ROSETTA-informed design of structurally stabilized cyclic anti-amyloid peptides

Abstract

β-amyloid oligomers are thought to be the most toxic species formed en route to fibril deposition in Alzheimer’s disease. Transthyretin is a natural sequestering agent of β-amyloid oligomers: the binding site to β-amyloid has been traced to strands G/H of the inner β-sheet of transthyretin. A linear peptide, with the same primary sequence as the β-amyloid binding domain on transthyretin, was moderately effective at inhibiting β-amyloid fibril growth. Insertion of a β-turn template and cyclization greatly increased stability against proteolysis and improved efficacy as an amyloid inhibitor. However, the cyclic peptide still contained a significant amount of disorder. Using the Simple Cyclic Peptide Application within ROSETTA as an in silico predictor of cyclic peptide conformation and stability, we investigated putative structural enhancements, including stabilization by disulfide linkages and insertion of a second β-turn template. Several candidates were synthesized and tested for secondary structure and ability to inhibit β-amyloid aggregation. The results demonstrate that cyclization, β-sheet structure and conformational homogeneity are all preferable design features, whereas disulfide bond formation across the two β-strands is not preferable.

Introduction

Alzheimer’s disease (AD) is characterized by neurofibrillary tangles of tau protein and intracellular plaques of β-amyloid peptide (Aβ) (Newell et al., 1999). Because Aβ is the primary component of the plaques in AD, and mutations in the Aβ precursor gene, APP, lead to early onset AD, many researchers have targeted sequestration and/or removal of Aβ as a promising therapeutic strategy (Reitz, 2012). Aβ monomer is disordered, but the peptide spontaneously self-assembles into soluble oligomers and ultimately fibrillar structures that deposit on tissue. Aβ oligomers adopt a diversity of conformations and sizes during the aggregation process, with the final fibrillar aggregates displaying the distinct cross-β sheet structure typical of amyloid (Ow and Dunstan, 2014). Although fibrils are deleterious to cells, the soluble, pre-fibrillar oligomeric species are widely believed to be the most toxic (Caughey and Lansbury, 2003; Ferreira et al., 2007; LaFerla et al., 2007). Compounds that specifically sequester these oligomers, then, may have therapeutic potential; indeed, this approach has recently been adopted by antibody therapeutics such as Aducanumab and Crenezumab (Cummings et al., 2018). Both of these antibody treatments were halted during Phase 3 trials due to a lack of desired efficacy, suggesting that the mechanism by which compounds bind to and clear toxic Aβ oligomeric species is not yet sufficiently understood.

Mice genetically engineered to overexpress the disease-related Swedish mutation, APP_sw, produce high levels of Aβ and numerous deposits in the brain, but do not show the neuronal dysfunction and death expected with amyloid deposits. An explanation for the apparent lack of disease was provided by the observation that transthyretin (TTR) synthesis was highly upregulated in these mice (Stein and Johnson, 2002). TTR appeared to provide neuroprotection against Aβ, and neutralization of TTR with antibodies eliminated the neuroprotective effects (Stein and Johnson, 2002; Stein et al., 2004; Buxbaum et al., 2008; Li et al., 2011). We and others have shown that neuroprotection is afforded by direct binding of TTR to Aβ (Du et al., 2012; Du and Murphy, 2010; Schwarzman et al., 1994), and we identified a strong binding site in strands G/H, located on the inner β-sheet lining the hydrophobic pocket of the TTR tetramer (Du et al., 2012). TTR’s normal physiological role is as a carrier for thyroxine and retinol; interestingly, TTR can also aggregate into amyloid deposits and is implicated in senile systemic amyloidosis (Westermark et al., 1990) and familial amyloid polyneuropathy (Saraiva et al., 1984). TTR aggregation proceeds via destabilization of the tetramer to the monomer, modest conformational change in the monomer, and then re-assembly into fibrils (Lai et al., 1996). A monomeric mutant of TTR, mTTR, was developed that did not aggregate and does not assemble into the native tetramer (Jiang et al., 2001). Previously, we showed that mTTR is a potent inhibitor of Aβ fibril formation and Aβ-induced cellular toxicity (Yang et al., 2013). Inhibition occurs at substoichiometric quantities of mTTR, via selective binding to oligomeric Aβ (Yang et al., 2013), indicating promise as a possible therapeutic compound. mTTR significantly reduces adherence of oligomeric Aβ to human iPSC-derived neurons (Pate et al., 2018), likely explaining its mechanism of protection against Aβ toxicity.

However, mTTR has two significant drawbacks: first, it is readily degraded by proteases, and second, its binding to Aβ is significantly diminished in the presence of nonspecific proteins (Pate et al., 2018). We have pursued an alternative approach, specifically, to develop peptides that mimic the Aβ-binding domain (strands G/H) of TTR. A linear peptide (G16) corresponding to residues 102-117 of TTR showed modest anti-Aβ activity (Cho et al., 2014). We hypothesized that restricting the conformational flexibility by insertion of the non-natural β-turn template ^DProPro and cyclization of the peptide backbone would produce a peptide that more closely mimics not only the sequence, but also the conformation, of TTR strands G/H. This peptide, cG3, was more effective than its linear counterpart at inhibiting Aβ fibrillogenesis and reducing Aβ toxicity (Cho et al., 2015). A further improvement in peptide design was obtained via analysis using TANGO, a statistical mechanical algorithm that predicts aggregation-prone regions from protein sequence (Fernandez-Escamilla et al., 2004). This analysis led to the development of cG8, a peptide with enhanced efficacy in inhibition of Aβ fibril formation that, upon binding, redirected Aβ towards protease-sensitive, nonfibrillar aggregates (Lu et al., 2016).

cG8 contains a mix of β-sheet and disordered secondary structure. We hypothesized that further improvement might be attained if we could design peptides that more closely mimic the structure of strands G/H of mTTR. In this report, we investigated methods for enhancing the stability of the two-stranded anti-parallel β-sheet secondary structure of cyclic peptides, as a means to more closely mimic the conformation of the Aβ-binding domain in mTTR. We proposed incorporation of disulfide bonds (Dombkowski et al., 2014) and additional β-turns (Haque and Gellman, 1997) to enhance stability, and we used the Simple Cyclic Peptide Application (SCPA) (Bhardwaj et al., 2016) within the ROSETTA software suite to rationally explore the design space. We then synthesized promising peptides and characterized their secondary structure and efficacy for inhibition of Aβ fibril formation.

Materials and Methods

Peptide synthesis and purification

All materials were purchased from Fisher Scientific (Pittsburgh, PA) except where indicated. Peptides were purchased from New England Peptide and either supplied fully cyclized (cG8_ss/dProPro and cG8_dProPro) or supplied on resin as a linear peptide with side chains protected (cG8 and G8_ss). Cleavage of the peptide from the resin was obtained by dissolving the resin in 20% HFIP in DCM and shaking for 1 hour, followed by centrifugation, recovery of the clear solution below the solid, and evaporation under nitrogen to remove HFIP. For solution-phase cyclization, benzotriazol-1-yl-oxytripyrrolidinophosphonium hexafluorophosphate (PyBOP) solution (25 mg PyBOP, 25 μL DIPEA in 2 ml DCM) was added to the peptide and incubated for 18 hours wrapped in foil, with periodic nitrogen flushing. The solution was slowly added to 30 ml ice-cold t-BME, incubated for 15 minutes on ice, then centrifuged and the supernatant removed. The pelleted peptide was washed with ice-cold t-BME again and centrifuged, then dried under nitrogen, dissolved in 3 ml of 40% acetonitrile, frozen and lyophilized. To remove side-chain protecting groups, lyophilized peptides were dissolved in 1 ml of Reagent K and incubated for 1–3 h at room temperature on a shaker. The solution was dripped into ice cold t-BME, incubated for 15 minutes on ice, and centrifuged at 2000 rcf at 4°C. Supernatant was removed and the pellet was washed with fresh ice-cold t-BME and centrifuged. The supernatant was removed and the pellet dried under nitrogen, then re-suspended in 40% acetonitrile and lyophilized. Lyophilized peptide was re-suspended in 20% acetonitrile and purified by RP-HPLC on a Vydac C18 column. Peptides were eluted from the column with a gradient of acetonitrile and water with 0.1% TFA at 2 ml/min. Purified peptide was collected and lyophilized before re-suspension in 0.22 μm filtered Milli-Q water. All cyclic and linear peptide concentrations were determined by absorbance with the extinction coefficient calculated from the peptide sequence as described elsewhere (Anthis and Clore, 2013). The calculated extinction coefficients at 205 nm were as follows: cG8, 81 150 M⁻¹ cm⁻¹; G8_ss, 80 570 M⁻¹ cm⁻¹; cG8_ss/dProPro, 76 870 M⁻¹ cm⁻¹; cG8_dProPro, 75 070 M⁻¹ cm⁻¹. Peptides were diluted to 0.6 mM aliquots, snap frozen and stored at – 80°C.

Oxidation protocols

cG8_sh (putative cG8_ss) was re-suspended in a variety of oxidizing conditions in order to promote disulfide formation. First, lyophilized cG8_sh was re-suspended in pure water and allowed to oxidize in air for 48 h at 4°C. Next, lyophilized cG8_sh was suspended in 100 mM ammonium bicarbonate (pH 8.3) for 24 hours at 4°C. Alternatively, 10 μM cG8_sh (final concentration) was suspended in 100 mM ammonium bicarbonate (pH 8.3) supplemented with a variety of additives intended to reduce potential peptide self-association. Peptides re-suspended with additives were allowed to oxidize at room temperature for 24 hours unless otherwise indicated. Additives were as follows: 2 M urea, 1 M guanidine hydrochloride, 1 M guanidine hydrochloride supplemented with 10 mM glutathione for 72 hours with gentle shaking at room temperature, 1 M guanidine hydrochloride supplemented with 10 mM glutathione and 50% v/v isopropanol for 72 hours with gentle shaking at room temperature. Lyophilized G8_ss and cG8_ss/dProPro were re-suspended in pure water and allowed to oxidize in air for 48 hours at 4°C. In all cases, MALDI mass spectrometry was used to detect disulfide formation.

Iodoacetamide reaction

cG8_sh was incubated in 50 mM TCEP for 1 hour at room temperature to ensure complete reduction of cysteine residues, then with 20 mM iodoacetamide for two hours at room temperature. Following incubation, the solution was dialyzed against MilliQ water for 30 hours and then analyzed by mass spectrometry.

Recombinant mTTR production

mTTR was produced recombinantly and purified as described previously (Du and Murphy, 2010). mTTR concentration was quantified by absorbance at 280 nm using an extinction coefficient of 19 400 M⁻¹ cm⁻¹.

Circular dichroism (CD)

Peptide stock solutions were diluted into phosphate buffer with sodium fluoride (10 mM Na₂HPO₄/NaH₂PO₄ and 140 mM NaF, pH 7.4) to a peptide concentration of 45 μM. Samples were filtered through a 0.45 μm filter then immediately loaded into a 1 mm cell. CD spectra were collected on an Aviv Model 420 CD spectrophotometer (Aviv Biomedical, Lakewood, NJ) at 25°C. Three scans were collected and averaged per sample. The mean spectra were then background corrected by subtraction of solvent spectra. Spectra were smoothed by a 10 point Savitzky–Golay function.

Simulation of peptide sequences in ROSETTA

Amino acid sequences of cG8, cG8_ss, cG8_ss/dProPro and cG8_dProPro (Table I) were input to the Simple Cyclic Peptide Application (SCPA) within the ROSETTA software suite. SCPA affords a convenient method of de novo, in silico generation of 3D model structures of an input cyclic peptide, including those with non-canonical amino acids (Bhardwaj et al., 2016). Generated structures were compared to an input PDB image of strands G/H (residues 104 to 123) of human TTR (PDB entry 1DVQ) to calculate an RMSD from the input structure. A sample input command string is shown in Supplementary Fig. S1, with required input text in blue/black, and comments for user understanding (not input to the terminal) shown in green. Unless otherwise indicated, 10 000 parallel simulations were run under identical input conditions. The linear peptide chain torsion angles were randomized prior to attempted loop closure. For peptides containing two cysteines, accepted ROSETTA solutions were not constrained to only those containing disulfide bonds. Program outputs include 3D atomic models of the converged solutions and the ROSETTA Energy Value (r.e.u.) and RMSD (Å) of each solved structure compared to strand G/H of TTR. These data are plotted as Energy (r.e.u.) vs RMSD of the structure solution to generate a scatter plot that displays a funnel towards minimized energies and RMSD values. The ROSETTA energy value does not have a direct quantitative meaning, but lower (more negative) r.e.u. correlates to greater peptide stability. The 3D models of the five lowest energy structures were overlaid in PYMOL for visualization.

Table I.

Peptide sequences

Peptide name	Sequence	TTR residue numbers	Calculated molecular weight (Da)
TTR strands G/H	PRRYTIAALLSPYSYSTTAVVT	102–123

^aAn N-C cyclic backbone is indicated by ‘c’ in peptide name and solid line.

^bSubscript ‘sh’ indicates cysteines are in reduced form.

^cDisulfide bond formed between cysteine side chains indicated by dashed line.

Aβ preparation

Lyophilized Aβ₄₀ (Bachem) was dissolved in 0.22 μm filtered 50% acetonitrile to a concentration of 1 mg/ml, frozen at –80°C and re-lyophilized. To prepare aliquots, lyophilized Aβ was dissolved in 0.02 μm filtered 8 M urea/100 mM glycine buffer, pH 10, for 1 hour at ~12 mg/ml. Samples were aliquoted and snap-frozen in ethanol/dry ice and stored at –80°C until needed. For each experiment, aliquots were thawed and diluted into 0.02 μm filtered 1 M NaOH at a 1:8 (v/v) peptide:NaOH for 20 minutes to break up any aggregates formed during snap freezing. Samples were diluted into the appropriate buffer and used immediately.

Thioflavin T (ThT) fluorescence assay

ThT stock solutions were prepared in PBSA and filtered through a 0.22 μm filter. ThT concentration was determined by absorbance at 416 nm, using an extinction coefficient of 26 620 M⁻¹ cm⁻¹. The stock solution was diluted to 10 μM in 0.22 μm filtered PBSA and stored at 4°C for no more than two days. Aβ alone (28 μM) and with peptides or proteins at various concentrations were prepared in 0.22 μm filtered PBSA and incubated at 37°C. After 1, 24 and 48 hours of incubation, each sample was mixed with ThT at a final ThT:Aβ molar ratio of 5:1, and ThT fluorescence emission was measured using a QuantaMaster spectrofluorometer (PTI, Birmingham, NJ), with excitation at 440 nm and emission spectra recorded from 460 to 500 nm. Each sample was scanned three times, and the resulting spectra averaged as data for one independent sample. Spectra from four independent samples were averaged at each time point. The background average signal of ThT in PBSA was subtracted from the averaged data to calculate a corrected mean ThT intensity. Data at 24 and 48 h were analyzed independently by a two-tailed t-test assuming unequal variances at a confidence level of 0.05.

Results

TTR is neuroprotective against Aβ toxicity in cell culture and in AD mice (Stein and Johnson, 2002; Stein et al., 2004; Buxbaum et al., 2008; Li et al., 2011) and neuroprotection is afforded via preferential binding of TTR to Aβ oligomers, disrupting their maturation to fibrils (Yang et al., 2013). The Aβ-binding domain was localized to specific residues on strands G/H in the inner β-sheet of TTR (Du et al., 2012). We hypothesized that peptides which mimic the sequence and conformation of strands G/H in TTR could be effective replacements for the protein (Fig. 1). Such peptides have advantages in protease stability and in specificity compared to proteins (Pate et al., 2018). We have previously shown that a linear peptide G16 (Table I), containing residues 102–117 of TTR with a single tyrosine to tryptophan substitution, was moderately effective at inhibiting Aβ aggregation and toxicity (Cho et al., 2014). Insertion of a single ^DProPro β-turn template and cyclization yielded cG3 (Table I) and improved efficacy (Cho et al., 2015; Lu et al., 2016). Cyclization should reduce conformational flexibility and help to lock in the anti-parallel two-stranded β-sheet structure of the strands G/H template. By circular dichroism, we observed that cG3 contained substantial disordered content (Cho et al., 2015). Three additional mutations, identified via analysis with the TANGO algorithm (http://tango.crg.es/), led to the design of cG8 (Table I, Fig. 1). Although cG8 represented a further improvement over cG3 in efficacy at inhibiting Aβ aggregation, analysis by circular dichroism indicated that the peptide was still not locked into a predominantly β-sheet structure (Lu et al., 2016). We hypothesized that peptide conformational rigidity plays a key role in its ability to mimic mTTR’s ability to bind to Aβ oligomers and inhibit aggregation; specifically, peptides that mimic as closely as possible not only the primary sequence but also the three-dimensional structure of strands G/H would be most effective therapeutic agents. An opposing view is that a more flexible structure that can adapt to several Aβ oligomeric conformations may show greater efficacy against an ensemble of Aβ species. We explored these two hypotheses through simulation and experimental characterization of structure-stabilized peptides.

Fig. 1.

Cyclic peptide design and transthyretin monomer structure. (A) cG8 ribbon diagram adapted from Lu et al. (2016) identifying key residue information and expected location on the β-strand: (green) residues on the putative Aβ-binding domain side (solvent exposed in TTR monomeric subunit); (blue) residues on the non-Aβ binding face; (red) residues changed for disulfide incorporation (I107/K120) or additional β-turn (P113/Y114). (B) Crystal structure of TTR monomeric subunit with Aβ binding strands G/H (residues 102-123) highlighted (red). Modified from PDB Entry 1DVQ.

To examine the experimental finding that cG3 and cG8 contain a significant amount of conformational disorder, we simulated the two peptides using the SCPA package in ROSETTA. SCPA is a recent addition to the ROSETTA software suite that uniquely can account for cyclization and non-canonical amino acids, thus making it particularly suitable for examining hyperstable peptides (Bhardwaj et al., 2016). Results for cG8 are shown in Supplementary Fig. S2. The five lowest energy structures were selected for closer examination and demonstrate a significant amount of flexibility and heterogeneity. cG3 simulations showed even greater conformational heterogeneity than cG8 (not shown), consistent with experiment and with our hypothesis that greater rigidity translates into greater efficacy.

Given these results, and because TTR can serve as a template for rational design, we attempted to design cyclic peptides with greater conformational rigidity and structure that more closely resembles strands G/H. Insertion of a disulfide bond linking the two β-strands is a common method for stabilizing β-sheets (Hutchinson et al., 1998; Santiveri et al., 2008), so we investigated sites for placement of cysteines in strands G and H, starting with cG8 as a basis. We sought to avoid disrupting the Aβ-binding face (Fig. 1A), which constrained us to four possible mutation pairs on the ‘back’ side: Y105C/V122C, I107C/K120C, K109C/Q118C or S111C/Y116C (numbers refer to position in native human TTR). TANGO analysis (not shown) identified I107C/K120C as the best site for disulfide incorporation, because K109C/Q118C showed propensity for aggregation, Y105C/V122C showed undesirable reduction in β-sheet character for strand G, and S111C/Y116C showed no major differences from cG8. We therefore designed G8_sh with cysteine residues located at I107C/K120C. We synthesized the linear peptide, cleaved it from the resin, cyclized in-solution, then air oxidized in pure water for several hours to promote disulfide formation. After exposure to conditions that should lead to N-to-C cyclization of the peptide and allow formation of the disulfide bond, mass spectrometry analysis (Fig. 2A) indicated a high purity product, with a major peak at 2434.2 Da. However, the expected molecular weight of cG8_ss is 2432 Da, a difference in mass of 2 Da, suggesting that the cysteines remained in their free sulfhydryl form. We hypothesized that air oxidation may not be sufficient to oxidize the cyclic-constrained peptide, so we investigated several other oxidizing conditions (Trivedi et al., 2009; Daly et al., 1999), such as adjusting pH to alkaline conditions in an ammonium bicarbonate buffer, increasing incubation time to several days, chemical denaturation, and treatment in 50% isopropanol, but none resulted in formation of the expected disulfide bond based on mass spectrometry analysis (not shown).

Fig. 2.

Mass spectrometry analysis of purified G8-derived peptides. (A) Product after exposure to cyclization and oxidation conditions. The major peak at 2434.2 corresponds to the expected molar mass of the cyclized and reduced peptide cG8_sh.(B) Linear peptide after exposure to oxidation conditions. The major peak at 2450.2 corresponds to the expected molar mass of the linear and disulfide-linked peptide G8_ss. (C) Product after reaction of cG8_sh with iodoacetamide.

As shown in Fig. 2A, a compound with molecular weight of 2450.2 Da was detected as an impurity in this preparation. This corresponds to the expected molecular weight of a linear, disulfide constrained peptide, G8_ss. We hypothesized that the trace amounts of the linear peptide remained after cyclization and that the linear peptide, unlike the cyclic, could be oxidized to form the disulfide. To test this, we omitted the cyclization step and subjected the linear peptide to mild air oxidation conditions in pure water. This treatment produced peptide with the expected molecular weight of G8_ss (Fig. 2B), supporting the hypothesis that cyclization interferes with disulfide bond formation. We next tested whether cG8_sh contained free sulfhydryls by incubating the peptide with excess iodoacetamide; under these conditions, free cysteine should undergo carboxamidomethyl addition, resulting in a + 57 Da shift per cysteine residue reacted. Two major compounds were observed in the reaction mixture, with molecular weights of 2434.1 Da (unreacted cG8_sh) and 2491.1 Da (one carboxamidomethyl addition) (Fig. 2C). We interpret this result to indicate that only one free cysteine is available for reaction in the cyclic peptide.

Taken together, these results show that we can obtain either cyclically constrained reduced peptide, or linear disulfide-constrained peptide, but not both.

To test whether the disulfide constraint stabilizes the anti-parallel β-sheet structure, we compared CD spectra of G8_ss to cG8 (Fig. 3). Incorporation of the disulfide bond in the absence of cyclization shifts the peptide conformation towards a more β-sheet rich structure. In an effort to understand why it was difficult to synthesize cG8_ss with both cyclization and disulfide bond formation, we returned to TANGO analysis. As shown in Fig. 4, the I107C mutation introduced an unwanted β-turn propensity within the strand G section. We speculated that this propensity could lead to significant distortion of the desired anti-parallel two-β-strand conformation.

Fig. 3.

CD spectra of G8_ss and cG8. Samples were prepared in phosphate buffer with fluoride (10 mM sodium phosphate, 140 mM NaF, pH 7.4) at a final peptide concentration of 45 μM. Data were taken at 25°C. Spectra shown are cG8 (black) and G8_ss (blue), reading from top to bottom at 220 nm.

Fig. 4.

TANGO analysis comparing cG8 and cG8_ss. Briefly, TANGO predicts from the linear sequence the propensity for forming β-sheet or β-turn secondary structure, or for aggregation. The mutations I107C/K120C (indicated by arrow) introduce a propensity for unwanted β-turn in cG8_ss that is absent in cG8. Neither cG8 nor cG8_ss show any aggregation propensity.

The TANGO algorithm cannot consider features such as cyclization or non-canonical amino acids. To obtain greater insight, we simulated cG8_ss using SCPA in ROSETTA and compared the results to cG8 (Fig. 5A). Disulfide bond formation reduced the average RMSD and narrowed the RMSD distribution, indicating overall lower conformational heterogeneity and a closer match to the native strands G/H for cG8_ss compared to cG8. However, the energy distribution shifts upwards (less negative), suggesting that the disulfide constraint introduces conformational strain that is not fully compensated for by energetically favorable H-bond or S–S bond formation. Our experimental results demonstrate that this energetic shift is sufficiently unfavorable to virtually eliminate population of the cyclized and disulfide-bonded structure.

Fig. 5.

SCPA Simulation of cG8_ss. The simulation consists of 10 000 independent completed closures. (A) Comparison of energy values and atomic RMSD for cG8_ss (blue) versus cG8 (black). (B) Cartoon overlay showing residue side chains and disulfide bonds (yellow) of the five lowest energy structures for cG8_ss. (C) Cartoons showing residue side chains and disulfide bonds of each of the five lowest energy structures individually. (D) Cartoon overlay of the five lowest energy structures with residue side chains hidden but disulfide bonds shown.

We examined the five lowest energy conformations of cG8_ss more closely (Fig. 5B, C and D). Of particular note is the conformational diversity, especially the kinked conformations shown in Fig. 5C. We used SCPA to screen the three other previously identified locations for incorporation of cysteine residues into the cG8 template for disulfide bond formation (Fig. 6). Overlays of the five lowest energy structures in all three alternative locations show that positions other than I107 C/K120C predict loss of β-sheet structure and increase in conformational heterogeneity and disorder. These results are consistent with TANGO analysis and show that incorporation of the disulfide bond at a different position than 107/120 on the non-Aβ binding face of the anti-parallel β-strands is unfavorable.

Fig. 6.

SCPA simulation of disulfide-constrained peptide structures of the four non-Aβ binding face positions. Each cartoon overlays the five lowest energy structures per position with residue side chains hidden but disulfide bonds shown. Each position was screened for conformational heterogeneity by ~1 000 independent simulations. (***) indicates I107C/K120 (cG8_ss).

We next looked for other modifications that could stabilize the cyclic backbone in the preferred anti-parallel β-strand conformation. Based on work by others in designing hyperstable small cyclic peptides (Bhardwaj et al., 2016), we considered incorporation of a second ^DProPro β-turn template by the double mutation P113^DP/Y114P (cG8_ss/dProPro, Table I). ROSETTA simulations indicated a favorable change in energetics for cG8_ss/dProPro compared to cG8_ss (Fig. 7). The five lowest energy structures show a higher degree of homogeneity compared to cG8_ss, although the kink in the β-strand is still possible (Fig. 7C). These encouraging results prompted us to attempt synthesis of cG8_ss/dProPro. Both cyclization and air oxidation in pure water proceeded efficiently (data not shown), indicating that the addition of the second β-turn template allowed easy disulfide linkage across the two β-strands. Circular dichroic analysis indicated a modest increase in β-sheet content compared to cG8 and G8_ss (Fig. 8), consistent with the simulation predictions.

Fig. 7.

SCPA Simulation of cG8_ss/dProPro. The simulation consists of 10 000 independent completed closures. (A) Comparison of energy values and atomic RMSD for cG8_ss/dProPro (red) compared against cG8_ss (blue). (B) Cartoon overlay showing residue side chains and disulfide bonds (yellow) of the five lowest energy structures. (C) Cartoons showing residue side chains and disulfide bonds of each of the five lowest energy structures individually. (D) Cartoon overlay of the five lowest energy structures with residue side chains hidden but disulfide bonds shown.

Fig. 8.

CD spectra of cG8_ss/dProPro, G8_ss and cG8. The samples were prepared in phosphate buffer (10 mM sodium phosphate, 140 mM NaF, pH 7.4) at a final peptide concentration of 45 μM. Data were taken at 25°C. Spectra shown are cG8 (black), G8_ss (blue) and cG8_ss/dProPro (red), reading from top to bottom at 220 nm.

Finally, we explored whether the two ^DProPro β-turn templates were sufficient, in the absence of the disulfide cross-strand linkage, to yield a stable anti-parallel β-sheet conformation. ROSETTA simulations of cG8_dProPro showed that removing the disulfides deepened the energy well and resulted in a high degree of conformational β-sheet homogeneity among the lowest energy conformers (Fig. 9). cG8_dProPro was synthesized and its secondary structure probed by circular dichroism. As shown in Fig. 10, the experimental data are strongly supportive of the ROSETTA-predicted structures, as evidenced by a deep trough in the spectrum indicative of strong β-sheet content.

Fig. 9.

SCPA Simulation of cG8_dProPro. The simulation consists of 10 000 independent completed closures. (A) Comparison of energy values and atomic RMSD for cG8_dProPro (green) compared against cG8 (black). (B) Cartoon overlay showing residue side chains of the five lowest energy structures from the full simulation. (C) Cartoons showing residue side chains of each of the five lowest energy structures individually. (D) Cartoon overlay of the five lowest energy structures with residue side chains hidden.

Fig. 10.

CD spectra of cG8_dProPro (green), cG8_ss/dProPro (red), G8_ss (blue) and cG8 (black). The samples were prepared in phosphate buffer (10 mM sodium phosphate, 140 mM NaF, pH 7.4) at a final peptide concentration of 45 μM. Data were taken at 25°C. Spectra shown are cG8 (black), G8_ss (blue), cG8_ss/dProPro (red) and cG8_dProPro (green), reading from top to bottom at 220 nm.

One goal of this study was to determine whether imposition of tighter conformational constraints will improve or disrupt the efficacy of these peptides at inhibition of Aβ fibril formation. To test, we incubated Aβ with peptides at 20-fold molar Aβ excess and measured fibril formation using Thioflavin T (ThT) fluorescence. Briefly, binding of ThT to amyloid strongly increases the dye’s fluorescence intensity, with the intensity proportional to the mass of amyloid aggregates in solution (Xue et al., 2017). We chose to carry out the experiment at large Aβ excess to enhance any distinctions among the various peptides. Results are shown in Fig. 11.

Fig. 11.

Aβ fibril formation, as measured by Thioflavin-T (ThT) fluorescence intensity, alone or in the presence of designed peptides at a 20:1 (Aβ:peptide) molar ratio, at 1, 24 or 48 h. Data were taken at 37°C. All data are background subtracted, and error bars indicate SEM. Reading from left to right, samples are: Aβ alone (gray); Aβ with cG8 (black); Aβ with G8_ss (blue); Aβ with cG8_ss/dProPro (red); Aβ with cG8_dProPro (green); Aβ with mTTR (white). (*) indicates different vs. Aβ alone (P < 0.05); (θ) indicates different vs. cG8 (P < 0.05). All peptides were ThT-negative in the absence of Aβ.

As expected, ThT fluorescence from Aβ alone increased with time, indicating formation of amyloid fibrils. cG8 significantly reduced the ThT emission compared to Aβ, consistent with previous reports (Lu et al., 2016). The linear disulfide constrained peptide, G8_ss, is noticeably less effective than cG8, and in early times (24 hours) does not appear to inhibit fibril formation at all. Incorporation of the second ^DProPro β-turn template enabled cyclization and disulfide bond formation (cG8_ss/dProPro), leading to partial (24 h) or full (40 h) recovery of peptide efficacy to similar levels as cG8. The double ^DProPro β-turn templates, with no disulfide constraint (cG8_dProPro), modestly improved efficacy compared to cG8. None of the peptides were as effective as mTTR at any time.

Discussion

Given the strong evidence that TTR is neuroprotective in cellular and animal models of AD and that neuroprotection is afforded via direct binding and sequestration by TTR to Aβ oligomers, we have attempted to design small stable peptides that mimic TTR’s efficacy at sequestering Aβ oligomers, inhibiting Aβ fibril formation, and providing protection against Aβ toxicity. Our peptide design strategy uses strands G/H of TTR as a template, because this protein region was identified as the Aβ-binding domain. Cyclization afforded improved activity compared to a linear peptide that mimicked only the primary sequence of TTR strands G/H, presumably by orienting the two strands in an anti-parallel arrangement (Cho et al., 2015). The cyclic peptide was superior to mTTR in terms of resistance to proteolysis and retention of activity in the presence of non-specific proteins, but was inferior to mTTR in terms of efficacy at low concentrations (Pate et al., 2018). We hypothesized that both sequence similarity and conformational similarity between the designed peptide and the native Aβ-binding domain in TTR are essential for optimum performance of the peptide. Given evidence that cyclic peptides cG3 and cG8 still contained a significant amount of disorder, we proposed that stabilization of the two-stranded anti-parallel β-sheet structure was needed to obtain conformational convergence to the TTR-derived template (Cho et al., 2015; Lu et al., 2016).

We first attempted stabilization via insertion of a disulfide bond linking the two strands. Somewhat surprisingly, we found that our attempt failed: cyclization precluded disulfide bond formation, and vice versa. We used the recent SCPA adaptation of ROSETTA that incorporates non-canonical amino acids and cyclization into simulations to assist in our evaluation of alternative designs (Bhardwaj et al., 2016). This analysis demonstrated that attempts to insert the disulfide bond into cG8 led to a loss in stability and an increase in heterogeneity of lowest energy conformations that included kinked structures, thus providing a rational explanation for the difficulty we faced in synthesizing cG8_ss. Further simulations led us to the solution of using not one, but two ^DProPro β-turn templates. cG8_ss/dProPro was readily produced with a disulfide bond, and CD analysis showed that this peptide had greater β-sheet content than cG8. Interestingly, deleting the disulfide bond produced a peptide, cG8_dProPro, with even greater stability, conformational homogeneity and β-sheet content. Based on ROSETTA simulations, disulfide bond formation results in a near-planar β-sheet while the absence allows for a slight twist, a more natural conformation that is likely the reason for cG8_dProPro’s deeper energy well (Santiveri et al., 2004; Alba et al., 1997).

We evaluated the relative efficacy of each of the designed peptides at inhibiting Aβ fibril formation. We asked whether conformational rigidity and homogeneity was optimal, or if it was preferable to retain some flexibility, allowing the peptide to adapt to the diverse size and conformation of Aβ oligomers. The results demonstrate that cyclization, β-sheet structure and conformational homogeneity are all preferable design features, whereas disulfide bond formation across the two β-strands is not preferable. Simulations of cG8 revealed a relatively large ensemble of conformations, which could include a mixture of strong Aβ binders, weak binders, or some conformations that do not bind at all: the observed activity against Aβ fibrillogenesis is due to binding averaged over all conformations. Thus, binding to Aβ and inhibition of Aβ fibril formation should be enhanced by peptide conformational homogeneity, but only if the predominant structure is a conformation that binds Aβ. cG8_dProPro was the most homogeneous in conformation, occupying the deepest energy well and containing the greatest β-sheet content. cG8_dProPro proved to be modestly more effective than cG8 at inhibiting Aβ aggregation, although the difference between cG8 and cG8_dProPro diminished with longer incubation times. The improvement might be attributed to cG8_dProPro adopting a stronger Aβ-binding conformation, elimination of weaker or non-binding conformations, or a combination of effects.

Our finding that conformational restriction is advantageous in the design of anti-amyloid peptides is consistent with other recent reports. For example, Jha et al. (2018) synthesized a small peptide containing a ^DProGly β-turn template, cyclized by a disulfide link. The disulfide link was necessary for both strong β-sheet signal in CD and for inhibition of Aβ aggregation. That work differs from our results in a few key features: inhibition of Aβ aggregation was observed at 10:1 peptide excess (versus ours which work at Aβ excess), and the proposed mechanism of action was via binding to and stabilizing disordered Aβ monomers, whereas our cyclic peptides preferentially sequester Aβ oligomers. Lu et al. (2019) searched computationally for pentapeptide sequences that bind specific domains in Aβ fibrils, then constrained these sequences to an extended conformation by placing them within a 42-mer macrocyclic scaffold. They found that the imposition of the scaffold constraint greatly enhanced the inhibitory effect against amyloid aggregation and toxicity.

However, cG8_dProPro is not as active as mTTR, and RMSD calculations show that it is not a perfect match for the native structure. A closer comparison of the lowest-energy conformers with the native structure of strands G/H might point to specific areas of mismatch and form the basis for further improvements in peptide design. Our work demonstrates the utility of using SCPA to design peptides that closely mimic natural protein interaction sites and that are conformationally homogenous. To our knowledge, this is the first use of SCPA for design of anti-amyloid agents. Use of SCPA can now be extended not only to optimize cyclic peptides as a structural mimic of the G/H strand on TTR, but also to generate other new hyperstable peptides. For example, it could be possible to take known structures of other proteins that bind to Aβ oligomers, search for and identify common motifs, and then rationally design peptides with enhanced stability that adopt a particular desired motif. In this way, simulation and experiment combined can contribute to the development of the next generation of high-affinity, highly specific and highly stable anti-amyloid peptide therapeutics.

Supplementary Data

Supplementary data are available at Protein Engineering, Design and Selection online.

Funding

Funding was provided by National Institutes of Health R01AG033493, the Wisconsin Alzheimer’s Disease Research Center, and the University of Wisconsin-Madison Biotechnology Training Program funded by National Institutes of Health 5 T32 GM008349. CD spectra were obtained at the University of Wisconsin – Madison Biophysics Instrumentation Facility, which was established with support from the University of Wisconsin – Madison and grants BIR-9512577 (National Science Foundation) and S10 RR13790 (National Institutes of Health), with technical assistance provided by Dr. Darrell McCaslin.

Author contributions

CBE ran simulations, purified peptides, collected circular dichroic spectra and thioflavin T fluorescence data, analyzed data, and wrote sections of the manuscript. PM synthesized and purified peptides and collected mass spectra. RMM developed the experimental strategy and wrote sections of the manuscript.