Four Mtb ALDH genes (Rv0223c, Rv0458, Rv2858c, and Rv3293) were previously identified as interacting with nucleosides using dye-resin chromatography and ligand-specific elution [24]. Each targeted ALDH gene was amplified by PCR from a M. tuberculosis H37Rv COSMID library as the template with Pfu proof-reading DNA polymerase (Stratagene), using the 5′ NdeI primer, 5′-AGATATACATATG + (N-terminal 21 bases of target sequence)-3′, and the 3′ BamHI primer, 5′-AATTCGGATCC + (C-terminal 23 bases of target sequence)-3′. The underlined bases represent the NdeI and BamHI sites, respectively. The PCR amplicon was digested with NdeI and BamHI restriction endonucleases (NEB), and cleaned using Qiaquick PCR spin column (Qiagen). The product was ligated into a modified pET-28 vector containing a C-terminal 6-His tag, in frame with the BamHI restriction site using T4 DNA ligase (New England BioLabs), and transformed into BL21(DE3) (Novagen). The expressed proteins contained the C-terminal tag GSHHHHHH, where GS is encoded by the BamHI restriction site (GGATCC). BL21(DE3) 3 ml cell culture was tested for the expression of heterologous protein by binding on a Cobalt-chelated Talon superflow bead slurry (Clontech) and SDS–PAGE analysis.

Cell culture was performed as described by Studier [25] with some modifications. Transformed cells were inoculated into 3 ml seed culture media (1 mM MgSO₄, 0.5% glucose, 17 amino acids of 100 μg/ml for each Na-Glu, Asp, Lys-HCl, Arg-HCl, His-HCl, Ala, Pro, Gly, Thr, Ser, Gln, Asn, Val, Leu, Ile Phe, Trp, metal mix of 50 μM Fe, 20 μM Ca, 10 μM Mn, 10 μM Zn, 2 μM for each Co, Cu, Ni, Mo, Se and B, 5 mM PO₄, 5 mM Na, 2.5 mM K, 2.5 mM NH₄ and 1.25 mM SO₄), and grown overnight at 37°C. From the seed culture, 500 μl was inoculated into 500 ml auto-induction media, containing 1 mM MgSO₄, metal mix (same as seed culture), 0.5% glycerol, 0.5% glucose, 0.2% α-lactose, NPS (same as seed culture), and 35 μg/ml kanamycin. After cells were grown at 37°C until OD₆₀₀ reaches 0.5, the growth was continued at 20°C for approximately 16 h until the OD₆₀₀ reached approximately 15. The cells were harvested and stored at −80°C.

The cell pellet was lysed by sonication in 10 ml of buffer A (20 mM Tris–HCl, pH 8.0, and 100 mM NaCl) per gram of cells for 10 min in 30 s pulses at 10°C. The cell debris was removed by ultra-centrifugation for 30 min at 38,000 rpm using a Ti 60 rotor (Beckman). The clear supernatant was filtered through a 0.45 μm pore membrane and loaded on a 5 ml Talon superflow affinity column equilibrated with buffer A. After washing with 30 ml buffer A and 20 ml buffer B (20 mM Tris–HCl, pH 8.0, 100 mM NaCl and 20 mM imidazole), the His-tagged Rv0223c (and the other ALDHs) was eluted from the cobalt affinity column using Buffer C (20 mM Tris–HCl, pH 8.0, 100 mM NaCl and 300 mM imidazole). The eluted fraction was dialyzed against Buffer D (20 mM Tris–HCl, pH 8.0, 100 mM NaCl and 10 mM β-mercaptoethanol) and purified by gel filtration on a Superdex-75 column (GE Healthcare Inc.) using Buffer D for equilibration and elution. The peak fractions (monitored at OD₂₈₀) were analyzed by SDS–PAGE and the pooled protein fractions were concentrated using a Centricon Plus-20 (Millipore) up to 35 mg/ml, which was measured by Bradford assay with IgG (Bio-Rad) as a standard. The purity of each protein was estimated to be higher than 95% based on densitometry of SDS–PAGE gels [26].

Recombinant proteins were evaluated for their ligand-binding properties using a modified affinity elution chromatography protocol [27]. Individual proteins were diluted to 2 mg/ml in column buffer (CB, containing 50 mM potassium phosphate, pH 7.5, 1 mM MgCl₂ and 2 mM DTT) and adsorbed to multiple small aliquots of F3GA resin (100 μg protein per 10 mg resin) in 2 ml spin-columns (Costar, Fisher Scientific). Binding was for 1 h at 4°C with very gentle vortexing, followed by recovery of unbound protein (flow-through fraction) and washing of the resin (4 × 0.4 ml washes with CB); spin-columns were micro-centrifuged for 30 s at 10,000×g, to recover fractions and change solutions. Individual spin-columns containing resin-bound proteins were then incubated (as for protein binding, above) with 50 μl 1 mM test ligand in CB, and the elution fractions recovered by centrifugation. Protein which remained bound to the resin was recovered by heating at 95°C for 5 min in 100 μl SDS sample buffer, and centrifugation (resin fraction). Aliquots of initial protein, spin-column flow-through and eluate fractions were diluted 1:1 with 2× SDS sample buffer, and loaded in equal proportion (equivalent to 1 μg input protein) on 15% gels and stained with silver.

Crystallization experiments were carried out by the hanging-drop vapor diffusion method [28] at room temperature (25°C) using 24-well plates. Recombinant Rv0223c was tested for the presence of bound nucleotides (see Supplementary Methods); the results indicated that the protein has at most 0.1:1 bound NAD or other nucleotide (see Supplementary Fig. 1). Each protein–ligand solution was prepared by mixing protein (0.68 mM in solution of Buffer D) with the corresponding ligand (20 mM in H₂O) at a molar ratio of 1:2 protein:ligand. The mixtures were incubated at room temperature for 30 min prior to setting up crystallization experiments. The final concentration of protein in each protein–ligand mixture was between 0.60 and 0.63 mM. The ligands used were NAD, NADH, NADP, NADPH, adenosine, AMP, ADP, ATP, GTP, FAD, FMN and Cibacron Blue F3GA (free dye). Crystals were grown from drops consisting of 1 μl protein–ligand solution mixed with 1 μl of reservoir solution against a reservoir containing 0.1 M MES (pH 6.0) and 0.8 M ammonium sulfate, for 3 days at room temperature. For some ligands (e.g. ATP), the effect of Mg⁺⁺ was tested for crystallization, but without any noticeable effect, crystallization experiments reported were executed without Mg⁺⁺. Native and SeMet Rv0223c-NAD complex crystals were flash-cooled in liquid N₂ with the addition of 10% glycerol in the crystallization buffer as cryoprotectant. Three-wavelength selenium multi-wavelength anomalous dispersion (MAD) data was collected at the beam line 5.0.2 at the Advanced Light Source (ALS). A native data set at a resolution of 1.8 Å was collected at the beam line 8.2.1 at the ALS. Both data sets were processed with the HKL2000 program suite [29].

Initial phasing was carried out with the program SOLVE [30] using the MAD data set. The resulting experimental map was density modified and traced using the program RESOLVE [31]. The protein model was further improved and built with the ARP/WARP package [32, 33] against the 1.8 Å native data. Manual model rebuilding was carried out with programs COOT [34]. The final model of this complex is deposited in the Protein Data Bank (http://www.rcsb.org) as entry 3B4W, and has R/R_free values of 0.18/0.20 at a resolution of 1.8 Å, after refinement with the phenix.refine program from the PHENIX software package [35]. Difference electron density maps were calculated for each Rv0223c-ligand complex by refining the structure of the Rv0223c protein (without ligands or solvent molecules) against the observed structure factor amplitudes for each complex. The resulting crystallographic phases were used to construct an (mF_o − DF_φ)e^iφc difference map [36]. The LigandFit algorithm for automated ligand-fitting in PHENIX [37] was used to identify the location of the largest contiguous regions of high density in the difference map. In this algorithm the contour level for identification of contiguous regions of density is set to a level such that the largest region is approximately the size of the anticipated ligand. In this way, the location of this largest region gives an indication of the location of the ligand. The difference electron density for each complex is shown in the region of the NAD in the Rv0223c protein-NAD complex and is displayed with PyMOL [38]. The overall structure comparison between Mtb Rv0223c and human mitochondrial ALDH-2 was performed by the DaliLite program [39].

In Table 2, we examine the degree of sequence identity among these four Mtb ALDHs and their closest human homologs in more detail. The sequence identities show that the four Mtb proteins have more sequence homology to specific human ALDHs than to each other; for example, Rv0223c, Rv0458, and Rv2858c show distinctively smaller e-values and higher percentages of identical residues when aligned with human ALDH class 1 and 2 protein (than when they are aligned with other Mtb ALDHs) by the NCBI Blastp program, and Rv3293 shows the same result with human ALDH class 7.Table 2

Homology analysis of Mtb ALDHs being compared with other Mtb ALDH or Human ALDH by the NCBI Blastp program

We used a modified version of our dye-resin/ligand-elution procedure to examine the specificity of ligand binding of one of the four Mtb ALDHs (see Materials and Methods). In this ligand-specific elution screen, recombinant Rv0223c was adsorbed on small aliquots of F3GA resin in spin-columns and assayed for elution by twelve nucleotides and nucleosides using one ligand per column. By using purified recombinant protein and one ligand per column, we identify effects of individual ligands on the stability of the dye-Rv0223c protein complex. We expect that those ligands that cause the Rv0223c protein to elute from the resin are likely to bind specifically to the protein, although non-specific interactions could potentially also cause elution to occur [24].

Figure 1 shows that the Rv0223c protein was eluted by NAD and NADH but only weakly by NADP and NADPH. This result is consistent with the dinucleotide preferences of homologous NAD-dependent aldehyde dehydrogenases (E.C. 1.2.1.3) from other organisms [40] and is addressed further below in the context of structural data. Additionally, the elution of Rv0223c protein by AMP/ADP/ATP is consistent with adenylate-binding by other dehydrogenases (cf. [3]).Fig. 1

Ligand-specific elution chromatography of recombinant Rv0223c, a putative Mtb aldehyde dehydrogenase. One-dimensional SDS–PAGE analysis of protein added (Rv0223c), not bound (FT, flow-through fraction) or released from Cibacron Blue F3GA resin by buffer or 12 different ligands shows the preferential release by NAD, NADH, ATP, and ADP

Ligands that bind specifically to proteins are potentially useful for protein crystallization and structure determination [41]. We therefore examined the effects of the nucleotide/nucleoside ligands shown in Fig. 1 on the crystallization of recombinant Rv0223c protein. We carried out a systematic screen for conditions that led to crystallization of Rv0223c in the presence of NAD, and found one condition that led to crystals of the Rv0223c+NAD complex diffracting to high resolution (1.8 Å; see Materials and Methods; Table 3). We then carried out crystallization trials of the purified protein under this condition, but in the presence of either no ligand or the various nucleoside ligands used in the dye-ligand chromatography assay. As shown in Fig. 2, ligand-free protein generally yielded either very small or no crystals in repeated tests for crystallization (in one experiment, crystals were obtained that diffracted to a relatively low resolution of about 3 Å). In contrast, co-crystallization with several ligands (NAD, NADH, NADP, adenosine, AMP, GTP and FAD) enhanced protein crystal formation, yielding crystals diffracting to high resolution suitable for structural analysis. Other ligands either hardly affected crystallization (NADPH, ADP and ATP) or promoted precipitation (FMN, and non-resin-bound F3GA). NMN was not tested. As summarized in Table 4, (although further screening of crystallization conditions may be required for the interactions between Rv0223c and ADP, ATP and F3GA) the strongly interacting ligands (NAD and NADH) enhanced crystal formation, leading to higher resolution data than the moderately or weakly interacting ligands (AMP, FAD, adenosine, GTP and NADP).Table 3

Data collection and refinement statistics

The structure of the Rv0223c+NAD complex was determined by MAD phasing at a resolution of 1.8 Å (PDB ID 3B4W; Table 3). The structures of all the complexes, Rv0223c+ligand (NADH, AMP, FAD, adenosine, GTP or NADP) were isomorphic with this one, with less than 1% changes in cell parameters. Consequently, we were able to use the structure of the Rv0223c protein (without ligand) from the Rv0223c+NAD complex to calculate phases and a difference electron density map for each of these structures (after refinement using the observed structure factors for each structure) using the diffraction data indicated in Table 4.

The crystals obtained with NAD, and with its reduced form NADH, were the largest and best-diffracting crystals (resolution range 1.8–2.0 Å) that we obtained in repeated experiments, and the ligand was most clearly defined in the difference electron density maps for these complexes (Fig. 3) although the crystals could potentially contain NAD because NADH showed ~50% conversion to NAD after about 3 days in the crystallization condition used. These data suggest that Rv0223c forms the most well-defined complexes with NAD and NADH, consistent with the gene annotation (putative aldehyde dehydrogenase).Fig. 3

Structure of Mtb Rv0223c with NAD (in center; resolution range 1.8–2.0 Å), and the difference electron density map of other nucleoside ligands that promoted crystallization of Rv0223c. NADH (resolution range 1.9–2.0 Å), NADP (2.7–2.8 Å), adenosine (2.3–2.6 Å), AMP (2.6 Å), GTP (2.5–2.9 Å) and FAD (2.5–3.0 Å) identified at the binding site of NAD

The structure of Mtb Rv0223c is the first reported structure from the Mtb ALDH family. We compared this structure with that of its human homologue, NAD-dependent mitochondrial ALDH, ALDH-2 (PDB ID 1CW3) [42], using the DaliLite server [39] to align the structures, as shown in Fig. 4. The choice of human ALDH-2 for comparison was based on sequence similarity (Table 2), and because Rv0223c has generally been categorized as a class 2 enzyme [43, 44]. The structure of Rv0223c closely overlapped that of human ALDH-2, with an overall rmsd of 1.5 Å over 184 residues in common. Only minor regions of incongruity were found, and these were in peripheral regions of the protein (Fig. 4).Fig. 4

Comparison of crystal structure of Mtb Rv0223c (green) with NAD (red) against human mitochondria aldehyde dehydrogenase-2 (orange) structure. The homology of two structures is very high with RMSD value 1.5 Å. In Rv0223c structure, the miss-aligned sequences (in magenta) are minor, consisting of residues; Pro360, Glu361, Gly362, and Leu363 in the catalytic domain, Asp9, Lys10, and Leu11 in the coenzyme domain, and Gly135, Ser136, Tyr137, Gly138, Gln139, and Ser140 in the oligomerization domain. The three domains are sectioned by purple lines. The adenine ring of NAD (and NADH) is located in the hydrophobic pocket formed by helices αD (gray) and αE (blue) in the coenzyme domain

The partial amino acid sequences of Rv0223c and human mitochondrial ALDH were aligned based on secondary structure for sequence homology analysis, as shown in Fig. 5. The full length Mtb and human sequences are shown (487 and 518 amino acids, respectively), and were determined by NCBI Blastp to be 39% identical and 53% similar. The two ALDH signature sequences identified by Prosite [46], and including cysteine and glutamic acid active site residues, were conserved in both proteins and found at identical locations.