Novel insight into the regulatory roles of diverse RNA modifications: Re-defining the bridge between transcription and translation

RNA modifications can be added or removed by a variety of enzymes that catalyse the necessary reactions, and these modifications play roles in essential molecular mechanisms. The prevalent modifications on mRNA include N6-methyladenosine (m6A), N1-methyladenosine (m1A), 5-methylcytosine (m5C), 5-hydroxymethylcytosine (hm5C), pseudouridine (Ψ), inosine (I), uridine (U) and ribosemethylation (2’-O-Me). Most of these modifications contribute to pre-mRNA splicing, nuclear export, transcript stability and translation initiation in eukaryotic cells. By participating in various physiological processes, RNA modifications also have regulatory roles in the pathogenesis of tumour and non-tumour diseases. We discussed the physiological roles of RNA modifications and associated these roles with disease pathogenesis. Functioning as the bridge between transcription and translation, RNA modifications are vital for the progression of numerous diseases and can even regulate the fate of cancer cells.


Introduction
In the 1950s, the first RNA nucleoside modification was identified [1]; since then, researchers have focused on updating the understanding of RNA modifications. At the very beginning, the 5'cap and the poly(A) tail, which represent cap and tail modifications, respectively, were discovered. However, with the limitations of technology, modifications of eukaryotic mRNA ends were considered the only posttranscriptional alterations to mRNA for a while. Fortunately, this situation did not last for a long time. Internal mRNA modifications have been investigated in succession in the last 50 years. The revealed mRNA modifications included but were not limited to N6-methyladenosine (m 6 A), N1-methyladenosine (m 1 A), 5-methylcytosine (m 5 C), 5hydroxymethylcytosine (hm 5 C), pseudouridine (Ψ), inosine (I), uridine (U) and ribose-methylation (2'-O-Me) [2][3][4] (Figs. 1 and 2). m 6 A is the most abundant modification and was therefore thoroughly investigated [5].
Analogous to mRNA modification, we also identified many modifications on transfer RNAs (tRNAs) and ribosomal RNAs (rRNAs), such as queuosine (Q) [6]. Eukaryotic tRNAs contain, on average, over 10 modifications per molecule. From elementary isomerization or methylation to complicated modifications of ring structures, the number of tRNA modifications is the largest and has the widest chemical variety. Moreover, there are over 200 modifications on human rRNAs. Thus, their less complicated nature and greater abundance led to more investigations of tRNAs and rRNAs, even beyond mRNAs. Early studies have demonstrated that this variety of modifications leads to extra cellular functions for diverse RNA species [7].

The regulatory role of RNA modifications
Modifications on different RNAs were found to regulate various cellular processes. Researchers demonstrated that these modifications can initiate translation, stabilize transcripts, splice pre-mRNA, facilitate nuclear export, etc. [8][9][10][11][12]. With respect to RNA modifications and technological advances in high-throughput sequencing and mass spectrometry, the mechanisms of different cellular processes influenced by RNA modifications are underexplored, including the less ubiquitous modifications on rare   6 A with a widespread distribution prefers to be located in the consensus motif in the 3'UTRs as well as the 5'UTRs, which closely correlate with translation. Although m 1 A-containing mRNA is 10 times less common than m 6 A-containing mRNA, m 1 A is discovered in every segment of mRNA, including the 5'UTRs, CDS and 3'UTRs and mostly in highly structured 5'UTRs. Analogous to m 1 A, m 5 C can occur in coding and non-coding regions of mRNA, especially in GC-rich regions. Nevertheless, m 5 C within different positions regulates transcription differently. Tet-family enzymes prefer to oxidize m 5 C modifications in coding regions, so hm 5 C has a greater possibility of being present in CDS. Subsequently, Ψ is demonstrated to have a diversified location, whereas I is present at a large number of sites in the CDS, and U accumulates in 3'UTRs. 2'-O-Me focuses on decorating specific regions of mRNA that encode given amino acids. Additionally, as reversible modifications, most have their own readers, writers and erasers RNA species. tRNAs, which have the greatest number of types of different chemical modifications, regulate molecular mechanisms by selecting and protecting the reading framework, promoting tRNA decoding capability as well as changing codon-anti-codon connections [13][14][15][16][17]. Moreover, the functions of 2'-O-Me, Ψ and m 5 C, which are abundant in rRNA, have been investigated in detail. Without any doubt, mRNA modifications play roles in modulating molecular mechanisms. Subsequently, RNA modifications contribute to tumorigenesis by regulating cell survival, differentiation, migration and drug resistance [18]. m 6 A RNA modification Introduction to m 6 A RNA modification m 6 A accounts for approximately 0.2~0.6% of total adenosines in mammalian RNA [2,5]. General m 6 A modifications occur in mammals, plants, bacteria and even other types of eukaryotic RNA [19][20][21][22]. In addition to their widespread distribution, there is no less than 1-2 methylated adenosines in every single mRNA [23]. Studies have reported that m 6 A is located in the 3' untranslated region (3'UTR), predominantly in a consensus motif, GGm 6 ACU [24][25][26]. Recently, m 6 A was also found in the 5' untranslated region (5'UTR), a region that closely correlates with translation. It has been reported that methylated adenosine in the 5'UTR of mRNA can support cap-independent translation commencement and can increase translation [27,28].

Regulatory role of m 6 A RNA modification in molecular functions
Accumulation of pre-mRNA and diminution of mature mRNA in cyclo-leucine-treated avian sarcoma virus-infected cells and neplanocin A (NPC)-treated SV40 RNA demonstrate that m 6 A is essential in pre-mRNA splicing [43,44]. Both cyclo-leucine and NPC are inhibitors of methylation that can be used to investigate m 6 A [45,46]. Subsequently, MTases and demethylases might be involved in regulating RNA splicing. By changing RNA structure and regulating the combination of RNA and reader proteins, HNRNPC can modulate the splicing of m 6 A-containing mRNAs [10]. More recently, by relying on the RGG region in the low-complication region of HNRNPG, a reader was reported to cooperate with modified pre-mRNA and the phosphorylated C-terminal domain of RNA polymerase II to modulate splicing [47]. Moreover, FTO is vital to mRNA splicing because it prefers to bind to introns of nascent mRNA [48]. Another splicing-related eraser is ALKBH5. Immunofluorescence analysis revealed that ALKBH5 was tightly related to splicing factors [11].
Writers, readers and erasers can all regulate mRNA export. By modulating the clock genes Per2 and Arntl, METTL3 regulates the export of mature mRNA [49]. By interacting with SRSF3 and regulating the combination of SRSF3 and NXF1 on RNA, YTHDC1 mediates the export of modified mRNA [50]. Subsequently, knockdown of ALKBH5 leads to acceleration of mRNA export, suggesting that m 6 A is essential to regulating mRNA export [11].
AU-rich element (ARE), iron-responsive element (IRE) and cytoplasmic polyadenylation element (CPE) represent functional domains and are responsible for mRNA decay in 3'UTRs [51]. Coincidentally, m 6 A accumulates in 3'UTRs. Thus, the neighbouring sites of m 6 A and Hu antigen R (HuR), which is supposed to bind ARE to increase the stability of mRNA, lead to weak HuR function and mRNA instability [52]. However, ELAV1/HuR, a potential m 6 A-binding protein, can stabilize transcripts with the cooperation of the ARE domain [53]. Subsequently, it was reported that the stability of mRNA was decreased slightly in cells lacking ALKBH5 [11].
The YTH domain family of proteins has a conserved m 6 A-binding pocket so that these proteins can tightly bind to m 6 A in a consensus sequence and directly transcribe the molecule [12,26,[34][35][36][37][38]. Specifically, YTHDF2 accelerates mRNA decay by transferring RNA from the translatable pool to processing bodies [12]. Under heat shock conditions, dysfunction of FTO in 5'UTRs, which is regulated by YTHDF2, contributes to the promotion of cap-independent translation [28]. Moreover, YTHDF1 can increase the efficiency of translation by binding m 6 A [37]. Subsequently, YTHDF3 can regulate translation by both interacting with ribosomal proteins with bound YTHDF1 and by decaying the translation-related mRNA region with bound YTHDF2 [54,55]. However, METTL3 can regulate translation flexibly because it can either recruit eIF3 to the initiation complex directly to increase translation or can inhibit translation efficiency [56,57]. The translation efficiency is increased when METTL3 is knocked out in mouse embryonic stem cells (mESCs) and embryoid bodies (EBs) [57] (Fig. 3).
Introduction of m 1 A RNA modification m 6 A has been reported to occur in DNA from a minor cluster of microorganisms and in RNA from an extensive range of organisms, and additionally, m 1 A was identified in the 1960s [58]. Rather than accumulating in mRNA, m 1 A is predominant in tRNA and rRNA, but we recently determined that it also exists in mRNA [59,60]. However, m 1 A-containing mRNA is 10 times less common than m 6 A-containing mRNA [61,62]. In tRNA and rRNA, m 1 A conserves the tertiary structure and affects translation [63,64]. In mRNA, m 1 A has been discovered in every mRNA segment, including the coding sequence (CDS), 5'UTR and 3'UTR, although it is mostly found in the highly structured 5'UTR [62]. As a result, the location of the m 1 A methylated atom determines the function and mechanism of this kind of modification. Because the distribution of m 1 A is imbalanced, the large number of m 1 A modifications on tRNA results in more tRNA m 1 A MTases than writers on mRNA. However, TRMT6/61A recognized a T-loop-like structure with a GUUCRA tRNA-like motif in mRNAs and decorated it with the m 1 A modification, TRMT61B installed m 1 A in mt-mRNA transcripts, and TRMT10C methylated the 1374 position of ND5 mt-mRNA [65,66]. All of these phenomena contribute to tRNA m 1 A MTases and can function as mRNA writers. By binding to m 1 A-bearing RNA, YTHDF1, YTHDF2, YTHDF3 and YTHDC1 act as readers [67]. Subsequently, similar to ALKBH5 functioning as an eraser for m 6 A, ALKBH1 and ALKBH3 were able to demethylate m 1 A mRNA modifications [62,68].

Regulatory role of the m 1 A RNA modification in molecular functions
It has been reported that m 1 A methylation occurs in highly structured or GC-rich regions of 5'UTRs (which Fig. 3 m 6 A RNA modification regulates physiological processes in cell. m 6 A RNA modification in mRNA plays an essential role in cellular processes, including mRNA splicing, mRNA export, mRNA stability and mRNA translation. Both readers (HNRNPC and HNRNPG) and erasers (FTO and ALKBH5) can modulate the splicing of mRNA. After splicing and combination, pre-mRNA evolves into mature mRNA. Regulated by ALKBH5, METTL3 and YTHDC1, mature mRNA is exported from the nucleus to the cytoplasm. Once exported to the cytoplasm, both ALKBH5 and ELAV1/ HuR can maintain mRNA stability. Finally, numerous enzymes contribute to the process of translation. YTHDF1, YTHDF2, YTHDF3, FTO and METTL3 together with eIF3 can regulate translation with different mechanisms individually is also the most frequent location) and may modify the predicted secondary structure, which hints at the potential of m 1 A to alter mRNA structural stability [61,62].Moreover, m 1 A methylation can not only increase translation by decreasing the binding of the releasing factor but also prevent effective translation of m 1 A-containing CDS in mt-mRNA [26,65]. Ultimately, it has been reported that the protein level is higher when a transcript carries the m 1 A modification around the initiation codon [69]. m 5 C RNA modification Introduction of the m 5 C RNA modification m 5 C is a long-standing DNA modification that is essential for gene expression and epigenetic regulation [70,71]. However, it can also be found in RNA. Although the m 5 C RNA modification can appear in both coding and non-coding regions, it has been reported to accumulate in the UTRs of mRNA and especially prefers to be located in GC-rich regions [72]. Since a number of studies have investigated the function of m 5 C in specific mRNAs, we concluded that m 5 C modifications in different locations (5'UTRs, 3'UTRs, coding regions) exert different transcriptional regulation activities [73].
It was revealed that m 5 C RNA modifications are catalysed by the NOL1/NOP2/SUN domain (NSUN) family of proteins (NSUN1, NSUN2, NSUN3, NSUN4, NSUN5, NSUN6 and NSUN7) as well as the DNA methyltransferase (DNMT) homologue DNMT2 [74][75][76]. However, among such diversified writers, only NSUN2 can install m 5 C on mRNA because rest of these proteins are writers of tRNAs and rRNAs. Subsequently, Aly/REF export factor (ALYREF), a specific mRNA m 5 C-binding protein that can read modifications, was identified as a reader of m 5 C [77]. According to liquid chromatography-tandem mass spectrometry analysis, YBX1 was defined as the other m 5 C reader that can maintain the stability of target mRNA [78]. Knowledge is limited about the protein factors responsible for removing modifications (Table 1).

Regulatory role of the m 5 C RNA modification in molecular functions
ALYREF, the reader of m 5 C, can adjust the export of transcripts by recognizing a unique RNA-binding motif [77]. Subsequently, NSUN2 adds m 5 C to both p27 mRNA at cytosine C64 in the 5'UTR and p21 mRNA in the 3'UTR [79,80]. Deleting NSUN2 in human diploid fibroblasts (HDFs) can induce the elevation of p27, and overexpressing NSUN2 results in contrasting outcomes [79]. These results suggest that the m 5 C catalysed by NSUN2 in the 5'UTRs can limit the translation of p27. However, the m 5 C modifications added by NSUN2 to the 3'UTRs of p21 mRNA coordinate with the m 6 A modifications added by METTL3/METTL14 together to enhance the expression of p21 [80]. With regard to m 5 C modification in mRNA coding regions, it was revealed that in both bacterial whole-cell extracts and HeLa cell extracts, m 5 C could diminish translation significantly [27,81]. Moreover, we demonstrated that when the m 5 C modification was present on interleukin-17A (IL-17A) mRNA, this modification could promote the translation of IL-17A [82]. The results of the above investigations revealed that the m 5 C RNA modification affects the expression of proteins by regulating both translation efficiency and transcript export ( Table 2).
Other RNA modifications hm 5 C m 5 C can be oxidized into hm 5 C via the function of the Tet-family enzymes [91][92][93]. Moreover, hMeRIP-seq showed that Tet-family enzymes prefer to oxidize m 5 C modifications in coding regions; these results indicate that hm 5 C is highly likely to be located in the introns and exons of coding transcripts. However, in contrast to m 5 C methylation in the coding regions of mRNA, which plays a negative role in translation, hm 5 C tends to associate with translation activation in Drosophila [69].

Ψ
As hm 5 C is analogous to the oxidization of m 5 C, Ψ is produced by the isomerization of U. Ψ is the most abundant RNA modification and prefers to accumulate in tRNA and rRNA; however, it has also been reported to be present on mRNA and snRNA [94,95]. Interestingly, the number of Ψ sites in mRNA ranges from 96 to 2084 in humans [84,[96][97][98].
However, by regulating U2 auxiliary factor (U2AF), Ψ, which is near the 3' splice site in the polypyrimidine tract, prevents pre-mRNA splicing [83]. Expression of heat shock-induced Pus7-dependent pseudouridylated  FTO FTO prefers to bind to introns of nascent mRNA [48] ALKBH5 ALKBH5 relates to splicing factors tightly according to the analysis of immunofluorescence [11] mRNA export METTL3 METTL3 regulates the export of mature mRNA by modulating clock genes Per2 and Arntl [49] YTHDC1 YTHDC1 mediates the export of decorated mRNA by interacting with SRSF3 and regulating the combination of SRSF3 an NXF1 on RNA [50] ALKBH5 Knockdown of ALKBH5 leads to acceleration in mRNA export [11] mRNA stability ALKBH5 The stability of mRNA was decreased slightly in RNA lacking ALKBH5 [11] N.A. Neighbouring sites of m 6 A and HuR weaken the function of HuR and increase the instability of mRNA [52] N.A. ELAV1/HuR, which is one of m 6 A-binding proteins and stabilizes transcripts with the cooperation of the ARE domain [53] mRNA translation YTHDF2 YTHDF2 regulates translation by transferring the bound RNA from the translatable pool to processing bodies to promote mRNA decay [12] YTHDF2 induces the dysfunction of FTO in the 5'UTRs and contribute to promoting capindependent translation [ When a separate Ψ modifies the special position of codon "UUU", mRNA translation can be limited [81] transcripts is higher in wild-type yeast than in Pus7knockdown yeast and indicates that Ψ has the capability to maintain RNA stability [84]. Nevertheless, modifications were examined at similar sequences, and compared to U-containing mRNA, Ψ-containing mRNA experienced an increase in translation by approximately 25% [84]. Such modifications could double the expression of translation when compared to blank control transcript without any modification [85]. Although Ψ can promote translation and enhance the lifespan of RNA, it has negative effects on protein expression [85]. It has been reported that Ψ-containing mRNA exhibits a 30% decrease in protein expression. Specifically, bacterial mRNA translation can be limited when a separate Ψ modification is present at a given position of codon "UUU", especially at the third codon position [81]. Moreover, both in vitro and in vivo, the Ψ modification might change the nonsense codons into sense codons [99,100]. Above all, some of these investigations were conducted by Ψ in artificial mRNA, and the function of Ψ in biological mRNA has yet to be elucidated.

I and U
Catalysed by adenosine or cytidine deaminating enzymes, RNA editing is a kind of programmed alteration [101]. However, rather than permanent DNA mutations or reversible RNA modifications, RNA editing has its own limited lifespan and results in more permanent modification [102]. Adenosine-to-inosine RNA editing (A-to-I editing), also called I, is catalysed by adenosine deaminases acting on RNA (ADARs) [101,103,104]. Recently, 1741 I sites have been reported in CD regions of transcripts from RNA-seq data of different human tissues [105]. Moreover, it has been reported that ADAR1 and ADAR2 act only on double-stranded regions, which limits the areas of mRNA that I can modify [106]. I can fasten pairs of nucleotides; thus, this widespread modification in metazoan mRNA can influence the native secondary structure of mRNA [86]. An in vitro translation system was implemented to scientifically test the decoding of I, revealing that guanosine, adenosine and uracil are the products decoded from I by translation machinery [87].
However, with regard to cytidine-to-uridine RNA editing (C-to-U editing), also called U, it has been reported that U accumulates in 3'UTRs, and over 70 new sites have been discovered by transcriptome-wide research [88,107]. Subsequently, after exploring several intestinal mRNAs, it was revealed that the protein level is altered by C-to-U editing of RNA [88]. However, there is little research on the relationship between the expression of transcripts and U. The biological influence of U has yet to be investigated.

2'-O-Me
Unlike how I and U are modifications on a base, 2'-O-Me is methylation of ribose at the 2' position [59]. It was revealed that by escaping the suppression mediated by IFN-induced proteins with tetratricopeptide repeats (IFIT), 2'-O-Me-modifiedviral RNA disrupts native host antiviral responses [89]. Surprisingly, 2'-O-Me focuses on modifying specific regions of mRNA where the encoded amino acids are immobilized; these amino acids include glutamate, lysine and glutamine [90]. This phenomenon hints at the hypothesis that 2'-O-Me has the potential to affect translation efficiency, which has previously been demonstrated in bacterial mRNA [81].

Regulatory roles of RNA modifications in pathogenesis Aberrant m 6 A RNA modifications in diseases
In acute myeloid leukaemia (AML), FTO decreases m 6 A abundance on ASB2 and RARA mRNA in several certain subtypes of AML, including t(11q23)/MLL rearrangements, t(15;17)/PML-RARA, FLT3-ITD, and/or NPM1 mutations [41,108]. Moreover, by constraining YTHDF2-mediated decay, FTO decreases m 6 A frequency on MYC mRNA [109], METTL3 promotes translation of BCL2 and PTEN mRNA by upregulating the m 6 A levels and supports expression of SP1 by binding to the unique region with the help of the  [90] transcription factor CEBPZ [110,111], and METTL14 enhances mRNA expression of MYB and MYC [112]. All pathological pathways contribute to carcinogenesis in AML. According to the datasets from The Cancer Genome Atlas, nearly 10.5% of AML patients carry copy number variations (CNVs) of ALKBH5, which predicts poor prognosis and p53 mutations [113].
In gastric cancer (GC), METTL3 can cause m 6 A to accumulate on HDGF mRNA, which indicates proliferation and poor prognosis and enhances the stability of zinc finger MYM-type containing 1 (ZMYM1) mRNA so that it accelerates epithelial-mesenchymal transition (EMT) and metastasis [114,115]. However, METTL3 can also reduce m 6 A on SEC62 with the help of MiR-4429 [116]. In hepatocellular carcinoma (HCC), METTL3 enhances the degradation of m 6 A-containing SOCS2 mRNA together with YTHDF2 [117]. Additionally, YTHDF2 supresses ERK/MAPK signalling cascades and cell proliferation by destabilizing the EGFR mRNA [118]. Regarding clinical diagnosis, downregulated METTL14 is detected in HCC patients, and the level of expression in metastatic HCC is further decreased [119]. In pancreatic cancer, m 6 A and METTL3 protein and mRNA levels were much higher in tumour specimens than in para-cancerous specimens [120]. Meanwhile, upregulation of YTHDF2 destabilizes YAP mRNA by initiating the AKT/GSK3β/cyclin D1 pathway, which promotes proliferation and inhibits the migration of pancreatic cancer [121].
In lung cancer, METTL3 enhances the translation of EGFR and TAZ mRNA [56]. Furthermore, SUMOylated METTL3 promotes non-small-cell lung cancer (NSCLC) by diminishing the amount of m 6 A [122]. Moreover, YTHDF2 enhances the translation of 6-phosphogluconate dehydrogenase (6PGD) mRNA by binding to a given region in lung cancer cells [123]. Additionally, FTO is overexpressed in human NSCLC tissues and stimulates lung cancer by stabilizing and increasing the expression of ubiquitin-specific protease 7 (USP7) [124]. In lung squamous cell carcinoma (LUSC), overexpressed FTO accelerates oncogene MZF1 expression by diminishing m 6 A and stabilizing mRNA as well [125,126]. For the nervous system, decreased levels of METTL3 or METTL14 determine the diminution of m 6 A on ADAM19 mRNA, which promotes protein expression [127,128]. Conversely, increased levels of ALKBH5 lead to decreased levels of m 6 A on FOXM1 mRNA and enhance protein expression [129]. Consequently, a high level of ALKBH5 predicts poor prognosis [130]. However, both pathways can contribute to glioblastoma. Subsequently, overexpressed METTL3 recruits HuR to modified SOX2 mRNA and enhances radio-resistance. Playing an oncogenic role in glioblastoma, METTL3 hints at poor prognosis and a potential therapeutic strategy as well [131].
In prostate cancer, reduced YTHDF2 elevates m 6 A contents dramatically, which suppresses proliferation and migration [132]. In bladder cancer, increased METTL3 predicts poor survival because with the help of pri-miR221/222, upregulated METTL3 results in downregulated PTEN and tumorigenesis of cancer [133].
Aberrant m 6 A modification can also lead to carcinomas in the reproductive system. It has been reported that m 6 A on KLF4 and NANOG can be suppressed by the cooperation of ZNF217 and ALKBH5, especially in a HIF-dependent manner, so that it enhances the stability of mRNA and contributes to breast cancer in a hypoxic microenvironment [134,135]. Increased METTL3 leads to enhancement of m 6 A on hepatitis B X-interacting protein (HBXIP) and proliferation of breast cancer stem cells (BCSCs) [136]. Moreover, elevated FTO leads to downregulated methylation and degradation of BNIP3. It is suggested that FTO enhances the colony formation and metastasis of breast cancer [137]; Nevertheless, in cervical squamous cell carcinoma (CSCC), high expression of FTO and low levels of β-catenin lead to chemoradiotherapy resistance, which hints that FTO is a potential target to increase the chemoradiotherapy sensitivity of CSCC [138]. In endometrial cancer, either mutated METTL14 or reduced METTL3 limits the expression of m 6 A. However, limited m 6 A activates the AKT signalling pathway and stimulates proliferation and tumorigenicity by decreasing the negative AKT regulator PHLPP2 and increasing the positive AKT regulator mTORC2 [139].
Besides the regular cancers with high incidence referenced above, aberrant m 6 A modifications also play roles in sensory organs. The fate of ocular melanoma can be modulated by m 6 A modifications. With the help of YTHDF1, the translation of methylated HINT2 mRNA, a tumour suppressor of ocular melanoma, was significantly accelerated, meaning m 6 A modification obviously inhibits the progression of ocular melanoma. Moreover, investigation of ocular melanoma samples indicated that decreased m 6 A levels were highly associated with poor prognosis [140].

Aberrant m 1 A RNA modification in diseases
Physiological functions lead to pathological impacts on diverse diseases. In ovarian and breast cancers, demethylation of m 1 A by ALKBH3 induces increased modified CSF-1 mRNA, which contains m 1 A in the 5'UTR near the translation initiation site. Hence, accumulated ALKBH3 means improved CSF-1 mRNA expression and invasion of cancer cells [141]. Subsequently, ALKBH3, considered the eraser of m 1 A, tightly correlates with the mTOR pathway in gastrointestinal cancer and is attributed to the limited expression of ErbB2 and AKT1S1 after ALKBH3 knockdown; the downstream genes of m 1 A are associated with cell proliferation according to Gene Ontology analysis [142]. Additionally, silencing of ALKBH3 arrests the cell cycle at the G1 phase and contributes to the progression, angiogenesis and invasion of urothelial carcinomas by modulating NADPH oxidase-2reactive oxygen species (NOX-2-ROX) and TNF-like weak inducer of apoptosis (TWEAK)/Fibroblast growth factor-inducible 14 (Fn14)-VEGF signals [143]. As a classical chemical modification of mRNA, the pathological pathways of m 1 A need to be elucidated.

Aberrant m 5 C RNA modification in diseases
Since m 5 C bridges transcription and translation, we propose a hypothesis that m 5 C can also regulate the pathological mechanisms of various diseases. For instance, diminishing NSUN2 leads to decreased levels of translation and an increased tumour initiating population in skin cancer [144]. In breast cancer, NSUN2 is reported to be upregulated as well at the mRNA and protein levels [145]. For patients with urothelial carcinoma of the bladder (UCB), m 5 C-modified 3'UTR in HDGF mRNA can be recognized by YBX1 and activate the oncogene of UCB [78]. m 5 C can also be regarded as a cancer biomarker because the amount of m 5 C RNA modification is increased in circulating tumour cells from patients with lung cancer [146].

Aberrant hm 5 C, Ψ, I, U and 2'-O-Me RNA modifications in diseases
Although the amounts of hm 5 C, Ψ, I, U and 2'-O-Me RNA modifications on mRNA are much lower than the three predominant types of modifications, their roles do not change and are vital to human disease. First, Ψ can function as a biomarker for prostate cancer because certain nucleolar RNAs (H/ACA snoRNAs) and the dyskerin (DKC1) protein can upregulate the transformation of U to Ψ and contribute to the advancement to cancer [147]. Regarded as the gene encoding the Ψ synthase, the mutation of DKC1 causes downregulated Ψ and Xlinked dyskeratosis congenita (X-DC) [148]. The risk for cancer development is higher in patients with X-DC than those without gene mutation [149]. Besides, H/ ACA snoRNAs are limited in acute leukaemia, lymphoma and multiple myeloma [150][151][152].
Subsequently, edited AZIN1 stimulates a serine to glycine (S/G) conversion in HCC and leads to proliferation and poor prognosis [153,154]. In HCC and in cervical cancer, increased editing of BLCAP activates the AKT/ mTOR signalling pathway or STAT3, which can increase cell proliferation and limit apoptosis [155][156][157][158]. In breast cancer, editing of DHFR transcripts at the 3'UTR by ADAR1 stabilizes the mRNA and enhances cell growth. Surprisingly, methotrexate, a chemotherapy agent, prevents cancer cell division by targeting DHFR.
It is suggests that downregulated ADAR1 can contribute to methotrexate treatment [159]. In gastric cancer, ADAR2 edits the CDS of PODXL, which induces a histidine to arginine conversion. The relationship between reduced ADAR2 and increased malignancy hints that transcript editing is essential to prevent cancer progression [160]. Additionally, adenosine deaminase RNAspecific B1 (ADARB1), a special type of ADAR, is expressed at low levels in H358 and A549 lung adenocarcinoma (LUAD) cells, which suggests that I might be a potential target in diagnostic and prognostic progression for patients with LUAD [161].
Finally, uridine phosphorylase 1 (UPP1) is another enzyme that can reversibly catalyse the phosphorolysis of uridine to uracil [162,163]. It has been reported that expression of UPP1 significantly depends on lymph node metastasis and tumour stage and size in patients with thyroid carcinoma [164] (Table 3, Fig. 4).

Clinical prospects of RNA modifications
RNA modifications and enzyme complexes exhibit upregulated and downregulated levels of expression in cancers, which means RNA modifications can serve as biomarkers to diagnose diseases in a manner that is helpful and precise. For example, upregulated YTHDF2 is found in pancreatic cancer, increased m 5 C is detected in lung cancer and accumulated Ψ contributes to the advancement of prostate cancer [121,146,147]. However, other biomarkers need to be elucidated. Besides biomarkers to diagnose cancers, RNA modifications are also biomarkers to predict patient prognosis. Since they stimulate or inhibit the progression of cancer, RNA modifications have therapeutic potential. 3deazaadenosine (DAA) interrupts METTL3/14 and inhibits the decoration of m 6 A by obstructing SAH hydrolase [165], SPI1 is considered a potential target for AML because of inhibition of METTL14 [112], and meclofenamic acid (MA), a non-steroidal anti-inflammatory drug, silences FTO by competing for binding sites [166]. Novel targets for treatment of cancer require further investigation.

Conclusion
In summary, chemical modifications in mRNA are vital for many processes of cell life, such as pre-mRNA splicing, nuclear export, transcript stability and translation initiation. Importantly, RNA modifications play a critical role in driving cell fate in cancer. The importance of the relationship between RNA modification and various diseases cannot be overly emphasized. In this review, we redefined the bridge between transcription and translation and applied it to physiological and pathological processes. To date, we have demonstrated 2 roles of mRNA modifications in transcription. Generally, one type is A levels [110] SP1 METTL3 supports the expression of SP1 by binding to the unique region with the help of the transcription factor CEBPZ [111] METTL 14 MYB/ MYC METTL14 enhances the expression of MYB and MYC mRNA in AML [112] ALKBH5 N.A. Approximately 10.5% of AML patients carry CNVs of ALKBH5, which predicts poor prognosis and p53 mutations [113] Gastric cancer METTL3 HDGF METTL3 causes m 6 A to accumulate on HDGF mRNA, which indicates proliferation and poor prognosis of gastric cancer [114] ZMYM1 METTL3 enhances the stability of ZMYM1 mRNA to accelerate EMT and metastasis [115] SEC62 METTL3 reduces m 6 A on SEC62 with the help with MiR-4429 [116] Hepatic carcinoma METTL3 SOCS2 METTL3 works with YTHDF2 together to enhance the degradation of SOCS2 m 6 A-containing mRNA, which leads to HCC [117] YTHDF2 EGFR YTHDF2 suppresses ERK/MAPK signalling cascades and cell proliferation via destabilizing the EGFR mRNA [118] METTL14 N.A. The expression of METTL14 is decreased in HCC, especially in metastatic HCC [119] Pancreatic cancer METTL3 N.A. METTL3 protein, m 6 A abundance and mRNA levels are much higher in tumour specimens than in para-cancerous specimens [120] YTHDF2 YAP Increased YTHDF2 promotes proliferation and suppresses migration of pancreatic cancer by destabilizing YAP mRNA [121] Lung cancer METTL3 EGFR/ TAZ METTL3 enhances the translation of EGFR and TAZ mRNA in lung cancer [56] Fig. 4 Regulatory roles of RNA modifications in pathogenesis. Applying physiology to pathology, RNA modifications redefine the bridge between transcription and translation and regulate disease pathogenesis. In AML, METTL3 and METTL14 enhance the expression of m 6 A modifications as well as the BCL2, PTEN, SP1, MYB and MYC genes, which lead to tumour progression. Simultaneously, FTO decreases m 6 A abundance on ASB2 and RARA mRNA.
In digestive system tumours, aberrant METTL3 leads to aberrant expression of HDGF, ZMYM1, SEC62 and SOCS2, which can regulate cancer cells in the stomach, liver and pancreas, respectively. In lung cancer, METTL3 enhances the translation of EGFR and TAZ, whereas SUMOylated METTL3 promotes NSCLC; aberrant YTHDF2 enhances the expression of 6PGD in lung cancer, and overexpressed FTO stabilizes and accelerates the expression of USF7 and MZF1 as well. In glioblastoma, METTL3, METTL14 and ALKBH5 promote the expression of ADAM19 and FOXM1 and predict poor prognosis. In prostate cancer, aberrant YTHDF2 suppresses proliferation and migration. In bladder cancer, METTL3 reduces the expression of PTEN and tumorigenesis of cancer. In the reproductive system, METTL3 and FTO contribute to the aberrant expression of KLF4, NANOG, HBXIP, BNIP3 and β-catenin, which induce proliferation of breast cancer and chemoradiotherapy resistance of cervical cancer separately. In sensory organs, YTHDF1 accelerates the translation of methylated HINT2 and inhibits the progression of ocular melanoma. Aberrant eraser ALKBH3 reduces m 1 A modifications, leads to aberrant expression of CSF-1, ErbB2 and AKT1S1, and induces the progression of ovarian cancer, breast cancer, gastrointestinal cancer and urothelial cancer. In UCB, YBX1 recognizes m 5 C-modified HDGF mRNA and leads to tumour advancement. Upregulated USUN2 is detected in breast cancer. Ultimately, aberrant ADAR1 edits AZIN1, BLCAP, and DHFR separately, which leads to hepatocellular carcinoma, cervical cancer and breast cancer. Additionally, together with Ψ, I and U, DKC1, ADAR1 and UPP1 can function as biomarkers to indicate prostate cancer progression, LUAD presentation and thyroid carcinoma status