A functional role of Rv1738 in Mycobacterium tuberculosis persistence suggested by racemic protein crystallography

Significance

Racemic protein crystallography was used to determine the X-ray structure of the predicted Mycobacterium tuberculosis protein Rv1738, which had been completely recalcitrant to crystallization in its natural l-form. Native chemical ligation was used to synthesize both l-protein and d-protein enantiomers of Rv1738. Crystallization of the racemic {d-protein + l-protein} mixture was immediately successful. The resulting crystals diffracted to high resolution and also enabled facile structure determination because of the quantized phases of the data from centrosymmetric crystals. The X-ray structure of Rv1738 revealed striking similarity with bacterial hibernation factors, despite minimal sequence similarity. We predict that Rv1738, which is highly up-regulated in conditions that mimic the onset of persistence, helps trigger dormancy by association with the bacterial ribosome.

Abstract

Protein 3D structure can be a powerful predictor of function, but it often faces a critical roadblock at the crystallization step. Rv1738, a protein from Mycobacterium tuberculosis that is strongly implicated in the onset of nonreplicating persistence, and thereby latent tuberculosis, resisted extensive attempts at crystallization. Chemical synthesis of the l- and d-enantiomeric forms of Rv1738 enabled facile crystallization of the d/l-racemic mixture. The structure was solved by an ab initio approach that took advantage of the quantized phases characteristic of diffraction by centrosymmetric crystals. The structure, containing l- and d-dimers in a centrosymmetric space group, revealed unexpected homology with bacterial hibernation-promoting factors that bind to ribosomes and suppress translation. This suggests that the functional role of Rv1738 is to contribute to the shutdown of ribosomal protein synthesis during the onset of nonreplicating persistence of M. tuberculosis.

Tuberculosis is caused by the bacterium Mycobacterium tuberculosis (Mtb), currently estimated to infect one-third of the world’s population (1). A major reason for this high rate of infection is the ability of Mtb to enter a dormant state known as nonreplicating persistence (2, 3). This occurs in response to engulfment of the bacterium by activated macrophages. A proportion of bacteria survives the immune system’s antimicrobial onslaught and persists in a dormant form, but can reemerge many years later as an active infection. In the nonreplicating persistent state, protein synthesis is drastically shut down (4), and the bacteria use host lipids, notably cholesterol, as a carbon source (5). Bacteria in this state are largely resistant to current drugs.

Microarray studies have identified a number of genes that are highly up-regulated under conditions thought to trigger the onset of dormancy, including hypoxia (low oxygen concentration) and exposure to nitric oxide (6, 7). The most highly up-regulated gene in hypoxic conditions, and the second-highest in response to NO, is Rv1738, suggesting this gene to be a key element in the onset of persistence. Rv1738 is also classed as an essential gene in the genome-wide transposon mutagenesis study of Sassetti et al. (8). The Rv1738 gene encodes a predicted 94-residue protein of unknown function.

In this research, we set out to determine the structure of the protein encoded by the ORF Rv1738 of Mtb H37Rv. Our hypothesis was that important clues to the function of the Rv1738 protein might be obtained from its 3D structure, as structure is much more highly conserved than sequence during evolution, and can reveal unsuspected functional and evolutionary relationships. There have been both failures and successes from this approach, but on discovering that Rv1738 could readily be expressed in soluble form in Escherichia coli, we initiated structural studies. Unfortunately, all attempts to crystallize recombinant Rv1738 failed, and the protein also proved unsuitable for structural analysis by NMR because of a tendency to aggregate.

Because of our inability to obtain crystals using protein expressed in E. coli, we turned to racemic protein crystallography, a recently introduced approach that can facilitate crystallization of recalcitrant proteins (9). In this method, a racemic mixture comprising equal amounts of the d-protein and l-protein forms of a target molecule are used for crystallization. A protein racemate can crystallize in centrosymmetric space groups, and can thus access many additional space groups, including some predicted to be highly favored for protein crystallization (10). In contrast, natural proteins are chiral, built only from l-amino acids and the achiral amino acid glycine, and thus cannot be incorporated into crystal lattices that include mirror planes or inversion centers. Evidence is accumulating that racemic protein mixtures can be much easier to crystallize in cases in which the natural l-protein is refractory to crystallization (9). d-proteins can only be made by total chemical synthesis (11, 12), but as Rv1738 is a small protein, we considered it a good candidate for total chemical synthesis using modern ligation methods.

Here we describe the total chemical synthesis of the d- and l-forms of Rv1738 and the successful crystallization of the {d-Rv1738 + l-Rv1738} protein racemate, in striking contrast to the complete failure of conventional crystallization approaches with the recombinant protein. The racemic crystals diffracted to good resolution, and the structure was solved using an ab initio approach that benefited from the quantized nature of centrosymmetric phases (13). Moreover, the structure of Rv1738 revealed a surprising similarity to a family of stress proteins known as hibernation-promoting factors (HPFs). This suggests that the functional role of the up-regulated Rv1738 protein in nonreplicating persistence of Mtb is to contribute to the shutdown of ribosomal protein synthesis.

Results

Attempts to Crystallize Recombinant Rv1738 Proteins.

Rv1738 was readily expressed in E. coli as a soluble, His₆-tagged protein at expression levels of 10 mg/L. After purification, crystallization trials were undertaken using a Cartesian nanoliter dispensing robot, with a 480-component screen (14). No crystals were obtained, either with or without the His₆-tag. Attempts were made to crystallize alternative constructs with truncations at the N- and C-termini; the truncations were intended to remove residues preceding the predicted N-terminal β-strand and following the predicted C-terminal α-helix. Other “rescue” strategies that change a protein surface and can alter the propensity for crystallization were also used, including reductive methylation of lysine residues (15) (Rv1738 has four lysines) and surface entropy reduction (16). In the latter approach, eight mutant proteins were prepared in which residues with long, flexible side chains were mutated to alanine. These proteins included single mutants K29A and K37A, double mutants K29A/K37A and K72A/K76A, triple mutants E36A/K37A/E38A and K29A/K72A/K76A, a quadruple mutant K27A/E36A/K37A/E38A, and a sextuple mutant K27A/E36A/K37A/E38A/K72A/K76A. The three proteins with the EKE mutation aggregated and were unsuitable for crystallization, but the other five proteins were used in crystallization trials. No crystals were obtained from any of these protein constructs.

Total Chemical Synthesis of d-Rv138 and l-Rv1738 Proteins.

The 94-residue Rv1738 polypeptide was synthesized by native chemical ligation of three unprotected peptides comprising residues 1–29, 30–65, and 66–94 (Fig. 1 and Fig. S1). Residues Ala30 and Ala66 were initially changed to Cys to enable native chemical ligation at those sites, and were then restored to native Ala residues by desulfurization (17, 18). The same synthetic scheme was used to make each of the full-length l- and d-polypeptide chains, which were folded by rapid dilution after reverse-phase HPLC purification (Fig. S2).

Fig. 1.

Synthetic strategy for preparation of Mtb Rv1738. The 94-residue polypeptide, with amino acid sequence MGGDQSDHVL QHWTVDISID EHEGLTRAKA RLRWREKELV GVGLARLNPA DRNVPEIGDE LSVARALSDL GKRMLKVSTH DIEAVTHQPA RLLY, was prepared from three synthetic peptide segments. Native Ala residues (bold, underlined) were initially replaced by Cys residues at the two ligation sites. The first native chemical ligation was followed by conversion of the N-terminal thiazolidine (Thz: 1,3-thiazolidine-4-carboxylic acid) to Cys. This was followed by a second native chemical ligation step. Once the full-length polypeptide was obtained, selective desulfurization was used for the conversion of Cys to Ala in the presence of Met.

Racemic Crystallization and Structure Determination.

Synthetic Rv1738 was crystallized as a racemic mixture consisting of {d-Rv1738 + l-Rv1738}. Conditions 25–96 of the Index crystallization screen (HR2-144; Hampton Research) were screened. These crystallization experiments gave immediate success within 4–7 d, yielding microcrystals in 10 of 72 conditions, for a success rate of 14%. After optimization, synchrotron data were collected for crystals from two conditions. Both crystal forms were in the centrosymmetric space group C2/c, with one molecule per asymmetric unit. Form-I crystals (26% solvent content; V_m = 1.66 ų/Da) diffracted to 1.65 Å resolution, whereas Form-II crystals, although having greater solvent content (45%; V_m = 2.23 ų/Da), diffracted to higher resolution (1.5 Å).

Fragment-Based Ab Initio Structure Solution.

To exploit the predicted advantages of racemic protein crystallography for structure solution (13), we tried a fragment-based ab initio approach based on that used in the program ARCIMBOLDO (19). Because the latter does not work on centrosymmetric space groups, we developed an analogous procedure, described here, and determined a near-complete Rv1738 model, starting with a single idealized α-helix as a search model. Both circular dichroism spectra and secondary structure prediction had indicated that Rv1738 contained at least one α-helix.

An initial structure was determined using 1.9 Å data from the Form-II crystals. Molecular replacement (MR) in PHASER (20) placed a single 14-mer poly-Ala α-helix in the unit cell, giving values of R/R_free of 76.4%/76.2%. R factors for a centrosymmetric crystal structure are ∼1.45 times higher than those from the same model in a noncentrosymmetric crystal structure (21). Thus, although an MR solution with R_free of 76.2% would normally seem unpromising, it is equivalent to R_free ∼50% for a noncentric structure. This solution, making up less than 8% of the total scattering matter of the final Rv1738 model, was refined with REFMAC (22) to improve the phases. Automatic “build” and “extend” model building cycles with BUCCANEER (23), iterated with REFMAC and removal of residues that did not fit the Rv1738 sequence, led to a model comprising 87 of 94 residues, with R/R_free of 41.4%/44.7%. The structure solution was then completed with conventional model-building with COOT (24) and refinement in PHENIX/REFINE (25), using all data to 1.5 Å resolution.

The final Form-II model comprised one Rv1738 protein molecule (residues 7–94), 88 waters, one glycerol, and a trifluoroacetate ion, and had R/R_free values of 22.8% and 25.3% (equivalent to R_free ∼18% for a noncentric structure). The Form-I crystal structure was then solved by MR, using the Form-II structure as a search model, followed by refinement in PHENIX/REFINE and model building in COOT. Validation checks were carried out with MOLPROBITY (26). Details of both final models, refined at 1.65 and 1.5 Å resolution, respectively, are found in Table S1.

Structure of the Rv1738 Protein.

The two crystal forms have the same space group, but distinct packing, with differing proportions of solvent. In each case, the unit cell contains four dimers, two l-dimers, and two d-dimers, with the l- and d-dimers related to each other by centers of inversion. No significant differences are apparent between the protein structures obtained from the two crystal forms, although the higher-resolution structure was more complete, with residues 7–94 well defined by electron density.

The Rv1738 polypeptide is folded into a three-stranded antiparallel β-sheet with strand order β1-β2-β3, packed against a long C-terminal amphipathic α-helix, residues 55–86 (Fig. 2A). The protein forms an intimately associated homodimer (Fig. 2B) in which the two monomers are related by a crystallographic twofold axis. The Rv1738 dimer is arranged in two layers. The antiparallel β-sheets are paired through their β1 strands to form an extended six-stranded β-sheet that provides a curved outer face for the dimer. This extended β-sheet wraps around the two antiparallel C-terminal α-helices, which pack against the β-sheet and make intimate side chain–side chain interactions with each other. The dimer core is roughly cylindrical, with dimensions 57 × 25 Å.

Fig. 2.

Structure of Rv1738. (A) Cartoon diagram, with β-strands in blue and the single C-terminal α-helix in green. The corresponding topology diagram is also shown. (B) The Rv1738 dimer, formed by twofold crystallographic symmetry. The two C-terminal α-helices pack antiparallel to one another, with the Rv1738 β-sheets combining to give an overall 6-stranded sheet that folds over the two helices.

Both crystal forms are in the same centrosymmetric space group and are dominated by interactions between the l- and d-dimers. In Form-I, residues 5–10 of the extended β1 strands, which are not involved in intramonomer or intradimer hydrogen bonding, are paired to form an antiparallel β-ribbon that links the l- and d-dimers across a center of inversion. Similar β-type hydrogen bonds also link the edge (β3) strands of the l- and d-dimers, again across a center of inversion. This gives a tightly packed (26% solvent) crystal built from layers of l- and d-dimers linked by a continuous network of hydrogen bonds. Similar crystal packing pertains in Form-II (Fig. 3), but with looser packing (45% solvent); the β3 strands are hydrogen-bonded between dimers (Fig. 3A), but the β1 strands are farther apart and do not interact directly.

Fig. 3.

Form-II crystal packing. (A) l- and d-dimers of Rv1738 are in contact across a center of inversion (o), forming antiparallel β-type hydrogen bonds (Inset). (B) The overall crystal packing, viewed down the b axis, is formed by rows of l-dimers (blue) and d-dimers (green), emphasizing its dependence on interactions between the two enantiomeric forms of Rv1738.

Structural Similarity of Rv1738 to HPFs.

Rv1738 is one of a small group of proteins from mycobacteria and other actinobacteria that form the PFAM (27) family DUF1876 (domain of unknown function number 1876), but outside this family, no significant sequence relatives were found. Searches of the Protein Data Bank were conducted with SSM (28), using the Rv1738 monomer as the query structure. The best match was another Mtb protein of unknown function, Rv2632c (PDB ID code 2fgg), with an rms difference (rmsd) of 1.23 Å for 80 equivalent Cα positions and 51% sequence identity. Other matches were all of low significance, but the top hits consistently identified similarities with domains that bind double-stranded RNA (dsRNA). These dsRNA-binding domains typically have an αβββα motif of ∼70 residues (29), and although Rv1738 lacks the first α-helix, its three-stranded β-sheet and C-terminal α-helix superimpose on the dsRBD βββα motif with rmsds of ∼2 Å for ∼55 Cα positions.

More insight came from comparisons with the Rv1738 dimer. These revealed close structural correspondence with a family of ribosome-associated proteins (S30Ae ribosomal proteins, PFAM family PF02482), which have a conserved βαβββα motif. These include the E. coli cold shock protein YfiA (29, 30) and the HPF proteins from E. coli and Vibrio cholerae (31, 32), which restrict protein synthesis during environmental stress. Although the HPFs and YfiA are monomeric, their four β-strands overlay four of the six β-strands of the Rv1738 dimer, and their two helices correspond spatially with the two helices of the Rv1738 dimer. Overall, ∼70 Cα atoms can be matched with an rmsd of 1.5–2.0 Å (Fig. 4A). There are differences, in that the β1 strand and α1 helix of HPF are antiparallel to the matching segments of the Rv1738 dimer, and the α1 helix diverges toward its N-terminal end, but the overall structures are of similar size and shape. Importantly, a band of positive charge around the waist of YfiA and HPF, which is one of their defining features and key to ribosome binding (33, 34), is also found in the Rv1738 dimer (Fig. 4B), and conserved basic residues in YfiA and HPF that contact the ribosome have counterparts in the Rv1738 dimer (Fig. 4A). In contrast, Rv2632c (see earlier) is not dimeric and shares none of these basic residues or positive charge, suggesting Rv1738 has a distinct biological role related to HPFs and the onset of dormancy in Mtb.

Fig. 4.

Relationship of Rv1738 to bacterial stress response proteins. (A) Superposition of the Rv1738 dimer (blue) on to E. coli YfiA (magenta). Side chains that match between Rv1738 and YfiA and are implicated in ribosome binding by YfiA (34) are indicated. (B) Cartoon and surface representations of the Rv1738 dimer (Left) and YfiA (Right), showing similar belts of surface positive charge (blue). (C) The YfiA binding site on the Thermus thermophilus ribosome. (Inset) Close-up of the Rv1738 dimer superimposed onto YfiA in this site.

Discussion

The work reported here dramatically illustrates the utility of racemic protein crystallography, enabled by modern chemical protein synthesis, for determining the structures of difficult-to-crystallize protein molecules. Racemic protein crystallography provides two major advantages over crystallographic studies on natural (chiral) proteins. The primary advantage is the relative ease with which protein racemates can be crystallized. With a racemic protein mixture, it is possible to access centrosymmetric space groups such as $P\overline{1}$, P2₁/c, and C2/c, which allow interactions between protein molecules that are highly favorable to crystallization (9, 10). Naturally occurring proteins, in contrast, are restricted to the 65 noncentrosymmetric space groups. This more facile crystallization of protein racemates is an advantage vividly demonstrated by the present example, for which exhaustive attempts were made to crystallize the recombinant protein, including many different constructs and the use of several widely used methods for modifying the protein surface. The racemic protein mixture, in contrast, crystallized very readily. Intriguingly, both the crystal forms we chose for optimization and analysis proved to be in C2/c, one of the space groups predicted to be favored, and all the principal crystal packing contacts were between the l- and d-protein dimers.

The second advantage of racemic protein crystallography concerns phase determination and refinement. Crystallization in centrosymmetric space groups results in quantized phases that are usually either 0° or 180°, and it has been predicted that this should make phase determination easier than for noncentrosymmetric data, for which the phases may take general values between 0° and 360° (13). The validity of this prediction may depend on the particular phase determination method used, but our results, supported by test calculations with synthetic data (SI Materials and Methods), appear to bear this out. For the ab initio method used here, starting with a model α-helix of 10–14 residues, a solution emerged much more quickly with centrosymmetric data than with noncentrosymmetric data. Similar advantages apply in phasing centrosymmetric data by isomorphous replacement (35). Structure refinement also converges faster with centrosymmetric data, as the correct phases can be more quickly established early in refinement (36). This is shown clearly by our test calculations with synthetic data (Fig. S3) and arises because given a model that is basically correct, even if incomplete, most of the phases will be exactly right, and refinement can progress in a manner more concerned with fitting the model-derived amplitudes to the experimental amplitudes than with phase improvement. The fragment-based ab initio structure solution described here is a demonstration of the ease of protein structure determination in a centrosymmetric space group.

The structure of Rv1738 revealed here provides an intriguing insight into the molecular events thought to be involved in the onset of dormancy of Mtb. Although its amino acid sequence classifies Rv1738 within a family of proteins of unknown function (DUF1876), restricted largely to mycobacteria, its folded structure identifies it as a member of a much wider family of proteins known to be associated with ribosomes; that is, the S30Ae proteins, PFAM family PF02482. Members of this family are also referred to variously as cold shock proteins or HPFs. Both the E. coli cold shock protein YfiA and E. coli HPF have been shown crystallographically to bind to a conserved site on 70S ribosomes, occupying a deep groove in the 30S subunit (33, 34). There they overlap the binding sites for tRNA, mRNA, and several initiation factors, thereby inhibiting translation and protein synthesis. This inhibition occurs under stress (cold shock) and is rapidly reversed at normal growth temperatures.

Dormancy in Mtb is known to be associated with a shutdown of protein synthesis and to be readily reversible by heat shock (4). It is known that entry to the dormant state in Mtb, and other mycobacteria, is triggered by hypoxia and controlled by the DosR response regulator, which controls the expression of some 48 genes in Mtb (6, 37). A study in Mycobacterium smegmatis has implicated two S30AE proteins from the dosR regulon as key mediators of ribosome stability during hypoxic stasis (38). Both have counterparts in Mtb.

The most highly up-regulated dosR gene under hypoxic stress is Rv1738 (6). Our structural analysis provides a compelling explanation for the up-regulation of this gene by revealing the striking similarity between Rv1738 and the ribosome-associated stationary-phase proteins HPF and YfiA. The latter are monomers, whereas Rv1738 is a dimer, but all have a belt of positive charge in common and most of the secondary structural elements match spatially, giving a similar size and shape. Modeling the Rv1738 dimer into the groove where HPF and YfiA bind in the bacterial 70S ribosome (34), using either HPF or YfiA as a template, shows that it fits extremely well (Fig. 4C). The major difference between them, divergence of one Rv1738 helix from helix α1 of YfiA, is well accommodated. The Rv1738 dimer also has basic side chains that match spatially with five of the six basic side chains by which YfiA makes key contacts with the ribosome (Fig. 4A).

By analogy with YfiA and HPF, whose effects on ribosome structure and function differ in detail while sharing a common binding site and shared ability to inhibit translation and protein synthesis (34), we propose that Rv1738 acts, very likely with other protein factors, to induce dormancy under conditions of low-oxygen stress by interaction with ribosomes.

Materials and Methods

Recombinant Protein Production and Modification.

Recombinant Rv1738 proteins were expressed in E. coli and purified by immobilized metal affinity chromatography and size exclusion chromatography, as described in SI Materials and Methods. The N-terminal His₆-tag could be readily removed by incubation with recombinant tobacco etch virus (rTEV) protease. Reductive methylation of lysine residues was carried out with dimethylamine–borane complex and formaldehyde (16), and the modified protein was purified by gel filtration. Mass spectral analysis showed the addition of eight methyl groups (mass increase, 112 Da), implying that all four lysines had been di-methylated. Potential sites for surface entropy reduction were identified with the surface entropy reduction prediction (SERp) server, and eight mutant Rv1738 proteins were made and expressed. Three mutants could not be purified because of aggregation, but the other five were successfully purified as for wild-type Rv1738.

Total Synthesis of d-Rv138 and l-Rv1738 Proteins.

Rv1738 peptides 1–29 thioester, Thz30-65 thioester, and Cys66-94 were prepared as in Fig. S1. Ligation reactions used previously published conditions: 200 mM sodium phosphate buffer containing 6 M guanidine hydrochloride, 20 mM tris(2-carboxyethyl)phosphine⋅HCl (TCEP) at pH 6.8, 2–4 mM of each peptide, 30 mM 4-mercaptophenylacetic acid, purged and sealed under argon. The one-pot sequential ligations were carried out on a 17.55-μmole scale for each of the three peptide segments. The Trp(CHO) formyl side chain-protecting groups were removed by adding equal volumes of piperidine and β-mercaptoethanol. Purification yield was 5.6 μmole (60 mg, 32% yield) of full-length polypeptide. Selective desulfurization to restore the native Ala residues at the ligation sites in the synthetic polypeptide was carried out as described in SI Materials and Methods. A total yield of 15 mg of each synthetic polypeptide was isolated and used for crystallization.

Crystallization.

Crystals of the d,l-Rv1738 racemic mixture were grown at 19 °C in hanging drops made up with 1 μL protein solution and 1 μL reservoir solution. The protein solution comprised 9 mg/mL of each Rv1738 enantiomer (18 mg/mL total protein) in water. Form-I crystals were obtained using 0.1 M Bis⋅Tris at pH 5.5, 0.2 M NaCl, 25% (wt/vol) PEG 3350 as precipitant, and Form-II crystals with 0.1 M Bis⋅Tris at pH 5.5, 25% (wt/vol) PEG 3350 (no NaCl). Full details of the screening procedures for both recombinant and racemic Rv1738 proteins are detailed in SI Materials and Methods.

Structure Determination.

Diffraction data were collected from both Form-I and Form-II crystals at the Advanced Photon Source, as detailed in SI Materials and Methods and Table S1. An initial model for the Rv1738 structure was obtained using Form-II data to 1.9 Å with the ab initio approach described in the Results section and in SI Materials and Methods. After reprocessing the Form-II data to higher resolution, the model was refined using data to 1.5 Å resolution. The Form-I crystal structure was then determined by molecular replacement with PHASER (20), using the Form-II structure as a search model (see Results section), and was refined using data to 1.65 Å resolution. Full refinement details are provided in Table S1.