Cracking the Code: Reprogramming the Genetic Script in Prokaryotes and Eukaryotes to Harness the Power of Noncanonical Amino Acids

Abstract

Over 500 natural and synthetic amino acids have been genetically encoded in the last two decades. Incorporating these noncanonical amino acids into proteins enables many powerful applications, ranging from basic research to biotechnology, materials science, and medicine. However, major challenges remain to unleash the full potential of genetic code expansion across disciplines. Here, we provide an overview of diverse genetic code expansion methodologies and systems and their final applications in prokaryotes and eukaryotes, represented by Escherichia coli and mammalian cells as the main workhorse model systems. We highlight the power of how new technologies can be first established in simple and then transferred to more complex systems. For example, whole-genome engineering provides an excellent platform in bacteria for enabling transcript-specific genetic code expansion without off-targets in the transcriptome. In contrast, the complexity of a eukaryotic cell poses challenges that require entirely new approaches, such as striving toward establishing novel base pairs or generating orthogonally translating organelles within living cells. We connect the milestones in expanding the genetic code of living cells for encoding novel chemical functionalities to the most recent scientific discoveries, from optimizing the physicochemical properties of noncanonical amino acids to the technological advancements for their in vivo incorporation. This journey offers a glimpse into the promising developments in the years to come.

1. Introduction

The central dogma of molecular biology states that genetic information flows from DNA to RNA to proteins, the latter serving as the primary functional units of cells. All life on Earth evolved a conserved genetic code for translating an RNA sequence into protein. This code, even present in the genome of viruses, defines how 64 base triplets are translated into a set of 20,¹ and in rare cases 22, so-called proteogenic or canonical amino acids (cAAs). The vast range of protein structures and functions is thus essentially based on 20 amino acids. There are very few exceptions for bypassing the conserved genetic code: those that do represent rather minor changes, for example, a handful of codons being interpreted differently in genomes originating from different kingdoms of life, or comparing the genetic codes of nuclei, mitochondria, and chloroplasts.²⁻⁵ Examples of these natural deviations include the reassignments of the CUG codon in the Candida yeast species from leucine to serine,⁶ of the UGA stop codon to encode tryptophan in Mycoplasma(7) or glycine in human microbiota,⁸ and of a single or both of the UGA and UAA stop codons to encode cAAs in ciliate organisms.⁹⁻¹¹ In some marine ciliates, all three standard stop codons are naturally recoded to cAAs at gene-internal positions while still being used as termination signals at the 3′ end of transcripts.¹² This recoding is context-specific and dependent on sequence elements in close proximity.¹³

The conserved set of 20 proteogenic amino acids is, in some cases, extended by selenocysteine (Sec) as the 21st and pyrrolysine (Pyl) as the 22nd cAAs.¹⁴⁻¹⁶ Genetic codes expanded with selenocysteine are found in all kingdoms of life, including in humans.¹⁷ The redox-active selenol side chains are found, for example, as part of the catalytic center of enzymes with antioxidant and redox-regulatory roles.¹⁸ Selenocysteine incorporation into proteins occurs in response to the UGA codon exclusively for transcripts that harbor cis-acting selenocysteine insertion sequence (SECIS) elements.^19,20 In contrast, pyrrolysine is encoded by the UAG codon in a few methanogenic archaeal and bacterial species²¹ and, similar to the selenocysteine systems, is only incorporated if a cis-acting pyrrolysine insertion sequence (PYLIS) element is present.²⁰

Notably, a vast variety of naturally occurring noncanonical amino acids (ncAAs) are not used as building blocks for proteins by the ribosomal machinery and instead serve other functions (e.g., as signaling molecules), or represent transient intermediates during cAA biogenesis and other metabolic reactions, or result from the post-translational modification (PTM) of cAA residues. The myriad of post-translational reactions, some of which are reversible, tremendously diversify the functionality of cAAs within a peptide chain. These reactions give rise to cAA derivatives that are phosphorylated, glycosylated, or lipoylated, to name just a few examples, and can induce changes in protein folding, stability, solubility, catalytic activity, or subcellular location.

Nature has evolved enzyme complexes capable of installing ncAAs within peptides and diverse secondary metabolites. These enzyme complexes consist of multiple distinct catalytic modules, each of which catalyzes a specific reaction. Found in archaea, bacteria, fungi, plants and marine organisms, such as plankton and sponges, the polyketide synthase (PKS) and nonribosomal peptide synthetase (NRPS) complexes give rise to highly diversified natural products, many of which are known for their antibiotic, anti-inflammatory, or other biological activity.^22,23 However, the high specificity of enzymatic PKS and NRPS modules for catalyzing only one particular reaction, and the non-templated nature of these reactions, limit the utility of PKS and NRPS components for the purpose of engineering proteins.

Making use of a technique referred to as genetic code expansion (GCE), ncAAs can be co-translationally installed into proteins, broadening the repertoire of building blocks for making proteins.^24,25 This includes naturally occuring, as well ascompletely artificial amino acids with designer properties. ncAAs are also often referred to as unnatural amino acids. This can be misleading since many do occur in nature, although instead of being used in ribosomal translation, they fulfill other biological roles as metabolites, hormones, and synaptic signaling molecules. Distinguishing proteogenic from non-proteogenic or canonical from noncanonical thus provides more precise classifications. The incorporation of ncAAs into proteins enables protein structure and function to be tuned beyond what the naturally evolved set of cAAs can achieve. The shape and stability of proteins can be altered, as can their dynamics, interactions, and enzymatic activities.²⁶ In addition, entirely new chemical or biological functionalities can be conferred to a protein of interest (POI). The ability to genetically encode ncAAs in prokaryotes and eukaryotes has therefore emerged as a powerful tool for unlocking the largely untapped potential of unnatural protein properties. This has opened, and will continue to open, new avenues of research in synthetic biology, protein engineering, material science, and precision medicine.

This review highlights the scientific milestones underpinning our ability to encode ncAAs into prokaryotes and eukaryotes, drawing knowledge mainly from E. coli and mammalian cells as major model systems, respectively.

By focusing on E. coli as a simple model organism that is most commonly used for developing GCE approaches, we first introduce the tools for performing GCE, detailing the diverse technologies that progress the encoding of single and multiple ncAAs at the same time. We explain the underlying molecular mechanisms of key technologies enabling specific GCE in prokaryotes and eukaryotes, ultimately leading to the latest breakthroughs and remaining challenges in mammalian cell biology and beyond. An inherent problem of GCE is the limited specificity for a defined RNA targeted for orthogonal translation, as naturally occurring codons are typically reprogrammed in parallel, which results in proteome-wide incorporation of the ncAA. As is discussed, this problem was recently solved for E. coli by exciting developments in whole-genome engineering, which enables the generation of strains with specific codons made “available” for custom reprogramming. In mammalian cells, whole-genome engineering cannot yet manage the complexity of the large multi-chromosome genomes. Here, we focus on alternative synthetic biology concepts for creating spatially distinct orthogonally translating systems that achieve RNA-specific translation and GCE. Our focus is on the methods applied to expand GCE in E. coli and mammalian cell culture, and we refer readers interested in GCE specifically applied in animals and multicellular organisms in general to other literature.²⁷⁻²⁹ Likewise, NRPS and PKS utilize amino acids in a different way to ribosomal translation, and thus we refer readers interested in this aspect of incorporating amino acids via enzyme modules into natural products to excellent reviews by Wang et al., Nivina et al., Hwang et al., and Jaremko et al.^23,30−32

In addition to reviewing methodology and technology within prokaryotes and eukaryotes, we devote section 4 to illustrating the chemical diversity of designer amino acids, their defined chemical properties, and how they can be genetically encoded in living cells. We categorize the emerging new applications enabled by ncAAs and mainly focus on those ncAAs which are used to study cell biology. We have mined the structures and properties of as many ncAAs as we found that have been genetically encoded in living cells to date, including the categorization of functional groups and measures of biocompatibility, together with fields of application and employed adoptive host organisms. Besides giving the reader a comprehensive background to the different chemical functionalities and example cases discussed in section 4, the mined Table S1 could also complement the iNClusive database, a powerful searchable online resource for genetic code expansion.³³

1.1. Discovery of the Genetic Code

In the 1960s, three independent groups of scientists cracked the genetic code of life.^1,34,35 Over a decade before the invention of DNA sequencing, they relied on simple but powerful genetics experiments to demonstrate that three-nucleotide ensembles, known as codons, carry the information of amino acids to define a protein sequence. The teams in vitro transcribed the first synthetic nucleic acids and crossed mutated bacteriophage T4 strains, measuring their ability to infect bacterial hosts. These experiments showed the existence of a degenerate genetic code with 64 codons, comprising 61 sense codons that encode 20 amino acids, and three stop codons. Sense codons are decoded by complementary anticodons on transfer RNA (tRNA) molecules according to the wobble hypothesis.^36,37 Most bacteria and all eukaryotes have three stop codons, named amber (UAG), opal (UGA), and ochre (UAA). These serve as termination signals that are not recognized by any tRNA (transfer-RNA) anti-codon. Instead, they are readily bound by release factor proteins that mimic tRNAs’ structure and binding properties but release the amino acid chain from the ribosomal subunits when the translation process is completed.

In his seminal paper, Francis Crick proposed that codons are likely allocated to tRNA-coupled amino acids by mere chance—the so-called “frozen accident”.³⁸ On the basis of previous postulation,³⁹ this genetic code is highly conserved due to any single change to the code having a high probability of being lethal as it would affect the whole proteome. In fact, any stable change of the genetic code would necessitate multi-layered, concerted mutations in almost all instances of a particular codon in the genome, in at least one tRNA species to generate a matching anticodon, as well as in its cognate tRNA synthetase (RS) for loading an amino acid onto this tRNA.

1.2. Expanding the Genetic Code

Stop codons can, in principle, be used to encode an amino acid. This is illustrated by the deviated and naturally evolved genetic codes of a few methanogenic bacteria and archaea that encode pyrrolysine using dedicated tRNAs which recognize the amber stop codon (UAG in RNA or TAG in DNA) as a sense codon when a cis-acting PYLIS element is present as in a handful of specifically recoded genes.^20,40 In the bulk transcriptome, UAG is still interpreted as a stop codon, in parallel to the recoding. In the active site of archaeal methylamine methyltransferases, pyrrolysine residues are critical for activating the N-methyl groups in methylamine substrates of the methanogenesis pathway, as a carbon source and to provide energy.⁴¹⁻⁴³ All known pyrrolysine-encoding organisms share common features. The genes of PylRS (pylS) and the tRNA^Pyl (pylT) are located in close proximity to the methyltransferase gene cluster.⁴⁴ Pyrrolysine is biosynthesized in an enzymatic cascade starting from two molecules of lysine and charged to the tRNA^Pyl by the catalytic activity of PylRS. Although pyrrolysine is encoded in only a few proteins that are mainly part of the methanogenesis pathway, it is essential for the cellular metabolism and physiology of the species using it.⁴⁵ The example of the pyrrolysine system illustrates how nature itself evolved systems for the recoding of (rare) stop codons as sense codons. This can serve as an inspiration for encoding ncAAs in general, by making them accessible to ribosomal translation via the expansion of the genetic code.

The incorporation of ncAAs with diverse chemical structures offers the chance to modify or add to the properties of a protein (Figure 1). The site-specific installation of ncAAs into dipeptides in vitro (by using ligation to couple ncAAs to tRNA molecules) was first reported in 1983⁴⁶ and of larger proteins in 1989.^47,48 Suppressor tRNAs were artificially aminoacylated by a purified tRNA synthetase, utilized as part of in vitro transcription–translation systems. This allowed for encoding the tRNA-coupled ncAA with an amber stop codon. In vitro-aminoacylated tRNAs could also be co-injected into Xenopus laevis oocytes to suppress up to three amber nonsense codons within a messenger RNA (mRNA) that matched the CUA tRNA anticodons.⁴⁹⁻⁵¹ This enabled the expression of an acetylcholine membrane receptor harboring up to three ncAAs incorporated for investigating the contribution of these sites to the structure and agonist/antagonist-binding functionality. The aforementioned methods necessitate loading a tRNA with the ncAA in vitro by synthetic aminoacylation, which can be achieved by chemical ligation, enzymatically with purified tRNA synthetases, or by using engineered RNA catalysts called flexizymes. Strategies for in vitro aminoacylation have been reviewed before,^52,53 and we refer the interested reader to those. In later sections of this review, we comprehensively expand on approaches to generating in vivo GCE systems that necessitate aminoacylation of heterologously expressed tRNAs in living cells.

Figure 1.

GCE for amino acid recoding. The conserved genetic code of life on earth and its expansion via stop codon suppression. Recoding of a stop codon to a sense codon allows the co-translational incorporation of ncAAs into proteins, harboring diverse chemical structure and functionality, as illustrated with selected structures from Table S1.

1.3. Genetically Encoding an Expanded Genetic Alphabet

In 1997, the Schultz laboratory reported the first genomic encoding of an ncAA in vivo, combining mutated components of the E. coli glutaminyl-RS/tRNA pair within its original host.⁵⁴ This study demonstrated the feasibility of expanding the genetic code in a living organism and marked the beginning of a new era in protein engineering and cell biology. The suppression of a stop codon to instead interpret it as a sense codon for the incorporation of a defined ncAA relies on engineering an orthogonal aminoacyl-tRNA synthetase/tRNA pair (aaRS/tRNA pair) (Figure 2). The respective tRNA anticodon pairs with the stop codon used, for example, the amber UAG triplet, thus enabling the cognate aaRS to charge this heterologous tRNA with the provided ncAA.

Figure 2.

Principle mechanisms of in vivo GCE. While natural translation terminates at stop codons, GCE systems enable stop codon suppression by introducing a heterologous tRNA synthetase/tRNA pair and by providing ncAAs. The anticodon of the heterologous tRNA thus encodes the stop codon to mediate co-translational ncAA incorporation.

The orthogonality of GCE systems is multilayered, relying on up to five prerequisites: (1) Synthetase orthogonality: the aaRS recognizes only the supplied suppressor tRNA and not endogenous tRNAs of the host cell. (2) Synthetase specificity: the aaRS recognizes only the ncAA of choice and no other amino acids or substrates available in the cytoplasm. (3) tRNA orthogonality: the heterologous tRNA is exclusively aminoacylated by the aaRS and not by other aaRSs of the host. (4) Codon orthogonality: the anticodon of the heterologous tRNA recognizes the stop codon targeted for suppression. (5) RNA specificity: recoding is confined only to the respective codon in the target mRNA, and not across other instances of the codon in the whole transcriptome.

Meeting these criteria facilitates the precise incorporation of a ncAA into a protein sequence of interest in a site-specific and transcript-specific manner and without interference of the translation processes endogenous to the host organism. For early efforts in ncAA recoding, the amber triplet (UAG) was chosen for two principal reasons: (1) it is the least abundant of stop codons in the E. coli transcriptome, found in approximately 8% of RNAs, compared to opal (UGA, 28%) and ochre (UAA, 64%) stop codons, and (2) it is depleted in transcripts of essential genes.^55,56 Mutation of the tRNA anticodon to CUA, which pairs with the UAG amber codon, enables amber codon suppression by the suppressor tRNA. Recoding requires aminoacylation of the tRNA with an ncAA, mediated by its cognate aaRS, which frequently allows for changes in the anticodon (see section 2.3 for tRNA motifs recognized by aaRSs). Of note, all three stop codons can in principle serve as sense codons, as illustrated by strains of E. coli which naturally evolved mutations in the tRNA anticodon for stop codon suppression to incorporate cAAs.^57,58 The groups of Yokoyama and Schultz established orthogonal aaRS/tRNA pairs in eukaryotes, expanding the genetic codes in mammalian cells in 2002⁵⁹ and the baker’s yeast Saccharomyces cerevisiae in 2003.⁶⁰ These inventions sparked innovation in several directions. Research efforts strived toward the screening and evolution of orthogonal aaRS/tRNA pairs to encode different ncAAs with novel chemical properties to, in turn, allow for new applications. In addition, the implementation of GCE systems in other model and non-model organisms and their improved efficiency and specificity to particular ncAAs turned GCE into a technique that is broadly used across diverse fields of biological research, allowing to also genetically encode multiple different ncAAs within the same protein.

2. Expanding the Genetic Code in Prokaryotes

In this section and the next, we review the developments regarding the genetic encoding of ncAAs in prokaryotes (section 2) and eukaryotes (section 3). We focus on developments in advanced prokaryote engineering mainly in E. coli (see also reviews (26, 61−68)) and illustrate how they were transferred to eukaryotes, and especially to higher eukaryotes, such as mammalian cells.^69,70 Or, if that was not possible, we show how new approaches overcame the challenges posed by the greater complexity of eukaryotic cells. The generation of efficient GCE systems requires the multiple layers of the translation process in the respective host organism to be considered, from the cellular uptake of ncAAs and the aaRS enzyme kinetics of the ncAA peptidyl transfer^71,72 to the abundance of tRNAs, codon usage and sequence context,⁷³⁻⁷⁶ transcript abundance and stability, general transcription and translation efficiency, and the availability of translation components.⁷⁷ These processes are interconnected and differ between GCE technologies and host organisms. Establishing GCE systems in E. coli has been addressed from various angles and by using different methodologies, to which we dedicate the following sections: Discovering and evolving orthogonal aaRS/tRNA pairs (section 2.1), engineering the aaRS enzyme (section 2.2), engineering the tRNA (section 2.3), methodologies for quantifying GCE performance (section 2.4), side effects of GCE systems on the host organism (section 2.5), enabling ncAA biogenesis (section 2.6), removal of release factors (section 2.7), quadruplet codons (section 2.8), engineered ribosomes (section 2.9), artificial base pairs (section 2.10), genome reprogramming (section 2.11), biocontainment and viral resistance (section 2.12), and genome synthesis (section 2.13).

2.1. Strategies for the Evolution of Orthogonal aaRS/tRNA Pairs

Given that the biosynthesis of proteins requires the precise interplay of more than a hundred different components, the process occurs with remarkably high accuracy, speed, and specificity in matching molecule interfaces between codon and anticodon to catalyze reactions as defined. The simple ribosomal machinery of E. coli is comprised of 54 ribosomal proteins and three ribosomal RNAs with a total of more than 4500 nucleotides. The bacterial translation process further requires a minimum set of additional components, a culmination of 33 tRNAs and 21 aaRS enzymes (in theory, 20 tRNAs and 20 aaRSs would be sufficient to cover the cAAs), 3 initiation factors, 3 elongation factors, 2 release factors, and 12 nucleotide-modifying enzymes.⁷⁸ Despite this complexity, an expanded genetic code is ideally accommodated without interfering with the endogenous machinery and physiology, that is, it is orthogonal at all levels. A major challenge lies in identifying or generating aaRS/tRNA pairs that work orthogonally in the desired host organism. Several methods have been developed, focusing on only the aaRS, only the tRNA, or both of them, and applied selection procedures to fulfill orthogonality on multiple layers (introduced in section 1.3).

2.1.1. Milestones in aaRS Evolution

In 1997, the Schultz lab reported the first orthogonal aaRS/tRNA pair working in vivo, based on the E. coli glutaminyl-RS (GlnRS) evolved through directed mutagenesis and DNA shuffling.⁵⁴ In parallel, eight mutations were introduced into the cognate glutaminyl tRNA, which was shown to be used by the evolved GlnRS with an approximately 1400-fold increased selectivity compared to the endogenous glutaminyl tRNA,⁷⁹ and with the possibility of tuning GCE based on the ratio between heterologous tRNA and aaRS.⁸⁰ One year later in 1998, Furter as well as the teams of Nishikawa and Yokogawa demonstrated that the yeast phenylalanyl-RS/tRNA and the tyrosyl-RS/tRNA pairs can be applied for orthogonal GCE in E. coli.^81,82 Because archaea are more closely related to eukaryotes than to eubacteria,⁸³ their aaRS/tRNA systems represent promising targets for exploring orthogonality in E. coli. Orthogonality for the Methanocaldococcus jannaschii (Mj) tyrosyl-tRNA synthetase (TyrRS)/tRNA pair was demonstrated in E. coli, with several features of the enzyme amenable for improving GCE reactions.^24,84 These include the large enzymatic binding pocket with a high number of substrate-interacting side chains that can be modified to accept ncAAs other than tyrosine, the minor contribution of the anticodon-binding domain for tRNA recognition,⁸⁵ and the lack of an editing domain. Many aaRSs contain editing domains capable of hydrolyzing wrongly aminoacylated tRNAs.⁸⁶⁻⁸⁸ To enable the efficient charging of heterologous tRNAs with ncAAs in GCE systems, aaRS/tRNA pairs are typically selected that lack editing domains or have been evolved and engineered to diminish editing reactivity. Editing reactions are classified into pre-transfer⁸⁹ and post-transfer editing,^90,91 occurring either before or after a “mis-activated” amino acid is loaded onto the 3′ end of a tRNA molecule.⁸⁶ In addition to the cis-editing reactions mediated by aaRS editing domains, most organisms also express factors whose functions in catalyzing the hydrolysis of mischarged aminoacyl-tRNAs resemble editing but in trans and without aminoacylation activity.^92,93 For example, the PrdX protein in the gram-positive bacterium Clostridium sticklandii acts as a homolog of the prolyl-aaRS editing domain, cleaving alanine-loaded tRNA^Pro molecules.⁹³

Further aaRS/tRNA pairs with orthogonality in E. coli include glutamyl,⁹⁴ leucyl,⁹⁵ and prolyl⁹⁶ aaRS enzymes of archaeal origins. The M. barkeri or M. mazei PylRS/tRNA pair⁹⁷ and its variants, most notably the PylRS Y306A Y384F variant,⁹⁸ are now among the most frequently used pairs for GCE in vivo due to their amenability to engineering and evolution and their broad substrate spectrum (discussed in detail in section 2.2.1).⁹⁹ An elegant strategy for establishing orthogonality of E. coli aaRS/tRNA systems in E. coli (similar host and component origin) has been achieved by first functionally replacing the endogenous aaRS/tRNA pair with a compatible pair of other origin, followed by the reintroduction of the now-liberated aaRS/tRNA pair for directed evolution and GCE with different ncAAs.^100−102 The components of an aaRS/tRNA pair can also be derived from distinct origins, as demonstrated by the M. mazei SerRS working orthogonally in E. coli in combination with the tRNA^Ser from either of the two thermophilic marine archea Archaeoglobus fulgidus or Pyrococcus horikoshii.¹⁰³ This includes chimeric synthetases–fusions of domains originating from different aaRSs. In 2020, Ding et al. combined key orthogonal components of the PylRS enzyme with components of other aaRS enzymes, generating chimeric histidine, phenylalanine, and alanine GCE systems.¹⁰⁴

2.1.2. Evolution Strategies

Directed evolution is a strategy involving the iterative diversification and functional screening of gene variants and is central to most evolution approaches with aaRS/tRNA pairs (Figure 3).¹⁰⁵ This typically involves the generation of a library (of aaRS or tRNA variants), followed by multiple rounds of selection for desired properties, such as high enzymatic efficiency and orthogonality.^106,107 There are numerous methods for generating gene variant libraries. Traditionally created by random gene mutagenesis (using UV light, chemical mutagens, or error-prone polymerase chain reaction (PCR))^108−110 and DNA shuffling,⁵⁴ contemporary variant libraries are mostly designed through targeted mutations based on a wealth of prior knowledge from databases and publications, often in a systematic and large-scale manner. Saturation mutagenesis can further be used to systematically create variants of particular gene regions that encode a protein domain or protein–protein interaction site. Saturation mutagenesis variant libraries are synthesized as DNA oligos and introduced into a vector construct for screening of a desired phenotype. Saturation mutagenesis represents one of the most frequently used techniques to screen aaRS variants for their ability to aminoacylate tRNAs with newly developed ncAAs. Critical regions of a given aaRS, such as the sites at which the tRNA and the (noncanonical) amino acid bind, can thus be targeted for introducing defined variants. This yields a library of plasmids comprising aaRS variants, with key positions changed to encode each of the 20 cAA residues, including combinations of multiple positions being varied. Running this approach in a nondegenerate way, with only a single codon for each of the 20 cAAs being designed (ideally one with high codon usage in the host) instead of including all 64 codons of the degenerate genetic code, can minimize library size and yet maximize the count of unique variants.¹¹¹ However, even by using this trick, variant libraries can easily be pushed to their limits due to the considerable combinatorial space. For example, the substitution of six codons with combinations of (not all 61 possible codons but limited to only) 20 codons for encoding the 20 cAAs results in 20⁶ (64 million) variants. This six-codon combinatorial space is feasible, as the very large variant libraries used for saturation mutagenesis studies to-date comprise around 10⁹ variants, and rarely exceeding much higher. The combinatorial space of substituting seven codons is challenging in a single library (20⁷ = 1.2 billion variants), and eight codons would require at least 26 large libraries (20⁷ = 25.6 billion variants). Limits in library size are given by transformation efficiency and, more importantly, by the number of cells that can be cultured and passed through selection. The latter depends on the optical density of the starting culture, the number of generations during the selection process, and the volume of growth media. Although the process can be scaled up by multiple transformations, by increasing media volume, and by a continuous-fed batch process, for most academic studies, the rather small gain in variant number would not justify the greater effort and cost. Saturation mutagenesis is therefore particularly useful if prior knowledge exists on certain residues that can be targeted for diversification, for example, those that are key in catalysis or substrate binding. Likewise, rational protein engineering uses X-ray structures obtained from crystallized aaRSs or aaRS–tRNA complexes, as well as computational homology models, to identify amino acid residues with critical functions in binding a particular substrate or catalysis.¹¹² This knowledge can then be used to make informed mutations for precisely tuning enzyme specificity, catalytic efficiency, or thermal stability.^64,113,114

Figure 3.

Directed evolution of orthogonal aaRS/tRNA pairs.

The aaRS variant library is then subjected to multiple positive- and negative-selection rounds, with the ncAA being typically supplied for positive selection and excluded for negative selection.^107,115 In E. coli, positive selection is typically performed using systems in which an antibiotic resistance gene with a stop codon confers survival only in cells expressing an efficiently recoding aaRS/tRNA pair. The positive selection marker initially used, β-lactamase, has been replaced with bacteriostatic chloramphenicol acetyltransferase selection due to its higher selection pressure.¹¹⁵ Instead of an antibiotic resistance gene, positive selection can also be achieved by using a fluorescent reporter gene with a stop codon, which is only recoded and fluorescent in cells with working aaRS/tRNA pairs.⁸⁴ Positive selection thus yields a variant population that is enriched for efficient pairs through multiple generations of growth (antibiotic resistance marker) or fluorescence-activated cell sorting (FACS).¹¹⁶ The population is then exposed to negative selection in the absence of the respective ncAA, often by using a stop-codon-harboring gene that encodes for a toxic gene product, such as barnase or uracil phosphoribosyltransferase. Variants capable of misincorporating cAAs at the stop codon site will produce the lethal toxin. Negative selection leads to the depletion of variants that are not specific to the desired ncAA from the population. Multiple iterations of this postive and negative selection process optimize the pair for use in the host organism. This approach was demonstrated by evolving the Mj TyrRS/tRNA pair toward incorporating O-methyl-L-tyrosine in E. coli.²⁴

2.1.3. Phage-Based, RNA-Guided, and In Vitro Evolution Technologies

Phage display can be used to direct particular antibodies toward an epitope generated via GCE.^115,117,118 The beauty of this system is that a stop codon can be introduced into a validated antibody epitope, fused to the N terminus of a phage coat protein, essentially making phage formation dependent on stop codon suppression to produce the coat protein. E. coli is first transformed with a phagemid comprising the aaRS variant library and tRNA and subsequently infected with phages. Working aaRSs enable the reproduction of phages, each presenting the coat-protein-fused antibody epitope and packing the phagemid that carries the enabling aaRS gene. Antibody affinity purification allowed specific extraction of epitope-presenting phages.¹¹⁵ The in vitro selection procedure avoids the false-positive hits of in vivo selections that can arise through the growth effects of strains that acquired secondary mutations. When incorporating ncAAs into proteins that are presented on the outer plasma membrane, cell display can be used for selection, either based on antibodies as outlined for phage display or based on bioorthogonal conjugation (e.g., for labeling with fluorophores and subsequent selection via FACS).

Phage-assisted continuous evolution (PACE)¹¹⁹ is another elegant method for random mutagenesis of a defined gene that can be used to evolve aaRSs. PACE is an inducible and highly effective strategy for mutagenesis in bacteria, and mutation rates of 322,000-fold versus basal levels have been observed, thus exceeding those of conventional treatments with ultraviolet light or chemical mutagens.¹²⁰ The approach exploits an engineered bacteriophage in which the gene for protein III (pIII), essential for infectivity, has been removed. An amber-codon-containing copy of the pIII gene is expressed from a plasmid in E. coli, coupling the propagation of the phage to functional GCE of its host, as performed by a co-expressed aaRS/tRNA pair. PACE has been used to generate variants of PylRS with up to 45-fold enhanced GCE efficiency compared to its wildtype ancestor PylRS.¹²¹ A modified version, referred to as phage-assisted noncontinuous evolution (PANCE) applies PACE in a serial flask dilution.¹²² PACE has been used to enhance quadruplet encoding (section 2.8) by a factor of up to 80.¹²³ Further strategies for in vitro directed evolution are established, such as making use of compartmentalization in liposomes to accomplish high-throughput variant library screening.¹²⁴

The Chin lab recently introduced a tRNA display strategy to select orthogonal aaRSs capable of acylating their cognate tRNAs with ncAAs that are poor ribosomal substrates. They demonstrated the application of tRNA display for the site-specific encoding of a range of β-amino acids, α,α-disubstituted amino acids, and β-hydroxy acids in E. coli.¹²⁵ A different approach with similar application was developed by the Chatterjee lab,¹²⁶ relying on unique sequence barcodes attached to the tRNA. If an active aaRS mutant acylates its cognate tRNA, it confers protection to the tRNA-linked sequencing barcode during an oxidative treatment that damages non-charged tRNAs at their 3′ end. Subsequent sequencing reads out the barcodes of aminoacylated tRNAs and their respective aaRS mutants. The Chin and Chatterjee groups used these approaches alongside engineered ribosomes to acylate even exotic monomers and doing so independent of the resulting acylated tRNA being accepted by the ribosome (since the direct barcode sequencing does not require translation as a readout).

Recent progress in the development of RNA-guided clustered regularly interspaced short palindromic repeats (CRISPR) and base-editing systems has led to the development of multiple methods for generating aaRS variant libraries. These include multiplexed precision genome editing,¹²⁷ base editing, and prime editing²⁸⁸ which allow defined mutations to be engineered. With the use of guide RNA libraries, these methods can be used to generate cell populations with single cells harboring a single target mutation. These cell populations can then be screened (in the pooled format) for desired phenotypes, such as the expression of a highly efficient aaRS/tRNA pair. Notably, a single guide RNA library can be designed to target both the aaRS and tRNA genes, although screening combinations with more than a single edit will be challenging in terms of the method for dual editing and the combinatorial space. Targeted DNA hypermutation can be used to mutate a defined gene region in a random manner by using an activation-induced cytidine deaminase coupled to Cas9¹²⁸ or T7 polymerase.¹²⁹ Although successful for targeted mutagenesis, these technologies have not yet been applied to generate aaRS or tRNA libraries for GCE applications.

2.2. Engineering the aaRS Enzyme Beyond the ncAA Recognition Site

In addition to the evolution workflows discussed in section 2.1, the aaRS enzyme can be further engineered to increase its affinity for a specific ncAA and tRNA (enhancing tRNA charging efficiency), as well as to decrease cross-reactivity with endogenous amino acids and host tRNAs (enhancing orthogonality; Figure 4).^130,131

Figure 4.

Engineering aaRS properties. Schematic of the aaRS–tRNA complex, indicating regions of special interest for mutations and engineering. These include residues in the aaRS that interact with ncAAs, tRNAs (identity elements, especially those that bind anticodons), the deletion of editing domains, the deletion of N- or C-terminal parts, and replacement with parts from an aaRS of different origin to yield a chimeric aaRS.

AaRSs can be categorized as class I or class II synthetases based on mutually exclusive sequence motifs.¹³² These sequence motifs primarily determine the conformation of the so-called activation or ATP-binding domain¹³³ and can further contribute to interactions with the identity elements of their cognate tRNA.¹³⁴ Because cAAs have limited diversity in their side-chain chemistries and structures, their recognition by aaRSs is imperfect, with some aaRSs accepting noncognate amino acids more readily than others. This results in mistranslation events that are generally rare, affecting only a small fraction of the proteome, and are tolerated by the respective organisms.^86,135−138 For example, E. coli has been shown to survive up to 10% mistranslation of some codons by increasing the expression of chaperones to alleviate proteotoxic effects.¹³⁹ The pathogenic yeast Candida albicans exploits misaminoacylation events of the CUG codon to encode leucine instead of serine for diversifying its cell-surface morphology to modulate flocculation and host interaction phenotypes.¹⁴⁰ The ambiguous aminoacylation of aaRSs can also be harnessed for technical applications. This has been demonstrated for the E. coli cysteinyl-RS, which is unable to discriminate between cysteine and selenomethionine. The stochastic incorporation of selenomethionine into the E. coli proteome via its endogenous cysteinyl-RS has been exploited for X-ray crystallography studies.^141−144

The site-specific incorporation of defined ncAAs generally necessitates the development of aaRSs that exclusively recognize the ncAA and disregard the cAAs. Notably, changing the tRNA anticodon to encode the amber stop codon can also lead to increased misaminoacylation of tRNAs and potentially decrease GCE efficiency.¹⁴⁵ As proposed by Fersht in the so-called double sieve model for aaRS enzymes, individual sites of an aaRS can promiscuously recognize amino acids with related structural features.¹⁴⁶ The interplay of two or more of these binding sites with distinct recognition features sufficiently enhances the fidelity of translation to ensure viability, and the editing domains of some aaRSs⁸⁶⁻⁸⁸ add another layer to enable the correct aminoacylation of tRNAs.⁸⁹ Whereas the promiscuity of aaRSs can be an advantage or a disadvantage for GCE, multiple tRNA or amino acid binding sites and editing domains generally pose a challenge for aaRS engineering. Frequently used orthogonal aaRS/tRNA pairs also include the E. coli TyrRS, which has been evolved to accept various ncAA chemistries as substrates for tRNA charging.⁸⁴ An elegant strategy for establishing orthogonality in a given aaRS/tRNA pair is the substitution of regions of the E. coli TyrRS with corresponding sequences of archaeal origin, essentially bypassing the evolution process.^{100−102,147}

2.2.1. Showcasing aaRS Engineering of the PylRS/tRNA Pair

The archaeal PylRS/tRNA pair, which evolved naturally for decoding the amber CUA codon,⁹⁷ found broad applications for GCE. On the basis of the polyspecific nature of PylRS, variants were evolved and engineered toward accepting a wide range of ncAAs as substrates, including many lysine derivatives, as well as a variety of amino acids recognized via α side-chains: α-hydroxyacids, non-α-aminocarboxylic acids, N^α-methylamino acids, and D-amino acids.^98,148−150 The PylRS(Y306A/Y384F) variant derived from Methanosarcina mazei was developed to accept a range of bulky lysine derivatives.⁹⁸ The crystal structure of the enzyme–substrate complex revealed its large binding pocket, which was found to have a rather low affinity for any specific residue side-chains and the N-terminal α-amino group.^151−154 In contrast to other aaRSs, PylRS has a unique N-terminal domain important for the recognition of its cognate tRNA and does not bind the tRNA anticodon loop, which can thus be modified for GCE purposes.^122,155 Elucidating the structures of 14 catalytic PylSc domains complexed with different ncAAs showed their different modes of substrate recognition.⁹⁹ Other mutants have been evolved for efficiently incorporating noncanonical derivatives of aromatic amino acids, such as phenylalanine derivatives (accepted as substrate by PylRS(N346A/C348A)),^156−158 histidine derivatives (accepted as substrate by PylHRS(L270I/Y271F/L274G/C313F/Y349F),¹⁵⁹ as well as α-hydroxy ncAAs.¹⁶⁰ Whereas the broad PylRS substrate spectrum is an ideal starting point for incorporating ncAAs with diverse chemistries, the evolved PylRS variants are frequently cross-reactive with each other, complicating their simultaneous use. Mutually orthogonal PylRS/tRNA pairs have been developed on the basis of enzymes truncated in their N-terminal domains that contribute to tRNA recognition.^161,162 The most recent efforts to generate orthogonal PylRS/tRNA pairs are based on combining N-terminal deletion with sequence similarity.¹⁶³ The Basic Local Alignment Search Tool (BLAST) algorithm was used to mine genomic PylRS sequences by their similarity to a reference PylRS. In total, 351 PylRS sequences were retrieved, converted to protein sequences, and hierarchically clustered by similarity. In most cases, the cognate tRNA sequences could also be identified and were clustered in a similar way to select PylRS/tRNA pairs that differed from each other and, thus, were likely orthogonal to each other. A subset of selected pairs was tested for orthogonality to each other, and to the endogenous pairs in E. coli. Combined with data from a previous study,¹⁶² the workflow ultimately yielded 924, 1324, 128, and 8 PylRS/tRNA pairs with mutual orthogonality within combinations of two, three, four, and five pairs, respectively, encoding the ncAA N⁶-[(allyloxy)carbonyl]-L-lysine (AllocK).¹⁶³ Encoding multiple different ncAAs will, in addition to mutual orthogonality, further require aaRS/tRNA pairs with different substrate specificity.

2.2.2. aaRS/tRNA Pairs for the Simultaneous Encoding of Multiple ncAAs

To make use of multiple orthogonal aaRS/tRNA pairs in E. coli, two stop codons in the same transcript (e.g., amber and ochre) have been efficiently recoded with multiple different ncAAs.¹⁶⁴ Chatterjee et al. genetically encoded both p-azidophenylalanine (pAzF) and p-propargyloxyphenylalanine (pPa) in E. coli to simultaneously incorporate them into a single protein.¹⁶⁵ To do so, they used two orthogonal aaRS/tRNA pairs: the Mj TyrRS/tRNA_CUA pair for pAzF and the Methanococcus maripaludis PheRS/tRNA_CUA pair for pPa. By dual encoding these ncAAs, the authors achieved the site-specific conjugation of two distinct chemical moieties to evaluate their influence on protein function and stability. This approach still results in the recoding of endogenous stop codons in the transcriptome. In addition, only three stop codons are available for recoding, and heterologously expressed tRNAs that recognise them compete with release factors. Several elegant approaches tackle these challenges, such as the removal of release factors, quadruplet codons, engineered orthogonal ribosome/orthogonal mRNA pairs, artificial base pairs, as well as reprogrammed and synthetic genomes and (all discussed in the following sections). Based on the combination of M. mazei (or M. barkeri) PylRS together with the TyrRS, LeuRS, and TrpRS from E. coli, the incorporation of up to three different ncAAs into proteins of E. coli has been achieved.^96,166

2.3. Engineering the tRNA

The tRNA is a crucial part of the GCE machinery. tRNAs naturally harbour so-called identity elements, which are recognized by their cognate aaRS.^167,168 Often found in the anticodon loop and acceptor stem, these identity elements can be mutated to prevent their aminoacylation by aaRSs in the respective host system. If the anticodon of the tRNA itself represents an identity element recognized by its cognate aaRS, mutation of the anticodon for the suppression of a stop codon can impair the recognition and aminoacylation of the tRNA by the aaRS. Whereas mutated aaRSs might still recognize the modified tRNA, most GCE systems rely on suppressor tRNAs with anticodons that do not partially or fullyrepresent an identity element.¹⁶⁹ A prerequisite for the aminoacylation of tRNAs is generally their expression in sufficient amounts, their processing and maturation, as well as the energy-consuming activation of their respective (noncanonical) amino acid. Therefore, the biogenesis and processing of tRNAs is of particular interest for engineering GCE systems.¹⁷⁰

2.3.1. Processing and Modification of tRNAs

An essential RNA-processing step in prokaryotes is the enzymatic addition of the trinucleotide CCA at the 3′ terminus of tRNAs. When transferring tRNA genes from archaea or eukaryotes to bacteria, such as for transferring the archaeal pyrrolysine tRNA, the CCA sequence generally needs to be added at the 3′ end of the tRNA gene to mimic bacterial tRNAs. Vice versa, the transfer of bacterial tRNAs into other kingdoms typically requires the removal of 3′-CCA sequences.^169,171 The ncAA activation step produces an amino acid adenylate, an intermediate which is then charged onto the CCA 3′ end of the tRNA molecule by the respective aaRSs. The orthogonality of tRNAs can be enhanced by preventing their recognition by host aaRSs. This can, in principle, be achieved by modifying all regions of the tRNA, including the anticodon, the acceptor stem, and the variable loop. Mutations in the anticodon are used to reassign stop codons, such as the amber UAG codon, to encode a defined ncAA, and mutations in the anticodon loop can increase or decrease decoding efficiency.¹⁷² Other aspects of the tRNA can be altered, including their post-transcriptional modifications to increase their stability and other properties.

Generally, tRNAs are heavily post-transcriptionally modified and processed, which is often essential for their functionality and thus poses challenges for the transfer of aaRS/tRNA pairs into new host environments.^169,170,173 For example, natural suppressor tRNAs in E. coli and S. cerevisiae require specific adenosines to be isopentenyl modified to be fully active.^57,174,175 In addition, incomplete modification of tRNAs can result in their targeted degradation by the nuclear surveillance turnover pathway.^176,177 The processing of tRNAs can involve the ligation of two tRNAs that originate from distinct genes into a final mature tRNA¹⁷⁸ or the splicing of intron sequences in eukaryotes. The processing and modification of tRNAs depends on the species. Such components were identified in E. coli by assaying the incorporation of the ncAA O-phosphoserine into a GFP reporter in 42 strains that harbored deletions of tRNA-modification genes.^179,180 Shutting down these tRNA modification components can further enhance GCE by altering tRNA affinity for specific aaRSs, elongation factors, or the ribosome.

2.3.2. Interactions with EF-Tu

Within an organism, or a certain genetic code system, tRNAs have a typical size and structural properties because they must be compatible with the respective RNA processing, base modification, enzymatic CCA-addition, elongation factor (EF) Tu, and ribosomal machineries.^5,77,180 GCE systems can therefore benefit from modifications to these components. EF-Tu binds tRNAs and promotes their ribosomal recognition. EF-Tu also enables proofreading of misaminoacylated tRNAs,^181−183 and EF-G, which catalyzes RNA translocation at the ribosome and suppresses frameshifting.¹⁸⁴ EF-Tu restricts GCE by limiting the delivery of ncAA-charged tRNAs to the ribosomal machinery. Engineering EF-Tu for abolishing its role in the quality control of tRNA aminoacylation was key to enabling its productive interaction with O-phosphoserine-charged tRNAs for GCE in E. coli.¹⁸⁵ Another study probed a library of M. jannaschii tyrosyl-tRNA_CUA, which was designed with diverse variants of a region proposed to interact with EF-Tu, for in vitro evolution experiments. These yielded 14 tRNA sequences that enhanced the yield of ncAA-equipped protein up to 25-fold using p-benzoylphenylalanine.¹¹⁶ In addition to modifying EF-Tu for ncAA compatibility, its substrate specificity has been extended by targeting specific residues of EF-Tu that have been diverged from its homologous selenocysteine-specific elongation factor selB.¹⁸⁶ Selenocysteine can be readily encoded in bacteria (and eukaryotes) that express the pathway components for its biogenesis and encoding. tRNASec variants,^187−189 as well as EF-Tu,^190,191 have been engineered to enable selenocysteine incorporation independently of the SECIS element, which is generally found close to the UGA codon as part of the coding sequence in bacterial selenoproteins.

2.3.3. Other Approaches for tRNA Engineering and Evolution

The Söll lab developed orthogonal initiator tRNAs (itRNAs) capable of initiating protein synthesis with ncAAs in E. coli. Through transplantation of identity elements into the formyl-methionine tRNA, they incorporated different aromatic ncAAs in response to the amber stop codon UAG with the Mj TyrRS.¹⁹² They further developed this system to encode ncAAs with the tyrosine codon UAU.¹⁹³ Importantly, this recoding of the initiation codon does not affect regular tyrosine incorporation at following UAU triplets.¹⁹³

Metagenome and meta-transcriptome analyses in search of selenocysteine tRNA species identified a group of tRNAs with unusual cloverleaf structures that harbor a long acceptor and a short T-arm. Whereas most canonical tRNAs have a branch configuration with a 7:5 acceptor stem/T-stem ratio, the newly identified tRNAs feature 8:4 or 9:3 configurations. Their overexpression in E. coli with different anticodons and identity elements resulted in the efficient recoding of serine, alanine, and cysteine in response to sense or stop codons.¹⁹⁴

Combining an optimized tRNA with expression of the same aaRS from both a constitutive and an inducible promoter was shown to increase intracellular protein concentration and further increase GCE efficiency.¹⁹⁵ However, accurate and efficient GCE depends on multiple components and processes that are in a dynamic interplay. This is illustrated by the finding that high expression levels of heterologous tRNAs can also decrease GCE efficiency in E. coli due to being affected by native RNA-modifying enzymes to a much higher extent than is observed at low expression.¹⁹⁶

Expanding on the screening of tRNA variant libraries for orthogonal tRNA adaptation,^121,165,197 the Chin group recently introduced a computational workflow that makes use of sequencing data for identifying orthogonal tRNA candidate sequences, combined with a fast and scalable evaluation pipeline to assess their aminoacylation status in E. coli.¹⁹⁸ This tRNA-extension strategy has been used to analyze more than 2.7 million sequences from a dedicated tRNA database with mapped identity elements^167,168 that are key for conferring selective recognition of the tRNA by its cognate aaRS.^{168,199−201} In total, 243 candidates were prioritized and expressed in E. coli, yielding 71 tRNAs of different isoacceptor classes that do not cross-react with endogenous aaRS/tRNA systems. For 23 of these, the corresponding aaRS enzymes were found to be functional, with five of them being also orthogonal to the host machinery. To further evolve also new substrate specificities for two aaRS enzymes, a matrix of 8 aaRS × 8 cognate tRNA combinations was profiled, revealing five novel and confirming three established pairs to be orthogonal in the bacterial host.¹⁹⁸

2.4. Efficiency Quantification of aaRS/tRNA Pairs

The efficiency of aaRS/tRNA pairs can be quantified with a wide variety of techniques. Mass spectrometry is a sensitive method that enables one to determine the ratio of modified to unmodified target proteins or measure peptides with incorporated ncAAs.^24,202 The yield of the purified target protein can also be quantified using Bradford or bicinchoninic acid assays, or semi-quantitatively by western blotting. If the ncAA is incorporated into an enzyme, its relative abundance could further be indirectly measured via its enzymatic activity in an established substrate conversion assay, although such quantification is rather imprecise. Ribosome profiling provides the sequence identity of ribosome-protected mRNA fragments and, as proposed by Liu and Schultz,⁶⁷ can be used to quantify the translation efficiency of a target gene by comparing ribosome occupancy at specific codons. Multiple studies have since used this technique to measure the efficiency of aaRS/tRNA pairs by directly quantifying ribosome binding events at sequences before and after a premature stop codon that has been targeted for suppression, as well as conditions with and without suppressor tRNA.^203−205 Finally, GCE efficiency can be measured with luminescence and fluorescence reporter genes that harbor one or multiple stop codons targeted for suppression. These reporters serve as a proxy for quantifying genetic recoding in vivo. The corresponding reporter protein is only made if the intrinsic stop codons of its mRNA are interpreted as sense codons to encode a defined ncAA (or by readthrough of the stop codon due to cAA misincorporation, which should always be controlled for), and positively selected using FACS.¹¹⁶ If an ncAA with a bioorthogonal handle is applied (section 4.3), click chemistry can be used to site-specifically label the modified protein with an additional fluorophore. Whereas the fluorescence signal can be measured with single-cell resolution using fluorescence microscopy and flow cytometry, and on the population level with fluorescence protein gels and well plate assays, luminescence is generally only assayed in a well plate format.

Unfortunately, there are currently no standardized assays for quantifying aaRS/tRNA pair activity and cross-reactivity in cells. Laboratories make use of their own workflows, limiting cross-study comparisons. As a minimal requirement for comparing values from different studies, we recommend including a simple set of three control and reference pairs, independent of the performed assay:

Control for ncAA Specificity: The newly developed aaRS/tRNA pair should be assayed in the presence of the ncAA and without it to evaluate cross-reactivity with cAAs.

Control for Orthogonality Regarding the tRNA and aaRS Components: The newly developed pair should be assayed in the presence of the respective ncAA with and without tRNA (mock tRNA) and with and without aaRS (mock aaRS). This allows for quantifying potential cross-reactivity with endogenous tRNAs and aaRS enzymes of the host.

Reference Measurement of Established Standards: In addition to assaying the newly developed aaRS/tRNA pair, at least one established aaRS/tRNA pair that is used frequently (ideally multiple) should be included in the measurement: (a) with the addition of the ncAA it was developed for, (b) with the addition of the ncAA used for the newly developed pair, and (c) without ncAA supplementation. This will enable results to be compared between studies that featured similar aaRS/tRNA pairs as reference standards, even if the assays are different. These controls allow the relative efficiency and specificity to be quantified for any given aaRS/tRNA pair with respect to one or, ideally, multiple widely established aaRS/tRNA pairs. Although further improvements based on the particular host organism used might make sense, the specific assay and the application, the outlined control, and reference samples represent a minimal set of criteria by which to standardize experimental quality and to ensure a level of comparability between similarly abiding studies.

2.5. Side Effects of GCE Systems on the Host Organism

We are just beginning to understand how GCE systems interact with the host in which they are established. From a bioengineering perspective, the host organism is often viewed as a chassis for the GCE reaction. The Rinehart lab recently applied fitness profiling and proteomics to systematically characterize the effects of an orthogonal phosphoserine GCE system on host E. coli, revealing global effects on transcription, proteome dysregulation, and changes in resource allocation, which, in turn, affect GCE performance.²⁰⁶ They revealed that whereas orthogonality of a given aaRS/tRNA pair is generally regarded as a static feature in a respective host, dynamic changes of the intracellular milieu, for example, in response to environmental conditions, can affect orthogonality. For example, the overexpression of tRNAs can lead to their misaminoacylation by endogenous aaRSs and thus compromise translation fidelity and host-cell fitness. Global changes in tRNA abundance, even with regard to only endogenous tRNAs, can give rise to misaminoacylated tRNAs. These can lead to ambiguous interpretations of the genetic code.^80,207

The overexpression of GCE components and their action increase demand for energy and cellular resources to the extent that elicits a stress response²⁰⁶ and reduces growth rates in bacteria.²⁰⁸ Strategies will be needed in future to control the production and activity of the GCE machinery precisely and shield them from interacting with endogenous components; this would minimize perturbations to cell physiology. In E. coli, GCE has been decoupled from growth, which also increased the yield of ncAA-modified proteins. This was achieved by routing unnatural protein expression to the T7 polymerase promoter and the arabinose-induced expression of the general RNA polymerase inhibitor gp2, originating from the T7 bacteriophage.²⁰⁸ Induction of gp2 just before the bacterial population plateaus in terms of cell density leads to the repression of endogenous transcription. Resources are thus reallocated mainly from growth to performing GCE. The yield of an enhanced GFP(Y40Azk) was increased from 16.9 to 21.0 rfu/mg (relative fluorescence units normalized by dry mass).²⁰⁸ We anticipate that efforts to better understand how different GCE approaches affect cellular physiology will be key to optimize the integration of the GCE reaction within diverse host cells.

2.6. Enabling ncAA Biogenesis for the Generation of “All-In Vivo GCE Systems”

Although most GCE systems in living cells or organisms require an exogenous supply or “feeding” of ncAAs, metabolic engineering has also been used to render organisms capable of producing the ncAA. The Schultz laboratory equipped E. coli with metabolic pathway genes from Streptomyces venezuelae and demonstrated that the strain successfully biosynthesized and site-specifically installed the ncAA p-aminophenylalanine into proteins.²⁰⁹ A similar system based on the biosynthesis of p-aminophenylalanine and its installation via GCE was later shown to efficiently produce fluorophore-conjugated antibody fragments.²¹⁰ Likewise, E. coli has been engineered to produce and incorporate 5-hydroxytryptophan,²¹¹ dihydroxy-L-phenylalanine,^212,213 different α-keto acids,²¹⁴ cysteine derivatives,²¹⁵O-methyltyrosine,²¹⁶ and some ncAAs resembling PTMs. The latter was demonstrated for phosphothreonine, which was produced from threonine by introducing the PduX kinase from Salmonella enterica in the E. coli host.²¹⁷ To bypass ncAA biogenesis, propeptides can also be supplied to bacteria for GCE, as shown for O-phosphotyrosine and its nonhydrolyzable derivative 4-phosphonomethyl-l-phenylalanine, to resemble or mimic phosphorylated residues (see section 4.1.1). Although integrating ncAA biosynthesis into the living organism can yield intracellular ncAA concentrations that cannot be achieved by simply supplementing the ncAA, these autonomous systems can also offer further advantages. This is illustrated by the biosynthesis and utilization of 5-hydroxytryptophan, which has been exploited to quantify oxidative stress over time; this could not be achieved by exogenously supplying the ncAA.²¹¹E. coli strains that require specific ncAAs for survival, for example, by means of antibiotic pressure, can also be used to further evolve enzymes that produce ncAAs from supplied precursors.²¹⁸ Heterologous expression of a carbamoylase from Sinorhizobium meliloti in E. coli grown in the presence of the synthetic ncAA precursor N-carbamoyl-L-3nY yielded enzyme variants with increased efficiency for converting this precusor into the ncAAs L-3-nitrotyrosine and L-3-iodotyrosine.

2.7. Releasing Release Factors

Stop codons are naturally recognized by release factors, which compete with any tRNA introduced for stop codon reassignment, and reduce the efficiency of ncAA incorporation. As a consequence, peptide chains are terminated, and truncated polymers accumulate in the cell, which can have toxic effects and hamper downstream applications. In E. coli, there are two release factors: RF1, terminating UAA and UAG, and RF2, terminating UAA and UGA codons.²¹⁹ Whereas RF deletions are lethal,²²⁰ temperature-sensitive RF1 variants have been used to transiently boost amber codon suppression.²²¹ However, this influences the suppression of UAG codons in the whole transcriptome, leading to excessive stop codon readthrough, potentially harming the integrity and physiology of the cell. The knock-out of RF1 in E. coli was reported viable if amber suppressor tRNAs were expressed along with seven essential E. coli genes, all of which terminate with TAA instead of the TAG codon.²²² Another strategy that enabled RF1 deletion was the removal of autoregulatory sequence elements of the RF2 gene to elevate its expression and introduction of the T246A mutation to increase UAA codon release.²²³ The approach was realized in an E. coli strain with a minimized genome with around 700 non-essential genes deleted and based on first introducing a plasmid for RF1 expression to allow removal of its chromosomal open reading frame (ORF), followed by the fixing of RF2 sequences and plasmid clearance. The resulting ΔprfA JX3.0 strain enables amber suppression without causing termination and without truncated side products. This allows true codon reassignment, as opposed to the sole recoding, by encoding the ncAA unambiguously through: a) the removal of release factors for stop codon reassignment and b) the removal of endogenous tRNAs that would decode a recoded sense codon for sense reassignment. Several other strategies have been developed that also allow RF1 deletion in genetically reprogrammed (see section 2.11)^224,225 and synthetic E. coli (see section 2.13)²²⁶ or that bypass RF1 dependence, such as with orthogonal ribosomes (see section 2.9).

2.8. Quadruplet Codons

Quadruplet codons represent four-base ensembles and expand the genetic code capacity to 4⁴ (256) codons. In analogy to naturally occurring frameshift suppression,^227,228 the anticodons of tRNAs can be artificially elongated to match quadruplet codons for encoding ncAAs.^229−235 Quadruplet codons can also be combined with stop codon suppression for dual ncAA incorporation into the same protein.²³⁶ To date, quadruplet codons have been used to extend the genetic code of E. coli to 68 codons and the recoding of up to four individual four-base triplets within a single transcript for incorporating four distinct cAAs¹²³ or ncAAs²³⁷ into the protein product. In E. coli, there is suggestion that a handful of mutations can turn any tRNA into an at least weakly efficient quadruplet anticodon tRNA.²³⁸ By contrast, aaRS enzymes were frequently (12 out of 20 aaRSs tested) found to work promiscuously, inefficiently, or not at all. Evolution of a four-base genetic code would have therefore very likely been possible and might not have occurred simply because a three-letter code is already sufficient, less complex (saving energy with fewer reactions and components needed), and thus more precise and robust.²³⁸ The evolution of an aaRS capable of efficiently loading quadruplet tRNAs is a prerequisite to recoding multiple quadruplet codons simultaneously.¹⁶²

Although tRNAs with extended anticodons recognize quadruplet codons, they can still retain ambiguous affinity for their respectively matching three-base codon, giving rise to frameshift products or in the case of three-base+1 stop codons, competing with release factors and resulting in truncated side products. Such extended four-base codons, starting with three-letter stop codons plus one extended base (with sequences UAGN, UAAN, and UGAN), can thus be released from the competition by deleting the respective release factor. For example, the deletion of RF1, which terminates peptide chain formation at amber stop codons, has been deleted from the E. coli genome to boost ncAA incorporation by UAGN quadruplet codons.²³⁹ Despite the powerful applications, the natural translation of quadruplet codons remains rather inefficient, with cells accumulating potentially toxic side products due to frameshift and impinging on the cellular proteome. To further enhance the efficiency and specificity of quadruplet decoding and to diminish release factor competition, they can be combined with engineered orthogonal ribosomes and matching orthogonal mRNAs (section 2.9).

2.9. Ribosome Engineering

Numerous strategies have been reported for engineering ribosomes with dedicated features beneficial for GCE, which can also be combined (Figure 5). Most notably, the small ribosomal subunit has been modified in concert with alterations of the ribosome binding site in specific transcripts to yield orthogonal ribosomes in E. coli. Specialized ribosomes with engineered anti-Shine Dalgarno sequences on the 16S ribosomal RNA (rRNA) have been developed by Hui and de Boer in 1987 to precisely translate RNAs with an altered Shine Dalgarno sequence match.²⁴⁰ On the basis of these specialized ribosomes, orthogonal ribosome/orthogonal mRNA pairs were developed by Rackham and Chin in 2005.²⁴¹ Expressed together with an aaRS/tRNA pair, these enabled the encoding of ncAAs in response to quadruplet codons in E. coli, such as encoding L-homopropargylglycine by the AGGA quadruplet²⁴¹. Orthogonal ribosomes were further evolved for optimized amber suppression fidelity using the U531G, U534A 16S rRNA mutant named “ribo-X”.¹⁹⁷ Ribo-X has translation rates comparable to natural ribosomes and was shown to enhance ncAA incorporation of single- and dual-amber codons around 3- and 20-fold, respectively. The enhancement was explained by the 16S rRNA mutation diminishing release-factor interactions to either reduce or completely avoid the competition of the orthogonal tRNA_CUA with RF1.¹⁹⁷

Figure 5.

Ribosome engineering. A mutated anti-Shine–Dalgarno sequence in ribosomal RNA of engineered ribosomes in concert with a mutated Shine–Dalgarno motif within orthogonal mRNA molecules are the basis for engineered orthogonal ribosomes that work with specific orthogonal mRNAs. These can further be combined with subunit tethering/stapling and quadruplet tRNAs to minimize or eliminate release factor competition with tRNAs.

Another orthogonal ribosome variant, named “riboQ”, was shown to efficiently decode both amber and quadruplet codons in cognate orthogonal mRNAs for GCE with multiple ncAAs in E. coli, while not affecting the translation of corresponding codons within endogenous transcripts.²³⁴ Through directed evolution of both the extended anticodon stem of pyrrolysine tRNAs, as well as the PylRS enzyme, GCE efficiency of the riboQ system has been improved and applied for the dual encoding of ncAAs compatible with bioorthogonal fluorophore labeling.²⁴² The combination of riboQ with several mutually orthogonal aaRS/tRNA pairs enabled efficient site-selective GCE with more than 10 ncAAs.^161,242 The riboQ/orthogonal mRNA system has been used for the simultaneous reassignment of up to two quadruplets plus the amber codon for ncAA incorporation in E. coli.^162,234

Ribosome "tethering" or “stapling” refers to the covalent coupling of the large and small ribosomal subunits through a short RNA linker, resulting in an orthogonal ribosome–mRNA system termed “Ribo-T”.^203,204 Specifically, the 16S and 23S rRNAs, which in the native ribosome are separated by a distance of more than 170 Å, were "opened up" in their hairpin loop structures and linked together by a double stranded RNA fragment of less than 20 nucleotides in total length. This approach was challenging, as the fusion product has to retain its native interactions with ribosomal proteins and biogenesis factors for assembling functional ribosomes without being a target for degradation by RNases. Additionally, the linker must be short enough to enable the natural cis association of subunits but also long enough not to impair subunit dynamics relative to each other; these are essential for translation initiation, elongation, and for the release of the completed peptide chain.²⁰³

The groups of Jewett and Mankin further evolved Ribo-T to overcome its low activity, developed new orthogonal Ribo-T/mRNA pairs that work in parallel and independently of the native ribosome population, and demonstrated the site-specific incorporation of pAzF at up to five amber codons of a GFP reporter transcript.²⁴³ Notably, E. coli cells equipped with Ribo-T have a slower growth rate, which has been traced back to the assembly of Ribo-T, which is slower than that of native ribosomes. Ribo-T assembly intermediates accumulate in cells with fewer rRNA and ribosomal protein modifications, as well as with incompletely processed rRNA. A ribosome profiling experiment found mildly elevated Ribo-T occupancy at start codons compared to stop codons, suggesting slight impairment of translation initiation and termination but did not detect any defects in translation elongation,²⁴⁴ which was corroborated by further translation analysis.²⁴⁵ Features of orthogonal and tethered (or stapled) ribosome versions can be combined to enhance activity and distinguish them further from the native ribosome pool.²⁴⁶ The Jewett lab developed an approach for evolving these, especially distant sequences of macromolecular machineries, such as the ribosome, and applied it to Ribo-T. One approach featured a computational pipeline for variant design that considered the structural stability of three-dimensional RNA designs and yielded Ribo-T variants with improved translation activity.²⁴⁷ The engineering of GCE and translational components can further broaden the spectrum of translation-compatible ncAA structures. Although the natural translation machinery only accepts α-L-amino acid configurations, these adjustments might allow extension of in vivo GCE to include ncAAs with exotic backbone structures, such as α-D-amino acids, N-methylamino acids, and β- and γ-amino acids, some of which have already been successfully encoded within in vitro GCE reactions.^{125,248−250}

2.10. Artificial Base Pairs for De Novo Codon Generation

Artificial nucleic acid base pairs refer to entirely new base pairs within DNA and RNA. These synthetic nucleic acids do not necessarily retain the chemical features of a base or the natural structure of a nucleotide and are labeled “artificial” simply due to their potential for substituting natural base pairs as the building blocks of nucleic acids.^65,251 Novel base pairs can thus be applied to generate codons de novo (Figure 6).

Figure 6.

Artificial base pairs for de novo codon formation. Examples of artificial base pairs are illustrated below the natural base pairs dA–dT and dC–dG. The chemical structures of the deoxy-(d-)nucleotides are shown in a double strand of DNA with base pairing indicated by dashed blue lines. The iso-dC–iso-dG base pair was the first artificial base pair that enabled GCE in vitro.²⁷⁵ The d5SICS–dNAM and dTPT3–dNAM base pairs form by hydrophobic interactions and have been successfully transferred into E. coli and also used for GCE.²⁷⁰ The dZ–dP and dS–dB pairs form the basis of Hachimoji DNA, which is yet to be applied for GCE.²⁷¹

In 1989, pioneering work from Benner and colleagues demonstrated the generation of constitutional isomers of C and G: the base pair iso-C and iso-G.²⁵²In vitro, the RNA, as well as the deoxyribonucleotides of DNA (iso-dC and iso-dG) paired selectively and could be used as a template for both a DNA and an RNA polymerase: the Klenow enzyme and the bacteriophage T7 RNA polymerase, respectively.²⁵² The iso-dG–iso-dC base pair allowed a six-letter genetic code to be generated in vitro, which was also used for site-specific GCE, installing the ncAA 3-iodotyrosine within a peptide chain in vitro(253). However, iso-dC was observed to undergo slow hydrolysis to uracil, potentially leading to a fraction of the iso-dG population containing its uracil-pairing tautomer.²⁵²

One year later, the Benner group developed a second artificial base pair without this potential ambiguity and compatible with the previously mentioned polymerases.²⁵⁴ The pair consisted of a pyrimidine and purine denoted dκ and dπ, respectively. In place of dπ, the natural base deoxyxanthosine (dX) was also considered as a dκ-pairing purine analog, exhibiting the matching acceptor–donor–acceptor hydrogen-bonding pattern. Due to the risk of deoxyxanthosine undergoing a depurination reaction, π was favored. Another artificial base pair consisting of the purine dS and pyrimidine dY has been developed, proving successful for installing 3-chlorotyrosine into the protein Ras in combination with a TyrRS/tRNA pair in a cell-free E. coli transcription/translation system.²⁵⁵ In 2006, the Benner group reported the dZ–dP nucleotide pair, a derivative of the iso-dG–iso-dC pair, which is prone to neither oxidation nor epimerization and features high fidelity rates as quantified by PCR.^251,256

In addition, base pairs with binding modes different to hydrogen bonding have been generated based on hydrophobic interactions²⁵⁷ and interactions with copper²⁵⁸ and silver²⁵⁹ ions. In addition to their outstanding pairing affinity and stability, these pairs are also highly orthogonal to natural Watson–Crick base pairs. Some artificial base pairs based on hydrophobic interactions have exhibited PCR fidelity of more than 99% (the dDs–dPa pair)²⁶⁰ and were developed further to achieve more than 99.9% fidelity (dDs–dPx).²⁶¹ The Romesberg group characterized further hydrophobic base pairs, yielding dNaM–d5SICS and dNaM–dTPT3, with an amplification fidelity of more than 99.9% independently of sequence context. These base pairs were successfully transcribed by DNA polymerases of the A, B, and X families.^262,263 Because natural enzymes for their biosynthesis do not exist, almost all artificial base pairs are derived from chemically synthesized (deoxy)ribonucleotide triphosphates (dNTPs and NTPs). Progress toward the biosynthesis of synthetic nucleotides from either natural or synthetic precursors has been made by demonstrating that some synthetic nucleotides can be used as enzyme substrates, for example, by kinases to mediate their phosphorylation in vivo.^264,265 The complete biosynthesis of two matching pair components has not been reported. Establishing artificial base pairs in vivo therefore necessitates the import of the respective synthetic (d)NTPs, which can be achieved through either high plasma membrane permeability or import channels/transporters. By screening a selection of NTP transporter proteins heterogeneously expressed in E. coli, the NTT2 importer from the diatom algae Phaeodactylum tricornutum was shown to efficiently import both d5SICSTP and dNaMTP supplied in the growth medium. The two uptaken dNTPs were subsequently used to faithfully replicate a template plasmid that contained d5SICS–dNaM at a defined position in vivo, mediated through endogenous DNA polymerases.²⁶⁶ Varying the bases flanking the unnatural base pair resulted in retention rates that ranged from almost complete retention to the total loss of the base pair. This could be explained by facilitated or masked recognition of the unnatural bases by the E. coli DNA repair machinery.²⁶⁷ By modifying the sequence context in which the unnatural base pair was embedded, it was possible to generate plasmids with high retention rates.²⁶⁸ This system allowed artificial base pairs to be used for GCE, incorporating the ncAA N^ε-[(2-propynyloxy)carbonyl]-L-lysine (PrK) into super folded (sf)GFP in live E. coli.^267−269 The synthetic E. coli strain was optimized by further mutation of the NTT2 transporter, chemical modification of the bases, and correction of sequence loci that lost the unnatural base pair. The improved strain exhibited high synthetic base-pair retention, enhanced import of precursor dNTPs, and increased rate of growth.²⁶⁸ In addition, de novo codons harboring unnatural base pairs have been systematically evaluated for encoding ncAAs, yielding nine codons for which ncAA incorporation was nearly complete and demonstrating that three of these codons are orthogonal, that is, they are specifically recognized by the exogenous tRNA with the matching anticodon that has the opposite unnatural base.²⁷⁰ Remarkably, the strain simultaneously and site-specifically incorporated two ncAAs and serine in response to three different codons with artificial base pairs.²⁷⁰

One of the most advanced artificial base-pair systems based on hydrogen bonding is referred to as “Hachimoji” (Japanese for “eight characters”) nucleic acids.²⁷¹ Establishing four additional base pairs, the Hachimoji code expands the standard genetic code from 4³ (64) to 8³ (512) unique codons. While this has great potential for unprecedented GCE, major challenges still constrain the broad applicability of artificial base pairs. These must have chemical and physical properties similar to those of natural base pairs to have comparable chemical stability and to not perturb the double helix structure of DNA. Their pairing should occur in an unambiguous match to prevent interference with natural base pairs. Their corresponding nucleotides need to be either membrane permeable, transported into cells, or, ideally, produced by the cell itself. The introduction and maintenance of artificial DNA and RNA into cells or whole organisms will further require the development of a new dedicated machinery for both the transcription of Hachimoji DNA and the translation of Hachimoji RNA bases. Therefore, a prerequisite of this endeavour is the engineering of specific DNA and RNA polymerases, ribosomes, tRNAs, and aaRSs. Notably, the naturally evolved cellular systems for DNA replication, repair, and recombination might not readily work with the new base pairs and would need to be adapted to establish a fully expanded alphabet of Hachimoji or other unnatural base pairs. We look forward to new developments in this field, which at some point in the future might enable us to stably implement synthetic base pairs into the genomes and transcriptomes of living cells to expand their genetic codes. In addition to focusing solely on novel base-pairing modes, xeno nucleic acids (XNAs) with synthetically altered backbone chemistry^272,273 by replacing the phosphate and/or (deoxy)ribose in DNA and RNA could be used in combination with new pairing modes. This strategy could yield fully orthogonal systems for replication, transcription, and translation of synthetic nucleotide sequences, to increase their stability, and to add diverse chemical functionalities to nucleotides for enabling further technical or biological approaches. However, establishing XNAs in living cells²⁷⁴ will require new types of cellular machinery to be developed for their replication, transcription, and translation.

2.11. Genome Reprogramming for Codon Substitution

Genome recoding has become a powerful method for systematically replacing redundant sense and/or stop codons within an organism’s entire genome to free up codons. The released codons are made available for encoding ncAAs without interfering with host translation. Examples exist of ciliate organisms, which naturally evolved genetic codes with only a single stop codon, reprogramming two stop codons into sense codons for cAAs;²⁷⁶ these serve as an inspiration for also recoding stop codons with ncAAs. Several methods have been developed for the programmed recoding of genomes.

Multiplex automated genome engineering (MAGE) is a method for high-throughput editing and evolution of bacterial genomes (Figure 7). By iteratively introducing small genetic modifications, such as indels (insertions and deletions of single or a few nucleotides), the technique allows for the simultaneous editing of multiple genomic loci and has been used to recode the genome of E. coli.²⁷⁷ MAGE can be applied to generate bacterial populations with a diversity of genotypes to evolve a desired trait. Conjugative assembly genome engineering (CAGE) is used after MAGE to combine modified genomic segments into a complete genome. CAGE relies on bacterial conjugation for transferring and recombining engineered DNA segments.^225,278

Figure 7.

Genome editing via MAGE and CAGE. The native E. coli genome is first divided into arbitrary regions. In each MAGE cycle, one of these regions is targeted for multiple edits with an oligo-DNA mixture, represented as colored dashes. The DNA oligos applied in a MAGE cycle induce multiple edits of the target genomic region, and we illustrate the dashes representing an oligo pool in the same color as the genomic region that underwent editing with exactly this oligo pool. For example, the dark red oligos are used in the first MAGE cycle to edit one of the genomic regions, resulting in the eventually edited dark red genomic region. MAGE cycles are performed for editing different genomic regions in parallel. Subsequently, CAGE cycles mediate the combination of edited genomic regions, finally merging the outcome of MAGE cycles into a completely edited E. coli chromosome.

Developed by Church and Isaacs, the MAGE and CAGE workflows have been used in an extensive genome-recoding approach in E. coli, replacing 314 TAG codons with synonymous TAA codons across a total of 32 strains.²⁷⁸ The MG1655 strain used served as an ideal starting point for recoding alleles at high frequency due to its deficient mismatch repair pathway (ΔmutS).²⁷⁹ Specifically, the genome was categorized into 32 regions, 31 of which contained 10 target sites each and a single region with four target sites. This separation was chosen to optimize the number and size of oligo pools (based on previous experiments and predictions) and minimize the number of MAGE cycles and thus the probability of acquiring secondary mutations (decreased incubation time in the absence of a mismatch repair pathway and presence of λ red proteins, which increase recombination and the de novo formation of mutations). In addition, because some genomic regions might be difficult to edit, the small subsets with 10 target sites were simple to evaluate for complete editing. A library of 314 DNA oligonucleotides for codon conversion mutations was computationally designed and iteratively applied in a total of 18 MAGE cycles for target mutations over 32 cultures; each oligonucleotide had 10 target sites. Two PCR-based methods were developed to evaluate single-base edits of TAG to TAA: multiplex allele-specific colony PCR (MASC-PCR) for identifying strains with the highest number of TAA substitutions, and multiplex allele-specific colony quantitative PCR (MASC-qPCR) for quantifying allele frequency at editing targets. Extraction of the best-performing clones across each of the 32 populations accumulated 78% of all designed edits. Clones with incomplete codon replacement were again used over 6 to 15 further MAGE cycles, eventually resulting in 32 strains with all of the desired codon conversions. Subsequently, their chromosomal fragments were assembled in five layers using conjugation-based CAGE. In this approach, a donor strain transfers its modified genomic fragment into the genome of a recipient strain via λ red-based recombination.²⁸⁰ The λ red protein recombines the oriT sequence that is fused to a kanamycin resistance (kanR) cassette upstream of the donor strain gene fragment into permissible loci of the recipient strain genome. Additional markers for positive selection were placed downstream of the donor fragment, either zeocin, spectinomycin, or gentamycin resistance genes. Recipient strains retained their own recoded regions, flanked by distinct positive and positive–negative selection markers, including tolC and galK. Recombination frequencies varied between 10^–7 and 10^–5 in the first round. Notably, the oriT-kanR and the positive–negative selection markers are not transferred, and the introduction of either the oriT-kanR or the positive–negative selection marker into the respective positive selection marker of any strain can turn this strain into either a donor or a recipient strain, respectively. By successfully overcoming unexpected, ambiguous tolC phenotypes, CAGE was iteratively applied to combine the modified regions of initially 32 strains into 16, then 8, 4, 2, and finally, a single genome.²⁷⁸ By overcoming technical hurdles with this recoded strain, and considering further instances of the TAG codon, a team effort of the Rinehart, Church, and Isaacs groups corrected and further reprogrammed its genome to achieve the recoding of all 321 amber codons with synonymous ochre codons, yielding the recoded C321.ΔA strain.²²⁵ The substitution of amber codons made it possible to delete the RF1 gene (ΔprfA) in C321.ΔA. The strain was shown to perform efficient and unambiguous amber suppression, installing either pAzF or 2-naphthylalanine (NapA) into a GFP containing up to three copies of the amber codon.²⁸¹ Regarding GCE efficiency, the deletion of RF1 in the reprogrammed C321.ΔA strain did not significantly affect single amber site suppression, but enhanced ncAA incorporation efficiency for transcripts containing multiple amber codons.²²⁴

Because this C321.ΔA strain exhibited slower growth and accumulated 355 off-target mutations, the teams of Yokoyama and Sakamoto attempted to recode 95 of the 273 amber codons to synonymous codons in the E. coli strain BL21(DE3), and in addition delete RF1.²⁸² These efforts yielded the B-95.ΔA strain, which had comparable growth rates to its ancestor in rich medium, and the B-95.ΔAΔfabR strain, which was better adapted to reduced temperature and nutrient conditions.²⁸² Efforts in sequencing and profiling C321.ΔA and its intermediate ancestor strains enabled the recoded genome to be identified and engineered further by selectively mutating translation machinery genes and repairing off-target mutations that manifested during the recoding process. These countermeasures rescued fitness defects and provided insight into the physiology of the recoded C321.ΔA strain.²⁸³

MAGE has also been used to construct an extensive library of chromosomally integrated TyrRS variants derived from M. jannaschii, bearing up to 12 mutations in the catalytic center and anticodon-binding domains.¹¹⁶ Compared to plasmid-based libraries, inserting variants into the genome strictly couples selection to cell viability and avoids gene-dosage effects that can arise if multiple plasmids enter a single cell. The recoding effort yielded aaRS variants with up to 25-fold enhanced incorporation of ncAAs (p-acetyl-L-phenylalanine and p-azido-L-phenylalanine) and modified selectivity toward 14 distinct ncAAs. These variants included one that faithfully aminoacylated p-azido-L-phenylalanine but disregarded 237 other ncAAs for GCE. Another variant was applied for GCE at 30 different amber sites with high accuracy and yield.¹¹⁶ MAGE can further be combined with RNA-targeted DNA cleavage by the CRISPR/Cas9 system to facilitate the efficiency and speed of multiplex genome editing.²⁸⁴

CRISPR/Cas systems can be used for RNA-guided genome cleavage, followed by a targeted homology-directed repair (HDR).²⁸⁵ The latter is achieved by supplying a donor DNA template, which is used by the HDR machinery to produce a defined editing outcome. The CRISPR/Cas9 system has been used to create genome-wide mutations, including codon replacements. However, editing multiple loci with CRISPR/Cas9 is still a considerable challenge for several reasons: (1) only sites containing a protospacer-adjacent motif (NGG in the case of Cas9) can be targeted for editing, (2) highly repetitive sequences and loci in heterochromatin are difficult to edit, and (3) off-targets accumulate with each gRNA used.^286,287 More recent DNA editing systems include prime editing²⁸⁸ and DNA base editing,²⁸⁹ which could potentially be used for genome recoding of defined sites to swap codons without causing double-strand breaks in the DNA. A recent example demonstrated the multiplex base editing of TAG to TAA conversions of selected essential genes in the HEK293T human kidney cell line.²⁹⁰ Up to 33 out of 47 target sites were successfully edited with a single transfection, yielding around 40 off-target C to T conversions in exons of the essential genes targeted. Finally, RNA editing and RNA base-editing tools, such as those based on Cas13, have great potential for recoding specific transcript sites.²⁹¹ However, CRISPR/Cas systems have not yet been used to recode codons on a large scale for GCE, also due to the complexity of precisely engineering hundreds of loci simultaneously. In principle, CRISPR/Cas systems can be applied in analogy to λ red recombination in MAGE cycles²⁷⁸ in groups of edits that can later be combined. Several challenges need to be overcome to achieve efficient multiplex editing for enabling the depletion or substitution of particular codons on the scale of an entire genome. These include, e.g., increasing the efficiency and specificity of inducing a double strand DNA break repair and its correct reconstruction via homology-directed repair. To avoid the simultaneous generation of double strand DNA breaks, nicking nucleases may facilitate multiple simultaneous genome-editing events. Precise editing of multiple sites could be improved by using CRISPR nucleases that act as nickases due to one of the enzymes nuclease domains being inactivated,²⁹² or by improving homology-directed repair efficiency,²⁹³ such as through the recruitment of donor DNA to loci targeted for editing,^127,294 or by inactivating the non-homologous end-joining repair pathway.^295,296

2.12. Altered Genetic Codes for Viral Resistance and Biocontainment

The critical dependence on the supply of ncAAs can be utilized as a biocontainment strategy to prevent the spread and viability of an organism in noncontrolled environments. Strategies including MAGE (see section 2.11) have been used to generate genomically recoded organisms, engineered with one or more stop codons within essential genes, and thus making recoding a requirement for survival.²²⁵ Biocontainment in bacterial^297,298 and viral systems²⁹⁹ yielded recoded organisms that do not survive without being supplied ncAAs and might be a useful strategy for developing future live vaccines. The efficiency and robustness of such a biocontainment system need to be balanced, as a higher number of stop codons for recoding will decrease the escape frequency while also decreasing both the recoding efficiency and thefitness of the recoded cell. Escape frequency can further be minimized if the protein target can be engineered to strictly require ncAA incorporation for functionality, as demonstrated for β-lactamase across different bacterial species^218,300 and for a sliding clamp protein in E. coli, where additional specificity, achieved through also engineering other residues near an essential protein–protein interaction site, diminished escape frequency to less than 10^–10.

Metagenome analysis revealed natural recoding of stop codons in bacteria and bacteriophages, suggesting that the evolution of genetic codes is also part of the arms race between viruses and their hosts.³⁰¹ By analogy, GCE can be exploited to confer viral resistance to engineered organisms.^302,303 The E. coli strain Syn61, which harbors a synthetically recoded genome (discussed in section 2.13), has been used for recoding three codons on which a bacteriophage relied for its reproduction, thus conferring viral resistance.³⁰² From another perspective, bacteriophages commonly infecting E. coli strains that possess an expanded genetic code have also been shown to acquire enhanced evolvability, i.e., exhibiting higher rates of neutral and beneficial mutations at codons reassigned in the E. coli host.³⁰⁴

2.13. Genome Synthesis

The construction and validation pipeline of MAGE is laborious, rather complex, and associated with potential off-target mutations.^205,225 Meanwhile, progress in accurate, fast, and low-cost methods of DNA synthesis means that the synthesis of whole genomes is feasible. In 2008, the synthesis of the Mycoplasma genitalium genome was achieved in a collaboration between the Smith, Hutchinson, and Venter groups and started from artificial DNA cassettes of 5–7 kb.³¹⁰In vitro recombination was then used to assemble these cassettes into three intermediate fragments of 24, 72, and 144 kb size that were maintained as BACs in E. coli. The final assembly of the synthetic genome was performed in S. cerevisiae, making use of its high homologous recombination rate. Two years later, the genome of Mycoplasma mycoides JCVI-syn1.0 was synthesized.³¹¹ A computational and experimental workflow was developed for synthesizing an E. coli genome with 57 utilized codons.³⁰⁵ This approach started from synthetic DNA fragments of two to four kilobases with homology overlaps. In total, 1256 of these fragments were assembled into 87 genome segments of about 50 kilobases in size into a low-copy-number plasmid backbone, making use of the highly efficient homology-directed repair machinery of S. cerevisiae. Replacement of genome segments in E. coli was performed via λ red recombination, and the CRISPR/Cas9 system was used to minimize non-integrated residual plasmid copies after the recombination. However, the resulting rE.coli-57 strain has not yet been applied for GCE.

A similar approach called “replicon excision method for enhanced genome engineering through programmed recombination (REXER)” has been developed to introduce large, synthetic genomic fragments from a bacterial artificial chromosome (BAC) by combining RNA-guided cleavage mediated by CRISPR/Cas9 endonuclease with λ red-mediated recombination. REXER can be iteratively applied by alternating positive- and negative-selection markers, a method named genome stepwise interchange synthesis (GENESIS; Figure 8).³⁰⁶ These methods have been applied to replace a 220 kb genomic region with a synthetic DNA fragment of 230 kb from a BAC in two steps.³⁰⁷ This was realized by first assembling synthetic DNA fragments of up to 10 kb in length, each flanked by 80–200 bp homology regions, into a linearized BAC backbone, by transformation into S. cerevisiae and thus making use of the highly efficient homology-directed repair in yeast for assembly. The BAC is then purified and transformed into E. coli along with a helper plasmid for expression of the λ red recombinase components (alpha, beta, and gamma), the Cas9 endonuclease, and a trans-activating CRISPR RNA that targets sites for excising the synthetic BAC sequence. The cleaved, synthetic fragment is subsequently inserted into the bacterial chromosome via λ red recombination, directed by homology regions flanking the fragment. Making use of different positive–negative selection cassettes in both the chromosomal and the BAC insert sequence allowed selection against the original chromosomal sequence and for the designed insertion sequence. These approaches have been used to study compression schemes of synonymous codons in E. coli.³⁰⁷ Variations of the REXER workflow introduced BAC delivery via conjugation along with the highly efficient assembly of large DNA fragments and the stepwise replacement of large genomic regions of around 100 kb in E. coli and have been used to assemble 1.1 Mb of human DNA in E. coli.³⁰⁸ Although BAC fragments have been introduced and expressed in human HEK293T cells, integration of their large DNA fragments is challenging. This could be facilitated by improved methods that enhance homology-directed repair, albeit with the compromise of low-efficiency transfection, as state-of-art viral delivery into mammalian cells is limited by DNA cargo size.³⁰⁹

Figure 8.

Assembly of synthetic genomes with REXER. Replacement of native DNA with large, chemically synthesized DNA fragments by iterative rounds of REXER. In each round, a new piece of synthetic DNA is inserted from a BAC together with a dual antibiotic resistance cassette, replacing the previous one. The colored boxes represent antibiotic resistance cassettes. Prior to using REXER, a dual antibiotic resistance cassette is introduced to the native genome. The residing antibiotic cassette can be removed after the final REXER iteration to yield a scarless synthetic chromosome. Alternatively, they can be retained in the genome to facilitate selection in later experiments.

The de novo synthesis of entire genomes has been successfully applied to a handful of organisms, including Mycoplasma genitalium (0.582 Mb),³¹⁰Mycoplasma mycoides JCVI-syn1.0 (1.08 Mb),³¹¹ and Escherichia coli (4 Mb)²²⁶. With regard to eukaryotes, an effort to synthesize the Saccharomyces cerevisiae genome is in progress (see section 3.1).

Iterative rounds of REXER were instrumental for the synthesis of the E. coli genome in 2019 that produced the synthetic strain Syn61.³¹² The Syn61 genome features a synonymous substitution of the amber and the serine codons UCG and UCA. The additional deletion of RF1 and of (previously essential) cognate tRNA genes yielded Syn61Δ3, and further rounds of random evolution were applied to almost completely rescue growth defects in, among others, the Syn61Δ3(ev5) strain. These strains were resistant to infection by a bacteriophage cocktail, produced ncAA-containing macrocyclic peptides, and site-specifically encoded up to nine copies of a single ncAA into a target protein. Remarkably, the Syn61Δ3(ev4) strain demonstrated incorporation of up to three different ncAAs, CbzK, BocK, and p-I-Phe, in ubiquitin by three orthogonal aaRS/tRNA pairs in response to the UCG, UCA, and UAG codons, respectively.³¹³

3. Recoding in Eukaryotes

3.1. Establishing Eukaryotic GCE Systems

GCE has advanced tremendously in prokaryotic systems over the past two decades, with E. coli being the workhorse for most of these developments. Great progress has also been made in eukaryotes, especially in baker’s yeast and mammalian cells. GCE in eukaryotes still holds substantial challenges, including that many strategies developed in prokaryotes are not readily transferable to eukaryotes because of the higher complexity of their genomes and cellular organization alongside differences in cellular processes (Figure 9). For example, the genomes of higher eukaryotes have billions of base pairs arranged in multiple chromosomes, compared to the single 4 Mb E. coli chromosome. Likewise, the complexity of tRNAs increases with the complexity of cells: 84 tRNA genes are encoded in E. coli,³¹⁴ 275 tRNAs in S. cerevisiae,³¹⁵ and more than 400 in humans.³¹⁶ Optimization of tRNA expression levels, especially for stable expression, as well as compensating for the difference in promoter elements between prokaryotes and eukaryotes, are some of the major challenges for transferring GCE-specific tRNAs of bacterial or archaeal origin to eukaryotic systems. Specific aaRS/tRNA pairs need to be evolved and engineered for orthogonality, to alter their substrate specificity, or to enhance their performance. Most synthetases are evolved in E. coli due to its fast growth, short generation time of less than one hour, high transformation efficiency, excellent retention of high plasmid copy numbers, and simple subcellular organization compared to eukaryotes. Several of the synthetases evolved in E. coli can also be used in eukaryotes. However, the need to develop eukaryotic evolution platforms for evolving and engineering aaRS/tRNA pairs is increasing with the number of newly discovered ncAA chemistries, and yeast has been recently applied as simple eukaryotic model and evolution chassis. Further considerations for eukaryotic GCE systems include the efficient delivery of genes encoding GCE components and the improvement of ncAA availability (cellular uptake or biogenesis), both of which can be challenging, especially in multicellular organisms. In addition, potential counter measures against host-specific cellular pathways that impair GCE, for example, the non-sense-mediated decay (NMD) pathway, need to be devised. GCE can also be combined with other techniques, such as RNA editing to tackle some of the challenges in eukaryotic hosts, such as the minimalistic labeling of endogenous proteins. Recently, Hao et al.³¹⁷ developed a nonheritable method for introducing stop codons into an endogenous RNA. They utilized RNA base editors to convert C to U in a target mRNA specified by a CRISPR Cas13 guide RNA. The target mRNA was subsequently translated by GCE-specific machinery to place a desired ncAA into endogenously expressed proteins.

Figure 9.

Transfer of GCE-enabling technologies from the prokaryotic to the eukaryotic world. Diverse GCE-enabling methods are denoted as either successfully established (green tick), partially established (light blue tick in parentheses), or not established (dark blue cross) across the model host systems of bacteria (E. coli), fungi (S. cerevisiae) and higher eukaryotes/mammalian cells (H. sapiens).

We dedicate the following sections to further discuss the advances that address the existing limitations of GCE and new methodologies for harnessing the power of GCE in eukaryotes.

3.2. tRNA Expression, Evolution and Engineering

A major challenge with the direct use of bacterial tRNAs for GCE in eukaryotic systems is the heterologous expression of a functional tRNA species in sufficient levels. Transcription and the subsequent processing steps vary significantly between prokaryotes and eukaryotes. Regarding phylogeny, bacteria are less related to eukaryotes than archaea, making bacterial aaRS/tRNA pairs generally better candidates for achieving orthogonality in eukaryotes compared to archaeal ones. Thus, GCE in eukaryotes frequently requires the expression of bacterial tRNAs. Eukaryotes transcribe tRNAs with RNA polymerase III, which requires the consensus sequences A-box and B-box elements within the tRNA sequence.^318,319 Most bacterial tRNAs lack these elements, hindering their efficient expression in eukaryotic hosts. In yeast, high expression of tRNA of bacterial origin is typically achieved via the RPR1 and SNR52 RNA polymerase III promoters.³²⁰ In mammalian cells, the U6, H1, and 7SK promoters are often used for tRNA expression.³²¹

Several strategies have been used to increase expression and ncAA charging of tRNA. Drabkin et al. added 5′ and 3′ flanking sequences from the human methionine initiator tRNA gene (hsup2) to the amber suppressor glutamine tRNA isolated from E. coli. They subsequently mutated cytosine at position 9 (present within the A-box element) to an adenine to mimic the consensus of most eukaryotic tRNAs.³²² This variant of the EctRNA gene, referred to as hsup2A9, was successfully expressed in Cos-1 and CV-1 cells after transient transfection. Another difference between eukaryotic and prokaryotic tRNA transcription is that the 3′ terminal trinucleotide CCA is enzymatically added in eukaryotes by the CCA-adding transfer RNA nucleotidyl transferase TRNT1. Hence, the CCA was removed from hsup2C9A to design hsup2A9ΔCCA; indeed, this was better expressed in CV-1 cells. Further, the tRNA transcribed from the hsup2A9ΔCCA gene and its cognate synthetase, EcGlnRS, could suppress amber in Cos-1 cells, whereas the gene product of hsup2A9 was not functional.

A similar strategy has been applied to the EcTyrtRNA, introducing three mutations at positions 9, 10, and 25 and adding the 5′ flanking region of the human tyrosyl tRNA.³²³ However, this rationally designed EcTyrtRNA (A9G10C25) failed to yield GCE in CHO cells. An alternative TyrtRNA from Bacillus stearothermophilus, which recognizes the corresponding EcTyrRS and also included the A and B consensus sequences, was successfully applied. Although functional, the heterologous pair yielded markedly lower suppression efficiency compared to other suppressor tRNAs in mammalian cells (20–40%).^324,325 Suppression efficiencies comparable to human tyrosyl tRNA have been achieved by transfecting a plasmid containing nine copies of the B. stearothermophilus suppressor tRNA in CHO cells.

Extragenic RNA polymerase III promoters have been used to drive transcription of prokaryotic tRNAs in eukaryotic host cells, including the aforementioned human H1 promoter, which does not necessitate post-processing of the tRNA 5′ end. Wang et al. showed its utility for GCE in HeLa cells with suppressors derived from the EcTyrtRNA gene that contain the 3′ flanking region of human methionine tRNAs without the 3′ terminal CCA triplet.³²⁶ In S. cerevisiae, the SNR52 and RPR1 promoters have been used to express EcTyrtRNA. Significantly higher expression of EcTyrtRNA was observed when the respective gene was supplemented only with the 5′ flanking sequence from yeast tRNA SUP4, compared to when transcription was driven by SNR52 and RPR1 promoters. However, these tRNAs demonstrate significantly lower protein translation, potentially due to partially dysfunctional tRNA processing.³²⁷

Hancock et al. achieved amber suppressor tRNA transcription in S. cerevisiae by manipulating the host’s endogenous machinery.³²⁸ The PyltRNA gene from M. barkeri, which lacks the A- and B-box promoter elements necessary for its transcription in S. cerevisiae, can be mutated to introduce these consensus sequences, although doing so did not yield detectable expression or amber suppression. The substitution of previously evaluated 5′ and 3′ flanking sequences partially rescued the expression and functionality of the MbtRNA for GCE in S. cerevisiae. GCE has been further enhanced by using an unconventional, dicistronic tRNA gene that is transcribed as a single precursor tRNA and subsequently processed into two mature tRNA species: arginine-tRNA_UCU and asparagine-tRNA_GUC. Substituting the asparagine-tRNA_GUC gene with the PyltRNA gene from M. barkeri facilitated its expression. Unfortunately, MbPyltRNA proved to be non-orthogonal in yeast and was also charged by the host alanine aaRS. Introducing a mutation converted the PyltRNA gene of M. barkeri to resemble the sequence of M. mazei, thus enabling orthogonal GCE in yeast using variants of the MbPylRS. This system enabled the incorporation of various ncAAs into S. cerevisiae proteins, including lysine derivatives mimicking PTMs, photocaged and photo-crosslinking lysine derivatives, as well as an ncAA with a clickable alkyne handle.³²⁸

Even if the appropriate promoters are used, low expression of tRNA can still be a major factor limiting the efficiency of ncAA incorporation in eukaryotes. Transfecting or genomically integrating multiple tRNA gene copies can help to increase tRNA expression.^329−331 Schmied et al. also demonstrated comparable levels of tRNA expression achieved with a single copy of tRNA driven by an optimized U6 promoter and four copies of suppressor tRNA, each under a U6 promoter and a CMV enhancer. These suppressor tRNAs also carried a mutation (U25C) in the anticodon stem. This mutation disrupts the wobble base-pair interaction of G:U by replacing the U with a C, thereby improving amber readthrough efficiency.³³²

The tRNA structure itself can also be a bottleneck to achieving high GCE efficiency and has therefore been optimized through directed evolution or rational design. The PyltRNA isolated from archaea is one of the most widely used suppressor tRNAs because of its orthogonality in both bacterial and eukaryotic hosts. However, its secondary structure varies significantly from mammalian endogenous tRNAs, bearing a limited degree of resemblance to the mitochondrial serine-tRNA_UGA. Serfling et al.³³³ hypothesized that the sub-optimal interaction of the archaeal PyltRNA with the mammalian translational machinery arises from this structural divergence. On the basis of this assumption, they designed two sets of mutated PyltRNAs guided by improving interactions with the mammalian translation machinery and mimicking conserved structural elements of human tRNAs, such as the mammalian mitochondrial serine tRNA. Inspired by previous studies,³³⁴ chimeric tRNAs were designed by adding combinations of PylRS recognition motifs to Bos taurus mitochondrial serine tRNA_CUA. After evaluation in HEK293T cells, selected mutants were further engineered by adding an A-box motif to the D-arm, as well as by combining successful mutations, particularly in the linker between the acceptor stem and T-arm, which is recognized by the elongation factor EF-Tu in bacteria. Incorporation efficiencies were determined for BocK and CbzK, yielding different results between the tRNA sets. The best-performing tRNAs from the two sets were named M15 and C15; these have in common a mutation between the D- and T-loops and differ in either containing (C15) or not containing (M15) mutations between the acceptor stem and T-arm. The substitution of the tertiary Watson–Crick pair U19:A56 with G19:C56 confers stability to both variants.^331,332 Compared to their ancestor PyltRNA, the M15 and C15 variants yielded a two- to fivefold increase in intracellular concentration, strongly contributing to improved amber suppression efficiency.³³³

Considering the impact suppressor tRNA engineering has on GCE, there is a pertinent need for a directed evolution platform for GCE-specific tRNAs in eukaryotes. This is somewhat challenging, because the strategies used for mammalian cells to date require stable expression of the gene of interest in the host and subsequent sequence randomization by mutagenesis, which is not fit for tRNA evolution. A single copy of tRNA per cell is not enough to achieve detectable ncAA incorporation,^331,335,336 and the low mutagenic frequency of these strategies is also not suitable for short DNA sequences like that corresponding to a tRNA. An additional constraint is that any mutation in a stem region generally needs to be complemented by a matching counterpart on the complement strand to retain base pairing and stem structure, which is often necessary for tRNA secondary structure recognition by processing components, aaRSs, elongation factors and the ribosome, and thus essential for activity. This has recently been addressed via synthetic site-saturation mutagenesis libraries in the virus-assisted directed evolution of tRNA (VADER) approach. VADER enables in vivo tRNA evolution in mammalian cells with the help of an adeno-associated virus (AAV2), which delivers a single tRNA variant to each cell.³³⁷ Similar to PACE described for prokaryote systems (see section 2.1.3), production of the viral replication machinery is inherently coupled to amber suppression and the expression of a GCE-performing suppressor tRNA variant. To validate whether a single viral infection per cell is sufficient for amber suppression, HEK293T cells were infected with AAV2, carrying an mCherry reporter and M. mazei PyltRNA genes. This was followed by the transfection of constructs for expressing the Rep gene, encoding AAV2 replication proteins, three viral capsid proteins harboring amber codons for suppression, as well as with further helper genes and the M. barkeri PylRS gene. The generation of progeny virus was observed only in the presence of the ncAA, thereby proving detectable amber suppression. In order to distinguish between orthogonal and cross-reactive tRNA variants and to isolate orthogonal ones selectively, amber suppression was performed with an ncAA containing a clickable azido group. Only orthogonal tRNAs would facilitate the production of virus with ncAA-incorporated capsid. These virus populations can bind to dibenzocyclooctyne (DBCO) conjugated with biotin via strain-promoted azide–alkyne reaction (SPAAC) between the azido handle of the incorporated ncAA and DBCO. Subsequently, these virus particles are pulled down via the biotin–streptavidin interaction. To conclusively prove the reliability of the above-described approach to identifying only orthogonal tRNAs, cells were infected with a predetermined mixture of orthogonal and non-orthogonal tRNAs, and VADER was subsequently used to show a >30,000-fold enrichment of orthogonal tRNA in a single round.

Jewel et al. applied VADER to evolve the PyltRNA further, targeting the A- and T-stems, which are involved in the recognition of the tRNA by translation components.^337,338 Coupling the directed evolution approach to sequencing allowed all members of the tRNA libraries to be identified, providing a better understanding of the effect the mutations have on stop codon read through efficiency. Because GCE efficiency and orthogonality are host-specific, effective tRNA variants generally have to be evolved for each host and may be transferred between related host organisms.

3.3. aaRS Evolution and Engineering

The GCE efficiency, orthogonality, and specificity for new ncAA substrates all depend on the properties of the aaRS. Although the majority of directed evolution efforts have been done in E. coli (see sections 2.1 and 2.2), the transfer to eukaryotes can be complicated by their subcellular complexity, translation machinery, the greater number of endogenous aaRS/tRNA pairs, as well as their closer phylogeny to archaeal aaRS/tRNA pairs. S. cerevisiae, as a simple eukaryotic model organism, is the workhorse for screening and establishing GCE systems.³³⁹EcTyrRS was a good starting point due to its orthogonality in S. cerevisiae. Five residues in the active site were randomly mutated to alter the substrate specificity of the yeast-orthogonal TyrRS from E. coli and selected based on the crystal structure of the homologous B. stearothermophilus TyrRS.³⁴⁰ Directed evolution involving iterative rounds of positive and negative selection was performed (see section 2.1) to yield an evolved synthetase that served for suppression of two amber sites in the transcription factor GAL4. GAL4, in turn, directed the synthesis of the HIS3, URA3, and lacZ gene products. Expression of HIS3 and URA3 genes enabled the yeast to synthesize histidine and uracil, respectively, to grow on amino-acid-depleted media auxotrophically. Expression of lacZ facilitated colorimetric selection of active amber-suppression variants in the presence of X-gal. For positive selection, surviving clones were selected from growth media containing the relevant ncAA and lacking histidine and uracil or supplied with 3-aminotriazole (3-AT), which is a competitive inhibitor of HIS3p. Negative selection was performed by growing cells in media devoid of ncAA and containing 0.01% 5-fluorootic acid (5-FOA), which is converted into a toxin by URA3. Five rounds of both positive and negative selection yielded a TyrRS variant that could incorporate five ncAAs into proteins, namely p-acetyl-L-phenylalanine, p-benzoyl-L-phenylalanine (pBpa), p-azido-L-phenylalanine (pAzF), O-methyl-L-tyrosine (OMeY), and p-iodo-L-phenylalanine (pIF). The same lab further evolved the EcTyrRS also to accept two ncAAs, an acetylene and pAzF.³⁴¹

Recently, Stieglitz et al. designed a high-throughput platform for engineering synthetases in S. cerevisiae based on FACS.³⁴² Starting from EcTyrRS and LeuRS, both known to be orthogonal in yeast,³⁴³ and led by structural studies, selected residues in the substrate binding pocket were subjected to saturation mutagenesis to engineer specificity for ncAAs with bulky or long side chains. For positive selection, cells were grown in the presence of the respective ncAA and those expressing a fluorescent reporter containing an amber site were collected by FACS. Negative selection was performed without ncAA, and cells not expressing the reporter were collected. Repetitive selection rounds were performed to yield aaRS variants for which GCE efficiency and selectivity was quantified in terms of relative readthrough efficiency (RRE) and maximum misincorporation frequency (MMF). RRE is an estimation of the aaRS efficiency. Its value ranges from 0 to 1, where 0 corresponds to the condition in which premature termination of translation occurred in all cases, generating only truncated reporter, and 1 represents wild-type protein translation efficiency. MMF provides an indication of misincorporation at the amber site by quantifying maximum cAA misincorporation in the absence of an ncAA. This method enabled isolation of synthetases with specificity toward a number of ncAAs, including 3,4-dihydroxy-L-phenylalanine (L-DOPA) and 4-borono-L-phenylalanine (BPhe). To further improve amber suppression efficiency, error-prone PCR, which had been successfully used for evolution of aaRSs in E. coli,^108−110 was applied to one of the evolved synthetases, further increasing L-DOPA incorporation and identifying mutations that have not been identified in previous directed evolution studies. Furthermore, it was also shown that, depending on the modifications in the selection strategy, it is possible to evolve polyspecific synthetases as well as synthetases that can distinguish between structurally similar ncAAs. With the established positive and negative selection procedures and a set of six structurally similar ncAAs, polyspecific synthetases are often evolved. To avoid polyspecificity, all five of the ncAAs except the one chosen for incorporation were added for the negative selection. The positive screen was left unchanged as it was performed in the presence of the desired ncAA (p-propargyloxy-L-phenylalanine, ProY), and after iterative selection it was possible to isolate synthetases with enhanced specificity for ProY over the other five ncAAs, despite their structural similarity to ProY. Vice versa, to enhance polyspecificity, two different strategies were developed. The first (track 1) involved the addition of all the selected ncAAs for the positive selection. For the second strategy (track 2), only one of the selected six ncAAs were added sequentially per round of selection. Track 1 is simpler than track 2, but it is impossible to control the specific ncAA being encoded per round of selection in track 1. On the other hand, track 2, although not as straightforward as track 1, allows fine tuning of ncAA specificity for each selection round. Using yeast as a eukaryotic screening platform thus enables fast and simple evolution and screening workflows for aaRS engineering.

While aaRS engineering has often focused on the ncAA binding pocket with the aim of changing the ncAA substrate spectrum, mutations made in other domains can modulate tRNA interactions.³⁴⁴ For systems like the EcTyrRS/tRNA, where the anticodon of the tRNA needs to be changed to CUA, it is not implausible that this deviation from the native condition adversely affects the synthetase–tRNA interaction. In the absence of the EcTyrRS–tRNA crystal structure, the homologous Thermus thermophilus TyrRS–tRNA complex³⁴⁵ was utilized by Takimoto et al.³⁴⁴ to design new relevant mutations. G34 in the anticodon loop of the EcTyrtRNA is mutated to C34 to change the anticodon to CUA, and because G34 of the TttRNA is recognized by Asp265 of the cognate synthetase, the same site was chosen for mutations in the EcTyrRS as well. The change from G to C creates a gap between the base and Asp265, thus hindering their interaction. To bridge this gap, five amino acids with longer side chains than Asp: Tyr, Arg, Gln, Phe, and Leu, were chosen to replace this Asp residue. Of the five possibilities, Asp265Arg performed the best, producing a 186% increase in the fluorescence intensity of the reporter in HeLa cells. Introducing the same mutation into other E. coli tyrosyl-derived synthetases was also explored, leading to improved ncAA incorporation efficiency. In general, this approach demonstrates the potential optimization of GCE by shifting focus toward improving suppressor tRNA–synthetase interaction.

The site-specific incorporation of minimalistic fluorescent labels, which reduce any potential perturbations of the biological systems to a minimum while emitting brightly, are perfectly suited for their use in super-resolution microscopy. However, unexplained unspecific nuclear labeling had severely hindered the utilization of GCE in super-resolution microscopy and for the visualization of nuclear or low-abundant proteins.^346,347 To solve this issue, Nikić et al.³⁴⁸ analysed the sequence of M. mazei PylRS and interestingly revealed that it possesses a nuclear localization signal (NLS). The NLS motif is responsible for facilitating nuclear import of the corresponding protein. The presence of the NLS, combined with the high affinity of the PylRS for its cognate tRNA, might have been the cause of the background fluorescence. Indeed, nuclear imported ncAA-charged PylRS bound to tRNAs can plausibly be click-labeled. To prove this hypothesis, immunostaining against PylRS and fluorescence in situ hybridization (FISH) for the cognate tRNA was performed in HEK293T and COS-7 cells. The results revealed nuclear localization of both the synthetase and the tRNA. Hence to ensure localization of the PylRS to the cytoplasm, a nuclear export signal (NES), stronger than the existing NLS, was added to the synthetase sequence. Cytoplasmic localization of the synthetase is not only necessary to eliminate the background nuclear fluorescence but also to increase the efficiency of amber suppression. Thus, with this engineered synthetase, a 15-fold increase in amber suppression was observed compared to the synthetase without NES, and reliable imaging of a nuclear protein jun-B fused with GFP was possible without interference from unspecific nuclear labeling.

3.4. Modulating Nonsense-Mediated Decay and Translation Components

The nonsense-mediated decay (NMD) pathway in eukaryotes is dedicated to surveilling the degradation of mRNAs with premature stop codons.³⁴⁹ Because GCE is most frequently achieved through the suppression and recoding of stop codons, the NMD-mediated decay of target mRNAs with site-specifically inserted stop codons can substantially decrease the expression of the ncAA-incorporated protein. Wang et al. showed that deletion of the gene UPF1 in S. cerevisiae,³⁵⁰ which represents an essential NMD pathway component, affords enhanced amber suppression.³²⁷ In yeast, NMD activity is more pronounced for stop codons closer to the 5′ ends of transcripts than those at the 3′ end. In line with this, the NMD-deficient upf1Δ strain performed amber suppression with more than two-fold increased ncAA incorporation if the TAG codon was positioned close to the 5′ end as opposed to almost no effect when it is closer to the 3′ end.³⁵¹ Similar strategies have also proved successful in other organisms. An NMD-deficient strain smg-2(e2008) was demonstrated to accomplish enhanced amber suppression.³⁵² Parrish et al. tested another strain of NMD-deficient C. elegans, produced by knocking down the smg-1 gene, essential for the NMD pathway.³²⁹ Compared to the control strain, the smg-1-knock-down strain expressed more of the modified POI, which was quantified using western blotting. A mean fold increase in POI expression of 5.6 was observed for the NMD-deficient strain versus the control strain. Recently, the utility of another such C. elegans strain, smg-6(ok1794), was investigated for the incorporation of photocaged tyrosines into nanobodies.³⁵³

Besides altering the suppressor tRNA and its cognate synthetase, translation components can be engineered to match specific needs for GCE. The eukaryotic release factor (eRF1) has been a suitable target for GCE optimization in mammalian cells, where the eRF1–eRF3 complex mediates translation termination and, in stop codon suppression systems, competes with suppressor tRNAs. Whereas bacteria use two distinct release factors to recognize UAG/UAA and UGA/UAA codons,^219,354 eukaryotic eRF1 terminates peptide chains at all three stop codons. Engineering eRF1 by rational design has focused on mutating its N-terminal domain, which interacts with amber codons, and leaving the C-terminal domain unchanged as it mediates interactions with eRF3 to yield functional translation termination complexes. The eRF1(Δ100) and eRF1(E55D) mutations, both at the N terminus, resulted in the formation of eRF1–eRF3 complexes with reduced stop-codon affinity, leading to five- to sevenfold higher incorporation of N^ε-(tert-butoxycarbonyl)-L-lysine in sfGFP in HEK293T cells than in cells not ectopically expressing the release factor. However, eRF1(Δ100) was observed to increase the readthrough of all three stop codons, whereas eRF1(E55D) was more specific to the amber stop codon and was thus preferred for further studies.³³¹ More recent studies have demonstrated the advantage of eRF1(E55D) in selective increments of amber codon readthrough.^355,356

Recently, Sushkin et al. demonstrated that modulation of the integrated stress response (ISR) pathway can be harnessed to boost GCE efficiency.³⁵⁷ A range of cellular stress conditions trigger the activation of protein kinase R (PKR), which phosphorylates eIF2α to reduce the global translation rate. Multiple strategies for inhibiting the PKR-dependent pathway were adopted. Of these, the use of a truncated PKR (ΔPKR), lacking the catalytic kinase domain, and using a mutated eIF2α (eIF2α S51A), with the main phosphorylation target of the PKR kinase mutated, were the most efficient. Combining both of these strategies to decrease PKR pathway activity yielded 2.8-, 3.2-, and 2.5-fold enhancements for single, double, and triple amber codon reassignment, respectively.

3.5. Codon Context Effects in GCE

The readthrough efficiency of stop codons depends on the position within a transcript and the surrounding nucleotide sequence, which is called the codon context effect.³⁵⁸ The base flanking the stop codon immediately downstream primarily affects its suppression. In eukaryotes, purines at this position tend to promote the termination of translation, whereas cytosine rather enhances stop codon readthrough.³⁵⁹ Multiple nucleotide positions further downstream of the stop codon have also been found to affect GCE, illustrating that local sequence context can be adjusted by introducing silent mutations to enhance ncAA incorporation at specific sites.³⁶⁰ While the aforementioned studies primarily looked at the effects associated with nucleotide positions 3′ of the stop codon, Bartoschek et al. assessed the entire local environment of the amber codon and developed a computational tool for predicting amber suppression efficiency based on the codon context effect. This tool identified codons of a target gene as candidates for substitution with stop codons.³⁶¹ Specifically, an evolved PylRS and four copies of the cognate tRNA from M. mazei was stably integrated into the genomes of mouse embryonic stem cells and HEK293T cells. Subsequently, a specialized version of SORT-E (section 3.12) was used to facilitate proteome-wide amber suppression and subsequent characterization of the modified proteome. The ncAA bicyclo[6.1.0]non-4-ynyl lysine carbamate (BCNK) incorporated at amber codons was bioorthogonally conjugated to a biotin-containing handle, which in turn, was used for the selective pull-down of interaction partners with streptavidin beads. Mass spectrometry identified ncAA incorporation sites and was used to derive the effects on amber suppression efficiency by analyzing the corresponding codon context. These data sets allowed development of a predictor based on linear regression, which had been evaluated by comparing predictions with experimental data of an in vivo GCE-dependent fluorescent reporter readout. The predictions correlated with experimental data, validating the pipeline across two cell lines, and predicting beneficial mutations around the amber codon.³⁶¹ However, many effects ultimately determine total protein yield, including mRNA stability and, potentially, secondary structure, the NMD pathway, final protein stability, and the ability of ribosomes to start translation at non-alternative sites. Designing mutation sites is a multi-parameter problem that is not easily addressed, and codon context is one of many parameters to optimize.

3.6. Strategies to Improve ncAA Availability

Making ncAAs bioavailable for intracellular GCE systems can represent a challenge based on the physicochemical properties of the ncAA. For example, charged molecules tend to have lower membrane permeability, as log P values indicate for all mined genetically encoded ncAAs (Table S1). At physiological pH, amino acids predominantly exist as zwitterions, which limits their availability.³⁶² Esterification of the carboxyl group can increase the availability of neutral forms of ncAAs and has the potential to improve ncAA permeability. However, hydrolysis of the ester bond by intracellular esterases can still lead to the incorporation of the original amino acid. Exploiting this property, Takimoto et al. synthesized methyl, ethyl, and acetoxymethyl (AM) esters of the ncAA dansyl alanine (DanA). A fluorescence-based reporter expression assay performed in HeLa cells demonstrated that all three esters performed better compared to unmodified DanA, with the AME (DanA-OAM) being the most efficient.

By measuring intracellular ncAA concentrations in HEK293T cells supplemented with DanA and DanA-OAM in the growth medium, a remarkable increase of 31-fold was measured in DanA concentration for the cells supplied with DanA-OAM. However, one disadvantage of AME modification is increased cell mortality due to the generation of formaldehyde upon ester hydrolysis.

Some ncAAs can be imported into cells because of the promiscuous nature of transporters and channels, especially if the ncAA has high structural similarity with a cAA.^351,363,364 In multicellular organisms, ncAA accessibility can be more complex, generally requiring ncAA uptake from food. For example, in C. elegans, a protective cuticle can exclude ncAAs, and the digestive system can actively degrade an ncAA. In order to study ncAA uptake in C. elegans and potentially improve it, Parrish et al. analysed the uptake of two ncAAs: OMeY, which is structurally similar to tyrosine, and DanA that has no structural similarity to any cAA. In contrast to OMeY, uptake of DanA was not observed in the amber suppression assay due to its sequestration in the C. elegans intestine. This was overcome by feeding the dipeptide DanA-A and exploiting the host’s dipeptide transporters, namely PEPT-1 and PEPT-2, for its import. Uptaken dipeptide is cleaved by cellular peptidases, the approach yielding efficient amber suppression with DanA. Notably, the arrangement of the dipeptide affects its uptake, with DanA-A being more readily taken up than A-DanA.³²⁹

A different approach for solving issues of ncAA uptake could be to manipulate the host machinery to synthesize the required ncAAs in vivo. Chen et al. used a cytosolic sulfotransferase (NnSULT1C1) from Nipponia nippon specific for tyrosine sulfonation for the biosynthesis of sulfotyrosine (sTyr) in HEK293T cells. The NnSULT1C1 gene was integrated into the HEK293T genome by the PiggyBac system to facilitate stable sulfotransferase expression. A higher yield of sTyr incorporation by GCE was observed in HEK293T cells biosynthesizing sTyr compared to those to which sTyr was externally supplied.³⁶⁵ In another study, Wu et al. used an O-methyltransferase (MfnG) to biosynthesize OMeY successfully in HEK293T cells and zebrafish. In a flow cytometry analysis using amber-suppressed EGFP as the reporter, it was observed that the mean fluorescence intensity for HEK293T cells stably expressing MfnG was 2.4 times higher than HEK293T cells that were externally supplied with 1 mM OMeY. A similar observation was made in zebrafish, demonstrating the applicability of this method to multicellular organisms.²¹⁶

3.7. Viral Delivery of GCE Components

Using viruses is an efficient approach for delivering GCE machinery into mammalian cells or genomes, generally yielding lower but also more homogenous expression levels than transiently transfected cell populations.³⁶⁶ Furthermore, the number of delivered gene copies per cell can be precisely controlled. For example, lentiviruses are an established system which can deliver precisely two copies of a gene (on the two lentiviral nucleocapsids) into host cells, which then can be controlled to incorporate only a single copy of the gene into a genomic target locus.³³⁰ Likewise, integration of a single copy of a gene via transient transfection is possible by appropriate dilution to adapt for the vector-to-cell ratio stochastically and in S. cerevisiae by using plasmids containing a centromeric element to ensure retention of only a single plasmid in the cell. As a prerequisite for delivery, the viral genome must have the capacity to accommodate all GCE components and, ideally, multiple tRNA copies, which must not be eliminated by recombination. Lentiviral vectors can carry exogenous DNA loads of up to 7.5 kb,³⁶⁷ whereas baculovirus-based systems offer a greater capacity of >30 kb, and have delivered GCE components into mammalian cells.³⁶⁶ Due to the lower transduction efficiency of various mammalian cells, these have combined viral envelope proteins, such as the vesicular stomatitis virus G glycoprotein (VSVG), for pseudotyping the viral envelope and improving transduction.^368−371 To minimize the chance of losing copies contained in the multiple-copy tRNA cassettes due to recombination, two different suppressor tRNAs derived from B. steareothermophilus and E. coli were encoded in alternating copies, both of which were recognized by the EcTyrRS. This allowed all of the GCE-specific components to be delivered in a single vector and was applied to study the dependence of ncAA incorporation on the relative amounts of various components.³³⁶ This approach was made possible by features of the viral delivery system that allow precisely tuneable population-wide gene expression. Compared with transient transfection, they showed a linear increase of reporter expression along with an increased amount of virus, whereas the corresponding observation for transfection was not linear. This study quantified the extent to which tRNA expression limits GCE efficiency, showing that more than 100 copies of tRNA gene per cell are required. Another relevant observation from this work was that overexpression of the suppressor synthetase negatively affects amber suppression under tRNA-limiting conditions. Notably, this viral vector can be used, as shown by Zheng et al., for the efficient transduction of more complex systems than cell culture, such as mouse brain tissue, indicating applicability to higher organisms. Other suitable viral delivery systems are the adenoviral vectors, which feature high cargo capacity, efficient transduction in various cell types, and sufficient stability to be readily condensed by cesium chloride density gradient centrifugation. Adenoviral vectors have been used for ncAA incorporation in several cell types, including HeLa, human tumor cell lines, and primary cells.³⁷²

3.8. Stable Expression of GCE Components

The stable expression of GCE components has clear advantages compared to transient transfection, which is generally limited to amenable cell types and results in heterogenous expression that is maintained only for a few days. The resulting variability and fluctuation in expression can interfere with the precision of measurements or the assay timeline, especially if long-duration treatments or cell differentiation are involved.³⁷³ A versatile vector based on the so-called PiggyBac inverted terminal repeats has enabled the genomic integration and stable expression of GCE-specific aaRS and multiple tRNA genes in mammalian cells. Flanked by transposon repeats, cassettes of up to 200 kb can be inserted at a genomic target locus with repetitive TTAA sequences by the PiggyBac transposase.³⁷⁴ This system has enabled high-efficiency gene integrations into several eukaryotic model organisms, including mouse embryonic stem cells (ESCs) in which GCE was used to probe the effects of histone acetylation on transcription. By integrating four copies of the PyltRNA, the PiggyBac system was shown to raise intracellular tRNA levels, which often limits GCE efficiency. PiggyBac-mediated GCE integration has been used in brain organoids formed from patient-derived induced pluripotent stem cells (hiPSC), which stably expressed the PylRS/tRNA pair from M. mazei. These primary brain organoids can, for example, be used to study the mechanisms of molecular pathways or PTMs and are powerful disease models for neurodegenerative disorders.³⁷⁵

A disadvantage of the PiggyBac transposition is that neither the copy number nor the integration site can be controlled, and cell populations that undergo transposition yield highly heterogenous mixtures. Typically, these populations require the selection and outgrowth of individual clones based on desired properties, such as the expression of a reporter gene that requires amber suppression, by using flow cytometry coupled with single-cell sorting. Alternatively, to transposition, the CRISPR/Cas9 system can be used with a donor DNA to yield the specific cleavage of a guide-RNA-targeted genomic locus, followed by the homology-directed repair of the site for mutations defined by the donor DNA template. This approach has been used to incorporate the gene of interest in the AAVS1 site on chromosome 9 of the human genome, a previously validated safe harbor site where genes can be inserted without adversely affecting cellular physiology.³⁷⁶ Specifically, the Streptococcus pyogens type II CRISPR/Cas9 system has been used in combination with a single guide RNA for stabilization of Cas9 at a defined locus, consisting of a fusion between the CRISPR RNA (crRNAs) and trans-activator RNA (tracrRNA). The crRNA-derived sequence consists of 20 nucleotides that are complementary to the defined target sequence, which further requires a 5′ NGG protospacer-adjacent motif (PAM) present on the nontemplate strand.²⁸⁵ Two nuclease domains in Cas9 mediate the double-strand DNA cleavage of three nucleotides upstream of PAM, and a defined donor DNA is generally supplied as a template for homology-directed repair. Although homology-directed repair is highly efficient in yeast, higher eukaryotes favor repair via non-homologous end-joining, which is not templated and typically results in small insertions and deletions.

Stable expression of a functional GCE machinery has further been achieved in the human hematopoietic cells, which were subsequently engrafted into mice.³⁷⁷ Because commonly applied gene integration strategies were not suitable for hematopoietic cells, an episomal vector system based on the Epstein–Barr virus has been used to deliver the MbPylRS/tRNA pair on a viral construct that is capable of replication in the host cell, as well as partitioning equally into daughter cells. This episomal property sustains the retention of the extra-chromosomal construct in the nucleus of mitotic primate cells for long-term transgene expression. Six tRNA gene copies were encoded on the vector, along with PylRS and a fluorescent reporter gene for amber suppression, to counteract low tRNA expression yields. After achieving stable ncAA encoding in a human hematopoietic system, CD34⁺ cells nucleofected with the same EBV-based vector system was engrafted into mice and about 20% of the engrafted cells incorporated the ncAA.

In a recent study in the Elsässer lab, Meineke et al. evaluated the GCE efficacy of transient transfection compared to stable expression of the necessary components in a range of mammalian cell lines.³⁷⁸ MmPylRS/tRNA and GFP^150TAG genes were integrated by the PiggyBac system into HEK293, HEK293T, HCT116, human melanoma A375, U-2OS, and COS-7 cell lines. HCT116 proved to be the best-performing cell line in expressing amber-suppressed GFP. Two other cell lines with appreciable GFP expression were HEK293T and HEK293. Furthermore, on transient transfection of the same GCE components, HEK293 and HCT116 cells showed poor transfection efficiency. However, a stable HCT116 cell line expressed comparable levels of amber-suppressed GFP to transiently transfected HEK293T cells, which were superior in GCE performance to their transiently transfected HCT116 counterparts. Stable cell lines exhibited reduced heterogenicity in expression of GFP^150TAG in all cases compared to transiently transfected variants. In addition to amber suppression, the readthrough of opal and ochre stop codons was also investigated in this study. It is notable that both opal and ochre suppression was successful only in the stable HCT116 cell lines, thereby highlighting the advantage of stably expressing GCE components. At the same time, since reporter expression is generally low in the case of stable expression, the level of nonspecific ncAA incorporation is more relevant and detectable in this case than in transient transfection, as demonstrated in this study by stable amber suppression of a variety of intracellular proteins in HCT116 cells. However, the stable and efficient expression of prokaryotic or archaeal tRNA in higher eukaryotes, such as mammalian cells, remains a major challenge; the problem can be overcome in transient transfections by increasing the number of plasmids delivered into the cell. A recent study solely focused on maximizing amber suppression yield for biotechnological applications used standard cell line generation tools (i.e., spontaneous insertion under strong survival pressure) to generate a CHO cell line that harbored more than 200 tRNA genes.³³⁵

3.9. Quadruplet Codons in Eukaryotes

The Schultz lab evolved a variant of the PyltRNA to recognize the AGGA quadruplet.³⁷⁹ Because the anticodon is devoid of identity elements in the PyltRNA, its alteration did not affect recognition by PylRS. However, mutating the anticodon to the quadruplet UCCU might affect tRNA affinity for the ribosome. Therefore, the respective tRNA was first evolved in E. coli by randomizing the four-base anticodon. A series of positive and negative selection strategies were designed to identify seven candidates. A chloramphenicol resistance gene containing the quadruplet codon AGGA was used for selecting the tRNAs. The first round of positive selection was performed in the presence of the ncAA BocK and chloramphenicol. It was followed by screening in the presence and absence of BocK. Hit candidates were subsequently subjected to further evolution based on the assumption that mutation in the anticodon stem would influence the interaction between the anticodon loop and the ribosome. The evolved tRNA incorporated BocK into GFP containing a frameshift mutation to accommodate the AGGA codon in HEK293T cells. A similar approach was followed to utilize tRNA evolved in E. coli directly in 293T cells to decode the UAGA codon. The evolved GCE machinery was further utilized to generate an HIV-1 variant dependent on UAGA decoding for its replication and propagation, with a final goal of developing HIV-1 vaccines.³⁸⁰ A more comparative study on 11 variants of quadruplet codon-decoding PyltRNAs was performed by Mills et al.³⁸¹ The 11 variants included those with anticodons mutated to NCUA (N = A/U/G/C), CUAG, UCCU, and five pre-validated quadruplet codon-decoding tRNAs.^379,380,382 It was observed that tRNAs with NCUA anticodons failed to work in HEK293 cells, and tRNA_UCCU(Ev2) developed by Niu et al. performed best.

The Greiss lab established the first quadruplet-based GCE system in multicellular animals, expressing the PylRS/tRNA pair in C. elegans. To improve ncAA incorporation efficiency, the PyltRNA was equipped with the UAGA anticodon and mutated so that its structure more closely resembled that of endogenous tRNAs.³³³ Hybrid tRNAs were designed that combined the optimized scaffold elements for eukaryotic translation with the previously evolved anticodon loop for efficient quadruplet codon decoding. These hybrid tRNAs facilitated the incorporation of a photocaged lysine into the Cre recombinase protein and demonstrated optical control of Cre target gene expression. Furthermore, the Greiss lab developed a method of cell ablation by incorporating a photocaged cysteine into caspase-3 and subsequently optically controlling the enzymes’s activity. Notably, mutations in the tRNA anticodon loop beneficial for quadruplet decoding were independent of the anticodon itself, implying that the hybrid tRNA sequences can be utilized to decode a diverse range of quadruplet codons.³⁸³

3.10. Strategies for Site-Specific Incorporation of Multiple ncAAs in Eukaryotes

Incorporation of multiple ncAAs into a POI can enable many functions, such as facilitating Förster resonance energy transfer (FRET) measurements and the study of multiple PTMs. There are two possibilities to achieve this in vivo: utilize all three available stop codons (amber, opal, and ochre) or use the same stop codon multiple times. Both strategies have been explored and have their respective advantages and disadvantages. A polyspecific synthetase derived from EcTyrRS was used by Xiao et al. to encode a maximum of three ncAAs into EGFP by repurposing the amber codon multiple times. Although it was possible to express EGFPs with one, two, and three ncAAs incorporated, yields for the double and triple amber mutants were lower than the single amber mutant EGFP.³⁸⁴ Nonetheless, several studies involving multiple ncAA incorporation have provided answers to pertinent biological questions.^385−387

In addition to codon availability, site-specific incorporation of multiple distinct ncAAs requires mutually orthogonal aaRS/tRNA pairs with specificities for those ncAAs (Figure 10). For example, to incorporate two ncAAs, the GCE-specific tRNA1 and tRNA2 must only interact with their cognate synthetases, RS1 and RS2, respectively. tRNA1 should not be charged by RS2 or any of the host’s endogenous synthetases. Similarly, tRNA2 should not interact with RS1 or endogenous synthetases. It is also essential that neither RS1 nor RS2 accepts cAAs and that they are selective for their specific ncAAs, ncAA1 and ncAA2, respectively. Xiao et al.³⁸⁴ demonstrated a combination of ochre-suppressing MbPylRS/tRNA and amber-suppressing EcTyrRS/tRNA pairs to simultaneously repurpose amber and ochre codons in HEK293T cells. However, the yield of the double-mutated POI (EGFP) was only 10% of that of the single mutant. The efficacies of other synthetase/tRNA pairs for dual stop-codon reassignment, for example, EcTyrRS/tRNA (TAG/TGA suppression) and PylRS/tRNA (TGA/TAG suppression) and EcLeuRS/tRNA (TGA suppressor) and PylRS/tRNA (TAG suppressor) have also been evaluated.⁷⁰ It has been consistently shown that neither opal nor ochre suppression works as efficiently as amber suppression.^70,388 Hence, the development of a practical synthetase/tRNA pair for reassigning one or both of these codons would considerably benefit GCE. To this end, Osgood et al. recently reported an EcTrpRS/tRNA pair for efficient opal suppression. They also tested the cross-reactivity of this pair with the other most commonly used suppressor synthetase/tRNA pairs in mammalian cells. EcTrpRS/tRNA showed orthogonality with the corresponding E. coli tyrosyl, leucyl, and archaeal pyrrolysyl pairs, facilitating more GCE machinery combinations for dual stop codon suppression.³⁸⁹ Osgood et al. also achieved suppression of all three stop codons in EGFP with EcTrpRS/tRNA (opal), EcTyrRS/tRNA (amber), and PylRS/tRNA (ochre) pairs. In this case, termination of the EGFP was ensured with a C-terminal TEV cleavage site followed by three ochre codons.

Figure 10.

Mutual orthogonality between GCE-specific and endogenous translational machinery. In order to facilitate the incorporation of multiple distinct ncAAs, it is essential that the GCE-specific tRNAs (tRNA1 and tRNA2) only interact with their cognate synthetases and that the GCE-specific synthetases (RS1 and RS2) selectively accept different ncAAs.

3.11. Strategies to Minimize Cross-Talk of the GCE Machinery with Host Translation

The site-specific GCE methods discussed so far in eukaryotes typically lack mRNA specificity. The GCE machinery cannot distinguish between the target stop codon and the naturally occurring stop codons in the host organism, thereby leading to proteome-wide modification. In E. coli, this has been overcome by including orthogonal ribosome (O-ribosome)/orthogonal mRNA pairs, artificial base pair codons, or by recoding or synthesizing entire genomes. Although orthogonal ribosome/orthogonal mRNA pairs have been established and optimized in prokaryotes (see section 2.9), the higher complexity of the eukaryotic translational machinery and cellular organization hinder transfer of such a strategy to eukaryotes. Likewise, the stable integration and decoding of artificial base pairs in eukaryotes will require adjustments of various DNA replication, RNA transcription, and protein translation components, which necessitates ground-breaking development on multiple levels (see section 2.10). Moreover, despite the lowering costs and rapid progress made in DNA editing and synthesis (see sections 2.11 and 2.13, respectively), and successful transfer of MAGE to S. cerevisiae,³⁹⁰ such endeavors still represent laborious and year-long multi-research-group efforts, as exemplified by the on-going assembly of the synthetic yeast genome, which started more than 10 years ago.³⁹¹ Diverse strains of S. cerevisiae with synthetic chromosomes, such as synIXR and semi-synVIL,^391,392 were developed and have demonstrated powerful features that include a controllable genome evolution system based on Cre recombinase shuffling of genes flanked by loxP sites. With regard to the potential for GCE, the Sc2.0 genome was compressed to 11.3 Mb, all TAG codons substituted with TAA, and all 275 tRNA genes relocated to a dedicated tRNA neochromosome.³⁹³ However, the full synthetic genome with 16 chromosomes has not yet been completed, and intermediate strains have not yet been reported in the context of GCE.³⁹² With at least 22 chromosomes and more than 3 billion base pairs, the human genome will be laborious and complex to synthesize and is not an immediately foreseeable achievement.

An alternative approach for reducing unspecific amber suppression of endogenous proteins in a time-dependent (or stimulus-dependent) way is the development of chemically inducible GCE systems. To this end, two strategies based on the T-Rex and Tet-On systems were explored by Koehler et al.³⁹⁴ For both systems, the synthetase and supressor tRNAs were placed under the control of an inducible promoter. The T-Rex system requires addition of tetracycline to induce transcription of a target gene by titrating the tetracycline repressor (TetR) protein away from the tetracycline operon (tetO) sites in the target gene promoter (this promoter is repressed in the TetR-bound state). In contrast, the Tet-On system is based on a reverse tetracycline-controlled transactivator (rtTA), consisting of a reverse TetR mutant, which binds to tetO sites in presence of tetracycline, fused to the C-terminal domain of the virion protein 16 originating from the Herpes simplex virus. Upon addition of tetracycline (or its potent analog doxycycline), rtTA binds to tetO sites of a minimal promoter to activate transcription of a downstream target gene. The Tet-On system allowed only for controlling the synthetase expression from a polymerase II promoter, while both synthetase and tRNA expression could be induced with the T-Rex system. Although the T-Rex system is not suitable for transient transfection (since the repressor protein TetR is required to be bound to the promoter before induction of gene expression), this can be overcome by the stable expression of TetR.

3.12. Residue-Specific Codon Reassignment Based on Natural aaRS Promiscuity

Residue-specific reassignment refers to the stochastic incorporation of a supplied ncAA in response to specific sense codons (Figure 11) occurring naturally in the whole transcriptome and making use of endogenous aaRS/tRNA pairs with promiscuous acceptance of the respective ncAA. This phenomenon can be exploited to identify newly synthesized proteins, for example, in specific cell types, at different developmental stages, or in response to external stimuli. Characterizing the cell-specific nascent proteome at specific time points is essential for understanding the molecular basis of dynamic processes such as environmental stimulation, stress responses, circadian oscillation,³⁹⁵ learning,³⁹⁶ and so forth.

Figure 11.

A comparison of site-specific stop codon and residue-specific sense codon reassignment. In the case of site-specific GCE, an amber codon introduced at a predetermined site in an mRNA of a POI is repurposed to introduce an ncAA at the corresponding site in the POI. In contrast, for residue-specific sense codon reassignment, a sense codon (denoted as XXX) is repurposed to incorporate an ncAA and all instances of the chosen codon occurring in the host transcriptome are attempted to be reassigned.

MetRS is naturally promiscuous in accepting ncAAs that are structurally similar to methionine. This property has been exploited by subjecting cells to methionine starvation and feeding with methionine analogues, such as azidohomoalanine (AHA). These analogues are then charged upon MettRNA by the endogenous MetRS and encoded proteome-wide in response to ATG codons. The bioorthogonal conjugation properties of the azido group arise from its click chemistry reactions with tags for affinity purification and subsequent protein identification through mass spectrometry. This technique, termed “bioorthogonal amino acid tagging” (BONCAT), has been developed by the groups of Schuman and Tirrell,³⁹⁸ and variations have emerged based on the subsequent detection methodology. Fluorescent imaging of the nascent peptides is possible by using clickable dyes compatible with AHA. This is termed “fluorescent noncanonical amino acid tagging” (FUNCAT). FUNCAT can also be utilized to label populations of nascent proteins synthesized at different time points with the help of multiple methionine analogues, distinct dyes, and a pulse–chase approach. Coupling FUNCAT with a proximity ligation assay (FUNCAT-PLA) enables the detection and visualization of specific newly synthesized peptides.³⁹⁹ In this case, a pair of primary antibodies are utilized. One of the antibodies is protein specific, while the other is targeted against the clickable moieties attached to the ncAA. The secondary antibodies carry the oligonucleotides, which are ligated once the two antibodies are in sufficient proximity, that is, on the same protein. After subsequent amplification of the ligated oligonucleotides by rolling-cycle amplification, these can be visualized by labeled complementary probes.

The significant efforts invested toward evolving diverse GCE-specific aaRS/tRNA pairs have facilitated the co-translational incorporation of hundreds of ncAAs in vivo. Codon-specific mutation in the anti-codon sequence of corresponding tRNAs can extend the concept of BONCAT to include all 61 codons and in principle any ncAA of desired functionality. Stochastic orthogonal recoding of translation with chemoselective modification (SORT-M) incorporates ncAAs with clickable chemical handles at diverse sense codons across the proteome of a specific cell type and has enabled tissue-specific proteins in Drosophila melanogaster to be visualized at precise developmental stages. SORT-M has also allowed the identification of proteins in specific parts of the fly without the need for dissection.⁴⁰⁰ SORT can also be coupled with enrichment of the ncAA-incorporated proteins, which can subsequently be identified and characterized by mass spectrometry (SORT-E).⁴⁰¹

Residue-specific sense codon reassignment is a powerful tool for proteomics studies. However, it lacks protein specificity and is not applicable to cases in which ncAA incorporation is desired at a precise predetermined site in a POI.

3.13. Orthogonally Translating Organelles for mRNA-Specific Translation in Eukaryotes

In contrast to E. coli, eukaryotic mRNA has no Shine Dalgarno sequence, and with this a conceptual copying of the O-ribosome approach, where the ribosome is engineered to manipulate its mRNA specificity, is not possible. Since mRNA is 5′ capped in eukaryotes, an analogy would be, if an orthogonal cap could be engineered to recruit a uniquely compatible ribosome. However, those ideas are currently out of reach.

mRNA-specific GCE, where ultimately the POI is selectively modified, has been realized in mammalian cells by means of synthetic membrane-less organelles. While eukaryotic cells confine a diverse range of functionalities within different membrane-enclosed compartments (organelles), a variety of phase-separated assemblies of proteins, RNAs, and other molecules have been discovered over the last decade; these assemblies are referred to as membrane-less organelles.^402−404 Fundamental to this was the realization that multivalent biomolecules can phase separate into a dense (reaching millimolar concentrations, here referred to as the condensate phase) and a dilute phase (sub-nanomolar concentrations), similar to the oil and vinegar in salad dressing. These natural condensates are enriched with specific components, such as proteins and RNAs, but are depleted in others, generating a unique biochemical environment that can be distinct from the surrounding cytoplasm. Provided the dense phase stays dynamic, it is, in principle, in thermodynamic exchange with the cytoplasm, that is, components are still exchanged between the dense and dilute phases.

Applying the concept of membrane-less organelles, Lemke and co-workers generated orthogonally translating (OT) designer organelles that confer mRNA specificity to the GCE reaction^405−407 (Figure 12). With translation being a highly complex process involving a multitude of components, a membrane-enclosed system would necessitate various mechanisms for molecular import and export of all translational components. Instead, an approach was chosen to construct a microenvironment that selectively enriched desired GCE components and also allowed for molecular traffic so that even ribosomes can locate into or at least very close to the designed microenvironment. The core building block of an OT organelle consists of a (1) the GCE-specific synthetase and its corresponding tRNA and (2) an mRNA-recruiting protein with specific affinity for a target mRNA containing the introduced stop codon. The key step is that the GCE-specific synthetase and the mRNA-recruiting protein are engineered with an assembler domain that highly concentrates the synthetase and the target mRNA to a spatially distinct site in the cell.

Figure 12.

A synthetic membrane-less organelle renders GCE mRNA-specific.

The ultimate working hypothesis is that this generates a sharp suppressor tRNA gradient only around the target mRNA and not elsewhere. An assembler can in principle be everything that leads to high concentration. In the seminal work, two approaches and combinations thereof were tested: (i) truncated kinesin domains that walk to the end of the microtubule network but cannot fall off and thus enrich and (ii) multivalent intrinsically disordered proteins that at higher expression levels (higher than the critical concentration) condense into a dense phase that is in equilibrium with the dilute phase. One of the best approaches was a combination of both assembler and enrichment methods. The system made use of kinesins as anchor proteins, the fused in sarcoma (FUS) protein as an assembler, PylRS, as well as the MS2 coat protein (MCP) as an RNA binding protein and yielded the recruitment of target mRNA tagged with MS2 hairpins for the efficient amber suppression of encoded mCherry reporter to what was termed an OT organelle. Amber suppression of the recruited mCherry(TAG) reporter gene was approximately eight fold higher than that of a nonrecruited GFP(TAG) reporter without MS2 loops expressed from the same plasmid.⁴⁰⁵ Note that infinite selectivity cannot be expected, since the suppressorPyltRNA is free to diffuse through the cell, and a sharp gradient around the mRNA of choice is only established by highly concentrating the cognate PylRS to the OT organelle. OT organelles were shown to facilitate the incorporation of the ncAA cyclooctyne-lysine (SCOK) into several proteins, including vimentin and nucleoporins, as well as membrane proteins, such as the insulin receptor.

Introducing anchors for subcellular membranes, OT organelles have been generated on the membrane surfaces of the inner plasmid membrane, mitochondria, the endoplasmatic reticulum, and the Golgi apparatus in the form of thin-film-like compartments of approximately 100 nm.⁴⁰⁶ In those systems, the membrane anchor, due to the reduction in dimensionality, is believed to reduce the critical concentration for phase separation so that OT organelle components are even more spatially enriched, and thus, ultimately, the suppressor PyltRNA gradient becomes even steeper. These thin-film OT organelles have been successfully utilized in HEK293T and HeLa cells for GCE with a selectivity of up to 12-fold comparing a recruited versus nonrecruited fluorescence reporter transcript. Beyond this, two OT organelles, anchored at different subcellular locations, have been expressed within the same cell to demonstrate the execution of two distinct genetic codes by suppression of the same stop codon (amber) in two distinct transcripts, which were recruited to different OT organelles for the incorporation of two distinct ncAAs in two distinct proteins. This was achieved in HEK293T cells, by generating one OT organelle at the plasma membrane and one at the mitochondrial surface, both of which are within <1 micron in HEK293T cells. Thus, within a 1-micron spatial distance, an OT organelle can be generated in which the amber codon is read out in one specific way to encode an ncAA, while another OT organelle encodes a different ncAA into a different protein in response to the amber codon. At the same time, the host mRNA is terminated whenever an amber codon is encountered.

The selective mode of action of these film-like OT organelles is based on their independent assembly without intermixing, selective target RNA recruitment, and distinct ncAA specificity. Recently, the application of mitochondria-anchored OT organelles has enabled the site-specific incorporation of two ncAAs into the disordered nucleoporin 98 in combinations of varied distances between each other; this benefitted from enhanced contrast due to less background ncAA incorporation compared to general cytoplasmic GCE. Bioorthogonal labeling with fluorescent donor and acceptor dyes enabled fluorescence lifetime imaging (FLIM) of genetically encoded FRET dye pairs to determine the distance distribution of the disordered protein conformation inside functional nanosized nuclear pore complexes in cells.³⁸⁷

4. Diversity of Noncanonical Amino Acids

The repertoire of synthetic ncAAs is vast and continuously expanding. Through advances in chemical synthesis, genetic engineering, and bioorthogonal chemistry, researchers can now design and introduce a range of ncAAs with diverse functionalities into proteins (Figure 13). When synthesizing ncAAs that work efficiently as part of in vivo GCE systems, chemists consider a host of factors, including their size, solubility, stability, and biocompatibility. In this section, we examine the chemical diversity and functional roles of ncAAs that have been genetically encoded, and sections are devoted to applications of ncAAs in the study of biological processes in their native cellular environments, specifically ncAAs that resemble PTMs (section 4.1), exhibit reactive groups for photo-controllable biological activity (section 4.2), undergo bioorthogonal chemistry for site-specific bioconjugation (section 4.3), harbor fluorophores or other moieties for probing conformational structure and protein dynamics (section 4.4), sense their molecular environment (section 4.5), have modified backbones (section 4.6), and have multiple functionalities (section 4.7).

Figure 13.

The expanding repertoire of noncanonical amino acids. Genetic encoding unleashes a multiverse of functional groups on the amino acid, paving the way for cutting-edge chemical biology innovations and various applications.

We have endeavored to provide a comprehensive overview of the chemical features of ncAAs as firstly encoded in prokaryotes and eukaryotes, along with some few related examples on their respective utility for in vivo application and the challenges that remain. The scope of this review is merely on the diversity of the chemical repertoire and does not cover the use of GCE for tailoring proteins for in vitro applications, but many excellent reviews have been published on those fields of research, which include enzymes with modified activity, stability, substrate spectra for biotechnological applications,^26,408 and new biomaterials.⁴⁰⁹ Since the incorporation of ncAAs into antibodies and drugs, such as antimicrobial peptides, has the potential to modulate molecular properties and diversify, e.g., their efficiency, solubility, bioavailability, stability or interactions,^410,411 the utilization of GCE for developing designer drugs is continuously increasing. Since we do not discuss this topic within the scope of this article in detail, we refer the reader to outstanding reviews that deal particularly with applications of GCE in diagnosis, pharmacology, and therapy.^412−416 In the main text, our focus lies on the diversity of chemical functionalities genetically encoded in vivo, both those already applied in vivo and those paving the way for further in vivo studies. Since not all the chemical entities of all ncAAs could be discussed in the main text, a table has been generated (Table S1), which, at our best, comprehensively compiles all ncAAs that have been genetically encoded in living cells as of to-date. This resource table of in vivo genetically encoded ncAAs serves as a customizable tool for exploring key functional groups to select candidate ncAAs with, e.g., high solubility and biocompatibility for translational research, or with specific chemical features for desired applications. For each ncAA, the table includes predictions of their cLogP value, reporting on hydrophobicity and membrane permeability, as well as of their log S value, reporting on solubility in an aqueous mileu. With the table published along with this manuscript, we thus provide a comprehensive resource that enables the comparison of biological parameters critical for biocompatibility across all mined genetically encoded ncAAs (Figure 14). In addition, organic chemists who plan to synthesize a novel ncAA can make use of this resource by predicting the log S and cLogP values from external source and use these values to get information about the new ncAA even before starting its synthesis. The newly designed ncAA can be visualized in comparison with all genetically encoded ncAAs (see sheet “New ncAA properties comparison” in the Supporting Information). The solubility and membrane permeability of a novel ncAA can thus be used to, from the realm of existing ncAAs, identify the ones with similar behavior. Notably, we observed an anticorrelation between log S and cLogP for our set of genetically encoded ncAAs (Figure 14a). This anticorrelation suggests that, during ncAA design, a compromise has to be made, choosing for the design of an ncAA with either a high log S, a high cLogP, or even better, an intermediate of both values. Moreover, by plotting log S and cLogP distribution, mean values of 0.3737 and −1,4494 were obtained, respectively (Figure 14b,c). In contrast to in vivo encoded ncAAs (reviewed in this manuscript in detail), ncAAs designed for GCE in vitro (reviewed elsewhere^52,53) have comparable requirements regarding their solubility (derived from log S), while no restrictions may be given by membrane permeability (derived from cLogP), unless used in the context of liposomes or vesicles. Taken together, our resource provides a powerful tool for selecting ncAAs out of more than 500 in vivo genetically encoded ones, based on their chemical groups and application field, and depending on the specific needs. Recently, a database, named iNClusive, has been published, focusing on all ncAAs integrated into proteins of interest. This database currently has 2432 entries of different protein residues recoded with 466 unique ncAAs and including diverse metadata.³³ Both our data resource and the iNClusive database offer data mining from a different point of view, providing the community with two complementary tools for optimized experiment design and ncAA development in future.

Figure 14.

Distribution of predicted chemical properties (log S and cLogP) of the set of more than 500 ncAAs genetically encoded in vivo. The distribution illustrates the overall trend and variability of these two chemical properties along the genetically encoded ncAAs. S-methylferrocenyl-l-cysteine has been treated as outlier and excluded from the plotting. (a) Scatter plot of the predicted cLogP and log S values, highlighting with individual data points (blue dots) the ncAAs given in Table 1 and providing a detailed view of the spread and correlation between these properties (Pearson r or corrCoeff = −0.688). (b) Gaussian distribution of predicted log S values, with mean value and standard deviation indicated (mean = −0.3737, SD = 1.0981). (c) Gaussian distribution of predicted clogP values, with mean value and standard deviation indicated (mean = −1.4493, SD = 1.2297).

4.1. “Post-Translationally” Modified Amino Acids

The decoration and conversion of cAAs post-translationally gives rise to an enormous variety of proteoforms, which is further increased by splicing isoforms, especially in eukaryotes.⁴¹⁷ Over 400 distinct types of PTMs exist, influencing an array of protein functionalities and finely tuning diverse cellular processes.^418,419 Perturbations in PTM profiles can cause disturbances in crucial biological processes, potentially impairing functions essential for viability, thus contributing to the emergence of a pathological condition. For example, several PTMs of α-synuclein, including phosphorylation and ubiquitination, have been found to contribute to its soluble and aggregated state within cells, while phosphorylation of Ser129 might initiate Parkinson disease pathogenesis, ubiquitination as well as SUMOylation favors a soluble-like state of α-synuclein.⁴²⁰ GCE enables the site-specific in vivo installation of ncAAs that resemble or mimic specific PTMs or represent intermediate precursors, facilitating controlled investigations of how a particular type of PTM affects the proteome and cellular homeostasis.^421−423 Given the distinct properties of PTMs, the direct incorporation of the envisioned ncAA might not be feasible because of instability or steric or electronic hindrance to their recognition during the translation process caused by bulky or charged side-chain moieties. Furthermore, in being reversible, most PTMs are inherently transient. For example, phosphorylation is one of the most abundant modifications found in human cells. The phospho-proteome of a cell is fine tuned at the single-residue level by the interplay of approximately 500 kinases and 140 protein phosphatases in response to the cellular physiology and environment.⁴¹⁷ The relevant endogenous enzymes can be genetically or chemically inactivated to ensure that synthetically recoded modifications are maintained.⁴²³ Alternatively, chemical approaches have been developed, including (1) masked post-translationally modified ncAAs, which can be photocleaved after translation (see section 4.2.1); (2) PTM mimics that are more stable than the natural PTM in physiological conditions; and (3) ncAAs bearing a chemical handle that can be bioconjugated to give the desired PTMs. The latter approach is mostly based on two-step procedures in which the incorporation of phosphoserine (pSer) is followed by its conversion to dehydroalanine (Dha), which can then be coupled with diverse chemical moieties through a radical-mediated C–C bond. This procedure is mainly utilized for in vitro biochemical studies of PTM, and we refer the reader to excellent reviews by Park and colleagues.^424,425

Despite recent developments, researchers still face significant challenges in investigating the complex landscape of PTMs within living organisms, mainly due to a lack of established methodologies for their incorporation. The following sections discuss these challenges and recent developments in the genetic encoding of PTMs crucial for cellular functions.

4.1.1. Phosphorylation

Phosphorylation is one of the most frequently found PTMs, regulating diverse cellular processes and cell states, including subcellular protein localization and signal transduction. Given the low permeability of cell membranes to organic phosphates and the reversible nature of phosphorylation, the phosphate ester is susceptible to hydrolysis in physiological conditions, the direct GCE incorporation of phosphorylated amino acids is challenging. Because serine (Ser) and threonine (Thr) are the more frequently phosphorylated residues, several efforts toward their genetic encoding have been made.

In 2011, the Söll group expanded the genetic code of E. coli with phosphoserine (pSer; Figure 15) by using a pSer-specific synthetase derived from Methanocaldococcus maripaludis, and MjCys-tRNA. An engineered EF-Tu was necessary to accommodate the pSep–tRNASep complex for efficient GCE.¹⁸⁵ Given the importance of pSer for a wide variety of cellular processes and the possibility to convert pSer into Dha as precursor of several post-translationally modified proteins through radical C–C bond formation,^425,426 several studies have tried to optimize the efficiency of pSer incorporation. Amber suppression was enhanced by knocking out release factor 1 (RF1),^427−429 and by further evolution of the pSerRS and EF-Tu.⁴³⁰ Metabolic engineering of endogenous components that affect the levels of the respective PTMs can further increase the population of modified proteins, especially if the modification is highly transient and dynamic. As this is the case for phosphorylation, knockout of the phosphoserine phosphatase serB was useful for boosting intracellular levels of phosphoserine and its site-specific incorporation into proteins.⁴³¹ Although this approach adjusted the concentration of phosphoserine to a level comparable with those of cAAs, the deletion of serB also increased the intracellular abundance of phosphothreonine (pThr) and phosphotyrosine (pTyr). In addition, the non-hydrolysable pSer analogue phosphonomethylene alanine (Pma) was genetically encoded in E. coli(432) and in mammalian cells through modification of eukaryotic elongation factor 1 alpha in combination with a mutant eRF1 and a metabolically engineered phosphoserine biosynthesis pathway.⁴³³ Given the similarity between Ser and Thr, directed evolution of the pSerRS/tRNA pair in E. coli was shown to switch substrate specificity, thus pThr was incorporated instead of pSer.²¹⁷

Figure 15.

Genetically encoded ncAAs, bearing PTMs as naturally occurring, as masked function, or as mimics. (a) Ser/Thr phosphorylation. (b) Tyr phosphorylation. (c) Tyr sulfation.

By genetically encoding pSer, live imaging of human cells revealed that phosphorylation of Thr308, but not Ser473, is required for cellular activation of proto-oncogene Akt, which plays a role in cell survival and rate of growth in response to extracellular signals.⁴³⁴ The substitution of serine (or threonine) for either glutamate or aspartate can mimic phosphorylation simply through their negative charge given by the acidic chain.⁴²³ However, Balasuriya et al. reported that the use of Glu and Asp, which however do not possess the real features of the native PTM, due to the different isoelectronic properties between phosphate monoester and Glu or Asp, is unsuitable for studying Akt signaling, showing the importance to encode the native PTM.⁴³⁴

Given the importance of ubiquitination (see section 4.1.5), incorporation of pSer on ubiquitin at different amino acid residues showed that phosphorylation of Ser20 can control the specificity of deubiquitinases by converting the polyspecific E3 ligase into a ligase that primarily synthesizes K48 chains.⁴³⁵

Phosphotyrosine (pTyr) (Figure 15) has been directly encoded using a synthetase (MjTyrRS) evolved to better accommodate pTyr, an engineered EF-Tu to bind the phosphotyrosylate-tRNA with higher efficiency, and the contemporary knock-down of highly active phosphatases from the E. coli genome.⁴³⁶ However, protein yield was rather low using this approach. In order to increase the intracellular level of pTyr, the Schultz group used the dipeptide Lys-pTyr. This dipeptide was efficiently transported into cells by the dipeptide transporter DppA, which selectively targets the N-terminal amino acid. Within the cell, nonspecific peptidase enzymes initiated hydrolysis to release free pTyr, enhancing GCE availability.⁴³⁷ Another approach to increase the intracellular level of pTyr is to improve its stability against phosphatases. The protection of the phosphate group as a charge-neutral phosphorodiamidate improved protein expression. However, the need to hydrolyze the phosphorodiamidate function at pH 2 was incompatible with in vivo experiments.⁴³⁸ Since dephosphorylation of the mutated full protein can occur, also after protein expression, ncAAs that mimic phosphorylation and are stable to hydrolysis in vivo have been developed. A phosphonate analog of pTyr, 4-phosphonomethyl-l-phenylalanine (Pmp), replaces the hydrolysable P–O bond in pTyr with a P–C bond and is similar to pTyr in both its structure and charge. Pmp was successfully incorporated in E. coli for characterizing the affinity between human Abl1 phosphorylated mutants and 3BP2, which regulates immunoreceptors signaling.⁴³⁷ However, differences in 3BP2-binding affinity between pTyr- and Pmp-mutated Abl1 were observed, likely due to the slight difference in pK_a values between the two and the loss of one hydrogen-bond acceptor in Pmp. In 2007, Schultz et al. synthesized a stable analogue of phosphotyrosine, p-carboxymethyl-l-phenylalanine (pCMF), which was genetically encoded in E. coli with a synthetase evolved from MjTyrRS.⁴³⁹ More recently, the engineering of EcTyrRS by replacing the native tyrosyl pair of E. coli with an archaeal counterpart¹⁰¹ allowed EcTyrRS mutants to be generated that efficiently incorporated pCMF.^440,441 This development extended the application of pCMF to mammalian cells. Although no in vivo studies using pCMF have been conducted to date, these recent advancements might serve as a springboard for future developments.

4.1.2. Sulfation

Tyrosine sulfation is essential for a range of pivotal biomolecular interactions, including in chemotaxis, viral infection, anticoagulation, and signaling. This PTM occurs principally in secreted and transmembrane proteins.⁴⁴² Tyr sulfation at protein interfaces can increase their interaction strength. For example, this plays a role in endogenous chemokine signaling, significantly impacting vital pathological processes such as the entry into cells of human immunodeficiency virus (HIV) and malarial parasites, and inflammatory diseases such as rheumatoid arthritis.⁴⁴³

Sulfotyrosine (sTyr; Figure 15) was site-specifically incorporated into hirudin with the MjTyrRS/tRNA pair; the resulting sulfohirudin had an approximately 10-fold higher affinity for the human thrombin.⁴⁴⁴ In other studies, genetic encoding of sTyr at specific sites in antibodies known to recognize the HIV envelope protein gp120 helped identify the residues critical for their respective interaction.^445,446 These studies showed how GCE can be exploited to understand molecular recognition and at the same time help to develop therapeutic proteins, containing sulfated residues, which display subnanomolar affinity.

To compensate for the low uptake of extracellular sTyr, the Xiao lab recently amplified the in vivo biosynthesis of sTyr from tyrosine and 3′-phosphoadenosine-5′-phosphosulfate using a sulfotransferase in both E. coli and human embryonic kidney HEK293T cells. The resulting biosynthesized sTyr was site-specifically incorporated into thrombin inhibitors with enhanced efficacy.³⁶⁵ The study used an EcTyrRS/tRNA pair, evolved for orthogonality in mammalian cells,⁴⁴⁷ which was also applied in further cell-based studies to investigate the role of a sulfation site in the activation of the chemokine receptor CXCR4 by its endogenous ligand.⁴⁴⁸

4.1.3. Acetylation

Proteins often undergo an array of reversible PTMs on lysine residues. These contribute to the regulation of DNA replication/repair, enzyme activities, chromatin structure, protein–protein interactions, protein stability, and cellular localization.^449−451 One of these PTMs is lysine acetylation, catalyzed by acetyltransferases and histone deacetylases. This modification is observed in transcription factors, nuclear regulators, and cytoplasmic proteins. Incorporation of acetylated lysine (AcK) via GCE has been accomplished in prokaryotes,⁴⁵² and eukaryotes,⁴⁵³ including animals.^28,29 Several analogs (Figure 16), including N-propionyl-, N-butyryl-, and N-crotonyl-lysine,^454,455N-2-hydroxyisobutyryl-lysine,⁴⁵⁶N-formyl-lysine,⁴⁵⁷N-benzoyl-lysine,^458,459N-lactyl, N-β-hydroxybutyryl-, N-lipoyl-lysine,⁴⁶⁰ and N-(7-octenoyl)-lysine⁴⁶¹ have been synthesized and encoded into recombinant histones. The stability of the acetyl function encoded in AcK might be compromised by sirtuins (e.g, SIRT1), widely studied for their histone deacetylase activity.⁴⁶² In such a scenario, the encoding of analogs with enhanced stability, such as trifluoroacetyl (TfAcK, Figure 16)⁴⁶³ and thioacetyl (ThioAcK, Figure 16)⁴⁶⁴ lysine derivatives, might be advantageous. Although many in vitro studies have focused on the function of lysine acetylation and its consequences for protein–protein and protein–DNA interactions, only a few in vivo studies have been conducted.

Figure 16.

Genetically encoded lysine derivatives for studying Lys acetylation. (a) ncAAs that have been encoded in vivo and (inside the box) on which PTM studies have been performed in living cells. (b) Deacetylase-resistant ncAAs that have been in vivo encoded.

Recently, acetylation-driven phase separation has been investigated for TDP-43, a transcription repressor associated with neurodegenerative pathologies⁴⁶⁵ as well as the transcription factors IRF3 and IRF7, the physiological phase separation of which is involved in immunoresponses.⁴⁶⁶ In particular, Garcia Morato et al.⁴⁶⁵ showed that acetylation at Lys84 affects the nucleus–cytoplasm trafficking of TDP-43, whereas acetylation at Lys136 induces a nuclear droplet-like distribution. Zhang et al. suggest that deacetylation of IRF3 and IRF7 by SIRT1 is required to drive transactivation of type I interferons (IFN-1) through phase separation, providing the innate antiviral response. Since histones can be modified with diverse epigenetic modifications to control chromatin structure and transcription, in vivo studies have the potential to dissect the role of specific histone marks in human diseases, such as cancers. Creation of stable mammalian cell lines (see section 3.8) allowed the incorporation of AcK in place of six lysine residues in histone H3.³⁷³ Similarly, modification of Lys56 of histone H3 (H3K56) in yeast with different acyl moieties (acetyl and crotyl) gave different responses to DNA-damage events, with acetylation leading to more efficient DNA repair than did crotonylation.⁴⁶⁷ By using a combined approach consisting of the genetic encoding of N-(7-octenoyl)-lysine (OcK; Figure 16a)⁴⁶¹ into histones as a PTM mimic and bioorthogonal chemistry, it was possible to determine which histone residues are targeted for deacetylation by the sirtuin SIRT6. Using an alkenoyl group as an acetyl mimic that is bioorthogonally reactive with a tetrazine-bearing fluorophore (see section 4.3), visualization of OcK-modified nucleosomes that did not undergo deacylation by SIRT6 was possible.

4.1.4. Ubiquitination

Ubiquitination is a PTM in which a specific lysine residue of an acceptor protein forms an isopeptide linkage with either the C terminus of a ubiquitin donor or one of the seven lysine residues (Lys6, Lys11, Lys27, Lys29, Lys33, Lys48, or Lys63) or the N terminus of ubiquitin itself.⁴⁶⁸ Although the role of ubiquitination in controlling protein stability through proteasomal targeting is well-established, the unique features of ubiquitin chains are thought to play distinct roles in a variety of biological processes, for example, signal transduction, intracellular trafficking, and the response to DNA damage.^469−471 Because direct incorporation of ubiquitin by GCE is challenging due to its large size, methods focused on its stepwise incorporation via non-native linkages have been mainly developed in vitro (Figure 17a).^472,473 Recently, genetically encoding the ncAA azidoacetyl-Gly-Gly-Lys (AzGGK), followed by Staudinger reduction and sortase-mediated transpeptidase, made possible the ubiquitination of proliferating cell nuclear antigen–cyclobutane pyrimidine dimers in E. coli and of superfolder green fluorescent protein (sfGFP) in mammalian cells (Figure 17b).⁴⁷⁴ To investigate if ubiquitination also occurs on acetylated lysine residues, α-amino-substituted lysines, such as N^ε-(l-methionyl)-l-lysine (MetK) and N^ε-(2-propargyl-R-glycyl)-l-lysine (PraK), were prepared (Figure 17c).⁴⁷⁵ Methionine (Met) was initially chosen as a branching linker (by incorporation of MetK), since Met is the first amino acid at the N terminus of nascent proteins, and thus might serve as a recruiter of other endogenous enzymes that modify lysine aminoacylation. PraK, carrying an alkyne group, was used as a bioorthogonal moiety (see section 4.3) to enrich a ubiquitinated protein via biotinylation. In the presence of the E2 ligase UBE2W, the α-amino group of the propargyllysine chain acts as a nucleophile for ubiquitination, thereby showing that ubiquitination also occurs at lysine residues extended with a reactive primary amine.

Figure 17.

Developments in genetically encoded amino acids for studying protein ubiquitination. (a) ncAAs that have been encoded in vivo, but for which deprotection and PTM studies have been performed only in vitro. (b) ncAAs that have been encoded in vivo, and for which deprotection and PTM studies have been performed in living cells. (c) ncAAs not requiring deprotection that have been encoded in vivo, for which PTM studies were performed in living cells.

4.1.5. Methylation

Lysine methylation is frequently observed in histone proteins⁴⁷⁶ and plays an important role in cellular physiology and pathology.⁴⁷⁷ However, several aspects regarding substrate spectra of protein methyltransferases, methylation sites, and biological downstream effects, are poorly understood. Despite the importance of lysine methylation, GCE of this PTM has not been achieved due to the marginal chemical differences between the ε-amino group of lysine and its methylated form. That similarity impairs the ability of the RS to effectively discriminate methylated lysine from lysine, which is highly abundant in the cell. Given the subtle chemical differences, the use of methyl lysine mimics such as arginine and methionine,^478,479 which do not share the critical biophysical properties of methylated lysine due to differences in steric bulk and H-bonding (Arg) and polarity and charge (Met), represents a suboptimal choice.⁴⁸⁰ For this reason, an approach to studying lysine methylation has been developed based on the site-selective incorporation of protected (sometimes referred to as “caged”) monomethylated lysine analogs. One of these analogs, a Boc-protected N-methyl lysine (Figure 18), was incorporated into a histone using GCE, but deprotection requiring strongly acidic conditions (trifluoroacetic acid) precludes in vivo studies.⁴⁸¹ Deprotections of other N-methyl lysine analogs have made use of ruthenium catalysis⁴⁸² or UV irradiation of a caged derivative,⁴⁸³ conditions that could be compatible with in vivo studies (Figure 18). Despite the latter compound being shown to function as intended in living cells, there is a remarkable dearth of such studies to refer to, pointing to the need for more developments in this area to address those gaps.

Figure 18.

Mimicking N-methylated lysine. Strategies for deprotecting genetically encoded N-methyl lysine derivatives.

4.1.6. Oxidative PTMs

Tyrosine nitration is a frequently observed oxidative PTM and has served as an indicator of oxidative stress in several conditions, including human disease cell models of neurodegenerative disorders, atherosclerosis, and cancer.^484,485 Oxidative PTMs form due to reactive oxygen species (ROS) and reactive nitrogen species arise from cellular stress and the malfunctioning of enzymes involved in redox control, such as peroxidases.⁴⁸⁶ Despite there being many sites on proteins that can be nitrated, oxidation of tyrosine to nitrotyrosine is the most frequently observed event.⁴⁸⁷ GCE offers a means to introduce single or multiple PTMs at specific sites precisely, allowing the consequences of those particular PTMs to be directly studied.⁴⁸⁸ In 2008, 3-nitrotyrosine (3-NT) (Figure 19) was genetically encoded in E. coli using a MjTyrRS/tRNA pair by the groups of Chin and Mehl.⁴⁸⁹ Studying the effect of nitration of heat shock protein 90 (Hsp90) in mammalian cells required the expression and purification of site-specifically modified Hsp90 from E. coli and the subsequent delivery into eukaryotic cells via protein transfection. This approach showed that nitration at a single protein site, either Tyr33 or Tyr56, turns Hsp90 into a toxic protein that triggers cell death.⁴⁹⁰ Furthermore, nitration of Tyr33 plays a role in the downregulation of mitochondrial metabolism.⁴⁹¹ Further studies demonstrated that nitration of single tyrosine residues can alter binding site affinity of 14-3-3 protein components⁴⁹² and that Tyr166 nitration in apolipoprotein A-I is abundant and dysfunctional in human atherosclerotic plaques.⁴⁹³ To examine the effect of oxidative damage directly in mammalian cells, a MbPylRS was evolved to accommodate 3-NT, halotyrosines,^494,495 and other ncAAs bearing the marks of oxidative conditions.⁴⁹⁶

Figure 19.

Oxidative-PTM ncAAs, genetically encoded in both prokaryotic and eukaryotic systems.

4.2. Photoresponsive Amino Acids

Biological processes are fine-tuned at various levels, and the ability to perturb and manipulate these processes with equal control enables researchers to understand their function in the cell. Optically controlled chemical tools have been developed forinducible and tuned perturbation, exploiting light as an external trigger.⁴⁹⁷ Light-responsive amino acids, such as photocaged, photoswitchable, and photo-crosslinking amino acids, have been incorporated into proteins using an expanded genetic code. This has enabled diverse proteins to be activated and deactivated with precision as well as protein conformational changes and protein–protein interactions to be visualized in cells and organisms. These considerable achievements have been accomplished with spatial and temporal control while minimizing undesired side effects.⁴⁹⁸

4.2.1. Photocaged Amino Acids

Protecting specific functional group of amino acids, as amino, hydroxy, thiol, and carboxylic acid groups with a protecting group, which can be released on demand, is advantageous for several reasons: the protecting group (1) can mask the activity site of the protein containing such amino acid and be activated, when necessary, (2) can help the noncanonical amino acid to be recognized by the tRNA synthetase during protein translation, and (3) can increase cellular uptake of the noncanonical amino acid itself. In particular, the incorporation of light-sensitive protecting groups, known as “photocages”, on specific amino acids within a protein sequence offer the precise control over protein function, allowing researchers to investigate and manipulate biological processes with exceptional spatiotemporal precision.⁴⁹⁹ Proteins carrying a light-removable protecting group within their sequence can lose their activity if the photocaged amino acids are located in an active site required for biological activity. In this way, a protein’s structure–activity relationships can be investigated in a light-inducible manner, by exposure to light at a specific wavelength.

Caged versions of tyrosine,^500−503 cysteine,^{499,504−511} lysine,^512,513 serine,^514−516 and recently selenocysteine⁵¹⁷ and histidine,⁵¹⁸ have proven useful for controlling the functions of side chains in protein recognition, signaling pathways, and the activity of proteases and luciferases (Figure 20a,b).

Figure 20.

Genetically encoded caged ncAAs. (a) Caged ncAAs genetically encoded. (b) Representative application of decaging in living cells. (c) Decaging efficiency of protecting groups, in order of increasing efficiency. (d) Depending on the benzylic substituents, deprotection of nitrobenzyl groups by light irradiation generates either an aldehyde or a ketone byproduct. (e) Genetically encoded polar and small ncAAs through post-translational photolysis.

Control over protein phosphorylation in E. coli and mammalian cells has been achieved by caging kinases at critical lysine residues.⁵¹⁹ Locating 2-nitrobenzyl caging groups to critical lysine residues in polymerase and recombinase enzymes allowed user control of gene expression and silencing,⁵¹³ DNA recombination,⁵⁰⁰ and protein localization in human cells.⁵¹² In order to optimize the properties of the caging group and improve its quantum yields at longer wavelengths, several analogues with better features have been synthesized (Figure 20c). In particular, phosphorylation of serine residues in the yeast transcription factor Pho4 was controlled by caging the hydroxy group with 4,5-dimethoxy-2-nitrobenzyl group, demonstrating control over nuclear export using 405 nm laser excitation.⁵¹⁴ In addition, 6-nitropiperonylmethyl groups are more suitable for live-cell applications, because electron-donating substituents on the aromatic ring result in a bathochromic shift of the absorption maximum and, as a consequence, reduce phototoxicity.⁵⁰² Methyl substitution at the benzylic position also improves decaging kinetics, and the photolysis byproduct is the less toxic ketone (Figure 20d).⁵⁰⁴ Other derivatives with a completely different caging group moiety include coumarin analogs. Irradiation at 405 nm decaged a 7-hydroxycoumarin-caged lysine (HOCouK); if 6-bromo functionalization was added, deprotection was achieved with two-photon excitation at 760 nm.⁵²⁰ CouK was used on zebrafish embryos, a model organism for embryo development and stem-cell differentiation, allowing temporal activation of the MEK/ERK pathway⁵²¹ and modulation of Cre recombinase activity.⁵²² A derivative of CouK bearing an amino substituent (NH₂CouK) showed faster decaging kinetics and enabled signaling in the early stages of embryo development to be visualized through the optical control of protein–nucleotide interactions.⁵²³ By caging the N-terminal residue of a degron, a tag that triggers protein degradation, with coumarin, proteins of interest could be targeted for degradation in response to light-triggered deprotection.⁵²⁴ Caging the residues critical to the catalytic activity of the tyrosine kinase LCK prevented its autophosphorylation, which is an essential step for initiating T-cell antigen receptor signaling.⁵²⁵ Recently, the genetic encoding of caged versions of two negatively charged amino acids, Asp and Glu, was achieved by leveraging nitroaryl photochemistry with 2-nitrobenzyl and 7-nitroindoline forms, respectively.^526,527

Although several ncAAs have been genetically encoded, some limitations include incorporating those amino acids which have small differences compared to the canonical ones, as we have seen for N-methyl Lys. Selenocysteine also resembles cysteine, and other small polar amino acids such as 2,3-diaminopropionic acid⁵²⁸ have not been incorporated yet, due to the lack of the desired steric and electronic properties required for their efficient discrimination. Caging the seleno and amino groups allowed efficient translation in mammalian cells (Figure 20e).^517,529N-Methyllysine and lysine are too similar to be discriminated (section 4.1.6) by, for example, PylRS, but using a caged N-methyllysine enabled the incorporation of this PTM site-specifically after light-triggered deprotection.⁵³⁰ Citrulline, the result of a PTM of arginine that is essential for epigenetic transcriptional regulation,⁵³¹ has been incorporated into proteins in mammalian cells with the same decaging approach using the EcLeuRS/tRNA pair.⁵³²

Given the high biocompatibility of light and the option of spatiotemporal activation, caged amino acids have found widespread application in studying complex signaling pathways, protein–protein interactions, and cellular dynamics in live cells and organisms and, if necessary, provide a useful strategy for the incorporation of challenging functional group with suboptimal electronic and steric properties, as charged residues, and small polar ones.

4.2.2. Photoswitchable Amino Acids

Photoswitchable ncAAs are capable of undergoing reversible changes in contrast to caged ncAAs. This confers the ability to switch on and off chemical properties and change between protein conformation states by exposure to light of a defined wavelength. Diazobenzene groups are highly biocompatible and undergo reversible photoisomerization between the cis and trans isomers upon irradiation and, therefore, widely used as photoswitches (Figure 21a,b). The first photoswitches were formed in calmodulin (CaM) with an azo bridge, which when irradiated with UV light induced a conformational change upon cis-bond isomerization (Figure 21c).⁵³³ With the intention of increasing thermal stability, photoconversion, and responsiveness to visible light, fluoro derivatives of the azobenzene ncAA were genetically encoded in E. coli and mammalian cells. The wavelength of light necessary to switch the configuration from trans to cis shifted from 365 to 530 nm for the monofluoro-substituted ncAAs⁵³⁴ and to 540 nm for the pentafluoro derivative.⁵³⁵ Longer wavelengths, which are lower in energy, are typically less harmful and thus preferred when working with living systems. The Deiters group controlled luciferase activity by placing the ncAA at a specific binding site, where overall protein destabilization occurred upon irradiation. Photocontrol over translation was also achieved through an arylazopyrazole-containing phenylalanine with cis–trans isomerization at 365/530 nm irradiation, respectively. The inability of the cis form to fit into the aaRS binding pocket permitted the controlled expression of a gene of interest in response to light (Figure 21d).⁵³⁶ Recently a dibenzo-fused azo compound, dibenzo[c,g][1,2]-diazocine-alanine (DBDAA, Figure 21a) has been encoded in both E. coli and mammalian cells.⁵³⁷ The particularity of its structure is the rigid photo-chromophore, which leads to bathochromic shift, high quantum yield, and photochemical robustness, enabling it to undergo multiple photoisomerization cycles. In this case the cis-isomer is the ground-state isomer. Irradiation at 405 nm induces isomerization to the trans-isomer, converting its bowl-shape into a more crown-like shape with high ring strain energy. This photoisomerization-induced conformational change has been leveraged to either regulate gene expression by evolving two different synthetases, cis-AARS and trans-AARS from MmPylRS, that can distinguish between the two isomers, cis-DBDAA and trans-DBDAA, or modulate the binding affinity of the expressed modified protein. The genetic encoding of photoswitchable ncAAs, PSCaa (Figure 21a), into the NMDA receptor enabled real-time detection of molecular rearrangements resulting from the reversible light-induced switching of individual side chains, thus introducing a dynamic dimension to site-directed mutagenesis of targeted protein residues.⁵³⁸

Figure 21.

Photoswitchable ncAAs. (a) Genetically encoded photoswitchable ncAAs. (b) Isomerization wavelengths of various azobenzene ncAAs, including heterocycle- and fluorine-containing derivatives. (c) Protein stapling and control over protein conformation by photoisomerization. (d) Control over protein translation by photoisomerization.

4.2.3. Photo-Crosslinking Amino Acids

Photo-crosslinking ncAAs contain a chemical moiety that can establish a covalent bond with a dedicated probe. Under irradiation at a specific wavelength, the ncAA forms a reactive radical intermediate, which then undergoes an addition reaction with biomolecules in close proximity. The suitability of these probes has been demonstrated mostly for trapping protein–protein interactions.⁵³⁹ In 2002, two photo-crosslinkers, p-benzoyl-l-phenylalanine (pBpa) and p-azido-l-phenylalanine (pAzF), were genetically encoded in E. coli using a mutant of MjTyrRS (Figure 22a).^540,541 In order to extend the application of these probes to eukaryotes, starting from Sakamoto’s work,⁵⁹ the Schultz group evolved the E. coli tyrosyl-tRNA synthetase, along with a tRNA derived from Bacillus stearothermophilus, to create an EcTyrRS/BstRNA pair that can specifically incorporate pBpa and pAzF into proteins in S. cerevisiae(339) and CHO cells.^542,543 On the basis of the orthogonality of the PylRS/tRNA pair in both prokaryotic and eukaryotic systems, crosslinkable azido lysine derivatives have also been prepared.⁵⁴⁴ The versatility of the PylRS/tRNA pair was demonstrated by evolved variants capable of efficiently incorporating phenylalanine-derived photo-crosslinkers, such as pBpa and pAzF.⁵⁴⁵ With the possibility to choose between phenylalanine and lysine derivatives, different kinds of protein–protein interactions could be studied on the basis of their distinct chemical features. Phenylalanine-derived photo-crosslinkers generally possess rigid side chains and a crosslinking radius suitable for investigating short-range interactions (3–5 Å) and for mapping protein interaction interfaces. The lysine-derived photo-crosslinkers have a larger crosslinking radius of about 12–15 Å, which enables the capture of interacting partners at both long and short distances because of the flexibility of the side chain.

Figure 22.

Photo-crosslinking ncAAs. (a) Genetically encoded ncAA photo-crosslinkers. (b) Light-dependent formation of active species and targeted residue sites. (c) Relative photo-crosslinking efficiency at hν > 345 nm in terms of photoactivation efficiency and half-time of active species.

Photo-crosslinking ncAAs are also characterized by the type of photoreactive group they contain, which determines the irradiation wavelength and time necessary to efficiently generate reactive species. Simple phenyl azides require short-wavelength UV light (265–275 nm), known to cause molecular damage, to be efficiently activated.⁵⁴⁶ Longer wavelengths (e.g., 365 nm) could be used, but as this wavelength is at the tail of the absorption spectrum of azido group activation, the photo-crosslinking reaction is low in yield (<1%).⁵⁴⁷ Despite the low efficiency, photoactivation of pAzF at 365 nm has been used to generate the active nitrene, which inserts without prejudice into C–H and heteroatom–H bonds within an estimated radius of 3 Å. Despite this lack of specificity, genetically encoded azides have enabled short-distance and transient interactions to be probed. For example, by encoding pAzF into the G-protein-coupled corticotropin receptor (CRF1R), insights into the mechanism of receptor activation upon ligand binding were revealed in mammalian cells.⁵⁴⁸ The use of pAzF has also enabled mapping of the site in the N-terminal domain of the human glucagon-like peptide-1 receptor (GLP-1R) that binds the peptide exendin-4.⁵⁴⁹

As previously mentioned, pBpa has also been used to probe protein–protein interactions,^543,550,551 through the formation of a benzophenone ketyl radical upon excitation at 350–365 nm. Its ability to react with nearby C–H bonds is similar to pAzF, allowing the binding site of the neurokinin-1 receptor to be mapped⁵⁵² and elucidation of the mechanism of cardiac repolarization involving the tetrameric voltage-gated ion channel KCNQ1 upon interaction with its β-subunit, KCNE1.⁵⁵³ However, a study comparing pAzF and pBpa showed differences in photo-crosslinking results.⁵⁵⁴ Multiple residue mutations by site-specific GCE with both ncAAs demonstrated that not all the residues underwent cross-linking with the same profile. The distinct profiles might be attributed to the smaller size of the pAzF side chain in comparison to pBpa’s; that could potentially favor the identification of receptor positions more susceptible to structural perturbations. Recently, the use of pBpa analogs bearing an electron-withdrawing group in the para position, such as fluorine in 4-FpBpa (Figure 22a), showed improved crosslinking efficiency compared to the unsubstituted analog, allowing detection of weak transient interactions.⁵⁵⁵

To expand the repertoire of photoaffinity labeling, a caged fluorine-substituted tyrosine (FnbY) was synthesized by installing a fluorine atom at the β position of O-(2-nitrobenzyl)-l-tyrosine, furnishing a light-reactive species with higher crosslinking efficiency than pAzF due to the longer half-life of the quinone methide intermediate.⁵⁵⁶In vivo experiments were enabled in both E. coli and mammalian cells, in which photoactivation of FnbY showed selective bond formation with nine natural amino acid residues, covering the majority of nucleophilic residues (Figure 22b). The study demonstrated proximity-enabled reactivity of Trp, Arg, Met, Gln, and Asn in vivo. FnbY has also been proved to be a good candidate for protein labeling; the highly active quinone methide species generated upon UV-irradiation reacts fast with amine derivatives. With the use of an amine bearing an N-oxide radical, it was possible to achieve the shortest linkage between the spin label and the protein backbone, an important requirement for high resolution in electron paramagnetic resonance (see section 4.4.1).⁵⁵⁷

The trifluoromethylphenyl diazirine-containing photo-crosslinker 4′-[3-(trifluoromethyl)-3H-diazirin-3-yl]-l-phenylalanine (TfmdF) also showed higher crosslinking efficiency compared to pAzF and pBpa (Figure 22c).^558,559 An alkyl diazirine linked to lysine as a urea, 3-(3-methyl-3H-diazirine-3-yl)-propaminocarbonyl-N^ε-l-lysine (DiZPK) was used to determine protein–protein interactions⁵⁴⁴ by profiling the native client proteins of the acid-stress chaperone HdeA with the periplasmic chaperones DegP and SurA.⁵⁶⁰ Other analogs, para- and meta-trifluoromethyl diazirinyl lysine (pTmdZK and mTmdZK), having a carbamate linker, which can engage in long-range crosslinking (15 Å) were encoded in mammalian cells using an adenovirus-based incorporation system.³⁷² An aliphatic diazirine with increased flexibility compared to the previous analogs, AbK, was investigated with long-wavelength UV activation for probing kinase Cdk5 partners.^561,562 It is worth noting that RNA–protein interactions are difficult to study since the UV light can damage RNA strands. Nevertheless, light of wavelength greater than 345 nm and the short irradiation time (1 min) necessary to activate DiZPK were sufficient to achieve a maximal extent of RNA photo-crosslinking.⁵⁶³ For the same purpose, red-light activation was possible with a furan-lysine derivative, which enabled deeper tissue penetration, paving the way for discovering and mapping other transient protein–RNA interactions with spatiotemporal resolution.⁵⁶⁴ The recently developed 2-aryl-5-carboxytetrazole-lysine analogs and their use with GCE in E. coli and mammalian cells with an MmPylRS mutant afford new opportunities for studying transient protein–protein interactions and their interfaces in living cells.⁵⁶⁵ Indeed, by screening different heteroaryl groups as substituents on the tetrazole ring, N-methylpyrroletetrazole-lysine (mPyTK, Figure 22a) was identified as the best candidate for site-selective photo-crosslinking reactivity in E. coli; it also exhibited higher photo-crosslinking efficiency in vitro than the diazirine analogue AbK. Moreover, these next-generation photo-crosslinkers showed improved ligation selectivity, with no reactivity toward C–H bonds but instead with strong nucleophiles such as thiol, hydroxy, and amino groups (Figure 22b). More recently, the design of a 2-nitrobenzyl alcohol moiety appended to the side chain of lysine (o-NBAK, Figure 22a) allowed photo-activated protein crosslinking with significantly improved ligation selectivity. The photogenerated 2-nitrosobenzaldehyde intermediate reacted only with the amino group of proximal lysine residues, allowing stable protein conjugates to form around a dihydro-3H-indazol-3-one core.⁵⁶⁶ An unusual ncAA, featuring a 3-trifluoromethylphenyl amino group, was developed to yield a reactive acyl fluoride species upon light-irradiation (311 nm, for 5 minutes) (mTFMAK, Figure 22a).⁵⁶⁷ Besides the selectivity of the reactive species toward lysine, histidine, and cysteine, the cross-linked protein conjugates could be interestingly identified directly in the living system, thanks to the blue light emission from the cross-linked chemical moiety (Figure 22b). Even if short wavelength irradiation and blue light emission are not optimal, progress in these directions might improve cell survival and identify new cellular processes.

4.3. Bioorthogonally Reactive Amino Acids

Bioorthogonal amino acids genetically encoded into proteins represent a revolutionary concept in chemical biology, offering unparalleled precision and selectivity in modifying proteins. The term “bioorthogonal” refers to the ability of these noncanonical amino acids to react selectively with unique chemical partners in a living biological system. Out of all these methodologies, the bioorthogonal azide-alkyne click chemistry stands out for its simplicity, emerging as the most extensively employed approach across various fields, a breakthrough recognized by the award of the 2022 Nobel Prize in Chemistry to Carolyn Bertozzi,^568,569 Morten Meldal,⁵⁷⁰ and K. Barry Sharpless.⁵⁷¹ These reactions are ideally rapid and undergo seamlessly in one “click”⁵⁷² to efficiently form stable covalent bonds under physiological conditions; reactants and byproducts should be inert to the various chemical functionalities encountered in the intracellular milieu.⁵⁷³ Although diverse click chemistries have been applied to living systems, not all of these reactions are suited to GCE, a process that demands the enduring stability of the bioorthogonally reactive handle mounted on the ncAA throughout protein translation, and the ability of its reaction partner to effectively traverse the cellular membrane in sufficient concentrations. Indeed, bioorthogonal reactions follow mainly second-order kinetics, which depend largely on the concentrations of both reaction partners and the intrinsic second-order rate constant k₂ (M^–1s^–1) of the reaction itself.⁵⁷⁴ Rate constants vary from 10^–4 to 10⁶ M^–1s^–1, surpassing those of many enzymatic labeling approaches (Figure 23).⁵⁷⁵ Considering critical factors such as kinetics, chemoselectivity, and stability, not all bioorthogonal handles are suited to labeling proteins within cells or living organisms. Our review focuses on bioorthogonal reactions that have found practical applications in living systems.

Figure 23.

Bioorthogonal reactions applied to study proteins in vivo.

4.3.1. Azido ncAAs for Staudinger Ligation/Reduction

Azides were among the earliest compounds investigated as bioorthogonal reporters. They exhibit true orthogonality in the sense that they are completely absent from biological systems and show high reactivity. Several ncAAs containing azido groups have been introduced into proteins via GCE and used in various chemical reactions.⁵⁷⁶ However, it is worth noting that a potential limitation of using azides for labeling proteins is their often observed reduction to amines. Azido-functionalized proteins have been used to react with phosphines in Staudinger ligations.⁵⁷⁷ Despite their broad use in labeling biomolecules in living cells and animals,^578,579 their in vivo application in combination with GCE has been mostly limited to the labeling of cell-surface proteins due to the tendency of phosphines to oxidize. Although this technology has the potential to be a universal tool for live-cell imaging, where better understanding of biological process can be achieved in real-time and in their native environment, addressing the membrane permeability of phosphine dyes remains an outstanding challenge.⁵⁸⁰ Cell membrane proteins, such as G-protein-coupled receptors, were labeled using pAzF. The covalent conjugation with a triarylphosphine-conjugated FLAG peptide allowed for single-molecule detection experiments of low-abundance signaling proteins in their native cellular membranes by using fluorescently labeled antibodies.⁵⁸¹ Considering the large size of antibodies compared to the target proteins, the use of antibodies might be a suboptimal choice for investigating native intracellular conditions. Azido groups are not the only functional groups that can participate in Staudinger ligations with phosphines: cyclopropenones can also be used (Figure 24a,c).⁵⁸² In addition to bioorthogonal labeling, the reactivity between azides and phosphines has been exploited in vivo to trigger chemically induced decaging, known as the Staudinger reduction (Figure 24d). With this approach, an azide-based ncAA with improved activation kinetics in the presence of a phosphine (>10^–3 M^–1s^–1) provided fast chemical control of protein activation, localization, and SUMOylation, by reducing its azido group to an amino group.^583−585 In a recent study, this reaction to generate a lysine residue (Lys173) from PABK (Figure 24a) afforded temporal control of endogenous enzyme activity of VP4 protease in zebrafish embryos.⁵⁸⁶

Figure 24.

ncAAs for Staudinger reactions. (a) Genetically encoded ncAAs for Staudinger reaction. (b) Mechanism of Staudinger ligation to an azido-ncAA. (c) Mechanism of Staudinger ligation to a cyclopropenone ncAA. (d) Mechanism of Staudinger reduction of aromatic azides.

4.3.2. Azide and Alkyne ncAAs for Cu^I-Catalyzed Alkyne–Azide Cycloaddition

Azides can serve as multifunctional compounds given their reactivity and biophysical properties. Azides can also undergo [3 + 2] cycloadditions with terminal alkynes by Cu^I catalysis (Cu^I-catalyzed alkyne–azide cycloaddition, CuAAC).^576,587 Both azido- and alkyne-containing ncAAs have been developed for CuAAC and genetically encoded.^588,589 Using concise functional groups, such as azido or ethynyl, to create the smallest possible linkages (triazoles in this example) minimizes the perturbations that might be caused by introducing ncAAs. CuAAC reactions are known to proceed at faster rates than Staudinger ligations, especially in the presence of a substantial amount of Cu^I catalyst, which is toxic to cells at higher concentrations. However, the use of water-soluble ligands, such as 2-{4-[(bis{[1-(tert-butyl)-1H-1,2,3-triazol-4-yl]methyl}amino)methyl]-1H-1,2,3-triazol-1-yl}acetic acid (BTTAA),⁵⁹⁰ tris(hydroxypropyltriazolylmethyl)amine (THPTA), and 3-[4-({bis[(1-tert-butyl-1H-1,2,3-triazol-4-yl)methyl]amino}methyl)-1H-1,2,3-triazol-1-yl]propanol (BTTP), enhances the reaction rate and inhibits oxidation of the metal, meaning that lower Cu^I concentrations are sufficient; this significantly lowers the toxicity of these reactions. Copper-chelating azides have been shown to yield faster click reactions than non-chelating azides.⁵⁹¹ In particular, a screen of Cu^I ligands showed that copper concentration in the cytoplasm was lower with sulfated ligands, likely due to the reduced uptake of the complex across the cell membrane, and that BTTP and BTTAA gave the highest CuAAC reactivity inside living cells.⁵⁹² Protein labeling with small molecules combining GCE with CuAAC represents a reliable tool for super-resolution microscopy, which requires very small and bright fluorophores, both of which are generally not readily achieved with conventional fluorescent protein labels.⁵⁹³ Using a tris(triazolylmethyl)amine (BTTAA) Cu^I ligand tethered to a cell-penetrating peptide, the Cai group reported a copper-catalyzed click reaction inside live human cells using homopropargylglycine (HPG, Figure 25).⁵⁹⁴ The concomitant treatment with N-ethylmaleimide to target intracellular thiols furthered increased the reaction yield threefold on cytosolic proteins. The authors cautioned that the limited efficacy of intracellular click labeling arises from various factors. These include the abundance of endogenous copper-binding molecules, such as thiols, which can poison the catalyst; constraints on reagent concentrations due to low cell uptake and cytotoxicity; and the high degree of biomacromolecular crowding in the cytoplasm.

Figure 25.

Genetically encoded ncAA for CuAAC labeling in living cells.

The use of HPG encoding has also enabled an imaging approach for measuring newly synthesized proteins in rat hippocampal neurons⁵⁹⁵ and, more recently, for analyzing the dynamics of mitochondrial translation at the single-cell level.⁵⁹⁶ Furthermore, using a Cy5-conjugated chelating picolyl azide, and a propargylated lysine carbamate (ProK, Figure 25) as genetically encoded ncAA, dual labeling in living mammalian cells was performed by simultaneous CuAAC and inverse-electron-demand Diels–Alder (IEDDA) labeling reactions, using trans-cyclooctene lysine (2′-TCOK) as second ncAA. As a prerequisite for the dual GCE reaction, the PylRS/tRNA pair derived from the methanogenic archaeon ISO4-G1 was engineered to yield a system orthogonal to the MmPylRS/tRNA pair used in parallel.⁵⁹⁷ By using a derivative with an extended aliphatic chain, pentynyl lysine carbamate (PenK), which likely provides a more sterically accessible alkyne function, successful labeling of the membrane protein EGFR in mammalian cells⁵⁹⁸ and hepatitis D virus surface protein was achieved.⁵⁹⁹ Similarly, CuAAC bioorthogonal labeling of EGFR in living mammalian cells was facilited by BTTAA-assisted catalysis, using genetic encoding of the azide instead, namely N^ε-p-azidobenzyloxycarbonyl lysine (PABK).⁶⁰⁰ Recently, the first copper-chelating azido amino acid (PazK, Figure 25) to be genetically encoded into proteins by amber suppression was used to label proteins on the surface of mammalian cells.⁶⁰¹ Incorporating clickable ncAAs bearing alkynyl groups potentially enables labeling with a wide range of functional markers. The Summerer and Drescher groups, for instance, demonstrated protein spin labeling in living cells using two p-ethynylphenylalanines (pEnF; Figure 25) as encoded ncAAs, an N-oxide spin label, and BTTAA as a Cu^I ligand. Dual labeling was required to measure in cellulo EPR distances. Nevertheless, the stability of the nitroxide spin label remains a limiting factor for in vivo labeling, restricting its application primarily to abundant proteins.⁶⁰² Multiple novel ncAAs, characterized by a Tyr backbone and varying lengths of aliphatic chains featuring terminal alkyne and azide functional groups, have been genetically incorporated into E. coli. These new side chains enable not only the conventional CuAAC bioorthogonal reaction but also facilitate a copper-catalyzed coupling between two alkynes, commonly referred to as the Glaser–Hay reaction. Given the study’s focus on in vitro investigations, further exploration of potential applications for in vivo labeling is desired.⁶⁰³

4.3.3. Azido and Strained-Alkyne ncAAs for Strain-Promoted Alkyne–Azide Cycloaddition

The toxicity associated with copper(I) catalysts often hinders the application of CuAAC to living systems. Another approach to expedite alkyne–azide cycloaddition involves placing strain on the alkyne in a ring, leading to the term “strain-promoted alkyne–azide cycloaddition” (SPAAC). The rates of these reactions are typically in the range from 10^–2 to 1 M^–1s^–1.^569,604 Over recent decades, numerous cyclooctyne derivatives have been developed with enhanced reactivity, solubility, and the ability to cross cell membranes. Their compatibility with bioconjugation in living cells is particularly advantageous, as they eliminate the need for additional reagents.^605,606 To augment cyclooctyne reactivity toward nucleophiles, structural modifications such as fluorination, sp² hybridization of ring atoms, or fusion to cyclopropanes have been examined.⁶⁰⁴ By fluorinating the carbon atom adjacent to the alkyne in a cell-permeable coumarin–cyclooctyne conjugate, efficient intracellular protein labeling was achieved in mammalian Rat-1 fibroblasts by the Tirrell group. The labeling process consisted of residue-specific labeling with AHA as a competing substrate of MetRS, achieved through pulse labeling with AHA. The efficiency of cycloaddition of difluorinated cyclooctyne (DIFO) analogs to AHA was higher than with the unsubstituted strained cyclooctyne (SCO) (Figure 26a).⁶⁰⁷ Copper-free SPAAC chemistry was, thus, used to improve the low reaction stoichiometries of the aforementioned Staudinger ligation⁵⁸¹ by site-specifically encoding pAzF into the human G-protein-coupled chemokine receptor CCR5.⁶⁰⁸

Figure 26.

(a) Genetically encoded azide ncAAs for SPAAC reactions and corresponding reactivity with strained alkynes. (b) Genetically encoded strained alkynes ncAAs for SPAAC reaction and the use of azides as external labeling handles.

The dibenzocyclooctynes (DBCOs) are highly reactive toward azides, with kinetics more amenable to labeling in living cells.⁶⁰⁴ However, DBCO derivatives have rather large and inflexible structures, and even if constraints result in increased reactivity compared to other strained alkyne (Figure 26), such features are less compatible with the binding pockets of synthetases. Because of these challenges in genetically encoding a DBCO-derivatized ncAA, DBCOs have been mainly used as external bioorthogonal handles for SPAAC reactions, principally for surface protein labeling. In 2014, the Petersson group used this strategy to investigate the effects of decreasing the aminoacyl transferase (AaT) activity inside E. coli through the site-specific encoding of pAzF. Since the bioorthogonal labeling reaction was necessary to efficiently inhibit enzyme activity, a cell-permeable rhodamine DBCO_TMR was applied to ensure sufficient labeling and subsequent protein inhibition.⁶⁰⁹

Despite the lower reactivity of cyclooctynes, such as SCO, compared to DBCO derivatives, they exhibit higher flexibility and solubility and thus represent better candidates to form the side chains of new ncAAs. In 2011, SCO-lysine was used for site-specific GCE labeling reactions with fluorogenic azides in E. coli (Figure 26b).⁶¹⁰ An advantage of incorporating azido and other nitrogen-rich functional groups, such as tetrazine, into fluorophores is that they serve as fluorescence quenchers (partial or total quenching varies case by case) before reacting with their respective bioorthogonal handles. Quenching can effectively reduce the background noise resulting from an unreacted excess of dyes.⁶¹⁰ The application has further been extended to living mammalian cells for labeling large and complex membrane proteins with cell-impermeable dyes.⁶¹¹ The bicyclo[6.1.0]non-4-ynyl lysine carbamate (BCNK), which exhibits superior reactivity and stability compared to SCOK because of the increased strain imparted by the fused cyclopropane and the absence of the oxygen group on the 3 position of the cyclooctyne, was developed and genetically encoded in both E. coli and mammalian cells by Lemke^347,612 and Chin⁶¹³ and extensively used as an optimal ncAA for SPAAC labeling.⁶¹⁴

4.3.4. Nitrile ncAAs for Nitrile–Aminothiol Click Reactions

Over the past decade, the condensation of 1,2-aminothiol moieties with cyano groups has been a valuable technique for labeling proteins in vitro.⁶¹⁵ Despite their favorable reaction rate constants (approximately 10 M^–1 s^–1) and without the need for a catalyst, their application in living cells has been generally restricted due to the tendency of 1,2-aminothiols to form adducts with intracellular metabolites. Despite the successful genetic encoding of the 1,2-aminothiol moiety, its reactivity with the abundant pyruvate found in biological systems prevents efficient bioorthogonal labeling. The genetic encoding of a caged 1,2-aminothiol functionality further limited its in vivo applications due to the need for chemical deprotection (Figure 27a,b).⁶¹⁶ Recent developments by Huber et al. paved the way for in vivo applications. Swapping the bioorthogonal handles and encoding the cyano group into proteins, either with meta-cyanopyridylalanine (mCNF) or para-cyanopyridylalanine (pCNF), set up a spontaneous reaction between the nitrile moiety and the bioaminothiol inside cells, the nitrile-aminothiol (NAT) click reaction; this achieved protein macrocyclization and protein stapling (Figure 27a,c).⁶¹⁷

Figure 27.

The nitrile–aminothiol (NAT) condensation. (A) Genetic encoded ncAAs for NAT condensation. (B) Genetic encoding of aminothiols for in vitro condensation. (C) Genetic encoding of a nitrile for inter- or intramolecular condensation in living cells.

4.3.5. Alkene ncAAs for “Photoclick” Cycloadditions

The light-induced coupling of tetrazoles and terminal alkenes to form fluorescent pyrazoline cycloadducts is an example of a 1,3-dipolar cycloaddition that offers excellent biocompatibility for in vivo applications.⁶¹⁸ This photo-triggered click chemistry⁶¹⁹ involves two steps: first, photolysis of the tetrazole moiety generates its corresponding nitrilimine, which proceeds with a first-order rate constant of approximately 0.1 s^–1; second, an alkene cycloaddition proceeds at a second-order rate constant of 10–60 M^–1 s^–1. The resulting photoclicked adduct not only facilitates site-specific labeling of target molecules but also emits fluorescence in the cyan–green spectral range, allowing the reaction to be monitored (Figure 28b). O-Allyltyrosine, encoded in E. coli,^620,621 and homoallylglycine (HAG), encoded in mammalian cells by human methionyl-tRNA synthetase in a methionine-deficient culture, served as bioorthogonal reporters of newly synthesized proteins upon photoclick cycloadditions (Figure 28a).⁶²² The modification of lysine with a pendant alkene broadens the applicability of this method to eukaryotic systems, offering enhanced flexibility for alkene-based modifications.

Figure 28.

ncAAs for photoclick cycloadditions. (A) Genetically encoded alkene-bearing ncAAs for photoclick cycloadditions. (B) Mechanism of photo-activation of the tetrazole ring. (C) Reactivity of alkenes as a function of substituents and strain, in order of increasing reactivity.

Strained alkenes, such as cyclopropane lysine (CpK), yielded faster bioorthogonal reactions than the terminal alkenes, as discussed above (Figure 28c).⁶²³ In addition, the incorporation of acryl lysine (AcrK) has expanded the scope of photoclick chemistry, for example, in the model plant organism, A. thaliana.⁶²⁴ Encoding of the tetrazole function in mammalian cells enabled protein lipidation in cells with spatio-temporal control.⁶²⁵ By contrast, the use of a strained alkene, spiro[2.3]hex-1-ene, encoded as a lysine analog (SphK) gave faster reaction rates than AcrK.⁶²⁶ However, since photoactivation generates intermediates with short half-lives, influencing crosslinking efficiency, a sterically shielded tetrazole was shown to stabilize the nitrile-imine intermediate and increase its half-life (102 s in aqueous media). However, the main drawback of this strategy is the need to activate the tetrazole moiety with UV light in the 300–365 nm range, which is harmful to living cells. Naphthalene-based tetrazoles were used in combination with genetically encoded AcrK. They have been used to image microtubules in mammalian cells through an efficient two-photon excitation with a 700 nm femtosecond pulsed laser.⁶²³

4.3.6. Alkene/Alkyne and Tetrazine ncAAs for Inverse-Electron-Demand Diels–Alder Reactions

The IEDDA [4 + 2] cycloaddition, involving the reaction between strained alkenes or alkynes, such as norbornene, cyclopropene, bicyclononyne, and trans-cyclooctene derivatives, with tetrazines offer several advantages.^573,627,628 These reactions do not require additional reagents, release only inert and harmless N₂ gas, and are exceptionally fast. Other electron-deficient dienes, such as substituted triazines, can be used, but are less reactive.⁶²⁹ The reaction proceeds via suprafacial/suprafacial interaction of 4π electrons of the diene with the 2π electrons of the dienophile. A key feature of IEDDA reactions is that, in contrast to the classical electron-demand Diels–Alder reaction, in which an electron-rich diene reacts with an electron-deficient dienophile, an electron-rich dienophile reacts with an electron-deficient diene, which is a result of the corresponding energy gap between the highest occupied molecular orbital (HOMO) and the lowest unoccupied molecular orbital (LUMO) of the reactants (Figure 29b). Even faster IEDDA reactions can be achieved by decreasing the energy differences between HOMO_dienophile and LUMO_diene. Electronic, steric, strain, stereochemical steric effect of the substrates, as well as solvent and pH effects interplay to impact the kinetics of IEDDA reactions, with kinetics spanning from 0.1 to 10⁶ M^–1 s^–1.⁵⁷³ The tetrazine moiety, much like the azido group, can serve as a bioorthogonal handle and has the potential to quench fluorescent dyes covalently linked to it. Fluorescence is usually restored after the IEDDA reaction between the strained alkene/alkyne and the tetrazine, as the resulting dihydropyridazines or pyridazines lack quenching properties. However, the efficiency of this quenching effect can vary between in vitro and in cellulo experiments. Background noise is a common issue in microscopy experiments on living cells, and washing steps do not always completely remove excess dye, which can then adhere to hydrophobic compartments such as mitochondrial membranes.

Figure 29.

ncAAs for IEEDA reactions. (a) Genetically encoded alkenes ncAAs for IEDDA. (b) Frontier molecular orbital of classical and inverse-electron-demand Diels–Alder reactions. (c) Reactivity of alkene and alkyne ncAAs with tetrazines. (d) Fast TCO decaging of a protein using a specially designed tetrazine. (e) Genetically encoded tetrazine-functionalized ncAAs. (f) Reactivity of tetrazine-functionalized ncAAs with TCO.

In 2012, the genetically encoded dienophiles NorbK1 and NorbK2 were reported (Figure 29a), having norbornene as a key group linked to the amino group of lysine by a carbamate bond.^347,630,631 With the use of the MmPylRS/tRNA pair, NorbK1 was incorporated into EGFR in mammalian cells and in Drosophila melanogaster.⁶³² Efficient cell-surface labeling was observed because of the use of a tetrazine-functionalized fluorophore, with the click reaction giving at least a sixfold increase in fluorescence intensity.^347,631 By evolving the MmPylRS/tRNA_CUA pair to become more promiscuous, norbornene derivatives with different chain lengths, stereochemistries, and strain properties were successfully encoded in E. coli by the Carell lab.⁶³³ Considering the outstanding research of the Bertozzi group in click conjugation of glucosides on membrane proteins^{578,634−636} with bioorthogonal reactions involving the insertion of the reactive handle on the glucoside function, a GCE approach has also been developed for visualizing the cell-surface display of mucins.⁶³⁷ Besides lysine derivatives, a class of norbornene-containing tyrosines showed reactivity toward not only tetrazines (kinetics > 10⁵ M^–1s^–1) but also to diaryl nitrilimines (a bioorthogonal NAT reaction, kinetics at 20 M^–1 s^–1). In particular, the better solubility of the NorbY derivative (Figure 29a) might have positively contributed to enhance GCE efficiency.⁶³⁸

Increasing the ring-strain of an alkene by using a cyclopropene also means reducing the size of the bioorthogonal product, an asset for biological labeling where size matters, such as for super-resolution microscopy. Although the reactivity of cyclopropenes is rather slow compared to other strained alkenes (Figure 29c), they still undergo click chemistry inside living cells within a single hour.⁶³⁹ The Chin group synthesized a 1,3-disubstituted cyclopropene ncAA, by modifying lysine derivative with a carbamate linker.⁴⁰⁰ Genetic encoding of CpK could be expanded to animals by co-translational CpK-tagging of proteomes in response to diverse codons in genetically targeted cells⁴⁰¹ in the mouse brain²⁸ and in human induced pluripotent cells.³⁷⁵ In order to further increase reactivity, the synthesis of trans-cyclooctene (4′-TCO), by light-induced isomerization of the cis-cyclooctene,⁶⁴⁰ enabled the synthesis of a lysine derivative (4′-TCOK) with an ultrafast bioorthogonal handle, with kinetic constants of up to 10⁶ M^–1 s^–1 when used in combination with tetrazines. In 2012, 4′-TCOK was firstly reported for site-specific GCE in mammalian cells using an evolved MmPylRS.³⁴⁷ However, a common drawback of 4′-TCOK is its instability, as it isomerizes to the cis configuration, induced by the presence of high thiol concentrations, possibly via a free-radical-mediated pathway.⁶⁴¹ Since the cis-form is not sufficiently reactive, click reactions might not occur efficiently.⁶¹³ With the purpose to optimize and prevent the trans-to-cis isomerization, other TCO derivatives have been prepared, positioning the carbamate linkage elsewhere on the cyclooctene^{348,642−644} and inserting a methylene bridge.⁶⁴⁵ These derivatives, substituted at the allylic position (2′-TCOK), provided higher stability probably due to the carbamate bond shielding effect. For this reason, 2′-TCOK (or TCO*K) was evaluated for live-cell labeling.⁶¹⁴ However, because 2′-TCOK exists as two atropisomers, with the carbamate substitution on the equatorial plane of the eight-membered cycle (2′-eTCOK or TCO*eK) or on the axial plane (2′-aTCOK or TCO*aK), both were further separated and investigated in detail. The results showed that 2′-aTCOK (k = 35 900 ± 420 M^–1 s^–1) had not only a much higher rate constant than 2′-eTCOK (k = 12 080 ± 150 M^–1 s^–1) in the IEDDA reaction but also increased stability toward thiols compared to 2′-eTCOK and any other TCO derivatives.⁶⁴⁴ 2′-TCOK can be used as a mixture of atropisomers,⁶⁴⁵ but because reactivity is greater in 2′-aTCOK, genetic encoding of this specific atropisomer and subsequent labeling is more efficient. 2′-aTCOK was used to label and visualize neuronal proteins^646,647 and also in “DNA-PAINT” technology.⁶⁴⁸ Specifically, the “Click-PAINT” approach enables high-contrast, residue-specific imaging of even low-abundance proteins using super-resolution microscopy.³⁴⁸ Despite its widespread use, removal of excess TCOK from cells can be challenging and requires rigorous washing. In some cases, TCOK is better suited to labeling surface proteins than to intracellular applications.³⁴⁶ In order to overcome what is a solubility issue, a more hydrophilic analog of TCOK has been synthesized to prevent its relatively sticky properties. The insertion of other functionalities into the cyclooctene ring, yielding DOTCOK, improved its solubility. In a comparison study, the time required to remove excess DOTCOK from cells was 72 times lower than for BCNK and 2′-TCOK. DOTCOK encoding in both prokaryotes and eukaryotes was used to label and track low-abundance intracellular proteins without background noise from free ncAAs.⁶⁴⁹ A recent study examined the stability, reactivity, and suitability in live-cell labeling of several TCOK derivatives and BCNK. In addition to those already discussed (2′-aTCOK, 2′-eTCOK, SCOK, and BCNK), new analogs were syntheized containing a urea linkage: eAmTCOK, aAmTCOK, and TCONK.⁶⁵⁰ The study revealed that in reaction with a tetrazine dye, eAmTCOK had the best rate constant (20000 M^–1 s^–1), but the product had one of the lowest maximum FRET fluorescence intensity (E_FRET-MAX), a parameter used to demonstrate the in vivo stability of the ncAA or its corresponding bioorthogonal conjugate. BCNK and 2′-aTCOK instead showed similar k values (10000 M^–1 s^–1), similar E_FRET-MAX, and no significant loss of fluorescence during the measurement window, demonstrating their comparable features and suitability in live-cell labeling.⁶⁵⁰ Tyrosine derivatives were also prepared in order to host TCO moieties and were encoded in E. coli by means of an engineered PylRS/tRNA pair.⁶³⁸

Even if TCOK has been proved to be a good candidate in live-cell labeling, in vivo isomerization and formation of unstable dihydropyrazine upon click reaction, might be influenced by parameters, e.g., labeling conditions and time required for labeling, usually longer times leads to TCOK degradation to Lys. As we have seen above for the Staudinger reduction (see section 4.3.1) and as we will seen in Pd-catalyzed reactions (next section), TCO as well as azido, propargylic, and allylic groups can undergo chemical-induced degradation, when inserted on the heteroatom of amino acid residues directly or through a proper self-immolative linker, like carbamate bonds. For instance, this phenomenon can be exploited to activate protein function or release a drug in its active form. From a chemical point of view and regarding in particular TCO, the presence of an electron-withdrawing group on the tetrazine is necessary to drive the IEDDA reaction, while for the following release step the presence of at least one electron-donating group is necessary to improve the deprotection yields (Figure 29d).⁶⁵¹ The deprotection occurs by rearrangement of the 2,4a,5,6,7,8,9,10-octahydrocycloocta[d]pyridazin-5-yl intermediate and by eliminating CO₂, with subsequent generation of the free amino group of lysine. That event is not desired for bioorthogonal fluorescence labeling, but it becomes useful if TCO is to be used to mask protein function. Indeed, if one wishes to activate a protein or study the function of a particular residue or site, combining TCOK with external treatment with an electron-rich tetrazine might provide a chemically induced method for doing so.^{586,652−654}

Despite the faster reaction kinetics of strained cyclic alkenes, terminal acyclic alkenes can also be exploited for IEDDA reactions (Figure 29a). Their low reaction rates are comparable to SPAAC reactions (≈10^–2 M^–1 s^–1), but their small size and stability make them favorable to use if fast labeling is not the priority. In 2015, several ncAAs bearing a terminal alkene function were investigated for bioorthogonal reactivity and encoded in E. coli using a PylRS mutant. Those containing an allylic oxygen or an α,β-unsaturated amide group showed higher reactivity than aliphatic alkene moieties. However, in studies using AlkY and AlkK (Figure 29a), even the labeling of surface proteins required quite long incubation times of around eight hours for efficient labeling.^655,656

Despite the lower kinetic constants of cyclooctynes compared to trans-cyclooctenes, the formation of a stable conjugate and single isomer upon tetrazine ligation makes cyclooctynes, particularly BCN, the best candidates for constructing homogeneous protein conjugates. Whereas both have already been discussed with regard to SPAAC reactions, their application in IEDDA reactions has increased their utility in living systems due to their faster kinetics when coupled with the electron-rich tetrazine.^347,599,613 Notably, a computational study suggested that the reaction of SCO with certain tetrazines might instead be a classical Diels–Alder reaction.⁶⁵⁷

SCOK has been labeled with a nitroxide radical as an EPR probe.^658,659 SCO derivatives of tyrosine were also prepared with different hydrophilic properties and encoded in E. coli using a mutated PylRS/tRNA pair. The ability of SCO to react with different nucleophiles, such as azides and tetrazine, was exploited for nitrilimine –alkyne cycloadditions along with IEDDA.⁶³⁸ Tsai et al. used BCNK as a bioorthogonal handle encoded into a kinase to react with an inhibitor equipped with a tetrazine moiety and an azo group for photocontrol. Cycloaddition of BCNK to the tetrazine brought the inhibitor into proximity with the kinase inhibition site; photo-isomerization of the azobenzene gave spatiotemporal control over inhibition of the mitogen-activated protein kinase (MEK) in mammalian cells.⁶⁶⁰ BCNK exists as two stereochemical forms, that is, endo and exo-BCNK. The endo form is more reactive and thus usually the preferred choice for labeling proteins in vivo.⁶¹¹ With this bioorthogonal handle in hand, which forms a stable adduct with fluorescent dye-bearing tetrazines, encoding of BCNK into short epitope tag, HA,⁶⁶¹ EGFR proteins, prototypic Kv channel, Shaker B,⁶⁶² and α-tubulin enabled real-time visualization of single molecules for localization and tracking.⁶⁶³

An alternative strategy to limit in vivo isomerization of TCO ncAAs involves encoding the tetrazine moiety as a bioorthogonal handle. This approach prevents TCO ncAAs’ prolonged exposure to free thiols in the cellular environment during protein expression (Figure 29e). In 2012, the Mehl group developed the first encoded tetrazine-based ncAA using a MjTyrRS/tRNA pair.⁶⁶⁴ The tetrazine featured an amino group as an electron donor in order to increase its stability against nucleophilic attack by thiols. The in vivo kinetic rate constant with external 4’-TCO feeding was measured at 330 M^–1 s^–1. However, it is known that electron donor groups decrease the IEDDA reactivity of tetrazines. To overcome this, new tetrazine-based ncAAs were developed containing a tetrazine moiety attached directly to the aromatic ring of a phenylalanine-derived ncAA (Tet-v2.0, Figure 29f). The ultrafast reaction (k = 7.8 × 10⁵ M^–1 s^–1) consumed the labeling reagent from the cell medium within minutes, eliminating the need to wash the cells before imaging.⁶⁶⁵ These outstanding reaction kinetics were further improved with the derivative Tet-v4.0Py, turning these ncAAs into suitable candidates for use with EPR probe labeling (see section 4.4.1), which requires fast, high-yielding labeling and minimal unused spin probe (Figure 29f).⁶⁶⁶ The evolution of a PylRS/tRNA pair that recognizes Tet-v4.0 ncAAs extends the utility of the fastest site-specific bioorthogonal labeling method known to date to eukaryotic systems.⁶⁶⁷

Nevertheless, the advantage to keep the tetrazine anchored to an external dye and not as a genetically encoded bioorthogonal reaction partner resides in its ability to fully/partially quench the fluorescence of the dye. Chemists are actively working to develop tetrazine-linked dyes that remain nonfluorescent in their unreacted form but exhibit high fluorescence upon reaction, offering a powerful tool for super-resolution microscopy, particularly for imaging less abundant and dynamic proteins.^{387,668−672} By significantly reducing background noise, the fluorescence signals can be attributed almost exclusively to the site-specifically labeled protein of interest. Therefore, whereas tetrazines have been genetically encoded and retain utility for other bioorthogonal functionalizations, the genetic encoding of strained alkenes/alkynes remains a more advantageous strategy for fluorescence microscopy studies.

4.3.7. ncAAs for Palladium-Catalyzed Reactions

Of the metal-catalyzed reactions for forming C–C bonds, Pd-catalyzed reactions exhibit several advantages for use in biological studies: (1) Pd is not naturally found in living systems, (2) Pd-catalyzed cross-coupling reactions demonstrate exceptional chemoselectivity, (3) Pd–ligand complexes can be custom-designed for cell permeability without significant cytotoxicity, and (4) reactions involving Pd are typically orthogonal to other bioorthogonal reactions. Iodoaryl group or alkynyl moieties have been installed on ncAAs to serve as reactive groups for various cross-coupling reactions, including Suzuki–Miyaura and Sonogashira reactions.^673−675 Notably, utilizing the water-soluble catalyst aminopyrimidine–palladium(II), a Sonogashira reaction was successfully performed inside living E. coli (Figure 30a). The reaction involved the genetic encoding of HPG into ubiquitin, and several exogenous aryl iodides, including an iodocoumarin derivative were evaluated.⁶⁷³ Although the catalyst did not ellicit cytotoxic effects at low concentrations, its low cell permeability necessitated high loading concentrations, limiting applications to labeling of cell-surface proteins.⁶⁷⁶ In order to extend the scope of Pd-catalyzed labeling, reactions inside cells were optimized by using the simple ligand-free Pd catalyst Pd(NO₃)₂. Its enhanced reactivity in comparison to other Pd salts and coordination complexes enabled the visualization of the type III secretion toxin OspF in Shigella cells with kinetics similar to copper-catalyzed reactions, but with significantly lower cytotoxicity.⁶⁷⁷ The use of two Pyl analogues bearing both alkynyl and iodoaryl bioorthogonal handles allowed a comparison of their reactivity toward exogenous iodoaryl and alkynyl compounds, respectively. Two years later, the concomitant use of PylRS for encoding these ncAAs extended the scope of cross-coupling reactions to mammalian cell surfaces.⁶⁷⁸ Pd-catalyzed reactions not only facilitate bioorthogonal coupling but also induce the decaging of specific substrates, such as allyl and propargylic groups (Figure 30b).⁶⁷⁹ This approach allows researchers to monitor protein functions similarly to photolytic decaging and chemical decaging of TCO.⁶⁸⁰ Through this gain-of-function strategy, allylic and propargylic pendants situated on heteroatoms can be readily cleaved by palladium, via the formation of a π–allyl intermediate, which typically undergoes elimination at the heteroatom–alkyl site during reductive elimination (Figure 30c). By functionalizing lysine residues harboring allyl or propargyl carbamates groups and genetically encoding them in various mammalian cell lines, it was observed that both double and triple C–C bonds exhibited similar reactivity in the presence of allyl₂Pd₂Cl₂ and Pd(dba)₂, without the need for additional ligand. However, when analyzing Pd uptake in various cellular compartments, uptake in the nucleus was at least fivefold lower than in the cytosol, limiting studies to cytosolic proteins, with reaction yields inside cells of 32% (calculated afterwards by quantifying the unreacted alkyne by CuAAC labeling).⁶⁸¹ With the purpose to improve the reactivity inside cells, an allenyl moiety (AlleY, Figure 30b) was installed on the phenolic oxygen atom of a Tyr and genetically encoded using the PylRS/tRNA pair as a way to study enzymatic functions and post-translational Tyr modification sites, masking the active site of proteins, including GFP and Src kinase. The allene group played a key role for functionality, since an allenyl species was postulated to be formed in a rate-limiting step during Pd-mediated depropargylation. Specifically, by using allene moieties, twofold higher yields were obtained during the deprotection compared to the propargyl derivative.⁶⁸²

Figure 30.

ncAAs for Pd-catalyzed reactions. (a) Genetically encoded ncAAs for Pd-catalyzed cross-coupling reactions. (b) Genetically encoded ncAAs for Pd-catalyzed decaging. (c) Mechanism for activating protein function (left) and catalytic cycle of decaging (right).

4.3.8. Electrophilic ncAAs for Chemical Crosslinking

While we have observed that the genetic incorporation of photo-cross-linkers enables the spatiotemporal control in the study of protein–protein interactions,⁶⁸³ UV-irradiation often generates highly reactive species that can form covalent bonds with carbonatoms or heteroatoms on unspecific amino acid residues. These challenges, including low yields due to potential quenching of highly reactive intermediates by water and other nucleophiles, have driven researchers to develop more sophisticated functionalities that spontaneously undergo chemical crosslinking through proximity-enabled reactions with adjacent nucleophiles (Figure 31a). The core concept of this approach is based on cross-linking reactivity occurring exclusively if effective local concentrations of reactants reach the order of 10⁵ M. Then, the high selectivity for proximal nucleophilic residues can capture transient protein–protein interactions in living cells (Figure 31b).^684,685 At first, simple alkyl halides were studied as crosslinker moieties. A physiologically stable fluoroacetamide tyrosine was site-specifically incorporated into the corticotropin release factor receptor type 1 (CRF-R1) in E. coli and mammalian cells by a EcTyrRS/BstRNA mutant. Its S_N2 reactivity toward nearby cysteine residues provided a map of the interface between CRF-R1 and its urocortin ligand.⁶⁸⁶ By using the more flexible derivatives of lysine, for example, the chloroacetamide (ClAcK),⁶⁸⁷ and by further elongation of alkyl halide chain (FSK, BrC6K, and BrC7K, Figure 31a), efficient inter- and intramolecular crosslinking of proteins was achieved, along with direct stapling of a protein α-helix and covalent binding of native membrane receptors in live cells.⁶⁸⁸ Bromo derivatives (BrC6K and BrC7K) expanded the diversity of reacting partners because of their improved reactivity toward histidine and lysine. Several derivatives of the same structure, with variable aliphatic chain length, were used to stabilize low-affinity interactions between Rab1b, GDP, and DrrA in bacterial and mammalian cells.⁶⁸⁹ Synthesis of three aza-Michael acceptor amino acids, N^ε-acryloyl-l-lysine (AcrK), p-acrylamido-l-phenylalanine (AcrF), and p-vinyl-sulfonamido-l-phenylalanine (VSF), were encoded into a fragment of the antibody drug trastuzumab (trastuzumab Fab) by either MbPylRS or MjTyrRS in response to the amber stop codon in E. coli. Their different abilities to react toward nucleophiles was evaluated and used to prove the antibody–receptor interaction by formation of covalent bonds between the aza-Michael acceptor encoded into the trastuzumab Fab mutant and the lysine amino group on the surface of ErbB2. Of the three ncAAs, only the vinyl sulfonamide resulted in rapid covalent crosslinking of the trastuzumab Fab–ErbB2 complex at physiological pH with >95% yield, whereas AcrF and AcrK resulted in modest crosslinking yields under basic conditions.⁶⁹⁰ In order to expand the variety of proteins amenable to crosslinking, the fluorosulfate group has been developed as a new functionality. Fluorosulfate esters are biocompatible and able to react with several residues, including proximal lysine, histidine, and tyrosine. Fluorosulfate-l-tyrosine (FSY) has been applied for GCE in E. coli via an evolved PylRS/tRNA pair, generating covalent inter- and intraprotein bridges by undergoing sulfur–fluoride exchange reactions in vivo.⁶⁹¹ By engineering the target protein with FSY at an appropriate distance from a target amino acid residue, it was possible to target Ser and Thr and exploit the hydroxy function as a nucleophile. Harnessing the sulfur–fluoride exchange approach, the generated unstable arylsulfate–Ser/Thr intermediate yielded the formation of dehydroalanine (Dha) and dehydrobutyrine (Dhb) in situ through partial hydrolysis, affording a substrate for a radical reaction with nucleophiles of interest (Figure 31c).⁶⁹² However, the relatively rigid nature of FSY limits its reactivity toward more hindered sites, and a more flexibile ncAA structure was required; this need was fulfilled by fluorosulfonyloxybenzoyl-l-lysine (FSK). FSK has been encoded in E. coli and mammalian cells, and by making use of multiple crosslinking ncAAs with different flexibility and reactivity, interacting positions of membrane receptor proteins have been mapped, including those of the epidermal growth factor receptors (EGFRs).⁶⁹³ A similar approach with several crosslinking ncAAs (including BprY, FSY, and FnbY) allowed mapping and identification of more than 250 peptides crosslinking to thioredoxin in vivo of the 205 residues involved in the binding.⁶⁹⁴ Alkyl sulfonyl fluorides, such as FSY and its derivative SFY, have also proved to be useful for studying more challenging interactions, such as those between proteins and RNA.⁶⁹⁵ The ability to perform the study in vivo enabled an N⁶-methyladenosine to be identified on RNA with single-nucleotide resolution; this modified nucleobase was shown to interact with the FSY encoded into the chaperone Hfq in E. coli. A lysine derivative, with similar features to mPyTK, described in section 4.2.3, but having a triazole instead of the tetrazole, showed in this case spontaneous crosslinking, without need of photoactivation.⁶⁹⁶ Selectivity toward lysine and tyrosine was observed, and thus, urea and carbamate bonds were formed after crosslinking and release of the 4-aryl-1,2,3-triazole (CATK, Figure 31). CATK also displayed increased reactivity compared to FSY, when reacting with vicinal Lys and Tyr at neutral pH. More recently, N⁶-{[2-(vinylthio)ethoxy]carbonyl}-l-lysine (VtK, Figure 31), which features a latent cross-linker was developed to respond to endogenous or exogenous oxidating species, such as photocatalyst and ROS. Acting as stimuli-responsive ncAA (see section 4.5), Vtk, upon activation, generates a vinyl sulfoxide, a Michael acceptor, which reacts with proximal nucleophiles such as Cys and Lys. With this ncAA in hand, short-transient PPIs of thioredoxin, which play an essential role in the redox regulation of mammalian cells through their highly conserved Cys residues, were trapped.⁶⁹⁷

Figure 31.

ncAA for crosslinking reactions. (a) Genetically encoded crosslinker ncAAs. (b) Proximity-induced crosslinking with a nucleophilic residue (inter- or intramolecular). (c) Dha/Dhb species generated in live cells by crosslinking.

In summary, bioorthogonal chemistry has revolutionized bioimaging techniques, enabling the creation of fluorescently labeled proteins or probes and illuminating complex biological structures and pathways in real-time with exceptional sensitivity and resolution. With the discovery of new ncAAs and bioorthogonal probes, new aaRS/tRNA pairs, and improved genetic engineering techniques, bioorthogonal chemistry will continue to make vital contributions to various fields of research.

4.4. Spectroscopic Probes

4.4.1. Non-Fluorescence-Based Probes

Incorporating genetically encoded ncAAs into proteins has afforded new opportunities to investigate protein structure, dynamics, and interactions using various spectroscopic techniques. Infrared (IR) spectroscopy gives us insights into the secondary structures of polypeptides, for example, α-helices, β-sheets, and random coils, by analyzing the vibrational modes of amide bonds.⁶⁹⁸ However, when applied to more complex systems such as whole proteins, various interactions, including inter- and intramolecular interactions, vibrational couplings, and energy degeneracies, complicate the use of these measurements to provide site-specific information. To address this challenge, genetically encoded ncAAs with finely tuned IR absorption bands provide a powerful approach to obtaining site-specific structural information (Figure 32a). Replacing native amino acids with noncanonical residues can enhance the specificity and precision of IR spectroscopic measurements in complex biological systems. With the introduction of site-specific ncAAs encoding, the Boxer lab⁶⁹⁹ in 2003 delineated four key properties that an IR probe should possess in order to provide information on the local or even wider protein environment: (1) it should occupy an uncongested region of the IR spectrum, as spectral overlap and vibrational couplings can complicate the interpretation of the spectrum; (2) its vibration extinction coefficient should be high; (3) it should feature good sensitivity to the local environment and a strong transition dipole moment;⁷⁰⁰ and (4) it should be stable in buffers. Thus, ncAAs functionalized with cyano,^701,702 azido,⁷⁰³ nitro,⁷⁰⁴ and/or carbonyl functional groups, all with characteristic vibrational stretchings, have been used, mostly in vitro, to probe the hydrophobicity or hydrophilicity of molecular environments in proteins of interest, with, in particular, the cyano group being the most promising candidate in terms of vibrational properties and tunability. More recently, a noninvasive mid-infrared metabolic imaging technique used azidohomoalanine (AHA) in combination with ¹³C- and ¹H-labeled amino acids to interrogate metabolic activities in cells, C. elegans, P. aeruginosa, and mice.⁷⁰⁵

Figure 32.

ncAAs used as probes for spectroscopy. (a) Genetically encoded IR probes for mainly in vitro applications. (b) Genetically encoded NMR probes for in vivo NMR. (c) Genetically encoded EPR probe for in vivo measurements.

Incorporating isotopes into ncAAs enables investigation of protein structure and dynamics, providing information on atomic-level interactions using nuclear magnetic resonance (NMR) spectroscopy.^706,707 In particular, ¹⁹F is a naturally abundant isotope with a stable and NMR-active nucleus and is often used as a reporter atom in NMR studies. ncAAs bearing fluorine atoms are ideal bioorthogonal probes for studying biomolecular interactions since fluorine atoms are extremely rare in biological systems.⁷⁰⁸ Considering that the size of the fluorine atom is comparable to hydrogen, many endogenous synthetases can incorporate fluoro-containing amino acids (Figure 32b), such as p-fluorophenylalanine (pFF), m-fluorotyrosine (mFY), or 6-fluorotryptophan (6FW), through selective pressure by using an auxotrophic E. coli strain (Figure 32b).^82,709−711 However, despite the widespread adoption of residue-specific reassignment, if multiple reassigned codons occur repeatedly along the POI, in principle, all the sites encoding that specific amino acid can potentially be labeled with ¹⁹F, resulting in a loss of specific site-selectivity. The orthogonal aaRS/tRNA pair that incorporates l-4-trifluoromethylphenylalanine (pCF₃F)^712,713 and l-amino-3-[4-(trifluoromethoxy)phenyl]propanoic acid (4-OCF₃F)⁷¹⁴ site-specifically in response to the amber codon UAG was reported in E. coli. This system enhanced the ¹⁹F signal, provided by the CF₃ moiety, and enabled the efficient differentiation of protein environments in vivo.^712,713 The structural and dynamic information was further extended to study CaM, α-synuclein, and GFP proteins.⁷¹⁵ Recently, ¹H NMR experiments were facilitated by genetically encoding a trimethylsilyl group [−Si(CH₃)₃] into a protein. In water, the trimethylsilyl group exhibits a signal in the ¹H NMR spectrum that is upfield (chemical shift, δ_H ≈ 0 ppm) and intense, typical of its nine equivalent shielded protons differentiating it from the protons of biomolecules (δ_H = 0–12 ppm). The trimethylsilyltyrosine ncAA probe (TMSiPhe, Figure 32b) has been applied to the characterization of multiple conformational states of phospho-β2 adrenergic receptor/β-arrestin-1(β-arr1) and other membrane protein signaling complexes.⁷¹⁶ Incorporation of the trimethylsilyl group with an evolved MjTyrRS has the potential for expanding in vivo applications with increased sensitivity.

The challenges of limited sensitivity and background noise in NMR studies and difficulties in conducting in vivo experiments have encouraged the use of complementary techniques such as electron paramagnetic resonance (EPR) spectroscopy.^717−719 EPR spectroscopy has emerged as a powerful and valuable technique for elucidating the structure, dynamics, and interactions of proteins and understanding their physiological functions.⁷²⁰ EPR can be combined with site-directed spin labeling (SDSL), to increase sensitivity and enable precise labeling of specific sites within the protein.^721−723 Bioorthogonal click reactions on GCE-incorporated ncAAs (see section 4.3) allow spin labels, such as nitroxide radicals, to be inserted at specific protein sites.⁷²⁴ However, bioconjugation, if not efficient and chemoselective, can prevent the study of proteins in their natural state, because EPR is highly sensitive to the background noise given from the unreacted spin label. Moreover, directly encoding spin-labeled ncAAs into proteins is the most appealing approach for tailored design of linker’s length and flexibility, which affects EPR measurements,⁶⁶⁶ without interfering with the protein’s inherent dynamism. The Summerer and Drescher groups genetically incorporated a lysine derivative with a nitroxide moiety into eGFP and thioredoxin in E. coli (Figure 32c).⁷²⁵ Singly labeled thioredoxin was selectively detected by continuous-wave EPR measurements in E. coli host cells, which were washed to remove excess spin label. However, a limited stability of the nitroxide in the reducing intracellular environment was observed. In vitro double electron–electron resonance (DEER) measurements on doubly labeled thioredoxin revealed the distance between the two spin probes, establishing the lysine-derived ncAA (Figure 32c) as a useful probe for such experiments.^659,726 Thus, combination of these spectroscopic techniques and constant optimization of the genetically encoded ncAAs holds immense promise for dissecting the structure and function of proteins in their native environments.

4.4.2. Fluorescent Amino Acids

The site-specific incorporation of ncAAs with distinct fluorescence properties is a useful technique in cell biology. Fluorescent ncAAs have been used to visualize protein localization, to study protein trafficking, and to understand cellular processes at the molecular level. In addition, by using the well-known FRET phenomenon, pairing a fluorescent ncAA with another fluorophore, either a small molecule or a fluorescent protein, can be used to quantify macromolecular interactions and conformational changes in a protein. The first studies in GCE refined the photophysical characteristics of fluorescent proteins, such as GFP and cyan fluorescent protein, through the substitution of essential chromophoric residues with various heterocycles.^727,728 In 2006, the fluorescent ncAA 7-hydroxycoumarin-alanine (HOCouA, λ_em = 450 nm) was encoded in E. coli using the MjTyrRS/MjtRNA pair (Figure 33).⁷²⁹ This was the first genetically encoded fluorescent ncAA. Encoding fluorophores at specific locations within proteins enabled changes of protein conformation to be studied during folding and unfolding. Introducing HOCouA into the bacterial tubulin homologue FtsZ enabled clear Z-rings to be visualized during cell division. Independent Z-ring formation could be observed when mitosis was disrupted using a β-lactam.⁷³⁰ The emission spectra and quantum yields of fluorescent ncAAs depend on several factors, including metal ion concentration⁷³¹ and the dielectric constant of the surrounding environment. The prodan derivative l-3-(6-acetylnaphthalen-2-ylamino)-2-aminopropionic acid (Anap, λ_em = 490 nm in water, λ_em = 420 nm in ethyl acetate) has been encoded in S. cerevisiae and mammalian cells with EcLeuRS/tRNA pairs for studying the dynamics of potassium channel pores. Changes in conformation lead to a more hydrophobic or hydrophilic environment around the ncAA, which fluoresces according to its surroundings. By monitoring changes in the fluorescence, the region of protein involved in the channel opening was identified.^732,733

Figure 33.

Genetically encoded fluorescent ncAAs with their corresponding fluorescence emission wavelengths.

Although the HOCouA can be encoded in procaryotes by the MjTyrRS/tRNA pair, a lysine derivative of coumarin (HOCouK, Figure 33) was required to extend the utility of this fluorophore to eukaryotes. HOCouK was recognized by the orthogonal MbPylRS/tRNA pair and genetically encoded in mammalian cells (see section 4.2.1). However, the coumarin moiety attached through a carbamate bond is unstable in the presence of light, and HOCouK is converted to lysine. In this regard, HOCouK has been mostly used as a self-reporting caged lysine derivative for the regulation of protein function (see section 4.2.1). However, an investigation of the spectral properties of CouK analogs showed that homologs with an additional methylene carbon between the coumarin and carbamate (Cou₂K) do not undergo photolysis and can be used as fluorescent probes.⁵²⁰ In one such study, FRET measurements between GFP fusion protein and encoded Cou₂K derivatives gave insights into deubiquitinase activity.⁷³⁴ More recently, a coumarin with increased brightness, optimized by replacing the hydroxy group with an amino group, generated an ncAA (NH₂Cou₂K, Figure 33) with greater quantum yield, red-shifted absorbance, and reduced pH dependence.⁵²³

A qualitative FRET analysis of protein–protein interactions was achieved by using 3-(6-acetylnaphthalen-2-ylamino)-2-aminopropanoic acid (Anap, λ_em = 490 nm)^735−737 and dansylalanine (DanA, λ_em = 540 nm).^326,738 FRET between Anap from a site-specifically mutated protein Bax and yellow fluorescent protein (YFP) from a Hsp70-YFP fusion protein revealed the mechanisms of apoptosis-induced dissociation of Bax from Hsp70.^739,740 Using the same approach, the apoptosis-inducing factor (AIF) translocation from the mitochondria to the nucleus was observed in real-time, using Anap as a FRET donor.⁷⁴¹ A fluorescent oxazole-based amino acid with a modified backbone has also been genetically incorporated (see section 4.6). Acridonylalanine (Acd, λ_em = 420–450 nm), an aromatic fused tricyclic chromophore, has also been successfully encoded in E. coli. Acd has proven useful for facilitating fluorescence lifetime imaging microscopy (FLIM) studies in mammalian cells because of its longer lifetime compared to other fluorescent probes and despite its blue-shifted fluorescence emission and low extinction coefficient.⁷⁴²

As outlined in section 4.3, bioorthogonal labeling enables site-specific protein labeling with fluorescent dyes, offering increased flexibility to modify and enhance the photophysical properties of external dyes for improved in vivo imaging resolution. However, bioorthogonal labeling of ncAAs may encounter challenges such as inefficient conjugation, degradation of the reactive handle, and background fluorescence. Consequently, the development of fluorescent ncAAs can obviate the need for additional labeling reactions. Moreover, in comparison to traditional tagging with large fluorescent proteins, the site-specific encoding of small and highly efficient ncAAs represents a minimally invasive system. Nevertheless, genetically encoded fluorescent ncAAs face significant drawbacks: (1) they are excited by light of shorter wavelengths (usually < 450 nm), which is cytotoxic; (2) they show low brightness compared to commonly used fluorescent dyes, such as cyanines and rhodamines; and (3) their excitation spectra may overlap with those of other fluorescent molecules within cells, for example, tryptophan, leading to background emission. Indeed, to date, only smaller dyes with suboptimal photophysical properties have been successfully incorporated into proteins via GCE, as larger, charged structures with superior properties are not yet compatible with protein translation, and do not fit into the active sites of protein synthetases. To overcome this issue, the development of red-shifted fluorescent ncAA, compatible with the translation machinery, is highly desirable.

4.5. Stimulus-Responsive Amino Acids

The integration of encoded ncAAs with finely tuned functional groups has ushered in a new era of biosensor development, enabling the precise detection of a range of critical cellular parameters. By incorporating redox-sensitive ncAAs into proteins, biosensors can be engineered that undergo conformational changes or fluorescence shifts in response to variations in redox potential,^{731,743−745} and new enzymes with improved metal-chelating⁷⁴⁶ and radical properties can be finely tuned.⁷⁴⁷ Given the difficulties in obtaining real-time insights into the cellular redox conditions if not coupled to a fluorescence output, internal-stimulus-responsive amino acids have been inserted into the chromophore of fluorescent proteins, such as GFP, to provide an on/off approach for sensing the concentration of a defined molecule (Figure 34a). This technique facilitates the real-time visualization of dynamic phenomena such as oxidative stress and redox signaling, pointing out the crucial factors for cellular health and disease progression. The insertion of the ncAA l-DOPA into the chromophore of GFP created a Cu^II biosensor, for which fluorescence intensity was indirectly proportional to Cu^II concentration (Figure 34b).⁷⁴⁸

Figure 34.

Stimulus-responsive ncAAs. (a) Genetically encoded stimulus-responsive ncAAs. (b) Metal-chelating ncAAs, in vitro studied. (c) pH-sensitive fluorescence of ncAA, HOCouA, in vitro studied. (d) GFP-fluorescence quenching and activation in the presence of internal stimuli, in vivo studied.

By exploiting the quenching or activation of fluorescent proteins, ncAAs have been used to sense diverse molecular species. For example, the azido group in pAzF, if inserted into the GFP chromophore, creates a nonfluorescent GFP species that is activated in the presence of H₂S. Hydrolysis of the azido to the amino group results in the formation of the chromophore given by the required electron-donating group.^749,750 With the same approach, p-boronophenylalanine (pBoF) was inserted into the chromophore and used as a sensor of oxidative stress caused by H₂O₂ (Figure 34d).⁷⁵¹ Because peroxynitrite (ONOO^–) is a highly reactive nitrogen species (hRNS) involved in the nitration of Tyr and oxidative damage (see section 4.1.6), the development of a specific ONOO^– sensor, which does not respond to other redox-active molecules at physiologically relevant concentrations is necessary. Using pBoF encoded into circularly permuted fluorescent proteins (pnGFP), and site-targeted random mutagenesis, a selective live-cell sensor for ONOO^– was constructed in mammalian cells.⁷⁵² Recently, further engineering of GFP, using structure-guided reactivity screening, furnished pnGFP-Ultra, encoding pBoF with a remarkable ∼110-fold fluorescence enhancement in the presence of ONOO^–.^753,754 Detection of other highly reactive oxygen species (hROS), such as radical OH^• species and hypochlorous acid (HOCl), was also possible by using aryloxyphenols, which can be O-dearylated in presence of hROS. By evolving an orthogonal ThyRS/PyltRNA pair able to incorporate thyronine (Thy) in both E. coli and mammalian cells, the concentration of hROS was monitored.⁷⁵⁵ However, compared to pBoF, Thy-based indicators need further improvement in detection limit, brightness, and specificity.

With regard to pH sensing, the fluorescent ncAA HOCouA, which exhibits pH-dependent fluorescence (HOCouA, neutral form λ_em = 380 nm; anionic form λ_em = 450 nm, pK_a of approximately 7.8), was used to report on environments more acidic than physiological pH (Figure 34c). Specifically, the phosphorylation status of Tyr in the STAT3 protein could be monitored because the presence of a phosphate group in the local environment of the ncAA resulted in a bathochromic shift as the anionic form of HOCouA predominated.⁷⁵⁶ These examples underscore the transformative impact of encoded ncAAs in biosensor design, collectively representing a versatile toolkit for probing complex cellular environments.

4.6. Backbone-Modified Amino Acids

Because the ribosome translation system is a complex and highly regulated molecular machinery able to discern, despite their subtle differences, l- from d-amino acids as well as α- and β-amino acids, d-amino acids and backbone mutations are rarely tolerated for in vivo translation. Thus, few examples of genetically encoding backbone-modified amino acids into proteins in vivo have been reported.¹⁴⁸ Unlike the already discussed ncAAs, which primarily contribute to side-chain diversity and functionality, backbone-modified amino acids structurally alter the peptide backbone directly, leading to a protein’s unique conformational and physicochemical characteristics (Figure 35). Given the important role of the amide nitrogen atom in forming hydrogen bonds, its replacement with another element might alter the folding properties of a protein. Substitution with oxygen to give esters results in the loss of one hydrogen-bond donor and a decrease in the basicity of the carbonyl oxygen atom (Figure 35b). In this regard, α-hydroxy acids have been synthesized and genetically encoded as probes of protein unfolding. It has thus been possible to incorporate p-hydroxy-l-phenyllactic acid into myoglobin with an evolved MjTyrRS/tRNA pair⁷⁵⁷ to quantify the thermodynamic contribution of individual backbone hydrogen bonds to protein stability. Furthermore, insertion of an ester along the backbone creates a cleavable function, readily hydrolyzed unlike an amide bond, offering a parallel tool to the intein-cleavage approach for protein purification.⁷⁵⁸ Similarly, treatment of proteins containing backbone ester bonds with hydrazide resulted in the cleavage and concomitant generation of a recombinant α-hydrazide-containing protein with increased stability compared to the classical α-thioesters.¹⁶⁰ More recently, the creation of a strain of E. coli with a synthetic genome, alongside specialized designed PylRS/tRNA pairs, was advanced to recognize multiple α-hydroxy acids. This innovative system facilitated the synthesis of nonnatural depsipeptide macrocycles, encompassing two noncanonical side chains and either one or two ester bonds.⁷⁵⁹ Simultaneously, Schepartz’s group described in vivo GCE of mCF₃PheOH, facilitated by a MaPylRS variant that retained activity for Phe derivatives.²⁵⁰ Notably, the first in vitro C–C bond formation within the ribosome, through a Claisen-type condensation, was reported from the same study. However, the labile nature of esters limits in vivo applications.

Figure 35.

(a) Genetically encoded α-hydroxy acids as ncAAs. (b) H-bonding perturbation by replacing the amide bond with an ester. (c) Genetically encoded β-amino acids. (d) Genetically encoded α,α-disubstituted amino acids. (e) Genetically encoded oxazole-containing ncAA for cell imaging.

In 2016, the first in vivo incorporation of a β-amino acid (β³-(pBr)Phe, Figure 35c), which exhibit higher stability than α-peptide bonds to proteases, was accomplished. By using the PheRS, which can cooperate with wild-type EF-Tu and ribosomes, which display mutant peptidyl transferase centers, β-amino acids were incorporated in E. coli.⁷⁶⁰ Recently the Chin group further expanded the repertoire of β-amino acids encoded in E. coli by using tRNA display to efficiently select orthogonal synthetases capable of acylating tRNA with ncAAs featuring modified backbones (Figure 35c).⁷⁶¹ This includes the first report of in vivo encoding of an α,α-disubstituted amino acid (Figure 35d). Recently, an unconventional amino acid moiety, containing a strongly fluorescent oxazole ring has been incorporated in living E. coli, using a specially engineered PylRS/tRNA pair and modification at the ribosomes in order to incorporate dipeptides (Figure 35e).⁷⁶² The intriguing chemical structure, resembling a dipeptide, unlocks the vast potential for in vivo-encodable ncAAs.

4.7. Polyfunctional Amino Acids

Polyfunctionalized amino acids offer precise integration of two or more distinct functionalities within a single amino-acid building block. Applying this innovation involves the incorporation of ncAAs that possess at least two orthogonally reactive groups. Such multifunctional proteins show unique applications, such as fluorescent labeling coupled with drug conjugation sites or PPIs studies combined with pull-down assays.^763,764 On the one hand, although it is possible to efficiently incorporate multiple ncAAs, each with distinct functionalities, into a single protein, there are potential advantages to consolidating all the required functionalities into a single ncAA. On the other hand, the simultaneous incorporation of several ncAAs into one protein, each bearing more functional groups, allows the preparation of site-specific multifunctional proteins in a “plug-and-play” manner.

Common bifunctional ncAAs possess a photoresponsive moiety, able to crosslink with vicinal proteins when irradiated with light, and a clickable pendant moiety for bioconjugation and detection (Figure 36a). As discussed in section 4.2.3, azirines can be activated on demand with light. Lysine derivatives, bearing an alkynyl function for copper labeling and an azirine moiety for crosslinking (PrDiAzK, Figure 36a), were encoded in multiple sites in response to a sense codon of interest, allowing programmable residue-specific proteome labeling.⁷⁶⁵ Crosslinked proteins could be easily tracked and identified after subsequent click labeling at the alkynyl function with a fluorescent azide. By retaining the alkyne moiety and substituting the γ-carbon atom in the lysine chain for a selenium atom (PrDiZASeC), the Chen group created a cleavable photo-crosslinker, which, after irradiation, transfers the bioorthogonal handle to the interacting protein directly in cells.⁷⁶⁶ By insertion of a photo-crosslinker on the “bait” proteins, the interacting partners (the “prey”) are labeled through transfer of the chemical handle upon crosslinking.⁷⁶⁷ With this photo-affinity capture and introduction of a clickable function via a releasable linker, prey proteins could be easily biotinylated and analyzed by proteomics (Figure 36b,c).^768−771 Another derivative, bearing a diazirine as a photoaffinity group and a crotonyl group as PTMs (K*cr) has been instead site-specifically inserted on histones and exploited to covalently capture enzymes and effectors protein of histone lysine crotonylation, enabling the identification of PTM-mediated histone-effector interactions.⁷⁷²

Figure 36.

(a) Bifunctional genetically encoded ncAAs for in vivo studies. (b, c) Mechanism of bait–prey protein photo-crosslinking and MS identification for DiZSeK (b) and DiZHSeC and PrDiZASec (c).

In addition to using light to activate crosslinking, the utilization of genetically encoded photoswitchable crosslinking amino acids has enabled the dynamic control of both conformational changes and CaM binding affinity with light.⁵³⁵ Crosslinking without the need of light for studying protein–protein interactions was also provided by an alkynyl derivative of pBpa, BPKyne.⁷⁷³ With the use of BPKyne⁷⁷⁴ and an alkynyl and alkyl halide Tyr derivative, binding partners of proteins of interest could be identified in a pull-down strategy based on CuAAC-mediated biotinylation.⁷⁷⁵ Biotinylating proteins is commonly used for enriching target proteins from complex biological systems, due to the high affinity of biotin for streptavidin.⁷⁷⁶ By contrast, multifunctional ncAAs are useful in drug development. The insertion of two orthogonal clickable residues⁷⁷⁷ further enabled the one-pot preparation of protein multi conjugates,⁷⁷⁸ which have powerful applications in positron emission tomography, fluorescence imaging, and targeted therapy.

Besides expanding the functional landscape of proteins, GCE has also served to improve other molecular methods, including the enhancement of CRISPR-based genome editing via different strategies. The incorporation of 4-(2-azidoethoxy)-l-phenylalanine (AeF) into the Cas9 protein has been applied for the specific bioconjugation of a minimal crRNA attached to a compatible handle, which enabled the reduction of the total size of the crRNA to 28 nucleotides, including 20 nucleotides for target DNA recognition, and without requiring further tacrRNA molecules.⁷⁷⁹ Applying a similar strategy to instead recruit a donor DNA molecule to a Cas9 with incorporated AeF was reported to facilitate the desired homology-directed repair of genome editing in mouse zygotes, increasing editing efficiency of the SOX2 gene from 26% with unconjugated donor DNA to more than 60% with conjugated donor DNA.⁷⁸⁰ Bioconjugation of crRNA molecules with the Cas12a nuclease have also been reported to increase genome editing efficiency also in a multiplexed fashion.⁷⁸¹ However, there are still limitations that need to be addressed to unleash the potential of GCE fully. Incorporating charged amino acids and very bulky ones has proved challenging due to low cellular uptake and the limited size of the synthetase binding pocket. Further engineering of key components for GCE with GCE itself might be the key to improve translation efficiency and go beyond limitations posed by subtle differences between ncAAs and cAAs. Sophisticated aaRS engineering is required to yield systems that effectively discriminate between these entities and, additionally, are efficient in generating polymers with variable numbers of ncAA building blocks. While progress in prokaryotic systems is relatively fast, the extension to higher eukaryotes and animals still bears significant challenges for developing orthogonal, site-specific, and transcript-specific GCE systems.

5. Conclusions and Outlook

The genetic code represents a universal concept that is essential to all life on earth. Inspired by the mechanisms and machinery that naturally evolved to interpret and execute the genetic code, scientists began to perturb the underlying molecular mechanisms, to purify and synthesize components, and to evolve and engineer them. The re-introduction of modified cellular machinery into living cells and their application in artificial environments has expanded a naturally set limit: the genetic code. In the last three decades, there were numerous approaches to establishing new genetic codes for incorporating diverse chemical functionalities in vivo. We give an extensive overview of how this field of research emerged, the methods applied to establish expanded genetic codes within living cells, the chemical diversity of genetically encoded ncAAs, and different strategies for GCE in the prokaryote and eukaryote kingdom. The described tools for the generation of new orthogonal aaRS/tRNA pairs bear the great potential to further increase the spectrum of accepted noncanonical amino acids and to establish orthogonal GCE systems in diverse host organisms, also considering that the majority of aaRS/tRNA pairs are still underexplored.^198,782 Inventive methods for achieving orthogonality, including genome editing and synthetic genomics, engineered orthogonal ribosomes working with an orthogonal mRNA, novel base pairs and orthogonally translating organelles, will likely enable groundbreaking further developments and applications of GCE. Given the plethora of functions achieved with the naturally evolved 20 (in rare cases 22) canonical AAs, we can only imagine how protein shape and functionality can be further engineered with the more than 500 ncAAs that have already been successfully encoded in living cells. The synthesis of ncAAs with novel chemical functionalities and biological properties, along with the parallel evolution and engineering of aaRS/tRNA pairs for genetically encoding them, represents a major future direction for expanding GCE applications in basic and applied research. With the comprehensive table of all ncAAs that have been genetically encoded in living cells to-date (Table S1), we anticipate to provide a resource for designing new ncAAs in the future. This resource can not only be used to compare the biocompatibility of ncAAs (such as the solubility and membrane permeability of an ncAA, as derived from predictions of the logS and cLogP, respectively) but also to design ncAAs with novel chemical functionalities and explore their predicted behaviour, as well as to identify existing ncAAs with similar behavior (see Figure 14). GCE has thus become a field that combines chemical and synthetic biology with genomics and molecular engineering to generate a freedom in protein diversity that is beyond current imagination and with great potential to tackle major problems in biology and medicine. Multiple drugs based on GCE are currently in phase II trials, to name just one area where the impact can already be projected into the very near future. Besides enhancing our basic and applied biomedical research ability, GCE further promises to revolutionize the generation of intelligent materials and lifelike matter by creating partially or even entirely synthetic protein polymers with designer functionality. With new developments facilitating GCE being made at a rapid pace alone within the last decade, we anticipate that the upcoming decade will also bring about technology solving the grand challenge of generating artificial polymers, consisting entirely of ncAAs and offering full control of their fold and function.

Glossary

Abbreviations

AA

amino acid

aaRS

aminoacyl tRNA synthetase

BAC

bacterial artificial chromosome

B. steareothermophilus

Bacillus steareothermophilus

BONCAT

bioorthogonal amino acid tagging

cAA

canonical amino acid

CAGE

conjugative assembly genome engineering

CaM

calmodulin

Cas9

CRISPR-associated protein 9

C. elegans

Caenorhabditis elegans

CRISPR

clustered regularly interspaced short palindromic repeats

crRNA

CRISPR RNA

CuAA

Cu(I)-catalyzed alkyne–azide cycloaddition

DNA

deoxyribonucleic acid

dNTP

deoxyribonucleotide triphosphate

E. coli

Escherichia coli

Ec

Escherichia coli

EF

elongation factor

FACS

fluorescence-activated cell sorting

FLIM

fluorescence lifeftime imaging

FRET

Förster resonance energy transfer

FUNCAT

fluorescent noncanonical amino acid tagging

GCE

genetic code expansion

GENESIS

genome stepwise interchange synthesis

H. sapiens

Homo sapiens

HDR

homology-directed repair

IEDDA

inverse-electron-demand Diels–Alder

ISR

integrated stress response

itRNA

initiator tRNA

ncAA

noncanonical amino acid

NES

nuclear export signal

NLS

nuclear localisation signal

NMD

nonsense-mediated decay

NMR

nuclear magnetic resonance

NRPS

nonribosomal peptide synthetase

MAGE

multiplex automated genome engineering

M. barkeri

Methanosarcina barkeri

Mb

Methanosarcina barkeri

M. jannaschii

Methanocaldococcus jannaschii

Mj

Methanocaldococcus jannaschii

M. mazei

Methanosarcina mazei

Mm

Methanosarcina mazei

MMF

maximum misincorporation frequency

mRNA

messenger RNA

MS

mass spectrometry

ORF

open reading frame

OT

orthogonally translating

PACE

phage-assisted continuous evolution

PANCE

phage-assisted noncontinuous evolution

PAM

protospacer–adjacent motif

PCR

polymerase chain reaction

PKR

protein kinase R PKS polyketide synthase

POI

protein of interest

PTM

post-translational modification

PYLIS

pyrrolysine insertion sequence

REXE

replicon excision enhanced recombination

RF

release factor

RNA

ribonucleic acid

rRNA

ribosomal RNA

RRE

relative readthrough efficiency

S. cerevisiae

Saccharomyces cerevisiae

SEC

selenocysteine insertion sequence

SORT-E

stochastic orthogonal recoding of translation with enrichment

SORT-M

stochastic orthogonal recoding of translation with chemoselective modification

SPAAC

strain-promoted alkyne–azide cycloaddition

tRNA

transfer RNA

tracrRN

trans–activating CRISPR RNA

T. thermophilus

Thermus thermophilus

Tt

Thermus thermophilus

UV

ultraviolet

VADER

virus-assisted directed evolution of tRNA

XNA

xeno nucleic acid

Biographies

Cosimo Jann graduated with a B.Sc. in Biological Sciences (2013) and an M.Sc. in Microbial and Plant Biotechnology from University of Kaiserslautern-Landau (RPTU), Germany (2015). His M.Sc. included a study and research semester at the Dept. of Bioengineering at University of California, Berkeley (UC Berkeley), USA. After graduation, he worked as a visiting researcher at University of Tartu, Estonia, and the Wellcome Trust Sanger Institute in Cambridge, UK, and performed an internship at BASF in Tarrytown, New York, USA. He earned a doctoral degree from ETH Zurich, Switzerland, and the European Biology Laboratory (EMBL) in Heidelberg, Germany (2020) under the supervision of Prof. Dr. Lars M. Steinmetz and Prof. Dr. Karsten Weis. During his doctoral studies, Cosimo established CRISPRa/i screens in baker’s yeast to identify gene functions in cellular growth, stress resistance and molecular signaling transduction pathways. In 2021, Cosimo joined the Johannes Gutenberg University in Mainz via the Institute of Molecular Biology (IMB) postdoctoral programme, where he is currently developing artificial membrane-less organelles for orthogonal translation and other cellular processes.

Sabrina Giofrè received her M.Sc degree in Pharmaceutical Chemistry (2016) from the Università degli Studi di Milano under the supervision of Prof. Maria L. Gelmi and her Ph.D. in Pharmaceutical Science from the Università degli Studi di Milano (2020), under the supervision of Prof. Egle M. Beccalli. After a period as a post-doc researcher in the group of Prof. Pierfausto Seneci, she currently holds a post-doc position in the group of Prof. Edward Lemke. Her research is mostly focused on metal-catalyzed reactions and synthesis of noncanonical amino acids and heterocycles, including fluorescent dyes, to interrogate the complex mechanisms and functions of biological systems and offer theraupetic tools.

Rajanya Bhattacharjee received her Bachelor of Engineering (BE) degree (2018) in food technology and biochemical engineering from Jadavpur University, Kolkata, India, and her Master of Technology degree (MTech) in bioprocess engineering from Indian Institute of Technology Roorkee (IITR), Roorkee, India (2020). She is currently a Ph.D. student in the IMB International Ph.D. Programme (IPP) at the Johannes Gutenberg University, Mainz, Germany, under the supervision of Prof. Dr. Edward A. Lemke. Her research is focused on the development of labeling strategies by genetic code expansion with a view toward improvement of high-resolution microscopy.

Edward A. Lemke spent several years as group leader of an Emmy Noether and ERC consolidator research group at the European Molecular Biology Laboratory (EMBL), and then the biophysical chemist has taken up a professorship for synthetic biophysics of protein disorder at Johannes Gutenberg University Mainz, where he has also became Adjunct Director at the Institute of Molecular Biology (IMB). Combining new research methods and expertise in chemical/synthetic biology, chemistry, biophysics and cell biology, his group innovates new approaches to studying intrinsically disordered proteins. He has been an EMBO member since 2022 and the spokesperson of networks that bring together scientists from polymer and life sciences (www.crc1551.com and www.spp2191.com), and his current work is also funded by an ERC advanced grant.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.chemrev.3c00878.

• In vivo site-specific genetically encoded noncanonical amino acids, chemical properties comparison of new ncAAs (XLSX)

Author Contributions

^⊥ C.J., S.G., and R.B. contributed equally to this review. CRediT: Cosimo Jann conceptualization, data curation, writing-original draft, writing-review & editing; Sabrina Giofré conceptualization, data curation, writing-original draft, writing-review & editing; Rajanya Bhattacharjee conceptualization, data curation, writing-original draft, writing-review & editing; Edward A. Lemke conceptualization, funding acquisition, project administration, supervision, writing-original draft, writing-review & editing.

The authors declare the following competing financial interest(s): E.A.L. holds several patents related to genetic code expansion and is a cofounder and consultant of Veraxa Biotech GmbH, a company specialized on generation of antibody drug conjugates via GCE.

Special Issue

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