Rational redesign of the folding pathway of a modular protein

Abstract

The modular structures of repeat proteins afford them distinct properties compared with globular proteins, enabling them to function in a large and diverse range of cellular processes. Here, we show that they can also have different folding mechanisms. Myotrophin comprises four ankyrin repeats stacked linearly to form an elongated structure. Using site-directed mutagenesis, we find that folding of wild-type myotrophin is initiated at the C-terminal repeats. However, close examination of the mutant chevron plots reveals that simple models are insufficient to describe all of the data, and double mutant analysis subsequently confirms that there are parallel folding pathways. Destabilizing mutations in the C-terminal repeats reduce flux through the wild-type pathway, making a new route accessible in which folding is initiated at the N-terminal repeats. Thus, the folding mechanism of the repeat protein is poised on a fulcrum: When one end of the molecule is perturbed, the balance shifts between the different nucleation sites. The vast majority of studies on small globular proteins indicate a single, well defined route between the denatured and native states. By contrast, the potential to initiate folding at more than one site may be a general feature of repeat proteins arising from the symmetry inherent in their structures. We show that this simple architecture makes it straightforward to direct the folding pathway of a repeat protein by design.

Repeat proteins, such as ankyrin repeats, leucine-rich repeats, and tetratricopeptide repeats, consist of tandem arrays of a structural motif of 20–40 aa that stack in a roughly linear fashion, creating elongated and superhelical architectures (1–3). Repeat protein structures are composed entirely of short-range interactions, either within a repeat or in adjacent repeats, in striking contrast to the more complex topologies of globular proteins that are stabilized by contacts between residues distant in sequence (4). They are ubiquitous, representing >5% of eukaryotic proteins in the Swiss-Prot database, and highly versatile, mediating molecular recognition in many different biological processes, but little is known about their folding mechanisms. The simple, modular nature of these structures has made them uniquely successful as scaffolds for engineering novel binding specificities (5–7). Does the repeat architecture likewise give rise to distinct folding properties, and are their folding pathways also amenable to design?

In this study, we have addressed this question using the small, well behaved ankyrin (ANK) repeat protein myotrophin. The folding and unfolding properties of ANK repeats are of particular interest because these motifs are thought to play a role in mechanical signal transduction by virtue of their putative elastic properties (8, 9). The ANK repeat is a 33-residue motif consisting of a pair of antiparallel α-helices linked to the neighboring repeat via an antiparallel β-loop. The 118-residue myotrophin is composed of four ANK motifs. We find a folding mechanism in which the structure in the rate-limiting transition state is polarized at one end of the molecule. However, the wild-type folding pathway is readily perturbed by single mutations, revealing an alternative folding mechanism in which the transition-state structure is polarized at the other end of the molecule. The concept of energy landscapes suggests that a protein can have multiple folding pathways, but to date, there has been very little experimental evidence to support this view (10–14). By contrast, the accessibility of multiple routes to the native state may be a general feature of repeat proteins that arises from the symmetry inherent in their structures. Consequently, we can take advantage of their modular nature and direct the folding flux by design by manipulating the stabilities of the individual repeats.

Results and Discussion

Myotrophin has a single tryptophan located at position 8. We and others found previously that urea-induced denaturation curves obtained by monitoring fluorescence, near-UV circular dichroism, and far-UV circular dichroism superimpose with each other, and that the calorimetric and Van't Hoff enthalpies of unfolding are in good agreement, indicating that myotrophin unfolds in a two-state manner at equilibrium (15, 16). The free energy of unfolding of myotrophin in water at 25°C is 5.8 ± 0.2 kcal·mol⁻¹.

Distinct Kinetic Properties of Mutants in the N-Terminal Versus the C-Terminal Moieties of Myotrophin.

The kinetics of folding and unfolding of wild-type myotrophin monitored by stopped-flow fluorescence and circular dichroism are in good agreement with each other (16). The urea dependence of the rate constant of unfolding of the wild type shows a pronounced downward curvature at high denaturant concentrations (Fig. 1). To map out structure formation during the folding reaction, 18 conservative mutations were made throughout the protein. The mutations destabilize the protein by between 0.7 kcal·mol⁻¹ and 3.1 kcal·mol⁻¹ [supporting information (SI) Table 1]. The chevron plots of the mutants show different degrees of curvature in the unfolding arm, which can be broadly grouped according to their location in the structure (Fig. 1). Mutants at the N terminus all unfold significantly faster than the wild-type protein, and they display unfolding limbs that converge with that of wild type at high denaturant concentrations. Mutants in the two C-terminal ANK repeats, as well as those in the second helix of ANKII, unfold at a similar rate to the wild type and have unfolding limbs that are parallel to or diverge from that of the wild type.

Fig. 1.

Chevron plots of wild-type myotrophin and 18 mutants. The kinetics of folding and unfolding were monitored by stopped-flow fluorescence. The mutants are group according to location in each of the four ANK repeats. The fits of the chevron plots to the sequential barriers model are shown as a guide for the eye.

The shapes of the wild-type and mutant chevron plots point to a sequential barriers model that describes a denaturant-induced switch between discrete rate-determining transition states separated by a high-energy intermediate (17). When there is curvature in the unfolding limb only, as appears to be the case for myotrophin, the presence of two transition states (TS^early and TS^late) is the minimal model and there are consequently two limiting slopes in the unfolding limb. The data are fitted to a sequential barriers model with a metastable intermediate (I*):

where

as described (17, 18). The values of k_I-U and m_I-U were fixed to 1 × 10⁴ s⁻¹ and 0 (kcal·mol⁻¹)·M⁻¹, respectively, for ease of fitting.

We fitted the wild-type data to this model and obtained β-Tanford values, β_T, for TS^early and TS^late of 0.47 and 0.93, respectively. The values of the equilibrium constant and m obtained by using this model were in good agreement with those obtained directly from equilibrium measurements.

Deviation of the Data for C-Terminal Mutants from the Simple Kinetic Scheme.

We fitted the mutant chevron plots to the sequential barriers model using two approaches. In the first, the data set for each mutant was fitted individually. There is a problem with this approach: For some of the mutants the curvature in the unfolding limb is such that, within the urea range accessible to experiment, there are insufficient data points to describe both of the transition states. To circumvent the problem, we used a second approach in which the data for wild type and all of the mutants were fitted globally by using a shared m value for each transition state. It is clear that the slope of the folding limb for a subset of mutations, those located in the C-terminal part of the protein such as A75G, is greater than that of the wild type, indicating anti-Hammond behavior (Figs. 1 and 2) (19, 20). When using the free fitting approach, the m values obtained for the folding limb (m_f) of mutants at the C terminus are indeed significantly higher than that of the wild type; when using the global fitting approach with shared m values, this behavior is manifest in a poor fit of the data for these mutants. There was no correlation between the equilibrium m values and the m_f values, which suggests that the higher m_f values for the C-terminal mutants do not arise from ground state effects.

Fig. 2.

Chevron plots and flux plots for representative mutants throughout the structure of myotrophin. The fit of the data to the parallel pathways model is shown in black. In blue is the hypothetical chevron corresponding to the wild-type pathway (pathway a), and in purple is the hypothetical chevron corresponding to the alternative pathway (pathway b). Below the main plots are the plots of the fractional fluxes through the wild-type (blue) and alternative (purple) pathways. The wild-type data are shown in gray for comparison with the mutants.

Mutations at the C Terminus Induce Pathway Heterogeneity.

It is clear that additional complexity in the model is required to adequately describe the whole data set of wild-type and mutant chevron plots. The apparent anti-Hammond behavior observed in the refolding limb of mutant chevron plots such as A75G suggests that there are parallel folding pathways. To assess whether pathway heterogeneity is consistent with the data, we fitted the kinetics of the mutants according to the following, minimal model. Rate constants were calculated assuming two parallel pathways. The observed rate constant is given by the sum of the rate constants for each pathway:

To fit the curvature in the unfolding arm of the chevron plots, one of the pathways is assumed to follow the sequential barriers model (pathway a). For pathway b, we used the simplest, two-state folding model:

The data for wild type and all of the mutants were fitted globally by using a shared m value for each of the transition states. A number of other constraints were used. Because the ground states (native and denatured) are the same for the two pathways, the value of the free energy of unfolding for each pathway was fixed at the value obtained by equilibrium measurement. Likewise, the overall m value for each pathway was fixed at the equilibrium-derived value (calculated as the weighted average for the mutants and repeat measurements of wild type).

The fractional flux (ρ) through each pathway at a given denaturant concentration was calculated according to

and

We were able to obtain good fits for all of the mutants to this model (representative mutants shown in Fig. 2). The β-Tanford value for pathway b was 0.81. The different shapes of the chevron plots can now be rationalized in terms of the pathway heterogeneity: The plots of ρ indicate that the wild-type pathway (pathway a) is maintained across the whole range of urea concentrations when a mutation is made in the N-terminal moiety (A9G, L10V, A18G, V22A, A23G, A28G, I50A). When a mutation is made at the C terminus (L39V, A42G, A43G, A63G, A75G, V76A, V84A, T103S, A115G, L116V), pathway heterogeneity is observed, the extent of which varies with urea concentration. At low denaturant concentrations, there is significant flux through the alternative pathway (pathway b), giving rise to a steeper folding limb than that of the wild type. Because the β-Tanford values for the transition states of the two pathways are different, they are differentially destabilized as the denaturant concentration is increased, and thus the wild-type pathway dominates at higher denaturant concentrations. We can infer from this pattern of behavior that the C-terminal part of the protein is highly structured and the N-terminal part of the protein is unstructured in the transition states for the wild-type pathway, and conversely that the C terminus is only weakly structured and the N terminus is more highly structured in the transition state for the alternative pathway. It is interesting to note the behavior of the mutant S74A. This mutation in the C-terminal moiety has a stabilizing effect (ΔΔG_eq = −1.67 ± 0.01 kcal·mol⁻¹), unlike all of the other mutations. Consequently, the wild-type pathway is stabilized and pathway homogeneity is maintained, so the mutant shows behavior similar to the destabilizing mutants in the N-terminal moiety.

Structure of the Sequential Transition States for the Wild-Type Folding Pathway.

Φ Values (21) are a measure of the degree of native structure in the transition state, and they are calculated from the effect of the mutation on the free energy of the transition state relative to that of the native state. A Φ value of 1 indicates that the interactions probed by the mutation are fully formed in the transition state, whereas a value of 0 indicates that the interactions are not formed. Φ Values were calculated by using

or

where k_f^wt and k_f^mut are the folding rate constants in water for the wild type and the mutant, respectively. k_u^wt and k_u^mut are the unfolding rate constants in water for the wild type and the mutant, respectively. ΔΔG_eq is the difference in free energy of unfolding between the wild type and mutant. Φ Values were calculated for TS^early and TS^late of the wild-type pathway by using the parallel pathways model described above, and they are shown schematically in Fig. 3 and listed in SI Table 2. It is clear from the pattern of Φ values that the structure is polarized at the two C-terminal repeats in both transition states and that there is consolidation of structure on going from TS^early to TS^late. The structure of TS^early consists of residues in ANK IV, ANK III, and the first helix of ANK II. A similar extent of structure formation was observed in the folding transition state of the 4-ANK protein p16 (22). That the transition-state structure comprises more than a single ANK unit is as expected, because the ANK unit has been shown to be intrinsically unstable, whereas the interaction between units is highly favorable (23).

Fig. 3.

Mapping the structures of TS^early and TS^late of the wild-type folding pathway. Shown are Φ values for TS^early and TS^late mapped onto the structure of myotrophin (prepared by using PyMOL). Mutants with low Φ values (<0.3) are indicated in red, mutants with medium Φ values (0.3 < Φ < 0.7) are indicated in magenta, and mutants with high Φ values (>0.7) are indicated in blue.

Mutant Variants Can Be Designed to Route the Flux Through a Single Folding Pathway.

In the case of multiple folding pathways, if a mutation is made that disrupts folding through one region, there is a switch to a pathway that proceeds through other regions, and little change in the folding rate constant is observed (10, 11). This behavior is what is observed when mutations are made at the C terminus of myotrophin, such as A115G, which has an apparent Φ value of 0.3 ± 0.1. However, when this mutation is made in the context of another mutation that destabilizes the alternative pathway, the single folding pathway is maintained, and the folding rate constant is now sensitive to the destabilizing mutation (shown schematically in Fig. 4). Thus, if we first destabilize the N terminus of myotrophin sufficiently by mutation (A9G, ΔΔG_eq = 2.48 ± 0.01 kcal·mol⁻¹), then folding proceeds through the C terminus even when we make a second, moderately destabilizing mutation in that region (A115G, ΔΔG_eq = 0.81 ± 0.01 kcal·mol⁻¹) [Fig. 5(top section) and SI Tables 3 and 4]; the Φ value, calculated from the folding rate constants, for the mutation A115G in the context of A9G is 0.9 ± 0.1.

Fig. 4.

Redesigning the folding pathway of myotrophin. The folding mechanism of myotrophin can be represented by analogy with weighing scales. The two folding pathways are shown schematically. Which pathway is followed is weighted by the relative stability of the two ends of the protein, represented as different sizes of weights on the scales that tip the balance accordingly. Mutations were designed to shift the flux through one or another folding pathway and to probe the transition-state structure for that pathway. Top left, pathway a: For the wild type, the scales are tipped in favor of pathway a. Structure formation at the N terminus can be probed by using a destabilizing mutation that does not perturb the flux. Top right, pathway a: A highly destabilizing mutation at the N terminus offsets the effect of a moderately destabilizing mutation at the C terminus made to probe structure formation in pathway a. Bottom left, pathway b: By greatly stabilizing the N terminus and destabilizing the C terminus, the flux is redirected through pathway b. The structure of the N terminus in pathway b can now be probed by using a moderately destabilizing mutation. Bottom right, pathway b: Structure formation at the C terminus can be probed by using a moderately destabilizing mutation in the context of the highly stabilized N terminus.

Fig. 5.

Analyzing the parallel folding pathways of myotrophin by using double, triple, and quadruple mutants. The top section depicts probing structure formation at the N and C termini of myotrophin in the wild-type folding pathway [mutation A9G (red) in the context of the wild-type protein (black), and the mutation A115G (blue) in the context of A9G (black)]. The bottom section depicts probing structure formation at the N and C termini of myotrophin in the alternative folding pathway [mutation A9G (blue) in the context of E17VD20LA115G (black), and the mutation A115G (red) in the context of E17VD20L (black)]. The transition states are represented schematically with one structured ANK motif at either the N or the C terminus; they are not intended as quantitative descriptions of the transition-state structures.

Redesigning the Folding Pathway: Stabilizing the N-Terminal Region of Myotrophin Routes the Flux Through the Alternative Folding Pathway.

We next designed a series of mutations to probe the alternative pathway (Fig. 4). We first mutated two positions, 17 and 20, in the most N-terminal ANK repeat, to the residues of the consensus ANK sequence. Both mutations stabilized the protein (ΔΔG_eq = −1.7 ± 0.2 kcal·mol⁻¹ and −0.5 ± 0.2 kcal·mol⁻¹, respectively), and the effect of the double mutant E17VD20L was approximately additive with respect to the single mutants, resulting in a stabilization of 2.32 ± 0.01 kcal·mol⁻¹. Because repeat structures are dominated by local interactions, the double mutant should stabilize only the N-terminal region of the protein and should not stabilize the C-terminal region; we reasoned, therefore, that this stabilizing double mutant should reroute the flux through the alternative folding pathway, since the single-mutant analysis had indicated the N terminus to be structured in the transition state for this pathway. Next, by making mutations in the context of the double mutant, we should be able to probe the structure in the transition state of the alternative pathway [Fig. 5 (bottom section) and SI Tables 3 and 4]. Φ Values were calculated by using the unfolding data because significant rollover was observed in the refolding arm of the chevron plots of these mutants (see discussion below); however, similar Φ values were obtained by using the refolding data outside of the region where there is curvature. The Φ value obtained for the mutation A115G at the C terminus (ΔΔG_eq = 0.89 ± 0.08 kcal·mol⁻¹) in the context of E17VD20L is 0.0 ± 0.1. The chevron plots obtained for mutations toward the N terminus indicated that the pathway flux was not homogeneous at all urea concentrations, presumably because these mutations destabilized the alternative pathway sufficiently for the wild-type pathway to become partially populated. We therefore sought to destabilize the wild-type pathway by making an additional, destabilizing mutation at the C terminus (A115G), and then we made an N-terminal mutation in the context of the triple mutant E17VD20LA115G to obtain a Φ value. Thus, the Φ value obtained for A9G (ΔΔG_eq = 1.93 ± 0.08 kcal·mol⁻¹) in the context of E17VD20LA115G is 0.8 ± 0.1. We can also calculate a Φ value for D20L (with the caveat that this type of mutation is not ideal for calculation of Φ) in the context of E17V. The Φ value of 1.0 ± 0.1 obtained for D20L concurs with that obtained for A9G, indicating a highly structured N-terminal moiety in the transition state of the alternative folding pathway.

Rollover is observed in the refolding arm of the chevron plots of these mutants and this could arise from the presence of an intermediate; however, we could not show unequivocally that the tail-off in the rate constants at low urea concentrations was not simply an artifact due to underestimation of the very fast folding rates (≈500 s⁻¹) because these rates were above the limits accessible to our stopped-flow apparatus.

It is clear from Fig. 5 that the chevron plot of each multiple mutant has the same shape as the pseudo-wild type with which it is compared, consistent with the pathway flux being the same in the two variants. The shape of the chevron plots of those mutants that have a stabilized N terminus is significantly different from the shape of the wild-type chevron plot, which is consistent with a change in the folding pathway. Furthermore, for the latter mutants the slope of the unfolding arm gives a β-Tanford value (0.8) that is within error the same as the value obtained for the alternative pathway from the global fit of wild type and mutants to the parallel pathways model.

Summary

In summary (Fig. 5, middle section), kinetic analysis of double and triple mutants of myotrophin reveals that the protein can fold along two distinct pathways. These mutants show that the structure is polarized at opposite ends of the molecule in the transition states for the two pathways. Thus, the folding mechanism of the ankyrin repeat protein is poised on a fulcrum (Fig. 4): A mutation destabilizes one end of the molecule, causing folding flux through the other end of the molecule. The initiation site for folding of wild-type myotrophin does not correlate with intrinsic helical propensity [as predicted by the program AGADIR (24, 25)], which is below 5% for all residues except those in the second helix of ANK II for which the propensity rises to ≈15%. Interestingly, folding appears to be initiated preferentially at the terminal rather than internal repeats of myotrophin. Theoretical studies predict this behavior also for other ankyrin repeat proteins, as a result of the lower entropic cost of folding the terminal repeats (26).

Experimental studies indicate that folding transition states of small globular proteins are represented by relatively homogenous ensembles of structures (excluding parallel folding reactions that result from heterogeneity in the denatured state, due to proline isomerization for example). Moreover, the folding mechanisms of globular proteins are generally robust; only a drastic change in the energetic balance, such as by circular permutation, can in some cases shift the folding nucleus from one part of the structure to another (27–30). By contrast, the experimental results for myotrophin presented here and for the 7-ANK protein Gankyrin (R. D. Hutton, A.R.L., and L.S.I., unpublished results) and the 12-ANK protein D34 (N. D. Werbeck and L.S.I., unpublished results) reveal parallel folding routes of similar energy, and simulations carried out on a number of ANK proteins also predict pathway heterogeneity (26). We therefore suggest that the potential to initiate folding at more than one site may be a general feature of repeat proteins that arises from the symmetry inherent in their structures. Consequently, the folding pathway of a repeat protein can easily be designed by taking advantage of the modular structure and manipulating the stabilities of the individual repeats.

Materials and Methods

Protein Purification.

A synthetic gene encoding myotrophin was cloned into a pRSET(A) vector (Novagen) modified for the expression of a six-histidine N-terminal fusion protein with a thrombin cleavage site, as described in ref. 16. Site-directed mutagenesis was performed by using the QuikChange kit (Stratagene) and was confirmed by DNA sequencing. Protein expression and purification were carried out as described in ref. 16. The buffer used for all subsequent measurements was 50 mM Tris·HCl (pH 7.5) containing 1 mM DTT, and the protein concentration was 2 μM.

Equilibrium and Kinetic Measurements.

Protein stability was determined by equilibrium urea-induced denaturation monitored at 25°C by fluorescence on an Aminco Bowman spectrophotometer and analyzed as described in refs. 16 and 31. The equilibrium data for wild type and mutants were fitted globally, sharing the m value. The change in the free energy of unfolding upon mutation (ΔΔG_eq) was calculated as the product of the m value and the change in the midpoint of unfolding. Stopped-flow fluorescence measurements were recorded at 25°C on an Applied Photophysics SX-18MV instrument as described in ref. 16.