2.1.1 Site of bound polypeptide

Following initial biochemical observations of association of non-native polypeptide with both mitochondrial Hsp60 and bacterial GroEL chaperonin assemblies (Goloubinoff et al. Reference Goloubinoff, Christeller, Gatenby and Lorimer1989; Ostermann et al. Reference Ostermann, Horwich, Neupert and Hartl1989), microscopy studies localized bound protein in the central cavity of an open GroEL ring. Hartl, Baumeister, and their co-workers, using negative stain electron micrsoscopy (EM), observed ‘mass’ within the central cavity in end views of substrate-bound as compared with empty complexes (Langer et al. Reference Langer, Pfeifer, Martin, Baumeister and Hartl1992). Wall, Hainfeld, Horwich, and their collaborators, using scanning transmission electron micrsoscopy, observed localization of a nanogold particle attached to a substrate protein within the central cavity in end views and at a terminal end of the GroEL cylinder in side views of substrate-bound complexes (Braig et al. Reference Braig, Simon, Furuya, Hainfeld and Horwich1993). The localization of bound substrate was confirmed by Saibil and co-workers during a study using negative stain EM to resolve ATP-directed movements of GroEL. They consistently observed ‘masses’ in the terminal ends of the GroEL central cavities in side views, abutting the GroEL apical domains (Saibil et al. Reference Saibil, Zheng, Roseman, Hunter, Watson, Chen, Auf der Mauer, O'Hara, Wood, Mann, Barnett and Ellis1993). The masses very likely comprised an averaged collective of substrate proteins that remained bound by GroEL during purification. This study was followed by one from the same group using cryo-EM that similarly localized a substrate protein, malate dehydrogenase (MDH), to one end of the GroEL cylinder (Chen et al. Reference Chen, Roseman, Hunter, Wood, Burston, Ranson, Clarke and Saibil1994), consistent with biochemical studies indicating a stoichiometry of one non-native polypeptide bound per GroEL double ring particle (Goloubinoff et al. Reference Goloubinoff, Christeller, Gatenby and Lorimer1989), implying a negative cooperativity between rings with respect to polypeptide binding.

At the same time as the early EM studies, a study from Horowitz, Lorimer, and their co-workers inferred that hydrophobic interactions might be involved in polypeptide binding, observing that addition of the detergent lauryl maltoside could modestly improve the low recovery of native rhodanese enzyme following dilution from urea at 25°C in the absence of chaperonin (Mendoza et al. Reference Mendoza, Rogers, Lorimer and Horowitz1991). While the recovery was well below the near quantitative recovery achieved by GroEL–GroES–ATP, it suggested that the chaperonin, like the detergent, might be binding-exposed hydrophobic surfaces of the non-native rhodanese, preventing them from directing aggregation. An additional inference of the importance of hydrophobic interaction came from a thermodynamic study observing a negative heat capacity associated with binding of non-native subtilisin, an indication of hydrophobic interactions as a driving force in polypeptide binding (Lin et al. Reference Lin, Schwarz and Eisenstein1995).

The X-ray structure of unliganded GroEL indeed revealed a hydrophobic cavity wall at the level of the apical domains at the terminal ends of the chaperonin cylinder, an apparent surface to which non-native polypeptides could become bound (Braig et al. Reference Braig, Otwinowski, Hegde, Boisvert, Joachimiak, Horwich and Sigler1994), apparently through their own hydrophobic surfaces that would be exposed specifically in the non-native state but that would be buried to the interior of native folded protein (for which GroEL does not have affinity). Consistently, mutagenesis studies showed that substitution of any one of eight different hydrophobic residues facing the GroEL central cavity at the level of the apical domains to hydrophilic or electrostatic character abolished polypeptide binding (Fig. 3; Fenton et al. Reference Fenton, Kashi, Furtak and Horwich1994). These residues were positioned on a tier of three different structures lining the cavity: helix H, helix I, and an underlying extended segment (see Fig. 3). Additional mutational studies used covalent rings that enforced a specific arrangement of subunits with wild-type and mutant apical domains (the mutant substituting a central residue in the binding site, Val 263, with Ser). It was observed that three or four consecutive intact wild-type apical domains were required in order to efficiently bind non-native substrate protein (Farr et al. Reference Farr, Furtak, Rowland, Ranson, Saibil, Kirchhausen and Horwich2000). Cross-linking studies validated this conclusion. Cysteines in a bound radiolabeled substrate protein, Rubisco, were assessed for ability to form disulfide bonds with multiple apical domains, each carrying a single cysteine substitution in the apical binding site. The radioactive cross-linked products were sized in an SDS gel to determine the number of GroEL subunits that had become cross-linked (Farr et al. Reference Farr, Furtak, Rowland, Ranson, Saibil, Kirchhausen and Horwich2000). This indicated that, indeed, bound substrate could simultaneously interact with three or four surrounding apical domains.

Fig. 3.

The apical residues that form the hydrophobic binding surface of a GroEL subunit. A ribbons model is shown, with side chains of residues whose mutation abolishes polypeptide binding shown as yellow sticks and labeled with residue numbers. The residues are grouped on three structures – helix H, helix I, and an underlying extended segment – from top to bottom, respectively. (a) View en face from inside the central cavity. (b) View from one side of a subunit, with the cavity to the left. From Horwich et al. (Reference Horwich, Farr and Fenton2006), with permission. Copyright 2006, American Chemical Society.

More recently, a multivalent binding topology has been directly observed by cryo-EM studies examining one of the well-studied substrate proteins, MDH, while bound to an open GroEL ring (Elad et al. Reference Elad, Farr, Clare, Orlova, Horwich and Saibil2007). Multiple image classes revealed MDH lodged against three or four apical domains of an open GroEL ring (Fig. 4). At the level of the individual domains, there were varying topologies observed. One image class, for example, showed electron density from MDH perched at the outside rim of the cavity (Fig. 4g–i), reflecting interaction with helices H and I, and another showed MDH density deeply buried in the cavity, abutting both helix I and the underlying extended segment (Fig. 4a–c). The GroEL apical domains themselves appeared to adjust to polypeptide binding, with those physically contacting substrate bunched together and those at the opposite aspect of the ring exhibiting a wider separation from each other (see also Falke et al. Reference Falke, Tama, Brooks, Gogol and Fisher2005).

Fig. 4.

Topologies of non-native MDH bound to GroEL. Three image classes from a cryo-EM reconstruction (Elad et al. Reference Elad, Farr, Clare, Orlova, Horwich and Saibil2007) are illustrated in side and end views. (a, d, g) Sections through the MDH-bound GroEL subunits in each class, showing the fitting of the GroEL apical domain secondary structural elements. Note the areas at the end of each domain without fitted elements, with density contributed by the bound non-native MDH. (b, e, h) Side views of a substrate-bound subunit (see thumbnail at top), with the GroEL rendered as a surface at 14 Å and the difference density corresponding to MDH colored yellow, green, or blue. (c, f, i) Top views of substrate-bound subunits (see thumbnail) rendered as in (b, e, h). Three GroEL subunits are shown in (c) and (f) and four in (i), corresponding to the number of subunits contacted by substrate in each of these classes. (j) The individual difference densities corresponding to MDH for each class, compared with the surface representation of an MDH subunit monomer (red) filtered to 20 Å; the exposed dimer interface of the MDH subunit in the bottom image is colored purple. Modified from Elad et al. (Reference Elad, Farr, Clare, Orlova, Horwich and Saibil2007), with permission. Copyright 2007, Elsevier.

An additional cryo-EM study of the large T4 phage-encoded capsid protein, gp23 (54 kDa), in complex with GroEL, has recently been reported, showing localization of density of the non-native capsid protein deep within the cavity, contacting five to seven surrounding GroEL subunits of an open ring (Clare et al. Reference Clare, Bakkes, van Heerikhuizen, van der Vies and Saibil2009). Bound substrates thus interact simultaneously with multiple surrounding hydrophobic apical domains and can interact at multiple points across the hydrophobic surface, although in general it appears that helix I represents a central shared position of association.

For polypeptides of size greater than 30–40 kDa, it seems clear that only a portion of the bound non-native chain can be contained within the central cavity, whereas other portions must reside in the bulk solution. Such portions have not been observable by EM but have been inferred to lie in solution immediately above the cavity opening. For example, the oblong shape of GroEL–rhodanese complexes as determined from neutron scattering studies suggested that the polypeptide occupied a topology like that of a ‘champagne cork’ (Thiyagarajan et al. Reference Thiyagarajan, Henderson and Joachimiak1996). Consistently, additional cysteine cross-linking studies employing cysteine scanning across the interior and exterior surfaces of GroEL do not observe GroEL-bound polypeptide forming cross-links with the top surfaces of the apical domains (Elad et al. Reference Elad, Farr, Clare, Orlova, Horwich and Saibil2007). Rather, cross-links are only formed within the cavity, occurring at all levels. That is, while the polypeptide chain is physically associated with the apical domains, it can occupy space throughout the cavity, including the equatorial zone. Thus, for example, the mass of C-terminal tails of the GroEL subunits (aa 523–547) bearing flexible GGM repeats, lying in the equatorial zone of the cavity and forming a mass (~20 kDa per ring) that is visible by EM but not crystallographically resolvable, can form contacts with the non-native polypeptide. The C-terminal tails presumably serve as a barrier that contains substrate within a ring, although their deletion does not produce any significant growth defect of Escherichia coli (Burnett et al. Reference Burnett, Horwich and Low1994; McLennan et al. Reference McLennan, McAteer and Masters1994). Notably, however, in Mycobacterium tuberculosis, there are two GroEL homologs, one of which, GroEL1, has C-terminal tails bearing a poly-histidine stretch. This segment might contribute to binding specific substrate proteins, such as KasA, involved in synthesis of mycolic acids that are required for biofilm formation, a property dependent on GroEL1 function (Ojha et al. Reference Ojha, Anand, Bhatt, Kremer, Jacobs and Hatfull2005).

Interestingly, the portion of bound polypeptide in solution outside the cavity appears in general to be also non-native in character, as opposed to occupying a folded state. For example, the large substrate protein, aconitase (82 kDa), was entirely susceptible to proteolysis and hydrogen exchange while bound by GroEL (Chaudhuri et al. Reference Chaudhuri, Farr, Fenton, Rospert and Horwich2001, unpublished observation). This, as discussed below, suggests that the chaperone binds various exposed hydrophobic surfaces in a spatially averaged as well as probably a time-averaged fashion.

2.1.2 Elements recognized in bound polypeptide

The proffering of hydrophobic side chains on the part of GroEL-bound non-native substrate proteins themselves has also been explored. For example, Itzhaki, Fersht, and their co-workers engineered single substitutions into a variety of residues in the protein CI2, observing the greatest reduction in binding with substitution of hydrophobic side chains with alanine or glycine (Itzhaki et al. Reference Itzhaki, Otzen and Fersht1995). In a further study from the Fersht group, the crystallization of an isolated apical domain led unexpectedly to the crystallographic resolution of seven residues, GLVPRGS, in a larger N-terminal His tag, bound in an extended conformation within a groove between helices H and I in a neighboring apical domain molecule in the asymmetric unit (Buckle et al. Reference Buckle, Zahn and Fersht1997). The leucine was positioned in a cavity between the two α-helices, while the C-terminal portion of the peptide aligned along helix I, forming numerous apolar contacts with the same hydrophobic side chains that had been identified by mutagenesis. In a more directed study, Chen & Sigler (Reference Chen and Sigler1999) used filamentous phage display to select a 12 residue peptide called SBP (strong binding peptide) that bound to isolated GroEL apical domain with ~10 μm affinity (compared with typically nanomolar affinity for polypeptides, bound multivalently). The SBP peptide, SWMTTPWGFLHP, was co-crystallized with isolated apical domain, binding as a β-hairpin with residues 7–12 associated with the cavity-facing surface of the apical domain via the groove between helices H and I, forming a number of hydrophobic contacts with residues that had been implicated in the earlier mutagenesis study (Fig. 5).

Fig. 5.

Topologies of three peptides bound to the apical domain of GroEL. The backbones of the GroES mobile loop segment (blue) (Xu et al. Reference Xu, Horwich and Sigler1997), the ‘strong binding peptide’ (green) of Chen & Sigler (Reference Chen and Sigler1999), and the N-terminal extension peptide from Buckle et al. Reference Buckle, Zahn and Fersht1997 (magenta) are shown overlaid on aligned Cα ribbon trace of apical domain helices H and I of GroEL (red) from the respective structures. Note that the bound regions of all three peptides appear to occupy nearly the same position in the groove between the two helices. From Fenton & Horwich (Reference Fenton and Horwich2003), with permission. Copyright 2003, Cambridge University Press.

These two studies of peptide–apical interactions provide the highest resolution to date of apical interactions. It has to be noted, however, that they resemble quite strongly the interaction formed by the hydrophobic mobile loop of GroES when it binds to the GroEL apical domains (Fig. 5; see Xu et al. Reference Xu, Horwich and Sigler1997; Shewmaker et al. Reference Shewmaker, Maskos, Simmerling and Landry2001). One might suggest that these interactions between peptide and the apical interhelical groove represent one physiologic mode of polypeptide binding, potentially a favored one, but, given the cryo-EM data, there are surely others that have yet to be resolved, not to mention that the ‘tracking’ of a non-native polypeptide across multiple apical domains has yet to be observed at high resolution. The problems of trapping such a state for X-ray studies remain daunting, as discussed earlier (Fenton & Horwich, Reference Fenton and Horwich2003).

One of the most informative studies concerning the role of substrate hydrophobicity in polypeptide binding has been carried out by the Gierasch group, systematically arranging hydrophobic side chains and secondary structure of 20 different synthetic peptides of 12–14 residues, assessing their overall hydrophobicity by retention time on reverse phase high-performance liquid chromatography and measuring affinity of GroEL for them by either line broadening or by the transferred NOE (trNOE) method of nuclear magnetic resonance (NMR) (Wang et al. Reference Wang, Hwa-Ping, Landry, Maxwell and Gierasch1999). First, reversal of chirality of an α-helical peptide using d-amino acids did not affect the extent of binding, and the trNOE pattern suggested that the helical character (now left-handed) was preserved. Interestingly, an l,d-mixed peptide was also recognized and could compete with the l-peptide but assumed a structure, as judged from the trNOE spectrum, that was not recognizable as either a right- or left-handed α-helix. Thus, the three peptides, exhibiting similar proffered hydrophobicity as judged from retention in reversed phase analysis, were bound with roughly similar affinity by GroEL, despite their substantially different secondary structures, pointing to hydrophobic side chains as a major recognized feature regardless of backbone context. Next, the binding of an amphiphilic α-helical peptide, placing hydrophobicity contiguously on one side, was compared with that of a non-amphiphilic one of the same amino acid composition. Here the non-amphiphilic peptide exhibited both less line broadening and smaller trNOEs, reflecting lower affinity of GroEL for this peptide. This would support the idea that clustering of hydrophobic side chains, here on one face of an α-helix, favors binding (see also Preuss et al. Reference Preuss, Hutchinson and Miller1999; Li et al. 2009). Finally, peptides whose trNOE spectra were consistent with extended structure while associated with GroEL (e.g. as in SBP) were also studied, and one with alternating hydrophobic/hydrophilic character was observed to be recognized by GroEL. Overall, when the retention time of the 20 peptides was compared with intensity of transfer NOEs as a measure of affinity, there was an excellent correlation, suggesting that extent of contiguous hydrophobic exposure determines affinity of GroEL for any given stretch of polypeptide chain, irrespective of structural context.

Indeed, GroEL has substrates that are all α-helical, like citrate synthase (Buchner et al. Reference Buchner, Schmidt, Fuchs, Jaenicke, Rudolph, Schmid and Kiefhaber1991), but also can bind all β-sheet proteins, like a partially disulfide-bonded F_ab fragment (Schmidt & Buchner, Reference Schmidt and Buchner1992). Similarly, studies of different structural states of α-lactalbumin indicate that GroEL does not efficiently recognize the disulfide-bonded molten globule state (formed by chelating calcium ion; Okazaki et al. Reference Okazaki, Ikura, Nikaido and Kuwajima1994) but prefers a reduced, less structured state, which exposes substantial amounts of hydrophobic surface (Hayer-Hartl et al. Reference Hayer-Hartl, Ewbank, Creighton and Hartl1994). Parenthetically, disulfide-bonded species are not normally acted on by chaperonins, because they are, for the most part, secreted proteins that fold in the oxidizing environment of the ER, where such ring assemblies have not been identified. Chaperonins, rather, reside in reduced compartments of the cell (cytosol, mitochondrial matrix, chloroplast stroma), where disulfide bonds are not usually formed. However, disulfide bonded proteins, because of their ability to form relatively stabilized intermediate states, have been of use in vitro as ‘probes’ of GroEL function.

2.3.1 GroEL-bound substrates lack stable secondary structure

Exchange studies on binary complexes of cyclophilin (Zahn et al. Reference Zahn, Spitzfaden, Ottiger, Wüthrich and Plückthun1994), disulfide-reduced α-lactalbumin (Robinson et al. Reference Robinson, Gross, Eyles, Ewbank, Mayhew, Hartl, Dobson and Radford1994), human DHFR (Goldberg et al. Reference Goldberg, Zhang, Sondek, Matthews, Fox and Horwich1997), MDH (Chen et al. Reference Chen, Walter, Horwich and Smith2001), and Rubisco (Park et al. Reference Park, Fenton and Horwich2005) showed at best only low levels of protection in the bound proteins. In several cases, potentially weakly structured native regions were observed, but protection factors in these regions were below 100. The observation of a lack of any significant hydrogen bonded structure raised the question of whether secondary structure present in species free in solution might be getting disrupted as they became bound to the open GroEL ring.

Yet observations on two well-studied substrates while free in solution have failed to indicate any substantial secondary structure. When MDH was diluted from denaturant and pulse exchanged while free in solution 5 s later (for 1 s), it exhibited little or no protection (Chen & Walter, unpublished observation; see also Chen et al. Reference Chen, Walter, Horwich and Smith2001). This corresponds to the timescale on which non-native MDH would be fully bound by GroEL, because its bimolecular rate constant for binding to GroEL has been measured at 5×10⁶ M⁻¹ s⁻¹, and GroEL concentration is ~1 μm (Ranson et al. Reference Ranson, Burston and Clarke1997). Similarly, a metastable non-native intermediate of Rubisco, formed after dilution from denaturant under chloride-free conditions and able to be efficiently bound by GroEL during a lifetime of ~1 h, exhibited no protection from pulsed hydrogen–tritium exchange (Park et al. Reference Park, Fenton and Horwich2005).

2.3.2 Binary behavior of secondary structure during a cycling reaction – progressive vs. all-or-none folding

There is an implication from the observations above that there may really be only two principal states of secondary structure of a substrate protein in a chaperonin-mediated folding reaction, either non-native intermediates, an equilibrium ensemble of poorly structured states that are rapidly captured by GroEL via exposure of hydrophobic surfaces, or the end-product native state, no longer recognized by the chaperonin. This implies that partially folded non-native intermediates with substantial secondary structure do not accumulate during the reaction. Indeed, such states, that would manifest stable secondary structure in pulsed hydrogen–deuterium exchange, have not been observed. Rather, the evidence at hand would support that stable secondary structure formation seems to occur as all or none. Indeed, hydrogen exchange behavior does not substantially change if one compares the behavior of the substrate in a starting polypeptide–GroEL binary complex with a binary complex isolated after a number of ATP–GroES-driven cycles – the not-yet-native molecules of polypeptide remain exquisitely sensitive to protease and exhibit the same hydrogen exchange sensitivity from cycle to cycle (Weissman et al. Reference Weissman, Kashi, Fenton and Horwich1994; Groß et al. Reference Groß, Robinson, Mayhew, Hartl and Radford1996; Horst & Fenton, unpublished observations). In addition, topology studies by single-molecule FRET indicate the same distribution of topologies at both the beginning and after minutes of the cycling reaction (Sharma et al. Reference Sharma, Chakraborty, Müller, Astola, Tang, Lamb, Hayer-Hartl and Hartl2008).

2.3.3 Secondary structure unbroken by GroEL on the physiologic timescale in two bound secretory proteins bearing disulfides (that are not normally substrates for GroEL)

There are two examples to date of GroEL-bound proteins that exhibit substantial stable secondary structure while bound to an open chaperonin ring. Gervasoni et al. (Reference Gervasoni, Staudenmann, James, Gehrig and Plückthun1996) reported that E. coli RTEM β-lactamase heated to 48°C formed a stable complex with GroEL, in which it had lost its native state as evidenced by loss of enzymatic activity. When the binary complex was subjected to hydrogen–deuterium exchange, however, the β-lactamase was exchanged to an extent resembling that of native enzyme at 25°C, reflecting that a native-like degree of secondary structure was preserved in the bound protein. Further examination of the binary complex using partial proteolysis suggested a bound topology in which the central β-sheet was separated from a C-terminal α-helical region, exposing hydrophobic surfaces that were recruited to GroEL (1998). RTEM β-lactamase seems likely to be stabilized against further unfolding by a disulfide bond in its N-terminal α-helical domain. This disulfide would normally be acquired upon entry of the protein into the oxidizing environment of the bacterial periplasm. As mentioned earlier, the normal cytosolic substrates of GroEL lack such stabilizing bonds, residing in the relatively reducing environment of the cytosol. Thus, this may represent a situation where further unfolding associated with recruitment of additional hydrophobic surfaces of the bound non-native state to the GroEL apical domains cannot occur due to covalent stabilization by the disulfide bond, a type of stabilization that GroEL does not normally confront both due to the lack of such bonds in its natural substrates and the failure to form such bonds in the reducing environment of the cytosol.

In a second case, Park and co-workers have produced a partially disulfide-bonded intermediate of the secretory protein, trypsinogen, that exhibits substantial hydrogen bonded secondary structure while free in solution as indicated by exchange protection (Park et al. unpublished observations). This intermediate, produced by partial reduction of native trypsinogen, contains four out of the six native disulfides of trypsinogen. It behaves as a monomeric species in gel filtration, but nevertheless appears to expose substantial hydrophobic surface, because when mixed with GroEL, it becomes efficiently bound. In free solution, when subject to pulsed deuterium exchange (1 s), the four disulfide intermediate acquired 75 deuterons, whereas in the binary complex with GroEL, it acquired 142 deuterons, relatively close to the theoretical maximum incorporation of 183 deuterons. Yet the time employed for binary complex formation in this experiment was 30 min! In contrast, under physiologic conditions during a cycling reaction, a GroEL ring is open and available for binding for only 1 s before ATP and GroES arrive (see Fig. 2). Indeed, when binding was allowed for only 1 or 10 s before pulsed exchange, there was no exchange beyond 75 deuterons. Thus, on the short timescale, only the same surface-accessible amide protons that were exchanged in free solution were apparently subject to exchange in the binary complex. Indeed, only modest exchange beyond this level was observed even when binding was carried out for 5 min.

Thus, GroEL has only a weak ability to unfold the disulfide-bonded intermediates described and does so slowly. Because it appears to rely simply on passive hydrophobic binding, it does not have the potency of, for example, the AAA+ pulling machines such as ClpX or ClpA, which use forceful ATP-driven axial-directed movement of loops lying in their central channels to pull bound segments of polypeptide through the channels, mediating an unraveling action (Sauer et al. Reference Sauer, Bolon, Burton, Burton, Flynn, Grant, Hersch, Joshi, Kenniston, Levchenko, Neher, Oakes, Siddiqui, Wah and Baker2004; Hinnerwisch et al. Reference Hinnerwisch, Fenton, Farr, Furtak and Horwich2005).

2.3.4 Direct NMR examination of stably bound substrate proteins – lack of secondary and tertiary structure and evidence for conformational exchange

Structure in bound polypeptide has also been directly probed by recent NMR studies using transverse relaxation optimized spectroscopy (TROSY) techniques, directly observing isotope-labeled human DHFR (Horst et al. Reference Horst, Bertelsen, Fiaux, Wider, Horwich and Wüthrich2005) and rhodanese (Koculi, submitted) while stably bound to GroEL. In both cases, the uniformly ¹⁵N-labeled and deuterated proteins in complex with ¹⁵N-depleted GroEL produced resonances in the random coil region of two-dimensional ¹⁵N, proton-correlated spectra, reflecting a lack of regular secondary or tertiary structure. Line widths were broad, reflecting either the slow tumbling of the large complex or internal motion of the bound substrate, or its presence in multiple bound conformations. With both uniform labeling and, in the case of ¹⁵N-leucine specific labeling of DHFR, the number of detected residues of any specific amino acid type appeared to be underrepresented. For example, in the case of the leucines of DHFR, only a small number of overlapped cross-peaks were observed, out of 19 that would be predicted (Horst et al. Reference Horst, Bertelsen, Fiaux, Wider, Horwich and Wüthrich2005). This suggests that parts of the bound polypeptide chain are NMR invisible. These could conceivably be parts of the chain that are stably associated with the cavity wall. An attempt to resolve this with site-specific assignment using mutagenesis of the glycine residues in rhodanese, observing spectra of various alanine-substituted mutant forms while in complex with GroEL, has not so far resolved which out of 20 glycines are producing the three to four broad cross-peaks observed in this region of the spectrum.

The dynamics of SR1-bound DHFR was also examined by comparing two different types of magnetization transfer, CRIPT and INEPT, measuring ¹H–¹⁵N-correlated cross-peak intensity in relation to different transfer times (Horst et al. Reference Horst, Bertelsen, Fiaux, Wider, Horwich and Wüthrich2005). The several-fold greater signal intensity with INEPT and persisting anti-phase character by both mechanisms of transfer indicated that DHFR is not rigidly bound to SR1 but is internally mobile. Because it is unknown whether bound portions of the chain are NMR visible, it is unclear whether such associated regions also undergo conformational exchange, e.g. with one region of the chain bound to an apical domain at one time but replaced by a different region of the chain at a subsequent time. This issue of whether apical bound segments are stably bound remains a fundamental question. Further NMR studies employing magnetization transfer between polypeptide and the apical domains might ultimately be able to address this. Such studies may become feasible in the near future considering both recent backbone assignment of the GroEL apical domains in the setting of an intact thermophilic GroEL assembly (Horst et al. unpublished observations) and recent successes in observing methyl side chains in large assemblies using methyl-TROSY techniques (e.g. Sprangers & Kay, Reference Sprangers and Kay2007). Meanwhile, fluorescence studies, discussed below, have provided additional measurements of chain behavior.

2.4.1 ‘Stretching’ action exerted on polypeptide topology in association with binding to an open ring – rescue of kinetically trapped monomers

Lin & Rye (Reference Lin and Rye2004) placed fluorophores near the N- and C-termini of Rubisco and studied the behavior of the ensemble following dilution from denaturant to 100 nm concentration at 4°C. Under these conditions, Rubisco behaved as a misfolded monomer, with a high fluorescence resonance energy transfer (FRET) efficiency reflecting a distance between its termini much shorter than in the native state, indicating an apparently collapsed misfolded state that brought the termini close to each other. Upon addition of GroEL–GroES–ATP, the monomeric substrate was efficiently refolded to the native state at 4°C. The step of polypeptide binding was examined by stopped flow mixing with GroEL alone. Binding to GroEL was associated with a rapid reduction of FRET efficiency, reflecting an increased distance between the substrate termini, occurring on a timescale corresponding to that of binding, i.e. approximately a second (Fig. 7). This stretching apparently rescued Rubisco from a kinetically trapped topology that could not on its own productively fold in solution at 4°C. This represents probably the most compelling evidence to date for the ability of GroEL to rescue a non-native monomeric species from a kinetically trapped conformation via binding to an open ring, albeit at a nonphysiologic temperature. In particular, while the same non-native state is likely to be populated to some degree at 23°C, at this temperature it would, as discussed later, be able to pursue additional fates in free solution, including the formation of aggregates, both low-order ones that are reversible and higher-order irreversible ones. In addition, at a higher temperature, it would have access to additional thermal energy to drive spontaneous reversal of this misfolded state, allowing production of the native state (Viitanen et al. Reference Viitanen, Lubben, Reed, Goloubinoff, O'Keefe and Lorimer1990; Apetri & Horwich, Reference Apetri and Horwich2008).

Fig. 7.

‘Stretching’ of non-native Rubisco upon binding to GroEL, as monitored by changes in FRET. (a) Schematic of the experiment. A kinetically trapped, apparently misfolded monomer of Rubisco (able to be converted to the native state by addition of GroEL–GroES–ATP), labeled near its N- and C-termini with the FRET pair AEDANS and fluorescein, was stable at 4°C and showed an interprobe distance of ~37 Å. After binding to GroEL (or SR1) to form a binary complex, the interprobe distance increased to >70 Å for the GroEL complex and to 46–47 Å for the SR1 complex. This indicates a significant stretching of Rubisco as a result of binding to the apical domains. (b) Stopped-flow experiment to determine the rate at which stretching occurs. The changes in FRET efficiency for Rubisco binding to both GroEL and SR1 were biphasic, requiring two exponential terms for fitting. Only the faster rate was dependent on chaperonin concentration, and the bimolecular rate constant with GroEL, 2×10⁵ to 4×10⁵ M⁻¹ s⁻¹, was essentially equal to the independently measured rate constant for Rubisco binding to GroEL under these conditions. This suggests that the stretching occurring during the faster step is an inherent feature of Rubisco binding. Redrawn from Lin & Rye (Reference Lin and Rye2004), with permission. Copyright 2004, Elsevier.

Yet a number of earlier studies carried out at 23°C and above have also observed situations where the substrate proteins Rubisco and MDH fail to reach the native state while free in solution and yet remain accessible to it upon subsequent addition of the chaperonin system. For example, following neutralization, acid-unfolded Rubisco failed to refold while free in solution over periods of 15 or 30 min, but upon addition of the chaperonin system, a fraction of the Rubisco (decreasing with time) proceeded to the native state (with identical kinetics) (Todd et al. Reference Todd, Lorimer and Thirumalai1996, reproduced in Fig. 8). This presumably involves recruitment of kinetically trapped monomers, unable to proceed to the native state, to open GroEL rings, followed by cis folding. Given the observations of Lin & Rye at 4°C, one must presume that binding to GroEL effectively stretched the Rubisco monomers, releasing them from a kinetic trap. Similar experiments have been reported as well for MDH (Peralta et al. Reference Peralta, Hartman, Hoogenraad and Høj1994; Ranson et al. Reference Ranson, Dunster, Burston and Clarke1995). This suggests, once again, that, even while there is little or no stable secondary structure in the trapped non-native states, their topologies can become sufficiently non-native that thermal motions will not provide sufficient energy to drive productive folding and that the action of multivalent hydrophobic binding by an open ring is able to relieve the trapped state.

Fig. 8.

Folding of Rubisco by GroEL added at different times after dilution from denaturant. Acid-denatured Rubisco was diluted directly into refolding buffer containing GroEL, GroES, and ATP (squares) or into buffer with GroES and ATP, with GroEL added at 15 (triangles) or 30 min (circles). In each case, the rate of recovery of Rubisco activity was the same, although the yield decreased with time before GroEL addition. In the absence of added GroEL, however, no Rubisco activity was recovered. This suggests that GroEL is binding a metastable, kinetically trapped intermediate and returning it to a folding pathway. Redrawn from Todd et al. (Reference Todd, Lorimer and Thirumalai1996), with permission. Copyright 1996, National Academy of Sciences, USA.

A more recent study has also observed that polypeptide binding by GroEL can be associated with passive stretching, using as substrate a protein called DM-MBP (double mutant V8G/Y283D of the mature maltose-binding protein; Sharma et al. Reference Sharma, Chakraborty, Müller, Astola, Tang, Lamb, Hayer-Hartl and Hartl2008). Several different pairs of fluorophore-labeled cysteine-substituted positions were used in ensemble FRET kinetic studies. There was a rapid rise of FRET efficiency within milliseconds of dilution from denaturant, whether GroEL was present or not, reflecting collapse of the polypeptide. In the presence of GroEL, however, this was followed by a rapid drop in FRET efficiency, exhibiting a t _½ of ~100 ms, which corresponds to a typical rate of substrate protein binding. This drop was attributed to an expansion of the substrate associated with GroEL binding.

2.4.2 Single-molecule FRET studies of stably associated substrate proteins reveals heterogeneity of substrate topology

Seeking to address the behavior of stably GroEL-bound substrate are two recent single-molecule FRET experiments, one examining rhodanese, the other examining DM-MBP. Both studies imply that stably bound substrates behave as a heterogeneous set of conformations. In the rhodanese study, Schuler and co-workers placed Alexa Fluor 488 and 594 FRET reporters at two different pairs of positions on rhodanese and measured energy transfer efficiency at the single-molecule level under denaturing conditions, after binding to GroEL, and after ATP–GroES-mediated refolding (Hillger et al. Reference Hillger, Hänni, Nettels, Geister, Grandin, Textor and Schuler2008). Both in denaturant and after refolding, they observed a relatively narrow distribution of transfer efficiency that corresponded in the former case to the distance along the primary structure separating the fluorophores, and in the latter to the distance determined from the X-ray structure of the native state. In the denatured state, studies indicated rapid intramolecular dynamics, as might be expected, on the timescale of ~70 ns. When the rhodanese was bound by SR1, a broad distribution of transfer efficiencies was observed, reflecting a heterogeneous ensemble of distances, with a maximum that resembled the distance that would be observed in the native state, suggesting possibly a bias toward a native topology. The dynamics for rhodanese in the binary complex, measured by several means, corresponded to that of GroEL itself, suggesting that rhodanese, or at least the fluorophore-bearing parts of the chain, is relatively static while bound (on a timescale up to 100 ms). Studies of singly labeled molecules indicated large increases of fluorescence anisotropy upon binding to SR1, with the rate of anisotropy decay corresponding to the rotational correlation time of SR1, suggesting orientational restriction under these conditions. Thus, it seems possible that bound rhodanese is exhibiting a static heterogeneity being reported by the fluorophores. Alternatively, however, the fluorophores themselves might be driving static heterogeneity of rhodanese by affixing to the cavity wall. The findings are somewhat at difference with the more rapid local dynamic mobility of at least the portion of the rhodanese chain that is visible in the NMR studies described above. Further studies, including those with NMR, should be of help to resolving the static vs. dynamic behavior of the various apical-associated and non-associated regions of bound substrate polypeptide.

A single-molecule study of GroEL-bound DM-MBP, as with the study of bound rhodanese, observed a broad distribution of FRET efficiency in the GroEL-bound state (Sharma et al. Reference Sharma, Chakraborty, Müller, Astola, Tang, Lamb, Hayer-Hartl and Hartl2008). In this case, however, there was a bimodal distribution where both highly expanded and more compact states were populated. The highly expanded states were suggested to comprise multivalently bound substrate.

In sum, the FRET studies examining polypeptide binding support the idea of topological stretching occurring on the timescale of binding to an open ring, producing an ensemble of topologies in the bound state, as reflected by the heterogeneity of FRET efficiencies. Presumably, both stretching and the heterogeneous efficiencies of the bound state reflect the nature of hydrophobic interactions between non-native substrate and the apical domains. It is still to be established, as indicated above, how static such substrate–apical interactions might be. It remains, however, that, in the absence of measurable secondary structure in physiologic substrates, such topology effects, both the selection by GroEL from the ensemble of states by thermodynamic partitioning and stretching action mediated by multivalent binding, comprise the major unfolding actions associated with polypeptide binding, and both are apparently passive actions.

3.5.1 A power stroke dependent on the γ-phosphate of ATP drives release

The evidence for forceful release of substrate protein into the cis chamber derives from the work of Motojima et al. (Reference Motojima, Chaudhry, Fenton, Farr and Horwich2004), who followed the rate and extent of apical movement during GroES encapsulation by monitoring FRET between a fluorophore placed on the stable equatorial ‘base’ of GroEL and a fluorophore placed on the mobilizable apical face (Fig. 11a). A displacement between these points from 52 Å in an open ring to 82 Å in the GroES bound end-state was measured from the respective X-ray structures. When either ATP–GroES or ADP–GroES was incubated with labeled GroEL or with labeled SR1, both the rate and extent of excursion of the apical domains were the same, occurring in <1 s (Fig. 11b, gray traces). If polypeptide substrate was present, however, as would be the case at this phase of rigid body movement during cis complex formation (R3 and beyond, see Fig. 10), the rate of movement produced by ATP–GroES was only slightly slowed relative to the absence of polypeptide, still completing within a second, but ADP–GroES required many seconds to reach completion (Fig. 11b, black traces). In the case of SR1+ADP–GroES, this movement was unable to be completed. The result with ADP–GroES correlates with the inability of ADP–GroES to release substrate protein from the cavity wall, as measured by fluorescence studies (Weissman et al. Reference Weissman, Rye, Fenton, Beechem and Horwich1996), and with failure of any productive folding to occur despite its ability to produce an encapsulated, protease-protected substrate protein. Remarkably, polypeptide release and productive folding by such ADP–GroES complexes could be subsequently triggered by addition of aluminum fluoride, simulating the γ-phosphate of ATP, triggering both an immediate further excursion of the apical domains as detected by FRET and immediate commencement of production of the native state (Chaudhry et al. Reference Chaudhry, Farr, Todd, Rye, Brunger, Adams, Horwich and Sigler2003; Motojima et al. Reference Motojima, Chaudhry, Fenton, Farr and Horwich2004; see Fig. 11). Thus, the γ-phosphate of ATP is apparently not required in order to form a potentially productive complex in which GroES and polypeptide share apical hydrophobic surface, and this state, produced in ADP, appears to be stable over many minutes. Whether it is identical to the collision state formed on the ATP-driven pathway remains to be established. The structure of this state is unknown, and in particular, whether the ADP–GroEL–GroES–polypeptide complex has elevated apical domains producing a fully domed state, is unknown. It is unlikely that a full excursion of the apical domains could proceed, as this would completely remove the hydrophobic surface from the cavity, perforce releasing substrate protein. Thus, it would seem that at least some residual hydrophobic surface must remain exposed to retain polypeptide substrate. Whatever the extent of excursion may be, it is clear that aluminum fluoride, binding at the position of the γ-phosphate as observed crystallographically by Chaudhry et al. (Reference Chaudhry, Farr, Todd, Rye, Brunger, Adams, Horwich and Sigler2003), drives a productive release of substrate protein. Perhaps, if the complex is already substantially domed in the ADP state, this is tantamount to an energetic shaking of the cis ring, releasing the substrate protein. In energy measurements performed by Chaudhry et al. (Reference Chaudhry, Farr, Todd, Rye, Brunger, Adams, Horwich and Sigler2003), it was observed that the binding of aluminum fluoride to an ADP complex (absent substrate protein) released ~40 kcal mol⁻¹ of free energy per ring, a total energy equal to that already produced by ADP–GroES binding. So it appears possible that a power stroke in the same direction as that mediated by ATP releases the substrate in this context.

Fig. 11.

Polypeptide acts as a load on the apical domains, restricting their rate of movement when ADP and GroES are added. (a) Ribbons representations of a subunit of unliganded GroEL (top) and a subunit of the cis ring of a GroEL–GroES–ADP–AlF_x complex (bottom), colored as in Fig. 10 (note that GroES is not shown in the bottom panel). The thumbnails at the left show the position of the subunits in the intact complexes. Indicated are the two Cys residues that were labeled with a FRET reporter pair, fluorescein and tetramethylrhodamine, to monitor movements of the apical domain relative to the equatorial domain; the distances shown were measured between the β-carbons of these residues in the crystal structure (Chaudhry et al. Reference Chaudhry, Farr, Todd, Rye, Brunger, Adams, Horwich and Sigler2003). (b) Stopped-flow measurements of FRET changes on addition of GroES and ATP to GroEL–rhodanese or SR1–rhodanese complexes (left) or on addition of GroES and ADP (right). Gray traces show FRET changes of the respective chaperonins in the absence of substrate protein. The rates of change in FRET (with presence of polypeptide) were fit to the sum of two exponentials, as indicated in the individual panels. ATP supports a rapid movement of the apical domains even with polypeptide bound, with half-times (1–2 s) that are physiologically reasonable, given the ~10-s lifetime of the folding-active complex. With ADP, however, the rates are at least 20-fold slower (when substrate protein is present) and the extent of movement is reduced. When AlF_x is added to the ADP complexes to form a folding-active ATP analog in the nucleotide binding site (right), the expected extent of the FRET change is achieved at a rate consistent with the refolding rate under these conditions. From Motojima et al. (Reference Motojima, Chaudhry, Fenton, Farr and Horwich2004). Copyright 2004, National Academy of Sciences, USA.

3.5.2 Concerted substrate release

The power stroke of rigid body movement that releases substrate from the apical hydrophobic surface is virtually certainly concerted in character, involving all of the apical domains at once. This is particularly enforced by GroES binding, which, beyond the collision state, involves tightening of the interaction of all of its mobile loops in 1:1 contacts with the apical domains of the bound ring, simultaneously moving all of the apical domains. As pointed out by Ma & Karplus (Reference Ma and Karplus1999), who carried out normal mode analyses, the excursion is so substantial that steric collisions would eventuate unless there was coordinated movement of all of the domains at once. The quantitative requirement for ATP binding at the level of a ring in order to trigger (with GroES) such movements has recently been determined using a small molecule inhibitor that selectively binds in the equatorial ATP pocket of a variant GroEL subunit, blocking ATP binding. Mixed rings were formed, composed of varying numbers of variant and wild-type GroEL subunits, with only wild-type subunits able to bind ATP in the presence of compound. In order to drive substrate release and productive folding, ATP was required to occupy at least five or six and, for full efficiency, all seven equatorial domains of a ring (Chapman et al. Reference Chapman, Farr, Fenton and Horwich2008).

The symmetry of the GroEL system and its concerted behavior contrasts with the type II eukaryotic cytosolic chaperonin, CCT, which has a heterooligomeric subunit composition and a ‘built-in’ lid attached to the apical domain of each subunit. In the CCT system, a sequential as opposed to concerted release of substrate domains may occur, enabling sequential domainwise folding of such proteins as actin and tubulin (Rivenzon-Segal et al. Reference Rivenzon-Segal, Wolf, Shimon, Willison and Horovitz2005). Interestingly, GroEL binds non-native forms of these proteins and GroES encapsulates them as well, but they never reach the native state, cycling on and off the bacterial chaperonin in a futile fashion (Tian et al. Reference Tian, Vainberg, Tap, Lewis and Cowan1995).

Recent fluorescence experiments have suggested that ATP–GroES-driven polypeptide release from the apical domains occurs in an ordered fashion, with locally less hydrophobic regions of polypeptide mobilizing first and more hydrophobic segments leaving last (Sharma et al. Reference Sharma, Chakraborty, Müller, Astola, Tang, Lamb, Hayer-Hartl and Hartl2008). This, of course, would parallel the relative affinities of the hydrophobic binding surface of the GroEL apical domains for such regions of the chain. The notion that hydrophobic collapse may occur last does not really seem a likely mechanistic requirement, given the myriad of in vitro studies diluting unfolded proteins from denaturant where ‘indiscriminate global collapse’ occurs yet productive folding nevertheless ensues. Considering the experiments discussed in the next section, collapsed ensembles of GroEL substrates, however arrived at (i.e., discharged from the hydrophobic cavity wall or diluted from denaturant into free solution) seem to take the same subsequent kinetic course to the native state. In sum, the mechanism of collapse seems to have little to do with kinetic outcome, at least in the case of the substrate proteins studied to date.

3.5.3 No evidence for unfolding during the step of ATP–GroES-driven release of substrate

Finally, polypeptides undergoing release from the forcefully moving apical domains might have been subject to stretching or ‘forced unfolding’ before final release, but, as mentioned earlier, hydrogen exchange studies carried out to date do not detect significant secondary structures in bound substrates, making it difficult to evaluate whether unfolding action occurs at the level of secondary structure. But FRET studies examining topology of Rubisco (Lin et al. Reference Lin and Rye2004) and DM-MBP (Sharma et al. Reference Sharma, Chakraborty, Müller, Astola, Tang, Lamb, Hayer-Hartl and Hartl2008) actually have observed compaction and not expansion of these polypeptides during this phase of cis complex formation.