The diverse structural modes of tRNA binding and recognition

Abstract

tRNAs are short noncoding RNAs responsible for decoding mRNA codon triplets, delivering correct amino acids to the ribosome, and mediating polypeptide chain formation. Due to their key roles during translation, tRNAs have a highly conserved shape and large sets of tRNAs are present in all living organisms. Regardless of sequence variability, all tRNAs fold into a relatively rigid three-dimensional L-shaped structure. The conserved tertiary organization of canonical tRNA arises through the formation of two orthogonal helices, consisting of the acceptor and anticodon domains. Both elements fold independently to stabilize the overall structure of tRNAs through intramolecular interactions between the D- and T-arm. During tRNA maturation, different modifying enzymes posttranscriptionally attach chemical groups to specific nucleotides, which not only affect translation elongation rates but also restrict local folding processes and confer local flexibility when required. The characteristic structural features of tRNAs are also employed by various maturation factors and modification enzymes to assure the selection, recognition, and positioning of specific sites within the substrate tRNAs. The cellular functional repertoire of tRNAs continues to extend well beyond their role in translation, partly, due to the expanding pool of tRNA-derived fragments. Here, we aim to summarize the most recent developments in the field to understand how three-dimensional structure affects the canonical and noncanonical functions of tRNA.

The need of an adapter molecule that decodes genetic information into function was predicted by Francis Crick in 1958 (1). In eukaryotes, tRNAs are transcribed by RNA Polymerase III and undergo several steps of maturation before being charged with their cognate amino acids (2, 3). Approximately 1000 individual and unique tRNA sequences have been identified by different sequencing methods in over 400,000 analyzed genomes (4). The human genome contains over 500 genes that encode functional tRNAs, which are able to decode 62 codons, including selenocysteine (4) and pyrrolysine (5). As the redundant pool of tRNAs is produced from several genomic loci, the number of clinically relevant mutations in cytoplasmic tRNA genes is relatively low (6). Nonetheless, mutations in mitochondrial tRNA genes and altered expression levels of individual tRNAs are associated with various human diseases (7, 8). Numerous reports of patient-derived mutations in genes coding for tRNA maturation, processing, and modification enzymes directly highlight the importance of accurate tRNA biogenesis for human health and disease (9, 10, 11, 12, 13, 14). Recently, tRNAs have been rediscovered as potential therapeutics to readjust protein synthesis in specific diseases (15). This idea has sparked interest from various research groups and highlights the importance of revisiting the structure of these small RNAs.

The general structure and shape of tRNAs

tRNA is a biopolymer composed of 76 to 90 RNA nucleotides (nt) linked together via 3ʹ-5ʹ phosphodiester bonds (16). Each nucleotide consists of a phosphate group attached via an ester bond to a five-carbon sugar (ribose), which is linked to one of the four nitrogenous bases, namely the purines (adenine and guanine) or the pyrimidines (uracil and cytosine). Specific hydrogen bonding networks between the bases define the characteristic clover-leaf secondary structure of tRNA, which was already predicted from the first sequence of yeast tRNA^Ala_AGC determined in 1965 by Holley et al. (17). The canonical secondary structure of tRNAs is divided into the following domains (Fig. 1A): (i) acceptor stem (also known as amino acid accepting stem (AAS) or acceptor arm) linked via (ii) the 2-nt “connector” to the (iii) D-stem loop (D-arm) closed by the D-loop, (iv) anticodon stem loop (ASL, anticodon arm), (v) variable region (variable loop, variable arm), and (vi) T-stem loop (T-arm or TΨC-arm) (18). The 3ʹ terminus includes a single-stranded CCA sequence (19), which constitutes the site of covalent amino acid attachment during aminoacylation (20). Nucleotides are numbered from 5ʹ to 3ʹ, and the posttranscriptionally added nucleotides are indicated by the number of the preceding nucleotide followed by a colon and a letter in alphabetical order. From extensive sequencing data, we know that U₈, A₁₄, G₁₈, G₁₉, A₂₁, U₃₃, G₅₃, T₅₄, Ψ₅₅, C₅₆, A₅₈, C₆₁, C₇₄, C₇₅, and A₇₆ are the most conserved nucleotides within tRNAs. Y₁₁, R₁₅, R₂₄, Y₃₂, R₃₇, Y₄₈, R₅₇, and Y₆₀ (R = purine; Y = pyrimidine) are semiconserved nucleotides found in many tRNA sequences. Differences in tRNA length are mainly a consequence of the varying sequence length in the variable loop that can accommodate “extra” nucleotides, while preserving the overall domain organization and three-dimensional shape (21). Some exceptions to the clover-leaf secondary structure are found in mammalian mitochondrial tRNAs (22, 23), flagellates (24), and nematodes (25, 26). In these rare cases, deviations occur in the D-stem/loop or T-stem/loop architecture or the complete D-loop is lacking (18, 27), which will be described in the following sections.

Figure 1.

The structure of tRNA at different levels and during translation.A, left: specific regions are indicated on yeast tRNA^Phe_GAA (PDB ID 1EHZ) as follows: 3ʹ and 5ʹ end; acceptor arm (light green); T-arm (yellow); D-arm (light blue); variable loop (gray); anticodon arm (pink); and 34, 35, 36 anticodon nt. The dimensions of folded tRNAs are approximately 60 × 70 × 25 Å. Middle: clover-leaf secondary structure representation of tRNA numbered according to the standard convention. Right: tertiary nucleotide interactions responsible for the L-shape tRNA structure. B, multiple structures of tRNA extracted from the context of elongating bacterial 70S ribosome determined by time-resolved cryo-EM. The schematic demonstrates local flexibility of the A-site (magenta), P-site (dark green), and E-site (purple) tRNA associated with the cognate mRNA (red) bound to the 70S ribosome [large 50S (blue) and small 30S (orange)]. The arrows highlight the bending of the 3ʹ end (silver) from the acceptor arm (light green) and the swinging of the anticodon arm (pink). The major sequential steps of the elongation cycle associated with the L-shape distortion are indicated with Roman numerals (I–IV). The two superimpositions on the left (A-site phenylalanine- and P-site methionine-tRNA) follow the steps and changes after cognate aminoacyl-tRNA delivery by elongation factor EF-TU (dark blue). The two superimpositions on the right depict AP proline- and PE methionine-tRNA translocation facilitated by elongation factor EF-G (cyan), which resets the system for the next elongation cycle. EF-Tu, Elongator factor-Tu.

The secondary structure of tRNAs not only depends on the RNA nucleotide sequence itself but also on the presence of other additional factors. For instance, coordination of magnesium ions is required for correct folding of tRNA^Glu_UUC in Escherichia coli (28). Despite sequence variations, all tRNAs fold into the characteristic L-shaped three-dimensional tertiary structure (29) (Fig. 1A). This compact shape comprises two orthogonal helical branches, each containing several Watson-Crick base pairs (bp). In detail, the AAS is stacked over the T-stem loop to form the acceptor domain that is linked by the so-called “elbow” region (30) to the anticodon domain formed by ASL and D-stem loop. The L-shape is further stabilized by additional non-Watson-Crick interactions between D- and T-arms. This uniform L-shaped conformation was initially proposed in 1974 based on the first crystal structure of yeast tRNA^Phe_GAA (31, 32, 33, 34) and then confirmed by other structural studies of yeast tRNA^Asp_GUC (35, 36), bacterial (37) and yeast (38) initiator tRNA^Met_CAU, and human tRNA^Lys_UUU, the primer of HIV reverse-transcriptase (39). The correct folding, stability, and functionality of tRNAs are also influenced by the posttranscriptional incorporation of chemical modifications (40).

Individual regions of tRNAs

In general, the size of acceptor (AAS), anticodon, and T-arms are highly conserved among functional tRNAs. The AAS is composed of a stem that is formed between the first 7 nt (5ʹ end) and nt 66 to 72 (3ʹ end), creating a molecular zipper (Fig. 1A). It is completed by a 4-nucleotide long, single-stranded overhang, harboring the so-called discriminator base at position 73 (41) followed by the CCA-tail. The exceptions to this general principle are all tRNA^His_GUG, which contain a posttranscriptionally added guanine nucleotide at the 5ʹ end, numbered “0” or “−1” (42). The discriminator base is one of the major identity elements that, in combination with several other nucleotides in the anticodon and the acceptor helix, ensures the very high specificity of aminoacyl-tRNA synthetases (43, 44). During the maturation process, 5ʹ leader sequences are excised by the endonuclease RNase P (45) prior to the 3ʹ trailer, although this order can be reversed for tRNAs harboring long 3ʹ trailer sequences (46). In the case of the 3ʹ end, the CCA sequence can be encoded in the tRNA gene (e.g., E. coli and related bacteria, certain Gram positive bacteria) or added to the tRNA during a separate maturation process in eukaryotes, certain Gram positive bacteria, and archaea (47). In the first scenario, 3ʹ trimming is primarily carried out by the exoribonucleases RNase T and RNase PH (48), but other exoribonucleases (e.g., RNase II and RNase D or RNase BN (49)) can act as a substitute, if needed. When the CCA is not encoded in the genomic sequence, the 3ʹ endonuclease RNase Z initially cleaves the overhang after the discriminator base and leaves a 3ʹ-OH group (50); subsequently the tRNA nucleotidyltransferase adds the 3′ terminal CCA residues without the need of a template (51). The fundamental role of the CCA sequence is to allow aminoacylation and therefore provide a charged tRNA to the ribosome. Moreover, inside the ribosome, the CCA sequence enables dynamic interactions with ribosomal proteins and rRNA elements as the tRNA is moved through the ribosome (52, 53) (Fig. 1B).

The 5ʹ end of the acceptor stem is connected to the D-arm by two nt (U₈ and N₉), which represents the second most variable region of tRNAs (Fig. 1A). The D-arm derives its name from the dihydrouridine (D) modification present at the 3ʹ and/or 5ʹ of conserved G₁₈ and G₁₉. Typically, the D-arm comprises a 4-bp stem (positions 10–13 pair with 22–25) that ends in a 7–11 nucleotide loop containing conserved nt A₁₄, A₂₁, G₁₈, and G₁₉. The 3ʹ part of the acceptor arm is directly connected to the T-arm, which is typically a 5-bp long stem (positions 61–65 pair with 49–53) with a loop composed of seven nt (positions 54–60). Due to the presence of a universally conserved motif [thymidine 54, pseudouridine (Ψ) 55, and cytosine 56], it is also called the “TΨC-arm”. C₅₆ and Ψ₅₅ form interactions with conserved residues G₁₉ and G₁₈ in the D-arm, which stabilizes the L-shaped form of tRNA (54). The D-arm is also responsible for interactions with the ribosome during all stages of translation (55).

The variable region, as its name indicates, can vary in length (56). In type I tRNAs, it usually consists of four or five nt, whereas in type II tRNAs (e.g., tRNA^Ser, tRNA^Leu, tRNA^Tyr), it consists of 10 to 24 nt. The longer variable regions are arranged in a double helical stem of 3 to 7 bp and a loop of 4 to 5 nt. Nucleotides in the variable region are located between position 45 and 46 and obey canonical base-pairing rules. The nt in the 5ʹ-strand and the 3ʹ-strand are numbered e11, e12, e13,... and e21, e22, e23,..., respectively. The prefix ‘e’ is specific for the variable region, and the second digit identifies the respective base-pair. In the case of a long variable region, the loop can be formed by up to 5 nt: e1, e2, e3, e4, and e5 (57).

The anticodon arm consists of a 5-bp stem (positions 27–31 pair with 39–43) and a 7-nt loop (positions 32–38), containing U₃₃, R₃₇, and the anticodon triplet N₃₄, N₃₅, N₃₆. The anticodon, together with a discriminator base, is one of the main identity elements of tRNAs (58). It is responsible for reading codon triplets in mRNAs and 61 different sense codons encoding 20 standard amino acids. The 21st amino acid, selenocysteine, is recoded by the UGA stop codon if assisted by a specific selenocysteine insertion sequence elements in the mRNA and is directly synthetized on already charged tRNA^Ser_UCA (59, 60). The 22nd amino acid, pyrrolysine, was discovered in Methanosarcina barkeri and is encoded by an UAG codon in the methylamine methyltransferase genes (61). The number of tRNA species in different organisms varies between 23 and 45, clustered in 20 groups of so-called iso-acceptors. These are tRNAs that are charged with the same amino acid despite having different anticodons. Apart from methionine and tryptophan, all amino acids have multiple iso-acceptors. The anticodon does not even have to be fully complementary to recognize a specific codon triplet. The nature of the wobble base at position 34 enables the accommodation of noncanonical interactions between the codon and anticodon. Furthermore, the ASL is the most heavily modified region of tRNA, with these modifications either restricting or facilitating the decoding process (62).

The overall L-shape of tRNA is maintained even though conformational changes occur during the different steps of the ribosomal translation cycle (Fig. 1B). Advances in time-resolved cryo-EM (63) have allowed the structural visualization of distinct states of the ribosome and tRNAs during translation (64, 65). The characteristic L-shape of tRNAs is further substantiated by numerous examples of structural “mimicry”, where other RNAs or proteins adopt a tRNA-like shape (66). For instance, viruses use the tRNA-shaped internal ribosome entry site in order to recruit ribosomes to initiate translation at non-AUG codons (67, 68). Furthermore, Staphylococcus aureus uses a riboswitch (yjdF) that was found to mimic the overall shape of a tRNA to activate translation (69). The translational elongation and ribosome recycling factors (e.g., EF-G, EF2, or RRF) also use structural mimicry to specifically interact with the ribosome (70, 71, 72, 73).

RNA modifications and their influence on the folding/stability of tRNAs

tRNAs fold into an L-shaped three-dimensional structure despite their high sequence variation (33, 74). At first glance, one would simply assume that the available sequences provide sufficient constraints for all tRNA transcripts to adopt the L-shaped conformation. This notion is also supported by very similar crystal structures of modified and unmodified tRNA^Phe_GAA from bacteria and eukaryotes (Fig. 2A) (74, 75, 76, 77). Several studies indeed have demonstrated that various “naked” in vitro–transcribed tRNAs are able to fold into functional L-shaped tRNAs that can be charged by aminoacyl-tRNA synthetases or used in in vitro translation systems (78, 79, 80, 81). However, other unmodified tRNA transcripts are found to dynamically switch between multiple conformations, and the equilibrium between misfolded and properly folded forms is regulated by additional factors, such as modifications or environmental conditions (82). For instance, unmodified human mitochondrial tRNA^Lys_UUU displays as an extended bulged-hairpin structure instead of the canonical cloverleaf structure. This finding is consistent with the observation that some hypomodified tRNAs are not functional in vivo and subject to degradation (83, 84, 85, 86, 87, 88, 89). Approximately 100 different RNA modifications are found in tRNAs, and on average ∼17% of nt are modified in each tRNA molecule (90, 91, 92). The occurrence of tRNA modifications can be divided into two groups, based on their functionality within the molecule: (i) modifications that are involved in the process of translational decoding and (ii) those that control tRNA folding and stability (91, 93, 94).

Figure 2.

Structure of unmodified and modified tRNA^Phe.A, comparison of prokaryotic and eukaryotic tRNA^Phe_GAA. Superimpositions of in vitro transcribed (transparent) and endogenous (opaque) tRNA^Phe_GAA from Escherichia coli (Ec) and Saccharomyces cerevisiae (Sc) obtained by crystallography. These tRNAs are depicted as cartoons using the same color code for each region as in Figure 1. Modifications that are important for the tRNA core structure are represented in sticks and colored according to their position in the tRNA. The PDB IDs and resolution limits of each tRNA are as follows: unmodified EctRNA (PDB ID 3L0U at 3.0 Å) and SctRNA (PDB ID 5AXM at 2.2 Å) and modified EctRNA (PDB ID 6Y3G at 3.1 Å) and SctRNA (PDB ID 1EHZ at 1.9 Å). The m⁷G₄₇ modification in EctRNA (PDB ID 6Y3G) is present in tRNA^Phe_GAA but is not visible in the electron density map. B, chemical structures of modified nucleosides that are shown in (A). The modifications are indicated in red and the ribose moiety is annotated with R. Base colors correspond to the modified target positions in (A), while the color (green or yellow) of the small circle indicates the presence of modification in E. coli or S. cerevisiae. RMSD values for E. coli tRNA^Phe and S. cerevisiae tRNA^Phe were calculated with ChimeraX using the backbone atoms of nt 3 to 74. E. coli: overall RMSD 2.669 Å², acceptor stem 4.426 Å², D-arm 1.595 Å², anticodon-loop 2.716 Å², variable loop 2.362 Å², T-arm 1.349 Å²; S. cerevisiae: overall RMSD 1.662 Å², acceptor stem 2.132 Å², D-arm 2.194 Å², anticodon-loop 1.628 Å², variable loop 0.801 Å², T-arm 0.665 Å².

In light of excellent reviews focusing on tRNA modifications (27, 40, 62, 95, 96, 97, 98, 99), we focus here on those modifications that seem important for the core structure of tRNA^Phe_GAA in prokaryotes and eukaryotes. In general, modifications that affect the tertiary structure of tRNA are rather simple, such as reduction or isomerization of uridine to D or Ψ, methylation of the base and/or ribose sugar, replacement of oxygen on the base with sulfur, or the addition of small functional groups, like aminocarboxypropylation (acp) (Fig. 2B) (91, 93, 100). These modifications modulate the structural plasticity of the tRNA core, where some regions become rigid and others are more flexible (75, 87, 101, 102, 103). These changes are achieved through increased hydrophobicity, change of charge, and stabilization of a 2ʹ-endo or 3ʹ-endo sugar conformation (87, 103, 104, 105, 106, 107, 108). Depending on the position on the nucleotide, modifications can alter hydrogen bonding interactions on all three edges (Watson-Crick, Hoogsteen, and sugar-edge), which in turn can change the base-pairing properties, leading to local or global structural rearrangements of the tRNA (85, 107, 109, 110). The introduction of small chemical groups typically maximizes the stabilizing effect while minimizing the entropic penalty of attaching larger chemical groups, which are more common in the anticodon region.

One of the most prevalent tRNA modifications is methylation, which can occur on the oxygen of the 2ʹ-OH moiety (2ʹ-OMe), on nearly all nitrogen sites of the nucleobase (with the exception of the N7 and N3 position of adenosine) and on the C2/C5 atom of pyrimidines and adenosine (104, 111, 112). The presence of a 2ʹ-OMe group induces the 3ʹ-endo ribose conformation, through repulsion between the 2ʹ-OMe and 3ʹ-phosphate, resulting in the stabilization of an A-form helical region (104, 106, 112). This effect leads to an increased melting temperature of tRNA and decreased susceptibility to alkaline and enzymatic hydrolysis (109, 113, 114). Methylations on the Watson-Crick edge appear predominantly on the 5ʹ-side of the tRNA and prevent canonical base pairing (40, 104). The m¹A modification at position 9 and 58 is common and has been linked to correct folding and structural stability of tRNA (85, 115). A prime example for global structural rearrangement is the well-studied m¹A₉ modification in human mitochondrial mt-RNA^Lys_UUU. It was shown that mt-RNA^Lys_UUU exists in an equilibrium between the classical L-shaped and an alternative extended hairpin conformation. Addition of the methyl group to N1 of adenosine disrupts A₉-U₆₄ base pairing, which supports the hairpin conformation and shifts the equilibrium towards the correctly folded structure through A₅₀-U₆₄ pairing (85, 115, 116). A similar effect was also described for mt-RNA^Asp_GUC and mt-RNA^Leu_UUR, demonstrating the importance of modifications for mitochondrial tRNA folding (117, 118, 119, 120, 121). Methylation of N1 at guanosine 9 (m¹G9) in cytoplasmic and mitochondrial tRNAs also disrupts canonical base pairing and is believed to be involved in structural reorganization (122, 123). The N1-methylation of A₅₈ does not disrupt the reverse-Hoogsteen base pairing with T₅₄ but introduces a positive charge that stabilizes the base pairing and thus the tertiary structure (124, 125, 126, 127, 128).

Single or double methylation of N2 in guanosine (m²G, m²₂ G) occurs frequently in the junctions between acceptor stem and D-arm (position 9 and 10) as well as between the D-arm and anticodon loop (position 26 and 27) in some eukaryotic and some bacterial tRNAs (129, 130, 131, 132, 133, 134). m²G can also be found at position 6 and 7 in the acceptor stem of higher eukaryotes (133, 135). The single methyl group on N2 of G₁₀ does not interfere with the triple base pairing between G₁₀-C₂₅-G₄₅, but the hydrophobic character of the methyl group might stabilize the D-loop (132, 136). Dimethylation of G₂₆ prevents base pairing with cytidine but still allows pairing with any other nucleotide (132, 137). Similar to m¹A₉, it was shown that m²₂G26 prevents an alternative folding of the human cytosolic tRNA^Asn_GUU (137). Interestingly, m²₂G₂₆ prevents mispairing with C₁₁ and controls pairing with A₄₄, which contributes to the stabilization of the hinge region between the D-arm and anticodon loop (132, 134, 137, 138).

The most prominent methylation on the 3ʹ-half of tRNAs is m⁵U₅₄/T or ribothymidine, and it occurs in the T-loop of nearly all tRNAs. m⁵U₅₄ together with m¹A₅₈ forms a reverse-Hoogsteen bp, which stabilizes the T-arm through additional base stacking with the conserved G₅₃-C₆₁ pair (125, 127, 139, 140). m⁵C modifications between the variable loop and T-arm most likely contribute to stabilizing the tRNA core structure. In a subset of eukaryotic tRNAs, C₄₈ in the Levitt bp (G₁₅•C₄₈) is methylated to m⁵C, which increases the hydrophobic character of the bp and has been suggested to promote base stacking (40, 128, 141). Another conserved modification in the variable loop is m⁷G₄₆, which introduces a positive charge that facilitates a triple base pairing with C₁₃ and G₂₂ and stabilizes the tertiary structure of tRNA (124, 142).

Two highly abundant uridine modifications are Ψ and D, which engender opposite structural properties. Ψ is formed through isomerization of the glycosidic bond between N1-C1 into the C5-C1 carbon-carbon bond (143, 144, 145). This repositioning alters the chemical properties of the base, resulting in a free N1 atom at the Hoogsteen edge that increases base-stacking properties of Ψ with its neighboring bases through the coordination of a water molecule (143, 144, 145, 146, 147). Furthermore, Ψ favors the 3ʹ-endo sugar conformation that reinforces the local structural rigidity and stability of RNA helices (143, 144). Ψ is present at several positions in tRNA and is commonly found in position 55 in the T-loop, where it pairs with G₁₈ and stabilizes the tRNA elbow region (30, 90, 148).

In contrast to Ψ and 2ʹ-OMe that confer rigidity, D promotes the more flexible 2ʹ-endo ribose conformation (103, 106, 143). This structural behavior originates from the nonplanar and nonaromatic nature of the D modification, where the double bond between C5-C6 of uridine is reduced to a single bond, shifting the atoms C5 and C6 on opposite sides of the N1, C2, N3, and C4 plane (149, 150). As D is no longer able to participate in stacking interactions with the neighboring aromatic residues, the modification increases the local flexibility of the D-loop. Moreover, D induces a 2ʹ-endo conformation in the adjacent 5ʹ-residue (74, 103, 151). The local flexibility of the D-loop may facilitate the interloop interaction of the conserved G₁₈ and G₁₉ residues with Ψ₅₅ and C₅₆, which are responsible for maintaining the structure of the elbow region (30, 103, 152). Because the lack of dihydrouridine in E. coli tRNA^Ser_GGA is associated with a lower melting temperature, the flexibility of the D-loop must somehow contribute to tRNA stability (110). The NMR-based analysis of an RNA hairpin derived from Schizosaccharomyces pombe tRNA_i^Met_CAU confirmed that the presence of D can indeed promote the formation of a stable RNA hairpin (152).

There are two additional uridine-based modifications that show a positive contribution to the structural stability of tRNAs. Thiolation of C4 at uridine 8 and 9 (s⁴U) in the acceptor stem is conserved in bacteria and archaea but has not been reported in eukaryotes (92, 110, 153, 154). s⁴U₈ is buried in the tRNA core and pairs with A₁₄ of the D-loop. It has been suggested that s⁴U is involved in tRNA stabilization, as lack of the modification decreased the melting temperature of E. coli tRNA^Ser_GGA by 5 °C compared to the fully modified tRNA (110). The 3-(3-amino-3-carboxypropyl) uridine (acp³U) modification prevents Watson-Crick base pairing and is located in the core region of tRNA (90, 151, 155, 156). acp³U was initially detected in E. coli tRNA^Phe_GAA and later on also found in several eukaryotic tRNAs (157, 158, 159). In E. coli, acp³U is located at position 47 in the variable loop (151, 160, 161). It was shown that the presence of acp³U₄₇ in tRNAs of E. coli increases the melting temperature by 3 °C (160). However, in eukaryotes, acp³U occurs in the D-loop at position 20 and 20a of several cytosolic tRNAs and at position 47 in plastid and mitochondrial-encoded tRNAs of various land plants (151, 160). Interestingly, tRNA^Lys_UUU from Trypanosoma brucei has acp³U at position 20 and a dihydrouridine derivative at position 47 (acp³D). Despite its prevalence, the exact mechanistic and functional consequences of this acp³D modification are not completely understood. It has been proposed that the presence of acp³U in the variable loop might destabilize alternative tRNA conformations, as the acp³ group blocks Watson-Crick pairing similar to m¹A⁹ (160). Furthermore, it was suggested that acp³U_20a might stabilize the local structure through Mg²⁺ coordination and that acp³U₄₇ could coordinate water or Mg²⁺ with the T-arm in a similar fashion to stabilize the local conformation (151, 160).

The various tRNA modifications can either happen as individual, independent events or as part of a complex modification circuitry that relies on a hierarchy of modifications (97, 98). These interdependencies between modification pathways are well described for the ASL, but recent work revealed the presence of a similar modification network also in the T-loop and tRNA core region (97, 98, 162, 163). In summary, the specific incorporation of tRNA modifications adds an additional layer of complexity that ensures the proper folding and structural stability of tRNAs. Foremost, the use of chemical modifications increases the number of RNA sequences that are able to fold into a functional tRNA and therefore the appearance of modification enzymes throughout evolution is directly correlated with the variation of tRNA sequences (164).

Recognition of the tRNA structure by modification enzymes and maturation factors

Although the shape of mature tRNAs must remain relatively rigid and stable to promote efficient translation of the ribosome, the structural characterization of precursor tRNAs that still need to undergo maturation suggests a very different scenario. In particular, analyses of tRNAs in complex with the proteins responsible for processing and modification provide insights into the dynamic nature of the tRNA structure itself, which could not be achieved by studying mature tRNAs alone or within the ribosome.

For several decades, X-ray crystallography and NMR have been the dominant methods for determining the structures of proteins bound to tRNAs. Numerous high-resolution structures of aminoacyl-tRNA-synthetases (type I aa-RS and type II aa-RS) in complex with their tRNA substrates have been reported and characteristic modes of tRNA recognition unveiled (Fig. 3A). Since 2010, eight crystal structures of tRNA-modifying enzymes bound to full-length tRNA have been reported, showing that these proteins use multiple approaches to recognize and bind substrate tRNAs. They can be either very specific to the target region or sense the global shape of tRNAs (Fig. 3A). Thermus thermophilus tRNA dihydrouridine synthase (TthDus) recognizes the elbow region of target tRNA and flips out the nucleotide at position 20 to present it to the catalytic active site (165). tRNA (cytosine (72)-C(5))-methyltransferase (NSUN6), responsible for 5-methylcytosine modification in C₇₂ of the AAS, recognizes the specific consensus sequence on the acceptor arm (166) but also senses the overall shape of target tRNAs by additional contacts. Similarly, the heterodimeric tRNA m¹A₅₈ methyltransferase (m¹A₅₈ MTase) buries the whole tRNA and breaks the interaction between D- and T-arms to gain access and to methylate the N1 position of A₅₈ (167). Although a tetrameric form of m¹A₅₈ MTase has been reported, it remains unclear if this represents a physiologically relevant oligomeric state of the complex.

Figure 3.

tRNA modification enzymes bind and recognize tRNAs using diverse strategies.A, gallery of available structures of tRNA modifying/processing enzymes solved with full-length tRNA (published since 2010). Precise modification sites are marked on the clover-leaf map of tRNA (left upper corner) and in the 3D structures (magenta). Elongator (PDB ID 8ASW), ADAT2/3 (PDB ID 8AW3), METTL1-WDR4 (PDB ID 8CTH), m¹A₅₈ MTase (PDB ID 5CD1), mitochondrial seryl-tRNA synthetase (PDB ID 7U2B), Dus (PDB ID 3B0V), Ribonuclease P (PDB ID 6AHU), TiaS (PDB ID 3AMT), Trm5a (PDB ID 5WT3), NSUN6 (PDB ID 5WWT), TSEN complex (PDB ID 7UXA/7ZRZ), RlmN (PDB ID 5HR6), QTRT1/2 (PDB ID 7NQ4), Cgi121 subunit of KEOPS (PDB ID 7KJT). All presented structures are cytoplasmic enzymes except for mitochondrial seryl-tRNA synthetase, which is marked by an orange asterisk. DusC from Escherichia coli (PDB ID 4YCP, 4YCO) is not shown as it is highly similar to TthDus (PDB ID 3B0V). The class I and class II aminoacyl tRNA synthetases are also included: arginyl-tRNA synthetase (ArgRS, PDB ID 5YYN) and glycyl-tRNA synthetase (GlyRS, PDB ID 7YSE). B, summary of the number of structures solved by different structural biology techniques together with full-length tRNA (2010–2022). KEOPS, Kinase, endopeptidase and other protein of small size complex.

Positions 34 and 37 on tRNAs, the modification hotspots in ASL, are modified with different moieties to guarantee a stringent decoding process. Despite these positions being in the same location on each tRNA, the mechanisms of binding and recognition for these positions differ dramatically. Such contextual recognition is essential for discrimination of nontarget from target tRNAs. For instance, tRNA^Ile-agm²C synthetase (TiaS) binds to the acceptor arm and the ASL and catalyzes the modification on C₃₄. Contacts on the ASL stem are key substrate-specificity determinants of TiaS for tRNA^Ile (168). tRNA methyltransferase (Trm5a) in Pyrococcus abyssi undergoes a conformational change to bind tRNAs via their elbow as well as ASL and executes the bifunctional methyl-transfer at position G₃₇ (169). The bacterial RNA methyltransferase (RlmN) modifies tRNA at A₃₇. It binds to the concave surface of tRNA, like most tRNA-modifying enzymes, suggesting that recognition of the entire tRNA molecule is key for this modification machinery. Its tRNA-binding mode resembles those of aminoacyl-tRNA-synthetases, which contact the 3′-end and D-arm of tRNA substrates (170). The “Kinase, endopeptidase and other protein of small size complex” (KEOPS) contacts the 3′-CCA tail of tRNA via its Cgi121 subunit, which is essential for t⁶A₃₇ modification (171). Details of how other subunits of the KEOPS complex orchestrate tRNA recognition is under investigation, but an initial tRNA–KEOPS complex model provides some insights into the potential reaction mechanism (172).

Because these tRNA-modifying enzymes act transiently and are highly dynamic, the chances of obtaining well-diffracting crystals are low thus scuttling structure determination. To overcome this problem, structural biologists have trimmed flexible regions of the proteins, crosslinked RNA to the complexes, or used point mutations to trap specific reaction intermediates. For instance, a crystal structure of heterodimeric human queuine tRNA-ribosyltransferase catalytic subunit 1 and 2 (QTRT1/2) crosslinked to a RNA stem-loop revealed details of substrate recognition and binding (173). These necessary compromises do raise concerns as to whether the obtained structures reflect the native architecture of the complexes. Foremost, the required quantities of homogenous tRNAs for macromolecular crystallography and NMR have complicated the determination of complexes with native tRNAs, reduced the variety of tRNA substrates used, and limited the number of structures of proteins in complex with full-length tRNA. For instance, out of the available repertoire of enzymes that modify tRNAs, the structures of only approximately 30% have been determined in complex with their tRNA target.

Recent developments in single-particle cryo-EM have boosted the field (Fig. 3B), allowing the structural characterization of enzymes in complex with tRNAs in their most natural conditions and requiring greatly reduced sample quantities. Moreover, advanced data analysis procedures allow delineation of different conformations and reaction states within the same sample, a feature that has the potential to provide a dynamic and more complete picture of the process of tRNA recognition and the underlying modification reaction itself. For example, the cryo-EM structure of the heterodimeric tRNA-specific adenosine deaminase 2/3 (ADAT2/3) complex bound to tRNA revealed that the anticodon loop, which harbors the modified nucleotide, is embedded in the catalytic site (174). The elbow region of the tRNA is recognized in a sequence-independent manner by distant regions of the protein located far from the catalytic core (174). Cryo-EM structures of the complex of methyltransferase 1-WD repeat–containing protein 4 (METTL1-WDR4) with a bound tRNA revealed that the noncatalytic subunit WDR4 interacts with the T-arm and is essential for the activity of METTL1, which further interacts with the elbow and modifies position 46 (175, 176).

Most of the known tRNA modification/processing enzymes exist as monomers or dimers and are usually relatively small (∼100 kDa) (62). One major exception is the 850-kDa eukaryotic Elongator complex, which is composed of two copies of its six individual subunits (e.g., Elp1, Elp2, Elp3, Elp4, Elp5, and Elp6). The Elongator complex binds tRNA via two contact points (177). The Elp1 subunit contacts T- and D-arms (178), while the catalytic Elp3 subunit interacts with the ASL and carboxymethylates uridines in position 34 (179). Elongator recognizes the overall tRNA shape and can bind target as well as nontarget tRNAs with similar binding affinities in vitro (177). Another family of tRNA-processing enzymes that have recently been studied using single particle cryo-EM are well-known aminoacyl-tRNA synthetases. For instance, the cryo-EM structure of mitochondrial seryl-tRNA synthetase (mt-SerRS) ultimately shows how this protein specifically recognizes this rather unusual tRNA that lacks the D-arm. These enzymes recognize the radically altered T-arm topology, which leads to the loss of the canonical L-shape (180). Very recently, the structures of arginyl-tRNA synthetase (ArgRS) (181) and glycyl-tRNA synthetase (GlyRS) (182, 183) from different organisms have been determined in the presence of full-length tRNA.

Several cryo-EM studies of human RNase P bound to a mature tRNA reveal how a catalytic RNA aided by 10 protein subunits recognizes the acceptor-T stem stack for accurate pre-tRNA processing (184). Recent cryo-EM structures of human tRNA splicing endonuclease complex together with a pre-tRNA in precleavage and postcleavage states revealed how the multisubunit complex coordinates substrate recognition and the two-step cleavage reaction (185, 186, 187). Three subunits of tRNA splicing endonuclease interact with the T-arm, D-arm, and the anticodon stem, which acts as a ruler to correctly orient the 5′-splice site and 3′-splice site at the catalytic pocket.

tRNA modification and processing enzymes alter the structure of tRNA

The structural studies mentioned above also illustrate that the bound tRNAs undergo clearly observable conformational changes while binding to the enzymes and during the intermediate steps of the modification reactions. These conformational changes aim to properly recognize and position the tRNAs and expose the specific target site (or region) to the active site of the modifying enzyme. The movements can be observed in almost every region of tRNA, either as local movements proximal to the target nt or large scale global conformational changes (Fig. 4). For example, NSUN6 methylates C₇₂ in tRNA^Cys_GCA and tRNA^Thr_UGU and distorts the conformation of the most distal 3ʹ region (166). Likewise, changes in the elbow region of tRNA can be observed upon binding of tRNA-dihydrouridine synthase (165, 188).

Figure 4.

Modification enzymes influence tRNA structure in various regions. Yeast tRNA^Phe_GAA (PDB ID 1EHZ) is aligned on each structure by the matchmaker tool in ChimeraX and shown transparently. Structures are sorted by the region affected structurally, and the names of the modifying enzymes are colored, accordingly. Modification sites are marked in magenta. Structures of full length-tRNAs are shown without their respective protein partners - Dus (PDB ID 3B0V), m¹A₅₈ MTase (PDB ID 5CD1), Elongator (PDB ID 8ASW), ADAT2/3 (PDB ID 8AW3), TiaS (PDB ID 3AMT), Trm5a (PDB ID 5WT3), NSUN6 (PDB ID 5WWT). ArkI (PDB ID 7VNV) is a crystal structure of tRNA alone, modified by TkArkI.

In contrast, the binding of human m¹A₅₈ MTase to tRNA not only changes the canonical shape of the elbow region (167) but also alters the shape of the ASL, which is distant from the modification site. Similarly, distant structural changes of the tRNA elbow were also reported in the T. brucei ADAT2/3 deaminase complex (174) and the yeast Elongator complex (178). A more drastic tRNA conformation change is observed in a complex with archaeosine tRNA-guanine transglycosylase, which forms a “λ form” of bound tRNA to allow the target site to be accessible to the enzyme (102). The phosphorylation of tRNA in position 47 by Thermococcus kodakarensis Arkl also influences the shape of the elbow region by restricting local backbone rotation (94) (Fig. 4). In this specific case, the described structural changes could be the consequence of the modification itself, as the observed differences are visible in the crystal structure of the phosphorylated tRNA in the absence of the modification enzyme. As these structures were obtained after trapping early reaction intermediates, these structural rearrangements are directly induced by the contacts between the tRNA and the protein complexes.

In all the cases presented above, the target nucleotides of the modification reaction were specifically flipped out from the canonical tRNA structure, facilitating access to the enzymatic active site. Although the three bases of the anticodon are already flipped out and accessible, this phenomenon is also visible in tRNA structures in complex with enzymes modifying residue 34 within the ASL (168, 174, 189) (Fig. 4). In all three mentioned cases, the conformation of the ASL is still further distorted, which may be necessary to discriminate between the other accessible bases and different tRNAs with similar anticodon sequences. In summary, the distortion of the tRNAs seems to be a common feature of almost all tRNA modification enzymes. Therefore, the respective proteins and protein complexes have evolved features that go beyond providing the enzymatic requirements for the respective modification reaction. They also guarantee the proper exposure of the specific target site. On the one hand, these additional conditions may have impeded modifications in certain positions on the tRNA. On the other hand, they may have supported the development of modification hotspots in certain domains (e.g., ASL).

Unexpected appearance, function, and locations of tRNAs

In the previous sections, we summarized the specific three-dimensional features of tRNA molecules and described the fundamental determinants that influence their maturation and folding. In contrast to the conserved features, tRNAs sometimes do not obey their own rules and lack features that otherwise would appear crucial. Furthermore, some tRNAs are specifically cleaved and the obtained products can fulfill additional functions. Finally, as tRNAs are highly abundant, rigid building blocks present in all living cells, it is not surprising that they are also used in other cellular pathways and are even exploited by viruses and pathogens for their own benefit. In the following section, we summarize the most recently discovered and surprising examples of tRNAs, tRNA-like molecules, and tRNA-derived fragments.

Virus hijacking or mimicking host tRNAs

Viruses are known to hijack host molecules, such as rRNA and tRNAs, to enhance transcription, translation, and virion assembly (190). For instance, HIV type 1 retroviruses package host tRNAs during virion assembly and employ them as primers to initiate reverse transcription after infecting human cells. A recent study revealed a novel manner by which viruses recognize and utilize tRNAs from host cells. For instance, HIV type 1 virions are assembled by incorporation of multiple copies of the structural Gag protein, which selectively binds to several human tRNAs, including tRNA^Lys_CUU, tRNA^Lys_UUU, tRNA^Glu_CUC, tRNA^Glu_UUC, tRNA^Gly_GCC, tRNA^Gly_CCC, tRNA^Val_AAC, and tRNA^Val_CAC. The binding selectivity is achieved via the N-terminal matrix domain (MA) of Gag, which harbors four critical residues to track the characteristic tRNA elbow (191). The U-shape curvature and the invariant G₁₉-C₅₆ pairing in the T-loop also contributes to tRNA recognition and selection. Of note, the binding of tRNAs occludes the strong basic surface area of MA, which prevents the interaction of Gag with membranes and thus prevents premature virion assembly. In stark contrast to the highly regulatory role of host tRNAs in the HIV-1 life cycle, human cytomegalovirus is also known to associate with human host RNA but the details remained unclear. Recently, Liu et al. used a combination of cryo-EM and deep RNA-seq to confirm the presence of tRNAs bound via pp150, a tegument protein forming a net-like density layer between the nucleocapsid and virion envelope (192). They identified at least six different cytosolic tRNA species, with tRNA^Glu_CUC appearing to be the most abundant bound tRNA. Structurally, one tRNA molecule bridges three tegument pp150 proteins (Fig. 5A). The interactions are mediated via the phosphate backbone instead of specific base-stacking, explaining the low sequence-specificity. Surprisingly, this way of incorporating host tRNAs directly into the capsid seems to be unique for human cytomegalovirus because the structures of β-herpesviruses murine cytomegalovirus and human herpesvirus 6B capsids reveal no sign of tRNA association between their pp150 counterparts (192).

Figure 5.

Examples of tRNAs with unexpected functions and atypical appearances.A, a cartoon representation of a cryo-EM structure of HCMV virion (PDB ID 5VKU) with tRNA bound to pp150 nt (inlet, PDB ID 7LJ3). The asymmetrical unit in the virion is colored in blue and the selected tRNAs are highlighted in pink, purple, and green. B, a cartoon representation of a cryo-EM structure of a full-length of Bacillus subtilis glyQ T-box-tRNA^Gly_GCC complex (PDB ID 6POM). The tRNA is illustrated as spaghetti representation, while the binding partners are shown with surface representation. C, two-dimensional representation of various tRNA folds, including the canonical and truncated forms. HCMV, human cytomegalovirus.

The term “tRNA-like structure” (TLS) refers to RNA motifs, which can fold into a tRNA fold. This type of a tRNA-mimicking motif was first discovered in the genome of positive-strand RNA Turnip yellow mosaic virus, and additional TLSs were identified later in other viruses and even mammals (193, 194). Apart from a predicted transcript with a sequence similar to tRNAs, several criteria need to be met to define a TLS. The potential TLS should (i) be able to fold into the full (or at least partial) shape of a tRNA (195), (ii) it should be able to interact with one or more tRNA-specific interactors (e.g., aminoacyl-tRNA synthetases, RNase P, RNase Z, CCA-NTase, or eEF1A). Several TLS that fulfill these requirements reside in viral genomic RNAs, mRNAs, and noncoding RNAs. Employing tRNA mimicry is a relatively simple strategy for viruses to hijack and to exploit the host’s cellular inventory for its own purposes. For instance, the Hepatitis C virus harbors a tRNA-like domain in its internal ribosome entry sites to facilitate binding to the human 80S ribosome (196). A tRNA-like element has also been identified in the 5′ untranslated region of the HIV-1 genome proximal to the primer-binding site. This U-rich element mimics the anticodon loop of human tRNA^Lys_UUU and competitively binds to lysyl-tRNA synthetase, thus increasing the availability of free tRNA^Lys_UUU, which serves as the primer for reverse transcription (197). As demonstrated in the crystal structure of Turnip yellow mosaic virus-TLS, the TLS folds into the characteristic L-shape, containing a pseudo-knotted acceptor stem (195). These TLSs can also be aminoacylated, like bona fide tRNAs. However, the selectivity of aminoacyl-tRNA synthetases limits aminoacylation of TLS to three amino acids, namely valine, histidine, and tyrosine. These various aminoacylated TLSs execute their function via different routes (198, 199). The TLS from Brome mosaic virus (BMV) has served as a model system to investigate the regulation of static and dynamic RNA structures for decades. For instance, the BMV TLS has been demonstrated to change conformation upon metal ion binding, which directly regulates viral replication (200). Recently, Bonilla et al. (201) presented the cryo-EM structures of the BMV TLS as well as the complex between TLS and tyrosyl-tRNA synthetase. Their work revealed how the BMV TLS undergoes conformational changes to interact with a host factor in a noncanonical fashion to achieve aminoacylation.

In stark contrast to the abundance of TLSs in viruses, only a limited number have been found in eukaryotic model organisms or humans (194). There are two known TLS in humans, namely mascRNA and Menβ, which are the cleavage products derived from two long noncoding RNAs, namely metastasis-associated long adenocarcinoma transcript 1 (MALAT1) and Menβ (also known as NEAT1). Both TLSs are CCA-edited at their 3ʹ-ends, but Menβ possesses an unstable acceptor stem which prompts the CCA-NTase to add tandem CCA motifs and trigger degradation (202). Although mascRNA has a 3ʹ-CCA, it is not aminoacylated. mascRNA has been linked to increased global protein synthesis and cell proliferation rates, as well as to several other cellular events, including antiviral and cardiovascular innate immunity regulation. A recent biochemical investigation demonstrated that mascRNA binds to glutaminyl-tRNA synthetase via its unique anticodon loop and upregulates glutaminyl-tRNA synthetase protein levels to promote protein translation (203). Although the aminoacyl-tRNA synthetase–enhanced protein translation is also observed in histidyl-tRNA synthetase (204), the details of this regulatory circuit remain elusive.

Y RNA-like A (YrlA) RNA represents a new class of tRNA mimicry, as it contains canonical tRNA motifs. YrlA RNA belongs to the family of Y RNAs, which are noncoding RNAs mostly found in bacteria and animal cells. The tRNA mimicry resembles the distinct tRNA L-shape (205), and these motifs are substrates of tRNA processing and modifying enzymes. The tRNA mimicry motif in Y RNA is believed to stabilize the structure and facilitate interaction with the Ro60 protein to regulate downstream cellular functions. Yet another kind of TLS is named nimtRNA for nuclear intronic mitochondrial-derived tRNAs. These nimtRNAs are transcribed from nuclear genomic DNA, which were originally derived from mitochondrial tRNA genes (206). These mitochondrial-tRNA look-alikes have been transferred and inserted into the human nuclear genome and are mainly found in the intronic regions of protein-coding or noncoding transcripts. Some of the sequences are identical to their mitochondrial predecessor, and others contain up to 25 mismatches while retaining the ability fold into the distinct tRNA structure (207). It has been shown that the presence of nimtRNAs in introns promotes pre-mRNA splicing and the tRNA-like motif seems crucial for the enhanced splicing efficiency.

tRNA sensing and metabolism

The T-box riboswitch is a regulatory noncoding RNA that binds to specifically uncharged tRNAs and thereby sense the availability of amino acids. This readout is used to attenuate downstream gene expression at the transcriptional or translational level (208). Over 23,000 T-box riboswitch domains are annotated according to the riboswitch database (209). These riboswitches are mainly found at the 5ʹ-leader regions of certain mRNAs. Meanwhile, a smaller number of T-box riboswitches can be found mostly in Actinobacteria, where they control translation. In general, riboswitches act in a two-step sensory and regulatory mechanism. In the past decade, many crystal and cryo-EM structures of T-box riboswitches with the appropriate tRNA target have been reported, but most structures only contain parts of the riboswitch. Therefore, we lack a complete picture of how each motif of the riboswitch orchestrates the conformational changes upon tRNA binding (210). Recently, an intermediate resolution cryo-EM structure of the complete Bacillus subtilis glyQ T-box riboswitch, which belongs to the class that controls transcription, revealed the overall conformation during tRNA recognition (211) (Fig. 5B). The overall U-shape architecture of the riboswitch “clamps” the cognate tRNA^Gly_GCC and the stem III, and discriminator domains sense the aminoacylation status (211). The regulatory domain can then switch between a terminator or antiterminator confirmation to govern the subsequent activation of transcription or translation (212). Overall, the recognition of cognate tRNA happens via the stem I domain of the riboswitch, which pairs the anticodon sequence of the tRNA with a “Specifier Sequence”. A crystal structure of the full-length of Mycobacterium tuberculosis ileS T-box, which belongs to the class that controls translation, shows that it recognizes the cognate tRNA similar to the glyQ T-box but does not contact the tRNA elbow (213). Another crystal structure of stem I and stem II domains of a Nocardia farcinica ileS translational T-box riboswitch in complex with its cognate tRNA^Ile_GAU reveals a different picture of how stem II reinforces the interaction of stem I and tRNA (214). In summary, the rigid structure of tRNAs is used by a variety of biosensors to determine the aminoacylation levels of tRNA, which is directly related to the metabolic state of the cell.

Glycosylated tRNAs

tRNA is heavily decorated with various chemical moieties (90), including the attachment of sugar groups. Queuosine (Q), a 7-deazaguanosine nucleoside, is obtained from gut bacteria and incorporated in the anticodon wobble position of cytosolic tRNA^Tyr_GUA, tRNA^His_GUG, tRNA^Asn_GUU, and tRNA^Asp_GUC in humans (215). The Q-modified position 34 in tRNA^Tyr_GUA and tRNA^Asp_GUC is further “sugar coated”. Unlike the polysaccharide conjugates on other noncoding RNAs, proteins, or lipid, tRNA^Tyr_GUA and tRNA^Asp_GUC carry a monosaccharide. Two glycosylated derivatives exist, namely galactose-Q (galQ)-tRNA^Tyr_GUA and mannose-Q (manQ)-tRNA^Asp_GUC. A method has been developed using an acid-denaturing gel and northern blot to detect and quantify the presence galQ-tRNA^Tyr_GUA and manQ-tRNA^Asp_GUC (216). These glycosylated variants are found in all tested human organs without any tissue specificity, and the expression levels of all three modifications (galQ, manQ, and Q) seem to be correlated with age. These modified tRNAs are located in the cytosol and the nucleus. The enzyme responsible for this modification and the functional significance of this modification are yet to be identified. The chemical synthesis and the structure of galQ (217) and manQ (218) have been reported, while the biochemical pathways leading to them in living organisms still remains unknown. In addition to these known sugar modifications of tRNA^Tyr_GUA and tRNA^Asp_GUC, a recent study employed a global labeling method and discovered that many tRNA species are attached to sialylated glycans in living cells (219). These so called “glyco-tRNAs” as well as other “glyco-RNAs” seem to be associated with membrane organelles and can also be detected on the cell surface. However, the modified sites on tRNAs and the detailed mechanism of RNA glycosylation are not known.

The atypical appearance of armless tRNAs

Despite the highly conserved L-shape structure of tRNA, there exists a group of structurally bizarre tRNAs (Fig. 5C) lacking D- and T-arms (220). They mostly exist in the mitochondria of nematodes, like Caenorhabditis elegans or Ascaris suum. Recently, it has also been found in the mitochondrial genome of arachnids (221). These structurally degenerate tRNAs are the consequences of the high mutations rates and genetic drift in the mitochondrial genome. Lack of either T-arm or D-arm (or both in extreme cases), as well as accumulation of weak base pairs and mismatches in stem regions, gives rise to the observed structural diversity (222, 223, 224). For instance, the smallest tRNA (with only 42 nt) has been identified in the mitochondria of Dermatophagoides farina, a mite. The predicted structure of two base-paired stems, connected by a variable bulge element, still resembles the dimensions and architecture of tRNAs (225). In the case of the T-armless tRNA, the canonical T-arm is replaced by a less structured 6- and 12-nt loop, named TV-replacement loop. Computational predictions show that some of these arm-less tRNAs retain the basic hairpin structure and adopt stable 3-D structures with “boomerang-like” shapes (224). Regardless of the missing parts, some of the miniaturized tRNAs have been proven to be functional, as they can be further edited by amino acid–specific aminoacyl-tRNA synthetases, CCA-NTase (226), and can be recognized by specific Elongator factor-Tu (EF-Tu) for mitochondrial protein translation (227). Of note, posttranscriptional modifications are also present on these arm-less tRNAs, including a methylation at position 9 (m¹A₉) (223). These modifications could stabilize mt-tRNA structure, enhance thermostability, and facilitate aminoacylation and translation. Correct aminoacylation of these arm-less tRNAs is crucial to ensure fidelity of mitochondrial translation. As the arm-less tRNA acts as an adapter (like the canonical tRNA), the mitochondrial EF-Tu1 and EF-Tu2 have undergone coevolutionary modulation to recognize features of these truncated tRNAs (228).

Mammals harbor mitochondrial tRNAs without D-arms, as exemplified by the mt-tRNA^Ser_GCU iso-acceptors that show this deviated architecture (229). Similar to armless tRNAs from nematodes, mammalian mt-tRNA^Ser_GCU also harbors RNA modifications (22), including three consecutive 5-methylcytosines (m⁵C) at positions C₄₇, C₄₉, and C₅₀. A recent cryo-EM structure revealed the details of the armless mt-tRNA^Ser_GCU in complex with human mitochondrial seryl-tRNA synthetase (180). Due to the lack of D arm, the mt-tRNA^Ser_GCU loses its tRNA core domain and unexpectedly adopts an acute-angled “Y” shape. Furthermore, this drastically reshaped topology results in the loss of identity elements in mt-tRNA^Ser_GCU. As a consequence of this highly degenerated architecture, the human mitochondrial seryl-tRNA synthetase has also coevolved to bind to the specific acceptor stem of mt-tRNA^Ser_GCU via its distinct helical arm and to charge the tRNA. Moreover, the stability of mt-tRNA^Ser_GCU is sensitive to its tRNA charging status (230). The loss of mt-tRNA^Ser_GCU is linked to retinal disease and class II mitochondrial seryl-tRNA-synthetase–related diseases (231).

Suppressor tRNAs

A subgroup of tRNAs exist as “suppressor tRNAs” in some species. These tRNAs can read through stop codons and nonsense mutations during translation (232, 233). They have been postulated to have resulted from mutations in the anticodon sequence, which enabled pairing with stop codons and decoding nonsense mutations in mRNAs. Recently, Kachale et al. discovered a “reassignment” of an in-frame stop codon to tRNA^Trp_CCA in a trypanosomatid, Blastocrithidia nonstop. A similar mechanism is also found in the ciliate Condylostoma magnum. Interestingly, tRNA^Trp_CCA in both species harbors a shorter tRNA anticodon stem (from 5-bp to 4-bp). When a shorter anticodon stem version of tRNA^Trp_CCA was engineered in yeast system, this specific tRNA indeed enhanced stop codon read-through (233). In principle, the simple engineering of the anticodon triplet of sense-coding tRNAs could transform them into suppressor tRNAs. However, only a few such mutant tRNAs have actually achieved high stop codon read-through activity and restored protein production (234). Various groups have taken different approaches (15, 235, 236) to understand what segments of tRNA are critical for suppression activity. Interestingly, redesigning the T-stem seems to be key strategy for suppressor tRNA to maintain a high decoding rate by strengthening the interaction with eEF1A.

The “+1-frameshift” (+1FS) suppressor tRNAs contain a single-nt insertion, which most often appears within the ASL (237, 238, 239). This extra nt generates a longer loop (8 nt), which differs from the canonical architecture. Nonetheless, the frameshift suppressor tRNA still decodes like a normal tRNA, as its anticodon is still defined by the universally conserved U₃₃ and m¹G₃₇ (240). The local ASL structure still undergoes distortion by disrupting the interaction between position 32 and 38 (239, 241, 242). A +1-frameshift takes place during translocation, which is mediated by +1FS suppressor tRNA, ribosome, mRNA, and elongation factor EF-G (241, 243, 244). These “slippery” tRNAs, which facilitate frameshifting of codon triplets, could be exploited biotechnologically (245).

tRNA in biotechnological and pharmaceutical applications

tRNA is not only important in fundamental regulation but also has been used in biotechnological and pharmaceutical applications. Ponchon and Dardel developed a system based on a camouflage strategy to fuse a target RNA with a tRNA scaffold (246). Due to the rigid and highly structured features of tRNA, the chimeric RNA was able to escape cellular RNase digestion. The target RNA can then be produced, purified, and released by RNase H treatment for downstream applications. Recently, Lee et al. (247) have taken the “tRNA-scaffold” approach and successfully determined the crystal structure of the ZIKA virus stem-loop A-tRNA to provide insights into the mechanism of stem-loop A–mediated RNA synthesis. As suppressor tRNAs naturally exist in a few species, the frameshift-suppressor tRNA or suppressor tRNA that decodes nonsense mutation have been employed in orthogonal translation systems (245). This strategy allows refactoring codons and incorporating noncanonical amino acids to expand the genetic code for various designed functions. For instance, an engineered suppressor tRNA can be charged with the desired amino acid containing the glycan modification and this artificial product is incorporated into the target protein at the chosen position for further analyses (248). Moreover, decades ago, suppressor tRNAs have been proposed and engineered as a therapeutic tool to read through the premature codon termination. This method has been applied to several premature codon termination–related disease models, such as β-thalassemia (249) and cystic fibrosis transmembrane conductance regulator (234). The selected engineered tRNA indeed display promising effects on rescuing target protein production and restoring the responsible cellular function in cell and animal models (250, 251).

Future perspectives and directions

Despite the fact that tRNAs are highly structured and rigid RNA molecules, their dynamic and transient interactions with many processing factors and modification enzymes have complicated the structural analyses of the different maturation steps of tRNAs. Research on the dynamic structures of RNA domains has recently been changed fundamentally by the advent of high-resolution single particle cryo-EM (252). Recent examples (175, 176) have shown that cryo-EM facilitates the structural analyses of several different reaction intermediates without the need for identifying novel crystallization conditions for each individual conformation. We will most likely obtain whole galleries of structural snapshots for most tRNA modification reactions in the near future. Such an advance will fundamentally change our understanding of the sequential order of individual reaction steps and allow us to explore the dynamic transitions between them. Cryo-EM requires small amounts of sample, an attribute that will enable us to conduct structural studies on low-abundant RNAs isolated from endogenous sources or from sufficient quantities of in vitro–modified RNA samples. It is not only a new method to study previously known samples but it also allows us to rethink the list of RNA samples that can be visualized at high-resolution.

Recent advances in the field have essentially solved the fundamental problem of predicting three-dimensional structures of proteins from the primary amino acid sequence (253). The relatively small number of experimentally determined RNA and protein-RNA structures hinder the development of reliable models to predict the structure of these dynamic assemblies reliably in silico. Nonetheless, we are certain that the rapid advances in machine learning and molecular modeling will offer us significant support to understand the interaction between tRNAs and their cellular environment. Recently, an intermediate resolution cryo-EM structure of a DNA G-quadruplexes has demonstrated that we can reconstruct maps even for a 28-kDa biomolecule using electron microscopes (254). This advance suggests that RNAs with similar mass (e.g., tRNAs, armless tRNAs, Y RNAs, or tRNA-like molecule) can also be analyzed by cryo-EM. It remains to be shown whether one can obtain near-atomic resolution structures of RNAs that are more dynamic than proteins. As certain RNA modifications are known to increase the stability/rigidity of RNAs, these modifications may be exploited to achieve higher resolution structures.

Apart from studies on tRNA and tRNA-like molecules, numerous functional studies have identified subpopulations of small noncoding RNAs that are derived from tRNA, called tRNA halves or tRNA-derived fragments (Fig. 5C). These tRNA derivatives, either as separated tRNA halves or as a nicked tRNA complex, were reported to be involved in a plethora of cellular functions and pathways (255, 256, 257, 258).

As mentioned above, it still remains unclear if and how individual tRNA modification cascades are coupled or even depend on each other (97, 98, 162, 163). The first time-resolved NMR studies have tried to dissect the hierarchy of tRNA decoration and highlight the exciting complementary opportunities to record multiple modifications steps in parallel (162, 259). As tRNA modifications can directly influence the folding of tRNAs, the efficiency of the underlying modification reactions might provide another route to control cellular translation. As some of the modification cascades directly rely on basic metabolic pathways (260, 261, 262), tRNA-related processes might represent molecular switches between the metabolic state and the regulation of protein synthesis in a cell. Therefore, we need to fully understand tRNA processing and modification pathways at the cellular and organismal level in the future.

A major question that needs to be addressed soon is the actual role of the processing and modification enzymes for the functionality of tRNAs. We lack a fundamental mechanistic understanding of whether the tRNA-interacting proteins can contribute more than “just” attaching chemical groups or if they also act as molecular chaperones or regulators. Here, the biophysical and structural analyses of tRNAs before and after a specific modification reaction might provide direct insights. Furthermore, it remains to be shown whether tRNA splicing and processing activate or respond to specific tRNA modifications. Finally, modified nt of tRNAs, mRNAs, and ribosomal RNA are proximal to one another during ribosomal decoding. We need to understand whether these molecules and their modifications communicate with one another to reveal yet another regulatory layer of gene expression.

Finally, tRNA-based technologies seem to have a very bright future in the field of personalized biomedicine, especially if suppressor tRNAs break the current limitations of highly selective aminoacyl-tRNA synthetases. Here, natural variations of tRNAs and translation systems can and should inspire us to develop novel tRNA-based tools and technologies that could enable us to control, correct, or restore cellular protein synthesis.

Conflict of interest

The authors declare that they have no conflicts of interest with the contents of this article.

Reviewed by members of the JBC Editorial Board. Edited by Karin Musier-Forsyth