Tackling Undruggable Targets with Designer Peptidomimetics and Synthetic Biologics

Abstract

The development of potent, specific, and pharmacologically viable chemical probes and therapeutics is a central focus of chemical biology and therapeutic development. However, a significant portion of predicted disease-causal proteins have proven resistant to targeting by traditional small molecule and biologic modalities. Many of these so-called “undruggable” targets feature extended, dynamic protein–protein and protein–nucleic acid interfaces that are central to their roles in normal and diseased signaling pathways. Here, we discuss the development of synthetically stabilized peptide and protein mimetics as an everexpanding and powerful region of chemical space to tackle undruggable targets. These molecules aim to combine the synthetic tunability and pharmacologic properties typically associated with small molecules with the binding footprints, affinities and specificities of biologics. In this review, we discuss the historical and emerging platforms and approaches to design, screen, select and optimize synthetic “designer” peptidomimetics and synthetic biologics. We examine the inspiration and design of different classes of designer peptidomimetics: (i) macrocyclic peptides, (ii) side chain stabilized peptides, (iii) non-natural peptidomimetics, and (iv) synthetic proteomimetics, and notable examples of their application to challenging biomolecules. Finally, we summarize key learnings and remaining challenges for these molecules to become useful chemical probes and therapeutics for historically undruggable targets.

Graphical Abstract

1. INTRODUCTION: DESIGNING AND DEPLOYING DESIGNER PEPTIDES, PEPTIDOMIMETICS, AND PROTEOMIMETICS FOR “UNDRUGGABLE” TARGETS

Over the past few decades, rapid development in the fields of molecular biology and chemistry have ushered in an era of tremendous progress for drug discovery. Modern techniques for protein and nucleic acid purification along with data-drivenomics approaches have identified an ever-expanding list of disease-causal targets—typically proteins—that may serve as starting points for the development of chemical probes and therapeutics. Specific protein classes have proven to be intrinsically druggable by current modalities, including specific receptor families like ion channels and GPCRs, as well as intracellular enzymes and nuclear hormone receptor transcription factors (NHR-TFs).^1,2 These protein targets are typically characterized by the presence of defined ligand binding sites and other hydrophobic pockets amenable to small molecule discovery and medicinal chemistry optimization.³ Other disease-associated protein targets are present on the surface of cells or are secreted into the extracellular space, thus enabling targeting by large, hydrophilic biologic modalities, including monoclonal antibodies (mAbs), recombinant proteins, and engineered peptides (Figure 1A). A significant number of disease-associated targets fall between these two druggable landscapes, with neither obvious ligand binding sites nor extracellular exposure lending themselves toward cell permeable small molecules or traditional biologics, respectively. This landscape of targets therefore has been referred to by many as being “undruggable” (Figure 1A).^4,5 Additional characteristic features shared by many undruggable targets generally include the lack of enzymatic activity and reliance upon large, dynamic surfaces and disordered structures that mediate protein–protein interactions (PPIs) and/or protein–nucleic acid interactions integral to their function (Figure 1A). While all protein families contain disease-associated targets that could be considered undruggable, several families have emerged as particularly challenging, including small GTPases, phosphatases, transcription factors (TFs), epigenetic enzymes, and chromatin regulatory factors.³

Figure 1.

Designer peptides and peptidomimetics for undruggable targets. A) The vast majority of putative disease drivers lack structural features conventionally targeted by small molecules or antibodies. B) Molecular dynamics simulations (50 ns) of unmodified and Diels–Alder cyclized loop designer peptide, demonstrating significant stabilization of structural topology. C) Classes and respective characteristics of designer peptides and proteomimetics, discussed in this review.

Molecules that blend some of the physicochemical properties of synthetic small molecules with the higher energy and specificity of protein therapeutics could serve as attractive starting points for undruggable targets. Peptides naturally embody this molecular middle ground, owing to their larger binding footprint and propensity for specific interactions at PPI and protein–nucleic acid interfaces in myriad natural biological processes. Indeed, dozens of natural and engineered peptides are currently in the clinic for diverse disease indications.^6,7 Relative to biologics, natural peptides are significantly smaller in size, which can confer increased stability, synthetic tractability, reduced manufacturing costs, and higher likelihood of developing peptide analogs with reasonable cell and tissue penetration for intracellular target engagement.⁸ However, as a class natural peptide sequences composed of linear α-amino acids generally exhibit poor cell membrane permeability, high in vivo clearance rates from blood, and susceptibility to proteolytic and other metabolic degradation outside and inside cells (Figure 1B).^9-14 Therefore, considerable research effort has been focused upon chemical strategies to improve the physicochemical properties of peptides to overcome these limitations. A hallmark approach is to use synthetic modifications to stabilize local and global structural motifs, which can theoretically provide enhanced affinity and specificity for intended molecular targets, as well as vastly improved pharmacological properties (Figure 1B). These chemically stabilized molecules, generally termed peptidomimetics,¹⁵ have a rich history in drug discovery as ligands for various receptors or enzymes as described in several reviews.^16-21 In this review, we consider these synthetically modified compounds “designer” peptides or peptidomimetics, as well as larger mimetics of protein domains, which encompass proteomimetics or “synthetic biologics.” In this review, we will specifically focus on the development for and application of these molecules to tackle undruggable targets (Figure 1C).

Designer peptides and peptidomimetics have previously been categorized into four classes.²² Class A represent minimally modified peptides wherein small changes to the backbone or side chains result in structurally stable or more active analogues. Class B are significantly modified with unnatural amino acids forming substantial backbone alterations such as foldamers. Class C are scaffolds that replace the peptide backbone entirely with an unnatural nonpeptidic scaffold but retain side chain functionality. Class D are mainly small molecule-like scaffolds that may mimic the mechanism of peptides but do not have a direct link to structure or design and therefore will not be discussed in this review and have been reviewed elsewhere.^22-24

In this review, we will focus on designer peptides and peptidomimetics in four general classes, including: (i) natural product-like macrocyclic peptides (MCPs); (ii) side chain stabilized peptides; (iii) peptidomimetics containing nonnatural backbones; and (iv) synthetic proteomimetics and biologics (Figure 1C). Sections (i) and (ii) mainly cover class A mimetics, whereas section (iii) will discuss class B and class C mimetics. Section (iv) focuses on larger mimetics of protein tertiary and quaternary structures for expanded recognition properties suitable for challenging protein and nucleic acid targets. We will focus on the defining structural features that characterize each class, the breadth of synthetic strategies previously employed to construct each class as well as notable examples of their applications in modulating undruggable disease targets. While there are numerous small molecule and biologic-inspired modalities developed for undruggable targets that may share properties with the classes above, such as class D mimetics and engineered recombinant proteins, we find these to be outside the scope of the current review and direct readers to other excellent recent reviews.^3,4,22-27 We will conclude our review with an outlook on the enormous potential of designer peptides, peptidomimetics, and synthetic biologics in targeting the “undruggable” as well as unmet requirements for advancing these promising modalities to the clinic.

2. MACROCYCLIC PEPTIDES

Peptide therapeutics are an attractive modality due to their synthetic accessibility and broad range of targetable biomolecules and interactions. However, naturally derived linear peptides suffer from pharmacological liabilities such as lack of cell entry and rapid degradation by cellular proteases, as well as reduced target affinity and specificity.^12-14 Structural preorganization through cyclization—either “head-to-tail” or via interchain cross-links—has been shown to significantly enhance several or all of these pharmacological pitfalls. Indeed, nature has devised numerous routes to biologically active macrocyclic peptides, including head-to-tail cyclized molecules like cyclosporin A, amantins, and somatostatin or depsipeptides such as romidepsin, FK228, and leualacin, and hybrids of similar structure.^28-30 Typically, these molecules are synthesized by nonribosomal peptide synthetase (NRPS) clusters and further modified with noncanonical peptide bonds, amino acid side chains, and other modifications.³¹ Researchers can take inspiration from the structure–function properties of these natural products to initiate campaigns to produce synthetic macrocyclic peptides (MCPs) with desirable properties.³² Due to their larger surface areas, arrayed amide backbone and chemically diverse side chains, synthetic MCPs could be developed to target traditional ligand binding pockets, as well as shallower allosteric pockets and grooves on the surface of or at the interface between proteins; these target features may be more prevalent in undruggable targets. Therefore, MCPs occupy a Goldilocks zone of physicochemical space to yield FDA-approved therapeutic molecules.^33-36 In this review, we focus on synthetic MCPs containing a natural peptide backbone with the formation of at least one covalent cyclic linkage to induce conformational constraints. While natural product-derived and synthetic MCPs such as cyclotide scaffolds have been developed to target classically druggable targets such as enzymes and receptors,^29,37-39 here we will focus primarily on their development for undruggable targets and surfaces.

2.1. Design Principles for MCPs

MCPs are generally constructed with ~5–20 amino acids to span a molecular weight range of ~0.5–3 kDa, making these one of the smallest classes of designer peptides (Figure 1C). This size confers small molecule character that is beneficial for biological applications. Conformational constraint and preorganization is often accomplished via head-to-tail cyclization with or without additional side chain linkages, which collectively introduce significant rigidity into MCP backbone conformations. In addition to protecting structures from metabolic degradation, the preorganization of active binding poses can reduce the entropic cost of folding during target binding. Common cyclization chemistries typically must proceed with high yield, chemoselectivity and operate under aqueous reaction environments. These include disulfide bridges, thioether linkages, amide bonds, unsaturated hydrocarbons, and cycloaddition adducts, for example through [3+2] azide–alkyne Huisgen “click” chemistry or [4+2] Diels–Alder cycloadditions (Figure 2).^40-42 These common approaches each serve to provide unique properties that have advantages and disadvantages dependent upon application and design.

Figure 2.

Common macrocyclization adducts (right) generated from common synthons (left).

In general, the design, selection, and optimization of MCPs for undruggable targets should aim for several attributes including: (i) strong binding affinity and specificity for the protein of interest (POI); (ii) the ability to penetrate cellular membranes and access intracellular targets; (iii) high metabolic stability, and (iv) high bioavailability.⁶ Owing to their larger size and chemical properties, many FDA-approved MCPs^33-36 do not adhere to Lipinski’s rule-of-5 or similar paradigms.^43-46 Instead, many bioactive MCPs are thought to exhibit cell permeability and high bioavailability through their potential for “chameleon-like” structural character. In this theory, MCPs can adopt a conformation with a high polar surface area (PSA) and more exposed backbone NH bonds to retain solubility in aqueous conditions but adopt an alternative conformation minimizing PSA and exposed NH bonds when diffusing into hydrophobic membranes.^47-50 Baker and colleagues explored this principle in examining MCPs from 6 to 12 amino acids in length, and found that internal hydrogen bonds and minimal exposed polar groups were critical to achieve cell entry.⁵¹ They designed cis/trans MCPs that isomerized around one peptide bond to achieve the chameleonic property, where the cis state is solvent exposed and the trans state leads to the formation of internal backbone hydrogen bonds. These findings agree with previous studies showing that backbone hydrogen bond matching as well as amide N-methylation and masking polarity contribute significantly to MCP cell uptake and oral bioavailability.⁵² Monovich and co-workers performed an extensive study assessing the properties of 6-, 7-, and 8-mer MCPs to determine that two factors are highly predictive of oral bioavailability and membrane permeability: (i) the lowest free energy of transition, ΔG*_transfer, between aqueous and membrane states and (ii) the lowest number of solvent exposed backbone NH groups in an MCP.⁵³ This study supported many others suggesting that the backbone plays a more critical role in permeability as opposed to the side chain, hence cell permeable MCP backbones with varied side chains could be used as a framework for large array library formats for discovering ligands to therapeutically relevant targets. For further discussion on MCP chameleonic design principles, refer to the excellent review by Monovich and co-workers.⁵⁴

Although the smaller size and constrained structure of MCPs generally improves cell uptake, this does not directly translate for every design. Therefore, a simple and accessible method that many designer peptides have adopted is appending a cellpenetrating peptide (CPP). CPPs are short peptides that can cross the cell membrane, which facilitates the transport of various attached cargo.^55-58 Typical CPP examples such as Tat,⁵⁹ penetratin,⁶⁰ or octa-arginine⁶¹ are popular choices for conjugating to MCPs, though amphipathic and anionic CPPs have also been reported.^55,62 Physicochemical properties such as formal charge and hydrophobicity correlate with cell penetration, and CPPs are internalized through clathrin- and caveolin-independent, energy-dependent endocytosis pathways involving sulfated cell surface proteoglycans.⁵⁶ Given that linear peptides suffer from protease digestion, cyclic CPPs have been developed that are more metabolically stable and membrane permeable than their linear counterparts and are nicely described in a recent review by Parangi, Tiwari, and co-workers.⁶³ While appending CPPs to MCPs is useful for improving bioactivity, there have been no therapeutics containing a CPP approved by the FDA to date, likely due to potential for cellular toxicity, adverse immune response, and poor efficacy arising from the lack of specificity due to the CPP moiety.⁶⁴

To create MCPs with desirable pharmacological properties without the need for external modifications such as CPPs, a team from Chugai Pharmaceuticals analyzed several properties inherent to MCPs that would yield the most pharmacologically viable molecules.⁶⁵ Using naturally cell permeable cyclosporine A as a model, they designed and tested 553 analogs to understand the critical properties that drive potential therapeutic efficacy. They determined desirable properties for metabolic stability and cell uptake include high hydrophobicity, a high permeability coefficient, a medium-sized ring between 8 to 11 amino acids, extensive backbone amide N-alkylation of 6 or more residues, and reduced side chain polarity (Figure 3). They also found that, for medium-sized rings, oxidative metabolism is dominant over enzymatic hydrolysis, and MCPs of 8 residues, while having higher average membrane permeability than MCPs of 11 residues, have higher oxidative metabolism. Smaller rings more susceptible to oxidative metabolism are unlikely to overcome this limitation without decreasing lipophilicity necessary for membrane permeability. However, ring size and residue identity also play important roles in target recognition that must be balanced with desirable metabolic stability.¹³ While these studies greatly improved our knowledge of MCP design, general strategies for designing therapeutically viable MCPs remain elusive owing to the greater complexity afforded by larger and more diverse molecules and interactions. However, interest in MCPs as a drug modality is increasing exponentially with many biotechnology firms exploring MCP therapeutic discovery platforms,^13,44 and an increased interest from large biopharma groups in exploring this modality for a range of targets.

Figure 3.

Summary of common structural features and modifications of drug-like MCPs. Representative features are highlighted on the structure of cyclosporin A, alongside a representative summary of structure–function correlations from a study⁶⁵ that systematically explored cyclosporin A-derived macrocycle libraries (right).

2.2. Discovery, Design, and Synthesis of MCPs for Challenging Targets

MCPs are versatile and can be expressed by natural machinery or produced synthetically utilizing the diverse chemistries. Synthetic optimization typically follows derivatization from existing scaffolds or de novo selection from large combinatorial libraries in high-throughput screening. In the following section, we will discuss how MCPs are developed through rational design or selection from large libraries in multiwell screening, one-bead-one-compound, DNA-programmable libraries, phage display, mRNA display, and in-cell formats. We will also discuss the common synthetic methods to form the MCP libraries and provide key examples that have focused on generating chemical probes and therapeutics for undruggable targets.

2.2.1. Rational Design.

The majority of MCP rational design campaigns begin with the identification of specific binding epitopes within naturally occurring proteins and peptides that are potentially amenable to ligand development and optimization (Figure 4A). Early structure-blind approaches involved synthesis of cyclic analogs of complementarity determining regions (CDRs) of antibodies or endogenous peptides such as enkephalin to target various POIs.^66-68 As more cocrystal structures of apo- or bound target structures have become publicly available, rational or structure-based drug design (SBDD) approaches have become more common. This methodology has been used to find MCPs that interfere with interactions such as TNKS-AXIN in WNT signaling,⁶⁹ RGS4,⁷⁰ and TGF-β-SMAD7.⁷¹ A notable recent example for a rationally designed MCP targeting an undruggable interaction is between the transcription factor TEAD1 and the coactivator YAP1, which represent a TF-co-activator complex that is central in the Hippo signaling pathway.⁷² The TEAD1/YAP1 interaction is highly implicated in many cancer types, and abolishing this interaction leads to the suppression of cell proliferation.⁷³ Hu and co-workers examined the structure of the TEAD1/YAP1 interaction and identified a looped-coil segment of YAP1 that interacts with TEAD1.⁷⁴ Using this structure as a guide, they generated a series of MCPs via disulfide bridging, and discovered optimized molecules that could bind to TEAD1 with greater potency than wild-type YAP1. Subsequent work has further focused on this interaction to identify small molecules that can disrupt this binding interaction,⁷⁵ highlighting an important byproduct of identifying peptide leads to ultimately develop more drug-like small molecules. In a separate study, an MCP derived from the BB loop in MyD88 was developed to interrupt protein dimerization.⁷⁶ MyD88 is an adaptor protein critical in the TLR and IL-1R signaling pathways that functions via dimerization and overexpression in cancers leads to overactivation of NF-κB and cancer metastasis.⁷⁷ Nussbaum and co-workers identified a sequence from the BB loop that controls dimerization and synthesized an MCP via an amide linkage to form c(MyD 4-4).⁷⁶ c(MyD 4-4) blocked the TLR/IL-1 signaling pathway both in vitro and in mice models. In a subsequent study, they sought to increase the bioactivity of c(MyD 4-4) and enable oral administration by myristoylating the MCP at the N-terminus.⁷⁸ Myristolyation led to a significantly improved cellular bioactivity and in vivo oral bioavailability while still retaining activity in mice models. In addition to these PPIs, there are many MCPs that have been rationally designed against amyloid peptides, which have recently been covered in an excellent review by Khairnar et al.⁷⁹

Figure 4.

Rational design of MCPs. A) MCP starting points can be directly extracted from key interacting loops discovered in natural protein–protein interactions. B) Structure of stonin2 (orange/green) bound to Eps15 (cyan), identified “hot loop” by LoopFinder highlighted in green. PDB: 2JXC. C) Chemical structure of Eps15 inhibitor ST1-oxy identified by LoopFinder. O-xylene macrocyclization linker is highlighted in green.⁸¹

One way that potentially druggable PPI interacting loops, also called “hot loops”, can be identified is through the program LoopFinder, designed by Kritzer and co-workers.⁸⁰ Methodologically, LoopFinder identifies “hot loops” if the interacting loop fit one of three parameters: (i) three or more hot-spot residues (defined as ΔΔG_residue ≥ 1 kcal mol⁻¹), (ii) two consecutive hot-spot residues, or (iii) the average ΔΔG_residue over the entire loop is greater than 1 kcal mol⁻¹. Most loops identified in this approach were secondary structures such as β-turns and β-hairpins, though many had noncanonical structures. In the updated version, they improved on the energy calculations and benchmarked it against known PPIs to refine the selection criteria for an identified “hot loop” to (i) an average ΔΔG_residue ≥ 0.6 Rosetta Energy Units (REU, roughly equal to 1 kcal mol⁻¹), (ii) three hot spot residues (ΔΔG_residue ≥ 0.6 REU), and (iii) a greater than 50% of binding energy contributed by the loop, represented as the sum of ΔΔG_residue.⁸¹ Loops that fulfilled all three criteria were seen as the best starting points for MCP development for a given target of interest. A close examination of the 210 “hot loops” that fulfilled all three criteria suggests that MCPs can only mimic some hot loops due to conformational constraints.⁸² It also suggests that, while larger MCPs can accommodate the native conformation of the hot loop, this must be balanced against the entropy loss of the larger MCP upon binding. One of the loops that fulfilled all three criteria is found in the interaction between Eps15 and stonin2, for which there are no known high affinity inhibitors (Figure 4B).⁸¹ Eps15 is a clathrin-associated protein and part of the endocytosis machinery⁸³ and its interaction with stonin2 leads to the recruitment of AP-2 and endocytosis, which is implicated in Ebola infectivity and potentially other diseases.⁸⁴ After identifying the loop on stonin2, cysteine or penicillamine residues were appended to both termini of the peptide sequence, and a panel of dithiol, bis-alkylation linkers were tested to cyclize the peptide by forming stable thioether linkages. The lead compound, ST1-oxy, was flanked by two cysteine residues and cyclized using dibromo-o-xylene to generate a selective and high affinity (K_d = 0.33 μM) ligand for Eps15 (Figure 4C). A 2D-NMR structure in water confirmed ST1-oxy recapitulated the structure of the stonin2 hot loop, displaying the powerful tools for computationally defining native protein structures for the discovery of designer peptide ligands.⁸¹

Expanding the discovery of cyclization-amenable loops at protein interfaces can identify starting points for chemical probe development. Building upon the LoopFinder work by the Kritzer group, Burgess and co-workers recently developed a virtual library approach termed backbone matching (BM) that is able to sample the conformations of cyclo-organopeptides composed of 4–10 residues cyclized with various organic elements through amide bonds.⁸⁵ Compounds capable of adopting the proper low energy conformation in solution corresponding to the target backbone structure are prioritized for further evaluation and development. Using this approach, they identified and validated MCP inhibitors for interactions between iNOS/SPSB2 and uPA/uPAR. They also discovered candidates for 1245 of the 1398 hot loops screened, suggesting a plethora of other target interfaces that may be interesting to pursue. The continued alignment of computational methods with experimentally determined protein structures for discovery and validation of interacting loop structures has been and will continue to be a powerful approach to rationally design MCP probes and therapeutics.

MCPs derived from natural protein surfaces can also be designed based on more active linear epitopes, which can be cyclized then reversed to develop pro-drugs. For example, on disulfide linked MCPs, a reversible cyclization strategy could leverage intracellular glutathione to reduce the disulfide bridge and linearize the peptide upon cell uptake (Figure 5A). The cyclic peptide gains stability and cell permeability afforded by rigidification but retains the linear binding epitope once inside of cells. Using this strategy, Dougherty et al. designed an MCP derived from a section in the C-terminus of CFTR to inhibit the intracellular CAL-CFTR interaction to prevent lysosomal degradation of the CFTR ion channel, a potential treatment strategy for cystic fibrosis.⁸⁶ Although symptomatic treatments using small molecules for dysregulated ion channels exist, direct modulation of intracellular CFTR mutant interactions remains challenging. In this design, a disulfide linked MCP, PGD97, containing a cell penetrating handle was generated that exhibited high affinity and specificity for the PDZ domain of CAL and an in vitro serum half-life over 24 h (Figure 5B). PGD97 acted as a stabilizer of F508del-CFTR and improved the CFTR function in cells. This strategy is also amenable to bicyclic peptides (BCPs). Qian et al. designed a disulfide BCP targeting the undruggable interaction between NEMO and IKK-β, which are part of the IKK kinase complex that mediates NF-κB signaling and is implicated in cancer and autoimmune diseases.^87-91 Similar to the CAL-CFTR inhibitor, a cyclic recognition domain was connected to a cyclic CPP to ensure entry into the cell to generate a BCP. The recognition domain binds to NEMO in linear form but not in cyclic form, thus acting as a prodrug to only function in reducing environments.⁹² Although only achieving modest inhibition (IC₅₀ ~20 μM), this strategy increases metabolic stability and cell permeability. Other approaches to create prodrug MCPs to function in reducing or oxidative environments take advantage of the redox lability of disulfides.⁹³

Figure 5.

Pro-drug disulfide linked MCPs. A) Dithiol cyclization improves cell permeability and is reduced by glutathione to produce active linear peptide in cells. B) Structure of stable, cell permeable pro-drug CAL-CFTR inhibitor PGD97. The reversible disulfide bond is highlighted in green.⁸⁶

Recent advances in computational approaches such as docking analysis and molecular dynamics simulations have contributed to the design of MCPs for challenging POIs. Baker and co-workers used computational analysis to design an MCP that could specifically bind HDAC6 over HDAC2, which is a notoriously difficult challenge to distinguish HDAC family members with small molecules or peptides.⁹⁴ They started with an “anchor” moiety ((2S)-2-amino-7-sulfanylheptanoic acid; SHA), based on a pan-HDAC inhibitor that coordinated with the zinc ion in the HDAC active site to situate the model peptides to the HDAC surface (Figure 6A). Their first attempt used two peptides of known structure and a library of previously generated MCPs of 7–10 residues that were docked to the anchor.⁹⁵ Using this method, they found an MCP that was decently potent in vitro (sub-μM IC₅₀) but was not specific for either HDAC. They then added an additional tryptophan anchor because indole moieties are present in some known small molecule HDAC inhibitors before extending the peptide to form new MCPs. This MCP also displayed good affinity but little selectivity. From there, they sampled and optimized the torsional angle of the anchors. This led to an MCP designated des4.3.1 that was 88-fold selective for HDAC6 compared to HDAC2 (Figure 6B).⁹⁴ Satisfying both binding and structural criteria is essential for generating de novo highly selective and efficacious MCP inhibitors. This and related studies are notable because they interrogate not only on-target^95-99 but also propensity for off-target specificity, which is essential for the concurrent development of specific probes and therapeutic leads.

Figure 6.

Docking analysis and molecular simulations select for HDAC6-specific inhibitors. A) Crystal structure of des4.3.1 bound to HDAC6 anchored with SHA. PDB: 6WSJ. B) Chemical structure of HDAC6-specific inhibitor des4.3.1.⁹⁴

The examples above demonstrate the key advantage of rational design: the lower barrier of entry to discover an MCP derived from a native protein structure. Identifying a potential interactor does not require lengthy and sometimes complicated screening approaches. However, there are a few limitations to the rational design approach: (i) requires prior knowledge of the target structure or interaction surface to identify potential ligands, (ii) identified ligands may not be amenable to structural stabilization after isolation, as proper folding may be dictated by interactions within the entire protein structure or interface,¹⁰⁰ and (iii) identified ligands may not represent the optimal binders as not all combinatorial space is explored compared to screening approaches. Many of these shortcomings are being addressed with the integration of computational design and virtual screening, which has and will continue to explode in coming years with improvements to computational methods and technologies. For example, Des3PI^101,102 and cPEPmatch,¹⁰³ have been used to predict MCP binding to classically undruggable PPIs but have yet to be substantiated with experimental results. Improved structural prediction methods such as AlphaFold¹⁰⁴ and RoseTTAFold^105,106 provide further tools for predicting ligand interactions. For further reading on the quickly evolving computational approaches to design macrocyclic peptide interactions, we point readers to other reviews.^107-110 While rational design approaches serve as solid starting points for MCP discovery particularly for identifying ligand-protein interactions, many undruggable targets are characterized by large, flat surfaces and disordered structures, and therefore unlikely to possess obvious domains from which to derive a potential ligand. For these challenging targets, large compound library screening is a more powerful approach for the de novo discovery of MCPs as described in the following sections.

2.2.2. Multiwell MCP Synthesis and Screening.

Multiwell screening involves plating a library of known compounds into a microtiter plate for subsequent biochemical or cellular testing, or concurrent synthesis and testing in multiwell formats. These methods are now anchor approaches to library screening in academia and pharmaceutical companies.¹¹¹ Tranzyme Pharma demonstrated an early approach to multiwell screening of an MCP compound library, with the main scaffold containing a tripeptide with a tether moiety bridging the two termini via amide linkages (Figure 7).¹¹² By diversifying the tethers used and incorporating ncAAs, they generated a 10⁴ member library that was used to design MCPs to targets such as the motilin receptor and the ghrelin receptor and others that made it into clinical trials.^113-117

Figure 7.

Scheme for the combinatorial synthesis of a diversified tripeptide MCP library in multiwell format, representing an early example reported by Tranzyme Pharma.¹¹² Bts: benzothiazole-2-sulfonyl, Ddz: α,α-dimethyl-3,5-dimethoxybenzyloxycarbonyl; DIAD: diisopropyl azodicarboxylate, MP: microporous, PS: polystyrene.

Many alternative approaches for the synthesis and screening of diverse MCP libraries have been developed in recent years. For example, Heinis and co-workers have reported the generation of low molecular weight MCPs through the cyclization of tri- and tetrapeptides via thioether linkages.^118-122 In one report they synthesized a core of N_a-bromoacetyl-activated linear peptides, which were reacted with diverse primary amines.¹¹⁹ Reduction of a masked thiol and introduction of a reactive linker moiety thus enabled cyclization with the newly formed secondary amine for MCP generation. This approach used acoustic liquid transfer to generate macrocyclic peptides at a picomolar scale, which minimizes the use of reagents and decreases the cost of producing the macrocyclic library (Figure 8A,B). This library of 2700 MCPs was tested against the classically undruggable p53/MDM2 interaction, which is a critical interaction in numerous cancers and a bellwether target used by many studies for designer peptide synthesis.^123,124 Screening the library in a competition format against a displacement peptide bound to MDM2 in a 384-well plate format resulted in the identification of several hit scaffolds with low-to-sub-μM inhibition, including the sulfone containing macrocycle 9 (Figure 8C).¹¹⁹

Figure 8.

Nanoliter (pM scale) MCP library preparation using acoustic liquid transfer. A) General reaction scheme to prepare dipeptide library. B) Schematic of acoustic liquid transfer for small volume mixing. C) Chemical structure of macrocycle 9.¹¹⁹

Recently, a new “blind” screening strategy termed Peptide Exploration Platform with Tag-Free Intramolecular Chemistry (PEPTIC), was introduced for label-free MCP library synthesis.¹²⁵ This platform uses “CyClick” chemistry to cyclize peptides, where an N-terminal amine reacts with an embedded C-terminal aldehyde to generate a 4-imidazolidinone cyclization group (Figure 9A).¹²⁶ After screening for hit compounds binding to a POI, MCPs are then linearized using a DeClick approach, wherein hydrolysis of the 4-imidazolidinone linker occurs at high temperatures and low pH (pH 2–3). Remaining hits are then deconvoluted using liquid chromatography tandem mass spectrometry (LC-MS/MS) to determine the MCP sequence. This selection strategy is beneficial as it eliminates the need to append an additional biomolecule tag for identification, which may interfere with the intended interaction and lead to false-positive or false-negative results. As a proof-of-concept, this strategy was used to select an MCP that binds to an HIV-1 capsid protein that is critical in viral replication to prevent its interaction with other proteins.¹²⁷ A generated hit molecule, cyFG-3, displayed submicromolar affinity for the capsid protein (Figure 9B).¹²⁵

Figure 9.

Tag-free PEPTIC selection strategy for MCPs. A) Schematic workflow of PEPTIC. B) Chemical structure of cyFG-3.¹²⁵

2.2.3. One-Bead-One-Compound MCP Libraries.

A classic approach to MCP library generation and screening is the use of “one-bead-one-compound” (OBOC) libraries.¹²⁸ Here, MCPs are synthesized through a split-and-pool model, where individual beads react with a single amino acid at a time before being split to react with another amino acid. This continues until a library of up to 10⁹-members is created that can then be screened against various targets, after which hit compounds are sequenced via Edman degradation or mass spectrometry.^129-131 An early example of OBOC used to select MCPs against intractable targets was the discovery of inhibitors specific to HDAC6, which is implicated in the development of several cancers and neurodegenerative disorders.¹³² Ghadiri and Olsen designed an OBOC MCP library based on a α₃β tetrapeptide scaffold cyclized via head-to-tail cyclization.¹³³ Initial attempts to use a fully α-amino acid tetrapeptide failed due to difficulty in lactam formation, presumably due to high ring strain.¹³⁴ Including an additional backbone methylene in the single β-amino acid residue led to a much higher yield of lactamization.^134,135 After two generations of focused OBOC selections determined by structure–activity relationship (SAR) analysis, a highly potent and HDAC6-selective MCP emerged, which was confirmed in cells by observing increased α-tubulin acetylation, a hallmark of HDAC6 inhibition. Finally, they explored inhibition of cell proliferation in HeLa cells and showed only compounds that also inhibited HDAC1 reduced cell growth, suggesting HDAC1 is much more important for cell viability than HDAC6.¹³³ Since then, OBOC has been used to discover MCP inhibitors for other difficult-to-drug targets including the RGS4/Gα₀ interaction,¹³⁶ PTP1B,¹³⁷ c-MYC,^138,139 hLysRS/CA,¹⁴⁰ and the calcineurin/NFAT interaction.¹⁴¹ For additional reading on OBOC and similar methods, we point readers to other reviews.^142,143

The early version of OBOC had two significant limitations that hampered its widespread use: (i) high false positive rates and (ii) difficulty in sequencing MCPs.¹⁴⁴ The first problem has been improved by lowering the density of peptide ligand on beads, testing diverse bead substrates and also requires ad hoc optimization based on the libraries and targets of interest. Solutions to the second identification problem have been approached through several strategies that incorporate a second molecular barcode onto each bead, including DNA, small molecules, and peptides themselves in a so-called one bead two compound (OBTC) approach. OBTC screens a linear peptide barcode that identifies the MCP sequence that is synthesized in the interior of the bead and does not interfere with selection (Figure 10).^145-147 Additional innovations include DNA-encoded solid phase synthesis to encode peptide information into DNA barcodes read by sequencing.¹⁴⁸ A seminal example using OBTC involved screening the Ras–Raf interaction, and more generally, the interaction of Ras with its effector proteins. The Ras family of proteins are a group of G-proteins (K-Ras, H-Ras, N-Ras) and are a hallmark of cancer, with 30% of human cancers having at least one Ras gene mutated.¹⁴⁹ Ras has over 50 different effector proteins that interact with and modulate Ras signaling, with the most famous of these being Raf kinase. Pei and co-workers focused on a specific K-Ras mutation, G12V and discovered a peptide termed peptide 12, that has submicromolar in vitro IC₅₀ values.¹⁵⁰ It bound specifically to K-Ras(G12V) but did not have activity in cells due to poor cell uptake. To improve cell uptake and increase potency, a second-generation MCP library was generated with OBTC.¹⁵¹ Optimization of the cyclic peptide led to cyclorasin 9A5, which exhibited an in vitro IC₅₀ value of 120 nM and an EC₅₀ value of 3 μM in cells and prevented Ras signaling events such as phosphorylation of Akt. To further increase cell uptake, they rescreened using OBTC with a bicyclic peptide framework where one of the cycles was a cell-penetrating peptide (CPP).¹⁵² With this BCP framework, they generated peptide 49 that selectively bound to K-Ras(G12V) and exhibited good cell penetration (Figure 11). Peptide 49 inhibited Akt phosphorylation in cellulo and reduced proliferation of H1299 lung cancer cells with an EC₅₀ value of 8 μM. However, peptide 49 displayed a higher affinity for inactive state Ras·GDP over active state Ras·GTP, leading to high concentrations required to inhibit cell proliferation. Following this, they sought to make a pan-K-Ras-GTP MCP inhibitor.¹⁵³ Using OBOC based on a previous design, cyclorasin B3-4, they screened to find a BCP that preferentially bound to Ras-GTP. After a comprehensive optimization campaign, cyclorasin B4-27 was developed as a 10-fold selective inhibitor of K-Ras-GTP over K-Ras-GDP (Figure 11). B4-27 was successful in acting as a pan-Ras inhibitor against various Ras mutant cancer lines and reduced proliferation with modest potency (single-digit μM EC₅₀ values). Importantly, B4-27 exhibited a > 24 h half-life in vitro in human serum and significantly reduced tumor growth in A549 xenograft models following subcutaneous administration.¹⁵⁴

Figure 10.

One-bead-two-compound BCP library preparation using standard SPPS chemistry and split-and-pool diversification. Peptide chain is extended on both inner and outer bead sites. Inner chain is used as the identified readout by LC-MS/MS. Outer chain is cyclized and interacts with POI. (a) 2% TFA in DCM; (b) Fmoc-OSu/DIPEA in DCM; (c) Pd(PPh₃)₄.¹⁴⁷

Figure 11.

Chemical structures of CPP-MCPs selected from OBOC or OBTC libraries. Peptide 49¹⁵² and cyclorasin B4-27¹⁵³ target Ras interactions, GPM-1d¹⁵⁵ targets Gαi1, and Peptide 7¹⁵⁶ targets NEMO/IKK interactions. CPP portion is highlighted in red. POI recognition portion is highlighted in green.

Another important example of using bead-based compound libraries was to target Gαi1 and Gαs, which are heterotrimeric G-protein subunits that function downstream of G-protein coupled receptors (GPCRs) to route signals from GPCRs to downstream effector pathways.¹⁵⁷ This pathway has been tied to many different cancers, such as thyroid and adrenal tumors, and directly targeting G-proteins represents attractive alternatives to GPCRs, yet as a class G-proteins remain challenging for conventional small molecules.¹⁵⁸ Following a previously discovered MCP binding motif for Gαi1 and Gαs that follows a guanine-nucleotide exchange modulator (GEM) motif of ΦζΦX[±]ΩL, (where Φ is a hydrophobic residue, ζ is a uncharged hydrophilic residue, X is any residue, [±] is any positively or negative charged amino acid, Ω is an aromatic residue, and L is a leucine),^159-163 Nubbemeyer et al. generated and screened a OBOC library to discover a peptide, GPM-1, that potently binds to Gαi/s.¹⁵⁵ Optimization of GPM-1 by cyclization and conjugation with a CPP to form GPM-1d increased the affinity, structural stability, and cell permeability to modulate the activity of Gαi/s in cell culture in a bifunctional GEM-like manner (Figure 11). Structural analysis using biomolecular simulations analyzed underlying MCP-protein interactions by identifying a network of important H-bonds that occur between the CPP and the protein at Arg14-Asn250 and Arg16-Asn251. In addition, the MCP has a stacking interaction between Tyr8 and the N-terminal Ipa, along with a hydrophobic interaction between Trp2 and binding site.¹⁵⁵ A second-generation design identified GPM-3, which binds to active-state Gαi1 with a submicromolar binding constant and was shown to act as a GTPase-activating protein (GAP) modulator.¹⁶⁴ In a separate example, Rhodes et al. developed a 2.4 × 10⁹-member OBTC BCP compound library and incorporated cell penetration and cyclization into the activity screen against the NEMO-IKKβ interaction.¹⁵⁶ Optimization of the CPP and binding interface produced a BCP, peptide 7, with low μM inhibition activity on viability in multiple ovarian cancer cell lines and reduced canonical NF-κB signaling (Figure 11). Interestingly, as in the previous example, the positively charged CPP portion contributed to the binding of negatively charged NEMO surface, suggesting a potential cooperative approach for improving the affinity and cell permeability for MCP-CPP constructs.¹⁵⁶

2.2.4. DNA-Encoded and DNA-Programmed MCP Libraries.

DNA-programmable MCP libraries were first reported using DNA-templated synthesis (DTS) by the Liu lab in 2001.¹⁶⁵ DTS uses synthetic oligonucleotide-small molecule conjugates to template ordered monomer coupling as a result of proximity induced DNA hybridization and either regio- or chemoselective reactivity. Sequential addition of DNA-fragment conjugates, which react in a sequence-specific manner based on DNA templates in solution, can generate diverse linear and cyclized peptide, peptide-like and small molecule libraries. This technology was first adapted to generate MCPs by amino acid coupling and cyclization by Wittig olefination to generate libraries ranging from 10⁴ to 10⁵ members (Figure 12A,C).^166-168 These libraries were used to screen for bioactive compounds against phosphatases (DEP1, MEG2, PRL2), GTPases (Cdc42, H-Ras-V12, RhoA), antiapoptotic proteins (BCL-X_L), nuclear receptors, PDZ and SH2 domains, along with assorted kinases and proteases.^168,169 The most potent molecules derived from these screens were focused within kinase and protease families, including Src, VEGFR2, and IDE. Other DNA-programmed MCP libraries, such as those generated by teams at Ensemble Therapeutics and Bristol-Myers Squibb using azide–alkyne cycloaddition (~160,000 members; Figure 12C), have also been successfully screened against the caspase inhibitor proteins cIAPs and XIAP, and could be applied to other challenging protein targets.¹⁷⁰ These screens produced highly potent compounds that exhibited modest pharmacokinetic properties (half-lives ~2 h and mean residence times ~2.5 h in mice), and bioactivity against MDA-MB-231 xenografts in mice, showcasing potential for these libraries to produce MCP therapies.

Figure 12.

DNA-programmable MCP libraries. Schematics of MCP library preparation by A) DNA-templated synthesis or B) DNA-recorded synthesis. C) Common DNA-compatible chemistries used to prepare MCP DNA-libraries. D) Chemical structure of UNP-6457.¹⁷⁶

A conceptually similar and widely used approach for library synthesis and screening is DNA-encoded library (DEL) technology. First proposed in 1992¹⁷¹ as “DNA-recorded synthesis,”^172,173 DEL libraries have been used extensively to record the synthesis sequence of complex and large MCP and small molecule libraries, now commonly reaching ~10⁶–10⁹ members. DEL synthesis uses small molecule or amino acid fragments ligated onto DNA oligonucleotides, which can be subjected to linear or parallel split-pool synthesis rounds for further fragment coupling, which is accompanied by DNA-ligation to encode the identity of each fragment in each round (Figure 12B). This process repeats multiple times until the library is synthesized and each compound has a unique DNA barcode for identification. Chen and co-workers generated a 10⁶-member MCP DEL by performing a DNA-compatible palladium-catalyzed S-arylation reaction to react the thiol groups on cysteine residues with noncanonical aryl iodide side chains.¹⁷⁴ As a proof-of-concept, they screened the library against p300, a transcriptional coactivator integral to oncoproteins function,¹⁷⁵ and discovered two MCPs with single-digit μM inhibition potency. As expected, the linear precursors to these MCPs showed very poor (>100 μM) inhibition of p300, highlighting the necessary conformational constraints to achieve high affinity binding. DELs have also been used to target the transcription factor p53 and its interaction with MDM2. Su and co-workers synthesized an MCP DEL library of 10⁹ tetrapeptides cyclized via azide–alkyne triazole formation to pan for molecules that disrupted the interaction.¹⁷⁶ They discovered a hit compound, UNP-6457, that binds potently (low nanomolar K_d) to MDM2 and projected similar hydrophobic groups into the surface groove that interacts with the key p53 residues Phe19, Trp23, and Leu26, further confirming these necessary interactions for potent binding (Figure 12D). This compound, while not passively permeable in cells, could be a scaffold for improving biophysical and pharmacokinetic properties down the line.

DNA-programmable libraries have many potential advantages for MCP synthesis and screening. First, the miniaturization of synthesis and screening is enabled by the robust, enzyme mediated amplification of reaction products, but also places specific requirements and limitations on the compatible chemistries that can be deployed. Furthermore, because no purification of library members is needed, entire libraries can be generated, screened, retrieved and reintroduced to subsequent “evolution” rounds all in solution. Collectively, these attributes make DEL- and DTS-like approaches very attractive for MCP generation and application to undruggable targets. The requirement for DNA tags does hinder the use of DEL libraries against some undruggable targets, for example large families of TFs and chromatin regulatory proteins that have intrinsic affinity for DNA. Also, advances in the chemistries that can be used in DEL construction could continue to advance the likelihood that MCP hits are likely to contain other necessary attributes like cell penetration and metabolic stability.^177-179 For additional reading on DTS and DELs, we refer readers to other in-depth reviews.^180-182

2.2.5. Phage Display MCP Libraries.

First conceived by Smith in 1985, phage display is a powerful in vitro selection method that uses a bacteriophage, commonly M13 or fd, to express and display a library of peptides by fusing them to a surface protein, commonly pIII.^183,184 This library is used to screen for binding to the POI typically through affinity purification of the phage particles. Iterative rounds of selection can be performed to isolate high affinity interactors that are identified by sequencing the phage DNA. This technology has been adapted to select for mono- and bicyclic peptides (Figure 13). The earliest examples of using phage display to discover MCP ligands date back to the early 1990s and was used to identify integrin-binding MCPs containing the RGD motif.^185-188 In 1995, phage display was used to identify an MCP ligand for DNA matrix associated regions (MARs), the first example of a nonclassical drug target.¹⁸⁹ Since then, phage display has been adapted to discover MCPs against various difficult-to-drug protein and nucleic acid targets including TEAD4,¹⁹⁰ the Grb7-HER3 interaction,¹⁹¹ interactions between βB2-Crystallin fibrils,¹⁹² and the LEDGF/p75-IN interaction.¹⁹³ For further reading on phage display or other related technologies capable of selecting peptide ligands, such as yeast display, bacterial display, or mammalian display, we point readers to other reviews.^194,195

Figure 13.

Representative schematic of bicyclic phage display for selection of BCPs. Chemical structure of β-catenin/ICAT inhibitor BC1 is shown.¹⁹⁶

An important set of intractable targets is selectively inhibiting the highly conserved B-cell lymphoma-2 (BCL-2) family members. These proteins that include the antiapoptotic members BCL-2, BCL-X_L, and MCL-1 regulate the intrinsic apoptosis pathway and are highly upregulated in cancer, thus represent important therapeutic targets.^197-200 These proteins are very structurally similar and often perform the same role in the regulation of apoptotic cell death. Wu and co-workers used phage display to discover an MCP to target BCL-2.²⁰¹ In their MCP design, they used 2-((alkylthio)(aryl)methylene) malononitrile (TAMM) conjugated with a chloroacetamide moiety as the scaffold molecule to cyclize via two included cysteine residues (Figure 14A).²⁰² TAMM specifically reacts with the N-terminal cysteine residue to generate 2-aryl-4,5-dihydrothiazole (ADT) (Figure 14A). From this selection, they identified cp1, which selectively bound to BCL-2 19-fold more than BCL-X_L (Figure 14 B,C).²⁰¹ Interestingly, when the scaffold molecule TAMM is not present and the MCP is instead cyclized via a simple internal disulfide linkage, these MCPs had no binding affinity to BCL-2, suggesting the size and conformation of the ring plays a role in binding despite the exact same amino acid sequence or the scaffold molecule directly interacts with the protein. To look closer at the binding interactions of cp1 with BCL-2 over BCL-X_L they solved a crystal structure and found that the hydrogen bond formed with BCL-2 at D111 and the backbone of cp1 at G6 is the critical differentiating factor (Figure 14C). In BCL-X_L, A104, the closest residue to cp1 G6, is 7.2 Å away, precluding the formation of a hydrogen bond. Generating a BCL-2 D111A mutant led to significantly decreased cp1 binding. Another interesting finding was that cp1 binds to the BCL-2 surface without inducing conformational changes of the α3- or α4-helices, unlike small molecule or BH3-derived ligands. This unique binding mode allowed cp1 to interact with drugresistant BCL-2 mutants G101V and F104L. From here, a phage display library constructed from the cp1 sequence was prepared and used to discover MCPs specific for BCL-X_L, which selected for cp3 that showed ~3-fold more selectivity for BCL-X_L compared to BCL-2 (Figure 14D,E). Interestingly, they found that the residue at position 6 is key for affinity and selectivity, suggesting that highly similar surfaces require very few distinct interactions for selectivity. Unfortunately, poor membrane permeability even when tagged by the CPP Tat severely limited the MCPs efficacy in live cells, and significant efforts to produce clinically relevant compounds are needed.²⁰¹ However, the ability to bind wild-type and drug-resistant mutant BCL-2 family proteins may provide starting points for additional ligand development for these challenging targets.

Figure 14.

Phage display selection of BCL-2 or BCL-X_L-specific MCPs. A) Schematic of cyclization approach to generate MCPs with 2-((alkylthio)(aryl)methylene) malononitrile (TAMM) linker. B) Chemical structures of BCL-2 specific cp1 (left) and BCL-X_L targeting cp3 (right). C) Crystal structure of cp1 bound to BCL-2. Key interacting H-bond between cp1(G6) and D114 of BCL-2 shown in yellow. PDB: 7Y90. D) Crystal structure of cp3 bound to BCL-X_L. Key interacting residues between cp3(P6) and BCL-X_L are shown. PDB: 7YAA.²⁰¹

Another key family of undruggable targets to overcome are gain-of-function mutants of K-Ras. Sakamoto et al. used phage display and panned an MCP library to target K-Ras(G12D), the most common K-Ras mutation in human cancers.²⁰³ This screen identified a molecule, KRpep-2d, which contains a disulfide cyclization linkage and exhibited very high affinity (K_d = 51 nM) and more than 10-fold selectivity for K-Ras(G12D) over WT K-Ras, while the linear version of the peptide displayed no binding (Figure 15A). In cancer cell lines with the G12D mutant K-Ras, KRpep-2d suppressed cell proliferation and the Ras signaling cascade, suggesting the MCP may have some cell permeability and stability. This is unsurprising given that KRpep-2d contains 8 arginine residues in close proximity, which has previously been identified as a CPP motif.^204,205 A 1.25 Å resolution crystal structure demonstrated that KR-pep2d recognizes a unique binding region not previously targeted by other small molecule inhibitors (Figure 15D).²⁰⁶ In a follow-up study by the same group, to increase the efficacy of KRpep-2d, they used SAR analysis to design a BCP that would have reduced molecular weight, not be susceptible to reducing conditions and remain protease-resistant.²⁰⁷ First, they started with the crystal structure of KRpep-2d binding to K-Ras(G12D) and introduced ncAA mutations at positions 6–14. They found that substituting linear aliphatic hydrophobic amino acids at positions 7 and 9, along with hydrophobic aromatic amino acids at positions 11 and 14 led to increased binding. In addition, (S)-2,3-diaminopropanoic acid (Dap) at position 10 forms a salt bridge with D69 of K-Ras(G12D), which increased binding approximately 5-fold. To eliminate bond cleavage under reductive conditions, they substituted the disulfide with an amide linkage between positions 5 and 15, which yielded a 0.6-fold weaker binding affinity. Combining these modifications led to KS-36, which bound to K-Ras(G12D) 30-fold stronger than KRpep-2d but had weaker than expected cell growth inhibition (Figure 15B). They identified the arginine-rich CPP region as the limiting factor due to its endocytosis-dependent uptake, so they eliminated the arginine residues and created BCPs by joining together positions 4 and 11, along with a head-to-tail amide linkage. By substituting in either Cys or d-Cys at positions 4 and 11, they determined that Cys4, d-Cys11, and a 1,4-diiodobutane linker led to the greatest cell uptake. Final optimization at positions 5 and 10 to introduce (S)-2-aminononanoic acid and a tryptophan residue, respectively, generated KS-58 (Figure 15C). Computational examination of KS-58’s mechanism for cell uptake revealed that it behaves like a chameleonic molecule, altering its polar surface area and internal hydrogen bond network depending on the solvent.²⁰⁸ KS-58 displayed improved K-Ras(G12D)-dependent cancer cell growth inhibition compared to KRpep-2d in both liver and lung cancer cells. This is likely because KS-58 inhibits the interactions of both SOS1 and B-Raf with K-Ras(G12D), whereas KRpep-2d only inhibits SOS1. In a PANC-1 mice xenograft model, administration of KS-58 led to a sizable decrease (~35%) of tumor growth. Co-administration with gemcitabine, a common chemotherapy drug, led to a larger decrease in tumor growth, showing similar efficacy in different xenograft models of colorectal and pancreatic cancer.^209-211 However, KS-58 has low water solubility and requires frequent administration. To address both issues, KS-58 was made into a nanoformulation with a biocompatible surfactant, Cremophor EL, to generate a cell permeable nanoparticle.²¹² This formulation reduced the cancer cell growth rate in mice xenografts of both colorectal and pancreatic models. A separate team from Merck also looked to optimize KRpep-2d to improve its pharmacologic properties.²¹³ Through SAR, they designed MCPs that are membrane-permeable, protease-resistant, and inhibit cell proliferation in K-Ras(G12D) cancer cells. However, these polyarginine-rich MCPs, which were included to improve cell uptake, induced mast cell degranulation (MCD) that in extreme cases can lead to patient death. They found that there was a strong correlation between the number of arginine residues and MCD, with MCPs containing greater than two arginine residues inducing MCD. Though these designs were not taken further, these studies demonstrated the potential to use MCP display approaches to identify high affinity and cell permeable ligands for a challenging and as-yet undrugged target like K-Ras (G12D).

Figure 15.

SAR optimization of KRpep-2d. Chemical structures and properties of KRpep-2d (A), KS-36 (B), and KS-58 (C). Cyclization linker is highlighted in green. D) Crystal structure of KRpep-2d bound to K-Ras (G12D). Key interacting residues are labeled. PDB: 5XCO.^203,206,207

One of the biggest limitations of the first generation of phage display is the inability to incorporate ncAAs. Although ambercodon suppression can only incorporate the one ncAA, it can lead to a genetically encoded, intramolecular thioether linkage. A phage display library of MCPs that incorporated this design was used to discover MCPs with single-digit micromolar dissociation constant and in vitro IC₅₀ values for HDAC8, a class I HDAC member.²¹⁴ Macrocyclic organo-peptide hybrid phage display (MOrPH-PhD) developed by Owens et al. works in a similar vein, but incorporates O-(2-bromoethyl)-tyrosine as the ncAA.²¹⁵ This new phage display technique was used to disrupt the KEAP1/NRF2 interaction, which is part of the stress response pathway.^216,217 Owens et al. discovered an MCP that interacted strongly with KEAP1 (K_d = 40 nM) and the amino acid sequence closely matches the tip of the NRF2 β-hairpin that interacts with a natural cleft in KEAP1.²¹⁵ Interestingly, though the most strongly binding sequences closely matched the NRF2 β-hairpin, there was one comparatively weak MCP (K_d = 2.8 μM) that contained a different sequence, suggesting that it interferes with the KEAP1/NRF2 interaction at a different location. Using a different variation of phage selection, Zheng et al. also generated an MCP against the KEAP1/NRF2 interaction.²¹⁸ In this approach, two cysteine residues were introduced and cyclized with a special linker, 2-cyanobenzothiazole conjugated with an α-cyanoacrylamide. The N-terminal cysteine reacts specifically with the 2-cyanobenzothiazole while the other cysteine reacts with the α-cyanoacrylamide to generate the MCP, similar to the reaction with the TAMM scaffold.²⁰² The highest affinity MCP (sub-μM K_d) also contained close sequence homology with the NRF2 β-hairpin as those selected by Owens et al. Interestingly, although using different linkers and ring sizes for generating MCPs, the top sequences between the two studies differed by only one residue within the β-hairpin sequence (DSETGE vs DIETGE), and obtained similar affinity. This suggests this sequence is likely highly specific for KEAP1 and can tolerate different substitutions, potentially to develop KEAP1-targeting MCPs with improved pharmacological properties.

To increase the diversity and versatility of phage display, a BCP selection strategy was developed by introducing three cysteines into the peptide sequence and adding 1,3,5-tris(bromomethyl)benzene (TBMB) to effectuate the production of BCPs via thioether bond formation.²¹⁹ Using this strategy, BCPs have been selected against two different historically undruggable targets such as the β-catenin/ICAT interaction and Dcp2.^196,220 β-catenin is a cell signaling molecule that is a central component of the WNT/β-catenin pathway, among other roles.^221-223 Mutations in this pathway can lead to cancer, and β-catenin is seen as a key therapeutic target. Bertoldo et al. discovered a β-catenin-targeting BCP using phage display by testing three different scaffold molecules: TBMB, TATA (1,3,5-triacryloyl-1,3,5-triazinane), and N,N′,N”-(benzene-1,3,5-triyl)-tris(2-bromoacetamide) (TBAB). All three scaffold molecules yielded MCPs with single-digit micromolar dissociation constants; however, MCPs cyclized by TBMB and TBAB interfered with ICAT binding, which inactivates the WNT pathway by binding to β-catenin.¹⁹⁶ In a separate study, Luo et al. selected a high affinity BCP (K_d = 116 nM) for Dcp2,²²⁰ which is an enzyme that removes the 5′ cap from a subset of mRNA molecules that is difficult to target due to an intrinsically disordered region (IDR) and intracellular delivery.^224-226 This BCP engages the IDR of Dcp2 and inhibits its activity in cells. The BCP was also used to find novel mRNA sequences that are recognized by Dcp2 in the decapping process, demonstrating how MCPs can be used as chemical probes to elucidate novel biology surrounding undruggable targets.

Multiple different strategies have been reported to form BCPs and tricyclic peptides using phage display and are summarized in this recent excellent review.²²⁷ This strategy has yielded several bicyclic peptides now in clinical trials with Bicycle Therapeutics, such as BT8009 targeting nectin-4,²²⁸ BT5528 targeting the EphA2 receptor,²²⁹ BT1718 targeting the zinc metalloprotease MT1-MMP,²³⁰ and BCY12491 targeting the CD137 receptor.²³¹ Even though none of these targets are classically undruggable, this selection strategy should be amenable to additional targets in the future. Further developments in incorporation of ncAAs may also yield MCPs with better pharmacological properties. Additional phage display technologies have been used to synthesize lanthipeptides, and these may be used to target classically undruggable proteins in the future.²³² Finally, yeast selection, though used much less frequently than phage display, can also be used to find MCP inhibitors. First reported in 2012, Bowen et al. used yeast display to generate MCPs cyclized via lactam formation to target the WW-domain and the N-terminal region of YAP1, an intractable transcriptional coactivator in the Hippo signaling pathway.^72,233,234

2.2.6. mRNA Display of MCPs.

Simultaneously designed by the Yanagawa lab and the Szostak lab in 1997, mRNA display involves transcribing a diverse DNA library into mRNAs before puromycin, an antibiotic that terminates translation in the ribosome, is conjugated at the 3′ end (Figure 16).^235,236 Peptides translated from the mRNA are covalently attached to puromycin to terminate translation, leaving an mRNA-peptide conjugate that can be biopanned against various POIs. After identification of interacting peptides, RT-PCR is performed on the RNA to identify the peptide sequence. In a move away from endogenous expression systems, modern mRNA display methods largely rely on in vitro translation, an early example of which is the protein synthesis using recombinant elements (PURE) system developed in 2001 from purified aminoacyl-tRNA synthetases, initiation factors, elongation factors, release factors, RNA polymerase and ribosomes to reconstitute translation.²³⁷ This approach can produce relatively large libraries of 10¹²-members. mRNA display has been used to select MCPs against numerous intractable targets and interactions such as p53/MDM2,²³⁸ SIRT2,²³⁹ SIRT7,²⁴⁰ TET1,^241,242 and KDM7,²⁴³ among others.^244-247

Figure 16.

General schematic of mRNA display for MCP selection.

An early example of using mRNA display to select MCPs against undruggable targets was performed by Millward et al. to target Gαi1·GDP.¹⁵⁹ The initial MCP, cycGiBP, exhibited high target affinity but was prone to protease cleavage. To overcome this limitation, they developed scanning unnatural protease resistant (SUPR) mRNA display (Figure 17A). In SUPR, the library incorporates ncAAs such as N-methylalanine and is exposed to a protease cocktail prior to affinity purification, thus only selecting for protease-resistant interactors.¹⁶¹ This method was used to identify Gα SUPR (Figure 17B), which maintained similar binding affinity to cycGiBP but exhibited a > 35-fold increase in serum stability a half-life of 115 h, thus highlighting the versatility of in vitro display selection approaches.¹⁶¹

Figure 17.

Scanning unnatural protease resistant (SUPR) mRNA display. A) Schematic of SUPR mRNA display selection for protease-resistant MCP ligands. B) Chemical structure of Gαi1 inhibitor Gα SUPR.¹⁶¹

Building off of the PURE in vitro translation system, the Suga lab designed the random nonstandard peptide integrated discovery (RaPID) system, where mRNA display is combined with a flexible in vitro translation (FIT) system to screen large (>10¹²-member) MCP libraries.^248,249 Along with the other recombinant elements, FIT involves flexizymes, which are tRNA-acylation ribozymes that can install ncAAs on to tRNAs that have been programmed to recognize different codons during translation (Figure 18A). The benefit of this system is that ncAAs can easily be incorporated into MCPs for testing via mRNA display, which expands the chemical diversity used to discover interactors. The most popular strategy for cyclization using RaPID incorporates a chloroacetamide moiety in the form of an N-chloroacetyl amino acid located at the N-terminus that reacts with an embedded cysteine to form a thioether bond.²⁵⁰ This strategy has been used to find MCPs against difficult-to-drug targets such as K-Ras-(G12D),²⁵¹ α-Synuclein amyloid fibrils,²⁵² and the retromer complex involved in endosomal membrane trafficking.²⁵³ For further reading on mRNA display or other similar display technologies, such as ribosome display, we point readers to other relevant reviews.^254-256

Figure 18.

Random nonstandard peptide integrated discovery (RaPID). A) General schematic of RaPID incorporating flexizyme technology to include ncAAs during in vitro translation. B) Chemical structure of ^K48Ub₄-selective MCP Ub4a generated by RaPID. Thioether linker used for head-to-tail cyclization is highlighted in green. C) Crystal structure of Ub4a bound to ^K48Ub₃ displaying ring-like arrangement (left) and H-bonding network (right, insert, yellow dashes) for recognition. PDB: 8F1F.^258,260

An interesting example of using RaPID to discover an MCP is for the recognition of poly ubiquitin (polyUb).^257,258 Ubiquitination is a post-translational modification where a small protein (ubiquitin, Ub) is attached to mainly lysine side chains on cellular proteins to perform different functions including degradation by the proteasome.²⁵⁹ Interfering with the Ub-proteosome interaction is an attractive approach for cancer therapy. PolyUbs have diverse linkages that yield different conformations and diverse biological effects that are difficult to distinguish with small molecules like ubistatins. Using RaPID, Nawatha et al. specifically targeted K48-linked tetra-ubiquitin (^K48Ub₄), which is important for 26S proteosome recognition.²⁵⁷ They discovered a single MCP, Ub4ix, that was capable of distinguishing Ub linkage and chain size by targeting ^K48Ub₄ with low nanomolar affinity. Ub4ix binding prevented USP2, a nonspecific deubiquitinase, from cleaving more than one Ub from the target, resulting in blocked deubiquitination of proteins like p53 and p27 that then accumulated in HeLa cells. To increase cell permeability and protease resistance, the same group performed another RaPID selection and incorporated ncAAs ClAc-d-Phe, N-methyl-Gly, N-methyl-Ala, N-methyl-Phe, d-Ala, d-Phe, and Aoc to replace Met, Glu, Asp, and Arg.²⁵⁸ This selection produced a cell permeable MCP, Ub4a, that bound to ^K48Ub₄ with high affinity (K_d = 9 nM; Figure 18B). Interestingly, structural NMR analysis of Ub4a bound to ^K48Ub₄ suggested that there is minimal interaction with the distal ubiquitin and that Ub4a is actually specific for ^K48Ub₃, which may explain the mechanism of only allowing a single Ub cleavage similar to Ub4ix. This was later confirmed by a solved crystal structure showing tri-Ub wraps around Ub4a in a ring-like arrangement driven mainly by H-bonding and hydrophobic interactions (Figure 18C).²⁶⁰ Significantly, Ub4a demonstrated anticancer activity in a mouse model of CAG myeloma by reducing tumor burden in a manner similar to the approved 26S proteasome inhibitor medication bortezomib.²⁵⁸

RaPID can also be used to explore physicochemical properties of MCPs and the effects on target binding, as ring size and residue identity play important roles in target recognition. For example, Lin et al. used two separate RaPID libraries and compared 7–9 residue MCPs to 10–14 residue MCPs targeting the stimulator of interferon genes (STING) adaptor protein.²⁶¹ These libraries were very diverse with 25 potential amino acids that could be incorporated. Surprisingly, they found that only 9-mer MCPs with 8-residue ring size exhibited measurable affinity using surface plasmon resonance (SPR). This is potentially due to the more thorough sampling of sequence space with fewer randomized positions or more likely the smaller loss of conformational entropy for shorter peptides to bind the target. Another key question is whether RaPID explores enough sequence space to select for the most efficient interactors. Suga and colleagues used RaPID to efficiently perform deep mutational scanning and explore the effects of ncAA incorporation for CP2, an MCP previously selected via RaPID to target the histone demethylase KDM4A-C.^262-264 Including ncAAs changed properties such as charge, steric bulk, N-methylation, or number of hydrogen bonding groups. Interestingly, they found few mutations resulted in increased binding affinity, while many substitutions had little effect on binding energy. Although likely to be target-specific, many MCPs can potentially be screened and optimized through numerous modifications to gain desirable properties.

While small molecules have difficulty distinguishing between different conformational states, MCPs are more likely to behave as state-specific inhibitors. Dai, Hu, and Gao et al. showed this by using RaPID to select for MCPs that distinguish between GTP- or GDP-bound Gαs.²⁶⁵ To promote selectivity, MCP libraries were exposed to negative selection for the undesired alternate states. Using this approach, they developed MCPs that could target only the Gαs·GTP state (MCP termed GN13) or Gαs·GDP state (MCP termed GD20), which were determined to be at least 100-fold selective for their respective target state (Figure 19A and C). The MCPs bind to the Gαs switch II-α3 pocket, which adopts a different conformation dependent on nucleotide to discriminate between the different states. Analysis of crystal structures showed that GN13 interacts with active state Gαs through necessary and complex hydrogen-bonding network and hydrophobic interactions, whereas an expected steric clash of Gαs R232 in inactive-state switch II domain likely explains the conformationally selectivity (Figure 19B). Conversely, GD20 recognizes the inactive state Gαs through important electrostatic interactions, hydrogen bonding, and hydrophobic interactions, and likely sterically clashes with R231, R232, and W234 in the active state switch II domain (Figure 19D). In addition, both GN13 and GD20 demonstrated excellent specificity for Gαs over other structurally similar G-protein members regardless of state. Additional modifications to both MCPs improved cell permeability and bioactivity, resulting in compounds capable of in cellulo disruption of Gαs interactions and activity, demonstrating the power and versatility of RaPID and generated MCPs.²⁶⁵

Figure 19.

Discrimination of active and inactive Gαs GTPase by RaPID-selected MCP. A) Chemical structure of active Gαs-selective GN13. B) Crystal structure of GN13 bound to active-state Gαs. H-bond between GN13 and R232 is highlighted as yellow dashed line. PDB: 7BPH. C) Chemical structure of inactive Gαs-selective GD20. D) Crystal structure of GD20 bound to inactive-state Gαs. Key interactions between GD20 and R231 and W234 of Gαs are highlighted. PDB: 7E5E. Thioether cyclization linkage highlighted in green in chemical structures.²⁶⁵

The team at Chugai Pharmaceuticals who identified the previously described desirable MCP properties for drug-like molecules used their guidelines to generate an mRNA display library to select for inhibitors to K-Ras.^65,266 Incorporating a high number ncAAs, particularly N-alkylated amino acids, and native chemical ligation for cyclization resulted in a library that could satisfy the criteria for drug-like MCPs. They discovered an 11-residue MCP, AP8784, that bound at the interface of KRas and SOS1 to potently inhibit the interaction with an IC₅₀ of 54 nM but showed no activity in live cells (Figure 20). Following their guidelines and a crystal structure to improve the physicochemical properties of AP8784, they embarked on a comprehensive SAR campaign.²⁶⁶ The crystal structure revealed AP8784 interacted with the Switch II pocket on GDP-K-Ras(G12D), which identified positions 8 and 11 as likely to be amenable to modification without disrupting binding, while positions 2, 5, 7, and 10 made direct surface contacts. Positions 7 and 10 specifically occupied the SII-hole, and likely could be modified to enhance bioactivity. To summarize the outcome of the SAR analysis, positions 7 and 10 were optimized from 3-chloro-phenylalanine at position 7 and N-methyl-valine at position 10 to homophenylalanine(4-CF₃,3,5-F₂) and N-methyl-cyclopentyl for positions 7 and 10, respectively to improve bioactivity. Further improvement to cell permeability was achieved by altering positions 8 and 9 from tryptophan and N-methyl-phenylalanine to a proline and cycloleucine, respectively. Rigidity was increased by modifying positions 3 and 4 from N-methyl-glycine to N-methyl-alanine at position 3 and N-ethyl-azetidine-2-carboxylic acid at position 4. Finally, replacing position 1 to N-methyl-leucine, position 5 with an N-ethyl-4-methyl-phenylalanine, and N-alkylation of position 11 resulted in lead compound LUNA18 (Figure 20). This lead MCP contained desirable drug-like properties such as a high Clog P, and membrane permeability via Caco-2 assay without the need to significantly alter the scaffold. LUNA18 displayed a > 25-fold improvement to potency compared to AP8784 and exhibited low picomolar affinity for GDP-K-Ras-WT and G12C, G12D, and G12V mutants. LUNA18 selectively reduced proliferation in K-Ras-G12X-dependent cancer cell lines and showed dose-dependent antitumor activity after oral administration in mice.⁶⁵ LUNA18 is currently undergoing a phase I clinical trial.

Figure 20.

SAR optimization of pan-K-Ras targeting MCPs. A) Chemical structure of AP8784. B) Chemical structure of LUNA18 after SAR optimization. Amide cyclization linkage is highlighted in green.²⁶⁶

mRNA display technologies are very powerful tools for selecting MCPs against various targets that have the potential to enter and succeed in clinical trials, key examples being FDA-approved zilucoplan^267,268 and more recently orally available PCSK9 inhibitor MK-0616 currently in phase III.^269-273 These examples and LUNA18 demonstrate how encodable designer peptides selected from large libraries can undergo synthetic optimization to produce viable clinical candidates. Including the ability of RaPID to incorporate multiple ncAAs during selection and SUPR to select for metabolically stable ligands, variations of mRNA display can be combined to generate MCPs with ideal pharmacological properties. Improvements to these technologies harness the power of computation to categorize the top hits into different binding modes on the POI, providing different starting points for further optimization.^274,275 Given the relatively large library size, these technologies may generate more pharmacologically viable lead MCPs that function in cells or animals with less required modification and optimization.

2.2.7. Split Intein Circular Ligation of Peptides and Proteins.

Split Intein Circular Ligation of Peptides and Proteins (SICLOPPS) is a selection technique for discovering potent MCP inhibitors in live cells (Figure 21).^276,277 In this approach, split inteins at the C- (I_C) and N-terminus (I_N) are used to catalyze head-to-tail peptide ligation (Figure 21B). After expression, trans-splicing of the peptide precursor occurs by the interaction of the termini to form a lariat structure, wherein the first residue at the N-terminus, usually a cysteine or serine, forms a thioester or ester linkage to cyclize the peptide, which is then spontaneously converted to the native lactam (Figure 21C). This naturally restricts the first residue of the MCP to either a serine or a cysteine but is amenable to various functional groups at the other positions in the MCP library. Since the cyclization process occurs in cellulo, the advantage of this selection is that it can be paired with bacterial or yeast reverse two-hybrid systems to select for MCPs that disrupt PPIs.^276,277 For further reading on the properties and applications of SICLOPPS we point readers to other reviews.^278,279

Figure 21.

SICLOPPS selection of macrocyclic peptides. A, B) Representative schematic of in cellulo selection strategy for cyclic peptides targeting HIF1α interactions. C) Representative example chemical structures of a cyclic peptide proceeding through SICLOPPS selection to produce HIF1α inhibitor cyclo-CLLFVY.²⁸⁵

The earliest use of SICLOPPS to identify an MCP against a target was in 2004 where it was used as a proof of principle to target HIV-1 protease and ribonuclease reductase.²⁸⁰ Since then, it has been adapted for several undruggable target interactions such as BCL6 dimerization,²⁸¹ IDOL E3 ligase homodimerization,²⁸² the interaction between HIV Gag protein and TSG101,²⁸³ between the GTPase Ras and p110α,²⁸⁴ and HIF1α/HIF1β heterodimerization.^285,286 In the latter example, an MCP SICLOPPS synthesis and screening strategy was developed where HIF1α was fused with the P22 bacteriophage repressor, and HIF1β was fused with bacteriophage 434 repressor (Figure 21A). In this system, HIF1α and HIF1β heterodimerization, which is an interaction necessary for transactivation and activation of pro-oncogenic gene programs,^287,288 controls the 434/P22 repressor binding to 434/P22 operator sites preventing transcription of selection genes necessary for survival. Disrupting this interaction with an MCP results in cell survival and propagation to be later sequenced to determine the MCP identity. After positively screening for activity and negatively screening against false positives and nonspecific inhibitors, they identified a peptide, Tat-cyclo-CLLFVY, that showed activity against HIF1α/β dimerization in two different cancer cell lines, MCF-7 and U2OS.^285,289

In another example, Tavassoli and colleagues used SICLOPPS to target dimerization of the C-terminal binding protein (CtBP) transcriptional repressor.²⁹⁰ There are two CtBP proteins, CtBP1 and CtBP2, that can either homodimerize or heterodimerize under high NADH conditions.^291,292 In their dimeric form, CtBPs associate with around 30 different transcription factors to recruit histone deacetylases and histone methyltransferase to silence gene expression at specific loci.^293,294 CtBP dimerization has been implicated as regulators of highly glycolytic tumor cells.²⁹⁵ Using a bacterial reverse two-hybrid system similar to the previous example, they first probed MCPs that disrupted CtBP1 homodimerization before taking those hits to probe their activity in disrupting CtBP2 homodimerization. They found a cyclo-nonapeptide, CP61, that inhibited CtBP1 and CtBP2 homodimerization both in vitro and in cellulo.²⁹⁰

Although it is uniquely capable of screening for interactions in live prokaryotic or eukaryotic cells, a current limitation with the SICLOPPS approach is that it is limited to the 20 naturally occurring amino acids with a few ncAAs able to be incorporated. Additional future studies should also focus more on targeting diverse undruggable sites with MCPs of varied ring sizes to explore the relationship between size and selectivity.²⁷⁸ Increasing the chemical diversity and library sizes within SICLOPPS technology platforms remain open areas to increase the utility of this approach for undruggable targeting.

2.3. Summary and Outlook of Macrocyclic Peptides for Undruggable Targets

MCPs are well-established therapeutic modalities with desirable pharmacological properties capable of modulating interactions in a variety of undruggable targets. Approaches to design and generate MCPs in silico, in vitro, and in cellulo highlight the versatility for this molecular class. Rational design and selection methods from diverse libraries have produced metabolically stable and cell permeable MCPs. Genetically encoded libraries (GELs) such as mRNA display, phage display, and SICLOPPs are particularly powerful for efficiently screening large compound libraries with reduced time and cost.²⁹⁶ Recent advances in genetic code expansion and late stage functionalization²⁹⁷ greatly diversify library members for selection, such as including reactive electrophiles^298,299 and multicyclization.²²⁷ Expanding these strategies for MCP generation provides novel avenues for targeting undruggable classes.

For MCPs to translate into therapeutically viable molecules they should bind tightly and specifically to the POI, permeate cell membranes, and retain activity and structure in cells and animals. The different strategies for the design and production of MCPs described have all been used to generate very potent MCPs for undruggable targets, but many are unable to enter cells without significant modification greatly limiting their therapeutic potential. Next steps would be to take these selection strategies and bias sampling toward MCPs that have chameleonic properties. This would increase the odds that MCPs selected would also have good pharmacological properties. An additional shortcoming is that the activity of many compounds found through in vitro methods does not translate directly to in cellulo or in vivo. Either improvements on in cellulo selection methods to incorporate ncAAs or changes in in vitro methods to select for only MCPs that perturb POI activity in a high throughput manner would help to address this issue.

MCPs can also serve a slightly different role: as a warhead for a PROTAC. Most PROTACs developed today use small molecules as warheads; however, a lack of selectivity for intractable targets may lead to off-target effects with potentially devastating responses. As MCPs and other designer peptides exhibit very specific interactions with target molecules, including these as PROTAC warheads could limit off-target effects. Though few MCP or designer peptide PROTACs have been reported,^300-302 this is a potentially powerful space to explore, as eliminating an undruggable protein rather than simply blocking function may provide a greater therapeutic effect. Additionally, as MCPs discussed in this review retain peptide-like structure and may be susceptible to proteolytic degradation, efforts to rationally derive nonpeptide based compounds (e.g., class D mimetics) from peptide-based structures may result in improved PROTAC warheads.

Though the primary reason to discover and generate MCPs against targets of interest was to develop them as therapeutics, they can also serve to explore basic biology. As a couple of examples above demonstrate, MCPs are great tools to acutely and specifically disrupt biological pathways to study the biological role of the target without requiring significant external manipulation of the cells. As more MCPs are discovered, they can be used to further elucidate key pathways in disease states or to study how disrupting the formation of a complex affects protein expression and function.

3. SIDE CHAIN STABILIZED SECONDARY STRUCTURES: DESIGNER HELIX AND HAIRPIN PEPTIDES

The hotspot regions involved in protein–protein and protein–nucleic acid interactions are often found within well-defined secondary structure motifs such as α-helices, β-sheets and β-hairpins. In particular, a majority of interactions found in the PDB involve α-helices, which adopt an i, i+4 hydrogen bonding pattern in the amide backbone with an average of 3.6 residues per turn (Figure 22A).³⁰³ 3₁₀ and π helices, which have i, i+3 and i, i+5 hydrogen bonding patterns, are also found in some protein interfaces. β-Sheets also have an extensive internal hydrogen bonding network between the amide backbone of adjacent protein strands that may run parallel or antiparallel. Smaller peptides generally form β-hairpins, consisting of two antiparallel strands where the i, i+2 side chains are oriented in the same direction (Figure 22B).³⁰⁴ Because these secondary structure motifs are common at interfaces, approaches to chemically mimic and stabilize these structures could serve as general methods for chemical probe development.

Figure 22.

Depiction of a typical α-helix, 3₁₀-helix, and β-hairpin. A) Top (left) and side (right) representation of an α-helix and 3₁₀-helix depicting hydrogen bonding patterns (dotted yellow lines), residue spacing, and helical turn distance. B) Representative β-hairpin showing hydrogen bonding pattern between antiparallel strands (dotted yellow lines).

Early attempts to stabilize peptide secondary structure folds relied on recreating the interactions used in native protein folding, namely hydrogen bonding, ionic and hydrophobic interactions, steric constraints, and covalent disulfide bridges.^305-314 These approaches were soon supplanted with the emergence of more stable cross-linking reactions such as lactam formation,^315-317 ring-closing metathesis,^318,319 and azide–alkyne cycloadditions,³²⁰ among many other novel strategies to stabilize secondary structures. Collectively, this class of side chain stabilized secondary structure mimetics has become broadly referred to as “stapled peptides.”

3.1. Design Principles for Secondary Structure Stabilization

Since it is known that certain amino acids have a stronger propensity toward helix formation or β-turn induction, strategic additions and mutations to the peptide sequence can also help to stabilize the desired secondary structure. Namely, Ala, Glu, Lys, Met, and Leu are traditionally considered helix-promoting residues (Figure 22A).³²¹ Conversely, residues like Gly, Pro, Ser and Thr generally disfavor α-helical structures, as Gly, Pro, and other non-natural turn inducing residues such as l-ornithine promote and stabilize β-hairpin structures (Figure 22B).^322,323 However, these residues are still useful in the design of stabilized α-helices as they can form kinked helical structures and initiate/terminate helical segments,³²⁴ which have been shown to be beneficial for potency in certain contexts. Fuchs et al. demonstrated this in a ribosome display screen of LXXLL peptides selected for binding to ERα and ERβ.³²⁵ After several rounds of enrichment, the consensus sequences for both isoforms were capped on either side with Pro, and subsequent investigation revealed that these residues limited the helix length and oriented adjacent charge clamp residues appropriately. Additionally, α,α-disubstituted amino acids reduce rotational freedom due to the potential steric clashes, which can contribute to overall rigidity. α-Aminoisobutyric acid (Aib) is a disubstituted amino acid first discovered in some fungal peptides, and its geminal methyl groups contribute to a steric effect that strongly promotes both 3₁₀ and α-helix formation (Figure 22A).^326,327 Its achirality also makes it suitable for inclusion in either l- or d-peptides. The α,α-disubstituted, olefin-containing amino acids used in hydrocarbon stapling (vide inf ra), S₅ and R₈, have a similar helix-inducing effect.^319,328

The formation of secondary structure generally helps to overcome the pharmacological pitfalls of poor cell permeability and proteolytic stability. A combination of hydrophobicity and positive charge, particularly in amphipathic helices, broadly correlate with better cell uptake, but beyond a certain threshold, these properties also lead to membrane disruption and cell lysis.^56,329 The presence of intramolecular hydrogen bonding, which typically shields the polar amide backbone from aqueous solvent interactions, makes entry into the lipid membrane more energetically favorable, so long as pendant side chains are appropriate for interaction with and diffusion through lipid bilayers.³³⁰ Balancing these properties is an area of great importance for designing and developing membrane permeable secondary structure mimetics for undruggable targets, as discussed above with MCPs. Consideration of these properties will be important in the discussion of exemplar stapled peptides and other stabilized mimetics in this section. Helix and hairpin motifs by definition have these internal hydrogen bond networks, and successful introduction of stabilizing mutations through rational structure optimization takes these patterns into account by incorporating constraints in positions that promote the desired hydrogen bonding pattern and “locking” the peptide into its proper conformation. This also reduces the exposure of the backbone to degradation by proteases and other metabolic enzymes. The amino acid positions that are covalently linked are generally ones that will be proximal to each other in 3D space when the peptide is folded in the desired conformation. For helical peptides, the most common staple positions are i, i+4 and i, i+7, as these positions span one and two turns of an α-helix, respectively (Figure 22A). For a hairpin structure, the staple is usually positioned across the two antiparallel strands or in some cases, across the termini to clamp the two strands together.³³¹ However, there are important caveats to consider when making amino acid substitutions and selecting staple positions. For example, while the introduction of d-amino acids generally confers proteolytic stability, as enzymatic cleavage is stereospecific for l-amino acids, d-amino acids may also destabilize the helices of predominantly l-peptides.³³² There is also a distinction between helix stabilization and introduction of structural rigidity. For example, a hydrogen exchange study of a small peptide library with i, i+4 hydrocarbon staples in various positions concluded that solvent exchange rates, which are directly correlated to the conformational freedom and solvent exposure of the amide protons on the backbone, were a better predictor of protease resistance than more classic measures of helicity.³³³ This highlights the fact that structural stabilization alone may not be enough to achieve ideal pharmacological properties, and side chain stabilization chemistry and placement must be carefully considered through iterative design and testing. We will consider these and other important elements of design and pharmacologic use for stabilized peptides throughout this section.

3.2. Early Side Chain Stabilization Strategies for Secondary Structure Mimetics

Early studies with model peptides and proteins demonstrated that properly positioned salt bridges could stabilize helical peptide stretches in solution. For example, studies with a short helical peptide derived from RNase A only exhibited helical character at the pH in which a protonated His residue at the i position interacts with the Glu carboxylate at the i+4 position, effectively stabilizing one helical turn.³⁰⁵ Similarly, attempts to form β-hairpin structures often used model peptides that normally adopted this conformation in the protein. Many early structural NMR studies and computational models reached the conclusion that the hairpin is nucleated by the presence of a β-turn that often utilizes specific residues such as Asn, Pro, and Gly.^314,334,335 Other early studies recognized the potential for metal chelation on specific helical faces or between helix-turn-sheet motifs commonly found in zinc-finger domains, to be highly effective at nucleating and stabilizing peptide secondary structure.^{307,308,336-338} Another early stabilization chemistry explored was natural intra- and interchain covalent stabilization via disulfide bond formation. The short cysteine side chains are more practical for hairpin and loop structures, but some examples with i, i+3 spacing between d-Cys and l-Cys have been reported.^339-341 In 1991, Jackson et al., reported a series of peptides with two orthogonally protected 2-amino-6-mercaptohexanoic acid residues at i, i+7 positions that were oxidized to form the disulfide in an aqueous solution (Figure 23A).³⁰⁹ CD spectroscopy confirmed an increase in the α-helical character for all peptides regardless of length. While the disulfide bridge spanned two helix turns in this system, it was shown to nucleate helix formation for longer sequences.

Figure 23.

Early examples of helical stapling. A) Oxidation to i, i+7 disulfide bridge reported by Jackson et al.³⁰⁹ B) General scheme for side chain lactam formation. Any standard coupling reagent and base typically used in SPPS workflows can be used. C) Early example of olefin metathesis to stabilize 3₁₀ helix formation by Blackwell, Grubbs, and colleagues.³¹⁸

Lactam formation between amino acid side chains had been extensively applied to macrocycle synthesis³¹⁵ specifically used in early helix stabilization studies, usually forming amide bonds between Asp and Lys, or Glu and Lys, with the capability to form multiple staples using orthogonal protecting groups (Figure 23B).³¹⁶ Phelan et al. took an alternate approach to this chemistry by using two acidic residues bridged with diamine linkers of variable length,³¹⁷ which permitted exploration of the effect of different staple lengths on helical structure. Shorter methylene bridges resulted in some helical bending, while longer bridges allowed more conformational freedom leading to transient unwinding of the helix. While lactam staples are more stable against redox conditions in different cell compartments, neither it nor disulfide staples are completely biorthogonal and can be reversed in biological environments.

Lactam staples are still widely used for the development of peptide ligands for undruggable targets. For example, Wang et al. designed a stapled peptide derived from the histone H3K56 acetylated peptide site, which is found within an extended α-helix.³⁴² They recapitulated a 12-residue stretch of this helix to contain a chelating hydroxamic acid warhead to target the HDAC active site Zn²⁺, and introduced a lactam staple between l-isoaspartic acid and 2,3-diaminopropionic acid as substitutes for the common Asp-Lys pair, a strategy they previously reported and validated in the context of ERα targeting.³⁴³ Molecules derived from these stable histone variants demonstrated antitumor activity in PA-1 xenograft models with intraperitoneal injection. Intriguingly, this lactam-stapled peptide outperformed vorinostat, an FDA-approved small molecule HDAC inhibitor, and did not exhibit toxicity.³⁴² Other similar examples include lactam-stabilized helical fusion inhibitors for RSV.³⁴⁴

3.3. β-Hairpin Stabilization

A common motif that enables PPI formation is a β-hairpin or a β-turn, and like the MCPs designed via rational design discussed in Section 2.3.1, these motifs can be identified and cyclized to create β-hairpin peptides with programs such as LoopFinder.^80,81 One of the common approaches toward designing β-hairpin peptides begins with identifying the key binding residues or a binding epitope and then grafting it onto a hairpin scaffold, where the signature turn is usually formed by a d-Pro-l-Pro segment or other turn-inducing structure, and the antiparallel strand may have a variable sequence that can be optimized as needed. This step is not entirely modular, as the introduction of the rigid turn may lead to backbone conformations that are not favorable to binding.^345,346 The loop is often closed using one of the aforementioned crosslinking chemistries (Figure 2). The original binding epitope does not necessarily have to be a hairpin; for example, the key residues on the binding face of a helix can be strung together as one of the strands of the β-hairpin, leading to the same presentation of side chain interactions in 3D space as the original helix.

Hairpins generally rely on specific turn inducing templates that may include covalent cross-links to stabilize the structure.^347-350 While it is known that the classic hairpin can be formed by introduction of a d-Pro-l-Pro segment, the introduction of the rigid turn may lead to backbone conformations that are not favorable to binding, so it is often important to sample multiple alternatives for the turn inducing residues in order to result in the ideal conformation, followed by staple incorporation. In one example reported by Pace et al., 4-mercaptoproline was cross-linked to another cysteine via meta-dibromomethylxylene to form an antiparallel β-hairpin structure that proved to be more resistant to proteolytic degradation compared to its unmodified counter-part.³⁵¹ This peptide, 4MP-m was computationally designed, using a novel “staple-first” approach. After identifying a set of cross-links that were most compatible with the desired β-hairpin structure, they introduced the turn and grafted on the rest of the sequence. The resulting hairpin peptides exhibited a higher degree of ordered structure than the parent sequence. The crucial hydrogen bonding interactions should also be considered during the design process and may alleviate the need for a natural or synthetic β-turn altogether, as recently reported by Nazzaro et al. using a hydrogen bond surrogate (HBS) approach.³⁵² However, natural hydrogen bonds within the turn may provide stability to the structure, as was the case in a report of a ULM peptide, a spliceosome assembly inhibitor, which naturally has a hydrogen bond between Glu340 and Arg337 residue supported by the Ser336 residue forming a hydrogen bond with Trp338 to form the β-turn. Therefore, these residues were left intact, while an adjacent Lys side chain was used to lactamize the ring with Glu to form the complete hairpin.³⁵³ Another hairpin example by Wendt et al. targets β-catenin/E-cadherin using dithiol and increasingly longer thioether linkers to stabilize a hairpin derived from E-cadherin.³⁵⁴ The structure–activity investigation through a series of FP assays revealed that the longest linkers showed higher inhibition, but only in certain positions along the hairpin, suggesting that not all staple positions are equally conducive to hairpin stabilization and binding affinity. In another example for this target, Blosser et al. report a β-hairpin targeting β-catenin/TCF4, using a Gly-Pro turn template and closing the loop with a thioether linkage.³⁵⁰ Computational modeling was used to select a linear strand that mimics TCF4 while the opposite strand is present for stabilization of β-hairpin structure. Because β-catenin has Cys residues near the TCF4 binding interface, it was possible to make a covalent peptide inhibitor using electrophilic groups on a Dap residue. Strømgaard and colleagues used a similar d-Pro-Gly turn closed with native chemical ligation to generate an nNOS/PSD-95 inhibitor.³⁵⁵ A positional heat map scan revealed which residues were most conducive to binding, followed by additional studies to determine binding mechanism. The resulting constrained hairpin exhibited nanomolar binding affinity and may have the potential to outperform linear peptides that are already in clinical development for targeting this interaction.³⁵⁶

A key example of a β-turn motif mediating recognition is in the interaction between KEAP1 and NRF2.²¹⁷ Since the discovery of this DEETGE loop, many research groups have developed cyclized β-hairpins capable of interacting with KEAP1 with high affinity.^357-360 However, most designs suffer from a lack of activity in cells due to low uptake. A general approach to improve permeability and activity is to reduce the size of the peptide required to bind to the target site. Ortet et al. examined the minimum structural motif required for a cyclic β-hairpin peptide to achieve the same affinity as wild-type NRF2, which has low nanomolar dissociation constants.³⁶¹ Further studies suggested that the terminal residues only serve to stabilize a favorable conformation, but head-to-tail cyclization of a 7-mer, based on residues 76–82, resulted in a 10-fold decrease in binding. The addition of a β-amino acid, d-β-homoalanine, at residue 76 to expand the β-hairpin cycle, along with a E78P mutation that leads to better preorganization resulted in a lead compound 23, which closely mimics the conformation of the NRF2-derived peptide (Figure 24A,B). Similarly, Lopez et al. performed all-atom, explicit-solvent molecular dynamics simulations on the CDEETGEC sequence cyclized using various thioether-forming linker molecules for binding to KEAP1.³⁶² To improve cell uptake of the NRF2 cyclic β-hairpin, Legre et al. studied various factors, including appending a CPP, appending a fatty acid chain, shortening the peptide, and masking the net negative charge.³⁶³ They found that by shortening the peptide from an 11-mer to a 9-mer and attaching a fatty acid tag induced transcription of the antioxidant response element (ARE) with an EC₅₀ of 740 nM in BEAS-2B cells.

Figure 24.

β-turn stabilized peptides for targeting KEAP1. A) Chemical structure of compound 23. B) Crystal structure overlay of compound 23 (blue; PDB: 7K2S) and the natural loop of NRF2^76-84 (yellow; PDB: 2FLU) bound to KEAP1 (cyan).³⁶¹

Beyond proteins, stabilized β-hairpins can target numerous biomolecular surfaces including nucleic acids. Initial studies targeted the transactivation binding response (TAR) RNA, which is found in the 5′ untranslated region (UTR) of mRNA of viruses such as HIV and bovine immunodeficiency virus (BIV).³⁶⁴ This loop interacts with a β-hairpin located in the Tat protein, and this β-hairpin was first stabilized to bind BIV TAR RNA in 2001 by Tok et al.^365,366 Additional studies examined HIV TAR RNA and designed stabilized β-hairpin peptides that achieved subnanomolar binding affinity.^346,367-370 Another way to design stabilized β-hairpin peptides are to employ evolved RNA-recognition motifs (RRMs) in a yeast display library using U1A RRM, a protein that contain multiple β-sheets, as a starting point.^371,372 This approach has yielded many stabilized β-hairpin peptides against HIV TAR RNA, though the best β-hairpin peptides designed using this method has a dissociation constant of 800 nM, 3 orders of magnitude worse than that of Shortridge et al.^373-375 Stapled β-hairpins have also been developed for miRNAs.³⁷⁶ Shortridge et al. first reported a β-hairpin mimic for targeting pre-miR-21.³⁷⁷ After screening a panel of existing stabilized β-hairpins, they identified a compound, L50, that bound to pre-miR-21 with a dissociation constant of 200 nM, which is still the strongest binding observed by a stabilized β-hairpin against an miRNA (Figure 25A). The crystal structure of L50 binding to pre-miR-21 was solved, which shows that the arginine residues (Arg3, Arg5, Arg8, Arg12) bind to the anionic miRNA backbone, whereas Dpr13/Pro14, along with Gly6/Lys7, form the β-turns (Figure 25B,C). L50 was shown to inhibit miR-21 formation both in vitro and in cells. Other miRNAs, such as pre-miRNA-20b, have also been targeted with stabilized β-hairpins, though with relatively weak affinity.³⁷⁸ A deeper examination of stabilized β-hairpins binding RNAs, along with other stabilized structures binding RNAs, have been covered in this excellent review by Pal and ‘t Hart.³⁷⁹

Figure 25.

β-Hairpin stabilized peptides targeting RNA. A) Chemical structure and binding properties of L50. B) Crystal structure of L50 (blue) bound to pre-miR21 (gray). PDB: 5UZZ. C) Insert highlighting the key residues for interacting with RNA (Arg3, Arg5, Arg8, and Arg21) or forming the β-turn (Dpr13/Pro14, Gly6/Lys7).³⁷⁷

3.4. Stabilized Helical Secondary Structures Using Hydrocarbon Stapling

A significant breakthrough in helical stabilization came with the application of ring closing metathesis for peptide stabilization as it introduced a stable, nonpolar stabilizing cross-link that could improve cellular uptake and activity of stapled peptides. Blackwell and Grubbs first reported the use of the ring closing metathesis (RCM) reaction to stabilize peptide helices in 1998, where they incorporated serine or homoserine O-allyl ether residues (Figure 23C).³¹⁸ The sequence, even prior to substitution with the olefinic residues, showed helical character, likely due to the prevalence of amino acids with high propensity for helix formation. An X-ray crystal structure suggested a predominantly 3₁₀-helix conformation, due to the i, i+3 hydrogen bonding pattern and the torsional angles. While the helical character showed little change with substitution and metathesis, the bioorthogonality and hydrophobic nature of the cross-link was an attractive strategy to stabilize α-helical conformations.

Schafmeister et al. further investigated the potential of RCM for peptide stabilization by using disubstituted amino acids with both R and S configuration and various side chain length to determine which combinations would give the best yield as well as the most stabilizing effect on helix formation (Figure 26).³¹⁹ Within a specific model system, they discovered that for i, i+7 staple placement between R- and S-amino acids, at least a 9-carbon cross-link was required to observe appreciable conversion of an RNase A derived peptide to the stapled product. For i, i+4, using R_i, R_i+4 and S_i, S_i+4 α-methyl amino acids, 8 carbons was the minimum length required for near complete conversion. From this library they concluded that an i, i+7 staple with R,S paired configuration and an 11-carbon cross-link resulted in the greatest increase in helicity, followed by the R,S paired i, i+7 staple with 12-carbon cross-link and S,S paired i, i+4 staple with 8-carbon cross-link.³⁸⁰ The proteolytic stability of these peptides was assessed using trypsin, and it was observed that the unstapled, disubstituted amino acids alone conferred some resistance, reducing cleavage rates 5-fold, which increased to 41-fold after metathesis.³¹⁹ The metathesis reaction can result in a mixture of cis and trans isomers that will have differential effects on peptide structure and biological activity,³⁸¹ though the general rules that govern this are not well understood.

Figure 26.

Design of bis-alkylated, hydrocarbon stapled peptides. A) Chain length, staple position, and stereochemistry are key considerations for generating helical hydrocarbon stapled peptides. B) Schematic of peptide functionalization using ring closing metathesis.

The initial protein family that was targeted using hydrocarbon-stapled peptides was the BCL-2 family, known for its central role in regulating apoptosis and ensuing importance in cancer.^197-200 Within this family, the pro-apoptotic members contain a conserved BH3 α-helical domain required for binding to other BCL-2 members and subsequent initiation of the apoptotic cascade. This BH3 domain served as the template for the development of stapled peptides targeting various proteins in this family.^382-388 In 2004, Yang et al. reported a lactam stapled peptide binder derived from BAK, a BCL-2 family member, and Walensky et al. described the development of a hydrocarbon-stapled peptide derived from the BH3 helix of BID, another member of the BCL-2 family, that bound BCL-X_L and has potential to broadly interact with other BCL-2 family members (Figure 27A).^389-391 In the case of the BAK-derived peptide, the i, i+4 lactam staple provided moderate helix stabilization, up to 56% in PBS.³⁹⁰ For the peptide derived from BID, a panel of analogs with varying positions of the i, i+4 olefin staple, including one with an i, i+7 staple, were synthesized and characterized for their helical characteristics, proteolytic stability, and binding affinity for BCL-2. All showed an improvement in helicity compared to the linear wild-type sequence. A lead compound, SAHB_A with 87% helical content, exhibited significant improvements in serum half-life and showed evidence of triggering cytochrome c release for apoptosis in vitro. Additionally, it was shown to be cell-permeable via endocytosis to a much greater degree than its linear counterpart. Importantly, SAHB_A suppressed growth of leukemia xenografts in vivo over the course of 7 days of treatment following IP administration, demonstrating for the first time that a hydrocarbon stapled peptide could be a viable therapeutic modality.³⁸⁴ Although the family wide BH3 binding motif is generally conserved among family members, selectivity can be attained by careful selection of the nonconserved residues. For example, a selective MCL-1 inhibitor based on the MCL-1 BH3 domain was designed by considering the unique residues present in MCL-1 and determining which mutations would abolish binding to other BCL-2 members. The MCL-1 BH3 domain contains a hydrophobic triad consisting of Leu213, Val216, and Val220, which are all crucial for binding to MCL-1 itself (Figure 27A-C). Treatment of cells with an MCL-1 designed SAHB in conjunction with other death receptor agonists showed that there was sensitization to the apoptotic cascade. The significance of this lies in the fact that MCL-1 overexpression is correlated to resistance against current drugs like Venetoclax that target the BCL-2 proteins, thus making selective and potent anti-MCL-1 inhibitors highly desirable. Following these initial studies, several groups reported BIM-derived sequences that also had selective binding for MCL-1, and Araghi et al. further modified these with i, i+4 and i, i+7 RCM staples to determine optimal positioning. Two of their compounds, SAH-MS1-14 and SAH-MS1-18 had binding affinities of 80 and 25 nM respectively. They verified that the lead compounds were not membrane lytic before proceeding to cell studies where they showed that treatment with SAH-MS1 peptides was only cytotoxic for MCL-1-dependent lines, despite the observation that SAH-MS1-14 also activated BAK to a certain extent.³⁸⁸ These stapled peptides have been applied for structural studies of the BCL-2 proteins as well. A BID SAH peptide binding to BAK was used as a tool for studying its activation mechanism, including NMR domain mapping and photo-cross-linking mass spectrometry.^386,392

Figure 27.

Stapled peptides derived from BH3 domains. A) Primary sequences of peptides derived from BAK, BID, and MCL-1 showing staple positions in designer peptides. B) Representative helix from MCL-1 SAHB_D. Hydrocarbon staple is highlighted in yellow, and key interacting residues are shown. C) Crystal structure of SAHB_D bound to MCL-1 with interacting residues labeled. PDB 3MK8.³⁸⁴

It must be noted that simply introducing the RCM staple does not always lead to the desired properties. Across the panel of stapled peptides synthesized by Walensky and colleagues, some only had a slight increase in helical character, and not all showed good cell permeability.³⁸⁵ Monte Carlo folding simulations suggest the existence of “decoy” states, where the peptide is partially folded or bent in a manner inconsistent with its natural structure in the binding domain.³⁹³ This also explains why i, i+7 staples are only sometimes more beneficial than the i, i+4 for helix formation: the effects of the hydrocarbon staple are highly context dependent. The tradeoff between the residues being mutated out and the new olefin side chains must be considered with regards to its effect on overall charge, polarity, and the potential loss of existing helixstabilizing interactions. Because of the need for extensive structure optimization with RCM stapled peptides, there have been many studies with hydrocarbon peptides that attempt to compile a general list of guidelines that would shorten lead time toward successful compounds.³²⁹ Bird et al. report their findings from a library of BH3 hydrocarbon stapled peptides, which included several observations that the peptides that exhibit high cell uptake generally contain properties in common such as (i) the hydrophobic staple is located along the border of the hydrophilic and hydrophobic face of the helix, (ii) overall high degree of hydrophobic character between 61 and 86% and (iii) pI between 8.8 and 9.34. Earlier, Chu and Moellering et al. used a quantitative, high-throughput imaging approach to assess the productive permeability and distribution of approximately 200 stapled peptides from diverse target proteins.⁵⁶ This study showed that peptides with net positive charge, though not hypercharged like many CPPs (e.g., Tat, Penetratin, and octa-arginine), were the most productively cell permeable and displayed distribution throughout cells. This was staple and context dependent and showed that much more penetrant and potent stapled CPPs could be developed, but that introducing staples alone was not always sufficient to improve cell uptake. Of course, the conclusions from these studies provided some general guidelines for active cellular uptake, but there are other design principles that must be balanced or even orthogonal to attain passive permeability.

Another example of a hydrocarbon stabilized peptide targeting a PPI in an undruggable protein involves targeting the NOTCH/CSL transcriptional complex. The Notch pathway involves activation of NOTCH transmembrane receptors (NOTCH1–4) that are cleaved in response to ligand binding, releasing its intracellular domain ICN1 that binds the transcription factor CSL and subsequently coactivator MAML, promoting downstream gene expression.^394-396 Domain mapping studies identified an extended stretch of the MAML1 N-terminus that acted in a dominant negative fashion to block MAML1 binding and transactivation by ICN/CSL complexes.³⁹⁷ Structural studies by Nam et al. showed that this dominant negative stretch formed an extended α-helix that bound to the interfacial groove formed by ICN1/CSL³⁹⁸ (Figure 28). Moellering et al. designed a series of stapled peptides derived from various segments of dnMAML1 and utilized hydrocarbon staples with i, i+4 spacing to stabilize the α-helical conformation.³⁹⁶ While all stapled peptides showed an increase in helicity, not all were cell permeable or strong binders of the ICN1/CSL dimer. A lead compound, SAHM1, showed significant inhibition of NOTCH1-dependent gene expression in T-cell acute lymphoblastic leukemia (T-ALL) cells, which was further established by an early example of using Gene Set Enrichment Analysis (GSEA) for specificity analysis. Furthermore, SAHM1 inhibited NOTCH1-dependent T-ALL cell lines, but not other lines with known resistance to NOTCH inhibition. Finally, SAHM1 demonstrated inhibition of systemic T-ALL xenografts in mice, further suggesting that stapled peptides could be viable in vivo regulators of TF activity and that they could be developed to target protein interfaces.³⁹⁶

Figure 28.

Design of stapled peptide NOTCH inhibitors. A) Crystal structure of the NOTCH1 ternary complex: CSL (gray), dnMAML1 (cyan), ICN1 (purple), MAML1 interacting domain (blue). PDB: 2F8X. B) Model of stapled peptide SAHM1 (blue) overlaid on CSL (gray) and dnMAML1 (cyan). Staple is shown in yellow. Underneath is a representation of the primary sequences of designed SAHM peptides against dnMAML1 and the primary sequence of SAHM1 showing the staple position.³⁹⁶

Another notable undruggable transcriptional regulatory pathway targeted by hydrocarbon stapled peptides is the WNT/β-catenin pathway. β-catenin is a central mediator of transcriptional regulation downstream of WNT signaling and does so via interactions with several proteins like BCL9, Axin and TCF transcription factors. Grossmann et al. directly targeted β-catenin using a series of Axin-derived peptides stabilized by hydrocarbon staples in both i, i+4 and i, i+7 positions and evaluated their helicity and binding affinity with CD spectroscopy and FP assays, respectively.³⁹⁹ A unique aspect of this effort was the integration of phage display screening to identify privileged consensus sequences to serve as starting points or optimizations of the WNT-Axin-derived sequences (Figure 29A). Improvements were made to initial stapled peptide leads by increasing the overall positive charge of the peptide by mutating nonessential positions of Asp or Glu and introducing Arg. Taking the staple design that worked best from the first rationally designed set of peptides, they stabilized the phage display-derived peptides (StAx peptides; Figure 29A) and showed cellular uptake by confocal fluorescence imaging. Several StAx peptides, including a lead StAx-35, inhibited TCF4-driven and endogenous target genes in colorectal cancer cell lines; however, these effects were not mediated by changes in β-catenin expression. This study also solved a cocrystal structure of StAx-35 peptide bound to β-catenin, highlighting conserved register within the Axin-targeted groove as well as new contacts introduced by phage-derived mutants (Figure 29B). Beyond Axin, several groups have reported stapled peptides derived from the BCL9 protein, which binds a unique site on the β-catenin armadillo repeat domain.⁴⁰⁰

Figure 29.

Hydrocarbon stapled peptide inhibitors of β-catenin selected from phage display. A) Schematic of phage display to select for consensus sequences of peptide ligands that are later stabilized by stapling. B) Crystal structure of StAx-35 peptide (blue) bound to β-catenin (cyan). Inset shows key interacting residues and staple (yellow).³⁹⁹

Perhaps the most studied “undruggable” target for which stapled peptides have been developed is tumor suppressor protein p53 and its negative regulators.²³ As one of the most mutated proteins in cancer and therefore desirable targets in oncology, many approaches have been taken to prevent suppression of p53 protein levels and transactivation by oncoproteins like the ubiquitin ligases MDM2 and MDMX. In 2007, Bernal et al. reported the development of SAH-p53-8, an α-helical peptide stabilized using RCM, derived from the MDM2 binding region of p53.⁴⁰¹ Since the initial designs were not cell permeable, they mutated negatively charged residues and specific residues known to contribute to ubiquitinylation or nuclear export, namely K24 and L14. The lead compounds were shown to bind MDM2 with high affinity and reactivate p53 expression. Three years later, it was shown that SAH-p53-8 was a dual inhibitor of MDM2 and MDMX, overcoming the limitation of Nutlin-3, a small molecule p53 activator that only targeted MDM2, and were therefore ineffective in MDMX overexpressing cancer models.⁴⁰² The emergence of crystal structures and phage display libraries yielded other potential peptide sequences that could disrupt binding of p53 and MDM2, which could then be structurally stabilized using a variety of chemistries.^403-406 Brown et al. used the crystal structure of p53 bound to MDM2 as well as the phage-derived sequence, PMI, reported by Pazgier et al. to design stapled peptides with the help of computational simulations to guide further mutations.^405,407 Their lead compound, sMTide-02, later called PM2, showed p53-activation in T22 reporter assays, outperforming SAH-p53-8 in the presence of serum. It also showed good potency in viability assays benchmarked against both Nutlin and SAH-p53-8.⁴⁰⁷ In mouse models, PM2 was tested in combination with radiotherapy and it was discovered that this regimen prolonged survival of HCT116 tumor bearing mice in a synergistic manner without adverse effects observed in treatment with small molecule MDM2 antagonists.⁴⁰⁸

Another phage display-derived peptide, pDI, became the core of perhaps one of the most successful stapled peptides and the first one to reach clinical trials, sulanemadlin.^404,409 While initial attempts were built around the SAH-p53-8 sequence, separate introduction of i, i+7 staples into the pDI sequence obtained from phage display by Hu et al. afforded significant increases in binding affinity. Notably, pDI preserved the Trp/Leu/Phe core binding residues but nearly every other residue was different from the native p53 sequence, which shows the advantage of peptide libraries in offering unexpected solutions that are often overlooked during rational design. Additionally, solubility and cell permeability were improved with the introduction of noncanonical amino acids and C-terminal extensions. The lead compound, termed ATSP-7041, was able to suppress tumor growth with a daily 15 mg/kg dose in SJSA-1 xenograft models, which has high expression of MDM2. Additional pharmacokinetic studies in mice, rats, and monkeys all showed a relatively low rate of clearance.⁴¹⁰ Before moving to clinical trials, some non-natural amino acids were reverted to the natural residue as a practical consideration for larger scale synthesis and a polyalanine tail was appended to the C-terminus for extension of the α-helix. This modification improved solubility as well, which was one of the disadvantages of the precursor ATSP-7041. The new and improved ALRN-6924, also called sulanemadlin, was found to be metabolized in vivo, via proteolytic cleavage of the polyalanine tail (Figure 30).^409,411 However, this metabolite was a 10-fold more potent binder of MDM2 and MDMX but was less cell permeable when used directly in cell proliferation assays leading to lower efficacy. Thus, ALRN-6924 functions as a cell permeable precursor drug that is metabolized to a more potent form in situ. ALRN-6924 was shown to have chemoprotective effects at low doses by inhibiting DNA synthesis and reduced cell viability at higher doses. In numerous clinical trials it also had favorable clearance rates and safety profile.⁴⁰⁹ Ultimately, the chemoprotective effects and prevention of cancer progression was not statistically significant and consistent efficacy was not observed, leading to the end of the sulanemadlin development. The potential benefits of stapled peptides, namely, the excellent safety profile and stability compared to larger biologics, were clearly demonstrated and serve as a promising first attempt for stapled peptide translation to patients. Other interesting attempts to protect p53 have emerged, such as all-D stapled peptides, where the original sequence was derived from the PMI sequence,⁴¹² and single, double, and stitched RCM staples were introduced in various positions and assessed for helicity, binding affinity, and p53 activation, with some of the analogs achieving comparable results to ATSP-7041, validating the “left-handed” approach to creating selective and highly stable peptide therapeutics.⁴¹³

Figure 30.

Hydrocarbon stapled peptide inhibitors of p53/MDM2. ALRN-6924 behaves as a cell permeable prodrug that is metabolized to more potent ALRN-8714 in cells. Crystal structure of ALRN-6924 in complex with MDM2. Inset shows key residues F19, W23, and L26, as well as staple (yellow) forming hydrophobic contacts with MDM2 surface. PDB: 8GJS.⁴⁰⁹

Initially developed by Arora, ring closing metathesis has also been used as a hydrogen bond surrogate (HBS) to stabilize α-helical peptides in many notable examples.^414-417 While technically distinct from side-chain stabilization, the HBS approach uses orthogonal chemistry to mimic the N-terminal hydrogen bonding pattern and nucleate α-helical structure. As such, it is powerful alone or in combination with other layered stabilization chemistries (discussed further below). For example, Arora and colleagues designed a stabilized helical peptide that mimics the p300/CBP coactivator of HIF1α with the purpose of inhibiting HIF1α driven transcription.^415,416 The construct was designed to disrupt the HIF1α-p300-CH1 interaction and subsequently downregulate hypoxia driven gene expression. It was stabilized using the hydrogen bond surrogate strategy, cross-linking a terminal 4-penetenoic acid to an N-allylalanine residue two amino acids away via RCM.⁴¹⁶ The p300/CBP coactivator regulates other transcription factors as well, so future work will likely involve building in specificity for a narrower mechanism of action. In one example against HIF1α, it was shown that 1 μM of the constrained peptide derived from CH1 domain of p300 could decrease VEGFA levels comparably to 200 nM chetomin, a molecule previously shown to antagonize HIF1α activity. Interestingly, the peptide was not cell permeable until the addition of an Arg residue on the C-terminus, which both improved helicity and permeability, presumably because of a potential salt bridge between it and a Glu i+4 residues away that further aided helix folding.⁴¹⁵

In a separate study targeting Ras, Patgiri et al. designed an α helical mimic of the interacting αH motif using an olefin metathesis linked HBS between i, i+4 residues on the N-terminus.⁴¹⁸ Nonessential residues were mutated to optimize solubility and improve helix propensity. The resulting compound had only moderate helicity but was shown to slow the Ras nucleotide exchange of GDP to GTP in vitro. NMR titration studies showed shifts for residues known to be part of the binding interaction and fluorescence microscopy confirmed cellular uptake. They also found that addition of the constrained peptide to HeLa cells attenuated ERK phosphorylation, which is triggered by Ras activation.⁴¹⁸ The activation of Ras is mediated by the SOS protein, which interacts with Ras via an α-helical hairpin motif. Another stapled peptide targeting the Ras-SOS interaction using the same helix template was reported by Leshchiner et al., also using olefin metathesis but through side chain cross-linking rather than the HBS approach.⁴¹⁹ They showed that their lead compound SAH-SOS1a, which had a central i, i+4 hydrocarbon staple, was able to bind both wild-type and mutant K-Ras, in both the GDP-loaded and GTP-loaded conformations. When incubated with K-Ras concomitantly with a fluorescent GTP analog, SAH-SOS1a inhibited GTP binding. They also showed that treatment with SAH-SOS1a reduced phosphorylation of downstream proteins both in cells and in vivo. These examples show the potential of targeting this PPI for broad applicability to the various mutant K-Ras variants that make it so difficult to drug by traditional means.

Beyond Ras family proteins, small GTPases represent a large, challenging family of proteins that are commonly considered undruggable. The first application of stabilized peptide approaches to a RAB-family small GTPase, RAB8a, was derived from the R6IP protein. An early inhibitor, StRIP3, was developed and bound RAB8a with a K_d of approximately 0.03 mM.^420,421 It was not proteolytically stable or cell permeable in its linear form, so an alanine scan was performed to assess which positions would be suitable for mutation to stapling residues. Additionally, an arginine scan was completed to identify compatible positions to introduce positively charged residues. While nearly all positions were sensitive to the mutations, the least perturbing positions were chosen, and a disubstituted Aib residue was strategically introduced in place of one of the key Leu residues, which improved affinity. After incorporation of single and double i, i+4 and i, i+7 staples, the peptides were again evaluated for affinity and proteolytic stability. StRIP14 containing two i, i+4 staples on the same helix face and an overall net negative charge of −4 was the most stable, but in order to reduce the negative charge, some Asp and Glu residues were mutated to Asn and Gln, respectively. The resultant StRIP16 still had an overall −2 charge but was able to colocalize with Rab8a in cells and demonstrated cell uptake comparable to the cell penetrating Tat peptide. In a separate study, Mitra et al., used hydrocarbon stapling to develop inhibitors derived from FIP-family adaptor proteins to target the constitutively active endosomal sorting proteins RAB11 and RAB25.⁴²² The latter protein, RAB25, is implicated as both an oncogene and tumor suppressor in specific tumor contexts like ovarian and breast cancers.^423,424 The FIP interacting domain regulates GTPase activity and binds as an extended α-helix-turn-3₁₀ helix pair, requiring a unique approach to design stabilized antagonists with desirable properties. A panel of FIPα-helix-derived peptides were synthesized, scanned for ideal staple positions, then further optimized with residue mutations to remove anionic residues for better cell uptake, thermal stability and oxidative stability.⁴²² In addition to stable helical character, the peptides showed pronounced phenotypic effects on RAB25-driven processes such as cell migration that were context specific to breast and ovarian tumor models, mimicking the effect of RAB25 overexpression or knockdown in these cellular systems. For example, RAB25 overexpression increases migration in MCF7 but decreases it in MDA-MD-231. The treatment of RFP14 peptide, which was the most RAB25 specific, led to the reversal of the respective RAB25 phenotypic effects in both cell lines.⁴²²

Beyond many of the classically undruggable protein classes discussed above, notable studies have developed stabilized peptides for many other challenging targets and interfaces. These include cell surface receptors,⁴²⁵ cytoskeletal proteins,^426,427 chromatin modifying enzymes,⁴²⁸ viral proteins^429-431 and even nonprotein targets like RNA and DNA.⁴³² A notable example of stapled peptide development for viral targets include the development of NYAD-1, which was based on a previously identified peptide sequence that is known to interfere with HIV capsid particle assembly but is not a viable treatment due to lack of cell permeability.⁴²⁹ Using an i, i+4 hydrocarbon staple with S₅ residues, Zhang et al. showed that constrained peptides were more α-helical and cell permeable and able to bind to the capsid protein but not able to prevent HIV-1 entry altogether.⁴²⁹ In follow-up studies, they tested i, i+7 variants of NYAD-1, which demonstrated some impaired HIV-1 infectivity in treated cells.^430,431 Another notable study by Walensky and colleagues used stapled peptides to help stabilize target epitopes from HIV for the development of adaptive immunity. This represents a novel example of using structural stabilization as a viral vaccine.⁴³³

Epigenetic enzymes, such as EZH2, have also served as targets for stapled peptide development. Dysregulation and mutation of EZH2, a histone methyltransferase, has been linked to the proliferation and metastasis of various cancers.⁴³⁴ While methyltransferases can be targeted with small molecules, selectivity is a concern in targeting, for example, the SAM-binding active site. Kim et al. reported a selective peptide inhibitor of the EZH2/EED complex that utilized hydrocarbon staples in an i, i+4 pattern.⁴²⁸ This sequence was derived from the EED-binding domain of EZH2, which adopts an α-helical conformation. The staples were strategically placed away from the interacting helix face, and different sequence lengths were assessed in vitro for EED binding. It was determined that an IQPV motif at the C-terminal end of the peptide was crucial for binding. The lead 27-residue peptide was further optimized for cellular uptake by mutating out negatively charged residues that were not expected to be crucial for binding to EED.⁴²⁸

Protein kinases are a large family of proteins that have been extensively studied for small molecule drug development, especially in oncology. While active-site directed small molecules have been enormously successful drugs for a subset of kinases, the ability to create selective inhibitors remains a challenge. Targeting allosteric sites could offer another path for selective modulation, and several stabilized peptides have been developed to regulate protein kinase effector binding as well as the conformational state. For example, Schepartz and colleagues developed hydrocarbon stapled peptides that allosterically inhibit EGFR kinase function by preventing the coiled coil interaction that occurs upon ligand binding.^425,435,436 The sequence was derived from the juxtamembrane (JM) segment that forms the coiled coil dimer and stapled in various positions opposite the binding interface of the helix. They identified multiple sequences that decreased cell viability in EGFR expressing cell lines, the most potent being peptide E1^S with a central i, i+7 hydrocarbon staple.⁴²⁵ Interestingly, this peptide exhibited potent bioactivity in its unstapled form close to that of the stapled E1^S. The unstapled peptide had greater helicity than the wild-type sequence but less than its stapled counterpart, which is not entirely unprecedented due to the previous studies of α,α-disubstituted amino acids promoting helix formation.⁴³⁵ Further structural studies revealed that the peptide functioned by interacting with both the JM segment and C-terminal tail of the kinase domain.⁴³⁶

In another example, hydrocarbon-stapled peptides were developed to target effector interactions with protein kinase A (PKA) and protein kinase C (PKC). A-kinase anchoring proteins (AKAPs) are a family of proteins that act as a scaffold to spatiotemporally control PKA activity.⁴³⁷ Disrupting the AKAP-PKA interaction in an isoform-selective manner has long been seen as a viable therapeutic strategy, with early peptides derived from the α-helical domain of AKAP that either specifically prevent AKAP interaction with PKA-R1⁴³⁸ or PKA-R2; however, these native peptides had poor pharmacologic properties.^439,440 Kennedy and co-workers designed hydrocarbon-stapled peptides by incorporating S₅ residues at i, i+4 positions in sequences from the α-helical AKB domain in different AKAP proteins.⁴⁴¹ The best hit, STAD-2, targeted PKA-RIIα with 40-fold selectivity over PKA-RIα and inhibited phosphorylation of PKA targets in cells. STAD-2 decreased parasitemia in malaria-infected red blood cells via cell lysis, with single-digit micromolar IC₅₀ efficacy.⁴⁴² Subsequent work demonstrated improved cellular uptake⁴⁴³ as well as design of compounds targeting AKAP-PKC interaction.⁴⁴⁴

The examples described showcase the immense potential of hydrocarbon stapling for stabilizing helical secondary structures and developing chemical probes and therapeutics for undruggable targets. Several aspects have to be considered to successfully incorporate hydrocarbon staples into a peptide as many designs are context and target specific. These include peptide sequence as well as staple linker length, stereochemistry, and position along the structure to impart desired properties. It is often necessary to sample the sequence space of the progenitor protein or peptide structure to discover the most active ligands as observed for the reported NOTCH inhibitors.³⁹⁶ It is also important to scan the staple placement across the sequence to identify the optimal stapling positions while ensuring the orientation of key interacting residues remains intact. This can greatly affect binding affinity and specificity to the target of interest and help reduce off-target effects, as well as generate highly specific mimetics capable of distinguishing closely related targets, for example the BCL-2 family. Future strategies may include library generation of hydrocarbon stapled peptides with the emergence of compatible chemistries⁴⁴⁵ that will become powerful strategies for screening against numerous challenging targets.

3.5. Thiol-Mediated Staples

While RCM chemistry has proven incredibly versatile and useful for the stabilization of α-helices, the requirement of custom, potentially toxic catalysts and relatively expensive amino acids could be a practical hindrance for scaling some probes and therapeutics. The thioether linkage is also widely used, especially in the macrocycle space, but its utility for secondary structure stabilization has been proven as well (Figure 31). It avoids the labile nature of the disulfide linkage, and can utilize natural cysteines, or cysteine analogs. Two of the common reactions are Michael addition to an α,β-unsaturated carbonyl⁴⁴⁶ or an S_N2 reaction with an α-halo acetamide.⁴⁴⁷ These electrophilic groups can be incorporated as non-natural amino acid side chains or as substrates that are added to dithiol peptides for a two-component staple. Jo et al. reported the use of thiol bis-alkylation to stabilize an α-helical calpain inhibitor.⁴⁴⁸ They screened several potential electrophilic cross-linking reagents to determine which would lead to the lowest energy conformations of the peptide. Benzyl halides performed the best, with dibromo-m-xylene yielding the highest conversion with the fewest side products and the greatest improvement in helicity, suggesting that yield and helical propensity are correlated, as in many other examples in the literature. They proceeded to apply this chemistry toward a calpastatin fragment sequence and showed dose-dependent inhibitory activity of the peptide.⁴⁴⁸ Dibromo-substituted xylenes remain a common choice for cysteine cross-linking, especially for use with phage display^449,450 and mRNA display libraries as demonstrated by Hartman and co-workers.⁴⁵¹ In this approach, they hypothesized that one reason library-generated peptides stapled with thiol-based linkers exhibit lower affinity than their hydrocarbon stapled counterparts was due to the loss of the stabilizing effect provided by the α,α-disubstituted amino acid. Thus, they synthesized a series of peptides using either hydrocarbon staples or thiol-alkylated staples via an α-methyl cysteine to examine the effects. The results showed that incorporating the disubstituted α-methyl cysteine at both ends of the staple with a dibromoxylene linker yielded a peptide with affinity close to that of the hydrocarbon stapled version. They then confirmed that this modified cysteine was able to be incorporated into an mRNA display library, which opens the possibility of more diverse libraries with potential to discover higher affinity hits. In addition, the diversity of bromo-substituted linkers allows for easy sampling of various peptide conformations. Sometimes this results in lead compounds with unusual staple patterns, such as the i, i+3 staple in the DD5-o autophagic activator peptide reported by Peraro et al., among other examples.^452,453 The less common spacing for helical stabilization may be uniquely compatible with d-amino acids.^339,452,454

Figure 31.

Representative example of the synthesis of bis-alkylated thioether stapled peptides using common building blocks.

A relatively new class of stabilized α-helices that utilize thioether stapling chemistry are “helicons,” which were originally reported by Li et al.⁴⁵⁵ They are generated and selected through phage display, with cysteines incorporated at i, i+7 positions in the helix (Figure 32A). The α-helix is stabilized with the addition of N,N’-(1,4-phenylene)bis(2-bromoacetamide). In this first report, a stabilized phage library was panned against β-catenin to both validate prior known α-helix binding sites as well as discover de novo sites. Through phage display selection, they discovered that helicons were able to recapitulate native α-helix interactions such as those known with Axin. An example from the library is Helicon FP01567, which binds in the same location as the α-helix Axin^466-482 and interfered with the β-catenin-Axin interaction with low micromolar affinity (Figure 32B). In addition, novel α-helical binding sites have been discovered that interferes with the ICAT binding site. In a follow-up study, Tokareva et al. developed a method to study helicon trimerizers, where the helicon facilitates interactions between two different proteins.⁴⁵⁶ They were able to form a novel PPI between E3 ligase CHIP and enzyme PPIA, CHIP and transcription factor TEAD4, and E3 ligase MDM2 and transcriptional coactivator β-catenin. To do so, they first screen for helicons that bind the E3 ligase, before performing a second phage display screen that is biased for residues that bind the relevant E3 ligase but retains flexibility in other residues. Only helicons that require both the E3 ligase and the relevant protein to bind were synthesized. An example is Helicon H330 (Figure 32B), which captured critical interactions H330-D18 with β-catenin H578 and R612, H330-D19 with β-catenin R582 and MDM2 H96, and the interaction between MDM2 V109 and β-catenin K433, collectively stabilizing the ternary complex.⁴⁵⁶ As the helicon technology has shown, phage display can select for stabilized α-helical peptides that can recognize novel sites on undruggable targets.

Figure 32.

Helicons selected by phage display for targeting diverse POIs. A) Schematic of phage display selection for Helicons targeting β-catenin. After translation and prior to affinity selection, displayed peptides are stapled between separate cysteine residues using a bifunctional alkyl bromide linker. B) Middle: Crystal structures of Helicon H330 (PDB: 8EIC) and Helicon FP01567 (PDB: 7UWI) bound to β-catenin (cyan) and MDM2 (gray). Left insert: Cooperative interactions promoted by Helicon H330 (blue) to the interface of MDM2 (gray) and β-catenin (cyan). Right insert: Overlay of FP01567 (blue) and Axin^466-482 (pink; PDB: 1QZ7) bound to β-catenin (cyan). Staples are shown in yellow.⁴⁵⁶

Another demonstration of dithiol stapling includes a study by Brown et al. with inhibitors of GABARAP. Here, they developed novel peptides that have unnatural modes of inhibition arising from a sampling of diverse structures, most commonly compatible with thiol mediated cross-links, rather than starting with a template derived from natural ligands and subsequently making modifications. The K1 peptide was found to be an inhibitor of GABARAP through phage display screening.⁴⁵⁷ It binds GABARAP and LC3 as a single 3₁₀-helical turn with high affinity, although the mode of binding is entirely different than natural interactors of GABARAP. In order to improve the pharmacokinetics of K1, various thiolreactive cross-linkers were tested with different spacing.⁴⁵⁸ The i, i+5 staples proved the most effective in improving binding affinity for GABARAP, and introducing penicillamine further enhanced affinity and specificity, yielding two peptides with completely different binding modes for GABARAP as revealed by crystal structures (Figure 33A-C). Pen3 mimicked K1, while Pen8, having the penicillamine on the opposite end of the peptide, more closely mirrored the natural ligands. In both cases, the staple itself was also involved in hydrophobic interactions between the peptide and GABARAP. Swapping the penicillamine for 4-mercaptoproline yielded a higher affinity peptide with better biological stability and generally improved membrane permeability.

Figure 33.

Stapled peptides targeting GABARAP. A) Primary sequence and staple locations of Pen3-ortho (left) and Pen8-ortho (right). Peptides were stapled between cysteine and penicillamine (denoted “X”) residues using ortho-dimethylbenzene linker. B) Crystal structure of Pen3-ortho (blue) bound to GABARAP (cyan). PDB: 7ZKR. C) Crystal structure of Pen8-ortho (purple) bound to GABARAP (cyan). PDB: 7ZL7. Interacting residues are labeled.⁴⁵⁸

Other variants of the thiol-based staples have been reported, such as the thiol–ene reactions and ketone-mediated modifications.^459-461 One interesting inversion of this chemistry involves incorporating a fluoroacetamide onto the peptide and using a benzyl thiol as the linker for proximity driven fluoride-thiol displacement, which can be done in an aqueous solution on unprotected peptides.⁴⁶² Another method that is compatible with unprotected peptides is perfluoroarylation.⁴⁶³ Jimenez-Macias et al. used this strategy to stabilize a CPP for cisplatin delivery to brain tumors.⁴⁶⁴ Additionally, it has been shown that introducing chiral centers into the staple linker leads to different effects on the helicity and thus, the cell permeability and stability of the peptides.^465-468 In a proof-of-concept study using thiol–ene chemistry to target ERα and MDM2, the authors introduced a chiral center on the alkene residue side chain and unexpectedly observed different structural and activity properties from unique diastereomers.⁴⁶⁵ A similar example was demonstrated by Speltz et al., where a γ-methyl substituent was incorporated into the hydrocarbon staple to mimic the branched side chain of an Ile residue involved in the native ligand interaction between SRC2 and ERα.⁴⁶⁹ This study demonstrated that the chiral staples slightly destabilized helix propensity, but bound ERα with higher affinity than the achiral hydrocarbon staple. These examples demonstrate ways in which additional functionality can be built into staple chemistries, furthering their potential for improving the pharmacological properties of peptide therapeutics.

A structurally interesting example of a RbAp48 inhibitor was reported by Hart et al., where a central trisalkylated linker templates a short one and a half turn helix, followed by a proline turn and an extended loop (Figure 34).⁴⁷⁰ The sequence was derived from MTA1, a scaffold protein that recruits RbAp48 into a histone modifying complex. Despite truncating two residues that were previously reported to have hydrophobic interactions with RbAp48, the weakened binding affinity was overcome by rigidifying the peptide using various thioether linkers, notably, 1,3,5-tris(bromomethyl)benzene, to form a bicycle peptide that formed a stabilized α-helix with two of the thioether links with i, i+4 spacing and a large constrained loop, resulting in compound 33, with a K_d value of 8.56 nM.⁴⁷⁰ Another recent example reported by Tennet et al. showcased the utility of thioether-mediated staples by stabilizing a less common polyproline II helix conformation.⁴⁵³ The highest affinity compound targeting the EphB2 receptor was stabilized via a 2,7-dimethylnaphthyl linker bridging two cysteines with i, i+3 spacing. However, trypsin digestion assays suggested the hydrophobicity of the linker increased susceptibility to degradation compared to lactam stapled analogs. This highlights the fact that potency may not always correlate with favorable pharmacologic properties. The use of thiol-based staples and others to stabilize structurally unique conformations such as constrained helix-loops or polyproline II helices opens new avenues for designer peptides beyond highly abundant α-helix or β-structures.⁴⁷¹

Figure 34.

Peptide stapling using trifunctional linker to generate bicyclic Peptide 33 to form a stabilized α-helix and constrained loop to increase affinity to RbAp48. Crystal structure of Peptide 33 bound to RbAp48 is shown. PDB: 6ZRD.⁴⁷⁰

3.6. Azide–Alkyne Click Reaction Staples

Cycloaddition click reactions such as CuAAC or RuAAC have the advantage of being bioorthogonal with fast reaction kinetics. RuAAC is not limited to terminal alkynes as in CuAAC and can include a variety of substituted alkynes (Figure 35A).⁴⁷² Ingale et al. reported a method for the incorporation of intramolecular CuAAC on-resin to stabilize α-helical peptides and used an HIV-1 gp41-derived sequence for proof-of-concept studies.³²⁰ In an ELISA against HIV-1 neutralizing antibodies, the peptides showed moderate binding. Strain-promoted cycloadditions (SPAAC) have been utilized for azide–alkyne cross-links as well, and the lack of a required metal catalyst enables these reactions to be biocompatible. Lau et al., demonstrated this advantage by reporting the in situ stapling and selection of peptides in cell culture.⁴⁷³ Using a p53-derived peptide functionalized with two azide containing residues, they treated p53 reporter cells and cotreated with the strained diyne that enabled SPAAC stapling. Reporter assays showed a dose-responsive increase in p53 activation for the stapled peptides (cotreated with diyne) while the same peptides without addition of the diyne showed no activation even at the highest dose. A crystal structure of the lead compound E1 in complex with MDM2 showed that the cycloadduct was part of the binding interface, possibly contributing to the interaction (Figure 35B).⁴⁷³ In a separate study, Kawamoto and co-workers identified two solvent-exposed residues on the β-catenin binding helix of BCL9 from the crystal structure that seemed suitable for modification with a triazole staple.⁴⁷⁴ The effect of the length of the azide side chain and location of the triazole itself was examined, and they found that minute differences caused significant changes in binding affinity, solubility and helix stabilization. Additionally, they designed and tested “double-stapled” peptides and showed that this dramatically improved the helical character. They also discussed the formation of “mixed staples” and how positioning of the azide and alkyne modified residues favors or disfavors these types of staples.⁴⁷⁴ This is a unique attribute of asymmetric staple chemistries.

Figure 35.

Synthesis of stapled peptides using azide–alkyne cycloaddition reactions. A) Schematic of peptide functionalization using Cu- or Ru-catalyzed or strain-promoted cycloaddition (SPAAC). B) Crystal structure of peptide E1 stapled through SPAAC using a strained dicyclooctyne linker (blue; staple shown in yellow) bound to MDM2 (cyan). PDB: 5AFG.⁴⁷³

While staples primarily serve to lock the peptide into a particular conformation, two-component staples are well-suited to the modular creation of diversified libraries that attempt to build the desired properties directly into the stapling chemistry and minimize necessary structural optimization to the primary sequence. The alkyne linkers for azide–alkyne cycloadditions can be functionalized in a number of ways to fine-tune properties such as polarity, steric, and electronic effects. A recent report from Sharma et al. created a library of functionalized diynes to investigate such effects on the p53 peptide bioactivity.⁴⁷⁵ Lau et al. report the use of dialkynyl linkers that are conjugated to a diazido peptide via CuAAC chemistry while having a third substituent, which can be a peptidic sequence, a fluorophore, or a small molecule.⁴⁷⁶ The practical application of this approach was demonstrated by multiple groups using the p53/MDM2/MDMX PPI for proof-of-concept. Wu et al. used azide–alkyne click chemistry to staple a p53-derived sequence in an i, i+7 fashion, and used diverse dialkyne linkers conjugated to CPPs as well as small molecule carriers.⁴⁷⁷ Despite similar binding affinities for MDM2 in fluorescence polarization assays, there was greater variability for p53 activation in cell based reporter assays, presumably due to differential uptake. Similarly, Charoenpattarapreeda et al. used a diazidopeptide clicked with a dialkyne linker containing an electrophilic, lysine reactive moiety.⁴⁷⁸ After confirming the helical character of the peptide via CD spectroscopy, they showed that a p53-derived peptide stapled in this way exhibited selective and covalent binding to MDM2 in vitro.

3.7. Diels–Alder Staples

Another cycloaddition in the toolbox of stabilizing chemistries is the well-known Diels–Alder reaction, favored for the spontaneous reaction occurring at room temperature or with mild heating with no catalyst and compatibility with a wide range of solvents and substitutions. In the interest of using diverse stapling chemistries to sample different peptide conformations, as discussed previously by Kritzer and others,⁴⁵² Diels–Alder chemistry is well-suited for this application, as it can utilize a variety of diene and dienophile combinations and has already proved its value as a bioorthogonal, biostable conjugation chemistry.^479,480 Montgomery et al. reported the use of Diels–Alder cycloaddition for stabilizing peptides.⁴⁸¹ The dienes or dienophiles were incorporated onto cysteine or lysine side chains postsynthesis or were built into the SPPS workflow using amino acid building blocks that already had the desired moiety. As the first proof-of-concept experiment, stabilization of an RGD loop was carried out using the Diels–Alder cycloaddition of hexadiene and maleimide, which was shown to increase trypsin resistance and enhance bioactivity. This was followed by the stabilization of the p53-derived α-helix in an i, i+7 fashion using furan and maleimide, which readily underwent near-complete cyclization in 15 min and showed increased helicity in an aqueous solution. The same strategy was applied to the development of an ERα inhibitor. ERα is a transcription factor with a ligand binding domain that contains a druggable pocket for estrogen binding that is typically targeted by selective estrogen receptor modulators (SERMs).⁴⁸² However, common point mutations such as Y537S are known to confer drug resistance by stabilizing the agonist conformation independent of estrogen binding, resulting in constitutively active ERα.⁴⁸³ An alternative strategy is to disrupt the binding of coactivator SRC2 to ERα, which binds via a hydrophobic helix with the LXXLL motif. Many groups have reported stapled peptides to target ERα, particularly by developing an SRC2-derived peptide sequence spanning the LXXLL motif that is responsible for the SRC2 coactivator binding and using various staple chemistries to stabilize the helix.^{340,469,481,484-487} Montgomery et al. used this well studied PPI as a proof-of concept for Diels–Alder peptide stabilization (Figure 36A).⁴⁸¹ During the SPPS workflow, furylalanine and lysyl-maleimide residues in i, i+4 positions were reacted on resin in DMSO. A small panel of Diels–Alder cyclized (DAC) peptides identified improved helical character and binding for several members. The mode of binding was confirmed by obtaining a crystal structure of ERα in complex with one lead compound, DAC6. Interestingly, although one of the Leu residues responsible for the binding interaction was substituted with the furylalanine, this was compensated by the interaction of the cycloadduct with the hydrophobic shelf of ERα, showing that Diels–Alder staples could not only stabilize secondary structure, but contribute to novel binding interactions (Figure 36B). Taking into consideration the modularity of the Diels–Alder reaction, this stapling approach could prove to be highly versatile for library synthesis and in combination with other stapling chemistries.⁴⁸¹ In addition, reverse Diels–Alder reactions have proved very useful for bioorthogonal crosslinking. The diene of choice is usually tetrazine, which can rapidly undergo Inverse Electron Demand Diels–Alder (IEDDA) followed by nitrogen loss, making the process irreversible and entropically favored.^488,489

Figure 36.

Diels–Alder stabilized peptides targeting ERα. A) Schematic of DAC peptide functionalization on resin. B) Crystal structure of DAC6 (blue; staple shown in yellow) in complex with ERα (cyan) bound to estradiol (red). PDB: 6PIT.⁴⁸¹

3.8. Staple Properties and Layered Stabilization Approaches

The emergence of complementary and orthogonal chemistry approaches for secondary structure stabilization allows the sampling of various strategies to achieve the optimal scaffold architectures and improvements in affinity, metabolic stability, and membrane permeability.^41,490 In a notable comparative study of several common stapling chemistries (lactamization, hydrocarbon, CuAAC, thiol alkylation, and perfluorarylation) using CD, NMR, and molecular dynamics of pentapeptides in water revealed that lactamization resulted in the most α-helical character, while the hydrocarbon staples and CuAAC staples showed characteristics of both α- and 3₁₀-helices.⁴⁹¹ Recently, Strømgaard and colleagues performed a comprehensive analysis of several stapling approaches by varying the position, chemistry, and linker length using amyloid precursor protein (APP) dodecamer peptide targeting Mint2 as a template.⁴⁹² They systematically tested lactam cyclization, hydrocarbon stapling, triazole stapling through CuAAC, and cysteine crosslinking, using various side chain substituents to alter the linker length for each approach. In general, increasing the linker length also increased affinity, and all stapled variants tested displayed increased resistance to tryptic digestion. Hydrocarbon stapling provided the most structured peptide in solution by NMR, followed by triazole and cysteine crosslinked peptides. While the hydrocarbon stapled peptides exhibited the highest affinity, cysteine cross-linked peptides displayed the greatest cell permeability, whereas the lactam and triazole linked peptides suffered from a lack of uptake and stability. Finally, metabolic stability and affinity was closely linked to the degree of structural order, and hydrocarbon stapled peptides most closely mimicked the native bound structure of APP to Mint2. A similar study by de Araujo et al. explored the effects of staple identity on the hMD2 binding peptide derived from p53.⁴⁹³ In this study, cysteine crosslinked peptides displayed the lowest protection from degradation, whereas lactam and hydrocarbon staples provided exceptional stabilization. However, the use of more hydrophilic linkers in lactams or triazoles greatly reduced cell uptake, as only modifications that increased hydrophobic surface area such as aromatic thioether and hydrocarbon staples led to increased permeability. These studies and others⁴⁹⁴ demonstrate the importance of systematically testing different scaffolds to balance affinity, stability, and permeability in a case-by-case basis, although in general it appears that hydrocarbon stapling enhances desirable properties.

Orthogonal stapling chemistries can also be used to layer multiple staples in a single scaffold to better improve structural stability. Showcasing the utility of layering multiple hydrocarbon staples, Bird et al. developed a double stapled peptide based off of the gp41 HR2 fusion domain, resulting in SAHgp41, which exhibited greater antiviral activity and better proteolytic stability compared to FDA-approved enfuvirtide, as well as some indication of oral bioavailability.³²⁸ Likewise, the concept of “stitched” peptides first reported by Hilinksi et al.⁴⁹⁵ These scaffolds utilized α,α-disubstituted residues with two olefin side chains to generate dual hydrocarbon staples to greatly improve the thermal and metabolic stability, as well as cell penetrating ability compared to single stapled counterparts (Figure 37C). Due to the requirement for two regioselective metathesis reactions, specific combinations of amino acids and spacing must be used; even with these designs a mixture of side products can be obtained, highlighting the value of complementary layered chemistries.

Figure 37.

Functionalization of peptides using stitched or layered stapling. Synthesis of peptide stapling using hydrocarbon and Diels–Alder (A), hydrocarbon and lactam (B), or dual hydrocarbon stapling (C).

We demonstrated an example of orthogonal and layered staple chemistries using hydrocarbon and Diels–Alder stapling in a double stapled peptide derived from SRC2 (Figure 37A).⁴⁸¹ The double stapled peptide displayed a strong α-helical signal by CD with classic minima at 208 and 222 nm, and suggests that this combination of chemistries could be developed further for hyperstabilized or larger stabilized peptides. Speltz et al. used a similar template to combine hydrocarbon and lactam stapling (Figure 37B).⁴⁸⁶ The bicyclic peptide exhibited increased α-helicity and proteolytic stability but comparable binding affinity to unmodified and single stapled counterparts. Fairlie and co-workers also explored dual hydrocarbon and lactam staples in a BH3 domain peptide.⁴⁹⁶ Including a lactam staple to induce stronger α-helicity and a hydrocarbon staple to improve cell permeability led to a cell permeable, metabolically stable dual inhibitor selective for BCL2A1 and MCL-1. We expect that as additional chemistries are added to the stapled peptide toolbox, more options will become available for layered synthesis of more stable and efficacious peptide probes and therapeutics.

3.9. Summary and Outlook for Side Chain Stabilized Secondary Structure Mimetics

Side chain stabilized secondary structure mimetics have shown immense promise in the development of chemical probes and therapeutics. Indeed, several “stapled peptides” have advanced to the clinic and have shown good safety profiles as a class in advancing the field of therapeutic biologics. However, despite the incredible breadth of studies performed to examine the types of staples, layering chemistries, and biophysical properties of various secondary structure mimetics, side chain stabilized peptides have failed to advance through late stage clinical trials. For example, hydrocarbon staples have been extensively applied to increase the cell permeability and affinity for peptide therapeutics and have shown good promise but have yet to be incorporated into a market drug. The consensus of the literature in this area seems to indicate that while some generalizations can be made regarding staple effects on pharmacologic properties (e.g., linker identity, length, and stereochemistry), there are no hard and fast rules applied to all cases. Alternatively, the field is moving further into the generation of diverse stapled peptide libraries, such as hydrocarbon DELs⁴⁴⁵ and phage display libraries.^497,498 Developing more compatible ways to incorporate stapling earlier into library generation and screening will be highly valuable going forward in identifying potential therapeutic candidates.

The number of useful reactions that can be used to stabilize secondary structures also enable the construction of larger combinations of motifs, or even different modalities. Peptides are easily conjugated to small molecules, other peptides, or larger biologics. Chemical stabilization enables them to maintain their structure and functionality in a range of environments, an advantage over larger natural biologics. As a testament to their versatility, many stapled peptides have been used in the development of PROTACs against difficult protein targets.^{301,343,499,500} Peptide library technologies are expanding to accommodate the selection of specific structural elements as well.⁴⁴⁹ While there are still many hurdles to be overcome in terms of pharmacokinetic properties, stabilized peptides generally have a good safety profile, and it has been repeatedly shown that with the right amount of tuning, better permeability, bioavailability, and efficacy in vivo is possible.

4. NON-NATURAL PEPTIDOMIMETICS

4.1. Non-natural Peptidomimetics Integrate Changes to the Peptide Backbone to Improve Activity

A common goal in the application of designer peptidomimetics is to recapitulate the folding or secondary structure of native peptides and protein domains. This is particularly important for binding to the large and disordered surfaces characteristic of undruggable targets. However, even with a well characterized ligand or surface, naturally derived peptides can still suffer from poor biostability and cellular uptake, which may stem partly from the peptide backbone core itself.^501,502 To alter the biophysical properties of peptides, many non-natural backbone analogs have been developed and implemented to create myriad structures that retain the ability to form helical or sheet-like structures. Most importantly, these non-natural backbones are not recognized by native hydrolytic enzymes, which can significantly improve biological stability (Figure 38).^501,502 Additionally, inclusion of substituents along different portions of the backbone can greatly improve solubility and cellular delivery, resulting in potentially improved pharmacological properties.⁵⁰²

Figure 38.

Non-natural backbones mimic native peptide secondary structure and provide resistance to proteolytic degradation.

This section covers non-natural peptidomimetics that may include features of both cyclic peptides and side chain stabilized peptides, but also incorporate changes to the peptide backbone that yield non-native species and structures. This includes foldamers, where a significant portion of the backbone is replaced by non-natural residues (class B mimetics), and structural mimetics that replace the entire backbone with nonpeptidic scaffolds that still retain side chain functionality and conformation (class C mimetics).^22,503-505 These modalities benefit from a repeating backbone unit or pattern for enhancing stability while maintaining side chain identity and position to promote interactions, as compared to side chain stabilized secondary structures that require key positions for modification. We will discuss the design and characteristics of these different backbone modifications and highlight examples for targeting undruggable classes.

4.2. Design and Characteristics of Non-natural, Backbone Modified Peptidomimetics

4.2.1. β-Peptides.

The addition of a methylene group to create a repeating aminoethyl carboxyamide backbone affords a β-peptide and is the most thoroughly studied backbone modification for non-natural peptidomimetics (Figure 39A,B).^22,506 Easily incorporated into SPPS protocols, β-amino acid monomers contain multiple locations for side chain substitution at the β₂- and β₃-positions, which can provide increased molecular recognition vectors and contribute to overall structure stabilization. Additionally, β-peptides have shown a propensity to fold into several unique helical conformations in both aqueous and organic environments, such as the 10/12-helix, 12-helix (2.5 residues per turn and 12 atoms between H-bonds), and 14-helix (3 residues per turn and 14 atoms between H-bonds) (Figure 39C). The position and stereochemistry of the side chains in a β-peptide play a strong role in the handedness and helix structure.⁵⁰⁶ In a 10/12-helix, alternating monosubstituted β₂- and β₃-positions result in the characteristic intertwined 10- and 12-membered hydrogen-bonded rings.⁵⁰⁷ 12- and 14-helix conformations are favorable with substituents at the β₃-position and including cyclic β-amino acids such as ACPC and ACHC can induce conformational preference to 12- or 14-helix geometries, respectively (Figure 39B). Cyclic β-amino acids also increase the tendency of β-peptides to fold into helices in aqueous environments, which is important to maintain activity in biological settings. For additional in-depth discussion of β-peptide folding and synthesis refer to several excellent reviews.^{22,506,508-512}

Figure 39.

A) Chemical structures of α-, β-, γ-peptide, oligourea, and peptoid backbones. B) Structures of helix-inducing cyclic β-amino acids. C) Hydrogen bonding patterns of β-peptide helices.

While the geometry and properties of homogeneous β-peptides are well characterized and provide excellent resistance to proteolytic degradation, it is more straightforward to design initial leads from natural peptide or protein (i.e., α-amino acid mediated) epitopes.^511,513 Thus, incorporating β-amino acids strategically within α-peptides, known as α/β-peptides, emerged as an alternative to mimic the native geometry and still provide enhanced resistance to proteolysis.⁵¹⁴ In these designs, α-amino acids are used for surface recognition, whereas β-amino acids typically induce and stabilize secondary structure conformations.⁵¹⁵ Many other conformations and properties depend upon the ratio and identity of included α- and β-amino acids.^{22,511,514,516} Most importantly, replacing even a small portion of a peptide mimetic with β-amino acids has been shown to increase metabolic stability in serum or in vivo half-life by orders of magnitude, both desirable traits necessary for development of useful therapeutics or chemical probes.^{501,509,517,518}

Early work by the Schepartz lab using β-peptides as antagonists of undruggable targets aimed to inhibit the interaction between the transcription factor p53 and the E3 ubiquitin ligase hDM2.⁵¹⁹ This interaction, which is broadly implicated in cancer, has become one of the hallmark models for designing and testing non-natural peptidomimetics. In addition to being an attractive therapeutic target, this is also due to the short α-helical motif from p53 that binds to a hydrophobic cleft on hDM2 involving three key hydrophobic residues: Phe19, Trp23, and Leu26, which can be incorporated into foldamers attempting to mimic this structure.⁵²⁰ Schepartz and co-workers utilized this motif to generate β³-peptides that adopt a left-handed 14-helix structure capable of binding hDM2 with submicromolar affinity.⁵¹⁹ Later use of computational modeling,⁵²¹ inclusion of non-natural or cationic side chains,^522,523 and stabilization by side chain stapling,⁵²⁴ led to β³-peptides with improved affinity for hDM2/X and cell permeability that may provide access to more active compounds in physiological disease settings.

Another common target for designing non-natural peptidomimetic antagonists is the highly conserved BCL-2 family of proteins.²⁰⁰ Gellman and co-workers first discovered a hybrid α/β-peptide that could mimic an α-helical BH3 domain necessary for many interactions involving BCL-2 family proteins.^525,526 A solved crystal structure of α/β-1 bound to BCL-2 indicated this architecture effectively mimics the display of side chains with good overlap of conserved hydrophobic residues h1–h4 compared to native BH3 domains (α-Bim), showcasing the potential for this design to specifically target protein–protein interaction interfaces using sequence-based design (Figure 40A,B).^527-529 This architecture was further improved by including chemical stabilization strategies such as side chain stapling to improve cell uptake and activity, while retaining the affinity and improved proteolytic resistance even over analogous stapled α-peptides.⁵²⁹ Other β-peptide and α/β-peptide mimetics have been developed to target viral infection and replication, showcasing the extraordinary potential for modulating biological processes involving PPIs and protein–nucleic acid interactions.^530-533 However, shortcomings including poor intrinsic cell permeability and bioavailability have limited the translation into clinically relevant compounds for targeting undruggable classes. Lessons learned from the immense research on structural stability as well as more recent studies to achieve prolonged half-life and efficacy in vivo against viral infection and modulating receptor activity may help to provide guidelines for adapting these structures toward undruggable intracellular targets in more complex disease settings.^533,534

Figure 40.

α/β-Peptides derived from BH3 domains for targeting BCL-2. A) Crystal structure of stapled α/β-1 (orange; staple shown in green) bound to BCL-2 (cyan). PDB: 5AGW. B) Overlay of reported α-Bim analogue (PDB: 3FDL), a stapled analogue α-1 (PDB 2YQ6), designed α/β-1 (PDB: 5AGW), and α/β-1-LIN (PDB: 5AGX) analogues showing close overlap of conserved hydrophobic residues h1–h4. Adapted from ref 529. Copyright 2015 American Chemical Society.

4.2.2. γ-Peptides.

The retention of the peptide-like backbone structure analogous to α-peptides but increasing the backbone by an additional two methylene groups yields a γ-peptide (Figure 39A). Similar to β-peptides, γ-peptides are easily incorporated into standard SPPS protocols and offer additional functionality along the backbone to fold into various conformations.^{509,516,535-537} Promising results in folding of homogeneous γ-peptides as short as tetramers into helical structures⁵³⁵ led to the creation of heterogeneous α/γ-, β/γ-, and α/β/γ-peptides to mimic protein secondary structure.^516,538-543 Although they are less widely used and studied than β-peptides, the additional functionality and complexity afforded by γ-peptides may offer significant opportunities for structural optimization and applications for targeting diverse biomolecular surfaces.^539,544 Additional information on mixed homologous α-peptides, β-peptides, and γ-peptides can be found in several interesting studies and reviews.^{509,516,536,541}

Although there are few applications in literature of using γ-peptides and hybrids, these analogues have been reported to target interactions of p53/hDM2 and amyloid formation.^540,545 Aitken and co-workers reported an α/β/γ-peptide foldamer as a selective inhibitor of the p53/hDM2 interaction. This α-helix mimetic contains alternating trans-2-aminocyclobutanecarboxylic acid (tACBC) β-amino acids and γ⁴-amino acids that adopt a 12,13-helix conformation capable of projecting the key hotspot residues involved in the natural p53 interacting helix (Figure 41A-B).⁵⁴⁰ Although this structure only achieved modest binding affinity and inhibitory activity, there was a significant increase in proteolytic stability. More recently, the Maillard group developed conformationally constrained heterocyclic γ-amino acids composed of 4-amino(methyl)-1,3-thiazole-5-carboxylic acids termed ATC monomers (Figure 41C). This design folds to adopt a 9-helix structure analogous to the native 3₁₀-helix (Figure 41D).⁵⁴⁶ They surmised that this architecture could be used to target transient 3₁₀-helices involved in amyloidogenesis to prevent aggregation. Using this design, they generated γ-ATC-peptides capable of modulating aggregation of both Aβ_1–42 and human islet amyloid polypeptide (hIAPP), which are desirable drug targets for Alzheimer’s disease and type 2 diabetes, respectively.⁵⁴⁵ They explored different side chain substitutions for effects on affinity and solubility, and found that ATC oligomers with at least 6 units containing cationic side chains to increase solubility was necessary for activity. This study demonstrated the importance for designs capable of binding large, disordered surfaces and the necessary requirements for balancing biophysical properties such as size and solubility to attain active compounds.⁵⁴⁵ For additional information on recent γ-peptide monomers and structures, refer to the excellent review by Legrand and Maillard.⁵³⁶

Figure 41.

Helical conformations of γ-amino acid containing foldamers. A) Hydrogen bonding patterns of α/β/γ-peptide mimicking p53 helix. α-Residues are highlighted in gray. β-tACBC-residues are highlighted in blue. γ-residues are highlighted in orange. Key residues Phe, Trp, and Leu are labeled. B) Molecular modeling simulation overlay of p53_16–29 (red) with α/β/γ-peptide mimic (yellow); RMSD = 0.89 Å for α-carbons. Adapted from ref 540. Copyright 2016 The Authors under CC BY https://creativecommons.org/licenses/by/4.0/.⁵⁴⁰ C) Hydrogen bonding pattern of a homogeneous γ-ATC 9-helix. D) Overlay of a native 3₁₀-helix (yellow/orange) and γ-ATC 9-helix (green/red) showing structural overlap of key residue positions. Adapted from ref 545. Copyright 2020 Wiley-VCH GmbH.

4.2.3. Oligourea Peptidomimetics.

Aliphatic N,N’-linked oligoureas are composed of a similar length backbone unit as γ-peptides but contain an additional nitrogen replacement of a carbon within the backbone to form the urea bond (Figure 39A).⁵⁴⁷ As in backbone modifications described above, oligourea foldamers are attractive mimetic scaffolds due to their synthetic accessibility, broad functionality, and folding propensity. Monomers are synthesized from available chiral substituted ethylene diamine or α-amino acid building blocks, and span a diverse chemical space beyond that of α- or β-amino acids.⁵⁴⁸ Oligoureas have been produced by both solution and solid phase methods; however, significant solubility issues resulted in solid phase becoming the routine approach.^549,550 Guichard and co-workers extensively studied the synthesis of oligoureas and eventually discovered a microwave-assisted solid phase protocol using azido-terminated succinimidyl-activated carbamate monomers that efficiently produced oligoureas and oligourea–peptide hybrids (Figure 42A).^551,552

Figure 42.

Synthesis and properties of oligoureas. A) Solid-phase synthesis of oligoureas using azide-terminated succinimidyl-activated carbamate monomers with microwave assistance. B) Hydrogen bonding pattern of homogeneous oligourea 12- ,14-helix. C) Crystal structure of p53-derived oligourea–peptide hybrid bound to MDM2 (cyan). Key interacting residues are labeled. PDB: 6HFA.⁵⁵⁶ D) Crystal structure of oligourea–peptide hybrid bound to ASF1 (cyan) with key interacting residues labeled. PDB: 6ZUF.⁵⁵⁷ Oligourea monomers are orange and α-amino acids are blue.

Oligourea peptidomimetics containing homogeneous or heterogeneous backbones with incorporated α-amino acids can adopt helical confirmations closely akin to an α-helix. This is independent of a primary sequence^553,554 and is caused by additional backbone conformational restriction and intramolecular hydrogen bonding stemming from the ethylene diamine moiety resulting in the presence of linked 12- and 14-membered hydrogen bonded rings (Figure 42B).⁵⁵⁵ This propensity to fold into helical conformations regardless of side chain identity make oligourea foldamers highly desirable for their potential to become active, biostable, and conformationally controlled therapeutics. For more detailed information regarding oligourea structure and synthesis, refer to the excellent perspective by Yoo et al.⁵⁴⁸

Although first synthesized and studied as foldamers in 2002, only a few examples of oligoureas targeting intractable PPIs have been reported, albeit nearly two decades later. Guichard and co-workers generated peptide–oligourea hybrids to target the transcription factors p53 and vitamin D receptor (VDR).⁵⁵⁶ Using the hallmark p53/MDM2 system and a peptide previously identified from a phage display library, PMI,⁴⁰⁵ as a blueprint, oligourea residues were gradually incorporated into the structure. Five or six residues on the C-terminus of PMI, of which contain the key interacting residues Leu and Trp, were replaced with a short oligourea triad, resulting in a minimal ~2-fold decrease in activity compared to PMI by a time-resolved fluorescence energy transfer assay. A solved crystal structure deduced that key hydrophobic contacts are maintained and the oligourea backbone closely mimics the contacts formed by the cognate peptide (Figure 42C). However, the position and spacing of side chains in the oligourea portion was critical to properly mimic the α-helix geometry, stemming in part from the larger diameter of the oligourea helix.⁵⁵⁶ The increased diversity for side chain positions available to oligourea peptidomimetics may provide more opportunities to improve structural mimicry.

The same group employed a similar approach to target an epigenetic modifying protein, antisilencing function 1 (ASF1).⁵⁵⁷ ASF1 is a histone chaperone protein involved in regulating gene expression that is overexpressed in some forms of cancer.⁵⁵⁸ Depletion of ASF1 paralogs A and B reduced cancer cell proliferation and increased sensitivity to some chemotherapeutic agents.⁵⁵⁹ ASF1 binds to the H3/H4 heterodimer over a large surface area containing several shallow pockets, which are key features involved in undruggable targets.⁵⁶⁰ A previously identified 26-residue peptide, ip4, binds to ASF1 with nanomolar affinity through four hotspot residues Lys2, Leu6, Arg9, and Ile10 derived from a helix on H3.⁵⁶¹ Mbianda et al. replaced the six central residues of this helical portion with four ureido residues (Figure 42D).⁵⁵⁷ This design was only able to recapitulate the positioning of three out of four key anchoring residues (Leu6, Arg9, and Ile10), resulting in a greater than 10-fold decrease in affinity (K_d = 2.7 μM vs 0.13 μM). However, the oligoureapeptide hybrid displayed much greater resistance to degradation in serum compared to the parent peptide, which agrees with other studies showing prolonged half-lives of oligourea peptidomimetics.^557,562,563

4.2.4. N-Alkylated Peptoids.

Peptoids are α-amino acids developed in the late 1980s that contain a side chain on the amide nitrogen of the backbone to produce oligomers of N-substituted glycines (Figure 39A).⁵⁶⁴ This substitution confers improved cell permeability and resistance to proteolysis.^565-567 Short oligomer peptoids closely mimic small molecules while providing peptide-like functional contacts and folds. Ramachandran-type plot calculations suggested peptoids contain more available conformational states compared to that of natural peptides.⁵⁶⁴ Structure-guided designs have shown that peptoids are capable of folding into helical and sheet-like conformations, closely mimicking the secondary structures seen in natural proteins.^564,568 Interestingly, these structures were unaffected by heat or urea denaturation, suggesting that steric forces drive structural stabilization.⁵⁶⁹ This is much different from the hydrogen bonding forces that apply for native and non-natural helices previously described. Additional strategies to conformationally constrain peptoids, such as the inclusion of chiral side chains or cyclization may improve activity.^570-572 This relationship between chemical structure and conformational preference is nicely outlined in a recent review by Eastwood et al.⁵⁷³

While peptoid monomers are capable of being incorporated into conventional SPPS methods for synthesis, a significant drawback to these standard protocols is the need for suitable quantities of protected building blocks. Moos and co-workers developed an alternative approach that uses readily available submonomers that sidesteps the need for backbone protecting groups (Figure 43A).⁵⁷⁴ Instead, synthesis proceeds by backbone elongation using haloacetic acid derivatives followed by introduction of side chains with a primary amine, thus greatly expanding the chemical diversity. Using this simple synthetic setup and a broad selection of substituents, large and chemically diverse peptoid libraries can be produced and screened against therapeutically relevant targets.^575-577

Figure 43.

Synthesis and structures of peptoids. A) Solid-phase synthesis of peptoids using a submonomer strategy. B) Chemical structure of constrained peptoid targeting MDM2. Key residues Phe, Trp, and Leu are highlighted in orange. C) Chemical structure of β-catenin/TCF interaction inhibitor macrocyclic peptide-peptoid hybrid compound 13.⁵⁷⁸

Peptoids have a rich history in drug discovery as described by Zuckerman in the commentary “Peptoid Origins”.⁵⁶⁸ Original designs centered around combinatorial libraries screened against various receptors.^579,580 However, the discovery of peptoids that are able to fold into higher order structures inspired interest in rational targeting approaches.⁵⁸¹ For example, the Sando group designed a conformationally constrained peptoid ligand for MDM2 by using an α-carbon substituted backbone to constrain the inherent flexibility through steric hindrance (Figure 43B).⁵⁸² This design provides a scaffold to display multiple functional groups in well-defined positions. Strategic placement of the hotspot residues Phe, Trp, and Leu, resulted in oligo-NSA 10, a peptoid capable of binding to MDM2 and competing with the native p53 helix peptide TAD. Importantly, they show that conformationally constraining the peptoid was essential for achieving efficient protein binding.⁵⁸² In a subsequent study, the group introduced side chain modifications to increase affinity and cell permeability, highlighting the programmability of peptoids for improvement via SAR analysis.⁵⁸³

In another study, Logan and co-workers applied computational design protocols to generate in silico cyclic peptoid–peptide hybrids to inhibit the β-catenin/TCF interaction, which plays a critical role in the cancer associated WNT signaling pathway.^578,584 Their identified macrocyclic hybrid, compound 13, was capable of inhibiting WNT-dependent luciferase activity in live cells better than commercially available WNT inhibitors, and decreased viability of WNT-dependent prostate cancer cell lines (Figure 43C). Compound 13 was confirmed to disrupt the β-catenin/TCF interaction in live cells by coimmunoprecipitation, suggesting inhibition of this PPI to the be main mechanism of action. Compound 13 was also shown to be efficacious in vivo, where it down-regulated WNT-dependent pathways to restore phenotypes in a zebrafish model. This example portrays the immense potential for peptoids and related compounds to address the myriad undruggable interactions implicated in human disease with the goal to advance therapies into the clinic. In fact, studies into the in vivo efficacy and oral bioavailability of cyclic peptoid hybrids are already showing promising results.⁵⁸⁵

4.2.5. γ-AApeptides.

Developed by the Cai group in 2011, γ-AApeptides are oligomers of γ-substituted N-acylated-N-aminoethyl amino acids (Figure 44A).⁵⁸⁶ Originally synthesized based on the γ-chiral peptide nucleic acid (PNA) backbone containing proteinaceous side chains,^587-589 this backbone modification is amenable to SPPS protocols with many different scaffolds and substituents.⁵⁹⁰ Importantly, the repeating unit projects an equivalent number of side chains comparable to repeating di-α-peptides, thus containing potential for mimicking natural structures. However, because of the side chain locations on the aminoethyl unit and proximal tertiary amide, significant differences in hydrogen bonding patterns, conformational flexibility, and side chain orientation creates a challenge in predicting the folding patterns of these mimetics. Therefore, homogeneous sulfono-γ-AApeptides and hybrid α/sulfono-γ-AApeptides were developed as helical mimetics capable of adopting left- and right-handed geometries.^591-593 These constructs adopt a few different helical conformations dependent upon monomer ratio, including a homogeneous left-handed 14-helix (Figure 44B) and a right-handed 1:1 l-α/l-sulfono-γ-AA 13-helix that most closely resembles a native α-helix (Figure 44C).

Figure 44.

A) Chemical structures of PNA, γ-AApeptide, sulfono-γ-AApeptide, and di-α-peptide backbones. B) Hydrogen bonding pattern of a homogeneous l-sulfono-γ-AApeptide left-handed 14-helix. C) Hydrogen bonding pattern of a heterogeneous 1:1 l-α/l-sulfono-γ-AA 13-helix. Sulfono-γ-AA monomers are highlighted in orange. l-α-amino acids are highlighted in gray.

Cai and co-workers generated γ-AApeptides to target the canonical p53/MDM2 interaction as a case study.⁵⁸⁶ Although only slight inhibition was achieved (IC₅₀ > 50 μM), the chemical diversity afforded to this design may provide better inhibitors in the future. Indeed, the group later developed a series of sulfono-γ-AApeptides, which have a greater helical propensity, and identified more potent inhibitors of p53/MDM2 interaction with submicromolar potency by fluorescence polarization (IC₅₀ = 0.891 μM).⁵⁹⁴ Computational simulations suggest this architecture retains good overlap with the key residues of the p53 peptide when bound to MDM2 (Figure 45A,B). In a separate study, the same group developed sulfono-γ-AApeptide inhibitors of the difficult-to-drug target β-catenin based on a helical interacting motif in BCL9.⁵⁹⁵ In this work, cell uptake of FITC-labeled sulfono-γ-AApeptides and associated activity in live cells was achieved, suggesting primary sequence may be important for cell permeability of this class of peptidomimetics, as also observed in a recent study targeting amyloid β (Aβ₄₂).⁵⁹⁶ Modeling studies again showed that a homogeneous sulfono-γ-AApeptide can project functional side chains and interact with a surface in a similar manner to native helical peptides (Figure 45C,D). Most notably, in all studies γ-AApeptides or sulfono-γ-AApeptides were shown to be stable against proteolytic degradation.^594,595,597

Figure 45.

Sulfono-γ-AApeptides recapitulate native helical peptides to bind surfaces. A) Overlay of side chains on a sulfono-γ-AApeptide mimic (red) with native Phe19, Trp23, and Lue26 of p53 (green). B) Crystal structure of sulfono-γ-AApeptide bound to MDM2 (CCDC: 1841094). C) Overlay of sulfono-γ-AApeptide mimic (purple) with critical residues of native BCL9 helix (red). D) Overlay of sulfono-γ-AApeptide mimic (purple) with BCL9 helix (red) on the binding surface of β-catenin (PDB: 2GL7). A,B Adapted from ref 594. Copyright 2020 American Chemical Society. C,D Adapted from ref 595.

Due to difficulties in predicting the γ-AApeptide structure as well as the large chemical space available, de novo selection screens using OBOC libraries could be necessary for compound identification, and were piloted for undruggable targets such as amyloid β and the transcription factor STAT3.^598,599 In the latter example, the selected γ-AApeptide inhibitors did not disrupt STAT3 dimerization but prevented STAT3-DNA binding, suggesting potential for this class to disrupt protein–nucleic acid interactions. Additional characterization of folding patterns and biological potential of γ-AApeptides for in vivo applications is necessary to further develop this class toward undruggable targets.⁶⁰⁰

4.2.6. Nonpeptidic Helix-Mimicking Scaffolds.

Many foldamers induce a secondary structure using backbone modifications that retain peptidic character; however, arguably the most important aspect of specifically targeting surfaces with high affinity is the position and orientation of the side chains to interact with key “hotspot” residues.⁶⁰¹ In 2001, Hamilton and co-workers envisioned replacing the peptide-like backbone with an entirely nonpeptidic scaffold that would mimic the side chain projection of an α-helix. Using a functionalized terphenyl scaffold, they showed that relative orientation of side chains from the i, i+3/4, and i+7 positions in an α-helix are closely mimicked (Figure 46A,B).⁶⁰² By functionalizing the aromatic rings at the ortho position, steric interactions induced a nonplanar, staggered conformation that mimics two turns of an α-helix.⁶⁰³ This low molecular weight scaffold with small molecule character could provide improved pharmacological properties such as proteolytic stability and oral availability. However, tedious synthetic routes and low solubility have thus far limited their use. To improve upon the design and production, many alternative nonpeptidic backbones were developed to include heterocyclic substituents and more accessible oligomerization strategies (Figure 46B). For a complete overview of nonpeptidic structural mimetics, refer to the corresponding reviews.^504,604-607

Figure 46.

Nonpeptidic backbone mimetics. A) Nonpeptidic backbones mimic the side chain orientation of natural α-helices. B) Chemical structures of common helix-inducing nonpeptidic backbones.

Initial designs by Hamilton and co-workers using terphenyl and terephthalamide derivatives focused on recapitulating the binding residues of p53/MDM2 and BCL-X_L/Bak.^603,608,609 They showed that this class had the potential to target these interfaces with submicromolar efficacy and activity in cells.⁶¹⁰ Since then, many different designs and improvements to nonpeptidic scaffolds have been developed for numerous undruggable targets including transcription factors,^611-614 GTPases,⁶¹⁵ and interactions governing amyloid aggregation,^616-620 antiapoptotic proteins,^621,622 and nucleic acids.^623-625 A scaffold that has generated much interest is based on aromatic oligoamides (Figure 46B). This design exhibits predictable folding patterns controlled by intermonomer hydrogen bonding and ortho substituent interactions and can be adapted to both solution and solid phase synthesis for library preparation.^613,626,627 The Wilson group generated a library of N-alkylated aromatic oligoamide helix mimetics and discovered potent and selective inhibitors of p53/hDM2 and MCL-1/NOXA-B interactions.⁶²⁸ Rather than screening for binding, they used a series of cell-based assays to look for phenotypic effects caused by library members, thus selecting for active compounds. Cell uptake, binding to hDM2 and disruption of the p53/hDM2 interaction were later confirmed for lead members. Interestingly, some lead compounds also bound to MCL-1, but not to structurally similar BCL-X_L, suggesting potential for specific dual inhibitors, though this may also give rise to off target effects and the source of specificity remains incompletely understood. The main purpose of the study was to determine relationship between biophysical and cellular potency and shows that it is important to evaluate the interplay between potency for binding and actual mechanisms in cellulo, taking into consideration both on- and off-target effects.⁶²⁸

Another scaffold that has shown potential for targeting undruggable space is based on the oligooxopiperazine backbone developed by Arora and co-workers (Figure 46B).⁶²⁹ This backbone is derived from chiral α-amino acids, which are more effective for recognizing chiral protein pockets and easily diversified with myriad functional groups. The nonpeptidic chiral scaffold was initially designed based on a helix derived from the transcription factor HIF1α that interacts with the CH1 domain of p300/CBP (Figure 47A,B).⁶¹⁴ By mimicking three key residues—Leu818, Leu822, and Gln824—oxopiperazine compounds (OHMs) were produced with submicromolar affinity for p300 by fluorescence polarization. Treatment with lead OHMs decreased gene expression associated with HIF1α targets in MDA-MB-231 cells. Importantly, this compound was shown to be well tolerated in vivo and modestly reduced tumor growth in a mouse xenograft model by intraperitoneal delivery, exhibiting the remarkable potential for nonpeptidic scaffolds as relevant therapeutics.⁶¹⁴ Incorporating computational design strategies show potential to improve oligooxopiperazines for targeting HIF1α/p300 and other therapeutically relevant PPIs.⁶³⁰

Figure 47.

Oligooxopiperazines designed to inhibit HIF1α/p300 interactions. A) Model of HIF1α bound to the CH1 domain of p300/CBP (PDB 1L8C). Key residues of HIF1α CTAD Leu818, Leu822, and Gln824 in the binding pocket of the CH1 domain are highlighted and presented with an overlay of OHM 1. Adapted from ref 614. B) Chemical structure of lead OHM 1 compound.

The modular structural and synthetic strategies within this class of mimetics provide many avenues to improve activity including generating multivalent ligands and inducing native protein interactions among other functional attributes.^631-634 The small molecule character and large chemical diversity afforded to nonpeptidic scaffolds situate this class for future development into inhibitors for a wide range of undruggable interactions. Further investigation into structural stability, specificity, bioavailability, and mechanisms in relevant in vivo disease settings is necessary. Many nonpeptidic scaffolds beyond those highlighted here exist and are in development as PPI or protein–nucleic acid interaction inhibitors, which are broadly covered in a recent review by Algar et al.⁶⁰⁷

4.3. Summary and Outlook for Non-natural Peptidomimetics

Non-natural peptidomimetics incorporate significant changes to the peptidic backbone yet retain the ability to fold into defined secondary structures. These peptidic and nonpeptidic mimetics are capable of folding into discrete architectures that portray key interacting residues in defined three-dimensional space, allowing the recognition of large, diverse surfaces characteristic of undruggable interactions. Incorporating nonnatural backbones, even in a small portion of the overall oligomer, leads to substantial improvement in metabolic stability, typically by preventing proteolytic degradation. Difficulties in solubility and production can be overcome by developing new building blocks and synthetic strategies as seen in microwave-assisted oligomerization approaches.^552,635,636 Additional studies exploring the pharmacological properties and efficacy of these mimetics in vivo would greatly improve the knowledge and design principles necessary to generate therapeutic compounds.⁶³⁷ This information, along with further improvements through chemical stabilization strategies or the development of new scaffolds to improve cell permeability could lead to improved mimetics capable of targeting various intracellular interactions associated with disease.

5. SYNTHETIC PROTEOMIMETICS AND BIOLOGICS

The vast majority of synthetic peptidomimetics developed to date have been focused on secondary structure stabilization to target relatively compact hotspot surfaces on protein targets. Indeed, the work reviewed in Sections 2-4 is almost entirely made up of secondary structure mimetics. However, many protein–protein and protein–nucleic acid interaction surfaces utilize extended, noncontiguous protein sequences for high affinity and specific binding. Therefore, the ability to move beyond peptide secondary structure to tertiary and quaternary structure stabilization is an important and still emerging challenge for undruggable targets. Proteomimetics that function through tertiary or quaternary folding analogous to protein domains would be able to achieve specific interactions of high affinity with myriad biomolecular surfaces. In this section, we discuss proteomimetics that are synthetically tractable and incorporate chemical modification to stabilize higher-ordered structures (Figure 48). We will specifically focus on the application of biorthogonal chemistries for the total or semisynthesis of protein mimetics that target large “undruggable” protein and nucleic acid surfaces. Engineered proteins and protein mimetics that are genetically encoded, including monobodies, affibodies, nanobodies, DARPins, and related scaffolds are outside of the synthetic scope of this review and have been extensively reviewed elsewhere.^27,638-644 Instead, our goal here is to focus on designs that include side chain functionality or backbone alterations to stabilize natural or non-natural tertiary or quaternary protein structure. In particular, we will focus on the role of these design elements in both biochemical and physicochemical properties required for increased stability and physiologic activity in targeting undruggable biomolecules.

Figure 48.

Designer synthetic proteomimetics utilize layered chemistries to stabilize higher-order tertiary and quaternary structure. Beyond encoding three-dimensional structure, synthetic modifications can significantly improve metabolic stability for cellular and in vivo applications. Structure derived from PDB:7RCU.

5.1. Design and Characteristics of Synthetic Proteomimetics and Biologics

Proteomimetics, or synthetic biologics, contain synthetically enhanced tertiary and/or quaternary structures that recreate features of a protein to drive function. In many applications, target synthetic biologics occupy a region of molecular space between short peptides and full length proteins; accessing these molecules therefore presents synthetic and pharmacologic challenges (Figure 1C). Research on stabilized secondary structures has shown how structural constraints can enhance pharmacologic stability and cell uptake.^56,329,502 Likewise, recent efforts have been made to produce structurally stable tertiary or quaternary mimetics that contain desirable biophysical properties. This can be accomplished through partial or complete artificial backbone replacements, altered side chain interactions that promote defined interfacial contacts, as well as the incorporation of non-natural interdomain covalent cross-linking and bridging to define tertiary structure (Figure 48).⁶⁴⁵ Early work focused on structure stabilization through the use of biselectrophilic agents for cross-linking amines on lysine side chains to explore tertiary folding and stability.^646,647 However, the abundance of these residues in natural protein domains reduces inherent selectivity, severely limiting amenable designs. Alternatively, total chemical synthesis of proteins emerged to site-selectively incorporate extensive natural and non-natural building blocks for defined intermolecular ligation,⁶⁴⁸ utilizing secondary structures such as sheets or helices to create higher order structures (Figure 49). These designs, which we refer to as synthetic biologics or proteomimetics, are more amenable to extensive chemical modifications to enhance biochemical and pharmacologic activity.

Figure 49.

General design and production of synthetic proteomimetics and biologics. Chemical structures of selected common linker chemistries for induced tertiary or quaternary formation are shown. Structure derived from PDB 1IAR.

Synthetic proteomimetics or biologics are assembled through convergent synthesis from multiple native or modified peptide domains to create higher order structures with defined local and global three-dimensional structures. These extracted domains or smaller fragments can include stabilized β-sheet⁶⁴⁹ and coiled-coil motifs^650,651 present across many protein–protein and protein–nucleic acid interfaces (Figure 49).⁶⁵² For example, Arora and colleagues identified the Nervy homology two (NHR2) domain of the AML1-ETO-containing transcription factor complex as a necessary coiled-coil motif that interacts with NHR2-binding (N2B) domains of E-proteins. They extracted the native primary sequence and synthetically produced the individual helices then covalently ligated the helix dimer to produce a synthetic proteomimetic capable of adopting a defined tertiary structure. Maintaining the key interacting residues resulted in a coiled-coil mimetic capable of binding to N2B peptides with much greater affinity than the native NHR2 domain.⁶⁵⁰ Alternatively, computational methods can produce entirely artificial domains and folds as seen with the CovCore approach (Figure 49).⁶⁵³ In many of these designs, intra- or intermolecular covalent linkages as well as optimized interfacial contacts between domains are essential for obtaining minimal structures that retain high affinity and stability.⁶⁵⁴ Common interdomain linking chemistries include those previously described in peptide macrocyclization and secondary structure stabilization such as thioether formation between bis- or triselectrophiles or Michael acceptors and the side chain of cysteine,^651,653-655 bis-triazole formation via azide–alkyne cycloaddition,⁶⁵⁰ and alkene formation from ring closing metathesis^649,656 (Figure 49). In these examples, synthetic tractability is paramount for including site-selective conjugation through layered bioorthogonal chemistries to produce defined higher order structures, which is a hallmark property of synthetic proteomimetics less feasible with larger engineered proteins. Additional linkages such as amide or lactam formation⁶⁵⁷ and more recently Diels–Alder cycloadditions⁶⁵⁸ may also prove useful as selective conjugation strategies.

In addition to interdomain conjugation approaches, partial or complete replacement of the backbone can result in structurally and metabolically stable tertiary and quaternary mimetics.^645,659 The challenging endeavor of producing protein-like folding patterns from non-natural architectures hints at the prospect of creating structures and functions currently unknown to nature. However, the difficulties associated with the requisite near de novo design have thus far limited the applications of non-natural synthetic proteomimetics to just a few structural classes, including zinc finger,^660-662 amyloid,⁶⁶³ and Z-domain mimetics.⁶⁶⁴ Moreover, most examples of tertiary domain mimetics have been focused on extracellular targets.^{645,657,665,666}

5.2. Designer Synthetic Biologics for Undruggable Targets

Many intracellular undruggable targets require high affinity, high specificity, cell permeable and metabolically stable chemical probes, and therapeutics for efficacy. Identifying chemical scaffolds with these properties has remained a challenge, and only a few examples exist within the class of synthetic biologics and proteomimetics. For example, a recent study by Sadek and Wuo et al. aimed to disrupt the interaction between NEMO, which is a scaffolding protein necessary for downstream NF-κB signaling, and the viral protein vFLIP (Figure 50A).⁶⁶⁷ Screening of a 40,000 member small molecule library against the extended coiled-coil NEMO/vFLIP binding interface identified only compounds with modest binding inhibition (IC₅₀ < 65 μM). These compounds also showed highly active submicromolar inhibition in a cell proliferation assay; however, there was no specificity between vFLIP-dependent and control cell types, suggesting nonspecific off-target effects. To develop more specific NEMO/vFLIP inhibitors, they rationally designed linear and HBS-constrained peptides to mimic the key hotspot residues on helix two of NEMO. Unfortunately, neither design inhibited this interaction at 100 μM, likely due to the requirement of the coiled coil partner to properly situate the hotspot residues for binding. They then turned to the cross-linked helix dimers (CHD) approach to generate a synthetic coiled coil mimetic.⁶⁵⁴ Design parameters obtained by computation and experimental characterization informed the creation of CHD3^NEMO, which incorporates both helix one and two of NEMO ligated by a bis-triazole linker with natural and noncanonical amino acids to potently and specifically disrupt the NEMO/vFLIP interaction (K_i = 6.9 μM) (Figure 50B). This synthetic proteomimetic also displayed increased resistance to thermal denaturation and proteolysis. Importantly, this design was cell permeable through an undetermined active uptake mechanism and capable of dose-dependent inhibition of vFLIP-mediated NF-κB activation and cell proliferation. Additional experiments in vivo resulted in delayed tumor growth and improved survival in a primary effusion lymphoma xenograft mouse model, showcasing the therapeutic potential for synthetic proteomimetics.⁶⁶⁷

Figure 50.

Synthetic proteomimetic inhibitor of NEMO/vFLIP interaction. A) Thus far, small molecules and single helix mimics have proven incapable of inhibiting the NEMO/vFLIP association. PDB: 3CL3. B) Structure and sequence of tertiary NEMO mimic CDH3^NEMO, which was shown to be a potent inhibitor of vFLIP.⁶⁶⁷

As the study by Sadek and Wuo et al. shows, affinity and specificity are important to develop robust inhibitors, but metabolic stability and cell permeability are equally if not more necessary for bioactive molecules. The larger size of synthetic proteomimetics complicates the delivery to intracellular organelles, as it is unlikely to permeate cells through passive diffusion, but rather active uptake mechanisms such as endocytosis or macropinocytosis. Fortunately, these mechanisms are often upregulated in disease such as cancer, which may provide selective uptake to certain cell types.⁶⁶⁸ Hong and Yoo et al. developed a synthetic proteomimetic pan-Ras inhibitor derived from the interacting helix–loop–helix domain of SOS that engages the nucleotide-bound Ras.⁶⁶⁹ In addition to increased conformational and proteolytic stability as well as high affinity target engagement, they also observed increased uptake into cell lines harboring Ras mutations, presumably through amplified macropinocytosis as previously shown.⁶⁶⁸ This suggests certain cancer cell lines are more amenable to proteomimetic therapeutics and that specificity can be achieved through selective uptake mechanisms. In a separate study, Yang, Zhang, and You et al. developed a BCL9/β-catenin interaction inhibitor based on the helix–loop–helix domain of ECRV stabilized by a thioether linker.⁶⁷⁰ To overcome poor cellular uptake and promote endosomal escape, the cECRV proteomimetic was attached to a poly-l-lysine coated gold nanoparticle (pAuNP). The pAuNP-cECRV construct induced apoptosis of the WNT-hyperactive cancer cell line HCT116, whereas free cECRV mimetic and cold pAuNPs showed no effect, highlighting the use of nanocarriers to overcome membrane permeability limitations for some synthetic proteomimetic designs.

5.2.1. Synthetic Biologics as Direct Modulators of Transcription Factor Activity.

Transcription factors (TFs) remain one of the most challenging classes of undruggable targets due to the reliance of TFs on extended PPI and protein–nucleic acid interactions to carry out their function in regulating gene expression. Aside from nuclear receptors, which can often be regulated by small molecule binding, many classes of TFs lack defined druggable pockets and function mainly through complex regulatory networks involving large surface interactions with highly disordered domains.^671,672 This makes most small molecules and small peptidomimetic designs based solely on secondary structure unlikely to retain necessary affinity and specificity to disrupt relevant TF interactions. On the other hand, the larger size and three-dimensional structural complexity of suitably designed synthetic proteomimetics could be promising for targeting undruggable TFs. Furthermore, tertiary domain mimetics derived from native TFs could be developed with two modes of action to antagonize TF function. First, they could be developed to mimic and/or block PPIs that are necessary for effector complex formation. Second, synthetic biologics could be developed to mimic the DNA binding domain of a TF, thereby enabling direct binding of target DNA sequences and inhibition of TF-dependent gene regulation at those sites. Displacement of TF-DNA interactions remains a sought-after approach for TF inhibition. However, TFs bind and recruit many protein cofactors to influence gene expression, and disrupting these interactions may provide a means to regulate transcription. The inherent complexity of these interactions involving flat and flexible protein surfaces reduces the efficacy of secondary structure mimics. Adihou et al. showcased this in the development of a two-helix synthetic proteomimetic to disrupt the TEAD/VGL4 interaction, wherein both helices derived from VGL4 were necessary for potent binding.⁶⁷³ Chemical cross-linking via lactam formation of the distinct helices considerably increased helicity in solution, affinity for hTEAD, and proteolytic stability. Unfortunately, this design showed poor cellular uptake, likely due to the overall negative charge. Subsequent terminal modification with a cell penetrating peptide, Tat, led to a bioactive mimetic that upregulated TEAD target genes and restored Hippo signaling pathway phenotypes.⁶⁷³ The versatility of synthetic proteomimetics extends beyond native helical domains, but also can incorporate stabilized sheet structures⁵⁰⁰ and non-natural domains. For example, Wilson and co-workers produced a synthetic proteomimetic derived from the C-terminal transactivation domain of an oncogenic bHLH member, HIF1α, that interacts with p300 to regulate transcription. They previously identified two helical domains that are required for the HIF1α/p300 interaction⁶⁷⁴ and developed a potent oligoamide non-natural peptidomimetic (helix mimetic 1).⁶¹² However, helix mimetic 1 also potently inhibited p53/hDM2 suggesting a nonspecific component to inhibition.⁶⁷⁵ To increase specificity, they synthetically fused the oligoamide helix mimetic and a necessary adjacent HIF1α-derived helix, resulting in a peptide–oligoamide mimetic hybrid. Although this construct resulted in reduced potency, it was accompanied by a significant increase in selectivity,⁶⁷⁵ and the natural/non-natural hybrid nature exhibits the extensive modularity afforded to synthetic proteomimetic designs.

A defining feature of TFs is the recognition and binding to distinct genomic DNA motifs to regulate gene expression. Early studies into the modularity of TF families demonstrated that TF DNA-binding domains (DBDs) could retain affinity and specificity for target DNA even when separated from the rest of the native TF protein.^676,677 This work inspired the creation of truncated DBDs derived from different TF families such as zinc fingers,⁶⁷⁶ homeodomains,^678,679 basic leucine zipper domains (bZIP),⁶⁸⁰ and basic helix–loop–helix domains (bHLH).^681,682 However, as discussed extensively in this review, the larger native peptide sequences in many early designs suffer from poor cellular uptake and significant proteolytic susceptibility. Coupled with the need to compete with high affinity endogenous TF-DNA interactions (usually in the low nM range), natural peptide or protein sequences struggle as pharmacologically viable chemical probes and therapeutics targeting TFs.⁶⁷⁷

An exemplar case study in the development of a natural TF-mimetic involves the exploration of the dominant negative, MYC-derived molecule Omomyc. A key attribute of many TFs is the requirement for homo- or heterodimerization to assemble the competent DBD. This is true, for example, in the basic leucine zipper (bZIP) and basic-helix–loop–helix (bHLH) TF families, the latter which holds the oncogenic protein MYC. MYC is a bHLH-family protein that requires heterodimerization with a partner MAX in order to assemble the intact DBD, bind “E-box” containing DNA motifs and regulate the expression of proliferative gene programs. Therefore, inhibition of dimerization interactions to prevent complex assembly of TFs like MYC and MAX is an approach that has garnered a lot of interest over the years. In 1998, Soucek et al. explored why MYC was incapable of homodimerization but could bind to the obligate partner MAX. By mutating only four residues in the leucine zipper helix of MYC, a structural analog termed Omomyc was developed that could homodimerize or heterodimerize with MYC or MAX to prevent DNA binding.⁶⁸³ Although initially expressed using natural protein machinery in E. coli, a team at Merck demonstrated a feasible solid-phase and native chemical ligation synthesis of the 91-amino acid miniprotein, albeit in limited quantities (10–15 mg) and yields (8–10% yield).⁶⁸⁴ Several classic studies in peptide total synthesis⁶⁸¹ as well as recent approaches using improved SPPS and cross-coupling chemistries have also described total synthesis of “native-like” MYC-derived protein mimetics⁶⁸⁵ with biological activity. Despite its large size, disordered structure and proteolytic susceptibility, Omomyc was unexpectedly shown to be internalized into cells by fluorescence microscopy, likely by clathrin-mediated endocytosis or macropinocytosis.⁶⁸⁶ In subsequent in vivo studies, it was rapidly cleared from mouse plasma where it was distributed to the liver and kidneys and quickly degraded, thus highlighting expected limitations of miniproteins derived from entirely native peptides.⁶⁸⁷ However, a separate study showed Omomyc was able to distribute to tumor tissue and antagonize MYC-dependent signaling.⁶⁸⁸ Nevertheless, nearly 30 years after its inception, Omomyc (OMO-103) has successfully completed a phase I clinical trial, which is a promising endeavor for synthetic miniproteins aimed at an important undruggable target.⁶⁸⁹

To overcome many of the existing challenges with natural polypeptide-derived miniproteins aimed at TFs, our lab recently developed a new class of modular synthetic biologics we termed synthetic transcriptional repressors (STRs).⁶⁹⁰ We aimed to created compact, fully synthetic and chemically stabilized TF DNA binding domain mimetics that could directly target TF-defined DNA binding sites, thereby blocking downstream gene expression (Figure 51A). As a first target, we developed a non-natural architecture to mimic the tertiary and quaternary bHLH domain structure of MAX. We wanted to avoid inefficient synthesis of long linear polypeptides that span the necessary regions of a functional bHLH domain (>80 AAs in MYC or MAX), which would further hinder the incorporation of non-natural amino acids and structural modifications. We identified minimal regions of the basic (B) and leucine zipper (Z) helices from MAX that, when ligated together at appropriate sites, could retain the potential for homodimerization to form a minimal, yet highly functional bHLH domain structure (Figure 51A). Intriguingly, we found that this dimerization could be preserved and promoted by covalently tethering B- and Z-helices from opposing dimerization partners, requiring a “sandwich-like” dimerization event to form the bHLH tetrahelix bundle, significantly departing from the scissor-like dimerization by natural bHLH polypeptides (Figure 51C,D). To stabilize these structures, we identified compatible, layered chemistries for local secondary structure stabilization within α-helical structures (e.g., side chain stapling mediated by RCM) and interdomain covalent construction of tertiary domain structure by orthogonally masked thiol-maleimide ligation, which can take place under facile aqueous conditions for relatively high yielding synthetic biologic construction (Figure 51B). These positions and chemistries were identified using structure-based design and iterative biochemical profiling, ultimately identifying representative lead compounds with low nM affinity and high specificity for canonical E-box DNA sequences (5′-NCACGTGN-3′). Importantly, these attributes enable direct competition with full-length MYC/MAX and MAX/MAX bHLH domains for DNA binding, the first such compounds with this activity and clear mechanism of action. In parallel, we demonstrated that synthetic stabilization confers significant protease resistance and enhanced cellular uptake, allowing lead STRs to bind genomic E-box DNA sites and inhibit expression of MYC-dependent proteins in MYC-driven Burkitt’s Lymphoma cell lines.⁶⁹⁰ This work demonstrated how synthetic optimization and layering of both secondary and higher-order stabilization strategies can simultaneously enhance biophysical and pharmacological properties of non-natural proteomimetics.

Figure 51.

Synthetic transcriptional repressors (STRs) are modular proteomimetics capable of sequence-specific DNA-binding. A) Convergent synthetic approach to prepare STRs derived from MAX. B) Convergent synthetic scheme to produce stabilized STRs purified and characterized by HPLC-MS. C) Crystal structure of bHLH STR on E-box DNA displaying the sandwich-like assembly. PDB: 7RCU. D) Overlay of crystal structures of STR (blue/pink) and MAX:MAX (orange/yellow) homodimers. PDB: 7RCU and 1HLO, respectively. E) Structure-blind design of STRs derived from distinct bHLH domains. Adapted from ref 690.

A second goal of this study was to determine whether the inherent modulatory properties found in natural bHLH domains (of which there are >100 in the human proteome)⁶⁹¹ could be retained in the STR architecture. Crystal structures of MAX-derived STR homodimerized and bound to E-box DNA confirmed the intended, non-natural “sandwich cross-dimer” fashion, yet faithful retention of overall domain register and specific side chain interactions with target nucleotides (Figure 51C,D). This shows the synthetic STR architecture mimics the DNA-binding specificity and structure of native bHLH domains and can be adapted to target other sequences. We therefore used the modular STR structure to recapitulate the binding activities of two other intractable oncogenic TFs, TFAP4 and OLIG2. Using a structure-blind design, we synthesized and ligated Z- and B-helices based on the primary sequence of TFAP4 or OLIG2 and showed that model STRs can approach or even exceed the DNA-binding specificity of native TF domains (Figure 51E).⁶⁹⁰

Recently, we extended the STR platform approach to mimic bZIP TFs, starting with the stress-responsive TF X-box binding protein 1 (XBP1).⁶⁹² While bZIP TFs bind DNA through homo- or heterodimerized helices that intercalate the major groove, the bHLH-derived binding approach deviates substantially enough from that of bZIPs to prevent bHLH-derived STRs from mimicking a bZIP TF. Thus, we developed a bZIP-mimicking STR architecture that involved (i) identification of the minimal, linear basic and leucine zipper domain residues required for DNA binding after (ii) covalent ligation of two DNA binding helix monomers (Figure 52). We further found that modification of helix–helix contact residues and incorporation of secondary structure-stabilizing macrocycles was necessary to create high affinity and high specificity XBP1-derived mimetics. Beyond development of a bZIP-derived architecture, there were several notable biochemical and biological discoveries in this work. First, we recognized that the target motif for XBP1—the UPRE—encompasses the canonical hypoxia-response element (HRE) motif bound by HIF1α and HIF2α in response to low oxygen levels (Figure 52A). Thus, we reasoned that an XBP1-derived STR should be able to bind and regulate hypoxia-regulated target genes and potentially oppose the action of oncogenic HIF proteins (Figure 52C). Indeed, we demonstrated that a lead XBP1-derived STR retains improved pharmacologic properties and is able to directly compete with hypoxia-dependent target gene binding (via ChIP-qPCR and ChIP-seq) and gene activation through global RNA-seq analyses. Importantly, these STRs could be delivered via intravenous injections where they modulated hypoxia-dependent gene expression in established 4T1, triple-negative breast cancer (TNBC) tumors. Repeated treatments did not exhibit overt toxicity or immunogenicity, suggesting in general that chemically stabilized proteomimetics may be designed without the worry of inherent immunogenic properties that may come from non-natural or non-human peptide sequences. Finally, hypoxia-targeting STRs showed significant inhibition of TNBC xenografts as well as spontaneous metastasis in syngeneic TNBC tumor models. Collectively, this work extends the synthetic biologic space to the bZIP family of TFs, demonstrates that TFs from distinct families can be targeted simultaneously and confirms that the pharmacologic stabilization of larger proteomimetics can engender in vivo activity for seemingly “undruggable” targets like XBP1 and HIF1α.⁶⁹²

Figure 52.

STRs derived from bZIP TF domains. A) Structures of bHLH TF HIF1α/ARNT and bZIP TF XBP1u DNA binding domains (DBD). The HRE DNA motif recognized by the bHLH transcription factor HIF1α is overlapped by the UPRE motif recognized by the bZIP transcription factor XBP1. Structures derived from PDB 4ZPK and 2H7H, respectively. B) Design and structural considerations for stabilized bZIP STRs derived from bZIP DBDs. C) bZIP STR-mediated HIF1α inhibition in the hypoxic tumor microenvironment prevents cancer cell survival and metastasis. Adapted from ref 692.

5.3. Summary and Outlook for Synthetic Proteomimetics and Biologics

Interactions driving many undruggable targets and processes mainly rely on highly dynamic and disordered surfaces for which small molecules and secondary structure mimetics struggle to attain affinity and specificity. Alternatively, proteins have precise recognition capabilities spanning large areas owing to tertiary or quaternary folding, but significant drawbacks including stability, cell permeability, and immunogenicity limit what can be feasibly targeted.⁶⁹³ Synthetic proteomimetics take advantage of the recognition abilities of proteins, but also have increased affinity and targetability, enhanced biostability and cellular uptake, and potentially improved pharmacology. The synthetic nature allows the reconstruction of stabilized higher-order domains to tune the biophysical properties and improve function. Altering the structural topology by site-specifically introducing backbone modifications or side chain interactions can recapitulate natural folding patterns or generate new structures not observed in nature yet retain ability to modulate biomolecular interactions. Although drug-like MCPs and secondary structure mimetics discussed above possess some advantages over larger proteomimetics for entering the clinic, these are not amenable to a wide variety of undruggable target classes. For example, certain PPIs, protein–nucleic acid interfaces and direct targeting of DNA motifs (a long-hypothesized approach with many potential advantages) are likely to require larger and potentially more polar binding footprints for efficacy. Thus, efforts to develop peptidomimetics or proteomimetics for these targets will require a balance of potency, specificity and pharmacokinetic and physicochemical properties. Stabilized proteomimetics may be the ideal, or perhaps the only solution for some targets. Future work to elucidate the mechanisms governing cellular uptake and pharmacologic properties for more examples of this class of molecules is needed to improve their design and hopefully contribute to their translation as therapeutics.

Although not a focus in this review, designer proteins based on structural scaffolds produced using genetically encoded methods are potentially promising tools for targeting undruggable classes. We recognize the immense potential of these engineered proteins, particularly through the genetic encoding of noncanonical amino acids, yet this is still restricted by the orthogonality and number of modifications that can be installed, thus limiting natural and engineered proteins in biological applications. Advances in synthetic approaches such as microwave-assisted solid phase synthesis, native chemical ligation, and alternative bioconjugation strategies, will allow these engineered scaffolds such as affibodies to be synthetically improved after selection strategies.⁶⁹⁴ Allied advances in computational de novo protein design will provide novel synthetic scaffolds and elucidate proteomimetic structures capable of disrupting previously intractable interactions.^695,696 The synthetic modularity associated with improved biophysical properties afforded to synthetic proteomimetics present exciting opportunities for this emerging class of molecules.

6. SUMMARY AND PERSPECTIVES

The ability to target truncal drivers of disease ultimately begins and ends with the ability to discover drug-like molecules that act on specific targets. While small molecules and biologics represent the two largest classes of therapeutics, a molecular middle ground presents immense promise for new chemical probe and therapeutic development. The peptidomimetics covered in this review have evolved from the historic success of bioactive peptides like insulin,⁶⁹⁷ the ~100 FDA approved peptide-based drugs, and the recent blockbuster status of “designer” peptides for extracellular targets like GLP-1 agonists.⁶ However, the hard-wired physical and chemical properties of natural peptides and biologics largely precludes their utility to target intracellular undruggable biomolecules.

Here, we have focused on four overlapping classes of synthetic, stabilized peptide- and protein-derived mimetics that draw upon structural and functional properties of small molecules and biologics. The combination of natural and non-natural amino acids with biorthogonal chemistry has enabled the development of molecules with augmented properties needed for therapeutics, namely: target affinity, specificity, metabolic stability, cell membrane permeability, and in vivo bioavailability. The unique ability to recognize extended, diverse surfaces with high affinity and specificity and ease of functionalization for enhancing desired properties situate this class of compounds suitably for targeting challenging biomolecules. This review covers a broad list of targets with some containing examples represented by most classes highlighted here, such as β-catenin and p53/MDM2, while others like small GTPases are overrepresented by MCPs and side chain stabilized peptides (Table 1). In the case of p53/MDM2, many approaches have been utilized to recapitulate the interacting domain of p53 to disrupt MDM2 binding. In general, mimetics stemming directly from native protein structures (e.g., MCPs and side chain stabilized peptides) displayed greater affinity and inhibition in the single-digit nanomolar range compared to non-natural mimetics, although these still achieved mid-to-high nanomolar potency. However, non-natural backbones generally confer increased resistance to proteolysis even compared to stapled counterparts,⁵²⁹ suggesting there may be a trade-off between potency, stability, and cell permeability that should be considered when designing and pursuing architectures for developing chemical probes and therapeutics.

Table 1.

Representative Undruggable Protein/Pathway Targets and Exemplary Application to Create Stabilized Peptidomimetics or Proteomimetic Modulators

Target	Biological Significance or Role	Target Mechanism of Action	Macrocyclic Peptides	Side Chain Stabilized Peptides	Non-natural Peptidomimetics	Synthetic Proteomimetics
Amyloid proteins	Fibrillar protein	Prevent aggregation and fibril formation	252		545, 596, 598, 616-620	663
BCL-2 family	Apoptosis regulator	Promote pro-apoptotic programs	201	383-391	525-529, 608, 609, 621, 622, 628
β-catenin	Transcription factor	Inhibition of Wnt signaling pathway	196	350, 354, 399, 400, 456, 474	578, 595	670
c-MYC	Transcription factor	Prevent Max dimerization; E-box DNA binding	138, 139		611	683, 685-690
ERα	Transcription factor	Block agonist functon; co-activator binding; DNA binding		325, 343, 465, 469, 481, 485, 486
Gα G-proteins	GTPase subunit protein	Prevent nucleotide exchange and enzymatic activity	136, 155, 162-164, 265
HDAC family	Histone deacetylase	Inhibit enzymatic lysine deacetylase activity	94, 132, 133, 214	342
HIF1α	Transcription factor	Prevent ARNT dimerization; HRE DNA binding	285, 286	415, 416	612, 614	675, 692
KDM family	Histone demethylase	Inhibit enzymatic lysine demethylase activity	243, 262-264
NRF2	Transcription factor	Prevent KEAP1-mediated degradation of NRF2	215, 218	357-363
NEMO	Inhibitor of IKK kinase	Inhibit activation of NF-κB pathway	92, 156			667
NOTCH complex	Transcription factor	Inhibit MAML co-activator binding; NOTCH-CSL binding; DNA binding		396
p53	Transcription factor	Prevent E3 ligase-mediated degradation of p53; promote stability and DNA binding of mutant p53	119, 176, 238	401-413, 465, 473, 475-478, 481, 493, 498	519, 521-524, 540, 556, 582, 586, 594, 603, 613, 628
Ras family	Small GTPase	Inhibit enzymatic activity	150-154, 203, 206-208, 211, 212, 251, 266, 284	418, 419		669
TEAD proteins	Transcription factor	Inhibit YAP/YAZ/TEAD interactions and Hippo pathway signaling	74, 190	456		673

While tremendous progress has been made across these categories, and molecules in these classes have even entered human clinical trials, there are still several challenges and horizons in the field to be addressed. First, the overarching limitation for many of these molecules is the fact the larger polar surface area, and therefore hydration, of peptides and protein mimetics limits their passive diffusion through cellular membranes, which is required for intracellular targets. Additional experimental and computational data sets should continue to refine our ability to hone-in on regions of molecular space that are likely to result in more permeable molecules, yet these regions are likely to be distinct between MCPs and synthetic biologics, for example. For molecules that may need to rely on active cellular uptake pathways, such as macropinocytosis and other endocytosis related mechanisms, the ability to properly traffic and access the cytosol is of paramount importance (Figure 53). For one, developing assays that quantitatively measure productive cellular uptake and intact distribution to specific cellular compartments can help in this regard. Methods like the CAPA assay⁶⁹⁸ and NanoClick⁶⁹⁹ assays represent recent advances that complement imaging-based approaches. Many studies, including our own recent work with synthetic mimics of TF DNA binding domains^690,692 have highlighted the importance of concurrent survival of cell permeable species, as many seemingly permeable peptides are rapidly degraded in cells and would thus be inactive.

Figure 53.

Designer peptides and peptidomimetics overcome limitations of unmodified peptides for modulating intracellular targets, such as transcription factor-mediated protein-protein and protein- DNA interactions.

Taking a page from the focus in macrocyclic peptide optimization, work in optimizing hydrocarbon stapled peptides is likely to advance to more active compounds through focused tracking of the mechanism(s) underlying cell permeability as well as the efficiency of this process. For example, several reported stapled peptides that are active in cells and in vivo utilize active uptake pathways, which then requires monitoring of compound stability within and through the endolysosomal pathway. Alternatively, trying to identify molecules that are permeable via passive membrane diffusion is more likely to bring cellular activity in line with biochemical potency. However, this may require the continued development and deployment of larger scale screening technologies, predictive computational models and identification of more compact binding epitopes. Balancing these properties and choosing which may be possible for a given target of interest will and should remain at the center of stabilized peptide and proteomimetic medicinal chemistry efforts and hopefully identification of development candidates and therapeutics.

Beyond cell permeability, recent research has begun to elucidate the necessary properties for bioavailability and efficacy in vivo;⁶⁵ however, much more work is needed to establish general rules and guidelines for effective designer peptide and peptidomimetic therapeutics.³⁰³ This includes looking into pharmacokinetic profiles in various models to elucidate serum stability and biodistribution, as well as determining optimal delivery routes and vehicles.⁷⁰⁰ The advances of nanoparticle delivery for nucleic acids has rapidly expanded the effective reach of genetically encoded modalities recently, and it is likely that opportunities exist to augment the in vivo pharmacokinetic profiles of stabilized peptides and proteomimetics targeting undruggable targets.⁷⁰¹

Another area of exploration could come in expanding the current target landscape for designer peptide and protein mimetics, as well as their mechanism of action. While many targets have been reviewed here, there is a common refrain of just a few prototypical protein targets across classes, such as the p53/MDM2 interface and the BCL2-family of apoptotic regulators (Table 1). This makes sense as these were early targets for the space; however, there is a large universe of undruggable targets and target biomolecules to be explored. For example, a less explored class of undruggable targets are epigenetic modifying proteins such as histone deacetylases and DNA methyltransferases. While a few peptidomimetic analogs targeting this class are reviewed here, such as the HDAC family and other chromatin regulators, these are mainly designs within MCPs or side chain stabilized peptide classes. This is not entirely unexpected due to the accessibility for these designs compared to wholly unnatural mimetics or larger synthetic biologics. However, expanding the target scope for these latter classes may yield molecules with better activity, which has been observed for targeting diverse classes of TFs.^{599,650,690,692} In this vein, further expansion to RNA and DNA as targets for upstream regulation of protein expression or activation of entire gene programs represents an exciting and perhaps fruitful landscape for these modalities. In parallel, the ability to target larger protein surfaces presents opportunities to create protein stabilizers, molecular glues and bifunctional modalities that could enhance the efficacy and scope of these molecules. Additionally, the capabilities of peptido- and proteomimetics extend beyond highly selective target interactors, and recent designs for generating artificial protein structures and biocatalysts has garnered much interest.^702,703 Enhancing designer peptidomimetics with additional functionality may serve to create chemical probes and therapeutics with a wide range of functions to modulate currently undruggable targets and pathways.

Finally, dramatic advances in structural biology and computation,⁷⁰⁴ for example with AI and ML modeling, have tremendous potential for the discovery and selection of target interfaces as well as the guided optimization of synthetic designer peptide- and protein-mimetics. These approaches coupled with synthetic stabilization, optimization and experimental validation of peptido- and proteomimetic constructs greatly improve upon the therapeutic potential of these modalities. Allied advances in computation and modeling to predict folding patterns and ligandability of biomolecule surfaces to elucidate targetable areas will help jumpstart the development of peptidomimetic ligands.^705,706 Alongside evergrowing advances in other therapeutic modalities, the molecules reviewed here hold tremendous promise to further reclassify and ultimately overcome a long list of “undruggable” targets in disease.

ABBREVIATIONS

ACHC

aminocyclohexanecarboxylic acid

ACN

acetonitrile

ACPC

aminocyclopentanecarboxylic acid

ADT

2-aryl-4,5-dihydrothiazole

AI

artificial intelligence

Aib

α-aminoisobutyric acid

AKAP

A-kinase-anchoring protein

AKB

protein kinase A binding domain

Akt

protein kinase B

Ala

Alanine

AML1-ETO

acute myeloid leukemia 1 protein-ETO fusion protein

Aoc

aminooctanoic acid

APP

amyloid precursor protein

Arg

arginine

ARNT

aryl hydrocarbon receptor nuclear translocator

ASF1

antisilencing function 1 protein

Asp

aspartic acid

ATC

4-amino(methyl)-1,3-thiazole-5-carboxylic acids

BAD

BCL-2 associated agonist of cell death

BAK

BCL-2 homologous antagonist killer protein

BCL-2

B-cell lymphoma 2

BCL9

B-cell CLL/lymphoma 9 protein

BCL-X_L

B-cell lymphoma extra large

BCP

bicyclic peptide

BEAS-2B

human bronchial epithelium cell line, non-tumorigenic

BH3

BCL-2 homology domain 3

bHLH

basic helix–loop–helix

BID

BH3 interacting-domain death agonist

BIM

BCL-2 interacting mediator of cell death

Boc

tert-butyl carbamate

B-Raf

serine/threonine protein kinase

Bts

benzothiazole-2-sulfonyl

bZIP

basic leucine zipper

CAL

CFTR-associated ligand

cAMP

adenosine 3,5-cyclic monophosphate

CBP

CREB binding protein

CD

circular dichroism

CD137

tumor necrosis factor receptor

CDR

complementarity determining regions

CFTR

cystic fibrosis transmembrane conductance regulator

CHD

cross-linked helix dimers

ChIP

chromatin immunoprecipitation

CHIP

C-terminus of Hsc70 interacting protein

cIAP1

cellular inhibitor of apoptosis protein 1

CPP

cell-penetrating peptide

CSL

CBF1, suppressor of hairless, Lag-1 (transcription factor in the Notch pathway)

CTAD

C-terminal transactivation domain

CtBP

C-terminal binding protein transcriptional repressor

CuAAC

copper catalyzed azide alkyne cycloaddition

Cys

cysteine

Dap

(S)-2,3-diaminopropanoic acid

DARPins

designed ankyrin repeat proteins

DBD

DNA binding domain

DCM

dichloromethane

Dcp2

mRNA decapping enzyme

Ddz

α,α-dimethyl-3,5-dimethoxybenzylcarbonyl

DEL

DNA-encoded library

DEP1

receptor-like protein tyrosine phosphatase

DIAD

diisopropyl azodicarboxylate

DIC

N,N’-diisopropylcarbodiimide

DIPEA

N,N’-diisopropylethylamine

DMF

dimethylformamide

DMSO

dimethyl sulfoxide

DNA

DNA

DTS

DNA templated synthesis

E-Box

enhancer box

ECRV

E-cadherin region V

EED

embryonic ectoderm development protein

EGFR

epidermal growth factor receptor

ELISA

enzyme linked immunosorbent assay

EphA2

Ephrin type-A receptor 2

Eps15

epidermal growth factor receptor pathway substrate 15

ERK

extracellular signal regulated kinase

ERα

estrogen receptor alpha

EZH2

enhancer of zeste homologue 2

FIP

Rab11-family interacting proteins

FIT

flexible in vitro translation

FITC

fluorescein isothiocyanate

Fmoc

fluorenylmethoxycarbonyl

FP

fluorescence polarization

GABARAP

GABA type A receptor associated protein

Gag

group-specific antigen

GAP

GTPase activating protein

GEM

guanine-nucleotide exchange modulator

Gln

glutamine

GLP-1

glucagon like peptide 1

Glu

glutamate

Gly

glycine

Gly-mal

glycine maleimide

gp41

glycoprotein 41

GPCR

G protein coupled receptor

GTP

guanosine-5′-triphosphate

HBS

hydrogen bond surrogate

HBTU

hexafluorophosphate benzotriazole tetramethyl uronium

HCT116

human epithelial cell line

HCTU

O-(1H-6-Chlorobenzotriazole-1-yl)-1,1,3,3-tetramethyluronium hexafluorophosphate

HDAC

histone deacetylase

hDM2

human double minute 2 protein (E3 ubiquitin ligase)

HER3

human epidermal growth factor receptor 3

hIAPP

human islet amyloid polypeptide

HIF1α

hypoxia induced factor 1 subunit alpha

HIF1β

hypoxia induced factor 1 subunit beta, encoded by ARNT

HIV-1

human immunodeficiency virus

HOBt

hydroxybenzotriazole

HPLC-MS

high performance liquid chromatography - mass spectrometry

HR2

heptad repeat 2

HRE

hypoxia-response element

hTEAD

human transcriptional enhanced associate domain

ICAT

inhibitor of β-catenin and TCF-4 protein

ICN1

intracellular domain of Notch1

IDE

insulin degrading enzyme

IDOL

inducible degrader of low-density lipoprotein receptor (LDLR)

IDR

intrinsically disordered region

IEDDA

inverse electron demand Diels–Alder

Ipa

isophthalic acid

IKK-β

inhibitor of nuclear factor kappa B subunit beta

IL-1

interleukin 1

IN

integrase

JM

juxtamembrane

KDM7

lysine demethylase 7

KEAP1

Kelch-like ECH-associated protein 1

KOTMS

potassium trimethylsilanolate

K-Ras

Kirsten rat sarcoma virus

LC-MS

liquid chromatography-mass spectrometry

LEDGF

lens epithelium-derived growth factor

Leu

leucine

Lys

lysine

mAb

monoclonal antibody

MAML1

mastermind like transcriptional coactivator 1

MAR

matrix associated region

MAX

MYC-associated factor X

MCD

mast cell degranulation

MCL-1

myeloid leukemia cell differentiation protein

MCP

macrocyclic peptides

MDA-MB-231

triple negative breast cancer cell line

MDM2

mouse double minute 2

MEG2

protein tyrosine phosphatase

Met

methionine

ML

machine learning

Mmt

monomethoxytrityl

MOrPH-PhD

macrocyclic organo-peptide hybrid phage display

MP

microporous

mRNA

messenger ribonucleic acid

MT1-MMP

membrane type 1 - matrix metalloproteinase

MTA1

metastasis-associated protein 1

MYC

bHLH transcription factor, oncogene

MyD88

Myeloid differentiation primary response 88

N2B

NHR2-binding motif

ncAA

noncanonical amino acids

NCKAP1

NCK associated protein 1

NEMO

NF-kappa-B essential modulator

NFAT

nuclear factor of activated T cells

NHR2

Nervy homology two domain

NHR-TF

nuclear hormone receptor transcription factor

NK-κB

nuclear factor kappa B

NMR

nuclear magnetic resonance

iNOS

inducible nitric oxide synthetase

nNOS

neuronal nitric oxide synthetase

NOTCH1

neurogenic locus notch homologue protein 1

NOXA-2

phorbol-12-myristate-13-acetate-induced protein 1 (pro-apoptotic BH3 only protein)

NRF2

nuclear factor erythroid 2-related factor

NRPS

nonribosomal peptide synthetase

OBOC

one bead one compound

OBTC

one bead two compound

OHM

oxopiperazine compounds

OLIG2

oligodendrocyte transcription factor

OSu

oxysuccinimide

p300

histone acetyltransferase

p53

tumor suppressor

PA-1

ovarian tetracarcinoma cell line

PANC-1

pancreatic carcinoma cell line

pAuNP

poly-l-lysine coated gold nanoparticle

PBS

phosphate buffered saline

PCSK9

proprotein convertase subtilisin/kexin type 9

PDB

protein data bank

PDZ

structural domain

PEPTIC

peptide exploration platform with Tag-free intramolecular chemistry

Phe

phenylalanine

PKA

protein kinase A

PKC

protein kinase C

PLSR

partial least-squares regression

PNA

peptide nucleic acid

POI

protein of interest

PolyUb

polyubiquitin

PPI

protein–protein interaction

PPIA

peptidylprolyl cis–trans isomerase

PRL2

phosphatase of regenerating liver 2

Pro

proline

PROTAC

proteolysis targeting chimera

PS

polystyrene

PSA

polar surface area

PSD-95

postsynaptic density protein

PTP1B

protein tyrosine phosphatase

PURE

protein synthesis using recombinant elements

Rab

Ras associated binding protein

RaPID

random nonstandard peptide integrated discovery

RCM

ring closing metathesis

REU

Rosetta energy units

RGS4

regulator of G protein signaling 4

RMSD

root-mean-square deviation

RNA

ribonucleic acid

RRM

RNA-recognition motifs

RSV

respiratory syncytial virus

RT-PCR

real time polymerase chain reaction

RuAAC

ruthenium catalyzed azide alkyne cycloaddition

SAH

stabilized α helix

SAR

structure activity relationship

SBDD

structure-based drug design

Seq

sequencing

Ser

serine

SH2

Src homology 2

SHA

shavenoid

SICLOPPS

split intein circular ligation of peptides and proteins

SIRT2

sirtuin 2

SJSA-1

osteosarcoma fibroblast

SMAD7

Mothers against decapentaplegic homologue 7

SOS1

son of sevenless homologue 1

SPAAC

strain promoted azide alkyne cycloaddition

SPPS

solid phase peptide synthesis

SPR

surface plasmon resonance

SPSB2

SPRY domain-containing SOCS box protein 2

SRC2

steroid receptor coactivator 2

STAT3

signal transducer and activator of transcription 3

STING

stimulator of interferon genes adaptor protein

STR

synthetic transcriptional repressors

SUPR

scanning unnatural protease resistant

tACBC

trans-2-aminocyclobutanecarboxylic acid

TAMM

2-((alkylthio)(aryl)methylene)malononitrile

TAR

transactivation binding response

TATA

1,3,5-triacryloyl-1,3,5-triazinane

TBAB

N,N′,N”-(benzene-1,3,5-triyl)-tris(2-bromoacetamide)

TBMB

1,3,5-tris(bromomethyl)benzene

TCEP

tris(2-carboxyethyl)phosphine

TCF

T-cell factor

TEAD1

transcriptional enhancer factor 1

TET1

ten-11 translocation methylcytosine dioxygenase1

TF

transcription factor

TFA

trifluoroacetic acid

TFAP4

transcription factor AP-4

TGF-β

transforming growth factor beta

THF

tetrahydrofuran

Thr

threonine

TIS

triisopropylsilane

TLR

toll-like receptor

TNBC

triple negative breast cancer

TNKS

tankyrase

Trp

tryptophan

TSG101

tumor susceptibility gene

ULM

U2AF ligand motifs

uPA

urokinase-type plasminogen activator

uPAR

uPA receptor

UPRE

unfolded protein response element

UTR

untranslated region

Val

valine

VDR

vitamin D receptor

VEGF

vascular endothelial growth factor

vFLIP

viral FLICE inhibitory proteins

VGL4

vestigial-like protein 4

WNT

wingless/Int1 signaling pathway

XBP1

X-box binding protein 1

XIAP

X-linked inhibitor of apoptosis protein

YAP1

Yes1 associated transcriptional regulator

γ-AA

γ-substituted N-acylated-N-aminoethyl amino acids

Biographies

Colin S. Swenson received his B.S. in Chemistry from the University of Utah in 2015. He then moved to Emory University where he completed his Ph.D. in Chemistry in 2020 with Prof. Jennifer Heemstra investigating amphiphilic peptide nucleic acids as stimuli-responsive biomaterials. In 2021, he joined the lab of Prof. Raymond Moellering at the University of Chicago as a NIH NRSA postdoctoral fellow, where he focuses on developing small molecule probes and synthetic biologics to study and target transcription factor activity.

Gunasheil Mandava received his combined B.S./M.S. degree in Molecular Biophysics and Biochemistry from Yale University as magna cum laude and earned the departmental award, the Sigler Prize, in 2021. He worked with Prof. Nikhil Malvankar to investigate the usage of electrically conductive bacterial nanowires in biomaterial applications. During his gap years, he continued his work in Prof. Nikhil Malvankar’s lab before moving to Prof. Sarah Slavoff’s lab in 2022 to design bicyclic peptides as viable PROTAC warheads to target undruggable proteins. He joined Prof. Raymond Moellering’s lab in 2023 as a graduate student in chemistry at the University of Chicago, where his current focus is to improve pharmacological properties of synthetic biologics that target transcription factor activity.

Deborah M. Thomas received her B.S. in Chemistry from Florida Atlantic University in 2019. She then moved to the University of Chicago in 2020 and is currently pursuing a Ph.D. in Chemistry with Prof. Raymond E. Moellering. Her work focuses on the development of novel chemistries for peptide stabilization, stabilized peptide therapeutics for transcription factors, and theranostic peptides for tumor imaging and treatment.

Ray Moellering is a Professor of Chemistry at the University of Chicago. He obtained Bachelor’s degrees in Chemistry and Biochemistry and Molecular Biophysics from the University of Arizona. He then earned a Ph.D. in Chemistry at Harvard University with Prof. Gregory L. Verdine as an American Association for Cancer Research Centennial Fellow, followed by postdoctoral training with Prof. Benjamin F. Cravatt, III as a Damon Runyon Postdoctoral Fellow at The Scripps Research Institute. Prof. Moellering started his independent program at UChicago in 2015, where his laboratory is focused on developing novel chemical probes and complementary technologies to expose and exploit novel signaling mechanisms in diseases like cancer, diabetes, and immunological disorders. Dr. Moellering has contributed to the development of stabilized peptide and protein mimetics as chemical probes and prototype therapeutics for diverse undruggable proteins in cancer.