The unfolding story of the Escherichia coli Hsp70 DnaK: is DnaK a holdase or an unfoldase?



We discuss recent experiments that have illuminated individual steps in the reaction cycle of the Escherichia coli Hsp70 molecular chaperone DnaK. Using this new information, we comparetwo distinctly different global mechanisms of action – holding versus unfolding – and argue that the available evidencesuggests that DnaK is an unfoldase.


The classic experiments of Anfinsen (1973) showing that the unfolded protein RNase A folds spontaneously in vitro to its active native conformation established a central tenet of chemical biology: a protein's primary sequence dictates its three-dimensional structure. As a rule, such in vitro experiments are carried out using small single-domain proteins at low concentrations in mild conditions of temperature, pH, ionic strength and solvent composition. But even with these ideal conditions, it is difficult to attain 100% folding efficiency because of unfavourable side-reactions such as misfolding and aggregation. In living cells, the conditions for folding are far from ideal. First, for nascent chains and for polypeptides during their translocation through membranes, folding is topologically restricted because folding conditions apply to some, but not all, parts of the chain (Rothman, 1989). Secondly, the cytoplasm of a living cell is so incredibly crowded with macromolecules, e.g. the total protein and RNA concentration in an Escherichia coli cell is ≈ 350 mg ml−1 (Zimmerman and Trach, 1991), that aggregation is a serious problem. Nature has endowed cells with complex and elegant protein molecular machines, referred to as heat shock proteins or chaperones, many of which are ATP dependent, that function to assist protein folding and to reverse or inhibit misfolding and aggregation.

The 70 kDa family of heat shock protein chaperones is the focus of this review. Although we possess extensive knowledge in the fields of cell biology, genetics, microbiology and the biochemistry of the 70 kDa family of molecular chaperones, the most important question – What is the molecular mechanism of Hsp70-assisted protein folding? – is still unanswered. In this review, we examine the reaction cycle of the E. coli Hsp70 molecular chaperone DnaK and its two co-chaperones, DnaJ and GrpE, and pose a testable answer to this question. Whatever the mechanism, Hsp70 molecules impart no steric information that influences the folding pathway of another protein; all the information necessary for folding is embedded in the protein's sequence.

Figure 1 illustrates the effect of kinetic partitioning on the distribution of protein states within a cell. Within the text of this article, nascent and folded states are designated by U n and F respectively; the free unfolded, misfolded and aggregated states are designated U, M and A respectively. Spontaneous protein folding occurs on the timescale of milliseconds. However, depending on the size of the protein and its energy landscape, a fraction of the unfolded molecules often misfolds and aggregates ( Thirumalai and Lorimer, 2001 ). Misfolded and aggregated proteins may be trapped in these states for minutes or even hours because the half-times for the return to folding-competent intermediates are very long. Under a variety of stressful conditions, or as a consequence of deleterious amino acid changes to a protein's sequence, misfolded and aggregated proteins may accumulate, causing a cell to be deprived of its essential proteins, which leads to cell death. A wealth of data indicate that Hsp70 molecular chaperones use free energy from ATP binding and/or hydrolysis to minimize the concentrations of M/A states in the cell. The question is, how do they do this?

Figure 1.

Schematic of protein synthesis, folding and misfolding and aggregation in an E. coli cell. Recent genetic and biochemical experiments have revealed that DnaK/DnaJ/GrpE function co- translationally in co-operation with ribosome-bound trigger factor, which is a prolyl isomerase ( Deuerling et al., 1999; Teter et al., 1999 ). An essential post-translational function of DnaK is to break up large protein aggregates, and DnaK accomplishes this in concert with the Hsp104 chaperone ClpB ( Goloubinoff et al., 1999 ; Mogk et al., 1999 ; Diamant et al., 2000 ). TF, K, J and E denote trigger factor, DnaK, DnaJ and GrpE respectively. I denotes a productive folding intermediate. The concentration of the free unfolded protein in a cell is vanishingly small.

Holding versus unfolding

One widely held view is that Hsp70 chaperones selectively bind short hydrophobic segments of nascent proteins or heat-denatured proteins and, by sequestering these segments, misfolding and aggregation is inhibited. Another way to state this is that Hsp70 chaperones possess antifolding or holding activity, in that they stabilize unfolded states of other proteins. Rothman (1989) formulated four essential properties that a chaperone must possess to be a holdase: (i) chaperones should bind unfolded segments of polypeptide chains; (ii) binding should persist through the time required to synthesize a protein or translocate it through a membrane (the duration of protein synthesis is ≈ 20 s in bacterial cells and 300 s in eukaryotic cells); (iii) dissociation must occur at a useful rate in order to keep pace with synthesis and translocation; and (iv) chaperone function is energy dependent. Hsp70 molecular chaperones possess the essential properties (i), (iii) and (iv). We argue below that complexes between DnaK and its substrate proteins are so short-lived in the presence of its co-chaperones that DnaK does not fulfil (ii); thus, it should seriously be considered that DnaK (and probably other Hsp70s) does not function via holdase activity.

An opposing view is that Hsp70 chaperones are unfoldases that use free energy from ATP binding and/or hydrolysis to unfold or pull apart misfolded and aggregated proteins to yield productive folding intermediates (M → I and A → I) (Rothman, 1989; Hubbard and Sander, 1991). Long-lived Hsp70–substrate complexes are not required for this mode of chaperone action. The productive folding intermediate then folds spontaneously in milliseconds to the native state (I → F). In this mechanism, Hsp70 proteins do not change the microscopic rate constant for the folding reaction (I → F); instead, they lower the activation energy barrier for the M → I and A → I reactions, and thus increase the microscopic rate constants for these reactions. Note that, during protein synthesis, a nascent chain can misfold (Un→ M). Acting as an unfoldase, a Hsp70 chaperone could reverse this reaction, thus keeping the polypeptide chain in a folding-competent state until synthesis is completed. It is intriguing that genetic evidence has been obtained showing that the mitochondrial heat shock protein (mtHsp70) unfolds preproteins before import into mitochondria in an ATP-dependent mechanism involving the protein Tim44 (Voisine et al., 1999). A model was proposed recently indicating how DnaK could potentially disentangle a misfolded polypeptide (Gisler et al., 1998). If DnaK is an unfoldase, then the term ‘chaperone’, which in our view implies a passive, holding activity, is a misnomer.

DnaK's structure, function

Composed of two functional domains, the ATPase domain (residues 1–385) and the polypeptide-binding domain (residues 393–537), which is capped by a lid domain (residues 538–638) (Fig. 2A), DnaK binds ATP tightly (Kd = 1 nM) but hydrolyses it incredibly slowly (k hyapp = 3 × 10−4 s−1 at 25°C) (Russell et al., 1998). The hallmark of DnaK and other Hsp70 chaperones is that nucleotide modulates peptide binding and release: in the absence of co-chaperones, ADP-bound DnaK binds and releases peptides over a timescale of minutes or even hours, whereas ATP-bound DnaK binds and releases peptides over a timescale of seconds or even milliseconds. The nucleotide exchange factor GrpE catalyses ADP dissociation from ADP-bound DnaK and thus promotes the high- to low-affinity transition (Packschies et al., 1997). The Hsp40 protein DnaJ catalyses the reverse transition, as described below. In an E. coli cell, DnaK, DnaJ and GrpE are in an approximate 10:1:3 molar ratio. Repeated cycles of substrate binding and release promote the folding, assembly, translocation and proteolysis of proteins in E. coli cells (see Bukau and Horwich, 1998; Ellis and Hartl, 1999; Witt and Slepenkov, 1999; Agashe and Hartl, 2000).

Figure 2.

A . The domain structure of DnaK.

B . Structure of the N-terminal ATPase domain of DnaK. DnaK contains only one tryptophan residue, located at position 102, which is shown in blue. The image was created using the Brookhaven PDB data file 1DKG.

C . Structure of the C-terminal polypeptide- binding domain of DnaK with a bound NR peptide (NRLLLTG) shown in purple. The two residues that comprise the hydrophobic arch (Met-404 and Ala-429) are shown in green. Residues that comprise the latch between the β-sandwich and lid are shown in red. The five helices that constitute the lid are marked A–E. The image was created using the Brookhaven PDB data file 1DKX.

Unfortunately, the three-dimensional structure of a full-length Hsp70 chaperone has not been solved to date; thus, the interface between the two functional domains, which regulates the transmission of conformational signals between the two domains, has not been elucidated. Structural information does, however, exist for the separate domains obtained from partial proteolysis. The structure of the ATPase domain (Fig. 2B) is defined by a central cavity for nucleotide binding at the base of two lobes (Flaherty et al., 1990; Harrison et al., 1997). Nucleotide binding is facilitated by the presence of two K+ ions and one Mg2+ ion.

The C-terminal peptide-binding domain of DnaK is composed of a unique β-sandwich domain arranged in two sheets of four antiparallel β-strands, and this domain is followed by an α-helical lid-like domain (αA–αE) (Fig. 2C) (Morshauser et al., 1995; Zhu et al., 1996). The bound peptide (NRLLLTG) interacts with a deep hydrophobic pocket of the β-sandwich domain through its Leu-4 residue and does not interact with the lid. A hydrophobic arch between residues Met-404 and Ala-429 encapsulates the bound peptide. The optimal substrate length for ATPase stimulation is eight residues (Jordan and McMacken, 1995), and the general motif for high-affinity binding is one in which the peptide has an interior hydrophobic core flanked by basic residues (Gragerov et al., 1994).

The structure depicted in Fig. 2C, with the lid blocking entry and exit from the peptide-binding pocket, is probably the high-affinity form of the C-terminal domain. ATP binding induces conversion to an open conformation in which the lid is thought to rotate via a hinge in the centre of B helix away from the top of the β-sandwich; a conformational change in the β-sandwich must also occur. As stated by Zhu et al. (1996), the picture of the NR peptide bound within a deep channel of the C-terminal domain makes one wonder how it or any other peptide could ever enter or exit the channel. It is even harder to figure out how an exposed hydrophobic loop on a protein could ever penetrate the site. Such a situation is guaranteed to produce interesting kinetics.

DnaK is a bidirectional, ligand-activated molecular switch

To appreciate the impact of ATP binding on the rate of conformational switching, one must know that peptides dissociate from nucleotide-free or ADP-bound DnaK (high-affinity state) with apparent first-order rate constants (k offapp) that fall in the range 10−3−10−4 s−1 at 25°C, whereas upon mixing ATP with nucleotide-free, DnaK-fluorescent peptide complexes, the peptide off-rates increase ≈ 104-fold, yielding k offapp of ≈ 3 s−1 at 25°C (Schmid et al., 1994; Gisler et al., 1998; Slepenkov and Witt, 1998). That ATP-triggered release of bound peptide is so much faster than DnaK-mediated ATP hydrolysis (k hyapp = 3 × 10−4 s−1) leads to the conclusion that ATP binding rather than hydrolysis causes the conformational switch in DnaK (Palleros et al., 1993).

Insights into the dynamics of ATP binding to Hsp70 chaperones have come about from stopped-flow experiments, in which changes in tryptophan fluorescence are observed upon mixing ATP with either nucleotide-free bovine brain Hsc70 (Ha and McKay, 1995) or nucleotide-free DnaK–peptide complexes (Theyssen et al., 1996; Gisler et al., 1998; Slepenkov and Witt, 1998). Our kinetic studies have shown that, when a solution of ATP is rapidly mixed with a solution of preformed DnaK–peptide complexes, ATP binds in a two-step sequential reaction in which the first step is the rapid formation of an intermediate complex, ATP·DnaK·P, and the second step is the rate-limiting conformational switch that expels the bound peptide from the site (and causes the reduction in tryptophan fluorescence). The minimal mechanism that accounts for ATP-triggered peptide dissociation is


where the asterisk denotes the reduction in tryptophan fluorescence. A minimal mechanism means that there is no direct evidence for an ATP·DnaK*·P intermediate. Strong support for this mechanism is that the apparent first-order rate constant for ATP-induced tryptophan fluorescence reduction in DnaK, kobs, obeys the expected relation kobs = k−2[P] + inline image. The kinetic constants at 25°C for the Cro peptide (MQERITLKDYAM) are K−1 =inline image = 22 µM, k2 = 3.3 s−1 and k−2 = 2.4 × 104 M−1 s−1. In

the reverse reaction, when excess unlabeled Cro peptide is mixed with low-affinity DnaK complexes (ATP·DnaK*), there is a rapid (kobs = 5–10 s−1) increase in tryptophan fluorescence that is equal in magnitude but opposite in sign to what occurs when ATP is mixed with DnaK–peptide complexes. This means that the second step in reaction (1), the conformational switch, is reversible, and that both forward and reverse reactions occur without ATP hydrolysis. It also means that the substrate enters and exits the reaction cycle at the same stage, at which the lid is open and the β-sandwich is exposed. ATP hydrolysis therefore occurs from the ATP·DnaK·P intermediate complex (ATP·DnaK·P → ADP·DnaK·P + Pi).

Substrate binding and ATP hydrolysis

With millimolar concentrations of ATP in an E. coli cell, at any instant the vast majority of DnaK molecules are probably ATP bound. An obvious question is, if ATP·DnaK is the low-affinity state, how does substrate binding ever occur? The key player is DnaJ. E. coli DnaJ is organized into four domains: the N-terminal ‘J’ domain is a highly conserved region that constitutes residues 1–78; a glycine/phenylalanine-rich linker region constitutes residues 79–142; a cysteine-rich zinc finger domain constitutes residues 143–200; and the C-terminal domain, which is not well conserved, comprises residues 201–376 (Kelley, 1998). Evidence is mounting that DnaJ itself binds to exposed short hydrophobic loops on the surface of partially denatured or unfolded proteins (Rudiger et al., 2001), and that ATP-bound DnaK then binds to the DnaJ–substrate complex. Karzai and McMacken (1996) showed that the J domain and the glycine/phenylalanine-rich linker of DnaJ are both required to stimulate DnaK-catalysed ATP hydrolysis. But maximal stimulation of DnaK-catalysed ATP hydrolysis occurs as a consequence of a bipartite signalling mechanism composed of two independent but simultaneous interactions: (i) the J domain of DnaJ binds to the ATPase domain; and (ii) the DnaJ-bound substrate binds to the polypeptide-binding domain of DnaK. Both events trigger the transition to the high-affinity state (ADP·DnaK·substrate).

Recent studies have revealed the powerful stimulatory effect of DnaJ on DnaK-mediated ATP hydrolysis. Using a quench flow apparatus to probe the hydrolysis of [32P]-ATP by DnaK in the absence of protein substrate under single turnover conditions ([DnaK] >> [32P-ATP]), it was shown that DnaK hydrolyses ATP in the absence of DnaJ with an apparent first-order rate constant, khy, equal to 1.6 × 10−5 s−1 at 5°C, whereas, in the presence of excess DnaJ, kJhy increases > 15 000-fold to 0.25 s−1 (Russell et al., 1999). The extrapolated value of khyJ is 5 s−1 at 25°C. Using the same technique, but more closely mimicking the conditions found in E. coli cells ([DnaJ] << [DnaK]), another study showed that DnaK rapidly hydrolyses ATP at 25°C (kJhy = 0.3 s−1) as long as saturating concentrations of protein substrate are present (Laufen et al., 1999). The ability of DnaJ to act catalytically is probably a general phenomenon because one molecule of the J domain of Sec63p activates up to 10 molecules of purified yeast Kar2p, an Hsp70 protein located in the endoplasmic reticulum (ER) lumen (Misselwitz et al., 1998).

One model, proposed by Russell et al. (1999), for the effect of DnaJ on DnaK is that DnaJ serves dramatically to increase the rate of DnaK-mediated ATP hydrolysis (kJhy >> khy) according to:


Russell et al. (1999 ) reported evidence that a low-affinity interaction between a DnaJ–substrate complex and an ATP·DnaK molecular complex occurs in the first step[ K d (25°C) =  inline image  > 20 µM]. They proposed that substrate

capture occurs in the second step of the above reaction as a result of a massive increase in DnaK-mediated ATP hydrolysis that occurs after complex formation (k Jhy >> k Joff), and that the principal source of specificity arises from this massive increase in khy by DnaJ. In the absence of DnaJ, substrate binding to ATP·DnaK does not trigger significant DnaK-mediated ATP hydrolysis because koff >> khy. It should be noted that two other studies, using surface plasmon resonance, have indicated a stronger interaction for this first step inline image =inline image (Mayer et al., 1999; Suh et al., 1999).

The discrepancy in the reported Kd values for DnaJ binding to DnaK·ATP might result from the different experimental techniques used. Russell et al. (1999) inferred Kd from the stimulatory effect of DnaJ on DnaK-mediated ATP hydrolysis. In this experiment, the reported Kd could be a constant for the binding of a specific domain of DnaJ (possibly the glycine/phenlyalanine-rich linker region) to the substrate binding site of DnaK (Laufen et al., 1999; Suh et al., 1999). This interaction, like the binding of added peptide, could in turn stimulate DnaK-mediated ATP hydrolysis. On the other hand, surface plasmon resonance experiments are subject to numerous artifacts. Although we use the Kd value of 0.5 µM in the simulations described below, we show in the Supplementary material that the DnaK/DnaJ/GrpE chaperone machine functions equally well when the initial interaction between DnaJ and DnaK·ATP is a rapid but weak equilibrium (Kd = 20 µM).

An intriguing aspect of these studies is that the transit time (τ) through the DnaK reaction cycle is shorter than previously thought. Earlier experiments had indicated that the transit time through the DnaK reaction cycle at 25°C was ≈ 17 s (Pierpaoli et al., 1997), which is close to the time required to synthesize a polypeptide chain in an E. coli cell. However, using the conservative value for kJhy of 0.3 s−1, the calculated transit time through the DnaK reaction cycle is ≈ 3 s (τ = 1/0.3 s−1 = 3.3 s) at 25°C, and probably an order of magnitude shorter at 37°C or at heat shock temperature. Such a short turnover time means that a DnaK molecule cannot be bound to a nascent chain throughout its synthesis. This seems to violate one of the Rothman postulates for antifolding or holding activity, namely that successful holding requires long-lived chaperone–substrate complexes. This discrepancy led us to conduct simulations that compare and contrast the two opposing mechanisms of chaperone activity, holding and unfolding.


The program kinsim (Barshop et al., 1983) was used to simulate various combinations of the reactions in Fig. 3. Reaction (i) shows an equilibrium between folding-competent unfolded (U) and misfolded (M) chains. Such an equilibrium could occur while a nascent chain is being synthesized on the ribosome. After synthesis, the nascent chain leaves the ribosome and folds spontaneously (ii). Reaction (iii) defines antifolding or holding activity: DnaK binds, holds and releases the unfolded and misfolded conformations. Reaction (iv) defines unfoldase activity: DnaK binds, holds and releases the unfolded and misfolded conformations and catalyses the M → U unfolding transition. With these definitions, we examined how these two different activities affect nascent chain synthesis (co-translational) and folding (post-translational). Because DnaJ and GrpE control the lifetime of a DnaK–substrate complex, their collective effect was modelled to a first approximation in the simulations by the appropriate choice of rate constants. Additional simulations that probe the effect of aggregation on protein folding are given in the Supplementary material.

Figure 3.

The chemistry. Reactions (i) and (ii) show misfolding and folding reactions respectively. The holdase (iii) and unfoldase (iv) reactions each have a substrate binding and dissociation phase, governed by k +1 and k −1 , respectively, that produce an intermediate; the intermediate rapidly converts to a terminal complex ( k it ); and then rapid ATP hydrolysis by DnaK occurs ( k cat ) with concerted product dissociation. Implicit in reactions (iii) and (iv) is that DnaJ presents the substrate to DnaK. It is assumed that the steps involving GrpE- mediated ADP dissociation and ATP rebinding are relatively fast (≥ 1 s −1 ); thus, DnaJ- stimulated ATP hydrolysis by DnaK is the rate-limiting step for peptide release.

Co-translational chaperone activity

The ability of DnaK to maintain a nascent chain in a folding-competent, unfolded state (U) during synthesis was simulated. Reactions (i and iii) and (i and iv), which define holding and unfolding activities (Fig. 3), respectively, were simulated The simulations were conducted with identical rate constants and initial conditions (kum = 10 s−1; kmu = 0.001 s−1; k+1 = 25 000 M−1 s−1; k−1 = 0.0125 s−1; kit = 10 s−1; kcat = 0.3 s−1; with [DnaK]0 = 40 µM, [U]0 = 10 µM and [M]0 = 0). Progress curves are shown in Fig. 4A. When DnaK functions as a holdase, within 1 s, ≈ 5% of the molecules are in the folding-competent conformation, whereas 95% are misfolded. In contrast, when DnaK functions as an unfoldase, within 3 s, ≈ 75% of the nascent chains are in a folding-competent conformation. The simulations with this set of parameters show an advantage of unfolding over holding in maintaining nascent chains in a folding-competent conformation.

Figure 4.


A . The ability of DnaK to maintain a nascent chain in an unfolded state was simulated. Holding and unfolding activities were assessed by simulating reactions (i and iii) and (i and iv) ( Fig. 3 ) respectively. Parameters: k um  = 10 s −1 , k mu  = 0.001 s −1 , k +1  = 25 000 M −1 s −1 , k −1  = 0.0125 s −1 , k it  = 10 s −1 and k cat  = 0.3 s −1 . Initial conditions: at time zero, [DnaK] = 40 µM, [U] = 10 µM and [M] = 0.

B . Plot of the percentage of unfolded molecules versus the misfolding rate ( k um ). The percentage of unfolded molecules after 20 s equals

100 × inline image .

Plots 1 and 2: DnaK functions with holding activity with inline image equal to inline image and inline image, respectively.

Plots 3 and 4: DnaK functions with unfolding activity with inline image equal to inline image and inline image, respectively. Other parameters:

k um  = 0.1, 1, 10, 100, 1000 s −1 ; k mu  = 0.001 s −1 ; k it  = 10 s −1 ; and k cat  = 0.3 s −1 . Initial conditions: at time zero, [DnaK] = 40 µM, [U] = 10 µM and [M] = 0.

C . Simulated protein folding. The effect of holdase and unfoldase activities on refolding was assessed by simulating reactions (i–iii) and (i, ii and iv) ( Fig. 3 ) respectively. Parameters: k um  =  k f  = 10 s −1 , k mu  = 0.001 s −1 , k +1  = 25 000 M −1 s −1 , k -1  = 0.0125 s −1 , k it  = 10 s −1 and k cat  = 0.3 s −1 . Initial conditions: at time zero, [DnaK] = 40 µM, [U] = 0 and [M] = 10 µM.

D . Plot of folding half-time versus folding rate ( k f ).

Plots 1 and 2: DnaK functions with unfolding activity with inline image equal to inline image and inline image, respectively. Other parameters:

k f  =  k um  = 0.1, 1, 10, 100 and 1000 s −1 ; k mu  = 0.001 s −1 ; k it  = 10 s −1 ; and k cat  = 0.3 s −1 . Initial conditions: at time zero, [DnaK] = 40 µM, [U] = 0 and [M] = 10 µM.

We found that, under some conditions, holdase activity effectively maintains nascent chains in a folding-competent conformation, i.e. holding inhibits misfolding. What conditions are needed? The key is that the rate of chaperone–substrate complex formation must be greater than or equal to the rate of misfolding (k+1[DnaK] ≥ kum). As reported second-order rate constants fall in the general range 104−106 M−1 s−1 (Gisler et al., 1998; Pierpaoli et al., 1998) and kJon ≈ 2.5 × 104 M−1 s−1 (Mayer et al., 1999; Suh et al., 1999), k+1 was varied from 2.5 × 104 to 2.5 × 106 M−1 s−1 at a fixed Kd (500 nM). Parameters and initial conditions were: inline image or inline image; kmu = 0.001 s−1; kit = 10 s−1; kcat = 0.3 s−1; with [DnaK]0 = 40 µM, [U]0 = 10 µM and [M]0 = 0. Plots 1 and 2 (Fig. 4B) show the percentage of unfolded molecules after 20 s when DnaK functions with holding activity, with k+1 = 2.5 × 104 M−1 s−1 and 2.5 × 106 M−1 s−1 respectively. Plots 3 and 4 (Fig. 4B) show the percentage of unfolded molecules after 20 s of reaction when DnaK functions with unfolding activity, with k+1 = 2.5 × 104 M−1 s−1 and 2.5 × 106 M−1 s−1 respectively. Holdase activity is ineffective at maintaining nascent chains in the unfolded state when k+1 = 2.5 × 104 M−1 s−1 (plot 1). On the other hand, increasing k+1 by two orders of magnitude results in a larger percentage of the molecules in the unfolded state over a wider range of misfolding rates (plot 2). The physical basis for this effect is that rapid binding of DnaK to unfolded molecules inhibits misfolding. In general, the simulations show that unfolding activity is more effective than holding activity at maintaining nascent chains in an unfolded state. For example, between misfolding rates of 1–1000 s−1, ≈ 75% of the molecules are maintained in the unfolded state when the chaperone functions as an unfoldase inline image (plot 3), whereas over the same range

of misfolding rates, 0% of the molecules are unfolded when the chaperone functions as a holdase with the same on and off rates (plot 1). DnaK might use both types of chaperone activity to keep nascent chains unfolded during their synthesis. The question is, can both types of chaperone activity promote protein folding?

Post-translational chaperone activity

Let us compare the effect of holding and unfolding activities on protein folding. The basic question we addressed was, if all protein molecules are initially misfolded, what type of chaperone activity promotes the most efficient refolding? Reactions (i–iii) and (i, ii and iv), which define holding and unfolding activities, respectively, were simulated. The two simulations were conducted with identical rate constants and initial conditions (kum = kf = 10 s−1; kmu = 0.001 s−1; k+1 = 25 000 M−1 s−1; k−1 = 0.0125 s−1; kit = 10 s−1; kcat = 0.3 s−1; with [DnaK]0 = 40 µM, [U]0 = 0 and [M]0 = 10 µM). Progress curves that show the percentage of folded molecules as a function of time are shown in Fig. 4C. Near-complete refolding occurs within 20 s when DnaK functions as an unfoldase, whereas no refolding occurs in 20 s when DnaK functions with holding activity. Reaction parameters were varied in several different ways to determine whether holding activity can, with different parameters, promote refolding, but no set of parameters was found in which holding activity promotes folding over the uncatalysed rate (kmu), which was 0.001 s−1 in these experiments.

Because peptide substrates exhibit a range of on- and off-rate constants for their interactions with DnaK, simulations should probe a range of these rate constants. To this end, simulations of unfoldase activity were conducted by varying k+1 at a fixed Kd(500 nM). Parameters and inital conditions were: inline image = inline image or inline image

k f  =  k um  = 0.1, 1, 10, 100 and 1000 s −1 ; k mu  = 0.001 s −1 ; k it  = 10 s −1 ; k cat  = 0.3 s −1 ; with [DnaK] 0  = 40 µM, [U] 0  = 0 and [M] 0  = 10 µM. Progress curves were analysed to determine the time ( t 1/2 ) at which 50% of the molecules are folded. Plots 1 and 2 in Fig. 4D show t 1/2 versus the folding rate, when k +1 equals 2.5 × 10 4  M −1  s −1 and 2.5 × 10 6 M −1 s −1 respectively. When the smaller k +1 -value is used (plot 1), the folding t 1/2 equals ≈ 7 s over a three order of magnitude range of folding rates (1–1000 s −1 ). In contrast, increasing k +1 by two orders of magnitude substantially increases the t 1/2 for folding, with folding rates ≤ 10 s −1 (plot 2). In general, folding is inhibited, that is t 1/2 increases, when DnaK binds rapidly to its substrates ( k +1 [DnaK] ≥  k f ). Of course, rapid folding (> 10 s −1 ) circumvents this inhibition. With the caveat that the simulations described here did not survey the entire parameter space that may prevail for the reactions in Fig. 3 , it appears that there is a distinct advantage of unfolding over holding as the basis for DnaK's activity. That said, DnaK, acting with enzymatic unfolding activity, is the only way to lower the activation energy barriers for the M → U and M → I transitions.

Possible unfolding mechanism

The hypothesis that Hsp70 chaperones unfold misfolded proteins has been in the literature for quite some time (Rothman and Kornberg, 1986; Rothman, 1989; Hubbard and Sander, 1991). The essence of the unfolding hypothesis is that DnaK unfolds misfolded proteins, or maintains nascent proteins in an unfolded state, by repeatedly binding and pulling apart the substrate protein; this type of chaperone-assisted unfolding has been referred to as ‘plucking’ (Hubbard and Sander, 1991). Because of the co-operative nature of protein folding/unfolding, relatively few cycles, or possibly even one cycle, would be needed to unfold or disentangle a substrate protein. Local unfolding of a substrate protein is proposed to occur either upon binding to DnaK (Fig. 3, reaction iv) or in concert with DnaK-mediated ATP hydrolysis (Pierpaoli et al., 1997). An example of chaperone-assisted unfolding follows. Suppose a misfolded protein displays on its surface an exposed, entangled loop containing a leucine residue in which the hydrophobic side-chain of leucine points into solution rather than into the hydrophobic core of the protein. DnaK, which contains a leucine pocket, binds rapidly to the leucine side-chain, forming an intermediate DnaK··M. In a subsequent step or steps, DnaK locks onto the backbone of the polypeptide chain, and this causes unfolding. The area over which unfolding of the substrate protein occurs may be enhanced by the action of DnaJ, which itself binds transiently to both exposed hydrophobic segments of other proteins and DnaK. Analogous to the GroEL/ES chaperone machine, the formation of a transient DnaJ–substrate–DnaK complex provides two distinct sites that can interact and fix the substrate protein. All DnaK's functions in the cell can be explained by an unfolding mechanism. A global mechanism for DnaK that synthesizes many of the reaction steps described above is shown in Fig. 5.

Figure 5.

Proposed reaction cycle of DnaK. P and P′ are the substrate and unfolded product of DnaK's unfoldase activity respectively. The P to P′ conformational change takes places in the binding step in this mechanism; it could also take place during the hydrolysis step ( Pierpaoli et al., 1997 ). J and E denote DnaJ and GrpE.

Bichaperone network

An essential post-translational function of DnaK is to break up protein aggregates, and DnaK accomplishes this in concert with the Hsp104 chaperone ClpB (Goloubinoff et al., 1999; Mogk et al., 1999; Diamant et al., 2000). It has been shown that catalytic amounts of DnaK/DnaJ/GrpE efficiently disaggregate relatively low-molecular-mass aggregates without help from ClpB, although the efficiency of disaggregation decreases as the size of the aggregates increases. In the case of high-molecular-mass aggregates, high folding efficiency can be restored by either increasing the concentration of the DnaK/DnaJ/GrpE chaperone machine or adding substoichiometric amounts of ClpB. Such results are in accordance with the DnaK/DnaJ/GrpE system functioning as an unfoldase without ClpB. It also means that DnaK and ClpB act co-operatively to rescue high-molecular-mass aggregates.

Future directions

The greatest challenge in the next few years will be to design experiments to test the idea that DnaK performs work on its substrates. The unfolding behaviour of GroEL has been probed via hydrogen–tritium exchange experiments (Shtilerman et al., 1999); possibly, the same technique can be used to probe this proposed aspect of DnaK's function. There is also much to learn about the regulation of DnaK's reaction cycle by DnaJ and GrpE. It is imperative to conduct presteady-state kinetic studies to determine the effect of DnaJ on the kinetic constants for peptide binding to ATP·DnaK. It is intriguing that GrpE also promotes peptide release from DnaK–peptide complexes, although the functional significance of this is not yet clear (Harrison et al., 1997; Mally and Witt, 2001; Mehl et al., 2001).


We thank Michael Sehorn and Dr Robert Smith for critical reading of the manuscript. Support for this work came from the NIH (GM51521).

Supplementary material

The following material is available from http:www. blackwell-science.comproductsjournalssuppmatmolemole3093mmi3093sm.htm.

Ancillary simulations

One simulation examines how the strength of the initial interaction between DanJ and an ATP-DnaK complex affects DnaK's putative unfoldase activity. The other ­simulation examines how aggregation affects the ability of DnaK, functioning as either a holdase or an unfoldase, to refold misfolded protein molecules.