Rewriting the Genetic Code

Abstract

The genetic code—the language used by cells to translate their genomes into proteins that perform many cellular functions—is highly conserved throughout natural life. Rewriting the genetic code could lead to new biological functions such as expanding protein chemistries with noncanonical amino acids (ncAAs) and genetically isolating synthetic organisms from natural organisms and viruses. It has long been possible to transiently produce proteins bearing ncAAs, but stabilizing an expanded genetic code for sustained function in vivo requires an integrated approach: creating recoded genomes and introducing new translation machinery that function together without compromising viability or clashing with endogenous pathways. In this review, we discuss design considerations and technologies for expanding the genetic code. The knowledge obtained by rewriting the genetic code will deepen our understanding of how genomes are designed and how the canonical genetic code evolved.

INTRODUCTION

A synthetic organism would revolutionize basic research and biotechnology. Such an entity would have additional protein constituents, noncanonical amino acids (ncAAs) assigned to their own codon in the genetic code. This would be the dream of protein engineers (80); it would allow the design of proteins with novel properties based on the presence of new building blocks in addition to the 20 canonical amino acids (109). Progress along these lines is being made, as codons have been successfully reassigned to encode ncAAs in Escherichia coli (69, 70, 95, 109) and genome synthesis projects aiming at rewriting the genetic codes of E. coli (113, 149), Salmonella typhimurium (71), and yeast (27, 124) are proceeding.

Rewriting the genetic code (Figure 1a, b) involves (a) engineering orthogonal translational components, (b) engineering endogenous translational components, (c) metabolome engineering, (d) massive genome/chromosome engineering for modulating global codon usage, (e) chemical synthesis or biosynthesis of ncAAs, and (f) motivating organisms to maintain the new genetic code and to evolve with it. Traditionally, several methods have been used for incorporating ncAAs into proteins (1) (Figure 1a). The selective pressure incorporation method replaces 1 of the 20 canonical amino acids with its ncAA analog. Expression of a variant of an aminoacyl-tRNA synthetase (aaRS) has facilitated ncAA incorporation into proteins (25). In Bacillus subtilis and E. coli, tryptophan (Trp) has been completely replaced by Trp analogs (such as [3,2]Tpa) in the proteome (1, 47). In contrast, stop or sense codon suppression methods (67, 152) as well as frameshift suppression (151) allow site-specific ncAA incorporation into proteins by using an orthogonal tRNA•aaRS pair specific for the ncAA and the codon (15, 77) (Figure 1b). Thus, the amino acid repertoire of the genetic code is expanded. During stop codon suppression, the stop codon is ambiguously translated as stop or sense (by a ncAA). Similarly, sense codon suppression simultaneously assigns one codon to encode two amino acids, and stochasticity in frameshift suppression results in a mixture of proteins translated in two different frames. All ambiguous decoding methods produce statistical proteins. In contrast, codons can be reassigned (5, 95) to unambiguously encode a ncAA by eliminating native tRNAs or the release factor (RF) originally decoding the codon to be redefined (Figure 1b). To accomplish this, the genomic usage of the codon should be reduced to alleviate the detrimental effects of codon reassignment (70, 112).

Figure 1.

Rewriting the genetic code. (a) Three methods used to augment the genetic code with ncAAs: selective pressure incorporation, site-specific incorporation, and codon reassignment. These methods are not mutually exclusive. (b) A proposed form of an organism having a new genetic code and amino acid repertoire. The ncAA (orange star) is either supplemented in the media and taken up by the cell through a transporter or produced by the cell. Enzymes that degrade the ncAA are inactivated, and an orthogonal aaRS charges the ncAA onto its devoted tRNA. Panel a adapted from Sakamoto (129). Abbreviations: aaRS, aminoacyl-tRNA synthetase; aa-tRNA, aminoacyl-tRNA; mRNA, messenger RNA; ncAA, noncanonical amino acid; RF, release factor; tRNA, transfer RNA.

In this review, we provide an update of recent in vivo genetic code and genome engineering studies in microbes, particularly E. coli, and a comparison with natural cases of noncanonical genetic codes. We focus on ncAAs that resemble canonical amino acids in size and hydrophilicity and may thus be biomolecule-friendly. These ncAAs are, however, often involved in cellular metabolism and posttranslational protein modification, as indicated in Figure 1b. This review does not cover the promising genetic code expansion studies using in vitro protein synthesis (e.g., 126, 137), some of the established applications of in vivo genetic code engineering (15, 57, 77), or the exciting work with supernumerary unnatural base pairs (6, 14).

NATURAL EXPANSION OF THE GENETIC CODE

Given the vast amount of new genomic and metagenomic DNA sequence information generated in the past few years, we document here the current knowledge of natural deviations from the standard genetic code (45, 61, 79, 92, 94, 98, 115, 125, 134, 157) (Figure 2) (Supplemental Figure 1). Previously, it was well known that Candida yeast reassigns the CUG codon from leucine (Leu) to serine (Ser) (60) and that many ciliates reassign either the UGA stop codon or both the UAA and the UAG stop codons to a canonical amino acid (12, 61, 88, 119) (Figure 2). A few bacteria reassign the UGA stop codon to either Trp or glycine (11, 76, 155) (Figure 2). The genetic code of mitochondria deviates significantly from the standard genetic code (61, 76) (Supplemental Figure 1), probably because there are only a few genes (encoding membrane proteins) in mitochondrial genomes. It was also known that the UGA stop codon is recoded as selenocysteine (Sec), the twenty-first genetically encoded amino acid, in the three domains of life when guided by Sec-insertion sequence elements (7, 33, 93, 158). In contrast, the UAG stop codon is translated as pyrrolysine (Pyl), the twenty-second genetically encoded amino acid, in some anaerobic archaea and bacteria (35, 93, 158), and as glutamine (Gln) in the late genes of some bacteriophages that kill bacteria whose UGA codon encodes Trp (54) (Figure 2).

Figure 2.

Deviation from the standard genetic code in nature. (a) Codon reassignment occurred in some bacteria and eukaryotes (nuclear genetic code), whereas dual or triple usage of a particular codon, including the assignment of selenocysteine (Sec) and pyrrolysine (Pyl), is found in all three domains of life. (b) Some bacteriophages change the genetic code of their host cells for late gene expression. The full map of codon reassignment (organisms and organelles) can be found in Supplemental Figure 1.

Recent eukaryotic genome/transcriptome analyses identified that (a) Pachysolen tannophilus, a yeast species distinct from Candida spp., reassigns CUG to alanine (Ala; 92, 125) (Figure 2); (b) diverse single-celled eukaryotes that are nonciliate also reassign stop codons to amino acids (17, 19, 59, 62, 63, 115, 157) (Figure 2); (c) not only Gln but also Leu, tyrosine (Tyr), and glutamic acid (Glu) are assigned to UAR [R denotes adenosine (A) and guanine (G)] (45, 134) or to UAG (115) (Figure 2); and (d) ciliates Parduczia sp. and Condylostoma magnum and trypanosomatids Blastocrithidia spp. reassign the three stop codons to amino acids, whereas one or three of the codons are still used as a termination signal at the end of open reading frames (45, 134, 157). In the ciliates Euplotes spp., the default function of the UAR stop codons may be frameshifting (79). Therefore, it was proposed that ciliates may have a special mechanism by which the polypeptide release factor eRF1 is tethered to the poly(A) tail of the mRNA, which may facilitate the context-dependent translation termination in ciliates (10, 79, 134). In contrast, rare distributions of UAG/UAA/UGA sense codons in the ribosomal protein genes of the Blastocrithidia spp. indicated that this alternative genetic code may have a young history (157). Interestingly, ciliates having the UGA cysteine (Cys) or Trp codon still assign Sec with UGA (Figure 2), probably in a context-dependent manner (134, 143).

Recent bacterial genome/metagenome/metatranscriptome analyses identified some bacterial species for which Sec is not assigned by UGA but by UAG and the UGU/UGC Cys codons (94) (Figure 2). In particular, all Geodermatophilaceae (actinobacteria) species sequenced so far use UAG for Sec (Figure 2). In contrast, it was predicted that the UGA stop codon would be recoded as both Sec and Cys in a few Deltaproteobacteria species such as Desulfococcus biacutus (98), although experimental validation is required (Figure 2). Furthermore, a group of potential missense and nonsense suppressor tRNA genes was identified in genome/metagenome/metatranscriptome sequences, derived mostly from Acidobacteria species (98), indicating that ambiguous decoding of a particular codon might be a popular mechanism in bacteria.

Taken together, these data make clear that the genetic code has more flexibility than was assumed at the time the code was determined. The basis for many of these code variations derives from the relative simplicity (i.e., one or two mutations in tRNA) by which tRNA identity may be switched (75, 76). Such codon reassignments also make the organism’s DNA refractive to horizontal gene transfer, a property that may be desirable in certain circumstances.

EXPANDING THE GENETIC CODE WITH ORTHOGONAL TRANSLATION SYSTEMS

Accurate protein biosynthesis is an immensely complex process involving more than 100 discrete components that must come together to translate proteins with high speed, efficiency, and fidelity. The E. coli ribosome alone is composed of 54 proteins and 3 RNAs, whereas other translation factors include 33 tRNAs, 21 aminoacyl-tRNA synthetases, 3 initiation factors, 3 elongation factors, 2 RFs, and 12 nucleotide-modifying enzymes (34). Efforts to expand the genetic code must introduce new translation components without compromising the function or fidelity of the endogenous translation system or relaxing the endogenous mechanisms of protein quality control. These efforts have predominantly focused on engineering aaRSs, tRNAs, elongation factor EF-Tu, and the ribosome (37) (Figures 1b and 3).

Figure 3.

Engineering the orthogonal translation systems. (a) aaRS engineering with an example of Methanocaldococcus jannaschii (Mj) TyrRS. (b) EF-Tu engineering. The changed residues are shown. Ser66 was modified to alanine to improve azido-phenylalanine recognition (37). (c) tRNA engineering. Colored residues were mutated to change the indicated properties. (d) Ribosome engineering. The PTC, A site, anti-SD sequence, and mutated ribosomal RNA residues are indicated. Abbreviations: aaRS, aminoacyl-tRNA synthetase; anti-SD, anti-Shine-Dalgarno; ncAAs, noncanonical amino acids; PTC, peptidyl transfer center; rRNA, ribosomal RNA; tRNA, transfer RNA; TyrRS, tyrosyl-tRNA synthetase.

Engineering of aaRS•tRNA Pairs

Orthogonal aaRS•tRNA pairs have allowed the introduction—to date—of more than 167 ncAAs into the genetic codes of bacteria, yeast, and animals (26) (Supplemental Table 1). Orthogonality requires that the aaRS•tRNA pair incorporate its cognate amino acid without cross-reacting with other added ncAAs, cellular amino acids, tRNAs, or aaRSs (78, 106). In general, the aaRS specifies an amino acid with its amino acid binding pocket and selects the tRNA species to be charged by recognizing a small number of bases or structural features (identity elements) in the tRNA (39). Although the aaRS enzymes discriminate superbly against any component of the cell’s metabolome, they have no mechanism to reject the many ncAAs that are used by synthetic biologists (31, 106); thus, aaRSs tend to be polyspecific for ncAAs (42). Many aaRS variants used in current work with ncAAs display (in vitro) greater than 100-fold reduced catalytic activity (42, 106). This poor catalytic activity can lead to low ncAA-tRNA levels that are successfully outcompeted by endogenous E. coli tRNAs that engage in near-cognate codon:anticodon interactions (107), with the consequence of inserting an undesired canonical amino acid at the codon of choice. This poor ncAA activation by the aaRS variant needs to be compensated for by overexpression of the orthogonal tRNA•aaRS pair and an elevated presence of the ncAA (42, 122). Clearly, aaRS variants with increased activity and specificity are required for future synthetic biological experiments like the reassignment of multiple codons with different ncAAs. In addition, orthogonal aaRSs equipped with a heterologous editing domain may improve amino acid specificity (110, 123).

Although tRNA identity elements are generally conserved across all three domains of life, several exceptions have been discovered, providing a starting point for evolving orthogonal aaRS•tRNA pairs. Archaeal systems are more likely to be orthogonal in bacteria than eukaryotes, whereas bacterial systems are more likely to be orthogonal in eukaryotes. For example, the tRNA^Tyr•TyrRS pair from the archaeon Methanocaldococcus jannaschii with a few tRNA modifications is orthogonal in E. coli (152), whereas bacterial tRNA^Tyr•TyrRS is orthogonal in eukaryotes (16, 130). The specificity of these aaRS•tRNA pairs can be tuned by generating large libraries of mutations in their amino acid binding pockets or at the residues involved in aaRS•tRNA interactions (Figure 3). Functional variants are identified by performing alternate cycles of positive selections in the presence of a ncAA and negative selections in the absence of a ncAA (77). Recent advances in directed evolution methods (114, 141) such as multiplex automated genome engineering (MAGE; 3, 132) and phage-assisted continuous evolution (29) may accelerate evolution. In addition to the traditional methods that use E. coli and yeast in vivo, positive in vitro selection systems were demonstrated as useful (28, 146), whereas negative selections are still limited to in vivo experiments (83).

Among others, tRNA^Tyr•TyrRS and tRNA^Pyl•PylRS pairs have been predominantly used for bacterial and mammalian ncAA incorporation (Supplemental Table 1). Likewise, yeast tRNA^Trp•TrpRS and tRNA^Phe•PheRS pairs and archaeal tRNA^Lys•LysRS pairs were used in E. coli, and bacterial tRNA^Leu•LeuRS and tRNA^Trp•TrpRS pairs were used in eukaryotes (Supplemental Table 1). tRNA^Tyr•TyrRS, tRNA^Leu•LeuRS, or tRNA^Pyl•PylRS systems are also used in gammaproteobacteria including enteropathogenic E. coli, species of Shigella and Salmonella, Yersinia ruckeri, Acinetobacter baylyi, and Pseudomonas syringae; in Synechococcus elongates (cyanobacterium); in gram-positive bacteria, including Mycobacterium tuberculosis, Streptomyces species, and Bacillus cereus; in yeasts, including Saccharomyces cerevisiae, Pichia pastoris, Candida albicans, and Schizosaccharomyces pombe; in the plant Arabidopsis thaliana; and in animals and animal cells (Supplemental Table 1).

Recent studies provided additional systems for use in E. coli: A part of the cysteinyl-tRNA^Cys synthesis machinery of methanogenic archaea was transplanted into E. coli (and Salmonella enterica), so that the intermediate product phosphoseryl-tRNA^Cys (Sep-tRNA^Cys) produced by SepRS could be used for inserting Sep into proteins (Supplemental Table 1). Although Sec naturally requires the dedicated elongation factor SelB (33), two separate tRNAs have been reengineered to incorporate Sec using the more conventional elongation factor EF-Tu (2, 44, 89, 139), which is not dependent on a Sec insertion sequence in the mRNA. Finally, it would be beneficial to develop orthogonal aaRS•tRNA pairs that can function across all domains, like the tRNA^Pyl•PylRS pair (97). E. coli strains are being developed that replace the endogenous tRNA^Tyr•TyrRS or tRNA^Trp•TrpRS pairs with heterologous alternatives, thereby allowing the original pairs to be repurposed for ncAA incorporation (50, 52).

Engineering of Elongation Factor EF-Tu and the Ribosome

AaRS•tRNA pairs are not the only crucial translation components for amino acid incorporation. EF-Tu also has amino acid recognition and tends to reject aminoacyl-tRNAs with bulky (23, 32) or negatively charged (44, 72, 116) amino acids. Directed evolution of the elongation factor’s amino acid binding region (Figure 3) made it possible to cotranslationally incorporate Sep (72, 116), Sec (44), and phosphotyrosine (pTyr; 32) into proteins in E. coli.

Meanwhile, ribosomes reject D-amino acids and define how incoming translation factors affect translation. The ribosome provides an intriguing target for accessing new genetically encoded polymers (30, 126). However, the key challenge that has prevented extensive ribosomal engineering has been the inaccessibility of orthogonal ribosomes. For instance, modifying the peptidyl transfer center of the 50S subunit permitted efficient in vitro translation with D-amino acids (20, 21) and β-amino acids (22, 86), but overexpression of the modified ribosome diminished fitness (20, 21). Several advances in the past decade have now provided a starting point for extensive ribosomal engineering (Figure 3). First, replacing the anti–Shine-Dalgarno (aSD) sequence at the 3′ end of the 16S rRNA produced an orthogonal 16S particle that only translates orthogonal mRNAs bearing the complementary synthetic ribosome binding site (48, 121). This in turn enabled evolution of the ribosomal A site in the orthogonal 30S subunit to improve UAG (150) and frameshift (104) suppression and UGA-to-Sec recoding (138). Subsequently, the orthogonal 16S rRNA was tethered to the circularly permuted 23S rRNA so as to link the 30S and 50S ribosomal subunits, thereby facilitating engineering of the peptidyl transfer center in the 50S subunit of orthogonal ribosomes (36, 111). Furthermore, the 23S rRNA has been engineered to recognize a noncanonical tRNA 3′-terminal tail (CGA or GGA instead of CCA) for in vitro translation (136, 137). Combining these methods will enable extensive engineering of ribosomal function in the near future.

SUSTAINED CODON REASSIGNMENT IN VIVO

In addition to reengineering the biochemical translation machinery, in vivo codon reassignment (Figure 1a) poses additional challenges in genetics and implementation. Synthetic biologists have faced two problems (95): (a) Some endogenous components essential for wild-type cell growth must be eliminated to achieve codon reassignment. (b) The codon to be redefined must be translated predominantly by a new decoding molecule (aa-tRNA) to avoid ambiguous assignment of the codon. Here, we summarize the state-of-the-art genome engineering technologies used to change the codon usage of E. coli.

Conditions for Codon Reassignment

How can an essential gene be knocked out? What if an essential component were depleted from the cytosol? The bacterial prfA gene encoding release factor RF1, responsible for termination at UAG (128), has been extensively studied because a partial RF1-deficiency drastically increases suppression of the UAG codon (128). First, the prfA gene was revealed as nonessential in particular genetic backgrounds (53, 55, 95) (Supplemental Table 2). UAG is the minor stop codon in many bacteria, being present in only 7% of all E. coli genes and 2–3% of essential E. coli genes. Furthermore, E. coli tolerates amber suppression to a significant level. RF1 is essential in E. coli K-12 strains that have RF2(Thr246) (128). However, RF1 is less essential in other E. coli strains and bacteria that contain RF2(Ala/Ser246), RF2 proteins with ten times higher activity than that of the K-12 RF2(Thr246) (55, 56, 69, 91, 108, 145), and is totally dispensable in E. coli expressing Salmonella RF2(L167K) (53) or E. coli RF2(T246A/A293E) (56, 69, 108) variants. These studies proved that RF2 partially substitutes for RF1. Furthermore, RF1 is not essential in E. coli cells expressing amber suppressor tRNA and four essential genes with UAG-to-UAA synonymous changes (95). These studies, together with a supporting theory (131), revealed that an essential component can be depleted from the cellular operating system if another component acts as a substitute. This idea was extended to reassign the rare sense codon AGG in E. coli; homoarginyl-tRNA_CCU partially substituted for arginyl-tRNA_CCU (99) (Supplemental Table 2) in accordance with the “similar replaces similar” rule (1).

The next question is more important: How is a new, unambiguous codon assignment achieved? Even in the absence of the original decoders, the new decoding factor must outcompete other cellular factors. Paused translation or ribosome stalling allows enough time for near-cognate tRNAs to mistranslate the codon (58, 107). Alternatively, stalled ribosomes on intact or cleaved mRNAs are rescued by three ribosome rescue pathways (46), resulting in the release of truncated proteins or modified proteins with a C-terminal ssrA degradation tag. In the case of UAG codon reassignment (in which the codon is located at the end of the native gene), the most straightforward way to prevent ribosome stalling is to express a strong amber suppressor tRNA (56, 100, 108). Actually, ribosome stalling was detected in an E. coli ΔRF1 strain expressing a weak UAG-decoding tRNA^Gln (SupE44) but was resolved by expressing a strong UAG-decoding tRNA^Gln (SupE3; 108). Similarly, near-cognate UAG decoding was observed in the E. coli ΔRF1 strain JX33 expressing RF2(T246A/A293E) and was resolved by efficient cognate UAG decoding (56). In a later study, UAG was eliminated from 95 genes, including the essential genes in E. coli BL21(DE3) with RF2(Ala246), to produce the E. coli B-95.ΔA strain (96). In B-95.ΔA, both near-cognate UAG decoding and ribosome stalling (or UAG decoding by RF2) were observed in the absence of SupE3 (96).

Recoding the Genome

The most promising method of in vivo codon reassignment is to eliminate particular codon assignments throughout the genome (Figure 4a–d). This was achieved in E. coli by using the genome engineering technologies MAGE and conjugative assembly genome engineering (CAGE; 51, 69) (Figure 4a, b, d). In short, the E. coli genome was conceptually split into 32 segments, each containing 10 UAG codons. MAGE was used to produce 32 separate strains, each with all 10 UAG codons in its target segment mutated to synonymous UAA stop codons (Figure 4a, b). CAGE was then used to hierarchically assemble these 32 recoded segments into one fully recoded chromosome (Figure 4d) (51). This genomically recoded organism, or the E. coli strain C321.ΔA (Supplemental Table 2), has no known UAG codon in its genome and lacks RF1 (69). Since the launch of this strain, several derivative strains have been developed for optimized assignments of each ncAA species (Supplemental Table 2). In separate work, coselection-MAGE (CoS-MAGE) was developed for enhancing scarless genome modification (69, 147) and was employed to eliminate all AGA/AGG arginine (Arg) rare codons (123 total) in the essential genes in E. coli (102) (Supplemental Table 2).

Figure 4.

State-of-the-art recombination methods used to engineer the codon usage in the Escherichia coli genome. For serial genome engineering, iterative multiplex oligo-mediated recombination of multiple alleles produced E. coli genomes having more than 120 intended mutations (99, 102) (a). Alternatively, a set of recoded genome segments was prepared by iterating oligo-mediated recombination (MAGE; 51) (b) or by de novo DNA synthesis subjected to assembling in yeast (113, 149) (c). These sets of recoded genome segments were assembled by hierarchical CAGE (51, 69), CAGE 2.0 (105, 113), or iterating REXER (149) (d). The latter two methods are optimized for assembling de novo synthesized segments (50–120 kbp). Hierarchical CAGE was used for assembling 32 segments, each having 10 UAG-to-UAA stop codon changes, into one genome devoid of any UAG stop codon to produce the E. coli C321.ΔA strain (69). Abbreviations: BAC, bacterial artificial chromosome; CAGE, conjugative assembly genome engineering; CoS, coselection; dsDNA, double-stranded DNA; MAGE, multiplex automated genome engineering; REXER, replicon excision for enhanced genome engineering through recombination; ssDNA, single-stranded DNA.

An apparent drawback of MAGE and CoS-MAGE is the accumulation of spontaneous mutations due to deficiency in the methyl-directed mismatch repair (MMR), a mechanism that is necessary to enhance recombination (18). Thus, one to four spontaneous mutations per one intended mutation were detected in the final strains (69, 102). Fortunately, the growth defects caused by the 355 spontaneous mutations and 321 intended mutation in C321.ΔA were largely restored by only six mutations (5 reversion, 1 de novo) to produce C321.ΔA.opt (66) (Supplemental Table 2). In contrast, spontaneous mutation rates dropped significantly in the wild-type MMR-proficient background (to 1 spontaneous per 10 intended; 96, 99) (Figure 4a), whereas recombination efficiencies dropped as well. Transient or local MMR-deficiency can serve as a compromise to boost on-target editing without increasing the spontaneous mutation rate (74, 148). However, the most essential drawback of oligo recombination is that it is currently not practical for generating more than hundreds or thousands of mutations, limiting MAGE to recode only a few rare codons (113).

Two strategies were demonstrated as valid for the de novo synthesis of recoded segments of the E. coli genome (113, 149) (Figure 4c, d). These projects aim to finally create an E. coli genome lacking up to seven codon assignments. In short, de novo synthesized double-stranded DNA fragments were assembled into a 50- to 120-kbp segment in yeast and cloned in an artificial chromosome. After introduction into E. coli, two routes could be used. The first one inserts these chromosomes into the genomic chromosome by λ-integration. In the second approach, the recoded segments are linearized by Cas9 and then inserted by homologous recombination (Figure 4c, d). To avoid aberrant recombination by RecA, λ-integrase–mediated CAGE (CAGE 2.0) was developed (105) (Figure 4d). These recoded segments will then be assembled into a fully recoded genome by iterating these methods. A major challenge will likely be troubleshooting design flaws that cause synthetic lethality as recoded segments are combined. Elimination of the original factors decoding these forbidden codons (68) would make seven codon assignments blank (113). Not only E. coli but also S. typhimurium is remarkably amenable to genome-scale modification (71). Similarly, the synthetic yeast chromosome project is aiming to eliminate the UAG codon (27, 124).

Simple synonymous substitution of a forbidden codon is not always successful, because codon sequence may simultaneously be defined by multiple constraints, including amino acid choice in the gene of interest, amino acid choice in an overlapping gene, DNA motifs affecting replication or transcription, and RNA motifs affecting translation (10, 68, 120) (Supplemental Figure 2). Furthermore, codon bias is involved in the fine-tuning of gene expressions (120). Global change in the codon usage would change the supply and demand of tRNA isoacceptors (24). A comprehensive mutagenesis of the AGA/AGG codons in the essential genes revealed that approximately 10% of these codons reject a simple synonymous replacement with CGU and need troubleshooting (102). Examples of troubleshooting are shown in Supplemental Figure 2. In most cases, these AGA/AGG codons were changed to synonymous Arg codons CGC/CGA/CGG (Supplemental Figure 2a, b). For overlapping genes, insertion of a short sequence allowed safe synonymous replacement by resolving the overlap (99, 113) (Supplemental Figure 2a). However, in a few cases, synonymous substitution was not possible. For example, the AGA codon at position 6 of the repY gene of a ColIb-P9 plasmid (4) was randomized and finally changed to the UUA Leu codon to maintain the plasmid copy number (99) (Supplemental Figure 2c). In another case, the AGG codon of the secE gene comprises the SD sequence for the nusG gene and was instead changed to the GAG Glu codon to maintain the activity of the SD sequence (102). A decrease in the transcription rate of an operon due to multiple codon replacements was compensated for by enhancing the promoter activity (113). Altogether, this knowledge is used to develop algorithms and rules for designing recoded genomes, genome segments, and episomal vectors such as plasmids and phages (81, 99, 113, 149). Oligo-mediated recombination allows one to inspect each case (51, 69, 96, 99, 102), whereas the replacement of genomic regions with recoded segments allows one to explore the feasibility of multiple codon replacements throughout the segments (68, 99, 113, 149). In this way, oligo-mediated recombination and genome synthesis are complementary technologies.

STABILIZING EXPANDED GENETIC CODES

Engineered organisms were traditionally forced to maintain the codon assignment for a ncAA lest the codon assignment be lost (73). Bacterial strains incorporating ncAAs based on the selective pressure incorporation method (Figure 1a) are auxotrophic for the amino acid being substituted (87) and thus are dependent on the substituting ncAA in the absence of the canonical amino acid (47, 65). One interesting example is the B. subtilis HR23 strain that was adapted to use 4-fluorotryptophan and can no longer grow without 4-fluorotryptophan even in the presence of Trp (84, 156), indicating that the functions of some essential proteins became sensitive to this ncAA (156). The same strategy is valid for stabilizing expanded genetic codes (Figure 1).

Reassigning a codon is only half the battle to change the genetic code; the other half is to stabilize the reassigned genetic code. Although tolerance for close analogs of a natural amino acid can be evolved by metabolic supplementation (73), codons can be reassigned to structurally diverse ncAAs only after first replacing all essential instances of the codon with synonymous codons (69, 95). Although incorporation of some amino acids at the remaining original sense codons may be deleterious (100, 107), diverse amino acids tend to be well tolerated (69, 95, 100). This means that the new genetic code remains fragile until cell fitness becomes dependent on the new translation function(s). Although this may occur over time due to natural genetic drift, it was also accomplished by reengineering essential proteins to be dependent on a specific ncAA for proper translation, folding, and function (82, 127). A similar strategy was used to establish ncAA dependence for a conditionally essential β-lactamase gene (antibiotic resistance) in E. coli and other gammaproteobacteria (135).

Aside from codifying the reassigned codon function, this strategy presents a mechanism for bio-containment that could be beneficial for industrial applications of recombinant organisms. First, ncAA dependence prevents escape of the recombinant organism into natural environments where the ncAA is not available. Second, utilizing the reassigned codon throughout the genome presents a firewall against horizontal gene transfer with natural organisms (82), preventing the transfer of functional genes between recombinant and natural populations. This same principle of genetic incompatibility applies to viruses, which depend on their hosts to properly translate the proteins that they need to propagate. Indeed, genomically recoded organism C321.ΔA exhibits increased resistance against multiple natural viruses (69, 81). However, viruses can rapidly evolve to match their host’s genetic code (54). In fact, a T7 bacteriophage mutant has been isolated that exhibits improved fitness as a result of incorporating iodotyrosine in gene 17.5 (43). Achieving true multivirus resistance will require the reassignment of additional codons.

BIOMOLECULE-FRIENDLY NONCANONICAL AMINO ACIDS

Despite the availability of aaRS•tRNA pairs that can incorporate more than 167 ncAAs, only a miniscule fraction of the potential chemical space has been explored. Although these ncAAs offer a diverse selection of functional groups, why are there not more studies that report their utility for improving protein function? The answer is not that the ncAAs lack adequate catalytic functions, as evidenced by Pyl and Sec, which perform functions inaccessible to the canonical 20 amino acids. One barrier may be structural, as the majority of available ncAAs are analogs of Pyl and Tyr (i.e., linked to large, hydrophobic side chains; 26) and do not fit into the functional positions of many natural proteins. If we could incorporate any side chain, what would it be? Smaller amino acids may be more versatile for packing in active sites, and hydrophilic side chains may be beneficial for expanding catalytic functions. However, aaRS design may prove more challenging for such amino acids for two reasons: (a) Smaller amino acids will be more difficult to distinguish from canonical amino acids in an orthogonal aaRS active site. (b) It will be challenging to produce aaRS variants that properly satisfy the hydrogen bonds required to accommodate hydrophilic side chains (99).

Although many ncAAs function only in protein translation, a subset interacts strongly with cellular metabolism. For example, 4-aminophenylalanine (pAF) (85), Pyl, and pyrroline-carboxy-lysine (Pcl) (13, 38) can be synthesized in E. coli with heterologous biosynthetic pathways (13, 38, 85). O-phosphoserine (Sep), O-phosphotyrosine (pTyr; 32), and N^ε-acetyllysine (103) are all common posttranslational modifications. These compatibilities with cellular systems, in turn, facilitate degradation and metabolism of free ncAAs, ncAA residues in proteins, and ncAA moieties of ncAA-tRNAs (41, 49, 116) (Figure 1b). Some ncAAs scramble cellular processes (118). However, ncAAs with a posttranslational modification can be cotranslationally incorporated when deacetylases (103) or phosphatases (32, 116) are deleted from the cell or inhibited with an inhibitor (Figure 1b). Furthermore, elimination of tyrosine/aspartate aminotransferases prevented conversion of p-hydroxy-L-phenyllactic acid to tyrosine (41). Depletion of two reductase genes (trxB and gor) prevented the reduction of azido-tyrosine to amino-tyrosine (49). A mutation in the arginine repressor ArgR(L70P) eliminated the toxicity of homoarginine in the E. coli B strain (96, 118). Future systems biology experiments may reveal additional interactions of ncAAs with the metabolome of an organism.

PREPARING FOR RADICALLY ALTERED GENETIC CODES

Protein Engineering Using More Than 21 Building Blocks

Emerging genome engineering technologies and plummeting DNA synthesis prices are now making it possible to rewrite entire genomes, raising interest in creating radically altered genetic codes (8, 113, 149). But why do we want to reassign more than one codon? It may seem a creative exercise to consider applications requiring more than 21 amino acids. Peptide chemists see expanded genetic codes as an opportunity to explore broader conformational and chemical landscapes (8), and the story need not be any different for proteins (80). For example, incorporating multiple bromo/chlorotyrosine residues into redox enzymes has increased enzyme thermostability due to better side chain packing (109). The above-mentioned T7 bacteriophage mutant uses an iodotyrosine residue for quicker propagation (43). Meanwhile, L-(7-hydroxycoumarin-4-yl)ethylglycine improved the activity of a phosphotriesterase beyond its supposed evolutionary limit with canonical amino acids (144). Furthermore, p-acrylamido-phenylalanine has induced a conformational change of an enzyme in a way that the canonical amino acids cannot (154). Covalent bond formation between a ncAA residue and a proximal Cys residue enables irreversible binding of two proteins (153). Directed evolution (140) and protein design (8, 90, 117) will increasingly integrate expanded genetic codes.

Controlling Translational Fidelity

Elimination of several codon assignments does not imply that these blank codons will be assigned to the same number of ncAAs. First, the number will be reduced to approximately half because of wobble pairing by tRNAs. Second, anticodons corresponding to sense codons are often the most important recognition element for aaRSs (39, 64); therefore, the three recoded genome projects are carefully focusing on Ala, Ser, and Leu codons whose cognate aaRS enzymes do not recognize the anticodon as an identity element (71, 113, 149). In addition, base modification of tRNAs is also important for accurate translation and restricted/extended codon decoding by tRNA (40) in not only a static but also a dynamic manner (142). Modified or unmodified U34 in an anticodon could be paired with any nucleotide at the wobble position of codons in the ribosome. In contrast, modified C34 derivatives can be paired with adenine (101, 133). Studies on modification of orthogonal tRNAs have just started (9), but they could provide opportunities to expand the genetic code by splitting anticodons (70).

OUTLOOK

Given the diverse strategies currently used in genetic code engineering, any firm prediction of future breakthroughs would be folly. However, if codon choice were to become limiting in the future, then efforts with quadruplet codons or with codons containing unusual bases would help build on the strategies described in this review. To engineer eukaryotic cells, context-dependent translation termination might be a practical method of stop codon reassignment. Once a variety of hydrophilic ncAAs and unnatural bases are confirmed to be suitable for cell engineering, they will be used to further expand the genetic code by developing new methods of RNA and amino acid recognition. Certainly, in vitro studies will provide the groundwork for successful in vivo systems.