A Designed System for Assessing How Sequence Affects α to β Conformational Transitions in Proteins*

The role of amino acid sequence in conformational switching observed in prions and proteins associated with amyloid diseases is not well understood. To study α to β conformational transitions, we designed a series of peptides with structural duality; namely, peptides with sequence features of both an α−helical leucine zipper and a β−hairpin. The parent peptide, Template−α, was designed to be a canonical leucine-zipper motif and was confirmed as such using circular dichroism spectroscopy and analytical ultracentrifugation. To introduce β−structure character into the peptide, glutamine residues at sites away from the leucine-zipper dimer interface were replaced by threonine to give Template−αT. Unlike the parent peptide, Template−αT underwent a heat-inducible switch to β−structure, which reversibly formed gels containing amyloid-like fibrils. In contrast to certain other natural proteins where destabilization of the native states facilitate transitions to amyloid, destabilization of the leucine-zipper form of Template−αT did not promote a transformation. Cross-linking the termini of the peptides compatible with the alternative β−hairpin design, however, did promote the change. Furthermore, despite screening various conditions, only the internally cross-linked form of the parent, Template−α, peptide formed amyloid-like fibrils. These findings demonstrate that, in addition to general properties of the polypeptide backbone, specific residue placements that favor β−structure promote amyloid formation.

Gross conformational transitions (or switches) in proteins are increasingly coming to light. Broadly speaking these may be classed into two types: those that have evolved to tailor or elicit specific normal protein functions, and those that lead to aggregated forms that render proteins defunct or even pathogenic. Examples of the first group include the large structural changes of certain viral-coat proteins that accompany virushost membrane interactions (1). The other type of structural change is associated with the prions and proteins that form amyloid (2,3).
It is accepted that misfolding of peptides and proteins cause the fibrillar aggregates known as amyloid, which characterize the diseases collectively known as amyloidoses (3)(4)(5). Therefore, the elucidation of the underlying molecular principles for the transformation of soluble proteins into amyloid has potential for understanding and tackling these diseases.
A diverse set of peptides and proteins form amyloid (6,7). This set is not limited to peptides and proteins that form amyloid deposits in vivo and are associated with the various diseases (8,9), and even non-natural, designed peptides and proteins can assume amyloid-like structures (10 -12). In these cases, the structural changes vary from slow conversions of random-coil peptides to gross structural changes of larger, natively folded, globular structures brought about by destabilization of the native state. Furthermore, for the latter, there are no clear themes in the types of native state that undergo transitions to amyloid: all-␣-helical proteins can be transformed into amyloid (13), as can all-␤-structures (14) and structural types between these extremes (6,7). Nonetheless, amyloid and amyloid-like structures have common cores based on ␤-structure (15,16): fibrils are assembled from protofilaments in which ␤-strands are aligned perpendicular to the long fiber axis. Thus, an increasingly adopted view is that the ability to form amyloid is largely a general property of the polypeptide backbone (9,17). However, the role, if any, of protein sequence in amyloid fibril formation is not clear.
A number of studies indicate that sequence, and therefore amino acid side chains, do influence the formation of amyloid (10, 18 -22). However, the question is whether sequence changes simply affect the relative stabilities of the various folded, partly folded, and unfolded states of the subject proteins and hence their propensity to aggregate (18,20,21,23); in this sense, the role of sequence in the formation of the amyloid structure itself may be regarded as passive. Alternatively, sequence may take a more active role in promoting ␤-structured elements in the unfolded states or within the amyloid-fibril structures themselves (10,19,22).
We are interested in addressing a specific issue in protein conformation switching, namely, how sequence brings about and influences the rather extreme ␣ to ␤ structural transitions in proteins. To do this, we set out to design a peptide with a structural conflict; namely, with sequence features compatible with both ␣-helical and ␤-hairpin structures. Such a system would allow us to assess the role of specific side chain placements in effecting the conversion between the two structures; in addition, if the ␤-structured form acted as precursor for amyloid formation, such a system may provide insight into the mechanism(s) of conversion to amyloid-like structures.
Others have engineered peptides and proteins with structural conflicts. Minor and Kim (24) show that an 11-residue peptide sequence can be accommodated at structurally distinct * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. sites within the same protein fold; in this case, overall tertiary context overrides local sequence and secondary structure preferences. At the level of a whole protein, Dalal and Regan (25) have met the Paracelsus Challenge and succeeded in transmuting a mixed ␣/␤ protein fold into an all ␣Ϫhelical fold by altering only Ϸ50% of the sequence. In addition, several groups have succeeded either serendipitously, or with reasoned designs to construct peptide sequences that do switch conformational state (11, 12, 26 -29).
To elucidate specific sequence features of natural proteins that drive ␣ to ␤ secondary structure switching, we designed and characterized a series of peptides in which positive design features for a dimeric, ␣-helical coiled coil and a ␤-hairpin were superimposed in a short sequence. The parent peptide, Template-␣, was confirmed as a stable, cooperatively folded, dimeric, helical structure consistent with the designed leucine zipper. As expected, mutation of three exterior Gln residues to Thr reduced the stability of this folded state. Surprisingly, however, thermal unfolding of the mutant was accompanied by conversion to ␤-structure on a time scale of tens of minutes. In this state the samples gelled and were shown to contain fibrils with tinctorial properties and the morphology of amyloid. Small sequence changes were made to probe the basis of the conversion to amyloid. We conclude that straight destabilization of the coiled-coil structure does not necessarily foster the change, but alterations geared to favor the alternative ␤-conformation do promote the structural switch.

EXPERIMENTAL PROCEDURES
Peptide Synthesis-Peptides were made on a Pioneer Peptide Synthesis System (Perseptive Biosystems) using standard Fmoc chemistry. Peptides were purified by reverse-phase high performance liquid chromatography and their identities confirmed by matrix-assisted laser desorption ionization-time of flight mass spectrometry. Purified peptides were stored at pH 2, Ϫ20°C in 7 mM DTT. 1 Oxidized peptides were prepared as follows: peptides (at Յ100 M) were agitated at room temperature overnight in 100 mM Tris (pH 8.5), 20% Me 2 SO, 2 M guanidine hydrochloride. Oxidized peptides were stored at pH 2, Ϫ20°C.
Circular Dichroism Spectroscopy-Circular dichroism measurements were made on a JASCO J-715 spectropolarimeter fitted with a Peltier temperature controller. Unless otherwise stated, the sample stocks were diluted into a standard buffer of 25 mM potassium phosphate (pH 7) containing 1 mM DTT (DTT was omitted for the oxidized peptides). All data were collected in 1-mm quartz cuvettes. Data points for CD spectra were recorded at 1-nm intervals using a 1-nm bandwidth and 4 -16-s response times. After baseline correction, ellipticities in mdeg were converted to molar ellipticities (deg cm 2 dmol res Ϫ1 ) by normalizing for the concentration of peptide bonds. Data points for the thermal unfolding curves were recorded through 1°C min Ϫ1 ramps using a 2-nm bandwidth, averaging the signal for 16 s at 1°C intervals.
Analytical Ultracentrifugation-Sedimentation equilibrium experiments were conducted at 5°C in a Beckman Optima XL-I analytical ultracentrifuge fitted with an An-60 Ti rotor. A Ϸ100-l sample of Template-␣ (100 M) in standard buffer containing 100 mM sodium chloride was used. The sample was equilibrated for Ϸ48 h at rotor speeds of 40,000, 50,000, and 60,000 rpm. Sedimentation curves were measured by absorbance at 240 nm (the E 240 for Template-␣ was calculated as 2240 M Ϫ1 cm Ϫ1 ). The resulting data sets were fitted simultaneously using routines in the Beckman Optima XL-A/XL-I data analysis software (version 4.0). Two fitting models were used: the first assumed a single ideal species; the second assumed a monomer-dimer equilibrium and fixed monomer molecular weight. The molecular weight and partial specific volume of the peptide were calculated from the amino acid sequence as 3118 and 0.755, respectively. The viscosity of the buffer at 5°C was taken to be 1.008 mg ml Ϫ1 .
Thioflavine T Binding-Emission fluorescence spectra of thioflavine T (10 M) with Ϸ10 M peptide added were recorded between 480 and 600 nm with an excitation wavelength of 435 nm using a Varian Eclipse spectrofluorimeter and 1-cm quartz cuvettes. A scan rate of 600 nm min Ϫ1 and data interval of 1 nm was used throughout.
Electron Microscopy-Droplets of peptide solution were applied to carbon-coated copper specimen grids and dried with filter paper before negative staining with 2% phosphotungstic acid at pH 7. Grids were examined in a Hitachi 7100 TEM at 100 kV and digital images were acquired with a (800 ϫ 1200 pixel) charge-coupled device camera.

RESULTS AND DISCUSSION
Design Principles and Characterization of Template-␣-The starting point for our study was a designed canonical ␣-helical leucine-zipper sequence (Template-␣). Using established rules for coiled-coil assembly (30 -33), Template-␣ was designed to form a leucine zipper (i.e. a dimeric, parallel coiled coil, Fig.  1A): the combination of Val at a and Leu at d sites of the heptad repeat was used to direct dimer formation; Lys at one a position and the juxtaposition of Lys and Glu at g and e, respectively, were placed to ensure parallel dimers; the b and c positions were filled with helix-promoting Ala residues and the outer f sites were made polar Gln. The heptad repeat was flanked by Cys residues, which could be oxidized to an intramolecular disulfide bond to favor the alternative ␤-hairpin conformation (Fig. 1B) as required. An N-terminal Tyr-Gly tag was added to allow peptide concentrations to be determined (34). The N and C 1 The abbreviation used is: DTT, dithiothreitol.
FIG. 1. Design principles and designed peptide sequences. A and B, schematic representations of a heptad sequence repeat (abcdefg) configured onto an ␣-helical wheel with 3.5-residues per turn, and a ␤-hairpin, respectively. In the latter the dashed lines indicate the intended inter-strand hydrogen-bonded sites; that is, positions where the backbones of the residues paired across the ␤-hairpin should hydrogen bond. C, designed peptide sequences. Key: standard one-letter codes are used for the amino acids; Ac, acetyl (CH 3 .C.O.-); Am, amidated C terminus (-NH 2 ). Color is used to highlight residues designed to interact in both structures: green, hydrophobic residues; dark blue, positively charged lysines; red, negatively charged glutamates. The f positions where the Gln to Thr substitutions were made are colored light blue.
termini were acetylated and amidated, respectively. The full sequences of Template-␣ and its derivatives are given in Fig. 1C.
The coiled-coil conformation of Template-␣, at 100 M concentration, 25 mM potassium phosphate, 1 mM DTT (pH 7), was confirmed experimentally using a combination of circular dichroism spectroscopy and analytical ultracentrifugation (Fig.  2). The CD spectrum of Template-␣ at 5°C had minima at 208 and 222 nm characteristic of ␣-helical structure ( Fig. 2A). From the intensity of the signal at 222 nm, we estimated that Template-␣ contained Ϸ70% helix (35). The thermal unfolding of Template-␣ was concentration dependent as expected for an oligomerizing system (Fig. 2B). Furthermore, the unfolding transitions were almost completely reversible ( Fig. 2A).
Sedimentation equilibrium studies of Template-␣ indicated a monomer-dimer equilibrium (Fig. 2C), consistent with the CD data and the design. For example, the data did fit an analysis that assumed a single ideal species, but returned a M r of 5273 (with 95% confidence limits of 5036 and 5503), i.e. between the molecular weights expected for monomer (3118) and dimer. More detailed analysis assuming a monomer-dimer equilib-rium model gave a dissociation constant of 66 M (with 95% confidence limits of 49 and 91 M). These data are thus consistent with the CD measurements, which indicated near, but not complete folding at 5°C and 100 M peptide concentrations. In summary, the CD and equilibrium sedimentation data for Template-␣ are comparable with those collected on other designed and natural leucine zippers under similar conditions (35) and are, thus, consistent with the peptide forming a dimeric helical structure as designed.
Promoting a Conformational Switch in the Template-␣ System-As a first step to introduce ␤-structure character into the designed system, the three Gln residues at the f sites of the heptad repeat of Template-␣ were replaced by Thr to give Template-␣T (Fig. 1C). The f sites lie away from and, so, should not compromise the leucine-zipper interface (Fig. 1A). As Thr has a lower ␣-helix propensity and a higher ␤-sheet propensity than Gln (36, 37) Template-␣T was expected to form a leucine zipper, but with a reduced stability compared with Template-␣. This was observed (compare Figs. 2, A and B, and 3, A and B). However, the thermal unfolding of Template-␣T was less reversible; that is, spectra recorded after returning to the starting temperature showed a lower helicity than was measured at the start of the experiment (Fig. 3A). When thermal unfolding was repeated at the higher peptide concentration of 300 M, Template-␣ again unfolded with a normal, reversible, sigmoidal transition (Fig. 2, A and B). By contrast, the unfolding of Template-␣T showed an inflection above 60°C, which suggested some refolding (Fig. 3B). Spectra recorded after cooling this sample were wholly different from anything recorded previously: the minima at 208 and 222 nm were absent and replaced by a single minimum at 218 nm, indicative of ␤-structure (Fig. 3A). It was possible to follow the transition of Template-␣T from a near-unfolded conformation to a ␤-conformation directly on the time scale of minutes by maintaining a freshly prepared 300 M sample of the peptide at 70°C (Fig.  3C), i.e. just beyond the inflection point in the unfolding curve. Coupled with these structural changes the samples of Template-␣T gelled. To test if this process could be reversed the gel was re-suspended in a 25 mM phosphoric acid buffer adjusted to pH 2; the rationale being that peptides, proteins, and protein complexes can be acid denatured, presumably because at low pH polypeptides carry net positive charges, which repel in the folded and/or associated states. The CD spectrum recorded for the Template-␣T gels recorded immediately after re-suspension indicated ␤-structure (Fig. 3D). Interestingly, however, with time the CD spectrum of this sample reverted to that for a partly folded ␣-helix (Fig. 3D).
Template-␣T Formed Amyloid-like Fibrils-Switches to ␤-structure accompanied by gel formation are indicative of amyloid-like fibrils. Therefore, we tested the gelled peptide samples for tinctorial properties characteristic of amyloid. The fluorescence of thioflavine T (38) was compared in the presence of Template-␣T in its ␣-helical and ␤-structured states. The fluorescence intensity at 480 nm for the latter was Ϸ2.5 times greater (Fig. 4A). This was evidence for the presence of amyloid-like fibrils in the heat-treated, 300 M Template-␣T sample. To confirm this we prepared samples of this state for electron microscopy. The resulting images revealed filamentous, 25-30-nm bundles of extended, non-branching fibrils (Fig.  4, B and C). The component fibrils were twisted with a similar periodicity and, on average, had a diameter of Ϸ10 nm. These features are all consistent with amyloid formation.
Manipulating the ␣ to ␤ Transition in the Template-␣ System-Having established a system that underwent an ␣ to ␤ transition with the added curiosity that the ␤-structured state formed amyloid, we set out to examine how small, specific and rational sequence changes influenced the transition. For instance, could the transition be eased and made to occur at lower temperatures by destabilizing the native coiled-coil conformation of Template-␣T? Capping the ends of free-standing ␣-helical peptides is known to stabilize helical structures (39), and this effect is enhanced in leucine-zipper and other coiled-coil peptides (40). 2 Thus, Template-␣T was re-synthesized with free N and C termini to give Template-␣Tu (Fig. 1C). This had the desired effect of destabilizing the leucine-zipper state; indeed, the peptide was random coil and did not form appreciable helical structure over a wide range of concentrations. It is not clear how uncapping the peptide might affect amyloid formation itself. Nevertheless, Template-␣Tu could be converted to a ␤-structured state. However, this still required elevated temperatures and, in fact, the transition was more difficult to effect than with Template-␣T: under the conditions used to convert Template-␣T (300 M, 70°C), Template-␣Tu did not transform over a 1-h time scale, but a transition was observed within 1 h at 80°C. Fibrils were observed for these samples by electron microscopy.
With Template-␣Tu the formation of amyloid was assisted by salt; 0.5 M KF lowered the temperature required for the conversion described above to 70°C, i.e. to that required for the capped peptide in the absence of salt. Curiously, however, salt also stabilized the low-temperature ␣-helical state of the peptide increasing the ␣-helical content from Ϸ33 to 50% at 300 M peptide. Thus, again for our system, there was no correlation between destabilization of the native state and easing fibrillogenesis. Salt also accelerated the conversion of Template-␣T at 70°C to the ␤-structured state; the CD-monitored transitions in 0, 0.5, and 1 M KF were complete in Ϸ60, 30, and 10 min, respectively.
Promoting a ␤-Hairpin Structured Intermediate?-Summarizing the above, simply destabilizing the leucine-zipper conformation of Template-␣T did not facilitate the ␣ to ␤ transition. Others have used ␤-hairpins to build a high-resolution structure for an A␤ amyloid fibril (15), and implicated their involvement in amyloid-like fibril formation by a peptide from OspA (41). Therefore, our next step was to determine if the transition was affected by increasing the ␤-hairpin propensity of Template-␣ (Fig. 1B). Although by no means an ideal ␤-hairpin sequence, Template-␣ was designed to be compatible with this conformation. The key features were the Cys residues flanking the central heptad-based sequence. In our design logic, oxidation of these to an intramolecular disulfide link would bring the termini of the peptide together. This should simultaneously promote the ␤-hairpin and destabilize the leucine zipper. The design principles for the ␤-hairpin follow from an understanding of amino acid pairings in anti-parallel ␤-sheets (42,43). On this basis, the Cys-Cys pair should occupy  . Solid lines are spectra recorded before thermal unfolding experiments and broken lines are for those recorded after thermal unfolding. Solid circles are for data recorded at 100 M peptide and open circles are for 300 M samples. B, thermal unfolding curves; the key is same as for part A. C, time-dependent change in the CD spectra of a 300 M sample at 70°C. Open squares, t ϭ 0; solid squares, t ϭ 120 min; intervening lines were recorded at 10, 20, 30, 40, 50, and 60 min. D, CD spectra of "gels" formed by heat-treated 300 M Template-␣T. The gels were re-suspended in 25 mM phosphate (pH 2) and spectra recorded immediately (solid line) and after 5 days (broken line). a so-called non-hydrogen-bonded site in the hairpin (Fig. 1B). In turn, this would lead to the alignment of complementary, inter-strand hydrophobic interactions (between the a and d positions of the heptad repeat) and electrostatic pairs (between e and g) (Fig. 1B). Furthermore, in this conformation, two of the Thr residues introduced in Template-␣T should also align across the structure at a favored non-hydrogen-bonded site (43). With these potential interactions, and if a ␤-structured precursor could seed/promote amyloid-like fibril formation, one might expect that forcing the intra-molecular Cys-Cys bridge would facilitate the ␣ to ␤ transitions in the Template-␣ peptides and promote fibrillogenesis.
To probe the role of the intramolecular cross-link, oxidized variants for Template-␣, Template-␣T and Template-␣Tu, which we distinguish with the suffix "-ox," were prepared. For each peptide, mass spectrometry confirmed the intramolecular disulfide bonds, and either sedimentation equilibrium experiments or analytical size-exclusion chromatography was used to show that the low-temperature states were monomers. At 100 M and 5°C Template-␣T-ox was ␣-helical and thermally stabilized (Fig. 5, A and B). Thermal unfolding converted the sample to a ␤-structured state (Fig. 5A). Compared with Template-␣T, this conversion of Template-␣T-ox to ␤-structured aggregates was facilitated; a 100 M sample could be transformed in Ϸ20 min at 70°C. The gels and fibrils from these samples had all the previously described characteristics consistent with amyloid. For Template-␣Tu-ox the disulfide link did not induce additional structure at low temperature, but the temperature required to observe the conversion to ␤-structured in the presence of salt was lowered to 50°C (compared with 70°C for the reduced peptide). Although we screened a number of conditions, only the oxidized form of Template-␣ showed any tendency to form amyloid-like fibrils (Fig. 5C). Thus, tethering the termini of the peptides using a disulfide bond increased their ability to form amyloid consistent with the possible involvement of a ␤-hairpin in the fibrillogenesis process.
Conclusions-We have shown that peptides designed and characterized as a canonical, dimeric leucine-zipper can be induced to switch to ␤-structured states that form amyloid-like fibrils. The switches were heat-induced, indeed they occurred from the heat-denatured states of the peptides, and they were facilitated by modifications that raise the ␤-propensity of the sequence. Straight destabilization of the native leucine-zipper state did not promote the switch per se; elevated temperatures were still required to induce ␤-structure in a mutant that did not form a leucine-zipper dimer at ambient temperatures. This contrasts with reports for certain globular proteins in which amyloid formation correlates with the destabilization of the native state (18,20).
What drives the formation of the ␤-structured states and amyloid-like fibrils in our designed system? The following are possible contributors. The fact that the transitions in Template-␣ and Template-␣T were induced by heat is consistent with other studies that show the requirement for the polypeptides to be partly or fully unfolded to transform to amyloid (9,20). However, in our system (as judged by fav-UV CD spectra), Template-␣Tu was largely unfolded under all conditions, but still required heating to switch state. This requirement for heat also suggests the involvement of the hydrophobic effect. This is consistent with our observation that the formation of the ␤-structured state was accelerated by salt, and foregoing work on other peptides that undergo structural switches (12,26), and most recently on insulin variants that form fibrils (22). The hydrophobic effect cannot be the only factor at play, however. This is for several reasons. First, the mutation from Gln to Thr, which first resulted in the structural switch, is a swap of one polar amino acid for another. Second, low pH abrogates the ␣ to ␤ switch: none of the peptides converted to the ␤-form at pH 2 (data not shown); and the soluble, helical form of Template-␣T could be recovered by re-suspending the fibril-containing gels in pH 2 buffer. This implicates electrostatic interactions in the formation and assembly of the ␤-structured state; either the designed cross-strand salt bridges (Fig. 1) are being broken at low pH, or the overall positive charge on the peptide destabilizes the assemblies. Third, the parent peptide only formed fibrils when its termini were tethered together. This cross-link should affect overall hydrophobicity only marginally, but it could favor the alternative ␤-hairpin conformation considerably (43). Similarly, the Gln to Thr changes should increase the designed alternative ␤Ϫhairpin structure: 1) by increasing the ␤-propensity of the sequences, which would bias the conformations populated by the "unfolded" state toward the ␤Ϫregion of Ramachadran space; and/or 2) by fostering favorable interstrand Thr-Thr interactions (43). Thus, an attractive hypothesis is that non-aggregated ␤Ϫhairpin intermediates seed the fibrillogenesis process that accompanies the structural switching.
Confirmation of involvement of a putative ␤-hairpin (or any nascent ␤-structure), and assessing how far along the fibrillogenesis process specific residue placements affect amyloid formation, will require detailed structural and kinetic studies on our system (3,44). However, what is clear from our present study is that small changes designed to favor a conflicting ␤-hairpin conformation in a leucine-zipper background promote conversion of peptides to amyloid. In this respect, our studies are consistent with, and build on recent studies from the Hecht (10) and Johansson (45) groups. Hecht and co-workers (10) generated semi-random sequences with alternating hydrophobic-polar patterns to promote amphipathic ␤-structures. Several of the resulting proteins reversibly formed amyloid-like ␤-structured fibrils. The group follow up this work with an examination of natural protein sequences and show that alternating hydrophobic-polar patterns are in fact disfavored by nature (46). Johansson and colleagues (45) present convincing evidence that amyloid formation is promoted in proteins with so-called ␣/␤ discordant regions; that is, patches of sequence with a potential structural conflict that are ␣-helical in the native structure, but predicted as ␤-structure. Our work demonstrates that in addition to such general features of ␤Ϫstructure, specific sequence features can also promote amyloid-like fibrils. It remains to be seen whether the system will provide a model for studying amyloid formation in general. Nonetheless, the peptides described should provide a means to unveil systematically features that contribute to ␣ to ␤ structural transitions in peptides and proteins. In addition, the approach that we advocate potentially offers routes to rationally designed conformational switches and even new biomaterials (17,47,48).