Genetic code flexibility in microorganisms: novel mechanisms and impact on physiology

Abstract

The genetic code, initially thought to be universal and immutable, is now known to contain many variations, including biased codon usage, codon reassignment, ambiguous decoding and recoding. As a result of recent advances in the areas of genome sequencing, biochemistry, bioinformatics and structural biology, our understanding of genetic code flexibility has advanced substantially in the past decade. In this Review, we highlight the prevalence, evolution and mechanistic basis of genetic code variations in microorganisms, and we discuss how this flexibility of the genetic code affects microbial physiology.

Pioneering work with Escherichia coli in the 1960s^1,2 ‘cracked’ the genetic code, which constitutes a cornerstone of molecular biology. The genetic code is established through accurate ligation of each amino acid to its correct (cognate) tRNA by a specific aminoacyl-tRNA synthetase (aaRS). The resulting aminoacyl-tRNA (aa-tRNA) products of the aaRSs read codons by codon–anticodon pairing between mRNA and tRNA on the ribosome, allowing precise translation of the genetic information from mRNA to protein (FIG. 1).

Figure 1. Mechanisms of genetic code flexibility.

a ∣ Each amino acid is attached to the corresponding tRNA by a specialized aminoacyl-tRNA synthetase (aaRS), in a reaction called aminoacylation. For example, threonyl-tRNA synthetase (ThrRS) selects Thr out of the amino acid pool and ligates it on to the 3′ end of tRNA^Thr. The resulting aminoacyl-tRNAs are then delivered to the ribosome by initiation or elongation factors to decode the matching codon. b ∣ There are multiple mechanisms of genetic code flexibility. Codon bias refers to selective usage of synonymous codons to encode the same amino acid. The frequency of codons in a given organism typically matches the cellular abundance of the corresponding tRNA. Codon reassignment requires evolution of a new tRNA to decode sense codons with a new amino acid, or a new tRNA that can decode stop codons with an amino acid. Ambiguous decoding refers to simultaneous decoding of the same codon by two or more amino acids in one cellular compartment; this could be caused by recognition of the same tRNA by more than one aaRS, by misacylation of a tRNA or by ribosomal decoding errors. Recoding traditionally refers to partial codon reassignment that is context dependent. For instance, in certain bacteria and eukaryotes, a subset of UGA stop codons with a nearby selenocysteine (Sec) insertion sequence (SECIS) element, in the presence of SelB, are recoded to Sec, whereas other UGA stop codons retain their ability to signal translational termination.

Originally, it was thought that all living organisms used a universal set of codons, with 61 of the possible 64 nucleotide triplets translating 20 amino acids (termed sense codons) and the 3 remaining codons (UAA, UAG and UGA) being responsible for termination of protein synthesis (termed stop codons). However, it was later discovered that the mitochondrial genetic code of yeast deviates from the standard code, with CUN codons assigned to Thr instead of Leu, and UGA used to encode Trp^3,4. Later sequencing, bioinformatics and biochemical studies of the protein synthesis machinery in microorganisms provided further insights into genetic code variations⁵ and revealed deviations from the standard genetic code in various microorganisms, including bacteria, archaea, fungi and viruses (see Supplementary information S1 (table)). These genetic code variants retain many features of the standard code, but the known exceptions are diverse, and they have evolved through distinct biochemical mechanisms (see Supplementary information S1 (table)). In this Review, we have broadly categorized these mechanisms into four types: biased codon usage, codon reassignment, ambiguous decoding and natural genetic code expansion (FIG. 1).

Codon degeneracy allows each amino acid to be decoded by more than one codon. These synonymous codons can be recognized by different tRNA isoacceptors. Some synonymous codons are preferentially used over others at higher frequencies, leading to biased codon usage, which is found in almost all sequenced genomes. Optimal codon usage typically correlates with high protein synthesis rates, particularly for highly transcribed genes⁶. Below, we provide updated views of how codon usage affects translation and the resulting biological consequences that are only beginning to be understood⁷.

Codon reassignment completely changes the meaning of a codon throughout the transcriptome. The most common example of codon reassignment occurs in microorganisms in which stop codons were reassigned to encode amino acids, but sense codons have also been reassigned. Codon reassignment events have been identified in viruses and in a wide range of microorganisms, including microbial eukaryotes and their mitochondria. Below, we focus on recently characterized codon reassignment mechanisms in yeast and bacterial systems.

Ambiguous decoding (also known as mistranslation) refers to simultaneous decoding of the same codon by two or more amino acids in one cellular compartment. Ambiguous decoding can be caused by recognition of the same tRNA by more than one aaRS or by misacylation of a tRNA with a non-cognate amino acid, which both result in different amino acids being loaded onto tRNAs recognizing the same codon. It can also be caused by ribosomal decoding errors, in which tRNAs are mismatched to their assigned codons. Because ambiguous decoding can lead to errors in protein synthesis, it is usually regarded as deleterious, but increasing evidence suggests that this is a widespread mechanism in nature and may increase microbial fitness under certain conditions, such as during stress⁸. Recent advances uncovering the causes and physiological impact of ambiguous decoding are discussed below.

Natural genetic code expansion refers to genetic codes that enable protein synthesis with more than the 20 canonical amino acids. The two known naturally evolved examples include selenocysteine (Sec) and pyrrolysine (Pyl). Whereas codon reassignment occurs at a genome-wide level, recoding refers to site-specific codon re-definition that is dependent on the mRNA sequence context⁹. One of the earliest recognized deviations of the standard genetic code involves recoding by frameshifting¹⁰, which changes the readout of mRNAs (for an extensive review, see REF. 9). Below, we discuss the mechanisms and physiological consequences of reassigning the stop codon UAG to Pyl and recoding the UGA codon to Sec in archaea and bacteria.

In this Review, we provide an update on recent studies of genetic code flexibility in microorganisms, with a focus on novel insights into the mechanisms of biased codon usage, codon reassignment, ambiguous decoding and recoding, and we discuss how these evolutionary events affect cellular fitness.

Biased codon usage

The most widespread mechanism of genetic code flexibility is the biased usage of synonymous codons, which is a universal feature of all microbial genomes (FIG. 2). Biased codon usage was first observed in a comparison of the degenerate codons of sequenced mRNAs, revealing that each gene in a genome exhibited similar preference for certain synonymous codons over others¹¹. For example, in E. coli, Arg codons CGC and CGU occur over ten times more frequently than AGA and AGG. Later studies show that codon bias is present in almost all sequenced genomes⁶. Both mutation (neutral) and selection hypotheses have been proposed to explain codon usage bias (these hypotheses are reviewed in REFS 6,12). According to the mutation hypothesis, codon bias results from mutational pressure that may vary in different organisms, leading to different codon usage across various genomes. By contrast, the selection hypothesis proposes that codon bias provides a selective advantage by optimizing the level and accuracy of protein expression. A combination of both mutation and selection hypotheses has led to a more widely accepted mutation–selection–drift model¹³, which posits that major (abundant) codons that are frequently used are preferred during selection, whereas mutational pressure keeps the minor (rare) codons in the genome.

Figure 2. Biased codon usage.

a ∣ Biased codon usage occurs when synonymous codons are decoded by different tRNA isoacceptors. As some synonymous codons are used at higher frequencies than others, this leads to biased codon usage. b ∣ Although the optimized usage of synonymous codons increases the protein synthesis rate under most conditions, non-optimal codon usage can improve bacterial fitness under certain conditions. For example, codon optimization of the frq gene in Neurospora crassa results in a loss of fitness. frq controls the circadian clock function, and optimizing the codons of frq increases the expression of Frq but leads to defects in folding of this protein. These folding defects impair Frq function in the regulation of the circadian feedback loop (middle panel). Similarly, the kaiB and kaiC genes in the cold-adapted cyanobacterium Synechococcus elongatus are also enriched for non-optimal codons. The kaiB and kaiC genes are critical for regulation of circadian rhythms, and codon optimization increases the protein levels of KaiB and KaiC and prevents the circadian rhythm from switching off, resulting in loss of fitness by the cyanobacteria (bottom panel). Part b of the image adapted from REF. 144, Nature Publishing Group.

Biased codon usage has been shown to affect the expression levels of endogenous and heterologous proteins, and usage of rare codons is generally considered to slow down translation⁶. However, several recent studies have reshaped our view of ‘optimal’ codon usage and protein expression. Ribosome profiling, which measures ribosomal densities at all codons in the transcriptome, of E. coli and Bacillus subtilis grown in rich media showed that rare codons recognized by tRNAs with low cellular abundance are translated at the same speed as abundant codons (FIG. 2). Instead of stalling at rare codons, ribosomes stall most often at Shine–Dalgarno-like sequences that may pair with the anti-Shine–Dalgarno sequence of the 16S ribosomal RNA¹⁴. Another study used a yellow fluorescent protein in E. coli to show that the protein synthesis rate is not affected by synonymous changes in abundant or rare codons in bacteria growing in amino acid-rich media¹⁵. High-throughput translation assays showed that the presence of rare codons located near the 5′ end of an open reading frame surprisingly increase protein expression about fourfold in E. coli¹⁶; this is primarily due to a decrease in the number of mRNA secondary structures, which facilitates the travelling of the ribosome along the mRNA.

Under amino acid starvation conditions, however, synonymous codons are indeed translated at distinct rates owing to the different aminoacylation levels of the corresponding tRNA isoacceptors¹⁵ (FIG. 2). It seems that the abundance of aa-tRNA, rather than the total level of tRNA (either charged or uncharged with amino acids), correlates better with the decoding rates in vivo. Under nutrient-rich conditions, the supply of aa-tRNAs meets the demand of protein synthesis, and all synonymous codons appear to be translated at similar speeds. By contrast, during amino acid starvation, tRNA isoacceptors are charged at various levels, leading to distinct translation rates of synonymous codons. Similarly, other conditions that perturb the relative abundances of charged tRNA isoacceptors, such as alterations in the tRNA expression level and aminoacylation efficiency, may also affect the translation rates of synonymous codons. In addition, non-optimal codons in over-expressed genes could sequester rare tRNAs, therefore reducing the charging levels of such tRNAs to decrease total protein expression. Given that microorganisms in the natural environment are not often supplied with sufficient nutrients, maintaining optimal codons with biased usage of synonymous codons probably has an important role in maximizing the overall translational speed and growth.

Although optimized usage of synonymous codons increases the protein synthesis rate under certain conditions, non-optimal codons used at specific locations sometimes lead to increased bacterial fitness. For example, in B. subtilis, non-optimal Ser codons are preferentially used in the gene encoding SinR, which inhibits biofilm formation¹⁷. Experimental substitution of non-optimal Ser codons with optimal Ser codons increases the synthesis rate of SinR, which leads to increased protein levels of SinR and suppresses biofilm development. Under physiological conditions, depletion of cellular Ser levels during stationary phase reduces translation of the native SinR to trigger biofilm synthesis. Therefore, the apparent non-optimal Ser codons in SinR serve as a molecular sensor for the cellular levels of Ser, and this strategy enables B. subtilis to control the expression of biofilm-associated genes in response to environmental conditions.

In the cold-adapted cyanobacterium Synechococcus elongatus, the kaiB and kaiC genes are critical for the regulation of circadian rhythms, and these genes are usually highly expressed¹⁸. The codon adaptation index of kaiB and kaiC is below average, which means that kaiB and kaiC are enriched in non-optimal codons. Notably, experimental codon optimization of kaiB and kaiC increases the protein levels of KaiB and KaiC at low temperature, which prevents the circadian rhythm from switching off, resulting in loss of fitness. Therefore, these data demonstrate that S. elongatus uses non-optimal codons as a mechanism to regulate bacterial fitness (FIG. 2). Similarly, in Neurospora crassa, the frq gene, which controls the circadian clock function, also exhibits non-optimal codon usage¹⁹. Optimization of the codons of frq increases the protein levels of Frq but also leads to protein folding defects and impairs its function in the circadian feedback loop. This suggests that the non-optimal codons serve as checkpoints to slow down translation of certain regions in Frq, allowing proper folding of the protein (FIG. 2). In these examples, non-optimal codons actually provide microorganisms with selective advantages for adaptation to environmental changes.

Codon reassignment

Codon reassignment events include stop-to-sense, sense-to-stop and sense-to-sense codon changes, and occur through multiple mechanisms (see Supplementary information S1 (table)). Whereas sense-to-stop reassignments only require loss of the cognate tRNA, stop-to-sense and sense-to-sense changes require evolution of new tRNAs. Such new tRNAs may arise from modification of an existing tRNA or by tRNA duplication²⁰. For example, reassignment of the UGA codon from stop to Trp is accompanied by either a mutation²¹ or post-transcriptional editing (C34→U34)²² that converts tRNA^Trp_CCA to tRNA^Trp_UCA, which reads both the UGA and UGG codons as Trp. In bacteria and archaea, both tRNA^Met_CAU and tRNA^Ile_CAU contain the same CAU anticodon. Whereas tRNA^Met_CAU recognizes the AUG codon, tRNA^Ile_CAU is modified at position C34 by addition of lysidine²³ or agmatidine^24,25. As a result, the modified tRNA^Ile_CAU specifically recognizes AUA instead of AUG. In the mitochondria, tRNA^Ile_CAU is absent and tRNA^Met_CAU contains a formylcytidine (f⁵C) modification at C34 (REF. 26), which allows tRNA^Met_CAU to recognize A at the third position (wobble position). A recent structural study confirmed that f⁵C of tRNA^Met_CAU pairs with both A and G at the wobble position, enabling it to read both AUA and AUG codons as Met²⁷. Consequently, AUA codons are reassigned from Ile to Met in the mitochondria.

A more complicated scenario involves tRNA evolution from duplication. For example, CUN codons are reassigned from Leu to Thr in Saccharomyces cerevisiae mitochondria owing to the loss of tRNA^Leu_UAG and the presence of tRNA^Thr_UAG, which has an enlarged eight-nucleotide (as opposed to seven-nucleotide) anticodon loop³. An unmodified U at the wobble position in mitochondrial tRNAs allows pairing with all four nucleotides²⁸. Interestingly, phylogenetic analyses reveal that tRNA^Thr_UAG is not related to either the ancestral tRNA^Leu_UAG or the isoacceptor tRNA^Thr_UGU but shows high similarity to the mitochondrial tRNA^His_GUG (REF. 29). Further biochemical studies verify that two nucleotide changes are sufficient to switch the tRNA^His_GUG from a His-accepting to Thr-accepting molecule²⁹. These results suggest that the tRNA^Thr_UAG is evolutionarily derived from a precursor tRNA^His_GUG. In support of this model, a duplicated copy of tRNA^His_GUG is found in the mitochondrial genome of Candida albicans and Kluyveromyces lactis, which are closely related to S. cerevisiae (FIG. 3).

Figure 3. Codon reassignment.

a ∣ CUN codons are read as Leu in the standard genetic code through the use of tRNA^Leu_UAG. However, in some microorganisms, CUN codons have been reassigned to Thr. b ∣ CUN reassignment in mitochondria involved several steps. In the mitochondrial genome of some Saccharomycetaceae species (for example, Candida albicans), the tRNA^His_GUG gene was duplicated. The CUN codons and a tRNA^Leu_UAG that decodes CUN then disappeared, leading to a reduced genetic code, as in Kluyveromyces lactis. In Saccharomyces cerevisiae, one copy of tRNA^His_GUG evolved to carry an anticodon UAG that reads CUN codons (tRNA^His_UAG). The threonyl-tRNA synthetase (ThrRS) co-evolved with the tRNA^His_UAG to recognize it as a tRNA^Thr_UAG. CUN codons then reappeared to complete the codon reassignment from Leu to Thr. In Ashbya gossypii, secondary mutations enabled the new tRNA^His_UAG to be recognized by AlaRS instead of HisRS or ThrRS, therefore generating tRNA^Ala_UAG and reassigning CUN codons to Ala in this species. c ∣ Codon reassignment can affect bacteria and phage physiology. In a phage from the human oral cavity, reassigning UAG codons to Gln may allow the phage to interfere with translation of host genes without affecting translation of the phage genes. In the early stages of infection, the phage early genes contain few in-frame UAG codons and are therefore efficiently translated by the host machinery. Among these genes, the phage expresses release factor 2 (RF2), which suppresses translation of UGA codons, including recoded ones. As the bacterial protein RF1, which suppresses translation of UAG codons, contains multiple in-frame UGA codons, expression of RF2 by the phage inhibits translation of host RF1. This inhibition allows the phage to translate late-stage phage genes, which are enriched in recoded UAG codons, while modifying the translation of host genes. Part b is adapted from Su, D., Lieberman, A., Lang, B. F., Simonovic, M., Söll, D. & Ling, J., An unusual tRNA^Thr derived from tRNA^His reassigns in yeast mitochondria the CUN codons to threonine, Nucleic Acids Res., 2011, 39, 11, 4866–4874, by permission of Oxford University Press. Part c is adapted from Ivanova, N. N. et al. Stop codon reassignments in the wild. Science 344, 909–913 (2014). Reprinted with permission from AAAS.

In another closely related species, Ashbya gossypii, a tRNA^Ala_UAG derived from either tRNA^His_GUG or tRNA^Thr_UAG reassigns mitochondrial CUA and CUU codons to Ala, and CUC and CUG codons are absent in mitochondrial genes³⁰. This leaves UUG and UUA as the only Leu codons in the A. gossypii mitochondrial genome.

In addition to tRNA, aaRSs have evolved to enable codon reassignment, as highlighted by CUN reassignment in S. cerevisiae mitochondria. The S. cerevisiae threonyl-tRNA synthetase (ThrRS) recognizes both tRNA^Thr_UAG and tRNA^Thr_UGU (REF. 31), whereas E. coli and the mitochondrial ThrRSs from Schizosaccharomyces pombe and C. albicans do not recognize tRNA^Thr_UAG, suggesting that the mitochondrial ThrRS from S. cerevisiae has co-evolved with tRNA^Thr_UAG to allow for CUN recoding (FIG. 3).

Together, these data suggest that CUN reassignment in S. cerevisiae and in A. gossypii involved several steps. First, in the mitochondrial genome of some Saccharomycetaceae species (such as C. albicans), the tRNA^His_GUG gene was duplicated. The CUN codons and a tRNA^Leu_UAG that decodes CUN then disappeared, leading to a reduced genetic code that does not involve CUN, such as in K. lactis. Following this reduction, in S. cerevisiae, one copy of tRNA^His_GUG evolved to carry a UAG anticodon that reads CUN codons (tRNA^His_UAG). The ThrRS co-evolved with the tRNA^His_UAG to recognize it as a tRNA^Thr_UAG. CUN codons then reappeared to complete the codon reassignment from Leu to Thr. In A. gossypii, secondary mutations enabled the new tRNA^His_UAG to be recognized by AlaRS instead of HisRS or ThrRS, thereby generating tRNA^Ala_UAG and reassigning CUN codons to Ala in this species (FIG. 3).

The driving forces and physiological consequences of codon reassignment are not completely understood. For example, reassignment of CUN codons in yeast mitochondria is considered to be neutral because the original CUN codons had been lost, new CUN codons appeared at non-conserved positions, and a new tRNA evolved, which apparently did not involve a period of ambiguous translation of mitochondrial proteins. However, a recent bioinformatic analysis of >5 trillion bp of metagenomic data revealed that stop codon reassignment is widespread in microorganisms and that differential reassignment between bacteria and phages could have physiological consequences³². For example, the authors found a phage from the human oral cavity that reassigned the UAG stop codon to Gln. Interestingly, the phage genes expressed at early stages contain few in-frame UAG codons and can therefore be efficiently translated by the host machinery. However, one of the phage proteins expressed during the early stages is release factor 2 (RF2), which suppresses translation of UGA codons. As the bacterial protein RF1, which suppresses translation of UAG codons, contains multiple in-frame UGA codons, expression of RF2 by the phage inhibits translation of host RF1. Inhibition of host RF1 allows the phage to translate late-stage phage genes, which are enriched in recoded UAG codons (FIG. 3). This intriguing model suggests that phages can use codon reassignment to interfere with translation of host genes without affecting translation of the phage genes, but this model needs to be experimentally validated in future studies.

Insight into the consequences of codon reassignment also comes from studies of microorganisms in which the genetic code has been re-engineered (BOX 1). Several groups have deleted RF1 in an attempt to reassign UAG to a sense codon in E. coli by subsequently introducing an orthogonal aaRS–tRNA^33–38 pair to incorporate a non-canonical amino acid. Notably, when the native UAG codons are present, deleting RF1 and reassigning UAG as a sense codon causes severe growth defects^34,35. Converting all endogenous UAG codons to UAA rescues the growth phenotype³⁶, suggesting that the observed toxicity is due to extension of proteins beyond their intended stop. In the RF1 deletion mutant, UAG is also translated by natural amino acids (predominantly Gln, Tyr and Lys), which compete with the non-canonical amino acid insertions, resulting in impure proteins³⁹. These studies indicate that radical codon reassignment may decrease fitness in some aspects, such as growth and proteome stability. Therefore, during the evolution of codon reassignment, the gained benefits may need to outweigh the potential negative effects. In one such example, phages growing in E. coli cells that reassign UAG from stop to a non-canonical amino acid (3-iodotyrosine) rapidly accumulate in-frame UAG codons from mutations; such speeded codon reassignment in the phage provides benefits to the phage by increasing its titre owing to the presence of the non-canonical amino acid in the proteome⁴⁰. An expanded genetic code can thus increase the evolvability of codons. Future studies are needed to identify the phenotypic and proteomic changes in organisms in which the genetic code has been re-engineered and to clarify the underlying molecular mechanisms of how a cell adapts to genetic code expansion.

Box 1. Emerging biotechnology for engineering the genetic code.

The flexibility of the genetic code as illustrated by natural codon reassignments has prompted researchers to expand the genetic code with non-canonical amino acids¹⁰⁸. Both site-specific^109,110 and residue-specific¹¹¹ approaches have been developed to co-translationally insert non-canonical amino acids into proteins of interest. The site-specific approach uses an orthogonal pair of aminoacyl-tRNA synthetase (aaRS)–tRNA to reassign a stop or quadruplet codon to a non-canonical amino acid. By contrast, the residue-specific approach allows ambiguous decoding of a sense codon by both the native amino acid and a non-canonical amino acid. Site-specific sense codon recoding is also being explored as a means to further expand the genetic code⁶². Genetic code expansion with non-canonical amino acids has been applied to site-specifically label proteins with fluorophores, study post-translational modifications and identify newly synthesized proteins¹⁰⁹. Given the flexibility of the active site and orthogonality in bacteria and eukaryotes, tyrosyl-tRNA synthetase (TyrRS) and PylRS have been most widely used for site-specific insertion of non-canonical amino acids (reviewed in REFS 108,112). Recently, non-canonical amino acids have been used as safeguards to contain recoded bacteria^113,114. In such synthetic microorganisms, stop codons are introduced into essential genes that require translation with a non-canonical amino acid. Therefore, the viability of the recoded bacteria depends on the presence of the non-canonical amino acid, preventing them from proliferating in natural ecosystems. As an additional safety layer, stop-codon-interrupted genes should be less susceptible to horizontal gene transfer (HGT), thus helping to prevent transmission of genes from synthetic organisms to natural genomes. A similar mechanism may also occur in nature. The bacterium SR1 encodes Gly with five codons, including the UGA ‘stop’ codon. This genetic code variation was hypothesized to bias HGT, because 85% of SR1’s genes contain UGA sense codons, which would be interpreted as stop codons by recipient microorganisms¹¹⁵.

Recent advances in genome editing and DNA synthesis provide powerful tools to engineer the genetic code and understand the evolution of codon reassignments. The goal of these efforts is to remove codons from the genome so that they can be reassigned to new and non-canonical amino acids. Multiplex automated genome engineering (MAGE)¹¹⁶ and conjugative assembly genome engineering (CAGE)¹¹⁷ approaches were developed to allow simultaneous and genome-wide editing of the Escherichia coli genome. MAGE uses synthetic single-stranded DNA fragments to simultaneously target multiple chromosomal sites and introduce precise genomic changes by recombination-based genetic engineering, whereas CAGE uses a conjugation strategy to move large regions of the chromosome between different strains. A combination of MAGE and CAGE has allowed complete reassignment of UAG to UAA in an E. coli strain³⁶. Another potent genome-editing tool is based on the CRISPR system, which permits convenient genome editing in bacterial and eukaryotic cells^118,119. Complete de novo synthesis of a Mycoplasma genome¹²⁰ and a yeast chromosome¹²¹ have also been achieved. Together, these technological advances set the stage for synthetic organisms (with potentially radically different genetic codes) that will provide novel opportunities for biotechnology and new insights into the evolution of the genetic code.

Ambiguous decoding

In contrast to biased codon usage in which a set of codons are translated by the same amino acid, a specific codon can also be ambiguously decoded by more than one amino acid⁴¹. Ambiguous decoding can result from multiple aaRSs recognizing the same tRNA, which leads to different amino acids being loaded onto tRNAs recognizing the same codon; from errors in the aminoacylation reaction carried out by a specific aaRS, which loads a tRNA with a non-cognate amino acid; or from ribosomal decoding errors (FIG. 4). Ambiguous decoding leads to a statistical pool of protein products with amino acid substitutions at various positions (statistical proteome). High levels of ambiguous translation lead to accumulation of misfolded proteins that are toxic to cells.

Figure 4. Ambiguous decoding.

a ∣ Ambiguous decoding refers to the process by which the same codon gives rise to incorporation of different amino acids in a nascent polypeptide chain. b ∣ Ambiguous decoding can result from errors in the aminoacylation reaction carried out by a specific aminoacyl-tRNA synthetase (aaRS), which loads a tRNA with a non-cognate amino acid; by multiple aaRSs recognizing the same tRNA, which leads to different amino acids being loaded onto tRNAs recognizing the same codon; or by ribosomal decoding errors. c ∣ In Candida albicans, altering the ratio of Ser to Leu incorporation at CUG codons introduces diverse cell and colony morphologies. Wild-type C. albicans encodes a CUG-decoding tRNA_CAG that is recognized by both seryl- and leucyl-tRNA synthetases. This ability of ambiguous decoding enables wild-type cells to display different colony morphologies, including smooth, ring, wrinkled and hyphae. Eliminating the ambiguity by the substitution of the tRNA_CAG with a Leu-specific tRNA results in loss of the smooth and ring morphologies. These changes also reduce cell adhesion and increase fungal susceptibility to macrophage killing by immune cells.

Multiple quality control mechanisms are used to ensure fidelity during protein synthesis. For instance, some aaRSs use editing sites to hydrolyse mismatched aa-tRNAs⁴². Similarly, the ribosome selects the cognate aa-tRNA based on both preferential binding of the cognate aa-tRNA at the decoding centre and kinetic proofreading of non-cognate aa-tRNAs^43,44. However, despite the existence of these complex mechanisms that maintain translational fidelity (reviewed in REFS 42,44), ambiguous decoding can still result from reduced fidelity during protein synthesis owing to either mutations⁴⁴ or stress⁴⁵. For instance, mutations in the ribosome allow mismatch between the mRNA codon and tRNA anticodon, and oxidative stress decreases fidelity during aminoacylation.

Increasing evidence suggests that ambiguous decoding may be used as an adaptive mechanism by microorganisms to survive harsh environmental conditions and gain a selective advantage during evolution. Some aaRSs use editing to ensure correct pairing between amino acids and tRNAs⁴⁶, but the editing domains of ThrRS, LeuRS and PheRS are either lost or defunct in Mycoplasma spp. and in mitochondria from yeasts^47–49, indicating that ambiguous translation occurs in these organisms and organelles. Indeed, analyses of mass spectrometry data reveal that the intracellular bacterium Mycoplasma mobile encodes a statistical proteome⁴⁸, which was suggested to benefit this organism by increasing the proteomic and phenotypic diversity of the bacterium, enabling it to escape host defences⁴⁸.

Direct experimental evidence of phenotypic diversity created by a statistical proteome comes from elegant studies of CUG codon ambiguity in C. albicans. Altering the ratio of Ser to Leu incorporation at CUG codons introduces diverse cell and colony morphologies, as well as distinct antifungal and immune responses^50,51. Wild-type C. albicans encodes a CUG-decoding tRNA_CAG that is recognized by both SerRS and LeuRS (FIG. 1b). The wild-type cells display colony morphologies, including smooth, ring, wrinkled and hyphae⁵⁰. Substituting the ambiguous tRNA_CAG with a Leu-specific tRNA abolishes CUG ambiguity as well as the smooth and ring morphologies⁵⁰ (FIG. 4). Enhanced CUG ambiguity increases cell adhesion by ambiguous translation of adhesins and reduces susceptibility to macrophage killing by decreasing surface exposure of β-glucans⁵¹. It has also been shown that, in Mycobacterium smegmatis, the level of ambiguous decoding at Asn codons increases during stationary phase and under low pH conditions⁵². Ambiguous translation increases resistance against rifampicin by producing RpoB variant proteins that are no longer recognized by this antibiotic, therefore epigenetically enhancing the fitness of M. smegmatis⁵² (FIG. 4).

Ambiguous decoding not only increases the diversity of the proteome but may also activate stress responses. A recent study reveals that ambiguous translation caused by a mutation in the ribosome activates the bacterial general stress response and enhances tolerance to hydrogen peroxide⁵³. In E. coli, oxidative stress induces ambiguous decoding at ACN codons, which are recognized by tRNA^Thr. ThrRS uses editing to hydrolyse misacylated Ser-tRNA^Thr, and the editing site of ThrRS contains an activated Cys that is hypersensitive to oxidation⁵⁴. Oxidative stress thus impairs ThrRS editing and causes incorporation of both Thr and Ser into ACN codons (FIG. 4). The reactive Cys and activating residues at the ThrRS editing sites are conserved among bacteria, suggesting that oxidative-stress-induced ambiguous decoding is a highly conserved mechanism. Switching between accurate and ambiguous decoding may allow bacteria to adapt to different environmental conditions quickly, but this possibility requires additional testing in future studies.

Natural genetic code expansion

In 1966, Crick stated: “A more serious problem is whether in a normal cell a triplet can be read in more than one way.” (REF. 55). Natural genetic code expansion encompasses changes in the genetic code — which evolved in natural organisms — that enable protein synthesis with more than the 20 canonical amino acids. Sec and Pyl are the only known cases of natural genetic code expansion. The mechanisms of Sec and Pyl translation are distinct and rarely co-occur in the same organism (BOX 2).

Box 2. Natural genetic code expansion.

Selenocysteine (Sec) biosynthesis evolved twice independently¹²² (FIG. 5). This is not in itself a unique feature of Sec, as two independently evolved aminoacylation systems exist for Lys, Gln, Asn, Gly, Cys¹²³ and possibly Ser¹²⁴. The fact that Sec is exclusively biosynthesized on its tRNA, however, is unusual. Sec is also the only amino acid for which a specific aminoacyl-tRNA synthetase (aaRS) does not exist. Initially, the normal seryl-tRNA synthetase (SerRS) ligates Ser onto tRNA^Sec, and the Ser-tRNA^Sec product is converted to Sec-tRNA^Sec (FIG. 5). This reaction is catalysed by a single pyridoxal-phosphate (PLP)-dependent enzyme in bacteria (SelA), which forms a stunning decameric ring structure that simultaneously binds to ten tRNA^Sec molecules¹²⁵. In archaea¹²⁶ and eukaryotes¹²⁷, a unique kinase (phosphoseryl(Sep)-tRNA^Sec kinase (PSTK)) phosphorylates the tRNA-bound Ser to form phosphoseryl (Sep)-tRNA^Sec, and a distant relative of SelA, a dimeric enzyme known as Sep-tRNA^Sec:Sec-tRNA^Sec synthetase (SepSecS), produces the Sec-tRNA^Sec substrate required to insert Sec into proteins. SepSecS is dimeric, like most members of the PLP enzyme family, and binds to two tRNA^Sec molecules¹²⁸. Both SelA and SepSecS require the selenium donor selenophosphate, which is the product of selenophosphate synthase (SelD), to complete the Sec synthesis reaction.

Adding to the complexity of selenoprotein biosynthesis, a specialized elongation factor (SelB in Escherichia coli, EF-Sec in eukaryotes) and a RNA recoding signal (Sec insertion sequence (SECIS)) are also required to convert the meaning of UGA from stop to Sec. In archaea and bacteria, the SECIS is found in the reading frame of the selenoprotein transcript, within a few nucleotides of the recoded UGA codon. In eukaryotes, the SECIS (of which there are two structurally divergent types) is found in the 3′ untranslated region (UTR), and additional proteins are required for Sec insertion (including SECIS-binding protein 2 (REF. 129)) or for regulation of the recoding event (including ribosomal protein L30 (REF. 130) and initiation factor 4a3 (REF. 122)). Despite numerous biochemical and structural studies on Sec synthesis and recoding in bacteria⁶¹ and eukaryotes¹³¹, the structural basis of UGA recoding on the ribosome remains unresolved. A structural model of the bacterial system¹³² indicates that the SECIS–SelB–Sec-tRNA^Sec complex may fit between the ‘body’ and ‘head’ regions of the 30S ribosomal subunit, which could alter the conformation of the decoding centre on the translating ribosome.

Following the discovery of Sec, pyrrolysine (Pyl) is the second known case of natural genetic code expansion. Genetic code expansion with Sec and Pyl arose by distinct evolutionary and biochemical mechanisms¹³³. Unlike Pyl, which appears in proteins as either a catalytic⁸⁴ or non-catalytic⁹³ residue, Sec nearly always has essential roles in the active sites of redox enzymes²⁸. Another feature that distinguishes Pyl from Sec is that incorporation of Sec requires a SECIS mRNA structure, whereas Pyl can effectively provide informational suppression¹³⁴ of UAG codons without a co-evolved mRNA secondary structure⁹⁷. The flexibility of tRNA^Pyl in UAG translation¹³⁵, together with the natural orthogonality of the PylRS–tRNA^Pyl pair in bacterial and eukaryotic hosts, and the large amino acid-binding pocket of PylRS^107,136, prompted multiple groups to engineer PylRS as a facile system for genetic incorporation of non-canonical amino acids in bacterial and mammalian cells^{109,112,137–141}. The Sec system was recently shown to be a powerful mechanism for recoding sense codons⁶². We anticipate that both systems will continue to drive further genetic code engineering that is inspired by nature (BOX 1).

Fascinatingly, Sec and Pyl may coexist in some organisms. Two genome sequences, that of Acetohalobium arabaticum and one from metagenomic sequencing of a bacterial symbiont of the worm Olavius algarvensis¹⁴² show evidence of genetic codes with 22 amino acids. Indeed, two Gly reductase selenoprotein B (grdB) genes in A. arabaticum contain both a SECIS-recoded UGA codon for Sec incorporation as well as one or two UAG codons, which may be translated as Pyl in the presence of trimethylamine. A. arabaticum was shown to decode UAG as Pyl¹⁰⁴ and because Sec in GrdB is required for Gly reductase activity¹⁴³, this organism may produce the only known natural proteins that contain 22 genetically encoded amino acids.

Recoding UGA to Sec

Most UGA codons signal termination of protein synthesis. Certain organisms, including E. coli and humans, have a naturally expanded genetic code and recode some UGA codons to insert Sec into selenoproteins. Therefore, UGA takes on two meanings in the same cell and sometimes even in the same open reading frame⁵⁶, illustrating the “more serious problem” foreshadowed by Crick⁵⁵. These naturally expanded genetic codes specify 21 rather than the ‘usual’ 20 amino acids.

Sec is chemically similar to Cys, but the lower redox potential and higher nucleophilicity of Sec explain why some selenoenzymes are far more reactive than their Cys-containing counterparts⁵⁷. For example, mammalian Met-R-sulfoxide reductases containing Sec in the active site are 100-fold more active than the Cys-containing variants⁵⁸. Sec is also found to be more resistant to irreversible oxidation than Cys⁵⁹ and has utility as a probe for protein structure and function⁶⁰.

Sec biosynthesis occurs on its cognate tRNA, beginning with a Ser-tRNA^Sec precursor (reviewed in REF. 61) (BOX 2). Particular UGAs are selected for recoding to Sec by the presence of a downstream Sec insertion sequence (SECIS) (FIG. 5). A special elongation factor (SelB) that binds to both the SECIS and Sec-tRNA^Sec on the ribosome is also required (BOX 2). Given the complexity of UGA recoding to Sec, it is not surprising that codon recoding is rare in nature. The benefit (and significance) of recoding, as opposed to codon reassignment, is that nature found a way to expand the genetic code that does not interfere with normal protein synthesis.

Figure 5. Expanding the genetic code with Sec and Pyl.

a ∣ Protein synthesis with selenocysteine (Sec). Sec is biosynthesized on its tRNA. This occurs in multiple steps, beginning with Ser-tRNA^Sec formation catalysed by the normal seryl-tRNA synthetase (SerRS). In bacteria, Ser-tRNA^Sec is converted to Sec-tRNA^Sec by the Sec synthase (SelA). In archaea and eukaryotes, Ser-tRNA^Sec is first phosphorylated by phosphoseryl(Sep)-tRNA^Sec kinase (PSTK) to generate pSer-tRNA^Sec and then an enzyme related to SelA known as Sep-tRNA^Sec:Sec-tRNA^Sec synthase (SepSecS) converts pSer-tRNA^Sec species to Sec-tRNA^Sec. A specialized elongation factor (SelB) simultaneously binds to Sec-tRNA^Sec as well as the Sec insertion sequence (SECIS) element to direct recoding on the ribosome of specific UGA codons in selenoprotein mRNAs. b ∣ Protein synthesis with pyrrolysine (Pyl). Pyl is biosynthesized as a free amino acid in the cell. PylRS ligates Pyl to tRNA^Pyl, which contains an anticodon (5′-CUA-3′) that reads UAG codons. Like canonical aminoacyl-tRNAs, Pyl-tRNA^Pyl is bound by elongation factor Tu (EF-Tu), enabling UAG translation with Pyl on the ribosome. The Pyl system does not require a specialized elongation factor.

Sec is not hardwired to UGA codons. Recent experiments demonstrate that the Sec machinery is able to recode nearly every codon in the genetic code table⁶². A tRNA molecule was engineered to insert Sec at the UAG codon using elongation factor Tu (EF-Tu), but even with optimal expression of the Sec synthesis components, the precursor Ser-tRNA^Sec was incorporated only 30% of the time⁶³. Ambiguous decoding of Ser and Sec in this SECIS-independent route to selenoprotein synthesis supports the notion that the elaborate Sec recoding machinery may have evolved to enhance the fidelity of Sec insertion into proteins⁶⁴. Further tRNA refinements created synthetic tRNAs^65,66 that, together with EF-Tu⁶⁷ mutants, led to efficient site-specific Sec incorporation⁶⁴.

Furthermore, Sec-tRNA^Sec is capable of wobble decoding, and a Sec-tRNA^Sec_CCA mutant has been shown to efficiently decode the UGA codon in a thioredoxin reductase⁶⁸. In E. coli, mutation of the tRNA^Sec to each of the other 63 anticodons shows that Sec can be inserted in response to 60 of the codons in the genetic code^62,66. In fact, for 15 codons, Sec-tRNA^Sec can completely out-compete cognate aa-tRNA and provide selenoprotein yields that are tenfold greater than when Sec is encoded by UGA⁶². These findings indicate that microorganisms could encode Sec with a codon other than UGA and bring into question as to why UGA is the ‘chosen’ Sec codon. It is, therefore, plausible that even in nature Sec is not hardwired to UGA.

Interestingly, Sec recoding is found in all three domains of life, but not in all organisms. Sec is normally found in redox enzymes, and these selenoproteins have important roles in central energy metabolism in archaea and bacteria as well as in cellular defence against reactive oxygen species.

Approximately 20% of sequenced bacterial taxa encode Sec⁶⁹. Some bacteria encode only a single selenoprotein, whereas others such as the δ-proteobacterium Syntrophobacter fumaroxidans encode 31 (REF. 69). Bacterial selenoproteins belong to 58 protein families⁷⁰. The number of selenoprotein genes in bacterial species differs greatly, in part because replacement of Sec with Cys is common in the evolution of bacterial selenoproteins⁶⁹. The majority of bacterial selenoproteins, including formate dehydrogenase-α subunit (FdhA), Gly reductase B and glutathione peroxidase, are homologous to thiol-based redox enzymes^69,71. The most common bacterial selenoproteins are selenophosphate synthase (SelD) (BOX 2) and FdhA. E. coli formate dehydrogenase (FDH_H) is one of the most studied selenoproteins⁷². FDH_H is a component of the formate hydrogen lyase complex and functions to reduce formate and shuttle electrons to the respiratory chain. The Sec residue in FDH_H is essential for its function, and replacement with Cys reduces the catalytic efficiency of the enzyme 200-fold⁷³.

Bacterial selenoproteins are involved in other central metabolic pathways, including purine degradation, and the energy-conserving acetogenesis pathway⁷⁴. Environmental sequencing of microbial genomes revealed many new potential selenoproteins and novel selenoprotein families, such as Ser proteases and formamidase regulatory proteins, in which the role of Sec is unknown⁷⁰.

Early work demonstrated that Sec is dispensable in E. coli and Salmonella enterica subsp. enterica serovar Typhimurium⁷⁵, but our understanding of how natural expansion of the genetic code with Sec affects bacterial physiology is still limited. A recent report found that concentrations of sodium selenite (a necessary precursor of Sec biosynthesis) in the millimolar range were detrimental to E. coli growth, but concentrations of 1–10 nM stimulated cell growth⁷⁶. Certain bacterial pathogens (such as Staphylococcus aureus) rely completely on the selenoprotein thioredoxin reductase to survive oxidative stress⁷⁷. Further work will be required to understand the selective value of genetic code expansion with Sec in bacteria.

Sec-decoding archaea are confined to closely related species⁷⁸, including Methanopyrus kandleri and all members of the order Methanococcales. The archaeal selenophosphate synthase is a selenoprotein, but most archaeal selenoproteins are involved in methanogenesis, which is the main energy production pathway for these microorganisms⁷⁸. Sec is found in FDH_H, formylmethano furan dehydrogenase, F₄₂₀ reducing and non-reducing hydrogenases, heterodisulfide reductase and a HesB-like protein, which has been implicated in iron–sulfur cluster assembly^79–81. These selenoproteins are involved in redox reactions and cofactor regeneration, facilitating the transfer of electrons in the reduction of formate to produce methane.

Genetic experiments in Methanococcus maripaludis indicate that archaeal selenoproteins are more active than Cys-containing homologues and that the metabolic range and efficiency of the organism decreases when their genetic code is reduced from 21 to 20 amino acids^82,83. In M. maripaludis JJ, genetic inactivation of the Sec-specific elongation factor SelB leads to overexpression of Cys-containing versions of each of its selenoproteins, suggesting that the Sec version is catalytically more active⁸². M. maripaludis JJ cannot sustain growth on formate because there is no Cys paralogue for FDH_H, and a 15% decrease in growth rate was observed in cells consuming H₂ and CO₂.

Reassigning UAG to Pyl

Pyl was first discovered in methanogenic archaea and identified as the twenty-second genetically encoded amino acid^84,85 (BOX 2). A recent search of the US National Center for Biotechnology Information and the US Joint Genome Institute sequence databases shows that there are 46 microbial species (21 archaea and 25 bacteria) that contain all of the genes that are necessary to synthesize Pyl-containing proteins. Recently, a new clade of Pyl-decoding methanogens was discovered in Methanomassiliicoccales species related to the euryarchaeal order Thermoplasmatales⁸⁶. The sequences hint at potential new roles for Pyl in CRISPR-associated protein Cas1 and digeranylgeranylglyceryl phosphate synthase⁸⁷, but it is still debatable whether all UAG codons in Pyl-decoding organisms lead to Pyl insertion.

The biochemical mechanism required to genetically encode Pyl is distinct and orthogonal to the system that evolved to insert Sec into proteins. Free Pyl is biosynthesized in the cell from two molecules of Lys by three enzymes that are the products of the pylB, pylC and pylD genes⁸⁸. Structural and biochemical work on Pyl biosynthesis was recently reviewed⁸⁹. Pyl is then ligated onto tRNA^Pyl by PylRS^90,91. The CUA anticodon of tRNA^Pyl reads the UAG codon and reassigns the meaning of UAG from stop to Pyl (FIG. 5).

Natural incorporation of Pyl was first experimentally verified in a monomethylamine methyltransferase (MtmB)⁸⁵, when crystallographic structures of Pyl-containing MtmB (purified from Methanosarcina acetivorans) provided an atomic-level view of Pyl in the active site⁸⁴. A hypothetical structure-based reaction mechanism suggested that Pyl may function to properly orient the methylamine substrate for transfer of the methyl group to a cognate corrinoid protein (MtmC). Related bacterial methyltransferases that lack Pyl are now known to use Gly betaine as a substrate⁹², a quaternary amine and a precursor of the tertiary amine trimethylamine (TMA). The authors argue that, because Gly betaine is a quaternary amine, it is already an ‘activated’ methyl donor, possibly providing a clue as to the role of Pyl in the methylamine methyltransferases. In the proposed reaction scheme⁸⁴, Pyl may form an adduct with TMA, converting TMA into a quaternary amine intermediate that is activated for methyl transfer.

Pyl is, however, not restricted to the active site of methyltransferases and was also identified and characterized in the tRNA-editing enzyme tRNA^His guanylyltransferase (Thg1) from M. acetivorans⁹³, in which Pyl is only required to read through an in-frame UAG codon in the Thg1 mRNA to generate a full-length protein; thus, Pyl serves as a ‘normal’ amino acid without a catalytic role. Furthermore, a recent proteomic investigation confirmed UAG translation or Pyl insertion in eight additional proteins in M. acetivorans⁹⁴. These proteins included methylcorrinoid:coenzyme M (CoM) methyltransferase (MtaA), His kinase, tRNA endonuclease, hypothetical proteins and methylornithine synthase (PylB), which is the enzyme that catalyses the initial step of Pyl biosynthesis. PylB is folded⁹⁵ and active without Pyl⁸⁸, but the 21 amino acid extension following Pyl readthrough may modulate PylB activity or participate in sensing or regulating Pyl synthesis.

Translation of UAG codons in M. acetivorans may involve competition between release factor and PyltRNA^Pyl readthrough, and it is possible that some UAG codons are translated more efficiently as Pyl than others. In a M. acetivorans tRNA^Pyl deletion mutant, three instances of UAG codons serving as stop codons were detected by mass spectrometry, with none detected in the wild-type strain⁹⁴. The fact that UAG codes for stop in the tRNA^Pyl deletion mutant indicates that the M. acetivorans release factor has activity towards the UAG codons. Indeed, an in vitro study using chimeric Methanosarcina barkeri–human release factor in an in vitro translation system found that the M. barkeri release factor had activity towards all three stop codons (UAG, UAA and UGA), but significantly reduced activity for UAG⁹⁶. Although no natural M. acetivorans transcripts have been shown to use UAG as a stop codon in wild-type cells, previous work using an E. coli β-glucuronidase gene reporter with a UAG codon showed translation with Pyl (20%) and stopping at UAG (80%) in wild-type M. acetivorans⁹⁷. Competition with release factor is not unique to Pyl. Trp decoding at UGG is well known to compete with release factor^62,98, and release factor is also active at SECIS-recoded UGA codons that specify Sec^62,99. Despite the fact that competition exists between translational readthrough and termination, Trp and Sec codons are not considered to be stop codons. The above studies indicate that UAG codons are not recoded to Pyl at specific loci, as is the case for UGA codons recoded to Sec (see above). Future studies will provide a more detailed picture of Pyl decoding, but the preponderance of the evidence indicates that in native M. acetivorans transcripts UAG is reassigned to Pyl.

Pyl recoding has physiological impacts on both archaea and bacteria. For example, in genetic studies, a M. acetivorans strain lacking tRNA^Pyl was unable to grow on TMA^94,100. This observation reaffirmed the notion that Pyl is evolutionarily connected to methylamine metabolism¹⁰¹. The fact that all Pyl-decoding organisms encode at least one methylamine methyltransferase with an in-frame UAG codon (placing Pyl in the active site of these enzymes)⁸⁴ suggests that the selective value of Pyl may be related to its role in methylamine metabolism. However, all organisms that putatively encode Pyl also have UAG codons in mRNAs that direct production of other proteins. Protein families encoded in the M. acetivorans genome and enriched with Pyl residues include transposases, recombinases, methylamine and methylcorrinoid:CoM methyltransferases, methyltransferases (with unknown substrates), radical S-adenosylmethionine (SAM) enzymes and His kinases involved in two-component signalling⁹⁴.

In M. acetivorans, there are 267 in-frame UAG codons in annotated genes, and Pyl is intricately linked with the proteome, cellular metabolism and fitness. Growth of a tRNA^Pyl deletion strain in minimal medium containing methanol as the sole carbon source resulted in significant increases in generation and lag time compared with the wild-type strain⁹⁴. This phenotypic defect corresponded to ~350 differentially abundant (mostly non-Pyl-containing) proteins. In the deletion mutant, the abundance of several enzymes required for methanogenesis from methanol were reduced 20-fold, indicating lower metabolic efficiency with methanol. Methyltransferases involved in methanogenesis from dimethylsulfides¹⁰² increased 100-fold, suggesting an altered or compensatory metabolism⁹⁴. Although anabolic or other defects in the deletion mutant cannot be ruled out, the data show Pyl has a far-reaching impact on the composition of the proteome and suggest that M. acetivorans is metabolically less efficient when the organism has a reduced genetic code.

The Pyl-decoding trait exists in 25 anaerobic bacteria from the Firmicutes phylum (in classes Clostridia and Negativicutes) and the δ-Proteobacteria class. Early work in the field showed that E. coli expressed Pyl-containing proteins with a recombinant Pyl-decoding system, indicating that bacteria are capable of decoding UAG codons as Pyl^90,103. However, there has been little work to show whether Pyl is actively decoded in bacteria in which the Pyl trait naturally occurs.

An investigation into several of these bacteria uncovered a novel genetic code variation referred to as dynamic genetic code expansion¹⁰⁴. Several bacterial strains thought to genetically encode Pyl and at least one TMA methyltransferase (MttB) with one in-frame UAG codon were examined for the ability to produce Pyl-tRNA^Pyl and Pyl-containing MttB. The Pyl-decoding bacterium Acetohalobium arabaticum genetically encodes 20 amino acids when grown on pyruvate; however, when grown on TMA, the cells dynamically expand their genetic code to 21 amino acids with Pyl¹⁰⁴. Only in the presence of TMA did A. arabaticum synthesize Pyl-tRNA^Pyl, produce Pyl-containing MttB and consume TMA. In the absence of TMA, A. arabaticum transcriptionally silences the gene encoding PylRS and downregulates pylB, pylC and pylD transcription. The mechanism by which A. arabaticum senses TMA and initiates the expression of the Pyl-decoding system remains unknown. Interestingly, despite encoding a functional PylRS^105,106 and tRNA^Pyl pair¹⁰⁷, Desulfitobacterium hafniense did not show detectable production of Pyl-tRNA^Pyl or expression of MttB. Desulfitobacterium dehalogenans expressed transcripts for tRNA^Pyl and PylRS, but no detectable transcription of the Pyl biosynthesis genes (pylB, pylC and pylD) was observed. This suggests that Pyl is either not decoded at all or simply not decoded under the conditions tested. Collectively, these data show two different ways that codon usage can be adapted to genetic code expansion with Pyl. Pyl-decoding archaea severely limit their use of UAG codons (~5% of all annotated genes ‘stop’ at a UAG codon)^87,104, whereas many other organisms, including all Pyl-decoding bacteria¹⁰⁴, use UAG codons abundantly (~25% of all annotated genes ‘stop’ at a UAG codon). Furthermore, the recently sequenced Thermoplasmatales-related methanogens display an extremely reduced use of UAG codons, as low as 1.6%⁸⁷. The Pyl machinery is thought to be the evolutionary driving force that maintains low UAG codon usage in these species⁸⁷. The archaeal strategy seems to be to limit UAG codons to locations where Pyl is required or at least tolerated. By contrast, the bacteria examined so far seem to be able to silence Pyl when Pyl-containing proteins are not needed, which may allow these species to use higher levels of UAG codons.

Conclusion and outlook

Recent developments in genome sequencing, biochemistry, bioinformatics and structural biology have facilitated the discovery of new genetic codes and provided crucial tools to analyse and gain insight into the physiological impact of genetic code flexibility. In this Review, we categorized the mechanisms of genetic code variation into four types — biased codon usage, codon reassignment, ambiguous decoding and natural genetic code expansion (FIG. 1b) — but this is certainly an incomplete list of the genetic code diversity in the living world. Increasing evidence suggests that genetic code flexibility is a selected trait during evolution to benefit microorganisms under certain conditions. However, little is known about the molecular mechanisms that lead to the improved fitness as a result of genetic code variation.

Our improved understanding of natural genetic code variations has also fostered a new exciting research area in expanding the genetic code with non-canonical amino acids and engineering synthetic organisms with new genetic codes (BOX 1). Single-cell model microorganisms with advanced genetic systems, such as E. coli and S. cerevisiae, serve as major vehicles for such engineering efforts. However, even in these well-studied organisms, we are only beginning to understand the physiological changes brought about by engineered genetic code variations.

For each genetic code variant described above, far more is known about how the genetic code changed (in terms of evolutionary or biochemical mechanisms) than about why the genetic code changed or the potential selection pressures that drive and maintain these changes. Therefore, these are areas where there is still great potential for future experimentation. Molecular genetic approaches have already revealed much about genetic code variants in terms of effects on growth, proteome status and cellular metabolism. Furthermore, engineered synthetic organisms with very different genetic codes may lay the basis for systematic experiments to explore the selective value and physiological impact of genetic code evolution.

Our knowledge of decoding mechanisms is far from complete. These gaps in knowledge have presented major obstacles to understanding the evolution of the genetic code and engineering synthetic cells with expanded codes. We envision that with advances in single-cell genome sequencing and systems biology, further genetic code variations will be uncovered. This will provide new insights into genetic code flexibility and novel tools for engineering organisms with altered genetic codes.

Glossary

Aminoacyl-tRNA

(aa-tRNA). A tRNA molecule with an amino acid attached to the 3′ end. It is used as a substrate by the ribosome to synthesize proteins.

Codon–anticodon pairing

During translation, the bases of the mRNA codon and the tRNA anticodon need to match each other. Watson–Crick pairing (A–U and G–C) in the first and second positions of the codon is required for efficient decoding, whereas the third position allows more flexible pairing, for example, between G and U or using modified bases.

Synonymous codons

Different triplet nucleotide sequences that decode the same amino acid.

tRNA isoacceptors

Different tRNA species recognized by the same aminoacyl-tRNA synthetase and ligated with the same amino acid.

Misacylation

Incorrect pairing of an amino acid and tRNA by an aminoacyl-tRNA synthetase. Errors resulting from misacylation, if left uncorrected, reduce the overall translational fidelity.

Frameshifting

Change in the reading frame during translation due to mutations in the DNA, errors during transcription or translation or specific mRNA structures, leading to new protein sequences.

Shine–Dalgarno-like sequences

mRNA sequences that share high similarity with the Shine–Dalgarno sequence, which pairs with the anti-Shine–Dalgarno sequence of the ribosomal RNA.

Codon adaptation index

A method for analysing usage bias of synonymous codons using a set of highly expressed genes from a species as a reference to assign a score to each gene.

Wobble position

The third position of a codon, which is more flexibly recognized by the tRNA compared with other positions.

Phages

(Also called bacteriophages). Viruses that infect and propagate within bacteria. Phages contain their own genome but hijack the translational machinery of the bacterial host for protein synthesis.

RpoB

The β-subunit of the bacterial RNA polymerase and target of the antibiotic rifampicin.

Nucleophilicity

The property to donate an electron in chemical reactions.

Elongation factor Tu

(EF-Tu). A bacterial elongation factor that delivers aminoacyl-tRNAs to the ribosome during peptide synthesis. The counterpart of EF-Tu in archaea and eukaryotes is EF-1A