Crystal Structure of the Nucleotide Exchange Factor GrpE Bound to the ATPase Domain of the Molecular Chaperone DnaK

See allHide authors and affiliations

Science  18 Apr 1997:
Vol. 276, Issue 5311, pp. 431-435
DOI: 10.1126/science.276.5311.431


The crystal structure of the adenine nucleotide exchange factor GrpE in complex with the adenosine triphosphatase (ATPase) domain ofEscherichia coli DnaK [heat shock protein 70 (Hsp70)] was determined at 2.8 angstrom resolution. A dimer of GrpE binds asymmetrically to a single molecule of DnaK. The structure of the nucleotide-free ATPase domain in complex with GrpE resembles closely that of the nucleotide-bound mammalian Hsp70 homolog, except for an outward rotation of one of the subdomains of the protein. This conformational change is not consistent with tight nucleotide binding. Two long α helices extend away from the GrpE dimer and suggest a role for GrpE in peptide release from DnaK.

Molecular chaperones play an essential role in protein folding by preventing the misfolding and aggregation of folding intermediates (1-3). Several classes of molecular chaperones have been conserved in evolution, including the members of the Hsp70, Hsp90, and Hsp60 (chaperonin) families. Whereas the chaperonins form large oligomeric ring structures, members of the Hsp70 and Hsp90 families function as monomers or dimers.

DnaK, the Escherichia coli homolog of Hsp70, and the various eukaryotic Hsp70s act by binding and releasing extended peptide segments enriched in hydrophobic side chains. DnaK and its homologs are composed of an NH2-terminal 42-kD ATPase domain and a COOH-terminal 25-kD peptide binding domain, the structures of which are known (4, 5). The binding and release of peptides from DnaK is controlled by conformational changes induced by adenosine triphosphate (ATP) binding and hydrolysis in a mechanism that is not understood. In this reaction DnaK does not act alone but cooperates with two other factors, the chaperone DnaJ and the nucleotide exchange factor GrpE (6), in a manner that is analogous to the regulation of many guanosine triphosphate binding proteins.

The following model of the DnaK reaction cycle in protein folding is now emerging (7): ATP-bound DnaK is characterized by rapid peptide binding and release (8). DnaJ stimulates the hydrolysis of ATP by DnaK, resulting in the adenosine diphosphate (ADP)–bound state of DnaK, which binds peptide tightly (6,7). Peptide release then requires the dissociation of ADP, which is catalyzed by GrpE, and ATP rebinding to DnaK then occurs (8,9). Interestingly, whereas the function of DnaK and of mitochondrial Hsp70 critically depends on GrpE, the Hsp70 homologs in the eukaryotic cytosol are GrpE-independent (10-12). For these Hsp70s, ADP dissociation is apparently not a rate-limiting step in the reaction cycle.

We now describe the three-dimensional structure of the ATPase domain ofE. coli DnaK in complex with GrpE. In addition to providing a structural explanation for the ability of GrpE to release nucleotide from DnaK, a striking architectural feature of the structure suggests a role for GrpE in influencing DnaK-peptide interactions.

The E. coli DnaK ATPase domain (residues 1 to 388) and full-length E. coli GrpE (residues 1 to 197) were overexpressed in E. coli. A mutant form of GrpE (G122D, in which Gly at position 122 is mutated to Asp) was used. The mutant confers a temperature-sensitive growth phenotype but is fully functional in mediating DnaK-DnaJ–assisted protein folding in vitro (13). Dimeric GrpE was digested with elastase, with loss of the NH2-terminal 33 residues, and a ∼164-residue proteolytic fragment was purified to homogeneity. Identical proteolysis results were seen in the presence and absence of the DnaK ATPase domain. A stable complex between one DnaK ATPase domain and a dimer of truncated GrpE was formed in the absence of ATP, purified, and crystallized (14).

Native crystals diffract weakly, to ∼3.5 Å on a rotating anode x-ray generator. Crystals derivatized with uranyl acetate diffracted better than native crystals; a single crystal derivatized with uranyl acetate was used for the final refinement after the structure solution by multiple isomorphous replacement (MIR) (Table 1). The final model has been refined at 2.8 Å [R value of 0.223 (F > 2σ) and R free of 0.317 (F > 2σ)]. The final model contains a dimer of GrpE and one DnaK ATPase domain in the asymmetric unit.

Table 1

Crystals (P41, a = b = 150.3 Å, c = 49.1 Å, α = β = γ = 90°) were grown by vapor diffusion of complex (40 mg/ml) and 3.5% PEG 8000, 15% ethylene glycol, 50 mM sodium acetate buffer (pH 4.6), 190 mM lithium sulfate, 5 to 10 mM dithiothreitol, and 10 mM (d,l)-methionine in a nitrogen atmosphere at 4°C. Data were collected at 100 K on an Raxis IIC (Rigaku) and indexed, integrated, and scaled with the DENZO package (26). On the basis of a preliminary model, three residues in DnaK (Arg235, Asp289, and Asp385) were mutated to cysteine. Derivatization of isomorphous crystals with ethyl-mercury phosphate (EMP) resulted in heavy atom substitution at each of the new cysteine sites as well as the native Cys17 for R235C and D289C DnaK. MIR phases were calculated to 3.6 Å with MLPHARE (13,042 reflections, figure of merit, 0.56) and solvent flattened with DPHASES in the GVX program suite (27). The model was built into the electron density maps by using O (28) and MAIN (29). The model was refined with X-PLOR (30). Experimentally determined seleno-methionine peaks confirmed the topology of the GrpE molecule (14). The final model includes residues 3 to 383 of DnaK (except residues 184 and 210 to 213, which are disordered) and residues 34 to 197 (except residues 108 to 115) for the GrpE monomer that is proximal to DnaK in the complex, and residues 38 to 195 (except residues 108 to 115) for the GrpE monomer that is distal to DnaK, 28 water molecules, and 8 uranyl acetate ions, modeled as single sites. The average B factor for the protein model is 54 Å2, the B factor estimated from the Wilson plot is 59 Å2, and the averageB factor for the waters is 42.5 Å2; n/a, not applicable.

View this table:

GrpE is a tightly associated homodimer that binds DnaK with 2:1 stoichiometry (15, 16). The overall identity between GrpE sequences ranges from 25 to 70%, and the conservation of buried side chains suggests that the structure (Fig.1) is likely to be conserved. The GrpE dimer interface encompasses two long, paired α helices that lead into a small four-helix bundle to which each monomer contributes two helices (Fig.2A). The two long helices and the four-helix bundle bury 2600 and 2100 Å2 of total solvent-accessible surface area, respectively, calculated with a probe radius of 1.4 Å. The loop connecting the long and short α helices is partially disordered in one of the GrpE monomers and totally disordered in the other. On the basis of weak electron density for one monomer, we favor a model without crossover of protein chains at the base of the four-helix bundle.

Figure 1

Secondary structure alignment ofE. coli GrpE with the mitochondrial GrpEs ofSaccharomyces cerevisiae and Drosophila melanogaster (GenBank numbers X07863, X78350, and U34903, respectively). Elements of β strands are drawn as arrows and α helices as cylinders. “Proximal” refers to the GrpE monomer that is closest to DnaK and makes the most contacts, and “distal” to the GrpE monomer farthest away from DnaK. Filled, half-filled, open, and dotted circles indicate residues with <20%, 20 to 40%, and >40% solvent-accessible surfaces and residues that were not modeled, respectively. Invariant residues in 20 GrpE sequences in the database (31) are shaded; some of the highly conserved residues are boxed; residues that contact DnaK have an asterisk; and invariant or highly conserved residues involved in intramolecular contacts have a diamond. Known mutations of GrpE or yeast mitochondrial homolog Mge1p are shown italicized and underlined (19-21). The G122D mutant used in this study is indicated. Lower-case letters indicate insertions (not shown) in a sequence relative to E. coli. Abbreviations for the amino acid residues are as follows: A, Ala; C, Cys; D, Asp; E, Glu; F, Phe; G, Gly; H, His; I, Ile; K, Lys; L, Leu; M, Met; N, Asn; P, Pro; Q, Gln; R, Arg; S, Ser; T, Thr; V, Val; W, Trp; and Y, Tyr.

Figure 2

(A) Ribbon drawing of the complex of GrpE and the ATPase domain of DnaK. Residue 122 (Gly mutated to Asp) and the loop connecting the long helix to the short helix in the distal monomer are shown. The loop is disordered in the proximal monomer. This figure was made with RIBBONS (32). (B) Overlap of DnaK and Hsc70 ATPase domains. The nucleotide of Hsc70 is in yellow. This figure, (C), and Fig. 3B were made with GRASP (33). (C) GrpE and DnaK are drawn as molecular surfaces and are colored according to area of contacts. Gly32 of DnaK, part of the loop that was described as necessary for GrpE interactions (34), is in region IV. C, COOH-terminal; N, NH2-terminal. Exact details of the contacts are available at

Whereas the topology of the polypeptide fold is similar in each monomer, the structure of the dimer is strikingly asymmetric, with the GrpE dimer being curved toward DnaK. The two long helices in the dimer do not form a canonical coiled-coil but instead lie nearly in the same plane and are translated slightly with respect to each other along the helix axis. The canonical i+3, i+4 heptad repeat of hydrophobic residues is not adhered to in GrpE. Instead there are two missteps [or stutters (17)] in the heptad repeat (i+3, i+4, i+3, i+4, i+4, i+3, i+4, i+3, i+4, i+4) that cause overwinding of the helices in some regions and a loss of superhelicity.

The GrpE monomer proximal to DnaK contributes the bulk of the residues at the DnaK interface. Starting at the NH2-termini of the long helices, the helix of the distal monomer continues unbroken for 19 turns, that is, ∼100 Å. In contrast, the helix of the proximal GrpE loses helicity between Phe86 and Leu88. Helicity of the proximal long helix is resumed at the start of the four-helix bundle. Two small β sheet domains emanate from the two COOH-terminal ends of the four-helix bundle of GrpE. The compact β domains are ∼60 residues in length, have six short β strands, and a limited hydrophobic core.

The ATPase domain actually consists of two large domains, each of which is composed of two subdomains (4). Subdomains IA and IIA, which lie at the base of the deep nucleotide binding cleft, are similar in topology and are related by pseudo twofold symmetry. Subdomains IB and IIB, which make up the sides of the nucleotide binding site, represent insertions in IA and IIA, respectively, and are not topologically related.

The ATPase domain from bovine brain heat shock cognate 70 (Hsc70) is 55% identical in sequence to that of DnaK. Insertions and deletions in DnaK relative to Hsc70 are all accommodated in surface-exposed loops (DnaK residues 43 to 47, 105 to 111, and 288 to 291), and none of these results in a changed topology or has any obvious consequence for GrpE interaction. We compared the structure of GrpE-bound, nucleotide-free DnaK with that of nucleotide-bound Hsc70. The close similarity in the structures of the nucleotide-bound forms of Hsc70 and actin (4), which share only ∼10% sequence identity, suggests that the structure of Hsc70 is a good model for the much more closely related DnaK.

GrpE induces a 14° rotation of domain IIB in DnaK relative to its position in the Hsc70 structure. In spite of this marked rotation, domain IIB itself is quite similar to the equivalent domain in the Hsc70 homolog [0.8 Å root-mean-square deviation (rmsd) for 79 Cα positions after superimposition]. The remainder of DnaK superimposes very well with Hsc70 and shows negligible changes because of GrpE binding (0.7 Å rmsd for 216 equivalent Cα positions) (Fig.2B).

There are six areas of contact between DnaK and GrpE (Fig. 2C). The two largest are between the two faces of the proximal β sheet domain of GrpE and domains IB and IIB of DnaK, on each side of the nucleotide binding cleft. The other areas are on separate regions of the GrpE long helix, on either side of the β sheet domain. The interface includes nonpolar, polar, and salt bridge interactions. The total solvent-accessible surface area contributed by GrpE and DnaK to the interface is about 2800 Å2, a typical value for tight protein-protein interactions (18). Most of the known mutations in GrpE that confer temperature-sensitive phenotypes (19-21) appear to be involved in structural stabilization and not in interactions with DnaK (Fig. 1).

The asymmetry in the structure of the GrpE dimer explains the 1:2 stoichiometry of the complex that is observed biochemically (15, 16). The binding of a second ATPase domain to the distal β sheet domain of GrpE would result in a deformation of our GrpE structure. Although not directly involved in DnaK interactions, Phe86 (proximal GrpE) is used to properly position the β sheet domain side chain Arg183, which forms an intermolecular hydrogen bond with Glu28 of DnaK. Phe86 (distal GrpE) would not be able to pack against the aliphatic portion of the Arg183 side chain of GrpE without the local distortion of the long helix that is seen in the proximal GrpE monomer. The four-helix bundle also bends toward the DnaK ATPase domain to optimize an area of GrpE-DnaK interaction (Fig.2C).

The GrpE-DnaK complex is free of nucleotide, whereas in the structure of the ATPase domain of bovine brain Hsc70, ADP is buried and tightly bound in the interdomain cleft. The GrpE-induced movement of domain IIB displaces by 2 to 3 Å three residues in DnaK that hydrogen bond to the adenine and ribose rings of ADP in Hsc70. These are Ser274, Lys270, and Glu267, which are equivalent to Ser275, Lys271, and Glu268 in bovine Hsc70 (Fig. 3). The nucleotide binding site is thus disrupted in the GrpE-DnaK complex by the mechanical opening of the DnaK structure, with GrpE itself remaining distant from the nucleotide binding site.

Figure 3

Superimposition of the nucleotide binding cleft of DnaK onto bovine brain Hsc70. (Left) Diagram of a close-up of the nucleotide binding pocket, with DnaK (red) and Hsc70 (cyan) and Hsc70’s ADP, Mg2+, and inorganic phosphate (yellow). Ser274, Lys 270, and Glu267 are shown to illustrate their displacement from the purine and ribose rings. (Right) To show the displacement of domain IIB of DnaK as a result of GrpE binding, GrpE is drawn as a molecular surface (yellow), with Hsc70 and the Hsc70 nucleotide superimposed onto the DnaK ATPase domain as described in the text. The temperature-sensitive point mutation for λ replication, Glu53 to Gly (19), is indicated with a cyan surface coloring. The COOH-terminus of the ATPase domain is directed toward the proposed location of the peptide binding domain.

We used surface plasmon resonance to determine the apparent association and dissociation rates for the complex of full-length GrpE and full-length DnaK (22). In the absence of nucleotide, the complex forms with an on-rate of ∼5 × 104M 1 s 1 and dissociates with a half-time (t 1/2) of ∼10 min, resulting in an apparent dissociation constant (K d) of ∼30 nM. In the presence of ADP, the on-rate for DnaK binding was reduced ∼20-fold, whereas ADP had little effect on the dissociation of the complex. In contrast, no complex formation was detected in the presence of ATP. Addition of ATP to a nucleotide-free complex caused instantaneous dissociation. We conclude that GrpE recognizes the ADP-bound state of DnaK and results in ADP dissociation.

At present, the only other structure of a nucleotide exchange factor in complex with its nucleotide binding protein is that of the E. coli elongation factor EfTu, a GTP binding protein, in complex with the exchange factor EfTs (23). Like GrpE, EfTs induces a conformational change by binding to EfTu and displaces secondary structure elements necessary for competent binding of the nucleotide. Unlike GrpE, which destabilizes the binding of the purine and ribose rings, EfTs acts by destabilizing the interactions of the phosphate groups of the nucleotide. EfTs also directly competes for the binding site of the divalent cation found adjacent to the nucleotide binding site. In contrast, the nearest approach of any side chain of GrpE to the nucleotide binding site is about 13 Å.

The asymmetry of the GrpE-DnaK interaction raises an intriguing question: Why is GrpE a functional dimer? Essentially all the interactions between GrpE and the ATPase domain of DnaK involve just the proximal GrpE monomer. A major reason for dimer formation would appear to be the need to stabilize the unusually long parallel α-helical structure. However, the long helices are only minimally involved in interactions with the ATPase domain.

The elongated structure of GrpE suggests that GrpE might interact with the peptide binding domain of DnaK. The COOH-terminus of the ATPase domain is positioned such that the COOH-terminal peptide binding domain of DnaK might be located in proximity to the NH2-terminal portion of the long helices of GrpE. A mutation at Glu53 of GrpE results in a temperature-sensitive phenotype for λ replication (19) (Fig. 3, right panel), and this residue in the long helix is located at least 22 Å from the ATPase domain and is not involved in interhelical stabilization. This suggests that the NH2-terminal portions of the long helices of GrpE are functionally important.

To further test this possibility we analyzed the effect of GrpE on substrate binding by DnaK. We found that binding of full-length GrpE to full-length DnaK causes the dissociation of a complex between DnaK and carboxymethylated α-lactalbumin (RCMLA), a permanently unfolded model substrate of DnaK (24, 25) (Fig. 4). Significantly, this dissociation was not observed with GrpE lacking residues 1 to 33, indicating that this presumably flexible segment is necessary for the effect. Consistent with this observation, it had previously been noted that addition of GrpE to a preformed complex of unfolded luciferase, DnaK-ADP, and DnaJ caused the dissociation of at least a fraction of luciferase from its bound chaperones (7). The significance of the peptide-dissociating effect of GrpE remains to be established, because the NH2 terminally truncated form of GrpE was functional in the luciferase refolding assay (13). It is possible that this function of GrpE is necessary for the efficient release of certain tightly binding peptide substrates of DnaK. On the basis of the finding that the NH2-terminal portion of GrpE facilitates peptide release, we propose that GrpE interacts with the peptide binding domain.

Figure 4

Effect of GrpE binding on the interaction between DnaK and unfolded polypeptide substrate. DnaK (9 μM) was incubated for 15 min at 25°C with 9 μM 3H-labeled, reduced, and carboxymethylated bovine α-lactalbumin (RCMLA) (35) in buffer A containing 0.005% Tween 20. Then 0 to 20 μM GrpE (closed circles) or GrpE(34–197) (open circles) was added for 15 min and the reactions analyzed by native polyacrylamide gel electrophoresis on 3 to 10% polyacrylamide gradient gels (36). Amounts of DnaK-bound RCMLA were quantified by densitometry.

The mechanism of nucleotide exchange by GrpE is particularly straightforward and involves a simple opening of the nucleotide binding cleft of DnaK. This mechanism is distinct from that observed in the EFTu-EFTs case, indicating that nucleotide exchange factors have evolved independent modes of action. The unusual structure of GrpE in which two long α helices extend beyond the ATPase domain, as well as the biochemical results presented here, expands the current view of GrpE function to include a role in peptide release. This dual function of GrpE was not previously anticipated.


View Abstract

Stay Connected to Science

Navigate This Article