Advances in mRNA 5-methylcytosine modifications: Detection, effectors, biological functions, and clinical relevance

Abstract

5-methylcytosine (m⁵C) post-transcriptional modifications affect the maturation, stability, and translation of the mRNA molecule. These modifications play an important role in many physiological and pathological processes, including stress response, tumorigenesis, tumor cell migration, embryogenesis, and viral replication. Recently, there has been a better understanding of the biological implications of m⁵C modification owing to the rapid development and optimization of detection technologies, including liquid chromatography-tandem mass spectrometry (LC-MS/MS) and RNA-BisSeq. Further, predictive models (such as PEA-m⁵C, m⁵C-PseDNC, and DeepMRMP) for the identification of potential m⁵C modification sites have also emerged. In this review, we summarize the current experimental detection methods and predictive models for mRNA m⁵C modifications, focusing on their advantages and limitations. We systematically surveyed the latest research on the effectors related to mRNA m⁵C modifications and their biological functions in multiple species. Finally, we discuss the physiological effects and pathological significance of m⁵C modifications in multiple diseases, as well as their therapeutic potential, thereby providing new perspectives for disease treatment and prognosis.

Graphical abstract

In this review, we discuss the detection methods and predictive models of mRNA 5-methylcytosine (m⁵C) modifications. We also surveyed the latest research progress on m⁵C-related effectors and their biological functions. Finally, we discuss the role of m⁵C modification in human disease pathology and its therapeutic potential.

Introduction

Since the discovery of the first post-transcriptional RNA modification, pseudouridine (ψ), in the 1950s,¹ researchers have explored the mechanisms underlying gene regulation at the RNA level. This emerging field has been designated as “RNA epigenetics” or “epitranscriptomics.”2, 3, 4 Compared with the chemical modifications of DNA, RNA modifications are more abundant, with a total of 170 identified to date,⁵ including N6-methyladenosine (m⁶A), 5-methylcytosine (m⁵C), N4-acetylcytidine (ac⁴C), N1-methyladenosine (m¹A), ψ, and 7-methylguanylate (m⁷G). These modifications constitute a relatively unexplored gene-expression regulatory layer, highlighting the complexity and diversity of RNA.⁴ The structural diversity of modified nucleosides plays a key role in maintaining RNA stability,⁶ translation,⁷ and splicing.⁸ Therefore, these modifications have critical roles in regulating the development and pathophysiology of several diseases.9, 10, 11

In m⁵C modifications, a methyl group attached to the fifth carbon of the cytosine ring in DNA and RNA molecules. This modification was first identified on DNA¹² and later on RNA in the 1970s.¹³ It is an important post-transcriptional modification (PTCM) that has significant roles in many biological processes.¹⁴ Hence, m⁵C modifications have recently garnered increased attention and have been reported in many species, including yeast, adenovirus, Phaseolus vulgaris, starfish, Tetrahymena thermophila, rat, Drosophila melanogaster, and wheat germ.15, 16, 17, 18, 19, 20, 21, 22, 23, 24

Studies on m⁵C modifications were initially focused on tRNA and rRNA.²⁵ In tRNA, m⁵C has been shown to participate in optimizing codon–anticodon pairing, maintaining homeostasis, regulating stress response, and controlling translation efficiency and accuracy.26, 27, 28, 29, 30, 31 Meanwhile, m⁵C modifications in rRNA play an important role in conferring bacterial drug resistance,^32,33 glioma sensitivity to bioactive substrates of the stress-related enzyme NQO1,³⁴ thermal adaptability (e.g., m⁵C modifications mediated by ribosomal RNA small subunit methyltransferase F (RsmF) in Thermus thermophilus reduce its thermal adaptability),³⁵ and structural stability of the tertiary rRNA–tRNA–mRNA complex under stress (i.e., promotion of adaptive translation).³⁴ However, although there is a large amount of data on m⁵C in tRNA and rRNA, little is known about its role in mRNA owing to the low abundance of mRNA and the lack of effective separation and purification technologies.³⁶ However, with recent advances in detection methods, studies on mRNA m⁵C modifications have gradually increased.

In mRNA, m⁵C modifications, along with multiple effector enzymes, such as NOP2/Sun RNA methyltransferase 2 (NSUN2),³⁷ NSUN6,³⁸ tRNA aspartic acid methyltransferase 1 (TRDMT1),³⁹ and Aly/REF export factor (ALYREF),⁴⁰ perform numerous functions, including the promotion of mRNA nuclear–cytoplasmic trafficking;⁴⁰ viral protein expression;⁴¹ DNA damage repair;⁴² mRNA stability;⁴³ cell tolerance,⁴⁴ proliferation and migration;³⁹ development, differentiation, and reprogramming of stem cells;^45,46 and regulation of mRNA splicing.⁴⁷ In addition, the distribution of m⁵C varies among cell types.²⁴ m⁵C modifications at specific positions of mRNAs exhibit different regulatory activities:⁴⁸ they can either promote or inhibit translation.^49,50 Therefore, abnormal mRNA m⁵C modifications have been associated with the development and progression of multiple diseases, including cancer,⁵¹ autoimmune diseases,⁵² and arteriosclerosis.⁵³

To date, no systematic review exists on the related effectors, specific biological functions in different species, or the clinical relevance of mRNA m⁵C modifications. Therefore, in this review, we comprehensively summarize the current detection methods, predictive models, related effectors, and roles of mRNA m⁵C modifications in normal physiology and diseased states, which may serve as a guide for researchers as well as new insights into the development of novel diagnostic methods, treatments, and prognostic biomarkers.

Detection methods

The identification of modified target genes is essential to investigating the biological functions of mRNA m⁵C modifications. With recent advances in sequencing technology and experimental methods, methylation modifications of mRNA can be qualitatively and quantitatively analyzed at the single-nucleoside level, providing novel insights into its biological functions. Currently, the primary detection methods for m⁵C modifications include: (1) physicochemical methods, such as chromatography, mass spectrometry (MS), high-performance liquid chromatography (HPLC; Figure 1), and liquid chromatography-tandem MS (LC-MS/MS; Figure 1); (2) chemical conversion, which combines next-generation sequencing (NGS) techniques, such as RNA bisulfite sequencing (RNA-BisSeq), with ten-eleven translocation (Tet)-assisted peroxotungstate oxidation sequencing (TAWO-seq; Figure 2); (3) immunoprecipitation combined with NGS techniques, such as aza-immunoprecipitation (5-Aza-seq; Figure 2) with m⁵C individual-nucleotide-resolution cross-linking and immunoprecipitation (miCLIP; Figure 2); (4) third-generation sequencing (TGS), based on the differences in electrical signals (Nanopore-seq; Figure 2); (5) predictive models, such as a machine learning-based m⁵C predictor trained with features extracted from the flanking sequence of m⁵C modifications (PEA-m⁵C), 5-methylcytosine sites via pseudo nucleotide compositions (m⁵C-PseDNC), Pm⁵CS-Comp-minimum redundancy maximum relevance (mRMR), identifying RNA 5-methylcytosine sites via pseudo nucleotide compositions (iRNA-m⁵C-PseDNC), and m⁵C-heuristic nucleotide physical-chemical property reduction (m⁵C-HPCR). These methods are commonly used in the detection of m⁵C modifications in target genes.

Figure 1.

5-methylcytosine (m⁵C) detection methods based on chromatography and mass spectrometry (MS)

Nuclease is used to digest the RNA molecule into its constituting nucleotides, which are dephosphorylated. These nucleosides are prepared as input for capillary electrophoresis (CE), high-performance liquid chromatography (HPLC), and liquid chromatography tandem MS (LC-MS/MS). CE: in high-voltage electric field drive, nucleoside samples are separated using a quartz capillary column as a separation channel. LC-MS/MS: the LC system is connected to the mass spectrometer. TOP-DOWN MS: without nuclease treatment, using a high-resolution mass spectrometer and various fragmentation patterns of ribonucleotides, RNA can be directly analyzed using MS to obtain information, including the identification of post-transcriptional modifications, relative quantification, and positional information.

Figure 2.

Transcriptome sequencing methods after m⁵C modification

(A) RNA-BisSeq: unmethylated cytosines of RNA fragments could be transformed into uridines when treated with bisulfite, while bisulfite does not affect m⁵C. (B) Peroxotungstate oxidation sequencing (WO-seq) and Tet-assisted WO-seq (TAWO-seq): for WO-seq, hm⁵C can be transformed into trihydroxylated thymine (^thT) by peroxotungstate, after which, the C-to-T mutation site was identified by sequencing. For TAWO-seq, m⁵C was converted to hm⁵C with NgTET1 oxidation, and hm⁵C was converted to ^thT according to the principle of WO-seq. However, the original hm⁵C of RNA was protected from being altered to ^thT by labeling with β-glucosyltransferase (βGT). The mutation of C to T was detected and identified as the original m⁵C but not the original hm⁵C. TGIRT, thermostable group II intron reverse transcriptase. (C) Nanopore-seq: a unique, scalable technology that enables direct, real-time analysis of long RNA fragments. The electrical current was monitored and recorded when nucleic acids passed through a protein nanopore. The modification signals were then decoded and identified along RNA fragments. (D) m⁵C-RIP-seq: RNA fragments containing m⁵C could be pulled down specifically by the anti-m⁵C antibody and then used to construct a sequencing library. The m⁵C peaks were identified against the input as background. (E) 5-Aza-seq: 5-azacytidine (5-Aza-C), a cytidine analog, is incorporated into the RNA molecule by RNA polymerase. RNA (cytosine-5)-methyltransferases (RCMT) can form a temporary intermediate with potential m⁵C residues; however, 5-Aza-C can inhibit the complex separation. The RCMT-RNA complex was pulled down with a specific anti-RCMT antibody, and the pulled-down RNA was used for library construction and sequencing to identify the m⁵C sites as C-to-G transversion (red underlined frame). (F) m⁵C individual-nucleotide-resolution cross-linking and immunoprecipitation (miCLIP)-seq: mutating cysteine (C271) into alanine (C271A) for RCMT inhibited the separation of the enzyme-RNA complex with strong covalent bonding, which was achieved by UV cross-linking, resulting in stop position during PCR. The RCMT-RNA complex was pulled down with anti-RCMT antibody, the RNA was washed to construct the library, and the significant truncation site was considered as the m⁵C site.

Physicochemical methods

In early studies, the detection of modified nucleosides was based on techniques, such as thin-layer chromatography (TLC) and capillary electrophoresis (CE).^54,55 Although these methods are primarily used for the detection of m⁵C in DNA, they are also suitable for RNA.^56,57 The advantages of these techniques include that they are well established, are easy to perform, and can be applied quantitatively.⁵⁸ Radioactive labeling has also been used to improve the low sensitivity associated with these early methods;⁵⁹ they remain unable to achieve single-base resolution, making quantitative assessment of methylation patterns in the natural sequence background difficult.

HPLC

The sensitivity of HPLC detection can reach the nanogram level; moreover, radioactive auxiliary compounds are not required for chromatographic separation, thus avoiding the effects of radioactive labeling on the nucleoside structure.⁶⁰ However, it remains necessary to develop HPLC for the simultaneous detection of multiple RNA modifications and to determine the appropriate HPLC conditions for tandem use with MS.

MS

MS is a physicochemical method that is used for detecting RNA modifications and has been used for more than 40 years.⁶¹ It does not require the conversion of RNA into complementary DNA (cDNA) by polymerase chain reaction (PCR); rather, RNA sequences are directly treated with ribonucleases to catalyze their degradation into nucleosides, which are separated and identified using MS.^62,63 Modified nucleosides are identified based on the increases in mass and comparison with the mass-to-charge ratio of normal nucleosides.⁶⁴ Thus, the stripping of all edited bases and epigenetic information from molecules⁶⁵ and the occasional introduction of substantial artifacts⁶⁶ for reverse transcription can be avoided. MS can also be used for direct RNA sequencing (RNA-seq) and transcriptomic analysis (modification type/level). In theory, MS is suitable for all modification types, including m⁵C and m⁶A.^67,68 However, it has certain limitations: (1) it requires relatively high-target RNA purity and concentration; (2) its sensitivity must be improved compared with other amplification-based detection methods;⁶⁸ (3) considering that MS is primarily used in a serial mode (one RNA sequence at a time), it is unable to meet the required upper limit of the RNA sequence number for analysis, even with multi-channel technology.⁶⁹ In spite of these limitations, MS remains a powerful tool for studying RNA modifications.

LC-MS/MS

Owing to its superior qualitative ability, MS is often used in tandem with LC, e.g., LC-MS/MS. In fact, Cui et al.⁷⁰ used this method to identify dynamic m⁵C mRNA modification in Arabidopsis thaliana. In LC-MS/MS, RNA is digested into oligonucleosides of different lengths by RNases, which are subsequently isolated by LC. Methylation modifications are identified according to the quality difference compared to standard nucleosides, and then modified residues are assigned to sequence sites in oligonucleosides according to gene sequence data.^71,72 Moreover, its sensitivity is at the femtomole level,⁷³ making it suitable for the detection of low-abundance modifications (e.g., mRNA m⁵C modifications). Besides screening for known modified nucleotides, it can also detect those that are unknown, such as the 3-(3-amino-3-carboxypropyl)-5,6-dihydrouridine in tRNA^Lys (tRNA lysine, UUU) from Trypanosoma brucei.^74,75 In principle, LC-MS/MS requires at least tens to hundreds of nanograms of purified molecules, which is not feasible for heterogeneous and low-abundance mRNA. However, as this method involves processing RNA into nucleosides using enzymes, RNA samples cannot exist in their full-length form, which can mask the detection of modified sites, cause loss of base sequence context information, and increase the signal/noise in low-quality areas.⁷⁶

Top-down MS

Unlike LC-MS/MS, top-down MS avoids the complications of enzymatic digestion and the need for oligoribonucleotide separation.⁵⁸ Thus, it provides the additional application in the detection of unknown modifications of RNAs. This method can also reveal site cytosine with altered mass and its modification type without labeling.⁷⁷ Although it uses a denaturing solution, reports have shown little effect on methylated bases.⁷⁸ Top-down MS has a unique potential for the detection, localization, and relative quantification, even though it relies heavily on specialized high-end mass equipment.

Nucleic acid isotope labeling-coupled MS (NAIL-MS)

Notably, the extent of m⁵C modifications differs during the course of development; thus, the fate of methylated RNA during growth should be tracked to provide data on dynamically changing modified nucleosides. NAIL-MS was developed by Matthias Heiss et al.⁷⁹ and can be used to assess the influence of the external environment on the apparent transcriptome. However, due to the difficulty associated with adding internal isotope standards to cultured human cells, its utility is limited for these applications.

Nanoscale liquid chromatography coupled to tandem mass spectrometry (Nano-LC-MS/MS/MS)

With the exception of early TLC and CE, these methods are suitable for the detection of single RNA modification types; however, they cannot simultaneously detect multiple RNA modifications. In 2015, Fu et al.⁸⁰ established a nano-flow LC-MS/MS coupled with a stable isotope-dilution method capable of simultaneously detecting four modified nucleosides (m⁵C, 2′-O-methylcytidine, m⁶A, and 2′-O-methyladenosine). Compared with LC-MS/MS, the new method has a clear chemical specificity as well as superior sensitivity for the identification of analytes.⁸⁰ Moreover, the addition of stable isotope markers in the nucleoside mixture produces more stable and accurate results; thus, it is applicable to studies on RNA modification-related effectors (writer, reader, and eraser).⁸⁰

2-dimensional mass-retention time hydrophobic end-labeling strategy into MS-based sequencing (2D-HELS MS Seq)

Another method for detecting multiple modification types is the 2D-HELS MS Seq. This method introduces 2D HEL into traditional MS, thereby improving the use efficiency of samples, while also broadening the application scope of RNA direct sequencing.^76,81

Chemical conversion methods

RNA-BisSeq

As the gold standard for the detection of m⁵C modifications,⁸² RNA-BisSeq can recognize cytosine methylation sites at single-base resolution.^83,84 Bisulfite treatment chemically deaminates unmethylated cytosine in single-stranded RNA into uracil; however, methylated cytosine remains unchanged. The m⁵C methylation sites on RNA can then be identified by constructing a library and sequencing^85,86 (Figure 2A). Currently, this method is widely used in the study of mRNA m⁵C modification in the tissues and cells of multiple animals and plants, including humans, mice, A. thaliana, and Oryza sativa.^40,44,87

Although RNA-BisSeq is a robust method for detecting m⁵C modification sites in high-abundance tRNA and rRNA, it is associated with high rates of errors for low-abundance mRNAs.²⁵ One cause of this is the permanent loss of base-modification information, as the modified bases are less abundant than unmodified bases. Additionally, RNA-BisSeq converts bisulfite-treated RNA into cDNA, requiring NGS after PCR amplification.^65,88,89 Moreover, the common generation of artifacts can cause the data to become biased.⁹⁰ RNA-BisSeq also cannot sufficiently distinguish m⁵C modifications from other cytosine modifications, such as hm⁵C (5-hydroxymethylcytidine), which interferes with accuracy and validation.⁷⁰ Further, experimental pH conditions damage RNA and even lead to its degradation, thus affecting subsequent experiments and limiting the enrichment of RNA containing m⁵C modifications.⁹¹ Lastly, incomplete deamination occurs in unmodified cytidine treated with bisulfite, which is often present in the RNA stem and double-stranded RNA regions, resulting in false positives.⁸⁷ Therefore, the application of RNA-BisSeq to detect m⁵C modification sites in mRNA requires further improvement.

TAWO-seq

WO-seq is a method for hm⁵C detection.⁹² Yuan et al.⁸⁶ used Tet enzymes to oxidize m⁵C to hm⁵C to establish a new detection method for m⁵C, termed TAWO-seq (Figure 2B). Here, m⁵C is first converted to hm⁵C by peroxotungstate oxidation with Naeglaria Tet-like oxygenase (NgTET1) oxidation, while the original hm⁵C is protected from oxidation by β-glucosyltransferase.⁸⁶ TAWO-seq causes less damage to RNA than does RNA-BisSeq. Furthermore, TAWO-seq does not alter unmodified cytosine but rather directly detects the transformation of modified cytosine, avoiding false-positive results caused by the incomplete transformation of unmodified cytosines in RNA-BisSeq. However, as it relies on the transformation of m⁵C, its incomplete transformation presents a new limitation. Therefore, its transformation efficiency requires further optimization to improve its applicability for mRNA.⁸⁶

Immunoprecipitation

To address the issue of RNA damage associated with chemical transformation methods, immunoprecipitation, primarily using anti-m⁵C antibodies that bind to modified RNA and form noncovalent complexes to locate RNA with m⁵C modifications, has been employed.⁹³

m⁵C RNA immunoprecipitation sequencing (m⁵C-RIP-seq)

Combining chromatin immunoprecipitation sequencing (ChIP-seq) and RNA-seq, this method can enrich for and identify m⁵C modifications in the transcriptome without altering RNA sequence information or requiring RNA-modifying enzymes (Figure 2D). It also avoids the harsh chemical and temperature conditions required for other methods (e.g., the temperature for PCR amplification and acid-base conditions for RNA-BisSeq).⁹⁴ Therefore, m⁵C-RIP-seq is theoretically suitable for detecting the distribution of multiple RNA modifications in the transcriptome⁹⁵ and has been applied to the analysis of bacterial, archaeal, yeast, and plant transcriptomes.⁷⁰ The limitations of m⁵C-RIP-seq include its high dependence on specific antibodies and the risk of nonspecific binding to RNA.⁹³ To address these issues, Weichmann et al.⁹⁶ developed an experimental toolbox capable of verifying antibody performance, while improving the accuracy and credibility of m⁵C-RIP-seq data. However, m⁵C-RIP-seq cannot accurately identify the location of single nucleoside sites, as the length of the sequence reads it produces is generally 100–150 nt and is hindered by the RNA secondary structure.^70,94

5-Aza-seq and miCLIP

These new derivatization methods (Figures 2E and 2F) have been developed based on immunoprecipitation. Compared with m⁵C-RIP-seq, 5-Aza-seq can recognize specific targets of enzymes and specific catalytic sites of RNA (cytosine-5)-methyltransferases (RCMTs) at single-base pair resolution. Specifically, 5-azacytidine (5-Aza-C) is incorporated into RNA to prevent the decomposition of the methyltransferase-RNA complex, transforming the target site of methyltransferase from C to G. The m⁵C modification site can then be identified.^36,37,97,98 This method can identify specific enzyme targets and the specific catalytic sites of RCMTs at the resolution of single bases, thus addressing the issue of nonspecific binding with MeRIP-seq, as well as the inability to accurately locate single nucleosides. Accordingly, Schaefer et al.²⁸ used 5-Aza-seq to characterize DNA methyltransferase (DNMT)2 function in the methylation of tRNA. However, 5-Aza-C destroys RNA integrity;⁹⁹ miCLIP forms irreversible covalent cross-linkages between the enzyme and RNA substrate by mutating the cysteine (Cys) in methyltransferase to alanine. Consequently, reverse transcription during amplification terminates at −1 of the methylation site, and the m⁵C site is mapped at the +1 of the read sequence data.100, 101, 102 Recently, Selmi et al.¹⁰³ mapped the m⁵C modifications in the human transcriptome using miCLIP.

Both 5-Aza-seq and miCLIP depend on the formation of covalent bonds between RNA methylase and its substrate; however, the processes for the formation of this bond differ. Unlike 5-Aza-seq, miCLIP does not affect the methyltransferase step or the structure or stability of RNA.³⁷ However, miCLIP is time consuming and costly and depends on the mutation rate of methyltransferase, with a high rate altering the methylation pattern.¹⁰⁴ Nevertheless, both 5-Aza-seq and miCLIP can be used to identify targets of m⁵C methylase. For example, in 2013, Khoddami and Cairns⁹⁷ discovered the target of methylase NSUN2 in noncoding RNAs, such as tRNA, using 5-Aza-seq. In the same year, Hussain et al.¹⁰² analyzed and confirmed using miCLIP that NSUN2 can act on non-coding RNAs. Thus, both of these techniques can locate targets of the m⁵C methylase NSUN2, thereby expanding the current knowledge regarding its scope of action and function.¹⁰⁵ Although these immunoprecipitation-based methods do not require PCR amplification, their application in trace-volume samples is limited. In addition, their reproducibility is poor due to batch-to-batch differences in antibodies and changes in their affinity.¹⁰⁶ Hence, it is necessary to develop more effective antibody-independent detection techniques in the future.

TGS

Although NGS can accurately localize modified nucleosides within the transcriptome,¹⁰⁷ sequence-by-synthesis technologies can only sequence short RNA reads,⁶⁵ limiting their ability to analyze long, repetitive sequences¹⁰⁸ and the subsequent plotting of genome-wide modification maps. Recently, third-generation, single-molecule long reads (TGS) generating long-read information have been developed, including Nanopore-seq technology as well as single-molecule real-time (SMRT) sequencing technology. These methods do not require RNA recombination or qPCR amplification when sequencing transcriptomes,¹⁰⁹ addressing the limitations of NGS and effectively improving continuity and integrity.¹¹⁰

Nanopore-seq technology offers unique advantages in the analysis of splice variants,¹¹¹ which provide more abundant biological information for RNA research. This technique is based on the principle that when RNA passes through nanopores, the bases cause characteristic changes in the current, which is unique to each normal and modified base. The corresponding bases can then be inferred from the signal-spatial data¹¹² (Figure 2C). In addition, SMRT technology uses reverse transcription to locate m⁶A modification sites on RNA¹¹³ and has been proposed to be capable of detecting other types of modifications in natural RNA.¹¹⁴ However, TGS faces certain challenges. For example, RNA molecules readily degrade and fold, causing inaccurate results when sequencing longer RNA molecules. Although TGS has a high error rate, it can obtain complete genome sequences and transcriptional landscapes,¹¹⁵ while addressing the issues associated with crosstalk and dependence among multiple RNA modifications in the same RNA molecule.⁶⁵ These advantages provide a solid foundation for further studies on biological etiologies and disease pathologies.

Predictive models

Although the above-mentioned experimental detection methods can obtain relatively accurate information on mRNA modifications, they are difficult to perform, time consuming, and costly; they are, therefore, often inappropriate for the in-depth analysis of m⁵C-related mechanisms and functions.¹¹⁶ In the post-genome era, computational tools for predicting m⁵C sites have become both important and feasible. Studies on the optimization of these tools primarily focus on the construction of improved datasets by optimizing mixed datasets and determining the gold standard for negative groups to eliminate the influence of false negatives (unmodified samples). However, studies also focus on determining optimal composite feature extraction methods and classifiers.¹⁴

Currently, 12 models have been developed to predict mRNA m⁵C modifications: PEA-m⁵C,¹¹⁷ a hybrid model for quickly and accurately identifying m⁵C sites from non- m⁵C sites in Homo sapiens RNA (Pm⁵CS-Comp-mRMR), ¹¹⁸ m⁵C-PseDNC, ¹¹⁹ iRNA-m⁵C-PseDNC,¹²⁰ m⁵C-HPCR,¹¹⁶ prediction of RNA 5-methylcytosine sites based on three different kinds of nucleotide composition (RNA-m⁵C-Pred),¹²¹ identifying the occurrence sites of different RNA modifications by incorporating collective effects of nucleotides into PseKNC (iRNA-PseColl),¹²² RNA-m⁵C-finder,¹²³ iRNA-m⁵C,¹¹⁹ a novel method for predicting m⁵C sites of RNA (m⁵C-Pred-SVM),¹²⁴ a new predictor for multiple types of RNA modification sites using deep learning (DeepMRMP),¹²⁵ and a platform for simultaneously identifying multiple kinds of RNA modifications (iMRM) (Table S1).¹²⁶ With the exception of PEA-m⁵C, which specifically predicts m⁵C methylation sites in A. thaliana,^104,117 all of these models can predict human m⁵C modifications. Compared with the m⁵C-PseDNC model, Pm⁵CS-Comp-mRMR is not affected by large or unbalanced datasets and adopts a mRMR algorithm with higher sensitivity;^118,119 however, these two models are not available on web servers.

In the early datasets used to construct predictive models, the numbers of positive and negative sequences were unbalanced, with the former far outnumbering the latter. In addition, these datasets contained minimal information on modifications and faced issues related to data redundancy. For example, iRNA-m⁵C-PseDNC, created in 2017, retains highly similar sequences, leading to higher predictive results, whereas data redundancy in training datasets leads to a decline in generalizability.¹²⁰ In 2018, Zhang et al.¹¹⁶ established a new calculation model based on Met1900 and Met1320. Their m⁵C-HPCR server employs a HPCR algorithm to optimize encoding;¹¹⁶ moreover, its area under the receiver operating characteristic curve, as well as the associated Matthews correlation coefficient, is superior to that of iRNA-m⁵C-PseDNC.¹¹⁶ In 2019, Fang et al.¹²¹ selected three different features, K-nucleotide frequencies (Kmer) + K-spaced nucleotide pair frequencies (KSNPF) + pseudo-dinucleotide composition (PseDNC), and established a new model, RNA-m⁵C-Pred, based on the Met935 database, with higher specificity and sensitivity than the models developed before 2019.

A total of seven site-prediction models have been developed to date: iRNA-m⁵C-PseDNC, iRNA-PseColl, RNA-m⁵C-finder, iRNA-m⁵C, m⁵C-Pred-SVM, DeepMRMP, and iMRM. Of the models capable of predicting m⁵C modifications in humans and mice, iRNA-m⁵C-PseDNC has the highest overall accuracy,¹²⁰ particularly when users want to detect more m⁵C sites in human samples; it is preferred for its high sensitivity. However, this method also has poor specificity.³⁶ For instance, although iRNA-PseColl can identify the methylation sites of three different RNA modifications (m¹A, m⁶A, and m⁵C), it is accompanied by a high false-positive detection rate.¹²²

Unlike iRNA-PseColl, the RNA-m⁵C-finder model can recognize m⁵C modification sites in eight tissues and cell types (mouse_embryonic stem cell, mouse_heart, mouse_kidney, mouse_liver, mouse_muscle, mouse_small intestine, mouse_brain, and human_HeLa);¹²³ however, it cannot predict methylation in fungi.¹¹⁹ Meanwhile, the iRNA-m⁵C model can detect m⁵C modification sites in four different species (Homo sapiens, Mus musculus, Saccharomyces cerevisiae, and A. thaliana) and has a higher predictive capacity than RNA-m⁵C-finder in terms of precision and accuracy.¹¹⁹ The m⁵C-Pred-SVM model can detect m⁵C modification sites in H. sapiens, M. musculus, and A. thaliana by introducing position-specific propensity-related features; its performance, including sensitivity, specificity, overall accuracy, and Matthews correlation coefficient, is superior to that of other existing methods¹²⁴. Similarly, DeepMRMP can predict m¹A, ϕ, and m⁵C modification sites in H. sapiens, M. musculus, and S. cerevisiae RNA,¹²⁵ while integrating multiple modification types in different species, making it both time and cost effective. In addition, iMRM, a recently reported prediction tool based on eXtreme gradient boosting (XGboost), can predict m¹A, m⁶A, m⁵C, ψ, and A to I in the RNAs of H. sapiens, M. musculus, and S. cerevisiae.¹²⁶

Predictive models for a single-modification type or species cannot provide information on the deep features or semantic information of such modification sites. Although these seven predictive models have effectively solved this problem, there remains much room for improvement in the use of these models in the prediction of the types of modification sites that may be involved. There is also a need to improve the predictive abilities of modification sites across different species. Therefore, additional experimental data are needed to improve the quality of datasets that can be used to establish more accurate prediction models.³⁶ However, this does not negate the importance of species-specific prediction models. Considering the high species specificity of eukaryotic m⁵C modifications,¹¹⁹ it is also necessary to optimize and develop more tailored models with higher performance for specific species.¹¹⁹ Currently, the predictive technology of traditional computational models is highly dependent on the effectiveness of relevant feature extraction, whereas it remains challenging to determine the most relevant species-specific feature combinations.¹²⁵ Therefore, it is necessary to compare and verify the predictive performance of models that integrate different feature extraction methods and avoid the impact of noise features on performance.

In summary, the establishment of computational models for predicting potential modification sites provides an excellent supplement to the experimental methods for identifying m⁵C sites. The continued development and optimization of bioinformatic tools will help researchers to better understand experimental data. In the future, predictive and analytical tools will be effectively applied to increasingly complex samples and experimental conditions.

Writers, readers, and erasers related to m⁵C modifications of mRNA

The PTCM of RNA primarily involves three types of effectors:^65,127,128 (1) writers that “write” specific chemical groups into mRNA, which subsequently mediates mRNA modifications; (2) readers that “read” the information contained in these mRNA modifications to maintain mRNA stability and participate in RNA translation and splicing; and (3) erasers that “erase” mRNA modification signals, mediate mRNA modifications, and convert them back into unmodified nucleosides. Currently, identified writers of mRNA m⁵C include NSUN2,³⁷ NSUN6,³⁸ TRDMT1,³⁹ the tRNA-specific methyltransferase 4B (TRM4B),⁷⁰ and Osnsun2,⁴⁴ and readers include ALYREF,⁴⁰ Y box binding protein 1 (YBX1),⁴³ and DNA repair protein RAD52 homolog (RAD52)⁴² (Figure 3). To date, there have been no reported erasers.

Figure 3.

Overview of the effectors (writers, readers, and erasers) related to m⁵C modifications of mRNA

NSUN2, NSUN6, TRDMT1, TRM4B, and OsNSUN2 are RNA methyltransferases responsible for m⁵C modification. (1) m⁵C reader RAD52 recognizes m⁵C-methylated RNA to promote reactive oxygen species (ROS)-induced atypical HR repair of DSBs through the TRDMT1-m⁵C-RAD52-RAD51 axis. (2) m⁵C reader ALYREF recognizes m⁵C-methylated mRNA and transfers mRNA from the nucleus to the cytoplasm. (3) In bladder cancer, m⁵C reader YBX1 recognizes methylated heparin-binding growth factor (HDGF) mRNA and recruits ELAV-like protein 1 (ELAVL1) to stabilize HDGF, finally promoting the proliferation and metastasis of bladder cancer cells.

Writers

Writing of m⁵C into RNA is primarily accomplished by methyltransferases from two families, namely the DNMT family, i.e., TRDMT1, and the nucleolar protein 1 (NOL1)/NSUN protein family, which has seven members in humans (NSUN1 to NSUN7).^129,130

NSUN2 is an enzyme with a wide target spectrum and was first discovered in association with a tRNA modification^131,132 and mRNA methylation.^133,134 It can catalyze the attachment of methyl to the 5^th carbon atom of cytosine in mRNA.⁸⁹ The RNA-recognition motif of NSUN2, as well as the Rossmann-fold catalytic core containing S-adenosylmethionine (SAM) cofactor, plays major roles in catalytic modification. They use the covalent connection between two Cys and the cytosine in mRNA to activate the electron-deficient pyrimidine heterocycle. The carbon 5 then nucleophilically attacks the methyl group of SAM, completing methylation.^135,136 One of the main biological functions of NSUN2-mediated mRNA m⁵C modification is its influence on protein translation.¹³⁷ Modifications at different mRNA sites can promote or inhibit translation. For instance, Xing et al.¹³⁴ reported that NSUN2 methylates the 3ʹ UTR (C1733) of cyclin-dependent kinase 1 (CDK1) mRNA, enhancing its translation by increasing the assembly of ribosomes on CDK1 mRNA. Similarly, Wang et al.¹³⁸ found that NSUN2 mediates the methylation of cytosine at site 466 of interleukin 17A (IL-17A) mRNA to promote its translation in T lymphocytes, mediating homocysteine-induced upregulation of IL-17A expression. Additionally, correlations between increased expression of src homology 2 domain containing (SHC) proteins (p66SHC, p52SHC, and p46SHC), p21, and ICAM-1 (intercellular adhesion molecule 1) and mRNA m⁵C modifications mediated by NSUN2 have been reported.^50,53,139 Tang et al.¹⁴⁰ reported that NSUN2 inhibits p27 translation by modifying C64 methylation (m⁵C) in the 5′ UTR of p27 mRNA. Similarly, Mei et al.¹⁴¹ found that m⁵C modifications in mRNA 3′ UTR mediated by NSUN2 inhibit the expression of p57^Kip2. Schumann et al.¹³⁷ showed that the abundance of m⁵C modifications in the coding sequence (CDS) region negatively correlates, whereas methylation enhancement in the 3′ UTR positively correlates, with mRNA translation efficiency. Therefore, the function of NSUN2 in m⁵C modifications may be closely related to the position of the methylation or the specific modified target gene. Fang et al.³⁸ discovered NSUN6, a new mRNA m⁵C methyltransferase, using CRISPR integrated gRNA and reporter sequencing (CIGAR-seq), a CRISPR-Cas9-based method for the unbiased screening of novel mRNA modification effectors. Using mRNA bisulfite sequencing in NSUN6- and/or NSUN2-knockout a near-haploid human leukemia cell line (HAP1) cells, they reported that the mRNA m⁵C sites targeted by the two methylases do not overlap and that these enzymes are responsible for nearly all of the m⁵C modifications in the mRNA.³⁸

TRDMT1, another type of RCMT that regulates mRNA methylation and inhibits the proliferation and migration of HEK293 cells,³⁹ can catalyze m⁵C modifications in tRNA.^28,130 This indicates that TRDMT1 has a wide target spectrum, similar to NSUN2, and that m⁵C methylase may modify many types of RNAs. Therefore, further investigation is required to determine which methylase mediates m⁵C methylation in different spatiotemporal states and RNA species. Unlike NSUN2, methylation by TRDMT1 only uses Cys at a single site.¹³⁶ Chen et al.⁴² found that TRDMT1 is also a writer of RNA m⁵C at a DNA damage site; its loss hampers homologous recombination (HR) and increases sensitivity to DNA double-strand breaks (DSBs). In addition, TRM4B in A. thaliana has mRNA m⁵C methyltransferase activity.⁷⁰ Another study identified Osnsun2 as an m⁵C methyltransferase targeting mRNA in O. sativa.⁴⁴

As m⁶A is the most common modification type in eukaryotic mRNA,^127,142 many methylases have been reported to catalyze its installation.143, 144, 145, 146, 147, 148, 149, 150 Huang et al.²³ found that certain m⁵C sites maintain high methylation levels in NSUN2 knockdown or knockout cell lines. Furthermore, these m⁵C sites have different sequences and structural characteristics from those dependent on NSUN2, suggesting the involvement of unknown methyltransferases in the catalysis of m⁵C sites, independent of NSUN2 in human mRNA.²³ Collectively, the diversity of m⁶A writers suggests that numerous m⁵C methylases remain to be discovered.

Interestingly, some studies have linked m⁵C to m⁶A. For example, m⁵C mediated by NSUN2 and m⁶A by METTL3/METTL14 have been shown to reciprocally promote each other and enhance p21 expression synergistically at the translation level during oxidative stress-induced cell aging.⁵⁰ Zhang et al.¹³³ found that NSUN2 acts on the 988^th adenine of p16 mRNA to produce m⁶A modification, suggesting that NSUN2 is an m⁶A methyltransferase of mRNA. In addition, crosstalk between m⁶A and m⁵C regulators in human cancers has been proposed.¹⁵¹ However, their precise interactions that are associated with the regulation of mRNA function remain unknown. It is also important to determine whether other types of modifications, apart from m⁶A and m⁵C, also interact. We speculate that these topics may become future research hotspots.

Readers

Reader proteins for m⁶A modification in mRNA have been extensively studied and found to play roles in mRNA splicing, output, stabilization, and translation.152, 153, 154 Relative to studies on mRNA m⁶A readers, those on mRNA m⁵C readers are in the early stages. For instance, ALYREF plays a role in the nucleoplasmic export of mRNA;¹⁵⁵ as a reader, it binds specifically to m⁵C-modified mRNA via K171 (lysine at the 171^st site) to regulate mRNA export.⁴⁰ Meanwhile, YBX1 participates in maintaining the stability of m⁵C-modified mRNA.^43,51 In addition, RAD52 has a higher affinity for hybrid strands containing m⁵C-modified RNA and DNA (complementary pair of RNA), indicating that it is an m⁵C reader of DNA damage sites. Finally, RAD52 promotes the atypical HR repair of DSBs induced by reactive oxygen species through the TRDMT1-m⁵C-RAD52-RAD51 axis.⁴²

Erasers

The reversibility of mRNA m⁶A modifications has been confirmed, with the mediation of RNA demethylases, such as fat mass and obesity-associated protein (FTO)¹⁵⁶ and AlkB homolog 5 (ALKBH5).¹⁵⁷ Early studies identified a demethylation pathway associated with m⁵C modifications of DNA; that is, during DNA modification, m⁵C is oxidized to 5-carboxycystine (5caC) by Tet and then cleaved by thymine-DNA glycosylase to complete the process.¹⁵⁸ The m⁵C modifications can be hydroxylated to form hm⁵C by the modification of mitochondrial (mt) tRNA methionine, with hm⁵C further oxidized to 5-formylcytosine (F⁵C).^159,160 However, the erasers of mRNA m⁵C remain unknown. Shen et al.¹⁶¹ reported that Tet2 mediates the oxidation of mRNA m⁵C. However, in mouse cells, the level of mRNA hm⁵C is lower than that of m⁵C, and the presence of Tet2 protein does not increase the mRNA hm⁵C content. Therefore, unknown proteins may reconvert hm⁵C to m⁵C,¹⁶¹ suggesting that m⁵C modifications may have a reversible erasure pathway. Further research is needed to determine whether mRNA m⁵C modifications are reversible.

mRNA m⁵C modification in disease

Cancer

Multiple mRNA m⁵C effectors, which regulate m⁵C modifications, have been identified as participants in the development and progression of cancer (Figure 4). Specifically, m⁵C plays an important role in cancer cell proliferation and metastasis, as well as cancer stem cell development, by regulating mRNA stability, expression, and translation (Table 1).^{51,141,166,167} Chen et al.⁵¹ showed that YBX1 recognizes m⁵C-modified mRNA via the W65 indole ring in the cold shock domain (CSD) and recruits ELAVL1 (ELAV-like protein 1) to stabilize HDGF (heparin-binding growth factor) mRNA, ultimately promoting the proliferation and metastasis of bladder cancer cells. Zhang et al.¹⁶⁸ found that the number of mRNA m⁵C methylation peaks in liver cancer tissues was significantly higher than that in adjacent normal tissues, with significant differences in the distribution of genes in different positions. These results suggest that the development and progression of liver cancer correlate with mRNA m⁵C modifications. High expression of m⁵C-related effectors NSUN4 and ALYREF is correlated with poor prognosis of patients with hepatocellular carcinoma (HCC).¹⁶⁷ Although the specific underlying mechanisms remain unclear, these results suggest that m⁵C modifications are closely related to the development, progression, and prognosis of HCC. Moreover, NSUN2 is reportedly significantly upregulated in gastric cancer tissues compared with levels in adjacent normal tissues and may promote the proliferation of gastric cancer cells in an m⁵C-dependent manner by inhibiting the expression of p57^Kip2.¹⁴¹ Collectively, these studies indicate that mRNA m⁵C modifications play a role in certain cancers; however, whether m⁵C is involved in the development and progression of other cancers needs to be investigated.

Figure 4.

The mRNA m⁵C modification associations with tumorigenesis and metastasis in multiple cancers

The red dotted line indicates that tumorigenesis and metastasis in multiple cancers are related to m⁵C modifications, whereas the purple dotted line indicates that the disease correlation with m⁵C modifications is uncertain.

Table 1.

Clinical relevance of aberrant m⁵C modifications

Disease	Effectors	Function	Regulation	Mechanism	Refs.
Bladder cancer	NSUN2	writer	upregulated	NSUN2 targets HDGF m5C modification	51
Gastric cancer	NSUN2	writer	upregulated	NSUN2 promotes cell proliferation by repressing p57Kip2 expression in an m5C-dependent manner	141
Gallbladder carcinoma	NSUN2	−	upregulated	NSUN2 promotes the proliferation and tumorigenesis cells by close cooperation with ribosomal protein L6 (RPL6)	[162]
Breast cancer	NSUN2	−	upregulated	hypomethylation of NSUN2 leads to its overexpression	163
Head and neck squamous carcinoma	NSUN2	−	upregulated	−	164
Oral cancer and colorectal cancer	NSUN2	−	upregulated	−	165
Skin cancer	NSUN2	−	downregulation	−	166
Systemic lupus erythematosus	NSUN2	−	downregulation	−	52
Bladder cancer	YBX1	reader	upregulated	YBX1 recognizes m5C-modified mRNA and recruits ELAVL1 to stabilize HDGF mRNA, finally promoting the proliferation and metastasis of bladder cancer cells	51
HCC	ALYREF	−	upregulated	high levels of ALYREF expression are associated with cell-cycle regulation and mitosis	167
Vascular endothelial inflammation	NSUN2	writer	−	NSUN2 promotes the expression of ICAM-1 by upregulating the methylation of ICAM-1 mRNA, thus promoting the development of vascular endothelial inflammation	53

−, unknown function.

It should be noted that only a few studies have reported associations between mRNA m⁵C modifications and cancer, most of which are focused on NSUN2, whereas no evidence has been presented to implicate NSUN2 in cancer development or progression by regulating m⁵C modification (Figure 4). Gao et al.¹⁶² found that a high expression of NSUN2 can promote the proliferation and tumorigenesis of gallbladder carcinoma cells both in vitro and in vivo by closely cooperating with ribosomal protein L6. Using immunohistochemistry, Yi et al.¹⁶³ reported that the expression of NSUN2 correlates with the clinical stage, tumor classification, and pathological differentiation of breast cancer. Furthermore, the hypomethylation of NSUN2 led to its overexpression, thus promoting the proliferation, migration, and invasion of breast cancer cells. The upregulation of NSUN2 has also been correlated with poor prognosis in head and neck squamous carcinoma, oral cancer, and colorectal cancer.^164,165 In contrast, other studies have reported that its deletion in mice enhances the self-renewal potential of tumor-initiating cells, thus promoting the occurrence of skin cancer. Furthermore, the overexpression of NSUN2 in human skin cancer negatively correlates with disease malignancy.¹⁶⁶ Hence, certain questions remain and require further investigation. It is important to determine whether NSNU2 promotes or inhibits cancer development and progression by regulating specific cancer-causing or cancer-inhibiting mRNAs in different cancers. Additionally, validating whether NSUN2 plays a role in cancer development and progression via the regulation of m⁵C modifications is necessary.

The m⁵C modification and its effectors, such as NSUN2 and YBX1, are involved in the development and progression of multiple cancers, suggesting new targets for disease treatment. For example, inhibition of sphingosine kinase (SPHK) in breast cancer can reduce the expression of NSUN2, suggesting that SPHK1 is a potential breast cancer marker.^169,170 The YBX1 phosphorylation inhibitors TAS0612 (multikinase inhibitor) and everolimus (rapamycin complex 1 inhibitor), as readers, can overcome the resistance to estrogen in progressive breast cancer cells.¹⁷¹ However, whether the role of YBX1 inhibitors in the treatment of breast cancer is related to the regulation of m⁵C modifications remains to be elucidated. In particular, whether mRNA m⁵C-related effectors can be used as markers to predict cancer prognosis and whether their inhibitors, combined with conventional chemotherapy drugs, represent effective anti-cancer strategies, remain to be confirmed.^65,166 Meanwhile, DNA modifications play an important role in regulating the characteristics of cancer stem cells. For instance, inhibitors of regulatory enzymes, such as those of histone deacetylase, e.g., vorinostat, and those of DNMT, e.g., azacitidine, decitabine, and SGI-110, are undergoing drug clinical trials for the treatment of multiple cancers.¹⁷² Therefore, increasing our understanding of mRNA m⁵C modifications may lead to improving the diagnosis, treatment, and prognosis of cancer.

Autoimmune diseases

Previously, we found that the 5-methylcytosine/cytosine (m⁵C/C) ratio of CD4⁺ T cells in patients with systemic lupus erythematosus (SLE) was lower than that in healthy controls. The distribution of m⁵C modifications in the mRNA of SLE patients and healthy individuals is highly conservative and primarily concentrated near the mRNA translation initiation site. Hypermethylated and upregulated genes in SLE function in the immune system, including cytokine and interferon signaling pathways.⁵² In addition, we observed the downregulation of NSUN2 and m⁵C modifications in SLE patients; however, many genes were hypermethylated in SLE patients, suggesting the presence of other m⁵C methylases.⁵² For instance, IL-17A is an important mediator in many autoimmune diseases.¹⁷³ In rat T lymphocytes, NSUN2 promotes the translation of IL-17A by methylating its C466 site;¹³⁸ however, it is unclear whether m⁵C modifications of IL-17A mRNA are involved in immune diseases. The specific mechanisms underlying mRNA m⁵C modifications in the remission and recurrence of SLE are also not well understood. Further research is needed to address these questions.

Vascular endothelial inflammation

Luo et al.⁵³ found that NSUN2 promotes the expression of ICAM-1 by upregulating the methylation of ICAM-1 mRNA, thus improving the adhesion between leukocytes and endothelial cells and promoting vascular endothelial inflammation, a key factor in the pathogenesis of multiple vascular diseases, including atherosclerosis, hypertension, restenosis, and ischemia/reperfusion damage.^53,174 Moreover, the lack of donor NSUN2 hindered the formation of allograft arteriosclerosis in a rat model of aortic allograft,⁵³ suggesting that the NSUN2-ICAM-1 regulatory axis is involved in vascular endothelial inflammation. Besides ICAM-1 mRNA, NSUN2 can also catalyze the methylation of other mRNAs and non-coding RNAs.^102,133 Therefore, the mechanism by which NSUN2 regulates the development and progression of vascular inflammation and atherosclerosis requires further investigation.

Clinical relevance and future directions

The clinical relevance of RNA methylation may potentially become a research hotspot in the coming years. However, current studies on the clinical effects of mRNA m⁵C modification tend to focus on neoplastic diseases, with few focusing on non-neoplastic diseases (Table 1).⁵² Hence, broadening the types of diseases in future studies will advance our understanding of the clinical consequences of m⁵C modifications. Additionally, mRNA methylation may be a potential biomarker in the progression of specific diseases. Thus, it is essential to develop novel techniques for the rapid identification of modifications that give rise to diseased states and the quantification of the protein levels associated with these modifications.¹⁷⁵ Further exploration of the role of m⁵C modifications in immune responses will provide a broader prospect for tumor immunotherapy¹⁷⁶ and tumor drug resistance.¹⁷⁷ There have been several companies targeting RNA apparent modifications.¹⁷⁸ The investigation of small-molecule inhibitors that are capable of targeting m⁵C modification-related effector proteins in the context of disease treatment may be a promising research direction.¹⁷⁹ As the preclinical efficacy of these targeted drugs is revealed, the development of RNA epigenetic drugs will enter a new era.

mRNA m⁵C modifications in other species

Plants

There is evidence for widespread mRNA modifications in plants.⁷⁰ Although the PTCMs of mRNA have been reported in A. thaliana, O. sativa, Helianthus annuus, Triticum aestivum, Glycine max, and Solanum tuberosum,¹⁸⁰ most studies on mRNA m⁵C modifications have been conducted on A. thaliana⁸⁷ and O. sativa⁴⁴ (Figure 5).

Figure 5.

Physiological effects of m⁵C modifications in different species

A. thaliana

In A. thaliana, mRNA m⁵C modification sites differ between tissue types, including seeding shoots, roots, siliques, and rosette leaves.^87,181 The low overlap between these sites indicates that mRNA performs tissue-specific functions through specific gene methylation patterns. Notably, the homologous protein TRM4B of human NSUN2 in A. thaliana and the homologous protein Osnsun2 in O. sativa exhibit high sequence conservation. These homologous proteins exist widely in green algae and flowering plants and contain a conserved catalytic domain.^40,44 In addition, a TRM4B mutant has been reported to suppress division in root tip meristem cells, resulting in the short root phenotype of A. thaliana.⁸⁷ The TRM4B mutation is speculated to affect mRNA stability; i.e., decreased levels of m⁵C modifications lead to decreased half-lives of SHY2 and IAA16, which are critical for root development.⁷⁰ Furthermore, m⁵C regulates the transport of TCTP1 (translationally controlled tumor protein 1) mRNA between the rootstock and scion in grafted seedlings, thus regulating root development.¹⁸¹ Not only does m⁵C influence plant development, but it also plays an important role in regulating environmental adaptation. In fact, its biological function is similar to that of m⁵C in mammals.^40,166 Compared with wild-type A. thaliana, the TRM4B mutant is more sensitive to oxidative stress,⁸⁷ the mechanism of which may be related to the loss of tRNA methylation, reducing its stability. However, considering that TRM4B can act on multiple RNAs, whether methylation modifications of mRNA and other types of RNA participate in the oxidative stress response of A. thaliana, needs to be confirmed.

O. sativa

Tang et al.⁴⁴ reported that the conservation of m⁵C sites between O. sativa and A. thaliana is relatively low, indicating that m⁵C modifications in plants may not be as conserved as those in mammals. However, the distributions of m⁵C in mRNAs of A. thaliana and O. sativa are similar, both primarily enriched in the coding sequence CDS located near the initiation and termination codons.^44,70,87 The mutant Osnsun2 presents with short roots;⁴⁴ however, the underlying mechanism remains unclear. Osnsun2 can also exhibit a heat-sensitive phenotype, as Osnsun2 primarily plays a role in O. sativa resistance to heat damage by selectively methylating mRNAs involved in photosynthesis and detoxification-related processes.⁴⁴ Although heat-adaptation mechanisms of Solanum lycopersicum,¹⁸² Citrullus Schrad,¹⁸² and Daucus carota¹⁸³ have been reported, it remains to be determined whether mRNA m⁵C modifications participate in them. Investigating the role of m⁵C modifications in the environmental tolerance of crops can help to increase yields.

Danio rerio

Recently, Yang et al.⁴³ reported that the reader YBX1 recognizes m⁵C-modified mRNA via a π-π interaction between the key residue Trp45 of the CSD and m⁵C. YBX1 then recruits the poly(A)-binding protein Pabpc1a to maintain the stability of maternal mRNA, thus ensuring maternal-to-zygotic transition and promoting early embryonic development. However, the deletion of YBX1 or Pabpc1a leads to stagnation in the blastula and gastrula stages⁴³ (Figure 5). Another study in D. rerio revealed that YBX1 plays a role in oocyte maturation and egg activation by binding to individual mRNAs to inhibit the translation of related proteins¹⁸⁴ (Figure 5), suggesting that the biological function of YBX1 is far more complicated than simply recognizing and stabilizing m⁵C-modified mRNA. Moreover, these roles of m⁵C modifications in the embryonic development of D. rerio suggest that they may also be involved in coordinating embryonic development in other species.

D. melanogaster

Ypsilon Schachtel (YPS), a human YBX1 homolog, reportedly promotes the maintenance, proliferation, and differentiation of D. melanogaster ovarian germline stem cells (GSCs) by preferentially binding to m⁵C-modified mRNA (Figure 5), whereas human YBX1 can functionally replace YPS to promote GSC development in the D. melanogaster ovary.⁴⁵ Considering that the function of YPS/YBX family proteins is highly conserved in D. melanogaster and humans, the D. melanogaster ovary is an attractive model to analyze the function of these proteins and the role of m⁵C modifications in germ-cell development.

Viruses

As a post-transcriptional regulator of HIV-1 mRNA splicing and functions, m⁵C has a direct impact on gene expression.⁴¹ Meanwhile, m⁵C modifications of murine leukemia virus (MLV) transcripts promote the replication and expression of viral genes.¹⁸⁵ m⁵C is more common in the genomic RNA of MLV than in the mRNA of uninfected cells, promoting virus replication in an ALYREF reader-dependent manner¹⁸⁶ (Figure 5). Therefore, since m⁵C regulates viral gene expression, viruses can activate the writers and readers of infected cells (hosts) to increase the expression and replication of their own genes. However, whether epitranscriptomic modifications have a beneficial role in viral life cycles remains controversial, as some studies have demonstrated that the methylation m⁶A inhibits the replication of HIV-1¹⁸⁷ and Zika virus.¹⁸⁸ Therefore, in-depth investigations are needed to reveal the roles of these RNA modification types. The regulation of m⁵C modifications during virus replication provides new insights into viral pathology and suggests the potential of m⁵C methylase inhibitors as antiviral drugs.¹⁸⁹

Future prospects and conclusion

In conclusion, m⁵C modifications have been detected in various organisms and exert unique biological functions in humans and plants,^44,87,89 with certain common features.¹⁹⁰ It is most abundant in tRNA, rRNA, and mRNA where it affects the RNA stability,²⁸ translational fidelity,¹⁹¹ cell differentiation,¹⁹² nervous system regulation, reproductive system development,^193,194 and viral viability.¹⁹⁵ Therefore, m⁵C modifications are important to understand a variety of physiological and pathological processes.

It is necessary to further improve and develop techniques for m⁵C detection to clarify whether m⁵C modifications actually exist in eukaryotic mRNA or are merely the result of experimental artifacts, as has been suggested by some studies.¹⁹⁶ We expect that TGS methods will improve the direct detection of RNA methylation. It is necessary to determine whether RNA m⁵C modifications can be converted into hm⁵C and F⁵C,^159,160 whether m⁵C is only a transient intermediate, and whether the dynamic balance between m⁵C and its oxidation products have any functional relevance. Lastly, it is necessary to determine whether additional m⁵C effector proteins exist and to investigate their potential mechanisms in disease to advance the current understanding. In summary, research related to the many mRNA m⁵C modification-related enzymes and their biological functions is in its infancy, with countless significant discoveries left to be made.