key: cord-263033-4790dhc5
authors: Laptev, I. G.; Golovina, A. Ya.; Sergiev, P. V.; Dontsova, O. A.
title: Posttranscriptional modification of messenger RNAs in eukaryotes
date: 2015-12-11
journal: Mol Biol
DOI: 10.1134/s002689331506014x
sha: 
doc_id: 263033
cord_uid: 4790dhc5

Transcriptome-wide mapping of posttranscriptional modifications in eukaryotic RNA revealed tens of thousands of modification sites. Modified nucleotides include 6-methyladenosine, 5-methylcytidine, pseudouridine, inosine, etc. Many modification sites are conserved, and many are regulated. The function is known for a minor subset of modified nucleotides, while the role of their majority is still obscure. In view of the global character of mRNA modification, RNA epigenetics arose as a new field of molecular biology. The review considers posttranscriptional modification of eukaryotic mRNA, focusing on the major modified nucleotides, the role they play in the cell, the methods to detect them, and the enzymes responsible for modification.

Posttranscriptional modifications of RNA were found rather long ago, but their biological functions remain obscure apart from few exceptions. Cap modi fications are the best understood in eukaryotic mRNA. Regions distant from the mRNA ends may contain N6 methyladenosine (m 6 A), 5 methylcytidine (m 5 C), pseudouridine (Ψ), and inosine (I), which were believed to play only a minor role because their pro portion in cell RNA is extremely low as compared with the standard nucleotides. As analytical methods devel oped, modified nucleotides proved to occur in certain positions, rather than being spread at random, and their occurrence in such positions sometimes reaches 100% [1] [2] [3] [4] [5] [6] . The review focuses on these four modifi cations, their occurrence, detection methods, and, when possible, the roles they play in the cell.

Higher eukaryotic mRNAs and many virus RNAs undergo 2' O methylation at one or two first nucle otides (methylated nucleotides are collectively desig nated N m ) [7, 8] . The most common are m 6 A m and other A m , accounting for approximately 70% of all nucleotides methylated at the ribose moiety. The G m proportion is 18%, and C m and U m together account for 12% [9] . Specific mRNA 2' O methyltransferases were isolated from a HeLa cell extract and character ized in 1981 [10] , but it was not until 2011 that their genes were identified and cloned [11] . N m are involved in self versus nonself RNA recognition [12] . Human and mouse coronaviruses mutated to lack 2' O meth yltransferases induce high level production of type I interferon via the MDA5 cytoplasmic protein, which is sensitive to dsRNA [13] .

N6 methyladenosine (m 6 A) is the most common mRNA modification. The first m 6 A detection in mRNA dates back to the 1970s [14] . Because mRNA accounts for only a minor proportion of total cell RNA, m 6 A detection in mRNA is rather problematic. By 2005, specific adenosine methylation sites were found only in two RNAs, the bovine prolactin (bPRL) mRNA [15, 16] and Rous sarcoma virus (RSV) RNA [17] . In the bPRL mRNA, methylation sites cluster in the 3' untranslated region (3' UTR) and in the vicin ity of the polyadenylation site. The m 6 A occurrence in the methylation sites is only ~20%. More than ten methylation sites were identified in the RSV RNA. Like with the bPRL mRNA, methylation of the RSV RNA is incomplete, varying from 20 to 90%. A muta tion analysis of the adenosine methylation sites in vitro and in vivo established the specific site sequence, RRm 6 ACH (where R is adenosine or guanosine and H is adenosine, uridine, or cytidine) [18] . Statistically, the sequence can be found in every 85 nt, so that approximately 30 adenosine methylation sites may occur in mRNA on average. Because m 6 A was not detected in all of the potential sites in the bPRL mRNA and RSV RNA, their adenosine methylation was not assumed to proceed quantitatively.

It is rather difficult to detect m 6 A in RNA because methylation does not affect the Watson-Crick base pairing (Fig. 1a) , and reverse transcriptase does not distinguish between modified and unmodified nucle otides. In 2012, m 6 A seq [1] and MeRIP seq [2] were developed to detect m 6 A in RNA with a high sensitiv ity. The gist of the method is as follows (Fig. 1b) . Poly adenylated RNAs are chemically cleaved into frag ments of approximately 50-100 nt. Fragmented RNA is divided into two portions. One is used for immuno precipitation with anti m 6 A antibodies to obtain a modification enriched fraction, and the other serves as a reference. Both of the portions are subject to deep sequencing. The m 6 A positions are identified by com paring the sequencing results for the two portions. Many RNAs proved to contain m 6 A in the well known context [18] . Generally, m 6 A is found mostly around the stop codon of the mRNA coding region. The sig nificance of this arrangement remains unclear. The m 6 A distribution differs among different tissues and in tissues exposed to stress. A virtually unique adenosine methylation profile is therefore characteristic of each cell type.

The method reports the methylated adenosine positions with a resolution of approximately 100 nt [1, 2] . Sequencing with greater overlaps and a more stringent bioinformatics data processing were used to identify the adenosine methylation sites in the total yeast transcriptome at a better resolution [19] . RNA was fragmented into shorter segments, and a more rig orous approach was used to eliminate false positive results. RNA from cells with the inactivated gene for methyltransferase responsible for adenosine modifica tion (which is impossible with mammalian cells because a METTL3 knockdown leads to apoptosis) and several RNAs synthesized in vitro were used as negative controls. About half of the identified m 6 A sites were considered to be false positives with these negative controls, indicating that the results reported in [1, 2] most likely need verification.

Recent single molecule real time sequencing showed that HIV reverse transcriptase is m 6 A sensitive [20] . Thermus thermophilus DNA polymerase I was found to act as reverse transcriptase in the presence of Mn 2+ , being sensitive to the m 6 A presence in these conditions [21] . The enzymes might be suitable for sequencing the transcriptome with the identification of m 6 A sites.

A method to verify the m 6 A presence in a particular RNA site was developed in our lab [22] . The method is based on analyzing the melting curves for DNA-RNA duplexes. To detect the modification in a given RNA site, two primers with end to end annealing are selected (Fig. 1c) . One primer contains a fluorescein (FAM) at the 5' end, and the other contains the black hole quencher BHQ1 at the 3' end. The BHQ1 con taining primer is designed to hybridize with a m 6 A con [1, 2] . (c) Primer annealing to check for the m 6 A presence [22] .

No. 6 2015 taining RNA region. A comparison of the differential melting curves for the control modification free duplex and a test sample reports whether m 6 A occurs in the given RNA site. The method can be used, for example, to identify the genes for methyltransferase that modifies a certain nucleotide. In the case of eukaryotic mRNAs, the method is suitable for probing the adenosine methylation status in a particular site of a particular RNA in various cell growth conditions. A method known as site specific cleavage and radioactive labeling followed by ligation assisted extrac tion and thin layer chromatography (SCARLET) [23] makes it possible to establish whether adenosine is methylated in a given position of a given molecule and to estimate the proportion of modified and unmodi fied nucleotides (Fig. 2) . In this method, a specific chimeric 2' O Me/2' H oligonucleotide is con structed and hybridized to the polyadenylated RNA fraction. The target RNA site is cleaved with RNase H to produce two RNA fragments so that the target ade nosine is at the 5' end of one of the fragments. The remaining RNA is phosphorylated to add [ 32 P] to the 5' ends of RNA fragments, and the fragment of interest is ligated to a long single stranded oligodeox yribonucleotide. The mixture is digested with RNases A and T1, which together cleave ssRNA after C, U, and G. The oligodeoxyribonucleotide with 32 P A/m 6 A, which remains intact, is purified by denaturing electrophore sis and digested with nuclease P1, which cleaves ssDNA and RNA to 5' monophosphates. The result ing sample is assayed for [ 32 P]A and [ 32 P] m 6 A by thin layer chromatography. The method is rather laborious and requires a radioactive label to be used.

Several methods were developed to detect m 6 A in RNA at various resolutions. Each method has its drawbacks and advantages. High throughput methods 

The protein product of METTL3 was the first to be identified as eukaryotic (adenine N6 ) methyltrans ferase [24] and is conserved among many organisms from yeast (IME4) to mammals. A knockdown or deletion of its gene exert various phenotypic effects, causing apoptosis in human cell lines, a lower survival in plants and Drosophila melanogaster, and sporulation defects in yeasts [25] .

Recent studies identified METTL14 as another protein that catalyzes adenosine methylation in RNA and forms a heterodimeric complex with METTL3 [26] . The two proteins belong to one methyltransferase superfamily and have 43% amino acid sequence simi larity. Both of them are catalytically active and methy late oligonucleotides in vitro. The findings indicate that the two proteins act as catalytic subunits of the complex.

The METTL3-METTL14 complex was shown to interact with WTAP (Wilms' tumor 1 associated pro tein) [26, 27] , which is involved in splicing [28] . A WTAP gene knockdown considerably reduces the m 6 A content, although WTAP does not display methyl transferase activity in vitro [26] . It seems that WTAP acts as a regulatory protein to facilitate methyltrans ferase activity or nuclear localization of methyltrans ferases [26, 27] .

In addition to RNA methylation, demethylation occurs as an opposite process. It is clear that the role the process plays in the cell is probably no less important. FTO (fat mass and obesity associated protein) belongs to a family of proteins homologous to Fe(II)/α ketogl utarage dependent dioxygenase AlkB. FTO was shown to catalyze oxidative demethylation of m 3 T and m 3 U in ssDNA and ssRNA [29, 30] , although its activity is lower than in AlkB family proteins [31] . Its Arg316 is essential for α ketoglutarate binding, and the R316Q substitution abolishes FTO catalytic activ ity in vitro [32] . Severe growth retardation is observed in organisms carrying this mutation. The FTO prefer ence for ssDNA and ssRNA is possibly explained by the presence of an additional loop, which covers a side of a conserved jelly roll motif and competes with the complementary stand of the duplex [33] . Surprisingly, the capability of m 6 A demethylation was observed for FTO [31] .

Physiological substrates of AlkB family proteins are not limited to N1 or N3 modified purines and pyrimidines, attention was consequently attracted to m 6 A as the most common mRNA modification [31] . FTO proved to convert m 6 A to adenosine in synthetic ssDNA and ssRNA with an efficiency comparable with that of m 3 U demethylation [31] . The m 6 A modi fication was not detected in higher eukaryotic DNA [34] , indicating that RNA acts as a physiological sub strate of FTO. To check this assumption, FTO was silenced via RNA interference or overexpressed in HeLa and HEK293FT cells. The results confirmed that m 6 A is a physiological substrate of FTO in these cell lines. FTO was observed to occur exclusively in the nucleoplasm together with the splicing factors SART1 (U4/U6.U5tri snRNP associated protein) and SC35 (serine/arginine rich splicing factor 2), implicating FTO in mRNA maturation [31] .

FTO demethylates m 6 A via the formation of two intermediates, N6 hydroxymethyladenosine (hm 6 A) and formyladenosine (f 6 A) (Fig. 3 ) [35] . The two intermediates are detectable in vitro, indicating that their dissociation and rebinding are involved in deme thylation.

FTO binds m 6 A and hm 6 A with comparable affin ity, and the generation rate of hm 6 A is higher than that of f 6 A. The intermediates (hm 6 A and f 6 A) are found in RNA in vivo. Their content was estimated at 0.5-1% of the total m 6 A content, but the intermediates might degrade during RNA isolation and fragmentation.

In 2013, ALKBH5 (alkylation repair homolog 5) was identified as another m 6 A RNA demethylase of the AlkB family [36] . Like FTO, ALKBH5 is localized in the nucleus together with the splicing factors CS35, Sm, and ASF/SF2 (alternative splicing factor/splicing factor 2). It is most likely that ALKBH5 directly inter acts with RNA because a granular pattern of ALKBH5 distribution in the nucleus is almost completely elimi nated by treating preparations with RNase A. An ALKBH5 knockdown increases the mRNA content in the cytoplasm, implicating the demethylase in mRNA export [36] .

Several proteins were found to selectively bind with synthetic oligoribonucleotides that mimic the adenos ine methylation site of the RSV RNA and contain m 6 as bait [1] , ELAVL1, YTHDF2, and YTHDF3 being the best binders.

ELAVL1, which is also known as HuR (human antigen R), belongs to the ELVAL family of RNA binding proteins and selectively binds AU rich regions in the 3' UTR of mRNA [37] . ELAVL1 stabilizes mRNAs that contain AU rich elements [38, 39] . Pub lished data on the functional relationship of ELAVL1 with mRNA methylation are discrepant. After ELAVL1 was initially reported to selectively bind m 6 A containing RNA [1] , a more efficient binding was demonstrated for total RNA from cells lacking m 6 A in mRNA [40] .

The two other proteins, YTHDF2 and YTHDF3, belong to the YTH domain superfamily of RNA bind ing proteins [41] . The YTH domain is conserved among eukaryotes and is widespread in plants. YTHDF2 recognizes m 6 A both in vitro and in vivo [42] . YTHDF2 binding with m 6 A containing oligori bonucleotides is ~15 times more efficient than with nonmodified oligonucleotides in vitro. YTHDF3 and YTHDF1 similarly bind the modified nucleotide in vitro with a 5 to 20 fold higher efficiency. More than 3000 YTHDF2 targets were found in human cell lines, and the majority of the targets occur in mRNA. YTHDF2 competes with ribosomes for mRNA bind ing in the cytoplasm. YTHDF2 bound mRNA is committed to degradation. Thus, YTHDF2 acts as a sorter of m 6 A containing mRNAs. When free ribo somes are available, mRNA binds with them and is translated; otherwise, YTHDF2 binds with mRNA and relocates it to a degradation site.

Predisposition to certain disorders was associated with polymorphic variants of the genes whose prod ucts are involved in mRNA modification (table) .

The effect of mRNA methylation and m 6 A produc tion on stem cell proliferation and differentiation has been discussed intensely in the past years. Published data are discrepant. Mettl3 and Mettl14 knockdowns reduce the m 6 A content in embryonic stem cells, thus decreasing their proliferative activity and inducing a loss of pluripotency markers; i.e., the effect is promot ing stem cell differentiation [40] . It is possible that methylation of the mRNAs coding for differentiation regulators facilitates their degradation in control stem cells. In contrast, a more recent study showed that complete Mettl3 silencing in mouse stem cells increased their self renewal potential and blocked their differentiation into cardiomyocytes and neurons [43] . Similar results were obtained when Mettl3 was inactivated via a Mettl3 knockdown rather than RNA interference [44] . Mettl3 knockout stem cells were via ble and proliferated normally, but their differentiation was distorted. When differentiation was induced, the cells still expressed genes characteristic of pluripotent stem cells. The effect was associated with stabilization of the mRNAs coding for pluripotency markers. Thus, it is commonly accepted that methylation destabilizes mRNA, but it remains unclear what mRNAs are more prone to methylation and destabilization in early development.

In the 1970s, m 5 C was found in mRNAs isolated from BHK 21 hamster cells [45] and certain virus RNAs [46, 47] , but not in HeLa cell mRNA [48] and SV40 RNA [14] . Fortunately, m 5 C proved far easier to map in RNA as compared with m 6 A. Bisulfite treat Disorders associated with m 6 A related genes [25] Gene

Function of the protein product Disorder (Cytosine C5 ) Methyltransferases Several RNA (cytosine C5 ) methyltransferases were found in eukaryotic cells: the NSUN protein fam ily (NSUN1-6 and NSUN2 homologs), DNMT2, and DNMT2 homologs. Methylating activity towards mRNA was observed only for NSUN2 as yet [3, 49] .

The human NSUN2 family includes nine proteins, most of which are highly conserved among mammals [3, 50] and possess a methyltransferase domain. Activ ity of the domain was studied only in NSUN2 [51] . Mouse NSUN2 is a component of chromatoid body and is necessary for testicular differentiation [49, 52] and a balance between self renewal and differentiation of skin stem cells [53] . NSUN2 is involved in tRNA methylation and modifies mRNA and noncoding RNAs as well [3, 49, 52, 54] .

Studies of m 5 C were initially limited to tRNA and rRNA. As for mRNA, the modification was poorly understood until bisulfite sequencing was applied to total cell RNA [3] and m 5 C was identified as a modifi cation common in mRNAs and noncoding RNAs of various, including human, cells. The method was based on conventional DNA bisulfite sequencing [55, 56] and revealed 10274 m 5 C sites in mRNAs and noncoding RNAs with m 5 C accounting for 0.4% of all cytidine residues. Noncoding RNAs had even a higher m 5 C proportion, 1.2%. Methylated cytosine occurred mostly in untranslated regions and in the vicinity of binding sites for Argonaute family proteins. It is pos sible that m 5 C plays a role in the miRNA mediated RNA degradation pathway [3] .

Two new methods were recently developed to iden tify the cytidine methylation targets in RNA, taking advantage of the methylation mechanism [57] . Many RNA methyltransferases are known to possess two highly conserved cysteine residues, which are essential for catalysis. One forms a covalent intermediate with the target cytidine, and the other is necessary for the covalent intermediate to be resolved after methylation [57] . The Aza IP method is based on a covalent bond ing of a cytosine analog (5 azacytosine) incorporated in nascent RNA with m 5 C methyltransferase (Fig. 4a) . 5 Azacytidine is not methylated, and the methyltrans ferase cysteine residues remains linked to the hetero cycle. The crosslinking is followed by immunoprecip itation and high throughput sequencing [54] . With this method, many tRNAs and noncoding RNAs were identified as NSUN2 substrates. A high frequency of the C → G transition was observed for presumably methylated C residues, helping to recognize specifi cally methylated cytidines in RNA targets. It is thought that C is incorporated in place of G during replication or reverse transcription because of cycle opening [58] (Figs. 4b, 4c ).

The other method is known as miCLIP (methyla tion iCLIP) and is based on the mechanism of NSUN2 mediated methylation. Cytidine methyla tion at C5 starts with a covalent bonding of Cys321 of NSUN2 and the pyrimidine ring of cytidine. After methylation, NSUN2 Cys271 plays a role in cleaving the enzyme-RNA covalent bond. The C271A muta tion of NSUN2 stabilizes the covalent RNA-protein intermediate, and a method taking advantage of this circumstance was used in place of iCLIP to identify the NSUN2 targets at a single nucleotide resolution. With this method, NSUN2 was found to methylate tRNAs, mRNAs, and noncoding RNAs [49, 52] . The two methods will find application in studying m 5 C methyltransferases.

The functional role m 5 C plays in tRNA and rRNA was the subject of many studies. Occurring in the vari able and anticodon loops of tRNA, m 5 C stabilizes its spatial structure and the codon-anticodon duplex [60] . A double knockout in DNMT2 and NSUN2 totally eliminates m 5 C from tRNA, thus destabilizing the tRNA structure and suppressing protein synthesis in mice [61] . The m 5 C residues found in rRNA are involved in translation and tRNA recognition [62] .

As already mentioned, m 5 C is one of the most common modified nucleotides in mRNAs and non coding RNAs, but its functions in these molecules are poorly understood. Bisulfite sequencing of the HeLa cell transcriptome showed that m 5 C accumulates in untranslated regions and that mRNA cytidine methy lation sites occur in the vicinity of binding sites for Argonaute, a major component of the miRNA/RISC complex [3] . The finding implicates m 5 C in miRNA mediated RNA degradation. The hypothesis is at vari ance with the fact that mRNAs identified as NSUN2 targets by miCLIP do not change in expression in the absence of NSUN2 [49, 52] ; i.e., an effect of m 5 C on mRNA stability is still unproven. It is of interest that noncoding vault RNA (vtRNA) is methylated by NSUN2 according to miCLIP data. When its m 5 C is lost, vtRNA is abnormally processed to small vtRNAs (svRNAs), which act as miRNAs to affect expression of several genes [49, 52] . Two long noncoding RNAs, HOTAIR and XIST, were found to contain m 5 C in the functional regions responsible for interactions with a complex of chromatin associated proteins or in the vicinity of these regions. There is evidence that m 5 C is capable of disrupting certain protein-protein interac tions in vitro [63] . PSEUDOURIDINE Pseudouridine (Ψ), which is also termed the "fifth nucleotide," was the first modified nucleotide discov ered in RNA almost 60 years ago [64] . Pseudouridine is found in many cell RNAs, from tRNAs and rRNAs to various small nuclear RNAs [65] . Such a broad dis tribution in RNAs indicates that pseudouridine is important for the cell function.

Pseudouridine is an uridine isomer (5 ribosylu racil) and forms via isomerization. First, the N1-C1' glycoside bond between uracil and ribose breaks. The base thus released rotates about the N3-C6 axis and forms a new, C glycoside bond between C5 and C1' [66] . As a result, Ψ is capable of forming a hydrogen bond, which, together with the C glycoside, rather than N glycoside, bond, differentiates Ψ from all other bases. Isomerization is catalyzed by enzymes of two types. The substrate is recognized by yeast Cbf5 and mammalian DKC1/Dyskerine with the aid of small nucleolar RNAs (snoRNAs) having a small region complementary to the target RNA [67] [68] [69] [70] . Pseudou ridine synthase (PUS) family proteins directly recog nize target RNAs [71, 72] .

As snoRNAs complementary to mRNAs rather than to noncoding RNAs were discovered, uridine isomerization was assumed to occur in mRNA as well [73, 74] . A total of 23 human genes were predicted to code for proteins similar to known pseudouridine syn thases, but their functions were not verified experi mentally [75] .

The RNA bases whose modification is subject to regulation are of particular interest. Isomerization of two uridines in the yeast U2 snRNA is regulated sepa rately in a stress depenent manner, for instance, in heat shock or nutrient deficiency [76] . Isomerization of two uridines in mammalian rRNA is regulated by the kinase mTOR [77] .

A transcriptome wide search for pseudouridine was recently reported for yeast and human cells [4] [5] [6] . Isolated RNA was treated with cyclohexyl 3 (2 mor pholino 4 ethyl)carbodiimide n toluenesulfonate (CMC). The CMC-uridine bond disrupts in an alka line milieu, while modified N3 CMC Ψ remains intact. The resulting RNAs were examined by reverse transcription (Ψ CMC is known to terminate reverse transcription [78] ) followed by deep sequencing. The method, which is known as Psi seq, Pseudo seq, or Ψ seq, reported ~50-100 Ψ sites for mRNAs of yeast cells cultured in optimal conditions and ~100-400 sites in human cell lines. Noncoding RNAs were also found to contain Ψ. The number of Ψ sites mapped in differ ent studies depended on the read depth and the criteria employed in site selection in a computer analysis. It is noteworthy that Ψ sites are regularly distributed throughout coding and noncoding regions, rather than clustering in particular regions of transcripts [4, 6] .

Stress dependence of the mRNA pseudouridyla tion level is among the most interesting findings of transcriptome wide Ψ mapping. The Ψ proportion in yeast mRNA considerably increases in heat shock [5, 6] . A total of 265 new pseudouridylation sites were identified, and the majority of them proved to be mod ified in heat shock by PUS7 pseudouridine synthase, like Ψ residues in the U2 snRNA [5] . The pseudou ridylation level is approximately doubled in nutrient defi ciency [4] . The majority of nucleotides subject to regu lated isomerization are isomerized only in the presence of active PUS1 and PUS7. Apart from PUS family pro teins, CBF5 is responsible for pseudouridylation of certain uridine residues, and its activity is stress inde pendent [5] . Pseudouridine synthases were found not only in yeast, but also in human cells, and transcrip tome wide mapping of pseudouridine residues in RNA was carried out for human cells growing in nor mal conditions [4, 5] . A total of 353 pseudouridylation sites were detected in mRNA, and the majority of them were DKC1 dependent. As for mRNA, 68 sites were observed in normal conditions and 92 sites, after 24 h serum starvation [4] .

The biological role of isomerization to pseudouri dine is unclear for the majority of uridine residues. Because Ψ is capable of hydrogen bonding with A, pseudouridine containing transcripts are translated to produce a functionally active protein without changes in amino acid composition [79] . A regulatory role might be possible for uridine to pseudouridine conversion in stop codons. It was shown that a stop codon is misread as a sense codon when artificially modified with pseudouri dine [80] . However, only one endogenous transcript was found to undergo pseudouridylation of the stop codon [5] ; i.e., this function is hardly a main one for Ψ.

The total number of pseudouridine residues in mRNA is lower than that of other modified nucleotides. However, stress dependent poseudouridylation sug gests a regulatory function for this modification.

RNA editing is a type of posttranscriptional modi fication and involves adenosine deamination to inosine (I). Adenosine deamination is a major type of editing in the case of mammalian RNA, in contrast to Trypanosoma mitochondrial RNA [81] . Inosine is rec ognized as G, rather than A, during splicing and trans lation, interacting mostly with cytidine to form a com plementary pair [82] .

Adenosine deamination occurs only in double stranded RNA regions and involves proteins of the ADAR (adenosine deaminase acting on RNA) family [81, 83, 84] . Three ADAR family proteins are encoded in the human genome: two ADAR1 isoforms (ADAR1p150, or ADAR1L, and ADAR1p110, or ADAR1S) [85] , ADAR2 [86] , and ADAR3 [87] . Their structures are schematically shown in Fig. 5 . The pro teins are highly conserved among vertebrates [88] . ADAR1 and ADAR2 occur in many tissues, while ADAR3 is found exclusively in brain tissues and is thought to be catalytically inactive [81] . All of the ADAR proteins have a dsRNA binding domain at the N end and a conserved catalytic domain at the C end [85] . The proteins are catalytically active only as homodimers, as was shown both in vitro and in vivo [89] , and are capable of specific and nonspecific edit ing of both noncoding and coding dsRNAs [81] . ADAR activity defects due to mutations or changes in expression are associated with various disorders, including cancer, neurology diseases, metabolic disor ders, virus infections, and autoimmune diseases [90] .

Before the advent of new generation sequencing, comparing the nucleotide sequences for cDNA and reference genomes was a basis of the majority of meth ods used to identify the deamination sites in RNA [91] . Only several tens of RNA editing sites were iden tified by this means [91] . High throughput sequencing was first used to find the adenosine deamination sites in 2009 [92] .

Several drawbacks are characteristic of the meth ods based exclusively on sequencing. Inosine cannot be distinguished from G appearing in cDNA as a result of a sequencing error or a single nucleotide polymor phism. To overcome this drawback, inosine chemical erasing (ICE) was developed taking advantage of the fact that cyanoethylated inosine terminates reverse transcription [93] . Combined with high throughput sequencing (ICE seq), the method was used to iden tify the adenosine deamination site in the human brain transcriptome [94] . 

Adenosine deamination in the pre mRNAs of the glutamate receptor subunit B (GluR B) and serotonin 2C receptor (5 HT2cR) are the best understood cases of RNA editing in the coding region.

Two sites where A → I deamination changes the codon were found in the GluR B mRNA. The R/G site affects the receptor desensitization kinetics [95] , and the Q/R site reduces the Ca 2+ channel permeabil ity [96, 97] . High level editing at the Q/R site is of immense importance in mammals. A decrease in edit ing level is associated with malignant glioma and lat eral amyotrophic sclerosis in humans and causes death almost immediately after birth in mice [98] [99] [100] [101] [102] . A low Q/R substitution rate increases the channel per meability to Ca 2+ and Zn 2+ , thus dramatically chang ing the membrane potential and affecting the cell sig naling pathways [103] . Changes in editing efficiency at the Q/R site of GluR B were observed in forebrain ischemia [103] and were presumably due to a decrease in ADAR2 expression.

Five adenosine residues are subject to deamination in the pre mRNA for G protein coupled 5 HT2cR, and their deamination changes three amino acid resi dues. A combinatorial editing yields 24 different iso forms [104, 105] . Mice that express only the unedited INI isoform of the receptor are normal, while mice that express the fully edited VGV isoform have a sub stantially reduced fat mass in spite of hyperphagia [106] . Changes in 5 HT2cR editing level are associ ated with anxiety, depression, and suicidal behavior [107] .

Adenosine residues in pre mRNA introns are also subject to deamination to thereby affect splicing. Edit ing may generate a new 5' GU splicing site and gener ate or eliminate a 3' AG splicing site [82] . As an exam ple of this editing, ADAR2 performs adenosine deam ination in its own pre mRNA, leading to a frameshift. This is an example of the regulation via negative feed back [82] .

Modification of internal mRNA regions has been known for a long time, but its function remains unknown in the majority of cases. N6 methyladenos ine is one of the most abundant and best studied of all modified nucleotides. Ample data are available for this modification, but the biological roles are still unclear for both modification itself and modification related proteins. The inosine function is well established in certain cases. Inosine generation changes the amino acid sequence of the mRNA encoded protein or regu lates splicing. Less is known about the other modifica tions, and further studies are necessary to better understand their functions.

Considering RNA modifications, we focused mostly on mRNA because its modifications are the least understood and the most interesting in terms of new mechanisms that regulate eukaryotic gene expres sion. Modification of eukaryotic rRNA was intention ally left beyond the scope of this review as a separate problem studied in great detail.

It is clear from the above that data on mRNA mod ification are fragmentary and are difficult to summa rize, especially with the purpose to focus on the roles of individual modified nucleotides. Studies of the rel evant processes will certainly bring many unexpected interesting discoveries.

Topology of the human and mouse m6A RNA methylomes revealed by m6A seq

Com prehensive analysis of mRNA methylation reveals enrichment in 3' UTRs and near stop codons

Widespread occurrence of 5 methylcytosine in human coding and non coding RNA

Pseudouridine profiling reveals regulated mRNA pseudouridylation in yeast and human cells

Transcriptome wide mapping reveals widespread dynamic regulated pseudouridylation of ncRNA and mRNA

Tran scriptome wide mapping of pseudouridines: pseudou ridine synthases modify specific mRNAs in S. cerevi siae

Terminal 7 methylguanosine in eukaryotic mRNA is required for translation

Methylated nucleotides block 5' terminus of HeLa cell messenger RNA

Blocked and methylated 5' termi nal cap structures of rat brain messenger ribonucleic acids

Post transcriptional modifications of mRNA. Purification and character ization of cap I and cap II RNA (nucleoside 2' ) methyltransferases from HeLa cells

2' O ribose methylation of cap2 in human: Function and evolution in a horizontally mobile family

2' O methylation of the viral mRNA cap evades host restriction by IFIT family members

Ribose 2' O methylation provides a molecular signature for the distinction of self and non self mRNA dependent on the RNA sensor Mda5

Identifi cation of methylated nucleosides in messenger RNA from Novikoff hepatoma cells

Mapping of N6 methyladenosine residues in bovine prolactin mRNA

Context effects on N6 adenosine methylation sites in prolactin mRNA

Precise localization of m6A in Rous sarcoma virus RNA reveals clustering of methylation sites: Implications for RNA processing

Comparison of methylated sequences in messenger RNA and het erogeneous nuclear RNA from mouse L cells

High resolution mapping reveals a conserved, widespread, dynamic mRNA methylation program in yeast meiosis

Analysis of RNA base modification and structural rearrange ment by single molecule real time detection of reverse transcription

Identification of a selective polymerase enables detection of N 6 methyladenosine in RNA

Method for site specific detection of m6A nucleoside presence in RNA based on high resolution melting (HRM) analysis

Probing N6 methyladenosine RNA modification status at single nucleotide resolution in mRNA and long noncoding RNA

Purification and cDNA cloning of the AdoMet bind ing subunit of the human mRNA (N6 adenisine) methyltransferase

A METTL3-METTL14 complex mediates mammalian nuclear RNA N6 adenosine methylation

Mamma lian WTAP is a regulatory subunit of the RNA N6 methyladenosine methyltransferase

Wilms' tumor 1 associating protein regulates G2/M transition through stabilization of cyclin A2 mRNA

The obesity associated FTO gene encodes a 2 oxoglut arate dependent nucleic acid demethylase

Oxidative demethylation of 3 methylthymine and 3 methylu racil in single stranded DNA and RNA by mouse and human FTO

N6 methyladenos ine in nuclear RNA is a major substrate of the obesity associated FTO

Loss of function mutation in the dioxygenase encoding FTO gene causes severe growth retardation and multiple malformations

Crystal structure of the FTO protein reveals basis for its substrate speci ficity

Undetectable levels of N6 methyl adenine in mouse DNA: cloning and analysis of PRED28, a gene coding for a putative mammalian DNA adenine methyltrans ferase

FTO mediated for mation of N6 hydroxymethyladenosine and N6 formyladenosine in mammalian RNA

ALKBH5 is a mammalian RNA demethylase that impacts RNA metabolism and mouse fertility

Cloning and characterization of HuR, a ubiquitously expressed Elav like protein

Interplay between microRNAs and RNA binding proteins determines developmental processes

HuR protein attenuates miRNA mediated repression by promoting miRISC dissociation from the target RNA

N6 methylade nosine modification destabilizes developmental regu lators in embryonic stem cells

YTH: A new domain in nuclear proteins

N6 methylad enosine dependent regulation of messenger RNA sta bility

m6A RNA modification controls cell fate transition in mammalian embryonic stem cells

m6A mRNA methylation facilitates resolu tion of naive pluripotency toward differentiation

The methylation state of poly A containing messenger RNA from cultured hamster cells

Methylation of Sindbis virus "26S" messenger RNA

The methylation of adenovirus specific nuclear and cytoplasmic RNA

Methylated, blocked 5 termini in HeLa cell mRNA

NSun2 mediated cytosine 5 methylation of vault noncoding RNA determines its processing into regulatory small RNAs

The RNA methyltransferase Misu (NSun2) mediates Myc induced proliferation and is upregulated in tumors

Identification of human tRNA:m5C methyltrans ferase catalysing intron dependent m5C formation in the first position of the anticodon of the Formula

The mouse cytosine 5 RNA methyltransferase NSun2 is a component of the chromatoid body and required for testis differentiation

The RNA methyltransferase Misu (NSun2) poises epider mal stem cells to differentiate

Identification of direct targets and modified bases of RNA cytosine methyltransferases

Deple tion of Saccharomyces cerevisiae tRNAHis guanylyl transferase Thg1p leads to uncharged tRNAHis with additional m5C

DNA methylation profiling of human chromosomes 6, 20, and 22

RNA methyltrans ferases utilize two cysteine residues in the formation of 5 methylcytosine

Mutagenicity of 5 aza 2' deoxycytidine is medi ated by the mammalian DNA methyltransferase

The nucleolar RNA methyltransferase Misu (NSun2) is required for mitotic spindle stability

Bringing order to translation: The con tributions of transfer RNA anticodon domain modifi cations

Design, biological activity and NMR solution structure of a DNA analogue of yeast tRNAPhe anticodon domain

Expanding the nucleotide repertoire of the ribosome with post transcriptional modifications

Long non coding RNAs as targets for cytosine methylation

Ribonucleic acids from yeast which contain a fifth nucleotide

RNA pseudouridylation: new insights into an old modification

Cocrystal struc ture of a tRNA Ψ55 pseudouridine synthase: Nucle otide flipping by an RNA modifying enzyme

Small nucleolar RNAs direct site specific synthesis of pseudouridine in ribosomal RNA

RNA guided RNA modification: functional organization of the archaeal H/ACA RNP

Pseudou ridylation of yeast U2 snRNA is catalyzed by either an RNA guided or RNA independent mechanism

Function ality and substrate specificity of human box H/ACA guide RNAs

Pseudouridine mapping in the Saccharomyces cerevi siae spliceosomal U small nuclear RNAs (snRNAs) reveals that pseudouridine synthase pus1p exhibits a dual substrate specificity for U2 snRNA and tRNA

Pseudouridylation (Ψ) of U2 snRNA in S.cerevisiae is catalyzed by an RNA independent mechanism

Identification of brain specific and imprinted small nucleolar RNA genes exhibiting an unusual genomic organization

RNomics: Identification and function of small, non messenger RNAs

InterPro in 2011: New developments in the family and domain prediction database

U2 snRNA is inducibly pseudouridylated at novel sites by Pus7p and snR81 RNP

28S rRNA is inducibly pseudouridylated by the mTOR pathway translational control in CHO cell cultures

Four newly located pseudouridylate residues in Escherichia coli 23S ribo somal RNA are all at the peptidyltransferase center: Analysis by the application of a new sequencing tech nique

Incorporation of pseudouridine into mRNA yields superior nonimmunogenic vector with increased translational capacity and biological stability

Converting nonsense codons into sense codons by targeted pseudouridyla tion

Functions and regulation of RNA editing by ADAR deaminases

Regu lation of alternative splicing by RNA editing

RNA editing by adenosine deaminases that act on RNA

RNA editing in reg ulating gene expression in the brain

Molecular cloning of cDNA for double stranded RNA adenosine deaminase, a candidate enzyme for nuclear RNA edit ing

Editing of glutamate receptor B subunit ion channel RNAs by four alternatively spliced DRADA2 double stranded RNA adenosine deaminases

A third member of the RNA specific adenosine deaminase gene family, ADAR3, contains both single and double stranded RNA binding domains

Comparative analysis of the DRADA A to I RNA editing gene from mammals, pufferfish and zebrafish

Require ment of dimerization for RNA editing activity of ade nosine deaminases acting on RNA

Adenosine to inosine RNA editing and human disease

A to I RNA editing: Current knowledge sources and computa tional approaches with special emphasis on non cod ing RNA molecules

Genome wide identification of human RNA editing sites by parallel DNA capturing and sequencing. Sci ence

Inosine cyanoethylation identifies A to I RNA editing sites in the human transcriptome

A biochem ical landscape of A to I RNA editing in the human brain transcriptome

Control of kinetic properties of AMPA receptor chan nels by nuclear RNA editing

Structural determinants of ion flow through recombi nant glutamate receptor channels

Identification of a site in glutamate receptor subunits that controls calcium permeability

Point mutation in an AMPA receptor gene rescues lethality in mice deficient in the RNA editing enzyme ADAR2

Deficient RNA editing of GluR2 and neuronal death in amyotropic lateral scle rosis

Underediting of glutamate receptor GluR B mRNA in malignant gliomas

Reduc tion of GluR2 RNA editing, a molecular change that increases calcium influx through AMPA receptors, selective in the spinal ventral gray of patients with amy otrophic lateral sclerosis

Glutamate receptors: RNA editing and death of motor neurons

ADAR2 dependent RNA editing of AMPA receptor subunit GluR2 determines vulnerability of neurons in fore brain ischemia

Regu lation of serotonin 2C receptor G protein coupling by RNA editing

RNA editing induces variation in desensitization and traf ficking of 5 hydroxytryptamine 2c receptor isoforms

Dysregulated editing of serotonin 2C receptor mRNAs results in energy dissipation and loss of fat mass

Serotonin receptor 2C and mental disorders: Genetic, expres sion, and RNA editing studies

This work was supported by the Russian Science Foundation (project no. 14 14 00072).