# Chapter 49. Biological Functions and Contexts of Pseudouridine, 2′-O-Methylation, Inosine, m5C, ac4C, Queuosine, m7G, GlycoRNA, and Other RNA Marks

## Scope Note

This chapter surveys major marks other than the m6A-centered adenosine methylations in [Chapter 48](chapter1044.md). It owns mark-by-mark chemical identity, RNA-class distribution, biological context, and evidence limits for pseudouridine, 2′-O-methylation, inosine, m5C, ac4C, queuosine, m7G, glycoRNA, and selected rarer marks. For glycoRNA, this ownership includes the evidence for covalent RNA-glycan linkage, acp3U-linked N-glycans, emerging O-glycosylation evidence, biosynthetic dependencies, substrate boundaries, and chemical-claim artifacts. [Chapter 107](chapter1102.md) owns cell-surface topology, extracellular ribonucleoprotein assemblies, intercellular and immune functions, and trafficking models; [Chapter 132](chapter1120.md) owns protocol-level glycoRNA detection and analytical workflows. [Chapter 47](chapter1163.md) owns the cross-mark comparison of enzyme folds, catalytic chemistries, guide-dependent recognition, cofactors, kinetics, fidelity, evolution, inhibitors, and biotechnology.

## Executive Summary

RNA modification biology is older and broader than the current focus on mRNA methylation. Transfer RNAs and ribosomal RNAs contain dense constellations of modified nucleotides that stabilize folding, tune decoding, organize active sites, and help ribonucleoprotein assembly. Small nuclear RNAs carry marks that support spliceosomal recognition. Messenger RNAs and viral RNAs can carry lower-stoichiometry or context-dependent marks whose functional interpretation is often more difficult. The central rule of this chapter is that a modification claim must name the chemical mark, the RNA class, the site or region, the enzyme or guide pathway when known, the occupancy, and the evidence connecting that mark to function.

Pseudouridine is the C-glycoside isomer of uridine. In ordinary uridine, the base is linked to ribose through N1; in pseudouridine, the base is linked through carbon C5, leaving an extra imino group that can participate in hydrogen bonding. This small rearrangement can stabilize local structure, alter hydration, and change protein recognition. Pseudouridine is installed by stand-alone pseudouridine synthases and by H/ACA small nucleolar ribonucleoproteins in guide-directed contexts. tRNA and rRNA pseudouridylation are long-established, while pre-mRNA, mRNA, plant RNA, and disease-linked pseudouridine maps are expanding rapidly.

Ribose 2′-O-methylation adds a methyl group to the 2′-hydroxyl of ribose. Because every ribonucleotide contains ribose, Nm marks can occur on A, C, G, or U. In eukaryotic rRNA and snRNA, many sites are installed by C/D box small nucleolar RNPs, where guide RNA base pairing positions fibrillarin near the target ribose. This guide-directed logic allows sequence-programmable selection without requiring a different enzyme for every site. Internal mRNA 2′-O-methylation and viral RNA 2′-O-methylation are active areas of investigation because ribose methylation can influence RNA stability, splicing, translation, and innate immune recognition, but site-specific functional claims require orthogonal validation.

Inosine is not installed by methylation or glycosyl transfer. It is produced when adenosine is deaminated in RNA, most prominently by adenosine deaminases acting on RNA, or ADAR enzymes, in double-stranded RNA regions and by ADAT-family enzymes in tRNA. Many polymerases and base-pairing interactions interpret inosine as guanosine-like, so A-to-I editing can recode codons, alter splice sites, change microRNA targeting, modify RNA structure, tune innate immune recognition, or expand tRNA decoding. [Chapter 50](chapter1046.md) treats editing enzymes in depth; this chapter frames inosine as a modification whose biology is inseparable from substrate class, site selection, structure, and readout.

Several cytidine and complex nucleoside marks illustrate the diversity of RNA modification logic. m5C is installed by NSUN-family enzymes and by DNMT2/TRDMT1 on selected tRNA, rRNA, mitochondrial RNA, and mRNA substrates. ac4C is established in tRNA and rRNA and has context-dependent mRNA assignments linked to NAT10. Queuosine is a hypermodified guanosine derivative at the wobble position of specific tRNAs; eukaryotes salvage a bacterial- or diet-derived precursor and can further glycosylate Q. Internal m7G is well established in stable RNAs, whereas reported internal mRNA sites remain method-sensitive. Dihydrouridine, thiolated bases, wybutosine, and domain-specific anticodon marks remain essential for complete RNA biology even when they receive less attention in mRNA-centered epitranscriptomics.

GlycoRNA extends the glycoconjugate concept to RNA. The 2021 discovery study provided convergent evidence that mammalian small noncoding RNAs associate with sialylated and fucosylated N-glycans through canonical glycan biosynthetic pathways. A 2024 chemical study then identified acp3U as an N-glycan attachment site using native-sialic-acid labeling, nuclease and glycosidase fractionation, isotope-sensitive and synthetic-standard mass spectrometry, and DTWD2 perturbation. Those results strongly support at least one covalent N-glycan-RNA linkage, but they do not establish that every glycan signal in an RNA preparation is covalent glycoRNA or that every reported RNA class carries the same linker. A 2024 preprint reported O-glycan biosynthetic dependencies and O-glycans released from small-RNA preparations, while explicitly leaving the atomic O-glycan-RNA linkage unresolved. A 2025 critical study showed that RNase-insensitive N-glycoconjugates can co-purify with RNA and can appear RNase-sensitive when post-digestion silica cleanup removes them. GlycoRNA is therefore chemically supported but analytically demanding: linkage-specific evidence must be separated from labeling, proximity, co-purification, and cell-surface function.

Detection is the major limiting factor for this field. Liquid chromatography coupled to tandem mass spectrometry can identify and quantify nucleosides but loses transcript and site information unless paired with targeted digestion or mapping. Reverse transcriptase signatures, chemical derivatization, bisulfite conversion, immunoprecipitation, enzymatic assays, direct RNA nanopore sequencing, and programmable nanopore or latch-based systems each measure different proxies. A peak, mismatch, deletion, current shift, or antibody-enriched region is not automatically a precise modified site. Strong claims combine chemical identity, site localization, stoichiometry, perturbation of the writer or guide machinery, and rescue or site mutation.

The chapter therefore uses a mark-by-RNA-class matrix rather than treating every occurrence of the same chemical group as one mechanism. The matrix distinguishes well-established, often high-occupancy marks in tRNA, rRNA, and small nuclear RNA from lower-occupancy or still-disputed assignments in messenger RNA and viral RNA. It also distinguishes endogenous site-specific modification from synthetic transcript-wide substitution. This organization prevents three common category errors: importing a tRNA decoding mechanism into mRNA, treating a global nucleoside measurement as a site map, and treating a writer-enzyme phenotype as proof for one proposed substrate.

## Concept Inventory

- **Pseudouridine:** an isomer of uridine in which the uracil base is attached to ribose by a carbon-carbon bond rather than the usual nitrogen-carbon glycosidic bond. The change creates a different hydrogen-bonding and hydration surface while preserving a uridine-like Watson-Crick edge. A pseudouridylated base can still pair with adenosine, but local RNA structure and protein contacts may change.
- **Pseudouridine synthases:** enzymes that convert uridine to pseudouridine without requiring an external energy source. Stand-alone pseudouridine synthases recognize RNA substrates through enzyme-RNA contacts. H/ACA guide RNPs use a guide RNA to base-pair with the substrate and position a target uridine in a catalytic pocket, with dyskerin or its homolog providing catalytic activity.
- **2′-O-methylation:** methylation of the ribose 2′-hydroxyl. The mark is often abbreviated Nm, where N means any base. Am, Cm, Gm, and Um specify the modified base. Ribose methylation changes local sugar chemistry and can reduce some forms of nucleolytic or immune recognition.
- **C/D box small nucleolar RNPs:** guide-directed methylation machines. The guide RNA contains conserved sequence motifs and a region complementary to the target RNA. Base pairing positions the target nucleotide a defined distance from the guide motif so fibrillarin can methylate the ribose.
- **Inosine:** the deamination product of adenosine. In RNA sequencing data, inosine is often detected indirectly because reverse transcriptases frequently read inosine as guanosine, producing apparent A-to-G differences relative to the genome. This signal must be separated from genomic variants, mapping artifacts, and editing in overlapping isoforms.
- **M5C:** 5-methylcytidine, a cytidine residue whose cytosine base is methylated at carbon 5. In RNA, m5C is installed by RNA cytosine methyltransferases such as NSUN-family enzymes and DNMT2/TRDMT1. The related base-level phrase 5-methylcytosine also appears in DNA, but DNA m5C and RNA m5C have different enzymes, substrates, detection workflows, and biological interpretations.
- **Ac4C:** cytidine acetylated at the exocyclic N4 position. NAT10 is the best-known mammalian enzyme associated with ac4C installation. ac4C claims in mRNA require caution because antibody and enrichment methods need validation against chemically specific site mapping.
- **Queuosine:** a complex modified nucleoside derived from guanosine and found at the wobble position of tRNAs with GUN anticodons, especially tRNAs for asparagine, aspartate, histidine, and tyrosine. Queuosine links RNA modification to metabolism, diet, microbiota, and decoding kinetics.
- **Wybutosine:** a wyosine-family guanosine hypermodification at position 37, immediately 3′ of the anticodon, in eukaryotic tRNA^Phe. The older spelling *wyebutosine* appears in biochemical literature. In budding yeast, stepwise elaboration of G37 through m1G37 and wyosine-family intermediates to mature wybutosine tunes anticodon-loop behavior and programmed frameshifting rather than acting as an all-or-none decoration.
- **Stoichiometry:** the fraction of molecules modified at a site. A tRNA position can be nearly fully modified, while an mRNA site may be present only in a small subpopulation of transcripts. Function must be interpreted in relation to occupancy and to the identity of the modified RNA population.
- **m7G:** 7-methylguanosine, a positively charged guanosine derivative best known as the cap nucleotide of eukaryotic mRNA but also present internally in selected tRNAs, rRNAs, and proposed mRNA sites. Cap m7G and internal m7G are topologically and functionally distinct.
- **Modification site:** a chemically defined nucleotide at a particular position in a particular RNA molecule or isoform. A nucleoside detected in a purified RNA class is not yet a site, and a mapped region is not necessarily single-nucleotide resolution.
- **RNA-class context:** the molecular constraints imposed by the substrate RNA. The same mark can support stable folding in a tRNA, active-site organization in an rRNA, spliceosome dynamics in an snRNA, or context-dependent RNA fate in an mRNA.
- **Modification circuit:** a bounded causal chain linking substrate recognition, chemical installation, fractional occupancy, a physical consequence, and a cellular phenotype. This chapter uses the term descriptively and does not assume that every mark has a dedicated reader or eraser.
- **GlycoRNA:** an RNA molecule bearing a covalently attached glycan. The term should be reserved for a molecular claim supported by linkage-sensitive evidence; a glycan signal that merely survives RNA extraction or lies near RNA is an RNA-associated glycoconjugate until covalent attachment is demonstrated.
- **Cell-surface RNA:** an RNA exposed on the extracellular face of a living cell, regardless of whether the RNA is glycosylated. Cell-surface RNA and glycoRNA are overlapping but non-equivalent classifications: glycan-centric assays do not by themselves establish extracellular topology, and topology-centric assays can recover non-glycosylated RNA.
- **Acp3U:** 3-(3-amino-3-carboxypropyl)uridine, a modified uridine found in defined tRNAs and supported as an N-glycan attachment site in mammalian glycoRNA. Evidence for this linker does not imply that acp3U is the only glycosylated RNA base or that all acp3U molecules are glycosylated.
- **SialoglycoRNA:** glycoRNA whose glycan carries terminal sialic acid. Sialic-acid-directed metabolic reporters and periodate labeling enrich this subset, so a sialoglycoRNA assay is not an unbiased inventory of all N- or O-linked RNA glycans.

## What to Know Before Reading This Chapter

The reader should distinguish base modification from ribose modification. Pseudouridine, m5C, ac4C, inosine, queuosine, and m7G alter the nucleobase or replace it with a more complex derivative. Ribose 2′-O-methylation alters the sugar. This distinction matters because base modifications can change pairing, stacking, reader recognition, or decoding, whereas ribose methylation often changes backbone geometry, nuclease sensitivity, and recognition by enzymes or immune sensors.

The reader should also distinguish enzyme-directed and guide-directed site choice. A stand-alone tRNA pseudouridine synthase or NSUN enzyme can recognize features of a folded RNA substrate directly. A C/D box or H/ACA guide RNP can use RNA-RNA base pairing to select a target in rRNA or snRNA. Guide-directed modification resembles a programmable recognition system, but it still depends on RNP assembly, guide expression, substrate accessibility, and nuclear compartment.

Finally, the reader should approach mRNA modification claims with the evidence standards introduced in [Chapter 46](chapter1043.md). Many tRNA and rRNA marks are abundant and structurally entrenched. Many proposed mRNA marks are lower occupancy, cell-state-specific, or method-sensitive. A valid discovery map is not automatically a functional mechanism. A valid functional perturbation of a writer enzyme is not automatically proof that a single modified site caused the phenotype.

Five ownership boundaries keep the comparison coherent. [Chapter 47](chapter1163.md) compares enzyme folds, catalytic chemistries, cofactors, kinetic strategies, specificity, evolution, inhibitors, and biotechnology across modification families. [Chapter 46](chapter1043.md) owns the general logic for deciding whether an epitranscriptomic claim is established, likely, or artifact-prone. [Chapter 52](chapter1048.md) owns cross-mark regulation and coordinated modification dynamics. [Chapter 107](chapter1102.md) owns cell-surface RNA topology, extracellular RBP assemblies, trafficking, and immune or intercellular functions, while [Chapter 132](chapter1120.md) owns protocol-level detection and quantitative measurement workflows. The present chapter teaches enough chemistry and assay logic to understand biological claims, then asks which RNA class carries the mark, at what occupancy, in which organism or cell state, and with what functional evidence.

## 49.1. Pseudouridine across rRNA, tRNA, snRNA, mRNA, and synthetic RNA

Pseudouridine is often called the fifth nucleoside because it is abundant in stable cellular RNAs and was recognized early in RNA chemistry. The term can be misleading if it suggests a single uniform function. Pseudouridine is a chemical state that can stabilize a tRNA anticodon stem-loop, shape a ribosomal RNA active region, support snRNA-dependent splicing, tune mRNA translation, or alter immune recognition depending on location and RNA context. The first interpretive step is therefore to name the RNA and site rather than asking what pseudouridine "does" in the abstract.

![Figure 49.1. One Chemical Name, Different Biological Objects](../assets/figures/chapter1045_figure1.png)

**Figure 49.1. One Chemical Name, Different Biological Objects.** Pseudouridine, 2′-O-methylation, inosine, m5C, ac4C, queuosine, and m7G alter different atoms or bonds. The same mark can be embedded in a tRNA anticodon loop, rRNA functional center, snRNA interaction surface, mRNA region, viral RNA, or synthetic transcript. Arrows lead from chemical identity through RNA class, site, occupancy, physical consequence, and phenotype.

The chemical reaction converts uridine into pseudouridine by breaking and remaking the glycosidic arrangement so the uracil base attaches to ribose through C5. Pseudouridine synthases catalyze this intramolecular rearrangement without consuming ATP or using a methyl donor. A conserved catalytic aspartate in many pseudouridine synthase families participates in the reaction, while surrounding protein or RNP contacts position the target uridine. Because the base remains uracil-like on its Watson-Crick edge, pseudouridine can pair with adenosine, but the altered base-ribose connection and added hydrogen-bonding potential can change local folding.

There are two broad ways to install pseudouridine. Stand-alone pseudouridine synthases are proteins that recognize features of the RNA substrate. Different enzyme families act on tRNA, rRNA, snRNA, or mRNA-like substrates, often using shape and sequence context together. Guide-directed pseudouridylation uses H/ACA guide RNPs. In eukaryotes, H/ACA small nucleolar RNAs or related guide RNAs base-pair with the target RNA so the target uridine is presented to dyskerin, the catalytic pseudouridine synthase. The guide does not supply chemistry; it supplies address information.

![Figure 49.2. Guide-Directed Addressing in Stable-RNA Modification](../assets/figures/chapter1045_figure2.png)

**Figure 49.2. Guide-Directed Addressing in Stable-RNA Modification.** Guide RNAs base-pair with a substrate and expose or position a target nucleotide. H/ACA RNPs direct pseudouridylation and C/D box RNPs direct ribose 2′-O-methylation. Guide complementarity supplies address information, while catalytic proteins supply chemistry; substrate accessibility, RNP assembly, compartment, and accessory factors determine whether a predicted target is modified.

tRNA is the clearest textbook example because tRNAs are compact folded adapters whose function depends on precise local geometry. Pseudouridines in tRNA can stabilize stems, tune anticodon-loop structure, and influence aminoacylation or decoding in site-specific ways. A recent review of tRNA pseudouridine synthases emphasizes that modification sites, enzyme families, and effects differ across domains of life and across tRNA positions. A pseudouridine in the T loop is not interchangeable with one in the anticodon loop, and an enzyme knockout phenotype may reflect multiple tRNA substrates.

rRNA and snRNA pseudouridines illustrate guide-directed RNP logic. Ribosomal pseudouridines cluster near functional centers where small changes in local structure can affect decoding, peptidyl transfer, or ribosome assembly. Spliceosomal snRNAs carry pseudouridines that can support base pairing and RNP conformational transitions during splicing. These stable-RNA examples explain why pseudouridine is often associated with RNA structure. They also provide a caution: a stable-RNA mark that is nearly stoichiometric and conserved is easier to connect to function than a low-occupancy mRNA site.

mRNA and pre-mRNA pseudouridylation have become more visible through transcriptome-scale mapping. A study of human pre-mRNA reported that pseudouridine synthases can modify pre-mRNA co-transcriptionally and influence pre-mRNA processing. Plant RNA maps likewise support multilayered translation control through rRNA, tRNA, and mRNA pseudouridylation. These examples broaden the field beyond stable RNAs, but they do not erase the need for site-level controls. The same enzyme may modify several RNA classes, and a processing phenotype may be indirect unless candidate sites are tested.

Regulated pseudouridylation need not use a dedicated signaling reader. In budding yeast, nutrient limitation changes mRNA pseudouridylation, and heat shock can relocalize Pus7 from nucleus to cytoplasm and expose new targets; stress-induced U2 snRNA sites also affect splicing and growth. These observations support regulation through enzyme abundance, localization, RNA structure, and substrate availability. They also expose a measurement boundary: transcript abundance and method sensitivity can make an apparent cell-type- or stress-specific site disappear or emerge without proving that its fractional occupancy changed.

PUS7 provides a disease-linked example. PUS7 is a pseudouridine synthase discussed in cancer reviews as a regulator of coding and noncoding RNA programs, including possible effects on translation, stemness, and stress responses. The evidence should be graded carefully. Elevated PUS7 expression, altered pseudouridine signal, and cancer phenotypes are not the same claim. Stronger evidence would show a direct substrate, loss of pseudouridine at a defined site after catalytic perturbation, restoration by rescue, and a plausible RNA fate pathway.

A recurring misconception is that pseudouridine always increases RNA stability or translation. In engineered mRNA, substitution with pseudouridine or N1-methylpseudouridine can reduce innate immune activation and improve protein output under some conditions, and N1-methylpseudouridine-incorporated mRNA outperformed pseudouridine-incorporated mRNA in a mammalian expression and immunogenicity study. Endogenous site-specific pseudouridylation is not the same as global substitution in synthetic RNA. Site, stoichiometry, neighboring sequence, purification, delivery, and RNA class determine the result.

The evidence hierarchy differs sharply across these substrates. For a conserved tRNA or rRNA site, purified-RNA chemistry, genetic loss of the relevant pathway, structural localization, and biochemical or translational phenotypes can converge on one molecular interpretation. For a pre-mRNA or mRNA site, the modified molecules may be a minor fraction of an isoform, and the same pseudouridine synthase may act on stable RNAs or have noncatalytic interactions. For synthetic RNA, the causal variable is often a transcript-wide substitution coupled to altered innate sensing, translation, purification, and delivery. Pseudouridine is therefore one chemical object embedded in at least three experimental regimes, and results should only be transferred across regimes when the physical comparison is explicit.

The enzyme chemistry underlying stand-alone and guide-dependent pseudouridylation is compared with other modification reactions in [Chapter 47](chapter1163.md). The biological ownership here is narrower: tRNA folding and decoding, rRNA functional centers and assembly, snRNA conformational transitions, pre-mRNA processing, context-dependent mRNA fate, and engineered-RNA immunobiology. That boundary keeps an enzyme-family comparison from replacing the RNA-class question that determines what a pseudouridine can actually do.

## 49.2. 2′-O-methylation in ribosome biogenesis, spliceosomal RNAs, tRNA, and immune recognition

Ribose 2′-O-methylation is chemically different from base methylation. The methyl group is attached to the ribose 2′-oxygen, not to the base. Because the 2′-hydroxyl is a defining feature of RNA, Nm marks alter a functional surface used by ribonucleases, polymerases, ribosomes, RNA-binding proteins, and innate immune sensors. The notation Am, Cm, Gm, and Um specifies the base whose ribose is methylated; the generic notation Nm means the base is unspecified or variable.

The most intuitive mechanism is C/D box guide-directed methylation. A C/D box small nucleolar RNA contains conserved C, D, C', and D' motifs and antisense elements that base-pair with a target RNA. In the assembled snoRNP, fibrillarin is the methyltransferase. The target nucleotide is positioned a characteristic distance from the guide motif, allowing one catalytic protein to modify many sites by changing guide RNA sequence. This architecture is a biological example of addressable RNA chemistry: guide base pairing chooses the site, while an enzyme shared across guides performs the reaction.

rRNA contains many 2′-O-methylated residues. These sites are enriched in structurally and functionally important ribosomal regions, and they can influence ribosome assembly, decoding, antibiotic sensitivity, translational fidelity, and cell-state-specific ribosome behavior. A study of rRNA 2′-O-methylation dynamics linked changing methylation patterns to cell fate decisions. Such work supports the idea that rRNA methylation is not merely static decoration, but evidence must still distinguish changes in methylation stoichiometry from changes in ribosome biogenesis, growth rate, or cell composition.

snRNAs are another guide-directed target class. Spliceosomal snRNAs require precise base-pairing and RNP interactions, so ribose methylation can affect spliceosome assembly and splicing decisions. GPATCH4 was reported to regulate rRNA and snRNA 2′-O-methylation through DHX15-dependent and DHX15-independent modes. This example is useful because it links guide RNP modification to helicase-associated RNP remodeling. It also warns against a simple one-guide one-site model: accessory factors, RNA remodeling, and subcellular localization can control whether a potential target is modified.

Internal mRNA 2′-O-methylation is an emerging area. A primary study reported that internal mRNA 2′-O-methylation can promote mRNA stability. Another study connected nuclear 2′-O-methylation with splicing through the binding protein FUBP1. These studies suggest that ribose methylation can influence mRNA fate beyond the cap1 and cap2 marks introduced in [Chapter 26](chapter1025.md). However, internal mRNA Nm claims are technically challenging because the modification is chemically similar across bases, may be low occupancy, and can be confounded by abundant rRNA or snRNA contamination if purification is imperfect.

Viral RNA 2′-O-methylation sits at the boundary between RNA chemistry and host defense. Many animal innate immune systems inspect RNA features that differ between self and nonself, including 5′ end state, double-strandedness, and ribose methylation. Viral cap methyltransferases or host-dependent methylation can help viral RNA resemble host RNA and evade sensors such as IFIT-family proteins in some systems. A review of viral 2′-O-methylation emphasizes that the effect depends on virus family, replication compartment, host cell, and timing. It is therefore too broad to say that Nm is simply antiviral or proviral.

Functional interpretation depends on mechanism. Ribose methylation can alter local structure, protect against cleavage, change reader binding, influence translation, or change immune recognition. These outcomes can occur at different sites in the same RNA class. A strong claim should specify whether the mark affects chemistry directly, such as reducing cleavage at a nearby phosphodiester, or acts through a protein, such as a binding factor that recognizes methylated RNA. It should also specify whether the evidence is biochemical, sequencing-based, structural, genetic, or inferred from perturbing a guide RNP component.

tRNA Nm adds a further boundary case. In tRNA, ribose methylation can contribute to local conformational control, decoding-loop behavior, and resistance to cleavage, but the relevant sites and enzymes vary across organisms and organelles. Loss of FTSJ1-dependent anticodon-loop Nm in animal models perturbs codon-specific translation and illustrates how a defined tRNA substrate class can connect modification loss to neurological phenotype. A phenotype attributed to “2′-O-methylation” is therefore incomplete unless it identifies whether the substrate is a ribosomal precursor, spliceosomal RNA, tRNA, messenger RNA, viral genome, or cap structure. The same chemical group can be constitutive in one stable RNA and inducible or low occupancy in another.

Innate immune recognition also depends on molecular topology rather than on an abstract methyl group. Cap-proximal ribose methylation helps some sensors distinguish host-like transcripts, internal Nm can change local cleavage or protein binding, and viral cap methylation can mask a nonself signature. These outcomes do not establish that every Nm residue suppresses immunity. [Chapter 108](chapter1103.md) owns sensor pathways and [Chapter 153](chapter1137.md) owns therapeutic-RNA chemistry; this section supplies the mark-specific bridge needed to interpret those systems.

## 49.3. Inosine as a modified nucleotide: decoding, structure, and RNA-fate consequences

Inosine is a modified nucleoside produced by deamination of adenosine. The exocyclic amino group of adenosine is replaced by a carbonyl oxygen, changing how the base pairs. Many cellular and viral machines interpret inosine as guanosine-like. In sequencing data, this often appears as an A-to-G mismatch between RNA reads and the genomic reference. The word "editing" is used because the RNA sequence information read by downstream machines can differ from the encoded DNA sequence.

The major mammalian A-to-I enzymes are ADAR proteins, which bind double-stranded RNA regions. ADAR1 and ADAR2 are the principal catalytically active enzymes in many discussions, while ADAR3 is generally treated as catalytically inactive or regulatory in most contexts. [Chapter 50](chapter1046.md) treats ADAR enzymes, isoforms, domain architecture, disease variants, and programmable editing strategies in detail. Here the essential point is that inosine is not randomly distributed: it is favored in dsRNA structures, including inverted repeats, structured untranslated regions, coding-region hairpins, and some viral or immune-relevant RNAs.

Inosine can change information at several levels. In coding sequence, A-to-I editing can recode an amino acid if the edited codon is translated as a G-containing codon. In splice sites or splicing regulatory elements, editing can alter splice choice. In untranslated regions, editing can change microRNA target sites, RNA-binding protein motifs, localization signals, or decay elements. In long double-stranded repeats, editing can weaken perfect dsRNA and reduce inappropriate innate immune activation. These mechanisms are distinct and should not be collapsed into one generic "editing changes expression" statement.

Cancer illustrates both the importance and the danger of broad inosine claims. A recent review summarizes A-to-I editing mechanisms and downstream targets in cancer. Editing can influence oncogene or tumor suppressor pathways, immune signaling, transcript stability, and antigenic landscapes. But cancer samples also contain changes in cell composition, inflammation, copy number, splicing, RNA abundance, and ADAR expression. A strong cancer claim therefore requires matched DNA and RNA data, editing-site quantification, enzyme perturbation, phenotypic rescue, and attention to whether the edited transcript is present in the relevant tumor cell population.

Detection of inosine has one apparent advantage: the A-to-G readout in RNA sequencing is intuitive. That same feature creates hazards. Genomic A-to-G variants, RNA-DNA mapping ambiguity, paralogous sequence, repetitive Alu-derived regions, reverse transcriptase biases, and low-quality bases can mimic editing. Proper analysis uses matched genomic DNA or variant databases, strand-aware mapping, replicate filters, editing-aware pipelines, and often orthogonal validation such as Sanger sequencing of cDNA, targeted amplicons, or inosine-specific chemical methods.

![Figure 49.3. Inosine Changes Information at Several Readout Layers](../assets/figures/chapter1045_figure3.png)

**Figure 49.3. Inosine Changes Information at Several Readout Layers.** Adenosine deamination produces inosine, which can be interpreted as guanosine-like in codon decoding, RNA-RNA pairing, reverse transcription, and some protein interactions. Consequences differ between tRNA wobble decoding, selected mRNA recoding or splice regulation, extensively edited double-stranded RNA, and sequencing-based detection.

Inosine also differs conceptually from reversible signaling marks. A-to-I editing changes the base identity of a specific RNA molecule. The edited molecule may be translated, degraded, exported, retained, or sensed differently. There is no simple "reader" equivalent to a YTH-domain m6A reader for every inosine site. Instead, readout occurs through ordinary base-pairing and protein-binding systems that now encounter a G-like base. This is why inosine biology belongs both to the modification chapters and to the editing chapter.

tRNA inosine demonstrates that deamination is not limited to mammalian messenger-RNA regulation. At the wobble position, inosine can recognize more than one codon-ending base and thereby expand decoding capacity. In human cells, ADAT2-ADAT3-dependent I34 is introduced during tRNA maturation, and structural work shows how the heterodimer presents different tRNA anticodon loops to the catalytic subunit. The physical object is a modified anticodon base inside a mature adapter RNA, not a transcriptome-scale A-to-G editing event. Its evidence base therefore comes from tRNA chemistry, decoding assays, genetics, and translation rather than from mismatch calling alone. This contrast is useful: inosine can alter information transfer in both tRNA and mRNA, but it does so from opposite sides of the codon-anticodon interaction.

The structural consequence of inosine is similarly context dependent. Replacing an A-containing pair with an I-containing interaction can weaken or remodel a double-stranded region, alter a protein-binding surface, or change competition between RNA conformations. Extensive editing of endogenous double-stranded RNA can mark a duplex as self-like and limit inappropriate immune activation, whereas editing at a coding site can create a discrete proteoform. Detailed ADAR and APOBEC substrate selection belongs to [Chapter 50](chapter1046.md), and organism-specific editing systems belong to [Chapter 51](chapter1047.md); this chapter owns the common product-level question of how inosine changes pairing, decoding, structure, and RNA fate.

## 49.4. m5C, ac4C, queuosine, m7G, and other marks by RNA substrate class

m5C is 5-methylcytidine, a cytidine residue methylated at carbon 5 of the base. In DNA, the related 5-methylcytosine mark is a central epigenetic feature. In RNA, the shared base chemistry should not import DNA assumptions. RNA m5C occurs in tRNA, rRNA, mitochondrial RNA, and some mRNA or noncoding RNA contexts. NSUN-family enzymes and DNMT2/TRDMT1 install m5C on selected substrates, with specificity governed by RNA structure, sequence, protein partners, and compartment.

tRNA m5C provides a concrete mechanism. Some m5C sites stabilize tRNA structure, protect tRNAs from stress-induced cleavage, or influence decoding. NSUN2 also enters mammalian mitochondria and modifies defined mitochondrial tRNAs, showing that cellular compartment is part of substrate assignment. If an NSUN enzyme is depleted, altered protein synthesis may reflect loss of modification on several tRNAs, altered tRNA stability, stress pathway activation, or changes in codon-specific translation. Direct assignment requires identifying the modified tRNA position, measuring occupancy, and separating the effect of methylation from the effect of losing the enzyme scaffold.

mRNA m5C claims often involve reader proteins and stability. One primary cancer study reported that YBX1 promotes esophageal squamous cell carcinoma progression through m5C-dependent stabilization of SMOX mRNA. This is a useful example of a proposed writer-mark-reader-output pathway: cytosine methylation is linked to an RNA-binding protein, mRNA stability, target expression, and disease phenotype. It should still be read as a specific case, not as proof that all mRNA m5C acts through YBX1 or stability.

Mitochondrial RNA m5C shows how compartment changes meaning. Mammalian mitochondria transcribe compact polycistronic RNAs that are processed into rRNAs, tRNAs, and mRNAs. A study reported that RNA m5C marks mitochondrial double-stranded RNAs for degradation and cytosolic release. This connects RNA modification to organellar RNA surveillance and innate immune exposure. It also illustrates why organism, compartment, and RNA class must be stated: mitochondrial dsRNA handling is not the same as nuclear mRNA stability.

ac4C is N4-acetylcytidine. NAT10 is the main mammalian enzyme associated with ac4C installation. Stable-RNA foundations are direct and site-defined: NAT10-dependent ac4C1842 in human 18S rRNA is linked to pre-18S processing, and conserved tRNA ac4C sites can be recovered by chemical mapping, with tRNA targeting requiring the adaptor THUMPD1 in humans.

The endogenous mRNA landscape is less settled. A landmark antibody-enrichment study reported ac4C-containing mRNA regions and associated coding-sequence acetylation with greater mRNA stability and translation efficiency after NAT10 perturbation. In contrast, quantitative chemical ac4C-seq recovered the expected human and yeast stable-RNA sites but detected no endogenous human or yeast mRNA sites in the tested conditions; the method could detect low-occupancy synthetic standards and did reveal mRNA ac4C after ectopic overexpression of acetyltransferase complexes. A current review notes that antibody enrichment cannot identify a nucleotide or quantify occupancy, chemical protocols can lose the base-labile acetyl group, and later chemical or metabolic-labeling studies support mRNA ac4C in additional contexts. A cardiovascular study further reported that NAT10 promotes vascular remodeling through mRNA ac4C acetylation. The appropriate conclusion is not that mRNA ac4C is absent or universal, but that transcript, site, occupancy, cell state, and assay chemistry must be resolved while separating direct RNA effects from NAT10's stable-RNA and protein-related functions.

Queuosine is not a small methyl or acetyl addition. It is a complex 7-deazaguanosine-derived modification found at position 34 of tRNAs for asparagine, aspartate, histidine, and tyrosine. Bacteria synthesize the queuosine pathway de novo, whereas eukaryotes salvage queuine or related precursors ultimately supplied by diet or microbes and insert the base through tRNA-guanine transglycosylase; animals can then add galactose or mannose to selected Q-tRNAs. In fission yeast, queuine availability changes both Q34-dependent codon translation and Dnmt2-dependent m5C38, demonstrating that salvage status can affect decoding and another anticodon-loop mark. A 2023 study reported that glycosylated queuosines in tRNAs optimize translational rate and post-embryonic growth. This makes queuosine a strong example of RNA modification as a link between metabolism, ecology, and gene expression, while the magnitude and direction of codon effects remain organism- and tRNA-dependent.

m7G is 7-methylguanosine. It is familiar from the mRNA cap, while internal m7G is well established at defined tRNA and rRNA positions. The METTL1-WDR4 complex methylates G46 in the variable loop of a subset of tRNAs. In a folded tRNA, m7G46 can participate in a tertiary base triple that supports structural integrity; structural work further shows that METTL1-WDR4 recognizes the tRNA body and bends the variable-loop region to present G46. In cancer models, increased METTL1-WDR4 activity raised the abundance of selected m7G tRNAs, especially Arg-TCT-4-1, and favored translation of AGA-rich growth-regulatory mRNAs; overexpressing that tRNA partly phenocopied oncogenic transformation. This is a defined tRNA-abundance and codon-demand mechanism, not evidence that m7G directly marks those mRNAs.

Internal mRNA m7G has been reported by antibody-crosslinking maps, including 5′-untranslated-region enrichment associated with translation, but chemical mutational profiling did not detect internal mRNA sites in deeply sequenced bacterial or yeast samples; these results define a system- and method-dependent controversy rather than a settled universal distribution. A method review highlights the need to distinguish cap-proximal, internal, abundant stable-RNA, and lower-abundance candidate sites. Comparative treatment of m7G writer chemistry belongs to [Chapter 47](chapter1163.md); the biological question here is which RNA substrate and topology carry the mark.

The molecular meaning of m7G depends especially strongly on topology. At the 5′ cap, m7G is the terminal recognition feature of a multipartite cap structure and participates in translation initiation, processing, export, and decay pathways. At an internal tRNA or rRNA position, the same base modification is embedded in a folded stable RNA and can support local structure or biogenesis. Proposed internal mRNA m7G sites require evidence that excludes cap-containing fragments and abundant stable-RNA carryover. “m7G abundance” is therefore not a biologically sufficient statement unless the assay resolves cap versus internal position and identifies the substrate class.

Less common marks include dihydrouridine, thiolated uridines, wybutosine, archaeosine, lysidine, agmatidine, 1-methyladenosine, 3-methylcytidine, and many others. These marks are not decorative exceptions; many solve specific decoding, folding, thermal-stability, or quality-control problems. This section treats them selectively because a complete atlas would require separate treatment of the chemical structure, biosynthetic pathway, enzyme genetics, organismal distribution, and disease links for every tRNA- and rRNA-centered family.

Dihydrouridine illustrates a folding-centered mark: reduction disrupts aromaticity and can favor local flexibility in tRNA. Thiolated uridines illustrate a decoding- and stress-linked chemical family in which sulfur substitution changes base properties and depends on cellular sulfur handling. Lysidine in bacteria and agmatidine in archaea are chemically distinct solutions that enable AUA decoding while excluding AUG; direct pairing tests for agmatidine support this identity switch. These examples should not be collapsed into an undifferentiated “other modifications” category, and their cataloged structures and distributions should be checked against curated evidence rather than inferred from naming alone.

Wybutosine, abbreviated yW and called *wyebutosine* in some older biochemical literature, gives a particularly instructive reading-frame example. yW is a wyosine-family guanosine hypermodification at position 37 of eukaryotic tRNA^Phe, immediately 3′ of the anticodon. In *Saccharomyces cerevisiae*, sequential pathway steps convert m1G37 through a tricyclic wyosine-family intermediate to fully elaborated yW. An isogenic reporter study compared yeast accumulating m1G37, the intermediate imG14-37, or mature yW and found progressively less programmed −1 ribosomal frameshifting as the position-37 base became more elaborated. The effect was strongest when tRNA^Phe occupied a phenylalanine-specific A-site codon and changed with the slippery sequence, so the result establishes a codon- and reporter-context-dependent fidelity effect rather than a universal frameshift suppressor. Equilibrium A-site binding did not reproduce the functional ordering of the three tRNAs, supporting a kinetic interpretation while leaving the exact ribosomal transition unresolved. The bounded conclusion is that many rare-looking marks have well-defined stable-RNA functions, while their comparative enzyme chemistry belongs to [Chapter 47](chapter1163.md).

Across m5C, ac4C, queuosine, m7G, and rarer marks, the substrate class is the first sorting variable. Stable-RNA sites often have a defined position, high occupancy, and a direct structural or decoding hypothesis. Messenger-RNA sites more often require purification controls, quantitative occupancy, and tests that separate direct site effects from broad enzyme perturbation. Mitochondrial and viral RNAs add compartment- and infection-specific constraints. This is why a single writer-centered narrative is insufficient: one enzyme phenotype can combine defects in several substrates whose biological consequences are not interchangeable.

## 49.5. GlycoRNA chemical identity, glycan linkage, biosynthetic evidence, and molecular boundaries

GlycoRNA names a proposed or demonstrated covalent conjugate between RNA and a glycan. The noun matters: an RNA molecule and a carbohydrate chain must be parts of the same chemical object, not merely present in the same extract, gel region, affinity fraction, membrane domain, or extracellular complex. Glycans are branched oligosaccharides assembled by glycosyltransferases and remodeled by glycosidases. They commonly modify proteins and lipids in the endoplasmic reticulum and Golgi apparatus. Extending this chemistry to RNA therefore raises three linked questions. Which atom or modified nucleoside in RNA carries the glycan? Which glycan class and linkage are present? Which biosynthetic and trafficking steps bring the RNA substrate into contact with the relevant machinery? These questions must be answered separately from what a glycoRNA might do at a cell surface.

GlycoRNA and cell-surface RNA are therefore overlapping but non-equivalent terms. GlycoRNA is defined by covalent glycan attachment and can be studied in extracted material without establishing where the conjugate was located in the intact cell. Cell-surface RNA is defined topologically by extracellular exposure and can include glycosylated RNA, non-glycosylated RNA, and RNA held in surface ribonucleoprotein assemblies. Glycan-centric enrichment does not by itself prove extracellular topology, while surface-selective labeling does not by itself prove glycosylation. [Chapter 107](chapter1102.md) owns the trafficking, anchoring, and cell-surface functions of these populations; the chemical distinction is retained here because treating the two labels as synonyms would overstate both assay classes.

GlycoRNA should not be confused with glycosylated queuosine. Galactosyl- and mannosyl-queuosine are chemically defined derivatives of the tRNA wobble nucleoside queuosine. Their sugar is part of a site-specific hypermodified nucleoside with established tRNA positions and translation-linked functions. The glycoRNA literature instead examines larger, often sialylated N- or O-glycan structures associated with small RNAs and secretory-pathway biosynthesis. Both are forms of carbohydrate-bearing RNA chemistry, but their substrates, enzymes, glycan scale, analytical methods, and biological interpretations differ. Galactosyl-queuosine can also react in some glycan-labeling workflows, making this conceptual distinction an experimental control rather than a matter of terminology alone.

The initial mammalian glycoRNA study used a metabolic chemical reporter related to the sialic-acid precursor N-acetylmannosamine. Cells processed the azide-bearing reporter through glycan biosynthesis, after which click chemistry allowed labeled material in RNA preparations to be visualized or enriched. The signal occurred in small noncoding RNA fractions, was sensitive to ribonuclease and sialidase under the reported workflow, and depended on components of canonical N-glycan synthesis. Glycomic analysis supported highly sialylated and fucosylated structures, and the phenomenon was observed across cultured mammalian cells, tissues, and species. These convergent observations established a serious case that mammalian RNA can participate in glycoconjugates. The study also associated the signal with Y RNAs and other small noncoding RNAs and reported cell-surface presentation. However, metabolic labeling identifies a reporter that entered a glycan; it does not by itself reveal the atom connecting that glycan to RNA, the fraction of a candidate RNA that is glycosylated, or whether every labeled species is covalent glycoRNA.

![Figure 49.4. From Glycan Signal to a Chemically Defined GlycoRNA](../assets/figures/chapter1045_figure4.png)

**Figure 49.4. From Glycan Signal to a Chemically Defined GlycoRNA.** A glycan signal in an RNA preparation is an entry point, not a complete molecular assignment. Nuclease and glycosidase controls constrain the material only when cleanup does not change recovery. Linker-sensitive mass spectrometry and synthetic standards support an asparagine-like N-linkage at the acp3U side-chain locus: the inferred native acceptor is amidated, whereas PNGase F yields the acp3U carboxylate used as a diagnostic product. Intact-conjugate analysis must ultimately join RNA sequence, attachment site, glycan composition, and occupancy. The co-purifying glycoconjugate route remains a competing explanation for bulk signals.

![Figure 49.5. Matched Mechanism Cases for m5C, ac4C, Queuosine, and Internal m7G](../assets/figures/chapter1045_figure5.png)

**Figure 49.5. Matched Mechanism Cases for m5C, ac4C, Queuosine, and Internal m7G.** Four matched lanes begin with a defined stable-RNA substrate and site, pass through the responsible writer or salvage pathway, identify the immediate physical consequence, and end at the strongest bounded output. Representative cases are NSUN2-linked tRNA m5C at position 48, NAT10/THUMPD1-linked tRNA ac4C at position 12, queuine-salvage/TGT installation of queuosine at tRNA position 34, and METTL1-WDR4 installation of internal m7G at tRNA position 46. These stable-RNA cases must not be projected automatically onto proposed messenger-RNA sites.

A subsequent study moved the N-glycan claim closer to atomic resolution. RNA-optimized periodate oxidation and aldehyde ligation, abbreviated rPAL, labels native terminal sialic acids after mild oxidation rather than requiring a metabolic reporter. The investigators enriched rPAL-labeled small-RNA material, digested exposed RNA with nucleases, released glycan-linked material with peptide:N-glycosidase F (PNGase F) or sialidase, and analyzed the products using data-independent tandem mass spectrometry. The modified nucleoside 3-(3-amino-3-carboxypropyl)uridine, or acp3U, was reproducibly enriched among PNGase F-released products. Retention time and fragmentation were compared with synthetic acp3U and acp3U-GlcNAc standards, and isotope-sensitive PNGase F reactions supplied an additional diagnostic of enzymatic release. In the proposed asparagine-like linkage, glycan attachment occurs through a carboxamide nitrogen on an amidated acp3U side chain; PNGase F hydrolysis converts that carboxamide to the carboxylate detected as acp3U. In vitro PNGase F activity on the synthetic glyconucleoside and reduced acp3U abundance plus an altered rPAL profile after loss of the acp3U-installing enzyme DTWD2 strengthened the linkage assignment.

The bounded conclusion is that the acp3U side-chain locus is strongly supported as at least one attachment site for N-glycans on mammalian RNA, with an amidated acp3U acceptor inferred from the N-linkage chemistry. Several steps beyond that conclusion remain models. The amidating enzyme and the native amidated RNA intermediate have not been identified. Oligosaccharyltransferase dependence and secretory-pathway glycan processing are consistent with the observed N-glycan chemistry, yet the responsible RNA-recognition and translocation mechanisms are unresolved. The same enrichment also yielded carboxyl-containing wyosine-family nucleosides yW-72 and yW-86 after PNGase F treatment, but acp3U was the most consistent and quantitatively enriched species. yW-86 is therefore a putative additional linker candidate, not a demonstrated native glyconucleoside, and should not be conflated with the established reading-frame role of mature tRNA^Phe wybutosine. The evidence also does not show that every cellular acp3U is glycosylated, that acp3U is the only RNA-glycan linker, or that a purified intact full-length tRNA bearing a compositionally resolved glycan has been characterized at single-molecule completeness. These distinctions separate a supported linkage locus from a complete biosynthetic pathway.

RNA-class assignment remains narrower than the umbrella term can suggest. acp3U is a known modification of selected tRNAs, and the 2024 linkage study consequently proposed tRNA as a major substrate class. Earlier and newer glycan-centric enrichment or sequencing studies repeatedly recovered tRNAs together with Y RNAs, small nucleolar RNAs, small nuclear RNAs, and other small noncoding RNAs, but a read from an enriched fraction does not prove that the sequenced molecule carried the glycan. Those findings can coexist if glycoRNA is a heterogeneous population, if mature tRNAs or fragments dominate one chemical fraction, or if multiple linker nucleosides and RNA carriers exist. They can also be distorted by differential recovery: a large glycan can make a short RNA migrate at high apparent molecular weight, while size-selection and silica-binding conditions can enrich or deplete conjugates unpredictably. Until an intact conjugate connects RNA sequence, exact modified nucleotide, linker atom, glycan composition, and occupancy in one molecular measurement, claims should specify “small-RNA-associated glycan,” “acp3U-linked N-glycan,” or the directly tested RNA class rather than projecting one carrier assignment onto all glycoRNAs.

Biosynthetic perturbations provide pathway evidence but not automatically direct enzyme-substrate evidence. In the N-glycan studies, perturbing oligosaccharyltransferase or N-glycan maturation altered glycoRNA-associated signal, and loss of DTWD2 reduced acp3U abundance and part of the sialoglycoRNA signal. Such results support dependency on the modified nucleoside and secretory glycan machinery. They do not yet show whether an oligosaccharyltransferase binds RNA directly, whether an RNA adaptor presents the substrate, or whether perturbing glycan synthesis changes the recovery or stability of an associated complex. Catalytic rescue, substrate-selective mutants, compartment-resolved intermediates, and reconstitution with a chemically defined acp3U-containing RNA would narrow these alternatives. Detailed enzyme assays and analytical implementation belong to [Chapter 132](chapter1120.md); the chemical claim here is that pathway dependence is one layer of evidence rather than the linkage itself.

O-linked glycoRNA is an emerging and less settled extension. A 2024 bioRxiv preprint reported that loss of GALNT1/2/3, C1GALT1, or its chaperone COSMC changed sialoglycoRNA labeling, whereas perturbations of O-glycan branching or sialylation altered signal intensity or migration in ways compatible with O-glycan structure. The study also released O-glycans from small-RNA preparations by reductive beta-elimination and characterized Core-1, Core-2, O-GalNAc, and O-fucose-related compositions by data-dependent and data-independent mass spectrometry. Cell lines and colon organoids showed different profiles, and the authors explicitly reported that the direct O-glycan-RNA linkage was not resolved. Because the study is a preprint and because glycan release identifies carbohydrate in an RNA preparation rather than the carrier atom, its strongest current interpretation is that O-glycan biosynthesis contributes to RNA-associated glycan signals and that O-glycan-like structures are recoverable from carefully processed small-RNA material. Direct covalent O-glycosylation of a defined RNA nucleotide, its transferase, and its occupancy remain open.

Probe chemistry determines which slice of the glycoRNA population is visible. A sialic-acid metabolic reporter depends on uptake, metabolic conversion, cell viability, glycosyltransferase tolerance, and incomplete labeling of the native sugar pool. rPAL preferentially detects terminal sialic acids and must suppress oxidation of the 2′,3′-vicinal diol at an RNA 3′ end. Galactose oxidase-based labeling can broaden access to terminal galactose or N-acetylgalactosamine but can damage RNA or react with glycosylated queuosine if enzyme preparation and reaction conditions are not controlled. Lectins and antibodies report binding specificity or proximity, not a unique glycan structure. Glycosidase sensitivity constrains the glycan class only within the enzyme's substrate range. Nucleoside mass spectrometry can support a linker identity after controlled release, while glycomics after glycan release describes carbohydrate composition but usually loses the RNA sequence and attachment site. Protocol-level strengths, calibration, and recovery controls are treated in [Chapter 132](chapter1120.md).

Contamination and co-purification create the most important chemical-claim boundary. A 2025 study reproduced a metabolically labeled N-glycoconjugate in RNA preparations but found that the detected species remained after extensive nuclease treatment when the digest was analyzed without a silica-column cleanup. With the commonly used post-digestion silica step, the same species disappeared because fragmented RNA no longer promoted its retention on the column, creating apparent RNase sensitivity. Increasing alcohol concentration or adding exogenous RNA restored recovery. The RNase-insensitive species was also sensitive to PNGase F and sialidase, so glycosidase response and gel migration did not distinguish it from the originally reported glycoRNA signal. The result demonstrates that proteinase treatment, phase separation, silica purification, RNase sensitivity followed by cleanup, and high-molecular-weight glycan signal are not independently sufficient evidence of covalent RNA glycosylation.

The practical cleanup-independent checkpoint is summarized in Box 49.3.

This ambiguity does not erase the linkage-sensitive acp3U evidence. It identifies a parallel molecular species that can mimic simpler glycoRNA assays and therefore raises the validation threshold. A discriminating workflow should verify that a glycan signal disappears upon nuclease treatment before any cleanup capable of changing recovery, track both digest supernatant and retained material, compare orthogonal purification chemistries, include unlabeled and no-probe controls, and test whether glycan and RNA fragments remain physically connected. Stronger linkage evidence then combines a chemically defined released glyconucleoside, isotope or synthetic-standard matching, genetic dependence of the linker nucleoside, and—ideally—intact conjugate mass or another method that joins RNA sequence, attachment site, and glycan composition. Proximity imaging, cell-surface RNase effects, and receptor binding can address topology or function only after the molecular species is bounded.

The evidence status is consequently tiered. Established: glycan-bearing material can be recovered with mammalian small-RNA preparations, and acp3U has strong direct chemical support as an N-glycan attachment site. Strong but incomplete: canonical N-glycan biosynthesis, DTWD2-dependent acp3U production, and sialic-acid-containing glycoforms contribute to the observed conjugates. Context-dependent: the relative contributions of tRNAs, Y RNAs, other small RNAs, N-glycans, and O-glycans vary by assay and biological system. Disputed or ambiguous: a bulk glycan band in an RNA preparation, even when apparently RNase-sensitive after cleanup, cannot be assumed to be covalent glycoRNA. Speculative: the acp3U amidating enzyme, RNA entry into the secretory lumen, the full set of attachment nucleosides, and direct O-glycan linkage chemistry. [Chapter 107](chapter1102.md) takes the chemically bounded species forward into cell-surface topology, extracellular assemblies, trafficking, receptor interaction, and immune or intercellular functions.

## 49.6. Comparative distribution, stoichiometry, cell state, and organismal context

tRNA is the densest modification platform in this chapter. A mature tRNA is a folded adapter that must be aminoacylated by the correct synthetase, bind elongation factors, enter the ribosome, decode codons, maintain reading frame, and survive quality control. Modifications in the anticodon loop tune decoding and frameshifting. Modifications in the D loop, T loop, and core stabilize folding and identity. Because one enzyme can modify many tRNA species, a writer defect can cause codon-biased translation changes, unfolded tRNA stress, or tRNA fragment production.

The mark-by-RNA-class matrix in Table 49.1 makes the chapter’s central comparison explicit. A filled cell means that the mark has a biologically relevant assignment in that RNA class; it does not mean that every candidate site is equally established. The matrix separately flags mature stable-RNA foundations, context-dependent messenger- or viral-RNA assignments, and engineered substitution. Reading across a row asks how one chemical mark changes meaning between substrates. Reading down a column asks how a particular RNA class combines several chemistries to solve folding, decoding, assembly, processing, sensing, or stability problems.

**Table 49.1. Mark-by-RNA-Class Coverage Matrix.** RNA class changes the meaning of a chemical mark. Stable-RNA foundations should not be transferred automatically to messenger, viral, mitochondrial, or engineered RNA contexts.

| Mark | tRNA | rRNA | snRNA | mRNA / pre-mRNA | Mitochondrial RNA | Viral RNA | Synthetic RNA |
| --- | --- | --- | --- | --- | --- | --- | --- |
| **Pseudouridine** | `F`: folding, anticodon-loop behavior, decoding | `F`: assembly and functional-center geometry | `F`: spliceosomal pairing and transitions | `C`: processing, translation, or stability at defined sites | `C`: organelle-specific substrates require assignment | `C`: virus- and host-specific consequences | `E`: pseudouridine or N1-methylpseudouridine substitution changes sensing and output |
| **2′-O-methylation** | `F/C`: local fold, cleavage resistance, decoding context | `F`: biogenesis, active regions, fidelity | `F`: spliceosomal RNP function | `C`: internal Nm must be separated from cap1/cap2 | `C`: organelle-specific sites and pathways | `C`: cap mimicry, replication, and immune discrimination | `E/C`: cap chemistry and internal design are distinct |
| **Inosine** | `F`: wobble decoding in defined tRNAs | `—`: not a major comparative owner here | `C`: editing can affect spliceosomal substrates or interactions | `F/C`: recoding, splicing, structure, RNA fate, and dsRNA self-recognition | `C`: organism- and compartment-specific editing | `C`: host or viral editing depends on strand and infection stage | `C`: programmable editing produces site-specific inosine rather than global substitution |
| **m5C** | `F`: folding, stability, cleavage, decoding context | `F/C`: biogenesis and local structure | `—`: not a principal substrate class here | `C`: proposed stability, translation, and localization pathways | `C`: processing and dsRNA handling in defined systems | `C`: virus- and host-specific assignments | `C`: may be incorporated or installed for design, but endogenous claims do not transfer automatically |
| **ac4C** | `F`: defined tRNA sites, including THUMPD1-assisted targeting in mammals | `F`: NAT10-dependent 18S sites and pre-rRNA-processing context | `—` | `C`: antibody, chemical, and metabolic-labeling studies give condition- and method-dependent landscapes | `C`: assignment must specify organelle and substrate | `C`: limited, system-specific evidence | `C`: possible design variable; not equivalent to NAT10-dependent cellular biology |
| **Queuosine** | `F`: wobble decoding, translational rate, diet/microbiota linkage | `—` | `—` | `—`: not a messenger-RNA mark in this chapter | `C`: organellar occurrence depends on organism and tRNA import or modification pathways | `—` | `—` |
| **m7G** | `F`: METTL1-WDR4-dependent G46, tertiary structure, tRNA abundance, and codon-biased translation | `F/C`: internal sites and biogenesis | `—` | `F` at the 5′ cap; `C` for proposed internal sites with conflicting mapping results | `C`: topology and transcript class must be resolved | `C`: cap and internal contexts depend on virus | `E/C`: cap design is distinct from internal m7G |
| **Dihydrouridine, thiolated bases, wybutosine, and other stable-RNA marks** | `F`: folding, decoding, reading-frame control, identity, stress response; yeast tRNA^Phe yW37 gives a codon- and slippery-sequence-dependent frameshifting example | `C`: selected marks with site-specific functions | `C`: selected marks; not a unified family | `C`: individual claims require mark-specific validation | `C`: domain- and organelle-specific repertoires | `C`: individual cases only | `C`: specialized design uses require separate evidence |
| **GlycoRNA** | `F/C`: acp3U-linked N-glycan strongly supports selected tRNA carriers; occupancy unresolved | `—`: bulk rRNA-region migration is not carrier assignment | `—`: no established principal substrate in this chapter | `—`: no established principal substrate; enrichment does not prove linkage | `—`: not established | `—`: not established | `C`: synthetic conjugates can calibrate linkage and nuclease behavior but are not endogenous biosynthesis |

rRNA modifications are embedded in ribosome biogenesis. Pseudouridine and 2′-O-methylation sites often occur near decoding, peptidyl transferase, subunit-interface, or ligand-binding regions. These marks are installed as pre-ribosomal particles assemble, so loss of a guide RNA, dyskerin, fibrillarin, or accessory factor can affect both mature ribosome chemistry and ribosome assembly. Functional ribosome heterogeneity claims should therefore separate changed modification stoichiometry from changed ribosome abundance, altered growth, and selection for different ribosome subpopulations.

snRNA marks support spliceosome function. Spliceosomal snRNAs participate in base-pairing networks with pre-mRNA and with one another. Pseudouridine and 2′-O-methylation can stabilize selected conformations or tune protein binding. GPATCH4-dependent effects on rRNA and snRNA methylation provide one example of how modification state can be linked to RNP remodeling factors. Cross-chapter handoff: [Chapter 27](chapter1026.md) treats spliceosomal RNP architecture, while [Chapter 50](chapter1046.md) treats editing events that can also affect splicing.

mRNA modifications are often lower occupancy and more condition-specific. That does not make them unimportant, but it raises the evidence threshold. mRNA pseudouridine, internal 2′-O-methylation, m5C, ac4C, m7G, and inosine have all been proposed to affect stability, translation, processing, or localization. The best mRNA examples combine transcriptome mapping with targeted validation and functional tests at defined sites. Broad writer knockdowns alone are usually insufficient because modification enzymes often have stable-RNA substrates and noncatalytic roles.

GlycoRNA adds a carrier-and-conjugate axis that a nucleotide-site matrix alone cannot capture. The strongest current chemical assignment joins acp3U to an N-glycan in mammalian small-RNA material and makes selected tRNAs plausible primary carriers. Earlier sequencing and enrichment implicated Y RNAs and other small RNAs, while O-glycan evidence remains preparation- and system-dependent. Occupancy is largely unresolved: neither the fraction of a given tRNA species carrying a glycan nor the fraction of acp3U sites converted into glycan linkers is established across cell types. The comparison must therefore track RNA identity, linker nucleoside, glycan class, and intact-conjugate abundance as separate variables.

Viral RNA is an arena of conflict. Viruses can carry, acquire, mimic, or avoid RNA modifications to improve replication, translation, packaging, or immune evasion. Host cells can also modify viral RNA or use modification-sensitive sensors to detect infection. Reviews of viral epitranscriptomic marks and viral 2′-O-methylation emphasize that modification effects vary by virus and host. A ribose methylation that helps one virus evade IFIT recognition may not predict the role of pseudouridine, m5C, or inosine in another virus.

Therapeutic RNA is related but should be treated separately from endogenous mRNA. Global incorporation of modified nucleosides, such as pseudouridine or N1-methylpseudouridine, changes every eligible position in an in vitro transcript and interacts with purification, cap structure, untranslated regions, coding sequence, and delivery. Endogenous pseudouridylation modifies selected sites on selected molecules. The same chemical family can therefore support two different design logics: cellular regulation through site-specific enzymes and therapeutic engineering through controlled synthetic chemistry.

Occupancy is the second sorting variable. A nearly stoichiometric tRNA position can influence most molecules of a decoding adapter, whereas a five-percent mRNA site can only act directly through the modified subpopulation unless it triggers amplification. Bulk changes can also arise from cell composition: a tissue sample may appear to alter a mark because the abundance of the RNA-bearing cell type changed. Time matters as well. A modification can be installed during stable-RNA maturation, induced during stress, lost when a precursor pathway is disrupted, or diluted when a rapidly produced RNA population enters the sample. Quantitative interpretation must distinguish altered occupancy at a site from altered abundance of the RNA that contains it.

Organismal context supplies a third axis. Guide-directed rRNA and snRNA modification is prominent in eukaryotic and archaeal systems, whereas individual tRNA marks and their biosynthetic or salvage routes can differ across bacteria, archaea, and eukaryotes. Animals obtain queuine through environmental and microbial sources, making nutritional ecology part of tRNA chemistry. Plant pseudouridine maps provide evidence across rRNA, tRNA, and mRNA. Viral RNAs encounter host modification machinery and encode or recruit their own cap-modifying functions. A claim should therefore name the organism or domain of life rather than projecting a mammalian writer-substrate relationship onto all cells.

The same restraint applies especially to glycoRNA. The defining experiments have centered on mammalian cells and tissues, although acp3U and glycan biosynthetic machinery have broader evolutionary distributions. Conservation of a possible linker or enzyme does not establish glycoRNA in another lineage. A non-mammalian claim requires direct biochemical evidence for an RNA-linked glycan rather than analogy from sequence conservation.

Cell state can tune both installation and consequence. Proliferation changes ribosome-biogenesis demand; differentiation changes the mixture of ribosomes, tRNAs, and transcript isoforms; nutrient state changes metabolites needed by some pathways; infection changes viral and host RNA populations; and stress can change RNA cleavage, localization, and sensor exposure. [Chapter 52](chapter1048.md) owns coordinated cross-mark dynamics. The role of this chapter is to insist that a reported dynamic change be assigned to a particular chemical mark, RNA class, site or region, occupancy range, cell state, and biological output before a coordinated program is inferred.

> **Box 49.1. Synthetic Nucleoside Substitution Is Not Endogenous Site-Specific Modification**
>
> - Source-note requirements: Use `Andries2015N1mΨ` as the bounded primary example; do not imply universal performance across formulations.

## 49.7. Evidence synthesis across orthogonal mark-specific assays

RNA modification detection asks several different questions. Is a modified nucleoside present in the sample? Which RNA carries it? Which exact nucleotide is modified? What fraction of molecules is modified? Which enzyme installs it? Does the mark change RNA fate or organismal phenotype? No single method answers all questions. A robust study chooses the method that matches the claim and adds orthogonal validation for the questions that the first method cannot answer.

Table 49.2 organizes evidence by the inference each method can support. The table is not a protocol guide; detailed workflows, calibration strategies, and computational pipelines belong to [Chapter 132](chapter1120.md). Instead, it shows why orthogonality matters. Two library preparations that depend on the same reverse-transcription signature do not provide the same independence as combining nucleoside chemistry, site localization, genetic dependence, and a site-specific functional test.

**Table 49.2. Orthogonal Evidence Matrix for Mark-Specific Claims.** Chemical presence, site identity, occupancy, pathway dependence, and biological function are separate claims. Strong studies combine methods based on independent physical principles.

| Evidence layer | Suitable method family | Strongest supported inference | Principal boundary | Mark-specific example |
| --- | --- | --- | --- | --- |
| **Chemical identity** | LC-MS/MS or other validated nucleoside chemistry | A modified nucleoside is present in the analyzed RNA preparation | Complete digestion loses transcript and site; positional isomers and sample-induced chemical changes require controls | Confirms queuosine or ac4C in purified RNA but does not by itself assign a transcript position |
| **Mass-spectrometric localization** | Bottom-up oligonucleotide mapping with multiple RNases or suitable top-down analysis | A modification can be constrained to a sequence fragment or intact RNA | Fragment ambiguity, incomplete coverage, spectral complexity, and RNA-class purity can limit site assignment | Stable-RNA ac4C or agmatidine can be localized when purified RNA and diagnostic fragments converge |
| **Candidate localization** | Chemical derivatization, reverse-transcription signature, enrichment, or native-RNA signal | A region or position is compatible with a mark | Sequence, structure, antibody, and model biases can mimic signal | Pseudouridine and Nm signatures require calibrated modified and unmodified standards |
| **Sequence-change readout** | Matched DNA/RNA and editing-aware analysis | A reproducible A-to-G RNA difference is compatible with inosine | Genomic variants, repeats, isoforms, mapping, and base-quality errors | Inosine calls require strand-aware and genomic controls |
| **Fractional occupancy** | Quantitative standards, calibrated targeted chemistry, or validated probabilistic models | A defined fraction of molecules carries the candidate mark | RNA abundance and cell composition can masquerade as occupancy change | Low-occupancy mRNA sites should be interpreted at the modified-subpopulation scale |
| **Pathway dependence** | Writer, guide, substrate, or catalytic perturbation with rescue | The site or abundance depends on a pathway component | Enzymes act on multiple RNAs and may have noncatalytic roles | NSUN, NAT10, PUS, or guide-RNP perturbation does not identify one causal mRNA by itself |
| **Functional causality** | Site mutation, targeted installation, biochemical reconstitution, reader test, or phenotype rescue | The defined mark contributes to a molecular output or phenotype | Perturbation can alter sequence, structure, or expression independently of modification | tRNA decoding, rRNA assembly, snRNA splicing, and mRNA fate require substrate-matched assays |
| **Covalent glycoRNA identity** | Cleanup-independent nuclease test, glycosidase class control, linker-glyconucleoside MS with isotope and synthetic standards, and intact-conjugate analysis | A glycan is covalently attached through a defined RNA linker and, at the highest tier, to a defined RNA | Co-purifying glycoconjugates can mimic gel migration and cleanup-dependent RNase sensitivity; released glycans lose RNA identity; PNGase F can convert an N-linked carboxamide acceptor into a carboxylate diagnostic product | The acp3U side-chain locus is strongly supported for an N-linkage, while yW-72/yW-86, direct O-glycan linkage, and intact-carrier occupancy remain unresolved |

Mass spectrometry is chemically powerful, but its inference depends on analysis scale. Complete digestion followed by LC-MS/MS can identify modified nucleosides and quantify global abundance with high specificity, yet it removes transcript and site information. Bottom-up mapping instead digests an RNA with one or more endoribonucleases and uses fragment masses and tandem spectra to localize a modification; multiple enzymes can be needed to resolve sequence ambiguity. Top-down analysis retains a larger or intact RNA but faces its own coverage and spectral-complexity limits. In every regime, purification of the claimed RNA class matters, positional isomers can require chromatographic separation and diagnostic fragments, and sample preparation can chemically alter labile modifications. Thus a global nucleoside change is not a site-occupancy change, and an mRNA claim remains vulnerable to abundant tRNA or rRNA carryover.

Reverse transcription signatures can localize candidate sites but are indirect. Some modifications cause stops, misincorporations, deletions, or altered readthrough under particular enzyme and chemical conditions. Pseudouridine mapping methods exploit selective chemical derivatization or reverse transcriptase behavior. Inosine is often inferred from A-to-G changes. Bisulfite sequencing has been adapted for RNA m5C, but RNA structure, incomplete conversion, RNA degradation, and sequence context can produce false positives or false negatives. Method-specific controls are not optional.

Antibody-based enrichment can be useful for discovery but is rarely sufficient alone for single-site claims. Antibodies may cross-react, show sequence bias, enrich fragments rather than single nucleotides, or depend on fragmentation and library preparation. Enrichment peaks should be described as candidate regions until higher-resolution validation is performed. This caution applies especially to lower-abundance mRNA marks, where a small amount of contaminating stable RNA or a highly expressed transcript can dominate signal.

Direct RNA nanopore sequencing is attractive because native RNA passes through the pore without reverse transcription, preserving some modification-dependent current information. The challenge is interpretation. Current shifts can reflect sequence context, structure, motor behavior, multiple modifications, RNA damage, or model limitations. Programmable nanolatch approaches and related methods aim to improve discrimination of single-nucleotide mutations and modifications. These methods are promising, but they still require benchmark RNAs with known modification status and transparent error models.

Site-localization must also handle transcript annotation. A genomic coordinate may belong to multiple isoforms, overlap repetitive sequence, or appear in pre-mRNA and mature mRNA. Cap-adjacent marks depend on transcription start sites. Viral RNAs may be nested, overlapping, antisense, or generated by replication intermediates. Mitochondrial RNAs are processed from polycistronic precursors. A modification map without isoform and RNA-class context can be chemically real and biologically misassigned.

The best evidence chain is layered. First, chemical evidence shows that the mark is present. Second, mapping evidence localizes the mark to a site or narrow region. Third, stoichiometry estimates the modified fraction. Fourth, writer or guide perturbation changes the mark without destroying the whole RNA population. Fifth, site mutation, catalytic rescue, reader perturbation, or biochemical reconstitution links the mark to an RNA fate or phenotype. Most studies achieve part of this chain; conclusions should be worded according to the strongest completed link.

The orthogonal chain must be adapted to the mark. Pseudouridine and 2′-O-methylation commonly require chemical or reverse-transcription behavior calibrated with known modified and unmodified RNAs. Inosine mapping requires genomic and alignment controls because the signature resembles sequence variation. RNA m5C assays must control incomplete conversion and structure-dependent protection. m7G studies must resolve the 5′ cap from internal positions. Queuosine and complex tRNA marks benefit from purified-tRNA chemistry and position-specific analysis because a whole-transcriptome enrichment paradigm is poorly matched to their biology. A generic “epitranscriptome-seq” label does not remove these mark-specific failure modes.

GlycoRNA requires a conjugate-specific chain. Metabolic or chemical labeling establishes a reactive glycan feature; nuclease response tests dependence on intact RNA only if the digest is examined without a recovery-changing cleanup; glycosidase response constrains glycan chemistry but not the carrier; released-glycan mass spectrometry profiles the carbohydrate but disconnects it from RNA; released-glyconucleoside analysis can identify a linker but may not identify the full RNA; and sequencing an enriched fraction identifies candidate carriers without proving that each read was covalently glycosylated. Intact-conjugate analysis or a set of orthogonal bridges is needed to join RNA sequence, attachment site, linkage atom, glycan composition, and occupancy. The co-purifying glycoconjugate identified by Kim and colleagues makes purification order part of the inference, not merely a technical detail.

Functional evidence must likewise match substrate scale. A tRNA modification can be interrogated with aminoacylation, decoding, frameshifting, ribosome transit, tRNA stability, and codon-biased proteomics. An rRNA or snRNA mark can be tested through assembly, structure, biochemical reconstitution, translation, or splicing. An mRNA site calls for isoform-aware occupancy, site mutation or targeted editing, and separation of RNA abundance from translation or localization. A viral mark requires attention to replication stage, strand identity, host cell, and immune state. [Chapter 46](chapter1043.md) provides the general claim ladder; the examples here translate that ladder into mark-specific biological tests.

## 49.8. Development, stress, immunity, disease, and mark-specific controversies

RNA modification function is easiest to prove when chemistry, site, and phenotype are close. A tRNA wobble modification that changes codon decoding can be tested with purified tRNAs, ribosomes, reporters, and codon-biased proteomics. An rRNA modification near a functional center can be linked to ribosome assembly or translation fidelity. An snRNA modification can be linked to spliceosome assembly or splice-site choice. mRNA and viral RNA claims often require more controls because modification sites may be sparse, transient, or confounded by changes in RNA abundance.

Developmental phenotypes can arise through several nonexclusive routes. A stable-RNA modification defect can limit the translational capacity needed by rapidly growing tissues, selectively change decoding of codon-biased programs, impair ribosome or spliceosome assembly, or activate surveillance and stress pathways. Queuosine supplies a metabolically sensitive example: glycosylated queuosines were linked to translational rate and post-embryonic growth in a defined experimental system. rRNA 2′-O-methylation dynamics have likewise been linked to cell-fate decisions. Neither example justifies a generic claim that modification changes cause differentiation. The causal chain must identify the RNA substrate, molecular defect, affected gene-expression process, developmental stage, and rescue.

Stress can change both modification state and the consequences of losing a mark. tRNA marks can stabilize folded molecules or influence cleavage into tRNA-derived fragments; ribosome-biogenesis stress can expose dependencies on rRNA pseudouridine or Nm; and oxidative, nutritional, or proteotoxic stress can alter the abundance of substrates and enzymes. A writer knockout that appears harmless in rich growth conditions may become consequential under stress because the modified RNA is part of a buffering system. Conversely, stress-induced changes in RNA turnover can make global modification abundance fall even when site occupancy on surviving molecules is unchanged.

Immunity adds a self-versus-nonself problem. Endogenous double-stranded RNA edited to inosine can be handled differently from an unedited duplex, viral cap-proximal 2′-O-methylation can help some viral RNAs resemble host transcripts, mitochondrial m5C has been connected to double-stranded RNA handling and cytosolic exposure, and synthetic modified nucleosides can change innate sensing when combined with suitable transcript purification and delivery. These mechanisms act at different sensors and molecular locations. The sensor pathways belong to [Chapter 108](chapter1103.md), but the mark-specific boundary belongs here: chemical identity alone does not determine whether an RNA is immunostimulatory.

Disease links fall into several categories. Germline defects in modification enzymes can perturb development, ribosome biogenesis, neurological function, mitochondrial function, or hematopoiesis. Somatic changes in writer expression can accompany cancer progression. Viral infections can expose or exploit modification-sensitive pathways. Cardiovascular, metabolic, and inflammatory disease studies increasingly report altered modification enzymes or marks. The evidentiary problem is directness. A writer enzyme can have many substrates and noncatalytic interactions, so disease association does not identify the causal RNA or site by itself.

Cancer examples in this chapter are useful but should be bounded. PUS7 has been reviewed as a cancer-linked pseudouridine synthase. YBX1-linked m5C stabilization of SMOX mRNA has been proposed in esophageal squamous cell carcinoma. NAT10-dependent ac4C has been linked to vascular remodeling rather than cancer in the local primary example. A-to-I editing has broad cancer relevance. Each case has its own enzyme, substrate, reader or readout, tissue context, and evidence chain. They should not be merged into a generic "RNA modifications drive cancer" claim.

Open controversies cluster around mRNA marks. Some proposed marks are difficult to detect reproducibly at single-nucleotide resolution. Some enzyme knockdowns produce large phenotypes that may reflect stable-RNA defects rather than mRNA site-specific regulation. Some reader proteins bind many RNAs without strict modification dependence. Some disease associations are based on expression correlations. These concerns do not mean the marks are false; they mean the field must use more precise language and stronger validation.

Another controversy concerns "epitranscriptomic" framing. The term can be useful when it highlights regulated chemical information on RNA. It becomes misleading when it implies stable, heritable, chromatin-like memory or universal writer-reader-eraser symmetry. Many RNA marks are installed during RNA maturation and remain until the molecule decays. Some are nearly constitutive structural features. Inosine is an editing product. Queuosine depends on metabolic availability. The biology is richer than a single regulatory metaphor.

Mark-specific controversies should stay separate. For pseudouridine, a central boundary is endogenous site-specific modification versus synthetic transcript-wide pseudouridine or N1-methylpseudouridine substitution. For internal mRNA Nm and m7G, localization and contamination controls are decisive; for m7G specifically, antibody-crosslinking and chemical mutational-profiling studies have reached different conclusions in different systems. For inosine, alignment and genomic variation complicate site calling, while the biological question spans both extensive repeat editing and selected recoding events. For RNA m5C and ac4C, some high-profile messenger-RNA claims depend on assay specificity, enzyme pleiotropy, and occupancy. For queuosine, the chemical reality in tRNA is established, but the relative contributions of diet, microbiota, salvage, and tissue context to particular organismal phenotypes remain context dependent.

For glycoRNA, the central controversy is not a binary choice between “all glycoRNA is proven” and “all signal is contamination.” acp3U-linked N-glycan evidence uses more chemically direct logic than a bulk glycan band, while co-purifying RNase-insensitive glycoconjugates show that the bulk band can contain a molecular mimic. O-glycan biosynthetic dependencies and released-glycan profiles are promising but remain short of an atomic RNA linkage and were reported in a preprint at the time of this review. Each claim should therefore be graded at its own resolution: association, pathway dependence, glycan composition, linker identity, intact carrier, topology, or function.

The current consensus is balanced. Pseudouridine, 2′-O-methylation, inosine, m5C, ac4C, queuosine, m7G, and other marks are real and biologically consequential in many RNA classes. GlycoRNA adds strong evidence for at least one covalent acp3U-linked N-glycan chemistry while retaining unresolved carrier, occupancy, O-linkage, and biosynthetic questions. The strongest foundations remain tRNA, rRNA, snRNA, well-validated editing sites, and chemically defined linkage evidence. mRNA, viral RNA, and bulk glycoRNA assignments require careful mapping, quantitative occupancy, purification controls, and functional tests. The field is moving from discovery catalogs toward mechanism, stoichiometry, intact molecular identity, and causality.

> **Box 49.2. Evidence Ladder for a Functional Modification Claim**
>
> - Boundary: A study can make a valuable discovery claim without completing the full ladder, but its language should stop at the strongest completed inference.
> - Source-note requirements: Original synthesis coordinated with the general evidence standards in [Chapter 46](chapter1043.md).

> **Box 49.3. RNase Sensitivity Is Only Interpretable If Cleanup Does Not Change Recovery**
>
> - Boundary: The control exposes a molecular mimic but does not refute linkage-specific acp3U evidence. A complete chemical claim still requires independent glycan, linker, RNA-carrier, and intact-conjugate evidence.
> - Source-note requirements: Original teaching box anchored by and contrasted with.

## Recent Consensus

Stable RNA modifications are foundational, not peripheral. tRNA, rRNA, and snRNA marks provide some of the clearest examples of RNA chemistry shaping structure, decoding, ribosome biogenesis, splicing, and stress responses.

Guide-directed modification is a major principle. H/ACA RNPs and C/D box RNPs show how RNA guides can direct enzymes to specific positions in large RNA substrates. This principle connects natural ribosome and spliceosome maturation to the broader logic of RNA-guided molecular recognition.

mRNA marks are plausible and increasingly supported, but their interpretation must be site-specific and quantitative. Pseudouridine maps, internal mRNA 2′-O-methylation studies, m5C examples, and ac4C studies support biological roles, while paired ac4C and m7G results show why assay principle, cell state, and organism-specific disagreements must remain visible.

Detection technology is a limiting scientific variable. Reviews and method papers emphasize that chemical identity, site localization, and function require different evidence types.

GlycoRNA is supported at more than one evidence level, but those levels must not be collapsed. Mammalian small-RNA preparations contain glycan-bearing material, the acp3U side-chain locus has strong chemical support as an N-glycan attachment site, and canonical glycan biosynthesis contributes to observed conjugates. The abundance of particular intact glycoRNA species, direct O-glycan linkage, complete biosynthesis, provisional wyosine-family linker candidates, and many carrier assignments remain unresolved. GlycoRNA and cell-surface RNA are overlapping rather than synonymous categories, and co-purifying non-RNA glycoconjugates make cleanup-independent nuclease controls and linkage-sensitive chemistry essential.

RNA class is part of the claim, not merely sample metadata. The same chemical mark can have a conserved structural role at a high-occupancy stable-RNA site, a context-dependent fate effect at a lower-occupancy mRNA site, or a design role when substituted throughout a synthetic transcript. Cross-class generalization requires a shared physical mechanism rather than a shared modification name.

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

Open questions:

- Which low-abundance mRNA sites carrying m5C, ac4C, internal m7G, queuosine-related signatures, or less common marks are reproducible, stoichiometric, and mechanistically meaningful?
- How can mapping methods distinguish true internal marks from cap-proximal signal, stable-RNA contamination, sequence bias, and modification-induced reverse-transcription artifacts?
- Which changes attributed to dynamic rRNA or tRNA modification reflect altered site occupancy, and which reflect altered RNA abundance, maturation, or cell composition?
- When does modification-dependent immune discrimination act directly at a sensor, and when is the phenotype mediated indirectly through RNA structure, turnover, localization, or contaminating double-stranded RNA?
- How broadly can diet- and microbiota-dependent queuine availability explain tissue-specific queuosine phenotypes across animals?
- Which RNA sequences and nucleotides carry intact N- or O-glycans in vivo, at what fractional occupancy, and does acp3U account for most N-glycoRNA or one major subclass?
- Which enzyme amidates acp3U or otherwise prepares the N-glycan acceptor, how does RNA enter secretory-pathway compartments, and does oligosaccharyltransferase contact RNA directly?
- Do the PNGase F-released wyosine-family nucleosides yW-72 or yW-86 represent native N-glycan linkers, and if so, which RNA carriers and linkage structures contain them?
- Can direct O-glycan-RNA linkage be demonstrated on a defined native nucleoside or intact RNA, independently of released-glycan profiles and O-glycan-pathway perturbations?
- How much of each bulk glycoRNA assay signal represents covalent RNA-glycan conjugates versus co-purifying or RNA-associated glycoconjugates?

Common misconceptions:

- "A writer-enzyme phenotype identifies one modified mRNA site." Many writers modify multiple RNA classes.
- "Every pseudouridine stabilizes every RNA." Position, context, occupancy, and RNA class determine the effect.
- "Internal mRNA 2′-O-methylation is equivalent to cap1 or cap2 ribose methylation." Topology changes mechanism and detection.
- "A-to-G RNA-seq mismatches prove inosine." Inosine can appear as A-to-G signal, but genomic variation, mapping artifacts, repeats, editing-site context, and isoforms must be controlled.
- "DNA m5C assumptions apply directly to RNA m5C biology." The enzymes, substrates, and readouts differ.
- "Queuosine is a minor decorative mark." It is a metabolically linked tRNA wobble modification with translational consequences.
- "m7G measurements automatically identify internal mRNA m7G." Cap-derived signal, stable-RNA sites, and internal mRNA candidates must be resolved experimentally.
- "Two sequencing assays are orthogonal because they use different software." Evidence is orthogonal when it depends on independent physical measurements or perturbations, not merely different analysis pipelines.
- "GlycoRNA and glycosylated queuosine are the same modification." Glycosylated queuosine is a defined tRNA wobble hypermodification, whereas glycoRNA denotes larger N- or O-glycan conjugates whose carriers and linkages must be specified.
- "GlycoRNA and cell-surface RNA are interchangeable terms." GlycoRNA is defined by covalent glycan attachment, whereas cell-surface RNA is defined by extracellular topology and can include non-glycosylated RNA or surface RNP assemblies.
- "An RNase-sensitive glycan band from an RNA preparation proves covalent glycoRNA." Cleanup-dependent recovery can make a co-purifying RNase-insensitive glycoconjugate appear RNase-sensitive; linkage-sensitive chemistry and cleanup-independent controls are required.
- "Evidence for acp3U-linked N-glycan proves every proposed glycoRNA carrier and O-glycan linkage." The acp3U result supports one N-glycan attachment route, while carrier breadth, occupancy, alternative linkers, and direct O-linkage remain separate questions.