKa values for all acidic/ basic residues based on desolvation, hydrogen bonding and charge-charge interactions were computed with PROPKA [29].

Expression and Properties of wt and Mutant Proteins
The wt Mt-NDPK and D93N mutant were expressed in E. coli. The D93N mutant enzymes displayed catalytic properties very similar to those of the wt enzyme, within experimental errors. Mutation aspartate 93 to asparagine decreases k cat from 264 s -1 in the wt enzyme to 232 s -1 for the D93N mutant, while the apparent K m for 8-bromoinosine 59-diphosphate increases from 159 mM to 230 mM (measured at a fixed concentration of 1.0 mM ATP). UV, fluorescence and CD spectra were identical for the wt and mutant enzymes ( Figures S1 and S2). Both proteins were hexameric as ascertained by size-exclusion chromatography ( Figure S2). This indicates that the mutation does not affect the global structure of the Mt-NDPK.

The Hexameric Structure is Necessary for Full Enzymatic Activity
The results of the fluorescence stopped flow experiments show that Mt-NDPK recovered the monomeric native structure within 1 second. At low protein concentration, when diluting the GuHClunfolded Mt-NDPK directly into the assay medium, the recovered specific activity of the enzyme was about 4 U/mg, which represents about 1% of the hexamer activity. Such a low activity could be attributed to monomeric or dimeric species. The hexameric structure is necessary for full enzymatic activity during its dissociation or assembly as with other NDP kinases [30,31].

Thermal Stability of the Wild-type Mt-NDPK and D93N Mutant
The differential scanning calorimetry (DSC) experiments ( Figure 2) display only one calorimetric peak with the two proteins. The thermal stability, as measured by DSC, dramatically decreased when mutating the Asp 93 into Asn. The T m of the D93N mutant was 48.4uC while that of the wt enzyme was 76uC. No reversibility was ever observed after heat denaturation, so no complete thermodynamic analysis of the thermograms could be performed. Very close T m values were obtained measuring the enzyme inactivation and the ellipticity at 222 nm (see below) under identical protein concentrations and heating rate. As the thermal denaturation was irreversible, it was less informative than the chemical denaturation.

Stability to Chemical Denaturation
We used both urea and GuHCl as denaturants since Arg 80 and Asp 93 interact via an intersubunit salt bridge. The two denaturants act differently since GuHCl is a salt, while urea is a neutral molecule. Figure 3A displays denaturation transitions of wt Mt-NDPK in urea as measured by the fluorescence intensity change of the single  tryptophan residue (indicative for the tertiary structure), as well as by enzymatic activity (indicative for the quaternary structure). Denaturation is parallel with inactivation. No dissociated species could be detected by size-exclusion chromatography. Renaturation occurred at much lower urea concentrations. Previous studies showed that hexameric NDPKs display similar hysteresis in urea after denaturation/renaturation experiments [32,33]. The denaturation curve describes the transition from native hexamer (N 6 ) to the unfolded protein (U) (Eq. 1), while the renaturation curve measured by intrinsic fluorescence intensity describes the transition from the unfolded protein to folded monomer (N) (Eq. 2).
The reactivation yield was low. Hexamer formation (measured by reactivation) is the result of at least three second-order reactions (Eq. 3). This process is very slow at the low protein concentration used here [19]. For this reason the reactivation of Mt-NDPK was not studied. When we refolded the unfolded Mt-NDPK by dialysis at a much higher concentration (300 mg/mL), full reactivation was obtained with a specific activity of 550 U/mg.
The UV and fluorescence spectra of the species that accumulated during refolding were characteristic for a native protein. Such species eluted essentially as a monomer (10% of hexamers) by size-exclusion chromatography with appropriate calibration standards ( Figure 4A). Second, it is not a folding intermediate since it does not bind BisANS, a dye having a high affinity for the folding intermediates [33]. Moreover, an oligomeric state could also be excluded by the double dilution experiment ( Figure 5).
In contrast, a very different pathway appeared when performing the denaturation/renaturation by urea with the D93N mutant ( Figure 3B). Inactivation occurred at very low urea concentrations (,0.5 M). This indicates dissociation without loss of tertiary structure. This was demonstrated by size-exclusion chromatography ( Figure 4A). Unfolding and refolding were reversible and had a midpoint of concentration for denaturation (c 1/2 ) of 2.4 M urea. The denaturation curve was reversible and a DG NU of 4.6 kcal/ mol calculated. This value was very close to the DG NU calculated for the wild-type Mt-MDPK by a double dilution experiment ( Figure 5). The dramatic decrease of the protein stability was therefore due to the decrease of subunit interaction. Overall, the dissociation/denaturation of the D93N mutant can be described by Eq. 4. As we suspected the intersubunit salt bridge Arg 80 -Asp 93 to be essential for the hexamer stability, we next studied the reversible dissociation/denaturation by GuHCl. In contrast with urea, GuHCl is a salt. It has been shown that ionic interactions are cancelled in GuHCl denaturation experiments while still present in the denaturation experiments by urea [34]. The wild-type Mt-NDPK unfolded and inactivated simultaneously in the presence of GuHCl ( Figure 3C) as with urea as denaturant ( Figure 3A). Refolding occured at much lower GuHCl concentrations. Again, the renatured species at 0-0.5 M GuHCl was the folded monomeric protein.
The D93N mutant displayed an unexpected behavior in the denaturation experiment with GuHCl. The hexamer was stable and active up to 2.5 M GuHCl ( Figure 3D) and refolded had a c 1/2 of 0.95 M GuHCl. Importantly it easily dissociated in low urea concentrations ( Figure 3B). The loss of activity along with unfolding in higher GuHCl concentrations indicate again that dissociated species did not accumulate. The refolding experiments in presence of urea and guanidinium show identical c 1/2 refolding for the wt and D93N monomers. This indicates that the thermodynamic stability of the monomer has not been affected by the D93N mutation. The thermodynamic stability of the isolated subunits did not change by the D93N, as in urea.

Salts Stabilize Mt-NDPK
The experiments shown in Figures 3B and 3D with the D93N mutant indicate that urea was very efficient in dissociating the hexamer to native monomer, while GuHCl was not. This is a nontrivial observation since in general GuHCl is more efficient than urea in both dissociation and unfolding of proteins. The only explanation is the stabilization of the hexamer by GuHCl at concentrations lower than denaturing. This was found indeed to be the case. 1.5 M urea dissociated the D93N mutant to folded monomers, as indicated by the size-exclusion chromatography on a calibrated column ( Figure 4A). The folded monomer has the smallest size among all possible species (unfolded proteins, folding intermediates and oligomeric structures) and therefore no misinterpretation of the elution profile is possible. The fluorimetric detection allows us to analyze the molecular mass distribution at very low protein concentrations, the same as used for activity measurements or steady-state fluorescence analysis. The undenatured mutant eluted as a hexamer in buffer but also in the presence of 1.5 M GuHCl. The hexamer was also the major species in the presence of combined 1.5 M urea +1.0 M GuHCl ( Figure 4A).  We took advantage of the fact that the full enzyme activity is essentially associated with the hexamer to investigate the dissociating effect of urea in the presence of salts. By incubating the D93N mutant at 10 mg/mL with 1.5 M urea, very little activity was present after 16 h of incubation. When 1.0 M GuHCl was present in the incubation mixture in addition to 1.5 M urea, the enzymatic activity reached that of the control ( Figure 4B) and the enzyme was hexameric ( Figure 4A). Increasing the GuHCl concentration at more than 1.5 M, activity declined since GuHCl unfolded the protein. Other salts stimulated the hexamer formation. The anion was kept constant as chloride and the cations were monovalent (sodium, ammonium and guanidinium) or divalent (calcium and magnesium) ( Figure 4B).
These experiments show that many salts have an important stabilizing effect on the hexameric structure of the D93N mutant. As the quaternary structure has a major contribution to overall stability of NDPKs, we studied next the effect of salts on the thermal stability of the wild-type Mt-NDPK and D93N mutant ( Figure 6).
The T m of wild-type Mt-NDPK was 73uC in the absence of salt, 80.5uC in the presence of 0.15 M sodium chloride and 84uC in the presence of 0.15 M GuHCl ( Figure 6A). With the D93N Mt-NDPK the salt effect was even more impressive. The T m was 50uC in the absence of salt, 63uC with 0.15 M sodium chloride and 67uC with 0.15 M GuHCl ( Figure 6B). It should be noted that the proteins did not unfold completely even well above the T m , as the ellipticity remained negative. The final CD far UV spectrum was quite similar to that of the native enzyme with reduced amplitude. Native protein was incorporated into the aggregate, or partially folded species were generated. For this reason, the quantitative thermodynamic analysis was not reliable and has not been performed.

The D93N Mutation in Mt-NDPK does not Alter the 3D Structure of the Protein
In the wild-type Mt-NDPK hexamer, the interface between adjacent subunits was stabilized by one salt bridge and four mainchain/side-chain hydrogen bonds (Figure 7) [13]. Besides the intersubunit salt bridge, Arg 80 made an intersubunit hydrogen bonds with main chain carbonyl L109 and loosely with amide Q96 ( Figure 7B).
The D93N mutant crystallizes along two different space groups ( Table 1). The two structures are similar (0.40 Å of rmsd for 135 Ca) and differ only slightly in the conformation of a A and a 2 helices due to crystal contacts. The overall B-factor values of the two D93N structures (64 Å 2 and 71 Å 2 ) are similar and higher than that of the wild type (31 Å 2 ). That could be related to the stability of the oligomeric assembly. Analysis of B-factor reveals that (1) two domains, the a A -a 2 region (40-70) and the C Term part (120-136), are highly flexible and (2) the beginning of the Kpnloop where the mutation is located appears more flexible in the mutant than in the WT structure ( Figure S3). In D93N structures, the intra-subunit salt bridge is obviously broken and consequently neither Arg 80 nor Asn 93 were involved in intersubunit hydrogen bonds ( Figure 7C). Accordingly, the Arg 80 and Gln 96 side chains protruding on the surface of the hexamer are not well defined in the electron density map and appear disordered. In the mutant structure, although the overall structures of the monomer and the hexamer are essentially unaltered, the intersubunit interactions were clearly weakened.

Discussion
The Quaternary Structure of Mt-NDPK is Essential for Enzymatic Activity and for Stability The Mt-NDPK active site is located between the a 2 /a A helices and the Kpn-loop (amino acids 89 to 114). Amino acids participating in nucleoside binding and catalysis are very conserved in NDPKs [14]. The Kpn-loop is also involved in the contact formed by the assembly of three dimers into the enzymatically active hexamer. It is likely that in the folded monomers since the Kpn-loop is not held in place by subunit interactions, it has some mobility. This will decrease enzymatic activity as substrate binds with less efficiency. It should be noted that the enzyme k cat as a monomer state is 1% of that of the hexamer.
NDPKs are made of small subunits (135-180 amino acids) displaying a high sequence similarity (.45% identity) (Figure 1). Eukaryotic NDPKs are hexamers, while bacterial NDPKs are hexamers [13] or tetramers [15,16,35]. In both hexameric and tetrameric NDPKs, subunit structure is identical and two subunits associate in an identical way to generate a ''dimer''. It should be noted that ''dimers'' refer to a partial NDPK subunit association seen in the oligomer X-ray structure (tetramers or hexamer). True dimers are easily formed by tetramer dissociation in solution and are probably the basic assembly in tetrameric NDPKs. In contrast, dimers have never been observed by dissociation or during association of hexameric NDPKs. In a similar way, ''trimer'' refers to the assembly of three subunits in the hexamer structure and not to a trimer in solution. The discussion below will be restricted to the stability of hexameric NDPKs only. The ''dimeric'' interface is highly conserved in eukaryotes and bacteria [14]. The assembly of three ''dimers'' generates hexamers. Due to the D3 symmetry, each subunit interacts with three neighbors [14]. This makes the hexamer assembly very cooperative i.e., it can be hardly dissociated into lower-order oligomers. Most contributions to the ''trimer'' interface are the Kpn-loop and the C-terminal residues. The C-terminal tail of 15 residues of Dd-NDPK and other ''long'' NDPKs is missing in Mt-NDPK. Deletion of a few C-terminal amino acids in Dd-NDPK has been shown to greatly decrease the hexamer stability [20,36]. The tail is devoid of secondary structure and interacts with the neighboring subunits. The ''dimer'' and ''trimer'' buried surface areas (bsa) are much lower in ''short'' NDPKs ( Table 2). The quaternary structure plays an essential role in protein stability to denaturation. This has been described for dimeric proteins [37,38] but is more predominant for higherorder oligomers [39,40]. As a consequence, the disruption of intersubunit interfaces requires conditions which are denaturing for the dissociated subunits. Loss of quaternary structure appears simultaneously with the loss of tertiary structure. While studying the denaturation of dimeric proteins two pathways are possible: (i) the dissociation into folded monomers followed by unfolding, or (ii) the unfolding without the accumulation of dissociated species [41]. In higher order oligomers the situation is similar. Hexameric NDPKs unfold without accumulation of dissociated species. While studying refolding/association, subunit association is very slow under our protein concentration since at least three second-order reactions generate the oligomers. An apparent hysteretic phenomenon therefore appears. This is a kinetic effect and not a true hysteresis generated by a slow conformational change of the monomer [39,42,43]. The absence of reversibility makes thermodynamic calculations unfeasible.
The contact area between subunits is much smaller in Mt-NDPK than in other hexameric NDPKs due to the absence of the C-terminal tail ( Table 2). For this reason, when the crystal structure of the Mt-NDPK was solved it was a surprise that it was a hexamer. Moreover, as complex protein thermostability is related to contact area, it was further surprising that Mt-NDPK is as stable, or even more stable, to heat denaturation than NDPKs having much more extensive intersubunit contacts. Careful inspection of the ''trimer'' interface composition failed to supply any explanation why Mt-NDPK is quite thermostable. Interfaces are not more hydrophobic than the corresponding interfaces in Dm-NDPK and Dd-NDPK. We suggest that hexamer stability is due to the intersubunit salt bridge Arg 80 -Asp 93 .
Preliminary experiments suggest the possibility to incorporate Mt-NDPK into hexamers made with human NDPK, despite the differences in sequence and the absence of an interaction domain. The observed transport of Mt-NDPK in the nucleus of human cells [11] could be due to a cargo effect of human NDPK subunits in a mixed hexamer.

Mt-NDPK Hexamer Stability
Salt bridges located on the protein surface have been suggested to stabilize proteins from thermophilic and hyperthermophilic organisms [21,44]. Intersubunit salt bridges have been shown to have a major contribution to overall stability of some proteins to denaturation [45,46]. One such in Mt-NDPK is the intersubunit salt bridge Arg 80 -Asp 93 [13]. This interaction is missing in most NDPKs ( Figure 1) but is present in all NDPKs from Mycobacteria. Ionic interactions may contribute to a large extent to protein stability since they are efficient at a much longer distance than van der Waals interactions. For these reasons we decided to study the contribution of the Arg 80 -Asp 93 salt bridge to the stability of Mt-NDPK, by mutating Asp 93 into neutral asparagine.
Replacement of the Asp 93 with the neutral asparagine showed a dramatically decrease of the thermal stability. The T m measured by DSC drops from 76uC to 48uC. Here again, the hexamer integrity has been followed measuring the residual activity, while DSC and CD signals were due to unfolding. The three techniques The chemical denaturation studies with urea and GuHCl as denaturants showed a large decrease in hexamer stability as a result of the D93N mutation, while the stability of the isolated subunit was not affected. This is not surprising since Asp 93 is located on the subunit surface.
The most significant information on the effect of the mutations on hexamer stability was obtained when studying Mt-NDPK denaturation by urea. The wild-type inactivated/unfolded c 1/2 was about 5.2 M. The inactivation (loss of quaternary structure) and the intrinsic tryptophan fluorescence intensity change (loss of tertiary structure) were concomitant, which suggests that the isolated subunits are not stable under the conditions needed for dissociation, or the hexamer unfolds without dissociation. The two patterns cannot be distinguished under our experimental conditions. Comparing the stability of the hexamer with that of isolated subunits reveals the important role of the quaternary structure to stabilize the overall protein native structure. We showed previously that in acidic conditions, isolated monomers of Dd-NDPK are unstable and form molten globule folding intermediates, while the hexamer conformation stays unchanged [47]. Moreover, isolated subunits of human NDPK-A cannot be native [48], while the native hexamer is quite stable. During Mt-NDPK renaturation, only the folded monomer was detected but no higher-order dissociated species such as dimers or trimers. One may speculate that evolution pressure acts on the hexamer stability and not on a single ''partial'' interaction. The D3 hexamer is very cooperative since each subunit has contacts with 3 other subunits. The hexamer is very stable even if all individual subunit-subunit interactions are rather weak.
In presence of urea, the mutation of the negatively charged Asp 93 into the neutral Asn had a dramatic effect on the hexamer stability. : the c 1/2 of the wt hexamer decreased from 5.2 M to less than 0.5 M for the D93N mutant. Folded monomers presented a c 1/2 of 2.5 M urea for the wt as well as for the mutant. The hexamer stability decrease was therefore not due to subunit destabilization (see also Figure 6). In the crystal structure of D93N mutant, no direct interaction of the Asn 93 exists with neighbouring subunit.
Long-range ionic interactions may also be involved with charged residues. Based on these interactions, PROPKA software calculates a rough estimate for the free energy of unfolding. When changing an amino acid residue, the interactions changes and so does the free energy of unfolding. The calculated DG for the Mt-NDPK hexamer was 142.9 kcal/mol. It decreased to 124.4 kcal/ mol for the mutant D93N, respectively. It appears that the stability calculated from PROPKA software actually corresponds qualita-tively with the measured T m or with the hexamer stability in urea as denaturant. The D93N mutation appears to decrease the protein stability because the negative charge of the Asp 93 interacts with distant protein charges and stabilizes the hexamer.

Role of Cation Binding for Hexamer Stability
The larger stability of the D93N hexamer in GuHCl than in urea is very surprising. Unexpectedly GuHCl stabilizes the hexamer, at low concentrations. Other monovalent (Na + and NH 4 + ) or divalent (Mg + and Ca + ) cations also stabilize the hexamer. The guanidinium cation is effective at slightly lower concentrations than Na + or NH 4 + , while keeping the anion constant. GuHCl is denaturing at higher concentrations, while it is stabilizing up to 1 M. This feature explains why GuHCl is much less efficient in dissociating the D93N mutant Mt-NDPK. Guanidinium cation is large and hydrated and should therefore interact better with the protein surface than small cations.
Protein stabilization by monovalent cations is not frequent but some examples have been described [49,50]. The stabilization mechanism of Mt-NDPK is unknown, but we suppose to be related to cation binding to the protein rather than an effect mediated by the change of global solvent properties. Indeed, cations which are on the opposite ends of the Hofmeister series are stabilizing, with similar efficiencies. Second, the effect is halfmaximum effect at about 100 mM. It is however too high for measuring binding affinity and stoichiometry. This is much lower than the common stabilization mediated by a global solvent effect, which appears at molar salt concentrations. The hexamer formation with the D93N mutant in 1.5 M urea was an original way to detect and quantify the stabilizing effect in a functional way.

Conclusions
The most important conclusion of this study was that the quaternary structure is essential for enzymatic activity and for the stability to denaturation. The lower contact surfaces between subunits, as compared to other NDPKs, are compensated by the intersubunit salt bridge Arg 80 -Asp 93 . This makes the Mt-NDPK quite thermostable. The thermal stability of proteins measured in vitro cannot be discussed as an adaptation for in vivo conditions, with the exception of the proteins from (hyper)thermophilic organisms. M. tuberculosis is a mammalian parasite and therefore lives at about 37uC. Instead, the thermal stability has been correlated with the kinetic stability of proteins in vivo [51,52]. A large number of proteins of M. tuberculosis have been shown to be relatively thermostable [53,54,55,56]. Mt-NDPK is not an exception in this context. Supporting Information Figure S1 UV and fluorescence spectra of wild-type Mt-NDPK (blue) and D93N mutant (red). (A) The UV spectra were recorded with a protein concentration of 1.19 mg/mL (WT) or 1.00 mg/mL (D93N) in 20 mM phosphate buffer, pH 7. Note that the shoulder at 300 nm is a specificity of native form of Mt-NDPK. (B) Tryptophan fluorescence (295 nm excitation) was measured at 25uC from 310 to 390 nm. The fluorescence spectra were recorded with a protein concentration of 10 mg/mL. mutant structures (red). Analysis of B-factor reveals that (1) two domains, the a A -a 2 region (40-70) and the C Term part (120-136), are highly flexible and (2) the beginning of the Kpn-loop where the mutation is located appears more flexible in the mutant than in the WT structure. The s(B) of the two D93N structures (Pdb_Id: 4anc, 4and) and of the WT structure (Pdb_Id: 1k44) are 29 Å 2 , 25 Å 2 and 13 Å 2 , respectively. (TIF)