Bent Into Shape: Folded Peptides to Mimic Protein Structure and Modulate Protein Function

Abstract

Protein secondary and tertiary structure mimics have served as model systems to probe biophysical parameters that guide protein folding and as attractive reagents to modulate protein interactions. Here we review contemporary methods to reproduce loop, helix, sheet and coiled-coil conformations in short peptides.

Introduction

Interactions between protein binding partners control essential cellular processes. Traditional drug discovery has focused on orally bioavailable small molecules that bind to defined protein pockets. Similar deep pockets have been hard to detect on large protein-protein interfaces.^1–3 Although, protein interfaces are large, analysis has shown that protein-protein interactions (PPIs) are often mediated through a small subset of “hot-spot” residues that contribute significantly to the free energy of binding.^4–6 This analysis suggests that it may be possible to develop low molecular weight competitive binding inhibitors if the binding hot spots are concentrated in a small region.^7–8

Protein secondary and tertiary structures often serve as hot spot hubs to mediate protein-protein interactions.^{7, 9–13} Peptide mimics of these secondary or tertiary structures can be rationally designed as PPI inhibitors.^12–13 A significant challenge in using short peptides as secondary and tertiary structure mimics is that short sequences rarely retain a defined fold once excised from the native protein. For example, short peptides containing less than 15 residues typically do not adopt stable secondary conformations;¹⁵ for tertiary motifs and miniproteins, this length range is roughly 30 residues.^16–17 Peptides are also prone to proteolytic instability, which limits their viability as reagents in molecular biology and drug discovery. The proteolytic instability of peptides is correlated with their propensity to adopt an extended conformation, as proteases selectively recognize and cleave peptides in this conformation.¹⁸ Structural stabilization of peptides allows display of hot-spot residues in their bioactive orientation and improves their proteolytic stability.

Several classes of molecules that mimic secondary and tertiary structures have been developed, including peptides and small molecules. Small molecules that reproduce protein surfaces are not discussed here but have been extensively reviewed in the literature.^{12, 19–21} In this review, we focus on peptide mimics that reproduce the conformation of α-helices, β-sheets, tertiary helical bundles, and non-regular loops (Figure 1).

Figure 1.

Examples of (A) α-helices, (B) β-sheets (C) helix dimers, and (D) irregular loops at protein-protein interaction interfaces. PDB codes: (A) 1BXL (Bcl-xL/Bak); (B) 1F3U (Rap30/Rap74); (C) 3CL3 (vFLIP/IKKγ); (D) 1NPO (Bovine neurophysin II/oxytocin)

I. α-Helix Mimics

The α-helix is the most prevalent secondary structure and features prominently in molecular recognition of biomolecules. The α-helix is composed of 3.6 residues per turn, resulting in a network of hydrogen bonds between every C=O at the i^th position and NH of the corresponding i+4^th residues. This repeating main chain hydrogen bonding pattern results in the display of side chain functionality on three different “faces” of the α-helix, such as the i, i+4, i+7, and i+11 side chains project from one face. Analysis of helix mediated interactions has revealed that 60% of helical interfaces feature hot spot residues on one face of the helix, one-third feature helices with hot spots on two faces, and roughly 10% require all three faces for interaction with their target protein (Figure 2).²²

Figure 2.

Interfacial α-helices may use one, two or all three faces to recognize partner proteins. Examples of each molecular recognition category are depicted. (PDB entries 1xl3, 1xiu, and 1or7).

Systematic investigation of protein-protein interactions in the Protein Data Bank indicates that the typical length of an interacting α-helix is 8–12 residues, or two to three helical turns.^22–23 This observation suggests that isolated 8–12 residue peptides, spanning 900–1400 MW range, that reproduce the native sequence should recapitulate the interaction affinity and specificity observed in the context of the full-length protein. However, short peptides rarely fold into a defined helical secondary structure because the entropic cost of nucleating a turn of the α-helix is not compensated by the free energy gained from main chain hydrogen bonding and various side chain interactions until the peptide length reaches 15–20 residues, depending on the sequence.¹⁵

Several groups have developed strategies to constrain peptides into helical conformation. In general, successful approaches involve strategic placement of a covalent bond that lowers the α-helix nucleation barrier and reinforces the hydrogen bonding network. These molecules can be classified based on the number of α-helix faces available for molecular recognition (Figure 3). Side chain crosslinked peptides have been extensively studied and are capable of presenting two faces of the α-helix, with the third face being blocked by the crosslinker. In a complementary fashion, mimicry of all three α-helical faces has been achieved using the hydrogen bond surrogate (HBS) approach to mimic a main chain hydrogen bond. Foldamer helices developed from β-amino acids or a mixture of α/β-residues can project some side chains along all three α-helical faces.

Figure 3.

Stabilized helices and non-natural helix mimetics.

Mimicking Two α-Helical Faces with Side Chain Crosslinked Peptides

A classic strategy for stabilizing the α-helical conformation in peptides utilizes covalent bonds between side chain functionality on the same α-helical face i.e., between i and i+4, i+7, or i+11 side chains (Figure 4). At the i and i+4 positions, these covalent crosslinks mimic salt bridge interactions present in native protein helices.²⁴ This approach was first described by Felix et al.,²⁵ who developed lactam bridged helix mimics of human growth hormone-releasing factor (GRF). The side chain crosslinking approach has been expanded since to include various chemistries, such as disulfide, thioether, triazole, and olefin bonds.

Figure 4.

Peptide stapling techniques for -helix stabilization with i to i+3, i+4, i+7 and i+11 linkers. Linkers that include disulfide, thioether, lactam, triazole and olefin bonds have been included into peptides.

Lactam Crosslinking

Crosslinking of amine- and carboxylic acid-bearing residues on the same face of the α-helix stabilizes its conformation.²⁶ The lactam-bridged helices are the oldest class of side chain crosslinked helices.²⁵ Detailed analysis of side chain length and relative orientation of the amine and carboxylic acid side chains by Fairlie et al. identified lysine at position i and aspartate at position i+4 as optimal side chains for the lactam staple to stabilize the α-helical conformation.²⁷ They also performed a comparative study of various stapling techniques using a model pentapeptide, showing that the Lys/Asp lactam stapled peptide exhibited greater α-helicity than any other stapled peptide tested (CuAAC-derived triazole, hydrocarbon, bis-thioether, or alkyl thioether stapling).²⁸

Two-component approaches have also been used to install lactam bridges that link position i to i+4, i+7, or i+11 to nucleate the α-helix conformation. In general, the two-component approach attaches two amino acid residues with a bifunctional linker to form the crosslink. Originally developed by Phelan and co-workers, an example crosslink spanned two turns of the α-helix, utilizing glutamic acid residues at positions i and i+7.²⁹ Direct bis-lactamization using a diaminoalkane resulted in poor yields and purity, so a stepwise procedure was developed. The authors used 9-fluorenylmethyl and allyl ester protecting groups on the Glu residues and a mono-Boc-protected diaminoalkane to create the two-component staple. Further analysis revealed four or five methylene units in the diaminoalkane linker provided the greatest level of helicity in two different peptide sequences. Fujimoto et al. performed a screening study which analyzed different linker lengths and rigidities of twelve disuccinimidyl ester linkers for coupling to peptides containing two alkylamine bearing side chain residues at positions i to i+4, i+7, and i+11.³⁰ Interestingly, they found that helical structures were optimized when the lengths of the cross linkers were shorter than the target distances between the two amino groups, ideally about 50–60% of the full pitch.

Lactam stapled peptides have been used to target a variety of PPIs, including MDMX/p53 and MDM2/p53 interactions with nanomolar potency in vitro.³¹ Philippe et al. overcame the poor cell permeability of these peptides by attaching a cytoplasmic transduction peptide (CTP) to their lactam stapled peptide. Several designs were tested with different lactam bridge locations as well as bis-lactam designs. All maintained nanomolar binding to MDM2 and MDMX, while the CTP linked designs also showed toxicity against both MM96L and HeLa cells. Flow cytometry indicated that the CTP-linked designs were capable of getting into cells. This work showed the potential of lactam stapled designs in specific binding, while also addressing the issue of cell penetration.

Disulfide Crosslinking

Disulfide crosslinking provided another early example of stabilizing short peptides in the α-helical conformation. In 1991, Schultz et al. reported the incorporation of two N-Fmoc-S-(acetamidomethyl)-2-amino-6-mercaptohexanoic acid residues into a model peptide at the i and i+7 position.³² Concomitant deprotection and disulfide formation yielded peptides that adopted stable α-helices based on circular dichroism (CD) spectroscopy. In 2003, it was determined that a disulfide formed between D-cysteine at position i to L-cysteine at position i+3 could stabilize a peptide in the α-helical conformation, though only when D-cysteine was placed near the peptide’s N-terminus in order to orient the connecting side chains towards one another.³³ The salient feature of the disulfide bond is that it can be reversibly formed, which enables biophysical studies on helix formation.^34–35

However, the reducing potential in the cytosol of eukaryotic cells is sufficient to reduce most disulfide bonds, limiting the use of disulfide-stapled α-helical peptides in biological applications.

Thioether Crosslinking

Thioether crosslinking takes advantage of the nucleophilicity of cysteine thiol groups under conditions where other common peptide nucleophiles (e.g., amine, carboxylate, hydroxyl, or guanidinium groups) are largely unreactive. Most thioether crosslinking approaches are two-component approaches (i.e., two cysteines reacting with a bis-electrophile). One exception from Brunel and Dawson uses a cysteine residue reacts with a bromoacetamide group installed on an amine-bearing side chain (e.g., lysine or ornithine) to generate the crosslink.³⁶ Peptides containing cysteine at position i and ornithine at position i+3 staples were found to exhibit significant α-helical character by circular dichroism relative to unconstrained analogs. Incorporation of this i-to-i+3 staple into a gp41 peptide epitope showed that the peptide with a centrally placed thioether staple bound more efficiently to a gp41-specific antibody 4E10 than the uncyclized peptide.³⁷

In addition to the single-component crosslinking approach above, many thioether designs use a two-component reaction between two peptide cysteine residues and a bis-electrophile to generate crosslinked peptides. One advantage of the two-component technique is the diversity of bis-electrophiles that can be used to form crosslinks with a single peptide. The most commonly used bis-electrophiles contain two bromides, especially benzylic and allylic bromides. In screening 24 different bis-electrophiles for i and i+4 cysteine alkylation on a model peptide, Jo et al. found that rigid crosslinkers, such as o-, m-, and p-xylene, were most successful in stabilizing helical structure.³⁸ Additionally, Pentelute and co-workers developed a two-component thioether staple that employs an S_NAr-based stapling technique between two cysteine residues placed at positions i and i+4 using hexafluorobenzene. Importantly, this approach was used on unprotected peptides.³⁹ This linker was incorporated into a peptide that interacts with the HIV-1 capsid assembly polyprotein and showed improvements in helicity, protease stability, in vitro binding affinity, and cellular uptake.

More sophisticated control of peptide conformation was described by Woolley et al. using a two-component thioether azobenzene crosslink to allow photoswitchable folding into an α-helix.^40–41 Cysteine residues at positions i and i+7 were reacted with two iodoacetamide groups flanking an azobenzene linker. Upon irradiation, the linker was designed and found to switch from trans to cis conformation, simultaneously driving the peptide from a non-helical conformation into a helical conformation. After several hours in the dark, the peptide reverted back to the non-helical conformation. Woolley and coworkers expanded this technique in both i, i+4 and i, i+11 stapling as well, determining the i, i+4 linkage promotes helicity while the i, i+11 staple is destabilizing if the azobenzene is in the excited cis form.⁴² Photoswitchable helices have been applied to bind several proteins, including the oncogenic protein Bcl-xL⁴³ and AP2, a key regulator of clathrin-mediated endocytosis.⁴⁴ The latter application permitted the use of light as a simple input to control protein endocytosis.

Cycloaddition Crosslinking: Triazoles and Beyond

Cu(I)-catalyzed azide-alkyne cycloaddition (CuAAC) was first applied for helical crosslinking by Chorev, D’Ursi and co-workers using model peptides derived from parathyroid hormone-related peptide.⁴⁵ Using solution-phase CuAAC between azide- and alkyne-bearing side chains at positions i and i+4 in a fully deprotected peptide, the authors determined that peptides with five or six methylene units in the crosslink were the most helical. Wang and co-workers further studied triazole-crosslinked peptides with the goal of inhibiting the oncogenic interaction of Bcl9 with β-catenin.⁴⁶ They created a library of triazole based stapled peptides and determined the optimal linker length to be five methylene units. They also were successful in incorporating two triazole staples into a single peptide. Their optimized designs improved in vitro binding to their target as well as proteolytic stability. On-resin stapling of triazole linkages has also been developed.

Two-component triazole crosslinks spanning two turns of the helix have also been synthesized by reacting two azide-bearing side chains at positions i and i+7 with a symmetrical benzene ring bearing two alkynes. Substitution at the meta position (relative to either alkyne) allows further functionalization of the linker and adds greater versatility to this stapled design.^47–48 For instance, these modified linkers have been modified with arginine residues to increase cell permeability. Spring and coworkers built on initial two-component designs using strained diynes for copper-free crosslinking. Impressively, this approach enabled the first example of in situ crosslinking in cell culture and was employed to inhibit p53-MDM2.⁴⁹

Beyond CuAAC cycloaddition, Madden et al. have utilized the UV-induced cyclization reaction between tetrazoles and alkenes to crosslink peptides for inhibition of the p53-MDM2/MDMX interaction.⁵⁰ Under UV irradiation, the tetrazole and alkene react to form a pyrazoline crosslink between the i and i+4 positions. A unique property of these crosslinks is their intrinsic fluorescence, enabling imaging of peptide localization in cells without conjugating an extra fluorophore. The crosslinked peptides, which target the p53/MDM2 interaction, exhibit higher affinity for their p53 target relative to the linear peptide. Addition of several arginine residues allowed the authors to obtain modest levels of cellular p53 activation. Moellering et al. utilized the Diels Alder [4+2] cycloaddition to generate a wide array of stabilized and macrocyclic peptides.⁵¹

Olefin Crosslinking

Olefin-based crosslinks have proven to be an extremely useful technique for α-helix stabilization, as well as PPI targeting and drug design. Early designs by Blackwell and Grubbs joined the terminal olefins of O-allyl serine or homoserine residues at positions i and i+4 using ring closing metathesis (RCM) to form the staple; however, these designs showed only a modest increase in helicity.⁵² To improve helical character, Schafmeister et al. used RCM to link α,α-disubstituted olefin-bearing amino acids, capitalizing on the helix-promoting effect of α,α-disubstituted amino acids like α-aminoisobutyric acid (Aib).⁵³ In these studies, optimized side-chain lengths and stereochemistry were identified for both i to i+4 and i to i+7 designs. Optimal designs display significant helical content and enhanced proteolytic stability compared to unstapled counterparts. In addition to improving α-helical character, more recent studies show that an olefin crosslink generally increases cellular uptake, particularly when it expands an existing hydrophobic peptide surface.⁵⁴ In 2013, it was shown that monosubstituted olefinic amino acids provide similar stabilization of the α-helical conformation, which proved important in generating the first stapled peptide inhibitor that binds with all three faces to its target protein.⁵⁵

Beyond single crosslinking, peptides with multiple hydrocarbon cross-links have been developed to further improve α-helical character and proteolytic stability. Bird et al. have presented a “double stapling” approach with two independent olefin crosslinks, which confers greater stability to long peptides like enfuviritide and enhances their bioactivity.⁵⁶ In another approach, Verdine et al. use a bis-olefin amino acid as a common junction point for simultaneous formation of two olefin crosslinks. These “stitched peptides” display a higher degree of chemical and proteolytic stability and an increase in cellular uptake relative to singly crosslinked analogs.^57–59 Similarly, Moore et al. have shown that overlapping staples enhance α-helicity, proteolytic stability for targeting nuclear receptor complexes.⁶⁰ They have also introduced γ-methylated olefin side chains to improve mimicry of branched hydrophobic side chains at PPI interfaces.⁶¹

Olefin stapled peptides are among the most extensively studied peptides for PPI inhibition.^{56, 62–70} PPI targets include EGFR dimerization, BID/Bcl-2, p53/MDM2, Rab6-interacting protein/Rab GTPase, EZH2/EED, estrogen receptor dimerization, HIV viral particle and gp41 assembly, WASF3/CYFIP1, and ExoS/14-3-3.Both BID BH3 and p53 peptides have been more deeply examined with xenograft mice. BID BH3 peptides proved capable of inducing apoptosis in leukemia cells and inhibiting the growth of human leukemia xenografts in mice. Crosslinked p53 peptides showed promising levels of in vitro binding, cellular activity and suppression of human xenograft tumors in animal models. Most importantly, p53-derived peptides have progressed to clinical trials for lymphomas and acute myeloid leukemia.⁶³

Mimicking α-Helical Faces with Foldamers

Beta Peptides

Foldamers are oligomers with enhanced propensity to adopt a defined conformation.^71–72 β-Peptide foldamers have been shown to mimic the α-helical conformation (also known as the 13-helix for comparison). Significantly, β³-peptides are poor substrates for proteases and thus resist proteolytic degradation.⁷³ Peptides synthesized entirely of β³-amino acids adopt a helical conformation known as a “14-helix.” The 14-helix contains a C=O(i)-N-H(i-2) hydrogen bonding pattern. The propensity to form the 14-helix can be increased using trans-2-aminocyclohexanecarboxylic acid (ACHC) residue.⁷⁴ β³-peptide 14-helices have been used to target the p53/MDM2⁷⁵ and MDMX^76–77 interactions, HIV gp41 mediated fusion^78–79 and GLP-1R activation with GLP-1⁸⁰ with sub-micromolar affinities. Another helix variant that can be formed by β-peptides is the 12-helix, which consists of C=O(i)-H-N(i+3) hydrogen bonds and 12-atom rings. This structure is promoted by five-membered ring trans-2-aminocyclopentanecarboxylic acid (ACPC) residues.⁸¹

Chimeric peptides that incorporate both α and β-amino acids into the peptide sequence have also been developed.^82–85 The ααβαααβ pattern is well studied and has been shown adopt conformations very similar to the α-helix with the β residues aligned along one helical face. Other patterns used to incorporate β-amino acids are ααβ and αααβ.⁸⁶ These patterns also show high α-helical propensities and have the β residues spiraling around the helix. β-Peptides have been used to target several PPIs, including BH3/Bcl-2⁸⁷ and gp41 assembly.⁸⁸

Peptoids

Peptoids are comprised of N-alkylglycine residues where the amide nitrogen is substituted with amine-derived “side chains”, as opposed to peptide side chains at the α-carbon.^89–91 Peptoids are another class of foldamers that aims to increase proteolytic stability while mimicking α-helical presentation of functional groups. In the absence of an amide NH group, peptoids do not feature main-chain intramolecular hydrogen bond networks and are often often stabilized by side chain interactions. For example, peptoids with chiral, bulky appendages can fold into helices which resemble the type I polyproline helix.⁹² Peptoid based designs have been shown to be promising as an antigen of vascular endothelial growth factor receptor 2 (VEGFR2), both in vitro and in vivo.^93–96

Mimicking Three α-Helical Faces with Hydrogen Bond Surrogate Peptides

Hydrogen Bond Surrogate

A drawback of using side chains to lock peptide conformation is that those side chains are no longer available for interactions with the target receptor. A second challenge is that the crosslinker may interact non-specifically with partner proteins, which is exacerbated when large hydrophobic crosslinkers are employed because hydrophobic and aromatic side chains drive molecular recognition in water.^{9–10, 22, 97} We have explored placement of the crosslink at an internal helix position as an alternative to side chain crosslinking. Our helix stabilization method is based on the helix-coil transition theory in peptides, which suggests that the energetically demanding organization of three consecutive amino acids into the helical orientation inherently limits the stability of short α-helices.^{15, 98–99} The theory points to the formation of the first main-chain hydrogen bond to be energy demanding. Our strategy for the preparation of artificial α-helices involves replacement of the main chain hydrogen bonds with a covalent linkage and is known as the hydrogen bond surrogate (HBS) approach.¹⁰⁰ This approach is inspired by the N-cap nucleation templates developed by Kemp, Muller, Bartlett and Satterthwait, among others.^101–102

In an α-helix, a hydrogen bond between the C=O of the i^th amino acid residue and the NH of the i+4^th amino acid residue stabilizes and nucleates the helical structure (Figure 3). To mimic the C=O--H-N hydrogen bond as closely as possible, we envisioned a covalent bond of the type C=X-Y-N, where X and Y would be part of the i and the i+4 residues, respectively.¹⁰³ Preorganization of amino acid residues in an α-turn is expected to initiate helix formation.^104–105 The exceptional functional group tolerance displayed by the olefin metathesis catalysts for the facile introduction of non-native carbon-carbon constraints in the preparation of peptidomimetics suggested that X and Y could be two carbon atoms connected through a ring-closing metathesis (RCM) reaction (Figure 3).^{52, 106–108} Later iterations used both thioether and disulfide bonds in place of the olefin. All three strategies stabilize the α-helical conformation based on NMR and circular dichroism.

HBS helical mimics have proven successful in the specific targeting of various PPIs both in vitro and in vivo. Specific targets include p53/MDM2,¹⁰⁹ BAK/Bcl-2,¹¹⁰ Ras/Sos,¹¹¹ HIF1α/p300,^112–113 and gp41 helix assembly in HIV.¹¹⁴ The most well-studied HBS mimics, derived from two helical fragments of the cancer-associated transcription factor hypoxia inducible factor 1α (HIF1α), bind to and inhibit the transcriptional activity of the partner protein p300/CBP. We demonstrated that HBS mimics of either helix form stable helical structures in solution, bind to the p300/CBP CH1 domain with submicromolar affinity, downregulate HIF1α-mediated transcription activation in cell culture, and reduce tumor growth in mouse xenograft models.

II. β-Sheet Peptide Mimetics

The β-sheet is a common regular tertiary structure in proteins composed of two or more β-strands.¹¹⁵ Each β-strand adopts a nearly extended conformation with preferred φ and ψ backbone dihedral angles of −135° and 135°, respectively. This combination of backbone dihedral angles and the uniform L-chirality of amino acids in proteins leads to an overall pleated geometry, positioning every other amino acid on the same side of the β-strand and forming two faces for molecular recognition.¹¹⁶ Detailed analysis of the Protein Data Bank shows that approximately one-quarter of β-strands at PPI interfaces engage their targets with a single face, one-third engage their targets with both faces equally, and the remainder have unbalanced contributions from both faces.¹⁰

Because the β-strand lacks local hydrogen-bonding interactions that stabilize protein helices, β-strands are rarely observed in isolation. Rather, they are components of tertiary motifs called β-sheets in which two or more β-strands engage in main-chain hydrogen bonding interactions. For any pair of β-strands in a β-sheet, the relative orientation of the peptide termini can be either the same (parallel β-sheet) or opposite (antiparallel β-sheet). Both theoretical and empirical evidence support the notion that antiparallel β-sheets are more stable than parallel β-sheets.¹¹⁷

In general, most studies on β-sheet peptides have focused on the antiparallel β-sheet in its simplest form – the β-hairpin. β-hairpins are composed of two β-strands in antiparallel arrangement with a short reverse turn of 2–4 amino acid residues connecting the two β-strands.^118–119 The β-hairpin has two faces: 1) the hydrogen bonding face occupied by side chains from all residues involved in cross-strand hydrogen bonding, and 2) the non-hydrogen bonding face occupied by side chains from all other residues. Early studies on β-hairpins focused on the assessment of folding stability in model sequences.^120–123 Building on this foundational work, more recent research has sought sequence-independent stabilization strategies that produce the desired β-sheet fold while maximizing the number of amino acids available for high-affinity, specific molecular recognition of various targets, including proteins and nucleic acids.^124–126 As in earlier reviews and the preceding section on α-helical peptides, we will classify β-sheet stabilization strategies based on which regions of the β-sheet retain functional groups for molecular recognition.

Compared with α-helices, specific PPI targeting with β-strand/sheet mimics is in its infancy.¹²⁵ As a result, we will discuss not only β-sheet stabilizing strategies that have successfully targeted PPIs but also various strategies for potential PPI targeting (Figure 5).

Figure 5. Three Classes of β-Sheet/β-hairpin Peptide Mimics for Molecular Recognition.

(A) Molecules in the top left have planar functional groups that position backbone amide hydrogen bond donors and acceptors in a β-strand arrangement. This β-strand mimic templates folding of a second segment of the peptide into another β-strand for molecular recognition. (B) Molecules on the right have either non-covalent or covalent side chain interactions between β-strands on one face to reinforce β-sheet hydrogen bonding, leaving a single face free for molecular recognition. (C) Molecules on the bottom left have functional groups that stabilize β-sheet formation independent of any amino acid side chains. This approach allows both β-sheet faces to participate in molecular recognition. A model β-hairpin structure is shown in the center.

Single Face, Single β-Strand Mimics

A wide variety of strategies have been developed to stabilize peptides into extended structures that mimic individual β-strands. Typically, these strategies involve a new covalent linkage between two side chains on the same side of a β-strand. As one face is occupied by this added crosslinker, only a single β-strand face remains available for molecular recognition. Backbone modifications that stabilize or mimic C5 hydrogen bonding have also been explored towards the same goal.¹²⁷ Many excellent reviews on this topic have been published,^128–129 and as our focus here is on β-sheets composed of two or more β-strands, we will not discuss these strategies in detail.

Mimicking Both Faces of an Individual Strand using a β-Hairpin

Many protein-protein interactions feature “edge-on” β-sheet interfaces, where amide NH and C=O groups from one protein’s β-strand form hydrogen bonding interactions with amide NH and C=O groups in an existing β-sheet in a different protein.¹³⁰ This so-called “β-sheet augmentation” is commonly observed in specific PPI domains, including PDZ and PTB domains.¹³¹ In this scenario, side chains from both faces of the β-strand can interact with the protein as well, providing additional binding affinity and forming the basis for interaction specificity.

To generate well-folded β-strands in which both faces remain available for molecular recognition, a templating strategy is generally employed.^132–133 In this approach, one or more functional groups are introduced within a defined peptide region to arrange amide NH and C=O groups to form a β-strand. Prearrangement of hydrogen bond acceptors and donors thus provides a template for a different peptide region to form hydrogen bonds and thereby fold into a β-strand. The templated β-strand can then interact further with a protein of interest. Key examples of β-strand templating functional groups include Hao (2-[3-(hydrazinecarbonyl)-4-methoxyanilino]-2-oxoacetic acid), MOPA (5-(aminomethyl)-3-methoxy-4-methyl-1H-pyrrole-2-carboxylic acid), and @-tide (1,6-dihydro-3(2H)-pyridinone).^134–136

Hao is a tripeptide mimic developed by the Nowick group and is derived from hydrazine, 5-amino-2-methoxybenzoic acid, and oxalic acid.¹³⁴ Hao replaces the central amino acid of a tripeptide with a methoxy-substituted aromatic ring to arrange hydrogen bond donors and acceptors in a planar geometry that mimics a β-strand. Incorporation of Hao into linear or cyclic peptides promotes formation of antiparallel β-sheets and has been leveraged to examine the role of β-sheet structures in various PPIs, including amyloid β, α-synuclein, and Tau amyloids.^137–139

A similar but smaller β-sheet template is MOPA, a dipeptide mimic developed by König et al. to stabilize β-hairpins.¹⁴⁰ In the context of a peptide, the MOPA C=O and pyrrole NH mimic the C=O and NH groups of an amino acid. These functional groups template β-sheet hydrogen bonding with a distal peptide region, even though the ψ-equivalent angles of MOPA (+70° and −176°) deviate from ideal β-sheet ψ angles. Two adjacent MOPA units template a β-hairpin-like structure, though this template has not been used to target PPIs yet.

An even smaller β-sheet template, @-tide, was developed by Bartlett et al. to replace the C5 hydrogen bond between a single amino acid’s NH and C=O group with a covalent linkage to enforce the β-sheet conformation.¹³⁶ @-tide was originally used to examine the role of various side chain interactions within a stabilized β-hairpin, taking advantage of @-tide’s unique, folding-sensitive spectroscopic signature. A derivative called aza-@-tide permits incorporation of substituents on the pyridinone ring to mimic amino acid side chains.¹⁴¹ Aza-@-tide peptides have been used as isolated β-strands to target PDZ domains, but no @-tide or aza-@-tide β-sheets targeting PPIs have been reported yet.

Mimicking a Single Face Across Both β-Hairpin Strands

Another commonly employed strategy to stabilize β-hairpins is the use of additional interactions to reinforce the native hydrogen bond network between β-strands. Analogous to crosslinking techniques in α-helices, these interactions require specific side chain residues such that a single face of the β-hairpin remains unhindered for molecular recognition. In general, stabilizing interactions within β-hairpins are introduced on the non-hydrogen bonding face because the side chain geometry on this face allows closer side chain association across the two β-strands.^{115, 132} Both covalent and non-covalent strategies have successfully stabilized β-hairpins.

Covalent crosslinks that have proven useful in β-hairpin stabilization include formation of a cystine disulfide from cysteine residues¹⁴² and 1,4-triazole linkages formed between azide- and alkyne-bearing side chains using copper(I)-catalyzed azide-alkyne cycloaddition (CuAAC).^143–144 In the case of the former, the geometry of cystine disulfides requires that the two cysteines be placed across from each other at non-hydrogen bonding positions to enhance β-hairpin stability. The placement of the triazole crosslink is more flexible because the amino acids used (typically propargylglycine and an azide-bearing side chain, often azidolysine) have longer side chains with more degrees of freedom. Still, these amino acids only stabilize β-sheets when the residues are across from each other.

Aromatic residues, especially tryptophan, feature significantly in β-strand-stabilizing non-covalent interactions. In early analyses of amino acid bias in protein secondary structure, it was determined that Phe, Tyr, and Trp are over-represented in β-sheets.¹⁴⁵ It was further demonstrated that interactions between aromatic residues clustered into specific geometries, some of which were consistent with β-sheet formation.^146–147 Inspired by these findings, researchers at Genentech designed the “tryptophan zipper,” or trpzip, that showed remarkably high β-sheet conformational stability.¹²² In these and subsequent studies,¹⁴⁸ it was shown that aromatic residues at non-hydrogen bonding positions prefer to stack in a stabilizing edge-to-face geometry, giving rise to unique spectroscopic signatures by NMR and circular dichroism. Andersen et al. later developed a shorter stable β-hairpin called HP7 using this approach.¹⁴⁹ In a similar vein and taking inspiration from cation-π interactions in natural PPIs such as bromodomains, the Waters group has shown that interactions between alkylamine-bearing amino acids and Trp stabilize β-hairpins when they are placed at opposite positions in a model β-hairpin.¹⁵⁰ Amine methylation and side chain length both strongly impact the stabilizing cation-π effect on conformational stability.

Combinations of these strategies have recently been employed to generate stable β-sheets that do not require a reverse turn to connect the β-strands. Andersen et al. demonstrated that peptides containing a central cysteine disulfide and terminal cation-π capping interactions fold into highly stable β-sheets.¹⁵¹

Despite the potential of the approaches above for mimicking a single β-sheet face for PPI inhibition, studies to date have focused largely on β-sheet folding and stability.

Mimicking Both Faces on Both β-Strands

Perhaps the most tunable approach to β-sheet stabilization for molecular recognition requires the development of stabilizing moieties that do not use side chains from either β-strand but covalently enforce the β-sheet hydrogen bond network by other means. In general, these approaches involve either stabilization of the inter-strand reverse turn, macrocyclization via the β-sheet N- and C-termini, or minimalist enforcement of C5 hydrogen bonding within β-strands. Such strategies allow manipulation of all β-sheet amino acid residues to finely tune protein binding affinity and specificity.

Reverse Turn Stabilization

A popular strategy to stabilize β-sheets is to enforce the reverse turn between β-strands. Early synthetic scaffolds that mimic reverse turns included dibenzofuran and azobenzene, with the latter having the advantage of photoswitchability.^152–154 Irradiation of an azobenzene-containing analog of GB1 yielded two major folded peptide conformations, one of which mimics a β-hairpin. However, neither of these strategies has been used so far to inhibit PPIs.

Alternatively, the use of D-amino acids for reverse turn stabilization was inspired by structural analyses that revealed an abundance of Gly residues in left-handed α-helical (α_L) conformation at specific positions in reverse turns.¹⁵⁵ Because of its conformational rigidity and compatibility with the α_L conformation, D-proline has become the amino acid of choice for enforcing reverse turns. The earliest example of D-amino acid stabilized reverse turns is DPro-LPro, first described and characterized by Balaram et al. in the 1970s.¹²³ Recognizing that DPro-LPro could be used not just in place of a native reverse turn but also as a macrocyclization approach, Robinson and co-workers have designed a number of DPro-LPro macrocycles, termed protein epitope mimetics (PEMs), that bind to specific protein and nucleic acid targets.^{125, 156} A similar DPro-Gly motif has been employed by Gellman and co-workers to similarly stabilize β-hairpins, but it has not been used for PPI inhibition yet.¹²¹

Macrocyclization

Instead of altering the composition of the reverse turn that naturally connects the antiparallel β-strands, another option to induce β-hairpin structure is to introduce a linkage between the peptide’s N- and C-termini, thus forming a macrocycle. One such macrocyclization uses L-ornithine, such that one of the β-strands originates from the Nε of the ornithine side chain.¹²⁶ The combination of L-ornithine macrocyclization with the Hao tripeptide mimic for β-strand stabilization has yielded peptides that allow detailed analysis and inhibition of amyloid PPIs.¹³⁷

Most recently, we have shown that direct replacement of the N- to C-terminal inter-strand hydrogen bond with a covalent hydrogen bond surrogate (HBS) linkage significantly stabilizes the β-hairpin conformation.¹²⁴ These studies were inspired by successful application of the HBS concept to stabilize peptides in the α-helical conformation by replacing a main-chain, i-to-i+4 hydrogen bond with a covalent carbon-carbon linkage. The HBS approach stabilized the β-hairpin conformation in diverse peptide sequences, including one from a designed PPI. This work also emphasizes the potential application of the HBS approach for stabilizing diverse hydrogen-bonded peptide motifs. Studies of HBS β-sheets for PPI inhibition are ongoing.

Mimicking C5 Hydrogen Bonds

The smallest β-sheet template reported to date involves the use of N-aminated amino acids to replace the C5 hydrogen bond.^157–158 Applied to the model β-hairpin GB1 at non-hydrogen bonding positions, this strategy preserves all side chains for molecular recognition but has not been used in such applications yet.

Coiled-Coils

Another commonly observed tertiary structure involved in PPIs is the coiled-coil.^159–161 They are characterized by a heptad repeat, (abcdefg)_n, where positions a and d are hydrophobic and form knobs-into-holes packing interactions.¹⁶² The helices wind around each other in a supercoil with varying twist. Electrostatic interactions often contribute additional stabilization both within each peptide and between the two peptides. Similar to β-sheets, the helices can be oriented with both N-termini together (parallel) or at opposite ends (antiparallel) of the coiled-coil (Figure 6). Traditional de novo design rules require long sequences (>3–4 heptads, or 21–28 residues).^163–164

Figure 6.

(A) Antiparallel and Parallel helix dimers. (B) Helical wheel representation of a parallel helix dimer. Positions a and d comprise the hydrophobic core and are represented by yellow spheres. The polar positions of the helical wheel are represented by blue spheres.

Two major goals of coiled coil design are control of oligomerization state and relative helix orientation. Studies into hydrophobic-core variations found in the leucine-zipper region of the yeast transcriptional activator, GCN4-p1 led to sequence-to-structure heuristics for controlling oligomeric states.^165–170 For instance, it was found that leucine and isoleucine residues at positions a and d in the heptad can determine the oligomerization state of the parallel coiled coil being formed. When a=Ile and d=Leu, a parallel dimer will form, a=d=Ile leads to trimer formation, and a=Leu and d=Ile will direct for tetramer formation. Woolfson and coworkers have developed elegant strategies for self-assembling helical bundles based on the patterning of hydrophobic and polar residues, as well as the identity of the hydrophobic residues. Gellman and Woolfson provided further insight into the packing of side chain residues in coiled-coil designs.¹⁷¹ The Gellman group has utilized their backbone thioester exchange (BTE) method to determine preferred packing interactions between buried side chains at the a positions of the heptad repeat for both antiparallel^172–173 and parallel¹⁷⁴ coiled-coils.

Based on these experimental studies, several computational tools were created to model^175–178 and design coiled-coil assemblies.^179–184 CCCP and CCBuilder are both web-based tools which have increased the accessibility of modeling and designing coiled-coils.^{181, 185} X-ray crystal structures and other biophysical analyses have confirmed computationally designed oligomeric states from dimers to heptamers. A range of pentamer designs, including natural and engineered peptides have been made,^186–191 as well as a variant of GCN4-p1 which forms an unusually spiraled heptamer.¹⁹² For larger oligomers (hexamers and larger), a solvent accessible channel is formed in the center of the bundle and is capable of binding small molecules.^193–195 Different oligomers show varying levels of sensitivity to mutation i.e., some designed hexamers formed both parallel and antiparallel tetramers, as well as open and collapsed hexamers, when mutations were incorporated.¹⁹⁶

Judicious incorporation of covalent bridges that mimic noncovalent contacts has also been employed to drive coiled-coil assembly. Work into covalently constrained helical assemblies includes template-assembled synthetic proteins (TASP) and covalently linked cores as miniproteins (CovCore). TASPs link peptide termini to a macrocyle scaffold,¹⁹⁷ reducing the entropic barrier for helical assembly.^198–199 CovCore miniproteins restrain computationally optimized helical bundles through a multivalent cross-linker.²⁰⁰

To investigate coiled coils in protein complexes as inspiration for PPI inhibitors, we performed an exhaustive search of the PDB.⁹ Two filters in our search – the first to identify pairs of helices with antiparallel or parallel geometry, and the second to ensure both helices participate in binding – allowed the identification of over 1000 helix pairs in protein complexes. A surprising finding from this dataset was that the first and last interaction hotspots along each helix are typically within 2 heptads, far closer than the 3–4 heptads typically required for stable coiled-coil formation.^163–164 This finding led us to envision the application of shorter helix dimers for PPI modulation.

Our initial goal for helix dimer design was to identify a crosslinker that would drive helical peptide folding and association, particularly through reinforcement of canonical knobs-into-holes interactions.¹⁷ The most successful design involved a salt bridge surrogate (SBS) that replaces a stabilizing inter-helix ionic interaction with a covalent linkage. Initial designs of antiparallel cross-linked helix dimer (CHD) used optimized knobs-into-holes packing at a and d positions¹⁷¹ and replaced a salt bridge interaction at positions e and e’ with a bis-triazole linkage formed by CuAAC between propargyl ether and an azide-bearing side chain from each peptide. The helix dimer structure was confirmed through CD and NMR studies. As proof-of-principle in using these short antiparallel helix dimers for molecular recognition, mimics of Nervy homology two (NHR2) domain of the AML1-ETO-containing transcription factor complex were designed and tested for interaction with the E-protein NHR2-binding (N2B) motif.¹⁷ Optimization of the NHR2 hydrophobic core and further reinforcement of the helix dimer with a disulfide bond yielded a peptide with 5-fold higher binding affinity relative to the native protein.

Expanding on this work, we have recently described crosslinking strategies to form parallel helix dimers.¹⁴ Covalent linkage of parallel helices is accomplished by sequential reaction of a linker with two benzyl bromide equivalents (joined by an ether linkage) with a unique cysteine on each peptide (Figure 7). As with the antiparallel helix dimer, this linker was designed to mimic an interhelical salt bridge with a similar number of atoms as the bistriazole linker The cysteine residues are placed at positions e/g’ to mimic this ionic interaction. As with the antiparallel designs, CD and NMR analysis confirmed a stable helical structure, including a buried hydrophobic core.

Figure 7.

(a) Helical wheel diagram of parallel dimer CHD-4 with the bis-triazole linkage. (b) CD spectra of parallel and antiparallel CHD dimers in 50 mM aqueous KF, pH 7.4. (c, d) Overlay of the 20 lowest conformations derived from parallel dimer NMR analysis and the lowest energy structure.¹⁴

Combining the antiparallel and parallel approaches to assemble short helical dimers, we also demonstrated the formation of stable helical trimers. In this case, a bistriazole linkage is used instead of an ether linkage to join the parallel helices.¹⁴ These studies enabled analysis of the hydrophobic core stability of helix trimers, which showed similar preferences for hydrophobic stacking as observed in the PDB for longer coiled coil trimers.

Non-regular Peptides

While roughly half of all amino acid residues at protein-protein interfaces are part of regular α-helical or β-strand structures, the remaining half are found in non-regular structures (sometimes called loops).²⁰¹ “Non-regular structure” is a catch-all term that generally describes a diversity of structures that lack local repetition of backbone dihedral angles. Importantly, antibodies, one of the hallmark classes of proteins for specific molecular recognition, typically interact with their partners through such non-regular loops, as do many other proteins.²⁰²

Recent surveys of the Protein Data Bank by Kritzer et al. revealed millions of loops of varying length at PPI interfaces.²⁰³ Many of these loops are predicted to play an important role in their respective interfaces based on the percent of the predicted binding free energy coming from interactions with loop residues. Loops which contribute significantly to protein-protein interactions have been termed “hot loops,” a name adapted from “hotspot” residues. These hot loops are found in PPIs with diverse functions, including transcription, oxidative stress response, the cell cycle, and proteolysis.

In addition to protein-derived loops, peptide natural products often have non-regular structures and have further inspired development of novel crosslinking strategies to explore peptide structural space through cyclization and bicyclization.^204–205 Beyond attempts to imitate natural products with synthetically simpler crosslinkers, total syntheses of peptide natural products with unusual covalent crosslinks also opens opportunities to explore these crosslinks as templates for stabilizing specific peptide loop conformations.^206–207

Here we will describe various crosslinking chemistries that have been used to generate loop structures that lack extended stretches of α-helical or β-sheet structure (Figure 8). Because the catch-all definition of a loop precludes categorization by structure, we will discuss peptides based on the chemistries used for peptide cyclization. Recent reviews by Fairlie, Hutton, and coworkers highlight many natural and synthetic peptides that bind to proteins, including several that competitively inhibit PPIs.^208–209 We will also mention novel chemical approaches that have been applied with the intent of stabilizing non-regular structures, even in the absence of structures.

Figure 8. Stabilizing Non-Regular Conformations for Molecular Recognition.

Peptide macrocyclization via lactam, thioether, and disulfide bonds stabilize non-regular peptide conformations in many natural product and synthetic peptides. Lactam bridges can be derived from the N- and C-termini, the N- or C-terminus and one side chain, or two side chains (e.g., Lys and Glu). Thioether bridges are formed from Cys residues, including linkage to an N-terminal acetyl group or electrophilic crosslinkers. Disulfide bonds are observed strictly between Cys residues. Other crosslinking chemistries include triazole, lactone, alkyne, biaryl, and unusual side chain crosslinks discovered in natural products.

Cyclization Chemistry

Lactam

Lactam macrocyclization is the most prevalent cyclization observed for peptide loops of known structure, including several well-known natural products like cyclosporin A and gramicidin S.^210–211 Head-to-tail cyclization, side chain-to-tail cyclization, and side chain-to-side chain (e.g., Lys-Glu) cyclization are all observed in forming the lactam. In light of the diversity of lactam placement, macrocycle size, and limited number of structures, it is largely impractical to suggest a link between macrocyclization approach and specific structure or structural constraints with one exception.

In the case of the natural product argadin, which binds the enzyme chitinase (PDB 1WAW),²¹² and a synthetic peptide containing the Arg-Gly-Asp (RGD) motif known to bind with high affinity to integrin (PDB 1L5G),²¹³ we observe a high overall structural similarity (RMSD = 1.08 Å for all backbone atoms). Both of these head-to-tail macrocycles are composed of 5 amino acid residues with overlapping γ- and β-turns. A key difference between the peptides is that argadin has a type II β-turn while the RGD mimic has a type II’ β-turn. The difference in β-turn geometry appears to be driven by a difference in placement of an amino acid that can adopt or prefers positive φ angles (relative to the γ-turn), namely D-proline in argadin at β-turn position i+2 and glycine in the RGD mimic at β-turn position i+1. It is thus tempting to suggest that attempts to mimic such combined γ,β-turn motifs can be guided in a straightforward manner by these peptides, with cyclization and strategic placement of glycine or D-amino acids favoring positive φ angles to generate the desired β-turn.

The protein targets of lactam-cyclized peptides are also diverse, including proteases, RNA polymerase, molecular chaperones, and many PPIs.

Disulfide

Aside from head-to-tail cyclization, the most commonly observed cyclization strategy in cyclic peptide natural products is disulfide formation. Many disulfide-cyclized peptide natural products adopt β-hairpin conformations, such that the cysteine residues occupy non-hydrogen bonding positions. Shorter peptides that cannot form a β-sheet instead fold into non-regular loops. Though only a limited number of protein-bound structures exist, two 6-amino acid macrocycles (oxytocin and a streptavidin-binding HPQ macrocycle) adopt remarkably similar conformation despite completely different sequences and disulfide dihedral angles (PDB 1NPO and 1SLD).^214–215 The conformational similarity is reflected in the backbone dihedral angles, which only differ at the terminal Cys residues. Thus, disulfide templating of short macrocyclic loops may be a viable strategy to target extracellular proteins with similar structures.

In bicyclic peptides, it was further demonstrated that cysteine-penicillamine disulfide bonds are formed selectively over cysteine-cysteine or penicillamine-penicillamine disulfides under oxidizing conditions.²¹⁶ Though no structures have been reported yet, the approach can theoretically provide access to diverse topologies, including some of the structures observed in cysteine-rich peptides.

Thioether

Thioether linkages are also commonly used to cyclize non-regular peptide structures, including the large class of lantibiotics.²¹⁷ The thioether bond is formed predominantly by S_N2 displacement of halides: from a N-terminal chloroacetyl residue using 1 cysteine,²¹⁸ from α,α’-dibromoxylene isomers using 2 cysteines,²¹⁹ or from α,α’,α”-tribromomesitylene using 3 cysteines.²²⁰ The last of these approaches yields a bridged 2-ring system in which each ring can be thought of as a cyclic peptide formed from reacting α,α’-dibromoxylene with 2 cysteines.

Unfortunately, the structural data available is insufficient to establish a clear relationship between cyclization strategy and peptide structure. Dihedral angles within the thioether linkage vary significantly, even between peptides with the same type of linkage (e.g., formed between N-terminal chloroacetic acid and cysteine). In fact, a peptide with two terminal cysteines that is cyclized using ortho-α,α’-dibromoxylene shows multiple conformations in a single crystal structure with lectin B (PDB 5NF0).²²¹

An interesting development in using thioether bonds for peptide cyclization is the use of two pairs of cysteines to form distinct crosslinks within the same peptide.²⁰⁵ Though structural data has not yet been reported, these highly constrained peptides show resilience towards proteases and nanomolar affinities for their targets, highlighting the use of structural constraints to stabilize non-regular conformations for protein interaction.

Other Cyclization Chemistries

Aside from lactam, disulfide, and thioether cyclization, various other linkages have been used for peptide cyclization, including ether, triazole, biaryl, olefin, and lactone bonds.^222–225 Some of these cyclic peptides have been structurally characterized but the dataset is generally too sparse to draw significant conclusions relating crosslinking strategy to specific structure(s).²⁰⁸

In addition, several natural products feature unusual crosslinks between amino acid side chains, such as the Leu-Trp-His linkage in celogentin C²²⁶ or the oxidized Trp-Cys linkage in α-amanitin.²⁰⁷ Recent total syntheses of these products, including solid-phase approaches, may enable peptide library synthesis and structural analysis to suggest novel methods for covalently driving peptides to adopt particular structures.

Conclusion

We have described a chemical toolbox to stabilize desired secondary and tertiary peptide structures including α-helices, β-sheets, helical oligomers, and non-regular loops for applications in biomolecular recognition. These molecules, in which functional groups are pre-organized to mimic PPI hotspots, efficiently target a variety of PPIs mediated by secondary and tertiary structures. The synthetic methods described offer routes to pre-pay the entropic penalty of folding upon binding by stabilizing the binding conformation. In many cases, folding also enhances proteolytic stability, making these peptide more attractive prospects for drug design. Table 1 lists examples of protein-protein complexes that have been targeted by designed mimics. Continued development of synthetic methods for other prevalent PPI structures and application to the hundreds of thousands of PPIs will allow efficient targeting of PPIs that were once believed to be “undruggable.” These studies will also facilitate chemical mapping of the protein-protein interactome.

Table 1.

A representative list of protein-protein interactions that have been targeted by secondary and tertiary structure mimics.

Complex	Target		Citation
α-Helix Targets
MDM2/p53 (PDB 1YCR)	MDM2/MDMX		31, 47, 49, 63, 75–77, 109, 227
Ras/Sos (PDB 1NVW)	Ras		111
HIF/p300 (PDB 1L8C)	p300		112–113
HIV gp41 assembly (PDB 1AIK)	gp41		37, 56, 78–79, 88, 114
Rab-interacting protein/Rab8a GTPase (PDB 3CWZ)	Rab8a		59, 64
BH3 domains/Bcl-2 (Bcl-xL) (PDB 5FMK)	Bcl-2 (Bcl-xL)		43, 46, 62, 87, 110
EZH2/EED (PDB 2QXV)	EED		67
WASF3/CYFIP1 (PDB 3P8C)	CYFIP1		69
EXoS/14-3-3 (PDB 4N7G)	14-3-3		70
NRCA/ER (PDB 2QGT)	Estrogen Receptor Dimerization		68
β-Sheet Targets
Tat/HIV-1 TAR RNA (PDB 2KDQ)	HIV-1 TAR RNA		228
nNOS/syntrophin (PDB 1QAV)	syntrophin		229
AMA1/RON2 (PDB 3ZWZ)	AMA1		229
CXCR4/CXCL12 (PDB 3OE0)	CXCR4		230
EPO/EpoR (PDB 1EBP)	EpoR		231
Plexin B1/semaphorin (PDB 5B4W)	Plexin B1		232
CdiA-CT/CdiI (PDB 4ZQW)	CdiI		233
MDM2/p53 (PDB 1YCR)	MDM2/MDMX (β-hairpin mimics an α-helix)		234
Coiled-Coil Targets
NHR2/N2B (PDB 4JOL)	N2B		17
Non-Regular Loop Targets
Grb2 SH2/phosphopeptide (PDB 1TZE)	Grb2 SH2 domain		235
Grb7 SH2/ErbB2 phosphopeptide (PDB 1MW4)	Grb7 SH2 domain		236
Hemagglutinin/antibody (PDB 3ZTN)	Hemagglutinin		237
Apelin/apelin receptor (PDB 5VBL)	Apelin receptor		238
CK2α/CK2β (PDB 1JWH)	CK2α/CK2β		239
CDK2/cyclin A/p27(Kip1) (PDB 1JSU)	Cyclin A		240
C5a/C5aR (PDB 6C1R)	C5aR		241
ephrinB2/EphA4 (PDB 2WO2)	EphA4		242
Integrin αVβ3/ECM ligand (PDB 1L5G)	Integrin αVβ3		213
Ras/Sos (PDB 1NVW)	Ras (allosteric)		243
MLL-menin (PDB 3U85)	Menin		236
PLK1/PBIP1 (PDB 1UMW)	Plk1		244
PD-1/PD-L1 (PDB 5O45)	PD-L1		245
UHM domain/ULM peptide (PDB 5LSO)	UHM domain		246
TLE/peptide (PDB 5MWJ)	TLE		247
WDR5/MLL (PDB 5VFC)	WDR5		248
TNFα/TNFR (PDB 4TWT)	TNFα		249