Pitfalls in RNA Modification Quantification Using Nucleoside Mass Spectrometry

Conspectus

In recent years, there has been a high interest in researching RNA modifications, as they are involved in many cellular processes and in human diseases. A substantial set of enzymes within the cell, called RNA writers, place RNA modifications selectively and site-specifically. Another set of enzymes, called readers, recognize these modifications which guide the fate of the modified RNA. Although RNA is a transient molecule and RNA modification could be removed by RNA degradation, a subclass of enzymes, called RNA erasers, remove RNA modifications selectively and site-specifically to alter the characteristics of the RNA. The detection of RNA modifications can be done by various methods including second and next generation sequencing but also mass spectrometry. An approach capable of both qualitative and quantitative RNA modification analysis is liquid chromatography coupled to mass spectrometry of enzymatic hydrolysates of RNA into nucleosides. However, for successful detection and quantification, various factors must be considered to avoid biased identification and inaccurate quantification. In this Account, we identify three classes of errors that may distort the analysis. These classes comprise (I) errors related to chemical instabilities, (II) errors revolving around enzymatic hydrolysis to nucleosides, and (III) errors arising from issues with chromatographic separation and/or subsequent mass spectrometric analysis.

A prominent example for class 1 is Dimroth rearrangement of m¹A to m⁶A, but class 1 also comprises hydrolytic reactions and reactions with buffer components. Here, we also present the conversion of m³C to m³U under mild alkaline conditions and propose a practical solution to overcome these instabilities. Class 2 errors–such as contaminations in hydrolysis reagents or nuclease specificities–have led to erroneous discoveries of nucleosides in the past and possess the potential for misquantification of nucleosides. Impurities in the samples may also lead to class 3 errors: For instance, issues with chromatographic separation may arise from residual organic solvents, and salt adducts may hamper mass spectrometric quantification. This Account aims to highlight various errors connected to mass spectrometry analysis of nucleosides and presents solutions for how to overcome or circumnavigate those issues. Therefore, the authors anticipate that many scientists, but especially those who plan on doing nucleoside mass spectrometry, will benefit from the collection of data presented in this Account as a raised awareness, toward the variety of potential pitfalls, may further enhance the quality of data.

Key References

• Kellner S.; Ochel A.; Thuring K.; Spenkuch F.; Neumann J.; Sharma S.; Entian K. D.; Schneider D.; Helm M.. Absolute and relative quantification of RNA modifications via biosynthetic isotopomers. Nucleic Acids Res. 2014, 42, e142.¹ This work details the biosynthetic production of a stable isotope labeled internal standard (SILIS) for MS quantification of RNA modifications.

• Cai W. M.; Chionh Y. H.; Hia F.; Gu C.; Kellner S.; McBee M. E.; Ng C. S.; Pang Y. L.; Prestwich E. G.; Lim K. S.; Babu I. R.; Begley T. J.; Dedon P. C.. A Platform for Discovery and Quantification of Modified Ribonucleosides in RNA: Application to Stress-Induced Reprogramming of tRNA Modifications. Methods Enzymol. 2015, 560, 29–71(2) This work proposes methods of RNA preparation and MS quantification of RNA modifications, including methods used within this Account.

• Jora M.; Borland K.; Abernathy S.; Zhao R.; Kelley M.; Kellner S.; Addepalli B.; Limbach P. A.. Chemical Amination/Imination of Carbonothiolated Nucleosides During RNA Hydrolysis. Angew. Chem., Int. Ed. 2021, 60, 3961–3966.³ This work explains how artifact formation can interfere with RNA modification analysis, proving the need for its systematic evaluation.

• Kaiser S.; Byrne S. R.; Ammann G.; Asadi Atoi P.; Borland K.; Brecheisen R.; DeMott M. S.; Gehrke T.; Hagelskamp F.; Heiss M.; Yoluc Y.; Liu L.; Zhang Q.; Dedon P. C.; Cao B.; Kellner S.. Strategies to Avoid Artifacts in Mass Spectrometry-Based Epitranscriptome Analyses. Angew. Chem., Int. Ed. Engl. 2021, 60, 23885–23893.⁴ This work reveals how artifacts can be mistaken for novel RNA modifications underscoring the importance of avoiding pitfalls within MS analysis of RNA modifications.

1. Introduction

It has long been known that RNA is the biological link between DNA and proteins.⁵ However, chemical modification of RNA is necessary, as the four so-called canonical nucleosides (adenosine, guanosine, cytidine, and uridine) are not sufficient to fulfill the myriad roles RNA plays within a cell’s biology. Thus, a large chemical variety of RNA nucleosides has evolved: So far, over 150 different nucleosides have been discovered in RNA across all domains of life,⁶ and the number is ever-growing.⁷⁻⁹ RNA modifications have been reported to be present in most types of RNA in various stoichiometries, with transfer RNA (tRNA) being the most extensively modified. For ribosomal RNA (rRNA), some positions are reported to be fully modified, while other positions are only partially modified.¹⁰ In tRNA^Asp of Schizosaccharomyces pombe, it was found that 5-methylcytidine (m⁵C) at position 38 is fully incorporated in the presence of the micronutrient queuine and substoichiometrically incorporated in its absence.¹¹

In addition to modification in native RNA, modification is also crucial for RNA-based therapeutics, ranging from siRNA and antisense oligonucleotides¹² all the way to the pandemic-resolving messenger RNA (mRNA) vaccines.¹³ For both native and synthetic RNA, the analysis of the modified nucleoside’s chemical structure as well as the accurate determination of absolute quantities and modification stoichiometries is important. Despite losing all sequence information, liquid-chromatography coupled to mass spectrometry (LC-MS/MS) of complete enzymatic RNA hydrolysates is the current gold standard, and various well-validated protocols for successful RNA analysis exist. Within the past decade, our lab has contributed to establish these protocols and found a whole range of sample preparation and analytical biases which have the potential to lead to inaccurate quantification results and wrong characterization of modified nucleosides. In this Account, we will give a detailed overview of the experimental workflow and highlight all potential biases that may falsify the analytical result. While we describe these biases from the perspective of subsequent LC-MS analysis of nucleosides, we want to emphasize that biases occurring during sample preparation will most likely lead to false results in other detection technologies such as next-generation sequencing, and we recommend consideration for all types of RNA modification analysis.

2. Workflow of LC-MS/MS Analysis of RNA Hydrolysates

The workflow can be subdivided into 7 major steps (Figure 1), which we briefly summarize here:

• Step 1: The process of RNA modification quantification starts at the stage of RNA isolation, e.g., from cells or tissue or from in vitro transcription/modification assays. The chemical environment of the RNA, such as organic solvent content, salts, and pH, dictates the success of the later stages.

• Step 2: The next step is the purification of the RNA of interest. Purification is possible by size or by sequence. Size separation is possible by gel electrophoresis¹⁴ or size exclusion chromatography^15,16 but also by exploiting the differential solubility during precipitation using, e.g., LiCl or spin columns. Purification by sequence requires the use of synthetic biotinylated DNA probes, which are reverse-complementary to the RNA of interest, and streptavidin-containing columns¹⁷ or streptavidin-coated magnetic beads.^7,18 This method might be biased as RNA modifications can interfere with the hybridization process and further by unspecific binding of highly abundant RNAs such as rRNA to the streptavidin-coated materials.

• Step 3: The enzymatic hydrolysis is done following either a two-step or one-pot protocol. The two-step protocol is similar to the original method published by Crain in 1990.¹⁹ Here, nuclease P1 (NP1) is used at slightly acidic pH to release nucleotides. NP1 is not capable of releasing, e.g., 2′-O-methylated nucleotides, but additional use of phosphodiesterase I (PDE1, historic snake venom phosphodiesterase) allows for complete release of modified nucleotides. The phosphate is removed in a second step using alkaline phosphatase (e.g., from calf intestine, CIP) at slightly alkaline pH.¹ The one-pot scheme utilizes Benzonase in conjunction with PDE1 for nucleotide release. Both enzymes accept pH 8, and thus the hydrolysis is performed under alkaline conditions in the presence of alkaline phosphatase and nucleosides are immediately released.² Commercial kits for RNA digestion are available, and their ability to release RNA modifications is further discussed in section 4.

• Step 4: After hydrolysis, enzymes might be removed by molecular-weight-cutoff filters, and a commonly known amount of internal standards is added for absolute quantification. For MS analysis, a stable isotope-labeled internal standard (SILIS) is considered best practice. In general, a SILIS is an isotopologue of the analyte containing the stable isotopes ¹³C, ¹⁵N, or D (²H) instead of the natural isotopes of carbon, nitrogen, or hydrogen (¹²C, ¹⁴N, or ¹H). The analyte and its respective SILIS have the same physicochemical properties and thus are equally affected by fluctuations in detection. The amount of the SILIS is constant across all samples and calibrant solutions, its signal is used to correct for potential signal fluctuation of the analyte, and thus accurate quantification of the analyte becomes possible. Several approaches for the acquisition of SILISs and their applications have been presented by us and others in the last years and are not further discussed.^1,2,20,21 If a SILIS is not available, isotope-labeled (deoxy)ribonucleosides or chemically similar modifications are often used as SILISs.^22,23 Here, the analyst must be highly confident in the sample preparation as small differences in, e.g., salt load can lead to deviations in the MS detection efficiency. If an online UV detector is available, the MS signals can be normalized to UV signals of canonical nucleosides.^9,24,25

• Step 5: For absolute quantification, the analyte of interest must be available in weighable quantities to produce calibrant solutions. For external calibration, 5–12 calibrant solutions of varying analyte concentration are prepared to ensure that the analyte signal of the sample is within the calibrated range. For external calibration and subsequent absolute quantification, the addition of SILIS to the calibrant solution is mandatory.¹ This step can be omitted for relative quantification and sample-to-sample comparison.

• Step 6: The LC-MS/MS instrument is equilibrated, and a method for detection of both the analytes of interest and their respective SILISs is loaded.²⁶ Afterward, the samples and the calibrant solutions are injected sequentially. Calibrant solutions are always injected from lowest to highest concentration.

• Step 7: Data analysis, by hand¹ or automated,²⁶ reveals the abundance of nucleosides, e.g., in femtomoles. For normalization, the absolute abundance of modified nucleosides can be divided by the absolute abundance of canonical nucleosides, and the results are ready for interpretation.

Figure 1.

General workflow of nucleoside analysis by LC-MS. The first step is RNA isolation from cultivated cells, tissues, or RNA synthesis, typically by phenol–chloroform extraction. Then, the RNA of interest is purified and enzymatically hydrolyzed into nucleosides. Optionally, molecular weight cutoff filtration can be performed to remove the enzymes. For quantification, SILIS (stable-isotope labeled internal standard) needs to be added to each sample and each of the serially diluted calibrant solutions. Next, samples and calibrant solutions can be analyzed on an LC-MS instrument. Finally, the obtained data can be analyzed and visualized.

Within this well-established workflow, we identified three classes of quantification biases, which require attention during sample preparation and analysis. Error class 1 comprises all errors connected to the chemical instability of modified nucleosides. Class 2 errors relate to the enzymatic hydrolysis step. Error class 3 revolves around all detection errors connected to chromatography and/or subsequent MS detection. An overview of the class 1, 2 and 3 errors found by us and others is given in Table 1. In this Account, we will summarize the consequences of all three error classes, and we present our strategies to circumvent them.

Table 1. Overview of Error Classes 1, 2 and 3.

cause	affected nucleoside(s)	ref
Class 1:Chemical Instability or Interaction
depurination at acidic pH	yW	(27)
amination in ammonium containing buffers	x5s2U	(3)
ring-opening at basic pH	ct6A, ms2ct6A, m7G	(28,29,34)
hydrolysis at basic pH	gluQ	(30)
isomerization	oQ	(31)
Dimroth rearrangement	m1A, m6A	(32)
deselenation	hypermodified uridine	(33)
chemical deamination	m3C, m3U	Figure 2
adsorption to filter	m6,6A, mcm5s2U, Am, i6A	Figure 3
eluting from filter	mimic of m6,6A	Figure 3
loss of RNA glycosylation	building block unknown	(34)
Class 2:Enzymatic RNA Hydrolysis
deaminases as contaminants	I, m3C, Cm	(2)
D as contaminant in THU	D
nuclease stability	Nm	(4)
Class 3:Sample Preparation with Impact on LC or MS Performance
salt adducts	all nucleosides
dimerization in MS	C and derivatives
residual organic solvent in RNA	Ψ, C, U	Figure 5

3. Chemical Properties of RNA Modifications

From a chemist’s perspective, nucleosides display electrophilic and nucleophilic sites and are able to accept or donate protons. This is true for the canonical nucleosides but even more so for modified nucleosides. Modified RNA nucleosides display a reactivity different from their canonical precursor, e.g., through the reactivity of a substituted or added atom (e.g., sulfur) or an added chemical group (e.g., threonyl) or through shifting the electron density of the nucleobase, e.g., by methylation. The chemical reactivity of nucleosides but also their adhesion to surfaces can lead to their under-quantification, complete loss, and/or misidentification. The reactions happen at the stage of either RNA isolation, purification, or hydrolysis and are very difficult to exclude.

3.1. Reactivity of Modified Nucleosides

The most prominent example is 1-methyladenosine (m¹A), which undergoes Dimroth rearrangement at mild alkaline pH and N⁶-methyladenosine (m⁶A) emerges (Figure 2A,C).³² This leads to an under-quantification of m¹A in the sample and, critically, to a false positive detection of m⁶A. This issue is commonly observed for human tRNA analyses,³⁵ where m¹A is found at position 58 of most tRNAs introduced by TRMT6/61A. Low quantities of m⁶A are observed in human tRNA by LC-MS/MS analysis, but no methyltransferase has been found. This makes a false positive occurrence of m⁶A as a class 1 error in human tRNA likely. Only recently, the new tRNA modification m^1,6A was described as a consequence of in vivo Dimroth rearrangement and renewed N1-methylation by TRMT6/61A.³⁶

Figure 2.

Chemical instabilities of nucleoside modifications. (a) Chemical structures of Dimroth rearrangement of 1-methyladenosine (m¹A) to N⁶-methyladenosine (m⁶A). (b) Chemical structures of 3-methylcytidine and 3-methyluridine. (c) Reaction of m¹A to m⁶A and m³C to m³U at various pH values at 21 °C for 24 h. (d) Time course of reaction from m¹A to m⁶A and m³C to m³U at 21 °C at pH 8. (e) Reaction progress after 2 h at pH 8 at 21 and 37 °C. (f) Hydrolysis of cyclic-N⁶-threonylcarbamoyladnosine (ct⁶A) into N⁶-threonylcarbamoyladenosine (t⁶A).²⁸ (g) Reaction of 2-thiolated compounds such as 5-methoxycarbonylmethyl-2-thiouridine (mcm⁵s²U) into the respective isocytidine derivative.³ (h) Potential solution to overcome quantification errors connected to instabilities using a second calibration (cal) or QC (quality control) samples alongside the regular calibration mixture (calmix).

Chemical rearrangement at an alkaline pH is also connected to difficulties in structure assignment. For example, the tRNA modification cyclic-N⁶-threonylcarbamoyladenosine (ct⁶A) was initially described in its noncyclic form as t⁶A due to alkaline ring-opening (Figure 2F).²⁸ This is also observed for other ct⁶-modified adenosine derivatives,²⁹ and covalent attachment of buffer components has been described for ct⁶A.³⁷ 7-Methylguanosine (m⁷G), an important modification not only in tRNA but also as the 5′-cap of both native mRNA and mRNA drugs, undergoes a ring-opening and decomposition at alkaline pH.³⁸

In summary, an alkaline pH is not suitable for some modifications, such as m¹A, m³C, m⁷G, and ct⁶A, and an acidic pH might be a favorable solution. However, some modifications such as the hypermodification wybutosine of tRNA are lost in acidic environments.²⁷ Due to the pH instabilities of some RNA modifications, we recommend careful consideration of solvent pH throughout the RNA handling process.

In addition to pH, deamination of both adenosine and cytidine modifications is similarly critical, as the reaction products are naturally occurring RNA modifications such as inosine or uridine modifications. Especially methylation of cytidine’s N3 leads to a reduced electron density at the neighboring C4 and thus a rapid deamination under alkaline conditions (Figure 2B). This reaction leads to a lowered signal of the tRNA modification 3-methylcytidine (m³C) and a false-positive signal of 3-methyluridine (m³U) occurs in these tRNAs (Figure 2C). Figure 2D shows the progress of the reactions over a time period of 24 h (tris (pH 8), 21 °C). The conversion is accelerated by higher temperatures, and incubation under conditions relevant to our hydrolysis protocol (37 °C, 2 h, pH 8) yielded product in the low single digit percent range (Figure 2E). Together with the Limbach lab, we described the inverse process of turning thiolated uridine into isocytidine (Figure 2G).³ In addition to native RNA modifications as a source of artifact formation, the environment of synthetic RNA drugs, namely, their lipid nanoparticle carriers, has been found to induce covalent RNA artifacts.³⁹ Thus, considerations concerning chemical reactivity must be extended to the emerging drug class of RNA therapeutics to ensure high product quality.

The reactions mentioned above are not limited to the sample but also occur within the calibrant solution. If the parallel quantification of, e.g., m¹A and m⁶A or m³C and m³U is required in an experiment, we recommend keeping separate calibration solutions containing m¹A and m³C alongside the calibration mix containing all the other nucleosides (Figure 2H). This separate solution can then be used as a quality control sample to estimate the error introduced into the calibration curve. In our experience, the m¹A and m³C calibrant solutions are stable for weeks to a few months at −20 °C, but not for years. Further, dissolving the nucleosides in mild acidic buffers (e.g., pH 5.3 ammonium acetate) can increase the stability of the solutions.

3.2. Adsorption of Modified Nucleosides during Filtration

Mass spectrometric analysis requires the use of purified samples, ideally containing only the analytes of interest. The removal of RNA hydrolysis enzymes reduces instrument contamination and enhances analytical sensitivity and instrument robustness. Commonly, MWCO filters (molecular weight cutoff) with a cutoff of ∼10 kDa are used for nucleoside analysis. They are available as single 1.5 mL tubes or 96 well-plate filtration plates. A hydrophobic material, poly(ether sulfone) (PES), which is regarded as inert, retains the enzymes, and most nucleosides pass through. We observe a loss of hydrophobic modifications such as N⁶,N⁶-dimethyladenosine (m^6,6A or m⁶₂A) or N⁶-isopentenyladenosine (i⁶A) from our native RNA hydrolysates due to the use of PES filters (Figure 3A). The same effect is observed by using synthetic standards (Figure S2A). Especially the loss of m^6,6A is critical as this known rRNA modification has been recently suggested as a quality control parameter to judge mRNA purity.¹⁴ Commonly, the addition of a SILIS prior to filtration is a suitable means to negate misquantification caused by analyte adhesion to filters. m^6,6A, however, appears to be especially problematic, as we unexpectedly find an unknown compound mimicking the retention time and MS/MS behavior of the m^6,6A SILIS after PES filtration (Figure 3B and Figure S2B, gen²-SILIS²⁶). Washing the PES filter with pure water prior to sample filtration partly removed the contaminant (Figure 3B). This m^6,6A mimic is highly abundant, is visible in the UV (Figure S2D), and leads to ion suppression of the native m^6,6A signal, which makes the use of the m^6,6A-SILIS mandatory for quantification. The use of a differently labeled m^6,6A SILIS from human cell culture partly avoids this challenge and allows improved quantification (Figure S2C). From our experience, two possible solutions to the problem remain: (1) filtration of samples is omitted if m^6,6A or i⁶A is of interest for the analysis, or (2) other filtration materials need to be used. For our analyses, we have validated composite regenerated cellulose (CRC) filters, commonly used for filtration of tryptic protein digests prior to LC-MS analysis. As shown in the heatmap of Figure 3A, CRC filters do not impact the quantification of modified nucleosides in our human rRNA and tRNA test system.

Figure 3.

Filtering samples prior to LC-MS analysis affects quantification of hydrophobic nucleosides. (a) Heatmap displaying the relative quantities of modified nucleosides after filtration of tRNA and rRNA hydrolysates in relation to unfiltered samples. Polyethersulfone (PES in tubes or well-plates, WP) filters and CRC filters (composite regenerate cellulose) were used. Cross hatches indicate the absence of the nucleoside in native RNA. (b) MS/MS peak of native m^6,6A (blue, m/z 296 → 164) and m^6,6A SILIS (red, m/z 313 → 176) of unfiltered, filtered and filter prewashed synthetic m^6,6A solutions. Peak areas are indicated below the respective chromatograms.

3.3. Avoiding Class 1 Errors

Class 1 errors occur due to the chemical reactivity and properties of modified nucleosides. As such, they can only be avoided if sample preparation is considered as a potential chemical reaction and the established protocols critically examined. If a change in the established protocol or its components is made, qualitative and quantitative validation is strongly recommended. Deviations from the protocol can lead to the true result of one modification, but other modifications might be wrong. In the case of class 1 errors, we thus come to the conclusion: There is no “one protocol fits all” approach for modified nucleoside analysis, and scientists must be aware of these chemistry-related biases in their biological data interpretation and choose the protocol carefully before analysis. Yet the possibility remains that even the published protocols might lead to currently unknown artifacts and critical data-evaluation is advised.

4. Enzymatic Hydrolysis of RNA Modifications

The chemical reactions outlined in the class 1 errors often occur in the step of enzymatic hydrolysis. Yet, the biological activity of the enzymes involved at this stage is also biased and comprise error class 2.

4.1. Enzyme Activities

For RNA hydrolysis, nucleases that cleave the phosphodiester bonds are needed. Nuclease P1 (and S1) is Zn²⁺-dependent, leaves a 5′-phosphate and a 3′-hydroxy group, and acts as an endo- and exonuclease on single-stranded DNA and RNA. Nuclease P1 (NP1) cannot cleave DNA or RNA containing 2′-O-methylated pyrimidines, 3-(3-amino-3-carboxypropyl)uridine (acp³U), m⁶A, or phosphorothioates (a nonbridging oxygen is replaced with a sulfur).⁴ The other common nuclease, Benzonase, is not able to cleave any phosphodiester with a 2′-O-methylated ribose and is further intolerant toward most base modifications. Also, note that Benzonase is unable to release single nucleotides as its products are oligonucleotides. Due to these limitations NP1 and Benzonase hydrolysis must be accompanied by phosphodiesterase 1 (PDE1) (Figure 4A). This self-made hydrolysis cocktail allows a quantitative release of modified nucleosides, even from structured RNAs. With commercial RNA hydrolysis kits, we experience an equally high release rate, e.g., for m¹A, m⁵U, Ψ, m⁷G, or mcm⁵s²U, but no release of, e.g., m⁵C, most 2′-O-methylations, m^2,2G, or acp³U.⁴ This leads to an under-quantification of modified nucleosides and even false-positive detection of the noncleaved dinucleotides as novel modified structures such as native RNA phosphorothiolation.⁴⁰ In terms of pH robustness, we have studied 19 common tRNA modifications and their releases in the one-pot reaction with Benzonase, PDE1, and CIP. In Figure 4B, we see no difference in quantification for these modifications. This means that (1) the 2 h of exposure to alkaline pH is not sufficient for quantitative Dimroth rearrangement and (2) the enzymes are highly robust with regards to pH variation and little quantification bias is introduced by pH inaccuracies.

Figure 4.

Impact of RNA hydrolysis on RNA modification quantification. (a) Abundance of canonical nucleosides [pmol] and modified nucleosides (per 1000 nucleotides = 10³ rN) using Benzonase (Benzo) or nuclease P1 (NP1) in the absence (−) or presence (+) of phosphodiesterase 1 (PDE1). (b) Absolute quantification of modified nucleosides using the one-pot scheme at various reaction pH. (c) Heatmap of modification abundances after hydrolysis without the deaminases pentostatin (Pento) or tetrahydrouridine (THU) as a fold-change compared to a hydrolysis in the presence of both pentostatin and THU.

4.2. Enzyme Contaminations

Enzymes are commonly purified from organisms and, as such, are prone to contamination with other enzymes. Depending on the enzyme’s source, more or fewer contaminations might be present. Deaminases are a common contamination, and their presence impacts the quantification results for adenosine and cytidine modifications. As a solution, the purine deaminase inhibitor pentostatin and the pyrimidine deaminase inhibitor tetrahydrouridine (THU) should be used.² As shown in the heatmap in Figure 4C, tRNA hydrolyzed with the one-pot cocktail in the absence of pentostatin shows a higher abundance of inosine, the deamination product of adenosine. Similarly, lower abundances of m⁵C and Cm are found in tRNAs hydrolyzed in the absence of THU. Interestingly, the respective reaction products, namely, m⁵U and Um, show no increased abundance in samples without THU (detailed quantification Figure S3). In this regard, THU is a useful addition to the hydrolysis cocktail. Yet, THU itself is another confounding factor as it is often substantially contaminated with dihydrouridine (D). For detection and accurate quantification of D from tRNA, THU must be omitted from both the sample hydrolysis and the SILIS hydrolysis. By omitting THU, quantification of D becomes possible, but m⁵C and Cm cannot be analyzed from the same preparation. If all three modifications are of interest, we recommend separate handling of sample aliquots to achieve true results.

4.3. Avoiding Class 2 Errors

Enzymes are biological products and as such are inherently prone to introduce bias into analysis if not carefully controlled, and batch-to-batch fluctuations are to be expected. It is common practice to employ a concentration of hydrolysis enzymes far exceeding the amount theoretically sufficient for the sample concentration. In this work, we suggest a protocol suitable for hydrolysis of as little as 10 ng and up to 10 μg without further adaptation.

With the high abundance of enzymes in the hydrolysis cocktail, we achieve a full release of modified nucleosides, even if the cleavage reaction is slowed by the modification. Enzyme activity depends not only on the substrate but also on an ideal environment of cations and pH. Thus, organic solvents and chelators of cations such as EDTA (ethylenediaminetetraacetic acid) must be avoided, and ammonium acetate/ethanol precipitation prior to sample hydrolysis might improve the analysis.

5. Technology-Related Errors

Regarding the instrumental analysis, organic solvents, salts, and insoluble particles within the sample must be avoided as they impact analysis. Class 3 errors emerge not only through sample impurities but also through inherent properties of the analytes.

5.1. Sample Preparation Impacts Chromatography

After RNA isolation, RNA is commonly precipitated in the presence of ethanol or acetone for the removal of salts and for concentration. After precipitation, the RNA pellet is dissolved in pure water for downstream LC-MS analysis. This step is critical as overdrying of RNA pellets might result in low solubility⁴¹ while remaining organic solvents impact the chromatography. Pseudouridine (Ψ) is one of the most polar modified nucleosides and elutes early in the common reverse phase chromatography setup. In Figure 5A, we display the impact of remaining ethanol in a sample on the retention time of pseudouridine. While we observe a symmetric peak at 0% ethanol, even low amounts of residual ethanol such as 1%, lead to a shift in retention time, peak splitting, and change in the signal ratio of analyte to SILIS (Figure 5B). Consequently, accurate quantification of Ψ and other early eluting nucleosides strongly depends on the sample solvent and purity. As shown in Figure 5C, the error of absolute quantification increases for Ψ, m³C, m¹A, ncm⁵U, and m⁷G with rising ethanol content with a deviation of approximately 40% from the expected value even after normalization to SILIS due to the broadening peak shape and thus inaccurate peak integration. For the quantification of these five modifications, special attention to the sample purity is advised. We expect similar effects from other residual solvents like acetone, isopropanol, acetonitrile, or methanol. Although late-eluting modifications are not impacted, the canonical nucleosides cytidine and uridine are. Because the normalization of modified toward canonical nucleosides is needed, the ethanol-induced misquantification of two canonicals will impact the absolute quantities of all modified nucleosides. A potential solution is the use of only guanosine or adenosine for quantification. Yet, for mRNA with the variable poly-A-tail, the normalization against guanosine is more accurate. Please note that in our setup the deaminase inhibitor pentostatin co-elutes with guanosine, which leads to ion-suppression of guanosine. In our hands, the quantification of guanosine is still possible due to the use of a SILIS, and thus we strongly recommend the use of a guanosine containing SILIS for mRNA analysis.

Figure 5.

Impact of residual ethanol (EtOH) on chromatography and accuracy of quantification. (a) Influence of varying levels of residual ethanol on the peak shape and retention time of pseudouridine (Ψ). (b) Observed relative abundance of early eluting modifications for varying concentrations of residual ethanol. (c) Observed relative abundance of later eluting modifications for varying concentrations of residual ethanol.

5.2. Salt Content during Sample Preparation Impacts Mass Spectrometry

Mass spectrometry weighs ions. Ions are made in modern ionization sources by either protonation or deprotonation. Nucleosides, and RNA in general, can also be charged with monovalent cations such as sodium (Na⁺m/z 23) and potassium (K⁺m/z 39). As shown in the upper mass spectrum of Figure 6 for m⁵C (M_w = 257 g/mol), the highest abundant peak arises at m/z 258 from protonation (M + H⁺). Yet, a clear peak with m/z 280 corresponding to (M + Na⁺) and even m/z 296 (M + K⁺) is detectable. The origin of monovalent cations can be from various sources: (1) from within the sample, e.g., samples precipitated with sodium acetate instead of ammonium acetate; (2) from the LC buffers if water or buffer salts of low quality are used; (3) from non-RNA samples run on the same instrument. Here, cations adsorb to all surfaces of the LC and MS and slowly attach to the RNA/nucleosides. It is important to be aware of these charged nucleoside species as they complicate spectrum interpretation, e.g., in the context of new structure discovery approaches, and disturb quantification in targeted analysis, which commonly focuses on the protonated compounds. Here, variation in salt content from sample to sample has the risk of under-quantification as the ratio of protonated and, e.g., sodium-charged nucleosides varies between samples. The use of SILISs mostly negates this risk, and thus we strongly recommend the use of SILISs for nucleoside quantification but also strict replacement of sodium buffers with ammonium.

Figure 6.

Observed m/z values for 5-methylcytidine m⁵C (M_w 527 g/mol) indicating adducts with salts and solvents and an isotope pattern.

5.3. Natural Carbon-13 Isotopes Impact Correct Identification

The discovery of modified nucleosides commonly uses the m/z as a first hint toward the occurrence of a novel nucleoside. As shown in the bottom mass spectrum of Figure 6, the mass analyzer separates the natural components by the abundance of their natural isotopes. Carbon-13 has a natural abundance of 1.1%, and sensitive mass analyzers easily detect the ¹³C₁- to ¹³C_N-signals next to the ¹²C-signal of any analyte. Similarly, sulfur-34 (natural abundance of 4.2%) might cause a +2 m/z signal in the spectrum. If full mass spectra are recorded, the natural isotopes can aid structure characterization of novel compounds. However, if targeted analysis using MS/MS transitions is performed, the information on the signal being an isotopologue of another nucleoside is lost. This bears the potential for misidentification of novel RNA modifications as recently observed for phosphorothioates in RNA.⁴⁰

5.4. Avoiding Class 3 Errors

Organic solvents are necessary for RNA precipitation, and we recommend a multiple step process for complete removal of solvent before drying the RNA pellet at room temperature. At first, the majority of solvent is removed using a 1 mL or 200 μL pipet until volumes below 10 μL are visible around the pellet. The tube should be left standing for ∼1 min to allow the discard of all solvent attached to the tube’s wall. The solvent is then removed with a 10 μL pipet before the pellet is dried at room temperature.

Salts, especially monovalent cations, must be avoided during sample preparation. If sodium buffers are required for, e.g., an in vitro reaction of the RNA, the RNA must be precipitated with ammonium acetate/ethanol before LC-MS/MS. Precipitation with lithium perchlorate is also well tolerated in our experience. Here again, a SILIS is key for precise and accurate quantification. Regarding the instrument, a dedicated LC-MS for RNA/nucleoside analysis is strongly recommended. If such an instrument is not available, we recommend purging the LC and MS source with boiling water for at least 1 h (safety concern: skin burns) to dissolve and reduce the cation load in the instrument.

6. Outlook

LC-MS/MS analysis of modified nucleosides and artifact determination is of high importance for both native RNA analysis^3,4 and analysis of RNA drugs.³⁹ Although we have determined and reported a variety of pitfalls within the current workflow, we anticipate the existence of numerous, yet undiscovered, biases for the analysis of RNA by LC-MS/MS. While we report the biases with their impacts toward mass spectrometric analysis, we want to emphasize that several of the biases, especially the chemical instabilities, will also lead to artifacts in other technologies such as third and next generation sequencing. We, and others, will continue our efforts to uncover pitfalls and thus improve the robustness and accuracy of RNA modification analysis technologies. We encourage all scientists working with RNA hydrolysates and LC-MS/MS analysis to carefully design all experimental steps, disseminate data, and thus uncover inconsistencies in the application. Thus, the reliability of the technology will ever increase, and epitranscriptome research and RNA drug quality assurance will greatly benefit.

Glossary

Abbreviations

CIP

calf intestinal phosphatase

Cm

2′-O-methylcytidine

CRC

composite regenerated cellulose

ct⁶A

cyclic N⁶-threonylcarbamoyladenosine

D

dihydrouridine

EDTA

ethylenediaminetetraacetic acid

EtOH

ethanol

i⁶A

N⁶-isopentenyladenosine

LC-MS/MS

liquid chromatography coupled tandem mass spectrometry

m¹A

1-methyladenosine

m³C

3-methylcytidine

m³U

3-methyluridine

m⁵C

5-methylcytidine

m^6,6A

N⁶,N⁶-dimethyladenosine

m⁶A

N⁶-methyladenosine

m⁷G

7-methylguanosine

mcm⁵s²U

5-methoxycarbonylmethyl-2-thiouridine

MWCO

molecular weight cutoff

mRNA

messenger ribonucleic acid

ncm⁵U

5-carbamoylmethyluridine

NP1

nuclease P1

PDE1

phosphodiesterase 1

PES

polyethersulfone

QC

quality control

rRNA

ribosomal ribonucleic acid

SILIS

stable isotope labeled internal standard

t⁶A

N⁶-threonylcarbamoyladenosine

THU

tetrahydrouridine

tRNA

transfer ribonucleic acid

Ψ

pseudouridine

Um

2′-O-methyluridine

Biographies

Gregor Ammann was born on the 3rd of February, 1994, in Feldkirch, Austria. He received his B.Sc. and M.Sc. degrees from Ludwig-Maximilians-University in Munich, Germany. He joined the group of Stefanie Kaiser first at Ludwig-Maximilians-University, Munich, and later at the Goethe University, Frankfurt, Germany, to work on his Ph.D. He is particularly interested in analyzing RNA modifications using various techniques, including LC-MS.

Maximilian Berg was born on April 11, 1995, in Wiesbaden, Germany. He completed the 2nd state examination in Pharmacy in 2019, followed by a Master of Science degree in Pharmaceutical Sciences at Goethe University Frankfurt. In 2021, he joined the group of Professor Stefanie Kaiser, where he is investigating the effects of drugs on the epitranscriptome in HEK cells.

Jan Felix Dalwigk was born on July 14th, 1999, in Munich, Germany. He completed a 2nd state examination in pharmacy and M.Sc. at Goethe University Frankfurt a.M. in 2021 and 2022, respectively. For his M.Sc. project, he joined Professor Stefanie Kaiser’s lab, where he investigated the effects of drugs on the epitranscriptomic modification patterns of human cell lines.

Stefanie Kaiser, born Kellner on June 6th, 1984, in Schorndorf, Germany, is a state-approved pharmacist from Ruprecht-Karls University Heidelberg. She joined Prof. Mark Helm at the Johannes-Gutenberg University Mainz for her Ph.D. After a 2-year postdoctorate fellowship with Prof. Peter Dedon at the Massachusetts Institute of Technology, she started her research group as Ludwig-Maximilians-University Munich. Since 2020, she is a full professor of Pharmaceutical Chemistry at the Goethe-University Frankfurt. Her current focus is on advancing MS-based technologies for the detection and quantification of RNA modifications to solve open questions in the field of RNA modification biology.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.accounts.3c00402.

• Additional experimental results and description of methods including critical instrument parameters (PDF)

Author Contributions

The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript. CRediT: Gregor Ammann conceptualization, data curation, investigation, methodology, validation, visualization, writing-review & editing; Maximilian Berg data curation, formal analysis, visualization; Jan Felix Dalwigk data curation, formal analysis; Stefanie Kaiser conceptualization, data curation, formal analysis, funding acquisition, resources, supervision, validation, writing-original draft, writing-review & editing.

This work was funded by the Deutsche Forschungsgemeinschaft (255344185-SPP 1784, 325871075-SFB 1309, KE1943/3-1) and the Horizon 2020 program (ID-952373, EPIVIRAL).

The authors declare no competing financial interest.

Special Issue

Published as part of the Accounts of Chemical Research special issue “RNA Modifications”.