# Chapter 132. RNA Modification Measurement and RNA Mass Spectrometry

## Scope Note

This chapter explains how RNA chemical modifications and RNA molecules are detected, localized, sequenced, quantified, calibrated, benchmarked, and reported. It owns antibody, chemical, enzymatic, sequencing, nanopore, chromatography, RNA-analyte mass-spectrometry, and isotope-tracing assay mechanisms. Its mass-spectrometry scope spans nucleoside analysis, bottom-up oligonucleotide mapping, intact-RNA measurement, and top-down RNA sequencing, including sample cleanup, separations, ionization, fragmentation, spectral assignment, database search, false-discovery control, and product-quality outputs. General terminology, claim tiers, evidence adjudication, and artifact reasoning hand off to [Chapter 46](chapter1043.md). Native ribonucleoprotein architecture hands off to [Chapter 59](chapter1054.md), protein-centric mass spectrometry to [Chapter 138](chapter1125.md), cross-mark biology to [Chapter 52](chapter1048.md), and manufacturing control systems to [Chapter 159](chapter1142.md).

## Executive Summary

RNA modification detection is difficult because most covalent changes do not create a simple new sequencing letter. A methyl group, isomerized uridine, deaminated adenosine, thiolated base, hypermodified tRNA nucleoside, or cap structure may alter antibody binding, chemical reactivity, enzyme behavior, reverse-transcription fidelity, ligation efficiency, chromatographic retention, mass-to-charge ratio, fragmentation spectra, or nanopore current. Each assay converts chemistry into a signal through a particular physical, chemical, enzymatic, or computational chain. The scientific claim must stay attached to that chain. An antibody peak is an enrichment signal over RNA fragments. A reverse-transcription stop is an enzyme behavior. A nanopore current anomaly is a native-molecule signal interpreted by a model. A nucleoside LC-MS/MS peak is strong evidence for a chemical species in a digested RNA pool, but the digestion usually removes transcript position.

The field therefore works best as a coordinated measurement toolbox rather than as a competition for one universal platform. Antibody and chemical assays can survey large transcriptomes or generate mark-selective signals. Enzymatic and sequencing assays can improve resolution when a modification produces a reproducible stop, mismatch, conversion, cleavage, ligation, or treatment-dependent difference. Direct RNA nanopore sequencing reads native RNA without cDNA synthesis or PCR and can, in favorable settings, connect modification-associated signals to transcript isoforms and individual molecules. Chromatography and mass spectrometry provide a complementary chemical axis. Complete digestion to nucleosides supports sensitive composition and abundance measurements but discards sequence position. RNase digestion to oligonucleotides can preserve local sequence and modification context. Intact and top-down analysis can measure whole-product mass and fragment selected RNA molecules, although increasing RNA length, salt adduction, overlapping charge states, and incomplete fragment ladders set practical limits. General promotion of these outputs through evidence tiers belongs to [Chapter 46](chapter1043.md); this chapter designs and validates the measurements.

The chapter uses six recurrent examples. N6-methyladenosine in mRNA illustrates antibody enrichment, motif-aware interpretation, writer perturbation, and the danger of treating broad peaks as single-nucleotide sites. Pseudouridine illustrates selective chemistry and reverse-transcription signatures. Inosine illustrates how a modification can masquerade as a sequence difference and why genomic variation and mapping artifacts must be excluded. Queuosine and other tRNA modifications illustrate dense modification patterns, hard-to-read stable RNAs, and how complementary RNases and fragment spectra can localize modifications. A synthetic single-guide RNA illustrates intact-mass and top-down sequence confirmation at approximately 100 nucleotides. Therapeutic mRNA illustrates why cap, poly(A)-tail, sequence, modification, truncation, and heterogeneity outputs often require a coordinated set of nucleoside, targeted digest, bottom-up, and intact measurements rather than one all-resolving spectrum.

Current consensus is conservative and pluralistic. Modification measurements should separate discovery, localization, chemical identity, stoichiometry, and biological interpretation. RNA mass spectrometry should likewise separate intact mass, sequence coverage, site localization, modification assignment, end-group assignment, and quantification. The strongest reports specify the RNA sample and purification, separation and ionization conditions, precursor charge and adduct handling, digestion and terminal-chemistry assumptions, fragment-ion series, search space and false-discovery procedure, standards and calibration, output uncertainty, software versions, performance limits, and known limitations. The recent reviews by Wetzel and Limbach (2016) and Santos and Brodbelt (2021), together with primary workflows for tRNA mapping, top-down guide-RNA analysis, mRNA sequence mapping, and target-decoy spectral search, anchor the mass-spectrometry treatment.

## Concept Inventory

- **RNA modification:** a covalent chemical change to a ribonucleotide in RNA. In common usage, the term includes base methylation, ribose methylation, isomerization, deamination, thiolation, acetylation, complex tRNA hypermodifications, cap modifications, and synthetic nucleotide substitutions. RNA editing is sometimes grouped with RNA modifications because it changes the chemical identity or pairing behavior of a base in RNA; editing is treated in detail in Chapters [50](chapter1046.md) and [51](chapter1047.md). RNA damage, oxidation, hydrolysis, and sample-handling artifacts can resemble modifications in assays but should not be assumed to be regulated biological marks.
- **Modification site:** a nucleotide position in a defined RNA molecule or transcript model that is inferred or shown to carry a modification. A site claim needs a coordinate system, strand, RNA class, transcript isoform or mature RNA definition, and an uncertainty statement. A modification stoichiometry is the fraction of RNA molecules carrying the modification at that site or within a specified RNA pool. Stoichiometry differs from peak height, enrichment fold change, global nucleoside abundance, or model probability. A modification map is a collection of candidate regions or sites; a map is not automatically a chemical inventory.
- **Direct detection:** that an assay measures a physical or chemical property of native modified RNA or a defined product of that RNA, such as mass, chromatographic retention, fragmentation, or nanopore current. Indirect detection means that a modification is inferred through enrichment, chemical treatment, enzymatic behavior, reverse-transcription errors, or altered coverage. The boundary is useful but not absolute. Direct RNA nanopore sequencing reads native RNA molecules, but chemical assignment is computational. LC-MS/MS directly identifies a modified nucleoside after digestion, but the original site may be lost.
- **Orthogonal validation:** testing the same modification claim with a method that has a different failure mode. An antibody-enriched N6-methyladenosine candidate can be checked by a site-resolving assay, mutation of the consensus motif, perturbation of a writer or eraser enzyme, and LC-MS/MS measurement of global N6-methyladenosine abundance. Orthogonality is weak if two assays share the same antibody, reagent, enzyme bias, computational model, or sample-preparation artifact.

## What to Know Before Reading This Chapter

Readers should know the basic structure of RNA: a 5′ to 3′ ribose-phosphate backbone, canonical bases, and many chemically modified nucleosides. Readers should also know the basic logic of RNA-seq: RNA is extracted, enriched or depleted, fragmented or captured, converted to a sequencing library or sequenced directly, aligned to a reference, and analyzed statistically. The new idea in this chapter is that a modification is a chemical state distributed across molecules. A single transcript position may be unmodified in some molecules and modified in others. The observed signal depends on RNA abundance, isoform structure, sequence context, RNA folding, protein binding, sample handling, and the assay's chemistry.

The chapter repeatedly asks five questions. First, what molecule or RNA pool was measured? Second, what physical or chemical signal was recorded? Third, what inference links the signal to a named modification? Fourth, what denominator supports localization or stoichiometry? Fifth, what independent evidence tests the claim? These questions prevent common overstatements such as treating enrichment as site localization, treating model probability as chemical proof, or treating detection as evidence of regulatory function.

## 132.1. Antibody-based and chemical modification detection

Antibody-based detection begins with a simple immunochemical idea: an antibody raised against a modified nucleoside or modified RNA epitope can enrich RNA fragments that contain or resemble that epitope. In a typical methylated-RNA immunoprecipitation workflow, RNA is fragmented, incubated with an antibody, captured on beads, washed, converted into a sequencing library, and compared with input RNA. The result is not a list of modified nucleotides. The immediate result is enrichment of fragments. Peak calling then identifies regions with more reads in the antibody-enriched library than in the input or control library. When the antibody is useful and the sample is well controlled, those peaks are candidates for modified regions.

The strength of antibody mapping is scale. Antibody enrichment helped make transcriptome-wide N6-methyladenosine mapping practical because N6-methyladenosine is common enough in mRNA to yield peaks, and many candidate peaks occur near sequence motifs and transcript regions that are biologically interpretable. Antibody mapping can also compare broad modification landscapes across conditions, cell types, or perturbations. Reviews by Zhang et al. (2022), Zhao et al. (2022), and Yu and Ueda (2026) treat antibody-based methods as useful discovery approaches when their resolution and specificity limits are acknowledged.

The weaknesses follow from the same immunochemical logic. An antibody may bind related chemical groups, RNA structures, contaminating molecules, or sequence contexts. Antibody lots may differ. Fragment size limits resolution: a 100-nucleotide enriched fragment may contain several candidate adenosines or cytidines. Highly expressed transcripts can dominate read counts, whereas low-abundance RNAs can be missed. RNA structure can hide or expose epitopes. Crosslinking-assisted antibody approaches can narrow the candidate position by creating truncation or mutation signatures near antibody binding, but crosslinking efficiency, antibody behavior, and reverse-transcription artifacts remain part of the measurement chain. A responsible antibody-based claim therefore distinguishes "an enriched region compatible with a modification" from "this nucleotide carries this modification at this stoichiometry."

> **Box 132.1. Reading an Antibody Peak Conservatively**
>
> **Reading an antibody peak conservatively**
>
> Ask four questions before calling a modified site. First, what RNA population entered the immunoprecipitation: polyadenylated RNA, total RNA, rRNA-depleted RNA, or a size-selected fraction? Second, what fragment length and transcript annotation define the peak? A broad peak may contain many candidate residues and may overlap isoforms or paralogs. Third, what controls showed that the antibody recognized the intended chemistry rather than a related nucleoside, RNA structure, protein-bound fragment, or abundant contaminant? Fourth, what independent evidence localizes and names the modification? Motif enrichment, writer perturbation, crosslink-induced signatures, targeted chemical assays, and LC-MS/MS each answer different parts of that chain. The careful conclusion is often "candidate modified region," not "this base is modified." Upgrade the wording only when the evidence also upgrades.

Chemical detection uses selective reactivity rather than immune recognition. A reagent may react with a modified base, fail to react with a modified base, install a bulky adduct, induce cleavage, change base-pairing behavior, protect a position from conversion, or create a reverse-transcription signature. Pseudouridine is a classic example because carbodiimide chemistry can create adduct-dependent reverse-transcription behavior under appropriate conditions. RNA cytosine methylation can be studied with bisulfite-like logic because unmodified cytidine and 5-methylcytidine differ in deamination susceptibility, although RNA bisulfite workflows can damage RNA and introduce conversion bias. Ribose methylation, cap modifications, thiolated bases, and other marks require different chemistries or indirect strategies; there is no universal reagent that makes all RNA modifications visible.

The central chemical lesson is that specificity is conditional. A reaction that is selective in a purified oligonucleotide may behave differently in folded rRNA, densely modified tRNA, protein-bound mRNA, or degraded clinical RNA. Conversion efficiency depends on pH, temperature, reagent concentration, solvent, time, accessibility, sequence context, and cleanup. Incomplete conversion produces false negatives. Off-target conversion and RNA damage produce false positives. A treated sample can also change library complexity by degrading some RNA classes more than others. Chemical modification detection should therefore include untreated controls, positive controls with known modified RNA, negative controls with matched unmodified RNA, conversion-efficiency estimates, and replicate modeling.

Figure 132.1 frames antibody and chemical assays as the first steps in an evidence ladder. Discovery, localization, identity, stoichiometry, and function are related but distinct claims.

![Figure 132.1. Signal-generation chains for RNA modification measurements](../assets/figures/chapter1120_figure1.png)

**Figure 132.1. Signal-generation chains for RNA modification measurements.** Each measurement converts an RNA chemical state into a platform-specific signal through sample selection, signal-generating chemistry or physics, acquisition, calibration, and computation. General evidence-tier adjudication hands off to [Chapter 46](chapter1043.md).

Table 132.1 compares the major method classes by what they measure. The table is meant to prevent a recurring interpretive error: methods with similar output labels, such as "modified site," may have very different underlying signals and failure modes.

**Table 132.1. Comparison of major RNA modification detection strategies.** RNA modification methods differ in what they measure. Antibody, chemical, enzymatic, nanopore, and LC-MS/MS methods should not be treated as interchangeable because their resolution, specificity, quantitative power, and artifacts differ.

| Method class | Measured signal | Typical resolution | Chemical specificity | Quantification strength | Main artifacts | Best use case |
| --- | --- | --- | --- | --- | --- | --- |
| **Antibody enrichment** | RNA fragment enrichment over input or control | Region-level, often tens to hundreds of nucleotides | Depends on antibody specificity and cross-reactivity | Weak for site stoichiometry; useful for relative enrichment | Antibody lot effects, nonspecific capture, RNA abundance, fragment-size bias | Transcriptome-scale discovery of candidate modified regions |
| **Crosslinking-assisted antibody mapping** | Antibody-linked truncation or mutation near a bound epitope | Potentially near-nucleotide but still antibody-dependent | Stronger than simple enrichment only when the antibody and crosslink signature are validated | Usually semi-quantitative | Crosslinking efficiency, reverse-transcription artifacts, antibody behavior | Narrowing antibody-enriched regions for follow-up validation |
| **Chemical derivatization sequencing** | Treatment-dependent stop, mismatch, cleavage, protection, or coverage change | Site-informative when conversion is selective and coverage is adequate | Conditional on reaction selectivity and conversion efficiency | Moderate if calibrated against known mixtures | Incomplete conversion, side reactions, RNA damage, structure-dependent accessibility | Mapping modifications with distinctive chemistry such as pseudouridine or selected cytosine marks |
| **Enzymatic or reverse-transcription signatures** | Stops, mismatches, deletions, ligation differences, nuclease cuts, or treatment-dependent read changes | Site-informative but enzyme- and context-dependent | Indirect; requires enzyme behavior to be benchmarked | Moderate when baseline error and treatment efficiency are modeled | Template structure, neighboring modifications, enzyme choice, buffer conditions, mapping artifacts | Detecting modifications that reproducibly perturb copying, ligation, cleavage, or end-processing enzymes |
| **Direct RNA nanopore sequencing** | Native RNA current, dwell-time, base-calling, or event-alignment differences | Long-read transcript context with model-dependent site inference | Native signal is direct; named chemistry is computationally inferred | Potentially per-read or site-level but requires calibration | Sequence context, pore chemistry, base-caller version, RNA structure, neighboring modifications | Linking modification-associated signals to isoforms, molecule context, tRNAs, viral RNA, or low-copy native molecules |
| **Nucleoside LC-MS/MS** | Chromatographic retention, precursor mass, product ions, and analyte abundance after complete digestion | RNA-pool level; site position is lost | High with authentic standards and diagnostic fragments | Strong with isotope standards and calibration curves | Co-eluting structural isomers, sample-induced analyte conversion, matrix effects, carryover, digestion bias | Chemical confirmation and global quantification of modified nucleosides or product composition |
| **Oligonucleotide LC-MS/MS** | Mass and fragmentation of controlled digestion products | Fragment-level and sometimes site-localizing | High when fragments are unique and spectra are interpretable | Targeted and quantitative when standards are available | Incomplete digestion, incomplete reference set, salt adducts, neutral-loss-dominated or ambiguous fragments, co-elution, limited standards | Confirming candidate sites or modified fragments in purified RNAs |

Concrete examples show why controls must be matched to chemistry. An N6-methyladenosine antibody peak near a stop codon in polyadenylated RNA may be biologically plausible because many mRNA N6-methyladenosine sites occur in recognizable motifs and transcript regions, but the peak still needs site-level support. A pseudouridine signal from a carbodiimide-treated sample may be strong when the same position is absent in an untreated control and present in a synthetic positive control, but RNA folding or nearby damage can alter reverse-transcription behavior. An apparent 5-methylcytidine call from bisulfite-treated RNA requires evidence that unmodified cytidines converted efficiently and that resistant cytidines did not simply escape because the RNA was structured or degraded.

Boundary cases are common. Some modifications are too rare for broad enrichment to be reliable. Some are present in highly abundant rRNA or tRNA contaminants and can appear in libraries intended for mRNA. Some antibody peaks may reflect RNA-binding proteins, fragment-end biases, or nonspecific capture. Some chemical reactions reveal accessibility as much as chemistry. In practice, antibody and chemical methods are best used as hypothesis generators or site-localizing tools within a validation plan, not as complete standalone inventories of the epitranscriptome.

Sample quality is part of the antibody or chemical assay, not a separate housekeeping issue. RNA oxidation, alkaline hydrolysis, prolonged storage, freeze-thaw cycles, metal contamination, endogenous nuclease activity, and incomplete removal of proteins or small molecules can all change apparent modification signals. Purification choices also matter. Poly(A) selection removes most tRNA and rRNA but enriches for mRNA and some long noncoding RNAs; ribosomal RNA depletion keeps more nonpolyadenylated RNA but can leave stable RNA fragments; size selection can remove small mature RNAs or degradation products. A modification map should therefore describe not only the detection chemistry but also the RNA population that entered the assay. Without that description, a signal assigned to "the transcriptome" may actually represent a biased subset of molecules that survived extraction, selection, and treatment.

## 132.2. Enzymatic and sequencing-based detection strategies

Enzymatic detection uses proteins as chemical sensors. Reverse transcriptases, ligases, nucleases, demethylases, methyltransferases, reader proteins, and polymerases can respond differently to modified and unmodified RNA. In sequencing-based assays, that response is converted into a readout: a stop, mismatch, deletion, readthrough change, coverage drop, cleavage site, ligation preference, barcode transfer, or difference between treated and untreated libraries. The method is indirect, but it can be highly informative when the enzyme behavior is reproducible and calibrated.

Reverse-transcription signatures are the most familiar enzymatic signals. A reverse transcriptase copying a modified RNA template may stop before a modified base, misincorporate opposite it, skip it, or read through with altered efficiency. The outcome depends on the enzyme, template sequence, neighboring modifications, secondary structure, magnesium concentration, temperature, dNTP concentration, and library protocol. A stop signature that works for one modification in one RNA class is not a universal chemical barcode. For this reason, assay developers often compare multiple reverse transcriptases and use synthetic modified templates to measure baseline error, position-specific signal, and context dependence.

Inosine illustrates both the power and danger of sequencing signatures. Adenosine-to-inosine editing changes pairing behavior because inosine is read like guanosine during reverse transcription and sequencing. An edited adenosine can therefore appear as an A-to-G difference in cDNA reads. That signal can be biologically meaningful, but the same pattern can be created by genomic variants, paralogous mapping, RNA damage, sequencing error, or alignment artifacts. A high-quality inosine or editing study uses matched genomic DNA when possible, filters known polymorphisms, models mapping ambiguity, and considers strand and transcript context. The broader lesson is that a mismatch is a measurement, not a conclusion.

Demethylase-assisted and writer-perturbation strategies use enzymes or genetic perturbations to create a differential signal. If removal of a methyltransferase reduces a candidate modification signal, the result supports a relationship between the enzyme and the mark. If an eraser enzyme treatment removes a signal in vitro, the result can strengthen chemical assignment. The logic is useful but not definitive. Writer loss can change RNA abundance, stress state, cell-cycle distribution, differentiation, or metabolism. Eraser treatment may be incomplete or off-target. Rescue experiments, catalytic-dead controls, dose-response behavior, and independent chemistry make the inference stronger.

Nuclease- and ligase-based assays are especially important for structured stable RNAs and ribose modifications. Some nucleases cut less efficiently near ribose-methylated residues or modified bases. Some ligases show terminal nucleotide preferences that can be exploited or must be controlled. Cap-analysis workflows can use enzymes that distinguish capped, decapped, phosphorylated, or hydroxylated RNA ends before sequencing. These approaches link modification detection to RNA end chemistry, processing state, and library construction. They are powerful precisely because enzymes are selective, but their selectivity creates bias that must be measured.

Sequencing-based assays also need denominators. A mutation fraction is the number of reads carrying a treatment-specific mismatch divided by all reads that confidently cover the position. A stop fraction is the number of terminated reads divided by the templates that could have reached the site. An enrichment fold change is antibody-captured signal relative to input. A differential signal is a change relative to a matched condition. Each denominator can be distorted by RNA expression, isoform usage, degradation, PCR duplication, mapping quality, base quality, and read length. A site with low coverage may show an impressive percentage that is statistically meaningless.

Table 132.2 summarizes representative chemical and enzymatic signatures. The table should be read as a logic guide, not as a protocol recipe, because each method family has variants that require method-specific validation.

**Table 132.2. Chemical and enzymatic signatures used in modification-aware sequencing.** Chemical and enzymatic assays convert modification state into stops, mismatches, cleavage, ligation differences, treatment-dependent changes, or coverage shifts. Each signature requires calibration because enzyme behavior and chemical conversion depend on context.

| Signature type | Example modification or context | Immediate assay output | Controls required | Main overinterpretation risk |
| --- | --- | --- | --- | --- |
| **Chemical adduct or altered reverse transcription** | Pseudouridine-sensitive carbodiimide-style workflows | Treatment-dependent stops, mutations, or readthrough changes | Untreated RNA, synthetic modified and unmodified templates, conversion-efficiency checks | Treating a reagent-induced RT pattern as a universal pseudouridine barcode |
| **Differential chemical conversion or protection** | 5-methylcytidine-style cytosine conversion logic | Converted and resistant cytidines in sequencing reads | Conversion controls, RNA integrity checks, structure-disruption controls | Calling resistant cytidines modified when inaccessible or damaged RNA escaped conversion |
| **Reverse-transcription mismatch** | Inosine read as guanosine during cDNA sequencing | A-to-G differences relative to reference sequence | Matched genomic DNA when possible, variant filtering, strand-aware mapping | Confusing editing or modification with genomic variation, paralogous mapping, or sequencing error |
| **Reverse-transcription stop or deletion** | Dense tRNA, rRNA, ribose methylation, or other modified-template contexts | Position-specific stops, deletions, or drop-offs | Multiple RT enzymes when feasible, known positive sites, unmodified controls, coverage thresholds | Assuming any RT obstacle is a named modification rather than structure or damage |
| **Enzyme perturbation or treatment difference** | Demethylase treatment, writer depletion, or catalytic perturbation | Loss, gain, or redistribution of candidate signal | Catalytic-dead controls, rescue, dose response, independent chemistry | Treating an indirect cellular response as direct evidence that one enzyme installs one site |
| **Nuclease, ligase, or end-chemistry gate** | Ribose methylation, caps, small-RNA ends, processed fragments | Cleavage differences, ligation bias, or end-selective library entry | End-state standards, enzyme specificity controls, input normalization | Mistaking library accessibility or end chemistry for internal base modification |

The strongest sequencing-based modification studies report the full chain from sample to call. That chain includes RNA extraction and enrichment, fragmentation or capture, treatment conditions, enzyme identity, library protocol, sequencing platform, alignment strategy, duplicate handling, baseline error model, candidate-site model, replicate concordance, multiple-testing control, and validation. If a method claims single-nucleotide resolution, the evidence should show that nearby candidate residues can be distinguished. If a method claims stoichiometry, the evidence should show how signal was calibrated against known modified fractions.

Several important method families need more local citation curation before this chapter can be considered publication-ready. Curation-deferred source need: add verified primary references for landmark N6-methyladenosine antibody maps, crosslinking-assisted antibody methods, pseudouridine carbodiimide sequencing, RNA bisulfite approaches, ribose-methylation mapping methods, A-to-I editing detection standards, cap-specific sequencing and analytical workflows, and newer enzyme-assisted direct modification assays. The absence of these citations in the chapter-local bibliography is a source-curation need, not a claim that the methods are unsupported in the broader literature.

RNA end chemistry deserves special attention because many sequencing libraries begin or end with ligation. A mature small RNA may carry a 5′ monophosphate, 5′ triphosphate, cap, hydroxyl end, cyclic phosphate, or other processing-dependent end. A 3′ end may carry hydroxyl, phosphate, cyclic phosphate, aminoacylation, tailing, or damage. Enzymatic pretreatments can make some ends ligatable and leave others invisible. A cap-analysis assay, a small-RNA library, and a degradation-fragment assay may therefore report different RNA populations even when they start from the same biological sample. This is not merely a library-preparation artifact; end chemistry is itself part of RNA biology and must be separated from internal base modification when interpreting sequencing evidence.

## 132.3. Nanopore base-calling and single-molecule inference

Direct RNA nanopore sequencing reads native RNA molecules by threading them through a nanopore and recording ionic current. A motor protein controls movement, and a sensor records current levels as successive RNA k-mers occupy the pore. The raw output is not a conventional sequence file. It is a time series of current values, dwell times, and signal shapes. Base-calling models convert those signals into RNA sequence, and modification-calling models or statistical tests ask whether the observed signal differs from the expected signal for unmodified RNA. Because the RNA molecule is native, modifications can influence the signal without reverse transcription or PCR.

This native measurement is the major attraction. Direct RNA sequencing can connect long transcript context, poly(A) tail information, isoform structure, and modification-associated signal on the same molecule more directly than fragmented short-read assays. It can also interrogate RNAs that are difficult to copy faithfully by reverse transcription. Stephenson et al. (2022) used single-molecule nanopore sequencing to study RNA modifications and structure, Zhao et al. (2022) reviewed direct RNA sequencing for modification detection, Furlan et al. (2021) reviewed computational methods for nanopore modification detection, Sun et al. (2023) applied direct RNA sequencing to queuosine and queuosine precursors in tRNAs, and Hewel et al. (2025) provides a recent local-reference example linking direct RNA sequencing with transcriptome assessment and modification tracking in medical contexts.

Nanopore modification inference is nevertheless computationally demanding. A pore senses several neighboring nucleotides at once, so a change in current cannot usually be assigned to a single base without modeling. Sequence context, RNA secondary structure, motor speed, pore chemistry, base-caller version, RNA damage, neighboring modifications, and transcript isoforms can all alter the signal. A modified base may cause base-calling errors, altered dwell time, shifted current, or subtler distributional changes. Different algorithms may use event-level alignment to a reference, per-read signal comparison, supervised classifiers trained on synthetic modified RNA, anomaly detection against an unmodified model, or differential analysis between biological conditions.

The phrase "direct RNA sequencing detects modifications directly" is therefore only partly true. The platform directly measures native RNA signal. The chemical label assigned to that signal is an inference. A current shift compatible with N6-methyladenosine, pseudouridine, 5-methylcytidine, inosine, queuosine, or another mark becomes a strong modification call only when the model has been trained or calibrated for that modification and context, and when matched controls rule out sequence, structure, isoform, and damage effects. Synthetic modified and unmodified oligonucleotides are useful controls because they provide known chemistry. Cellular writer knockouts or knockdowns are useful biological controls, but they can have indirect effects. LC-MS/MS can confirm global chemical presence, but it may not prove the exact nanopore site.

> **Box 132.2. Native Signal Is Not a Chemical Name**
>
> **When a nanopore read becomes a modification claim**
>
> A direct RNA read supplies a native current trace, dwell-time pattern, base-called sequence, and long-read molecule context. A chemical call requires more. The analysis must define the unmodified expectation for the same sequence context, show that the model recognizes the named modification rather than a generic anomaly, and test whether the signal persists across pore chemistry, base-caller, and alignment choices. Synthetic modified and unmodified RNAs help establish chemical truth. Biological perturbations test whether the candidate signal follows a writer, eraser, or pathway in cells. Orthogonal assays such as targeted sequencing or LC-MS/MS check whether the inferred chemistry exists in the sample. Per-read scores are most useful when calibrated against known mixtures; otherwise, aggregate site-level evidence may be stronger than confident molecule-by-molecule labeling.

Figure 132.2 follows the nanopore measurement chain from molecule to current trace to base call to modification model. The key teaching point is that a native-molecule signal still needs a model and a validation set.

![Figure 132.2. Direct RNA nanopore sequencing and modification inference](../assets/figures/chapter1120_figure2.png)

**Figure 132.2. Direct RNA nanopore sequencing and modification inference.** Native RNA molecules pass through a nanopore and generate ionic-current traces. Base-calling, event alignment, expected-signal modeling, and modification classification convert traces into sequence and candidate modification calls. Matched controls, synthetic standards, perturbations, and orthogonal assays are needed to convert current anomalies into confident chemical claims.

Single-molecule language requires special care. Direct RNA sequencing produces reads from individual molecules, but a per-read modification probability is not always a reliable statement that a specific molecule definitely carries a specific mark. Per-read calls can be useful for studying co-occurrence, allelic differences, isoform-specific modification, poly(A)-tail relationships, or viral RNA heterogeneity, but those claims demand stronger calibration than aggregate site-level tests. Aggregating reads across a site can detect a reproducible signal even when individual reads are too noisy for confident molecule-level classification.

tRNA and rRNA show both promise and difficulty. Stable RNAs carry dense modification patterns that strongly affect reverse transcription and can be biologically important. Direct RNA sequencing can, in principle, read native stable RNAs and reveal modification-dependent signals without forcing cDNA synthesis through difficult templates. In practice, dense nearby modifications create overlapping signal effects, and mature tRNAs are short, structured, end-processed, and often difficult to map uniquely. Sun et al. (2023) is useful here because queuosine and its precursors provide chemically defined tRNA examples where direct RNA sequencing can be evaluated against known modification biology and complementary analytical evidence.

Nanopore data also expose coordinate and transcript-model problems. A signal assigned to a genomic coordinate may belong to one transcript isoform, a retained intron, an overlapping transcript, a mature processed RNA, or a paralogous gene. Long reads can reduce ambiguity by spanning isoform structures, but coverage may be uneven and read accuracy remains lower than many short-read platforms. For viral RNA, the same issue appears as subgenomic RNAs, replication intermediates, defective genomes, and sequence diversity. For clinical samples, degraded RNA, low input, and mixed cell populations can change both signal quality and interpretation.

The most reliable nanopore modification studies make their assumptions testable. They report pore and chemistry versions, base-caller versions, model training data, alignment parameters, read filters, control samples, known positive and negative sites, performance metrics, and whether the claim is site-level, transcript-level, or molecule-level. They avoid presenting a model score as chemical proof unless the model has been benchmarked for the named modification in an appropriate context.

Benchmark design is one of the limiting problems for nanopore modification calling. Fully synthetic RNAs can encode known modified and unmodified positions, but synthetic molecules may lack native folding, protein occupancy, damage patterns, neighboring modifications, and isoform complexity. Cellular knockout or knockdown samples preserve biological context, but they rarely produce a perfect unmodified ground truth because writer enzymes can have indirect effects and redundant pathways may remain. In vitro transcribed controls can provide matched sequence, but incorporation efficiency and transcript purity must be verified. A useful benchmark panel should therefore include more than one control type when possible: synthetic standards for chemical truth, biological perturbations for cellular relevance, and orthogonal analytical chemistry for global abundance.

## 132.4. LC-MS/MS, nucleoside quantification, and isotope tracing

Chromatography and mass spectrometry answer a different question from most sequencing assays. Instead of asking where reads accumulate or where an enzyme stops, LC-MS/MS asks what chemical species are present and how much of each species is present under specified analytical conditions. Liquid chromatography separates analytes by interactions with a column and solvent gradient. Mass spectrometry ionizes molecules and measures mass-to-charge ratios. Tandem mass spectrometry selects ions for fragmentation and records product ions that provide structural information. With authentic standards, retention time, precursor ion mass, product ion spectra, and calibration can provide strong chemical identification and quantitative confidence.

A mass shift is evidence about composition, not automatically a chemical name. Most post-transcriptional modifications change the mass of a canonical nucleoside, but pseudouridine is a structural isomer of uridine and therefore does not create a precursor-mass difference. Other structural isomers can share an elemental composition and exact mass even when their bond arrangement and biological meaning differ. Precursor mass narrows the candidate set; retention against an authentic standard, diagnostic product ions, and sometimes an additional separation dimension are needed to distinguish isomers. Wetzel and Limbach (2016) use this boundary to explain why mass spectrometry is chemically direct but not inference-free.

Nucleoside LC-MS/MS is the most common global strategy. RNA is purified, digested to nucleosides, spiked with internal standards when available, separated by chromatography, and analyzed by mass spectrometry. The output can be a ratio such as N6-methyladenosine per adenosine, pseudouridine per uridine, 5-methylcytidine per cytidine, or a modified nucleoside per microgram of RNA. This is powerful for comparing global modification abundance after writer perturbation, verifying the presence of a rare nucleoside in a purified RNA fraction, monitoring disease-associated changes, and testing synthetic RNA product composition. Zhang et al. (2022) supports LC-MS/MS as a core technology in the RNA modification detection landscape.

Digestion and cleanup must also preserve the chemical species being claimed. A modification can hydrolyze, oxidize, rearrange, or be lost during extraction and enzymatic processing. A historically important example is cyclic N6-threonylcarbamoyladenosine in tRNA, which can undergo ring opening during sample preparation and appear as the noncyclic product. The resulting peak is real analytical signal but may describe processing chemistry rather than the original in vivo nucleoside. Process blanks, standards added before digestion, recovery experiments, and comparisons of preparation conditions are therefore part of identification, not merely quantification.

The cost of complete digestion is loss of position. Once an mRNA population is converted into individual nucleosides, the analyst may know that N6-methyladenosine is present but not which transcript or site carried it. This distinction is often misunderstood. A global reduction in N6-methyladenosine after a writer knockout supports a global chemistry effect, but it does not identify every affected transcript. A nucleoside signal for pseudouridine in a purified RNA fraction confirms chemical presence in that fraction, but it does not localize the modified uridine unless the fraction contains a single RNA or additional site-localizing evidence is provided.

Oligonucleotide LC-MS/MS can preserve partial positional information. Instead of complete digestion to nucleosides, RNA can be cleaved with nucleases that produce predictable fragments. A modified oligonucleotide can then be separated, weighed, fragmented, and mapped to a sequence. If the fragment is unique and contains one candidate residue, localization can be strong. If the fragment contains several candidate residues, maps to multiple RNAs, or co-elutes with another molecule, localization is weaker. Fragmentation spectra, retention behavior, targeted standards, isotope labeling, and alternative digestion enzymes can improve confidence. The approach is technically more demanding than nucleoside quantification because RNA fragments can be large, charged, structured, salt-adducted, incompletely digested, and isobaric.

Figure 132.3 contrasts total nucleoside analysis with site-informative oligonucleotide analysis. The visual should emphasize that chemical specificity and positional information are gained or lost at different steps.

![Figure 132.3. LC-MS/MS paths for RNA modification analysis](../assets/figures/chapter1120_figure3.png)

**Figure 132.3. LC-MS/MS paths for RNA modification analysis.** Complete digestion to nucleosides supports sensitive global quantification with internal standards but loses transcript position and can transform labile analytes during preparation. Precursor mass alone does not distinguish uridine from pseudouridine or other structural isomers. Controlled digestion to oligonucleotides can preserve partial sequence context and support reference-guided site localization, but interpretation depends on fragment uniqueness, candidate sequences, digestion completeness, chromatography, fragmentation, and standards.

Quantitative LC-MS/MS depends on analytical chemistry vocabulary. An internal standard is a known molecule added in a known amount to correct for extraction, injection, ionization, and instrument variation. A stable-isotope-labeled internal standard has the same chemical behavior as the target analyte but a shifted mass, making it especially useful when available. A calibration curve relates measured signal to amount across a defined range. The limit of detection is the lowest amount distinguishable from background. The limit of quantification is the lowest amount that can be measured with acceptable precision and accuracy. Matrix effects occur when salts, proteins, detergents, buffers, nucleases, or co-eluting molecules change ionization efficiency. Carryover occurs when analyte from a previous run contaminates a later injection.

Stable-isotope standards can be prepared one analyte at a time or as multi-analyte pools. Synthetic labeled nucleosides provide well-defined calibrants, whereas RNA produced in isotopically labeled culture can yield a broader collection of labeled modified nucleosides through the organism's own modification machinery. The pooled strategy expands coverage when individual standards are unavailable, but each isotopologue's identity, enrichment, concentration, and response still require characterization. A labeled biological pool is not automatically a traceable quantitative standard merely because its peaks are mass-shifted.

Table 132.3 defines the quantitative terms that should appear in serious mass-spectrometry reports. These terms are not bureaucratic details; they determine whether a modification abundance claim is qualitative, semi-quantitative, or quantitative.

**Table 132.3. Quantitative terms for LC-MS/MS modification reports.** Quantitative mass-spectrometry claims require analytical vocabulary. Internal standards, calibration curves, detection limits, quantification limits, matrix effects, carryover checks, and peak-integration criteria determine whether a modification measurement is qualitative, semi-quantitative, or quantitative.

| Term | Definition | Why it matters | Reporting minimum |
| --- | --- | --- | --- |
| **Internal standard** | Known molecule added in a known amount before or during analysis | Corrects for extraction, injection, ionization, and instrument drift | Identity, amount, addition step, and whether it matches the analyte chemically |
| **Authentic chemical standard** | Independently characterized analyte used to match retention and product-ion behavior | Distinguishes exact-mass or nominal-mass candidates that precursor mass alone cannot separate | Source, purity, retention-time tolerance, product ions, and whether the standard experienced the same preparation and matrix |
| **Stable-isotope-labeled standard** | Isotopically shifted version of the target or a close surrogate | Provides the strongest routine correction for analyte-specific losses and ionization | Isotope label, supplier or synthesis route, purity, concentration, and monitored transition |
| **Calibration curve** | Relationship between analyte amount and measured signal across a defined range | Determines whether reported abundance is quantitative rather than only detectable | Concentration range, curve model, weighting, replicate points, and acceptance criteria |
| **Limit of detection** | Lowest amount distinguishable from background | Makes negative or trace-level claims interpretable | Calculation method, matrix used, signal threshold, and analyte identity |
| **Limit of quantification** | Lowest amount measured with acceptable precision and accuracy | Defines the range where abundance estimates can be trusted | Calculation method, precision and accuracy criteria, and whether samples fall inside the range |
| **Matrix effect** | Change in ionization or recovery caused by salts, buffers, proteins, detergents, nucleases, or co-eluting molecules | Can make equal analyte amounts produce unequal signals across samples | Sample cleanup, matrix-matched standards or dilution tests, and quality-control behavior |
| **Carryover** | Residual analyte from one injection appearing in a later run | Can create false low-level positives for abundant or sticky modified nucleosides | Blank injections, run order, wash method, and carryover acceptance threshold |
| **Peak-integration criteria** | Rules for defining and quantifying chromatographic peaks | Small boundary choices can dominate low-abundance estimates | Retention-time window, fragment or transition criteria, manual edits, and quality-control examples |
| **Diagnostic fragment or transition** | Product ion, fragment pattern, or monitored transition used to identify an analyte | Helps separate true analyte from isobars, adducts, and in-source fragments | Precursor mass, product ions, collision settings, retention time, and standard match when available |

> **Box 132.3. Minimum Evidence for a Quantitative LC-MS/MS Claim**
>
> **Minimum evidence for a quantitative LC-MS/MS claim**
>
> A mass-spectrometry peak becomes a quantitative RNA modification claim only when the report defines five linked items. The analyte must be named by retention time, precursor mass, product ions or transitions, and preferably an authentic standard. The denominator must be explicit: modified nucleoside per canonical nucleoside, per microgram of RNA, per purified RNA class, or per manufactured product lot. The calibration must cover the observed range and include an internal standard, ideally a stable-isotope-labeled form of the same analyte. Quality controls must show acceptable blanks, carryover, matrix effects, digestion completeness, and peak integration. Limits of detection and quantification must be stated so negative or low-level results can be interpreted. A calibrated nucleoside ratio is strong chemical evidence for an RNA pool, but it is not site stoichiometry unless positional information is preserved by an additional method.

Stable-isotope tracing extends mass spectrometry from measurement to kinetics. Cells, organisms, or in vitro systems can be supplied with labeled methyl donors, nucleosides, amino acids, carbon sources, nitrogen sources, or other precursors. LC-MS/MS then measures labeled and unlabeled modified nucleosides over time. If a methyl group becomes labeled, the experiment can connect a modification to one-carbon metabolism or methyl donor usage. If the base or ribose becomes labeled, the experiment can distinguish newly synthesized RNA from preexisting RNA. If a hypermodified tRNA nucleoside incorporates label into a side chain, the experiment can test precursor pathways.

Tracer interpretation is not automatic. Precursor pools may label incompletely. Salvage pathways can recycle unlabeled nucleosides. Growth rate, cell death, RNA class half-life, and stress responses can change apparent kinetics. Stable RNAs such as tRNA and rRNA turn over differently from many mRNAs. A decrease in a labeled modified nucleoside could reflect reduced installation, slower RNA synthesis, faster RNA decay, altered precursor supply, or dilution by cell growth. Tracing experiments should therefore report precursor enrichment, labeling duration, RNA class, normalization strategy, and kinetic model.

Manufactured RNA highlights the analytical side of the chapter. For a synthetic mRNA product, modification detection may ask whether uridine has been replaced by a modified uridine, whether the cap structure is correct, whether the poly(A) tail and full-length product are intact, whether double-stranded RNA impurities remain, and whether degradation products or residual reagents are present. LC-MS/MS can quantify nucleoside composition and some cap-related analytes, while chromatographic, electrophoretic, sequencing, and immunological assays address other product attributes. These measurements overlap with epitranscriptomics, but the standard of evidence is closer to product release testing than to discovery biology.

Mass spectrometry also has its own localization pitfalls. Isobaric or nearly isobaric modifications can share nominal mass and require chromatographic separation or diagnostic fragments. In-source fragmentation can create ions that look like biological analytes. Adducts with sodium, potassium, ammonium, or solvents can shift apparent mass. A modified nucleoside standard may not be commercially available, and an inferred elemental composition is weaker than a match to an authentic standard. For low-abundance analytes, a small integration decision can change the estimated amount substantially. These issues do not undermine LC-MS/MS; they explain why analytical reports should include retention time, transitions or fragments monitored, standard identity, calibration range, quality-control samples, and criteria for peak integration.

## 132.5. Oligonucleotide mapping, intact-RNA analysis, and sequence-level mass spectrometry

RNA-analyte mass spectrometry asks more than whether a modified nucleoside exists. It can ask whether a purified RNA has the expected intact mass, which sequence variants or truncations are present, which ends and terminal groups it carries, where a modification lies, and how product forms are distributed. Those questions require three distinct measurement levels. Nucleoside analysis completely digests RNA and reports chemical composition without sequence position. Bottom-up analysis uses one or more ribonucleases to make oligonucleotides that retain local sequence. Intact-RNA analysis measures unresolved or partly resolved molecular forms without digestion, whereas top-down analysis isolates an intact RNA precursor and fragments it to obtain sequence information. The terms describe sample level and fragmentation logic, not guaranteed performance: an intact mass is not automatically a sequence, and a top-down experiment is not automatically de novo complete.

The workflow begins before the mass spectrometer. RNA must be recovered without introducing cleavage, oxidation, unwanted dephosphorylation, or cation contamination. Detergents, chaotropes, glycerol, nonvolatile salts, metal ions, nucleases, and affinity-elution reagents can suppress ionization or create broad adduct distributions. A preparation suitable for gel electrophoresis may therefore be unsuitable for high-resolution RNA mass spectrometry. Purification can combine denaturing or native chromatography, size exclusion, ion exchange, reversed-phase cleanup, ultrafiltration, precipitation, or solid-phase extraction, but every operation can select among full-length RNA, truncations, duplexes, and structures. Desalting is a measurement step rather than a cosmetic one: excessive sodium or potassium broadens each charge-state envelope into multiple adducted species, complicates deisotoping and deconvolution, and can hide low-abundance variants. Crittenden et al. (2023) showed that online reversed-phase material operated in a size-exclusion-like cleanup mode could remove salts and metals before top-down analysis of approximately 100-nucleotide single-guide RNAs.

Separation reduces the number of species presented to the ion source at once. Ion-pair reversed-phase liquid chromatography is widely used for oligoribonucleotides because hydrophobic ion-pair reagents partially mask the polyanionic backbone and permit length- and sequence-dependent retention. Mobile-phase composition changes both chromatography and gas-phase behavior. Volatile amines, fluorinated alcohols, organic solvent, pH, and temperature can alter retention, charge-state distribution, metal adduction, and source contamination. Strezsak et al. (2022), for example, tuned a hexafluoroisopropanol, triethylammonium acetate, triethylamine, and ethanol system to reduce charge states and metal adducts in messenger-RNA characterization assays for cap efficiency and poly(A)-tail distribution. The broader rule is that a mobile phase is part of the ion-generation model; a recipe optimized for peak shape may not be optimal for sensitivity, adduct control, or instrument cleanliness.

Capillary zone electrophoresis (CZE) provides a complementary separation based principally on electrophoretic mobility. CZE can use very small sample amounts and can resolve analytes that are difficult to separate chromatographically, but electroosmotic flow, adsorption to the capillary, injection reproducibility, sheath-flow dilution, and MS-compatible buffers require control. Microfluidic CZE-MS studies of transfer RNA and a larger viral RNA region show an important boundary: species can comigrate in a single electrophoretic peak yet still yield useful bottom-up sequence coverage when high-resolution mass and tandem spectra distinguish their ions. Liquid chromatography and CZE are therefore not interchangeable winners. Separation choice should follow RNA length, hydrophobicity, sample amount, expected heterogeneity, digestion products, quantitative goal, and the ion source that follows.

Electrospray ionization (ESI) is usually operated in negative-ion mode for RNA because deprotonation of backbone phosphates generates multiply charged anions. Multiple charging places a large RNA within a usable mass-to-charge range and supports online LC or CZE coupling and precursor-selected tandem MS. It also creates an interpretation problem: one molecule appears in several charge states, and each charge state can carry isotopic peaks and different numbers of metal or solvent adducts. Charge-state assignment, deisotoping, and deconvolution convert those envelopes into a neutral-mass estimate, but the result inherits every unresolved overlap. A deconvoluted peak is thus a model-based summary of observed ions, not a unique molecular identity. Two sequence isomers, positional modification isomers, or compensating truncation and adduct combinations may remain indistinguishable without retention, fragmentation, standards, or another separation dimension.

Matrix-assisted laser desorption/ionization (MALDI) occupies a different niche. MALDI commonly produces simpler, lower-charge spectra and can rapidly profile purified oligonucleotides, digest products, or mass fingerprints. The simplicity can make a presence-or-absence screen or approximate mass map easier to inspect. Yet matrix-derived background, spot-to-spot crystallization heterogeneity, alkali-metal adducts, in-source decay, mass discrimination, limited online separation, and less routine precursor-resolved fragmentation constrain sequence-level analysis, especially for larger heterogeneous RNA. MALDI and negative-mode ESI are complementary boundaries rather than universal substitutes: MALDI is often attractive for rapid profiling, whereas ESI more naturally supports online separations, charge-state selection, and systematic MS/MS.

Bottom-up mapping converts a long RNA into a tractable set of sequence-bearing oligonucleotides. The nuclease is a chemical reagent with sequence and end-group rules. Ribonuclease T1 preferentially cleaves after accessible unmodified guanosine and typically leaves a 3′ phosphate, whereas other enzymes provide complementary base preferences or broader specificity. RNA structure, neighboring modifications, enzyme concentration, reaction time, and terminal chemistry can create missed cleavages or unexpected products. A missed cleavage is not simply failed preparation: it may join two short nonunique fragments into a longer unique fragment and improve localization. Conversely, uncontrolled partial or nonspecific digestion expands the candidate space and makes spectral assignment harder. The digest design should specify intended and allowed cleavage rules, missed-cleavage count, terminal phosphates or hydroxyls, cyclic-phosphate possibilities, and whether modification-dependent protection is expected.

Most bottom-up modification maps are reference-guided. Known or predicted RNA sequences define the digest products against which precursor masses and tandem spectra are matched; the experiment then places modified residues within that candidate sequence set. This design is efficient, but it is not equivalent to unconstrained de novo sequencing. An unrepresented paralog, sequence variant, contaminant, or unexpected end can redirect an otherwise convincing spectrum to the wrong RNA. Reports should therefore state the reference database and include plausible variants and contaminants when they can affect localization.

Complementary nucleases turn coverage into an experimental design variable. A tRNA digested only with RNase T1 may produce repeated short fragments or leave a modification in a fragment with several candidate sites. A second digest, such as RNase U2 or an engineered broader-specificity nuclease, can create overlapping fragments with different boundaries. The intersection of those fragments can localize a modification even when either digest alone is ambiguous. Ross et al. (2016) established a practical sequence-mapping workflow for modified tRNA, and Solivio et al. (2018) showed that an RNase U2-E49A mutant could improve modification-mapping coverage through less sequence-specific cleavage. For long manufactured RNAs, deliberately partial complementary digests can avoid an unmanageably dense set of tiny oligonucleotides; Welbourne et al. (2025) combined RNase T1 and U2 maps for messenger-RNA sequence characterization. The cost is a larger, more assumption-sensitive search space.

Metabolic isotope labeling can provide another constraint on digest assignments. A defined carbon- or nitrogen-isotope shift can restrict the elemental composition of a candidate oligonucleotide and can support relative comparison between RNA samples. The isotope pattern does not supply sequence order or site position by itself; it strengthens an assignment only when digestion boundaries, reference sequence, retention, and fragment ions are also consistent.

After separation and precursor selection, tandem mass spectrometry must break the phosphodiester backbone in an interpretable way. Collision-induced dissociation (CID) and higher-energy collisional dissociation (HCD) are widely accessible and can yield terminal a, a–base, b, c, d and w, x, y, z ion series, with distributions that depend on RNA sequence, precursor charge, collision energy, length, and terminal groups. Labile modifications can instead undergo dominant base or side-chain neutral loss. Such a loss may be diagnostically useful for recognizing a modification class, yet it can suppress the backbone ladder needed to localize that modification. Ultraviolet photodissociation (UVPD) can provide complementary and sometimes denser coverage. Electron-based methods require careful names and polarity logic. Conventional electron-transfer dissociation was developed mainly for multiply protonated cations; for RNA anions, negative electron-transfer dissociation (NETD) and activated-ion NETD are relevant analogues. NETD can favor d- and w-type ions, preserve some labile modifications, and reduce certain neutral-loss or internal-fragment channels relative to collision methods. Activated electron photodetachment (a-EPD), radical-transfer dissociation, and related electron- or radical-based approaches further broaden the toolbox. No fragmentation method is universally superior: the useful question is which combination yields confident sequence-discriminating ions while retaining the modification or terminal group under study.

Precursor charge and end chemistry can change the same sequence's spectrum. Highly charged anions experience different Coulombic repulsion and dissociation pathways from lower-charge precursors. A 3′ phosphate, 5′ phosphate, hydroxyl, cyclic phosphate, cap, or phosphorothioate changes precursor mass, charge localization, neutral losses, and fragment-ion interpretation. Sun et al. (2023) and Zuo et al. (2024) showed that charge state and terminal phosphate groups materially affect collisional dissociation of RNA oligonucleotide anions. A search that assumes 5′ hydroxyl and 3′ phosphate cannot safely assign fragments produced by a nuclease or phosphatase workflow with different end chemistry. Terminal-state standards and deliberate enzymatic treatments can test the assignment.

Fragment annotation must distinguish terminal ladders from internal fragments. A terminal fragment contains one original RNA end and supports a directional sequence series. An internal fragment is bounded by two fragmentation events and lacks both original ends. Internal ions become increasingly important as RNA length grows because long precursors may not yield complete terminal ladders. Kenderdine et al. (2023) showed that including internal fragments can recover sequencing information from larger oligonucleotides. The benefit comes with an ambiguity penalty: many subsequences can share the same elemental composition, and the number of possible internal assignments grows quickly. A software score should reward mass accuracy, expected ion series, sequence coverage, complementary ions, and reproducibility while penalizing unexplained peaks and promiscuous matches. Manual labeling of a visually plausible ladder is insufficient when many alternatives exist.

Database and spectral search make those assumptions explicit. The search space may contain one expected product sequence, all RNAs in an organism, a therapeutic construct plus specified variants, or a combinatorial library. Search engines must define enzyme specificity, missed cleavages, terminal groups, allowed modifications, isotope errors, adducts, precursor and fragment tolerances, charge states, neutral losses, and fragment series. If the modification list is unconstrained, almost any mass difference can acquire a chemical name. If the sequence database omits a contaminant or synthesis variant, a high-scoring incorrect target can still win. NASE and Pytheas demonstrate automated RNA-MS workflows that incorporate sequence databases, modification-aware fragments, and target-decoy analysis. Their larger lesson is methodological: identification confidence is conditional on the candidate universe that the analyst allowed.

False-discovery rate (FDR) estimation for RNA MS borrows target-decoy logic from proteomics but cannot be copied uncritically. Decoy sequences may be shuffled or reversed while preserving properties needed to model chance matches. A score threshold or q-value then estimates the fraction of accepted matches expected to be incorrect under that target-decoy model. Small databases, short repetitive fragments, dense modification lists, nonspecific cleavage, and chemically impossible decoys can make the null distribution unrepresentative. A reported one-percent q-value does not validate the identity of a modification, prove database completeness, or correct wrong terminal chemistry. Reports should give the decoy-generation rule, database size, target-decoy competition method, score components, filtering order, and whether FDR was estimated at the spectrum, oligonucleotide, site, or RNA level.

Intact-RNA measurement has a simpler experimental diagram but a harder spectrum. A purified guide RNA, transfer RNA, aptamer, small interfering RNA strand, or other moderate-length transcript enters the source without nuclease digestion. The analyst observes a charge-state envelope and deconvolves it to candidate neutral masses. Expected full-length product, n−1 truncations, additions, depurination products, oxidation, incomplete deprotection, end variants, and adducts may be detectable if resolution and abundance permit. Intact mass is strong evidence that a product lies near an expected composition, but it cannot generally distinguish sequence isomers or locate an equal-mass substitution. Macias et al. (2023) and Crittenden et al. (2023) used top-down strategies to assess guide-RNA spacer fidelity and sequence coverage, illustrating both the power and practical length range of this approach.

Top-down fragmentation can upgrade an intact precursor from a mass check to sequence-confirming evidence. For an approximately 100-nucleotide single-guide RNA, combined CID, HCD, UVPD, and a-EPD can produce overlapping terminal and internal evidence across much of the molecule. Yet coverage percentage must be interpreted carefully. A reported fraction of cleaved backbone positions is not the same as the probability of detecting every possible substitution, and abundant fragments from one region can coexist with a blind region elsewhere. Modified guide RNAs add another difficulty: labile modifications or phosphorothioates may be lost, rearranged, or selectively retained depending on dissociation. A sequence-fidelity statement should therefore report the actual discriminating ions for the variable or risk-critical region, not only total coverage.

Therapeutic messenger RNA lies beyond the range where a single deconvoluted intact spectrum routinely provides complete de novo sequence. Thousands of nucleotides generate broad, overlapping charge-state, isotope, adduct, conformer, truncation, and poly(A)-length distributions. Intact or limited-digest measurements can still report product envelopes and gross heterogeneity, while targeted nuclease release can isolate a capped 5′ fragment or tail-containing 3′ fragment for cap-efficiency and poly(A)-distribution assays. Bottom-up complementary digests can test coding-region identity and modification placement. Complete nucleoside digestion measures overall modified-nucleoside incorporation. These outputs answer different quality questions and should be integrated, not collapsed into the claim that “mass spectrometry confirmed the mRNA.” Double-stranded RNA impurities, higher-order assemblies, and biological potency require additional assays; the broader product-control framework belongs to [Chapter 159](chapter1142.md).

RNA mass spectrometry can also be applied to RNA recovered from ribonucleoprotein particles, but the ownership boundary follows the analyte. This chapter owns purification and sequence/modification analysis of the RNA molecule. Experiments that preserve native ribonucleoprotein stoichiometry, architecture, assembly states, or RNA–protein contacts hand off to [Chapter 59](chapter1054.md). Identification, quantification, post-translational modification, and interaction analysis of the protein components hand off to [Chapter 138](chapter1125.md). Keeping those boundaries explicit prevents a bottom-up RNA digest from being mistaken for a native-complex measurement.

Quantification requires standards matched to the level of inference. Stable-isotope-labeled nucleosides support nucleoside ratios; isotopically labeled oligonucleotides can correct digestion, recovery, retention, and ionization for targeted fragments; full-length RNA standards or well-characterized mixtures can calibrate intact variant abundance and deconvolution bias. External calibration of mass accuracy is not a substitute for an internal quantitative standard. Response factors differ across sequence, length, charge state, modification, and mobile phase, so raw peak areas from different oligonucleotides are rarely directly comparable. Appropriate quality controls include process blanks, digestion blanks, system-suitability mixtures, carryover injections, replicate preparations, spike recoveries, retention-time and mass-accuracy windows, fragment-ion acceptance rules, and dilution or standard-addition tests for matrix effects.

Figure 132.7 presents the causal RNA-MS workflow from sample cleanup to a qualified molecular output, while Table 132.5 contrasts nucleoside, bottom-up, intact, and top-down analysis. Figure 132.8 shows how RNase boundaries and terminal versus internal fragment ions constrain a tRNA modification site. Table 132.6 lists reporting fields for sequence-level RNA mass spectrometry, and Box 132.4 explains why deconvoluted mass alone is not molecular identity.

## 132.6. Site localization, stoichiometry estimation, and orthogonal assay design

Site localization asks which nucleotide in which RNA molecule carries the modification. That statement requires more than a signal near a transcript. The coordinate system must be defined: genomic coordinate, transcript coordinate, mature tRNA position, rRNA numbering, viral genome coordinate, synthetic construct coordinate, or cap-proximal position. Isoforms must be considered because a site in one transcript model may be absent from another. Reads must map uniquely enough to assign the signal. The assay must have enough resolution to distinguish neighboring candidate residues. If an enriched fragment contains several possible adenosines, the evidence is region-level unless another method localizes the site.

Stoichiometry asks what fraction of molecules carry the modification. The denominator might be all molecules of one transcript isoform, all reads covering a site, all molecules in a purified RNA class, all nucleosides in a total RNA sample, or all product molecules in a manufactured RNA lot. These denominators are not interchangeable. A nanopore per-read model may estimate the fraction of reads with a signal at a site. A targeted chemical assay may estimate the fraction of molecules producing a treatment-specific signature. LC-MS/MS may estimate modified nucleoside abundance relative to canonical nucleosides in an RNA pool. Each measurement has its own calibration problem.

Enrichment intensity is not stoichiometry. A strong antibody peak can arise from high RNA abundance, high antibody affinity, favorable fragment size, or sequence context. A weak peak can arise from low expression even if the modified fraction is high. Sequencing mutation fraction is closer to a site denominator but still depends on treatment efficiency, baseline error, RT bias, and coverage. Nanopore model probability can be converted to a fraction only if model calibration is valid for the modification and context. LC-MS/MS can be quantitatively rigorous for bulk analytes but may average across many RNAs.

Orthogonal assay design combines measurements whose decisive failure modes differ. An mRNA N6-methyladenosine candidate discovered by antibody enrichment can be paired with a site-resolving signature, defined modified and unmodified standards, writer-depleted material, and global LC-MS/MS. A pseudouridine candidate can combine selective treatment, untreated controls, synthetic standards, and an independent signal class. A tRNA queuosine call can combine native sequencing with nucleoside or oligonucleotide mass spectrometry. [Chapter 46](chapter1043.md) determines how those outputs support a general evidence tier; this section specifies how the assays are made technically independent and quantitatively comparable.

Figure 132.4 shows a validation lattice rather than a single linear path. Some claims need chemical identity most urgently; others need localization or stoichiometry. The figure should help readers choose validation according to the claim being made.

![Figure 132.4. Orthogonal assay-design lattice](../assets/figures/chapter1120_figure4.png)

**Figure 132.4. Orthogonal assay-design lattice.** Orthogonal assay design combines different signal-generating principles, truth standards, and quantitative outputs. Shared sample identity, coordinates, denominators, and calibration mixtures make the measurements comparable without sharing the decisive failure mode.

RNA class changes assay design. mRNA and long noncoding RNA require expression normalization and isoform-aware coordinates; tRNA and rRNA require structure-aware enzymes, mature-end definitions, and controls for dense neighboring modifications; viral RNA requires separation of host RNA and viral transcript forms; manufactured RNA requires lot identity, full-length product, cap and tail attributes, impurity testing, and product-specific acceptance criteria. These are measurement constraints. Biological comparison among marks, states, and diseases belongs to [Chapter 52](chapter1048.md).

Figure 132.5 focuses on boundary cases across RNA classes. It should teach that the same assay label can mean different evidence in mRNA, tRNA, viral RNA, and synthetic RNA.

![Figure 132.5. RNA-class measurement constraints](../assets/figures/chapter1120_figure5.png)

**Figure 132.5. RNA-class measurement constraints.** mRNA and lncRNA require abundance and isoform normalization; tRNA and rRNA require mature-coordinate and dense-modification controls; viral RNA requires transcript-form separation; manufactured RNA requires lot, full-length product, cap, tail, and impurity measurements. Cross-mark biological synthesis belongs to [Chapter 52](chapter1048.md).

When perturbations are included in an assay panel, the measurement report should show that the modification signal itself changes, quantify RNA abundance and composition, and document rescue or matched controls. Whether those observations establish writer, eraser, reader, or functional mechanism is adjudicated in [Chapter 46](chapter1043.md), while the cross-mark biological consequence belongs to [Chapter 52](chapter1048.md).

## 132.7. False discovery, assay benchmarks, standards, and reporting requirements

False discovery in RNA modification detection has many sources. Antibody cross-reactivity can create peaks unrelated to the intended mark. Chemical conversion can be incomplete or nonspecific. Reverse transcriptases can stop at structured templates or damaged bases. Nanopore models can confuse sequence context, neighboring modifications, pore chemistry, or isoform differences with a named modification. LC-MS/MS can misassign co-eluting isobars or suffer from matrix effects and carryover. Alignment errors, unrecognized genomic variants, paralogs, RNA editing, short reads, repetitive elements, transcript isoforms, and sample contamination can create convincing but wrong calls.

Statistical false-discovery control must be paired with experimental false-discovery control. A multiple-testing-adjusted p value cannot rescue an antibody with poor specificity, a chemical reaction with unknown conversion rate, or a nanopore classifier trained on the wrong context. Conversely, beautiful controls do not replace replicate-aware statistics when thousands of sites are tested. A modification map should report biological replicates, independent library preparations when feasible, negative and positive controls, spike-ins or standards, read-depth thresholds, multiple-testing correction, peak or site-calling parameters, and validation criteria.

An assay benchmark should define the intended use, truth set, analyte range, sequence and structural contexts, RNA classes, input amounts, operators or laboratories, and performance metrics before evaluation. Sensitivity, specificity, precision, recall, calibration error, localization accuracy, quantitative bias, reproducibility, and limits of detection or quantification answer different questions. Synthetic standards test controlled chemistry but may not reproduce endogenous folding or neighboring marks; knockout or writer-depleted samples preserve biological complexity but may not eliminate every site or indirect change. A useful benchmark therefore uses complementary standards, held-out contexts, versioned software and models, and acceptance thresholds tied to the intended claim.

Figure 132.6 organizes false-discovery sources by stage: sample, chemistry, library, sequencing or instrument, computation, and interpretation. The stage model helps readers diagnose whether a proposed fix addresses the actual artifact.

![Figure 132.6. False-discovery sources by workflow stage](../assets/figures/chapter1120_figure6.png)

**Figure 132.6. False-discovery sources by workflow stage.** False positives can arise during sample handling, RNA purification, antibody capture, chemical conversion, enzymatic treatment, library construction, sequencing or instrument acquisition, alignment, model calling, and interpretation. Effective controls target the stage that can create the artifact.

Standards should match the intended measurement. A discovery assay needs a stated candidate-calling threshold. A site-localization workflow needs positional accuracy. A chemical-identity workflow needs modification-specific chemistry or analytical confirmation. A stoichiometry workflow needs calibration, denominator, bias, precision, and reportable range. A differential workflow needs matched conditions and normalization. A single-molecule co-occurrence workflow needs per-read calibration and error modeling. General claim-tier adjudication belongs to [Chapter 46](chapter1043.md).

Table 132.4 provides a reporting checklist. It includes sample identity, organism, cell type or tissue, RNA purification, RNA integrity, depletion or enrichment strategy, assay chemistry, enzyme or antibody details, standards, controls, calibration, coordinate system, transcript annotation version, statistical thresholds, validation assays, stoichiometry denominator, software versions, data deposition, and known limitations.

**Table 132.4. Reporting checklist for RNA modification studies.** Transparent reporting lets readers distinguish discovery signals from validated modification claims. The checklist includes sample identity, RNA purification, assay chemistry, controls, standards, calibration, coordinate systems, statistical thresholds, validation, stoichiometry denominators, data deposition, and known limitations.

| Reporting item | Why it matters | Minimum information | Common omission |
| --- | --- | --- | --- |
| **Sample identity and RNA population** | Defines which molecules the claim can cover | Organism, cell type or tissue, condition, RNA class, enrichment or depletion strategy, input amount | Reporting a transcriptome-wide claim without stating the RNA fraction assayed |
| **RNA extraction, storage, and integrity** | Sample handling can create damage, loss, or biased recovery | Extraction method, storage conditions, RNA integrity or size profile, contamination controls | Treating degradation or purification bias as unrelated to modification signal |
| **Assay chemistry, antibody, enzyme, or instrument settings** | The measured signal depends on reagent and platform behavior | Reagent lot or enzyme identity, treatment conditions, library protocol, nanopore chemistry, LC-MS/MS method | Naming a method family without enough detail to reproduce its signal chain |
| **Positive, negative, and spike-in controls** | Controls define specificity, conversion efficiency, sensitivity, and baseline error | Known modified and unmodified controls, untreated samples, input controls, spike-ins or standards | Using only biological replicates without chemical or analytical controls |
| **Coordinate system and annotation version** | Site claims depend on transcript models and mature RNA numbering | Genome build, transcript annotation, strand, isoform, mature RNA coordinate convention, mapping filters | Reporting a site without enough coordinate information to locate it unambiguously |
| **Denominator and calibration** | Stoichiometry and differential claims require a defined reference population | Reads covering the site, input fraction, RNA pool, canonical nucleoside normalizer, calibration curve, model calibration | Presenting enrichment, model score, or peak area as stoichiometry |
| **Replicates and statistical thresholds** | Large modification maps need replicate-aware false-discovery control | Biological and technical replicate counts, filtering thresholds, multiple-testing method, peak or site-caller parameters | Reporting candidate sites without read-depth, replicate, or false-discovery criteria |
| **Orthogonal validation and claim level** | Validation should match discovery, localization, chemical identity, stoichiometry, or function | Independent assay, perturbation or rescue design, LC-MS/MS support, targeted site test, stated evidence grade | Using language for a higher claim level than the evidence supports |
| **Software, models, and data deposition** | Computational calls change with aligners, base-callers, models, and training data | Software versions, model source, training or benchmark data, parameter files, raw and processed data accessions | Omitting model version or raw signal data for nanopore modification calls |
| **Known limitations and negative-result boundary** | Readers need to know what the study could not detect | Sensitivity limits, coverage gaps, excluded RNA classes, unresolved isobars, ambiguous sites, failed controls | Presenting absence of signal as absence of modification without detection limits |

The sequence-level RNA-MS visual set consolidates the workflow and its main interpretive boundaries. The first figure follows a sample through cleanup, separation, ionization, precursor processing, fragmentation, search, and molecular reporting. The second uses overlapping tRNA digest fragments to distinguish cleavage boundaries, terminal-ion ladders, internal fragments, and a localized modification. The tables compare analysis levels and define the reporting information needed to reproduce an assignment. The boxed caution focuses on a common category error: neutral mass constrains molecular composition but rarely establishes unique sequence or modification position.

![Figure 132.7. Causal workflow for sequence-level RNA mass spectrometry](../assets/figures/chapter1120_figure7.png)

**Figure 132.7. Causal workflow for sequence-level RNA mass spectrometry.** RNA-analyte mass spectrometry begins by defining whether the intended output is nucleoside composition, a bottom-up oligonucleotide map, intact molecular mass, or top-down sequence evidence. Cleanup and desalting constrain adducts; LC or CZE constrains mixture complexity; negative-mode ESI or MALDI constrains ion populations; charge assignment and fragmentation constrain observable evidence; and the search model constrains identification and FDR. Each output retains different information and uncertainty.

![Figure 132.8. Complementary RNase maps and RNA fragment-ion logic](../assets/figures/chapter1120_figure8.png)

**Figure 132.8. Complementary RNase maps and RNA fragment-ion logic.** RNase T1 and a complementary RNase create overlapping oligonucleotides across a tRNA modification. Terminal ion ladders retain one original RNA end and constrain direction; internal fragments can fill gaps but have more candidate assignments. Localization is strongest where the modified mass shift is bracketed by sequence-discriminating ions from more than one digest.

**Table 132.5. Information retained across RNA-MS analysis levels.** Nucleoside, bottom-up, intact, and top-down RNA-MS workflows do not form a simple ladder of quality. Each sacrifices and preserves different information, so the analysis level must match the biological or product question.

| Analysis level | Input transformation | Immediate measurement | Information retained | Strongest outputs | Principal ambiguities | Representative standard |
| --- | --- | --- | --- | --- | --- | --- |
| **Nucleoside analysis** | Complete enzymatic digestion to individual nucleosides | Retention, precursor/product ions, and analyte abundance | Chemical identity and RNA-pool composition; sequence position is lost | Modified-nucleoside presence, calibrated abundance, isotope incorporation | RNA source, site, transcript, and neighboring sequence | Stable-isotope-labeled nucleoside and matrix-matched calibration curve |
| **Bottom-up oligonucleotide mapping** | Defined complete or partial RNase digestion | Oligonucleotide precursor masses and tandem spectra | Local sequence, digest termini, and some modification position | Sequence map, site-localized modification, end-group and targeted cap/tail fragments | Repeated fragments, missed cleavage, nonspecific cleavage, co-elution, isomers | Isotopically labeled or synthetic oligonucleotide plus complementary digest |
| **Intact-RNA mass analysis** | No nuclease digestion | Charge-state envelope and deconvoluted mass distribution | Whole-product mass linkage and resolved product variants | Expected mass, gross truncations/additions, end variants, heterogeneity | Sequence isomers, modification position, adduct/truncation overlap, response bias | Full-length reference RNA or characterized product mixture |
| **Top-down RNA MS/MS** | No nuclease digestion before precursor isolation and gas-phase fragmentation | Intact precursor plus terminal and internal fragment ions | Linkage among observed sequence regions, ends, and retained modifications | Sequence confirmation, risk-region fidelity, selected modification or terminus localization | Incomplete ladders, length-dependent transfer, internal-fragment ambiguity, labile groups | Full-length reference RNA and site- or variant-specific controls |

**Table 132.6. Minimum reporting fields for sequence-level RNA mass spectrometry.** A reproducible RNA-MS assignment requires enough information to reconstruct the molecular candidate space from sample preparation through inference. Instrument mass accuracy alone is not a complete confidence statement.

| Workflow field | Minimum report | Why it changes interpretation | Failure if omitted |
| --- | --- | --- | --- |
| **RNA input and recovery** | Construct or biological RNA identity, purification, integrity, amount, storage, and recovery selectivity | Defines which intact, truncated, duplex, or contaminating forms could enter analysis | A missing product may reflect preparation loss rather than true absence |
| **Cleanup and mobile phase** | Desalting method, volatile salts or ion-pair reagents, solvents, additives, pH, and metal-control steps | Determines ion suppression, adduct burden, retention, and source behavior | Broad or shifted peaks may be misread as biological heterogeneity |
| **Separation** | LC column and gradient or CZE capillary, buffer, voltage, flow interface, temperature, injection, and retention/migration criteria | Determines mixture complexity and co-elution | One feature may contain multiple unresolved RNA forms |
| **Ionization and precursor processing** | ESI polarity or MALDI matrix, source settings, charge assignment, deisotoping, adduct rules, and deconvolution software/parameters | Determines which ions contribute to neutral mass and abundance | A deconvoluted peak may merge wrong charge states or adducts |
| **Digestion model** | RNase, reaction conditions, specificity, allowed missed cleavages, partial-digest policy, and terminal chemistries | Defines theoretical oligonucleotides and sequence uniqueness | Search candidates and coverage cannot be reproduced |
| **Fragmentation and annotation** | Dissociation method and energy, precursor charge, ion series, neutral losses, internal-fragment policy, mass tolerance, and acceptance rules | Defines which peaks support sequence, end, or modification position | Visually plausible annotations may be nonunique |
| **Search and FDR** | Sequence database, contaminants and variants, modification/adduct list, scoring, decoy construction, competition, filtering, and inference level | Defines the candidate universe and null model | A q-value cannot be interpreted or compared |
| **Quantification** | Internal and external standards, addition step, curves, response model, recovery, matrix effects, carryover, precision, accuracy, and reportable range | Converts peak area into a defensible amount or fraction | Relative response may be presented as unbiased abundance |
| **Molecular output** | Separate calls for sequence, modification/site, termini, cap, poly(A), truncation, and heterogeneity with uncertainty | Prevents evidence for one attribute from being promoted to all attributes | “MS-confirmed RNA” hides unmeasured or unresolved properties |

> **Box 132.4. Deconvoluted Mass Is a Constraint, Not an Identity**
>
> - Conceptual home: [Section 132.5](chapter1120.md), after intact-RNA analysis and charge-state deconvolution.
> - First placeholder placement: [Section 132.7](chapter1120.md), after `[Chapter 132](chapter1120.md).table.06`, to preserve monotonic stable visual-ID order.
>
> **Deconvoluted mass is a constraint, not an identity**
>
> An RNA molecule observed by negative-mode electrospray appears as several charge states, often accompanied by isotopes and sodium, potassium, ammonium, or solvent adducts. Deconvolution groups those observations into candidate neutral masses. A close match to the theoretical full-length mass supports product composition, but several alternatives can survive. Sequence isomers have the same mass. Positional isomers of many modifications have the same mass. A truncation combined with an adduct or addition can overlap another form within practical resolution. Low-abundance variants can be absorbed into a broad envelope or respond differently during ionization. Upgrade an intact-mass match with retention behavior, end-specific enzymatic treatment, discriminating tandem fragments, authentic or full-length standards, and orthogonal measurements. Report the mass tolerance, charge states, adduct model, software, unresolved envelope, and alternative assignments.

## Recent Consensus

The current consensus is that no RNA modification or RNA-MS method is universal. Antibody methods remain useful for broad discovery but require specificity controls and follow-up. Chemical and enzymatic assays can be site-informative when their reaction or enzyme signature is calibrated. Direct RNA nanopore sequencing is a major emerging approach because it can read native long RNA molecules and potentially link modifications to isoforms and molecule-level states, but modification calling is model-dependent. Nucleoside LC-MS/MS remains a chemical and quantitative anchor, especially for global abundance and product composition, but complete digestion usually does not localize sites.

Sequence-level RNA mass spectrometry is now a distinct analytical tier rather than a minor extension of nucleoside quantification. Complementary RNase digests and high-resolution LC-MS/MS can map tRNA and long-RNA fragments, including modifications and termini, when cleavage and search assumptions are explicit. Intact and top-down analysis can characterize selected oligonucleotides, guide RNAs, tRNAs, and other moderate-length products, but increasing length makes adduct control, charge-state deconvolution, fragment assignment, and full sequence coverage progressively harder. Collision, ultraviolet, and electron- or radical-based dissociation provide complementary evidence. Automated search and target-decoy strategies improve reproducibility, yet the reported FDR remains conditional on database, decoy, modification, adduct, and enzyme models. Orthogonal validation remains essential because it lets independent limitations cancel rather than compound.

## Open Questions, Controversies, Deprecated Models, and Common Misconceptions

Open questions:

- How many low-abundance transcriptome-wide modification calls are reproducible across laboratories and platforms?
- Which nanopore models generalize across pore chemistry, base-callers, organisms, RNA classes, and neighboring modifications?
- How should site-level stoichiometry be benchmarked for low-input samples?
- What reporting standards should journals require for disease-associated modification maps?
- How should synthetic standards be designed so that they represent cellular RNA structure and neighboring modifications without becoming too complex to interpret?
- Which direct RNA sequencing signals are robust enough for clinical or manufacturing use?
- How far can top-down RNA sequencing scale before incomplete ladders, internal-fragment ambiguity, adducts, and precursor congestion outweigh the benefit of preserving an intact molecule?
- Which decoy strategies provide calibrated oligonucleotide-, site-, and RNA-level FDR estimates for small product databases and large biological search spaces?
- How should laboratories standardize RNA fragment-ion nomenclature, end chemistry, spectral libraries, and confidence reporting across CID, HCD, UVPD, NETD, and radical-based workflows?
- Which full-length or fragment standards best measure quantitative bias across digestion, recovery, separation, ionization, charge-state selection, and deconvolution?

Minimum reporting should make negative measurements interpretable. If a study does not detect a modification, readers need the coverage, conversion efficiency, sensitivity, mass-spectrometry detection limit, or model power needed to see the expected signal. A negative result in total RNA does not exclude a mark in a rare transcript, and a negative nucleoside LC-MS/MS run may remain compatible with a site below the detection limit. Reporting limits and denominators make absence of signal comparable across studies.

Common misconceptions:

- "A peak is a modified nucleotide." A peak is an assay signal over a region. It becomes a site claim only after localization evidence, and it becomes a chemical identity claim only after modification-specific support.
- "Direct RNA sequencing automatically identifies modifications." Direct RNA nanopore sequencing measures native RNA current. The modification name is assigned by a computational model that requires controls and benchmarking.
- "Mass spectrometry always solves the problem." LC-MS/MS is chemically powerful, but complete nucleoside digestion usually removes transcript position. Site-localizing mass spectrometry is possible but technically demanding.
- "An intact deconvoluted mass identifies the RNA sequence." Neutral mass constrains composition, but sequence isomers, positional modification isomers, adducts, and compensating product changes require separation, fragmentation, standards, or orthogonal evidence.
- "A one-percent RNA-MS FDR proves that the modification is correct." A q-value estimates chance matches under a specified target-decoy and search model; it does not validate an omitted contaminant, wrong end group, implausible modification list, or incomplete database.
- "One RNase digest provides an unbiased sequence map." Cleavage depends on sequence, structure, modification, enzyme conditions, and allowed missed cleavages. Complementary digests can reveal blind or ambiguous regions.
- "ETD and NETD are interchangeable names." Conventional ETD acts on cations; RNA is commonly analyzed as anions, for which NETD or activated-ion NETD is the polarity-appropriate electron-transfer family.
- "Detection proves function." A detected modification can be structural, constitutive, damage-related, metabolic, or incidental. Function requires causal evidence.
- "Two positive assays are automatically orthogonal." Two assays that share the same antibody, conversion chemistry, enzyme bias, model training set, or sample artifact may reproduce the same error. Orthogonality depends on independent failure modes.