The central role of tRNA in genetic code expansion

Abstract

Background

The development of orthogonal translation systems (OTSs) for genetic code expansion (GCE) has allowed for the incorporation of a diverse array of non-canonical amino acids (ncAA) into proteins. Transfer RNA, the central molecule in the translation of the genetic message into proteins, plays a significant role in the efficiency of ncAA incorporation.

Scope of Review

Here we review the biochemical basis of OTSs for genetic code expansion. We focus on the role of tRNA and discuss strategies used to engineer tRNA for the improvement of ncAA incorporation into proteins.

Major Conclusions

The engineering of orthogonal tRNAs for GCE has significantly improved the incorporation of ncAAs. However, there are numerous unintended consequences of orthogonal tRNA engineering that cannot be predicted ab initio.

General Significance

Genetic code expansion has allowed for the incorporation of a great diversity of ncAAs and novel chemistries into proteins, making significant contributions to our understanding of biological molecules and interactions.

Graphical abstract

1. Introduction

Expansion of the genetic code has resulted in the incorporation of >160 non-canonical amino acids (ncAAs) into proteins and allowed for the creation of organisms encoding more than the canonical repertoire of 22 amino acids. The engineering of aminoacyl-tRNA synthetases (aaRSs) and their corresponding transfer RNAs (tRNAs) to produce orthogonal translation systems (OTSs) has long been the focus of genetic code expansion (GCE) studies. However, the efficiency of ncAA incorporation can be further improved through the engineering of the elongation factor, ribosome, and release factors, resulting in an optimized system including nearly all steps in translation. To produce an ideal OTS, orthogonal aaRS (o-aaRS) must recognize the orthogonal tRNA (o-tRNA), but none of the host tRNAs, while the o-tRNA must not be recognized by any of the host's aaRSs. It is also necessary to have a codon in the messenger RNA (mRNA) to direct incorporation of the desired ncAA. The number of developed OTSs has rapidly expanded to include a large number of orthogonal aaRS•tRNA pairs [1-3]. While most GCE studies have focused on the suppression of the three stop codons (UAA, UAG, UGA), an increasing number of studies have targeted sense codons for reassignment [4, 5].

The ability and efficiency to incorporate ncAAs depend on multiple steps of translation. The o-tRNA plays a central role in each step of translation, interacting with numerous nucleic acid and protein macromolecules during translation (Fig. 1). Thus, engineering of tRNA and its interaction with components of the translation apparatus play a major role in the optimization of GCE. Here we aim to provide a broad overview of the many aspects of tRNA in GCE. We highlight the many interactions of tRNA with other components of the translation machinery and how these interactions have been engineered to improve GCE applications.

Figure 1.

Orthogonal tRNAs (cyan) participate in all stages of translation. Spheres indicate nucleotides critical in the process. Terminal nucleotides are depicted as dark blue spheres (important in 5' and 3' end processing (1)), positions 34 and 37 depicted in orange (often modified in all 3 domains of life (2)), the anticodon is highlighted in lemon green (important in cognate aaRS recognition (3)), T-stem nucleotides depicted as red spheres are important in EF-Tu binding (4). Deacylated tRNA is subjected to another round of aminoacylation and elongation on the ribosome (5). Orthogonal aaRS is shown in pink, EF-Tu in green, and the ribosome in tan.

2. The cycle of tRNAs in translation

tRNAs serve as the adaptor molecules that carry the amino acids to the mRNA template, linking information encoded in the genome to protein synthesis. As the key molecules in deciphering the genetic code, tRNAs are highly regulated, processed, and modified (Fig. 1). Transcription of tRNA genes results in precursor tRNAs (pre-tRNAs) containing 5′ leader and 3′ trailer extensions, which must be precisely cleaved to yield tRNAs of appropriate length [6]. In eukaryotes, archaea and some Gram-positive bacteria, the tRNA genes generally do not encode the universally conserved 3′ CCA sequence, which instead is added post-transcriptionally by the CCA-adding enzyme after cleavage of the 3′ trailer. These immature tRNA molecules are then subjected to multiple site-specific modifications that are fundamental in tRNA recognition and decoding [7].

Once tRNAs are fully processed, AAs are attached to the 2′- or 3′-OH group of the terminal adenine residue (A76) by tRNA-specific aaRSs [8]. AaRSs provide one of the first quality control check points in translation by preferentially selecting cognate pairs of tRNAs and AAs while discriminating against near- and non-cognate substrates [9]. The specific recognition and selection of tRNA by the cognate aaRS occurs through the use of sequence and structural elements in the tRNA [7]. To ensure that mischarged aa-tRNAs are not released to the elongation factor, a number of aaRSs inherently proofread the two steps of the aminoacylation reaction by catalyzing the hydrolysis of either misactivated aminoacyl-adenylates (aa-AMPs, pre-transfer editing) or mischarged aa-tRNAs (post-transfer editing) [10]. After aminoacylation and editing, aa-tRNAs form a ternary complex with an activated, GTP-bound form of elongation factor (EF-Tu in bacteria and EF-1α in eukaryotes), which subsequently delivers the aa-tRNA to the ribosome. Both the AA and tRNA make variable thermodynamic contributions to EF-Tu binding, which results in correctly paired aa-tRNAs binding to EF-Tu with similar affinity, while mischarged aa-tRNAs vary in affinity over several hundred fold [11]. Once delivered to A-site of the ribosome, pairing of the mRNA codon and tRNA anticodon occurs in the decoding center of the small ribosomal subunit and, if correct, results in release of the aa-tRNA by EF-Tu. The aa-tRNA then moves into the A-site of the large ribosomal subunit for peptide bond formation [12]. During subsequent rounds of elongation, the tRNA moves into the P-site, transfer of the nascent peptide to the new A-site aa-tRNA occurs, and the uncharged tRNA then moves to the E-site of the ribosome and is released for subsequent aminoacylation and decoding events. At the end of the mRNA coding region, the translating ribosome encounters a stop codon, where release factors compete with near cognate aa-tRNAs for interactions with the stop codon. Upon binding to the ribosome and interactions with the stop codon, release factors terminate translation and release of the peptide chain occurs.

3. Expression and processing of orthogonal tRNA

As cells commit significant energy and resources to the production of tRNAs, their expression and processing are highly regulated [13]. While there are differences between how these events are carried out and regulated in bacteria and eukaryotes, the need for an ample supply of mature, processed tRNA is essential for GCE in all organisms.

3.1 Bacteria

In bacteria, tRNA genes are typically present within transcription units either containing tRNA genes only, or together with rRNA or protein-coding genes [14]. This organization allows several tRNAs to be transcribed from a single promoter and processed from a single, large precursor [15]. E. coli tRNA genes are encoded with their 3′ CCA sequence [14]. To ensure adequate processing, most o-tRNAs are introduced into E. coli with their 3′ CCA sequence, regardless of their original gene sequence. Expression of o-tRNA genes in E. coli is often carried out using an lpp promoter, although strong E. coli tRNA promoters (such as proK [16]) or their variants are also used [17].

Both natural and orthogonal tRNA containing polycistronic transcripts are endonucleolytically processed immediately after transcription ([18] and references within). Because bacterial RNase P recognizes some general tRNA features (acceptor stem, T-arm and the 3′ CCA end) [19], it is assumed that o-tRNAs are processed correctly. Although in some cases expressing o-tRNAs from several gene copies increases the ncAA-containing protein yield [20], endoribonucleases RNase E and RNase III may be required to cleave long precursors into smaller pieces prior to RNase P processing [18]. Thus, an increased in transcribed tRNA in the cell and the number of properly processed molecules may not correlate accordingly [21]. Both RNase E and RNase III have broad specificities and process the 3′ ends of the precursors leaving a few extra bases downstream the CCA end [18]. Final, accurate trimming of the 3′ CCA end is a complex process involving several exonucleases [22].

3.2 Eukaryotes

While bacterial tRNA genes are typically present in clusters containing multiple genes, eukaryotic tRNAs are often encoded as individual transcription units transcribed by RNA polymerase III (Pol III). The promoter structure of Pol III transcriptional units varies, with tRNAs being expressed from promoter elements located within the tRNA sequence itself (A and B boxes) [23] or through upstream promoters elements where the tRNA sequence may contain partial internal promoters [24]. Transcriptional termination of tRNA genes is mainly mediated by a stretch of T residues in the noncoding DNA strand, generating a pre-tRNA with a 3′ oligo U tract of variable length. The poly U sequence is bound by the La protein, which protects the 3′ end of the tRNA from exonucleases [6]. Processing of the pre-tRNA proceeds with RNase P catalyzing the cleavage of the 5′ leader sequence from the pre-tRNA with processing of the 3′ trailing sequence occurring by RNase Z cleavage thereafter [25]. If present, introns are then spliced out by the splicing endonuclease and the resulting tRNA halves ligated together ([26] and references within). Finally, the 3′ CCA trinucleotide is added by the CCA adding enzyme [27].

The use of promoters within the tRNA sequence in eukaryotes limits the ability to utilize tRNAs derived from bacteria and archaeal species, as most of them do not contain recognizable A and B box promoter sequences. To overcome this challenge, GCE studies in eukaryotes have taken several approaches for expression of bacterial and archaeal tRNAs. In an early GCE experiment, expression of E. coli tRNA^Gln only required the mutation of a single base (C9A) to generate internal promoters for expression in mammalian cells [28]. This single nucleotide change did not alter the tRNA structure and did not affect aminoacylation. This sharply contrasts with E. coli tRNA^Tyr, where the nucleotide substitutions needed to insert internal promoters impact the overall tRNA structure. Two different approaches were taken to overcome this problem; the direct transfection of aminoacylated E. coli tRNA^Tyr_CUA [29] and the use of Bacillus stearothermophilus tRNA^Tyr_CUA, which naturally contains an internal promoter sequence [30]. Improvements in heterologous expression of bacteria and archaeal tRNAs in higher eukaryotes came with the use of upstream promoters from the upstream Pol III promoters of H1RNA (RNA component of RNase P ribonuclease), U6 snRNA (RNA component of U6 snRNP), and CeN74-1 (stem bulge RNA loci CeN74-1) [31-35].

While expression of E. coli tRNA^Tyr_CUA has been shown to support UAG read-through in the single cell eukaryote Saccharomyces cerevisiae, even with incomplete internal promoters [36, 37], o-tRNA expression and UAG suppression was improved by placing Pol III promoter sequences upstream of the o-tRNA gene. This has been done using the SUP4 [38, 39], the SNR51, and PRP1 promoters [40]. Interestingly, although using the SUP4 promoter increased E. coli tRNA^Tyr_CUA expression by approximately 100 fold relative to the SNR51 and RPR1 promoters, UAG read-through was ∼10 fold lower [40]. Therefore, it is possible that while the SUP4 promoter is able to efficiently induce transcription of tRNA^Tyr_CUA, large amounts of this tRNA may overwhelm or impair the proper processing of the pre-tRNA^Tyr_CUA. In lower eukaryotes, another effective strategy for o-tRNA expression is the use of natural dimeric tRNA transcription units [32, 41, 42]. In these systems, the o-tRNA is used to replace the downstream tRNA gene of a dimeric transcription unit, with the native A and B box promoters of the upstream tRNA gene driving expression of the entire transcriptional unit including the downstream tRNA lacking a consensus A and B box.

While tRNA expression and modification (discussed in section 7) have been subjected to engineering for GCE applications, tRNA decay has not been explored or fully appreciated for its role in tRNA levels for GCE. These processes are unescapably connected and can have multiple effects on each other. For example, as suggested above, tRNA aminoacylation levels relate to the amount of accurately processed and modified tRNA. In S. cerevisiae, mature tRNAs can be degraded via the rapid tRNA decay (RTD) pathway if they are hypomodified (especially if they lack m^2,2G on position 26 and m⁵C) or have destabilizing mutations. Furthermore, EF-1α overexpression suppresses tRNA degradation indicating that EF-1α competes with the RTD pathway for tRNAs [43].

4. tRNA dictates orthogonality: recognition by aaRSs

Faithful translation of the genetic code requires aaRSs to attach amino acids to their cognate tRNAs. The tRNA specificity of aaRSs is dictated by a set of idiosyncratic features embedded in each tRNA, which are known as “identity elements” [44]. tRNA recognition elements extend across the tRNA L-shape structure, but are generally found in the AA acceptor stem and anticodon loop where N73 and anticodon bases N35 and N36 commonly act as identity elements [44]. For example, all aaRSs from E. coli, except for glutamyl-tRNA synthetase (GluRS) and threonyl-tRNA synthetase (ThrRS), rely on the identity of N73, whereas only alanyl- , histidyl-, seryl- and leucyl-tRNA synthetases (AlaRS, HisRS, SerRS, and LeuRS, respectively) do not use anticodon bases for specific recognition. Identity elements can also consist of post-transcriptional modifications as in E. coli tRNA^Ile, tRNA^Glu, tRNA^Lys, and S. cerevisiae tRNA^Phe, or structural motifs as is the case for E. coli tRNA^Cys and tRNA^Leu (reviewed in [44]). The most comprehensively understood identity element set is that of the tRNA^Ala, which encodes a wobble base pair in the acceptor stem (G3:U70) that is indispensable for aminoacylation by AlaRS as substitution with a Watson-Crick base pair A3:U70 or G3:C70 results in a decrease in aminoacylation efficiency of 147- and 468-fold, respectively [45-47]. While the tRNA^Ala G3:U70 element is conserved through evolution, most tRNA identity sets for a particular aaRS•tRNA pair are usually species-specific, and these variations in aaRS•tRNA recognition are the fundamental basis for the orthogonality of translational systems for GCE applications. For example, the Methanocaldococcus jannaschii TyrRS•tRNA^Tyr is widely used in E. coli due to its lack of cross-reactivity with the E. coli TyrRS•tRNA^Tyr pair as M. jannaschii tRNA^Tyr harbors a C1:G71 base pair at the top of the acceptor stem that is an identity element for M. jannaschii TyrRS but a negative element for E. coli TyrRS [48-50]. Similarly, because the C1:G72 base pair is also present in eukaryotic tRNA^Tyr, the E. coli TyrRS•tRNA^Tyr is orthogonal in eukaryotic cells and has been used as a vehicle for incorporation of various ncAAs in yeast and mammalian cells [36, 51]. In addition to TyrRS, the evolutionarily divergent identity elements of tRNA^Leu allowed the development of an E. coli LeuRS•tRNA^Leu pair for GCE in S. cerevisiae [52]. The basis for the orthogonality of the E. coli LeuRS•tRNA^His pair in S. cerevisiae relies on recognition of a core tertiary feature (a reverse Hoogsteen base pair U8:A14) and the lack of anticodon recognition by E. coli LeuRS, preventing aminoacylation of S. cerevisiae tRNA^Leu [53, 54]. Although S. cerevisiae LeuRS can aminoacylate wild-type E. coli tRNA^Leu (because it contains S. cerevisiae tRNA^Leu identity elements A35 and G37), the anticodon mutation that convert E. coli tRNA^Leu to an amber suppressor and the substitution of G37 with A prevents aminoacylation of E. coli tRNA^Leu by S. cerevisiae LeuRS. Due to the recognition of E. coli tRNA^Leu by S. cerevisiae LeuRS, the yeast LeuRS•tRNA^Leu_CUA pair is unfit for GCE experiments in E. coli [54].

The tryptophanyl-tRNA synthetase (TrpRS)•tRNA^Trp has also been employed for GCE in different organismal contexts. For instance, S. cerevisiae TrpRS•tRNA^Trp_CUA was introduced into E. coli, as the S. cerevisiae tRNA^Trp_CUA contains A73, G1:C72, and the C35U mutation that together prevent aminoacylation by E. coli TrpRS [55, 56]. Similar identity rules applied to the incorporation of Bacillus subtilis TrpRS•tRNA^Trp_UCA into human 293T cells [57]. More recently the suppression efficiency of the S. cerevisiae TrpRS•tRNA^Trp_CUA pair was improved by creating libraries of tRNA^Trp_CUA targeting the five first bases pairs of the acceptor stem [58], both acceptor and anticodon stems [59], and entire tRNA body [60].

The recognition of tRNA^Pyl by PylRS has become of special interest in the field of GCE. That tRNA^Pyl is a natural amber suppressor and that PylRS largely interacts with the G73, G1:C72, and unique core tertiary motifs of tRNA^Pyl but not with its anticodon [61, 62] has made the PylRS•tRNA^Pyl pair popular in GCE applications in both bacteria and eukaryotes [1, 63]. More importantly, the lack of tRNA anticodon recognition of PylRS has been exploited to not only incorporate ncAAs in response to stop codons but also in response to sense [64] and quadruplet codons [65].

Finally, tRNA identity elements have been at the center of the development of the GCE field, which has thrived on comprehensive studies that were mostly carried out over two decades ago for aaRS•tRNA pairs from model organisms. Therefore, for GCE to expand, discovery and investigation of new sets of aaRS•tRNA pairs with unique recognition elements are needed.

5. Alteration of aaRS editing activity for tRNA charging with ncAAs

As discussed in the previous section, the large number of nucleotides and interactions between the tRNA and aaRS allows for very accurate selection of tRNAs by their corresponding cognate aaRS, with a discrimination factor greater than 10^-6 [7]. In contrast, AAs do not display a sufficient range of functional groups to allow aaRSs to efficiently discriminate between those that are structurally similar, which can jeopardize the accuracy of protein synthesis. To prevent accumulation of mischarged aa-tRNAs, several aaRSs have evolved editing activities to hydrolyze mischarged aminoacyl adenylates (pre-transfer editing) or to hydrolyze mischarged aminoacyl-tRNA (post-transfer editing) [10]. While pre-transfer editing occurs in the aminoacylation active site (either by enzymatic hydrolysis or by selective release) [66], post-transfer editing activity can occur in cis through the use of a dedicated editing domain attached to the aaRS or in trans by freestanding deacylaces [12, 67, 68].

Only a few GCE studies have made use of aaRS editing activities to enhance ncAA incorporation. For example, in an effort to enhance formation of ncAA-tRNA, the Pyrococcus horikoshii phenyl-tRNA synthetase (PheRS) editing domain was inserted into an E. coli TyrRS variant (iodoTyrRS) developed for charging with 3-iodo-L-tyrosine (3-I-Tyr) [69]. While iodoTyrRS not only charges 3-I-Tyr to produce 3-I-Tyr-tRNA^Tyr, it also retains the ability to produce Tyr-tRNA^Tyr in the absence of supplied 3-I-Tyr [70]. To enhance production of the 3-I-Tyr-tRNA^Tyr and to clear Tyr-tRNA^Tyr, the editing domain of the P. horikoshii PheRS was inserted into iodoTyrRS, creating a variant (iodoTyrRS-ed) that was able to efficiently produced 3-iodo-L-tyrosine-tRNA^Tyr. A recent series of studies also employed the PheRS editing domain inserted into a TyrRS variant; however, in this case the goal was to change the enantioselectivity of TyrRS from L-Tyr-tRNA^Tyr to D-Tyr-tRNA^Tyr [71, 72], resulting in an approximately 2.6-fold shift in the enantioselectivity of the enzyme towards the formation of D-Tyr-tRNA^Tyr [71, 72].

6. Delivery of ncAA-tRNA by elongation factor

The activated form of EF-Tu (EF-Tu•GTP) binds all cellular aa-tRNAs with similar affinities. Relative affinities vary by about an order of magnitude and are negatively correlated with the relative abundances of individual aa-tRNAs [73]. To achieve similar affinities for structurally different substrates, EF-Tu relies on the combination of specific thermodynamic contributions of the AA (up to 2.5 kcal/mol) and cognate tRNA body (up to 3.6 kcal/mol) [74]. It is evident from the crystal structures of Thermus aquaticus EF-Tu in complex with S. cerevisiae Phe-tRNA^Phe (PDB ID: 1TTT) and E. coli Cys-tRNA^Cys (PDB ID: 1B23) that EF-Tu interacts with the acceptor-and T-stem of both tRNAs [75, 76]. A detailed study of acceptor- and T-stem single base pair mutants revealed that EF-Tu relies mostly on base pairs 49:65, 50:64, and 51:63 for specific recognition of the tRNA body and that thermodynamic contributions at the three sites are additive [77]. Acceptor-stem base pairs contribute much less to the sequence specificity, although a C1:A72 mismatch acts as a negative determinant in E. coli initiator tRNA^Met [78] and U1:A72 in tRNA^Gln was proposed as a determinant that defines its weak binding to EF-Tu [77].

As all ncAA-tRNA are effectively misacylated, it is apparent that inefficient binding by EF-Tu may cause poor incorporation of ncAAs into proteins. In E. coli, under normal growth conditions the concentrations of EF-Tu and endogenous aa-tRNAs are similar [79, 80], and the ncAA-tRNA is likely to be outcompeted for the EF-Tu binding. The easiest way to circumvent this may be by overexpressing EF-Tu as even (mutated) weaker-binding aa-tRNAs decode their cognate codons similarly to wild-type tRNAs at saturating EF-Tu concentrations [81]. Further, lowering the incubation temperature may aid the EF-Tu ncAA-tRNA binding [82] as the affinity of EF-Tu toward aa-tRNAs increases inversely with temperature [74]. Lastly, the affinity of EF-Tu for ncAA-tRNA may be increased by mutations, both in the o-tRNA and amino acid binding pocket of EF-Tu (see below).

In a number of GCE studies, acceptor- and T-stems of o-tRNAs were mutated to obtain optimal binding to EF-Tu. Although T-stem base pairs that lead to tighter (G51:C63, C50:G64 and G49:U65) or weaker (A51:U63 or U51:A63, G50:U64 or U50:G64, and G49:C65 or C49:G65) EF-Tu binding are known [77, 81], the exact combination of strong and weak sequences will depend on the ncAA employed. For example, mutations in a tRNA^Pyl variant, tRNA^Pyl_opt [83], enabled 5-fold better incorporation of N^ε-acetyl-lysine without influencing aminoacylation efficiency. Yet, the relevant T-stem sequence of tRNA^Pyl_opt harbors a U49:A65 (intermediate strength, replaced the weak C49:G65), G50:C64 (intermediate, replaced the intermediate C50:G64) and G51:C63 (naturally present, strong). Thus, the tightest binding of tRNA^Pyl to EF-Tu was not the most advantageous one. The frequently used M. jannaschii tRNA^Tyr [1] was also mutagenized in the same manner and different targeted ncAAs were incorporated by these mutants with varying success [84].

In in vitro translation systems, specific o-tRNA mutagenesis might not be needed as entire tRNA bodies can be swapped. An ncAA-tRNA can be (bio)chemically synthesized (typically by ligation to pdCpA-ncAA-NVOC [85] or by flexizyme [86]) and added directly to the reconstituted translation mixture. An early study showed that an E. coli tRNA^Asn-derived suppressor is the best candidate for inserting both proteinogenic valine and nonproteinogenic homoglutamate [87]. The same tRNA^Asn proved to be better than E. coli tRNA^Phe for the incorporation of N-methylated alanine, and facilitated insertion of the N-methylated phenylalanine equally well as E. coli tRNA^Phe [88]. Another tight binding tRNA, tRNA^AlaB, outperformed tRNA^Phe in the incorporation of allyl-glycine, methyl-serine and biotinyl-lysine by increasing the overall affinity to EF-Tu•GTP and, unexpectedly, incorporation of N-methyl-phenylalanine by increasing the rate of dipeptide formation by an unknown mechanism [89].

In the EF-Tu•GDPNP•Phe-tRNA^Phe structure (PDB ID: 1TTT) the phenyl ring of Phe is stacked on the side chain of His67. Amino acids Phe229, Asp227, Glu226, and Thr239 line the EF-Tu's pocket and create a space structurally and chemically suited for any of the 20 canonical AAs [75]. To form a pocket suitable for the very negatively charged ncAA, phosphoserine (Sep), EF-Sep (EF-Tu-His67Arg/Glu216Asn/Asp217Gly/Phe219Tyr/Thr229Ser/Asn274Trp) capable of Sep-tRNA^Sep binding was evolved [90]. In case of p-azidophenylalanine binding, a triple EF-Tu mutant, EFBM1 (EF-Tu-Ser66Ala/Asp217Ala/Val275Ala), was shown to improve ncAA incorporation [91]. EF-Tu residues Glu216 and Asp217 are sometimes mutated to Ala when bulky ncAAs are employed in GCE experiments. Glu216Ala or Asp217Ala EF-Tu mutants improve the incorporation of D,L-2-anthraquinonylalanine (anqAla), L-2-pyrenylalanine, (2pyrAla) and L-1-pyrenylalanine (1pyrAla) into proteins [92]. Moreover, a double mutant can rescue the EF-Tu's lack of affinity for misacylated N^ε-Bodipy576/589-lysyl-tRNA^Lys (BOP-Lys-tRNA^Lys) resulting in EF-Tu-Glu216Ala/Asp217Ala binding to BOP-Lys-tRNA^Lys being almost as stable as that of wild-type EF-Tu to native Lys-tRNA^Lys [93].

Although increasing the stability of EF-Tu•GTP•ncAA-tRNA complex may help ncAA incorporation, the binding affinity of the complex must be balanced to ensure efficient release of ncAA-tRNA in the ribosome. In the case of EF-Tu-Glu216Ala•BOP-Lys-tRNA^Lys, EF-Tu-Glu216Ala/Asp217Ala•BOP-Lys-tRNA^Lys [93], and probably EF-Tu-Glu216Ala/Asp 217Ala with pyrAla-tRNA^Phe and anqAla-tRNA^Phe [92], stabilization of the ternary EF-Tu•GTP•ncAA-tRNA complex was achieved by reducing the k_off, which then resulted in an ineffective ncAA-tRNA release into ribosomal A-site. Thus, an ideal way to stabilize the EF-Tu•GTP•ncAA-tRNA complex would be to “speed up” its formation (increasing k_on) rather than to slow down its dissociation.

As both the ncAA and the tRNA are responsible for EF-Tu overall affinity, ideally, the o-tRNA variants should be co-evolved along with EF-Tu's amino acid binding pocket. Taking into account only three critical T-stem base pairs and six residues in the EF-Tu's amino acid binding pocket the library size would still be within the technical limit. As with endogenous tRNAs, K_d of an co-evolved EF-Tu_mut•o-tRNA_mut pair [94] would depend inversely on the relative abundance of the ncAA-tRNAs in the host [73]. Because ncAA-tRNA levels depend on the strength of o-tRNA promoter, proper o-tRNA processing, and the efficiency of o-aaRS charging, it is possible that different optimal EF-Tu_mut and o-tRNA_mut may evolve in different experimental set-ups.

7. Ribosome decoding with ncAA-tRNA

Between the initial EF-Tu•GTP•ncAA-tRNA binding and ncAA incorporation there are several kinetically detectable steps each of which can determine the fate of ncAA incorporation (reviewed in [95]). During initial selection, the ternary complex is brought to the A-site by the ribosomal stalk. The C-terminal domain of the ribosomal stalk binds to a cleft between domains 1 and 3 of aEF1α in archaea [96] and EF-Tu in bacteria [97] but there is so far no evidence on how the (mis)aminoacylation status of a ncAA-tRNA might influence this interaction. Once brought to the A-site, still within the ternary complex, the tRNA anticodon interacts with the codon in the 30S subunit. If a correct set of codon-anticodon interactions occurs, a conformational change takes place in EF-Tu and stimulates the hydrolysis of the bound GTP. A GDP-bound conformation then leads to structural rearrangement of EF-Tu, which loses its affinity for the ribosome, and dissociates from it, freeing the ncAA-tRNA. The ncAA-tRNA can then move into the A site of the 50S subunit (the process of accommodation). Once accommodated, ncAA-tRNA participates in the peptidyl transferase reaction.

During initial association of ternary complex (EF-Tu•GTP•aa-tRNA) with the ribosome, codon–anticodon interactions are formed between aa-tRNA and mRNA. Correct Watson-Crick base pairing is established through the interactions with the ribosome at positions 1 and 2 of the codon (bases 36 and 35 of the anticodon) while a set of unorthodox interactions may occur at the third position, including wobble decoding [98] and other unexpected base pairings [99, 100]. For all codons to be read by canonical wobble decoding a minimal set of 32 tRNAs is needed. However, some bacteria and organelles have smaller tRNA sets and tentatively employ the “2 out of 3” rule (interactions in case of G:C pairs on the first two positions are sufficiently strong [101]), the “superwobble” (an unmodified U34 reads any nucleotide in the third position of an mRNA codon [102]) or have the position 34 post-transcriptionally modified to allow nonstandard pairing and/or fitting the opposing bases into a codon-anticodon mini-helix (reviewed in detail in [103]).

Aberrant decoding rules can be adapted for codon emancipation in a model organism, such as E. coli, for GCE. For instance, unmodified Mycoplasma mobile tRNA^Ile_UAU, able to decode AUA codons, can rescue the lethal tilS deletion in E. coli. tRNA^Ile-lysidine synthetase (encoded by tilS) modifies tRNA^Ile2_CAU by adding a lysyl moiety onto C34, which allows this E. coli tRNA to decode the rare AUA codon. Thus, by introducing an aberrant decoder (here M. mobile tRNA^Ile_UAU), direct AUA decoding is possible, and if accompanied by a fully orthogonal isoleucyl-tRNA synthetase (IleRS), AUA codons may be reassigned to a ncAA [104]. In nature, reassignment of Ile AUA codons to Met exists in metazoan and S. cerevisiae mitochondria where a 5-formyl-cytidine (f⁵C34) containing Met-tRNA^Met reads both AUG and AUA [105, 106], as well as in ascidian mitochondria where the reassignment is mediated by a 5-taurine-2-thio-methyl-2-thiouridine on position 34 (Tm⁵s²U34) and an unmodified U37 in Met-tRNA^Met [107].

Very little data is available on the o-tRNA heterologous modification status. It is well known that positions 34 and 37 are (hyper)modified in all three domains of life, yet the identity of chemical groups attached differs and so do the corresponding modification enzymes. Presumably even suppressor tRNA anticodons can be modified, which may, in turn, modulate their efficiency. In cytosolic tRNA^Tyr of S. cerevisiae, Drosophila melanogaster and Nicotiana rustica ([108] and references within) pseudouridine (Ψ) 35 is critical for successful UAG decoding. Conversely, presence of queuosine (Q) on position 34 effectively destroys the suppressor capacity of this tRNA [109]. In E. coli, a theoretical spectrum of anticodon modifications of an amber suppressor would include 2′-O-methylcytidine (Cm), 2-lysidine (k²C) or N⁴-acetylcytidine (ac⁴C) at position C34 [103], while U35 and A36 should remain unmodified (MODOMICS database [110]). While Cm34 and ac⁴C normally read G3 in the codon, k²C is, as mentioned, responsible for A3 recognition in the Ile AUA codons. Thus, theoretically, Cm and ac⁴C-containing tRNA_CUA should be able to suppress amber stop codons, while a k²C-containing one should not.

The wobble base has a prominent role in decoding but also acts as an identity determinant for many aaRSs [44]. In cases where the AA activation depends on tRNA binding, information on the wobble base recognition is transmitted to the distant aaRS active site [111]. In case of the M. jannaschii Tyr-OTS, the G34C mutation, needed to create a suppressor o-tRNA derivative, increases affinity for cognate tyrosine 30 times [112]. For M. maripaludis SepRS, cognate tRNA improves discrimination over glutamate and phosphothreonine in vitro [113]. The role of wobble base in successful ncAA incorporation is further accentuated when their corresponding modifications serve as both aaRS determinant and promote correct codon-anticodon pairing (GluRS, IleRS, lysyl-tRNA synthetase (LysRS), see [114]). However, a sense codon reassignment in E. coli with M. jannaschii Tyr-OTS revealed an o-tRNA variant (o-tRNA^Tyr_AUG) capable of Tyr incorporation in response to histidine CAU and CAC codons with similar efficiencies [115]. Analysis of in vivo expressed o-tRNA^Tyr_AUG revealed modification of A34 into inosine, generating o-tRNA^Tyr_IUG which is then capable for the decoding of CAC codons [116]. Although E. coli has only one copy of tRNA^His (tRNA^His_GUG containing queuosine at position 34) capable of decoding both CAC and CAU codons, unexpected modification of o-tRNA^Tyr renders it capable of dual decoding in an alternative manner. The modification is presumably executed by TadA, a tRNA-specific adenosine deaminase responsible for generating tRNA^Arg2_ICG [117].

Most of the popular o-tRNAs used for GCE are suppressor derivatives of the original tRNA or natural suppressors such as tRNA^Pyl [1]. As such, they inherit their original N37 identities, which may influence the decoding properties of the o-tRNA in the host organism. The major role of an invariant purine 37 is to increase stacking interactions over the first base pair of the mRNA-tRNA mini-helix and promote stacking to the 3′-adjacent residues on positions 38 and 39, thereby ensuring correct codon-anticodon recognition ([103] and references within). Apart from eukaryotic and archaeal tRNA^Phe [118] G37 modification includes only a methyl group. However, bacterial and archaeal methyl transferases work with different tRNA identity determinants [119], thus, archaeal derived o-tRNAs might not be suitable substrates for E. coli enzymes. For instance, an archaeal o-tRNA adapted to genetic encoding of phosphoserine (tRNA^Sep, an M. jannaschii tRNA^Cys derivative [90]) performs better in E. coli once changed to contain a G37A mutation [21, 120]. Because unmodified G37 has been identified as a determinant for cognate O-phosphoseryl-tRNA synthetase reaction in vitro [121] and the effect is further strengthened by the G37 methylation [122] the improved G37A variant performance appears to reflect this o-tRNA's properties as a decoder. Interestingly, the identity and modifications of base 37 appear to depend on the identity of the first base pair of the codon-anticodon minihelix [103]. The presence of cyclic N⁶-threonylcarbamoyladenosine or 2-methylthio-N⁶-isopentenyladenosine (stabilizing modifications) in E. coli tRNAs (MODOMICS database [110]) matches A1:U36 or U1:A36 pairing [103]. The same pair is by definition present in all of the stop codon suppressors so the apparent conservation of A37 between various successful suppressor o-tRNAs might not be coincidental [21, 83, 120, 123, 124].

The anticodon stems normally contain fewer modifications, mostly simple pseudouridine or methylated bases to ensure a more rigid structure [103]. This rule is violated in several in vitro evolved tRNAs. In case of quadruplet decoding tRNA^Pyl variants [65], the base pairing in the anticodon stem is almost entirely destroyed. In some of the evolved o-tRNA^Sep variants [120] the canonical Watson-Crick's base pairs are changed to wobble (29:41 and 30:40) thereby suggesting an alternative conformation of tRNA^Sep anticodon-stem loop (ASL). On the other hand, ASL variants selected from S. cerevisiae tRNA^Trp [60] and M. jannaschii tRNA^Tyr [123] strictly abide the canonical packing rules needed to establish a helical structure of the ASL.

Comparison between the ASL modifications conserved in Euryarchaeaota (often used Pyl-OTS originating from Methanosarcina species and Tyr-OTS from M. jannaschii) and Gram-negative bacteria [118] reveals that, at least location wise, some of the equivalent ASL modifications occur in E. coli. Both groups modify tRNAs at positions 32, 34, 37, 38 and 39, while bacteria might also employ modifications of position 40. In contrast, cytosolic tRNAs of eukaryotic single cell organisms, Fungi and Metazoa also modify U and G on positions 27, 28, 30 and 31, as well as Us and Cs on position 40. More importantly, anticodon uridines 35 and 36 might be transformed into Ψ or contain a yet unknown modification (U36) [118].

8. Competition between o-tRNAs and release factors

Translation termination occurs when the A-site of the translating ribosome encounters one of the three stop codons. Upon encountering a stop codon, the release factor binds to the ribosomal A-site and promotes hydrolysis of a P-site bound peptidyl-tRNA [125]. While eukaryotic organisms utilize a single release factor (eRF1) to recognize all three stop codons, bacteria use two release factors that partial overlap in their substrate range: release factor 1 (RF1) recognizes UAA and UAG while release factor 2 (RF2) recognizes UAA and UGA [126]. The use of a stop codon for GCE results in competition between stop codon suppression by ncAA-tRNAs and stop codon recognition by the endogenous release factors, often leading to premature termination of the growing peptide chain without incorporation of the ncAA. As RF1 is the only release factor in bacteria that recognizes the UAG stop codon, deletion of the RF1 gene in E. coli has been used as a means to reduce ncAA-tRNA_CUA competition for the UAG codon [127, 128]. Interestingly, while RF1 deletion does lead to higher UAG codon suppression and an increase in ncAA incorporation, it also results in an increase in competition between the ncAA-tRNA and near-cognate suppression by endogenous aa-tRNAs, which increases off target AA incorporation [129, 130].

In addition to premature chain termination, near-cognate suppression is one of the main culprits of poor ncAA insertion and/or recombinant product yield. In E. coli, the notoriously high near cognate suppression of UGA stop codons is facilitated by tRNA^Trp. The effect is further exacerbated by the presence of a G24A mutation in this tRNA (Hirsh's mutation, [131]). A recent crystallographic study revealed that the mechanism by which this D-stem mutation influences the UGA suppression relies on internal base pairing of tRNA^Trp rather than with the ribosome itself. The mutation enables base pairing between bases 24 and 44 which allows mutant tRNA^Trp to adopt the A/T conformation more easily than the wild-type tRNA^Trp thereby facilitating the reading of the stop codon [132]. This example highlights the active role of o-tRNAs in the process of decoding/stop codon suppression. In S. cerevisiae, a recent analysis showed that near-cognate tRNAs could mispair at position 1 or 3 of nonsense codons and that during read-through the same endogenous AAs are inserted [133]. Gln, Tyr, and Lys were inserted at UAA and UAG codons, while Trp, Arg, and Cys were incorporated at UGA [133]. This analysis further illustrates how inflating read-through events by chemical means does not induce alternative mispairing but only enhances endogenous mistranslation [133].

9. Conclusions and outlook

O-tRNAs play a vital role in successful ncAA incorporation by GCE, directly influencing (i) efficiency of o-aaRS binding, (ii) cognate o-aaRS's affinity for the ncAA, (iii) thermodynamic and kinetic behavior of the EF-Tu•GTP•ncAA-tRNA, (iv) efficiency of decoding targeted codons.

As illustrated by the M. jannaschii Tyr-OTS [112], the anticodon recognition impact on the active site's affinity for the ncAA might be underappreciated. Ideally, o-aaRSs should be mutagenized both in ncAA binding pocket and anticodon binding domain simultaneously. As library sizes needed for rational mutagenesis of both o-aaRS elements are beyond current technical limit and iterative rounds of both rational or random mutagenesis are labor intensive [134], continuous and rapid techniques such as phage-assisted continuous evolution (PACE) [135] might prove invaluable in the future.

Highlights.

• Transfer RNA is the central molecule in the translation of nucleic acid into protein.

• Engineering of the interaction between orthogonal tRNAs (o-tRNAs) and cellular translation machinery optimizes genetic code expansion (GCE).

• There are many unexpected outcomes of o-tRNA engineering.

• Full orthogonality of tRNAs devoted to GCE is vital for cellular fitness.