Engineering tRNAs for the Ribosomal Translation of Non-proteinogenic Monomers

Abstract

Ribosome-dependent protein biosynthesis is an essential cellular process mediated by transfer RNAs (tRNAs). Generally, ribosomally synthesized proteins are limited to the 22 proteinogenic amino acids (pAAs: 20 l-α-amino acids present in the standard genetic code, selenocysteine, and pyrrolysine). However, engineering tRNAs for the ribosomal incorporation of non-proteinogenic monomers (npMs) as building blocks has led to the creation of unique polypeptides with broad applications in cellular biology, material science, spectroscopy, and pharmaceuticals. Ribosomal polymerization of these engineered polypeptides presents a variety of challenges for biochemists, as translation efficiency and fidelity is often insufficient when employing npMs. In this Review, we will focus on the methodologies for engineering tRNAs to overcome these issues and explore recent advances both in vitro and in vivo. These efforts include increasing orthogonality, recruiting essential translation factors, and creation of expanded genetic codes. After our review on the biochemical optimizations of tRNAs, we provide examples of their use in genetic code manipulation, with a focus on the in vitro discovery of bioactive macrocyclic peptides containing npMs. Finally, an analysis of the current state of tRNA engineering is presented, along with existing challenges and future perspectives for the field.

1. Introduction

Transfer RNAs (tRNAs) are the central implementers of the genetic code. First conceptualized by Francis Crick in his 1955 letter “On Degenerate Templates and the Adaptor Hypothesis”, tRNAs map each of the 22 genetically encoded proteinogenic amino acids (pAAs) to a specific set of nucleotide triplets in messenger RNA (mRNA) known as codons.¹ Along with the 20 l-α-amino acids present in the standard genetic code, selenocysteine (Sec) and pyrrolysine (Pyl) constitute the 22 pAA building blocks employed in translation.² This mapping of the 64 mRNA triplet codons, which contain 61 sense codons and 3 termination (stop) codons, ensures that mRNAs are translated into proteins in a template-directed manner.

The canonical tRNA cloverleaf structure consists of five different sections: the acceptor arm, containing the 5′- and 3′-ends; the D arm; the anticodon arm, containing the anticodon; the variable loop; and the T arm (Figure 1a).³ Within each of these arms, paired bases are termed the stem, while unpaired bases are the loops; for instance, the D stem consists of base pairs 10–25, 11–24, 12–23, and 13–22, while the D loop consists of bases 14 to 21. In a conserved network of tertiary interactions, tRNAs fold into an L-shaped three-dimensional structure, though each tRNA can be slightly different (Figure 1b).^4,5 D arm and T arm interactions are important for the formation of the L-shaped architecture, creating the hydrophobic elbow region of the tRNA which separates the acceptor arm and anticodon arm.

Figure 1.

Structure of tRNA with structural domains highlighted in different colors. (a) The canonical cloverleaf folding structure of tRNA consists of the acceptor arm (pink; nucleotides 1 to 7 and 66 to 76), D arm (green; nucleotides 10 to 25), anticodon arm (blue; nucleotides 27 to 43), variable arm (yellow; nucleotides 44 to 48), and T arm (orange; nucleotides 49 to 65). Sections where base pairing occurs are called the “stems”, while the end of the stem without base pairing is the “loop”. ARS and ribosome interact with the different parts of the tRNA, while EF-P interacts with the D arm and EF-Tu interacts with the T stem and part of the acceptor stem. (b) The three-dimensional structure of yeast tRNA^Phe using the same color scheme (PDB: 1EHZ).⁵

As expected from their fundamental role in translation, tRNAs interface with many other major translational components through these conserved structural elements. Amino acids (AAs) are attached to the tRNA CCA-3′ end by enzymes known as aminoacyl-tRNA synthetases (ARSs). ARSs recognize their cognate tRNAs by idiosyncratic structural and sequence motifs known as identity elements that occur throughout the entirety of the tRNA.⁶ For example, phenylalanyl-tRNA synthetase (PheRS) specifically charges phenylalanine (Phe) to the 3′-end of its cognate tRNA^Phe_GAA with the anticodon GAA to form Phe-tRNA^Phe_GAA (the prefix denotes aminoacylation status while the superscript corresponds to tRNA identity). The resulting aminoacyl-tRNA (AA-tRNA) is shuttled to the ribosome via elongation factor thermo unstable (EF-Tu); EF-Tu binds to the T arm, acceptor arm, and aminoacyl moiety of the AA-tRNA.⁷ The ribosome (70S in prokaryotes) contains three different sites—aminoacyl (A), peptidyl (P), and exit (E)—which bind tRNAs through a number of direct molecular contacts.⁸ When the AA-tRNA arrives at the ribosome, the tRNA anticodon attempts to decode the mRNA codon in the A site.⁹ The small ribosome subunit (30S in prokaryotes) contacts the anticodon arm of bound tRNAs and mediates the interaction between tRNA and mRNA. When cognate codon–anticodon duplexes are formed, the AA-tRNA is fully accommodated into the ribosome A site.⁹ Then, the acceptor arms of the A site AA-tRNA and P site peptidyl-tRNA are positioned by the large ribosome subunit (50S in prokaryotes) to catalyze peptide bond formation between the amine of the A site amino acid and the ester of the P site peptidyl-tRNA.^10,11 In this process, the nascent peptide chain is both elongated and transferred between tRNAs. Subsequent translocation of the P and A site tRNAs to the E and P site, respectively, by elongation factor G (EF-G) frees the A site for the elongation cycle to recommence.¹²

This complex interplay between tRNAs and translational machinery has been a focal point for bioengineering. Engineered tRNAs have been especially important for the development of reprogrammed genetic codes that enable the ribosomal incorporation of non-proteinogenic amino acids (npAAs: l-α-amino acids with non-proteinogenic side chains, d-AAs, N-alkyl-AAs, β-AAs, etc.) into polypeptides.¹³⁻¹⁵ Technological advancements in the field of genetic code manipulation have also allowed for the ribosomal incorporation of non-amino acid exotic monomers (exMs: α-hydroxy acids, α-thio acids, etc.);¹⁵ as such, we find it helpful to use the term non-proteinogenic monomers (npMs), which includes both npAAs and exMs (Figure 2). When introduced within a protein, site-specifically encoded npMs can be used to interrogate its function and regulation in vivo, such as by cross-linking interacting partners or mimicking post-translational modifications.¹⁶ Meanwhile, npMs can drastically improve the pharmacokinetic profile of bioactive peptides, an emerging therapeutic modality.^17,18In vitro genetic code manipulation enables the incorporation of a wider range of npAAs and exMs, leading to the generation and screening of large mRNA-encoded libraries (greater than 10¹² unique sequences) of npM-containing peptides for drug discovery (Figure 3).

Figure 2.

Explanation of terms used throughout the review. The 22 proteinogenic amino acids (pAA: the 20 standard l-α-amino acids, selenocysteine and pyrrolysine) and non-proteinogenic monomers (npMs, everything except the 22 pAAs) are the broad terms which encapsulate all possible monomers. The two subset terms of npMs are non-proteinogenic amino acids (npAAs: l-α-amino acids with non-proteinogenic side chains, d-AAs, N-alkyl-AAs, β-AAs, etc.) and non-amino acid exotic monomers (exMs: α-hydroxy acids, α-thio acids, thionoesters, etc.).

Figure 3.

Selected examples of non-proteinogenic monomers (npMs) ribosomally elongated (noninitiator npMs) using the FIT system.

In this Review, we will discuss the progress made in engineering tRNAs for the incorporation of npMs into peptides and proteins. This Review is organized according to the key biomolecular partners of tRNA: (1) ARSs, (2) the ribosome and its associated translation factors, and (3) mRNA codons. A variety of in vivo and in vitro methodologies have been used to engineer tRNAs, each presenting a distinct set of challenges—and opportunities—for genetic code manipulation.

In section 2, we will discuss the efforts to engineer the recognition of tRNAs by ARSs. As this interaction determines the set of npMs available for translation, it is the key to diversifying the chemical repertoire accessible to proteins. Within living organisms, adding a new npM to the genetic code requires the development of a dedicated ARS/tRNA pair. The foremost challenge in this endeavor is one of orthogonality: the introduced ARS must only charge a specific (cognate) amino acid onto its specific (cognate) tRNA, and the introduced cognate tRNA must not be a substrate for any noncognate ARSs. For in vitro purposes, the ability to introduce precharged npM-tRNAs directly into the translation system removes the requirement for a new and orthogonal ARS, allowing for an expanded scope of npMs.

Peptide bond formation will be the focus of section 3, highlighting the poor ribosomal translation efficiency of npMs and the two major tRNA engineering solutions. Mutation of the tRNA T stem for optimal interaction of npM-tRNAs with EF-Tu can improve accommodation into the ribosome A site, and modification of the tRNA D arm to recruit elongation factor P (EF-P) can increase peptidyl transfer reactivity through repositioning of the peptidyl-tRNA.¹⁹ Significant progress has been made in the development and use of engineered tRNAs within reconstituted cell-free translation systems. For example, the “custom-made” Flexible In vitro Translation (FIT) system utilizes tRNA aminoacylation catalysts known as flexizymes and allows for near-total control of system components, greatly facilitating the introduction of npMs into polypeptides.²⁰

Section 4 examines the decoding of mRNA codons by tRNA anticodons and the task of sourcing blank codon–anticodon pairs for genetic code manipulation. Regardless of the ability to incorporate npMs into polypeptides, the combinatorial diversity of sequence-defined polymers remains limited by the standard genetic code. To attain an expanded genetic code, any new codon–anticodon pair must be orthogonal to the existing codon–anticodon pairs. The three major approaches for generating new codon–anticodon pairs will be described: breaking the degeneracy of the genetic code by sense codon reassignment; using codons formed from four nucleotide bases, which are read by tRNAs harboring the corresponding “quadruplet” anticodons; and introducing artificial nucleotides that form unnatural base pairs into codons and anticodons.

After reviewing these distinct engineering projects, we will turn to the applications of the resulting engineered tRNAs for genetic code manipulation in section 5. After briefly discussing in vivo applications of npMs, we will focus on the in vitro incorporation of backbone-modifying npMs (non-l-AAs such as d-AAs, N-alkyl-AAs, β-AAs, etc.) into peptides. Combination of in vitro translation technologies with genetically encoded peptide library display methods, such as mRNA display, has resulted in the discovery of exotic peptides with high-affinity to biological targets. These npMs are often encoded into peptide libraries to confer beneficial pharmacological properties, such as promoting macrocyclization or other rigid structural features,²¹⁻²⁵ increasing protease resistance,²⁶ or enhancing membrane permeability.²⁷ We will survey examples of exotic peptides arising from mRNA display-based affinity selections in cases involving engineered tRNAs and assess the effect of npMs on their pharmacokinetic properties.

Finally, we will conclude with an outlook on the field of tRNA engineering for genetic code manipulation and suggest future directions to advance the synthesis and screening of encoded npM-containing biopolymers.

2. Orthogonal tRNAs for Acylation with Non-proteinogenic Monomers

2.1. Basis of Translational Orthogonality

The first step in protein synthesis is aminoacylation, wherein an ARS catalyzes the esterification of a cognate amino acid to its set of cognate, isoacceptor tRNAs to form AA-tRNAs (Figure 4).²⁸ Isoacceptors are tRNAs that accept the same amino acid but have different anticodons. First, the ARS recognizes its cognate amino acid and activates it with ATP to form an aminoacyl-adenylate (AA-AMP) intermediate. Then, the AA-AMP moiety is transferred to either the 2′- or 3′-hydroxyl of the 3′-terminal adenine of the cognate tRNA, forming the final AA-tRNA.²⁹ While the 3′-O-aminoacyl ester is the active species during peptide bond formation, it is generally accepted that the esters can be readily interconverted via transacylation in an aqueous environment.³⁰⁻³² Most organisms utilize over 30 different tRNAs for translation that are unique for each organism.³³ Additionally, 23 individual ARS classes have been discovered: one for each canonical amino acid (with the exception of Lys, for which there are two unrelated forms of ARS), pyrrolysyl-tRNA synthetase (PylRS) and O-phosphoseryl-tRNA synthetase (SepRS).^6,34−36 However, the recent discovery of multiple distinct classes of PylRS may change this classification in the future.³⁷ Sec is charged onto tRNA^Sec through a separate biosynthetic pathway.³⁸ The exceptional robustness of the genetic code relies on every ARS being mutually orthogonal; in other words, each ARS must selectively recognize its cognate amino acid and tRNA with no cross-reactivity.

Figure 4.

Synthesis of pAA-tRNAs, also known as aminoacylation. (1) The pAA (blue) is activated by ATP (red). Although the tRNA is not usually necessary for activation, some ARSs do require tRNA binding before pAA activation. (2) The pAA-AMP substrate is transferred to the 2′- or 3′-hydroxyl of the 3′-terminal adenine of the tRNA (green), resulting in a pAA-tRNA. This aminoacyl ester is in dynamic equilibrium between the 2’ and 3′-hydroxyl of A76.³²

Recognition of the cognate pAA poses a difficult challenge, as the number of possible contacts with the ARS is low due to its relatively small size. Thus, a variety of strategies have evolved, such as the use of metal ions, π-stacking, hydrogen bonding, and exclusion via size or charge.^34,39 For example, GlyRS contains a number of bulky, charged amino acids in the active site that effectively exclude noncognate pAAs from entering.⁴⁰ Nevertheless, in cases where the cognate pAA has sufficient similarity to certain noncognate pAAs, misacylation can occur. Thus, a subset of ARSs also contain an AA-tRNA deacylase editing site that allows for in cis cleavage of noncognate pAAs to minimize aminoacylation errors.⁴¹⁻⁴³ Finally, in trans editing enzymes can also act as another layer of security, such as the widespread, standalone homologue of the editing domain of AlaRS, AlaXp, that deacylates mischarged Ser-tRNA^Ala.⁴⁴

The complex, three-dimensional structure of tRNAs allows for a much more specific ARS/tRNA recognition mechanism via idiosyncratic tRNA sequence or structural features known as identity elements. Features whose mutations are accompanied by loss of cognate ARS recognition are known as specificity determinants or positive identity elements.^6,29,45 Conversely, nucleotides that prevent tRNA recognition by noncognate ARSs are called antideterminants. Other than specific bases, tertiary interactions due to the folding of tRNA can also influence ARS/tRNA orthogonality and thus act as identity elements. While identity elements for each ARS/tRNA pair can be motifs unique to various species or isoacceptor classes, especially in organellar tRNAs, nucleotides located in the acceptor arm or anticodon loop usually contribute significantly toward tRNA recognition in model organisms from all three domains of life.⁴⁶

Nucleotide 73 (N73, where N is A, U, C, or G) in the acceptor arm is known as the discriminator base and plays a key role in differentiating cognate from noncognate tRNAs (Figure 1).²⁹ Simple point mutation of the discriminator base can result in drastic changes in ARS recognition. For instance, the G73Y mutation (where Y is U or C) in Escherichia coli (E. coli) tRNA^Asn abolishes aminoacylation by its cognate AsnRS.⁴⁷ Due to the proximity of the acceptor arm to the aminoacylation site, many tRNAs have other important recognition elements in the acceptor arm. One of the most well-studied examples is the E. coli tRNA^Ala isoacceptor class, where the major identity determinant is the G3○U70 wobble pair (○ denotes non-Watson–Crick base pairing).⁴⁸ Mutation of this pair to the Watson–Crick (WC) A3•U70 or G3•C70 pair abolishes alanylation (• denotes WC base pairing).⁴⁹ Furthermore, incorporation of this nucleotide pair into E. coli amber suppressors tRNA^Tyr_CUA, tRNA^Phe_CUA, and tRNA^Cys_CUA was sufficient to elicit their alanylation by AlaRS.^50,51 A notable acceptor arm identity element occurs in tRNA^His. Unlike the majority of other tRNAs, where the discriminator base N73 is unpaired, tRNA^His possesses a G(−1) nucleotide in most organisms, which forms a G(−1)•C73 pair.⁶ Removal of G(−1) or mutation of C73 to any other nucleotide significantly decreases the rate of aminoacylation for E. coli and Saccharomyces cerevisiae (S. cerevisiae/yeast) tRNA^His, demonstrating the importance of this noncanonical base for His-specific function.^52,53

Given the centrality of the anticodon in defining tRNA function, it is no surprise that the anticodon is a key identity element for most tRNAs. Often, mutation from the canonical anticodon changes aminoacylation specificity or decreases the aminoacylation rate.²⁹ For example, recognition of tRNA^Gln_YUG by GlnRS is particularly sensitive to changes. Point mutation of the second anticodon base U35 decreases catalytic aminoacylation activity by up to 10,000-fold.^54,55 Moreover, the anticodon loops of many tRNAs contain a significant number of modified nucleotides; these can also confer strong specificity, with key examples being tRNA^Ile and tRNA^Glu. As one of the isoacceptors of tRNA^Ile contains the anticodon CAU, it must be discriminated from tRNA^Met_CAU. In many organisms, the modified base lysidine (k²C34) in tRNA^Ile_k2CAU acts as the necessary antideterminant. Point mutation to unmodified C34 confers misacylation activity by MetRS and decreases IleRS activity.⁵⁶ Meanwhile, E. coli tRNA^Glu contains the hypermodified residue 5-carboxymethylaminomethyl-2-thiouridine (mnm⁵s²U34) as an identity element.^57,58 Although GluRS can recognize unmodified transcripts of tRNA^Glu_UUC, removal of the mnm⁵s²U34 modification decreases the kinetics of aminoacylation by over 500-fold.⁵⁹

Three of the canonical ARSs that do not recognize the anticodon are AlaRS, LeuRS, and SerRS. AlaRS recognizes almost all nucleotides in the acceptor arm and G20 in the D arm instead of the anticodon.⁶ For E. coli, bases C72, G2•C71, and G3•C70 in the acceptor arm and C11•G24 in the D stem are necessary for SerRS recognition.^60,61 Mutation of the antideterminant discriminator base G73 causes significant misacylation by both LeuRS and TyrRS.^60,61 Furthermore, transplantation of tRNA^Ser positive identity elements to tRNA^Leu conferred complete serylation specificity.⁶² For the variable arm of tRNA^Ser, only stem length is the specificity determinant for SerRS, not sequence. Insertion of an additional three bases into the tRNA^Tyr variable arm, along with the mutations G9U and G73A, was enough to switch identity from tyrosination to serylation.^63,64 Finally, identity elements for LeuRS are mostly dependent on the discriminator base A73 and the presence of specific tertiary interactions between the D and T loops.⁶⁵ LeuRS can still aminoacylate minimal substrates lacking the variable arm as long as the D and T loops are still present.⁶⁶

2.2. Development of In Vivo Orthogonal ARS/tRNA Pairs

To perform in vivo genetic code manipulation, an additional ARS/tRNA pair must be introduced into the organism. Just as all proteinogenic ARS/tRNA pairs are mutually orthogonal, the introduced ARS/tRNA pair must be orthogonal with respect to all endogenous ARS/tRNA pairs (Figure 5a). Additionally, this new pair must mediate specific incorporation of an npM at a “blank” codon that is not already assigned to pAAs (Figure 5b).

Figure 5.

Orthogonality required for genetic code manipulation. To allow for site-specific incorporation of npMs into polypeptides, orthogonality in the genetic code must be strictly maintained. (a) Natural ARSs recognize their cognate pAA and tRNA to generate pAA-tRNAs. They show high substrate specificity (double-sided black arrows) and do not cross-react with other pAAs or tRNAs (double-sided gray dotted arrows). Thus, introduced ARS/tRNA pairs must be similarly orthogonal to all other endogenous ARS/tRNA pairs. (b) Incorporation of npMs is accomplished through creation of “blank” codons which are not decoded by pAAs. Details of other types of blank codons other than nonsense suppression are addressed in detail in section 4.⁶⁸

A large body of work with in vivo orthogonal tRNAs focuses on nonsense suppression for genetic code expansion. Three codons, UAA (ochre), UAG (amber), and UGA (opal), are usually reserved for terminating translation via release factors (RFs).⁶⁷ A nonsense suppressor tRNA contains an anticodon that recognizes a stop codon instead of one of the 61 sense codons. This allows for the incorporation of an npM at a stop codon, though the nonsense suppressor npM-tRNA is still in competition with the native RFs for termination.⁶⁸ As only one stop codon is necessary per polypeptide, two stop codons are “blank” and can be decoded by these suppressor tRNAs. As ARS identity elements vary between organisms, and especially between the three domains of life, suppressor tRNAs from certain organisms can be genetically encoded into a different organism (heterologous usage) to form orthogonal ARS/tRNA pairs.^6,68

2.2.1. Early Efforts

Early work demonstrated that E. coli PheRS and yeast tRNA^Phe were nearly orthogonal in vitro as E. coli PheRS aminoacylated yeast tRNA^Phe at less than 1% efficiency compared to E. coli tRNA^Phe.⁶⁹ Additionally, the yeast amber suppressor PheRS/tRNA^Phe_CUA pair was shown to be translationally active in a mammalian in vitro protein synthesis system when incorporating Phe.⁷⁰ Having established that the PheRS/tRNA^Phe_CUA pair was readily orthogonal for the in vitro heterologous translation of pAAs, the groundwork was laid for incorporation of npMs.

Noren and co-workers in 1989 demonstrated the first in vitro site-specific incorporation of an npM into a ribosomally synthesized protein by amber codon suppression.⁷¹ First, yeast suppressor tRNA^Phe_CUA was chemoenzymatically aminoacylated with either p-nitro-l-phenylalanine (pNO₂Phe), p-fluoro-l-phenylalanine (pFPhe), or l-homophenylalanine (HPhe). This was accomplished by chemical synthesis of the dinucleotide conjugated npM (pCpA-npM), followed by T4 RNA ligation to suppressor tRNA missing the terminal CA-3′ nucleotides. Then, a β-lactamase gene containing a Phe66 to amber mutation (UUU to UAG) was translated using a crude E. coli S30 cell extract, allowing for the incorporation of each npM into the protein. Later that year, Bain and co-workers used a similar strategy to incorporate 3-iodo-l-tyrosine (oITyr) into a 16-mer polypeptide containing a UAG codon using an E. coli suppressor tRNA^Gly_CUA in rabbit reticulocyte lysate.⁷² Analysis of the translation products from both experiments unambiguously demonstrated site-specific incorporation of the desired npM with no detectable misincorporation from pAAs. Using this chemoenzymatically aminoacylated yeast suppressor tRNA^Phe_CUA and E. coli S30 cell extract system, the Schultz group was able to incorporate a wide variety of npMs into proteins throughout the 1990s.⁷³ However, as chemoenzymatic aminoacylation was quite laborious, an orthogonal ARS/tRNA pair would be preferable for charging npMs.

The first report of in vivo genetic code expansion using a genetically encoded orthogonal ARS/tRNA pair was in 1998 by Furter.⁷⁴ Using the yeast PheRS/tRNA^Phe pair in E. coli, site-specific incorporation of pFPhe into dihydrofolate reductase (DHFR) was achieved, with up to 75% of the expressed protein containing the encoded pFPhe and about 20% misincorporation of Phe. While E. coli PheRS and yeast tRNA^Phe are orthogonal, yeast PheRS and E. coli tRNA^Phe are not; thus, the contamination of Phe was likely due to this cross-species charging.^69,74 This revolutionary work paved the way for further experimentation with heterologous pairs in vivo while simultaneously emphasizing the need for completely orthogonal ARS/tRNA pairs.

2.2.2. The TyrRS/tRNA^Tyr Pair from Methanocaldococcus jannaschii

Based on the known ARS identity elements, TyrRS/tRNA^Tyr pairs appeared to have several positive traits that could lead to heterologous orthogonal pairs. First, the major identity elements for TyrRS located in the acceptor arm and variable arm differ between bacteria and eukarya/archaea. While bacterial positive identity elements are G1•C72 and a longer variable arm, many eukarya and archaea instead recognize C1•G72 and a shorter variable arm.^6,75,76 Thus, it was hypothesized that bacterial and eukaryotic TyrRS/tRNA^Tyr pairs were naturally orthogonal. Second, only part of the anticodon is recognized by TyrRS, so point mutation from the natural GUA anticodon to the suppressor CUA anticodon was presumed to have less impact on aminoacylation activity.^77,78 Finally, as TyrRS does not contain an in cis editing domain, a misacylated npM cannot be hydrolyzed after aminoacylation, allowing for efficient production of npM-tRNAs.⁷⁹

It was first noted in 1999 that the thermophilic methanogenic archaea Methanocaldococcus jannaschii (Mj) TyrRS was able to in vitro aminoacylate unmodified Mj tRNA^Tyr but not E. coli tRNA^Tyr (Figure 6a).⁸⁰ The knowledge that Mj TyrRS could aminoacylate unmodified Mj tRNA^Tyr was important, as heterologous tRNAs may lack their naturally occurring modified nucleotides necessary for ARS recognition.^81,82 Subsequently, it was experimentally confirmed that the Mj TyrRS/tRNA^Tyr_CUA pair was nearly orthogonal in E. coli when charging Tyr, and the pair had significantly higher aminoacylation efficiency when compared with contemporaneously developed orthogonal yeast suppressors.⁸³⁻⁸⁵

Figure 6.

2-Dimensional representations of the secondary structures of Mj mutRNA^Tyr_CUA, Mb tRNA^Pyl_CUA, and E. coli tRNA^AsnE2_NNN. Primary sequences without modified bases are displayed. Important identity elements for their cognate ARS are shown in blue, mutations from the wild type to the engineered tRNAs are shown in red, and important structural features are underlined. (a) Mj mutRNA^Tyr_CUA contains five point mutations: C17A, U17aG, U20C, G37A, and U47G. In addition, the anticodon is changed from GUA to CAU to allow for amber codon suppression. (b) Mb tRNA^Pyl_CUA contains unique structural elements: short variable arm, long anticodon stem, short D loop, a deleted base connecting the D arm and acceptor stem, and compact core structure. (c) tRNA^AsnE2_NNN was developed as an orthogonal tRNA to E. coli ARSs by introducing point mutations U1G, C2G, G71C, and A72C; additionally, any anticodon can be used, as it is not charged by a cognate ARS. (d) Superimposition of yeast tRNA^Phe (PDB: 1EHZ) andD. hafniense tRNA^Pyl (PDB 2ZNI). D arm is in green, variable arm is in yellow, anticodon is in blue. The lighter colors correspond to yeast tRNA^Phe while the darker colors are D. hafniense tRNA^Pyl.^{86,100,102,203}

In 2001, Wang and co-workers demonstrated in vivo genetic code expansion of E. coli using an engineered Mj TyrRS/tRNA^Tyr_CUA pair.^86,87 As Mj RNA^Tyr_CUA was not completely orthogonal to endogenous E. coli ARSs,⁸⁵ the orthogonality of tRNA^Tyr_CUA was targeted first for improvement. Eleven nucleotides in Mj tRNA^Tyr_CUA that do not interact with Mj TyrRS were mutated, and these libraries were passed through a so-called “double-sieve” selection, consisting of sequential rounds of negative and positive selection.^86,87 In negative selection, amber codons were introduced into the toxic protein barnase. If a mutant Mj tRNA^Tyr_CUA library was aminoacylated by endogenous E. coli ARSs, barnase would be translated and the cell would die. Surviving colonies would contain Mj tRNA^Tyr_CUA variants that were orthogonal to E. coli ARSs. In positive selection, amber codons were introduced into β-lactamase, and Mj tRNA^Tyr_CUA libraries were selected for their ability to be efficiently recognized by Mj TyrRS. Growth of E. coli in ampicillin media resulted in surviving colonies which contained active Mj TyrRS/tRNA^Tyr_CUA pairs that were also orthogonal to E. coli ARSs. The resulting Mj mutRNA^Tyr_CUA (mutations C17A, U17aG, U20C, G37A, and U47G) afforded almost complete orthogonality (Figure 6a).^86,87 Second, in order to change the specificity of Mj TyrRS from Tyr to an npM, an Mj TyrRS library consisting of mutated active site residues was screened for specific incorporation of O-methyl-l-tyrosine (Tyr^OMe) into chloramphenicol acetyltransferase in the presence of mutRNA^Tyr_CUA. The resulting mutant TyrRS (Y32Q, E107T, D158A, L162P) was used in combination with mutRNA^Tyr_CUA to site-specifically incorporate Tyr^OMe at an amber stop codon in DHFR.^86,87

Given the high level of orthogonality and translation efficiency, Mj TyrRS/tRNA^Tyr_CUA pairs in E. coli are among the most common heterologous systems currently in use.^13,14 Recently, Mj TyrRS/tRNA^Tyr_CUA derivatives have also been used to initiate translation.^88,89 Generally, to use mutRNA^Tyr_CUA with other npMs, directed evolution (double-sieve selection) is performed on Mj TyrRS libraries (usually consisting of 10⁷–10⁸ variants) to discover mutants capable of charging other npMs.^90,91 Although many substrates are para-substituted Phe derivatives, a variety of other phenyl derivatives and heteroaromatics have been incorporated using evolved Mj TyrRSs.^{13,14,92−94}

Due to the similarities between the acceptor arm of E. coli tRNA^Pro and mutRNA^Tyr_CUA, background misacylation of E. coli tRNA^Pro by evolved Mj TyrRS variants has been reported, albeit at low levels; in this case, misacylation was prevented through overexpression of endogenous E. colipro S.^95,96 In a separate study, Maranhao and Ellington implemented a more stringent negative selection scheme to discover a new mutant of Mj tRNA^Tyr_CUA termed S7.⁹⁷ S7 contained numerous mutations in the acceptor arm, T stem, variable loop, and anticodon stem, leading to it being over 2-fold more orthogonal in E. coli than Mj mutRNA^Tyr_CUA. Thus, orthogonality can be restored through further double-sieve selections of the tRNA, though such cases may be npM specific.

2.2.3. The PylRS/tRNA^Pyl Pair from Archaea

In 1998, it was observed that in the methanogenic archaeon Methanosarcina barkeri (Mb), the gene encoding a monomethylamine methyltransferase (MtmB) contained an amber stop codon that did not encode termination.⁹⁸ Subsequent crystallization of this protein led to the discovery of Pyl at the position encoded by the amber stop codon, and genome analysis uncovered PylRS and amber suppressor tRNA^Pyl_CUA, unambiguously establishing Pyl as the 22nd pAA.⁹⁹⁻¹⁰¹ Co-translational incorporation of Pyl is highly conserved in the family Methanosarcinacea, with homologous genes being rapidly discovered in Methanosarcina mazei (Mm), Methanosarcina acetivorans, and Methanococcoides burtonii.³⁵ PylRS and tRNA^Pyl_CUA genes are also present in some bacteria, with the most well-studied being Desulfitobacterium hafniense (D. hafniense).¹⁰² Given that tRNA^Pyl_CUA is naturally a suppressor tRNA, it was quickly verified that the Mb PylRS/tRNA^Pyl_CUA pair was an orthogonal amber suppressor in E. coli when using Pyl to translate MtmB in response to the UAG codon.^101,103

Generally, tRNA^Pyl sequences contain a variety of unique features that distinguish them from most canonical tRNAs and form the basis of their orthogonality in bacteria and eukarya (Figure 6b). In Mb tRNA^Pyl_CUA, these include one base between the acceptor and D stem instead of two, a five base D loop instead of eight, deletion of the highly conserved G18-G19 in the D loop, an anticodon stem containing six base pairs instead of five, a three base variable loop instead of five or more, and the mutations T54U and C56A in the T54-Ψ55-C56 motif (where Ψ is pseudouridine).^45,100 Depending on the organism, a wide variety of other structural anomalies have also been discovered.³⁷ Orthogonality originates from a combination of these features. In general, the numerous nucleotide deletions and mutations cause tRNA^Pyl to adopt a more compact tertiary core structure when compared to canonical tRNAs (Figure 6d).^102,104 PylRS forms an extensive network of contacts with this tRNA^Pyl core, sterically excluding tRNAs with bulkier cores. These interactions share little homology to that of canonical ARS/tRNA interactions and explain why the PylRS/tRNA^Pyl_CUA pair is naturally orthogonal in many organisms.^102,104 Positive identity elements are the G73 discriminator base and the G1•C72 pair, though some PylRS may recognize the T stem, a different discriminator base, or a different overall shape.^6,105−108 Fortuitously, tRNA^Pyl_CUA only contains two modified nucleotides, meaning that lack of these nucleotides in vitro or in vivo barely affects aminoacylation efficiency.¹⁰¹ Additionally, PylRS does not contain an editing domain.¹⁰⁴ These two additional features make PylRS/tRNA^Pyl_CUA pairs attractive candidates for genetic code expansion.

Thus, PylRS/tRNA^Pyl_CUA pairs, usually from Mm or Mb, have emerged as the other main choice for in vivo genetic code expansion due to their natural orthogonality with E. coli, as well as yeast, Caenorhabditis elegans (C. elegans), Drosophila melanogaster, and mammalian cells.^109−114 PylRS is amenable to mutation through directed evolution schemes to recognize a wide variety of npMs with most substrates being Lys or Phe derivatives.^13,14,68 Finally, because PylRS does not recognize the anticodon of tRNA^Pyl, mutants of tRNA^Pyl bearing modified anticodons have been used to incorporate npMs in response to diverse codons.^115−120

2.2.4. Other Common Orthogonal Pairs

Various other orthogonal ARS/tRNA pairs have been developed to try and expand npM substrate scope, increase suppression efficiency, or perform genetic code expansion in different organisms.

As previously mentioned, bacterial and eukaryotic identity elements for TyrRS differ significantly. Analogous to the orthogonality of the Mj TyrRS/tRNA^Tyr_CUA pair in E. coli, the E. coli TyrRS/tRNA^Tyr_CUA pair is orthogonal in yeast.^121,122 The first report of genetic code expansion with the E. coli TyrRS/tRNA^Tyr_CUA pair was in 2003 by Chin and co-workers wherein five different para-modified Phe derivatives were incorporated into amber codons in human superoxide dismutase 1 (hSOD).¹²³ This pair has also been used to incorporate Phe and Tyr derivatives in various other eukaryotes such as Candida albicans,^124,125Pichia pastoris,¹²⁶Schizosaccharomyces pombe,¹²⁷ and C. elegans.¹²⁸

Since LeuRS does not recognize the anticodon of tRNA^Leu isoacceptors, the E. coli LeuRS/tRNA^Leu_CUA pair has been explored as an orthogonal amber suppressor in various eukaryotes. In 2004, Wu and co-workers reported the incorporation of Tyr^OMe, α-aminocaprylic acid, and o-nitrobenzyl-l-cysteine into hSOD in yeast.¹²⁹ Due to the promiscuity of mutant LeuRSs, both aliphatic and bulky aromatic substrates (including metallocenes) can be tolerated.^{13,14,130−133} Use of this pair has also been reported in mammalian cells and C. elegans.^128,134

Heterologous TrpRS/tRNA^Trp pairs have also been used to generate orthogonal pairs in bacteria or eukaryotes. This is because bacterial TrpRSs recognize G73 in tRNA^Trp as a positive identity element, whereas archaeal and eukaryotic TrpRSs recognize A73 and the G1•C72 pair as positive identity elements.^135,136 Various TrpRS/tRNA^Trp combinations have been used for the incorporation of bulky Trp derivatives.^13,14,137 Hughes and Ellington used rational design and in vivo optimization to design the yeast TrpRS/tRNA^Trp_CUA-AS3.4 and -AS3.5 pairs which contained numerous anticodon stem mutations and were orthogonal in E. coli.¹³⁸ Subsequently, Chatterjee and co-workers further optimized the acceptor stem to incorporate Trp derivatives into green fluorescent protein (GFP) in E. coli.¹³⁹ Evolved Bacillus subtilis TrpRS/tRNA^Trp_UCA has also been used for opal suppression to incorporate 5-hydroxy-l-tryptophan in mammalian cells.¹⁴⁰

Finally, as SepRS is a unique ARS devoted to Sep-tRNA^Cys formation in methanogenic archaea, the mesophilic Methanococcus maripaludis SepRS and Mj tRNA^Cys cross-species pair are naturally orthogonal in E. coli.^141,142 Evolved derivatives of this pair have been used for the incorporation of Sep, 2-amino-4-phosphonobutyric acid (a nonhydrolyzable analogue of Sep), and l-phosphothreonine in response to amber codons in E. coli.^143,144

2.2.5. Multiple Non-proteinogenic Monomer Incorporation Using Mutually Orthogonal Heterologous Pairs

Two (or more) heterologous ARS/tRNA pairs can be used simultaneously to encode two (or more) distinct npMs into the host proteome provided that three requirements are fulfilled. First, all introduced pairs must be mutually orthogonal to one another, in addition to being orthogonal to all host organism ARS/tRNA pairs. Second, the ARSs must recognize different npMs as substrates. Third, the tRNAs must be assigned to distinct blank codons (as elaborated upon in section 4).¹⁴⁵ To date, there are over 30 reports of multiple npM incorporation in both prokaryotes and eukaryotes.¹⁴⁵

The addition of two distinct npMs to the genetic code of an organism was first realized by Anderson and co-workers in 2004 through combination of the Mj TyrRS/mutRNA^Tyr pair and an uncommonly used Pyrococcus horikoshii LysRS/tRNA^Lys pair for double incorporation of Tyr^OMe and l-homoglutamine (HGln), respectively, into myoglobin in E. coli.¹⁴⁶

In 2010, two different groups reported the simultaneous incorporation of p-azido-l-phenylalanine (pAzPhe) and N^ε-Boc-l-lysine (BocLys) or N^ε-propargyloxycarbonyl-l-lysine (AlkLys) with Mj TyrRS/tRNA^Tyr and Mb PylRS/tRNA^Pyl, respectively, in E. coli.^117,147 The dual use of Mj TyrRS/tRNA^Tyr and PylRS/tRNA^Pyl is the most common methodology for the double incorporation of two different npMs in prokaryotes.^88,148−159 Meanwhile, in mammalian cells, the E. coli TyrRS/tRNA^Tyr and Mb PylRS/tRNA^Pyl pairs^160−162 or the Mm PylRS/tRNA^Pyl and Candidatus Methanomethylophilus alvus (CMa) PylRS/tRNA^Pyl pairs^163,164 are among the most frequently used combinations.

The first instance of triple incorporation of three distinct npMs into a protein in vivo was by Italia and co-workers in 2019, when they used Mj TyrRS/tRNA^Tyr_CUA, Mb PylRS/tRNA^Pyl_UUA, and E. coli TrpRS/tRNA^Trp_UCA for simultaneous incorporation of three different npMs at all three stop codons into superfolder GFP (sfGFP) in an E. coli strain possessing altered translational machinery.¹⁶⁵ Triple incorporation was later extended to mammalian cells by Shi and co-workers in 2022 by using E. coli LeuRS/tRNA^Leu_CUA, E. coli TyrRS/tRNA^Tyr_UUA, and Mm PylRS/tRNA^Pyl_UCA.¹⁶⁶ Finally, Dunkelmann and co-workers in 2021 reported the first in vivo quadruple npM incorporation by using three different PylRS/tRNA^Pyl pairs and an Archaeoglobus fulgidus TyrRS/tRNA^Tyr pair in E. coli.¹⁶⁷

2.2.6. Recent Methods to Discover New Naturally Occurring Orthogonal Pairs

Advances in computational and experimental methodologies have facilitated the discovery of further orthogonal ARS/tRNA pairs. An early example by Neumann and co-workers used Mj TyrRS/tRNA^Tyr_CUA as a starting point for iterative cycles of structure guided design of combinatorial libraries and genetic selections to evolve a new mutually orthogonal pair for E. coli, termed XTyrRS/XtRNA.¹⁶⁸ A recent approach used genome mining and clustering analysis of homologous PylRS and tRNA^Pyl sequences to identify sets of pairs with a high degree of natural orthogonality.^37,169,170 This revealed substantial evolutionary divergence of PylRS/tRNA^Pyl pairs, including tRNA features at the PylRS/tRNA interface in previously cocrystallized pairs, such as the discriminator base, variable loop, and D arm.^102,104 A total of five distinct sequence classes were identified; tRNA engineering and directed evolution was then used to generate quintuply orthogonal PylRS/tRNA^Pyl pairs.³⁷ Notably, some of the discovered PylRS/tRNA^Pyl pairs, such as the PylRS/tRNA^Pyl pair from CMa, have proven more efficient than the standard Mb or Mm PylRS/tRNA^Pyl_CUA pairs for incorporation of various npMs.^{169,171−173}

Efforts have also been devoted to sourcing orthogonal ARS/tRNA pairs from other isoacceptor classes, which may expand the range of genetically encodable npMs. In 2020, Cervettini and co-workers developed an orthogonal ARS/tRNA discovery pipeline based on the tRNA Extension (tREX) technology which can rapidly determine the aminoacylation status of tRNAs.¹⁷⁴ Computational screening identified 71 tRNA candidates for characterization by tREX, and subsequent mutation of anticodons to CUA and evolution of the ARS/tRNA pairs led to the identification of three orthogonal, amber suppressor ARS/tRNA pairs in E. coli: Sorangium cellulosum AspRS(C4)/tRNA^Asp(10)_CUA, Ilumatobacter nonamiensis GlnRS(S9)/tRNA^Gln(A1)_CUA, and Archaeoglobus fulgidus TyrRS(G5)/tRNA^Tyr(A01)_CUA.¹⁷⁴

2.3. Improving Orthogonality and Substrate Scope for In Vitro Translation Systems

Early methods for in vitro translation used a crude cell lysate (or extract) that was supplemented with a chemoenzymatically aminoacylated, heterologous, orthogonal npM-tRNA (as mentioned in section 2.2.1).¹⁷⁵ These cell-free protein synthesis (CFPS) systems were capable of ribosomally polymerizing a wide variety of npMs, but there were significant limitations such as poor suppression efficiency and the technical difficulty of aminoacylation.⁷³

A key breakthrough for in vitro translation came in 2001 when Shimizu and co-workers reported a new CFPS system called the protein synthesis using recombinant elements (PURE) system. This system consisted of 32 individually purified E. coli translation factors, along with the other necessary components for transcription and translation.¹⁷⁶ Because the system is constituted of individually purified components, the presence of factors detrimental to translation efficiency, such as DNase, RNase, or protease, is minimal. While the PURE system can be used for genetic code expansion (for example at the amber codon), sense codons can also be liberated for sense codon reassignment by removal of specific pAAs and their cognate ARSs. Then, orthogonal npM-tRNAs can be supplemented into the PURE mixture to decode empty codons, allowing for translation with minimal background misincorporation from the natural translation machinery.

Several orthogonal ARS/tRNA pairs commonly used for in vivo genetic code expansion have also been used in vitro.^{173,177−179} However, these pairs are often heterologous to the organism from which the in vitro translation system is derived, such as using yeast suppressor tRNA^Phe_CUA with the E. coli PURE system. This heterologous usage can result in lowered translation efficiency, due to poor compatibility of the tRNA with endogenous translation factors.¹⁸⁰ An attractive alternative involves the use of homologous tRNAs precharged with an npM (chemoenzymatically or otherwise). This significantly broadens the range of useable tRNA sequences because they do not need to be specifically recognized by an orthogonal ARS; the only requirement is that the tRNA is orthogonal to the current ARSs present in the translation system. Because several tRNA isoacceptor classes can be made orthogonal to their cognate ARS by simple mutation of the tRNA anticodon or discriminator base, it becomes feasible to use a tRNA derived from the CFPS chassis organism for in vitro genetic code reprogramming.^6,181,182

In the early 1990s, amber suppressor mutants derived from over 20 natural E. coli tRNAs were constructed and characterized in E. coli with suppression efficiencies ranging from 0 to 100%.^181−183 In other words, the inactive E. coli suppressor tRNAs were both homologous and orthogonal in E. coli. In 1996, Cload and co-workers tested whether some of the inactive variants could be transcribed in vitro, chemoenzymatically aminoacylated with Val or HGln, and then translated in response to UAG codons located in T4 lysozyme or E. coli chorismate mutase using E. coli S30 extracts.¹⁸⁴ Compared to other heterologous suppressor tRNAs from Tetrahymena and yeast, the homologous E. coli suppressor tRNA^Asn_CUA showed significant improvement in translation efficiency.¹⁸⁴ As the only major identity elements of tRNA^Asn are the anticodon GUU and the discriminator base G73, mutation to the amber anticodon CUA afforded orthogonality.^6,47

In 2003, Forster and co-workers demonstrated one of the first examples of sense codon reassignment using a derivative of the PURE system using an E. coli tRNA^Asn derivative called tRNA^AsnB.¹⁸⁵ tRNA^AsnB was used (mutations U1G, C2G, C3G, G69C, G70C, A71C), as it contained a more stable acceptor arm compared to tRNA^Asn_CUA, and extensive mutations to the native acceptor arm also contributed to orthogonality.¹⁸⁶ First, tRNA^AsnB with the anticodons CAG, UUG, or UGG was transcribed using in vitro transcription, resulting in tRNA transcripts without any modified nucleotides.¹⁸⁵ Next, chemoenzymatic aminoacylation was used to charge l-allylglycine (VinGly) onto tRNA^Asn_CAG to decode the GUU codon, l-propargylglycine (PrgGly) onto tRNA^Asn_UUG to decode the AAC codon, and O-methyl-l-serine (Ser^OMe) onto tRNA^Asn_UGG to decode the ACC codon. Finally, the pentapeptide N^α-formyl-l-methionine (^fMet)-PrgGly-Ser^OMe-VinGly-Glu was translated in this system with minimal read-through by near-cognate pAA-tRNAs.

Numerous routes for in vitro aminoacylation have been developed, such as the previously mentioned chemoenzymatic aminoacylation,⁷³ naturally promiscuous ARSs,^187−189 and postaminoacylation modifications.¹⁹⁰ While most are still currently in use, they often suffer from low yields, technically challenging chemical syntheses, or poor substrate scope.

In the period of 2001 to 2006, the Suga group developed flexizymes (Fxs), which are small, artificial RNAs that catalyze the aminoacylation of an activated acyl substrate to the CCA-3′ end of a tRNA (Figure 7).^191−196 Three types of Fxs have been developed that recognize different elements: dinitro-flexizyme (dFx) recognizes 3,5-dinitrobenzyl esters (DBEs), enhanced flexizyme (eFx) recognizes p-chlorobenzyl thioesters (CBTs) or the combination of an aromatic side chain and a cyanomethyl ester (CME) group, and amino-flexizyme (aFx) recognizes 4-[(2-aminoethyl)carbamoyl]benzyl thioesters (ABTs).^192,197,198 As the substrate only needs to contain a carboxylic acid that can be activated, npMs without an amino functional group are able to be acylated. Thus, highly modified monomers can be used, allowing for the acylation of a wide array of unique exMs.^199,200

Figure 7.

Different types of activating groups used with its related flexizyme. DBE (red) is recognized by dFx, eFx recognizes monomers with an aromatic side chain and CME (blue, left) or monomers with CBT (blue, right), and ABT (pink) is recognized by aFx.¹⁹²

Custom-depleting pAAs and ARSs from the PURE system and supplementing it with orthogonal tRNAs charged by Fxs is called the FIT system.²⁰¹ The FIT system has been widely used for the translation of peptides containing exMs and will be discussed more in-depth later. To complement the Fxs, Ohta and co-workers created tRNA^AsnE2 (mutations U1G, C2G, G71C, and A72C, similar to tRNA^AsnB), which can be used for sense codon reassignment (Figure 6c).^191,202,203 The FIT system is also amenable to initiation reprogramming using the natural E. coli tRNA^fMet_CAU, which contains a single C1G point mutation to increase transcription efficiency.^204,205 Regarding other orthogonal tRNAs, Katoh and co-workers designed tRNA^Pro1E2 to improve the translation efficiency of exMs (detailed discussion in section 3).¹⁹ The orthogonality of tRNA^Pro1E2 is derived from the E. coli tRNA^Pro isoacceptor class, whose main recognition elements are the anticodon and the discriminator base A73, with minor positive identity elements of G72, G49, and G37.^206,207 Thus, mutations G72C and A73G in the acceptor arm and G49A in the T stem prevented recognition by E. coli ProRS, allowing for complete orthogonality in the FIT system.¹⁹

3. Engineered tRNAs for Improved Polymerization of Non-proteinogenic Monomers

3.1. Peptide Bond Formation in Prokaryotes

In prokaryotes, delivery of noninitiator pAA-tRNAs to the ribosomal A site is facilitated by a ternary complex with EF-Tu-guanosine-5′-triphosphate (GTP) (Figure 8a).⁶⁷ All 20 pAA-tRNAs bind to EF-Tu-GTP with less than a 20-fold difference in equilibrium dissociation constants (K_D) between them, ranging from approximately 0.5 to 10 nM.^208,209 Similarly, pAA-tRNA/EF-Tu-GTP complexes also bind with a uniform affinity to the ribosome with K_D values around 1.0 to 3.0 nM to mediate uniform translation.^208−210 Subsequent polypeptide polymerization occurs in the peptidyl transferase center (PTC) to form peptide bonds. Here, the peptidyl-tRNA in the P site acts as the peptidyl donor and reacts with the peptidyl acceptor α-amine of the pAA-tRNA in the A site (Figure 8b).^{10,67,211,212} The ribosome facilitates peptide bond formation by ordering water molecules, positioning tRNA and ribosomal RNA (rRNA) nucleotides, and electrostatic shielding.^213,214 This complex interplay within the PTC accelerates the peptidyl transfer by about 10⁷-fold through entropic trapping.²¹⁴ Currently, there are two different proposed models for the peptide bond formation mechanism.^10,11 Though slightly different, both models describe the concerted movement of three protons in the rate-limiting transition state; one proton is always contributed by the 2′-hydroxyl of A76 of the peptidyl-tRNA.^10,11 Although peptidyl transfer rate constants can vary by a few orders of magnitude (approximately 0.1 s^–1 to 100 s^–1), the ribosome is generally able to form most peptide bonds in a facile manner.^210,215,216

Figure 8.

Schematic of peptide elongation. (a) First, a pAA-tRNA is brought to the ribosome through a ternary complex with EF-Tu-GTP and accommodated into the ribosome A site. Second, the lone pair of the amino group of the A site pAA attacks the carbonyl group of the peptidyl-tRNA in the P site. As a result, a peptide bond is formed and the peptide chain. Finally, EF-G translocates the A site peptidyl-tRNA to the P site, the now deacylated P site tRNA is translocated to the E site, and a new pAA-tRNA is recruited to the A site by EF-Tu. In this way, the nascent peptide chain is iteratively elongated. (b) Simplified mechanism of peptidyl transfer.⁶⁷

3.2. Low Translation Efficiency of Polypeptides Containing Non-proteinogenic Monomers

Unfortunately, attempted incorporation of npMs, especially backbone-modifying npMs such as C^α-modified AAs and chain elongating AAs, often results in minimal to no detectable product.^217−224 A variety of studies have investigated the use of N-alkyl and α,α-disubstituted AAs using in vitro amber suppression; while a few of these npMs had sufficient translation efficiency, such as pipecolic acid with 43% in vitro amber suppression efficiency into T4 lysozyme,²²⁵ most had 15% to undetectable levels of incorporation.^{190,225−231}

Using E. coli S30 extract, Dedkova and co-workers only observed 3% and 5% amber suppression for d-Phe and d-Met, respectively, into DHFR using yeast suppressor tRNA^Phe_CUA.²¹⁹ Using a similar system, when Maini and co-workers attempted to incorporate (S)-β³-homoalanine (β³-Ala), racemic β²-homoalanine (rac-β²-Ala), β^3,3-homoaminoisobutyric acid (β^3,3-Aib), (S)-β³-homophenylalanine (β³-Phe), or (S)-β³-(p-bromophenyl)alanine (β³-pBrAla), less than 2% amber suppression efficiency for all npMs was observed.²²¹ To solve such issues, one strategy has been to engineer ribosomes with mutations in the PTC.²³² In the aforementioned studies, an engineered ribosome was reported to increase the yield of d-Phe and d-Met by approximately 4-fold,²¹⁹ while a different engineered ribosome was reported to improve translation of some β-AAs with up to 18% suppression efficiency.²²¹

As the FIT system is highly malleable and does not contain components that negatively impact translation like other CFPS or in vivo systems, it has allowed for more detailed studies on the ribosomal translation of backbone-modifying npMs using sense codon reassignment. Kawakami and co-workers used the FIT system to classify the translation efficiency of 23 N^α-methyl-l-amino acids (^MeAA) charged onto tRNA^AsnE2_GGU for incorporation at the ACC (Thr) codon in the 12-mer peptide ^fMet-Arg-^MeAA-Arg-Flag.²³³ Flag is an octapeptide tag (DYKDDDDK; D = Asp; Y = Tyr; K = Lys) commonly used for antibody purification or polypeptide visualization during radioisotope experiments.²³⁴ The relative expression levels of each ^MeAA were compared to Thr-tRNA^AsnE2_GGU incorporation and could be divided into three groups: nine with over 40% incorporation (^MeGly, ^MeAla, ^MeSer, ^MeCys, ^MeHis, ^MePhe, ^MeTyr, ^MeTyr^OMe, and ^MepNO₂Phe), six with 10–40% incorporation (^MeThr, ^MeMet, ^MeGln, ^MeTrp, ^MeNorvaline [^MeNva], and ^MeNorleucine [^MeNle]), and eight with below 10% incorporation (^MeVal, ^MeIle, ^MeLeu, ^MeAsn, ^MeAsp, ^MeGlu, ^MeArg, and ^MeLys).

Similarly, Fujino and co-workers classified the relative incorporation of 19 d-AAs charged onto tRNA^AsnE2_GGA into the 13-mer peptide ^fMet-(Lys)₃-d-AA-Flag at the UCC (Ser) codon relative to their corresponding l-AA.²³⁵ The overall relative incorporation efficiency of d-AAs was lower, with eight having over 40% incorporation (d-Ala, d-Ser, d-Cys, d-His, d-Phe, d-Tyr, d-Met, and d-Thr), four having 10–40% incorporation (d-Gln, d-Val, d-Leu, and d-Asn), and seven showing no incorporation (d-Asp, d-Glu, d-Arg, d-Lys, d-Ile, d-Trp, and d-Pro).

In a separate study, Fujino and co-workers evaluated both (S)-β³-AAs and (R)-β³-AAs (β³-d-AAs) using the same peptide and tRNAs with the previously mentioned d-AA study, though the translation conditions were slightly altered for optimal incorporation.²³⁶ Relative to β³-Gly, seven had over 40% incorporation (β³-Ala, β³-Ile, β³-Gln, β³-phenylglycine [β³-Phg], β³-Met, β³-d-Leu, and β³-d-Phg), six had incorporation efficiencies between 10% and 40% (β³-Leu, β³-Phe, β³-Trp, β³-Asn, β³-Lys, and β³-d-Ala), and three showed no incorporation (β³-Trp, β³-Pro, and β³-Glu).

When comparing these three studies, there is a similar trend that aromatic or small, uncharged side chains were preferable for elongation into the peptide backbone; however, efficiency varied considerably and was often below 40% relative to the respective controls. Additionally, multiple incorporation of npMs further reduced yields, and consecutive incorporation often resulted in truncated or mistranslated peptides, even for highly efficient npMs.^233,235,236 The low translation efficiency of backbone-modifying npMs can generally be attributed to two distinct issues: (1) slow kinetics of EF-Tu-related processes due to decreased affinity of the npM-tRNA to EF-Tu (Figure 9a) and (2) poor peptidyl donor and/or acceptor abilities of the npM, which results in failed peptide bond formation and mistranslocation (Figure 9b).

Figure 9.

Reasons for the low incorporation efficiency of npMs. (a) Poor binding to EF-Tu by npM-tRNAs prevents recruitment to the ribosome and accommodation into the PTC. (b) Slow peptidyl transfer stalls the ribosome, resulting in mistranslocation, peptidyl-tRNA drop-off, and peptide reinitiation.^242,274.

3.3. Engineering tRNA for Increased EF-Tu Affinity

Thorough studies investigating the pAA-tRNA and EF-Tu binding interaction have demonstrated that the thermodynamic contributions (ΔG°) from the esterified pAA and tRNA add independently to around −8.0 to −9.4 kcal/mol for all pAA-tRNAs/EF-Tu-GTP complexes.^7,237−240 The main contribution from the tRNA is the T stem, especially base pairs 49–65, 50–64, and 51–63, with minor contributions from the acceptor arm (Figures 1 and 10).²⁴¹ These bases have evolved to compensate for the differential binding affinities of pAAs to EF-Tu: weak EF-Tu-binding pAAs are assigned to strong EF-Tu-binding cognate tRNAs, while strong EF-Tu-binding pAAs have weak EF-Tu-binding cognate tRNAs.^7,242

Figure 10.

Crystal structures of EF-Tu. (a) Crystal structure of Thermus aquaticus EF-Tu complexed with yeast Phe-tRNA^Phe and 5′-guanylyl imidodiphosphate (GDPNP), a nonhydrolyzable analog of GTP (omitted for clarity) (PDB: 1TTT). EF-Tu interacts with the pAA Phe (red), the acceptor stem (pink), and the T stem (orange). (b) View of the amino acid binding pocket of E. coli EF-Tu interacting with yeast Phe-tRNA^Phe; kirromycin and GDPNP are omitted for clarity (PDB: 1OB2). The chirality of d-AAs causes suboptimal interactions with the amino acid backbone and side chain. The negatively charged residues Glu216 and Asp217 of EF-Tu decrease the binding affinity of negatively charged pAAs and npMs.²³⁹

Similar to pAAs, npMs have a wide range of binding affinities toward EF-Tu, and the lower affinity of some npMs can account for decreased translation efficiency.²⁴³ For instance, bulky npMs such as N^ε-biotinyl-Lys (BioLys) or l-naphthylalanine (Nap) can exhibit over a 100-fold loss of affinity compared to pAAs.^244−247d-AAs also suffer from loss of EF-Tu affinity, potentially due to the inverted chirality causing the hydrogen bonding from the α-amino group to be lost.^248,249 EF-Tu shows a 24-fold decreased affinity for d-Tyr-tRNA^Tyr (K_d = 1.2 μM) as compared to l-Tyr-tRNA^Tyr (K_d = 50 nM).²⁴⁸ Similarly, many ^MeAA have such low affinity for EF-Tu that ΔG° becomes unmeasurable with conventional assays.²⁵⁰ Finally, as the binding pocket of EF-Tu contains two negatively charged residues (Glu216 and Asp217), the negatively charged pAAs Glu and Asp have the weakest ΔG° out of all pAAs (+0.5 and +1.5 kcal/mol compared to the third weakest pAA, Ala, respectively).^237,239 This explains the difficulty of polymerizing negatively charged npMs such as ^MeAsp, ^MeGlu, d-Asp, d-Glu, and β³-Glu.^233,235,236 Thus, npM affinity for EF-Tu can exhibit high variability depending on the exact monomer due to bulkiness, chirality, or charge.

To the increase translation efficiency of npMs with low EF-Tu affinity into polypeptides, the T stem of the orthogonal tRNA should be engineered to possess a higher binding affinity toward EF-Tu. However, the thermodynamic contribution from both the npM and tRNA needs to be tuned appropriately; npM-tRNAs with too strong affinity for EF-Tu can lead to the slow release of EF-Tu-GDP after hydrolysis, decreasing the overall yield of polypeptide formation.^251−253

3.3.1. Evolution of tRNAs with High Affinity for EF-Tu In Vivo

Directed evolution is the most commonly used technique to engineer tRNAs for improved in vivo translation efficiency, though de novo design has recently been explored.^{97,139,254−256}

Guo and co-workers performed a two-step, rational evolution of Mj mutRNA^Tyr_CUA to create tRNAs with stronger affinity toward EF-Tu in E. coli.²⁵⁷ First, five base pairs of the T stem were randomized (pairs 49–65, 50–64, 51–63, 52–62, and 53–61) and were subjected to negative and positive selection. Negative selection removed nonorthogonal tRNAs by translation of a toxic barnase gene containing an in-frame amber codon. In positive selection, translation of an amber mutant of chloramphenicol acetyltransferase produced tRNAs that increased translation efficiency of either Tyr, Nap, pAzPhe, p-iodo-l-phenylalanine (pIPhe), p-acetyl-l-phenylalanine (pAcPhe), or p-benzoyl-l-phenylalanine. Second, the best tRNAs from the first selection were subjected to the same selection scheme but with the eight bases in the acceptor arm known to interact with EF-Tu randomized (2, 3, 6, 7, 66, 67, 70, and 71). As a result, five tRNAs (tRNA^Nap1, tRNA^Nap3, tRNA^pIPhe1, tRNA^pAzPhe1, and tRNA^Tyr1) consistently gave the highest fold increase in protein expression when compared with mutRNA^Tyr_CUA using the previous six npMs and seven additional npMs: p-hydroxy-l-phenyllactic acid, l-bipyridylalanine, l-hydroxyquinolinylalanine, l-sulfotyrosine, p-azobenzyl-l-phenylalanine, o-nitrobenzyl-l-tyrosine, and 7-hydroxycoumarinyl-l-ethylglycine. The evolved tRNAs were predicted to bind to EF-Tu 0.3–0.9 kcal/mol more tightly than mutRNA^Tyr_CUA. Finally, purified yields of GFP containing the Asn149 to amber mutation (AAU to UAG) suppressed by Mj tRNA^Nap1_CUA in E. coli were 3.8- to 11.8-fold higher than those for Mj mutRNA^Tyr_CUA. Overall, the npMs with the bulkiest side chains benefitted the most from the optimized tRNAs, consistent with the aforementioned finding that bulky npMs have lower affinity for EF-Tu.²⁵⁷ Subsequently, Young and co-workers used the consensus sequence of these tRNAs to create Mj tRNA^opt_CUA (mutations G49A, C50G, U51G, G63C, G64C, and C65U relative to Mj mutRNA^Tyr_CUA), which gave rise to the highest translation efficiency.²⁵⁸ Surprisingly, Mj tRNA^opt_CUA only binds to EF-Tu around 0.5 kcal/mol more tightly than Mj mutRNA^Tyr_CUA, which is lower than some of the other evolved mutants. Thus, improved yield in vivo does not simply entail maximizing EF-Tu affinity, and it may be necessary to consider other factors such as cell viability.⁹⁷

Fan and co-workers rationally evolved Methanosarcina tRNA^Pyl_CUA in E. coli using a similar strategy to the previous Mj tRNA^Nap1_CUA.²⁵⁹ The major differences in methodology were the use of a three-step library selection (first randomizing pairs 2–71 and 3–70, then pairs 6–67 and 7–66, and finally pairs 49–65 and 50–64) and optimization of N^ε-acetyl-l-lysine (AcLys) instead of Tyr variants. The most active variant was termed tRNA^Pyl-opt_CUA (C7G, C49U, C50G, G64C, G65A, and G66C) and improved double AcLys incorporation into sfGPF in E. coli by 5-fold compared to wild type tRNA^Pyl_CUA. Additionally, tRNA^Pyl-opt_CUA increased the double incorporation of BocLys, o-iodo-phenylalanine, Tyr^OMe, O-tert-butyl-l-tyrosine, o-cyano-l-phenylalanine, and o-nitro-l-phenylalanine by 1.4- to 2.7-fold. Variants that contained stronger-binding T stems compared to tRNA^Pyl-opt_CUA performed worse with AcLys incorporation, though this might be attributed to the fact that AcLys is not a bulky npM so overtuning the T stem may cause decreased translation efficiency.^241,259

Recently, in 2023, Jewel and co-workers reported a new method of tRNA directed evolution called virus-assisted directed evolution of tRNAs (VADER) which was used to optimize Mm tRNA^Pyl_CUA and E. coli tRNA^Tyr_CUA for use in mammalian HEK293T cells.²⁶⁰ The incorporation efficiency of AcLys was almost 2.5-fold higher when using the evolved tRNA^Pyl_CUA-A2.1 compared to Mm tRNA^Pyl_CUA in response to a UAG codon in enhanced GFP. Similarly, use of the evolved tRNA^Tyr_CUA-NGS6 for incorporation of Tyr^OMe into the same protein was almost 3-fold higher when compared to E. coli tRNA^Tyr_CUA. While mammalian cells do not contain the bacterial EF-Tu, it would be interesting to see if the increased suppression efficiency of evolved tRNAs can be attributed to increased affinity for the analogous mammalian elongation factor EF1a.²⁶¹

Finally, a variety of EF-Tu mutants with higher binding affinities for bulky or negatively charged npMs have also been developed.^{246,262−264} This involves engineering the AA-binding pocket of EF-Tu, which itself contains several bulky and negatively charged residues (Figure 10).^242,262 These two negatively charged residues (Glu216 and Asp217) occlude negatively charged pAAs, such as Sep, Asp, and Glu, and npMs, such as ^MeAsp and ^MeGlu.^233,235 To this end, Park and co-workers developed an EF-Tu mutant, referred to as EF-Sep, which contained six mutated residues in the binding site (H67R, E216N, D217G, F219Y, T229S, and N274W). This allowed for the incorporation the negatively charged Sep into human mitogen-activated ERK activating kinase 1 in E. coli.¹⁴¹ Subsequently, Katoh and Suga used EF-Sep in conjunction with the FIT system to perform single incorporation of ^MeAsp, ^MeGlu, d-Asp, and d-Glu using tRNA^Pro1E2_CGG in response to the CCG codon in the 20-mer peptide ^fMet-(Tyr-Lys-Lys)₂-Tyr-Lys-npM-(Gly)₂-Flag.²⁶⁵ Additionally, consecutive double incorporation of ^MeAsp was possible with a 3.4-fold increase in translation efficiency when EF-Sep was added into the translation mix.

3.3.2. Rational Design of tRNAs with High Affinity for EF-Tu In Vitro

The FIT system has been utilized extensively for the incorporation of npMs. As previously mentioned, Fujino and co-workers were able to successfully perform single incorporation of 12 different d-AAs into peptides using the FIT system with tRNA^AsnE2. However, single incorporation was not possible for the other seven d-AAs, and any successful double incorporations proceeded with extremely low efficiency.²³⁵ When comparing binding ΔG° values for 21 different valylated E. coli tRNAs to EF-Tu, the observed range of ΔG° was 3.6 kcal/mol (−8.1 kcal/mol for Val-tRNA^Tyr to −11.7 kcal/mol for Val-tRNA^Glu).²⁶⁶ As the T stem of tRNA^AsnE2 is derived from the natural E. coli tRNA^Asn, which exhibits low affinity for EF-Tu (Val-tRNA^Asn has ΔG° = −8.9 kcal/mol), it was hypothesized that a tRNA with stronger binding affinity to EF-Tu could be used to compensate for npMs with weak EF-Tu binding and thereby increase incorporation efficiency.²⁶⁶

Achenbach and co-workers used unmodified E. coli tRNA^Gly_CCC (fourth strongest EF-Tu binder) for the incorporation of 18 different d-AAs (Cys was omitted) into a polypeptide at the GGG codon with the FIT system.²⁴⁹ In comparison with the previous study, 16 d-AAs showed successful single incorporation (all except d-Glu or d-Pro), and several permitted consecutive double or triple incorporation, albeit in low yields. In addition, the T stem base pairs 49–65, 50–64, and 51–63 from tRNA^Asp (third strongest EF-Tu binder) were transplanted into an E. coli opal suppressor tRNA^Tyr_UCA (weakest EF-Tu binder) to create a chimeric tRNA^TyrTP_UCA. The translation of six d-AAs was tested (d-Ala, d-Arg, d-Ile, d-Lys, d-Pro, and d-Trp), and the T stem-enhanced tRNA^TyrTP_UCA showed significantly higher single incorporation efficiencies for each of these d-AAs compared to tRNA^Tyr_UCA, except for d-Trp; this may be attributed to steric clashes within the PTC.

In 2017, using a similar strategy, Katoh and co-workers designed a new orthogonal tRNA with increased EF-Tu affinity for the FIT system called tRNA^GluE2, based on tRNA^Glu (highest EF-Tu affinity tRNA) (Figure 11a).^267,268 As the major positive identity elements for GluRS are the anticodon bases, tRNA^GluE2 is naturally orthogonal when not using the Glu codon. Under optimized translation conditions, tRNA^GluE2 was able to increase the consecutive double incorporation of d-Ala into the 14-mer peptide ^fMet-(Lys)₃-(d-Ala)₂-Flag at the UCC codon by 2.9-fold compared to tRNA^AsnE2.²⁶⁷ Additionally, up to 10 consecutive d-Ser residues could be incorporated at the ACU codon in the same peptide.²⁶⁷ Finally, the macrocyclic peptide ^fMet-(Lys)₃-cyc[d-Cys-d-Ser-d-Ala-d-Ser-d-Cys]-Flag (closed by a disulfide bond) was able to be translated with d-Cys-tRNA^GluE2_GUG, d-Ser-tRNA^GluE2_GGU, and d-Ala-tRNA^GluE2_GAU at the CAU, ACU, and AUU codons, respectively.²⁶⁷ Thus, the increased affinity for EF-Tu granted by tRNA^GluE2 makes it especially well-suited for the incorporation of d-AAs.

Figure 11.

Different types of tRNA engineered for use in the FIT system. Engineered tRNAs are synthesized by in vitro transcription and thus lack modified nucleotides. Primary sequences of E. coli tRNA^Pro1_CGG and E. coli tRNA^fMet2 without modified nucleotides are displayed. Important identity elements for its cognate ARS are shown in blue, and mutations from the wild type to the engineered tRNAs are shown in red. The T stem shown in orange is engineered for stronger EF-Tu binding affinity, while the D arm shown in green is engineered for EF-P recognition. (a) tRNA^GluE2 has a T stem that has tighter EF-Tu binding compared to the original tRNA^AsnE2. (b) E. coli tRNA^Pro1_CGG has natural affinity EF-P but is not engineered for high EF-Tu affinity. (c) tRNA^Pro1E2 has combined the elements of tRNA^GluE2 and tRNA^Pro1_CGG for both high EF-Tu affinity and EF-P recognition. (d) Sequence of E. coli tRNA^fMet2. (e) The engineered tRNA^iniP which bears a chimeric structure taken from tRNA^fMet2 and tRNA^Pro1_CGG.^6,19,313.

Previous methods to improve recruitment of EF-Tu focused solely on the tRNA T stem. However, too strong affinity for EF-Tu can be detrimental; thus, the contribution from the esterified npM should also be considered so the overall affinity of each npM-tRNA falls within the uniform range of the −8.0 to −9.4 kcal/mol for pAA-tRNAs. Thus, Iwane and co-workers tuned the binding affinities between EF-Tu for various ^MeAA-tRNAs using four different T stems (base pairs 49–65, 50–64, 51–63, and 52–62).²⁵⁰ In this study, the original tRNA^AsnE2 was termed tRNA^AsnE2#2 and the relative difference in free energies (ΔΔG°) of mutant tRNAs was theoretically calculated. tRNA^AsnE2#1 had a ΔΔG° of +1.4 kcal/mol, tRNA^AsnE2#3 had a ΔΔG° of −0.6 kcal/mol (taken from tRNA^Glu), and tRNA^AsnE2#4 had a ΔΔG° of −1.2 kcal/mol (rationally designed). While the ΔG° of many ^MeAA-tRNA^AsnE2#2 for EF-Tu was greater than −6.0 kcal/mol (^MeVal, ^MeLeu, ^MeThr, ^MeMet, ^MeAsp, ^MePhe, ^MeTyr, ^MeNva, ^MeNle, ^MeTyr^OMe), use of tRNA^AsnE2#3 or tRNA^AsnE2#4 decreased the ΔG° of each ^MeAA-tRNA, down to even −10 kcal/mol in the case of ^MeSer-tRNA^AsnE2#4. As expected, all ^MeAA-tRNA^AsnE2#4 showed higher affinity for EF-Tu than ^MeAA-tRNA^AsnE2#3. As the ΔG° of each ^MeAA-tRNA reached the uniform range of pAA-tRNAs, the translation efficiency and fidelity of the 14-mer peptide ^fMet-(Lys)₃-^MeAA-Tyr-Flag using the GUC codon for incorporation increased. Finally, a 32-mer peptide containing 9 different ^MeAAs was translated using two strategies. The first method charged all ^MeAAs onto the highest affinity tRNA^AsnE2#4 to maximize EF-Tu binding, whereas the second method tuned the ΔG° for all ^MeAA-tRNA to between −7.9 and −8.8 kcal/mol. While the “all tRNA^AsnE2#4” translation resulted in only unknown byproducts, the “uniform tuning” method showed a clean translation product. This highlights that excessive enhancement of the binding affinity might prevent the release of npM-tRNA from EF-Tu after GTP hydrolysis and decrease translation fidelity and efficiency.

3.4. Accelerating Peptidyl Transfer In Vitro via Recruitment of EF-P

Inefficient polymerization due to poor reactivity either as the peptidyl donor in the P site or as the peptidyl acceptor in the A site can be attributed to the incompatibility of the npM in the PTC, especially for backbone-modifying npMs.²⁶⁹ This poor peptide bond formation step can be observed with Pro, as it is polymerized significantly more slowly than all other pAAs.^210,215,216 For non-Pro pAAs occupying the A site, the α-amine is directed toward the peptidyl-tRNA with a distance of approximately 2.9–3.3 Å, and the side chain flips away from the P site.¹⁰ However, crystal structures showed that the Pro α-amine is further away from the P site by about 1 Å more than normal and that the side chain orients toward the P site.²⁷⁰ These two factors may prevent optimal Bürgi–Dunitz angles for peptide formation even though the α-amine is still accessible. Similar issues have also been observed for d-AAs. Due to steric clashes with the universally conserved U2506 base in the PTC, d-AAs are unable to optimally position their α-amine for nucleophilic attack.²⁷¹ In addition, P site-bound peptidyl-d-AA-tRNAs have been reported to stabilize inactive ribosome conformations, preventing sufficient peptidyl donor activity.²⁷² Thus, npMs must be carefully oriented within the PTC to retain sufficient peptidyl donor and/or acceptor activity.

When peptidyl transfer is slow, the ribosome can stall, causing other ribosomes to stack behind the stalled ribosome and lowering translation efficiency.²⁷³ Furthermore, EF-G can prematurely translocate the P site and A site tRNAs to the E site and P site, respectively, resulting in drop-off of the peptidyl-tRNA and reinitiation of translation with the new P site AA-tRNA, termed drop-off reinitiation (Figure 9b).²⁷⁴ To prevent ribosomal stalling during polyproline sequences, bacteria utilize a specialized translation factor known as elongation factor P (EF-P).^275−278 EF-P, which structurally mimics both the acceptor and anticodon stems of tRNAs, binds into the ribosomal E site and stabilizes the peptidyl-tRNA via contacts with the ribosome L1 stalk, the P site tRNA, and the E site codon (Figure 12).²⁷⁹ Often, Lys34 is post-translationally modified to N^ε-(R)-β-lysyl-l-hydroxylysine, such as in E. coli, though other modifications include 5-aminopentanolylation of Lys or rhamnosylation of an analogous Arg residue.^280−287 Recently, it was confirmed that Actinobacteria EF-P contain unmodified Lys and instead rely on a more rigid loop sequence.²⁸⁸ Regardless, this lysine or arginine residue extends into a space adjacent to the CCA-3′ end of the peptidyl-tRNA and forms hydrogen bonds with both rRNA and the peptidyl-tRNA.²⁸⁹ This stabilization position forces the nascent peptide chain to adopt an alternative conformation, placing the peptidyl-tRNA into a favorable position for peptidyl transfer.^279,289 As an additional effect of increased polymerization kinetics, EF-P alleviates peptidyl-tRNA drop-off caused by EF-G-driven mistranslocation, decreasing unwanted translation byproducts.²⁷⁴

Figure 12.

Mechanism of EF-P-mediated peptide bond formation. (a) When peptidyl transfer is slow during polyproline sequences, EF-P enters the E site and recognizes the D arm of tRNA^Pro. Through stabilization of the peptidyl-tRNA by post-translationally modified Lys34, peptidyl transfer activity is increased and promotes peptide bond formation. (b) Crystal structure of E. coli EF-P in the ribosomal E site (green), peptidyl-Pro-Pro-tRNA in the P site (salmon), and Pro-tRNA in the A site (blue) (PDB: 6ENJ). The Pro-Pro in the P site was modeled into the structure by aligning the 6ENJ P site CCA to a Pro-Pro-bound yeast 80S ribosome (PDB: 5DGV). The Lys34 modification of EF-P in E. coli is N^ε-(R)-β-lysyl-l-hydroxylysine, which interacts and stabilizes the CCA end of the peptidyl-tRNA at the PTC.^279,312

While EF-Tu has affinity for all elongator tRNAs and its cognate esterified pAA, E. coli EF-P recognizes only the unique D arm of tRNA^Pro isoacceptors and initiator tRNA^fMet (Figure 11b, d).^279,290 These features consist of a nine nucleotide D loop closed by a stable four base pair D stem with two G-C pairs at positions 12–23 and 13–22.^291,292 Transplanting this D arm motif into unmodified E. coli Pro-tRNA^Ser or Pro-tRNA^Ala allowed for EF-P recognition and increased translation of polyproline sequences.²⁹⁰ Thus, this methodology can be extrapolated for the increased translation efficiency of npMs.

3.4.1. Impact of EF-P on Elongation of Non-proteinogenic Monomers

To investigate whether fully modified E. coli EF-P could be utilized to increase the reactivity of backbone-modifying npMs, Katoh and co-workers used the FIT system to charge backbone-modifying npMs onto an E. coli tRNA^Pro isoacceptor, tRNA^Pro1_CGG.^19,293 As tRNA^AsnE2 and tRNA^GluE2 are transcribed in vitro and do not contain modified nucleotides, unmodified tRNA^Pro1 (instead of wild type) was employed to evaluate the baseline effect of EF-P. First, the consecutive double incorporation of d-Ala into the peptide rP1-(npM)₂ (whose sequence is ^fMet-(Tyr-Lys-Lys)₂-Tyr-Lys-(npM)_n-Gly-Flag) at the CCG, ACU, and CAC codons using tRNA^Pro1_CCG, tRNA^Pro1_GGU, and tRNA^Pro1_GUG, respectively, was tested.¹⁹ Depending on the codon, the expression level was increased 2.6- to 12.0-fold by addition of EF-P. The addition of EF-P also improved the consecutive double incorporation at the CCG codon of the same peptide for d-Ser, d-Cys, d-His, 2-aminoisobutyric acid (Aib, α,α-dimethyl substituted amino acid), β³-Gln, β³-Met, and β³-Phe, ranging from a 1.3- to 20-fold increase.^19,293 Thus, inclusion of EF-P into the FIT system increased the translation efficiency of backbone-modifying npMs charged onto a tRNA that is recognized by EF-P.

While tRNA^GluE2 increases translation efficiency through tighter EF-Tu binding via the T stem, its D arm is unable to be recognized by EF-P. Thus, the T stem from tRNA^GluE2, which has a strong affinity for EF-Tu, was transplanted into tRNA^Pro1, resulting in the chimeric tRNA^Pro1E1 (G49A, A51G, and U63C relative to tRNA^Pro1).¹⁹ Additionally, to prevent recognition of tRNA^Pro1E1 by ProRS, the acceptor arm was further mutated to create tRNA^Pro1E2 (C1G, G72C, and A73G relative to tRNA^Pro1E1) (Figure 11c).^6,19,206,294 Thus, tRNA^Pro1E1 and tRNA^Pro1E2 have T stem and D arm optimizations, allowing for recognition by both EF-Tu and EF-P.

To determine the effect of these newly engineered tRNAs, the relative expression level of rP1-(d-Ala)₂ at the CCG codon with or without EF-P was investigated.¹⁹ Without EF-P, tRNA^AsnE2_CGG had the lowest absolute expression level of approximately 0.05 μM, while tRNA^GluE2_CGG, tRNA^Pro1E1_CGG, and tRNA^Pro1E2_CGG had similar expression levels (about 4.0- to 5.0-fold higher than that of tRNA^AsnE2_CGG) due to enhanced affinity toward EF-Tu. Upon addition of EF-P, tRNA^AsnE2_CGG and tRNA^GluE2_CGG showed no change in expression level, as they are not recognized by EF-P. However, an increase of 4.1-fold for tRNA^Pro1E1_CGG and 5.0-fold for tRNA^Pro1E2_CGG was observed, demonstrating the combined effects of EF-Tu and EF-P.¹⁹ Usage of tRNA^Pro1E2 was also able to increase the expression level of the macrocyclic peptide ^fMet-(Tyr-Lys-Lys)₂-Tyr-Lys-cyc(d-Cys-d-Ser-d-Ala-d-Ser-d-Cys)-Flag (closed by a disulfide bond) by almost 8.0-fold compared to tRNA^GluE2, showing the impact of improved peptidyl transfer of consecutive d-AAs on the peptide expression level.¹⁹ When using β³-Met instead of d-Ala, similar results were observed with a 7.3-fold increase in expression upon addition of EF-P.²⁹³ With optimized translation conditions, up to seven consecutive β³-Met could be incorporated into one peptide. This also marked the first successful consecutive double incorporation of any β-AA. Finally, tRNA^Pro1E2 enabled the expression of a peptide containing five consecutive d-AAs and β³-AAs with β³-Phe-tRNA^Pro1E2_GUG, β³-Phe-tRNA^Pro1E2_GAU, and d-Ser-tRNA^Pro1E2_GGU being encoded at the CAU, AUU, and ACU codons, respectively.²⁹³

Previous attempts to incorporate cyclic β^2,3-AAs into peptides using the PURE system were extremely inefficient, and multiple incorporation was not possible.¹⁸⁹ Katoh and co-workers used tRNA^Pro1E2_GGC to perform single incorporation of the four stereoisomers of 2-aminocyclopentanecarboxylic acid (2-ACPC) and the four stereoisomers of 2-aminocyclohexanecarboxylic acid (2-ACHC) into the peptide rP1-(npM)₁ at the CCG codon (Figure 3).²⁴ Consecutive double incorporation was possible for all stereoisomers of 2-ACPC, and (1R,2R)- or (1S,2S)-2-ACHC into rP1-(npM)₂, and even up to 10 consecutive incorporations of (1S,2S)-2-ACPC were possible for rP1-(npM)₁₀. Interestingly, only (1R,2S)-2-ACPC showed an increase in translational efficiency when using tRNA^Pro1E2 with EF-P, while the other three stereoisomers showed a decrease in efficiency with EF-P.²⁴ Thus, tRNA^Pro1E2 is not a universal promoter of translation efficiency as EF-P may actually promote an unfavorable conformation in the PTC for some npMs. Use of tRNA^GluE2 versus tRNA^Pro1E2 should therefore be evaluated for each npM.

Aromatic cyclic β-AAs based on 2-aminobenzoic acid (2-Abz) have also been incorporated using tRNA^Pro1E2 (Figure 3).²⁹⁵ Recent cryo-EM structures and molecular dynamic simulations of various 2-Abz derivatives within the PTC have indicated that their nucleophilic amine is located 4 Å or more away from the P site carbonyl and that their relatively rigid aromatic backbone sterically occludes optimal fit into the A and P sites, potentially disrupting the proton transfer chain.^296,297 These issues may contribute to the poor translational efficiency of these npMs.^296,297 Katoh and Suga used tRNA^Pro1E2_CGG to perform single incorporation of 2-Abz and seven other derivatives into rP1-(npM)₁ at the CCG codon.²⁹⁵ Inclusion of EF-P increased translation efficiency, ranging from 1.4-fold to 6.8-fold. Consecutive double incorporation was not possible, indicating that 2-Abzs are more difficult to elongate than their fully saturated cyclic β^2,3-AAs counterparts. However, the use of EF-P and tRNA^Pro1E2_CGG still allowed for double incorporation of 2-Abz into one peptide when there were two pAAs separating them in the sequence.²⁹⁵

Like cyclic β^2,3-AAs, Katoh and Suga showed that cyclic γ^2,4-AAs can benefit from the use of tRNA^Pro1E2 with EF-P (Figure 3).²⁹⁸ Using tRNA^Pro1E2_CGG at the CCG codon, cis- and trans-3-aminocyclobutanecarboxylic acid (3-ACBC), all 4 stereoisomers of 3-aminocyclopentanecarboxylic acid (3-ACPC), and (1R,3R)- and (1R,3S)-3-aminocyclohexanecarboxylic acid (3-ACHC) were successfully introduced into a 20-mer peptide. Use of EF-P increased the translation yield of cis-3-ACBC by 1.8-fold and that of trans-3-ACBC by 1.7-fold. Again, consecutive double incorporation was not achieved, but double incorporation of trans-3-ACBC was possible when a single pAAs spacer was introduced.

Finally, tRNA^Pro1E2 has been used to elongate exMs whose backbone is extended by heteroatoms instead of carbons. Katoh and Suga were able to perform single incorporation of six α-aminoxy acids (α-aminoxyacetic acid [^NOGly], l-α-aminoxypropionic acid [l-^NOAla], d-α-aminoxypropionic acid [d-^NOAla], α-aminoxy-isobutyric acid [^NOAib], l-α-aminoxy-β-phenylpropionic acid [l-^NOPhe], and d-α-aminoxy-β-phenylpropionic acid [d-^NOPhe]) and three α-hydrazino acids (N^α-methyl hydrazinoacetic acid [^NNMeGly], l-hydrazinoproline [l-^NNPro], and d-hydrazinoproline [d-^NNPro]) into a 20-mer peptide at the CCG codon with tRNA^Pro1E2_GGC (Figure 3).²³ Although none of the exMs benefited from use of EF-P for single incorporation, five of the α-aminoxy acids used for consecutive double incorporation showed a 1.7- to 10-fold increase in translation efficiency when using EF-P. However, l-^NOPhe showed decreased translation with EF-P, similar to the case with cyclic β^2,3-AAs.²⁴

Recently, Daskalova and co-workers designed a similar tRNA^Pro1_CGG derivative, termed tRNA^Hyb, which recognized EF-P; tRNA^Hyb also contained mutations for increased orthogonality and EF-Tu recognition.²⁹⁹ Instead of the FIT system, the authors utilized E. coli S30 cell extract to incorporate the npMs site selectively in response to the UAG codon at Val10 in DHFR via nonsense suppression. To this end, tRNA^Hyb_CUA was chemoenzymatically aminoacylated with either an adenine-based l-α-AA, l-phosphotyrosine, β³-Phe, β³-BrAla, a fluorescent thiazole dipeptidomimetic, a constrained l-dipeptide, or a constrained hydrazino dipeptide. Additionally, as EF-P recognizes polyproline motifs, two different DHFR templates were tested: template 1 contained the Ala9-UAG10 motif, while template 2 also mutated Ala9 to create the Pro9-UAG10 sequence. As seen with previous studies, EF-P does not always facilitate peptide bond formation; inclusion of EF-P did not significantly improve the incorporation efficiency for the β³-AAs, thiazole, or the constrained l-dipeptide. However, a 5-fold increase in yield was observed for the adenine-based l-α-AA with template 1, while a 5-fold to 7-fold increase was observed for the adenine-based l-α-AA, l-phosphotyrosine, and the constrained hydrazino dipeptide with template 2.²⁹⁹ Thus, EF-P dependent increases in npM translation yield are also possible using cell extract systems with longer polypeptides.

3.4.2. Adapting EF-P for N-Terminal Initiation with Non-proteinogenic Monomers

In bacteria, the canonical initiation event begins with the formation of a 30S initiation complex (30S IC).³⁰⁰ In the 30S IC, an mRNA containing a Shine–Dalgarno (SD) sequence and an AUG start codon is bound to the 30S ribosome, along with three initiation factors (IFs) and the initiator ^fMet-tRNA^fMet_CAU (Figure 13).⁶⁷ While there is no definitive order of binding, IF2 and IF3 generally bind to the 30S subunit first, followed by IF1 and finally ^fMet-tRNA^fMet_CAU.^301,302 Subsequent recruitment of the 50S subunit and dissociation of the IFs form the mature 70S IC. During these complex processes, the overall role of IF2 is to recognize initiator ^fMet-tRNA^fMet_CAU to increase the k_on and stabilize the 70S IC; IF3 increases the k_off for incorrect tRNAs and rate-limits premature inter-subunit ribosomal bridging; and IF1 shapes the 30S subunit and contributes to the kinetic checkpoint of 70S IC formation.^{300,303−306} Once the A site codon is available, the standard cycle of elongation occurs.

Figure 13.

Mechanism of N-terminal drop-off reinitiation during formation of the first peptide bond with an npM. During the 30S initiation complex (30S IC) assembly, IF3 acts as a kinetic checkpoint of npM-tRNA^fMet_CAU in the P site. Either failed positioning or complete dissociation of npM-tRNA^fMet_CAU in the ribosome means the first peptidyl transfer cannot occur if an aberrant 70S IC is formed, causing drop-off of the initiator tRNA and reinitiation of peptide polymerization without the N-terminus. However, quick disassembly of the ribosome by RRF and EF-G can allow for new attempts at correct formation of the mature 70S IC.³⁰⁷

Analogous to drop-off reinitiation during elongation, drop-off of the initiator peptidyl-tRNA can lead to translation initiation occurring at the second position of the mRNA, resulting in the production of peptides lacking their N-terminal amino acid.³⁰⁷ As IFs play important roles in the kinetic association of initiator ^fMet-tRNA^fMet_CAU, it was hypothesized that these factors could be involved in these reinitiation events. Initial biochemical studies have suggested that destabilization of the initiator pAA-tRNA^fMet_CAU by IF3 prevents peptidyl transfer and that this effect is exacerbated when using an npM-tRNA^fMet_CAU with low peptidyl donor activity.³⁰⁷ This seems to be in agreement with previous kinetic studies demonstrating that aberrant 30S IC complexes can still associate with the 50S subunit and form mature 70S complexes, either with incorrectly aminoacylated tRNA^fMet_CAU or in the complete absence of tRNA^fMet_CAU.^304,308

One method to increase expression yields is through increased levels of IF3, EF-G, and ribosome recycling factor (RRF).^307,309 As the npM-tRNA^fMet is not consumed during drop-off reinitiation, fast ribosome recycling of aberrant 70S ICs by EF-G and RRF and tRNA dissociation from the 30S subunit allows for more chances of correctly assembled 70S ICs.³⁰⁷

While ^fMet-tRNA^fMet can also recruit EF-P, there is some debate as to whether one of its natural functions is to participate in forming the first peptide bond.^310−312 Regardless, if EF-P could be used to properly reposition the initiator tRNA in the P site and enhance the peptidyl transfer reaction, then drop-off reinitiation could potentially be suppressed, especially in the case of npMs.

In the standard FIT system, unmodified E. coli tRNA^fMet2_CAU is charged with an npM and subsequently used for initiation at the AUG codon.¹⁸ While the body of tRNA^fMet2_CAU is quite different from that of tRNA^Pro1, the D arm used for EF-P recognition only differs in the 11–24 base pair: A11-U24 for tRNA^fMet2_CAU and C11-G24 for tRNA^Pro1 (Figure 11b, d).^291,292 Thus, Katoh and Suga created the mutant tRNA^{iniG1/C11/G24} as a starting point to test the effect of EF-P on initiation (Figure 11e).³¹³ The A11C and U24G mutations modify the tRNA^fMet2_CAU D stem to be identical to tRNA^Pro1 for optimal EF-P recruitment, while the C1G mutation is for improved in vitro transcription efficiency. These mutations also have the additional benefit of reducing recognition by methionyl-tRNA formyltransferase, allowing for a more orthogonal initiator tRNA.^314−316

When initiating translation with N^α-acetyl-l-proline (^AcPro) at the AUG codon, expression levels with tRNA^{iniG1/C11/G24}_CAU were 5.1-fold higher without EF-P than tRNA^fMet2_CAU and 13.0-fold higher in the presence of EF-P.³¹³ When comparing the full-length peptide to the drop-off reinitiated peptide expression levels, 17% and 34% full-length peptide was observed for tRNA^fMet2_CAU and tRNA^{iniG1/C11/G24}_CAU, respectively, in the absence of EF-P, and 24% and 43% full-length peptide in the presence of EF-P.³¹³ Interestingly, even without the use of EF-P, the triple mutant tRNA^{iniG1/C11/G24}_CAU showed improved expression yields and full-length peptide translation, potentially indicating that these mutations may lead to some productive conformational change within the PTC for ^AcPro incorporation.

To further improve full-length peptide yield, different combinations of five anticodon stems, five acceptor arms, five T stems, and five variable loops were introduced into tRNA^{iniG1/C11/G24}_CAU.³¹³ These mutant stems and loops were derivatives of the tRNA^Pro1 sequence. The best combination of these was tRNA^iniP (Figure 11e), which contained 11 tRNA^Pro1-specific mutations: two in the D stem, five in the anticodon stem, one in the variable loop, and three in the T stem. Use of ^AcPro-tRNA^iniP_CAU resulted in an absolute expression level of 1.12 μM (16.9-fold increase relative to tRNA^fMet2_CAU) and a full-length peptide percentage of 55% (2.3-fold increase relative to tRNA^fMet2_CAU) in the presence of EF-P.³¹³

In addition to using tRNA^iniP, the concentrations of various translation factors that influence N-terminal drop-off—such as IF3, EF-G, and RRF—were optimized, as well as the Shine–Dalgarno sequence present before the start codon.³¹³ Finally, as the FIT system does not require the AUG codon for initiation, a variety of other initiation codon–anticodon pairs were tested, with tRNA^iniP_CUU decoding the AAG codon found to be the most efficient combination. With these optimized conditions, drop-off reinitiation of ^AcPro was completely suppressed and an absolute expression level of 59.1 μM achieved, an over 1000-fold improvement compared to the standard conditions used in the FIT system (tRNA^fMet2_CAU with no EF-P).³¹³

To examine the scope of npMs that could be initiated by tRNA^iniP_CUU, ^Acd-Trp, ^Acd-Tyr, ^Acβ³-Phg, and ^Ac3-Abz were introduced at the N-terminus as representatives of d-AAs, β³-AAs, and cyclic γ^2,4-AAs.³¹³ Consequently, addition of EF-P led to an 81-fold, 390-fold, 7.1-fold, and 1.7-fold improvement of expression levels, respectively, and almost no detectable drop-off reinitiation. Finally, tRNA^iniP_CUU could be used to initiate translation with N-chloroacetyl-l-proline (^ClAcPro) and ^ClAc3-Abz, leading to spontaneous cyclization reaction between the N-chloroacetyl group and a downstream Cys to create thioether-closed macrocyclic peptides with a rigidified cyclization motif.

4. Decoding Fidelity and Degeneracy of the Genetic Code

4.1. Decoding of the Codon-Anticodon Duplex

In the standard genetic code, most pAAs are encoded by at least two distinct codons. The concept that one pAA is encoded by multiple different codons is termed the degeneracy of the genetic code, and different codons that encode the same pAA are known as synonymous codons. As all known organisms have less than 61 distinct tRNAs for each of the 61 elongator codons (for example, E. coli have 43 tRNAs with 40 distinct specificities), many tRNAs necessarily decode more than one codon.^33,317 This leads to isoacceptor tRNAs which encode the same pAA but have different anticodons. Thus, accurate protein synthesis requires rapid and efficient decoding of these degenerate mRNA messages by the tRNAs. Overall, translation errors are estimated to occur on the order of 10^–3–10^–5, which is a net accumulation from transcription (∼10^–4–10^–5), aminoacylation (∼10^–4–10^–5), and ribosomal decoding (∼10^–3–10^–4).^318−322 Understanding how the ribosome reads the codon–anticodon complex to reject near-cognate (one mismatch) or noncognate (two or more mismatches) pAA-tRNAs is crucial in designing engineered tRNAs that can properly decode cognate codons and efficiently incorporate npMs.

4.1.1. Recognition of the First Two Codon Bases

Once the pAA-tRNA has been delivered to the ribosome by EF-Tu-GTP, the anticodon stem-loop (ASL) is inserted into the A site decoding center. Here, nucleotides 3′-N36-N35-N34-5′ of the tRNA anticodon form three base-pair interactions with nucleotides 5′-N′1-N′2-N′3-3′ of the mRNA codon in a stepwise manner, starting from N′1 (Figure 14).³²³ Cognate tRNAs allow for local rearrangements of rRNA bases, inducing 30S domain closure, subsequent hydrolysis of GTP, and tRNA accommodation into the A site.³²⁴ Most near- and noncognate pAA-tRNAs are unable to form WC interactions in the rigid decoding center and 30S domain closure does not occur, encouraging dissociation from the ribosome.^{323,325−329} While a subset of near-cognate pAA-tRNAs can form nonstandard base pairs that may sample WC-like conformations, GTPase activation and accommodation is generally slower, promoting dissociation before peptidyl transfer.^{21,327,330−332} Overall, these initial proofreading steps closely monitor the WC geometry of N36•N′1 and N35•N′2 via the minor groove. To facilitate these conserved WC interactions, N36 and N35 are almost never modified in any organism.^333,334

Figure 14.

Codon–anticodon interaction in the ribosome decoding center. Nucleotides 3′-N36-N35-N34-5′ of the tRNA anticodon form three base-pair interactions with nucleotides 5′-N′1-N′2-N′3-3′ of the mRNA codon, respectively. The anticodon stem is bases 27–31 and 39–43; the “extended anticodon” is bases 31–33 and 37–39.³³⁴.

4.1.2. Influence of the Third Codon Base—Wobble Base Pairs

Compared to the N36•N′1 and N35•N′2 pairs, where WC pairs are almost a necessity, there are significantly fewer contacts between the N34•N′3 pair and the ribosome.³²⁹ This allows for the formation of wobble bases. The original wobble hypothesis states that U34 can form U34•A′3 or U34○G′3 pairs, G34 can form G34•C′3 and G34○U′3 pairs, C34 can form the C34•G′3 pair, and A34 can form the A34•U′3 pair.³³⁵ However, tRNA is the most highly modified type of RNA with over 100 modifications currently known.²⁹¹ As many of these modifications occur on N34, the wobble rules have been revised to describe how modification of N34 influences which N′3 bases can be decoded (reviewed in detail in refs (33, 334, 336, and 337)). As different organisms can exhibit distinct modification patterns, the analysis in sections 4.1.2, 4.1.3, and 4.2 will be limited to the 43 E. coli elongator tRNAs (Figure 15).

Figure 15.

Codon table for E. coli. tRNA anticodons are on the left and are connected to their cognate codon sequences on the right. An asterisk indicates a modified tRNA base. More detailed information on tRNA anticodon modifications can be found in refs (291, 334, and 337).

The distribution of codons can be grouped into three different categories: eight unsplit boxes where four synonymous codons code for the same pAA, five 2:2-split boxes with two synonymous codons for each of two pAAs, and three special boxes for Tyr/stop, Ile/Met, and Cys/stop/Trp. Generally, unsplit boxes contain GC-rich codon–anticodon duplexes with larger ΔG° (around −5 kcal/mol), while 2:2-split and special boxes contain AU-rich codon–anticodon duplexes with smaller ΔG° (around −1.8 kcal/mol).^334,338 To maintain accurate decoding, each codon must be decoded with similar efficiencies. Thus, different strategies have evolved for each type of codon box to stabilize AU-rich codons or destabilize GC-rich codons.

In unsplit boxes, no modifications are found on G34 or C34. As these boxes are composed of more thermodynamically stable codon–anticodon duplexes, modification of G34 or C34 is not necessary. However, as the unmodified U34○G′3 wobble pair is less thermodynamically stable than the standard G34○U′3 wobble pair, U34-containing tRNAs are often modified.³³⁴ Thus, modification of U34 to uridine 5-oxyacetic acid (cmo⁵U34) allows for improved recognition of G′3, A′3, and U′3,^339−341 while 5-methylaminomethyluridine (mnm⁵U34) only increases G′3 and A′3 decoding efficiency.^342−344 The one unique case is the Arg unsplit box, where an A34-containing tRNA is deaminated to inosine, which has an expanded decoding ability to read C′3, U′3, and A′3.³⁴⁵

The 2:2-split and special boxes require more complex N34 modifications to stabilize weak codon–anticodon duplex interactions. As G34-containing tRNAs can decode both NNC and NNU codons, the upper half of these boxes only have one tRNA to decode both codons. While some tRNAs contain unmodified G34, it can also be modified to either queuosine (Q34) or glutamyl-queuosine (gluQ34).^346,347 Q34-containing tRNAs are able to decode both U′3 and C′3, and gluQ34 may prevent misreading of N′2.^{337,348−351} In the lower half of these boxes, there are two unmodified C34-containing tRNAs and two modified 2′-O-methylcytidine (Cm34)-containing tRNAs. The Cm34 modification may increase the decoding fidelity of G′3, which is important for tRNA^Trp_CmCA, as recognition of the near-cognate UGA stop codon would cause readthrough translation.^325,352,353 All five U34-containing tRNAs are modified to 5-iminomethyl-uridine derivatives (xnm⁵U34). Although modification to xnm⁵U34 only allows for the standard wobble decoding of A′3 and G′3, these codons are now read with similar efficiencies.^345,354,355 Finally, the special Ile/Met box employs two unique modifications: N4-acetylcytidine (ac⁴C34) to decode the Met AUG codon and k²C34 to decode the Ile AUA codon. While ac⁴C34 actually decreases the affinity of elongator tRNA^Met_ac4CAU for its cognate AUG codon, the modification prevents misreading of Ile AUA codon.^356,357 On the other hand, k²C34 completely changes the hydrogen bonding abilities of C34, allowing for formation of k²C34•A′3 WC pairs instead of wobble pairs.^358,359

As most tRNAs used in genetic code manipulation are unmodified, the removal of anticodon loop modifications can cause unintended loss of translation efficiency and fidelity. Early experiments using a 70S ribosome in vitro showed that the dissociation rate is often significantly faster for unmodified pAA-tRNAs than for modified pAA-tRNAs in both the ribosome A site and P site.³⁶⁰ Additionally, some organisms (or organelles) take advantage of superwobbling, wherein unmodified U34-containing tRNAs can decode all four N′3 bases; this phenomenon has also been observed in vitro.^361−365

While unmodified U34-containing tRNAs do not exist in E. coli, tRNAs used for genetic code manipulation may contain unmodified U34 and are more likely to mis-decode near-cognate codons.³⁶⁶ Conversely, G34 and C34 are often unmodified in natural tRNAs. Thus, using unmodified G34- or C34-containing tRNAs may result in lower levels of misincorporation. Recent work has demonstrated that all 31 possible combinations of unmodified E. coli elongator tRNAs with the SNN anticodon were able to correctly decode and translate their cognate NNS codons with similar levels of efficiency (where S is G or C).³⁶⁷

4.1.3. Covariance between the Anticodon and Neighboring Bases

In the early 1980s, Yarus proposed that there was a strong covariation between the last anticodon base N36 and various nucleotides within the E. coli ASL, introducing the concept of the “extended anticodon”.³⁶⁸ Multitudes of studies have corroborated this apparent but indirect link through mutational studies of bacterial tRNAs.^{334,369−372} As more organisms have had their tRNAs elucidated and a similar covariation observed, it has become more evident that this concept is particularly relevant for bases N31, N32, N37, N38, and N39.^334,373 The major exception is the near invariant U33, which is necessary for proper anticodon loop formation, though a few eukaryotic initiator tRNAs contain C33.^374−376

N31•N39 pairs always form WC interactions and denote the end of the anticodon stem.³³⁴ In unsplit boxes for N31•N39, all tRNAs contain the relatively strong C•G pair, while only two contain the weaker U31•A39 pair. Conversely, 70% of tRNAs in 2:2-split and special boxes have the weaker A•U or A•Ψ pairs, and only 30% have the stronger C•G pair.³³⁴

The first pair in the anticodon loop, N32○N38, is unable to form proper WC interactions due to suboptimal nucleoside orientation.^334,369,370 Over 80% of tRNAs contain a bifurcated hydrogen bond between the N32 carbonyl (over 90% are Y32) and an amino group of N38 (over 70% are A38).^370,371 Approximately 10% of tRNAs contain the U○U pair that forms a single hydrogen bond, and the remaining less than 10% consist of nonisosteric pairs.^370,371 As with the actual anticodon, unsplit boxes seem to have evolved different strategies for decoding compared to 2:2-split and special boxes due to codon–anticodon duplex energetics. GC-rich codon–anticodon duplexes tend to have tRNAs with stronger pairs (A○U, Y○Y), while AU-rich codon–anticodon duplexes tend to have weaker pairs (Y○Y, C○A).³³⁴

Thus, the general trend appears to be that tRNAs harboring stronger codon–anticodon duplexes contain stronger N31•N39 and N32○N38 pairs, while weaker codon–anticodon duplexes contain weaker N31•N39 and N32○N38 pairs. Strong interactions between N31•N39 and N32○N38 may antagonize the canonical anticodon loop conformation, preventing optimal binding of GC-rich codons.^{366,371,372,377,378} This is likely to uniformly tune the translation rates of each codon and prevent binding to near-cognate codons.^210,377 Conversely, weaker base interactions allow for an open anticodon hairpin and easier base pairing interactions in the A site.³³⁴

The final important base is R37 (about 85% A37 and 15% G37), which stabilizes N36 and rigidifies the anticodon hairpin to prevent frameshifting and mis-decoding of near-cognate codons.³³⁴ As R37 is by far the most modified nucleoside in the ASL (about 75% modified), removal of these modifications when using engineered, unmodified tRNA could be problematic.^379,380 For example, when the modification of 2-methylthio-N6-isopentenyladenosine (ms²i⁶A37) was removed from E. coli tRNA^Phe_GAA, Phe was misincorporated into the near-cognate CUU codon during in vitro translation of β-lactamase.³⁸¹ Unmodified R37 can also cause the anticodon loop to be overly flexible, collapsing into a structure with unwanted base pairing and reduced translation efficiency.^3,382−385

Predicting the aggregate effect of removing all modified nucleotides from a tRNA on decoding and translation efficiency can be difficult, with some unmodified tRNAs still recognizing their cognate codon and performing translation accurately, while others may lose most (or all) of their native activity.^384,386,387 However, a general rule which may be useful is that tRNAs which decode GC-rich codons are modified to tune down binding affinity, while tRNAs which decode AU-rich codons require modifications to increase binding affinity to cognate codons. Thus, to retain the overall structure of the ASL, one solution is to transplant the entire anticodon loop into an already functioning orthogonal, unmodified tRNA instead of just mutating the anticodon.³⁸⁸ While this may reduce translation efficiency or cause mis-decoding, optimization of bases in the ASL can potentially rescue these engineered tRNAs.^138,255,371

4.2. Misincorporation of Proteinogenic Amino Acids by Near- and Noncognate tRNAs

Accurate decoding of cognate codons by npM-tRNAs is important for translation fidelity as competitive mis-decoding by near-cognate pAA-tRNAs will hinder suppression efficiency. Thus, it is necessary to understand the mis-decoding patterns of pAA-tRNAs when determining which codons to suppress during genetic code manipulation.

One of the most common methods to determine the mis-decoding profile of a specific codon is to starve the translation system of the cognate pAA, exacerbating misincorporation by noncognate pAAs.^389−393 Recently, Katoh and Suga performed a comprehensive in vitro analysis of all potential 1,159 mis-decoding possibilities (61 sense codons and 19 noncognate pAAs per codon) using a FIT system where an arbitrary pAA was removed to create a vacant codon, termed the 19-aa FIT system.³⁹⁴ Model peptides were translated from genes containing a single instance of each vacated codon and were analyzed by LC-MS to determine the extent of misincorporation by each pAA (Figure 16). In this study, replication and transcription errors were minimal, as misincorporation was enforced by the complete removal of particular pAAs, though misaminoacylation errors could not be fully discounted.

Figure 16.

Misincorporation frequencies observed for all 1,159 different possibilities in the FIT system. An asterisk indicates a modified tRNA base. (a) Asn-tRNA^Asn_QAA reading the Ser AGC codon would result in Asn misincorporation via U35○G′2 mismatch, leading to a green arrow from the NAN to NGN box. In a separate example, Ile misincorporation into the Phe CUU codon could occur by two different tRNAs: Ile-tRNA^Ile_GAU via U36○U′1 mismatch, or Ile-tRNA^Ile_k2CAU via U36○U′1 mismatch and k²C34○C′3 mismatch. This would result in a red arrow for mis-decoding within the NUN box. Finally, Met-tRNA^Met_ac4CAU misincorporation into the Thr ACG codon would occur by A35○C′2 mismatch, leading to a brown arrow from the NUN to NCN box. (b) Graphical representation of the most common mis-decoding events observed using the FIT system. A pAA in the NUN or NCN codon box to a different pAA within the same NUN (87.5% of 64 possible combinations, red arrow) or NCN (91.7% of 48 possible combinations blue arrow) codon box was extremely prevalent. This would be due to mismatches at N36○N′1 and/or N34○N′3. NUN codons were frequently mis-decoded by pAAs in the NCN box (75.0% of 64 possible combinations, purple arrow); similarly, NCN codons were also mis-decoded by pAAs in the NUN box (53.8% of 80 possible combinations, brown arrow). This is necessarily due to mismatches at N35○N′2, but N36○N′1 and/or N34○N′3 were also observed. Mis-decoding events in the NAN and NGN codons were less frequent. Thus, only instances of mis-decoding with >50% frequencies are shown. Yellow arrows indicate mis-decoding caused by G○U/U○G base pair formation at either N36○N′1 or N34○N′3; for example His-tRNA^His_QUG mis-decoding the Tyr UAC codon results in a G36○U′1 mismatch. Green arrows indicate mis-decoding caused by G○U/U○G base pair formation at N35○N′2, for example Tyr-tRNA^Tyr_QUA mis-decoding the Cys UGC codon results in a U35○G′2 mismatch.³⁹⁴

The analysis indicated that mis-decoding of a pAA in the NUN or NCN codon box to a different pAA within the same NUN or NCN codon box was extremely prevalent (87.5% of 64 possible combinations and 91.7% of 48 possible combinations, respectively), indicating that mismatches at N36○N′1 and N34○N′3 are tolerated by the ribosome.³⁹⁴ Additionally, it was also observed that pAAs in the NUN box were frequently mis-decoded by pAAs in the NCN box and vice versa (75.0% of 64 possible combinations and 53.8% of 80 possible combinations, respectively), indicating that second-base G35○U′2 and A35○C′2 mismatches are also possible. Conversely, pAAs in the NGN and NAN codon boxes showed significantly lower mis-decoding frequencies for all three codon bases (<35% for all combinations). When mis-decoding did occur for NGN and NAN codons, it was most often due to a G○U or U○G pair at N36○N′1 or N35○N′2. As expected, misincorporation occurred more frequently for codon–anticodon pairs with only one mismatched base compared to those with two or three (49.4%, 18.7%, and 14.7%, respectively). However, it is surprising that mis-decoding can occur even with three mismatches, such as between tRNA^Asn_QUU and nine different noncognate codons (Leu UUA, Leu UUG, Leu CUA, Leu CUG, Val GUA, Val GUG, Pro CCA, Pro CCG, and Arg CGA).³⁹⁴ Large-scale in vivo studies have uncovered similar mis-decoding rules for the ribosome; misincorporation is most often due to a combination of G○U/U○G mismatches at N36○N′1 or N35○N′2 and wobble mismatches at the third codon base (notably U34○G′3, U34○C′3, and U34○U′3).^395−397

In these studies, there is a recurrent statistical bias of G appearing within the codon (most often at G′2) and mismatching to U within the anticodon.^394−398 In the case of the prevalent U35○G′2 mismatch, this bias seems to stem from mis-decoding tRNAs which are highly modified and can replace stronger, but less modified, cognate tRNAs.³⁹⁸ Additionally, some differences in misincorporation abundance between synonymous codons have been observed.^386,394,397 For example, both in vitro and in vivo studies showed that the Cys UGC codon is statistically more likely to be mis-decoded by tRNA^Tyr_QUA than the Cys UGU codon.^394,397 These results potentially indicate that neighboring bases can directly affect the codon–anticodon duplex stability during mis-decoding events.^329,386

A plethora of crystallographic studies have begun to characterize such base-pair mismatching within the ribosome decoding center.^{325,328,329,398−402} The most recent studies of 70S ribosomes with G○U mismatches at N36○N′1 or N35○N′2 revealed that they adopt Watson–Crick-like geometry instead of the expected Watson–Crick wobble geometry. While previous studies identified these interactions as having wobble geometry, this discrepancy could likely stem from older studies conducting crystallography using only isolated 30S subunits instead of complete 70S ribosomes.^{324,328,329,398} Regardless, G○U mismatches are stabilized by the conserved key nucleotides A1493 and A1492/G530 of the 16S rRNA, similar to canonical base pairs, likely explaining why this mis-decoding event is most often observed (Figure 17).³⁹⁸ Other mismatches, such as A○A, C○A, and U○U, adopted isosteric or very similar Watson–Crick geometry, allowing for recognition by the ribosome decoding center; however, the nitrogenous bases of these mismatches were shown to either not interact or be destabilized, explaining why these mis-decoding events are less common.^399,400

Figure 17.

A comparison between the canonical Watson–Crick G•C pair (PDB: 6GSJ) and the wobble G○U pair (PDB: 6GSK) using Thermus thermophilus 70S ribosome at N35/N′2. (a) Cognate mRNA codon ACC and E. coli tRNA^Thr_GGU form a Watson–Crick G35•C′2 pair. (b) Near-cognate mRNA codon AUC and E. coli tRNA^Thr_GGU form a wobble G35○U′2 pair. Black dashed lines indicate the interatomic distance ≤3.3 Å, while the red solid line indicates the G and U carbonyls which are 3.5 Å apart. These crystal structures show that these pairs are structurally nearly indistinguishable, though some energetic differences may be present.³⁹⁸

Various studies attempting to incorporate exMs into easily mis-decoded codons in the FIT system have corroborated these mis-decoding rules. When attempting to reprogram the Thr ACC codon to α-thio Ala (^HSAla) using tRNA^GluE2_GGU, Maini and co-workers reported persistent Cys misincorporation.⁴⁰³ Lee and co-workers noticed significant background mis-decoding to Ser when attempting to incorporate α-hydrazino acids at the ACC codon using tRNA^Pro1E2_GGU.⁴⁰⁴ These results are consistent with the Thr ACC codon being decoded by 10 different pAAs, with the most common being Ser or Cys. In a separate experiment, Katoh and Suga attempted to perform consecutive double incorporation of ^NNMeGly and l-^NNPro at the Pro CCG codon using tRNA^Pro1E2_CGG.²³ Instead of the desired peptides, overwhelming Ala and Gly misincorporation was observed. Additionally, while d-^NNPro was translated efficiently with tRNA^Pro1E2_CGG, switching to the less efficient tRNA^AsnE2_CGG resulted in Gly misincorporation. These results are consistent with Gly being one of the easier pAAs to misincorporate into the Pro CCG codon.

Misincorporation has also been observed during in vivo amber codon suppression. Johnson and co-workers identified that Tyr, Gln, and Trp were often misincorporated at the UAG codon when attempting to incorporate pAcPhe with an Mj LW1RS/mutRNA^Tyr_CUA pair in some E. coli strains.⁴⁰⁵ When Odoi and co-workers used the unengineered Mm PylRS/tRNA^Pyl_UCA pair for opal UGA codon suppression for incorporation of BocLys into sfGFP in E. coli BL21(DE3) cells, only Trp misincorporation was observed.¹¹⁵ Finally, while Tang and co-workers successfully incorporated a photocaged 2,3-diaminopropionic acid (Dap) derivative into the high-temperature requirement protein A2 at an amber UAG codon in the HEK293T human cell line using an evolved Mb DapRS/tRNA^Pyl-opt_CUA pair, near-cognate suppression by Gln and Tyr was also observed.⁴⁰⁶

Although the mechanistic understanding of misincorporation remains incomplete, breakthrough experiments in the past decade have uncovered the main patterns and structural basis of mis-decoding, which in turn can inform decisions on how to better perform genetic code manipulation. For example, it may be beneficial to choose codons without G′2 for npM incorporation. These results not only demonstrate the necessity of choosing the correct codon at which to incorporate npMs but also the importance of using engineered tRNAs with sufficient affinity to EF-Tu and EF-P.

4.3. Liberating Codons for Genetic Code Manipulation

Assignment of new npMs to the genetic code is complicated by the fact that all 64 existing codons are already mapped to pAAs or termination signals by natural tRNAs or RFs. Therefore, methods to either repurpose existing codons or design new codons are a prerequisite for genetic code manipulation. As stated previously, these codons must be efficiently decoded by their cognate tRNA and minimally mis-decoded by near- or noncognate tRNAs. Assignment of a well-matched codon–anticodon pair is clearly essential but does not constitute the sole factor for efficient assignment. The entirety of the tRNA body, including nucleotide modifications, contributes to decoding fidelity and has been subject to engineering efforts (Figure 18).

Figure 18.

Different types of genetic code manipulation. (a) Nonsense suppression utilizes one or more of the three stop codons: Amber (UAG), Ochre (UAA), or Opal (UGA). (b) Sense codon reprogramming frees sense codons through removal of pAAs, tRNAs, and/or ARS; then, an npM is reintroduced at the blank codon, allowing for up to 61 different options. (c) Quadruplet codons introduce a fourth base pair within the anticodon to create up to 256 different codon/anticodon combinations. (d) Unnatural base pairs insert novel base pairs, which allows for an expanded codon table with new interactions.

4.3.1. Artificial Division of Codon Boxes by Sense Codon Reassignment to Incorporate Non-proteinogenic Monomers

Due to codon degeneracy, removal of certain pAAs from the genetic code provides multiple potential free codons for reassignment to npMs (Figure 18b).^{250,407−409} For instance, by completely removing Arg from the FIT system, Passioura and co-workers reassigned three of the liberated codons (AGG, CGC, and CGG) to ^MeTyr^OMe, O-methyl-l-threonine, and d-Ala, respectively.⁴⁰⁷ Removal of further pAAs from the translation system enabled the creation of a radically reprogrammed genetic code containing 11 npMs and 12 pAAs. Similarly, by removing Phe, Leu, Ile, Val, and Ala from the genetic code, Iwane and co-workers in 2021 introduced nine ^MeAAs at the freed codons.²⁵⁰

An alternative approach for in vitro sense codon reassignment involves removing a synonymous set of natural tRNA isoacceptors and replacing them with orthogonal npM-tRNAs.¹⁵ Notably, by reassigning only a subset of degenerate sense codons to npMs, concurrent removal of pAAs from the genetic code may be avoided. In 2016, Iwane and co-workers created a fully modular FIT system using only synthetic tRNAs harboring SNN anticodons and showed that all pAAs could be efficiently translated from all 31 NNS sense codons (Figure 19).³⁶⁷ This system was used as the starting point for synonymous codon reassignment to npMs. Synonymous codon boxes of Gly (GGS), Arg (CGS), and Val (GUS) were each split to decode both their original cognate pAA and an npM, with the npMs introduced by tRNA^AsnE2.³⁶⁷ This enabled the creation of a genetic code with 23 distinct monomers: three npMs and 20 pAAs. Subsequent reports by Hibi and co-workers and Fujino and co-workers reduced the synthetic tRNA-based FIT systems to a minimal set of 21 tRNAs and demonstrated their flexibility with the translation of proteins such as GFP or DHFR, providing a future basis for even more extensive reprogramming.^410,411

Figure 19.

Sense codon reassignment via breaking the degeneracy of codon boxes. NNS codons are decoded by synthetic tRNAs containing the cognate SNN anticodons, where N = A, U, C, or G and S = C or G. This results in 31 sense codons (including initation) and one stop codon that code for 20 pAAs. To incorporate npMs while still retaining 20 pAAs, a degenerate codon for each of Gly, Arg, and Val was reassigned to decode p-iodo-l-phenylalanine (pIPhe), N-methyl-l-serine (^MeSer), and N-methyl-l-tyrosine (^MeTyr), respectively. This created a genetic code with 23 unique monomers instead of only the standard 20 monomers. Finally, a peptide containing three distinct npMs was synthesized using the synthetic tRNA FIT system.³⁶⁷

Although these reports demonstrate that synthetic tRNAs can be used for in vitro translation, some tRNAs, such as tRNA^Lys_CUU and tRNA^Ile_GAU, show relatively poor decoding, likely due to the absence of nucleotide modifications in the native tRNAs.³⁶⁷ Therefore, other work has focused on replacing only a subset of endogenous tRNAs with synthetic versions. In 2016, Lee and co-workers demonstrated the use of colicin D, a tRNA^Arg specific RNase, to selectively remove all tRNA^Arg species from E. coli lysate, creating six blank codons.⁴¹² This lysate was then supplemented with in vitro transcribed tRNA^Arg_CCU to restore Arg to the genetic code at the AGG codon. After confirming that degeneracy was broken, the CGN Arg codons could each be separately reprogrammed to AcLys or N^ε-methyl-l-lysine for incorporation into enhanced GFP using tRNA^AsnE2, albeit with background misincorporation by some pAAs. In addition, CGU and CGG could be simultaneously recoded to distinct npMs.

An alternative method for replacement of specific endogenous tRNAs is to use DNA hybridization chromatography, as reported by Cui and co-workers in 2017.⁴¹³ DNA oligonucleotides complementary to subsequences of certain tRNAs were prepared and immobilized on sepharose resin to remove the targeted tRNAs from the total E. coli tRNA mixture. Unlike the colicin D-based depletion system, this method does not depend on the existence of tRNases of suitable specificity and is therefore potentially applicable to diverse sets of tRNA isoacceptors. Hybridization-based depletion of tRNA^Arg_mnm5UCU and tRNA^Arg_CCU reduced translation of GFP containing a single AGG codon by approximately 90%. Supplementation of the depleted lysate with an orthogonal Mj TyrRS/tRNA_CCU pair or an in vitro preaminoacylated tRNA^Cys_CCU enabled the incorporation of pAzPhe or BODIPY FL, respectively, into GFP and calmodulin at the AGG codon. However, hybridization-based depletion of tRNA^Ser_GCU, which decodes the much more abundant AGC and AGU codons, only reduced translation of the corresponding GFP template (containing a single AGC codon) by 60%. A subsequent alteration of this strategy used 2′-OMe-modified antisense oligonucleotides to sequester endogenous tRNAs directly within the lysate instead of using hybridization chromatography to remove the tRNAs. Simultaneously sequestering tRNA^Ser_GCU and supplementing the lysate with an orthogonal ARS/tRNA_ACU pair allowed for one-pot sense codon reassignment at the Ser AGU codon.¹⁷⁹

In the above works, the reassigned tRNAs must still be synthetically produced and are thus lacking nucleotide modifications. In 2022, McFeely and co-workers reported a strategy for sense codon reassignment in the PURE system which involved only wild type, post-transcriptionally modified tRNAs.⁴¹⁴ By combining liquid-phase DNA hybridization with fluorous affinity chromatography, individual wild type E. coli tRNAs were able to be selectively purified. Using this method, 23 out of 27 tested tRNAs could be isolated, including the five leucyl isoacceptors which carry multiple unique modifications in the ASL. By assigning each wild type Leu-tRNA^Leu (carrying isotopically labeled Leu) and its synthetic version (carrying nonisotopic Leu) simultaneously to each corresponding Leu codon, a 3- to 4-fold preference for decoding by the wild type tRNAs was observed in all cases. Purified wild type tRNAs or unmodified tRNAs were charged with two different npMs or Leu and subsequently incorporated at three consecutive Leu codons within a peptide. The desired reprogrammed peptide was obtained as the major product only when using wild type (and not synthetic) tRNAs.⁴¹⁴ In a recent follow-up study, McFeely and co-workers found that, for both wild type and especially synthetic tRNAs, the orthogonality of codon-tRNA interactions could be increased by using hyperaccurate ribosome mutants.⁴¹⁵ With these ribosomes, the Leu codon box was further divided to simultaneously encode four npMs and one pAA into one peptide.

For in vivo sense codon reassignment, there is a substantial, additional challenge, which is to avoid toxic misreading of the host organism proteome. For example, if the Ser UCA codon was reassigned to an npM, every occurrence of the UCA codon would be decoded by the npM-tRNA, including within proteins essential for survival. To allow for reassignment only within the protein of interest, all other instances of the UCA codon would have to be reassigned a different synonymous codon which encodes Ser, such as AGC. Using this concept, Fredens and co-workers in 2019 created a synthetic genome to generate an E. coli strain, named Syn61, where all instances of the Ser UCA and UCG codons and the UAG stop codon were reassigned to synonymous codons AGU, AGC, and UAA, respectively.¹¹⁹ Subsequently, Robertson and co-workers were able to simultaneously remove the genes for tRNA^Ser_UGA, tRNA^Ser_CGA, and RF1 from Syn61 and introduce triply orthogonal engineered PylRS/tRNA^Pyl pairs to encode distinct npMs at the now blank UCA, UCG, and UAG codons.³⁸⁸

4.3.2. Quadruplet Codons

Other than nonsense suppression, another approach to obtain blank codons for genetic code expansion involves radically altering the translation system to read codons consisting of four nucleotide bases (Figure 18c). In theory, a total of 256 so-called “quadruplet” codons could be used for encoding npMs.⁶⁸ However, in the presence of native pAA-tRNAs, decoding of canonical triplet codons competes strongly with quadruplet codon decoding, so careful engineering is required for efficient genetic code manipulation with expanded codons (Table 1)

Table 1. Summary of the Progress Made Using Quadruplet Codons for the Incorporation of npMs.

Year	Achievement	Quadruplet codon(s)	Example monomer(s)	Reference
1993	First use of a quadruplet codon/anticodon pair for in vitro translation	AGGU or UAGG	Ala	Ma et al. Biochemistry 1993432
1996	First incorporation of npMs using quadruplet-decoding tRNAs in vitro	AGGN	pNO2Phe, L-2-naphthylalanine, L-2-anthrylalanine, or L-p-(phenylazo)phenylalanine	Hohsaka et al. J. Am. Chem. Soc. 1996435
1999	First use of mutually orthogonal quadruplet codon/anticodon pairs to encode two distinct npMs in vitro	AGGU	pNO2Phe	Hohsaka et al. J. Am. Chem. Soc. 1999436
		CGGG	l-2-prenylalanine
2001	Evolution of efficient quadruplet codon- decoding tRNAs in vivo using randomized anticodon loop libraries in E. coli	AGGA, CCCU, CUAG, UAGA	Ser	Magliery et al. J. Mot. Biol. 2001439
2003	Integration of flexizyme and quadruplet codon tRNA for double incorporation in vitro	GGGU	plPhe	Murakami et al. Chem. Biol. 2003202
		UAG	l-p-biphenylalanine
2005	Use of triply orthogonal quadruplet codon/anticodon pairs to encode three distinct npMs in vitro	CGGG	l-2-acridonylalanine	Ohtsuki et al. FEBS Lett. 2005438
		CUCU	pNO2Phe
		GGGU	l-2-anthrylalanine
2010	Evolution of an orthogonal ribosome to efficiently decode quadruplet codons in E. coli	AAGA	Ser (for evolution)	Neumann et al. Nature 2010147
		AGGA	pAzPhe (for npM incorporation)
2014	Improved suppression efficiency of UAGN codons by using a modified E. coli strain lacking UAG codons and RF1	UAGN	pAcPhe	Chatterjee et al. ChemBioChem 2014640
2021	Use of an orthogonal quadruplet-decoding ribosome and quadruply orthogonal ARS/tRNA and anticodon/codon pairs to encode four distinct npMs at four distinct quadruplet codons in E. coli	AGGA	Nπ-methyl-l-histidine,	Dunkelmann et al. Nat. Chem. 2021167
		AGUA	N6-((benzytoxy)carbonyl)-l-lysine
		UAGA	N6-((allyloxy)carbonyl)-l-lysine
		CUAG	plPhe
2021	Phage-assisted continuous evolution of efficient quadruplet-decoding tRNAs in E. coli to decode up to six consecutive UAGA codons	Continuous evolution on UAGA- decoding tRNAs	Various pAAs due to aminoacylation by endogenous ARSs	DeBenedictus et al. Nat. Commun. 2021442

In the late 1960s, it was discovered that various natural tRNAs possess anticodon loops expanded by a single base.^416−419 These frameshift suppressor tRNAs decode quadruplet codons that naturally arise at specific positions in specific genes as a result of nucleotide insertions in the coding sequence. Such mutations are known as frameshifts; in reading the quadruplet codon, the tRNA “suppresses” the frameshift and ensures correct translation of the original coding sequence. Some of the most well-studied natural suppressors are tRNA^SufA6 and tRNA^SufB2, derivatives of two Salmonella typhimurium isoacceptor tRNA^Pros, which both contain a G37a insertion to the 3′-side of the anticodon.^420−424Using these model tRNAs, crystallography and kinetic assays provided strong evidence that there is no actual quadruplet pairing or +1 frameshifting in the A site, just canonical three base codon–anticodon duplexes. Instead, +1 frameshifting occurs during translocation from the A site to the P site and/or within the P site, with potential quadruplet codon interactions occurring during this time frame.^425−428 One proposed reason is that the A site forms multiple obligate rRNA–anticodon contacts, while the P site only has one direct rRNA–anticodon contact, allowing for tRNA structural rearrangements (Figure 20).⁴²² While many engineered quadruplet codon tRNAs that will be discussed instead contain an inserted N33.5 nucleotide to the 5′-side of the anticodon, it is proposed that a similar quadruplet codon decoding mechanism is present.^428−430

Figure 20.

Interactions between the frameshift-prone tRNA^SufA6 and the CCC anticodon within the Thermus thermophilus 70S ribosome. (a) Structure and schematic of tRNA^SufA6 in the A site. Various rRNA interactions (A1492/93, A1913, and C1054, gray) stabilize the tRNA (blue) and mRNA (green) duplex. Even though the anticodon loop is expanded to eight nucleotides by the extra G37.5, there is only the standard triplet codon decoding in the zero frame (PDB: 4L47). (b) Structure and schematic of tRNA^SufA6 in the P site. Only the nucleotide C1400 directly monitors the codon–anticodon duplex, allowing for engagement of the mRNA in the +1 frame. Within these snapshots, there is no direct evidence of a quadruplet codon–anticodon interaction (PDB: 5VPO).⁴²²

After the initial discovery of suppressor tRNAs, Curran and Yarus showed in 1987 that tRNAs engineered to have quadruplet anticodons could be used to read quadruplet codons artificially introduced into genes in vivo.⁴³¹ Subsequently, Ma and co-workers demonstrated quadruplet codon decoding using fully synthetic tRNAs possessing cognate quadruplet anticodons in vitro.⁴³² The Val GUG codon at position 75 of DHFR was replaced with either AGGU or UAGG, and active full-length protein was produced in an E. coli S30 cell-free system in the presence of a chemically acylated pAA-tRNA possessing the WC complementary quadruplet anticodon. These quadruplet codons were based on AGG, a rarely used triplet codon for which the decoding isoacceptor tRNAs are in low abundance,^413,433 and UAG, the most commonly suppressed stop codon, respectively.⁴³⁴

In a series of studies, Hohsaka and co-workers examined the utility of tRNAs bearing different quadruplet anticodons for CFPS.^435−437 These anticodons were all derived from rare triplet codons, thus minimizing competitive decoding by canonical triplet-reading tRNAs. Although the decoding efficiencies of different quadruplet codons varied significantly, the best, GGGU, directed the incorporation of pNO₂Phe into streptavidin with a yield of 86% compared to wild type production levels.⁴³⁷

The practical utility of quadruplet decoding was further demonstrated by Ohtsuki and co-workers with the discovery of up to triply orthogonal quadruplet decoding tRNAs.⁴³⁸ This enabled the incorporation of three distinct Phe-derived npMs into a single protein in vitro at the codons CGGG, CUCU, and GGGU. Interestingly, the efficiency of quadruplet decoding for different codons was affected by the tRNA scaffold into which the cognate quadruplet anticodons were inserted, with each scaffold having different anticodon preferences.

In vivo systems offer the opportunity to explicitly search for tRNAs that efficiently decode quadruplet codons, using positive selection assays. Starting from E. coli tRNA^Ser_CGA, Magliery and co-workers prepared a library of tRNAs containing randomized 8- or 9-base anticodon loops and transformed them into E. coli cells harboring an antibiotic resistance gene containing a randomized quadruplet codon (NNNN) within the coding sequence.⁴³⁹ By sequencing colonies capable of surviving in the presence of antibiotic, tRNAs capable of efficiently decoding the quadruplet codons AGGA, CCCU, CUAG, and UAGA, which are all based on rare codons, were discovered. Curiously, the anticodons of the best quadruplet decoding tRNAs had full WC complementarity to their corresponding codon, including at the fourth base. However, this and other studies still show much greater noncognate recognition at the fourth anticodon base compared with the first three, which may limit the number of free codons obtainable from quadruplet decoding.^440−442

Further efforts have highlighted the effectiveness of in vivo tRNA selections for identifying efficient quadruplet suppressors.^{118,151,441−443} Recently, DeBenedictis and co-workers reported a pair of studies on the propensity of different E. coli tRNA scaffolds to tolerate quadruplet anticodons.^442,443 First, a selection was performed for read-through translation of a randomized quadruplet codon at defined positions of an antibiotic resistance gene by E. coli tRNAs containing a concomitantly randomized anticodon loop. Decoding efficiencies of the obtained hits varied substantially based on the codon and parent tRNA scaffold, ranging from near 0% translation efficiency for nine tRNAs to 6% translation efficiency with E. coli tRNA^His_AGGG, compared to the wild type triplet codon. To specifically explore the effect of the tRNA body on a single codon, UAGA-decoding tRNAs were generated using scaffolds drawn from 10 of the 20 isoacceptor classes and the best five were evolved using phage-assisted directed evolution.⁴⁴⁴ Overall, beneficial mutations found in evolved tRNAs were located in multiple regions of the tRNA, showing the utility of the unbiased approach. The evolved tRNAs enabled translation of mRNAs containing up to six consecutive UAGA codons. In the second study, the phage-based selection approach was expanded to identify suppressors of other quadruplet codons and, thus, create a set of ten mutually orthogonal quadruplet codon/tRNA pairs.⁴⁴³ Although none of these codons differed exclusively at the fourth anticodon base, this nevertheless suggests that a large number of quadruplet codons can be recruited for genetic code expansion.

The efficiency of quadruplet decoding is also limited by the ribosome, which has evolved to read triplet codons. To address this issue, Neumann and co-workers evolved the ribosome for improved reading of quadruplet codons.¹⁴⁷ Because mutating the natural ribosome would be highly toxic, a previously generated orthogonal ribosome was used as the starting point for saturation mutagenesis of nucleotides proximal to the ribosomal decoding center. Mutants were selected based on readthrough of an antibiotic resistance gene containing an in-frame AAGA quadruplet codon in the presence of known AAGA-decoding tRNA^Ser_UCUU.⁴³⁹ The resulting mutant ribo-Q1 was able to enhance quadruplet decoding for a variety of codon–anticodon pairs while maintaining wild type levels of translational fidelity for triplet decoding.¹⁴⁷ This evolved ribosome was then used in combination with an orthogonal ARS/tRNA pair to mediate the incorporation of pAzPhe in response to AGGA codons.

Subsequent improvements to orthogonal translation systems have further expanded the utility of the orthogonal quadruplet-decoding ribosome. For instance, orthogonal PylRS/tRNA pairs are especially valuable for quadruplet decoding, as the anticodon is not a recognition element for PylRS and may therefore be mutated to direct npM incorporation to various quadruplet codons. Recently, mutually orthogonal PylRS/tRNA pairs and computationally optimized orthogonal mRNAs were used with the quadruplet decoding ribosome to incorporate four distinct npMs into a protein at four distinct quadruplet codons.^167,445 Further tRNA and ribosome engineering may also enhance the fourth-base specificity of quadruplet decoding or even improve decoding of five-base codons, which has previously been shown to occur with lower efficiency to quadruplet decoding.^446,447

4.3.3. Unnatural Base Pair Systems

While sense codon reassignment and quadruplet codons feature tRNAs and mRNAs composed of natural nucleotides, a separate strategy has emerged using unnatural nucleotides that form unique base pairs. Introduction of an unnatural base (UB) pair that does not interact with A•T(U) or C•G could expand the genetic alphabet and theoretically create hundreds of new codons available for incorporation of npMs (Figure 18d).⁴⁴⁸ The major challenge faced by this field is that replication of DNA, transcription of RNA, and translation of proteins are all built upon natural nucleotides; thus, each of these processes must be made compatible with UBs.⁴⁴⁹ While each of these processes come with their own inherent challenges and have been discussed in detail in other reviews (see refs (448−451)), this section focuses on UBs which have been used in translation.

The first successful strategy for translation from UB-containing codons featured derivatives of the natural bases (Figure 21a). These UBs utilized different hydrogen bond geometry while still retaining the overall WC geometry. In a series of experiments in the late 1980s and early 1990s by the Benner group, they demonstrated that isoguanosine (isoG) and its pair isocytidine (isoC) were able to be replicated and transcribed.^452,453 Subsequently, it was reported that an in vitro translation system using rabbit reticulocyte lysate was able to site-specifically incorporate oITyr in response to the (isoC)AG codon using the CU(isoG) anticodon introduced into E. coli tRNA^Gly with 90% efficiency.⁴⁵⁴ Generally, such hydrogen-bonding UBs are not extensively used for translation, but more recent versions of hydrogen-bonding UBs, called Hachimoji DNA and RNA, have emerged as a potential new option.^455,456

Figure 21.

Summary of work within the field of unnatural base pairs. (a) Original unnatural base pair pioneered by the Benner group. (b) Second generation base pair used by the Hirao group. (c) Optimized base pairs used by the Romesberg group. (d) Semisynthetic organism (SSO) developed by the Romesberg group that is capable of replication, transcription, and translation using UB pairs NaM and TPT3. dNaMTP = deoxyNaM triphosphate, dTPT3TP = deoxyTPT3 triphosphate, NaMTP = NaM triphosphate, TPT3TP = TPT3 triphosphate, AK = AzLys, AF = pAzPhe, Ser = Ser. Crystal structure of sfGFP is from the PDB file 2B3P.⁴⁷³

In the early 2000s, the Hirao group created the x-y pair which interacted through two hydrogen bonds; the bulky dimethylamino group of x prevented noncognate pairing with the natural bases.⁴⁵⁷ Subsequent optimizations led to the dimethylamino group of x being replaced with a heterocyclic thienyl group to create the s base (Figure 21b).^458−460 The s-y pair was then applied to in vitro translation using E. coli. S30 extract to site-specifically incorporate 3-chloro-l-tyrosine at the yAG codon using S. cerevisiae tRNA^Tyr containing the CUs anticodon. However, it was noted that the yAG codon could be mis-decoded by E. coli Lys-tRNA^Lys_UUU and Gln-tRNA^Gln_CUG, leaving room for improvement.⁴⁶⁰ Later work by the Hirao group focused on hydrophobic base pairs that are often used for DNA aptamers.⁴⁶¹

In 1999, the Romesberg group developed a self-complementary UB with an isocarbostyril scaffold that drove base-pairing interactions through hydrophobic and van der Waals interactions.⁴⁶² After 15 years of structure–activity relationship studies, two optimized combinations of UBs were created: 5SICS-NaM in 2009 and TPT3-NaM in 2014 (Figure 21c).^{451,463−469} In 2014, Malyshev and co-workers reported the first semisynthetic organism (SSO) that could propagate stably using an expanded genetic code with the 5SICS-NaM or TPT3-NaM pair.^450,470

Using the second generation E. coli SSO, YZ3, Zhang and co-workers demonstrated in 2017 that all three critical steps of DNA replication, transcription, and translation were possible with UBs.^471,472 In this work, Ser was incorporated at position 151 into sfGFP using E. coli tRNA^Ser_GYU and codon AXC (where Y is TPT3 and X is NaM). SSOs transformed with the plasmid encoding both sfGFP(AXC) and tRNA^Ser_GYU exhibited fluorescence that was nearly equal to that of control cells expressing wild type sfGFP(AGU), though a minimal amount of mis-decoding by natural tRNAs was observed. Additionally, AlkLys was site-specifically incorporated into position 151 of sfGFP using Mm tRNA^Pyl_GYU for the AXC codon or tRNA^Pyl_GYC for the GXC codon. Although initial tests showed a potentially high background misincorporation due to mis-decoding of NaM by natural tRNAs, these events were sufficiently suppressed using optimized conditions. This resulted in a 96.2% final purity of AlkLys-sfGFP for the AXC codon and 97.5% for the GXC codon, indicating highly successful decoding of these two codon–anticodon pairs. Finally, the incorporation of pAzPhe using the AXC-GYU codon–anticodon pair was achieved by using an evolved Mj pAzPheRS/tRNA^pAzPhe pair.

A third generation E. coli SSO, ML2, was utilized in 2020 to explore a variety of codon–anticodon combinations (Figure 21d).^473,474 The decoding efficiency of different combinations was evaluated by incorporation of N^ε-(2-azidoethoxy)-carbonyl-l-lysine (AzLys) into position 151 of sfGFP using the engineered Mm chPylRS^IPYE/ tRNA^Pyl pair. Using this methodology, nine codon–anticodon pairs were identified that could successfully produce active AzLys-sfGFP. Specifically, the codon–anticodon pairs AXC-GYU, GXU-AYC, and AGX-XCU were mutually orthogonal. As a proof-of-concept, sfGFP containing the codons AXC, GXU, and AGX was decoded by tRNA^pAzPhe_GYU to incorporate pAzPhe at position 151, tRNA^Ser_AYC to incorporate Ser at position 190, and tRNA^Pyl_XCU to incorporate AzLys at position 200, respectively.⁴⁷³ Recently, Zhou and co-workers have reported initial experiments of utilizing UBs in eukaryotes.⁴⁷⁵

5. Applications of Polypeptides Containing Ribosomally Translated Non-proteinogenic Monomers Using Engineered tRNAs

5.1. In Vivo Expression of Proteins and Peptides Containing Non-proteinogenic Monomers

Engineered tRNAs have been used for the in vivo incorporation of over 150 npMs into proteins for a wide variety of functions, such as for site-selective labeling, spectroscopic probes, post-translational modifications, and cross-linking (reviewed in refs (13, 68, 476, and 477)). A newer avenue for engineered, npM-containing proteins is for therapeutic purposes, with several currently in clinical trials.⁴⁴⁸ A common strategy is to genetically encode an npM with a bioorthogonal handle at an amber stop codon and then conjugate on a polyethylene glycol (PEG)-based moiety.⁴⁷⁸ For example, the biopharmaceutical company Ambrx utilized this strategy to express an anti-HER2 antibody with pAcPhe at position 121 and conjugate on a nonhydrolyzable hydroxylamine-based PEG linker with a potent synthetic drug.⁴⁷⁹ This antibody–drug conjugate (ARX788) has successfully completed its phase I trial for treatment of HER2-positive metastatic breast cancer and a global, phase 2 study is underway (clinical trial ID NCT04829604).⁴⁸⁰

Along with biologics, peptides are excellent pharmaceutical candidates, as they can selectively bind to relatively large and flat surfaces involved in clinically important protein–protein interactions (PPIs).⁴⁸¹In vivo peptide display systems such as SICLOPPS, phage display, and yeast display aim to discover novel bioactive peptides through creation and screening of diverse, combinatorial libraries. While most selection schemes use only the 20 pAAs, some have started to utilize amber codon suppression to incorporate npMs during affinity maturation for improved pharmacological properties.^17,482−487 Using phage display, Young and co-workers used the Mj p-benzoyl-l-phenylalanine ARS/tRNA^opt_CUA pair to discover backbone macrocyclic peptide inhibitors against HIV protease with the best candidate having an in vitro IC₅₀ of 0.85 μM.⁴⁸³ Similarly, Wang and co-workers used an Mm N^ε-acryloyl-l-lysine ARS/tRNA^Pyl-opt_CUA pair to discover thioether macrocyclic peptides against TEV protease and histone deacetylase 8 with all peptides having K_D values in the low micromolar range.⁴⁸⁴ One of the more recent and promising platforms is the macrocyclic organo-peptide hybrid phage display system (MOrPH-PHD) developed by the Fasan group in 2020.^485,488 Using phage display, O-(2-bromoethyl)-l-tyrosine was incorporated at UAG codons using an engineered Mj TyrRS/tRNA^opt_CUA, followed by a downstream Cys to form thioether-bridged macrocyclic peptides. Subsequent affinity selection against streptavidin produced two macrocyclic peptides with K_D values of 20 nM and 150 nM, and affinity selection against the keap1 kelch domain produced binders with K_D values of 40 nM and 43 nM.⁴⁸⁵ Iannuzzelli and Fasan have also explored different electrophilic npM cyclization handles which could be applied in future selections.⁴⁸⁷

Overall, in vivo technologies excel at the expression of large quantities of npM-modified proteins which can be used for a plethora of applications. However, the range of npM analogs is mostly restricted to natural l-α-AA derivatives and only two or three unique npMs per protein.^13,14 While some ARSs can aminoacylate backbone-modifying npMs, until 2024, the only backbone-modifying npMs that had been successfully translated by wild type ribosomes in vivo were α-hydroxy acids.^{173,489−493} When using engineered ribosomes in vivo, an oxazole-based dipeptide mimetic and (S)-β³-(p-bromo)-l-phenylalanine were ribosomally elongated with yields sufficient for purification of recombinant proteins.^222,223 Recently, a report by Dunkelmann and co-workers in 2024 demonstrated in vivo site-selective incorporation of three β³-AAs [β³-pBrAla, (S)-β³-(m-bromophenyl)alanine , (S)-β³-(m-trifluoromethyl)-l-phenylalanine] and one α,α-disubstituted amino acid [(S)-α-methyl-4-iodo-l-phenylalanine] into GFP(150UAG) using E. coli.⁴⁹⁴

Compared to in vitro methodologies, there remains much room for improvement for in vivo peptide display platforms utilizing npMs, as many hit peptides have weak, micromolar activities and backbone-modifying npMs typically cannot be included.⁴⁹⁵ As all current platforms only use a single npM during affinity maturation, the most likely area that could be significantly improved is the use of mutually orthogonal ARS/tRNA pairs to incorporate multiple npMs into peptide libraries. The Mj TyrRS/tRNA^Tyr and CMa PylRS/tRNA^Pyl pairs would be a good avenue for exploration, as the recently discovered CMa PylRS/tRNA^Pyl pair has been shown to be more active than the standard Mm and Mb PylRS/tRNA^Pyl pairs for some npMs.^169,171,172

Regardless, as in vitro methodologies allow for much larger initial libraries and the incorporation of backbone-modifying npMs and exMs, in vitro display platforms are necessary to fully realize the applications of tRNA engineering and genetic code manipulation for peptide drug discovery.

5.2. Discovery of Bioactive Peptides Containing Non-proteinogenic Monomers Using In Vitro Discovery Platforms

The most prevalent use of engineered tRNAs in vitro is for the discovery of bioactive peptides. Linear and macrocyclic peptide therapeutics derived from natural products often contain an extensive number of backbone-modifying npMs which can drastically improve peptide pharmacokinetics.^{17,18,495−497}d-AAs are cleaved minimally or not at all by proteases, allowing for a significant increase in peptide stability.^498−501N-Alkylation of exposed secondary amine groups removes hydrogen bond donors, enhancing oral bioavailability and membrane permeability.^502−504 Especially in macrocyclic peptides, conformationally constrained npMs such as cyclic β-AAs or cyclic γ-AAs can impart structural rigidity to allow for increased binding affinity.²⁵ Heterocyclic, amide bond isosteres such as oxazoles, oxazolines, thiazoles, and thiazolines generally stabilize secondary structures and are less recognized by proteases, serving multiple purposes for increased peptide stability and membrane permeability.^25,505−507 One of the most notable natural products is the backbone macrocyclic peptide cyclosporin A, which contains two l-AAs with non-proteinogenic side chains, seven ^MeAAs, and one d-AA.⁵⁰⁸ These npMs contribute to the superior oral bioavailability and membrane permeability of cyclosporin A compared to other macrocyclic peptides.⁴⁹⁶

First reported independently by the groups of Yanagawa and Szostak in the late 1990s, mRNA display quickly became one of the most common methods for discovery of bioactive peptides.^509,510 In the mid-2000s, this technology was further improved to allow for the incorporation of npMs using sense codon reassignment due to the advent of the PURE system.¹⁸⁷ A model example of a macrocyclic peptide derived from mRNA display which emphasizes the utility of backbone-modifying npMs is a proprotein convertase subtilisin-like/kexin type 9 (PCSK9) inhibitor discovered by Merck & Co.^511,512 The initial hit peptide had a K_i of 956 nM and contained three fluorinated l-Trp derivatives which were essential for binding to PCSK9 through halogen interactions. Removing the N-terminal tail region increased potency to a K_i of 110 nM, but this peptide was quickly degraded by serum proteases. Mutation of a Pro residue at a protease cleavage site to the backbone-modifying α-methyl Pro not only increased stability by almost 5-fold but simultaneously increased potency to a K_i of 14 nM due to rigidification by the α,α-disubstituted amino acid. Subsequent mutation of a Gly residue to d-Ala removed another vulnerable protease cleavage site, and methylation of the Tyr hydroxyl removed a hydrogen bond donor, potentially increasing cell permeability. After various other optimizations such as cyclization and PEGylation, the final peptide MK-0616 (patent WO2023023245A1, formula I) became orally bioavailable in primates using an enabled formulation-based approach, making it one of the first clinical candidates to result from an mRNA display selection.^513,514 In May 2023, MK-0616 successfully completed its phase 2b clinical trial for treatment of hypercholesterolemia, and a phase 3 trial began in late 2023 (clinical trial ID NCT06008756).⁵¹⁵ This case study demonstrates the importance of backbone-modifying npMs for improving peptide pharmacological properties and further suggests that directly integrating backbone-modifying npMs into mRNA display screening platforms may potentially decrease the time needed for lead optimization.^516−518

To this end, the major use of the FIT system is through integration with mRNA display in the Random nonstandard Peptides Integrated Discovery (RaPID) system (Figure 22).¹⁷ The RaPID system can be used to screen large and diverse macrocyclic peptide libraries against proteins of interest to discover bioactive macrocyclic peptides. First, a randomized DNA library is transcribed to create an mRNA library. Generally, this mRNA library contains an initiation AUG codon, followed by a randomized region of six to 15 NNK codons (where K is G or U), a Cys codon, codons encoding a pAA linker, and a UAG stop codon. Occasionally, l-Cys is reprogrammed to d-Cys. When combining each NNK library of different lengths, an initial diversity of over 10¹² unique sequences is possible. After covalent linkage of puromycin (Pu) to the 3′ end of the mRNA, in vitro translation using the FIT system without RF1 is performed. During translation, the ribosome will stall at the UAG stop codon as RF1 is removed, allowing Pu to enter the ribosome A site and become covalently linked to the peptide at the C-terminus. Thus, each individual peptide can later be directly identified by its cognate mRNA. Next, reverse transcription of the mRNA into cDNA, followed by screening the peptide-Pu-mRNA-cDNA complex against an immobilized target protein, separates the binding peptides from the nonbinding peptides. Finally, cDNA from binding peptides can be recovered and amplified by PCR, which can then be subsequently transcribed for a new round of iterative selection. After multiple rounds of RaPID, peptides with high affinity for the target protein are enriched, and the peptides can be identified by sequencing their corresponding cDNA by next-generation sequencing.

Figure 22.

General scheme of RaPID selection. (1) Transcription of a large diversity DNA library, followed by (2) puromycin ligation. (3) Translation using the FIT system results in a randomized library of peptides containing npMs. Often, translation is initiated with a chloroacetyl npM (^ClAcnpM), resulting in (4) spontaneous thioether macrocyclic peptide (teMP) formation. (5) The mRNA sequence is reverse transcribed, and (6) the teMP library is then screened against a protein of interest immobilized on magnetic beads. (7) Binding peptides are recovered, and the cDNA is amplified by PCR for a new round of RaPID. (8) Alternatively, if sufficient enrichment has occurred, the hit peptides are identified by next generation sequencing.¹⁸

Compared to a standard mRNA display, the FIT system allows inclusion of engineered npM-tRNAs, leading to high diversity, npM-containing peptide libraries. Notably, initiation is often reprogrammed to an N-chloroacetylated npM or exM (^ClAcnpM or ^ClAcexM) charged onto an in vitro transcribed tRNA^fMet2_CAU. After elongation, the ^ClAcnpM undergoes a spontaneous ring-closing reaction with a downstream Cys thiol to form a thioether linkage. These macrocyclic peptides are referred to as thioether-closed macrocyclic peptides (teMPs) and are the most common method for discovering bioactive peptides via the RaPID system.¹⁸

To date, over 150 unique npMs have been translated using engineered tRNAs in the FIT system or RaPID selection. In the following sections, the applications of the FIT and RaPID systems will be discussed with specific emphasis on the impact of backbone-modifying npMs.

5.2.1. Initiation with Non-proteinogenic Monomers

Although polypeptide translation typically begins with ^fMet in nature, the initiation event tolerates a wide variety of npMs, even in the absence of tRNA^fMet_CAU engineering.^67,519,520 Examples translated in the FIT and RaPID systems include l-npMs,^521−523d-AAs,^524,525N^α-acyl-AAs,^204,526N^α-alkyl-AAs,^524,527 γ⁴-AAs,⁵²⁸ δ⁵-AAs,⁵²⁸ ε⁶-AAs,⁵²⁸ ζ⁷-AAs,⁵²⁸ farnesyl-AAs,⁵²⁹ malonyl-AAs,⁵³⁰ biotinylated AAs,^{526,531−533} (amino)benzoic acids,^{22,521,527,530,534−536} heteroaromatics,⁵²¹ oligopeptides,^537,538 foldamers,^539−541 bipyridines,⁵⁴² and carborane.⁵⁴³ As mentioned in section 3, the engineered tRNA^iniP has been used to initiate translation with relatively more unreactive substrates such as a β³-AA and a cyclic γ^2,4-AA.³¹³ Although the effects of these more constrained npMs and exMs have not yet been tested in RaPID, it is hypothesized that they will contribute to higher proteolytic stability and improved membrane permeability of discovered peptides.

5.2.2. Elongation with Non-proteinogenic Amino Acids and Non-Amino Acid Exotic Monomers

5.2.2.1. l-Amino Acids

Similar to in vivo genetic code expansion, the FIT system is able to easily translate side chain-functionalized l-AA derivatives. Generally, tRNA^AsnE2 is used, as EF-Tu affinity is not an issue, though for some bulkier npMs the optimized T stem of tRNA^GluE2 may be necessary (refs (198, 202, 268, 367, 407, 409, 424, 511, 523, and 544−578)). In a similar manner to RaPID, the PURE system has also been combined with Mj TyrRS/tRNA^opt_CUA pairs for mRNA display selections by Iskandar and co-workers in 2023.^178,177 As the cotranslational incorporation of l-AA derivatives is more efficient than backbone-modifying npMs and has been extensively reviewed elsewhere, it will not be discussed in detail.^17,18,579 However, these npMs are nevertheless important and are often used in RaPID selections for their beneficial side chains. For instance, chemical warheads can be incorporated to generate privileged peptide binders.^{550,552,553,562} Morimoto and co-workers encoded a single incorporation of N^ε-trifluoroacetyl-l-lysine (TfaLys) at the AUG elongator codon decoded by tRNA^AsnE2_CAU to discover isoform selective teMPs against human deacetylase mammalian sirtuin 2 (SIRT2), which is a target for Alzheimer’s disease and Parkinson’s disease.^550,580 The top teMP had an IC₅₀ of 3.2 nM and was 10-fold and 100-fold more selective for SIRT2 than for SIRT1 and SIRT3, respectively.⁵⁵⁰

A notable example of more exotic l-AA-based npMs is the translation of aromatic foldamers based on quinoline (Q) and pyridine (P) monomers which fold into stable helices (Figure 3).^539,581 These foldamers were previously shown to initiate translation, and they constitute the largest npM translated to date. Based on this precedent, Tsiamantas and co-workers elongated various aromatic foldamers appended to the side chain of 4-(aminomethyl)-l-α-phenylalanine (Amf) using tRNA^Pro1E2_GGU at the ACC codon.⁵⁸² Foldamers containing one positively charged [Amf(-Gly-Q^Dap-Ac)] or one negatively charged [Amf(-Gly-Q^Asp-Ac)] monomer were able to be translated. Finally, simultaneous incorporation of Amf(-Gly-Q^Dap-Ac) at the CCU codon using tRNA^Pro1E2_GGU and of Amf(-Gly-Q^Asp-Ac) at the AUG elongator codon using tRNA^Pro1E2_CAU and initiation by ^ClAcQ^Dap-Gly-Phe using tRNA^fMet_CAU resulted in a teMP containing three aromatic, helical foldamers. The ability for the ribosome to translate such bulky substituents is facilitated by the use of engineered tRNAs; future efforts may aim to determine whether even larger monomers can be ribosomally elongated using tRNA^Pro1E2.

5.2.2.2. C^α-Modified Amino Acids: d-Amino Acids and α,α-Disubstituted Amino Acids

While tRNA^AsnE2 can efficiently translate some d-AAs, tRNA^GluE2 and tRNA^Pro1E2 are more often used for incorporation of d-AAs during RaPID selections (Figure 3) (refs (23, 24, 295, 298, 534, 551, 555, 558, 564, 574, 575, and 583−587)). Imanishi and co-workers used tRNA^Pro1E2 to create a teMP library that contained d-Ser, d-His, d-Tyr, and d-Ala at the AUU, ACU, GCU, and CAU codons, respectively.⁵⁸⁶ Additionally, ^ClAcd-Trp was used as the initiator npM. This library was then used for RaPID selection against human epidermal growth factor receptor (hEGFR), a well-known biomarker protein of various cancers. Overall, 11 out of the 18 most enriched sequences included consecutive and/or alternating d-amino acids, indicating that the d-AAs were efficiently translated in the RaPID system using tRNA^Pro1E2. The most promising d/l-hybrid teMPs were synthesized, with peptides 2D (5 out of 16 residues were d-AAs) and 18D (6 out of 16 residues were d-AAs) having a K_d of 523 nM and 998 nM, respectively (Figure 23a). Importantly, the d-AAs imparted significant proteolytic stability. Neither peptide showed degradation by proteases after 24 h of incubation in human serum, but when all d-AAs were mutated to their corresponding l-AAs (peptides 2L and 18L), the serum half-life decreased to 5.0 and 15.5 h, respectively.

Figure 23.

Selected peptides discovered by the RaPID system containing npMs, along with their pharmacological properties.

α,α-Disubstituted amino acids are unique npMs that can induce both 3₁₀- or α-helical structures in linear and macrocyclic peptides, allowing peptides to bind efficiently to helical PPIs.^588−591 As well, α,α-disubstituted amino acids can increase proteolytic stability.^{511,591−593} However, they are relatively unexplored, with only Aib,^19,551,558 1-aminocyclopropane-1-carboxylic acid (Ac₃c),⁵⁶⁴ and 1-aminocyclopentane-1-carboxylic acid (Ac₅c)^534,575 being used in the FIT system or RaPID selections (Figure 3). Thus, it would be interesting to investigate the incorporation of other α,α-disubstituted amino acids into the RaPID system.

5.2.2.3. N-Alkyl Amino Acids

Generally, tRNA^AsnE2 and tRNA^GluE2 are used to incorporate a variety of N-methyl (refs (250, 265, 295, 298, 367, 403, 407−409, 511, 534, 551, 555−559, 563, 564, 575, 584, 585, and 594−600)), linear N-alkyl,^563,564,601 branched N-alkyl,^563,601 functionalized alkyl chain (azido, alkyne, cyano, etc.),^595,601 and cyclic N-alkyl amino acids (Figure 3).^560,564,602 Given their overall ease of translation, many different RaPID selections have been conducted using N-alkyl npMs.

The first demonstration of RaPID selection was by Yamagishi and co-workers in 2011 targeting the ubiquitin E3 ligase E6AP HECT domain.⁵⁹⁴ Aberrant E6AP function is associated with a variety of cancers and neurological disorders, especially papillomavirus-induced cancers.⁶⁰³ In this selection, teMPs were initiated with ^ClAcd-Trp, and tRNA^AsnE2 was used for incorporation of ^MePhe, ^MeSer, ^MeGly, and ^MeAla at the UUU, CUU, AUU, and GCU codons, respectively.⁵⁹⁴ Three teMPs were synthesized, CM₁₁-1, CM₁₁-3, and CM₁₁-5, which contained an average of 25% N-methylated residues (Figure 23b). While all CM₁₁ peptides had sub-nanomolar to 1 nM K_d values, removal of N-methylation from CM₁₁-1 completely abolished its binding ability, demonstrating the importance of ^MeAAs for peptide potency. In in vitro ubiquitination assays, CM₁₁-1 at 1 μM almost completely inhibited charging of ubiquitin onto the HECT domain of E6AP. This work provided the proof-of-concept for RaPID selections and served as the basis for other ^MeAA selections.

Using the strategy developed by Iwane and co-workers where each ^MeAA was matched to a different T stem version of tRNA^AsnE2 based on EF-Tu affinity, van Neer and co-workers constructed a library with ^MeTyr^OMe, ^MePhe, ^MeSer, ^MeNle, ^MeTyr, and ^MeGly, and at the UUC, UUG, CUC, CUG, AUC, and AUG codons, respectively.^250,598MeSer and ^MeGly used the T stem from E. coli tRNA^Asn, ^MePhe and ^MeTyr required the stronger T stem from E. coli tRNA^Glu, and ^MeNle and ^MeTyr^OMe needed the strongest T stem, which was rationally designed.⁵⁹⁸ Subsequently, RaPID selection was conducted against three prokaryotic orthologs of the cofactor-independent phosphoglycerate mutase (iPGM) enzymes derived from Staphylococcus aureus, Helicobacter pylori, and Mycoplasma orale for the treatment of bacterial infections. Out of the 10 discovered teMPs, six contained more than four ^MeAAs, and one peptide had over 50% methylation in the randomized region. The high percentage of ^MeAA incorporation shows that the T stem tuning strategy was successful when compared to using just tRNA^AsnE2. For example, Mo-D2 contained 5 out of 16 ^MeAAs residues (three ^MeTyr, one ^MeSer, and one ^MeTyr^OMe) and had an IC₅₀ of 10.2 nM against Mycoplasma orale iPGM and low nanomolar against various other iPGMs (Figure 23c). Finally, incorporation of ^MeAAs greatly increased the stability of various teMPs. For example, Mo-D2 was completely resistant to proteases after 24 h, but removal of the N-methylation decreased serum stability half-life to around 10 h.

The RaPID system is also amenable to npM deep mutational scanning to quickly and efficiently optimize lead peptides.^{551,554,558,563} Alteen and co-workers first conducted a RaPID selection against the N-terminal domain of O-linked-N-acetylglucosamine transferase.⁵⁶³ After obtaining peptide D3, which had an IC₅₀ of 190 nM, in an in vitro fluorescence-based assay, the sequence was subjected to saturation mutagenesis. To accomplish this with RaPID, the D3 peptide sequence was converted back to an mRNA library where each residue was mutated from its original codon to the AUG codon. Then, either one of the 18 pAAs (excluding Met and Cys), 13 l-AA-based npMs, seven ^MeAAs, or two N-alkyl AAs was incorporated at that codon using tRNA^GluE2_CAU and subjected to one round of RaPID selection. Finally, a relative enrichment score was determined for each mutation, which indicated its beneficial or detrimental effect relative to the starting D3 peptide sequence. Mutation of the Pro residue at position 9 to either N-butyl Gly or N-isopentyl Gly resulted in about a 6-fold increase in inhibition activity compared to the parent peptide. Further rational mutation to N-(2-cyclopentylethyl) Gly further increased potency 10-fold to a final IC₅₀ of 3.3 nM for peptides D3–15. Overall, this study demonstrates that the inclusion of linear and branched N-alkyl npMs into RaPID could be beneficial for future selections. Additionally, the saturation mutagenesis RaPID scheme can be used to identify mutant teMPs with increased binding affinity through ribosomal incorporation of both N-methyl and N-alkyl npMs.

5.2.2.4. β-Amino Acids

Due to the insufficient affinity for EF-Tu and slow peptidyl transfer of many β-AAs, tRNA^Pro1E2 is most often employed for translation of these backbone-modifying npMs. However, tRNA^AsnE2 and tRNA^GluE2 may be required in the few circumstances that EF-P decreases translation efficiency. Regardless, tRNA^Pro1E2 has been used for the incorporation of β³-AAs,^293,575,604 β²-AAs,^584,564,605 β^2,3-amino acids,⁶⁰⁵ and cyclic β^2,3-AAs (Figure 3).^{24,295,583,606}

Wakabayashi and co-workers constructed a RaPID library using tRNA^Pro1E2 with β³-Phg, β³-Ala, β³-Gly, and β³-Gln at the codons AUU, ACU, GCU, and CAU, respectively.⁶⁰⁴ This library was screened against hEGFR with discovered teMPs containing up to 35% β³-AAs per sequence. L2β contained three β³-Phg residues and bound to hEGFR with a K_D of 47.0 nM and IC₅₀ of 66.8 nM, while D3β had one of each of β³-Phg, β³-Ala, and β³-Gln and possessed a K_D of 34.1 nM and an IC₅₀ of 20.3 nM. Backbone shortening of any singular β³-AA in either peptide by mutation to an α-AA abolished binding, and for the β³-AA-containing teMPs that were tested for proteolytic stability, backbone shortening decreased serum half-life anywhere from 1.2-fold to almost 10-fold. Thus, these experiments demonstrated that the increased backbone length was essential for activity and stability.

Cyclic β^2,3-AAs are attractive building blocks for peptide-based therapeutics, as they act as particularly strong helix/turn inducers, imparting unique secondary structures and structural rigidity.^607−610 Additionally, cyclic β^2,3-AAs have been shown to significantly improve proteolytic stability.⁶¹¹ To discover bioactive teMPs containing cyclic β^2,3-AAs, Katoh and co-workers performed RaPID selections against human factor XIIa (hFXIIa) and interferon-gamma receptor 1 (IFNGR1).²⁴ hFXIIa inhibitors can act as anti-thrombosis drugs with a low bleeding risk, while IFNGR1 inhibitors could help treat autoimmune diseases.^612,613 To construct this library, tRNA^GluE2 was used for (1R,2R)-2-ACPC and (1S,2S)-2-ACPC, and tRNA^Pro1E2 was used for (1S,2S)-2-(ACHC) at the CAU, UGU, and AUU codons, respectively.²⁴ For the hFXIIa selection, four teMPs containing either one or two cyclic β^2,3-AAs had K_D values ranging from 0.58 nM to 15.5 nM and low K_i values ranging from 3.7 to 20.3 nM (Figure 23d). While the teMPs from the IFNGR1 selection were not evaluated for inhibitory activity, the strongest binding peptide I1-5, which had a K_D of 1.87 nM, was evaluated for serum stability, along with the hFXIIa peptides. These peptides were extremely proteolytically stable, with the longest half-life being 285 h. Alanine mutations of the cyclic β^2,3-AAs decreased affinity by over 20-fold and serum stability by up to 50-fold. A cocrystal structure between hFXIIa and one of the hFXIIa-binding peptides, F3, revealed that it adopted a compact β-hairpin structure. (1S,2S)-2-ACHC at position 8 induced a pseudo γ-turn followed by an inverse γ-turn, while(1S,2S)-2-ACHC at position 13 participated in a standard β-turn. The crystal structure showed how the overall macrocyclic peptide structure is affected by cyclic β^2,3-AAs and that structural stabilization imparted by the cyclic β^2,3-AAs was important for binding affinity.

Aromatic cyclic β^2,3-AAs, such as 2-Abz and its derivatives, also create unique folding patterns in macrocyclic peptides, though in this case due to the planarity and rigidity of the aromatic ring.^614−616 These backbone-modifying npMs have been observed in bioactive macrocyclic peptide natural products and impart proteolytic stability.^617−620 Using a combination of aliphatic and aromatic cyclic β^2,3-AAs, Katoh and Suga performed more RaPID selections against hFXIIa and IFNGR1.⁵⁸³ In this iteration, UGU, GGU, CAG, and GAG codons were reprogrammed to (1S,2S)-2-ACPC, (1R,2S)-2-ACPC, 2-Abz, and 3-aminothiophene-2-carboxylic acid (Atp) respectively. (1S,2S)-2-ACPC was incorporated using tRNA^GluE2, while the other npMs utilized tRNA^Pro1E2. Overall, all eight IFNGR1 peptides contained one 2-ACPC derivative and either zero or one Abz or Atp residues. Interestingly, peptide ArβI-4 even had a consecutive (1S,2S)-2-ACPC and 2-Abz motif, demonstrating the increased translation efficiency imparted by using engineered tRNAs (Figure 23e). Additionally, seven out of eight peptides showed strong binding affinity, with K_D values ranging from 0.44 nM to 130 nM, and reasonable inhibitory activity, with IC₅₀ values ranging from 9.7 nM to 400 nM. The importance of 2-Abz and Atp was confirmed through point mutations to either Ala or β³-Ala. Most mutant peptides showed more than 2-fold weaker binding affinity. However, the impact of 2-Abz and Atp was most noticeable for the serum stability, as even one of these npMs was able to increase the half-life by 2- to 5-fold in all cases. Overall, this study showed that incorporation of even one aromatic cyclic β^2,3-AA is sufficient to grant high potency and protease resistance.

5.2.2.5. γ-Amino Acids

Along with issues of low EF-Tu affinity and slow peptidyl transfer, incorporation of linear γ-AAs faces an additional challenge. Compared to most other npMs, linear γ-AAs undergo much more rapid self-deacylation by intramolecular nucleophilic attack of the γ-amino group to the esterified carboxyl group of the tRNA.⁵⁴⁶ Various strategies have been developed to overcome this issue and allow incorporation of γ-AAs into peptides (Figure 3).

Kuroda and co-workers decided to first translate linear α- or β-hydroxy acids containing an azide at the γ- or δ-position at the CAG codon using tRNA^GluE2_CUG, such as (3S,4S)-4-azide-3-hydroxy-6-methyl-heptanoic acid (azido-statine, γ⁴-N₃-Sta) (Figure 24).⁶²¹ Once introduced into the peptide, the azide can then be demasked to an amine via a Staudinger reduction and will spontaneously undergo an O-to-N acyl shift in alkaline conditions. Although the translated exMs were α- or β-hydroxy acids, the final peptide contained hydroxyhydrocarbon (Hhc)-based γ⁴- or δ⁵-AAs. Using this system, single incorporation of four different Hhc γ⁴-AAs and one Hhc δ-AA was achieved. Additionally, a synthetic β-secretase 1 inhibitor peptidomimetic (P10-P4′ statV) containing statine was translated in the FIT system.

Figure 24.

Incorporation of linear, hydroxyhydrocarbon (Hhc)-based γ⁴- or δ⁵-AAs into macrocyclic peptides in the FIT system. First, an azide-containing hydroxy acid was incorporated into a macrocyclic peptide. Treatment with TCEP caused reduction of the azide to a primary amine which then underwent spontaneous O-to-N acyl shift to rearrange the peptide backbone. This resulted in peptides containing Hhc-based γ⁴- or δ⁵-AAs.⁶²¹

As self-deacylation occurs more rapidly in neutral environments, Adaligil and co-workers performed Fx-mediated aminoacylation at pH 9.0 and were thereby able to aminoacylate both branched γ⁴-AAs and Hhc γ⁴-AAs.⁵⁸⁴ Subsequent translation using tRNA^Pro1E2_GGG at the CCU codon allowed incorporation of linear γ⁴-Gly, (S)-γ⁴-Nva, (S)-γ⁴-Ala, (R)-γ⁴-Leu, and four Hhc stereoisomers β³-OH-γ⁴-Ala (γ^4S-Ala^3R–OH). Additionally, double consecutive incorporation of (R)-β³-OH-(S)-γ⁴-Ala and (R)-β³-OH-(R)-γ⁴-Ala was possible, though some truncated peptide side products were observed.

The final strategy is to use conformationally constrained, cyclic γ^2,4-AAs that are too sterically hindered to undergo self-deacylation. Cyclic γ^2,4-AAs are interesting building blocks for peptide therapeutics, as they induce unique conformations such as C₁₂ turns and improve cell permeability and proteolytic stability.^25,622−624 As mentioned in section 3, Katoh and co-workers were able to incorporate cis- and trans-3-ACBC, four stereoisomers of 3-ACPC, two stereoisomers of 3-ACHC, and 3-aminobenzoic acid (3-Abz) at the CCG codon using tRNA^Pro1E2_CGG.^295,298 Using these previously optimized conditions, Miura and co-workers performed a RaPID selection against the SARS-CoV-2 main protease (M^pro) using two cyclic γ^2,4-AAs, cis- and trans-3-ACBC, and two cyclic β^2,3-AAs, (1R,2S)-ACPC and (1S,2S)-ACPC, at the GAG, GUG, GUU, and UGU codons, respectively.⁵⁸⁷ While cis- and trans-3-ACBC and (1R,2S)-ACPC used tRNA^Pro1E2, (1S,2S)-ACPC required the use of tRNA^GluE2.²⁴ The RaPID library was designed to have at least one cyclic γ^2,4-AA appear in the sequence to ensure their incorporation into every peptide.⁵⁸⁷ Four of the seven synthesized teMPs (GM1, GM2, GM4, and GM5) with binding affinities ranging from 2.3 to 71 nM contained the conserved four residue motif d-Tyr-Phe-His-γ^2,4-AA (where γ^2,4-AA was either cis- or trans-3-ACBC) at the N-terminus (Figure 23f). GM1, GM4, and GM5 also had low IC₅₀ values of 40, 50, and 40 nM, respectively, against the hydrolytic activity of M^pro. These teMPs were also extremely resistant to serum proteases with half-lives of 90, 126, and 32 h, respectively. Mutation of the γ^2,4-AA to Ala in the conserved motif decreased proteolytic stability by 1.5-fold to over 10-fold. A cocrystal structure between M^pro and GM4, which contains cis-3-ACBC, revealed that the carbonyl of the amide linking His at position 3 and cis-3-ACBC at position 4 is positioned in the oxyanion hole in a similar manner to the nitrile of the inhibitor nirmatrelvir (Figure 25).⁶²⁵ Further optimization by mutation of His at position 3 to Glu (GM4H3Q) caused a 6-fold increase in binding affinity (K_D = 0.86 nM) and 5-fold more potent inhibitory activity (IC₅₀ = 10 nM).⁵⁸⁷ Thus, cyclic γ^2,4-AAs can not only significantly help with proteolytic stability but can directly interact with the binding pocket of proteins in a unique manner to improve binding affinity.

Figure 25.

Macrocyclic SARS-CoV-2 main protease (M^pro) inhibitor (PDB: 7Z4S). (a) Complex of M^pro (gray) and the macrocyclic peptide GM4 (cyan). (b) The catalytic dyad amino acids of M^pro, Cys145 and His41, are labeled. GM4 penetrates deep into the substrate binding cleft. The tetrapeptide segment Phe2-His3-γ¹4-Leu5 of GM4 is shown, where γ¹ is cis-3-ACBC. The key amino acids are His3 and γ¹4, which occupy the S1 and S1′ pockets, respectively. Notably, the His3 backbone carbonyl is positioned in the M^pro oxyanion hole, forming a hydrogen bond with the Cys145 backbone amide (red dashed line).⁵⁸⁷

5.2.2.6. Non-Amino Acid Exotic Monomers

Other than amino acid-based npMs, the ribosome is able to incorporate a variety of backbone-modifying exMs without an amine or carboxylic acid group, such as α-hydroxy acids (refs (198, 203, 537, 546, 575, 600, 626, and 627)) and α-thio acids with tRNA^AsnE2;^21,585,628 thionoesters with tRNA^GluE2;^403,576 and α-aminoxy acids and α-hydrazino acids with tRNA^Pro1E2 (Figure 3).^23,404

As previously mentioned, nearly all RaPID selections are performed using teMPs. However, Takatsuji and co-workers took advantage of α-thio acids to form backbone macrocyclic peptides via native chemical ligation that were compatible with RaPID (Figure 26).²¹ First, a linear peptide was initiated using the dipeptide of (R)-thiazolidine-4-carboxylic acid (Thz) and (S)-2-amino-4-(2-chloroacetamido)butanoic acid (Cab). Downstream, the α-thio acid analog of p-chlorophenylalanine (^HSF^Cl) was incorporated at the AUG codon using tRNA^AsnE2_CAU, followed by a Cys residue. Next, the side chain thiol of Cys underwent spontaneous thioester exchange with ^HSF^Cl, forming a new thioester bond between the Cys side chain and the peptide main chain. Subsequently, the freed thiol of ^HSF^Cl attacked the chloroacetyl group of Cab, forming a teMP. Finally, deprotection of Thz with sodium cyanoborohydride converted it to ^MeCys, which underwent spontaneous native chemical ligation. The final backbone macrocyclic peptide still contained all the translated pAAs and npMs, meaning that the Pu linker necessary for RaPID selection was still attached. Thus, this methodology can be used in the future to discover backbone macrocyclic peptides containing backbone-modifying exMs using RaPID.

Figure 26.

Methodology for creation of backbone macrocyclic peptides in the RaPID system using α-thio acids and native chemical ligation. First, thioether exchange between ^HSPhe^Cl and a downstream Cys occurs, followed by spontaneous thioether macrocyclization. Second, treatment with NaBH₃CN reduces the thiazolidine to ^MeCys, allowing for native chemical ligation. As Pu is still attached to the backbone macrocyclic peptide, this methodology is applicable for RaPID selection.²¹

5.2.2.7. Enzymatic, Post-translational Modifications to Form Natural Product-like Peptides

By directly adding purified enzymes to the FIT system, various enzymatic post-translational modifications can be carried out on npM-containing peptides.^{573−575,577} Thiopeptides are a class of ribosomally synthesized and post-translationally modified peptides with their defining feature being the presence of a six-membered heterocycle in the peptide backbone, usually accompanied by other npMs such as azoles and dehydroalanines.⁶²⁹ Lactazole A is a thiopeptide from Streptomyces lactacystinaeus that is biosynthesized from five enzymes and a precursor peptide.⁶³⁰ Vinogradov and co-workers combined the Laz enzymes with the FIT system to create the FIT-Laz system.^575,631 Translation of the leader peptide and treatment with the Laz enzymes resulted in a thiopeptide scaffold containing a dehydroalanine, thiazole, oxazole, and pyridine moiety. To demonstrate that npM-containing core peptides could be converted into the thiopeptide scaffold, npMs were first incorporated into the AUG codon of the Laz leader peptide using tRNA^Pro1E2_CAU.⁵⁷⁵ A wide variety of npMs such as ^MeGly, ^MeAla, d-Ala, d-Ser, Ac₅c, pentafluoro-l-phenylalanine [Phe(F₅)], 5-hydroxy-l-tryptophan, α-hydroxy Ala, β³-Met, and β³-Leu were all compatible with the FIT-Laz system, resulting in thiopeptides containing a single incorporation of an npM. Additionally, a 16-mer peptide containing Ac₅c, Phe(F₅), ^MeGly, and ^MeAla at the CAU, UGG, AAG, and UUU codons, respectively, was successfully translated and post-translationally modified into the thiopeptide scaffold, resulting in a peptide containing three heterocyclic rings and four backbone-modifying npMs. This work demonstrates the potential for the discovery of pseudonatural products as peptide therapeutics using the RaPID system.

6. Conclusion and Future Perspectives

Since the elucidation of the quasi-universal genetic code in the 1960s, diverse methodologies have been used to ribosomally synthesize polypeptides containing npMs. Early efforts were fraught with difficulties caused by native translation systems, resulting in low translation efficiency and high misincorporation by pAAs. Thus, the engineering of translational machinery has been important in advancing the field of genetic code manipulation.

The key advancement that is fundamental to genetic code manipulation and is shared between almost all modern techniques is the use of orthogonal tRNAs. With these orthogonal tRNAs, hundreds of npMs have been incorporated into polypeptides both in vitro and in vivo. Major improvements have come through rational design and directed evolution of tRNAs: mutations of ARS recognition elements have increased orthogonality, fine-tuning recognition by EF-Tu and EF-P has increased npM translation efficiency, and creating or repurposing codons for assignment to npMs has allowed for expanded genetic codes comprising more than 20 monomers.

Engineering of tRNAs for the in vitro incorporation of npMs using the FIT system has been particularly successful. As a result, not only have l-AAs with non-proteinogenic side chains been ribosomally elongated, but also N-modified-AAs, d-AAs, α,α-disubstituted amino acids, β-AAs, and γ-AAs. Additionally, non-amino acid substrates that can form unique non-amide backbones have been elongated, such as α-hydroxy acids, α-thio acids, thionoesters, α-aminoxy acids, and α-hydrazino acids. Inclusion of backbone-modifying npMs can impart important characteristics to polypeptides, such as proteolytic stability, structural rigidity, binding affinity, and membrane permeability. To this end, screening of diverse peptide libraries against proteins of interest using RaPID has resulted in a multitude of bioactive teMPs with favorable pharmacological properties.

Given that polypeptide synthesis is a complex process that necessitates the cooperative function of a multitude of different mechanisms, further progress in genetic code manipulation should attempt to integrate recent advances in all areas of engineered translational machinery. For in vitro sense codon reprogramming, low translation efficiency of highly exotic monomers such as γ-AAs limits the number of npMs that can be simultaneously incorporated into a single peptide. Conversely, the major hurdle for in vivo genetic code expansion is to incorporate backbone-modifying npMs into proteins within the context of complex cellular systems.

There are multiple avenues for the improvement of in vitro translation systems. One area of potential development would be to combine engineered ribosomes with engineered tRNA^GluE2 and tRNA^Pro1E2. Recent work by McFeely and co-workers demonstrated that synonymous codon boxes could be orthogonally translated with in vitro transcribed, unengineered tRNAs and mS12 hyperaccurate ribosomes.^415,632 Integration of mS12 ribosomes into the FIT system could allow for more efficient incorporation of backbone-modifying npMs by significantly reducing mis-decoding events. Other engineered ribosomes, such as those evolved for incorporation of d-AAs or β-AAs, could also be utilized.²³² Similarly, EF-Tu could be further engineered to improve npM incorporation. While engineered EF-Tu in the form of EF-Sep has been employed in the FIT system for translation of negatively charged npMs, no other reports of using recombinant translation factors in the FIT system have been reported.²⁶⁵ Thus, directed evolution of EF-Tu to specifically recognize other classes of npMs would be a compelling avenue for exploration.

Another approach would be to incorporate modified nucleotides into orthogonal tRNAs in vitro to further improve translation efficacy. Modified nucleotides play an important role in many aspects of translation such as tRNA stability and decoding, including in the tRNA core.³⁸⁷ For example, Gamper and co-workers used flexizymes with fully modified suppressor tRNA^SufB2 to incorporate Pro analogs using quadruplet codons.⁴²⁴ Thus, it would be valuable to ascertain the extent to which these orthogonal tRNAs could be substrates of tRNA-modifying enzymes and whether the installed modifications could improve translation fidelity or efficiency at various codons.

To increase the incorporation efficacy of ^MeAAs, the binding affinity for EF-Tu of each ^MeAA-tRNA was tuned for uniform binding.^250,598 An analogous concept would be the design of unique tRNAs tuned for specific npM/codon pairs. This approach relies on the fact that each tRNA has a unique three-dimensional structure that dictates its interactions with the ribosome and that every monomer adopts slightly different conformations within the PTC. While this task is theoretically possible, the development of a high-throughput workflow to explore different tRNA mutations for each npM/codon permutation would be beneficial to fully examine this concept.

Turning to in vivo genetic code manipulation, perhaps the most pressing technological advancement would be the ability to effectively aminoacylate monomers that are not l-α-AA derivatives. Some attempts have been made to discover a ribozyme that is capable of in vivo aminoacylation analogous to the flexizyme, such as an evolved T-box riboswitch by Ishida and co-workers.⁶³³ However, it is more likely that continued evolution of orthogonal ARS/tRNA pairs will yield the best results due to significant advances in in vivo continuous evolution and the advent of AlphaFold, machine learning, and other computational techniques. This was recently realized by Dunkelmann and co-workers in 2024 through successful in vivo incorporation of three different β³-AAs and one α,α-disubstituted amino acid into GFP(150UAG) using E. coli.⁴⁹⁴ This new discovery platform, called bio-mREX, was built upon the Chin group’s previously established tREX platform described in section 2.2.6.¹⁷⁴ Briefly, this specialized tRNA display selects for PylRS variants that acylate their cognate tRNAs with npMs, independent of ribosomal translation efficiency.⁴⁹⁴ As such, there is significant potential for the rapid expansion of available npMs for in vivo genetic code manipulation.

Further in vivo tRNA engineering could continue to explore the impact of mutations throughout the tRNA body on translation efficiency. While this Review has mainly focused on translational mechanisms in prokaryotes, such mutations should also especially be investigated in eukaryotes. To date, there have only been a handful of in vivo studies—all in E. coli—that investigate the effects of D arm mutations which would recruit EF-P. In-depth analysis on whether such mutations would improve the ribosomal elongation of npMs is yet unreported. In eukaryotes, the homologues of EF-Tu (eEF1A) and EF-P (eIF5A) could similarly be investigated and potentially harnessed for improved translation efficiency.^634,635 Similarly, newer structural and kinetic studies of eIF2, eEF2, and other eukaryotic translation components could shed light on how to prevent peptidyl drop-off or mis-decoding in eukaryotic genetic code manipulation.^636−639 The recent report of the VADER platform by Jewel and co-workers is an impressive demonstration of directed tRNA evolution in mammalian HEK293T cells; approaches such as this one could be used to expand the scope of ribosomally translatable npMs in various model organisms.²⁶⁰ For instance, it would be highly valuable to be able to produce backbone-modifying macrocyclic peptides directly in human cells and thus evaluate their bioactivity in situ.

For efficient quadruplet codon decoding in vivo, it is still necessary to reduce competitive decoding by canonical triplet-decoding tRNAs. One strategy might be to use orthogonal ribosome/tRNA pairs as a basis for generating ribosomes that exclusively accept quadruplet tRNAs, enabling the simultaneous coexistence of triplet and quadruplet genetic codes within one system.²⁶⁸ Alternatively, quadruplet decoding could be combined with reassignment of the triplet codon corresponding to the first three bases of the quadruplet.⁶⁸ For instance, reports have shown that decoding efficiency of UAGN quadruplets is substantially increased in the absence of RF1, which competes for recognition of the UAG stop codon.^441,640 Such approaches could help maximize the diversity of the in vivo genetic code.

Finally, in vivo sense codon reassignment has seen significant advancements in the past few years, notably the development of the E. coli strain Syn61Δ3 that enables simultaneous reassignment of two Ser codons and one stop codon using engineered Pyl tRNAs.³⁸⁸ Future efforts to extend this approach to additional synonymous codons should enable the development of even more diverse genetic codes in vivo.

In the past few decades, new understanding of the complex biochemical interactions underlying translation has given rise to engineered orthogonal tRNAs, ARSs, ribosomes, and translation factors. Concomitantly, there has been significant growth in the ability to incorporate npMs into polypeptides. These npM-containing polypeptides have long-term potential in pharmaceuticals, protein labeling, studying of protein–protein interactions, protein localization, enzymatic activity and cellular signaling, and other biochemical applications. With increasing interest being garnered for genetic code manipulation, we expect that interdisciplinary collaborations will give rise to more robust and diverse methods for incorporation of npMs in the near future.

Glossary

Abbreviations

npM

Non-proteinogenic monomer; any monomer used in genetic code manipulation

npAA

Non-proteinogenic amino acid; l-α-amino acids with non-proteinogenic side chains, d-AAs, N-alkyl-AAs, β-AAs, etc.

exM

Exotic monomer; non-amino acid exotic monomers such as α-hydroxy acids, α-thio acids, thionoesters, etc.

pAA

Proteinogenic amino acid; the 20 standard l-α-amino acids, selenocysteine, and pyrrolysine

Backbone-modifying npMs

Non-l-α-AAs

Genetic code manipulation

Any kind of genetic code manipulation to incorporate non-proteinogenic monomers, inclusive of genetic code expansion and sense codon reassignment

Genetic code expansion

Expanding the genetic code outside of the standard 61 elongator codons: nonsense codon suppression, quadruplet codons, and unnatural base pairs

Sense codon reassignment

Utilizing one of the 61 elongator codons for incorporation of non-proteinogenic monomers

ARS

Aminoacyl tRNA synthetase

AA-tRNA

Aminoacyl-tRNA

AA

Amino acid

EF

Elongation factor

EF-P

Elongation factor P

EF-Tu

Elongation factor thermo unstable

EF-G

Elongation factor G

FIT

Flexible in vitro translation

mRNA

Messenger RNA

tRNA

Transfer RNA

○

Denotes non-Watson–Crick base pairing

•

Denotes Watson–Crick base pairing

AA-AMP

Aminoacyl-adenylate

E. coli

Escherichia coli

k²C or L

Lysidine

mnm⁵s²U

5-Carboxymethylaminomethyl-2-thiouridine

Pyl

Pyrrolysine

S. cerevisiae

Saccharomyces cerevisiae

Nle

Norleucine

Sec

Selenocysteine

WC

Watson–Crick

RF

Release factor

UAA

Ochre codon

UAG

Amber codon

UGA

Opal codon

DHFR

Dihydrofolate reductase

HPhe

Homophenylalanine

oITyr

3-Iodo-l-tyrosine

pFPhe

p-Fluoro-l-phenylalanine

pNO₂Phe

p-Nitro-l-phenylalanine

Mj

Methanocaldococcus jannaschii

Tyr^OMe

O-Methyl-l-tyrosine

D. hafniense

Desulfitobacterium hafniense

C. elegans

Caenorhabditis elegans

Mb

Methanosacrcina barkeri

Mm

Methanosacrcina mazei

MtmB

Monomethylamine methyltransferase

GFP

Green fluorescent protein

hSOD

Human superoxide dismutase 1

AlkLys

N^ε-Propargyloxycarbonyl-l-lysine

BocLys

N^ε-Boc-l-lysine

CMa

Candidatus Methanomethylophilus alvus

HGln

l-Homoglutamine

pAzPhe

p-Azido-l-phenylalanine

sfGPF

Superfolder GFP

ABT

4-[(2-Aminoethyl)carbamoyl]benzyl thioesters

aFx

Amino-flexizyme

PrgGly

l-Propargylglycine

CBT

p-Chlorobenzyl thioesters

CFPS

Cell-free protein synthesis

CME

Cyanomethyl ester

DBE

3,5-Dinitrobenzyl esters

dFx

Dinitro-flexizyme

eFx

Enhanced flexizyme

^fMet

N^α-Formyl-l-methionine

Fx

Flexizyme

PURE

Protein synthesis using recombinant elements

Ser^OMe

O-Methyl-l-serine

VinGly

l-Allylglycine

BioLys

N^ε-Biotinyl-l-lysine

GTP

Guanosine triphosphate

K_D

Equilibrium dissociation constant

PTC

Peptidyl transferase center

^MeAA

N^α-Methylated amino acid

^MepNO₂Phe

N-Methyl-p-nitro-l-phenylalanine

^MeTyr^OMe

N-Methyl-O-methyl-l-tyrosine

Nva

Norvaline

Phg

Phenylglycine

rac-β²-Ala

Racemic β²-homoalanine

β²-Ala

(S)-β²-Homoalanine

β^3,3-Aib

β^3,3-Homoaminoisobutyric acid

β³-Ala

(S)-β³-Homoalanine

β³-pBrAla

(S)-β³-(p-Bromophenyl)alanine

β³-d-Ala

(R)-β³-Homoalanine

β³-Phe

(p-Bromo)-l-phenylalanine

Nap

Naphthylalanine

rRNA

Ribosomal RNA

ΔG°

Thermodynamic free energy

ΔΔG°

Relative difference in thermodynamic free energy

AcLys

N^ε-acetyl-l-lysine

pIPhe

p-Iodo-l-phenylalanine

pAcPhe

p-Acetyl-l-phenylalanine

VADER

Virus-assisted directed evolution of tRNAs

2-Abz

2-Aminobenzoic acid

2-ACHC

2-Aminocyclohexanecarboxylic acid

2-ACPC

2-Aminocyclopentanecarboxylic acid

3-ACBC

3-Aminocyclobutanecarboxylic acid

3-ACHC

3-Aminocyclohexanecarboxylic acid

3-ACPC

3-Aminocyclopentanecarboxylic acid

Aib

2-Aminoisobutyric acid

d-^NNPro

d-Hydrazinoproline

d-^NOAla

d-α-Aminoxypropionic acid

d-^NOPhe

d-α-Aminoxy-β-phenylpropionic acid

l-^NNPro

l-Hydrazinoproline

l-^NOAla

l-α-Aminoxypropionic acid

l-^NOPhe

l-α-Aminoxy-β-phenylpropionic acid

^NNMeGly

N^α-Methyl hydrazinoacetic acid

^NOAib

α-Aminoxy-isobutyric acid

^NOGly

α-Aminoxyacetic acid

IC

Initiation complex

^AcPro

N^α-Acetyl-l-proline

^ClAcPro

N^α-Chloroacetyl-l-proline

IF

Initiation factor

RRF

Ribosome recycling factor

SD

Shine–Dalgarno

ASL

Anti-codon stem-loop

ac⁴C

N4-Acetylcytidine

Cm

2′-O-Methylcytidine

cmo⁵U

Uridine 5-oxyacetic acid

gluQ

Glutamyl-queuosine

mnm⁵U

5-Methylaminomethyluridine

Q

Queuosine

xnm⁵U

5-Iminomethyl-uridine

ms²i⁶A

2-Methylthio-N6-isopentenyladenosine

Ψ

Pseudouridine

Dap

Diaminopropionic acid

Thr^OMe

O-Methyl-l-threonine

UB

Unnatural base

isoG

Isoguanosine

SSO

Semisynthetic organism

isoC

Isocytidine

AzLys

N^ε-(2-Azidoethoxy)-carbonyl-l-lysine

PCSK9

Proprotein convertase subtilisin-like/kexin type 9

PEG

Polyethylene glycol

PPI

Protein–protein interaction

^ClAcnpM

N-Chloroacetylated non-proteinogenic monomer

Pu

Puromycin

RaPID

Random nonstandard peptide integrated discovery

teMP

Thioether closed macrocyclic peptide

TfaLys

N^ε-Trifluoroacetyl-l-lysine

Ac₃c

1-Aminocyclopropane-1-carboxylic acid

Ac₅c

1-Aminocyclopentane-1-carboxylic acid

hEGFR

Human epidermal growth factor receptor

iPGM

Cofactor-independent phosphoglycerate mutase

Atp

3-Aminothiophene-2-carboxylic acid

hFXIIa

Human factor XIIa

M^pro

SARS-CoV-2 main protease

IFNGR1

Interferon-gamma receptor 1

3-Abz

3-Aminobenzoic acid

Cab

(S)-2-Amino-4-(2-chloroacetamido)butanoic acid

Thz

(R)-Thiazolidine-4-carboxylic acid

Hhc

Hydroxyhydrocarbon

^HSF^Cl

α-Thio-l-p-chlorophenylalanine

Phe(F5)

Pentafluoro-l-phenylalanine

A

Adenine

T

Thiamine

C

Cytosine

G

Guanine

U

Uracil

N

A, T/U, C, G

Y

C or T/U

S

G or C

R

A or G

Ala

Alanine

Arg

Arginine

Asn

Asparagine

Asp

Aspartate

Cys

Cysteine

Gln

Glutamine

Glu

Glutamate

Gly

Glycine

His

Histidine

Ile

Isoleucine

Leu

Leucine

Lys

Lysine

Met

Methionine

Phe

Phenylalanine

Pro

Proline

Ser

Serine

Thr

Threonine

Trp

Tryptophan

Tyr

Tyrosine

Val

Valine

Biographies

Maxwell Sigal received his B.A. in Chemistry from Northwestern in 2020. He received his M.Sc. degree in Chemistry from The University of Tokyo under the supervision of Professor Hiroaki Suga and subsequently began his Ph.D. in 2023. His research mainly focuses on the ribosomal synthesis of peptide libraries containing non-proteinogenic monomers.

Satomi Matsumoto received her B.Sc. degree in Chemistry from Tokyo University of Science under the supervision of Professor Motoyuki Shimonaka. After joining the group of Professor Hiroaki Suga at The University of Tokyo in 2021, she received her M.Sc. degree in Chemistry in 2023 and is currently a Ph.D. candidate. Her research mainly focuses on the ribosomal synthesis of peptide libraries containing exotic monomers.

Adam Beattie is undertaking a Ph.D. in Chemistry in the group of Professor Hiroaki Suga at The University of Tokyo. His research mainly focuses on the ribosomal synthesis of peptide libraries containing exotic monomers.

Takayuki Katoh received his Ph.D. degree in Engineering from the University of Tokyo under the guidance of Prof. K. Watanabe and Prof. T. Suzuki. His Ph.D. study focused on RNA interference and biogenesis of mammalian microRNAs. After postdoctoral research at Japan Biological Informatics Consortium, he was appointed as an Assistant Professor at the University of Tokyo in 2009 and promoted to Associate Professor in 2018. His research now focuses on the engineering of tRNAs that enable ribosomal incorporation of diverse non-proteinogenic amino acids.

Hiroaki Suga received his Ph.D. degree in Chemistry in 1994 from the Massachusetts Institute of Technology. Subsequently, he was an Assistant Professor at the Department of Chemistry in the State University of New York at Buffalo from 1997 and was promoted to Associate Professor in 2002. He became Professor of the Research Center for Advanced Science and Technology at the University of Tokyo in 2003 and, in 2010, of the Department of Chemistry, Graduate School of Science. He is the recipient of the Akabori Memorial Award 2014, the Japanese Peptide Society and Max-Bergmann Gold Medal 2016, the Nagoya Medal 2017 Silver, and a Wolf Prize Laureate in Chemistry in 2023.

Author Contributions

CRediT: Maxwell Sigal visualization, writing-original draft, writing-review & editing; Satomi Matsumoto visualization, writing-original draft, writing-review & editing; Adam Beattie visualization, writing-original draft, writing-review & editing; Takayuki Katoh conceptualization, funding acquisition, project administration, supervision, writing-review & editing; Hiroaki Suga conceptualization, funding acquisition, project administration, supervision, writing-review & editing.

The authors declare no competing financial interest.

Special Issue

Published as part of Chemical Reviewsvirtual special issue “Noncanonical Amino Acids”.