Cell culture NAIL-MS allows insight into human RNA modification dynamics in vivo

In the last years, studies about the dynamics of RNA modifications are among the most controversially discussed. As the main reason, we have identified the unavailability of a technique which allows to follow the temporal dynamics of RNA transcripts in human cell culture. Here, we present a NAIL-MS (nucleic acid isotope labeling coupled mass spectrometry) scheme for efficient stable isotope labeling in both RNA and DNA (>95% within 7 days) in common human cell lines and growth media. Validation experiments reveal that the labeling procedure itself does neither interfere with the isotope dilution MS quantification nor with RNA modification density. We design pulse chase NAIL-MS experiments and apply the new tool to study the temporal placement of modified nucleosides in e.g. tRNAPhe and 18S rRNA. In existing RNAs, we observe a constant loss of modified nucleosides over time which is masked by a post-transcriptional methylation mechanism and thus not detectable without NAIL-MS. During alkylation stress, NAIL-MS reveals an adaptation of tRNA modifications in new transcripts but not existing transcripts. Overall, we present a fast and reliable stable isotope labeling strategy which allows a more detailed study of RNA modification dynamics in human cell culture. With cell culture NAIL-MS it is finally possible to study the speed of both modification and demethylation reactions inside human cells. Thus it will be possible to study the impact of external stimuli and stress on human RNA modification kinetics and processing of mature RNA.


Introduction
Most RNAs studied to date were found to be covalently modified by dedicated enzymes in a site specific manner. In addition to the placement of RNA modifications by RNA writer enzymes, their direct removal through e.g. demethylation by RNA erasers was reported. In human cells, the α-ketoglutarate dependent dioxygenases ALKBH5 and/or FTO were found to catalyze the demethylation of e.g. (2'-O-methyl-)N6-methyladenosine (m 6 A(m)) in mRNA 1,2 and thus influence e.g. the stability and translational function of mRNA 1,3-8 .
For human tRNAs, a similar relationship of RNA writers and erasers was observed. E.g. ALKBH1 demethylates 1-methyladenosine (m 1 A) and appears to be responsive to glucose starvation in some cell lines 9 . Considering the half-life of mammalian tRNAs (~ 100 h 10 ), a fast adaptation by removal of modified residues appears beneficial to react to changes in the cellular environment 11 .
Unfortunately, it is currently not possible to analyze the speed of both modification and demethylation reactions inside human cells. Thus it is not possible to study the impact of external stimuli and stress on human RNA modification kinetics and processing of mature RNA.
tRNA is the most extensive and chemically diverse modified RNA with ~10-15% of all nucleosides being modified 12 . Recent studies showed that certain modified nucleosides in specific tRNAs are only partially modified 13,14 and that tRNA modification abundance differs among tissues 15,16 . This would allow for an adaptation of translation by tRNA modification as recently suggested 17 . While the speed of tRNA amino acid charging 18 and tRNA transcription and half-live are known 10 , the speed of modification processes is difficult to study. For example, tRNA Phe is heavily posttranscriptionally modified and in addition one of the best studied RNAs [19][20][21] . By using stable isotope labeled tRNA Phe substrate and cellular extracts, the modification dynamics and hierarchy was recently solved in S. cerevisiae using NMR spectroscopy 22 . Under the influence of chemical stress, S. cerevisiae was reported to adapt its abundance of tRNA modifications and thus influence its translation and the term stress induced tRNA reprogramming was coined 11,23 . Similar evidence has been observed in other organisms, including mammals 24 . In this context, the question remains by which mechanism and how fast tRNA modifications respond to external stimuli.
In contrast to tRNA, 18S rRNA is mainly modified by methylation of ribose and altogether only 2.05 % nucleosides are modified. While tRNA modifications are easily accessible for potential RNA erasers, rRNA modifications are placed in the functional regions of the ribosome 25 . Although modified sites in rRNA have been reported to regulate translation initiation by promoting the recognition of different mRNA subsets 26 their inaccessibility in mature ribosomes makes them a difficult target for RNA erasers.
Current studies of RNA modifications are limited to either mass spectrometric analysis 16 or sequencing 27,28 . Both techniques provide information on the modification status at the time point of sample harvest and give no details on the mechanisms of RNA modification adaptation. To overcome this limitation, we have recently developed NAIL-MS (nucleic acid isotope labeling coupled mass spectrometry) in bacteria 29,30 and yeast 31 , which reveals the dynamics of RNA modification processes. The technique is based on metabolic stable isotope labeling of RNA using simple nutrients with e.g. carbon-13, nitrogen-15 or sulfur-34. By combining differentially labeled media in a pulse chase set-up, we recently succeeded to observe tRNA demethylation through AlkB in E. coli in vivo.
Currently, NAIL-MS studies are not available for human cell lines as a monoisotopic labeling of all four canonical nucleosides is highly complex and thus not available.
Here, we report a fast and reliable method for monoisotopic stable isotope labeling in both RNA and DNA (>95% within 7 days) in common human cell lines and growth media. We apply the cell culture NAIL-MS method and reveal the dynamics of human tRNA and 18S rRNA modifications in depths unreachable by any other tool for RNA modification analysis. Furthermore, we resolve the mechanism of stress induced tRNA modification reprogramming in the presence of methylation stress. With cell culture NAIL-MS it is finally possible to study the speed of both modification and demethylation reactions inside human cells. Thus it will be possible to study the impact of external stimuli and stress on human RNA modification kinetics and processing of mature RNA.

Absolute quantification of human tRNA Phe modifications
tRNA Phe is heavily post-transcriptionally modified and in addition one of the best studied RNAs [19][20][21][22] .
Thus it is an ideal model to study the temporal dynamics of its modifications. In a first step, we purified tRNA Phe GAA from HEK 293 cells using a complementary DNA probe 13 . We used our established isotope dilution LC-MS/MS analysis for absolute quantification of modified nucleosides and plotted the modification profile in Figure 1 16  is lower compared to the literature, presumably due to partial modification of the respective sites.
Partial modification has been suggested to play a role in stress induced reprogramming of tRNA modifications 17 . The abundance of 5-methylcytosine (m 5 C) is slightly higher than expected and can be explained by the additional methylation of C48 by NSUN2 33 .
Although 1-methylguanosine (m 1 G) is not reported in tRNA Phe GAA, we found around 0.3 m 1 G per tRNA. This observation can be explained by the fact that m 1 G is a precursor during the biosynthesis of wybutosine (yW), a hypermodified nucleoside reported at position 37 of tRNA Phe GAA 34,35 . Due to the unavailability of a synthetic standard, yW could not be quantified in this study. Additionally, we also quantified the abundance of other modified nucleosides (Table S1). We found around 0.3 6-methyladenosine (m 6 A) per tRNA, potentially caused by intracellular dimroth rearrangement of m 1 A 36 . In addition, we found 0.063 inosine (I) and 0.026 1-methylinosine (m 1 I) per tRNA Phe . These are most likely artefacts from A and m 1 A deamination. All other modified nucleosides, were found with an abundance of less than 1.6% (e.g. 0.016 N 6 -threonylcarbamoyladenosine (t 6 A) per tRNA) which indicates a high purity of isolated tRNA Phe .
Overall the detected quantities of modified nucleosides from purified tRNA Phe GAA are in accordance with the reported values and thus it is a suitable model to study the temporal placement of modified nucleosides.

Stable isotope labeling of RNA in human cell culture
For this purpose, a method is needed which allows the discrimination of mature RNA from new transcripts. NAIL-MS (nucleic acid isotope labeling coupled mass spectrometry) relies on the metabolic incorporation of stable isotope labeled nutrients into RNA and allows the distinction of original RNA and new RNA within a pulse chase experiment. With this tool, we studied the temporal placement of modified nucleosides in S. cerevisiae total tRNA 31 and the demethylation during tRNA repair in E. coli 37 . Both organisms are rather simple and they can be grown in minimal media with controlled availability of stable isotope labeled nutrients.
In contrast, human cell culture medium is highly complex and requires the addition of fetal bovine serum (FBS). FBS is a natural product of undefined composition and variable concentration of metabolites. Thus a complete and monoisotopic labeling of nucleosides and even nucleobases for a pulse chase NAIL-MS assay is challenging.
From our experience, the target isotopologue of a nucleoside must be at least 3 u heavier compared to the naturally occurring nucleoside to avoid false positive results by the detection of the natural carbon-13 signals.
De novo synthesis of nucleosides utilizes several amino acids such as glutamine or aspartic acid ( Figure S1A and S1B) 38 . Hence, we supplemented the growth media with stable isotope labeled glutamine. After 5 days (2 passages), we observed the expected stable isotope labeling of RNA ( Figure S1C). Cytidine, guanosine and adenosine got a mass increase of +2 whereas uridine just increased by +1. Due to the overlap with naturally occurring ( 13 C)-isotopologues, this mass increase was not sufficient for our planned experiments.
As recently described, it is possible to use glucose-free growth medium and supplement with 13 C6glucose 37 . The feeding with 13 C6-glucose leads to the formation of nucleosides with a variable number of ( 13 C) per nucleoside ( Figure S1C). During method development, we utilized the non-monoisotopic nature of 13 C6-glucose labeling to test the incorporation efficiency of various unlabeled metabolites.
Addition of aspartate and pyruvate did not allow the envisioned monoisotopic labeling ( Figure S2).
The addition of the nucleobases adenine and uracil resulted in ribose labeled purines but undefined labeled pyrimidines. This indicates a direct usage of adenine from the medium which is then enzymatically connected with 13 C5-ribose followed by further processing to guanosine and the respective triphosphates ( Figure S1B). RNA supplemented with the nucleosides adenosine and uridine showed undefined labeled purines and only unlabeled pyrimidines ( Figure S3). This indicates that uridine is taken up by the cells and immediately utilized for cytidine and RNA synthesis ( Figure   S1A). In summary, our data indicates that the addition of adenine and uridine blocks de novo purine and pyrimidine synthesis ( Figure S1A and S1B) and 13 C6-glucose medium is not necessary for our labeling strategy as unlabeled nucleosides remain visible in the mass spectra (Figures S3). The high resolution mass spectra of the resulting RNA nucleosides showed the desired labeling for >95% of all canonical nucleosides after 7 days (Figure 2b). A +7 mass increase is observed for cytidine and uridine and a +5 and +4 mass increase for adenosine and guanosine, respectively. By using dialyzed FBS, the signal of unlabeled adenosine could be further reduced in comparison to normal FBS ( Figure S5). Similarly, DNA nucleosides become stable isotope labeled ( Figure S6).
With these metabolites, a pulse chase NAIL-MS study is possible in human cell culture. To this end, we achieve excellent labeling in HEK 293, HAP and HeLa cell lines using supplemented DMEM RPMI or IMDM medium ( Figure S7).
In HEK 293 cells, the signals of new tRNA transcripts became detectable and quantifiable after 120 minutes of labeling ( Figure S9/S10).
Most modified nucleosides in RNA carry one or more methylations. To follow the fate of these methylated nucleosides in the context of RNA maturation and methylation damage response, we used CD3-labeled methionine. Methionine is the precursor amino acid of S-adenosylmethionine (SAM) which in turn is cofactor of most RNA methyltransferases. In the presence of CD3-methionine, methylated nucleosides get a mass increase of +3 and can thus be distinguished from nucleosides modified in the presence of unlabeled methionine. High resolution mass spectra of fully labeled m 5 C, m 7 G and m 1 A are exemplarily shown in (Figure 2c). In order to achieve complete labeling of methylgroups methionine depleted medium has to be used. We chose DMEM D0422 (from Sigma-Aldrich) which lacks glutamine, cystine and methionine ( Figure S11). Neither cell shape nor growth speed were influenced by the labeling and both were comparable to standard DMEM (e.g. D6546, from Sigma-Aldrich) ( Figure S12).
The combination of nucleoside and methyl-group labeling allows the design of elegant pulse chase studies to follow the fate of RNA in human cells.

Validation of human cell culture NAIL-MS
After finding a suitable way for monoisotopic labeling of RNA in human cells, we wanted to rule out the possibility that the labeling itself impacts the abundance of RNA modifications. For this purpose, cells were grown in labeled or unlabeled media for 7 days. Both media contained adenine, uridine and methionine as either unlabeled or labeled nutrients. Cells were harvested with TRI reagent and split into two aliquots. One aliquot (2/3 Vol) was used for immediate RNA isolation and purification, while the remaining aliquot of the labeled and unlabeled cells were mixed and RNA was co-isolated and co-purified ( Figure 3a and Figure S13). The total tRNA was enzymatically digested to nucleosides and their abundance determined by isotope dilution mass spectrometry 16 . In the aliquot from unlabeled samples, only unlabeled nucleosides were detectable, while the aliquot of the labeled cells showed mainly signals (>98%) for labeled nucleosides. As expected from the mixed sample, we detected unlabeled and labeled isotopologues of all canonicals in equivalent amounts ( Figure 3b).
Next, we quantified the abundance of modified nucleosides. For normalization, unlabeled modifications were referenced to unlabeled canonicals and labeled modifications were referenced to labeled canonicals. The calculated quantities of modified nucleosides present in tRNA Phe are plotted for the unlabeled against labeled tRNA in Figure 3c. This validation revealed that the quantities of modified nucleosides are independent of the media and that the labeling procedure itself does not interfere with the isotope dilution MS quantification. The deviation from the expected values is the error of this NAIL-MS experiment and the limitation to detect differences in a biological setup (also see Figure S14). E.g. In total tRNA, 2'-O-methyluridine (Um) has the largest error as its abundance deviates 1.6 fold in labeled and unlabeled media.
The promising results from the validation experiments allowed the design of pulse chase experiments.
Such experiments start with cells seeded in medium-I and upon experiment initiation, the medium is exchanged to medium-II with different isotopically labeled nutrients. The concept is shown in Figure   3d. To rule out possible differences in the results in dependence of the starting medium, we designed a brief validation experiment. In the forward experiment, cells are seeded in unlabeled medium and switched to labeled medium while the reverse experiment starts in labeled medium (after a 7 day labeling period) before switching to unlabeled.
For analysis of modified nucleoside quantities, we harvested the cells and extracted total tRNA after switching to medium-II (time points 0, 6, 24 and 48 hours). To assess the suitability of the method for temporal placement of modified nucleosides into the total tRNA, we focused on the abundance of new modified nucleosides in the newly transcribed tRNA. For direct comparison, the ratio of found (6, 24, 48 hnew transcripts) and expected (0 horiginal transcripts) modified nucleoside quantity was formed and plotted over time. As expected, we observed the incorporation of modified nucleosides into the new tRNA after medium exchange. While the timing of the tRNA modification process was comparable in the reverse and forward experiment, the start values were obscured in the reverse experiment due to low, but detectable signals of unlabeled nucleosides. For this and economic considerations, we decided to perform forward pulse chase experiments in the future to avoid the excessive use of labeled medium.

Temporal placement of modified nucleosides in RNA
From a biological perspective, we observed that most modified nucleosides reach their final abundance (100 % compared to the starting point) within 48 h (Figure 3e). Some modified nucleosides, such as m 1 A, m 5 C, Ψ and m 5 U, are already > 90 % after 6 h which indicates a fast incorporation after transcription. These modified nucleosides are located in the structure-stabilizing positions of the tRNA's D-and TΨC-loops and thus a fast and reliable modification is to be expected 39 . m 7 G is also involved in structure stabilization 40 and yet, this methylation is placed rather slowly in total tRNA. Other modified nucleosides such as Cm, Gm, and the base-methylated G derivatives While the modified nucleosides of total tRNA are placed by various enzymes at various positions, we were interested to observe the modification process of defined enzymes in a defined substrate. For this purpose, we performed a pulse chase experiment and purified tRNA Phe GAA after 0, 2, 4, 6, 24 and 48 hours. The abundance of modified nucleosides in new tRNA transcripts is shown in Figure 4. We observe an immediate high abundance of Ψ, which argues towards an immediate isomerization of e.g.
U55 to Ψ55 as observed in yeast 22 . In fact, we observe 1.5 fold more Ψ in the early lifetime of tRNA Phe GAA as is expected from mature tRNA Phe GAA (Figure 1). At these early time points, the abundance of new tRNA Phe GAA transcripts is low and thus the MS signal intensity is close to the lower limit of quantification (LLOQ). Uridine and its modifications have a low ionization efficiency and thus a higher LLOQ compared to other modified nucleosides. Thus biological interpretation of Ψ and m 5 U ( Figure S15) quantities must be conducted carefully. D is not included in this analysis, due to its artificial addition to the samples through the deaminase inhibitor tetrahydrouridine (which was omitted for analysis in Figure 1 and thus allowed quantification of D). While m 7 G is the next modified nucleoside placed in yeast tRNA Phe , our data hints towards a fast incorporation of m 5 C followed by m 1 A and finally m 7 G. Here, the dynamic placement of modifications in the TΨC-loop seems to be slightly different between yeast and human. The slow incorporation of m 2 G in the D-loop is in accordance with the reports from yeast. In the anticodon-loop (ac-loop), we observe a rather slow formation of Gm and Cm. These modified nucleosides are not involved in structure stabilization but codon-anticodon binding 41,42 and protein translation. Our data implies that structure stabilization by modified nucleosides is a key necessity and must thus happen early on, while ac-loop modifications are not immediately needed and are potentially placed on-demand. One exception is the formation of wybutosine (yW). Its precursor modification m 1 G is immediately incorporated into tRNA Phe before its abundance drops at later time points, presumably due to its further processing into yW.

Dynamics of tRNA and 18S rRNA modifications
With the design of our pulse chase NAIL-MS assay, we can observe RNA maturation processes by quantifying the abundance of modified nucleosides in new transcripts. In addition, we can follow the fate of original RNA (unlabeled nucleosides in forward experiment) and observe methylation or demethylation events.
In Figure 5a, we plotted the abundance of exemplary modified nucleosides from original total tRNA, which were present before the medium exchange. Other modified nucleosides are shown in Figure   S16. Similar to our initial observations in S. cerevisiae 31 Figure S18).

Impact of methylation stress on tRNA modification processes
We have recently applied NAIL-MS to profile bacterial tRNA damage by methylating agents 29  From these samples, we purified tRNA Phe GAA and quantified the abundance of canonical and modified nucleosides. By comparison of canonical nucleosides, we could observe a higher ratio of new transcripts over original transcripts in the unstressed samples compared to the stressed samples ( Figure S20). This is to be expected as stressed cells stop growing and thus less transcription and translation are needed. In addition, the prolonged abundance of original tRNA suggests that methylation stress does not lead to extensive degradation of tRNAs.
The quantification of methylated nucleosides derived from direct MMS methylation, indeed showed formation of the known damage products m 7 G and potentially m 1 A. In comparison to the natural abundance of these modified nucleosides (~ 0.5 m 7 G and 1 m 1 A per tRNA Phe GAA), the damage accounts for less than 1 % of these methylated nucleosides (Figure 6b) Figure   5).
Finally, we studied the abundance of modified nucleosides in new tRNA transcripts in dependence of stress. For methylated guanosine derivatives (m 7 G, m 1 G, m 2 G and m 22 G), we observed a slightly reduced, but statistically significant (e.g. m 7 G p6h = 0.0096) incorporation into tRNA Phe GAA under stress compared to the control samples (Figure 6c and Figure S21). For Cm and Gm we observed a higher abundance under stressed conditions while m 1 A or m 5 C were comparable. Our results imply that human cells i) adapt their tRNA modifications to methylation stress by differentially modifying new transcripts and ii) consider tRNA modification as a highly important process and thus continue even during stress exposure.

Discussion
Current analyses of the epitranscriptome are limited to snapshot moments and cannot truly follow dynamic processes inside cells. While NAIL-MS allows the observation of RNA modification adaptation processes 37,44 it was not possible to apply the technique in human cell culture due to the complexity of culture medium. 13 C6-glucose is a reasonable and economic option for stable isotope labeling (28 € per 50 mL medium) 45 but it suffers from the formation of multiple isotopomers which complicates its application especially when additional feeding with CD3-methionine is required. In such studies, the signals of partially 13 C-labeled nucleosides and CD3-methylated nucleosides can overlap and quantification becomes impossible. In contrast, supplementation of various media with 15 N5-adenine and 13 C5 15 N2-uridine results in monoisotopic labeling with no overlap with naturally occurring 13 C-isotopomers or artificially CD3-methylated nucleosides (305 € per 50 mL medium).
Thus a broad applicability and even quantification by isotope dilution mass spectrometry is possible.
While we observe best results with dialyzed FBS, it is also possible to use regular FBS instead if it is preferable to the cells. If the nucleoside of interest is a G or A derivative, 13 C6-glucose labeling can be combined with supplementation of unlabeled adenine. This approach is less costly and produces monoisotopically labeled A and G derivatives with a 13 C5-ribose moiety ( Figure S3 and S4).
An important consideration for any NAIL-MS study is the constant supplementation with adenine and uridine, even when unlabeled medium is used to prevent activation of de novo synthesis pathways. Independently of the chosen nucleic acid labeling scheme, we strongly recommend validation experiments as shown in Figure 3c. Such an experiment is crucial to later judge the statistical significance of e.g. pulse chase studies. For example, our validation experiment indicates that a less than 1.6 fold change in Um would not be biologically significant ( Figure S14). In such a case we recommend the direct comparison to a control sample (such as those in Figure 6) to judge the accuracy of the received NAIL-MS data.
Furthermore, we suggest careful interpretation of new transcript data at early time points of pulse chase experiments. As described for Ψ and m 5 U ( Figure S15), it is possible that some modified nucleosides are early on too close to the lower limit of quantification (LLOQ) in new transcripts and thus the received quantities must be carefully interpreted.
We have studied the temporal placement of modified nucleosides in tRNA Phe as a model. Our data implies that structure stabilization by modified nucleosides is a key necessity and must thus happen early on, while anticodon-loop modifications are not immediately needed and are potentially placed on-demand. One exception is the formation of wybutosine (yW). Its precursor modification m 1 G is immediately incorporated into tRNA Phe before its abundance drops at later time points, presumably due to its further processing into yW. By NMR spectroscopy in combination with stable isotope labeling, Barraud et al. recently observed an inhibition of m 22 G formation by m 2 G 22 . In our hands, m 22 G is placed into tRNA Phe as fast as is m 2 G, but as both modifications are incorporated slowly it is possible that m 22 G is placed in a non-m 2 G modified sub-population. This question might be approached by combining NAIL with oligonucleotide MS.
With NAIL-MS we are not limited to RNA modification studies in new transcripts. In addition, we can follow the fate of RNA modifications in mature transcripts. In human cells, we observe a constant loss of modified nucleosides from tRNAs, similar to our initial report in S. cerevisiae 31

Competing financial interest
None declared

Additional information
Supplementary information is available. Correspondence and requests for materials should be addressed to S.K..

Data availability
The data that supports the findings of this study are available from the corresponding author upon reasonable request.  Background signals are marked with asterisks. c, Merged high resolution mass spectra of three exemplary modifications (m 5 C, m 7 G and m 1 A) in total tRNA after stable isotope labeling of HEK 293 cells.

Salts, reagents, media and nucleosides
All salts, reagents and media were obtained from Sigma-Aldrich (Munich, Germany) at molecular biology grade unless stated otherwise. The isotopically labeled compounds 13 C5, 15  For Labeling in RPMI R0883, dialyzed FBS, glutamine, methionine, uridine and adenine were added in the same concentrations as for DMEM D0422.
HAP1 cells were either labeled using DMEM D0422 as described above or IMDM I3390 where FBS, glutamine, uridine, adenine and methionine were added in the same concentrations as used for DMEM D0422 medium. Cells grown in RPMI or IMDM medium were kept at 5% CO2 for proper pH adjustment.
Mouse embryonic stem cells (mESC) were cultured as recently reported 50 . Isotopically labeled compounds were added as described for regular cell culture labeling.
For biological replicates one culture was split into several flask at least 24h prior to experiment initiation.

Cell lysis and RNA purification
Cells were directly harvested on cell culture dishes using 1 mL TRI reagent for T25 flasks or 0.5 mL TRI reagent for smaller dishes. The total RNA was isolated according to the supplier's manual with chloroform (Roth, Karlsruhe, Germany). tRNA and 18S rRNA were purified by size exclusion chromatography (AdvanceBio SEC 300Å, 2.7μm, 7.8x300mm for tRNA and BioSEC 1000Å, 2.7μm, 7.8x300mm for 18S rRNA, Agilent Technologies) according to published procedures 31,51 . The RNA was resuspended in water (35 μL).

tRNA digestion for mass spectrometry
Total tRNA (300 ng) in aqueous digestion mix (30 μL) was digested to single nucleosides by using 2 U alkaline phosphatase, 0.2 U phosphodiesterase I (VWR, Radnor, Pennsylvania, USA), and 2 U benzonase in Tris (pH 8, 5 mM) and MgCl2 (1 mM) containing buffer. Furthermore, 0.5 µg tetrahydrouridine (Merck, Darmstadt, Germany), 1 µM butylated hydroxytoluene, and 0.1 µg pentostatin were added to avoid deamination and oxidation of the nucleosides. When quantification of dihydrouridine was intended tetrahydrouridine was omitted. After incubation for 2 h at 37 °C, 20 µL of LC-MS buffer A (QQQ) was added to the mixture and then filtered through 96-well filter plates (AcroPrep Advance 350 10 K Omega, PALL Corporation, New York, USA) at 3000 ×g and 4 °C for 30 min. A stable isotope labeled internal standard (SILIS) was produced in S. cerevisiae using 13 C and 15 N rich growth medium (Silantes, Munich, Germany, Product# 111601402) following recently described procedures 16,31 . 1/10 Vol. of SILIS was added to each filtrate before analysis by QQQ mass spectrometry. For each sample 10 µL were injected (~90 ng of sample tRNA)

High resolution mass spectrometry
The ribonucleosides were separated using a Dionex Ultimate 3000 HPLC system with a Synergi, 2.5 μm Fusion-RP, 100 Å, 100 x 2 mm column (Phenomenex®, Torrance, California, USA). Mobile phase A was 10 mM ammonium formate and mobile phase B was 80% acetonitrile containing 2 mM ammonium formate. Gradient elution started with 0 % B and increased to 12% B after 10 min and to 80% after 12 min. After 4 min elution at 80 % B and subsequently regeneration of starting conditions to 100% A after 5 min, the column was equilibrated at 100% A for 8 min. The flow rate was 0.2 mL/min and the column temperature 30 °C. High-resolution mass spectra were recorded by a ThermoFinnigan LTQ Orbitrap XL operated in positive ionization mode. The parameters of the mass spectrometer were tuned with a freshly mixed solution of uridine (10 μM). Capillary voltage was set to 20 V and capillary temperature to 300 °C. Sheath gas and sweep gas flow rate was set to 0, and auxiliary gas flow rate to 35. Source voltage was set to 4.0 kV and tube lens to 75 V.

QQQ mass spectrometry
For quantitative mass spectrometry an Agilent 1290 Infinity II equipped with a diode-array detector  (Table S2).
For separation a Synergi, 2.5 μm Fusion-RP, 100 Å, 100 x 2 mm column (Phenomenex®, Torrance, California, USA) at 35 °C and a flow rate of 0.35 mL/min was used in combination with a binary mobile phase of 5 mM NH4OAc aqueous buffer A, brought to pH 5.6 with glacial acetic acid (65 μL/L), and an organic buffer B of pure acetonitrile (Roth, Ultra LC-MS grade, purity ≥99.98).
The gradient started at 100% solvent A for 1 min, followed by an increase to 10% solvent B over 4 min. From 5 to 7 min, solvent B was increased to 40% and maintained for 1 min before returning to 100 % solvent A in 0.5 min and a 2.5 min re-equilibration period.

Calibration
For calibration, synthetic nucleosides were weighed and dissolved in water to a stock concentration of 1-10 mM. The calibration solutions ranged from 0.025 to 100 pmol for each canonical nucleoside and from 0.00125 pmol to 5 pmol for each modified nucleoside. Each calibration was spiked with 10% SILIS. The sample data were analyzed by the quantitative and qualitative MassHunter Software from Agilent. The areas of the MRM signals were integrated for each modification. The values of integrated MS signals from target nucleosides were set into relation to the respective MS signals of the respective isotope labeled SILIS nucleosides after Equation (1) to receive the nucleoside isotope factor (NIF): = Results from Equation 1 were plotted against the expected molar amount of nucleosides and regression curves were plotted through the data points. The slopes represent the respective relative response factors for the nucleosides (rRFN) and enable an absolute quantification. The principle is described in more detail in our published protocol 16 . The plotting of these calibration curves is done automatically by the quantitative MassHunter Software and should be checked manually for linearity.

Data Analysis
Molar amounts of nucleosides in samples were calculated after Equation In the case of NAIL-MS experiments, the different isotopomers were referenced to their respective labeled canonicals, so that original (unlabeled) modifications were referenced to original tRNA molecules and new (labeled) modifications were referenced to new tRNA molecules (see Equations (4) and (5)). Table 1 gives a summary of the calculations exemplarily for m 7 G.

Statistics
All experiments were performed at least 3 times (biological replicates) to allow student t-test analysis.
P-values of student t-test (unpaired, two-tailed, equal distribution) were calculated using Excel.