Triggering Protein Folding within the GroEL-GroES Complex*

The folding of many proteins depends on the assistance of chaperonins like GroEL and GroES and involves the enclosure of substrate proteins inside an internal cavity that is formed when GroES binds to GroEL in the presence of ATP. Precisely how assembly of the GroEL-GroES complex leads to substrate protein encapsulation and folding remains poorly understood. Here we use a chemically modified mutant of GroEL (EL43Py) to uncouple substrate protein encapsulation from release and folding. Although EL43Py correctly initiates a substrate protein encapsulation reaction, this mutant stalls in an intermediate allosteric state of the GroEL ring, which is essential for both GroES binding and the forced unfolding of the substrate protein. This intermediate conformation of the GroEL ring possesses simultaneously high affinity for both GroES and non-native substrate protein, thus preventing escape of the substrate protein while GroES binding and substrate protein compaction takes place. Strikingly, assembly of the folding-active GroEL-GroES complex appears to involve a strategic delay in ATP hydrolysis that is coupled to disassembly of the old, ADP-bound GroEL-GroES complex on the opposite ring.

To fold, many essential proteins require the assistance of specialized molecular chaperones known as chaperonins (1). The GroELS chaperonin system of Escherichia coli is one of the best-studied examples of the chaperonin class of molecular chaperones (for review see Refs. 2,3). GroEL is a tetradecamer of fourteen identical 57-kDa subunits arranged into two heptameric rings (4). Each ring contains a large, central, open cavity, and the two rings are stacked back-to-back to create a double toroid. Maximally efficient folding of most proteins that possess a strict dependence on GroEL (so-called stringent substrate proteins) requires that they be encapsulated within a closed chamber formed by a GroEL ring and the separate cochaperonin GroES (5)(6)(7)(8). Binding of the GroES lid over a captured substrate protein seals the GroEL cavity and releases the non-native protein into the confined space of the GroEL-GroES chamber (the cis complex). Protein folding proceeds in this isolated space for several seconds until the GroEL-GroES complex is dismantled and the substrate protein, folded or not, is ejected into free solution (5)(6)(7)(8)(9)(10).
For stringent substrate proteins, ATP is required for the assembly of the GroEL-GroES cis complex and the associated steps of substrate protein encapsulation, release and folding (5,(11)(12)(13). Binding of ATP to a GroEL ring drives a large-scale rearrangement of the GroEL subunits, resulting in a dramatic elevation and rotation of the GroEL apical domains away from the central cavity (14,15). GroES binding and substrate folding within the cis complex depend upon these ATP-induced structural rearrangements (5,16,17). However, despite a wealth of structural and biophysical information, precisely how assembly of the GroEL-GroES cis complex leads to substrate protein encapsulation, release, and folding remains poorly understood. Because they bind to the same part of a GroEL ring, how does GroES binding lead to substrate protein encapsulation before the non-native protein escapes into solution? In addition, how is substrate protein release and folding coordinated with ATP hydrolysis?
Large, stringent substrate proteins not only bind to sites on the GroEL apical domains that overlap with the binding sites for GroES but are also likely to fill and spill out of the open GroEL ring (14, 16, 18 -22). This raises a serious issue: how does GroES locate and bind to its binding sites on a GroEL ring without the substrate protein prematurely escaping into solution? In fact, ATP must bind before GroES, and the same, highly cooperative structural rearrangements of the GroEL ring that lead to GroES binding also weaken the interaction between the substrate protein and the GroEL ring (13,23,24). Based on a detailed examination of changes in Trp fluorescence of an engineered GroEL mutant, it has been suggested that GroES loading and substrate encapsulation could occur prior to substrate protein release (25). In this case, however, a small and non-stringent substrate protein was employed and conformational changes in the substrate during encapsulation could not be assessed. Other work has suggested that premature substrate release is prevented by a secondary timer within the assembled GroEL-GroES complex, implying that the initiation of ATP hydrolysis and substrate release is tightly coupled (26).
Here we have examined these questions using a novel, chemically modified mutant of GroEL that has a small organic probe attached to an engineered Cys residue at the bottom of the GroEL cavity. Designated EL43Py, this modified GroEL variant displays a much higher rate of steady-state ATP hydrolysis than wild-type GroEL (wtGroEL) 3 but is dramatically compromised in its ability to fold stringent GroEL substrate proteins. We demonstrate that EL43Py populates, but fails to efficiently exit, a key intermediate allosteric state of the GroEL ring. We show that the ATP-driven formation of this intermediate conformation of the GroEL ring is necessary not only for GroES binding, but also for the forced unfolding of the non-native substrate protein. Premature escape of a large, non-native substrate protein from the GroEL ring is inhibited through the formation of this intermediate allosteric state, which can simultaneously bind both substrate protein and GroES with high affinity. Additionally, even though population of this allosteric state is sufficient to commit the bound ATP to hydrolysis, ATP turnover is delayed during a normal GroEL folding cycle. This slowing in ATP hydrolysis is linked to the disassembly of the ADP-bound GroEL-GroES complex on the opposite ring and ensures that GroES binding and substrate encapsulation are complete prior to the initiation of ATP hydrolysis.

EXPERIMENTAL PROCEDURES
Proteins-The GroEL variants containing single Cys residues S43C (EL43C), A529C (EL529C), A535C (EL535C), and E315C (EL315C) were created using standard site-directed mutagenesis techniques in the background of a cysteine-free version of GroEL (27). All GroEL variants were expressed and purified using a combination of previously described methods (5,27,28; see also supplemental "Methods"). Wild-type GroES, GroES98C, wild-type Rubisco and the various Rubisco Cys mutants were expressed and purified as previously described (5,27,28). Bovine rhodanese was purchased from Sigma and purified as previously described (10; see also supplemental "Methods"). Hexokinase, creatine kinase, rabbit pyruvate kinase, lactate dehydrogenase, and protease K were purchased from Roche Applied Science.
Labeling of GroEL and Rubisco with Fluorescent Dyes-Labeling of GroEL and Rubisco variants was performed as previously described (27,29). The thiol-reactive dyes used in this study were N-(1-pyrene)maleimide, 5-iodoacetamidofluorescein, 5-(2-acetamidoethyl)aminonaphthalene 1-sulfonate, and tetratmethylrhodamine 5-iodoacetamide. All dyes were obtained from Invitrogen and were prepared fresh from dry powder in anhydrous dimethylformamide immediately prior to use. Following separation from unreacted dye, EL43Py was further purified over a Sephacryl-300 column. The extent and specificity of dye conjugation was examined as previously described (29 -31). Unless otherwise stated, the extent of dye conjugation to specific Cys residues was between 95 and 100% for GroEL and Rubisco variants. The labeling efficiency of ES98F was ϳ15% (ϳ1 dye/heptamer).
Refolding and Enzymatic Assays-The refolding of Rubisco and rhodanese was assayed essentially as previously described (2,5,10). See supplemental "Methods" for additional details. The measurement of steady-state ATP hydrolysis was conducted using a coupled enzymatic assay (28,32). The change in absorbance at 340 nM was measured at 25°C using an Amersham Biosciences Pharmacia Ultrospec 3100 pro spectrophotometer. Pre-steady-state measurements of ATP hydrolysis were conducted with a rapid-mixing, quench-flow apparatus (KinTek, Austin, TX) using [␥-32 P]ATP. Time points were quenched with 4 M formic acid, and samples were analyzed and quantified by PhosphorImager analysis (Amersham Biosciences Storm 860, Sunnyvale, CA) following TLC separation of free phosphate, ADP, and ATP (33,34).
Asymmetric GroEL-GroES Complex Formation and Protease Protection-The binding of Rubisco to the trans ring of a GroEL-GroES complex was conducted by first creating an asymmetric GroEL-ADP-GroES complex (see supplemental "Methods"). For protease protection experiments, the asymmetric GroEL-ADP-GroES complex was mixed (100 nM GroEL and 350 nM GroES, final) with acid-urea-denatured fluoresceinlabeled Rubisco (Rub454F, 100 nM). The sample was then supplemented with either buffer or ATP (1 mM) and rapidly mixed (within ϳ5 s) with hexokinase (0.05 unit/l) and glucose (20 mM). Protease K (0.05 g/ml) was added to each sample, aliquots at each time point were removed, and the protease was deactivated with phenylmethylsulfonyl fluoride (0.2 mg/ml). Samples were then run on SDS-PAGE, and the level of undigested Rubisco was determined by quantifying the amount of intact and fluorescent Rubisco using a Storm 860 system. For the protease experiments shown in Fig. 3C, GroES binding was blocked at different times following the addition of ATP (1 mM) with excess SR1 (600 nM). In all cases, the ATP was depleted after 5 s with hexokinase and glucose. Samples were supplemented with proteinase K (0.9 g/ml) for 2 min, and then separated and quantified.
Gel Filtration-Analytical gel filtration of chaperonin complexes was conducted with a Superose 6 (GE Healthcare Systems) column connected to a high-performance liquid chromatograph (Waters, Milford, MA) configured with an in-line fluorescence detector. All component mixing was performed in 50 mM Hepes (pH 7.6), 10 mM KOAc, 5 mM Mg(OAc) 2 , 2 mM dithiothreitol (Buffer A). For experiments examining the dissociation rate of GroES, the column running buffer was 50 mM Hepes, pH 7.6, 100 mM KOAc, 5 mM Mg(OAc) 2 , 200 M ADP, 2 mM dithiothreitol. For both wtGroEL and EL43Py, excess unlabeled GroES (4 M) was added to mixtures of each chaperonin containing fluorescently labeled GroES and ATP after 2 min of steady-state cycling. At various times, the ATP was rapidly quenched with hexokinase and glucose, and the samples were loaded onto the Superose gel-filtration column. The amount of free and chaperonin-bound fluorescent GroES was then measured.
ADP Release-ADP bullets with radiolabeled ADP in the cis ring were made by mixing GroEL (7 M), GroES (14 M), and radiolabeled ATP (245 M, and 20 Ci of [␣-32 P]ATP) in Buffer A. The mix was incubated for 10 min at 25°C to form ADP bullets (see supplemental "Methods"). The ADP bullets were then diluted to 100 nM and mixed with denatured Rubisco (100 nM) and incubated at 25°C for 10 min to allow non-native Rubisco binding to the trans ring. Cold ATP (1 mM) or buffer was then added, and at varying times the solution was applied to a Zeba micro spin desalting column (Pierce) and centrifuged at maximum speed (16 ϫ 10 3 relative centrifugal force) in a tabletop microcentrifuge for 4 s. The sample flow-through was analyzed with a scintillation counter (Wallac).
Stopped-flow Fluorescence and Data Analysis-Stoppedflow experiments were performed essentially as previously described (29), using an SFM-400 rapid mixing unit (BioLogic, Claix, France) equipped with a custom designed, two-channel fluorescence detection system. The time-dependent change in donor-side FRET efficiency of the labeled Rubisco monomer was extracted from matched sets of donor-only and donor-acceptor stopped flow experiments as previously described (29). Fitting of experimental data were accomplished with either Igor Pro (Wavemetrics, Portland, OR) or Origin (OriginLab, Northampton, MA).

RESULTS
EL43Py Is a Fully Productive, but Slow, Protein-folding Machine-We recently carried out a targeted screen for amino acid positions at the base of the GroEL cavity that could be used to attach small fluorescent probes. We introduced a series of conservative Cys mutations into a cysteine-less GroEL background (27) at locations around the bottom of the GroEL cavity (Fig. 1A). Labeling one of these GroEL mutants (S43C) with the organic fluorophore N-1-pyrenemaleimide (Fig. 1B) serendipitously resulted in a GroEL variant (EL43Py) that demonstrated unexpected alterations in assisted protein folding. While EL43Py productively folds substrate proteins like Rubisco (Fig.   1C) and rhodanese (Fig. 1D), it does so much more slowly than wtGroEL.
We reasoned that a modified chaperonin like EL43Py might represent a powerful new tool for studying the mechanism of GroELmediated protein folding, especially given the ongoing debate about the importance of the GroEL C-terminal tails, which are located in a similar region at the base of the GroEL cavity (35)(36)(37). To establish the utility of EL43Py, we therefore examined whether the folding deficiency of EL43Py can be explained by a trivial and inhibitory interaction between exogenous dyes localized at the bottom of the GroEL cavity and non-native folding intermediates. We employed two strategies for this analysis: 1) conjugation of different, modestly hydrophobic dyes to the same EL43C position and 2) development of additional GroEL variants with labeled Cys residues at nearby but distinct positions within the GroEL cavity. When we attached a range of chemically and physically dissimilar substituents to position 43, most showed little or no alteration in assisted protein folding (see supplemental "Methods"). For example, modification of the 43C position with the moderately hydrophobic dye AEDANS (Fig. 1B), results in a GroEL variant (EL43Ed) that displays no significant alteration in Rubisco folding (Fig. 1C). More importantly, conjugation of pyrene to unique Cys residues at positions 529 and 535 results in chaperonins (EL529Py and EL535Py) that, while possessing pyrene dye rings at the base of the GroEL cavity similar to EL43Py, show no significant perturbation in Rubisco folding ( Fig. 1C and supplemental Fig. S1). EL43Ed and EL529Py do display a somewhat reduced capacity to refold rhodanese ( Fig.  1D), suggesting that the presence of the dyes can have a modest influence on the folding of certain proteins. However, this effect is relatively small in comparison to the more dramatic effects observed with EL43Py. Overall, these observations strongly suggest that the perturbed folding behavior displayed by EL43Py is not primarily caused by inhibitory interactions between the pyrene dyes and a substrate protein.
EL43Py Executes a Full Hydrolytic Reaction Cycle More Quickly Than GroEL-Because efficient protein folding by GroEL depends upon a carefully coordinated, ATP-driven reaction cycle (Fig. 2), we next examined how EL43Py binds and hydrolyzes ATP. Unexpectedly, EL43Py demonstrates a substantially elevated rate of steady-state ATP hydrolysis, both in  S1). Thus the reduction in protein folding observed with EL43Py appears to be linked to a substantial increase in steadystate ATP hydrolysis. Interestingly, EL43Py displays an allosteric response to ATP that is very similar to that exhibited by wtGroEL (supplemental Fig. S3). At low ATP concentrations, where ATP binding occurs to only one ring of the tetradecamer complex (38 -40), EL43Py displays a positive cooperative transition that is very similar to wtGroEL. EL43Py also displays changes in ATP hydrolysis at higher concentrations of ATP that are similar to wtGroEL (supplemental Fig. S3), whereas EL43Py displays a consistently higher steady-state rate of ATP turnover at high ATP concentrations. Overall, these observations suggest that ATP binding, and the attendant structural changes that lead to hydrolysis, are similar in EL43Py and wtGroEL, but that EL43Py completes a cycle of ATP hydrolysis (Fig. 2, Phase II) more quickly than wtGroEL.
The elevated rate of steady-state ATP hydrolysis by EL43Py implies that the rate-limiting step of the EL43Py reaction cycle is different from that exhibited by wtGroEL (Fig. 2). To establish the nature of this difference, we first examined the intrinsic, single-turnover rate of ATP within the EL43Py-GroES cis complex. Using a rapid mixing, quench-flow apparatus, we directly measured the rate of ATP turnover for both the EL43Py-GroES and wtGroEL-GroES complex (Fig. 3B), employing only enough ATP to support one round of GroES binding and ATP hydrolysis (Fig. 2, Phase I).
Remarkably, the k cat values observed for the EL43Py and GroEL cis complexes are essentially identical (0.11-0.13 s Ϫ1 per active subunit) and are well matched to the values obtained from the steady-state hydrolysis data in supplemental Fig.  S3, as well as measurements from other studies (5,28,38). Thus, the increase in steady-state ATP turnover by EL43Py cannot be due to a stimulation of intrinsic ATP hydrolysis within the cis complex.
We next examined the presteady-state ATP hydrolysis kinetics of EL43Py and wtGroEL, in the presence of GroES, and at levels of ATP sufficient to allow multiple rounds of GroES association and release (Fig. 3C). With wtGroEL, we observe a burst of ATP hydrolysis followed by a linear, steady-state reaction phase. The burst is a direct consequence of the rapid formation of a committed, GroEL-ATP-GroES cis complex (Fig. 2, Phase I), followed by the slow decay of the post-hydrolysis GroEL-ADP-GroES complex upon ATP binding to the trans ring and initiation of the steady-state cycle (Fig. 2, Phase II (9,27,38). The pre-steady-state kinetics of ATP hydrolysis by EL43Py are distinct from wtGroEL. Although EL43Py also displays a burst of ATP hydrolysis that is followed by a linear steady-state reaction, the amplitude of the burst from EL43Py is smaller than that observed with wtGroEL while the linear phase is significantly faster (Fig. 3C). In the context of a simple two-step kinetic model that describes the pre-steadystate hydrolysis burst of GroEL (see supplemental "Methods"), these observations suggest that the rate-limiting transition that controls cis complex disassembly (Fig. 2, rate-limiting step) proceeds much more quickly with EL43Py than wtGroEL. This conclusion further predicts that EL43Py should release GroES more rapidly than wtGroEL in the course of a steady-state reaction cycle. We tested this prediction using an assay capable of tracking the dissociation rate of GroES during the course of a steady-state GroEL hydrolysis reaction (Fig. 3D). In strong sup- The hydrolytic reaction is divided into two phases, where the first phase (Phase I) illustrates the pre-steady-state behavior of the system when apo-GroEL is first mixed with ATP and GroES. An apo-GroEL ring first fills cooperatively with ATP, followed by rapid binding of GroES (2, 3). The resulting asymmetric GroEL-ATP-GroES complex (an "ATP bullet") hydrolyzes the bound ATP with an intrinsic turnover rate of 0.12 s Ϫ1 , yielding an asymmetric GroEL-ADP-GroES complex (an "ADP bullet" (38)). When ATP is limiting, the reaction stops at the ADP bullet following a single hydrolytic turnover. In the presence of excess ATP, the binding of ATP to the open trans ring of the ADP bullet initiates the disassembly of the cis complex, leading to release of GroES and ADP (5,9,27). Nearly simultaneous binding of another GroES heptamer (indicated by the brackets) results in the formation of a new cis complex, regenerating the ADP bullet upon hydrolysis of the ATP within the cis complex (5,9,27). The steady-state cycling between ATP and ADP bullets constitutes the second phase of the reaction (Phase II), and disassembly of the ADP-bound cis complex is the rate-limiting step of this cycle in the absence of substrate protein (27,38). Under in vivo conditions, the GroEL-GroES system persists almost exclusively in the steady-state cycle (Phase II). port of our model, the rate of GroES release from EL43Py is 2.6-fold faster than that observed with wtGroEL, an increase that is very similar to the 2.3-fold increase in the steady-state hydrolysis rate of EL43Py in the presence of GroES.
EL43Py Fully Supports Protein Capture and Encapsulation Beneath GroES-EL43Py thus displays two key characteristics: 1) a considerably faster hydrolytic cycle and 2) a substantially slower rate of assisted protein folding. The behavior of EL43Py could thus be interpreted as satisfying a key prediction of one model of GroEL action, in which the primary stimulatory action provided by GroEL derives from substrate protein confinement within the GroEL-GroES cavity. A faster reaction cycle in this model should result in a reduced cis-cavity lifetime and therefore slower folding. The strength of this conclusion, however, rests upon whether EL43Py: 1) captures non-native substrate protein, 2) encapsulates the substrate protein beneath GroES, and 3) releases the substrate protein into the GroEL-GroES cavity as efficiently as wtGroEL. We therefore examined whether EL43Py correctly binds non-native substrate protein.
Because the open trans ring of a GroEL-ADP-GroES complex (an "ADP bullet") captures the vast majority of the non-native substrate proteins of GroEL in vivo (8,27,41), we tested whether an EL43Py ADP bullet can capture non-native Rubisco. Based on both gel filtration and protease protection assays (Fig. 4, A and C), EL43Py displays no observable defect in its ability to bind non-native Rubisco. Additionally, the extent of bindingdriven, passive unfolding of the Rubisco monomer (29) is the same for both EL43Py and wtGroEL (Fig.  4B). EL43Py thus displays no deficiency in substrate protein binding, and the average conformational states of the bound substrate protein on an EL43Py and wtGroEL ring are similar.
We next examined the capacity of an EL43Py ring to support non-native substrate protein encapsulation beneath GroES. Fluorescently labeled, non-native Rubisco bound to the trans ring of an ADP bullet, made from either wtGroEL or EL43Py, is highly susceptible to rapid proteolysis (Fig. 4C) (31). However, when additional GroES and ATP are added, followed by rapid enzymatic depletion of the ATP to prevent cycling and allow only a single round of GroES binding to the substrate-occupied trans ring, the non-native Rubisco becomes resistant to proteolysis (Fig. 4C). Notably, EL43Py appears to be as efficient in GroES capture and Rubisco encapsulation as wtGroEL (Fig. 4D). Based upon these and other protease protection experiments (see supplemental "Methods"), as well as complementary gel filtration measurements (data not shown), GroES encapsulation of nonnative Rubisco on the trans ring of EL43Py bullets is 80 -95% efficient within the 45-to 60-s window of these measurements.
EL43Py Fails to Release Substrate Protein into the GroEL-GroES Cavity-We next examined the ability of EL43Py to release non-native Rubisco in the presence of ATP and GroES. If EL43Py executes normal cycles of substrate protein binding and release (9, 10), the addition of excess wtGroEL to an EL43Py reaction should out-compete the EL43Py and facilitate normal folding. Strikingly, the addition of a large excess of wtGroEL to an EL43Py-Rubisco folding reaction does not result in normal Rubisco folding (Fig. 5A). This observation suggests that EL43Py does not correctly execute a normal GroEL cycle of substrate protein binding and release.
To more directly examine substrate protein release with EL43Py, we developed a single round encapsulation and release assay based on a fluorescently labeled Rubisco variant whose conformational state can be monitored by fluorescence reso-FIGURE 3. EL43Py completes a hydrolytic cycle more quickly than wtGroEL. A, the steady-state hydrolysis of ATP by wtGroEL, EL43Py, EL43C, EL529Py, and EL43Ed are shown in the presence of excess ATP. In each case, 250 nM GroEL tetradecamer was mixed with 500 nM GroES and 0.5 mM ATP, and the rate of ATP hydrolysis was monitored. B, in limiting ATP, the single-turnover kinetics of ATP hydrolysis by wtGroEL and EL43Py are identical. The rate of ATP hydrolysis from a single ring of wtGroEL and EL43Py was examined by rapidly mixing 3 M chaperonin and 12 M GroES with a limiting amount of radiolabeled ATP (21 M) in a quench-flow apparatus at 25°C. The single round hydrolysis data were fit to a single-exponential rate law with k cat ϭ 0.110 Ϯ 0.006 s Ϫ1 for wtGroEL and k cat ϭ 0.128 Ϯ 0.008 s Ϫ1 for EL43Py. Error bars represent the standard deviation of n ϭ 3 experimental replicates. C, the pre-steady-state kinetics of ATP hydrolysis by EL43Py in the presence of GroES show a reduced burst amplitude. The ATPase activity of wtGroEL and EL43Py (3 M) in the presence of GroES (6 M) was examined by rapidly mixing each chaperonin with excess radiolabeled ATP (1 mM) in a quench-flow apparatus at 25°C. Error bars represent the standard deviation of n ϭ 5 experimental replicates. The pre-steady-state bursts observed with wtGroEL and EL43Py, obtained by extrapolation of the linear reaction phase to zero time, are illustrated with dashed lines. D, the release of fluorescently labeled GroES (ES98Fl, 200 nM) from wtGroEL and EL43Py (220 nM) during steady-state ATP turnover was monitored by gel filtration. In both cases, the dissociation of GroES was observed to follow a single-exponential decay with rate constants of k obs ϭ 0.030 Ϯ 0.003 s Ϫ1 for wtGroEL and k obs ϭ 0.079 Ϯ 0.004 s Ϫ1 for EL43Py. Error bars represent the standard deviation of n ϭ 3 experimental replicates. NOVEMBER 14, 2008 • VOLUME 283 • NUMBER 46 nance energy transfer (FRET, Fig. 5B) (29,31). Release of the Rubisco monomer into the enclosed GroEL-GroES cavity is observable as a distinctive FRET signature in single round encapsulation experiments (29,31). Following encapsulation and compaction of the non-native Rubisco intermediate on the trans ring of a wtGroEL bullet complex, the release and folding of the Rubisco monomer can be detected as an increase in distance (decrease in FRET efficiency) between the labeled domains of the monomer as it matures inside the wtGroEL-GroES complex (Fig. 5, B and C). When the same experiment is conducted with EL43Py, by contrast, even complete GroES binding does not lead to substrate release and folding, as shown by the persistently high FRET signal observed when Rubisco is enclosed in the EL43Py-GroES complex (Fig. 5C). This result is consistent with a failure of EL43Py to release the vast majority of the non-native Rubisco from the cavity walls despite full substrate encapsulation following ATP and GroES binding.

The GroEL Protein Folding Trigger
EL43Py Only Slowly Executes a Key ATP-driven Conformational Change-The ATP-driven conformational shift of the EL43Py ring thus appears to stall at a critical point following GroES binding. Completion of this step seems to be central to the ability of GroEL to trigger protein release and folding. Previous kinetic studies suggested that an ATP-saturated GroEL ring populates an intermediate conformation that simultaneously binds substrate protein and GroES (25). EL43Py could thus provide additional evidence for the existence of this state, as well as offer the opportunity to trap and study this critical and otherwise fleeting intermediate conformation in detail. We therefore sought to determine whether the allosteric transitions of an EL43Py ring differ from those of wtGroEL. Because we previously demonstrated that the conformation of a bound Rubisco monomer is linked to the conformational state of the GroEL ring (29, 31), we employed labeled Rubisco in a FRET-based assay to examine the conformational shift of the wtGroEL and EL43Py rings (Fig.  6A). Upon ATP and GroES binding to the Rubisco-occupied trans ring of a wild-type ADP bullet, we observed an initial decrease in FRET efficiency, followed by a rise (Fig. 6B). The initial drop in FRET efficiency, reflecting a rapid increase in the distance between the labeled segments of the Rubisco monomer, is the result of a forced unfolding event linked to the ATPdriven movement of the GroEL apical domains (31). The subsequent rise in FRET efficiency is caused by the compaction of the non-native Rubisco monomer as GroES binds and encapsulates the substrate protein (29,31). Qualitatively, ATP and GroES binding to the trans ring of an EL43Py bullet leads to a similar sequence of Rubisco conformational modifications (Fig. 6B). However, the amplitude of the FRET signal change seen with EL43Py is only one third that observed with wtGroEL.

. EL43Py correctly binds and encapsulates non-native substrate protein beneath GroES.
A, binding of a non-native substrate protein to the trans ring of an EL43Py ADP bullet complex is comparable to binding to the trans ring of a wtGroEL ADP bullet complex. Each bullet complex (120 nM) was mixed with denatured, fluorescently labeled Rubisco (100 nM) for 5 min at 25°C, and the samples were then loaded onto a gel-filtration column. The elution profile of each sample is shown. The elution position of the fluorescent, non-native Rubisco bound to the trans rings of each bullet complex is indicated. The total integrated area of each peak in arbitrary units (a), reflecting the amount of non-native Rubisco bound and retained by each complex over the course of the column run are indicated in parenthesis. B, the conformational state of the bound Rubisco monomer was examined by time-resolved, donor-side FRET using a previously described fluorescently labeled Rubisco variant (29,31). In each case, fluorescently labeled Rubisco samples were denatured in acid-urea and mixed (100 nM) with either EL43Py or wtGroEL ADP bullets (120 nM) to allow trans ring binding (shown schematically in the inset). The donor intensity decay curves for non-native Rubisco carrying either donor-only (fluorescein; D) or donor and acceptor probes (fluorescein/rhodamine; D ϩ A) are shown. The essentially identical decay curves demonstrate that the observed FRET efficiency of the labeled Rubisco monomer and, therefore, the average conformation of the non-native protein, on the two trans rings is the same. C, GroES encapsulates non-native Rubisco on an EL43Py ring. Denatured, fluorescein-labeled Rubisco (Rub454Fl, 100 nM) was bound to the trans ring of both GroEL and EL43Py ADP bullets (100 nM), formed in the presence of excess GroES. The protease sensitivity of the non-native Rubisco was examined either following (cis-3°) or without (trans) the subsequent addition of ATP, permitting GroES binding. To prevent cycling and allow a single round of GroES binding, added ATP was quenched after 5 s with hexokinase and glucose. Samples were supplemented with proteinase K for varying times and then subjected to SDS-PAGE. The location of the non-native Rubisco protein in the cis and trans complexes is shown schematically above the gel. The quantity of undigested Rubisco was examined using a laser-excited fluorescence gel scanner. The amount of intact Rubisco was quantified and plotted as a function of time in D. Error bars represent standard deviation of n ϭ 4 experimental replicates.
The shallower FRET decrease seen with EL43Py could, importantly, be the result of a slowing in the forced conformational expansion of Rubisco on the EL43Py ring. With wtGroEL, the initial unfolding phase, and its associated decrease in FRET, occurs much faster than GroES binding, which causes an increase in FRET. The two steps are kinetically separable and therefore fully observable with wtGroEL. In the case of EL43Py, however, if the step that leads to substrate unfolding slows and proceeds at a rate that is similar to the rate that GroES binds, then the average change in FRET efficiency will be much lower, because the drop in FRET caused by unfolding in a subset of the population would be simultaneously canceled by the binding of GroES to another subset. If correct, this implies that the ATP-driven conformational transition of the GroEL ring that results in forced protein unfolding precedes, and is required for, GroES binding. To test these ideas, we conducted a series of experiments where the consequences of ATP binding alone to the substrate-occupied trans ring can be observed (Fig. 6C). For these experiments, we prevented substrate protein encapsulation by adding a large excess of a single ring version of GroEL (SR1) as a GroES trap (8). SR1 efficiently binds both ATP and GroES but cannot release the bound GroES under the conditions of this experiment. In contrast to the simple, fast change observed with wtGroEL, the FRET signal measured with EL43Py displays a biphasic decrease upon ATP binding (Fig. 6D). The most rapid phase occurs at a rate very similar to wtGroEL (t1 ⁄ 2 ϭ 0.2 s), whereas the slowest phase, constituting ϳ65% of the total observed amplitude, proceeds much more slowly (t1 ⁄ 2 ϭ 13.9 s). These observations strongly suggest that roughly two-thirds of the EL43Py bullet population only slowly completes the ATP-driven allosteric transition that unfolds the substrate protein and permits GroES binding. Two additional observations are consistent with this conclusion. First, GroES binding and substrate encapsulation on an EL43Py trans ring requires more time to reach completion than does GroES binding to a wtGroEL trans ring (Fig. S4). Second, experiments that directly probe the conformational shift of the chaperonin ring by following changes in pyrene fluorescence suggest that the ATP-driven allosteric transition of the EL43Py ring is slowed (supplemental Fig. S5). Importantly, pyrene fluorescence has previously been used to map the allosteric transitions of GroEL (38). At low pyrene substitution ratios, where alterations in assisted protein folding and ATPase rate cannot be detected (supplemental Fig. S1), ATP-induced changes in pyrene fluorescence are very rapid. However, these same fluorescence changes become significantly slower in the fully labeled EL43Py (supplemental Fig. S5). Additionally, a rise in pyrene fluorescence observed at low pyrene substitutions, most likely reflecting a subsequent structural transition of the GroEL ring, appears to fail completely at high pyrene substitution (supplemental Fig. S5).
ATP Hydrolysis by a Newly Formed GroEL-GroES Complex on a trans Ring Is Delayed-Our observation of slow GroES binding to the EL43Py trans ring raises an important question: if a GroEL ring commits to hydrolysis immediately after binding ATP, why does substrate protein encapsulation with EL43Py succeed at all? For GroES to bind to a GroEL ring and encapsulate a large substrate protein like Rubisco, the ring must remain filled with ATP (2, 3). Thus, if ATP hydrolysis proceeds unimpeded, and the rate of GroES binding is slowed sufficiently, then GroES binding should fail as the EL43Py ring prematurely depletes its store of ATP. However, despite a substantial slowing in the rate of GroES binding, the EL43Py trans ring FIGURE 5. EL43Py fails to release non-native Rubisco upon GroES and ATP binding. A, the addition of excess wtGroEL to an EL43Py folding reaction does not rescue Rubisco folding. Denatured Rubisco (100 nM) was mixed with EL43Py (250 nM), and the sample was then supplemented with a 5-fold excess of wtGroEL (1.25 M). Folding was initiated by the addition of ATP (2 mM) and excess GroES (3 M, EL43Py rescue). The refolding curves of Rubisco (100 nM) in the presence of either wtGroEL or EL43Py (250 nM in both cases, with 500 nM GroES) are shown for reference. B, schematic of a FRET experiment designed to examine conformational changes in Rubisco following a single round of GroES binding to the trans ring of an ADP bullet. Fluorescently labeled, nonnative Rubisco (29,31) was first bound to the trans ring of an ADP bullet in the presence of excess GroES. Subsequent addition of ATP permits GroES binding and Rubisco encapsulation. To prevent cycling and permit long time observation of the complex, the excess ATP was quenched with hexokinase and glucose (Hex/Glc) ϳ 4 s after the addition of ATP. C, the change in FRET efficiency was monitored as a function of time following GroES binding to the Rubisco-occupied trans ring of wtGroEL and EL43Py ADP bullets. Rubiscobound ADP bullet complexes in the presence of excess GroES were manually mixed (60 nM) with ATP (1 mM) in a standard fluorometer cuvette with magnetic stirring. Excess ATP was then quenched with hexokinase and glucose to prevent cycling. This mixing sequence precludes observation within the first ϳ8 s of the experiment. displays no loss in its ability to bind GroES (Figs. 4 and supplemental Fig. S4). This observation implies that ATP hydrolysis by a GroEL trans ring is inhibited or slowed until assembly of a new cis ternary complex is complete. The observation that the stretched conformation of the Rubisco monomer on an ATP-bound, wtGroEL trans ring is stable for 7-10 s (Fig. 6D) is also consistent with retarded ATP hydrolysis. Ten seconds is enough time for the majority of the bound ATP to hydrolyze to ADP at the intrinsic turnover rate of a GroEL ring (Fig. 3B). Hydrolysis should lead to a relaxation of the ring and a change in Rubisco conformation. By contrast, the stretched Rubisco conformation stably persists for several seconds, suggesting that the rate of ATP hydrolysis by the trans ring is reduced. We tested this prediction by examining the pre-steady-state kinetics of ATP hydrolysis by the ADP bullet trans ring (Fig. 7). To push the ADP bullet through the rate-limiting transition of the GroEL cycle (27), we saturated the trans ring with non-native Rubisco prior to the addition of ATP. Upon the addition of ATP, we observe a near total loss of the pre-steady-state burst of ATP hydrolysis from the ADP bullet trans ring (Fig.  7C). For apo-GroEL, the burst of hydrolysis is caused by slow decay of the post-hydrolysis GroEL-ADP-GroES complex following the faster step of phosphodiester bond cleavage (Fig. 2) (9,27,38). However, the nearly linear pre-steadystate behavior seen with the ADP bullet strongly suggests that ATP hydrolysis by the ADP bullet trans ring is, at a minimum, dramatically slowed in comparison to hydrolysis by an apo-GroEL ring.

DISCUSSION
Using a novel GroEL variant, we have examined the sequence of events that lead to substrate protein release and folding inside the GroEL-GroES cis cavity (Fig. 8). We have shown that a chemically modified mutant of GroEL, EL43Py, binds non-native substrate protein, ATP, and GroES and initiates a substrate protein encapsulation reaction. EL43Py fails, however, to efficiently complete the steps of the process needed to trigger substrate protein release and folding, allowing us to examine a key but poorly understood allosteric state of the GroEL ring. EL43Py executes a series of non-productive and nearly futile GroES binding and release cycles that only rarely result in substrate protein release and folding. The failure of the EL43Py ring to populate the folding-competent conformation of the cis ring, in addition, bypasses the slowest step in the GroEL reaction cycle (GroES release from a folding competent, ADP-bound asymmetric complex (27,38)), resulting in a faster chaperonin cycle and an elevated rate of steady-state ATP turnover (Figs. 3 and 8). The perturbations exhibited by EL43Py further allowed us to identify a central and previously unrecognized role for ADP dissociation in controlling the progression of the GroEL reaction cycle.
Chemical Modification and Altered Allostery in GroEL-Our results with EL43Py highlight the utility of tethering small molecules to a target protein as a method for probing complex allosteric systems. The site-specific chemical attachment of exogenous functional groups offers a much wider array of physical and chemical moieties than afforded by standard mutagenesis. Indeed, this basic idea has been developed over the last several years into a rational drug discovery strategy that has yielded a set of novel small molecules that allosterically modulate the activity of several caspases (42)(43)(44). Although the creation of EL43Py was serendipitous, this modified chaperonin has nonetheless provided several very important insights into the complex allostery of GroEL. We do not currently understand how the addition of particular substituents at position 43 causes the observed alterations of the allosteric transitions of GroEL. The dye moiety at position 43 is relatively distant from the GroEL ATPase site, and the GroEL equatorial domains overall display only small conformational shifts upon binding ATP (14,45). However, the loop that contains position 43 does  31 and are re-plotted here for reference. C, schematic of a stopped-flow FRET experiment designed to follow conformational changes in a substrate protein upon the binding of ATP alone to the trans ring (31). For this experiment, an excess of the single ring GroEL variant SR1 is used as a GroES trap (8). D, the allosteric transition triggered by ATP binding that causes forced unfolding proceeds slowly on an EL43Py trans ring. Stopped-flow FRET experiments were conducted in essentially the same manner as in B, except that SR1 was present in a 20-fold excess (1.2 M). The inset shows the FRET change over the first second of data. The decrease in FRET efficiency, reflecting an increase in the intra-probe distance, was fit to a double exponential rate law for EL43Py with rate constants of k fast ϭ 2.9 Ϯ 0.4 s Ϫ1 and k slow ϭ 0.05 Ϯ 0.02 s Ϫ1 for EL43Py. The wtGroEL data shown are from Ref. 31 and are re-plotted here for reference. must occur as a direct result of simple GroES binding before the protein is released. The compaction of the Rubisco monomer appears, therefore, to reflect something akin to an actual compression event and not a relaxation or "snapping back" of an expanded substrate protein upon its release into the cis cavity.
Synchronizing ATP Hydrolysis and Protein Release-Our study of EL43Py also sheds light on how GroEL's folding trigger is coordinated with the overall progression of the chaperonin ATPase cycle. It has been suggested that the transition from the R 2 to the R 3 state of the ATP-saturated GroEL ring is responsible for the release of the substrate into the enclosed GroEL-GroES cavity (25). Our observations with EL43Py are consistent with this model and indicate that EL43Py stalls in the R 2 state, rarely making the transition to the R 3 state. Strikingly, despite failing to populate the R 3 state, ATP hydrolysis by an EL43Py ring appears to proceed normally (Fig. 3B). This observation suggests that the cis ring ATP commits to phosphodiester bond cleavage prior to population of the R 3 state. Our observations with EL43Py are thus not consistent with the GroEL cycle being controlled by a double timer as has been suggested (26). The two-timer model was formulated as a way to explain how GroES binding to an open GroEL ring could occur with-out premature substrate protein release. This model proposes that ATP hydrolysis is not possible until the R 3 transition is complete and the substrate protein is released into the GroEL-GroES cavity. However, EL43Py rarely populates the R 3 state but hydrolyzes ATP with singleturnover kinetics that are identical to wtGroEL (Fig. 3B), suggesting that commitment to ATP hydrolysis is achieved prior to formation of the R 3 state of the ring, most likely in either the R 1 or R 2 states.
Nonetheless, we do find strong evidence that the initiation of ATP hydrolysis by a newly formed cis complex is subject to a key delay during a steady-state reaction cycle. The ability of the EL43Py trans ring to complete GroES binding, despite a slowing of the transition that makes this binding reaction possible (Figs. 4, 6, and S4), suggests that ATP hydrolysis is substantially slowed on a trans ring that is in the process of forming a new cis complex. The dramatic reduction in the pre-steady-state burst of ATP hydrolysis by a wtGroEL trans ring also strongly supports this conclusion (Fig. 7).
How could such a slowdown in hydrolysis be encoded? In the course of a typical GroEL reaction cycle, the assembly of a new GroEL-ATP-GroES complex on a trans ring is linked to the ejection of GroES and ADP from the other ring, because the cis cavity left over from the previous phase of the cycle is dismantled (27). Binding of both ATP and non-native substrate protein to the trans ring causes GroES to be released from the other GroEL ring very rapidly (27). Additionally, ADP bound to one GroEL ring has been shown to be an effective non-competitive inhibitor of ATP hydrolysis on the other ring (46). Thus if ADP release is delayed behind GroES and substrate protein release by even a few seconds, ADP release could function as a simple and effective throttle, slowing the rate of ATP hydrolysis by the other ring. Under our experimental conditions, we observe this effect as a dramatic slowing of ATP hydrolysis by a newly formed cis complex (Fig. 7C). At higher and more physiologic ADP concentrations, where non-competitive inhibition of ATP hydrolysis is stronger (46), ADP could function more like a true hydrolysis delay gate.
To test this model, we examined the rate of both GroES and ADP release from an ADP bullet under the same conditions (Fig. 7D). Using a previously described FRET assay (27), we first confirmed that binding of non-native Rubisco and ATP to the trans ring of an ADP bullet causes rapid (t1 ⁄ 2 ϳ 0.5 s) dissociation FIGURE 8. The stages of the protein folding trigger of GroEL. A simplified reaction cycle is shown (left), illustrating the sequential progression of allosteric states (R 1 , R 2 , and R 3 ) of the ATP-saturated GroEL cis ring required to trigger protein folding. Initial binding of ATP leads to population of the R 1 state. The R 1 state was not directly observed in the current work but has been previously described (25). The transition from the R 1 to the R 2 state, involving the directed movement of the GroEL apical domains, is responsible for forced substrate unfolding (1). The movement of the apical domains also permits GroES binding and the subsequent compaction of the non-native protein (2). The R 2 state simultaneously binds the substrate protein and GroES (2 and 3). Following GroES binding to the R 2 state, the GroEL-GroES complex undergoes an additional transition to the R 3 state, which releases the substrate protein into the stable GroEL-GroES cavity and initiates folding. ATP hydrolysis in the cis ring is delayed by an event on the opposite GroEL ring, most likely ADP release, allowing GroES binding to reach completion. The stages of substrate, ATP, and GroES binding for EL43Py are similar to wtGroEL. However, the EL43Py ring only rarely populates the R 3 state, usually stalling in R 2 (3, left arrow). Because the bound ATP is committed to hydrolysis, and these EL43Py rings usually fail to populate the foldingcompetent allosteric state of the GroEL ring (R 3 ), the slowest step of the GroEL steady-state ATPase cycle (GroES and ADP release from the folding competent GroEL-GroES cis complex) is usually bypassed by EL43Py, resulting in faster steady-state ATP hydrolysis and much less efficient folding.