During development of the central nervous system (CNS), a wide variety of unique cell types are generated with temporal and spatial precision. In the early embryo, neural stem cells sequentially produce neurons and glial cells that orderly migrate to assemble into neural circuits (1). These processes are accurately curated by dynamic gene expression programs that guide the patterning of the differentiation potential of progenitors and the divergence of neuronal/glial lineages. Epigenetic mechanisms, such as DNA methylation, histone modification, and changes in chromatin architecture, have been extensively investigated in neural development over the last decades (2). In addition, post-transcriptional regulation mediated by chemical modification on RNA provides an additional control of fine-tuned gene expression patterns requiring for the proper development and activity of the nervous system (3). Especially, the advance of high-throughput sequencing approaches and quantitative mass spectrometric analysis revealed the existence of more than 160 types of RNA modifications, including N6-methyladenosine (m6A), 5-methylcytosine (m5C), pseudouridine (ψ), N6,2’-O-dimethyladenosine (m6Am), N1-methyladenosine (m1A) and N4-acetylcytidine (ac4C) (4, 5). Different RNA species, such as transfer RNAs (tRNAs), ribosomal RNAs (rRNAs), messenger RNAs (mRNAs), and non-coding RNAs (ncRNAs) are often post-transcriptionally modified by chemical modifications.
The transcriptomic plasticity, conferred by post-transcriptional regulation including RNA modification, editing, and alternative splicing, is recognized as a fundamental mechanism driving proteomic diversity (6). Notably, recent studies suggested that RNA modifications influence almost all aspects of RNA metabolism, including stability, splicing, localization, and translation. These regulations by RNA modifications have physiologically important functions in many different biological contexts in cell- and/or tissue type-specific manners, eventually opening a newly emerging field, known as the epitranscriptomics.
Several epitranscriptome mapping studies showed that the embryonic and adult tissues that build up the mammalian CNS contain relatively higher abundance of RNA modifications than other organs. For example, higher level of m6A and m6Am in mRNA exist both in human and mouse brain tissues compared to non-brain tissues (7, 8), and it was also reported that either m5C or ψ level was significantly higher in the brain than those in other tissues (9, 10). These results imply that fine-tuning of gene expression by RNA modifications may regulate development and function of the CNS. Therefore, dysregulation of these post-translational regulatory programs often manifests as malformation or dysfunction of the normal CNS, which is implanted in various human brain disorders. In this review, we will overview the recent advances in our understanding of the transcriptome plasticity by RNA modifications in neurodevelopment, and how the alterations in these RNA regulatory programs lead to human brain disorders.
RNA modifications are tightly regulated by specialized RNA-binding proteins. Similar to epigenetic regulatory proteins, epitranscriptomic “writers” catalyze the installation of chemical modifications on RNA, “erasers” reverse the modified chemical groups into the original form, and “readers” recognize the modified RNAs to affect various aspects of RNA metabolism (Fig. 1).
First of all, we will discuss the major epitranscriptomic marks predominantly found in the CNS and their regulatory machineries, and then focus on recent progress to highlight the physiological roles of epitranscriptome in development and disorders of the nervous system.
m6A is the most abundant internal modification in mRNA and noncoding RNA, that affects various aspects of RNA metabolism, including stability, splicing, translation, localization, and biogenesis of specific small regulatory RNAs (4, 5). Genome-wide mapping studies showed that m6A modification generally prefers to be installed at the DRACH (D = A, U or G; R = Purine; H = A, U or C) consensus sequence, and is highly enriched either in the 3’ UTR around stop codons and long exons (7, 11), though sometimes it can be also found in 5’ UTR region with lesser levels (12). The m6A modification is catalyzed by a core methyltransferase complex which is composed of two heterodimeric subunits, corresponding to Methyltransferase like-3 (METTL3) and Methyltransferase like-14 (METTL 14), both of which are essential for precise action of the complex (13). Furthermore, other regulatory proteins, including Wilms tumor 1-associating protein (WTAP), Vir like m6A methyltransferase associated (VIRMA), KIAA1429, zinc finger CCCH-type containing 13 (ZC3H13), and RNA binding motif protein 14 (RBM14/14B) interact with core METTL3-METTL14 complex to contribute to RNA binding specificity and nuclear localization of the core complex (5). On the other hand, two major Fe2+-α-ketoglutarate-dependent m6A demethylases, Fat mass and obesity-associated protein (FTO) and alkB homolog 5 (ALKBH5) are known for their function in erasing m6A marks from target RNAs (14, 15), although the substrate preference of FTO is still controversial due to its demethylase activity either to internal m6A and 5’ cap-specific terminal m6Am modification in different contexts (16, 17). Until now, a dozen of m6A reader proteins have been identified and characterized. Among them, YT521-B homology (YTH)-domain containing protein family, including three YTHDF proteins (YTHDF1/2/3) plus two YTHDC proteins (YTHDC1/2) selectively recognize and directly binds to m6A tag on RNA. YTHDF1 is well-known to directly promote translation of target mRNAs by the recruitment of eIF3, a key component of translation initiation complex, to m6A-modified transcripts (18). On the other hand, YTHDF2 has been known for its significant role in accelerating mRNA degradation through direct recruitment of the CCR4-NOT deadenylase complex (19) or HRSP12–RNase P/MRP complex (20). Next, it was reported that YTHDF3 not only promotes protein synthesis in synergy with YTHDF1, but also affects methylated mRNA decay mediated through YTHDF2 (21). However, a common theme of YTHDF proteins in recent studies is that all three YTHDF proteins would have similar functions and compensate for each other, which is supported by their highly conserved sequence similarity, similar localization in the cytoplasm, as well as their tendency to bind the same targets (22, 23). Thus, the functions of all three YTHDF proteins should be carefully re-examined in order to resolve the controversy whether YTHDF proteins serve distinct or redundant roles in different biological contexts.
Next, YTHDC1 is widely distributed in the nucleus and appears to regulate alternative splicing, by recruiting RNA splicing factor SRSF3, while blocking SRSF10 from binding to mRNAs (24). Moreover, it was also reported that YTHDC1 interacts with SRSF3 and NXF1 to promote the m6A-dependent mRNA nuclear export (25), as well as regulates the transcriptional inactivation of X chromosomal genes mediated by XIST (26). On the other hand, YTHDC2, a putative RNA helicase, has been characterized as a key protein to enhance the translation efficiency (27).
Apart from YTH proteins, a number of other m6A readers have been identified. For instance, HNRNPC/G and HNRNPA2B1, which bind target RNAs through m6A-mediated destabilization of secondary structure, have been shown to regulate pri-miRNA processing and translation (28). Additionally, insulin-like growth factor 2 mRNA-binding proteins 1-3 (IGF2BP1/2/3), as well as Fragile X mental retardation protein (FMRP) were also shown to preferentially bind to m6A-modified mRNAs through evolutionarily conserved RNA recognition element, such as K homology (KH), RNA recognition motif (RRM) and arginine/glycine-rich (RGG) domains (29). Lastly, proline rich coiled-coil 2A (Prrc2a) has more recently been identified as a novel m6A reader protein, which regulates oligodendroglial specification and myelination (30). Collectively, multiple types of m6A reader proteins are specifically recruited by m6A-tagged RNAs to regulate their metabolism. How the similar m6A moieties in RNA are differentially interpreted by reader proteins in specific biological contexts require additional studies to be understood.
Unlike other modification, Nm does not require a specific nucleotide rather it occurs on any kind of bases by adding a methyl group to the 2’-hydroxyl of the ribose molecule (31). In general, Nm can influence RNAs in different ways as it can increase hydrophobicity, protect RNAs from nuclease attacks, stabilize helical structures and affect interaction between modified RNAs and proteins (32). Nm is frequently deposited at internal region of rRNAs and small regulatory RNAs, though tRNAs and mRNAs also have considerable sites for this modification (33). To date, several Nm methyltransferases have been identified. For example, FTSJ1 is a tRNA 2’-O-methyltransferase that targets the C32 and N34 positions in the anticodon loop of tRNAPhe and tRNATrp (34). TRMT44 is a putative 2’-O-methyluridine methyltransferase predicted to methylate residue 44 in tRNASeq (35). In addition, Fibrillarin (FBL) is localized in the dense fibrillar component (DFC) of the nucleolus where newly synthesized pre-ribosomal RNAs reside and methylates specific rRNA targets with help from C/D box family snoRNA (36). Last, CMTR1 is a 2’-O-methyltransferase that modifies the first transcribed nucleotide of the mRNA (37). On the other hand, it remains unclear whether Nm modification is reversible and recognized by specialized proteins for further downstream pathways due to lack of knowledge of defined eraser or reader proteins.
Although methylation of cytosine has been described and characterized as a major epigenetic mark that is frequently added to CpG region in eukaryotic DNA, m5C can also be detected across various RNA species, especially at tRNA and rRNA (38). Previous studies have reported that m5C is related to nuclear export (39) and translation efficiency (40) for certain target RNAs, but general regulatory functions of m5C on gene expression and its precise mechanism still need more further investigation. Additionally, deposition patterns of m5C on RNA are enriched at CG dinucleotides adjacent to transcription initiation sites of mRNA (38). Among several eukaryotic m5C methyltransferases, two key writer proteins, DNMT2 and NSUN2, are have been separately focused on their functions. Originally identified as eukaryotic cytosine-5-DNA methyltransferase, DNMT2 also serves as a RNA m5C methyltransferase, mainly affecting stability and biogenesis of tRNA (41). Conversely, NSUN2 shows broader target specificity, including mRNAs, long non-coding RNAs (lncRNAs) and other small regulatory RNAs, such as vault RNAs, 7SK and Y-RNAs with non-overlapping manner to DNMT2 (38). Very similar to removal pathway of DNA, RNA m5C can be oxidized by ten-eleven translocator (TET) family proteins, giving rise to 5-hydroxymethylcytosine, then 5-formylcytosine and 5-carboxycytosine (42). Although it should be addressed whether oxidized form of cytosine in RNA can be changed to uracil that is suitable for appropriate base-excision repair process, the fact that C-to-U conversion is common phenomenon in RNA and the emergence of key enzyme, SMUG1, which remove 5-methlyuracil in RNA support the possibility that reversible m5C metabolism is likely to be similar to that of DNA (43). Unfortunately, less about exact reader protein for m5C is known so far, except for ALYREF nuclear exporter (39), leaving a question to be addressed.
ψ is known as one of the most prevalent RNA modifications dominantly found in non-coding RNA, though it is also detectable in a subset of mRNAs at a low level (44). Intrinsically, isomerization of uridine to ψ, in which additional hydrogen bonds are adopted, enhances the stability of RNAs by increasing base stacking interactions, resulting in stable secondary structure formation (45). Moreover, ψ is thought to be a key molecular regulator that affects translation efficiency, processing of rRNA and snRNA, as well as telomerase activities (44). Installation of ψ is modulated by two different pathways: guide RNA-dependent, in which the guide H/ACA box snoRNA is required for base pairing to target RNA to be modified, and guide RNA-independent pathway, in which stand-alone protein searches target RNAs and catalyzes the formation of ψ. For instance, human Dyskerin (DKC1) belongs to guide RNA-dependent writers of ψ that is also involved in sno-ribonucleo-protein complex (46), whereas pseudouridine synthases (PUS) family proteins, including PUS1, PUS3, PUS7 and PUS10, are included in guide RNA-independent writers (45) and some of these PUS proteins directly recognize target RNAs in structure-dependent manner (47). Unfortunately, there are no direct eraser and reader proteins identified so far, leaving the reversibility of pseudouridylation and the downstream mechanisms unclear. Taken together, further studies to screen regulatory proteins will be required for better understanding of this epitranscriptomic mark.
The neurodevelopment of animals is composed of a lot of biological events, including early fate decision of neural stem/progenitor cells, generation of neurons and glial cells, migration of post-mitotic immature neurons, formation of synapses, programmed cell death, synaptic rearrangement, as well as maintenance and reconstruction of neurons throughout the postnatal stages. In this section, we will overview the dynamic changes of epitranscriptomic modifications and their functional contribution of transcriptome plasticity in different steps of neurodevelopment (Fig. 2).
In addition, m6A was shown to also regulate histone modifications at genome-wide level (49). In detail, NPCs derived from
In addition, m6A reader proteins also play a significant role in brain development. For example, YTHDF2 mediated the decay of certain transcripts involved in JAK-STAT signaling pathway, which significantly contributes to normal neuroprotection and neurite outgrowth. Consistently,
Gliogenesis proceeds not only in a prenatal but also in a postnatal development.
Cerebellum proceeds its development until postnatal period in chronological order (67). Mettl3-depleted cerebellum showed the enhancement of mRNA stability related to apoptosis, leading to premature cerebellar granule cell death in external granular layer and subsequent cerebellar hypoplasia. In addition, the neuronal layer and structures of Purkinje cell and Bergmann glia cells malformed, which can be resulted from dysregulation of splicing of synapse-associated genes like Grin1 (68). The other study uncovered that balanced expression of m6A writers and readers fine-tunes mRNA methylation in a time-specific way, which leads to normal cerebellar development (69). In addition, under hypoxia condition, Alkbh5-deletion caused abnormal cell proliferation and differentiation in the cerebellum by impaired nuclear export of the hypermethylated RNAs, suggesting that the dynamic regulation of m6A epitranscriptome by Alkbh5-mediated demethylation has important physiological roles depending environmental conditions
After birth, adult neural stem cells (aNSCs) reside in the subgranular zone (SGZ) of the dentate gyrus in the hippocampus and subventricular zone (SVZ) of the lateral ventricles in mouse (70). aNSCs are also regulated in a genetic and epigenetic mechanism upon environmental stimuli and neuronal activity (70), as well as epitranscriptomic regulation. For example, Mettl3 depletion in aNSCs led to reduced proliferation of aNSCs and alteration of differentiation potential toward glial lineage (71). Depletion of an eraser protein, Fto, in aNSC also resulted in reduced proliferation and differentiation of aNSCs in the SGZ, leading to impairment of learning and memory (72). Interestingly, m6A was present on the transcripts of histone methyltransferase Ezh2, and its protein level and consequent H3K29me3 level were markedly dysregulated upon Mettl3 knockdown (71), suggesting highly interconnected regulation between epigenome and epitranscriptome in neurodevelopment. As we mentioned already, interactions between epigenetic and epitranscriptomic regulations were also identified in other systems during neurodevelopment (8, 49).
Complex tasks in the CNS including learning, memory, and cognition require the precise and accurate formation of neural networks through synapse formation and function (73). m6A modification has been suggested to play a role in neuronal maturation and synapse formation (74) as well as in synaptic plasticity (75). Mettl14 depletion led to downregulation of m6A level on mRNAs encoding synapse-specific proteins (76). In cerebellum, Mettl3-mediated m6A regulation affects alternative splicing of synapse-associated pre-mRNAs (68). Ythdf1 regulates activity-dependent neural responses related to learning and memory, and depletion of Ythdf1 showed impairment of synaptic transmission and long-term potentiation in the mouse hippocampus (77). In addition, depletion of Ythdf1 or Ythdf3 in cultured hippocampal neurons led to dysregulation of excitatory synaptic transmission as well as immature spine morpho-logy (78). In another study, Mettl3-depletion in adult mouse brain caused impaired long-term memory formation without morphological alteration of the brain (79). These findings imply that m6A-mediated regulations have crucial roles at synapse for both development and activity-dependent modulation of neural networks.
Taken together, several types of RNA modifications appear to play important roles in controlling RNA metabolisms, such as splicing, degradation, translation, as well as crosstalk with epigenetic mechanisms to regulate gene expression, all of which are essential for proper development of the CNS.
Neurodegenerative disorders, such as Alzheimer’s Disease (AD) and Parkinson’s Disease (PD), are highly correlated with aging (80). While several factors including genomic instability, malfunction of mitochondria, and cellular senescence are hallmarks of aging that affect neurodegeneration (80), it was also reported that mRNA modifications such as m6A and m5C regulate stability of mRNAs which encode senescence-associated proteins such as AGO2 (81, 82). In addition, another study showed that defect in ALKBH8, tRNA methyltransferase was related to senescence with downregulation of seleno-protein synthesis and elevation of reactive oxygen species in mouse embryonic fibroblasts (83). Regarding these points, the perturbation of epitranscriptomic regulation might be a possible mechanism of neurogenerative disorders (Fig. 3).
Lesion in the sciatic nerve induces the elevation of m6A levels on regeneration-associated genes and protein translation machinery components in adult mouse dorsal root ganglion (DRG) to enhance injury-induced protein translation essential for axon regeneration. Indeed, loss of
Stroke is an acute focal injury in the CNS, but if not cured immediately, causes poststroke neurodegeneration and further AD (85). Using transient middle cerebral artery occlusion (MCAO), a popular model of stroke in mouse, a study found that global level of m6A on transcripts related to inflammation, apoptosis, and transcriptional regulation increased after 12 to 24 hours of MCAO (86). The m6A writers were unaltered, but the m6A eraser Fto decreased significantly after stroke, but the functional contribution of m6A in stroke and injury response is yet unclear.
Alzheimer’s disease (AD) is a one of the most common neurodegenerative disease in the old, and progresses into dementia and cognitive impairment (87). In human genetic studies, different genetic variants of
Parkinson’s Disease (PD) is one of the most commonly diagnosed neurodegenerative movement disorder with the symptoms such as tremor and rigidity (94). Several genetic factors and mechanisms are known to be associated with PD, such as loss of dopaminergic neurons and accumulation of misfolded alpha-synuclein (94). FTO overexpression or m6A reduction in dopaminergic neurons upregulated GRIN1 expression, leading to subsequent elevation of oxidative stress and Ca2+ influx and apoptosis (95). On the other hand, Fto depletion in dopaminergic neurons caused impaired dopamine receptor type 2 mediated signaling, which led to attenuated activity of G protein-coupled inwardly-rectifying potassium channel (GIRK channel) and dysregulation of behaviors controlled by dopaminergic transmission (96). Recently, entacapone, known as catechol-o-methyltransferase inhibitor used for PD treatment, has been suggested as FTO inhibitor (97). Entacapone directly bound to and deactivated FTO, then inhibited the activity of transcription factor forkhead box protein O1 (FOXO1). As genetic and epigenetic factors have been suggested to play a role in regulating neurodegenerative diseases, specific molecular and cellular mechanisms in epitranscriptomic way are yet to be more investigated.
Intellectual disability (ID) is a complicated developmental disorder concluding cognitive impairment, learning failure, maladjustment to social environment, and so on. Several genetic factors including
Among X-linked genes, Fragile X Mental Retardation Protein (FMRP) is related to synaptic function, and depletion of this protein caused degradation of its target mRNAs (100). Interestingly, FMRP interacts with YTHDF2, m6A reader protein to enhance stability of its target mRNAs. Moreover, frameshift mutation on
Homozygous mutation of truncation form in
NSUN2, m5C writer, was also suggested as the cause of ID. For example, mutations in
From 1990s, FTJS1, Nm writer, has been reported for its mutation to be associated with nonsyndromic X-Linked mental retardation (NS-XLMR) in human genetic studies (109, 110). Moreover, single nucleotide polymorphisms (SNPs) (111, 112) or copy number variations (113) of
FTO, which demethylates both m6A and m6Am (m6A/m, collectively), has been actively investigated in mental disorders even before generally known as being related to RNA modification. Exposure to arsenite led to decrease of Fto expression, which increased m6A level and subsequent dysregulation related to deficits in dopaminergic neurotransmission (116). Several SNPs of
Glucocorticoid response upon chronic psychological stress, which underlies several mental disorders, induces profound time-specific alteration of m6A/m landscape (75, 119). Deletion of
Neuroblastoma is known to be the most common solid tumor found in early children and derived from several genetic aberrations. Interestingly, polymorphisms in
Glioma, which is one of the most common malignant tumor of astrocytes in brain, has four different grades defined by WHO (I-IV): low-grade and anaplastic astrocytoma (WHO grades I-III) and glioblastoma (WHO grade IV) (126). The prognosis of glioma patients is quite poor even its treatment has been evolved (127). Among the m6A-related proteins, Wilms’ tumor 1-associating protein (WTAP) is implicated for the marker of glioblastoma (128). WTAP was over-expressed in glioblastoma with 169 clinical samples with glioma patients, and the higher expression of WTAP was correlated with poorer prognosis (128), suggesting m6A regulatory proteins may have important roles in tumor progression.
Glioblastoma stem cells (GSCs) were suggested as a primary source that contributes to tumor propagation, maintenance, and treatment resistance (129), which m6A-mediated regulation also takes part in. For example, GSC-mediated tumorigenesis was markedly promoted by knockdown of
As altered histone modification and their modifiers are related to glioma genesis, these epigenetic regulators have been regarded as therapeutic targets and applicable biomarkers (133). Similarly, epitranscriptomic factors have been suggested as appli-cable biomarkers and therapeutic targets in gastrointestinal cancer (134) and renal cancer (135). To identify biomarkers of glioblastoma, m6A-lncRNA co-expression networks were constructed through statistical analysis of primary glioblastoma patients, which screened four lncRNAs with their co-expressed functional genes. In that the lncRNA expression was correlated with the effect of m6A modification, it is suggested that post-transcriptional regulation of noncoding RNAs may have a significant role in dynamic gene expression control of glioblastoma (136). In addition, twenty-four lncRNAs were explored to be prognostic m6A-related lncRNAs, which shows the distinct m6A status between low- and high-risk subgroups of lower-grade glioma patients (137). Moreover, it was reported that FTO inhibition enhanced effectiveness of chemotherapy of glioma (138), and a newly synthesized inhibitor of ALKBH5 successfully kept specific glioblastoma cell line from migration and invasiveness (139), suggesting that epitranscriptomic regulators will be considered as important targets for development of potential therapeutic intervention of brain cancers in the future.
In summary, we have reviewed overall mechanisms of epitranscriptomic gene regulation and endeavors to identify the physiological functions of various epitranscriptomic modifications and their mechanisms. Indeed, the dynamic regulations of transcriptome plasticity via RNA modifications have been demonstrated to be involved in prenatal and postnatal neurodevelopment ranging from neural stem cell establishment to adult neurogenesis. However, only few of writer, eraser, and reader proteins modulating these neurodevelopmental processes have been investigated in detail. Further studies on regulator proteins of epitranscriptome and cell-type specific reader proteins will be necessary to appreciate the whole picture of dynamic transcriptome plasticity in development and physiology of the brain. In addition, it is still challenging to detect a specific RNA modification with high resolution and low background signals by current epitranscriptome mapping technologies that majorly rely on the sensitivity and the specificity of modification-specific antibodies. Therefore, recently developed antibody-independent approaches (140, 141) will greatly improve our understanding on the landscape of epitranscriptome upon various cellular and environmental context at single base-pair resolution.
In addition, we examined several brain disorders including neurodegenerative diseases, intellectual disability, mental disorder, and brain cancer in the context of epitranscriptomic regulation. Accumulating evidences suggest that the dysregulated epitranscriptome is a potent pathological mechanism of brain disorders, together with genetic and epigenetic factors. From this perspective, novel biomarkers to predict disease progression as well as new therapeutics are currently being developed by investigating and modulating epitranscriptomic regulation. Through continued efforts to advance epitranscriptome mapping technologies, to uncover functional mechanisms of key regulatory proteins of RNA modification, and to develop novel reagents to control disease-related features of transcriptome plasticity, we believe our expanding understanding of epitranscriptome will significantly contribute to deciphering the daunting complexity of brain disorders.
We thank for wonderful illustrations by Jaewook Lee, and for careful reading and suggestions by Huiseon Hwang. This work was supported by the National Research Foundation of Korea (NRF) grants (2018R1A5A1024261, 2019R1C1C1006600, 2017 M3C7A1047654 to K.-J.Y., the Brain Pool program to N.-S.K.) funded by the Korean Ministry of Science, ICT, and Future Planning (MSIP), and the Young Investigator Grant from the Suh Kyungbae Foundation (to K.-J.Y.),
The authors have no conflicting interests.