The HX MS Folding Experiment.

When denaturant-unfolded Escherichia coli MBP is diluted into folding conditions not complicated by aggregation (21) (concentration 1 μM, unfolding by acid urea, no Cl⁻, pH 7.4, 22 °C; more details in SI Appendix), it rapidly collapses (<40 ms) into a dynamic polyglobular ensemble and then more slowly folds to its native state (29). The course of folding was probed by exposure to a brief high D₂O acidity (pD) H-to-D HX-labeling pulse. This strategy labels, in a structure-sensitive way, folding intermediates that are present at any time during folding as the protein moves to its native state. Amide sites in segments that are not yet protected by H-bond formation at the time of the D-labeling pulse become fully deuterated (heavier) while sites that are already folded and protected remain undeuterated (lighter). To resolve the positions of the HX-labeled sites, samples were quenched (pH 2.4, 0 °C) to halt further exchange and injected into an online flow-analysis system (SI Appendix, Fig. S1). The flow system proteolytically fragments the protein, then separates and analyzes the resulting peptides, roughly by reverse-phase liquid chromatography and then at high-resolution by MS.

MS/MS analysis identified 519 peptide fragments, including 375 unique peptides (SI Appendix, Fig. S2). We chose 92 that redundantly cover the entire length of the 370-residue MBP protein and could always be measured with high quality when analyzed by the ExMS program (27). Examination of the many H/D-labeled peptides reveals the identity of folding intermediates and the folding rate, sequence of formation, stability, and dynamic properties of the protein segments that they represent. The comparison of many overlapping peptides provides increased structural resolution and internal consistency checks. HX MS experiments were supplemented by the observation of global MBP folding kinetics by intrinsic tryptophan fluorescence with and without ATP turnover.

Spontaneous Folding of MBP^WT.

Fig. 1 illustrates pulse-labeling HX MS data for the folding of a peptide (22–44) that covers one segment of an initial MBP folding intermediate and another peptide (170–193) that monitors a segment that folds much later. The bimodal isotopic MS envelopes indicate that, during folding, each segment switches from an unfolded unprotected condition (heavier) to folded and protected (lighter) in a concerted reaction, as has been found for other proteins (30⇓–32). At each folding time point, the fraction of the protein population that has already folded any particular segment can be read out as the fraction of the lighter (protected) peptide isotopic envelope. The measurement has high accuracy because all of the necessary information, the relative envelope areas, is wholly contained in each bimodal MS trace.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

MBP folding. Upper panels show the time-dependent folding of two MBP segments observed by HX MS H-to-D pulse labeling (20-ms pulse). One segment (22–44) participates in the MBP early folding intermediate (∼1 s half-time), the other (170–193) folds much later (∼40 s). (A) The WT protein in free solution (collapsed ensemble). (B) The V9G mutated variant (MBP^V9G) in free solution. (C) MBP^V9G encapsulated in GroEL^D398A. Lower panels track the folding of the early intermediate (blue) and global folding (multicolor), measured by many peptides.

Pulse-labeling data for many peptides placed within the early intermediate, the blue curves in Fig. 1, trace its time-dependent folding. In MBP^WT, the half-time for formation of the early intermediate in free solution is 1.2 s (Fig. 1A; 1.0–1.5 s total range among the various peptides). The protein segments that form the H-bonded intermediate structure, residues 22–44 and 264–279, are remote in the amino acid sequence but they are in direct contact in the native protein and fold at identical rates, implying that they come together into a relatively native-like configuration, as has been seen for folding intermediates in other proteins (31, 32). The nearly identical behavior of many overlapping peptides provides multiple consistency checks. The difference in folding rate compared with a previous report (29) is due to the different conditions used (7 s half-time at pD 9 vs. 1.2 s at pH 7.4 used here).

The rest of the protein encounters a large kinetic barrier and folds much more slowly. Unlike the tightly determined time course for the many peptides that show formation of the early intermediate, the slow-folding segments that track later global folding are spread over some width on the time axis, with half-times of 20–40 s, suggesting more than one folding step. The suggestive groupings indicated in color represent sets of protein segments that are adjacent in the native protein (SI Appendix, Fig. S3). Sequential multistep folding pathways have been seen for other proteins (31, 32). However, in the present case, a large kinetic barrier after formation of the early intermediate is essentially rate-limiting, which makes the separation of later steps ambiguous.

Examination of the cognate structure of the early folding intermediate in the native protein shows that it is stabilized, in part, by a system of 24 hydrophobically interacting side chains, shown in space-filling representation in Fig. 2. The hydrophobic cluster contains the protein segments just noted, the sequentially distant sequences 22–44 and 264–279, which form the H-bonded HX-protected early folding intermediate, and also the neighboring 9–21 and 45–60 segments which fold later. Six amino acid substitutions are known to slow MBP folding (34, 35); all are placed within the segments pictured. These segments all exhibit weak protection against HX labeling in the initial collapsed ensemble, well before the later formation of the strongly protected H-bonded intermediate just described. Many other segments also exhibit weak protection in the collapsed ensemble. However, unlike the other peptides, the segments pictured in Fig. 2, some near and some far in the sequence, are all destabilized by a single mutation at sequence position Val9 (see below). These several observations consistently indicate that some approximation to this structure preexists as a preintermediate in the collapsed ensemble.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

The preintermediate hydrophobic cluster. The extensive hydrophobic side-chain interactions that stabilize the preintermediate in the initial collapsed ensemble are shown in space-filling representation as they are seen in the native protein. They are as follows: L8, V9, I10, W11, I12, L21, V24, F28, I34, V36, V38, L44, F48, V51, I60, I61, F62, W63, L276, F280, L281, Y284, L285, and L286.

Spontaneous Folding of MBP^V9G.

We prepared MBP with the substitution Val9Gly (MBP^V9G) placed within the cluster in Fig. 2. When MBP^V9G is in free solution, the early intermediate folds more slowly, by 20-fold with a half-time of ∼25 s (Fig. 1B). As expected for an obligatory on-pathway intermediate, they remain the first elements to fold (blue segments in Fig. 1B). Because subsequent steps are much slower, global folding is delayed by about the same absolute amount, making it slower by approximately twofold.

The same HX MS methods were used to observe the folding of MBP^V9G when it is inside the GroEL/ES cavity (Fig. 1C). In these experiments, a modification of the online analysis system was used to remove GroEL and GroES before the MBP proteolysis step to avoid overwhelming the MS analysis with many peptide fragments (SI Appendix, Fig. S1). The additional time needed in the online peptide analysis (10 min) adds trivially to D-back exchange during the analysis (SI Appendix, Fig. S4).

Encapsulation restores the fast folding rate of the early intermediate. Average measured half-time is 1.1 s (0.9–1.3 s total range) compared with 1.2 s for folding of MBP^WT in free solution. This result is not masked by the time (∼1 s) required to ensure initial encapsulation (10, 36). The course of folding is followed for much longer times (Fig. 1). To ensure that MBP^V9G remains in the cavity and folds only therein during the experiment, we used a mutant GroEL, D398A, which hydrolyzes ATP so slowly (∼30 min half-time) (37⇓–39) that ATP turnover and substrate protein cycling do not occur during the folding experiment. ATP binding triggers substrate protein encapsulation with GroES capping, and folding occurs in the static cavity (Fig. 1C) without substrate protein cycling. The results therefore measure the effect of static encapsulation alone.

The Role of Substrate Cycling.

To directly compare the effect of encapsulation with and without substrate cycling, we employed intrinsic MBP^V9G fluorescence (Fig. 3). These experiments compare the global folding of MBP^V9G in GroEL^WT with ATP-dependent substrate cycling to folding in GroEL^D398A without cycling. (GroEL/ES has no tryptophan and no interfering fluorescence.) In both cases, global folding of MBP^V9G returns to the rate found for MBP^WT in free solution. ATP-dependent substrate cycling does not produce faster folding than static encapsulation alone. Fig. 3 also shows the absence of any folding when MBP^V9G is bound to the GroEL apical domains before encapsulation.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Global folding by fluorescence. The global folding of mutant MBP^V9G followed by the intensity of intrinsic tryptophan fluorescence when free in solution (blue), when bound to the GroEL apical domains (red), and in the GroEL cavity with ATP-dependent cycling (in WT GroEL; orange) and without cycling (in mutant D398A GroEL; purple). GroEL/ES has no tryptophan and therefore no interfering fluorescence. (Excitation at 295 nm, observation through a 305-nm long-pass filter, Biologic SFM400 stopped flow; Grenoble.) AU, arbitrary units.

The approximately twofold acceleration of MBP^V9G global folding, which might otherwise be thought to represent a direct GroEL effect, is fully accounted for as an indirect effect of the recovered fast formation of the prior on-pathway intermediate. Encapsulation itself, with or without ATP-dependent cycling, has no effect on the large kinetic barrier that limits the MBP global folding rate. Similarly misleading behavior can be expected for any other case that ignores pathway intermediates and focuses only on the effect of GroEL on the final global folding step.

The Collapsed Ensemble.

To understand the GroEL mechanism, we pursued a more thorough analysis of MBP folding. Protein folding studies often detect a fast initial collapse of an unfolded protein into a molten globule type of ensemble before folding. Much attention has been directed at the condition of structure and stability in the collapsed ensemble, its possible role in guiding subsequent native-state formation, and the relationship with various physical measurements [FRET, small-angle X-ray diffraction, circular dichroism (CD)] (40, 41).

In previous work, HX protection has been connected to specific amide hydrogen bonding in native secondary structures (42, 43) and commonly exhibits protection factors in the range of Pf ∼10³ and far higher. These numbers indicate free energy of stabilization over 4 kcal/mol [protection factor (Pf) = HX = HX reference rate for fully unprotected amides (44)/measured HX rate]. We report here HX MS observations on the structure and dynamic properties of MBP in the collapsed ensemble where many peptides display a much lower level of HX protection.

To assess weak HX protection, unfolded MBP^WT was diluted into folding conditions to populate the collapsed ensemble and then exposed to increasingly long 10- to 80-ms H-to-D labeling pulses. Representative peptides are shown in Fig. 4. The segment 22–44 is a part of the fast intermediate. In the 0.2-s delay before the labeling pulse, ∼10% have already formed the well-protected intermediate (left envelope), as expected (t_1/2 = 1.2 s). The major not-yet-folded population fraction (unprotected, heavy) can be separately observed. Some amides exchange within the weakest 10-ms pulse (Pf < 5). A few sites are more resistant but become labeled by longer pulses, emphasized by the blue and red vertical dashed lines, which serves to quantify their low-level protection (Pf < 50, stabilization free energy <2 kcal/mol). Many other peptides, like 45–62 and 179–196, also show low-level protection in the collapsed ensemble. A few, like 293–302, have little or no residual protection.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Low-level HX protection in the collapsed ensemble. Unfolded MBP^WT was diluted into folding conditions and brief H-to-D labeling pulses of increasing length were applied (10–80 ms at pD 9.8, 22 °C where intrinsic free peptide half-time is 0.5–5 ms). Weak HX protection is seen for hydrophobic cluster peptides, like 22–44 (A), which forms the early H-bonded intermediate (∼10% folding has already occurred), and for 45–62 (B), which does not, and also for many noncluster peptides (C) but not for some others (D). Vertical lines are added to aid visualization of increasing deuteration with longer but still weak labeling pulses. Other peptides with EX1 character are shown in SI Appendix, Fig. S5.

Peptides in four sequence regions (90–114, 150–160, 240–265, 341–351) show broadened or clearly bimodal HX EX1 isotopic envelopes (SI Appendix, Fig. S5). EX1 kinetics measures cooperative segmental unfolding reactions, suggesting α-helical secondary structure due to burial in the collapsed state (45). The EX1 peptides do have high hydrophobic content, but other segments that have high hydrophobicity do not show EX1 behavior (SI Appendix, Fig. S6). They seem likely to produce CD and other spectroscopic signals in the collapsed ensemble, although they are mainly in β-sheet conformation in the native MBP protein. Even the apparently structured EX1 segments show only weak stability; the EX1 folded to unfolded ratio extrapolated to zero pulse-labeling length (K = 1) indicates a stabilization free energy close to zero. Interestingly, encapsulation often slows the EX1 unfolding rate as well as other EX2 HX rates (SI Appendix, Fig. S5B). Also interesting is the fact that these EX1 peptides, with some structure in the unfolded state, do not contribute to the early intermediate.

Preintermediate in the Collapsed Ensemble.

The V9G mutation has no effect on measurable HX in the collapsed ensemble, with one notable exception. It selectively abolishes the weak residual HX protection of the hydrophobic cluster segments that will become strongly protected in the early intermediate (22–44, 264–279) (Fig. 5B). Other peptides that monitor these same protein segments confirm the effect. Cluster segments that will fold much later (9–21, 45–62) are partially but not completely destabilized (SI Appendix, Fig. S7). All other peptides are wholly unaffected.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

Intermediate and preintermediate formation. WT and mutant MBP were exposed to H-to-D labeling pulses of increasing length, either in free solution (A and B) or in the cavity of noncycling GroEL^D398A (C). (A) After 0.2 s of folding (delay time before HX pulse), H-to-D pulse labeling shows that a small fraction of two hydrophobic cluster segments in MBP^WT has folded (fast folding rate as expected from Fig 1A), and the not-yet-folded fraction has some low-level protection indicating presence of the preintermediate. The vertical lines emphasize the response to increasing but still weak labeling pulses. (B) The V9G mutation selectively abolishes the low-level stability of the preintermediate peptides and eliminates the fast folding of the early intermediate (as in Fig. 1B). (C) Encapsulation restores low-level stability in the preintermediate peptides and fast formation of the early intermediate (more folding due to a 2-s delay before the pulse, needed to allow full encapsulation). (D and E) Line plots summarizing the response—added D labeling—of the hydrophobic cluster peptides to the increasingly long HX-labeling pulse for the following: MBP^WT in free solution (blue), MBP^V9G in free solution (red), and MBP^V9G in GroEL (orange). Normalized centroid means the degree of H-to-D labeling plotted as the fractional D value between the protected native value (lighter) and the unprotected unfolded value. This normalization cancels any effect of a difference in back exchange between different runs. (See SI Appendix, Fig. S7 for other cluster peptides.) (F) Line plot for a typical noncluster peptide not affected by the V9G mutation but nevertheless endowed with HX protection by encapsulation.

This observation requires that, well before the formation of the HX-protected intermediate (1.2 s; Fig. 1A), the hydrophobic cluster segments, although distant from each other in sequence, are already spatially grouped together immediately adjacent to the Val9 position. The preintermediate cluster appears to be held together by the cooperative more or less native-like interaction of its many hydrophobic side chains (Fig. 2), even before it clicks into its more strictly correct format enforced by the more demanding H-bond constraints observed by strong HX protection.

The Role of Encapsulation.

In Fig. 1, the main variable is folding time. Fig. 5 extends those observations by varying pulse strength to measure the weakly stable preintermediate. There is a short delay before the H-to-D labeling pulse is applied, 0.2 s for the collapsed ensemble (Fig. 5 A and B), and 2 s for MBP^V9G in GroEL^D398A (Fig. 5C; to ensure full encapsulation). During the delay time, some population fraction folds and some does not. The resulting bimodal envelopes separately measure the population fractions already folded and not yet folded. The folded fraction is ∼10% of the MBP^WT population in A, zero for MBP^V9G in B, and over half for encapsulated MBP^V9G in C. These results match the folding results in Fig. 1.

The pulse-labeling MS data in Fig. 5 also measure HX protection of the still-unfolded population fraction. These molecules are not affected by the prior folding time (Fig. 1). The hydrophobic cluster peptides are weakly protected in A, destabilized in B, and reprotected in C. The relative stability of the preintermediate is displayed more quantitatively by the line plots in Fig. 5 D and E and in SI Appendix, Fig. S7. The curves plot the fractional D-uptake of the not-yet-folded peptides as a function of pulse time. For each noted peptide, these curves compare the amount of D-labeling found for MBP^WT in free solution (blue; from panel A), for MBP^V9G in free solution (red; from panel B), and when MBP^V9G is statically encapsulated in the GroEL^D398A cavity (orange; from panel C). In the collapsed ensemble of MBP^WT in free solution, the cluster segments show their characteristic weak protection (blue), and they fold rapidly. They are selectively destabilized by the V9G substitution (red) and then fold more slowly. Restabilization upon encapsulation (orange) restores their fast folding.

A noncluster peptide is also shown (Fig. 5F), typical of many others that display weak protection in the collapsed ensemble. Their red and blue curves are identical because they are not destabilized by the V9G mutation. Like the cluster peptides, encapsulation stabilizes them (less D-labeling). Unlike the cluster peptides, their folding rate is not affected by encapsulation. Thus, the stabilization seen for the mutant preintermediate is due to a general nonselective encapsulation-dependent effect.

In these experiments, the preintermediate was identified by virtue of the selective destabilization of the hydrophobic cluster peptides by the V9G mutation. The placement of the mutation was guided by knowledge of the early on-pathway MBP folding intermediate (Fig. 2). Does the weak protection seen for many other peptides in the collapsed ensemble reflect similar native-like association or simply random hydrophobic association? Some more general mutational scanning together with appropriate HX methods might help to illuminate this interesting question.