Beyond 20 in the 21st Century: Prospects and Challenges of Non-canonical Amino Acids in Peptide Drug Discovery

Abstract

Life is constructed primarily using a toolbox of 20 canonical amino acids—relying upon these building blocks for the assembly of proteins and peptides that regulate nearly every cellular task, including cell structure, function, and maintenance. While Nature continues to be a source of inspiration for drug discovery, medicinal chemists are not beholden to only 20 canonical amino acids and have begun to explore non-canonical amino acids (ncAAs) for the construction of designer peptides with improved drug-like properties. However, as our toolbox of ncAAs expands, drug hunters are encountering new challenges in approaching the iterative peptide design–make–test–analyze cycle with a seemingly boundless set of building blocks. This Microperspective focuses on new technologies that are accelerating ncAA interrogation in peptide drug discovery (including HELM notation, late-stage functionalization, and biocatalysis) while shedding light on areas where further investment could not only accelerate the discovery of new medicines but also improve downstream development.

Peptides have been referred to as the “Goldilocks” chemical modality due to their intermediate size which combines favorable attributes of both small molecules and biologics, such as high target specificity, absence of off-target effects, and differentiated pharmacokinetic profiles. Yet only modest gains have been observed in FDA approvals despite a rich legacy of peptides impacting human health for over 100 years, from insulin to vasopressin and, most recently, tirzepatide.^1,2 One reason for this discrepancy is that the majority of approved peptide drugs are inspired by native peptide structures and are linear in nature with high canonical amino acid (AA) content, resulting in poor gastrointestinal stability and low permeability/bioavailability that often restricts the route of administration to intravenous or subcutaneous. However, a recent Perspective in ACS Med. Chem. Lett. from Blanco et al.³ highlighted the creativity and innovation of “new generation” macrocyclic peptides (MPs, 5–14mers) that are showing great promise in clinical studies owing to their improved biopharmaceutical properties, thereby enabling modulation of protein–protein interactions (PPIs), once considered undruggable.⁴ While macrocyclization addresses some of the liabilities associated with linear peptides, expanding to the vast space beyond canonical AAs provides a means to selectively access both intra- and extracellular targets while fine-tuning a range of properties such as solubility, metabolic stability, and importantly, oral bioavailability.⁵

The inspiration for applying non-canonical amino acids (ncAAs) to macrocyclic peptide drug discovery has origins in peptide natural products, specifically non-ribosomal peptides (NRPs) and ribosomally synthesized and post-translationally modified peptides (RiPPs).⁶ Notable examples include cyclosporin A (an NRP) and, most recently, darobactin A (a RiPP),⁷ where biological activity can be attributed to multifaceted, unique AA architectures (Figure 1A).⁸ While ncAAs are established in small molecule and peptidomimetic drug discovery,⁹ not until recently have advances in affinity-based hit-finding platforms, such as mRNA display, provided a means to generate >10¹² unique de novo ncAA-containing macrocyclic peptides (ncAA-MPs) in a single round of screening.¹⁰ Pioneering efforts by the Suga and Szostak laboratories have enabled ncAA incorporation into mRNA display screening libraries;^11,12 however, limitations on codon table modifications remain, providing rich opportunities for drug hunters to further optimize the hits to lead-like chemical matter. As one can imagine, the chemical space of ncAAs for medicinal chemists to explore can be limitless; nevertheless, inspiration from NRPs and RiPPs, such as cyclosporin A, darobactin A, and tryglysin A, have provided several common strategies to design and optimize peptidic structures (Figure 1B). For instance, d-configured building blocks, substitution at the N-, α-, or β-centers, or simply changing the side chain functionality can greatly enhance chemical diversity while improving the pharmaceutical properties.¹³⁻¹⁵ It should be noted that other modifications, such as cross-links or amide bond surrogates, can also add structural diversity by influencing the resulting three-dimensional conformation.¹⁶ Recently, macrocyclic peptides with origins in mRNA display have emerged from such institutions as Chugai and Merck, leveraging the engineered introduction of ncAAs for accessing either intra- or extracellular targets, respectively (Figure 1C). For Chugai, their first clinical macrocyclic peptide for intracellular rat sarcoma virus (RAS) inhibition was identified through incorporation of multiple N-substituted ncAAs, reducing the polar surface area and improving membrane permeability.¹⁷ In turn, Merck recently disclosed the structure of MK-0616, an oral proprotein convertase subtilisin/kexin type 9 (PCSK9) inhibitor, which in a recent Phase 2b study, significantly reduced LDL-C in patients with hypercholesterolemia.¹⁸ Potency and protease stability were obtained through introduction of a fluorinated tryptophan, d-Ala (AA2), α-Me-Pro (AA8), and cross-linking, ultimately yielding an oral MP inhibitor at a fraction of the size of current monoclonal antibody injectables.

Figure 1.

Nature serving as inspiration for peptide drug discovery (A), with ncAA classes from NRPs and RiPPs now appearing in promising peptide candidates with improved pharmaceutical properties (B, C).

Several promising macrocyclic peptides are on track to be the first approved drugs derived from mRNA display,¹⁹ resulting in a renewed focus on understanding the roles and impacts of ncAAs in drug discovery and development.⁵ However, the “peptide toolbox” for drug discovery is still being defined, and its understanding remains extremely limited compared to that of small molecules and biologics which continue to dominate FDA approvals.²⁰ For instance, both informatics and prediction software tend to focus primarily on either small molecule or biologics discovery, largely stemming from the historical lack of interest in peptides, coupled with their topological complexity and shortage of relevant training data for machine learning. And while ncAAs can be sourced commercially for initial structure–activity relationship (SAR) exploration, there are gaps in structural diversity, and availability becomes increasingly limited on larger scales, negatively impacting early development timelines. Moreover, current synthetic technologies, such as solid-phase peptide synthesis (SPPS), have historically employed excess AAs to improve synthetic efficiency, which exacerbates the cost of goods (COG) for ncAA-rich peptides. Thus, to seize the opportunity that peptides, both linear and cyclic, can bring to human health, we need to capitalize on ncAAs in peptide drug discovery, investing in informatics and synthesis to realize the full potential of this highly versatile modality.

Informatics and Prediction for ncAAs

Computational analysis and prediction are key aspects of drug discovery. While a wealth of tools exist for both small molecule and biologic modalities, ncAA-MPs do not necessarily fit into either category and thus require significant adaptations. For example, regarding textual representation, small molecules typically use simplified molecular-input line-entry system (SMILES) strings which use atoms and bonds as the primary units, whereas biologics are represented as a sequence of canonical AAs written in the standard single-letter codes (typically FAST-All (FASTA) format). Neither approach works well for ncAA-MPs, since the SMILES strings become far too long and complex, while FASTA single-letter AA codes only allow for linear structural representations using the standard 20 canonical AAs (Figure 2). However, the flexible hierarchical editing language for macromolecules (HELM) has proven to be an excellent fit.²¹ HELM can represent diverse ncAAs as simple, human-legible text symbols (analogous to canonical AA single-letter codes) as well as cross-linking or cyclization architectures. This simplified monomeric representation can reduce assembly errors commonly arising from manually misdrawing these large biomolecules that contain many atoms, stereocenters, and cross-links. Additionally, the monomeric representation is a convenient model for SAR analysis and representation. Despite these advances, there remains a need for further standardization of HELM in both symbol naming and processing.

Figure 2.

Textual representations for various chemical modalities. ncAA-MPs (center) are well-represented using HELM which concisely depicts ncAAs as well as cross-linking and cyclization architectures.

Sequence alignment is fundamental for protein analysis, as positional assignment is necessary to study SARs; however, performing these alignments is a non-trivial task for ncAA-MPs. For example, the similarity matrices, which are required for performing alignments, typically use well-established 20×20 monomer–monomer similarity matrices, such as BLOSUM (blocks substitution matrix) scores.²² However, for ncAAs, where the total count is limitless, the matrices become not only much larger but also highly dynamic. To address this issue, Merck has developed a method called Peptide Sequence Alignment (PepSeA) for ncAA-MPs using a dynamic monomer similarity matrix (Figure 3).²³ This has enabled downstream peptide SAR analysis using alignment and visualization tools along with sequence-based descriptors for machine learning. Sequence-based analysis is particularly useful to directly interpret the impact of single AA mutations through direct linear coefficients, simple generative estimators such as Free-Wilson analyses,^24,25 or alternatively, more complex machine-learning algorithms. One might conjecture that ncAA-MP sequence alignment could enable structure prediction, such as what is done in AlphaFold2 or RoseTTAFold,^26,27 but the dearth of molecules topologically similar to ncAA-MPs in the Protein Data Bank prohibits practical training and deployment of such deep-learning models at this time. Thus, structure determination is relegated primarily to experimental techniques or less reliable physics-based predictors.²⁸⁻³⁰

Figure 3.

Monomer similarity matrix using atom/distance pairs with chirality (APC/DPC) descriptors.^31,32 Subset of a large similarity matrix, which must be generated dynamically, to perform sequence alignments for ncAA-MPs. For example, dW is more likely to align to W (0.85 similarity) than to F (0.55).

Physics-based modeling of peptides is necessary for predicting properties of novel ncAA-MPs which lack sufficient experimental data to fuel machine learning. For typical physics-based models using atomic representation, molecules are parameterized distinctly as either proteins or small molecules, by simple heuristics but which may not natively handle ncAAs. In silico residue scans consisting of three-dimensional conformational searches using molecular force fields typically handle side chain substitutions well but struggle with backbone modifications. Molecular dynamics (MD)-based free energy calculations (free energy perturbation, thermodynamic integration, etc.) are optimized for minimal changes in framework—either small molecules with limited cores or proteins with minimally moving backbones; however, the conformationally flexible backbones of peptides and macrocycles can be highly non-trivial to predict with these relatively short simulation timescales. Potentially, this problem can be partially addressed by combining the binding enthalpy estimates with separate conformational preference estimates by utilizing solution-state MD simulations (vanilla MD or enhanced sampling to study the conformational behavior).^33,34 Still, there remains a need for robust unified physics-based binding free energy solutions for peptides.

These adaptations of biologic and small molecule informatics utilities, along with other custom informatics tools built for ncAA-MPs, are necessary to properly leverage the wealth of data generated by peptide drug discovery teams. However, peptide tools are often less mature; therefore, opportunities to maximize the unique aspects of informatics within this space are abundant.

Advances in Peptide Medicinal Chemistry: Leveraging Late-Stage Functionalization for ncAA Incorporation

The incorporation of ncAAs into peptides is typically approached via two methodologies: the first is through traditional SPPS, and the second is through late-stage functionalization (LSF) of fully assembled sequences. Fluorenylmethyloxycarbonyl (Fmoc)-based SPPS requires either commercial availability or preliminary synthesis of discrete Fmoc-protected AAs, which can be readily integrated through standard amide bond-forming protocols. However, commercial availability of diverse, readily accessible ncAAs is limited and often exorbitantly expensive. For instance, if we consider Fmoc-protected phenylalanine analogs, at the time of this Microperspective, there were ∼250 l-configured building blocks, ∼150 d-configured analogs, and ∼40 compounds containing homologated aryl side chains commercially available, with the number of available AAs inversely proportional to the price.³⁵ In addition to accessing diverse ncAAs for SAR investigations, as medicinal chemistry teams look to scale-up promising peptide leads for safety and pharmacokinetics studies, securing bulk quantities of Fmoc-ncAAs can be a non-trivial, costly, and time-consuming endeavor.

Alternately, a ncAA bearing a reactive handle can be incorporated during linear chain assembly, serving as a point of further diversification through LSF (Figure 4). An established technology used in small molecule drug discovery, LSF enables rapid and extensive chemical/structural diversity to be accessed from a single precursor molecule.³⁶ For peptide drug discovery, the advantages of pursuing LSF are even greater, as the precursor peptide can typically be obtained from commercially available building blocks. However, the biggest challenge remains in developing chemo-, regio-, and stereoselective LSF chemistries compatible with peptides and larger biopolymers both on-resin and in solution.

Figure 4.

Peptide LSF protocols to access ncAAs through diversification of aryl side chains (A, B) and installation of de novo functionality (C, D).

While native chemical ligation along with bioconjugation and cross-linking methodologies (click, ring-closing metathesis, etc.) have gained widespread adoption within the peptide community,⁵ LSF on peptides is still an emerging field. Recent developments highlight both on-resin and in solution methodologies and can be split into two general categories: those that incorporate a distal decoration on an existing AA side chain and those that allow for installation of de novo functionality in a customizable, modular fashion (Figure 4). Although the diversity of contributions in this space are vast, we will focus on highlighting a selection of representative LSF technologies reported over the past 12–24 months in this Microperspective. In 2022, Merck described the late-stage diversification of solid-phase-supported peptides containing p-bromophenylalanine residues using Suzuki–Miyaura cross-coupling chemistry (Figure 4A). This work nicely illustrated how leveraging the power of high-throughput experimentation (HTE) within SPPS can enable rapid access to multidimensional peptide libraries with expanded chemical space through distal modifications on aryl side chains.³⁷ Similarly, the Ackermann lab expanded the tryptophan LSF toolbox through a manganese-catalyzed C–H hydroarylation at the indole C2 position, which provided access to both ligation and macrocyclic peptide products (Figure 4B).³⁸ Lastly, Mallek et al. have described a palladium-mediated, chemoselective N-arylation of p-aminophenylalanine on a fully deprotected peptide in the presence of other aliphatic amino groups.³⁹

As we broaden our field of view and expand beyond chemistries that modify existing aryl AAs, progress has also been made in synthetic methodologies that allow access to diverse, de novo side chain functionalities. For instance, the Thomas laboratory reported on a nucleophilic substitution reaction via on-resin iodination of homoserine, followed by subsequent functionalization with aliphatic and aromatic thiols, pyridines, and carboxylic acids, showing great promise for HTE and combinatorial approaches.⁴⁰ More recently, the Xu lab described a visible-light-mediated deaminative hydroalkylation on a series of alkenes, including dehydroalanine (Dha), providing an efficient way to prepare β-alkyl-substituted ncAAs either as discrete building blocks or within an elaborated peptide under mild and catalyst-free conditions (Figure 4C).⁴¹ Common to many Dha modifications for ncAA synthesis is the lack of stereocontrol which can cause complex mixtures of epimers; however, new reports have shown that chiral catalysts can invoke higher stereoselectivity.^42,43

Recently, alkyl Katritzky salts have emerged as stable and versatile radical precursors, providing an alternative way to access diverse and customizable side chain functionality in peptides.⁴⁴ Conversion of lysine, and related AAs with primary amine-containing side chains, into the corresponding pyridinium salts involves straightforward, high-yielding, and robust chemistry. In addition, the resulting Fmoc-protected building blocks are often resistant to S_N2 and epimerization reactions and can be seamlessly incorporated into peptides through standard SPPS and trifluoroacetic acid (TFA) deprotection workflows. Openy et al. showcased the first example of utilizing a side chain Katritzky salt for on-resin peptide LSF, accessing diverse ncAAs via a photochemical Giese reaction using commercially available Michael acceptors.⁴⁵ With line of sight to readily transform abundant canonical AA feedstocks into unique ncAAs, an academic–industrial collaboration between Merck and the Watson group at the University of Delaware recently reported a cross-electrophile coupling between lysine-derived Katritzky salts and aryl halides to access aryl alanines and homologated derivatives (Figure 4D).⁴⁶ Through this work, facile access to LSF substrates was achieved, ultimately providing diverse, boutique side chain functionality on fully elaborated peptidic structures.

As investment in this space continues, we believe the next wave of peptide LSF chemistries should not only strive to identify unique reactive handles but also leverage canonical AA side chain functionality to access complementary reactivity. One could envision an aspirational future state where the rapid buildup of complexity occurs from a single parent peptide in a multidimensional format from distinct side chain vectors to produce large peptide libraries of vast chemical diversity. Ultimately, these LSF chemistries should allow us to continue to mimic and expand upon the structural features and functional group diversity, both aryl and alkyl, that we observe in NRPs/RiPPs.

Prospects for ncAA Scale-Up: Streamlined Biocatalytic Processes to Access ncAA Classes of Interest

In comparison to conventional small molecules, peptides typically possess higher production costs due to their larger size (>1000 Da) and longer synthetic sequences. To this end, many improvements have been documented for peptide synthesis on a manufacturing scale, with SPPS remaining the most common method, where excess AAs are usually required.⁴⁷ For ncAA-containing peptides, the cost of synthesizing each ncAA must also be considered, as delivering excess of synthetically complex ncAAs could significantly impact overall production costs. While there are efficient chemical processes to access some ncAAs, enzymatic processes are quintessential because they proceed with high stereocontrol and obviate protecting group manipulations, culminating in reduced step counts. Frances Arnold and co-workers recently shared a perspective on this topic, bringing attention to new biocatalytic methods for accessing ncAAs, including β-AAs, substituted prolines, and α-deuterated AAs.⁴⁸ Herein, we highlight recent biocatalytic transformations and remaining opportunities toward accessing β-Me-, β-OH-, and α-Me-branched α-AAs (Figure 5). These represent several examples of ncAA classes which are known to stabilize peptides to protease-induced amide hydrolysis, a particularly challenging barrier to overcome for the development of oral drugs. At the same time, they maintain a relatively small profile, which helps to minimally impact linear peptide chain assembly. These ncAA classes can also improve potency, conformational rigidity, and other drug-like properties.

Figure 5.

Established routes and existing opportunities toward β-Me-, β-OH-, and α-Me-substituted AAs of interest.

In addition to having the capacity to confer peptide stability, β-Me-branched α-AAs are of interest as the presence of the β-Me group can increase binding affinity due to an entropic gain from restricted bond rotation between the α-carbon and the β-carbon.¹⁵ Highly stereoselective routes to access both (2S,3S) and (2S,3R) β-Me-branched diastereomers are sought, as the optimal diastereomer is case-dependent. Indole-substituted β-Me-tryptophans constitute one subclass, and engineered variants of tryptophan synthase (e.g., PfTrpB 7E6) coupled with pyridoxal phosphate (PLP) cofactor have been reported to deliver diverse (2S,3S) β-Me-tryptophan analogs in one step from threonine and the requisite indole (Figure 5A).⁴⁹ Currently only a single example of accessing the complementary (2S,3R) β-Me-tryptophan exists, which proceeds through a PLP-dependent cascade from tryptophan and methyl iodide via α-AA transaminase (TA), α-keto acid methyltransferase (MT), halide methyltransferase (HMT), and S-adenosylhomocysteine (SAH).⁵⁰ Considering aromatic α-AAs beyond tryptophan, the Renata lab has recently engineered PLP-dependent aromatic AA aminotransferases (ArATs) to access (2S,3R) β-Me-arylalanines from ketoacids with broad scope (Figure 5B).⁵¹ A complementary enzymatic process to access (2S,3S) β-Me-arylalanines would be desirable yet remains elusive. For aliphatic AAs, (2S,3R) β-Me aliphatic AAs (Figure 5C) can be accessed by the aforementioned enzymatic cascade used to make (2S,3R) β-Me-tryptophan;⁴⁸ however, an enzymatic process to access (2S,3S) β-Me aliphatic AAs has yet to be reported.

While β-OH-branched α-AAs can impart benefits akin to the β-Me group, the hydroxy group can also participate in favorable intra- or intermolecular hydrogen-bonding interactions. PLP-dependent threonine aldolases (L-TA), threonine transaldolases (L-TTA), and serine hydroxymethyl transferases (SHMT) have each been utilized to access diverse β-OH aryl- and alkylalanines (Figure 5D).⁵² However, these reactions typically produce mixtures of (2S,3S) and (2S,3R) diastereomers. There have been recent engineering efforts to develop diastereoselective processes,⁵³ yet many opportunities remain to access single diastereomers in high diastereomeric excess and conversion, a necessity for manufacturing processes. On a final note, α-Me-branched α-AAs can also improve peptide stability while potentially influencing peptide secondary structure due to the location of the α-Me group on the peptide backbone. Biocatalytic processes to access this subclass would therefore be of value but are currently elusive.

Conclusion

As a scientific community, our understanding of how best to leverage ncAAs to access a seemingly endless source of peptide chemical diversity is still in its infancy. Computational tools can be employed to accelerate the design of ncAA-containing peptides, but significant adaptation and innovation are required in terms of both informatics and physics-based modeling. However, with investment in novel synthetic methodologies and manufacturing platforms for both ncAAs and peptides, we aspire to a future state in which accessing chemical diversity is not rate-limiting and peptide scale-ups proceed with costs and timelines akin to small molecules. Consequently, we must come together to truly capitalize on the opportunities that ncAA-rich peptides present to human health, utilizing industrial consortia and academic–industrial collaborations to address the challenges outlined in this Microperspective. Only time will tell how influential ncAAs will be in the discovery of new peptide therapeutics; nevertheless, we look forward to playing an active role on this exciting journey, with the ultimate goal of positively impacting human health well beyond the 21st century.

Glossary

Abbreviations

AA

amino acid

APC/DPC

atom/distance pairs with chirality

ArATs

aromatic AA aminotransferases

BLOSUM

blocks substitution matrix

COG

cost of goods

Dha

dehydroalanine

FASTA

FAST-All

Fmoc

fluorenylmethyloxycarbonyl

HELM

hierarchical editing language for macromolecules

HMT

halide methyltransferase

HTE

high-throughput experimentation

LSF

late-stage functionalization

L-TA

l-threonine aldolase

L-TTA

l-threonine transaldolase

MD

molecular dynamics

MPs

macrocyclic peptides

MT

methyltransferase

ncAA

non-canonical amino acid

ncAA-MP

non-canonical amino acid-containing macrocyclic peptide

NRP

non-ribosomal peptide

PCSK9

proprotein convertase subtilisin/kexin type 9

PepSeA

Peptide Sequence Alignment

PLP

pyridoxal phosphate

PPIs

protein–protein interactions

RAS

rat sarcoma virus

RiPP

ribosomally synthesized and post-translationally modified peptide

SAH

S-adenosylhomocysteine

SAR

structure–activity relationship

SHMT

serine hydroxymethyl transferase

SMILES

simplified molecular-input line-entry system

SPPS

solid-phase peptide synthesis

TA

transaminase

TFA

trifluoroacetic acid

The authors declare no competing financial interest.

Special Issue

Published as part of the ACS Medicinal Chemistry Letters virtual special issue “New Enabling Drug Discovery Technologies - Recent Progress”.