Determining the effects of pseudouridine incorporation on human tRNAs

To investigate the structure, folding, and stability of human tRNAs, we produced 10 human tRNA sequences individually by T7 run-off in vitro transcription (IVT; Appendix Fig. S1A,B). We annealed and purified tRNAs (Fig. 1A) to obtain pure and homogenous samples displaying the expected molecular size and similar mobility in a native electrophoresis (Fig. 1B; Appendix Fig. S1C,D). Next, we measured the melting temperature (Tm) of each human tRNA sample by monitoring the increase in fluorescence intensity as RiboGreen^TM binds to tRNA regions that become single-stranded upon unfolding under a temperature gradient. While applying a temperature gradient, 8 out of 10 purified transcripts indeed folded into stable conformations that displayed Tm values well above human body temperature (Fig. 1C; Appendix Fig. S1E). tRNA^Pro_UGG and tRNA^Val_UAC display very low Tm values and their hydrodynamic radii are the largest compared to others, suggesting they did not fold properly, and they may rely on modifications for folding. We noticed that their hydrodynamic radii are still significantly smaller than the predicted size of a fully unfolded tRNA, indicating partial folding.

To gain a comprehensive understanding of the folding and unfolding mechanisms of a tRNA, we analyzed the behavior of the folded tRNA^Gln_UUG at different temperatures by monitoring changes in dynamic light scattering (DLS) and in silico simulation (Fig. EV1A,B). DLS of the folded tRNA^Gln_UUG shows a relatively small increase in the hydrodynamic radius over an increasing thermal gradient, suggesting the presence of remaining structural elements even at very high temperatures. Our molecular simulations show that the conserved canonical G₁₉-C₅₆ pair at the tip between D- and T-loops represents the molecular “Achilles heel” of tRNAs. Once this contact is lost, the remaining non-canonical interactions between D- and T-loops break easily, further destabilizing the characteristic tRNA tertiary structure (Fig. EV1C and Movie EV1). However, the individual stem structures of the tRNA arms remain to a large extent base-paired even at higher temperatures, suggesting that the observed “unfolding” transition during the thermal ramp is caused by the separation of D- and T-loops (Fig. EV1D,E and Movie EV2). This observation is consistent with chemical probing experiments using bacterial and yeast tRNAs (Yamagami et al, 2022; Wilkinson et al, 2005). We further conclude that GC content is a very weak criterion for predicting the stability of folded RNA domains (Fig. 1C) and the formation of functional tertiary RNA structures seems to instead depend on individual canonical and non-canonical base pairs.

To characterize the three-dimensional structures of individual human tRNAs, we vitrified purified tRNA^Asp_GUC, tRNA^Arg_UCU, tRNA^Gln_UUG, tRNA^Glu_UUC, tRNA^Gly_CCC, tRNA^His_GUG, tRNA^Lys_UUU, tRNA^Ser_UGA on cryo-EM grids. After grid screening and sample optimization, we collected larger datasets for human tRNA^Gln_UUG and tRNA^Gly_CCC. After applying basic image correction, particle picking, and classification procedures, we observed numerous 2D classes displaying tRNA-like particles with the expected diameter of approximately 90 Å for human tRNA^Gln_UUG and tRNA^Gly_CCC (Fig. EV2; Appendix Fig. S2). Iterative rounds of 3D classification yielded homogenous sets of particles, allowing us to reconstruct cryo-EM maps at nominal FSC_0.143 resolutions of 5.3 Å and 4.9 Å, respectively (Table 1). Conventional methods of resolution estimation are challenged by very small objects (Wu and Lander, 2020), but the obtained maps clearly resolve distinct structural features (e.g., RNA phosphate-backbone) that indicate resolutions better than 6 Å (Fig. 1D). Each of the structures displays a canonical L-shaped tRNA, wherein the individual domains are distinguishable. In both structures, the ASL region is well defined, and the major and minor grooves of the anticodon and acceptor stems are recognizable. In tRNA^Gln_UUG, the elbow region appears compact and the interaction between the D- (G₁₉) and T-loop (C₅₆) is visible. In tRNA^Gly_CCC, the CCA at the 3ʹ-end is visible, but the T-arm is less well resolved. Next, we used restrained flexible fitting implemented in SimRNA (Boniecki et al, 2015) to check whether sequence-based models can be reasonably fitted into the maps at the obtained resolution. For both tRNAs, we obtained ensembles of atomic models, which fit well into the cryo-EM maps (Fig. 1D). The top-scoring models for each tRNA are highly similar (Fig. EV3A) and slight variations appear only at the 3ʹ-end and in the anticodon loop. Hence, the combination of structural modeling together with spatial restraints from intermediate resolution cryo-EM maps is sufficient to obtain reliable models of human tRNA^Gln_UUG and tRNA^Gly_CCC. In summary, we present structural models of two unmodified human tRNAs by combining single-particle cryo-EM and map-assisted semi-supervised modeling, representing the smallest asymmetric particles (~25 kDa) analyzed by cryo-EM to date.

We next asked how the introduction of the most common tRNA modification, pseudouridine, influences the structures of tRNAs. Interestingly, the recently developed mRNA vaccines use total substitution of all uridines by Ψ or N1-methyl-Ψ (m¹Ψ) (Sahin et al, 2021) to enhance mRNA stability and reduce undesired immunogenic responses (Karikó et al, 2008). However, the mRNA molecules in the vaccines do not need to fold into a defined, stable tertiary structure. We expected that saturating tRNAs with these modifications would negatively affect their structures. To verify this, we synthesized tRNA^Gln_UUG by IVT with complete replacement of all 19 uridines with either Ψ or m¹Ψ. While tRNA^Gln_UUG was well-folded in its unmodified state (Tm 61.7 °C), its Tm decreased by more than 23 °C (U → Ψ; Tm 38.1 °C) and 29 °C (U → m1Ψ; Tm 32.1 °C) when fully substituted with either Ψ-variant (Fig. 2A). To address the initial question in a biologically meaningful way, we tested whether introducing Ψs exclusively at their five naturally occurring sites in tRNA^Gln_UUG (Cappannini et al, 2021) would stabilize it. We produced and purified wild-type or catalytic mutants (carrying an Asp-to-Ala (DA) substitution (Huang et al, 1998)) of all five human PUS enzymes expressed in bacteria (PUS1_30-427, TRUB1/PUS4_1-349, PUS7_98-661 and PUS10_1-529) or insect cells (PUS3_1-481) (Figs. 2B and EV4A). PUS1, PUS3, TRUB1/PUS4, PUS7, and PUS10 catalyze formation of Ψ_27/28, Ψ_38/39, Ψ₅₅, Ψ₁₃, and Ψ₅₄, respectively (Rintala-Dempsey and Kothe, 2017). All of the catalytic mutants bind to tRNAs with comparable affinity to the wild-type versions (Fig. EV4B). Using a N-cyclohexyl-N’-β-(4-methylmorpholinium)ethylcarbodiimide (CMC)-based primer extension assay, we show that the purified PUS enzymes catalyze highly specific pseudouridinylation of their expected target sites while leaving other uridines unmodified (Fig. 2C). Combining all five enzymes into one reaction allowed us to produce tRNA^Gln_UUG with all known naturally occurring Ψ-sites (Ψ₁₃, Ψ₂₈, Ψ₃₉, Ψ₅₄, and Ψ₅₅) (Cappannini et al, 2021) in vitro. We repurified the tRNA after the modification reaction and subjected it to a temperature gradient. In stark contrast to the complete Ψ-substitution, we observed an increased thermostability (Tm +3.4 °C) for tRNA^Gln_UUG carrying Ψ₁₃, Ψ₂₈, Ψ₃₉, Ψ₅₄, and Ψ₅₅ (Fig. 2A).

Next, we performed in vitro pseudouridylation reactions using individual PUS enzymes or various combinations thereof (Fig. 2C). We found that the activities of PUS1 and PUS3 on their own contribute very little to the overall increase in stability, whereas TRUB1/PUS4, PUS7, and PUS10-dependent Ψs are principally responsible for the elevated thermostability of modified tRNA^Gln_UUG. The Tm value is further increased when Ψ₁₃ (PUS7) is combined with Ψ₃₉ (PUS3; Tm +2.4 °C) or Ψ₅₅ (TRUB1/PUS4; Tm +2.6 °C), respectively. Incubation with a mixture of all inactive PUS enzymes shows no detectable Ψ modifications and no impact on the thermostability of tRNA^Gln_UUG (Fig. 2D), supporting the specificity of the stabilizing effect by the Ψ modification itself. Of note, we noticed that PUS10 can catalyze pseudouridinylation of both, U₅₄ and U₅₅. Pseudouridinylation at position 55 by PUS4 does not seem to affect the activity of PUS10 in vitro or vice versa (Fig. EV4C). Therefore, it is difficult to analyze the impact of Ψ₅₄ on tRNA stability individually and we focused on the analyses on Ψ₁₃, Ψ₂₈, Ψ₃₉, and Ψ₅₅ for all tested human tRNAs.

We tested whether tRNA^Glu_UUC, tRNA^Asp_GUC and tRNA^Gly_CCC are also stabilized by post-transcriptional pseudouridinylation. Those human tRNAs naturally carry fewer Ψ-sites, because they have other nucleotides (e.g., A, C, G) or uridine modifications (e.g., m⁵U) in the respective position. Our experiments with tRNA^Glu_UUC (naturally modified at Ψ₁₃, Ψ₅₄, and Ψ₅₅) and tRNA^Asp_GUC (naturally modified at Ψ₁₃ and Ψ₅₅) reveal similar stabilization patterns. The introduction of Ψ₁₃ and Ψ₅₅ in tRNA^Glu_UUC is most beneficial whereas the addition of Ψ₅₄, which is rare in eukaryotes (Roovers et al, 2021), does not contribute to the stability of tRNA^Glu_UUC by itself and even lowers the Tm value. tRNA^Asp_GUC only carries Ψ₁₃ and Ψ₅₅, which both improve the stability and again show a synergistic effect in the double-modified tRNA. We introduced Ψ₁₃, Ψ₃₉, and Ψ₅₅—the only three modifications to naturally occur in tRNA^Gly_CCC—which are located on each of its three domains (D-arm, ASL, and T-arm, respectively) (Fig. EV4D). The presence of all three Ψs in tRNA^Gly_CCC increases its Tm value by 5.2 °C (Fig. 2E). The presence of Ψ₁₃ or Ψ₅₅ influenced the Tm value more than the presence of Ψ₃₉ and combining Ψ₁₃ with Ψ₅₅ is as efficient as modifying all three sites. We confirmed that only the simultaneous modification of Ψ₁₃ with Ψ₅₅ on the same tRNA molecule leads to the observed synergistic effect, as mixing tRNAs carrying either Ψ₁₃ or Ψ₅₅ at an equimolar ratio, did not lead to an increase in Tm (Fig. 2F). Generating heterogenous mixtures of modified samples by titration leads to intermediate effects on the Tm. Time course experiments confirm the almost complete level of Ψ incorporation (Fig. 2G), which is in line with the known fast reaction kinetics of PUS enzymes in vitro (Purchal et al, 2022). Finally, we show that Ψs are highly resistant to heat and that the Tm changes can be observed even after several cycles of folding and unfolding (Fig. 2H). Our analyses suggest that incorporation of Ψ at positions 13 and 55 in the D- and T-arm, respectively, consistently increases the stability of in vitro transcribed human tRNAs. The incorporation of Ψ in alternate sites, namely Ψ₂₈ and Ψ₃₉, seems to be mostly dispensable for the stability of these four human tRNAs.

Our ability to resolve individual tRNA structures by cryo-EM for the first time gives us the opportunity to dissect the structural mechanism by which pseudouridinylation influences and stabilizes tRNAs. To this end, we determined the structures of human tRNA^Gln_UUG, tRNA^Gly_CCC, tRNA^Glu_UUC and tRNA^Asp_GUC before and after introducing Ψ₁₃ and Ψ₅₅ by single particle cryo-EM (Fig. 3A–D; Appendix Figs. S2–9). We obtained improved maps after modification, which enabled us to place consistent ensembles of models for all four human tRNAs with Q-scores expected at these resolutions (Pintilie et al, 2020) and small RMSD differences among the top-scoring models for each reconstruction (Figs. 3A–D and EV3A). Comparing modified tRNA^Gln_UUG and tRNA^Gly_CCC to their unmodified counterparts, we observed changes in the elbow region and more distinct densities for the phosphate groups along the RNA backbone of the ASL and the AAS, indicating reduced disorder in the presence of Ψ₁₃ and Ψ₅₅. In the case of tRNA^Glu_UUC and tRNA^Asp_GUC, the increase in rigidity provided by Ψ₁₃ and Ψ₅₅ modification led to an even more dramatic improvement of the cryo-EM densities: while the unmodified tRNAs produced featureless maps of L-shaped blobs at intermediate resolution above 8 Å, stabilization of the D- and T-arms allowed us to resolve the individual arms of both tRNAs. Moreover, we observed a featureless map towards recognizable tRNA features at the elbow region of tRNA^Asp_GUC upon modification (Fig. 3D). Of note, none of the observed conformational changes leads to a visible difference in migration behavior during native PAGE analyses (Fig. EV3B).

We also attempted to generate cryo-EM structures of these tRNAs with different patterns of pseudouridinylation. We obtained an interpretable map of tRNA^Gln_UUG after simultaneously introducing Ψ₁₃ and Ψ₃₉. In this map, no compaction of the elbow region is observed, highlighting the importance of Ψ₅₅ for the flipping of the nucleotide and formation of the elbow (Fig. EV4E). In summary, we conclude that pseudouridinylation, particularly at the Ψ₁₃ and Ψ₅₅ sites, improves the interaction between D- and T-arms. Pseudouridinylation in other tRNA positions (Ψ_27/28 and Ψ_38/39) only marginally confers additional stability to tRNAs, showing that Ψ acts highly position- and context-specific.

Having established that Ψ₁₃ and Ψ₅₅ are key determinants of tRNA stability, we next asked whether these findings would hold true in other tRNA species that lack one or more commonly modified uridines or do not typically carry these canonical modifications. To address this, we extended our biophysical analyses to human tRNA^His_GUG, tRNA^Lys_UUU, tRNA^Ser_UGA and tRNA^Arg_UCU (Fig. 4A; Appendix Fig. S10), all of which display alternative Ψ patterns in cells. We found that tRNA^His_GUG, which is naturally modified by PUS1 (Ψ₂₈), TRUB1/PUS4 (Ψ₅₅), and PUS7 (Ψ₁₃), is stabilized by Ψ₁₃ and Ψ₅₅, but in this case Ψ₂₈ does not contribute to additional stability. By contrast, tRNA^Lys_UUU, which contains three natural modification sites (Ψ₂₇, Ψ₃₉, and Ψ₅₅), is strongly stabilized by the presence of Ψ₂₇ and Ψ₃₉, while Ψ₅₅ seem to play a less prominent role. We speculate that the uridine-rich ASL loop may require additional stabilization with Ψ₂₇ and Ψ₃₉, two modified residues located within the ASL. tRNA^Ser_UGA lacks U₁₃ and exhibits only a moderate stabilization following the introduction of Ψ₅₅, suggesting that its longer variable region makes this tRNA less dependent on the presence of D- and T-arm modifications. Instead, the stability of tRNA^Ser_UGA seems to be dominated by the stabilization of the ASL by Ψ_27/28 and Ψ₃₉. Finally, tRNA^Arg_UCU shows only a minimal change in Tm (+1.2 °C) after introducing Ψ at positions 28, 39, or 55. Hence, it seems that the stability of tRNA^Arg_UCU is not dependent on the presence of Ψ, which raises the question of whether the modifications could be beneficial for this tRNA in some other way.

As the conversion from U to Ψ would not lead to altered features of the coulomb potential map of the nucleobase, we would not expect to visualize the modifications in our structural data even at high resolution. Nonetheless, we reasoned that we may be able to detect the resultant local structural consequences of pseudouridinylation in our cryo-EM maps. We calculated and compared local resolution maps of tRNA^Gln_UUG and tRNA^Gly_CCC before and after introducing different sets of modifications, namely Ψ₁₃/Ψ₃₉ or Ψ₁₃/Ψ₅₅ and Ψ₁₃/Ψ₅₅ or Ψ₁₃/Ψ₃₉/Ψ₅₅, respectively. As explored above, we find that pseudouridinylation impacts on local conformation and improves the local resolution in regions around the respective modification sites to a greater extent than more distal regions of the same tRNA (Fig. 4B; Appendix Figs. S11 and S12; Movie EV3 and EV4). More specifically, the presence of Ψ₃₉ reduces the disorder in the ASL of tRNA^Gln_UUG, whereas Ψ₅₅ induces the compaction of the elbow region in both analyzed tRNAs. For instance, in tRNA^Gln_UUG, the elbow region seems to get loosened when U₁₃ and U₃₉ are modified to Ψ. Upon U₅₅ modification, the resultant region gets compact again (like in the unmodified state), highlighting an example where Ψ provides a local structural switch on the modified residue itself. The conditional nature of U₅₅ accessibility shows that certain modifications may only be possible or functional in the presence of others (Sokolowski et al, 2018).

To gain further insight of how the incorporation of Ψ can lead to stabilization, we carried out molecular dynamics simulations of folded tRNA^Gln_UUG carrying no modifications, Ψ₁₃, Ψ₅₅ or Ψ₁₃/Ψ₅₅ (500 ns, three independent simulations for each of the four variants). We analyzed the interactions of all tRNA nucleotides during the simulation with a specific focus on the respective Us and Ψs and their close surrounding. In all modified variants, Ψ₁₃ and Ψ₅₅ residues formed additional water-mediated hydrogen bonds between the N1-H group and the phosphate oxygen of the preceding 5ʹ residue, and Ψ₁₃ additionally tended to form an additional hydrogen bond to its own phosphate. These interactions, missing for the U residues, lowered the mean Ψ₁₃ and Ψ₅₅ stacking energies by ~4.1 and ~1.8 kcal/mol, respectively, thereby improving tRNA stability (Fig. EV5).

To complement our structural analyses, we measured the Temperature-related Intensity Change (TrIC) of tRNA variants carrying site-specific labels over an increasing temperature gradient (Fig. 4C). For these experiments, we first synthesized full-length tRNA^Gln_UUG with a Cy5-probe positioned either at the 5′-end or close to the interface between D- and T-arm (U₄₈^Cy5). The Tm value of the 5′-labeled tRNA (without Ψ) agrees well with our earlier measurements using RiboGreen^TM (Fig. 2D). We then measured Tm values of the two fluorescently labeled versions of tRNA^Gln_UUG treated with either PUS7 to install Ψ₁₃ [which caused the largest Tm-increase when applied individually (Fig. 2D)], or with the mixture of five PUS enzymes to catalyze the introduction of all naturally-occurring Ψ sites. We observed that both 5′- and U₄₈-labeled tRNAs display a Ψ-induced thermostability change. The relative Tm shifts after Ψ incorporation are almost identical for RiboGreen^TM and the 5′-probe, whereas the relative Tm difference for the U₄₈^Cy5-labeled tRNA^Gln is significantly higher (Fig. 4C). Therefore, Ψ₁₃ indeed stabilizes the proximal nucleotides around the core to a higher extent than the residues in other elements of the tRNA. In summary, Ψ enhances tRNA structural rigidity locally, resulting in a site-specific stabilization effect that depends on its surrounding structural context (Voegele et al, 2023).

Our work up until this point has relied on in vitro transcribed tRNAs that provide an ideal pure and controllable substrate, but we next asked whether these methodologies could extend to endogenous tRNAs as a more biologically relevant system. However, it is hard to isolate homogeneous samples of a single tRNA iso-decoder because of the similar shape, sequence and charge of the cellular tRNA pool. We isolated thiolated human tRNA^Arg_UCU and tRNA^Lys_UUU from HEK293T cells using a modified chaplet column chromatography (CCC) method (Suzuki and Suzuki, 2007) and analyzed their structures by cryo-EM. We used thiolated tRNAs to take advantage of APM gels (Fig. 5A) that allowed us to confirm the high purity of our CCC preparations. Both samples yielded maps resembling the canonical tRNA size and shape that could be used to fit sequence-based models. The obtainable resolution (~10 Å) for the endogenous tRNAs was limited in comparison to the well-defined maps of the in vitro-derived tRNAs and the reasons could be manifold. Nonetheless, the modification patterns might also be heterogenous for the same iso-decoder tRNA, which can hamper high-resolution reconstructions. The incorporation of additional, larger modifications (e.g., mcm⁵s²U₃₄, t⁶A₃₇), might decrease the stability of a given tRNA again, leading to an increased entropy and flexibility of fully modified tRNAs (Fig. 5B).

Future studies will need to focus on solving these issues by optimizing purification protocols for each individual tRNA, by customizing sample preparation methods, by further improving cryo-EM data collection strategies and data analyses pipelines. In addition, our study focuses exclusively on Ψs and the used in vitro transcribed tRNAs lack other modifications. Hence, future studies need to include the full range of tRNA modification to understand how they contribute to folding and stability of tRNAs and how they affect the stabilizing effects of Ψs, determined in the study. Nonetheless, our proof-of-concept data paves the way toward structural characterization of endogenous tRNAs and other similar-sized structured RNA molecules from endogenous sources.

In vitro transcribed tRNAs were produced as previously described (Lin et al, 2022). PCR was used to generate the DNA templates with T7 polymerase recognition sequence at 5′ end. The tRNA sequences were retrieved from tRNA database (http://gtrnadb.ucsc.edu/) (Chan and Lowe, 2016) and the CCA-end was also introduced in the template sequences (Appendix Fig. S1A). To obtain sufficient material during T7-mediated IVT (Dunx et al, 1983), we introduced additional Gs at the 5ʹ-end of tRNA^Asp_GUC, tRNA^Glu_UUC, tRNA^Ser_UGA, and tRNA^Val_UAC (Appendix Fig. S1A). As it has been reported that the addition of 5’-nucleotides can impacted on the biochemical properties in vivo (Preston and Phizicky, 2010), we confirmed that the additional of 5ʹGs has no severe effect on the experimentally determined Tm values (Appendix Fig. S1D,E). For those low yield tRNA transcripts, we introduced at least 2Gs at the 5′ end to facilitate the production efficiency. Briefly, the DNA template of each tRNA was generated using the Primerize method (Tian et al, 2015) and the T7 promoter sequence was included for in vitro transcription driven by T7 RNA polymerase (Chramiec-Głąbik et al, 2023). An overnight T7‐driven transcription reaction was performed at 37 °C with the following: 40 mM Tris, pH 8.0, 5 mM DTT, 30 mM MgCl₂, 1 mM spermidine, 20 mM NTPs, linear DNA, RNasin (Thermo Fisher Scientific), T7 RNA polymerase (Thermo Fisher Scientific), and pyrophosphatase (Thermo Fisher Scientific). The Cy5-labeled CTP was added to the reaction to generate fluorescent RNA substrates (20% of CTP are Cy5-labeled which randomizes the labeled position and minimizes the labeling-related binding disruption). RNase‐free DNase I (Thermo Fisher Scientific) was added to digest the DNA template, and the reaction was stopped by the addition of EDTA and protease K. The reaction was subjected to a FPLC system using a DEAE weak anion exchange column (GE) and followed by temperature‐gradient‐based annealing. Refolding of the RNA was carried out by heating the RNA solution to 80 °C for 2 min and slowly cooling to room temperature. The annealed RNAs were further purified using a Superdex 75 Increase gel filtration column (GE) in the buffer (20 mM HEPES, pH 7.5, 1 mM MgCl₂, 150 mM NaCl) and the fractions of interest were pooled, concentrated, and stored at −20 °C.

The ORFs of PUS enzymes were synthetically generated and cloned in suitable expression vectors for bacteria or insect cells, such as pETM30 (with N-terminal GST-His-tag) or pFastBac. However, human PUS10 ORF was cloned to pETM11 with N-terminus His-tag. To generate catalytically dead variants of PUS enzymes, a site-directed mutagenesis method was applied. PUS1, TRUB1/PUS4, PUS7 and PUS10 are known to be produced using the bacterial expression system, while PUS3 can be expressed using the insect cell expression method according to our previous work (Lin et al, 2022). Depending on the expression system, the cells were grown according to standard protocols. All plasmids, including the PUS1, TRUB1/PUS4, PUS7 and PUS10 constructs, were transformed into the strain BL21 (DE3) CodonPlus-RIL competent cells. Transformed cells were grown in TB broth and protein expressions were induced by isopropyl β-D-1-thiogalactopyranoside IPTG (1 mM) for overnight at 18 °C. Cells were collected and lysed in lysis buffer (50 mM Tris, pH 8.0, 300 mM NaCl, 5 mM DTT, 5% glycerol, 1 mM EDTA, 5 mM MgCl₂) containing protease inhibitors and DNase. The solution was sonicated, and the soluble part was separated from the cell debris using centrifugation. The protein containing supernatant was subject to affinity purification, such as NiNTA beads or GSTPrep purification. The eluate was collected and incubated with the glutathione S-transferase (GST) fused or Histidine (His) tagged Tobacco Etch Virus (TEV) protease overnight at 4 °C and followed by removal of TEV and GST-tag or His-tag. The solution was applied to a gel filtration step. An optional heparin step was introduced prior to the gel filtration depending on whether there is nucleic acid contamination.

Proteins at various concentrations (0.01–10 µM) were mixed with Cy5-labeled tRNA^Gln substrate (30 nM) in a 10 µL reaction volume. The buffer contains 20 mM HEPES, 100 mM NaCl, 2 mM MgCl₂ and 2 mM DTT. The reactions were performed at 4 °C for 30 min to equilibrate the complex formation. The samples were subjected to capillaries for the Monolith device (NanoTemper Technologies GmbH) to determine the Kds. To determine Kds, samples were loaded into the premium capillaries (NanoTemper Technologies GmbH; cat. #MO-K025) and the run was controlled with MO.Control v1.4.4 (NanoTemper Technologies GmbH; cat. no MO-S001C). The Kds were calculated based on the measurement results (n = 3) using MO.Affinity Analysis v1.4.4 (NanoTemper Technologies GmbH; cat. no MO-S001A) and the binding response graphs were plotted using GraphPad Prism (GraphPad Software).

Human PUS enzymes (5 µg for PUS1, PUS3, PUS4 and PUS7 or 25 µg for PUS10) were mixed with tRNA of interest (8 µg) in a 25-µL reaction volume in buffer containing 100 mM ammonium acetate, 100 mM NaCl, 20 mM Tris-Cl, pH 8.0, 5 mM MgCl₂, 5 mM DTT, incubated at 37 °C for 1 h and followed by phenol/chloroform extraction and precipitation overnight in EtOH at −80 °C (Lin et al, 2022). In the cases with PUS10 treatment, PUS10 was added last to the mixture and the incubation was performed at 37 °C for further 8 h. The tRNA was further treated with or without CMC to form specific CMC-Ψ conjugation, cleaned up again using phenol/chloroform extraction and precipitation in EtOH at −80 °C. The recovered RNAs were redissolved in 15 µL H₂O and 1 µL of tRNA solution was subjected to reverse transcriptions using SuperScript III (Invitrogen) with Cy5-labeled RNA-specific primers for 1 h at 50 °C and followed by Protease K treatment (37 °C for 30 min). The reverse transcribed cDNA products were then resolved in a 15% urea denaturing gel and run at 200 V for 60 min. The bands corresponding to with and without CMC-Ψ conjugations were visualized using the Bio-Rad GelDoc imaging system.

For time- course pseudouridinylation reaction, human PUS enzymes (2.5 µg) were mixed with tRNA of interest (4 µg) in a 25-µL reaction volume in the reaction buffer (100 mM ammonium acetate, 100 mM NaCl, 20 mM Tris-Cl, pH 8.0, 5 mM MgCl₂, 5 mM DTT), incubated at 37 °C for dedicated time, namely: 1 min, 30 min, 1 h and 4 h. Reactions were stopped by adding phenol/chloroform and following with extraction as described above.

HEK293T cells (Dharmacon) were cultured at 37 °C in 5% CO₂ in Dulbecco’s modified Eagle’s medium (Lonza) supplemented with 10% (v/v) FBS (EurX). Cells were cultured and collected from nine T75 flasks with 90% confluence. Cells were washed with PBS and lysed in 335 µL lysis buffer (10 mM Tris-HCl pH 7.5, 100 mM NaCl, 10 mM MgCl₂, 1% Triton X100, 0.5 mM DTT and 0.5% sodium deoxycholate). An equal volume of water was added to the lysate and the total RNA was isolated by three subsequent extractions of one volume of acid phenol‐chloroform (Acid‐Phenol:Chloroform, pH 4.5 (with IAA, 125:24:1)) and followed by final extraction with one volume of chloroform. During each extraction, the mixture was vortexed thoroughly and followed by centrifugation at 4500 × g for 10 min at 4 °C. The upper aqueous phase was transferred to a new tube and the extraction was performed again. The RNA, in the upper phase, was precipitated with 0.1 volume of 3 M sodium acetate (pH 6.3), 3 volumes of 96% EtOH, and 10 µg glycogen. The solution was incubated overnight at −80 °C and RNA was spun down at 7100 × g for 30 min at 4 °C. The RNA pellet was washed in 70% EtOH and air‐dried for 2 min. The pellet was then dissolved in RNase‐free water and subjected to total tRNA isolation. A NucleoBond AX100 column was equilibrated with 10 mL of equilibration buffer with Triton X100 (10 mM BisTris HCl, pH 6.3, 200 mM KCl, 15% EtOH, and 0.15% Triton X100). Total RNA was dissolved in 2 mL of equilibration buffer without Triton X-100 and applied to the column. The column was washed twice with 12 mL of wash buffer (10 mM Bis-Tris HCl, pH 6.3, 300 mM KCl, 15% EtOH). Bound tRNA was eluted with 12 mL of elution buffer (10 mM BisTris HCl, pH 6.3, 750 mM KCl, 1% EtOH) and followed by precipitation using 2.5 volumes of 96% EtOH and 10 µg of glycogen. The solution was incubated overnight at −80 °C and tRNA was spun down at 7100 × g for 30 min at 4 °C. The pellet was washed twice in 70% EtOH and air‐dried for 2 min. The tRNA pellet was dissolved in 20 µL of RNase‐free water. Subsequently, bulk tRNA was subjected to isolation of specific single tRNA using Dynabeads MyOne Streptavidin C1 magnetic beads coupled with specific oligonucleotide at 7.5 µM concentration. Bulk tRNA (350 µg) was mixed with 200 µL magnetic beads with immobilized oligonucleotide in binding buffer (1.2 M NaCl, 30 mM HEPES pH 7.5, 15 mM EDTA). The sample was incubated 20 min at 65 °C with shaking (350 rpm), followed by incubation for 1 h at room temperature with shaking. tRNA was washed 3 times with washing buffer at 37 °C (100 mM NaCl, 2.5 mM HEPES pH 7.5, 1.25 mM EDTA). Elution was done first with high salt buffer (150 mM NaCl, 0.5 mM HEPES pH 7.5, 0.25 mM EDTA) 3 times for 5 min at 65 °C with shaking and then with low salt buffer (100 mM NaCl, 0.5 mM HEPES pH 7.5, 0.25 mM EDTA) 3 times for 5 min at 65 °C with shaking. Elution fractions were combined and precipitated with 2.5 volumes of 96% EtOH at −80 °C. The sample was spun down at 7100 × g for 30 min at 4 °C and washed twice with 70% EtOH. The pellet was dissolved in 15 µL RNase-free water. The quality of extracted tRNA was checked by resolving the samples in a 10% urea gel, staining with ethidium bromide solution, and visualizing with a BioRad ChemiDoc imaging system.

Thiolated tRNAs were separated from non-thiolated tRNAs on a 10% denaturing poly-acrylamide gel (7 M urea, 0.5% TBE) supplemented with 130 μM [N-acryloylamino]phenyl)mercuric chloride (APM) (Sokołowski et al, 2024). APM interacts with exposed thiol groups and leads to an increased retardation of thiolated tRNAs in comparison to non-thiolated tRNAs. In gels without APM thiolated and non-thiolated tRNA are not separated and run at the same migration speed.

The thermal stability of the tRNAs was evaluated with the use of a Quant-iT RiboGreen™ RNA fluorescent probe (RiboGreen) (R11491 ThermoFisher Scientific). 200 nM of tRNA produced in-house was mixed with 200x diluted RiboGreen probe in 50 µL of the assay buffer containing 20 mM HEPES, pH 7.5, 150 mM NaCl, 1 mM MgCl₂, 5 mM DTT. The assay volume was mixed, short-spined, and incubated for 10 min at room temperature protected from light. 10 µL of the assay mix was loaded into capillaries (NanoTemper Technologies GmbH, cat. no AN-041001)—a total of 4 capillaries per the condition set. The capillaries were sealed with oil-based liquid rubber (NanoTemper TechnologiesGmbH, cat. no PR-P001). Capillaries were loaded into the Andromeda system from NanoTemper Technologies, allowing monitoring the fluorescence signal in the temperature gradient. The data was collected with AN.Control Software v1.1 (NanoTemper TechnologiesGmbH, cat. no AN-020001). The fluorescence excitation was adjusted to maintain total intensity at 510 nm below 2000 counts. The temperature ramp was set to 1 °C/min and fluorescence signal was collected for the temperature range 25–95 °C. Collected raw data were exported, processed and plotted (box plot) with Origin Pro 2023 (OriginLabs). For the line plots representing change of the fluorescence signal as a function of temperature the signal was further processed for comparisons across conditions (different tRNAs, PUS enzyme reactions). The signal processing involved linear interpolation to obtain equal size of the temperature vector, Min-Max normalization of the fluorescence signal, and first derivative determination with baseline correction using PeakAnalyzer from OriginLabs.

tRNA^Gln_UUG labeled with a Cy5 was obtained from Metabion (Metabion International AG, Germany). The Cy5 label was placed at the 5′ end and U₄₈ position of the tRNA. The lyophilized tRNAs were resuspended in miliQ grade water at 100 µM. Before the enzymatic reaction, tRNA refolding was carried out by heating the RNA solution at 80 °C for 2 min and slowly cooling to room temperature. A 10 nM solution of enzymatically modified tRNAs was prepared in 100 µL of 20 mM HEPES, pH 7.5, 150 mM NaCl, 1 mM MgCl₂, 5 mM DTT. The assay volume was mixed, short-spined, and incubated for 10 min at room temperature. 10 µL of the solution was loaded into capillaries (NanoTemper Technologies GmbH, cat no. AN-041001)—a total of 4 capillaries per condition set. Thermal unfolding analysis was performed with the use of Andromeda system (NanoTemper Technologies GmbH) using Cy5 as a source of fluorescence signal; run parameters and data analysis were performed as described above for the Quant-iT Ribogreen™ probe.

The tRNA of interest, modified individually with PUS4, PUS7, and the combination PUS3,4,7, was pre-diluted to 1 µM in 30 µL of the assay buffer comprising 20 mM HEPES (pH 7.5), 150 mM NaCl, 1 mM MgCl₂, and 5 mM DTT. Subsequently, titrations of PUS4 against PUS-3/4/7 and PUS7 against PUS-3/4/7 were prepared in separate tubes according to the provided tables. The total molar concentration of tRNA was maintained at 250 nM, with a total volume of 40 µL in the assay buffer. In each tube, 10 µL of a 40x pre-diluted RiboGreen probe was added, resulting in a final concentration of the tRNA of interest at 200 nM, with the fluorescent probe being 200x diluted. As a reference, 200 nM of non-modified tRNA of interest was mixed with a 200x diluted RiboGreen probe in 50 µL. This reference was employed to calculate ΔTms during the subsequent analysis. The reactions were incubated for 10 min at room temperature with RiboGreen, followed by the determination of tRNA stability using the PR.Andromeda system, as previously described.

Non-modified tRNA, tRNA incubated with PUS-3/4/7, and tRNA treated with inactive mutants of PUS-3/4/7 enzymes were separately mixed with a RiboGreen probe using the following procedure. A 200 nM concentration of in-house-produced tRNA was combined with a 200x diluted RiboGreen probe in a 50 µL assay buffer containing 20 mM HEPES (pH 7.5), 150 mM NaCl, 1 mM MgCl₂, and 5 mM DTT. The assay mixture was thoroughly mixed, briefly spun, and then incubated for 10 min at room temperature, protected from light. Subsequently, 10 µL of the assay mix was loaded into capillaries (NanoTemper Technologies GmbH, cat. no AN-041001), with a total of 4 capillaries per condition. These capillaries were sealed with oil-based liquid rubber (NanoTemper Technologies GmbH, cat. no PR-P001) and then inserted into the Andromeda system from NanoTemper Technologies. This system allowed for the monitoring of the fluorescence signal in a temperature gradient. Data collection was performed using AN.Control Software v1.1 (NanoTemper Technologies GmbH, cat. no AN-020001). The fluorescence excitation was adjusted to maintain a total intensity at 510 nm below 2000 counts. The temperature ramp was set to 1 °C/min, and the fluorescence signal was collected over the temperature range of 25–95 °C in three consecutive cycles of unfolding-refolding, with continuous monitoring of fluorescence changes. The analysis of the melting profiles and ΔTm calculations were conducted as described in the section on the determination of tRNA thermal stability changes.

The hydrodynamic radius of tRNAs was assessed with a Prometheus Panta nanoDLS module (NanoTemper Technologies GmbH). 20 µM of in-house produced tRNA was prepared in 30 µL of 20 mM HEPES, pH 7.5, 150 mM NaCl, 1 mM MgCl₂, 5 mM DTT. The assay volume was mixed, short-spined, and incubated for 10 min at room temperature. 10 µL of the solution was loaded into the capillaries (NanoTemper Technologies GmbH, cat no. PR-C006)—a total of 3 capillaries per condition set. The capillaries were loaded onto the Prometheus Panta system and sealed with oil-based liquid. Data were collected using PR.Panta Control software v1.4.4 (NanoTemper Technologies PR-S061). For the isothermal size analysis of tRNA the nanoDLS run was set to 5 s and 10 acquisitions per capillary. The hydrodynamic radius of tRNA was determined using the regularization model implemented in the software and the intensity plots were exported. For the thermal unfolding analysis of the tRNA hydrodynamic radius the nanoDLS run was set to a 500 ms acquisition per capillary. The thermal ramp was set at 1 °C/min and the temperature analysis range was adjusted to 20–95 °C. Hydrodynamic radius changes in function of temperature were determined using the cumulant radius model implemented in the software. The raw data were exported, analyzed and plotted using OriginLabs. Predictions of rH based on the tRNA molecular weight were calculated using available online converter (www.fluidic.com/toolkit/hydrodynamic-radius-converter/).

QUANTIFOIL® R 2/1 copper grids (200 mesh) were glow discharged on a Leica EM ACE 200 glow discharger (8 mA, 60 s). 2.5 µL of 15 µM tRNA in assay buffer (20 mM HEPES, pH 7.5, 150 mM NaCl, 1 mM MgCl₂, 5 mM DTT) was plunge-frozen using a Vitrobot Mark IV (Thermo Fisher Scientific) set to 100% humidity and 4 °C with the following blotting parameters: wait time 1 s, blot force 5 s and blot time 2 s. The datasets were collected on a 300 keV Titan Krios G3i (Thermo Fisher Scientific, Solaris, Poland) equipped with a Gatan BioQuantum energy filter and a Gatan K3 BioQuantum direct electron detector. The under-focus range was −0.8 to −2.4 µm for a total of 40 frames accumulating an overall dose of 40 e⁻/Ų. Respective numbers of collected micrographs for each dataset can be found in the Table 1. The used pixel size was 0.415 Å for in vivo tRNA^Arg_UCU and 0.86 Å for the remaining datasets.

All micrographs were subjected to motion correction using WARP5 with subsequent CTF estimation. Then the micrographs were subjected to the selection protocol to fish out only the best quality images for further analysis. This task was done using the ExposureCuration job in cryoSPARC with a set of parameters as follows: image resolution <8 A, defocus tilt angle <20°, astigmatism <500 nm. Selected micrographs were subjected to further processing steps. Picked particles together with averaged micrographs were imported to cryoSPARC and analyzed in detail. As a first step, reference free 2D classification was used to select the best possible particles and create templates for the template picker. As a validation of selected template, template picker was always run on a small subset of micrographs (typically 500-1000). Template-picked particles were 2D classified in reference-free 2D classification(s) until best particles were selected (typically 3-to-5 rounds). The particles from the best 2D classes (resolution ≤ 8 Å, ECA ≤ 3) were used to build initial 3D models, from which the one contains all essential features was used to train TOPAZ (Bepler et al, 2019) to increase the number of good particles that could have been missed out during previous steps. TOPAZ picked particles were submitted to few rounds of reference free 2D classification to clean the particles set. The particles with all required density were used to create new 3D models, of which only the best one containing all essential features was refined and determined the solution. To refine the best model, the Homogenous Refinement (HomoRef)(Punjani et al, 2017) followed by Non-Uniform refinement (Punjani et al, 2020) protocols were used. The FSC correlation curve was calculated for each final reconstruction using a gold standard FSC = 0.143 threshold. Local resolution was calculated in LocalResolutionEstimation job implemented in cryoSPARC and implements a local windowed FSC method for estimating local resolution, similar to the blocres program (Cardone et al, 2013).

The initial models of human tRNA 3D structures were built using comparative modeling with ModeRNA (Rother et al, 2011), using a collection of experimentally determined tRNA structures as templates (with RCSB PDB code 6r7q_A as the main template). These models were superimposed onto the corresponding cryo-EM maps, with the aid of the Fit in Map tool in ChimeraX (Pettersen et al, 2021). The models were flexibly refined in the context of the cryo-EM maps using a coarse-grained modeling tool SimRNA (Boniecki et al, 2015), which performs Monte Carlo simulations and evaluates RNA conformations with a statistical potential. For the cryo-EM tRNA structure determination, distance restraints were used to maximize the preservation of canonical and non-canonical base-pairs and stacking interactions conserved in tRNAs. 10 models that scored best according to the SimRNA scoring function had their all-atom representation reconstructed. These models were further refined against the map using Phenix real_space_refine (Afonine et al, 2018) with macro_cycles=1 and restraints derived from the initial model using doubleHelix (Chojnowski, 2023). Finally, we used QRNAS (Stasiewicz et al, 2019) to minimize steric clashes and improve local geometry. The fit of models to the map was assessed using QScore with the MapQ plugin (Pintilie et al, 2020) in ChimeraX.

Starting with the folded 3D structure of tRNA^Gln_UUG, a series of isothermal single-replica SimRNA simulations were run, without any restraints, to assess the stability of the global tRNA structure and the pairwise contacts made by individual ribonucleotide residues. The probability of accepting or rejecting a move in the SimRNA simulation is based on a Metropolis criterion (Metropolis et al, 1953), which depends on the T parameter, analogous to the physical concept of temperature in statistical mechanics. If the new state in the SimRNA Monte Carlo simulation corresponds to a higher (less favorable) energy, it is accepted with a probability governed by the Boltzmann factor, exp(-ΔE/kT), where ΔE is the difference in energy between the two states, k is the Boltzmann constant, and T is the temperature. Therefore, the temperature parameter T modulates the probability of accepting higher-energy states, and consequently, it controls the degree of exploration of the conformational space. A higher temperature allows for more frequent acceptance of higher-energy states, thereby enabling a broader exploration of the conformational space, while a lower temperature narrows this exploration to energy minima, by prioritizing lower-energy states. To simulate the structural stability of tRNA^Gln_UUG, we used T ranging from 0.6 (where the tRNA remained folded and very similar to the structure determined by cryo-EM) to 1.4 (where the tRNA unfolded quickly). For all frames from each simulation, all-atom models were reconstructed and compared with the cryo-EM structure as a reference. For every residue in each conformation, pairwise contacts were identified and compared with the reference structure using 1D2DSimScore (Moafinejad et al, 2023), to determine whether a given residue retains native-like interactions or if its native-like contacts are broken. The hydrodynamic radii of the models were assessed using HullRad (Ortega et al, 2011; Fleming and Fleming, 2018). To compare hydrodynamic radii from simulations with the experimental data, the simulated radii values were recalibrated by subtracting the factor of 0.5 nm that accounts for the underestimation of the hydrodynamic radii by the cumulant radius model. All values were exported to xls files.

Molecular dynamics (MD) simulations were performed for four RNA variants of tRNA^Gln_UUG: without any modifications, with single modifications Ψ₁₃ or Ψ₅₅, and both Ψ₁₃ and Ψ_55.. The structure of tRNA^Gln_UUG built into the cryo-EM density map was used as the starting model. The modifications were introduced using ModeRNA (Rother et al, 2011). The system was prepared using the Amber χOL3 force field (Pérez et al, 2007; Zgarbová et al, 2011; Banáš et al, 2010), and the modified nucleotide parameters for Ψ were generated using Antechamber (Wang et al, 2006, 2004). Both systems were solvated in a TIP3P water (Sun and Kollman, 1995) box with at least a 10 Å buffer, and counterions were added for neutralization. Initial minimization was performed in two stages to relieve any steric clashes and relax the system’s energy. Heating was conducted using a Langevin thermostat to gradually raise the system temperature from 0 K to 300 K under constant volume conditions, with heavy atom restraints applied to maintain RNA structural integrity.

Equilibration followed in three stages, each with progressively reduced positional restraints, allowing the system to adjust to the target temperature and pressure. Constant pressure periodic boundary conditions were used to mimic physiological conditions, with a pressure relaxation time of 2 ps. After withdrawing all constraints, 1 ns MD simulations were carried out under the conditions of constant temperature and constant volume (NVT) as well as constant temperature, constant pressure (NPT). In NPT condition, a pressure of 1.0 atm was kept using the Berendsen barostat (Berendsen et al, 1984) with a relaxation time of 2 ps. The SHAKE (Ryckaert et al, 1977) algorithm was employed to constrain all bond lengths involving hydrogen atoms. The final production run, conducted for 500 ns, utilized the GPU-accelerated pmemd.cuda engine of AMBER22 (Götz et al, 2012) with a 2 fs time step. The choice of parameters, such as Langevin thermostat (Turq et al, 1977) (γ = 1 ps⁻¹) and particle mesh Ewald (PME) (Darden et al, 1998) or long-range electrostatics, was made to balance computational efficiency and accuracy, ensuring stable temperature and pressure control throughout the simulation. Three independent runs for each RNA variant were calculated to mitigate the potential influence of random initial conditions and stochastic fluctuations that could arise during the simulations. MD trajectories were analyzed by cpptraj (Roe and Cheatham, 2013). Stacking energies were calculated using the LIE (de Amorim et al, 2008) utility of AmberTools (Case et al, 2023).

All details related to statistical analysis can be found in figure legends. Graphed datasets are represented in box plots. The central line inside the box indicates the median and the square the mean. The lower and upper edges correspond to the first (Q1) and third (Q3) quartiles, respectively, defining the interquartile range (IQR). The whiskers extend to the minimal and maximal values within the 1.5IQR. Points beyond this range are considered outliers and are shown as transparent points. All statistical analyses were performed using one-way ANOVA (α = 0.05) with a Bonferroni multiple comparisons test. The exact p values are shown in each panel.