De novo mapping of α-helix recognition sites on protein surfaces using unbiased libraries

The α-helix is one of the most common protein surface recognition motifs found in nature, and its unique amide-cloaking properties also enable α-helical polypeptide motifs to exist in membranes. Together, these properties have inspired the development of α-helically constrained (Helicon) therapeutics that can enter cells and bind targets that have been considered “undruggable”, such as protein–protein interactions. To date, no general method for discovering α-helical binders to proteins has been reported, limiting Helicon drug discovery to only those proteins with previously characterized α-helix recognition sites, and restricting the starting chemical matter to those known α-helical binders. Here, we report a general and rapid screening method to empirically map the α-helix binding sites on a broad range of target proteins in parallel using large, unbiased Helicon phage display libraries and next-generation sequencing. We apply this method to screen six structurally diverse protein domains, only one of which had been previously reported to bind isolated α-helical peptides, discovering 20 families that collectively comprise several hundred individual Helicons. Analysis of 14 X-ray cocrystal structures reveals at least nine distinct α-helix recognition sites across these six proteins, and biochemical and biophysical studies show that these Helicons can block protein–protein interactions, inhibit enzymatic activity, induce conformational rearrangements, and cause protein dimerization. We anticipate that this method will prove broadly useful for the study of protein recognition and for the development of both biochemical tools and therapeutics for traditionally challenging protein targets.

Recent advances in identifying human disease targets have not been matched by advances in the ability to drug these targets. This actionability gap is largely due to the fact that neither of the two main classes of approved therapeutics – biologics and small molecules – can simultaneously address target accessibility and selective target engagement. Biologics, despite an impressive ability to engage diverse target proteins, are largely restricted to an extracellular operating theater, as their size and polarity render them unable to cross biological membranes. Small molecules, in contrast, can access the intracellular space, but cannot bind with high affinity and specificity to the vast majority of proteins that are found there (1).This disconnect between the ability to identify disease targets and the ability to drug them with high strength and specificity has created an impetus to develop new classes of drugs – ones that can engage intracellular proteins that lack the deep hydrophobic pocket ordinarily required for small-molecule binding. In nature, such “undruggable” proteins are often targeted with macrocyclic molecules, frequently peptidic in structure, whose large size compared with small molecules enables them to bind with high affinity and specificity to protein surfaces.Significant efforts have been made to elucidate the mechanisms of cell entry for these natural products, which possess molecular weights of 700 to 1,200 Da or higher, well beyond the typical range for cell penetration in small-molecule drug discovery (2). While the mechanisms of cell entry are complex and vary from molecule to molecule, a substantial body of research on peptidic macrocycles has highlighted the importance of desolvating amide protons and reducing their exposure to the membrane interior as a key driver in passive, thermal diffusion across the lipid bilayer (2, 3) – a phenomenon we refer to as amide-proton cloaking. The amide proton, present between every residue in a polypeptide chain, is highly electropositive and forms a strong hydrogen-bonding interaction with water. This poses a substantial hurdle for membrane permeability, since tightly bound solvent water molecules must be shed prior to entering the lipid bilayer. Exposed amide groups incur a further energetic penalty upon membrane entry due to unfavorable electrostatic interactions with the low-dielectric environment of the membrane interior. Consequently, most peptides and proteins are unable to cross membranes.For peptide macrocycles that are able to permeate the membrane, these problematic amide protons are typically removed either by replacing the amide with an ester, replacing it with a methyl group, or cloaking it from solvent water through the formation of intramolecular hydrogen bonds between the amide proton groups and a hydrogen bond-accepting group elsewhere in the molecule, often a carbonyl. Indeed, the paradigmatic example of a natural peptide macrocycle that exhibits robust cytosolic exposure, cyclosporine A (CsA), employs both N-methylation and cloaking through transannular hydrogen bonding (4). Extensive work by several research groups has shown that these strategies can be applied as design principles to endow artificial macrocycles with the ability to cross membranes (5–7).In the context of folded proteins, nature has offered an alternative structural solution to the problem of amide proton cloaking: the α-helix, a protein secondary structure that is defined by repeating intramolecular hydrogen bonds between the amide proton group of one residue and the carbonyl of the amino acid located four residues N terminal to it. The intrinsic ability of α-helices to cloak their own amide protons explains their widespread prevalence in natural transmembrane proteins (8). Nuclear-encoded transmembrane proteins in eukaryotes are almost exclusively α-helical, and the only alternative transmembrane fold found in nature is the bacterially derived β-barrel, a helical structure that also cloaks amide protons via an intramolecular hydrogen bonding network, albeit in a significantly larger structure than single α-helices that is impractical for the development of synthetic drugs.Just as CsA has served as the inspiration for the design of mimetic head-to-tail cyclized peptide ligands, so have proteinaceous α-helices inspired efforts to recapitulate nature’s design features in small, synthetic, α-helically constrained peptides (Helicons) that are hyperstabilized through the incorporation of a structural brace, also known as a “staple” (9–12). One of these, the all-hydrocarbon staple formed by ring-closing metathesis, has been extensively studied and is the basis for a drug candidate that targets the challenging proteins MDM2 and MDMX, currently undergoing Phase II clinical trials (13, 14).Rational design of Helicons is difficult given the inability to systematically define the α-helix binding sites on a protein’s surface, and to identify Helicons that bind to those sites. This limitation has restricted research on Helicons to only those protein targets for which naturally occurring or previously characterized α-helical binders were known, with the Helicons generated from fragments of the known binders (3). Here, we report a rapid, high-throughput screening platform utilizing phage display that enables an unbiased mapping of the α-helical interactome of a given protein without any prior knowledge of its structure or known binding partners. We show that this platform is capable of identifying α-helix binding sites on the surfaces of a range of protein folds, including many for which no α-helical binders are known to exist. Helicons that bind these sites are able to impact diverse protein functions, including inhibiting protein–protein interactions, inhibiting enzymatic activity, inducing dimerization, and inducing conformational changes. Analysis of 14 high-resolution crystal structures of Helicon–protein complexes across six different protein domains reveals a range of binding modes, all of which are “side-on”, i.e., mediated exclusively by Helicon side-chains rather than involving main chain amide interactions. This screening platform significantly expands the universe of proteins that can be bound by Helicons, and furthers the pursuit of targeting undruggable proteins.