Heterogeneous-Backbone Foldamer Mimics of a Computationally Designed, Disulfide-Rich Miniprotein

Abstract

Disulfide-rich peptides have found widespread use in the development of bioactive agents; however, low proteolytic stability and difficulty exerting synthetic control over chain topology present barriers to applications in some systems. Here, we report a method that enables creation of artificial backbone (“foldamer”) mimics of compact, disulfide-rich tertiary folds. Systematic replacement of a subset of natural α-residues with various artificial building blocks in the context of a computationally designed prototype sequence leads to “heterogeneous-backbone” variants that undergo clean oxidative folding, adopt tertiary structures indistinguishable from the prototype, and enjoy proteolytic protection beyond that inherent to the topologically constrained scaffold. Collectively, these results demonstrate systematic backbone substitution as a viable method to engineer the properties of disulfide-rich sequences and expands the repertoire of protein mimicry by foldamers to an exciting new structural class.

Graphical Abstract

Systematic alteration of peptide backbone chemical composition in a prototype sequence with a compact, disulfide-rich folded architecture yields mimics with identical folds and enhanced resistance to degradation by protease enzymes. The approach described constitutes a valuable new method to engineer properties in disulfide-rich peptides and proteins.

Introduction

Peptides and small proteins containing multiple disulfide linkages play diverse biological roles.^[1] Many act through specific binding to a target receptor, a process facilitated by their complex topologies and constrained architectures. Extensive structure-activity relationship studies carried out on disulfide-rich sequences derived from animal venoms have led to approved drugs (e.g., ziconotide for chronic pain, hirudin for anticoagulation) as well as a number of compounds in clinical development.^[2] Expanding beyond their natural functions, disulfide-rich scaffolds have also been engineered as de novo ligands for protein surfaces,^[3] with sequences derived from rational design,^[4] computational,^[5] as well as directed evolution approaches.^[6]

While peptides can exhibit improved target selectivity and potency compared to small molecules,^[7] they often suffer pharmacological drawbacks including low metabolic stability, poor oral bioavailability, and rapid clearance.^[8] Disulfide-rich sequences enjoy some advantages over similar size linear chains in this regard, as the conformational constraint imposed by multiple side-chain linkages can mask efficient recognition by proteolytic enzymes; however, this protection is not complete and varies considerably with sequence. As an example, the spider toxin ProTx-I is degraded rapidly by trypsin, while the closely related ProTx-II from the same organism (<1.5 Å C_α rmsd)^[9] is completely stable under the same conditions.^[10] The need for improved properties in some disulfide-rich systems has motivated chemists to develop and apply various chemical alterations;^[11] examples include backbone cyclization or replacement of the disulfides with other moieties.^[12] A key challenge in such work is maintaining the precise 3-dimensional fold of a prototype after chemical modification.

One approach for generating analogues of peptides with improved biostability is the modification of the covalent structure of the peptide backbone. A broad body of precedent studies have shown that many well-defined folds are possible in sequence-specific oligomers containing artificial building blocks. These agents, referred to as “foldamers,”^[13] have been shown to mimic an array of natural peptidic structures and functions.^[14] A subset of work in this area has focused on “heterogeneous” backbones that combine natural and artificial building blocks in chains that display biologically derived primary side-chain sequences.^[15] Our lab has a longstanding interest in the development of design principles for the construction of heterogeneous-backbone foldamers with complex protein-like tertiary folding patterns.^[16] In prior work, we validated design rules toward such entities in the context of the B1 domain of streptococcal protein G and a zinc finger domain from the human protein specificity factor 1.^[17]

Given the myriad of unique folds and functions available to disulfide-rich sequences, we saw the pursuit of mimicry of this class by heterogeneous backbones an attractive testbed for expanding the scope of existing design strategies. The effect of backbone alteration on disulfide-rich scaffolds has not been extensively explored, but some precedent exists that inspired the present work. Bulaj showed the incorporation of ethylene glycol and methylene spacers in place of a Gly-Gly motif in a loop of the μ-conotoxin SIIIA led to hybrids exhibiting bioactivities comparable to the native toxin.^[18] More recently, Pentelute reported heterochiral variants of Ecballium elaterium trypsin inhibitor II, a disulfide-rich miniprotein that has been retargeted toward cell-surface receptors for tumor imaging applications.^[19] Replacement of a large exposed loop in the prototype with a D-α-peptide segment led to mimics of the natural protein with similar folds but improved stability to proteolytic degradation.

Here, we explore the effects of heterogeneous backbone substitution in the context of NC_HEE_D1 (1, Fig. 1), a computationally designed 27-amino acid de novo protein recently reported by Baker.^[20] This sequence was created as part of a broader effort to develop a family of hyperstable miniproteins as scaffolds for biomedical applications.^[5] The fold of 1 is comprised of an α-helix and a β-hairpin stapled by two disulfide bonds (Cys⁴⁻¹⁸ and Cys¹⁴⁻²⁷). Of note, 1 already contains a backbone alteration at a single site in the form of a D-Pro introduced to stabilize the hairpin turn. This motivated us to assess the amenability of 1 to more extensive modification.

Figure 1.

Sequence, secondary structure map, and NMR structure (PDB 5KVN) for peptide 1 and sequences of backbone modified variants 2-6. The identity of “R” groups, when present, match the side chain specified by the corresponding α-residue denoted by the single letter code.

In seeking to expand the scope of protein tertiary structure mimicry by heterogeneous-backbone foldamers to disulfide-rich scaffolds, we were faced with a key question unique to the new target class. Would extensive modification to the backbone have an impact on the oxidative folding pathways? Synthesis of peptides with multiple disulfide bonds can be complicated by the formation of products with the incorrect disulfide-bond connectivity that differ in receptor target or biological activity.^[21] Though significant work has been put toward developing stable disulfide-rich peptide scaffolds that minimize such byproducts, these problems can be exacerbated by even subtle alterations to peptide sequence.^[11a]

Results and Discussion

Design and Synthesis of Turn Variants

Our initial goal was to establish the applicability of the previously reported synthesis of 1 by statistical aerobic oxidation to variants modified in both turn regions. The prototype contains two backbone reversals: a type-II’ turn in the center of the β-hairpin and a type-I turn connecting the α-helix to the β-hairpin. As noted above, the hairpin turn already has a backbone alteration, so it was left unchanged. As we evaluated potential modifications to apply in the other turn, we were faced with a conundrum. There are many known backbone modifications that stabilize turns in hairpins; however, these moieties remain largely unexplored in turns from other structural contexts. Due to the role of the helix-to-hairpin turn in 1 juxtaposing the secondary structures that pack to form the core of the tertiary fold, an incompatible substitution would likely disrupt the structure considerably. Given the similarity of the type-I turn to that commonly found in hairpins, we hypothesized that hairpin-stabilizers might be well accommodated in this region. Thus, we synthesized three variants of 1 (Fig. 1) in which Pro¹²-Asn¹³ was replaced with D-Pro-Gly (peptide 2), Aib-Gly (peptide 3), or δ-linked Orn (peptide 4). In terms of known folding behavior, D-Pro-Gly tends to nucleate type II’ or type I’ turns,^[22] Aib-Gly is more diverse in its conformational preferences,^[23] and δ²-Orn promotes a backbone reversal similar to that formed by D-Pro-Gly.^[24]

Reduced linear precursors corresponding to 1-4 were synthesized by microwave-assisted Fmoc solid-phase methods as C-terminal carboxamides, purified by preparative HPLC, and the identities of each confirmed by MS (see ESI for details). Following literature precedent for 1,^[20] the disulfide bonds were generated in solution through random aerobic oxidation (~75 μM peptide, pH 8.3 ammonium bicarbonate); these reactions were monitored by analytical HPLC and MS.

In the case of prototype 1, conversion of the linear precursor to a single intermediate and final product was observed, accompanied by characteristic retention time shifts for each new species (Fig. S1). A peak eluting later than the starting material was observed in the early phase of the reaction with mass corresponding to the presence of a single disulfide. At later time points, a peak eluting earlier than starting material was formed with a mass corresponding to the fully oxidized peptide. Aerobic oxidation of turn-modified variants 2-4 proceeded in a similar manner as 1, yielding a single disulfide bond intermediate that transitioned to a single fully oxidized product (Fig. S1). The observation of similar oxidative folding behavior among prototype 1 and turn variants 2-4 suggests identical disulfide topologies and similar folded structures. Purification by preparative HPLC generated samples that were subjected to further characterization to provide direct evidence bearing on this hypothesis.

Structural Analysis of Turn Variants

To assess the impact of the turn modifications in 2-4 on the overall fold, we subjected each to circular dichroism (CD) spectroscopy (60 μM peptide in 10 mM phosphate buffer, pH 7.0). The spectrum for prototype 1 showed a minimum at 218 nm and a shoulder at 210 nm (Fig. 2), consistent with published results.^[20] Spectra obtained for 3 and 4 were similar in shape and magnitude to 1, suggesting a similar fold. The spectrum for peptide 2 differed more significantly from the prototype, with a reduction in intensity of the negative peaks. This may indicate a change in overall secondary structure content; however, the complex relationship between CD spectral shape and backbone composition in heterogeneous-backbone foldamers make these results somewhat inconclusive.^{[17a, 25]}

Figure 2.

CD spectra of 1-6 at 60 μM concentration in 10 mM phosphate buffer, pH 7.

To obtain more incisive data on the folded structures of 1-4, we subjected each to multidimensional NMR spectroscopy. We acquired ¹H^/1H NOESY, COSY and TOCSY spectra (90% H₂O / 10% D₂O, pH 4.0, 298 K), enabling full assignment of proton chemical shifts. Qualitative analysis of the NMR results (Fig. 3) provides insights into secondary structure content and disulfide connectivity. The α-helix in prototype 1 (Asp²-Tyr¹¹) is indicated by strong sequential i→i+1 H_N-H_N NOEs, medium-range i→i+3 H_α-H_N NOEs, small ³J_Hα-HN coupling constants, and contiguous negative chemical shift index (CSI) values versus random coil. The hairpin region in 1 (Cys¹⁴ to Cys²⁷) is apparent from cross-strand H_N-H_N or H_α-H_α NOEs, large ³J_Hα-HN coupling constants, and contiguous positive CSI values. Finally, disulfide connectivity can be gleaned from long-range NOEs between remote Cys residues. Analyzing the above factors for 2-4 indicates that each retains the native disulfide topology as well as a defined hairpin region (Fig. 3). The NMR observables in the putative helix region for 3 and 4 are very similar to those for prototype 1; however, data for 2 show little evidence supporting a well-formed helix. Indeed, the grouping of three positive CSI values preceding the D-Pro may point to a partial strand character.^[26]

Figure 3.

Summary of NMR data (chemical shift index values, vicinal HαHN coupling constants, and NOE correlations indicative of fold) for peptides 1-4. A putative secondary structure map is provided for each sequence based on the analysis of the aggregate data.

We generated high-resolution structures of well-folded variants 3 and 4 by simulated annealing with NMR-derived restraints (Fig. 4, Table S3). The ensemble of low energy structures obtained showed excellent internal agreement (backbone rmsd < 0.75 Å); we used the lowest energy model from each ensemble in the structural analysis below. As the qualitative NMR data obtained for 1 (a C-terminal carboxamide) were entirely consistent with the previously reported structure of the prototype sequence as a terminal carboxylic acid,^[20] we used the latter published coordinates (PDB 5KVN) as the basis for comparison. Analysis of the NMR structures obtained for 1, 3, and 4 shows that the turn modifications have no significant impact on the overall tertiary fold (Fig. 4); however, closer comparison of the region of the helix-to-hairpin turn bearing the backbone modification reveals some subtle differences. As mentioned above, the Pro¹²-Asn¹³ segment in 1 adopts a canonical type-I turn. Examination of the same region in the NMR structure of 3 shows that the Aib¹²-Gly¹³ segment adopts a type II turn. The high-resolution structure of peptide 4 shows that δ²-Orn as a dipeptide replacement, while not able to be categorized as a canonical turn type, still positions the flanking residues in the helix-to-hairpin turn at similar positions as seen in 1.

Figure 4.

Comparison of the overall folding pattern and helix-to-hairpin turn from NMR structures of peptides 1 (PDB 5KVN), 3 (PDB 6E5H), and 4 (PDB 6E5I). Spheres in the cartoon representation indicate positions of a Cys residue or artificial backbone unit; residues are colored by type, matching the scheme in Fig. 1.

Taken together, the CD and NMR results obtained for 1-4 suggest that Aib-Gly and δ²-Orn are well-accommodated modifications in the helix-to-hairpin turn and that D-Pro-Gly, although a well-established stabilizer of turns in β-hairpins, is unable to effectively mimic this region in disulfide-rich scaffold 1. One possible origin of this finding is an incompatibility of the mirror image turn types (I’, II’) available to D-Pro with the helix-to-hairpin connection. This hypothesis is further supported by the observation that Aib-Gly, which has four turn types available to it, preferentially adopts a type-II turn in this context.

Design and Synthesis of Variants with Modified Helix, Turns, and Hairpin

The successful application of backbone modification in both turns of the prototype spurred us to seek variants with a greater density of artificial residue content. Given the short sequence, complex topology, and compact folded architecture of 1, choices of potential modification sites were limited. Cys residues were left untouched to minimize the potential impact on disulfide bond formation. We incorporated Aib-Gly in the helix-to-hairpin turn based on the results above. For modification to the hairpin, we made side-chain retaining α→N-Me-α substitutions at cross-strand non-hydrogen bonding positions (Val¹⁶, Val²⁵). Looking to the helix, prior work has shown that both β³-residues and C_α-Me-α-residues like Aib are well accommodated as helix modifications in proteins^[27]—even stabilizing to the tertiary fold in some contexts.^[17b] As the differences in flexibility between Aib and β^3-residues impact folding thermodynamics,^{[25a, 25b]} we sought to compare these two residue types side-by-side in the present system. Guided by the above considerations, we designed two heterogeneous-backbone variants of prototype 1. Peptides 5 and 6 each have ~25% artificial residue content and differ only in the composition of the helix segment. Peptide 5 incorporates side-chain-retaining α→β³ substitutions in place of Lys³, Glu⁶ and Lys⁹; α→Aib substitutions were made at the corresponding positions in peptide 6.

Linear precursors to 5 and 6 were synthesized and purified as described above. We analyzed the time course of oxidative folding reactions and compared the results to prototype 1 (Fig. 5A, Fig. S1) in an effort to ascertain if backbone modification led to a change in the rate and order in which the disulfide bonds form. Each reaction progressed similarly in terms of observable intermediates, with initial formation of a single intermediate followed by clean conversion to a single fully oxidized product. The rates of oxidative folding varied somewhat with backbone composition: oxidation of 1 was complete in 10 hours vs. 18 and 24 hours for heterogeneous-backbone counterparts 5 and 6, respectively.

Figure 5.

(A) Time course of the oxidative folding to form 1, 5, and 6 from the corresponding fully reduced linear precursors. Reactions were 0.25 mg/mL peptide in 0.1 M ammonium bicarbonate, pH 8.3. (B) Scheme showing the preparation of 1 through sequential chemoselective disulfide bond formation from protected precursors 1a and 1b (top) as well as random aerobic oxidation of fully reduced precursor 1_red (bottom). Key HPLC retention time changes in product vs. starting material for each reaction that informed the assignment of the structure of single-disulfide intermediate 1_int are shown. Full HPLC chromatograms can be found in the ESI.

The above observations led us to seek a better understanding of the folding mechanism. Thus, we prepared parent peptide 1 by two parallel routes in which the disulfide bonds were formed sequentially. Two linear precursors were prepared with the same sequence as 1: one with acetamidomethyl (Acm) protecting groups at Cys⁴ and Cys¹⁸ (1a) and the other with Acm groups at Cys¹⁴ and Cys²⁷ (1b). Aerobic oxidation of these peptides produced two structural isomers with different disulfide bond connectivity, 1a_ox and 1b_ox (Fig. 5B). Comparison of the HPLC properties of the cyclic products to the corresponding linear precursors (Fig. S2) shows that formation of the Cys⁴⁻¹⁸ disulfide shifts retention time earlier, while formation of the Cys¹⁴⁻²⁷ disulfide shifts retention time later. The above difference is due to the altered conformational preferences of 1a_ox and 1b_ox compared to their linear precursors due to the difference in disulfide-bond connectivity between the two species.

Recall, the single disulfide intermediate (1_int) observed in the aerobic oxidation of the fully reduced precursor to 1 (1_red) eluted later than starting material. Based on the above analysis, we assigned 1_int as corresponding to initial closing of the hairpin to form the linkage at Cys^14–27. Of note, both 1a_ox and 1b_ox could be cleanly converted to 1 by treatment with iodine to cleave the Acm groups and form the second disulfide.

The oxidative folding to form 5 and 6 proceeded in a similar manner as 1, with the observation of a single intermediate. In the case of 5, the intermediate eluted slightly later than starting material as observed for the prototype, while the intermediate en route to 6 eluted slightly earlier than the linear precursor. Despite this difference, the fully oxidized final products of both foldamer variants eluted much earlier than starting material. We interpret this as indicating the extensive burial of residues in the hydrophobic core resulting from the juxtaposition of the helix and β-hairpin upon formation of the Cys⁴⁻¹⁸ disulfide bond second.

Collectively, the above results demonstrate that even extensively-modified heterogeneous-backbone foldamer analogues of a disulfide-rich peptide sequence can undergo oxidative folding with similar mechanism and efficiency as their native counterpart.

Structural Analysis of Variants with Modified Helix, Turns, and Hairpin

We assessed the folded structure of 5 and 6 by CD and NMR spectroscopy, as detailed above for the turn variants. CD spectra of 5 and 6 were similar to each other but qualitatively different than the spectrum of 1, with a red shift of the minimum from 218 nm to ~222 nm and a decrease in signal magnitude (Fig. 2). Qualitative interpretation of the NMR data is complicated somewhat in these sequences by the density of artificial building blocks; however, the results support the hypothesis that the secondary structure content and disulfide topology of the prototype sequence are both retained in the variants (Fig. S3). High-resolution structures determined by simulated annealing with NMR-derived restraints bear out this hypothesis (Fig. 6A). Heterogeneous-backbone variants 5 and 6 show overall tertiary folds that are virtually identical to 1 (backbone rmsd values of 1.1 Å and 1.5 Å, respectively).

Figure 6.

(A) Comparison of the overall folding pattern (center) and zoomed views of identical sites in the helix (left) and hairpin (right) from the NMR structures of peptides 1 (PDB 5KVN), 5 (PDB 6E5J), and 6 (PDB 6E5K). Spheres in the cartoon representation indicate positions of a Cys residue or artificial backbone unit. (B) Packing of Tyr²³ in the structure of 1 and 6. Residues in both panels are colored by type, matching the scheme in Fig. 1.

In both sequences, the added methyl moieties of N-Me-Val¹⁶ and N-Me-Val²⁵ orient toward solvent as designed and do not disrupt the hairpin or the packing of the side chains at these sites into the hydrophobic core. A type-I turn conformation is preferred for the Aib-Gly motif in the helix-to-hairpin loop in 5 and 6, matching prototype 1 but contrasting with the type-II turn preferred for this moiety in peptide 3. The β³-residues in 5 and the Aib residues in 6 are well accommodated in the helix, leading to a native-like fold and hydrogen-bonding pattern. Residues Asn¹ and Asp² are somewhat disordered in each structure; however, this region in peptide 6 does not appear to be as dynamic as in the prototype.

In terms of subtle differences, inspection of the structure of 6 showed that the helix is juxtaposed at a slight angle in relation to the hairpin—an orientation not pronounced in the prototype peptide or other variants examined in this work. This unique packing may be partially caused by the loss of a side chain resulting from substitution of Lys³ with Aib (Fig. 6B). In the structure of the prototype 1, the side chain of Tyr²³ orients toward the hairpin turn and forms hydrophobic contacts with the side chains of Lys³. Replacement of Lys³ with Aib eliminates this interaction, leading to a change in the orientation of Tyr²³ in the structure such that it engages with Leu⁷.

Proteolytic Susceptibility

One practical motivation for developing foldamer mimics of peptides and small proteins is the prospect of such modification reducing susceptibility to degradation by protease enzymes.^[14] Disulfide-rich sequences as a class raise an interesting question in this regard, as one characteristic driving their use in biomedical contexts is high intrinsic proteolytic stability.^[1] With the above considerations in mind, does backbone alteration provide a measurable functional benefit in an already topologically constrained scaffold? In order to probe this question, we subjected prototype 1 and variants 5 and 6 to digestion in the presence of trypsin (Fig. 7, Fig. S4).

Figure 7.

Time courses for proteolytic degradation of peptides 1, 5, and 6 in the presence of trypsin; each reaction was 100 μM peptide with 2.8 mg/mL enzyme in 0.1 M ammonium bicarbonate buffer, pH 8.5.

Prototype 1 was degraded by the enzyme, albeit slowly, with ~20% of the starting material remaining after 24 h. In comparison of the proteolysis profiles for 5 and 6 under identical conditions, the heterogeneous-backbone variants were degraded to a lesser degree—~75% and ~40% remaining after 24 h for 5 and 6, respectively. The observation of greater protection in the case of the helix modified with β³ residues vs. Aib contrasted previous results from a side-by-side comparison of these residue types in an intrinsically disordered chain;^[28] the behavior in the present system may be due to subtle differences in hydrophobic core packing (Fig. 6B). Overall, the above results show that backbone modification provides an added degree of protection above and beyond that inherent to the disulfide-rich scaffold.

Conclusions

In summary, we have reported here that sequence-guided incorporation of artificial backbone units into a computationally designed disulfide-rich miniprotein can generate variants with ~25% artificial residue content and identical tertiary folds to the prototype. The helical region was effectively mimicked by replacing every third α-amino acid with a β³ analogue or Aib, and N-methyl amino acid substitutions were effectively incorporated at cross-strand positions in the β-hairpin. Comparison of three artificial turn inducers in the helix-to-hairpin loop showed that Aib-Gly and δ²-Orn were both effective mimics of the type-I turn in the prototype, while incorporation of D-Pro-Gly significantly disrupted the fold. All variants showed similar oxidative folding behavior as the starting sequence, converting cleanly to a single disulfide isomer under aerobic conditions. Investigations into the mechanism suggest the order of disulfide formation is the same in each case. Backbone modification provided a measurable improvement in stability to proteolytic degradation relative to the starting scaffold.

Disulfide-rich peptides have garnered growing interest as starting points for the generation of pharmaceutical leads,^[2] particularly given their potential role in the development of novel treatments for chronic pain.^[29] The approach we demonstrate here adds to the arsenal of methods for tuning the properties of disulfide-rich sequences through chemical alteration.^[11] Importantly, sequence-guided backbone modification of the type described is fully complementary to existing approaches such as disulfide replacement and cyclization. From the standpoint of protein tertiary structure mimicry by foldamers, the present work represents an important expansion in the demonstrated scope of existing design principles through their application to a new target class (disulfide-rich sequences) as well as a new sequence origin (de novo designed rather than taken from nature).

Experimental Section

Peptide Synthesis and Cleavage.

All reduced linear precursor peptides were synthesized via microwave-assisted Fmoc solid-phase methods (CEM MARS microwave reactor) on NovaPEG rink amide resin. The resin was swelled in DMF overnight prior to the initiation of synthesis. Amino acids were activated in situ by mixing N-α-Fmoc-protected amino acid (4 equiv), HCTU (4 equiv), and DIEA (6 equiv) in NMP (final concentration 0.1 M amino acid) unless otherwise specified. Microwave couplings were performed at 70 °C for 4 min. Fmoc-Cys(Trt)-OH, Fmoc-Cys(Acm)-OH, and Fmoc-His(Trt)-OH were coupled at room temperature for 30 min to prevent racemization. Fmoc-N-Me-Val-OH and Fmoc-Aib-OH were coupled using PyAOP (4 equiv) in place of HCTU. Microwave-assisted deprotection of the Fmoc group was achieved using 20% v/v 4-methylpiperidine in DMF at 80 °C for 2 min. Post-synthesis, the resin was sequentially washed with DMF, CH₂Cl₂ and methanol, then dried under vacuum. Cleavage of the peptide from resin was achieved by treatment with a cocktail composed of TFA/TIS/anisole/EDT/H₂O (85:4:4:4:3 by volume) for a period of 3 h. After filtration, peptide was precipitated with cold diethyl ether and pelleted by centrifugation at 6000 rpm for 2 min. Crude peptides were purified by preparative HPLC on a Phenomenex Jupiter Prep C₁₈ column (particle size 10 μm, 300 Å pore size) using gradients between water and acetonitrile containing 1% TFA (solvent A and solvent B, respectively). The identity of each peptide was confirmed by MALDI-TOF MS and purity was assessed by analytical HPLC with a Phenomenex Jupiter Analytical C₁₈ column (particle size 10 μm, 300 Å pore size). Pure peptides were lyophilized prior to oxidative folding, as detailed below.

Oxidative Folding.

Lyophilized peptides were dissolved in 0.1 M NH₄HCO₃ (pH 8.3) at a 0.25 mg/mL concentration and allowed to stir at room temperature open to air. The reaction was monitored by analytical HPLC and the identity of the intermediates and final products confirmed by MALDI-TOF MS. After completion, the reaction was quenched by addition of 50% acetic acid / H₂O to a pH ~4. The quenched samples were then diluted with H₂O, lyophilized, and purified by preparative HPLC as detailed above.

Circular Dichroism (CD) Spectroscopy.

Stock solutions of each peptide were prepared in H₂O, and the concentration quantified UV-vis spectroscopy (ε₂₇₆ = 3190 M⁻¹ cm⁻¹). CD spectra were acquired on an Olis 17 UV/VIS/NIR spectrophotometer. Samples consisted of 60 μM peptide in 10 mM sodium phosphate buffer at pH 7.0 at 20 °C. Each spectrum was acquired with wavelength scans from 200–260 nm in 1 nm increments. Sample data were exported in ASCII format and visualized in GraphPad. A 2^nd order polynomial smoothing function was applied to the nearest 5 neighboring points for each spectrum.

Oxidative Folding Kinetics Analysis.

Lyophilized linear precursors of peptides 1, 5 and 6 were diluted to a concentration of 0.25 mg/mL NH4HCO3 buffer (pH 8.3) and stirred open to atmosphere at room temperature. At each time point, 15 μL of the reaction mixture was aliquoted and quenched with 5 μL of 50% acetic acid. Evolution of the oxidation reaction was monitored by analytical HPLC using a Jupiter Analytical C18 column (particle size 5 μm, 300 Å pore size). Each sample was run at unique gradients used in the initial purification of the peptides (peptide 1: 10–30 %B, peptide 5: 15–30 %B, and peptide 6: 20–33 %B over 40 min). Identity of the fully oxidized final product was determined by HPLC retention time shift and MALDI-TOF MS. The final time points were determined by the complete disappearance of the linear precursor and the single disulfide intermediate.

Iodine Oxidation of 1_aox and 1b_ox.

Purified and lyophilized 1_aox or 1_box were dissolved in a solution of 4:1 acetic acid / H₂O at a concentration of 2 mg/mL. To this was added a solution of 0.1 M iodine in acetic acid (25 equiv). The reaction was vortexed for 10 min and then quenched by addition of 0.2 M ascorbic acid in water until the solution turned clear. A small sample of the quenched reaction was desalted by a C₁₈ resin ZipTip, and the identity of the fully-oxidized peptide species confirmed by MALDITOF MS. The oxidized peptides were diluted with H₂O and lyophilized prior to being purified by preparative HPLC as detailed above.

NMR Data Acquisition, Analysis, and Structure Determination.

NMR samples were prepared with 1–5.7 mM peptide in 1:9 D₂O/H₂O and 0.2 mM DSS at pH 4 (uncorrected). Spectra were obtained on a Bruker Avance 700 MHz spectrometer at 298 K. Two dimensional experiments were comprised of NOESY (200 ms mixing time, 32 scans), TOCSY (80 ms mixing time, 16 scans), and COSY (16 scans). Spectra were acquired with 2048 data points in the direct dimension and 512 data points in the indirect dimension. Each spectrum was processed in TOPSPIN, using DSS as an internal standard for calibration. NMR data were analyzed using the NMRFAM-SPARKY software package,^[30] and the proton resonances in each sequence assigned by standard methods. Hα chemical shifts for canonical α-residues were compared against random-coil chemical shift values generated through the Poulsen IDP/IUP Random Coil Chemical Shift Server [https://spin.niddk.nih.gov/bax/nmrserver/Poulsen_rc_CS/]^[31] to generate a chemical shift index (CSI) plots for each peptide.^[26]

Determination of high-resolution structures was carried out by simulated annealing using ARIA (Ambiguous Restraints for Iterative Assignment, version 2.3)^[32] in conjunction with CNS (Crystallography & NMR System, version 1.2).^[33] The software was modified to handle nomenclature and geometric restraints for the unnatural residues, with parameter and topology definitions based on analogous atom types already present for the natural α-residues. Base parameters for each ARIA run were modified from program defaults to improve model quality and convergence, as previously described.^[34] Each sequence showed unambiguous long-range NOEs consistent with the native disulfide topology, so corresponding covalent bonds were introduced in the calculation. A set of initial H-bond restraints was generated based on qualitative inspection of the NOESY spectrum: existence of long-range H_N(i)→H_N(j) cross peaks for the hairpin, strength of H_N(i) →H_N(i+1) cross peaks and existence of H_α(i) →H_N(i+3) cross peaks for the helix. Backbone φ dihedral restraints were prepared based on measured ³J_Hα-HN coupling constants for well-resolved amide doublets in the 1D ¹HNMR spectrum (φ = −65°±25° for J ≤ 6.0 Hz and φ = −120°±40° for J ≥ 8.0 Hz). NOE distance restraints were generated automatically by ARIA in iterative fashion over the course of the calculation, starting from an unassigned set of integrated peaks from the NOESY spectrum and a list of proton chemical shift values. The ensemble of 10 lowest energy structures resulting from the above calculation was then used as the input structure for a second ARIA run. Parameters used were the same as above, with additional restraints added for any new H-bonds observed in the first ensemble. The final set of 10 lowest energy structures resulting from the second run was taken as the NMR ensemble for that peptide. The lowest energy entry from each ensemble was used for structure comparisons, figure generation, and analysis presented in the main text. Ensemble coordinates and additional experimental data are deposited in the PDB (accession codes 3: 6E5H, 4: 6E5I, 5: 6E5J, 6: 6E5K) and BMRB (accession codes 3: 30496, 4: 30497, 5: 30498, 6: 30499).

Proteolytic Degradation by Trypsin.

Stock solution of trypsin was prepared by dissolving 2.8 mg bovine trypsin (MW 23,290 Da) in 1 mL of 0.1 M NH₄HCO₃ solution with 1 mM CaCl₂ (pH 8.5). Lyophilized samples of peptide 1, 5, and 6 were dissolved in deionized H2O, and their concentrations determined by UV spectroscopy (ε₂₇₆ = 3190 M⁻¹ cm⁻¹). Each reaction contained 10 μM enzyme and 100 μM peptide in 0.1 M NH₄HCO₃ (pH 8.5), with a total volume of 120 μL. Reactions were mixed and allowed to incubate at room temperature. At 6, 12, and 24 h time points, 40 μL of the reaction solution was quenched with 5 μL of 10% TFA and diluted by addition of 55 μL of 0.1 M NH₄HCO₃. The quenched samples were analyzed by analytical HPLC (90 μL injection) using a Jupiter Analytical C18 column (particle size 5 μm, 300 Å pore size) and a 10–30% B gradient. Identity of the peptides of interest were confirmed by MALDITOF MS, and degradation monitored by integration of the corresponding peak on the HPLC chromatogram. Peak areas were normalized to the area of the starting material and the amount of peptide remaining over time was plotted via GraphPad Prism.