Human tRNA^Lys3_UUU is Pre-Structured by Natural Modifications for Cognate and Wobble Codon Binding through Keto-EnolTautomerism

Abstract

Human tRNA^Lys3_UUU (htRNA^Lys3_UUU) decodes the lysine codons AAA and AAG during translation, and also plays a crucial role as the primer for HIV-1 reverse transcription. The post-transcriptional modifications 5-methoxycarbonylmethyl-2-thiouridine (mcm⁵s²U₃₄), 2-methylthio-N⁶-threonylcarbamoyladenosine (ms²t⁶A₃₇)and pseudouridine (ψ₃₉) in the tRNA'santicodon loop are critical for ribosomal binding and HIV-1 reverse transcription. To understand the importance of modified nucleoside contributions, the structure and function of this tRNA's anticodon stem and loop domain were determined with these modifications at positions 34, 37 and 39, respectively (hASL^Lys3_UUU-mcm⁵s²U₃₇;ms²t⁶A₃₇;ψ₃₉). Ribosome binding assays in vitrorevealed that the hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉bound AAA and AAG codons, whereas binding of the unmodified ASL^Lys3_UUU was barely detectable. The UV hyperchromicity, the circular dichroism and the structural analyses indicated that ψ₃₉ enhanced the thermodynamic stability of the ASL through base stacking while ms²t⁶A₃₇ restrained the anticodon to adopt an open loop conformation that is required for ribosomal binding. The NMR-restrained molecular dynamics derived solution structure revealed that the modifications provided an open, ordered loop for codon binding. The crystal structures of the hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ bound to the 30S ribosomal subunit with each codon in the A site showed that the modified nucleotides mcm⁵s²U₃₄ and ms²t⁶A₃₇ participate in the stability of the anticodon/codon interaction. Importantly, the mcm⁵s²U₃₄•G₃ wobble base pair is in the Watson-Crick geometry, requiring unusual hydrogen bonding to G in which mcm⁵s²U₃₄ must shift from the keto to enol form. The results unambiguously demonstrate that modifications pre-structurethe anticodonas a key prerequisite for efficient and accurate recognition of cognate and wobble codons.

INTRODUCTION

Accurate ribosome-mediated protein synthesis requires proper recognition of mRNA codons by specific aminoacylatedtRNAisoacceptors. The influence of post-transcriptional tRNA modifications on translational fidelity has been widely investigated.^1–4 As a result of such studies, 93 different, naturally-occurring modification chemistries have been identified in tRNAs.⁵ These modifications, particularly in the anticodon stem and loop domain (ASL), have been shown to significantly impact ribosome binding affinity,^6–8 codon discrimination, thermodynamic stability,¹¹ speed of translation,¹² recognition by cognate aminoacyl-tRNAsynthetases¹³ and reading frame maintenance.^14–17 Specifically, modifications to the wobble nucleoside at position-34 and the conserved purine residue at position-37, 3'-adjacent to the anticodon, are essential for pre-structuring the ASL, permitting proper anticodon conformation and efficient codon recognition. Human tRNA^Lys3_UUU (htRNA^Lys3_UUU; Fig. 1a,b) is of particular interest because of its dual role in promoting proper endogenous translation and acting as the primer selected for initiation of reverse transcription (RT) by human immunodeficiency virus 1 (HIV-1) and other lentiviruses.¹⁹ Indeed, the ASL modifications present in htRNA^Lys3_UUU have been shown to be required for efficient initiation of RT in HIV-1.^20–22

Figure 1.

Human tRNA^Lys3_UUU, its anticodon stem and loop domain and its modified nucleosides. (a) Sequence and secondary structure of human tRNA^Lys3 (htRNA^Lys3_UUU) with all of its known modified nucleosides: N²,N²-dimethylguanosine at position 10, m²₂G₁₀; dihydrouridine at positions 16, 20 and 48, D; pseudouridine-27 (ψ), 39 (ψ) and 55 (ψ); 5-methoxycarbonylmethyl-2-thiouridine-34, mcm⁵s²U₃₄; 2-methylthio-N⁶-threonylcarbamoyladenosine-37, ms²t⁶A₃₇; N⁷-methylguanosine, m⁷G; 5-methylcytidine at positions 48 and 49, m⁵C; 2'-O-methylribothymidine (2'-O-methyl-5-methyluridine), Tm; N¹-methyladenosine-58, m¹A. The modifications,mcm⁵s²U₃₄, ms²t⁶A₃₇ and ψ₃₉are in red. The ASLs were synthesized with a G₂₇•C₃₄ terminal base pair instead of the ψ₂₇•A₄₃ in order to stabilize the stem. (b) Chemical structures of the modified nucleosides within the anticodon stem and loop domain of htRNA^Lys3_UUU: mcm⁵s²U₃₄; ms²t⁶A₃₇; and ψ₃₉.

The anticodon stem and loop (ASL^Lys_UUU) domain of the sole E. coli tRNA^Lys_UUUhas been studied extensively as a model due to its modified nucleotides being similar to those of the human tRNA^Lys3_UUU and its anticodon loop nucleoside sequence being identical.²³ The ASL loop modifications of the E. colitRNA^Lys_UUU, 5-methylaminomethyl-2-thiouridine (mnm⁵s²U₃₄) and N⁶-threonylcarbamoyladenosine (t⁶A₃₇), pre-structure the loop into the canonical U-turn conformation that is necessary for efficient binding to synonymous codons in the ribosomal A-site. Both studies in vitro^27–29 and in vivo³⁰ have shown that a 2-thiolation, and not 5-methylaminomethylation, at the wobble position acts as a strong positive identity element for recognition of tRNA^Lys_UUU and tRNA^Glu_UUC by their respective cognate aminoacyl-tRNAsynthetases. X-ray crystallography of the ASL domain of E. colitRNA^Lys on the 30S ribosomal subunit demonstrated clearly that the t⁶A₃₇ modification further stabilizes the unusually low enthalpy of binding in the A–U rich codon-anticodon interaction by forming a cross-strand stack with the first adenosine residue of the codon.²⁵ The t⁶A₃₇ modification and its derivatives are found in the tRNAisoacceptors that decode codons beginning with A in all three Domains of life.

htRNA^Lys3_UUUis one of three lysine isoacceptors found in mammalian species and is responsible for decoding the minor lysine codon AAA, as well as the major AAG codon.³² The ASL domain of htRNA^Lys3_UUU is more extensively modified than that of the E. colitRNA^Lys_UUU. The anticodon loop of hASL^Lys3_UUU has a 5-methoxycarbonylmethyl-2-thiouridine at position-34 (mcm⁵s²U₃₄)and a 2-methylthio-N⁶-threonyl-carbamoyladenosine at position-37 (ms²t⁶A₃₇).³⁵Apseudouridineis present at position-39 (ψ₃₉) in the ASL stem, immediately adjacent to the loop (Fig. 1a,b). In contrast to the htRNA^Lys3_UUU, the ASLs of the isoaccepting htRNA^Lys1,2contain an unmodified C₃₄ and a more simply modified A₃₇ (t⁶A₃₇). The htRNA^Lys1,2isoacceptors represent two-thirds of all lysine tRNA in mammalian cells, and are specific for decoding only the dominant AAG codon.In vitro ribosome binding studies using the hASL^Lys3_UUU with the E. coli modifications, mnm⁵U₃₄ and t⁶A₃₇, demonstrated that the insertion of the individual modifications in the ASL at positions 34 or 37 respectively, would rescue codon binding to only the cognate AAA.With both modifications present the ASL recognized both AAA and AAG.⁸ This highlights the importance of the two-hypermodificationsbeing present for dual codon recognition and specificity.

The importance of the hypermodified nucleotides mcm⁵s²U₃₄ and ms²t⁶A₃₇to thestructure and functions of htRNA^Lys3_UUUhave been modeled and some empirical data reported.^8,21,39–41However, tangible structural evidence of their functional contributions to the fully modified hASL^Lys3_UUU is still lacking. Here, using NMR, X-ray crystallography and otherbiophysical and biochemical methods, we report the key structure/function relationship for these naturally occurring modifications to the anticodon domain of htRNA^Lys3_UUU. To the best of our knowledge, the modified nucle otides mcm⁵s²U₃₄, ms²t⁶A₃₇ and ψ₃₉ were introduced simultaneously into the hASL^Lys3_UUU sequence for the first time through chemical synthesis. Codon binding characteristics and atomic resolution structures, both in solution and on the 30S ribosomal subunit, of the doubly modified hASL^Lys3_UUU-mcm⁵s²U₃₄,ms²t⁶A₃₇ and triply modified hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ show that the modifications are instrumental in the pre-structuring of the loop into an open loop conformation and facilitating keto-enoltautomerismthat is essential for cognate and wobble codon binding.

RESULTS

Recognition of lysine codons at the ribosomal A-site

The triply modified hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ was chemically synthesized in order to determine the effect of the native modifications on its structure and function in binding the cognate and wobble codons AAA and AAG. The site-selected introduction of the complex modifications mcm⁵s²U₃₄ and ms²t⁶A₃₇ is problematic and required a novel chemical synthesis of the oligonucleotide in which the modified nucleoside functional groups as well as the major bases were transiently protected. The protecting chemistries had to be compatible with each other and with that of the major nucleosides such that all are readily and quantitatively removed when the oligonucleotide synthesis was complete. The cyclic chemistry involves the repeated coupling of a nucleoside phosphoramiditeto the growing sequence and, at each addition, the oxidation of the trivalent phosphate into a pentavalent phosphate. Removal of the oligonucleotide from the solid support and deprotection of the C2'-OH are accomplished under conditions that would normally alter the modified nucleosides. A novel use of protecting agents and an alteration of the deprotection protocol maintained the native integrity of the modifications (Supplementary Fig. S1). The presence of the modifications in stoichiometric amounts was quantified by nucleoside composition analysis with HPLC (Supplementary Fig. S2), and observed in NMR and X-ray crystallography. The terminal base pair ψ₂₇•A₄₃ was substituted with a G₂₇•C₄₃ pair to enhance stability of the ASL constructs (Fig. 1a).

The triply modified hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ bound AAA and AAG at the A-site of the E. coli ribosome with a high affinity, dissociation constants (K_d) of 3.1 ± 0.4 μM and 3.9 ± 0.8 μM, respectively (Fig. 2a,b). In contrast, the unmodified ASL^Lys3_UUU exhibited very poor binding to both lysine codons. The equilibrium binding of the unmodified hASL^Lys3_UUU to either of the lysine codons AAA or AAG was nearly undetectable (Fig. 2a,b). Previously, we had observed that the single atom substitution of sulfur for oxygen at the 2-position of U₃₄ rescued ASL^Lys_UUU binding of the cognate and wobble codons, and that the introduction of only t⁶A₃₇ enabled AAA binding, but not AAG binding or translocation to the P-site.^6,8,15,25,42 Our experiments with the unmodified and modified hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ revealed that anticodon domain modifications enhanced codon binding, similar to results reported for the variously modified anticodon domain of the bacterial tRNA^Lys_UUU (ASL^Lys_UUU-s²U₃₄, ASL^Lys_UUU-mnm⁵U₃₄, ASL^Lys_UUU-t⁶A₃₇ and ASL^Lys_UUU-mnm⁵U₃₄;t⁶A₃₇).^7,8

Figure 2.

Ribosome-mediated codon binding and thermodynamic analysis.(a) Equilibrium binding of the cognate codon (a) AAA and (b) AAG in the aminoacyl, or A-site of the Escherichia coli 70S ribosome by hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇ (▲) and the unmodified hASL^Lys3_UUU (■). (c) UV-monitored, thermal denaturations of the hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇ (red,-—), hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ (green,—) and the unmodified hASL^Lys3_UUU (blue,—). Thermal denaturations/renaturations are the averages of three separate experiments. Results are presented after baseline correction and normalization to 1.00 at maximum absorbance. Thermodynamic parameters extracted from these experiments are found in Table 1 and reflect errors as one standard deviation. (d) Circular dichroism (CD) spectra of the hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇ (red,-—), hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ (green,—) and the unmodified hASL^Lys3_UUU (blue,—). Spectra were collected at 25 °C, and are shown as the averages of three separate experiments after baseline correction.

Anticodon loop modifications alter thermal stability

A comparison of the thermal stabilities of the differently modified hASL^Lys3_UUU constructs by UV-monitored thermal denaturation revealed both similarities and distinct differences in their thermodynamic properties. While the enthalpy (ΔH), entropy (ΔS) and standard free energy (ΔG) did not vary significantly, the introduction of modifications caused a reduction in the melt temperature, T_m, at which half of the RNA molecules are denatured.The hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇ and the hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉exhibited a T_m of ~53 °C in comparison toa T_m of ~57 °C for the unmodified hASL^Lys3_UUU (Fig. 2c, and Table 1). Even so, the hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉exhibited the greatest degree of hyperchromicityin comparison to the unmodified hASL^Lys3_UUU and the hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇; (Fig. 2c and Table 1). The degree of hyperchromicity is a measure of base stacking and overall molecular order.

Table 1.

Thermodynamic properties derived from UV-monitored thermal denaturation and renaturation.

	hASLLys3-Unmodified	hASLLys3-mcm5s2U34;ms2t6A37	hASLLys3-mcm5s2U34;ms2t6A37;ψ39
ΔH (kcal/mol)	−45.66 ± 4.32	−41.69 ± 1.76	−46.73 ± 7.74
ΔS (cal/K*mol)	−139.90 ± 13.30	−127.97 ± 5.46	−143.10 ± 23.66
ΔG37 (kcal/mol, 37 °C)	−2.76 ± 0.28	−2.00 ± 0.15	−2.34 ± 0.42
Tm (°C)	56.82 ± 1.37	52.67 ± 1.12	53.37 ± 1.16
Hyperchromicity (%)	13.3 ± 0.80	11.0 ± 0.80	13.9 ± 1.50

Modifications may affect the molecular order of the anticodon through base stacking interactions. Relative base stacking interactions are readily observed by comparison of circular dichroism spectra. The circular dichroism(CD) spectra of the ASLs were analyzed at the maximum positive ellipticityof 264 nm. A negative ellipticity at a wavelength of λ = 330 nm and positive ellipticity at λ = 310 nm for both the doubly and triply modified ASLs was also observed (Fig. 2d). These ellipticities are in agreement with characteristics typical of a 2-thio modified uridine.⁴³ The hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ exhibited a small but reproducibly greater positive ellipticityat 264 nm than the other two ASLs, suggesting again a higher degree of order.

Structure of the ASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇in solution

Sequence-specific NMR resonance assignments

The results of the UV-monitored thermal denaturations and CD spectroscopy implied that the mcm⁵s²U₃₄, ms²t⁶A₃₇ and ψ₃₉ modifications may play a role in ordering the dynamic structure of the hASL^Lys3_UUU. To study and understand the extent to which these modifications affect the anticodon conformation and dynamics, the three dimensional (3D) solution structure of the doubly modified hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇ was determined through NMR and restrained molecular dynamics. In addition, the structure of its unmodified counterpart was fully characterized by NMR. In achieving this goal, the identification and sequence-specific assignment of the nucleotide spin systems was completed by using combinations of homo-(¹H-¹H)-nuclear and natural abundance hetero-(¹H-¹³C; ¹H-³¹P)-nuclear 1D and 2D NMR experiments conducted in H₂O or D₂O solvent.Well-established protocols were employed for the assignment of nucleic acid resonances in the absence of isotopic labeling.^44,45 The NMR spectra were recorded in the absence of Mg²⁺ because previous experiences revealed that the addition of a physiological amount of Mg²⁺ ions (1–5 mM) to the modified ASL samples did not significantly affect the resulting NMR spectra. The addition of larger concentrations of Mg²⁺ induced spectral line broadening in some ASLs.^11,46,47

Exchangeable protons

The imino and amino protons of the doubly modified hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇ were identified and assigned using ¹H 1D and ¹H-¹H 2D NOESY NMR spectra that were recorded in H₂O at 2 °C.In assigning the exchangeable protons,we used the initial assumption that the stem of the ASL was reasonably close to A-type RNA conformation, as was indicated by CD spectroscopy (Fig. 2d). Four resonance signals were observed in the low field region (12.50 – 14.00 ppm) of the ¹H 1D spectrum of the hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇. These NMR signals were characteristic of imino protons that are engaged in Watson-Crick base pairs and were sequentially assigned to the G₂₇H1, G₄₂H1, U₄₁H3 and G₃₀H1 (Supplementary Fig. S3). The resonance peak of the G₂₇H1 displayed a low intensity and broad line width which is expected from the terminal base pair G₂₇•C₄₃. The resonance signal corresponding to the H1 imino proton of U₃₉ that is engaged in the A₃₁•U₃₉ base pair adjacent to the loop could not be identified. Therefore, this imino proton was not assigned. The assignment of the imino protons allowed us to fully assign the amino protons of the nucleotides of the stem. Under our NMR experimental conditions (see Methods and Materials) the rapidly exchanging amino and imino protons of the loop nucleosides could not be identified.

Interestingly, the HN proton of the amide group of the threonyl moiety of ms²t⁶A₃₇ (Fig. 1b) was clearly identified due to the distinctive chemical shift of its resonance peak observed at 9.66 ppm (Fig. 3). Nuclear Overhauser effect (NOE) cross-peaks were observed between HN and s²-CH₃, between HN and Hβ of the methylene group and between HN and the gamma methyl group, CγH₃, of the threonylmoiety (Fig. 3). However, a very weak NOE connectivity between HN and Ha was hardly visible at 2 °C. Additionally, at 2 and 22 °C, NOE signals between the water molecules and the assigned imino and amino protons were observed, except in the case of HN. Also an NOE was detected between CγH₃ and s²-CH₃.

Figure 3.

Two dimensional homonuclear (¹H-¹H) NOESYNMR spectrum of hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇. The spectrum demonstrates the unambiguous identification ofthe modified nucleoside residues. NOE cross-peaks were observed between the atoms of mcm⁵s²U₃₄ and ms²t⁶A₃₇ and those of the residues that are in their respective vicinity.This spectrum of hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇ (1.5 mM in 90% ¹H₂O + 10% ²H₂O at pH 6.8, 295 °K) was recorded with a mixing time of 400 ms and at 500 MHz.

Non-exchangeable protons

The H6 and H5 aromatic protons of the nine unmodified pyrimidines of the doubly modified hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇ could readily be identified as a result of the presence of intense cross-peaks observed in the COSY and TOCSY spectra (between F₁ = 5.18 – 5.90 ppm and F₂ = 7.30 – 7.90 ppm). The cross-peaks of the cytidines and uridines were differentiated from each other by using their respective ¹³C chemical shifts detected in the ¹H-¹³C HSQC spectrum. The H6 proton of the modified uridinemcm⁵s²U₃₄ was identified on the ¹H-¹³C HSQC and later assigned on the NOESY spectra. Sequential assignments were obtained within the aromatic to H1' resonance region of the NOESY spectra (F₁ = 5.18 – 6.60 ppm and F₂ = 6.90 – 8.40 ppm). An NOE walk (Supplementary Fig. S4), that is characteristic of A-form RNA conformation, could be observed through intra/inter-nucleotide NOE cross peaks occurring from G₂₇H1'-H8 to G₄₃H1'-H6. Disruptions in the NOE walk only occurred in the anticodon loop between mcm⁵s²U₃₄ H1'-H6 and mcm⁵s²U₃₄ H1'-U₃₅H6 and between ms²t⁶A₃₇ H1'-H8 and ms²t⁶A₃₇H1'-A₃₈H8.

The relatively small size of the heptadecamerASL, and the high level of purity of the NMR samples combined with the good resolution and dispersion of the NMR cross-peaks allowed us to achieve a near-complete assignment of the non-exchangeable proton resonances. The H5'/H5” protons were not entirely identified and assigned as a result of spectral overlap. Despite the disruptions in the NOE walks, NOE connectivities that were important for structure determination were observed between the unmodified and modified nucleotides of the loop region (Supplementary Fig. S4).

Modifications may affect the presence of a canonical U-turn in the anticodon loop at the invariant U₃₃. Therefore, the solution structure of the doubly modified ASL was investigated for proton and phosphorus resonances that characterize the U-turn. The canonical U-turn conformation in RNAs is confirmed through a distinctive ¹H-NMR “fingerprint.”⁴⁸ This fingerprint usually consists of NOE signalsbetween the anomeric H1' proton of the conserved U₃₃ and its own H5/H6, and between the anomeric H1' proton of U₃₃ and the U₃₅H5/H6. In addition, U₃₃H2' has NOE connectivities to H5/H6 of bases U₃₄ and U₃₅. Also, the phosphorus of the dinucleotide U₃₃33pU₃₄ has a strong down-field chemical shift which reflects the fact that the a dihedral angle is trans. In our study, an NOE was observed between U₃₃H1' and its own aromatic protons for the doubly modified hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇. Another NOE connectivity was detected between U₃₃H2' and mcm⁵s²U₃₄H6. Interestingly, CγH3 of the methylene group of mcm⁵s²U₃₄ gave a medium strength NOE signal to U₃₃H1'. The ¹H-³¹P HETCOR NMR experiment showed that the chemical shifts of the resonances for the ³¹P nucleus were narrowly confined (Supplementary Fig. S5), as commonly observed in the case of A-RNA.^44,49

Solution structure of the hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇

The distance and angle restraints required for the structure calculation were determined following the NMR spectral analyses.⁴⁴ A total of 552 NOE-derived distance restraints were used for the structure calculation. These distance restraints were constituted by 392 and 134 NOE distance restraints that were assigned to the stem and the loop, respectively. In addition, 26 hydrogen bond (hb) and five planarity restraints were also used to define the structure of the stem. A total of 126 dihedral angle restraints were also used to determine the entire structure of the ASL. The results of the UV thermal denaturation, the CD experiments and the pattern of the anomeric-aromatic NOE walk strongly suggested that the stem of the doubly modified ASL adopted an A-RNA conformation. Therefore, the NMR torsion-angle and distance restraints from the stem were complemented with distance restraints that yield the conventional A-form helix RNA conformation, as previously described and employed by others.^47,48

The ten lowest energy structures determined from the torsion angle molecular dynamics calculations that were in agreement with the NMR data were selected to represent the final ensemble (Fig. 4). The final set of solution structures was chosen on the basis of displaying not a single NOE restraint violation superior to 0.5 Å and no dihedral angle violations greater than 5°. In addition to these selection criteria, structures were rejected if their root-mean square deviations (rmsds) of bonds and angles varied from ideal values, i.e. by greater than 0.02 Å and 2.0°, respectively. The pairwise rmsd for all atoms is 0.82 ± 0.20 Å.

Figure 4.

Solution structure of hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇. Ensemble of the ten lowest energy structures of hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇ obtained from the NMR-derived, restrained molecular dynamics. The pairwise rmsd for all atoms is 0.82 ± 0.20 Å from the average structure. The locations of the modified residues are indicated by red circles for clarity.

As expected, the stem of the ASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇ exhibited typical features characteristic of A-RNA conformation with Watson-Crick base pairs⁵⁰ (Supplementary Fig. S6). The unmodified nucleotides of the stem adopted an anti-parallel geometry with a glycosyl C1'-C1' distance of ~10.6 Å. The sugar moieties adopted the C3'-endo conformation except those of residues U₃₃ and mcm⁵s²U₃₄ showing C4'-exo and C2'-exo puckers, respectively. The nucleotides of the loop appeared to be stacked. The distance between the horizontal planes of ms²t⁶A₃₇ and U₃₆ was wider by 2 Å in comparison to that between ms²t⁶A₃₇ and A₃₈. Residues 33–37 of the loop displayed local base step parameters (tilt and roll) that deviated from the standards of A-form helices, but resulted in a stacked anticodon. As a result the labile proton of the acid group of ms²t⁶A₃₇ is in close proximity (2.8 Å) to U₃₄ and U₃₅O4. The distance between A₃₈N6 and C₃₂O2 proximal to the stem was determined to be 3.5 Å, thereby invoking the potential of hydrogen bonding and the limited extension of the stem (Supplementary Fig. S6).

Crystal structures of the hASL^Lys3_UUUinteraction with mRNA on the ribosomal 30S subunit

The crystal structures of the triply modified hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ bound to the cognate 5'-AAA-3' and wobble 5'-AAG-3' codons were obtained by soaking native Thermusthermophilus 30S ribosomal subunit (30S) crystals with the relevant RNA oligonucleotides. The A-site binding antibiotic paromomycin was included to improve both the resolution of the data and the occupancy of the ASL. This has been shown not to change the structure of the codon-anticodon complex itself.⁵¹ The recognition of the cognate codonanticodon complex in the 30S decoding center is identical to structures reported previously.⁵¹ As already observed for other cognate recognition structures,^25,52 the structure of the 30S itself is identical within experimental error compared to its published structures. This invariance of the 30S allows the differences in ASL structures to be unambiguously attributed to the modifications of the ASL alone.Additionally, the high resolution of the structures, 3.1 Å and 2.8 Å for AAA and AAG respectively (Supplementary Table S1), allowed for unambiguous determination of the geometry of the ASL and the codon-anticodon interaction.

Contributions of ms²t⁶A₃₇ and mcm⁵s²U₃₄ to tRNA^Lys3_UUU and codons recognition

The N⁶-threonylcarbamoyl (t⁶) modification of A₃₇ functions in codon binding, using three different mechanisms as observed previously.²⁵ First, the t⁶ modification abrogates the possible U₃₃•A₃₇ base pair, in effect pre-ordering the hASL^Lys3_UUU in a conformation for codon recognition.⁵³ This open loop conformation of the anticodon seems to be critical for codon binding, at least for the case of the UUU anticodon.⁸ Additionally, the ureido group of the t⁶ modification forms a pseudo-heterocycle stabilized by a hydrogen bond from N to N1 of the modified adenine base and by the significant charge delocalization within the group. This planar structure is used to increase stacking of the ms²t⁶A₃₇ with A₃₈, its 3' neighbor (Fig. 5a). Finally, the threonyl moiety is too bulky to be incorporated into the helical stack (Supplementary Fig. S7), so it serves to sterically position the pseudo-three cycle base in a very specific location in order to make a cross-strand stack which bridges from the ASL's 3' stack to the codon itself. This cross-strand stack greatly enhances the stability of the codon-anticodon complex.

Figure 5.

Crystallographic structures of hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ decoding cognate and wobble codons, AAA and AAG, respectively. (a)The ms²t⁶A₃₇ stacks strongly with A₃₈, making use of the N⁶-threonylcarbamoyl moiety to increase area of interaction. (b) The ms²t⁶A₃₇ does not stack with U₃₆ (both with carbons colored slate), but instead forms a cross-strand stack involving its 2-methylthio modification with A1 of the codon (carbons colored cyan), which is hydrogen bonded to U₃₆ (black dotted lines). (c) The positioning of the t⁶ and ms²-groups within the structures of the hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ (carbons colored slate) and the ASL^Lys3_UUU-t⁶A₃₇ (carbons colored grey). The ribosomal crystal structures bound to the AAA codon are superimposed using the backbones as a register. The modified A₃₇ residues including the t⁶-moieties (carbons colored pink) superimpose with the ms²-moiety clearly directed away from the ASL.

The 2-methylthio (ms²) modification of A₃₇ functions to enhance the stacking interactions with A1 of the codon, as predicted by a fortuitous crystal packing interaction in the crystal structure of the tRNA alone.⁵⁴ The sulfur of the ms² modification is positioned to stack proximally to the ring of the codon base A1. It has been shown that placement of a stacking sulfur such that it can interact with an N1 of the stacking partner maximizes the strength of the interaction,⁵⁵ and in the current case, it also stacks well with an S₂-N characteristic, albeit with the N7 rather than the N1 of the codon base A1 (Fig. 5b). This provides strong evidence that the sulfur of the ms² modification acts primarily to enhance the stacking of ms²t⁶A₃₇ with the first base of the codon. The two modifications act in concert to optimize the structure of the anticodon loop in an open conformation prior to codon binding, and to maximize the strength of the codon-anticodon interaction by forming a sulfur-enhanced cross-strand stack by the precisely-positioned pseudo-three cycle base. In the absence of the ms², the t⁶ takes the same position in the 3'-stack of the ASL as it does in its presence. This conserved positioning of the t⁶ is clearly seen in a superimposition of the crystal of what would be the bacterial ASL^Lys_UUU-t⁶A₃₇ and ASL^Lys_UUU-mnm⁵U₃₄;t⁶A₃₇²⁵ with that of the hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ (Fig. 5c).

Although this is the first structure of the base mcm⁵s²U₃₄ bound to its cognate and wobble codon bases A and G, respectively, there are studies that have provided important evidence as to how the modifications function. It has been shown that the 5-methoxycarbonylmethyl (mcm⁵) modification promotes the binding of G-ending codons, but not A-ending codons. In contrast, the 2-thio (s²) modification has been shown to promote binding to both codons,³⁹ but particularly to A.^7,56The s², and not the 5-position modification, produces the significantly more stable, anti C3'-endo, g⁺ conformation, and transfers a similar restraint to the 3'-neighbor.⁵⁷In studies of ASLs, it has been demonstrated that mcm⁵U₃₄ has little measurable effect on structure, but that s²U₃₄ increases the amount of stacking between U₃₄ and U₃₅.⁴¹ This is evident when comparing the anticodon structure of the A-site hASL^Lys3_UUU-mnm⁵U₃₄;t⁶A₃₇ with that of the hASL^Lys3_UUU-mcm⁵s²U₃₄,ms²t⁶A₃₇ψ₃₉ (Fig. 6). The mcm⁵s²U₃₄ is more strongly stacked with s² on the nucleobase of U₃₅ than is the mnm⁵U₃₄. In fact, s² was shown to stack directly on the aromatic ring of U₃₅ in a crystal structure of the tRNA alone.⁵⁴ The s² modification is used to maximize stacking interactions with U₃₅ (Fig. 6), stacking close to N1 of the ring, the optimal position for maximal stacking interactions.⁵⁵ The current crystal structures explain all of these observations and offer an integrated explanation for the functioning of these two modifications.

Figure 6.

The anticodon structure of hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ in the ribosomal A-site compared to that of the ASL^Lys3_UUU-mnm⁵U₃₄;t⁶A₃₇. The crystallographic structures of the hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ (blue) and that of the ASL^Lys3_UUU-mnm⁵U₃₄;t⁶A₃₇ (grey)in the ribosomal A-site are superimposed to reveal differences in the anticodon conformations. The s²-moiety of mcm⁵s²U₃₄ stacks on the nucleobase of U₃₅. The positioning of the methyl groups of the mcm⁵ and mnm⁵ do not correspond due to the weak electron density at the extremities of the side chains.

The structure of the mcm⁵s²U₃₄•A₃ base pair is in the conventional Watson-Crick geometry, as expected (Fig. 7a). The mcm⁵ modification does not appear to play a significant role in decoding AAA. However, the s² modification is unambiguously participating in a stacking interaction with U₃₅, as anticipated from the tRNA crystal structure. This rationalizes the results for binding to A in previous studies,^6,7,38,56,58 where a position 34 modification was shown to display a higher binding affinity to A but not G; whereas, s² enhanced the binding affinity to both A and G.

Figure 7.

Cognate and wobble codon recognition by hASL^Lys3_UUU employs keto-enoltautomerism. (a) hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ bound to AAA in the ribosomal A-site. The mcm⁵s²U₃₄ (carbonsslate) base pairs with A₃ of the codon in the canonical Watson-Crick geometry. (b)The hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ (carbons blue) recognition of the wobble codon AAG. When paired with G3 (cyan), the Watson-Crick geometry of a C•G base pair is maintained for this mcm⁵s²U₃₄•G3 pair. This requiresthe tautomericenol form of mcm⁵s²U₃₄. The electron density of the 2f_o-f_c maps, shown at 1.5σ, clearly indicates the positioning of the modifications.(c) Comparison of the hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ and ASL^Lys_UUU-mnm⁵U₃₄;⁶A₃₇ recognitions of the wobble codon AAG. The mnm⁵U₃₄ (carbons grey) in the mnm⁵U₃₄•G3 base pair²⁵ is superimposed on the mcm⁵s²U₃₄ (carbons blue) of the mcm⁵s²U₃₄•G3 (cyan) using the G3 as a reference. The mcm⁵-moiety alters the electronic structure of the ring producing a shift toward the enol tautomer. The positioning of the mcm⁵s²U₃₄ can be attributed to the stereochemical restrictions imposed by the larger sulfur atom at position 2 of the uridine as compared to that of the oxygen. For lack of sufficient electron density to accurately place the extremes of the 5-position moieties, the differing geometries of the methoxy- of mcm⁵s²U₃₄ and the aminomethyl of mnm⁵U₃₄ are not considered significant.

The structure of the mcm⁵s²U₃₄•G3 base pair is again in the Watson-Crick geometry, but this is more unusual because to pair with G3 through hydrogen bonding in this geometry, mcm⁵s²U₃₄ must shift from its keto to an enol form (Fig. 7b). The keto to enoltautomerism has been observed in the case of uridine 5-oxyacetic acid, cmo⁵U, in tRNA^Val, binding to G.⁵²Therefore, it is not surprising for this base pair configuration to occur for mcm⁵s²U. However, atautomerismof mnm⁵U₃₄ has not been observed in its binding of G3.²⁵ The geometries of the mcm⁵s²U₃₄ and mnm⁵U₃₄ base paring to G3 are significantly different with the size of the s²-group a contributing factor to the positioning of the thiolated base (Fig. 7c). Whereas the thiol positions the mcm⁵s²U₃₄, the main function of the mcm⁵ modification is to modify the electronic structure of the ring. This shift of the keto-enol equilibrium far enough towards enol enables hydrogen bonding of the mcm⁵s²U₃₄•G3 base pair.

DISCUSSION

Modifications structure and thermodynamically restrain the anticodon loop

The pre-structured and constrained tRNA anticodon is energetically advantageous for a cell's protein synthesis. Every anticodon (other than that of initiator tRNA) must conform to the constraints of the ribosome's A-site, and yet present a physicochemical face that repeatedly reads only the appropriate codons. Repeated use of a conformed and dynamically restricted anticodon negates the need for remodeling the RNA each time an aminoacyl-tRNA is used for the decoding of its synonymous codons. Thus, pre-structuring of the anticodon enhances the accuracy and efficiency of protein synthesis. Though this is true of a great many tRNAs, the tRNA^Lys_UUU species is of particular interest for two reasons. First, approximately half of all bacteria have one lysine tRNAisoacceptor, the tRNA^Lys_UUU that has a 5-methylaminomethyl-2-thiouridine modification at wobble position 34 and a N⁶-threonylcarbamoyladenosine, which allows it to respond to both lysine codons AAA and AAG.²³ In contrast, mammalian cells contain three lysine isoacceptingtRNA species for cytoplasmic protein synthesis. The mammalian tRNA^Lys1_CUU and tRNA^Lys2_CUU constitute the majority of lysine tRNAs in the cell and respond only to AAG, the prominent lysine codon. The tRNA^Lys3_UUU is the least abundant of the three isoacceptors in the cell and responds to AAA and AAG. Second, the host cell tRNA^Lys3_UUU is selected by HIV as the primer for the initiation of reverse transcription as predicted from the viral RNA sequence.^59–61

The individual modificationsof mnm⁵U₃₄ at the wobble position and t⁶A at position-37 enhanced the cognate codon binding of the unmodified ASL^Lys3_UUUby approximately seven-fold; however, the modifications in combination increased codon binding to a level equivalent to that of endogenous E. colitRNA^Lys_UUU.⁸ The combination of t⁶A₃₇ and mnm⁵U₃₄ or s²U₃₄ also allowed efficient wobble codon recognition, but the s²-modification was required for proper translocation to the P-site.^6,8,25,56The 2-thiolated uridine at the wobble position of many tRNAs, as well as that of htRNA^Lys3_UUU, causes a slight enhancement of ribosome binding to A over G at the third position of the codon.^6,38,58 Molecular dynamics simulations of the xm⁵s²U mononucleoside bound to A or G also showed a marked preference for binding to A.⁶² Previous NMR experiments investigating xm⁵s²U as a mononucleoside and in short polymers showed a strong steric preference for the modified uridine to adopt an anti, C3'-endo sugar pucker conformation, effecting the same in 3'-neighboring nucleosides. Thus, the thio-moiety causes a more rigid binding geometry that could account for selection against pyrimidines at the third position of the codon.^57,58,63,64

Molecular dynamics simulations also confirmed that the ms²t⁶A₃₇ reduced thermal stability via a stereochemical effect that widened the loop and prevented the A₃₇•U₃₃ cross-loop interaction.⁴⁰ In contrast, structural models of the unmodified ASLs of tRNA^Lys_UUU and those singly modified with ψ₃₉ that had been generated using a subset of NMR distance restraints indicated a thermally stable, closed UUU triloop structure.⁴¹ Consistent with previous in silico models and empirical studies, the addition of ms²t⁶A₃₇ to models of the singly modified ASL^Lys_UUU-ψ₃₉ also enhanced the stacking interaction of A₃₇ with U₃₆ and created a more defined positioning of U₃₆.⁴¹ According to the models, neither the mcm⁵- nor s²-moieties of U₃₄ were sufficient to promote a canonical U-turn conformation. However, the singly modified ASL with mcm⁵s²U₃₄ showed many U-turn NMR indicators.⁴¹ These models suggested a role for the ms²t⁶A₃₇ in negating intraloop interactions, ordering of the anticodon loop conformation and stabilizing the weak U₃₆•A1 pair while mcm⁵s²U₃₄ enhanced anticodon base stacking and promoted the canonical U-turn conformation.

Empirical and in silico structural studies conclude that these modifications enhance codon recognition by acting to remodel the highly constrained and unfavorable conformation of the unmodified ASL toward a more canonical shape.^{6,26,41,53,65,66} Although these studies have shown that the modifications push the equilibrium toward a U-turn structure, it has been unclear as to whether the solution conformation of the fully modified ASL would be sufficiently pre-structured to explain the high level of endogenous codon specificity and fidelity. Here we conclusively show that the fully modified solution structure conforms to a structure that shows only minor differences from the stably bound crystal structure in the ribosomal A-site. In fact, our data also supports the notion that the pre-structuring has underlying thermodynamic causes and allows for a novel keto-enol mechanism of exclusively recognizing purines at the third codon position.

The thermodynamic causes of modification-induced pre-structuring are evident in the thermal melting and circular dichroism spectra. The negative 330 nm and positive 330 nm signals observed in the CD spectrum of the modified ASLs were due to the presence of the 2-thio modification of uridine as previously observed,⁴³ confirming the presence of this modification in the doubly and triply modified ASLs. The slightly higher value of the positive ellipticityat 264 nm found in the case of the hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ indicated that the bases were involved in stacking interactions that exceeded those of both the unmodified hASL^Lys3_UUU and the doubly modified hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇ (Fig. 3d). These results are not surprising as the introduction of mcm⁵s²U₃₄ and ms²t⁶A₃₇ negated intra-loop hydrogen bonding, reducing the melting temperature and enthalpy of base stacking.Indeed it has been shown that modifications most likely affect codon bias by contributing significantly to the thermodynamic stability and structural characteristics of htRNA^Lys3_UUU. Modifications of the Watson-Crick face of the invariant purine 37 decrease the thermal stability of the ASL by negating intra-loop hydrogen bonds and by widening the loop, as was initially found for the m¹G₃₇ of yeast tRNA^Phe.^11,65,67 The modification of A₃₇ to t⁶A₃₇ and to ms²t⁶A₃₇ in an otherwise unmodified anticodon stem and loop domain of htRNA^Lys3_UUU resulted in decreases in thermal stability of 2–3 and 5 °C, respectively.^53,68{Bajji, 2002 #465}However, by negating the intra-loop hydrogen bonding, the base stacking at the base of the stem is alleviated, which resulted in only minor differences in ellipticity between the constructs. ψ₃₉ is known to stabilize tRNA's anticodon domain,^66,69 but does not appear to effect codon binding.⁶⁹The presence of pseudouridine at position-39 enhanced the interaction with A₃₁ that constitutes the closing base pair of the stem^40,41,66 and stabilized the canonical U-turn conformation,⁴¹ thereby raising the thermal stability of the ASL and increasing the base stacking of the anticodon residues.

The variations in ¹H NMR chemical shifts observed when comparing the unmodified and the modified ASLs (Supplementary Table S2) suggested that the spatial electronic environments of the aromatic atoms of mcm⁵s²U₃₄, U₃₆, A₃₈ and U₃₉ were affected by the introduction of the modified residues within the ASL. The invariant chemical shifts were mostly seen for the residues of the stem attesting that the presence of mcm⁵s²U₃₄ and ms²t⁶A₃₇ in the loop did not structurally influence the helical A-RNA conformation of the stem. The greater distance observed between ms²t⁶A₃₇ and U₃₆ could have been induced by the large atomic radius of the sulfur group trying to accommodate itself between the above residues. The short distances found between the labile proton of the acid group of ms²t⁶A₃₇ and mcm⁵s²U₃₄/U₃₅O4 and between A₃₆N6 and C₃₂O2 invoked the potential hydrogen bonding that restrained the threonyl group of ms²t⁶A₃₇ to be positioned across the loop region. As a result the loop extension was limited. The NOE observed between C₁₀H₃ and s²-CH₃ indicated the close spatial proximity (4 – 5 Å) of these two groups. As a result, a clear orientation of these methyl groups was obtained during the solution structure determination of the doubly modified ASL^Lys3_UUU. The lack of a complete set of the signature NMR resonances for a U-turn indicated that the loop of the modified hASL^Lys3_UUU did not have a canonical U-turn, but nonetheless exhibited a backbone turn at U₃₃ in the NMR-restrained molecular dynamics derived structure.

Here, we have shown that the extensive post-transcriptional modifications of the htRNA^Lys3_UUU pre-structure the anticodon domain for codon binding, and alter the chemistry of the wobble position uridine to engage in a Watson-Crick base pair with the G3 of the AAG codon. Previously, we had determined and compared the codon binding properties, NMR-derived solution structures and crystallographic structures of codon binding of the ASL^Lys_UUU with and without the bacterial modifications, mnm⁵- and t⁶A₃₇.^8,53 Now, we can compare the structures of the hASL^Lys3-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ presented here with a previous structure in solution,⁵³ and to that of the crystal structure of the entire htRNA^Lys3_UUU.⁵⁴

Comparison of the solution structures

A structural comparison of the solution structure of hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇ to that of the singly modified hASL^Lys_UUU-t⁶A₃₇⁵³ reveals a significant, modification dependent rearrangement of the loop region (Fig. 8a).However, the nearly superimposable nature of the stems renders the overall rmsd quite low (Supplementary Table S3b).At the base of the stem, the possibility for the classically bifurcated cross-loop hydrogen bond is slightly perturbed from that of the singly modified structure (Fig. 8b). In the doubly modified hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇, the distance between the bases are increased from 3.0 Å in the singly modified structure to 3.4 Å. The stacking interaction between ms²t⁶A₃₇ and A₃₈ in hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇is in a more favorable geometry, presumably stabilizing A₃₈ into a less favorable cross-loop interaction, resulting in a more open and more canonical loop structure (Fig. 8b).

Figure 8.

Comparison of the NMR-derived solution structures.(a)A comparison of the NMR structure of the singly modified hASL^Lys3_UUU-t₆A₃₇ (orange)⁵³to that of the doubly modified hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇ (green) shows the high degree of deviation in the backbone in the anticodon loop. (b) A zoomed view of the potential bifurcated hydrogen bonding between C₃₂ and A₃₈ indicates that the prospective of stacking with the heterotricyclic ring system of A₃₇ competes with the cross-loop interaction, widening the distance between these nucleosides.

Although the anticodon of doubly modified hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇ in solution did not contain a canonical U-turn conformation, the hypermodified nucleosides shift the structure into a structure that is more indicative of a U-turn than the singly modified hASL^Lys3_UUU-t⁶A₃₇.⁵³ The two solution structures differ dramatically in the anticodon (Fig. 8a). While the anticodon nucleosides in the singly modified construct are mostly unstacked, in hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇ the anticodon bases are clearly ordered into a well stacked conformation and exposed to the solvent for proper codon recognition. The anticodon stacking, abrupt backbone turn at U₃₃ and the nicely ordered and stacked residues along the 3' side of the loop indicate that the doubly modified ASL has structural characteristics suggestive of a canonical conformation that would be significantly more favorable to ribosomal A-site binding.

Comparison of the crystal structures

A comparison of the crystal structures of the full length, fully modified tRNA^Lys3_UUU⁵⁴ to the co-crystal structures bound to synonymous codons in the ribosomal A-site reveals minor differences. This comparison strongly supports the hypothesis that modifications pre-structure tRNA`s anticodon domain for codon binding.¹The crystal structures reported here, bound to AAA/G, are nearly superimposable, with a heavy atom rmsd of 0.45 Å (Supplementary Fig. S8,and Table S3a). The most notable difference between these two structures and the anticodon domain of the full length, fully modified tRNA^Lys3_UUU occurs at the base of the stem. In the crystal structure of tRNA^Lys3_UUU, A₃₁ forms a planar, stable base pair with U₃₉,⁵⁴ but in the co-crystal structures on the ribosome, the A₃₈ residue is rotated out of plane indicating a less stable base pair and possibly contributing to the opening of the loop. Interestingly, in the solution structure we observe that the ms²t⁶A₃₇ causes a slight perturbation of the C₃₂•A₃₈ mismatch base pair by enhancing the stacking interaction of A₃₈ with A₃₇ and causing a displacement of ~0.4 Å. These data suggest that both ψ₃₉ andms t⁶A₃₇ cooperatively extend the distance between C₃₂ and A₃₈ to abrogate the mismatch pair, forcing the loop into the open conformation necessary for a stable U-turn. Indeed, the distance and geometry between C 32 and A₃₈ in all three crystal structures would not be amenable to a stable cross-loop interaction.

Comparison of the Solution and crystal structures

As expected from the Watson-Crick geometry of the U₃₄•G₃ wobble base pair, the crystal structures of the ASL bound to AAA and AAG codons were nearly identical (Supplementary Fig. S8). However, a superimposition of the NMR-derived solution structure of the hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇ with that of the crystal structure of the ASL region of the entire tRNA⁵⁴ and the co-crystal structure of hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ bound to the AAG codon indicates some salient differences (Fig. 9a, Supplementary Table S3c). The crystal structures show a very well defined U-turn conformation in which there is an abrupt change in direction at U₃₃.The anticodon bases are ideally stacked with their Watson-Crick faces solvent exposed and the bases on the 3' side of the loop form a very canonical stacking pattern (Supplementary Fig. S8). In contrast, in the solution structure, the common features of the U-turn are not evident and the structure is clearly dynamic (Fig. 9a). This feature is also evident in the thermal denaturation, where the doubly modified construct has a much lower hyperchromicity and enthalpy change than the triply modified hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ (Table 1). These are both measures of the relative order of a molecule and indicate that these structural differences are empirical and not artifacts.

Figure 9.

Comparison of the crystal structure bound to the A-site AAG codon to the NMR-derived solution structure of the hASL^Lys3_UUU-mcm⁵s²U₃₄,ms²t⁶A₃₇. (a) The NMR-derived solution structure (green) has an rmsd from the AAA bound crystal structure of 1.91 Å (not shown) and the AAG bound crystal structure of 1.81 Å (blue). (b) The C₃₂•A₃₈ cross-loop interaction shows the difference between the NMR and crystal structures that is induced by the weakened base pairing between the stem residues A₃₁ and ψ₃₉.

CONCLUSION

A lack of significant differences between the ASL of the full length tRNAcrystal structure and the co-crystals of the ASLs bound to the ribosome reveal that the functional binding results associated with using the truncated anticodon domain are a reasonable measure of the decoding capacity of the full length tRNA. It should also be noted that although an NMR-derived solution structure of the fully modified ASL was not obtained, it can be assumed that the addition of the ψ at position 39 would further drive the formation of a canonical U-turn in solution.^66,69 An induced fit model of decoding, in which the ASL would conform to a U-turn upon entering the ribosomal A-site is not consistent with the data due to the clear evidence from the crystal structure of the full length tRNA of a pre-structured, U-turn containing anticodon domain prior to entering the A-site.⁵⁴ Curiously, the high degree of structural conservation (rmsd = 0.65 − 0.77 Å, Supplementary Table S3a–c) between the crystal and the solution structures of the hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇;ψ₃₉ bound in the A-site suggests that in comparison, the anticodon domains of the partially modified E. coli ASLs incurred a degree of conformational change during codon recognition.²⁵

The AAG codon binding of the hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇ exhibits an mcm⁵s²U₃₄•G3 pairing that is a Watson-Crick pair with three hydrogen bonds, one of which involves the s²-modication. The hydrogen bond between the iminonitrogens of mcm⁵s²U₃₄ and G3 of the codon requires a keto-enoltautomerism similar to the mechanism by which cmo⁵U₃₄ enables ASL^Val3 to respond to all four codons in its degenerate box.⁵² While the current mcm⁵s²U₃₄•G3 base pair structure agrees well with that of cmo⁵U₃₄•G3.⁵² it stands in stark contrast to the structure of mnm⁵U₃₄•G3,²⁵ in which the modified U makes a three-centered hydrogen bond to G3 and partially stacks on U₃₅ (Fig. 7c). The equilibrium shift from the keto to enol tautomer inherent to these two modified nucleosides may suggest either a role of the cmo⁵ or the s² and mcm⁵ modifications in altering the electronic structure of the aromatic base or a local pH change in the environment in the immediate vicinity of the wobble base pair. It is critical to note that the mnm⁵U₃₄ is missing the naturally-occurring 2-thio modification, so its stacking potential and electronic structure is not that which occurs in nature. The difference in binding modes could be due to the fact that the previous work is a different modification that does not affect the keto-enol equilibrium and thus does not allow a Watson-Crick base pair as for cmo⁵U₃₄ and mcm⁵s²U₃₄ and that the observed structure is what will occur for the fully-modified ASL. The partially modified ASL^Lys_UUU-mnm⁵U₃₄ binds the cognate AAA, but not the wobble codon AAG, while the unmodified E. coliASL^Lys_UUU binds neither codon.⁸ We find it more likely that the previous, partially modified structure behaves differently from the fully modified mnm⁵s²U₃₄•G3, and that the fully modified structure would be very similar to the current structure. The thio modifications have been shown to be used to stack with adjacent rings for modifications at position-34 and 37, and were mnm⁵U₃₄ to be fully thiolated, stacking on U₃₅ would require moving into Watson-Crick geometry. This is an excellent illustration of how finely tuned codon-anticodon interactions are, with the lack of one sulfur atom and presence of an oxygen disrupting the true structure of the codon-anticodon complex.⁶

Differences in modification between the bacterial and eukaryotic tRNA^Lysperhaps indicatethe co-evolution of anticodon modifications with codon bias and the use of less common tRNAisoacceptorsfor control of gene expression. Half of all bacteria have but the one tRNA^Lys with the UUU anticodon that reads the predominant codon AAA. The same tRNAreads the rarer codon AAG. In contrast, the additional isoacceptors tRNA^Lys1&2 in mammalian cells have the anticodon CUU and respond only to AAG, the preferred lysine codon by a factor of two. The amount of tRNA^Lys3 is only a third of the total amount of tRNA^Lysrequired for protein synthesis in mammalian cells. This leads one to the concept of codon bias being linked to tRNAisoacceptor abundance in the cell.¹⁵tRNA abundance, in turn is also correlated to environmental responses.¹⁵By pre-structuring the nucleoside geometry and backbone conformation, the tRNA^Lys3 modifications certainly alter the presentation of anticodon nucleosides for codon recognition. tRNA^Lys1,2 do not require t⁶A₃₇ for binding of the AAG codon,⁸ but the anticodon modification t⁶A₃₇ that is common to all isoacceptors appears to be a recognition determinant for the lysyl-tRNAsynthetases. The difference in modifications between tRNA^Lys3 and tRNA^Lys1,2 implicate tRNA^Lys3 modifications in regulatinggene expression inmammalian cells through codon usage. Since lysine codons share a mixed codon box with the two asparagine codons, the major difference between isoacceptorsis located at the wobble position. Thus, the chemistries, structures and physical volumes of post-transcriptional modifications alter tRNA's anticodon conformation and restrict its dynamics for protein synthesis as had been proposed twenty years previously in a modified wobble hypothesis.¹

METHODS AND MATERIALS

RNA sample preparation

In the solid-phase syntheses of the hASL^Lys3_UUU-mcm⁵s²U₃₄;ms²t⁶A₃₇ with and without ψ₃₉, the 5′-O-dimethoxytrityl- and 2′-O-t-butyldimethylsilyl-protecting groups were used for both modified and unmodified phosphoramidite monomers. The exocyclic amines of A, G and C were protected with the 4-tert-butylphenoxyacetyl (tac) group. The mcm⁵s²U phosphoramidite had no additional protecting groups. The ms²t⁶A phosphoramidite had the carboxyl group of the threonyl residue protected as a 4-nitrophenylethyl ester and the secondary hydroxyl group of the threonyl residue was protected as a tert-butyldimethylsilyl (TBDMS) ether.

Synthesis was performed on rC(tac) CPG (27 μmole and 31 μmole scale for non-ψ₃₉ and ψ₃₉ variants) and coupling was performed with 0.1 M monomer solutions and 0.3 M BTT in acetonitrile for 6 minutes (min) with U and C monomers, 7 min for A and G monomers, and for 10 min with the mcm⁵s²U and ms²t⁶A monomers. For standard RNA monomers 5 molar equivalents were used per coupling, but for the modified monomers only 3 equivalents were used. The capping step was performed for 2 min with tac anhydride. To prevent desulfurization, the oxidations were performed with 1 M cumenehydroperoxide in toluene for 3 min.

After assembly, the 5'-O-trityl group was left on. The dried support was treated with 10% DBU in THF under argon at 40°C to remove the cyanoethyl and 4-nitrophenylethyl protecting groups. The CPG was then washed twice with dry THF and treated with 10% DBU in methanol under argon for 18 hours at room temperature. The supernatant was removed and stored under argon. After washing with dry methanol, the washings and supernatant were combined and dried in a Speedvac. The residue was further dried under high vacuum (0.01–0.001 Torr) over separate containers of phosphorus pentoxide and potassium hydroxide pellets for 48 hours to remove residual DBU. The residue was then dissolved in triethyleneaminetri(hydrofluoride) under argon to cleave the TBDMS protecting groups and the DMTr from the 5'-terminal G. After quenching with water, the RNA was precipitated with 1-butanol and dried under vacuum. Purification was by preparative anion-exchange HPLC on a Source 15Q 10/10 column using a gradient of sodium chloride in a phosphate buffer containing 50 μM EDTA and 8 M urea. Desalting was performed on a Sephadex G25 column. Nucleoside composition analysis was conducted by HPLC with on-the-fly UV diode array spectroscopy.⁷⁰

The ASLs were further purified by extensive dialysis (using Slide-A-Lyzer MINI Dialysis Units, 3.5K MWCO from Pierce) against 20 mM phosphate buffer (pH 6.8), 50 mM Na⁺, and 50 mM K⁺. The oligonucleotides were heated to 80 °C followed by slow cooling to form a solution homogeneous in ASL conformation. After cooling, the samples were lyophilized using a freeze-dryer (Thermo Savant SPD Speed Vac, Thermo Scientific) and then dissolved in 99.996% ²H₂O or a 90% ¹H₂O+10% ²H₂O mixture to give a final volume of 300 μL. Samples in ²H₂O were re-dissolved in ²H₂O at least twice more and lyophilized. NMR samples of the ASLs were generally at concentrations of 1.5 mM.

Ribosomal binding assay

The ribosomal binding assays consisted of reaction mixtures of purified E. coli (MRE 600) 70S ribosomes⁴²and chemically synthesized mRNAs and ASLs. The 27-nt mRNA oligonucleotide sequences were derived from T4 gp32 mRNA⁷¹ and were purchased from Dharmacon (ThermoFisher, Lafayette, CO). In order to study binding at the ribosomal A-site, the P-site needed to be blocked. Therefore, the mRNA was designed with the Methionine (Met) codon AUG at the P-site. The E. colitRNA^fMet was then used to saturate the P-site. Possible secondary structure folding of each mRNA sequence was assessed with the program RNA Structure 4.2⁷² and resulted in a low probability of any secondary structure that may obstruct binding of the mRNA to the ribosome. The mRNA sequences are as follows (Lys codons AAA and AAG are in bold and underlined):

• 1)5'-GGCAAGGAGGUAAAAAUGAAAGCACGU-3';
• 2)5'-GGCAAGGAGGUAAAAAUGAAGGCACGU-3'.

The 70S ribosomal subunits were isolated as previously described.⁴² The ASLs were 5'-end ³²P-labeled using [γ-³²P] ATP (MP Biomedicals). Unlabeled ASLs in a range of concentrations (0–2.5μM) were mixed with insignificant, but radiochemically detectable amounts (2,000–5,000 CPM) of 5'-end, ³²P-labeled ASLs in a fixed ratio of unlabeled ASL to labeled ASL. The assay was performed in ribosomal binding buffer [50 mM HEPES, pH 7.0; 30 mMKCl; 70 mM NH4Cl; 1 mM DTT; 100 μM EDTA; 20 mM MgCl₂). Ribosomes (0.25μM) were activated by heating to 42°C, incubating for 10 min and then slowly cooled to 37°C. The ribosomes were programmed with 2.5 μM mRNA for 15 min at 37°C. The P-site was saturated with 1.25 μME. colitRNA^fMet (Sigma-Aldrich), which binds to AUG, for 15 min at 37°C; see underlined codons of the mRNA sequences above. Binding of ASL^Lys3_UUU to the A-site was allowed to proceed for 30 min at 37°C. The reaction mixtures (20μL each) were then placed on ice for 20 min, diluted with 100 μL buffer per reaction mixture, and filtered through nitrocellulose in a modified Whatman Schleicher and Schuell (Brentford, U.K.) 96-well filtration apparatus.⁷³ Prior to filtration of experimental samples, the nitrocellulose filter was equilibrated in binding buffer at 4°C for at least 20 min and each well of the filtration apparatus was washed with 100 μL of cold binding buffer. After filtration of reaction samples, each well was then washed twice with 100 μL of cold binding buffer. The nitrocellulose was blotted dry with Kimwipes (Kimberly-Clark), and the radioactivity was measured using a phosphorimager (Molecular Dynamics, GE Healthcare). Data were measured for radioactive intensity using ImageQuant (Amersham). Nonspecific binding was determined by the binding of ASLs to ribosomes without mRNA and subtracted from the experimental data. The final data is a result of at least two separate experiments, each done with samples in triplicate, i.e. at least six results for each binding point.

Thermal denaturation

The thermal stabilities of the ASLs were monitored using a Varian Cary 3 UV-visible spectrophotometer (Agilent) controlled with WinUV version 3.00 software. All samples were first adjusted to an absorbance of ~0.2 OD units at 260 nm in 20 mMsodium/potassium phosphate buffer (10 mM Na₂HPO₄ and 10 mM KH₂PO₄, pH 6.8). Absorbance at 260nm was measured over the temperature range of 5°C to 95°C. The temperature was ramped at 1°C/min and data was collected at a rate of 1 data point per minute, averaged over 2 seconds. Thermodynamic parameters and melting temperatures were calculated from the melt profiles using MeltWin v3.5. All data were baseline corrected using a control containing buffer only.

Circular dichroism

Circular dichroism spectra were collected on a Jasco J600 spectropolarimeter. All samples were adjusted to ~0.2 OD at 260nm in 20mM sodium/potassium phosphate buffer (10 mM Na₂HPO₄ and 10 mM KH₂PO₄, pH 6.8) prior to CD analysis and analyzed in a 1 cm path length quartz cuvette. Samples were temperature controlled to 25°C during data collection. All data were baseline corrected using a control containing buffer only.

NMR experiments

Bruker DRX500 and Varian Inova-600 instruments were used to conduct the experiments. NMR data were processed with either XWINNMR (Bruker Inc., Rheinstetten, Germany) or NMRPipe,⁷⁴ and the analysis was conducted with SPARKY.⁷⁵ WATERGATE⁷⁶ method was used to suppress the water signal resulting from the samples prepared in ¹H2O whereas a low-power presaturation technique was used for samples dissolved in ²H₂O. NOESY⁷⁷ experiments of the samples in ¹H2O were performed at low temperatures (2, 5, and 10 °C) to observe the exchangeable proton resonances. For NMR resonance assignments:¹H-¹H-COSY, ¹H-¹H-DQF-COSY, ¹H-¹H-TOCSY, natural abundance ¹H-¹³C HSQC, and ¹H-³¹P HETCOR experiments were performed.⁴⁹ To achieve structures determination, NOESY experiments were conducted in ²H₂O with different mixing times (50, 75, 100, 200, 300, and 400 ms) at 22 °C without removing the samples from the magnet. The spectra were acquired with spectral widths of 5000 Hz in both dimensions, 1024 points in t₂, 360 points with 64 scans per block in t₁, and a recycle delay of 1.5 s. The FIDs were processed with 60° phase-shifted sine bell apodization functions and third-degree polynomial baseline correction in both dimensions. To improve the digital resolution for the cross-peak integration, the FIDs were zerofilled to 2048 × 2048 points.

Structure Determination

Structure calculations were performed using CNS and were based on published protocols.^47,78,79The backbone and the ribose torsion angles were restrained to be within ±15° of the standard A-form values. NOESY mixing time studies also provided distance restraints between nonexchangeable protons in the loop region. The NOE cross-peaks were integrated by using the peak fitting Gaussian function and volume integration in SPARKY. The distances were determined and normalized to the non-overlapped pyrimidine H5–H6 cross-peaks with a distance of 2.44 Å. Upper and lower bonds were assigned to ± 20% of the calculated distances. The distances involving the unresolved protons, i.e., methyls of ms²t⁶A₃₇ and mcm⁵s²U₃₄ as well as its methylene group, were subjected to the pseudoatom correction automatically computed by CNS. Residues with non-observable H1′–H2′ cross-peaks on the DQF-COSY spectra or with ³J_{H1′–H2′} values of <3 Hz were restrained to the C3′-endo conformation, whereas in those with ³JH1′–H2′ values appearing to be between 4 and 5 Hz, the ribose was left unconstrained. The α and ζ torsion angles were restrained to exclude the trans conformation. The β, ε, and γangles were restrained on the basis of the ¹H-³¹P HETCOR spectra as previously described.^44,49 The structures were analyzed using 3DNA.⁸⁰

Crystallographic methods

Purification, crystallization and cryoprotection of Thermusthermophilus 30S ribosomal subunits was conducted exactly as has been described.⁸¹ The mRNA oligonucleotides were chemically synthesized (Dharmacon,ThermoFisher, Lafayette, CO) with the sequences 5'-AA(A/G)AAA-3' (codons underlined). After cryoprotection, the empty crystals were soaked in cryoprotection solution augmented with 80 μM paromomycin, 300 μM ASL and 300 μM mRNA oligonucleotide for at least 48 hours. Crystals were flash-cooled by plunging into liquid nitrogen and stored for data collection.

Crystals were pre-screened at European Synchrotron Radiation Facility (ESRF, Grenoble, France) beamline ID14-2 and data was collect at ESRF beamline ID14-4. Processing was performed using XDS,⁸² CCP4 (Collaborative Computational Project Number 4:1994) was used for format manipulation, COOT⁸³ for visualization and model building, CNS 1.2⁷⁹ for refinement, HIC-Up⁸⁴ for novel constraints generation, 3DNA⁸⁵ for RNA analysis and PyMOL⁸⁶ for figure production.

Abbreviations

ASL

anticodon stem and loop

hASL^Lys3_UUU

ASL of human lysine tRNAisoaccepting species 3 with anticodon UUU

CD

circular dichroismspectropolarimetry

mcm⁵s²U₃₄

5-methoxycarbonylmethyl-2-thiouridine at position 34

ms²t⁶A₃₇

2-methylthio-N⁶-threonylcarbamoyladenosine at position 37

HPLC

high performance liquid chromatography

NOE

nuclear Overhauser effect

Tm

temperature at the mid-point in the UV-monitored major thermal transition