Stapled beta-hairpins featuring 4-mercaptoproline

Abstract

Peptides constrained by intramolecular cross-links, especially stapled α-helices, have emerged as versatile scaffolds for drug development. However, there are fewer examples of similarly constrained scaffolds for other secondary structures. Here, we used a novel computational strategy to identify an optimal staple for antiparallel β-strands, and then we incorporated that staple within a β-hairpin peptide. The hairpin uses 4-mercaptoproline as a novel staple component, which contributes to a unique, kinked structure. The stapled hairpins show a high degree of structure in aqueous solution, excellent resistance to degradation in cell lysates, and cytosolic penetration at micromolar concentrations. They also overlay with a unique subset of kinked hairpin motifs at protein-protein interaction interfaces. Thus, these scaffolds represent promising starting points for developing inhibitors of cellular protein-protein interactions.

Graphical Abstract

Constrained peptides are a rich source of protein ligands, often binding protein surfaces with high affinity and specificity.^1,2 Many of the methods used to discover constrained peptides have difficulties incorporating non-proteinogenic amino acids and chemical cross-links. Computational design could be a more direct route to advanced scaffolds for drug development. However, prior design efforts have struggled to incorporate non-proteinogenic amino acids and cross-links.^3,4 Here, we describe the computational design of β-hairpins organized by four non-proteinogenic amino acids. The hairpins are further stabilized by a novel intramolecular cross-link (a “staple”) involving the conformationally constrained amino acid 4-mercaptoproline (4MP). The designed β-hairpins are well-structured in aqueous solution, resist degradation in cell lysates, and penetrate the cytosol of cultured cells at micromolar concentrations.

Previously, several groups have employed cysteine alkylation using ortho-, meta- and para-dibromomethylbenzene, the “CLIPS” reaction pioneered by Timmerman and colleagues,⁵ to stabilize loop and α-helical structures.^6–10 Cysteine, D-cysteine, homocysteine, and penicillamine have all been incorporated into structure-stabilizing staples using this chemistry. We envisioned using the same chemistry to staple 4MP (Fig. 1a). 4MP is a hydroxyproline analog that was previously applied to native chemical ligation and studies of proline conformation,^11–14 but not to peptide stapling. Using 4MP for stapling is appealing for several reasons, including the increased torsional rigidity of 4MP relative to cysteine, and the potential to use up to four different stereoisomers (all commercially available and synthetically accessible)¹³ to vary staple geometry.

Figure 1. Stapled peptides with 4-mercaptoproline with β-hairpin structure.

(a) Peptide stapling using thiol alkylation with 4-mercaptoproline (4MP). (b) Hairpin peptides analyzed by circular dichroism (CD) spectroscopy. The predicted β-turn within each sequence is underlined. Lowercase letters denote D-amino acids, X denotes 1-aminocyclopropane-1-carboxylic acid, and 4 denotes (2S,4R)-4-mercaptoproline. Degree of structure is calculated from the Trp-Trp exciton coupling band at 228 nm. Degree of structure is reported relative to HP7, and T₅₀ values represent the temperature at which each peptide was observed to retain 50% structure (Fig. S7–S8); n/m denotes value not measured. (c) CD spectra for HP7, ΔHP7, ΔHP7-GC-p (initial design), Δ4MP-a (computational design, unstapled), and Δ4MP-m (computational design, stapled). (d) CD spectra for HP7, ΔHP7, 4MP-a (computational design, unstapled), and 4MP-m (computational design, stapled). CD data shown are the average of three independent trials.

Given the prominent role of proline in stabilizing β-turns,^15,16 we turned to β-hairpin peptides as a model system to evaluate 4MP-based staples. HP7 is one of the smallest well-structured β-hairpins, and its truncated analog ΔHP7 is less well-structured because it lacks two stabilizing charge pairs (Fig. 1b).¹⁵ ΔHP7 and HP7 have been used previously to test the structural effects of β-hairpin modifications.¹⁷ Our initial strategy employed diversity-oriented stapling guided by simple modeling: we substituted (2S,4R)-4MP directly for proline within ΔHP7, then produced analogs with the 4MP stapled to cysteine substituted at one of two nearby positions (Table S1). Bis-thiol alkylation with ortho-, meta-, and para-dibromomethylbenzene was performed on fully deprotected, unpurified peptides as described,^5,6 with near-quantitative yields. These results demonstrate that 4MP is fully compatible with stapling unprotected peptides in solution. The secondary structures of the peptides were analyzed by circular dichroism (CD) spectroscopy (Fig. S6). These peptides ranged in the extent of their β-hairpin structure, but none were more well-structured than the parent peptide ΔHP7. The most well-structured, ΔHP7-GC-p, had a positive Cotton effect of 8,900 near 228 nm, implying roughly half the level of structure of parent peptide ΔHP7 (Fig 1b,c).

We next sought to apply a more systematic design approach. However, prior large-scale computational design approaches could not be applied to 4MP stapling due to the diversity of possible staple geometries and the non-proteinogenic nature of 4MP. Thus, prior approaches designed scaffolds with well-folded secondary structure, then grafted disulfide bonds onto the already-designed scaffold (Fig. 2a).⁴ In this work, we took a novel approach. We systematically modeled many different staples to find those most compatible with a specific secondary structure, and then the rest of the peptide was grafted onto the designed staple (Fig 2b). We started by systematically identifying which combinations of residues and cross-linker would fulfill the hydrogen bonding geometry of antiparallel β-strands. Conformers were generated for all possible permutations of 4MP (2S,4R and 2S,4S stereoisomers), cross-linker (ortho-, meta-, and para-dimethylbenzene, alkyl chains, and disulfides), and second thiol-containing residue (cysteine and homocysteine) using RDKit (Fig. S1–S2).¹⁸ 10,000 conformers for each permutation were generated and subsequently energy-minimized using AIMNet(SMD)-D4.^19,20 By evaluating the energies of each conformational ensemble, we identified which staples were most compatible with antiparallel β-strands (Fig. S2–S4, Fig. 2b Step 1). Next, we attached the stapled strands to 29 different, crystallographically verified β-turns. A turn containing D-Pro and 1-aminocyclopropane-1-carboxylic acid (Acpc)^21,22 was the only β-turn which templated two amide bonds between residues 1 and 4 of the β-turn, which superimposed closely with the stapled strands. Grafting the D-Pro-Acpc turn onto the stapled strands produced a stapled β-turn (Fig. 2b Step 2). Finally, we sought to introduce the stapled β-turn into the model sequence HP7, but simple substitution was complicated by the 4MP residue which disrupted the canonical β-hairpin structure. We compared models with different L- and D-amino acids on the strand opposite the 4MP, and we observed that a D-amino acid allowed extension of the β-hairpin (Fig. 2b Step 3). The complete β-hairpin was assembled by superimposing the stapled β-strands, the β-turn, the D-amino acid, and the antiparallel β-strands from HP7. Side chains for the final design, 4MP-m, were selected for torsional compatibility with the template (D-Tyr4, Thr6, Gly9, and Gly11), or to match model β-hairpin HP7 (residues 1-3 and 12-14).

Figure 2. Computational design of β-hairpins stapled with 4-mercaptoproline.

(a) Prior design approaches grafted a staple onto an already-designed scaffold. (b) The novel approach described here employed a systematic search for staples that were compatible with the target structure (antiparallel β-strands), and then grafted compatible structural motifs onto the already-designed staple. (c) Chemical structure of designed peptide 4MP-m, with structural elements colored as in (b).

We synthesized 4MP-m and its unstapled analog 4MP-a, which was alkylated with allyl bromide instead of a bifunctional cross-linker. These were compared to parent β-hairpin HP7 using CD spectroscopy (Fig. 1d). 4MP-a and 4MP-m both had the same overall CD signature as HP7, with characteristic minimum at 215 nm indicating β-sheet structure and a strong maximum at 228 nm from exciton coupling of cross-strand Trp side chains.²³ Due to the incorporation of multiple non-natural amino acids, including D-amino acids, we used the strength of the Trp-Trp exciton coupling as an indicator of extent of β-hairpin structure. 4MP-a had a 26% more intense Cotton effect near 228 nm compared to HP7, suggesting that the non-proteinogenic amino acids in the unstapled analog (D-Tyr, D-Pro, Acpc, and 4MP) increased the extent of β-hairpin structure at room temperature. 4MP-m had a 41% more intense Cotton effect near 228 nm compared to HP7, suggesting that adding the staple increased the extent of β-hairpin structure even further. These results indicated that the non-proteinogenic amino acids supported robust β-hairpin structure that was fully compatible with the novel 4MP staple.

To examine the relative roles of the staple and the β-turn, we prepared analogs of 4MP-a and 4MP-m with either of the turn residues D-proline or Acpc substituted with glycine (Fig. S7). These analogs had reduced hairpin structure relative to 4MP-a and 4MP-m. Also, unlike 4MP-m, analogs with altered β-turns had less hairpin structure when stapled. These results indicated cooperativity between the designed staple and the β-turn.

To see if the designed features could stabilize β-hairpin structure within the shorter peptide ΔHP7, we prepared the corresponding analogs Δ4MP-a and Δ4MP-m and compared them to ΔHP7 using CD spectroscopy. The unstapled analog Δ4MP-a had a Cotton effect at 228 nm with similar intensity to that of ΔHP7, and the stapled analog Δ4MP-m had an intensity 41% greater than that of ΔHP7 (Fig. 1c), further suggesting the designed staple promotes β-hairpin structure.

We used CD thermal melts to compare the thermodynamic stability of the designed hairpins and analogs. T₅₀ values of for HP7, 4MP-a, and 4MP-m were 58.7°C, 50.8°C, and 58.5°C, respectively, and similar trends were observed for the truncated analogs ΔHP7, Δ4MP-a, and Δ4MP-m. These data indicated that, despite being more well-structured at room temperature, the stapled peptides do not have significantly different thermal stability. van’t Hoff analysis revealed that, compared to HP7 and ΔHP7, unstapled analogs 4MP-a and Δ4MP-a had greater enthalpic contributions to folding. Also, the stapled analogs had a smaller entropic barrier to folding than their corresponding unstapled analogs (Fig. S11–S12, Table S3). Further, the analogs with glycine substitutions in the β-turn lost favorable enthalpic contributions to folding relative to 4MP-m, further indicating cooperativity between the designed staple and the β-turn. These data are consistent with the design strategy.

Next, we used NMR spectroscopy to examine the aqueous solution structure of 4MP-m at higher resolution. We identified 56 NOEs between non-consecutive residues, including many cross-strand interactions (Fig. 3a and Table S5). These NOEs were used to constrain customized simulated annealing simulations that included molecular dynamics in explicit water.²⁴ These simulations produced an ensemble with a low-energy cluster that was consistent with 92% of the observed NOEs (Fig. 3b) and no violations greater than 0.6 Å (Table S5). Comparing the lowest-energy structure of this cluster to the predicted structure resulted in a backbone RMSD of 0.91 Å, and the major difference was the type I′ β-turn at residues 7 and 8 instead of the type IIʹ β-turn that was predicted (Fig. S18). As an independent measure of extent of structure and intramolecular hydrogen bond formation, we performed variable-temperature NMR experiments (Fig. S16, Table S6). We observed protected chemical shifts consistent with hydrogen bond formation for the amide protons of D-Tyr4, Thr6, Gly9, and Gly11 (Fig. 3c). These data are fully consistent with the results from CD and from NOE-guided simulated annealing. All together, these data indicated that, in aqueous solution, 4MP-m adopts a unique, kinked structure in which the 4MP staple and the D-Tyr residues position the β-turn roughly perpendicular to the rest of the β-hairpin.

Figure 3. 3D structure of 4MP-m in aqueous solution.

(a) Chemical structure of 4MP-m with selected medium-range NOEs shown as blue double-headed arrows. Complete NOEs are shown in Table S6. (b) Predicted structure of 4MP-m (magenta) overlaid with a low-energy cluster of 12 structures derived from simulated annealing (gray lines). (c) Lowest-energy calculated solution structure of 4MP-m. Side chains not involved in the kinked turn are omitted. Magenta dotted lines indicate hydrogen bonding independently supported by variable temperature NMR data (Fig. S16 and Table S6).

One of the potential advantages of incorporating non-proteinogenic amino acids and covalent staples is resistance to proteolytic degradation. To test whether the designed β-hairpins were stable to degradation, we adapted a cell lysate stability assay recently applied for development of olefin-stapled α-helices.^25–27 Because HeLa cell lysate contains all cellular proteases including lysosomal proteases, this represents a relatively rigorous test for metabolic stability of peptide therapeutics. Control peptides HP7 and ΔHP7 were degraded rapidly, with roughly 0% and 19%, respectively, remaining after only two hours (Fig. 4a). Few discrete degradation products of these peptides were detected by mass spectrometry. 4MP-a and 4MP-m, the unstapled and stapled analogs of HP7, were also degraded relatively rapidly (roughly 22% and 7%, respectively, remaining after 4 hours). However, even after 24 hours in HeLa cell lysate, the primary degradation products for 4MP-a and 4MP-m were intact except for the two N-terminal amino acids. The truncated analogs Δ4MP-a and Δ4MP-m were much more stable to degradation in cell lysates, with calculated half-lives of 30 and 21 hours, respectively. No sulfide oxidation was observed for any hairpin peptides, even after 24 hours of incubation with cell lysates.

Figure 4. Biological degradation and cytosolic penetration of designed β-hairpins.

(a) β-hairpin peptides were incubated for selected time points in HeLa cell lysate at 37 °C. Enzymatic reactions were quenched by adding methanol and supernatants were analyzed by analytical HPLC. Areas under each peptide chromatogram peak were normalized to the area under the zero timepoint chromatogram peak. Average values and standard errors of the mean were calculated from three independent replicates. Asterisks indicate zero or near-zero values. (b) Cytosolic penetration of chloroalkane-tagged peptides as measured by the chloroalkane penetration assay.^30,31 Error bars denote standard errors of the mean from three independent replicates. Curves (dotted lines) were fitted and CP₅₀ values were calculated as described.^30,31

Another sought-after property for constrained peptide scaffolds is the ability to penetrate cells and localize to the cytosol.^28,29 While cytosolic penetration depends greatly on identity of side chains, some constrained peptides (such as olefin-stapled α-helices) represent privileged scaffolds that can be engineered for cytosolic penetration, while some others (such as lactam-stapled peptides) are rarely observed to be cell-penetrant.^28,29 Thus, it was unclear whether stapled β-hairpins represented a scaffold with measurable cytosolic penetration. We examined the extent to which stabilized β-hairpins penetrate the cytosol using the recently reported Chloroalkane Penetration Assay (CAPA).^30,31 We produced hairpin peptides with chloroalkane tags on their N-termini and measured their cytosolic penetration after a 4-hour incubation using CAPA (Fig. 4b). Surprisingly, chloroalkane-tagged versions of HP7 and ΔHP7 had CP₅₀ values of 4.7 and 0.9 μM, respectively. These data could be confounded by degradation.³¹ The less degradation-prone 4MP-a and 4MP-m showed essentially no cytosolic localization when incubated with cells up to 20 μM. By contrast, chloroalkane-tagged Δ4MP-a and Δ4MP-m had CP₅₀ values of 5.4 and 9.6 μM, respectively, indicating substantial cytosolic penetration after only 4 hours at these concentrations. Notably, these are only two- to three-fold higher than the analogous CP₅₀ value for Tat (3.1 μM),³⁰ demonstrating that these stapled hairpins have a moderate ability to penetrate to the cytosol. We also observed no signs of toxicity for any peptide up to 30 μM, as measured by the proportion of live cells in each sample when gating for CAPA.³¹ Overall, the CAPA data indicated that the stapled β-hairpin was more cell-penetrant when it was smaller in size and had fewer charges. This observation matches prior work on stapled α-helices.^32,33 Moving forward, these scaffolds will allow us to examine whether the biophysical properties known to promote cytosolic localization and bioactivity for stapled α-helices^32,33 will apply in the context of stapled β-hairpins, considering they are similar in size and physicochemical properties.^34–37

Relative to previous work in the field,^3,4 this work provides novel computational methodology that can be applied to any staple and any secondary structure. The designed β-hairpins 4MP-m and Δ4MP-m display a high degree of structure, matching or exceeding the level of structure observed for previously described β-hairpins based on HP7 and ΔHP7.^15,17 In fact, the small temperature dependences of the chemical shifts of D-Tyr4 and Gly9 amide protons are within the range typically observed for the folded core of globular proteins.^38,39 The kinked structure of 4MP-m is unique compared to HP7 and other model β-hairpins (Fig. S19).^15–17,23 To assess whether the kinked hairpin mimics protein-like structures, we aligned the backbone of 4MP-m with all possible 12-residue fragments in a non-redundant subset of the Protein Data Bank. We found many close alignments, including numerous hairpins at protein-protein interfaces (Table S7, Fig. S20–S21). These results indicate that 4MP-m and related hairpins will be useful starting points for the design of unique protein-protein interaction inhibitors.

In this work, we also demonstrate the versatility of 4MP as a novel component of structure-stabilizing staples. While we employed the (2S,4R) and (2S,4S) stereoisomers, we anticipate the other two stereoisomers will be equally useful as building blocks for peptide stapling, especially as analogs of D-proline. Proline and D-proline are key structure-promoting residues for β-hairpins, but also for many other secondary structures including other β-turn-nucleated loops, α-helix caps, polyproline helices, and collagen-like helices and assemblies.^40–42 Additionally, a computationally designed set of over 200 well-structured head-to-tail cyclic peptides was previously reported, and over 98% of these cyclic peptides had one or more prolines or D-prolines.³ Overall, we anticipate that the application of 4MP staples to these and other structures will provide straightforward ways to generate highly structured cyclic and multicyclic peptides, spanning many diverse structural classes and providing unique scaffolds for drug development.