Mycobacterium tuberculosis Chaperonin 10 Heptamers Self-Associate through Their Biologically Active Loops

Abstract

The crystal structure of Mycobacterium tuberculosis chaperonin 10 (cpn10_Mt) has been determined to a resolution of 2.8 Å. Two dome-shaped cpn10_Mt heptamers complex through loops at their bases to form a tetradecamer with 72 symmetry and a spherical cage-like structure. The hollow interior enclosed by the tetradecamer is lined with hydrophilic residues and has dimensions of 30 Å perpendicular to and 60 Å along the sevenfold axis. Tetradecameric cpn10_Mt has also been observed in solution by dynamic light scattering. Through its base loop sequence cpn10_Mt is known to be the agent in the bacterium responsible for bone resorption and for the contribution towards its strong T-cell immunogenicity. Superimposition of the cpn10_Mt sequences 26 to 32 and 66 to 72 and E. coli GroES 25 to 31 associated with bone resorption activity shows them to have similar conformations and structural features, suggesting that there may be a common receptor for the bone resorption sequences. The base loops of cpn10s in general also attach to the corresponding chaperonin 60 (cpn60) to enclose unfolded protein and to facilitate its correct folding in vivo. Electron density corresponding to a partially disordered protein subunit appears encapsulated within the interior dome cavity of each heptamer. This suggests that the binding of substrates to cpn10 is possible in the absence of cpn60.

The 10-kDa chaperonin of Mycobacterium tuberculosis (cpn10_Mt) (3) is strongly implicated as an important virulence factor during infection (10, 28). Investigation of the structural and functional characteristics of the protein is therefore an imperative in the search for routes to combat this persistent human pathogen. While the fundamental role of chaperonins as facilitators of protein folding within the cell is well established (13, 41), the secretion of cpn10_Mt into the extracellular space (1, 4) and its pathogenic properties are poorly understood.

cpn10 (GroES) acts in concert with the 60-kDa chaperonin (cpn60), GroEL, ATP, and a repertoire of other molecular chaperones to catalyze protein folding (39, 47). The genome for M. tuberculosis contains the two GroEL homologues cpn60.1_Mt and cpn60.2_Mt (17) that presumably function in concert with cpn10_Mt to enable protein folding in the mycobacterium to occur and to contribute to the stress response of the organism that resides in the hostile interior of the macrophage. Mitochondrial cpn60 is also located on the surface of cells, suggesting that it could also serve as a receptor for extracellular cpn10 (42).

cpn10_Mt has a potent bone-resorbing activity by promoting the recruitment of osteoclasts and inhibiting the growth of osteoblasts (29). This is believed to be a central feature of the pathology of spinal tuberculosis, one of the more severe diseases associated with M. tuberculosis infection, and represents a cell signaling function of the secreted protein. Healthy bone is maintained by a dynamic equilibrium between the bone matrix-forming osteoblast cells and the bone-resorbing osteoclast cells (32). Promotion of osteoclast and inhibition of osteoblast cell proliferation will therefore cause bone resorption, resulting in collapse of the vertebrae. Bone serves as a potential source of nutrients, such as iron (35), and of macrophages, where the bacterium primarily resides in humans (38). By using synthetic peptide fragments, the sequences causing bone resorption in cpn10_Mt were identified to include residues 26 to 35 in the base loop and residues 65 to 70 (29).

We have previously reported the cpn10_Mt crystallization conditions, data collection, molecular replacement (MR) solution with Escherichia coli GroES, and initial rigid body refinement (37). The MR model had all loop extensions and nonidentical side chains omitted. At that stage the appearance of new electron density clearly showed that two heptamers are linked as a tetradecamer by interleaving base loop interactions that extend between them (37). More recently, a lower-resolution (3.5 Å) model for the cpn10_Mt structure was reported by Taneja and Mande (43, 44) that is in disagreement with our model. Electron density at the ends of the base loops were absent, making their loop connectivity ambiguous. Here we describe our completed cpn10_Mt structure and compare it with the structure reported by Taneja and Mande. We conclude here that Taneja and Mande have been investigating the same molecular assembly, which they have interpreted differently.

Our X-ray structure of cpn10_Mt shows the determinants of bone resorption to share similar structural features, suggesting a common receptor. In addition, we propose a role for the cpn10 tetradecamer in encapsulating substrate that may be important as part of the folding process with cpn60.

MATERIALS AND METHODS

Model building and refinement.

An overall B value of 30 Ų and no σ cutoff was applied to the 30-2.8 Å data used in the initial rigid body refinement of the MR model, giving a starting R value of 0.52 (37). From this initial model, noncrystallographic symmetry (NCS) averaging of the 14 subunits and phase extension from 8.0 to 2.8 Å was carried out with the program DM in the CCP4 suite (7). This enabled model building into the remaining electron density missing from the MR model by using the program SwissPdbViewer (12). The cpn10_Mt model was further refined by simulated annealing (2), torsion angle (36), and positional refinement by using strict NCS constraints on all 14 subunits and an overall B value of 30.0 Ų, with the programs X-PLOR (5) and CNS (6) used in between rounds of further NCS averaging, phase extension, and model building. At this stage, the cpn10_Mt model had an overall B value of 53 Ų and an R value of 0.39. When the cpn10_Mt model was defined as far as possible by strict NCS, the NCS constraints were relaxed to NCS restraints on the β-strands in the β-barrel domains during refinement. Subsequent electron density maps visualized additional details and differences in the individual subunits that were hidden by the NCS averaging. An overall anisotropic B value and bulk solvent correction was applied. Further, model building, and NCS-restrained positional refinement and B-refinement were carried out. Other B-refinement protocols were tried, but restrained individual B-refinement was found to be the best option, giving the lowest R values and the most accurate indication of the true B values. The stage at which restrained individual B-refinement was applied to the model with an overall B value of 54 Ų resulted in a drop from R_free of 0.317 and R_work of 0.303 to R_free of 0.284 and R_work of 0.263. Furthermore, the resulting average B value gives the closest agreement to that estimated from the Wilson plot (Table 1). To avoid NCS bias, a 5% R_free test set was selected from a series of thin-resolution shells by using DATAMAN (16). All data from 30-2.8 Å were used in the refinement, calculation of electron density maps, and solvent addition to give an R_free value of 0.280 and an R_work value of 0.259 for the final model (Table 1). The solvent molecules of 2-methyl-2,4-pentanediol (MPD) were added as R-enantiomers. In the S-enantiomeric form the MPDs resulted in a slightly higher R_free value (0.282). The geometry for the final model is better than that for an average protein structure determined at this resolution (Table 1). Buried molecular surface calculations used the algorithm of Lee and Richards (25) in CNS with a probe radius of 1.4 Å. Figures were prepared by using SwissPdbViewer. Simulated annealing omit maps were calculated on models with all solvent molecules and the residues in question omitted prior to torsion angle refinement with slow cooling from 1,000 K in steps of 10 K.

TABLE 1.

Data collection and refinement statistics

Data collection

Refinement

Unit cell parameters at 100 K, 76.5 by 87.9 by 124.4 Å (90°, 106.8°, 90°); resolution, 30.0-2.8 Å; measurements, 134,674; no. of unique reflections, 38,321 multiplicity, 3.5 (3.5)c; completeness, 98.0% (99.7%)c; Rsyma, 5.0% (61.1%)c; <I>/<sig(I)>, 34.4 (1.6)c

Target, maximum likelihood function; anisotropic B-correction, applied to 6.0-2.8 Å data; bulk solvent, correction applied with overall density level of 0.26 eÅ−3; no. of reflections (working/test), 34,993/1,900 with Fobs > 0; RworkbRfree, 25.9%/28.0%; total no. of atoms: 10,690; waters, 64; MPD molecules, 18; calcium ions, 1; bond length deviation (rms), 0.011 Å; bond angle deviation (rms), 1.4°; dihedral angle deviation (rms), 26.1°; improper angle deviation (rms), 1.0°; avg B values for: main chain, 74 Å2; side chain, 77 Å2; MPD, 89 Å2; waters, 72 Å2; calcium, 73 Å2; whole model, 76 Å2; overall B value from Wilson plot: 80 Å2; Brmsds for: main chain bonds, 3.0 Å2; side chain bonds, 3.6 Å2; main chain angles, 4.8 Å2; side chain angles, 5.2 Å2; PROCHECK (24) data: 90.2 % in core, 8.6 % in allowed, 1.2 % in generous (Glu 9), and none in disallowed regions of the Ramachandran plot; overall G-factor, 0.16; SFCHECK (46) data: Luzzatti coordinate error (26), 0.46 Å; Matthews coefficient (28), 2.67; solvent content, 53.66 %; correlation factor, 0.89

^aR_sym = Σ_hΣ_j|I_hj − <I_h>|/Σ_hΣ_jI_hj, where I_hj is the intensity of observation j of reflection h.

^bR_work = Σ_h||F_o| − |F_c||/Σ_h|F_o|, where F_o and F_c are observed and calculated structure factors, respectively.

^cData in parentheses correspond to the highest resolution shell (2.82 to 2.80 Å).

The <I>/<sig(I)> value of 1.6 in the highest-resolution shell (Table 1) is an indication of the falloff in intensity of the X-ray diffraction data with increasing resolution. This is reflected in the high B value (80 Ų) of the data from the Wilson plot (Table 1), suggesting that the intensity falloff is due to a degree of disorder at the molecular level, such as the loop regions. However, despite the weaker signal-to-noise ratio at higher resolutions, the completeness (99.7%) and multiplicity (3.5) remain high in the outer shell (Table 1). This may be because during data collection at 100 K with cryoprotectant we observed a definite increase in resolution from 3.5 to 2.8 Å and in crystal stability compared to results from room temperature data collection. Under these cryogenic conditions it is known that unit cell shrinkage occurs with the formation of crystalline substructures with more internal order but oriented to each other with increased mosaicity (18). This may explain why the extra data in the outer shells obtained at low temperature has a low signal-to-noise ratio even though it is complete. Analysis of the diffraction spot profiles revealed a mosaicity of 0.1° (full width at half maximum). The inclusion of all data to 2.8 Å was justified on the basis of the quality of the resulting electron density map, which showed more details in the loop regions and side chains that were not in the initial MR model compared to that of a phase extension performed with the same data truncated to 3.5 Å. The relevant diffraction data and electron density maps are available on request.

Dynamic light scattering (DLS).

Protein concentrations were estimated from the measured absorbance at the λ_max of 278 nm, and the extinction coefficient was calculated from the amino acid sequence at www.expasy.org together with absorbance measurements at 750 nm by using reagents from the Bio-Rad D_c Protein Assay. Solutions of purified GroES and cpn10_Mt were filtered through a 0.1 μm-pore-size membrane and then were measured at 21°C in a 12-μl cell with the DynaPro-801 MSTC Molecular Sizing Instrument (Protein Solutions, Ltd., High Wycombe, Buckinghamshire, United Kingdom). The recombinant GroES was supplied by Sigma as a lyophilized powder. The recombinant cpn10_Mt was expressed as described before (37) but was purified by anion exchange chromatography following heat treatment of the cell lysate (9). A laser wavelength of 8,288 Å was used, and the scattered light was measured at an angle of 90° to the incident radiation. Volume shape hydration molecular weight models of GroES and cpn10_Mt were calculated from their frictional ratios (ƒ/ƒ_o) and partial specific volumes (ν), where ƒ is the frictional coefficient of the protein (ƒ = RT/D_TN_A), ƒ_o is the frictional coefficient of a rigid sphere with a volume identical to that of the protein [ƒ_o = 6πη(3Mν/4πN_A)^1/3], v was calculated from the amino acid composition of the protein (34), R is the gas constant, T is the absolute temperature, D_T is the translational diffusion coefficient, N_A is Avogadro's number, η is the viscosity of the solvent (0.01 poise), and M is the molecular weight of the protein (40). The program HYDROPRO (11) was used to calculate D_T for cpn10_Mt and GroES from their crystal structures by using a 2.9-Å hydration layer as suggested by Perkins (34). For both proteins the calculated hydrodynamic radii (R_H) were derived from the Stokes-Einstein equation (R_H = k_BT × 10¹⁷/6πηD_T), where k_B is Boltzmann's constant. The fluctuations in the intensity of the scattered light from the protein molecules in solution were used to estimate D_T by an autocorrelation function from which the experimental values of R_H were subsequently obtained. These calculations and the molecular weight calculations based on the observed R_H values and volume shape hydration models were processed with the DYNAMICS software supplied with the instrument in monomodal and bimodal analysis. Every R_H value has an associated polydispersity (C_P), or range of possible R_H values, and this was used to estimate the range of possible oligomers in association with the predominating oligomer defined by R_H in Table 2. Measurements with a baseline error outside the range 0.95 to 1.05 were excluded.

TABLE 2.

Estimation of molecular size in solution by DLS^a

Protein and bufferb	Total mass (%)	RH (nm)	CP/RH (%)	RH range (nm)c	Subunits per oligomer	Predominating subunits (oligomer range)	Baseline error
cpn10Mt
A	99.99	3.70	10.4	3.51-3.89	8.0,d 5.9e	7	1.012
B	95.9, 4.1	4.31, 36.77	29.8, 54.1	3.67-4.95	13.3f	14 (7-14+)	1.001
GroES
C	99.9	3.70	10.2	3.51-3.89	8.0,d 5.9e	7	1.003
D	99.7, 0.3	4.27, 29.10	24.8, 21.2	3.74-4.80	12.9f	14 (7-14+)	1.003

^aVolume shape hydration (VSH) models with hydrodynamic radii (R_H) calculated from crystal structure data. R_H for cpn10_Mt heptamer without base loops, 3.54 nm; R_H for cpn10_Mt heptamer, 3.92 nm; R_H for cpn10_Mt tetradecamer, 4.39 nm; R_H for GroES heptamer without base loops, 3.49 nm.

^bBuffers: A, 2.3 mg of cpn10_Mt per ml in 100 mM TrisHCl (pH 6.8), 50 mM KCl; B, 2.65 mg of cpn10_Mt per ml in 40 mM sodium acetate (pH 5.4), 8 mM CaCl₂; C, 0.56 mg of GroES per ml in 40 mM sodium acetate (pH 7.4); D, 0.5 mg of GroES per ml in 36 mM sodium acetate (pH 7.4), 20 mM CaCl₂.

^cEstimated from polydispersity.

^dCorresponding to VSH model a.

^eCorresponding to VSH model b.

^fCorresponding to VSH model c.

RESULTS AND DISCUSSION

Overall description of the structure.

Our first cpn10_Mt model was initially deposited in the Protein Data Bank (PDB) in June 2001 under code 1JH2. The cpn10_Mt model described in this paper has now been greatly improved in terms of its geometry, Ramachandran plot, B values, and B_rmsds (Table 1), although the overall protein fold remains the same (root mean square [rms] deviations of the models were 0.6 Å for 1,365 out of 1,379 Cα positions following least-squares superposition) and is deposited in the PDB under the new accession code 1P3H. The construction of our model was not aided at all by the conflicting cpn10_Mt model described by Taneja and Mande (43, 44). The latter was referred to only at the final stage for comparative purposes.

The crystal structures of three other cpn10s are known (14, 15, 27, 49) with subunit folds similar to those of cpn10_Mt heptameric assemblies. E. coli GroES has 44% sequence identity with cpn10_Mt, and 65 Cα positions per subunit superimpose on GroES (15, 49) as a heptamer, with rms deviations of 1.3 Å (Fig. 1 and 2) and with unmatched residues at 1 to 2, 18 to 37, 51 to 56, 79 to 83, and 99. While the base loops of uncomplexed GroES are disordered (15), the roof loops 51 to 56 are raised and the turn 79 to 83 is lowered with respect to GroES (Fig. 1 and 2). Mycobacterium leprae cpn10 has 90% sequence identity with cpn10_Mt. cpn10_Mt is a tetradecamer with a spherical cage-like structure composed of two copies of the dome-shaped heptamer, each with a hydrophilic inner cavity and loops extending outwards from the base (Fig. 3). These base loops are referred to as mobile loops in cpn10 structures where they are disordered (15, 27). In the M. leprae cpn10 (27) and T4 phage Gp31 (14) structures the base loops also make contact with their symmetry-related heptamers, although the interheptamer contacts are not as extensive as those in cpn10_Mt. Comparison of the symmetry-related heptamers of Gp31 and cpn10_Mt shows each tetradecameric assembly to have similar molecular dimensions, although their polypeptide chains do not superimpose very well. However, despite the fact that there is only 14% sequence identity between the two chaperonins, least-squares superposition of the individual Gp31 and cpn10_Mt heptamers gives an rms deviation of 1.6 Å for 61 residues per subunit (Fig. 4); i.e., at residues 4 to 17, 35 to 48, 61 to 73, 75 to 77, and 82 to 97. The Gp31 roof loops are absent and the base loops are oriented differently, but the β-barrel domains superimpose quite well (Fig. 4). Like Gp31, the two cpn10_Mt heptamers interact through their base loops, which become visible as a result of their immobilization on binding (Fig. 3A). The cpn10_Mt base loops have a significant area of contact with the subunits of the opposite heptamer that is slightly less than that of the subunit interactions within the heptamer. Both heptamers are positioned on the same sevenfold axis and are related by twofold NCS as the tetradecamer that defines the asymmetric unit of the cpn10_Mt crystal structure (37). This means seven NCS twofold axes run perpendicular to the sevenfold axis in between pairs of base loops, relating subunit pairs in opposite heptamers. There is a staggered arrangement of heptamers about the sevenfold axis, with a half-subunit rotation of heptamer 1 (subunits A to G) with respect to heptamer 2 (subunits H to N) (Fig. 3B). The core domain of each subunit is a β-barrel of two orthogonal antiparallel β-sheets (Fig. 5A). The β-sheet interactions on the β-barrels extend across the subunit interfaces of the heptamer between the N- and C-terminal β-strands of neighboring subunits in a fashion similar to that of other cpn10 structures (14, 15, 27) (Fig. 1). Each heptamer has a narrow oculus at the dome roof surrounded by a cluster of acidic residues (Fig. 6) which are held together by chelated calcium ions and protonated side chains at the acidic pH of the crystal (5.4). Compared to GroES, cpn10_Mt has two additional acidic residues on each loop forming the dome roof at Asp 51 and Asp 53 (Fig. 5 and 6). cpn10_Mt is therefore more favorable for calcium binding in this region (Fig. 2).

FIG. 1.

Superposition of Cα backbones of cpn10_Mt heptamer (subunits A to G in red) and GroES (blue) structures with residues 18 to 34 (cpn10_Mt) and 16 to 32 (GroES) deleted. Cα atoms outside the rms fit of 1.3 Ų are in green. Stereo views are from the side (A) and top (B).

FIG. 2.

Least-squares superposition of cpn10_Mt subunit B (red) on one GroES subunit complexed with GroEL (blue) and including the calcium binding site of cpn10_Mt on the roof loop at the top of the figure.

FIG. 3.

The cpn10_Mt tetradecamer. (A) The tetradecamer viewed along the twofold NCS axis. (B) Sketch of the staggered arrangement of subunits viewed down the sevenfold NCS axis. (C) Electrostatic potential from +2.5 kT/e (blue) to −7.5 kT/e (red) of the molecular surface of subunits D, M, G, and J. The internal diameter at the rim of the heptamer base (21 Å) is marked by Tyr 73 side chains and at the midsection (30 Å) by Glu 35 side chains. Viewed along the twofold NCS axis.

FIG. 4.

Least-squares superposition of heptamers of cpn10_Mt (red) and Gp31 (blue), shown as a Cα plot. Stereo views are from the side (A) and top (B).

FIG. 5.

Structure of the cpn10_Mt subunit. (A) Subunit B facing the intraheptamer interface showing the amino acid side chains lining the interior cavity. (B) Primary and secondary structure of cpn10_Mt with turn types (48) are indicated in Greek letters underneath the sequence. The 333 stretch from 90 to 92 indicates a 3₁₀-helix. Sequences active in bone resorption are shaded in black.

FIG. 6.

cpn10_Mt heptamer 1 (subunits A to G) viewed along the sevenfold NCS axis, with a closeup of the roof from the dome interior.

Each of the seven twofold axes in the tetradecamer defines two types of interheptamer subunit interaction, exemplified by subunits B and I, creating an average buried accessible surface area of 751 Ų, and subunits A and I, creating an average buried accessible surface area of 668 Ų (Fig. 7). Therefore, each subunit interacts directly with four neighboring subunits, two in the same heptamer and two in the other heptamer.

FIG. 7.

Contacts between cpn10_Mt subunits at the interheptamer interfaces showing the buried interacting amino acid residues viewed along the twofold NCS axes. For clarity, only interacting side chains are labeled. (Top) Subunits B (black labels) and I (orange labels). Buried accessible surface area/subunit is 750 Ų (average, 751 Ų). (Bottom) Subunits A (black labels) and I (orange labels). Buried accessible surface area/subunit is 700 Ų (average, 668 Ų).

On average, each subunit has a total surface area of 6,969 Ų and a total buried accessible surface area of 2,884 Ų, 1,464 Ų of which is buried by subunit interactions within the same heptamer and 1,420 Ų of which is buried by subunit interactions with the neighboring heptamer. This indicates that the extent of bonding interactions between subunits within heptamers is slightly greater than subunit interactions across heptamers, an average of 41% of the total surface area of each subunit being buried.

The Ramachandran plot shows no residues in the disallowed regions. Only one residue, Glu 9, occurs in a generous region, possibly because it is situated at a tight 2-residue turn (Fig. 5B). This turn shortening appears to be associated with thermostable proteins (45), which would be a required property of cpn10s as heat shock proteins. Other glutamates are also found in this location of the Ramachandran plot (8).

Each cpn10_Mt subunit has an associated MPD molecule bound at the opening to the hydrophobic interior of the β-barrel and in contact with the main chain from Val 3 to Ile 5 and the Tyr 85 side chain. Therefore, 14 out of the 18 MPDs are related by the NCS.

The base loops.

The base loop (amino acids 17 to 33) extends out of the β-barrel domain to interact with subunits in the neighboring heptamer (Fig. 7 and 8A). The three types of subunit interface connect where the hydrophobic triplet Leu 27-Val 28-Ile 29 in the base loop of heptamer 1 forms a part of the β-sheet with heptamer 2 (or vice versa). This forms a hydrophobic patch made up of contributions from three subunits (Fig. 8A). Val 28 in the base loop of subunit I, for example, forms a β-sheet with Lys 79 of subunit A. This positions Leu 27 of subunit I to interact with Val 3, Ile 5, and Ile 78 of subunit A and Ile 69 and Val 97 of subunit B, reinforcing the hydrophobic interactions between subunits in the heptamer. These interactions are crucial to bind the conserved residues of the base loop sequence (26-35) active in bone resorption (29) into the cleft of the subunit interface of the symmetry-related heptamer. The downward displacement of the residue 79 to 83 turn relative to GroES clearly plays a role in facilitating these interheptamer base loop interactions, resulting in tetradecamer formation, and this seems to be connected to the upward displacement of the cpn10_Mt roof loops (Fig. 1 and 2).

FIG. 8.

Base loop stereo views (A) from cpn10_Mt subunit K (yellow) of heptamer 2 in between subunits F (pink) and G (blue) of heptamer 1. The solvent-exposed exterior (T21-S25) is in the foreground, while the bone-resorbing sequence (G26-P30) is buried behind in the subunit interface by hydrophobic contacts. One of the NCS-related MPD molecules situated at the hydrophobic entrance of each β-barrel is shown in grey. (B) Cα superimposition of base loop regions for cpn10_Mt subunit G (blue) and E. coli GroES complexed with GroEL (red). (C and D) Bone resorption sequences. (C) superimposition of sequence 27 LVIPDT 32 of cpn10_Mt subunit G on sequence 25 IVLTGS 30 of GroEL-bound E. coli GroES. rms fit of all Cα atoms is 0.7 Å. (D) cpn10_Mt sequences 26 to 32 and 66 to 72 of subunit K superimposed.

The base loops are essential for complexation with cpn60 (23, 49). These are mostly disordered in uncomplexed GroES (15) and M. leprae cpn10 (27). This is the least conserved sequence among the cpn10s, and yet the 5-residue segment in the middle of the loop consistently adopts a β-turn structure when it is immobilized in tetradecameric cpn10_Mt, Gp31 (14), or the GroEL-GroES-ADP complex (49). This is corroborated by nuclear magnetic resonance measurements, which show turns in the base loops of GroES and Gp31 as part of a β-hairpin structure when bound to GroEL (22). X-ray crystallographic and nuclear magnetic resonance data indicate that the base loop of human cpn10 also preferentially samples a hairpin conformation (21). However, there is conformational variability in the base loop sequences surrounding the β-turn in all these structures. This is illustrated by the superimposition of subunits from cpn10_Mt and GroEL-bound GroES in Fig. 2, where the base loop can exist in two possible orientations, depending on whether it binds cpn60 or another cpn10. However, superimposition of the base loops from the cpn10_Mt and GroEL-GroES-ADP structures shows that cpn10_Mt 27 to 32 and GroES 25 to 30 have the best least-squares fit, with rms deviations of 0.7 Å between the Cα atoms (Fig. 8B and C). This suggests that although there is a structurally conserved region in the base loops, the flanking residues can serve as hinge regions to allow movement of these loops.

Comparison of two cpn10_Mt models.

A 3.5-Å resolution model of the cpn10_Mt structure has also been reported by Taneja and Mande (43, 44) as a standalone heptamer. There are some similarities in the crystallization conditions that produced these models: sodium acetate buffer with calcium chloride, but at pH 4.0 in their case (43, 44) and pH 5.4 in ours (37). Electron density was absent in their model in all the base loops from resides 17 to 21 and 31 to 36, leaving a set of seven disconnected turns from 22 to 30. Despite the absence of electron density, Taneja and Mande assumed that these disconnected loops were joined to each of the seven subunits in a novel conformation (43, 44). A closer inspection of their model revealed quite a different story. The symmetry operation −x, y, [1/2]-z applied to the coordinates of their model in their chosen space group C222₁ generated a tetradecamer that is essentially identical to ours. The rms deviation of the superimposed Cα coordinates of the two tetradecameric models is 0.97 Å (Fig. 9a and b). An inspection of the disconnected ends of the Taneja and Mande model shows that each of the seven base loops is associated with the wrong crystallographically related heptamers. We have calculated simulated annealing omit maps at the main chain junctions in question, and these show unequivocally that the base loops extend towards the symmetry-related heptamer (Fig. 9c). In other words, the base loops should be in the trans, not cis, orientation. Even more solid evidence for this is seen in the electron density map based on DM phase extensions with the initial rigid-body-refined MR model (37) before any model building was done. Although the MR model had gaps between residues 15 to 35, 48 to 58, and 69 to 73, the connectivity of the base loop is unambiguous in the electron density (Fig. 9d and e). For the Taneja and Mande model to be viable, it would have to undergo a mobile loop flip from trans to cis at their slightly lower pH. However, since the authors have not deposited their structure factors in the PDB, there is no way of verifying this from electron density maps. With the mobile loops retracted as they have proposed (43, 44), there would be little basis for tetradecamer formation in this specific orientation of heptamers.

FIG. 9.

Affirmation of the connectivity of the base loops. (a) Superimposed Cα traces of our cpn10_Mt model (red) and that of Taneja and Mande (43, 44) (white) viewed at the interface between twofold symmetry-related heptamers. The crystallographically related white heptamer is generated by the symmetry operation −x, y, [1/2]-z in their chosen space group C222₁. The rms deviation of all superimposable Cα atoms in all 14 subunits is 0.97 Å. Primed labels mark one base loop related to the unprimed base loop by twofold NCS in our model. In the Taneja and Mande model the base loop connections are assumed to exist between A16-T22′ and P30′-P37 (bottom heptamer) and in the crystallographically related heptamer (top) between A16′-T22 and P30-P37′. (b) Cα plot of one cpn10_Mt subunit from least-squares superposition of the whole 14-mer cpn10_Mt model (red) with that of Taneja and Mande (blue) (43, 44) after regeneration of the second heptamer from crystallographic symmetry and swapping the base loops to their correct positions in the other heptamer. (c) Simulated annealing σ_A weighted |F_o| − |F_c|φ_c omit map of adjacent base loop sections 31 to 36 of subunits B and I contoured at 2.2 σ. Initial electron density maps obtained from phase extension with DM on the initial MR model viewed on subunit I at residues 16 to 22 contoured at 0.4 σ (d) and residues 30 to 37 contoured at 0.8 σ (e).

Despite the evidence in the electron density for the trans-loop configuration, we decided to test out the cis model by deleting the attachments to the base loops at residues 18 to 22 and 31 to 35 and rebuilding these attachments to the alternative cis-loop configuration. The rebuilt sections were minimized and were subjected to slow cool-simulated annealing and B refinement. The newly calculated electron density map showed no density for these rebuilt sections but still indicated the trans-loop density. Furthermore, the R values and Ramachandran plot worsened considerably, with a number of steric clashes. Consequently, the cis-loop model was abandoned.

A comparison of the two 14-mer cpn10_Mt models resulting from their least-squares superposition shows that all residues from 5 to 16, 23 to 30, and 37 to 97 match with rms deviations of 0.97 Å for Cα positions and 1.53 Å for all atoms (Fig. 9a). Our model is shifted towards the neighboring heptamer with respect to the other (43, 44), particularly at the 79 to 83 turn and the base loops, both of which are important for tetradecamer formation. These differences probably arise from a stronger phase bias to the E. coli GroES MR model in their case (43, 44).

We therefore conclude that Taneja and Mande have crystallized the same cpn10_Mt tetradecamer, and there is not any evidence for the existence of mobile loops in a partially stable conformation for a stand-alone seven-subunit structure of cpn10 as they have claimed (43, 44). It should also be stressed that the bone resorption sequences (29) do not form a single conformational unit as previously stated (44) but are located on quite separate parts of the structure. Furthermore, a calcium ion was found coordinated to the acidic side chains of two roof loops in our model, with Glu 55 identified as one of the coordinating side chains (Fig. 2). This contrasts with the binding of a mercury ion to the indole ring of Trp 50 in the other cpn10_Mt model (44). The latter therefore cannot be used as a model for the binding of other divalent cations to cpn10_Mt, such as calcium.

Polypeptide encapsulation.

Curiously, electron density resembling a partially folded polypeptide is located in the interior cavity of each heptamer (Fig. 10). The density has a tubular appearance rather than discrete spheres typical of solvent, with bends separated by 3.8 Å in places. While this satisfies the twofold NCS, it results in a symmetry mismatch with the sevenfold NCS. The level of disorder in this molecule is too high to allow an unambiguous sequence assignment to the density. Extensive analysis of the purified protein and the contents of the crystallization tray by reversed-phase high-performance liquid chromatography and mass spectroscopy did not reveal any other component in significant amounts (data not shown). At this level of electron density (0.6σ), any heterogeneous protein would have shown up in the analysis. It is therefore highly likely that the encapsulated density corresponds to a partially folded subunit of cpn10_Mt. The distribution of density in the cavity of the tetradecamer shows two separate regions, each about the size of the cpn10_Mt β-barrel.

FIG. 10.

A 10-Å cross-section through the internal cavity of the cpn10_Mt tetradecamer and its crystal packing contacts of σ_A weighted 2|F_o| − |F_c|φ_c electron density contoured at 0.6 σ to show the features in the cavity not accounted for by the model.

Solution studies.

The DLS data were processed with the DYNAMICS software through different volume shape hydration (VSH) models (Table 2) to get a best fit for the number of subunits per oligomer. The base loops in the cpn10 heptamer are usually mobile and disordered when not bound to neighboring cpn10 or cpn60 oligomers. To simulate this, VSH models were also calculated for cpn10_Mt and GroES without the base loops (residues 17 to 33 and 16 to 32, respectively). At neutral pH and without calcium (buffers A and C), the R_H values for both cpn10_Mt and GroES (3.7 nm) are in between those calculated for the cpn10 heptamer with base loops (3.9 nm) and without (3.5 nm). Consequently, the base loops may be partially disordered. The polydispersity range of 3.5 to 3.9 nm suggests that different states of base loop disorder exist. Therefore, the total number of subunits per oligomer is calculated as 6 or 8 depending on which VSH model is used. The best estimate that can reasonably be given is approximately 7 subunits per oligomer. DLS studies on GroES were done recently in another laboratory (47) which showed an R_H value of 3.9 nm, in agreement with our calculated value based on the heptamer with extended base loops.

The presence of calcium in the buffer (solutions B and D) causes R_H to increase to 4.3 nm, close to the tetradecamer VSH model (4.4 nm), with an increase in polydispersity and a small proportion of higher-molecular-weight aggregate. The polydispersity range (3.7 to 4.9 nm) suggests that heptamers are in equilibrium with tetradecamers and that low levels of higher oligomers are also present. As the polydispersity has a bell-shaped distribution, the tetradecamer would be the predominating oligomer. Molecular weight calculations in solutions B (cpn10_Mt crystallization buffer) and D with the tetradecameric VSH model give about 13 subunits per oligomer, slightly less than that for a tetradecamer. This may be due to mobility in the roof loops or a smaller than expected hydration layer.

The role of metal ions in cpn10 tetradecamer formation.

We have shown that calcium binds to the acidic side chains of the roof loops in cpn10_Mt and is coordinated to Glu 55. Interestingly, the Glu 55 site in cpn10_Mt shows the largest displacement relative to GroES (Fig. 1 and 2) that appears to affect the positioning of the 79 to 83 turn that interacts with the trans-base loops resulting in tetradecamer formation. In addition to revealing the existence of tetradecameric cpn10_Mt in the crystal structure, the states of oligomerization of both cpn10_Mt and GroES in solution were investigated by DLS. This revealed an increase in molecular size consistent with the assembly of both cpn10_Mt and GroES heptamers into tetradecamers on the addition of calcium (Table 2). Substitution of calcium for magnesium in the crystallization buffer produces cpn10_Mt crystals that are isomorphous with the unit cell parameters reported in Table 1 (a = 77.02 Å; b = 89.95 Å; c = 125.98 Å; β = 105.63°). Crystal growth was not possible in our buffer conditions without either divalent cation, so these make it possible for tetradecameric assembly and crystal packing to occur. There was no evidence in the crystal structure for the binding of calcium in the vicinity of the base loops. We therefore have to consider the possibility that calcium coordination by the roof loops draws the subunits upwards in that location, resulting in subunit rotation to favor the mobile loop interactions required for tetradecamer formation. GroES Glu 53 matches Glu 55 in the cpn10_Mt sequence, so this could be a crucial residue for calcium binding in both proteins.

A careful analysis of the molecular packing in M. leprae cpn10 (27) and Gp31 (14) shows that the base loops of one heptamer interact with the subunit interfaces of the symmetry-related heptamer, although their association as tetradecamers has not been demonstrated in solution.

The coexistence of the cpn10 heptamer and tetradecamer in solution (Table 2) together with the slightly smaller proportion of buried accessible surface area between subunits across heptamers in the cpn10_Mt crystal structure indicates that the bonding interactions between heptamers are weaker than those between subunits within heptamers. This would ultimately need to be confirmed with a more detailed thermodynamic evaluation of the bonding interactions.

Is there a biological function for the tetradecamer?

The existence of cpn10_Mt and GroES tetradecamers in solution with divalent metal cations (Table 2) and the extensive contacts between the subunit interfaces across neighboring heptamers strongly suggest that they may have biological significance. This is because in the cytosol of the bacterium there is a substantial concentration of divalent metal cations, including calcium, to support the existence of cpn10 tetradecamers. The cpn10 tetradecamers may play a role in the protein folding pathway either by transporting ATP into the cpn60 chamber or, after the release of protein substrate from the cpn10-cpn60 complex, by transporting proteins out of the cpn60_Mt cavity to complete the folding process of intermediates in a protected environment. The possible existence of a polypeptide in the interior of the cpn10_Mt structure (Fig. 10) supports this hypothesis. The detection of both cpn10_Mt and GroES tetradecamers in solution suggests that this oligomeric form may be biologically important in other cpn10s.

Sequences active in bone resorption.

The pathology of bone resorption in tuberculosis patients is determined by the fact that cpn10_Mt is secreted from the bacterium and not E. coli GroES (33). The secretion of cpn10_Mt into the extracellular space therefore appears to be a crucial component of its pathogenic potential. The possible rationale behind this is that bone serves as a source of nutrients for the bacterium, such as iron (35), and of macrophages, which are the primary location of M. tuberculosis in infected patients (38). The principal sequences that were identified as being active in bone resorption include 26 to 35 and 65 to 70 in cpn10_Mt and the equivalent in GroES, 23 to 33 (29). The only conserved sequence relating peptides cpn10_Mt 26 to 35 and GroES 23 to 33 that might be crucial for bone resorption activity is 26 GLVIPDT 32 in cpn10_Mt. Interestingly, this contains the cpn10_Mt sequence 27 to 32 that has structural homology with GroEL-bound GroES 25 to 30 (Fig. 8B and C). Furthermore, superimposition of the 26 to 32 and 66 to 72 sequences do reveal common structural features (Fig. 8D). This suggests that the cpn10_Mt bone resorption sequences could be structurally conserved when bound to a common receptor and may represent the conformation of the active sequence. Since mitochondrial cpn60 is located on the cell surface (42) and since the early pregnancy factor is located extracellularly as a homologue of cpn10 (30), either could serve as a receptor for cpn10_Mt in bone resorption. This ligand-receptor association is supported by the fact that E. coli GroES and GroEL both stimulate bone resorption, but when they are added together they are mutually inhibitory (31). The bone resorption receptor for cpn10_Mt could also be another protein having structural features in common with those of the binding pockets for the cpn10_Mt 26 to 35 sequence on cpn10 or cpn60.

At this stage it is important to mention that the osteoclast differentiation factor plays a key role in osteoclast proliferation by binding to its receptor RANK (receptor activator of NF-κB) (19). One of the principal sites of receptor binding is centered around the sequence TSIKIPSS (20), which bears some similarity to the cpn10_Mt sequence SGLVIPDT, although at this point one can only speculate on the significance of this similarity as an indicator of osteoclast proliferation by direct binding of cpn10_Mt to RANK.

It has been stated that since cpn10_Mt binds calcium, the direct action of cpn10_Mt on bone should be considered a mechanism for bone resorption (44). This is highly unlikely to be the case. We found calcium binding at the cluster of acidic side chains in the roof loops (Fig. 2 and 6). We have clearly ruled out the roof loop sequence as having bone resorption activity (29). While mercury binding to the Trp 50 side chain has been shown (44), this in no way represents the mode of calcium binding, which has a very different coordination chemistry. Furthermore, we have clearly shown that the bone resorption sequences 26 to 35 and 65 to 70 act indirectly by stimulating the growth of osteoclasts rather than acting directly on bone (29). Neither is there any evidence from our electron density maps that these sequences bind calcium ions.

Conclusions.

We propose that the tetradecameric form of cpn10 could exist in vivo and play a role in the folding of individual structural protein domains and in the transport of proteins, ATP, peptides. This is supported by the observation that each heptamer of the cpn10_Mt tetradecamer encapsulates a polypeptide-like region of electron density that is probably a misfolded cpn10_Mt monomer. The cpn10_Mt sequence 25 to 35, which is active in bone resorption, lies in a protected location embedded in the subunit interface of the cpn10_Mt structure and has some common structural features with other cpn10 bone resorption sequences. The density in the central cavity could be better characterized with higher-resolution data, and we believe there is room for improvement here. First, we found that the resolution of the diffraction data were enhanced by optimizing the cryoprotectant treatment of the cpn10_Mt crystals. Second, the crystal used in this study was grown directly from reversed-phase high-performance liquid chromatography-purified and lyophilized protein (37). It would therefore be interesting to try growing crystals of native-purified cpn10_Mt without lyophilization as well. This may give a better characterization of the protein in the cavity and establish whether cpn10_Mt monomers can coexist with heptamers in solution and become trapped during tetradecamer formation.