The roles and implications of RNA m⁶A modification in cancer

Abstract

N⁶ methyladenosine (m⁶A), the most prevalent internal modification in eukaryotic messenger RNA (mRNA), has been extensively studied in recent years. Dysregulation of m⁶A and its associated machinery (including “writers”, “erasers” and “readers”) has been frequently observed in various cancer types, and their dysregulation profiles may serve as prognostic/predictive biomarkers in cancers. The dysregulated m⁶A modifiers have been shown to function as oncogenes or tumour suppressors, and play essential roles in cancer initiation, progression, metastasis, metabolism, drug resistance, and immune evasion, as well as in cancer stem cell self-renewal and tumour microenvironment, highlighting the therapeutic potential of targeting the dysregulated m⁶A machinery for cancer treatment. In this review, we illustrate the mechanisms by which m⁶A modifiers interplay and orchestrate the fate of target RNAs. We also describe the state-of-the-art methodologies for mapping global m⁶A in cancer epitranscriptomes. We further summarize the recent discoveries regarding the dysregulations of m⁶A and the associated machinery in cancers and their pathological roles and underlying molecular mechanisms in cancers. Finally, we discuss the findings on m⁶A related molecules as prognostic/predictive biomarkers in cancers, as well as the recent development of small-molecule inhibitors targeting oncogenic m⁶A modifiers and their preclinical results in treating cancers.

Introduction

Tumourigenesis is triggered by altered gene expression. Besides alterations of DNA sequence, gene expression is also controlled by multiple layers of gene regulations at the DNA, RNA and protein levels, such as epigenetic, epitranscriptomic, transcriptional and translational regulations. Epitranscriptomics represents a novel layer of gene regulation at the RNA level, which post transcriptionally regulates the target RNA with over 170 types of chemical modifications such as N⁶ methyladenosine (m⁶A), pseudouridine (Ψ), 5-methylcytidine (m⁵C), N¹ -methyladenosine (m¹A), and N⁷ -methylguanosine (m⁷G). Among these chemical modifications, m⁶A, as the most abundant one for mRNA and long non-coding RNA (lncRNA), is a leading regulator of this layer, which affects RNA fate and expression via regulating RNA splicing, nuclear export, stability, and translation. Dysregulation of m⁶A induced gene alterations has been recently widely studied and identified as a critical player in various cancers. With advanced transcriptome-wide sequencing techniques, m⁶A was found evolutionarily conserved and dynamically reversible, but its profile is very cell context-dependent. Consistently, dysregulated m⁶A and its modifiers (including “writers”, “erasers” and “readers”) have been shown to play oncogenic or tumour suppressive roles in a cancer-type-dependent manner.

Mechanisms and biological impacts of m⁶A modifiers

RNA m⁶A modification is dynamically regulated by its “writers” and “erasers” (Figure 1). The majority of mRNA m⁶A is deposited co-transcriptionally by a “writer” complex called methyltransferase complex (MTC)^1–3. The MTC is comprised of 3 key components: METTL3, METTL14, and WTAP^2–4. Additional accessory subunits of the MTC, including VIRMA, RBM15/RBM15B, and ZC3H13, assist with recognition of specific RNA binding sites and nuclear localization of the complex^5–8. Apart from the MTC, METTL16⁹, METTL5¹⁰, and ZCCHC4¹¹ have also been recently reported to possess m⁶A methyltransferase activity for targets such as mRNAs, lncRNAs, small nuclear RNAs (snRNAs), and ribosome RNAs (rRNAs). So far only two m⁶A “erasers”, FTO and ALKBH5, have been identified. FTO, the first discovered RNA demethylase, removes the methyl group from m⁶A or m⁶Am in mRNAs and m¹A in transfer RNAs (tRNAs)^12,13. ALKBH5, unlike FTO, only targets m⁶A¹⁴.

Figure 1. RNA m⁶A machinery and various aspects of RNA fates regulated by m⁶A methylation.

The installation of m⁶A methylation for most mRNA transcripts is fulfilled by the methyltransferase complex (MTC) comprising a core complex and other accessory subunits. Methyltransferase (METTL) 3, METTL14, and WT1 associated protein (WTAP) constitute the core complex. Other “writers” apart from the MTC, such as METTL16, phosphorylated CTD interacting factor 1 (PCIF1), and METTL5, govern m⁶A deposition for their own specific targets. Two “eraser” proteins, FTO alpha-ketoglutarate dependent dioxygenase (FTO) and alkB homolog 5, RNA demethylase (ALKBH5), are responsible for demethylation of m⁶A-modified RNAs. The chemical structure of m⁶A is shown (bottom left), with its methyl group highlighted. A variety of m⁶A reader proteins, represented by the YT521-B homology (YTH) family, the insulin-like growth factor 2 mRNA-binding protein (IGF2BP) family, and the heterogeneous nuclear ribonucleoprotein (HNRNP) family, specifically recognize m⁶A marks and mediate the regulation of various aspects of target RNA fates, including RNA stability, translation, splicing, and nuclear export. Respective reader proteins involved in these aspects are listed and indicated for their roles when applicable (upward arrow indicates a promoting role and downward arrow indicates an inhibitory role).

The executors of m⁶A signalling, m⁶A readers, can be divided into three major groups. The first group are YT521-B homology (YTH) domain family proteins including YTHDF1/2/3 and YTHDC1/2. YTH readers directly recognize m⁶A sites via their YTH domains, impacting target mRNA fate by regulating their stability^15–17, translation^16–18, splicing¹⁹, and nuclear export²⁰ (Figure 1). Specifically, YTHDF2 localizes its targets from the translatable pool to mRNA degradation machineries, resulting in decay of the targets¹⁵. YTHDF1, on the other hand, facilitates ribosome loading of its targets to expedite translation¹⁸. YTHDF3 assists YTHDF1 in enhancing translation efficiency and modulates YTHDF2-mediated mRNA decay¹⁶. Nuclear-localized YTHDC1 governs splicing and nuclear export of m⁶A-decorated mRNAs^19,20 whereas the cytoplasmic reader YTHDC2 accelerates translation and impairs stability of target transcripts¹⁷. Insulin-like growth factor 2 mRNA-binding protein 1/2/3 (IGF2BP1/2/3) represent another distinct class of readers that contain four common RNA binding K homology (KH) domains, which are required for the selective binding of m⁶A-modified RNAs²¹. The mutations in KH3–4 domains completely abolished the m⁶A recognition and binding capacity of IGF2BPs, yet KH3–4 peptides alone display little m⁶A selectivity²¹, suggesting that the KH domain flanking regions contribute to m⁶A selectivity. Nevertheless, the detailed mechanism underlying the m⁶A selective binding of IGF2BPs requires further investigation (e.g., by structural studies). IGF2BPs increase the stability of m⁶A-containing transcripts through recruiting RNA stabilizers such as ELAVL1, MATR3, and PABPC1, and also promote target mRNA translation likely through recruiting eIF factors^21,22. There is also evidence that IGF2BPs interfere with microRNA (miRNA)-dependent downregulation of target mRNAs^23–25, but whether this is m⁶A-dependent is elusive. It has been shown in the case of SRF that IGF2BP1 protein selectively binds SRF mRNA via its m⁶A-modified 3’UTR and protects it against miRNA-directed mRNA decay, suggesting that IGF2BPs stabilize at least some of their m⁶A-marked targets through an miRNA-dependent mechanism²³.FMR1 is proposed to bind m⁶A sites via a similar mechanism using its three KH domains and one Arg-Gly-Gly [RGG] box to facilitate mRNA nuclear export, but whether FMR1 favors m⁶A-marked RNAs over unmethylated ones is still controversial²⁶. The third group of readers recognize “m⁶A-switch”, the remodelled local RNA structure induced by m⁶A, and some heterogeneous nuclear ribonucleoproteins (hnRNPs), such as hnRNPA2B1, hnRNPC, and hnRNPG^27–29. The ongoing search for RNA binding proteins that prefer m⁶A probes over unmethylated ones (or unmethylated probes over m⁶A probes in the scenario of antireader) is still expanding the collection of m⁶A readers (antireaders).

The process of m⁶A deposition is tightly regulated by both methyltransferase-intrinsic specificity and extrinsic factors. Despite the well-established DRACH (D=A/G/U, R=A/G, H=A/C/U) consensus motif where the MTC selectively installs m⁶A, only ~5% of all DRACH-containing transcripts are m⁶A-modified³⁰. Further, m⁶A does not display random distribution across the transcriptome, but rather is enriched near stop codons and in internal exons longer than 200 nt^30,31, implying that factors beyond primary sequence contribute to the determination of methylation sites. External determinants (m⁶A modulators), such as transcription factors and RNA binding proteins that recruit the MTC to certain transcripts^32–34, alters m⁶A specificity and stoichiometry (i.e. the fraction of a given site that is m⁶A-modified) in a cellular context-dependent manner, supporting a dynamic and responsive nature of m⁶A. It is reasonable to speculate that m⁶A stoichiometry has crucial effects on gene expression and cellular functions, although our knowledge of such effects is still very limited due to the lack of effective quantitation methods previously. The newly developed transcriptome-wide quantitative detection techniques^35–37 are expected to reveal the critical connection between m⁶A stoichiometry and transcript abundance/biological consequences. Moreover, even when m⁶A status remains the same for a particular site, as dictated by both intrinsic and extrinsic factors, divergent expression patterns and activities of m⁶A readers across diverse cell types and cell states may lead to differential recognition of m⁶A and thereby distinct functional outcomes, adding yet another layer of regulation of m⁶A biological effects.

Methodologies for characterizing and mapping the landscape of m⁶A methylation

Despite the fact that global m⁶A levels can be reliably and readily measured by methods such as RNA immunoblotting and liquid chromatography-tandem mass spectrometry (LC-MS/MS), transcriptome-wide profiling of m⁶A by next-generation sequencing exclusively provide information about locus-specific m⁶A changes and have advanced our understanding of the wide-ranging roles of m⁶A in gene expression regulation. Supplementary Table 1 summarizes the merits and limitations of current popular high-throughput m⁶A-specific sequencing techniques.

The m⁶A epitranscriptome was first mapped by two independent studies in 2012 through combining antibody-based enrichment of m⁶A-containing RNA fragments with deep sequencing, a technique termed m⁶A-seq or MeRIP-seq^30,38. This method produces m⁶A maps with a resolution of around 100–200 nt and requires a starting amount of at least 300 μg total RNA. Thereafter, various improved versions of antibody-based approaches have been developed with advantages of higher resolution, less starting materials, and/or capacity for m⁶A quantitative detection, such as PA-m⁶A-seq³⁹, miCLIP/ m⁶A-CLIP^40,41, m⁶A-LAIC-seq⁴², m⁶ACE-seq⁴³, m⁶A-seq2⁴⁴, and SLIM-seq⁴⁵. However, antibody-based methods still fail to provide m⁶A stoichiometric information at single-base resolution to study its biological functions and demonstrate inherent antibody-related limitations: (1) anti-m⁶A antibodies are unable to discriminate RNA m⁶A modifications from DNA 6mA modifications, nor can it differentiate between m⁶A and m⁶Am⁴⁰; (2) anti-m⁶A antibodies may be biased toward specific RNA sequences and secondary structures, as shown in a yeast study⁴⁶. Consequently, a variety of antibody-independent m⁶A profiling strategies have been developed, including MAZTER-seq/m⁶A-REF-seq^47,48, m⁶A-SEAL⁴⁹, m⁶A-label-seq⁵⁰, and DART-seq⁵¹. Each of these techniques has its own advantages, ranging from the capability of profiling m⁶A from low quantities of input RNA to less false-positive signals. Three recently developed antibody-independent techniques, m⁶A-SAC-seq, GLORI, and eTAM-seq, offer new options to quantitatively map transcriptome-wide m⁶A at a single-base resolution^35–37. m⁶A-SAC-seq takes advantage of MjDim1, an enzyme that converts m⁶A into N⁶-allyl,N⁶-methyladenosine (a⁶m⁶A), which subsequently generates mutations during reverse transcription³⁵. m⁶A-SAC-seq can be applied to as little as ~30 ng of poly(A)⁺ RNA to produce quantitative m⁶A maps but displays a preference for the Gm⁶AC over the Am⁶AC motif. Unlike m⁶A-SAC-seq, GLORI and eTAM-seq are conceptually similar to DNA 5-methylcytosine bisulfite sequencing and detects unmethylated A via conversion of A-to-I chemically (GLORI) or enzymatically (eTAM-seq)^36,37. Both methods calculate m⁶A stoichiometry by subtracting unmethylated A from total A, and eTAM-seq requires relatively low starting materials (50 ng of poly[A]+ RNA) for high-throughput sequencing and ultra-low RNA inputs (250 pg of total RNA) for site-specific m⁶A quantification (Supplementary Table 1)^36,37. However, neither GLORI or eTAM-seq can differentiate m⁶A from other adenine modifications, and considering the much higher abundance of A than m⁶A (over 200-fold), an approach that directly detects m⁶A instead of A is still preferred. Direct sequencing of native RNA by Oxford Nanopore technology, the so-called “third-generation sequencing”, is sensitive to RNA modifications and can thus become a potential detection method for such modifications including m⁶A in the future⁵². Overall, tremendous advancement in detection techniques that allows precise and accurate quantitative detection of m⁶A at single-nucleotide resolution have considerably raised the ability to evaluate the precise distributions and effects of m⁶A in cancer biology to an unprecedented level. Notably, the m⁶A signals detected for the same transcripts are largely consistent across different methods³⁷, and the newly developed single-nucleotide quantitative methods (m⁶A-SAC-seq/GLORI/eTAM-seq) can provide accurate and consistent qualification of m⁶A methylation at the transcriptome-wide level^35–37. Nevertheless, more advanced methods that allow for transcriptome-wide m⁶A quantitative detection at single-base and single-cell resolution are warranted.

m⁶A modification in cancer

Due to the enormous impact of m⁶A on gene expression and cellular function, the dynamic equilibrium of methylation is under tight control, and disruption of which may lead to diseases such as cancer^53–56. In fact, it has been widely reported that the levels of m⁶A modification and its modifiers are dysregulated in various types of cancer, contributing to tumour initiation, progression and metastasis by regulating m⁶A-modified coding and non-coding transcripts and related pathways, which is summarized in Figure 2. Notably, recent studies revealed that some m⁶A enzymes also function in m⁶A-independent manners in certain cancers, which are also discussed below.

Figure 2a. Oncogenic functions of m⁶A modifiers in cancer.

Tumour-promoting m⁶A modifiers (including writers, erasers and readers) and their representative downstream targets, including coding and non-coding RNAs, are illustrated. Positive targets of m⁶A modifiers are in red, while negative targets of m⁶A modifiers are in blue. METTL3, methyltransferase 3; METTL14, methyltransferase14; WTAP, WT1 associated protein; METTL16, methyltransferase16; FTO, FTO alpha-ketoglutarate dependent dioxygenase; ALKBH5, alkB homolog 5; YTHDF1, YTH N6-Methyladenosine RNA Binding Protein 1; YTHDF2, YTH N6-Methyladenosine RNA Binding Protein 2; YTHDF3, YTH N6-Methyladenosine RNA Binding Protein 3;YTHDC1, YTH domain-containing protein 1; YTHDC2, YTH domain-containing protein 2; IGF2BP1, Insulin-like growth factor 2 mRNA-binding protein 1; IGF2BP2, Insulin-like growth factor 2 mRNA-binding protein 2; IGF2BP3, Insulin-like growth factor 2 mRNA-binding protein 3.

Oncogenic roles of m⁶A modifiers in cancers

Oncogenic roles of m⁶A writers

METTL3 has been predominantly reported to play oncogenic function in hematopoietic malignancies and solid tumours, except for endometrial cancer⁵⁷. METTL3 expression was significantly higher in cancerous tissues than in corresponding paracancerous tissues in colorectal cancer (CRC), gastric cancer (GC), lung cancer, breast cancer and hepatocellular carcinoma (HCC) etc., especially in metastatic tissues^58–62. High level of METTL3 often correlates with unfavourable prognosis of cancer patients⁵⁷. The dysregulation of METTL3 in cancer could be attributed to different mechanisms, including transcription activation by transcription factors and histone modifications, post-transcription regulation by miRNAs and post-translation regulation by SUMOylation and lactylation^57,63,64. In most cases, METTL3 plays critical roles in cancer as an m⁶A methyltransferase. For instance, overexpression of METTL3 in the wild-type (WT) but not the catalytically dead form in acute myeloid leukemia (AML) cells promotes cell growth⁶⁵. METTL3 promotes leukemogenesis through promoting the stability and translation of MYC, BCL2, PTEN, SP1, and SP2 mRNAs in an m⁶A-dependent manner^32,65. Treatment of AML cells with STM2457, a selective inhibitor that blocks METTL3’s methyltransferase activity, effectively kills leukemia cells and prevents AML progression in preclinical animal models⁶⁶, further suggesting the essential role of METTL3 as an m⁶A methyltransferase in AML. Meanwhile, METTL3 was also reported to function as an m⁶A reader to enhance the translation of its target mRNAs and thereby contribute to the progression of human lung adenocarcinoma (LUAD)⁶⁷. Mechanistically, METTL3 binds simultaneously to the m⁶A-modified 3’UTR of target mRNAs and to the eukaryotic eIF3h in proximity to the 5’ cap of target mRNAs, thus facilitating ribosome recycling and accelerating translation⁶⁰. Recently, METTL3 was also reported to promote translation of non-m⁶A mRNAs in GC through the interaction among METTL3-PABPC1-eIF4F⁶⁸. As the cytoplasmic mislocalization of METTL3 is also observed in other malignancies^68,69, the translation-promoting function and other non-enzymatic functions of METTL3 might be more prevalent than appreciated.

METTL14 exerts oncogenic effects in leukemia and solid tumours. In AML, we reported that METTL14 is overexpressed in AML and promotes the development/maintenance of AML and the self-renewal of leukemic stem/initiating cells (LSCs/LICs) through the SPI1-METTL14-m⁶A-MYC/MYB axis⁷⁰. In pancreatic cancer, METTL14 markedly promotes cancer cell proliferation and migration, via modulating m⁶A and stability of PERP mRNA and the associated p53 pathway⁷¹. METTL14 could also promote the progression of cervical cancer, oral squamous cell carcinoma (OSCC), prostate cancer, glioma, and breast cancer through regulating the expression of MYC, MALAT1, THBS1, ASS1, and CXCR4/CYP1B1 in an m⁶A-dependent manner⁷². Interestingly, a viral-encoded latent oncoprotein EBNA3C could transactivate METTL14 and stabilize its protein product to regulate the expression of Epstein-Barr virus (EBV) latent transcripts and thereby promote EBV-mediated tumourigenesis⁷³.

Previously known as the WT1 associated protein, WTAP was found to be aberrantly expressed in glioblastoma (GBM), AML, and cholangiocarcinoma^54,74. In AML, WTAP facilitates m⁶A deposition and enhances stability of MYC and WWTR1 mRNAs to promote leukemogenesis, while its expression is regulated by METTL3 and HSP70/90 via protein synthesis and degradation, respectively^69,74. In HCC, WTAP modulates G2/M cell cycle progression and autophagy by regulating expression of ETS1 mRNAs in an m⁶A-dependent manner⁷⁵. Beside coding RNAs, miR-200 miRNA and DIAPH1-AS lncRNA have also been characterized as WTAP targets whose m⁶A and expression levels were increased by WTAP, resulting in vigorous glycolysis in ovarian cancer and nasopharyngeal carcinoma (NPC)^76,77. In addition, WTAP particularly mediates mTORC1-promoted cell proliferation in mTORC1-driven cancers⁷⁸, in which mTORC1 stimulates both the synthesis of the methyl donor S-Adenosyl methionine (SAM) and the translation of WTAP, therefore affecting global RNA m⁶A modification and MYC pathway^78,79.

METTL16 was reported to be upregulated in GC, HCC, pancreatic cancer and AML, and its high expression is associated with poor prognosis^80–84. Similar to METTL3, METTL16 plays oncogenic function via both methyltransferase-dependent and -independent mechanisms. In nuclear, METTL16 functions as an m⁶A methyltransferase responsible for m⁶A deposition on hundreds of mRNAs and non-coding RNAs^9,81. For instance, METTL16 mediates m⁶A modification and enhanced stabilization of cyclin D1 mRNA to promote cell cycle progression and proliferation in GC⁸², while deposits m⁶A on tumour suppressive lncRNA RAB11b-AS1 to accelerate its decay in HCC⁸³. In cytosol, METTL16 facilitates translation initiation in an m⁶A-independent manner through directly interacting with eIF3A, eIF3B and rRNAs to promote translation of thousands of mRNAs⁸¹. METTL16 is also overexpressed in AML and is required for AML development/maintenance and LSC self-renewal⁸⁴.

Oncogenic roles of m⁶A erasers

Dysregulation of m⁶A demethylases was also frequently observed in cancers. FTO is upregulated and exerts an essential tumour-promoting role in various cancer types^85,86. We reported that FTO is overexpressed and plays critical oncogenic roles in AML by blocking myeloid differentiation, promoting AML development/maintenance and LSC/LIC self-renewal, and enhancing AML aerobic glycolysis and immune evasion^87–90. Mechanistically, FTO promotes the expression of MYC, PFKP, LDHB, CEBPA, and LILRB4, while suppressing expression of ASB2 and RARA, in an m⁶A-dependent manner in AML^87–90. Consistently, small molecule inhibitors (e.g., FB23/FB23–2 and CS1/CS2) targeting the demethylase activity of FTO showed remarkable anti-tumour efficacy in vitro and in vivo^87,91. In keratinocytes, FTO is degraded by selective autophagy, which is impaired by low-level arsenic exposure during arsenic-induced malignant transformation⁹². Targeting FTO genetically or pharmacologically could inhibit arsenic-induced tumourigenesis, supporting FTO as a potent prevention target of skin cancer⁹². FTO is indeed upregulated in primary (stage I-IV) and metastasis melanoma, and could promote melanoma growth and decrease response to anti-PD-1 blockade immunotherapy by demethylating PDCD1 (PD1), CXCR4, and SOX10 mRNAs⁹³. In addition, FTO regulates the m⁶A level and stability of BNIP3, CAV1(caveolin1), EIF4G1, and PDGFC mRNAs, as well as lncRNAs MALAT1 and LINC00022, to affect apoptosis, mitochondrial fission/fussion, autophagy, cell growth and metastasis in breast cancer, GC, OSCC, pancreatic cancer, bladder cancer, and esophageal squamous cell carcinoma (ESCC)⁸⁵.

ALKBH5 is also frequently dysregulated in cancers and acts as an oncogene or tumour suppressor depending on the cancer type. In particular, ALKBH5 has been reported to induce cancer stem cells (CSCs) self-renewal in many malignancies such as AML, breast cancer, GBM and gynaecologic cancers^94–96. We and others reported that ALKBH5 is overexpressed and associated with poor prognosis in AML patients, and ALKBH5 is required for AML development/maintenance and LSC/LIC self-renewal but is dispensable for normal haematopoiesis and self-renewal of hematopoietic stem cells (HSCs)^96,97. Mechanically, the transcription of ALKBH5 is activated by KDM4C-mediated H3K9me3, while ALKBH5 demethylates and affects the stability of target mRNAs such as TACC3 and AXL^96,97. In breast cancer and endometrial cancer, exposure of cancer cells to hypoxia stimulates HIF-dependent ALKBH5 expression, which mediates enrichment of CSCs in the tumour microenvironment via regulating the expression of core pluripotency factors NAONOG and SOX2⁹⁴. In GBM, the expression of ALKBH5 is also induced by hypoxia, leading to uncontrolled CSC growth through regulating FOXM1, and the emergence of an immunosuppressive microenvironment facilitating tumour evasion through regulating lncRNA NEAT1^95,98. In sum, ALKBH5 is upregulated in CSCs by epigenetic regulation or environment factors (e.g., hypoxia) and facilitates CSC self-renewal/expansion as an m⁶A eraser. Together with the research of FTO, it is suggested that demethylases are potent therapeutic targets for eliminating CSCs and preventing tumour recurrence.

Oncogenic roles of m⁶A readers

As the earliest identified m⁶A readers, YTH family proteins have been well studied in many cancer types, including AML, CRC, breast cancer, GBM, and lung cancer. In AML, YTHDF2 and YTHDC1 play oncogenic roles and were proposed as potential therapeutic targets^99–102. YTHDF2 is overexpressed in human AML and is required for AML development/maintenance⁹⁹. Suppression of YTHDF2 selectively compromised LSCs/LICs in AML, while promoting expansion and maintenance of functional HSCs, through increasing the half-life of a group of mRNAs critical for stem cell-fate determination^99,103. YTHDC1 protects m⁶A-modified mRNAs from the PAXT-complex and exosome-associated RNA degradation in a phase-separated nuclear body in leukemia cells¹⁰⁰. YTHDC1 is also overexpressed in AML and essential for the proliferation and survival of human AML cells^100,101. Notably, Ythdc1 is haploinsufficient for the self-renewal of LSCs¹⁰¹. In breast cancer, YTHDF1 functions through the HIF1A-miR-16-YTHDF1-PKM2 axis and thereby induces hypoxia-dependent cancer cell growth and metastasis¹⁰⁴, while YTHDF3 regulates the translation of m⁶A-enriched transcripts for ST6GALNAC5, GJA1, and EGFR, and thereby promotes brain metastasis through enhancing the interaction between cancer cells and the brain microenvironment¹⁰⁵. In GBM, high expression of YTHDF2 is clinically correlated with poor glioma patient prognosis, and YTHDF2 promotes GBM cell proliferation, invasion, and tumourigenesis^106,107. Interestingly, although YTHDF2 has been known as a destabilizer of m⁶A transcripts in many cancers and mediates mRNA decay of LXRA and HIVEP2 in GBM cells¹⁰⁷, YTHDF2 also stabilizes MYC and VEGFR mRNAs in GSCs by an unknown mechanism¹⁰⁶, suggesting the function of m⁶A reader is complicated and likely context dependent. It should be pointed out that although the functional compensation of YTHDF members has been reported in certain context during development^108–110, YTHDF proteins may not function redundantly in cancer. Specifically, depletion of YTHDF1/2/3 individually in CRC and other cancers is sufficient to suppress cancer cell survival and induce apoptosis, by targeting different sets of transcripts^111–113.

IGF2BP1/2/3 are among the most upregulated RNA binding proteins (RBPs) across The Cancer Genome Atlas (TCGA) panels and emerging as essential players in human cancers¹¹⁴. IGF2BP1 and IGF2BP3 are known as oncofetal proteins that are not expressed in adult tissues and can serve as prognostic biomarkers in ovarian cancer, CRC and neuroblastoma^115,116. Although IGF2BP2 is ubiquitously expressed in most normal tissues, its overexpression could also be detected in many cancers, especially in high-stage or high-grade tumours^115,116. Recently, IGF2BPs have been extensively exploited in cancers with respect to their roles as m⁶A readers in regulating expression of classical oncogenic targets, such as MYC and KRAS, as well as many growth factors and epithelial-mesenchymal transition (EMT)-promoting genes^21,117. IGF2BP1 was previously identified as a stabilizer of MYC mRNA through binding to the coding region stability determinant (CRD) region¹¹⁸. We showed that MYC mRNA, especially the CRD region, has high abundance of m⁶A, which facilitates IGF2BPs’ binding and thus enhances the stabilization and translation of MYC mRNA²¹. IGF2BP1 enhances CBX8 mRNA stability to regulate CSC self-renewal and chemosensitivity in colon cancer¹¹⁹. IGF2BP3 modulates aerobic glycolysis through targeting HK2 and PDK4 mRNAs in cancer cells¹²⁰. We recently demonstrated that IGF2BP2 promotes AML development and self-renewal of LSCs/LICs by upregulating expression of MYC, SLC1A5, and GPT2, and pharmacological inhibition of IGF2BP2 shows promising therapeutic efficacy in treating AML²². It is noteworthy that noncoding RNAs (ncRNAs) play crucial roles in mediating the pro-tumoural effects of IGF2BP family, not only as downstream targets (e.g., MALAT1, ZFAS, circNSUN2, DANCR, PACERR, circMDK, circMAP3K4), but also as upstream modulators of IGF2BPs (e.g., THOR, LIN28B-AS1, RPSAP52, GHET1, circNDUFB2, Let-7)^121–125.

Studies on the oncogenic functions of other m⁶A-binding proteins (such as hnRNPC, hnRNPA2B1) are just starting, and emerging evidence also suggests that they might play important roles in mediating the effect of m⁶A modifications in human cancers^126,127, which is worth elucidating in future studies. Taken together, the vast majority of the m⁶A modifiers predominantly play oncogenic functions in caner, providing potentially valuable targets for cancer treatment (Figure 2a).

Tumour-suppressor roles of m⁶A modifiers in cancers

Several m⁶A modifiers also function as tumour suppressors in multiple types of cancers (see Figure 2b). For instance, METTL14 plays a tumour-suppressor role in multiple solid tumours, including HCC, endometrial cancer, clear cell renal cell carcinoma (ccRCC), skin tumour, bladder cancer, CRC, and GC^54,72. METTL14 suppressed HCC metastasis through regulating DGCR8-mediated pri-miR-126 processing and RNA degradation of EGFR and USP48 in an m⁶A-dependent manner^128,129. METTL14 deficiency leads to proliferation and tumourigenicity of endometrial cancer cells, accompanied by a substantial decrease of global m⁶A level, by activating the AKT pathway through suppressing PHLPP2 expression and increasing expression of mTORC2¹³⁰. Remarkable decreases of METTL14 were also observed in renal cell carcinoma (RCC), associated with metastasis in RCC patients¹³¹. Downregulated METTL14 in RCC reinforces EMT process and distal lung metastasis, by reducing m⁶A on BPTF and PTEN mRNAs and NEAT1 lncRNA ^131–133. In CRC, METTL14 suppresses tumour dissemination via m⁶A-mediated degradation of SOX4 and ARRDC4 mRNAs and XIST lncRNA^134–136. In bladder cancer, METTL14 inhibits the proliferation, self-renewal, metastasis, and tumour initiating capacity of bladder tumour initiating cells (TICs) through regulating m⁶A and stability of NOTCH1¹³⁷. In sum, the tumour-suppressor properties of METTL14 are associated with its m⁶A writer’s function.

Figure 2b. Tumour suppressive functions of m⁶A modifiers in cancer.

Tumour-suppressing m⁶A modifiers and their representative downstream targets, including coding and non-coding RNAs, are illustrated. Positive targets of m⁶A modifiers are in red, while negative targets of m⁶A modifiers are in blue. METTL3, methyltransferase 3; METTL14, methyltransferase14; WTAP, WT1 associated protein; METTL16, methyltransferase16; FTO, FTO alpha-ketoglutarate dependent dioxygenase; ALKBH5, alkB homolog 5; YTHDF2, YTH N6-Methyladenosine RNA Binding Protein 2; YTHDC1, YTH domain-containing protein 1.

The demethylase ALKBH5 is also downregulated and displays tumour-suppressor roles in pancreatic cancer, LUAD, HCC, bladder cancer, and osteosarcoma, and its downregulation is associated with poor prognosis in such cancer patients^138–144. In pancreatic cancer, ALKBH5 prevents tumourigenesis by removing m⁶A and promoting expression of PERP and WIF1 mRNAs and KCNK15-AS1 lncRNA^138,145,146. In lung cancer, ALKBH5 suppresses growth and 3D-spheroid generation of cancer cells and intra-pulmonary tumour formation in mice by suppressing glycolysis and the Hippo-YAP pathway, via the demethylation of m⁶A on critical mRNAs such as ENO1 and YAP^140,141. In HCC, bladder cancer, and osteosarcoma, ALKBH5 inhibits tumour progression via reducing m⁶A in critical target transcripts (e.g., LYPD1, CK2α, pre-miR-181b and YAP)^142–144.

FTO also plays tumour-suppressor roles in HCC, ovarian cancer, and papillary thyroid cancer^147–150. Post-transcriptional regulation by SIRT1-mediated ubiquitination and the circGPR137B-miR-4739 axis has been shown to cause reduced expression of FTO in human HCC^147,148. FTO expression is suppressed in ovarian CSCs, and FTO augments the second messenger cyclic adenosine 3,5-monophosphate (cAMP) signalling and suppresses stemness features of ovarian cancer cells by reducing m⁶A on the mRNAs of two phosphodiesterase genes PDE1C and PDE4B¹⁴⁹. In papillary thyroid cancer, FTO inhibits glycolysis and cell growth via decreasing stability of APOE mRNA in an IGF2BP2-dependent manner¹⁵⁰.

In addition, m⁶A reader proteins YTHDF2, YTHDC1, and YTHDC2 have been proposed to act as tumour suppressors in certain types of cancer. YTHDF2 could be specifically induced by hypoxia in human HCC, in which it was expressed at a low level, accompanied by a hyper-m⁶A and high-mRNA profile¹⁵¹. Overexpression of YTHDF2 suppressed cell proliferation, tumour growth, and activation of MEK and ERK in HCC cells¹⁵². Conversely, deletion of YTHDF2 in mouse hepatocytes provoked tumour growth, vasculature remodelling, and metastatic progression through inducing RNA decay of m⁶A-containing IL11 and SERPINE2 mRNAs¹⁵¹. Smoking-induced downregulation of YTHDC2 is associated with tumourigenesis of lung cancer¹⁵³. In human pancreatic cells, YTHDC1 mediates the maturation of miR-30d to attenuate aerobic glycolysis¹⁵⁴.

The complexity of m⁶A epitranscriptomics in cancers

As aforementioned, RNA m⁶A modifiers are often dysregulated and function as oncogenes or tumour suppressors in cancers, in most of cases through regulating m⁶A modification and expression of different target sets, adding the complexity to cancer epigenetics (Figure 2). The complexity of m⁶A epitanscriptomics in cancers exhibits in several scenario: (1) certain modifiers (e.g., METTL14 and ALKBH5) play oncogenic roles in some cancer types, while functioning as tumor suppressor in the others. Thus, the pathological role of a given m⁶A modifier is cell context-dependent and it may play distinct roles in different types of cancers; (2) both writers and erasers play similar pathological roles in the same cancer types. For example, METTL3/14/16 and FTO/ALKBH5 are all upregulated and play oncogenic functions in AML. Similar phenomena have also been observed for DNA methylation writers and erasers (e.g., both DNMT3A and TET2 function as tumor suppressors in AML)^155,156. Thus, it appears that either increased or decreased epigenetic modifications could contribute to tumorigenesis. The changes of m⁶A levels on certain critical target transcripts, rather than the changes of global cellular m⁶A level, may be critical for tumorigenesis^22,53,54,86. Both m⁶A writers and erasers play similar pathological roles in some cancer types usually through targeting distinct set of critical downstream targets and driving different signalling pathways; and (3) nevertheless, both writers and erasers could also target the same transcripts and cause similar patterns of dysregulation of the targets. For instance, both METTL3/14 and FTO could target of MYC transcript and promote its expression as m⁶A writer and eraser, respectively^65,70,90. According to our unpublished data (Deng et al. unpublished) and as discussed by us previously⁵³, overexpression of FTO can preferentially remove m⁶A from the first two MYC exons and avoid YTHDF2 (also preferentially binding here)-mediated RNA decay and thereby promotes MYC expression; overexpression of METTL3/14 could preferentially increase m⁶A on the last MYC exon and thus enhance MYC stability and translation via an IGF2BP (preferentially binding to the last exon²¹)-dependent mechanism. Consistently, both YTHDF2 and IGF2BP2 are also overexpressed and play oncogenic roles in AML^22,99. So, it is apparent that m⁶A writers, erasers and readers work together to orchestrate epitranscriptome regulation in cancer. It is important to systematically reveal the molecular mechanisms by which different m⁶A modifiers orchestrate epitranscriptome regulation in a given type (or sub-type) of cancer, especially at the single cell level.

Roles of m⁶A modifiers in cancer metabolism, tumour microenvironment and tumour immunity

Cancer cells undergo extensive metabolic reprogramming and exhibit distinctive metabolic features. Considering that cancer cells reside in a tumour microenvironment (TME) with abnormal vasculature and limited availability of nutrients, such metabolic reprogramming has profound impacts not only on the growth of tumour itself, but also on non-tumour cells in the TME. Specifically, rapidly proliferating cancer cells hijack various metabolic pathways to constitutively adopt an anabolic metabolic phenotype and deprive tumour-infiltrating immune cells of essential nutrients to create an immunosuppressive milieu. RNA m⁶A and the modifiers play multifaceted roles in modulating cancer metabolism and shaping the tumour microenvironment (Figure 3).

Figure 3. Regulation of cancer metabolism and tumour immunity by m⁶A in the tumour microenvironment.

(a) Cancer-intrinsic m⁶A methylation plays a critical role in modulating metabolism of 3 key nutrients: glucose, amino acid, and lipid. Cancer cells demonstrate aerobic glycolysis and primarily metabolize glucose to lactate as the end product. The m⁶A modifiers involved in regulation of glucose metabolism are listed in red. m⁶A also guides the metabolism of vital amino acids that are indispensable for tumour growth, and respectively implicated modifiers are listed for the metabolism of serine (yellow), glutamine (purple), and branched-chain amino acids (BCAAs) (green). Those modifiers engaged in cancer lipid metabolism are listed in brown. (b) Anti-tumour immunity is controlled by both cancer-intrinsic and immune cell-intrinsic m⁶A modifiers. Cancer-intrinsic modifiers (dark red) maintain PD-L1 expression and regulate functions of certain tumour-infiltrating immune cells. Immune cell-intrinsic modifiers regulating the functions of natural killer (NK) cell (blue), dendritic cell (DC) (orange), macrophage (purple), T cell (green), and regulatory T cell (Treg) (turquoise) are listed under the corresponding cells. METTL, methyltransferase; FTO, FTO alpha-ketoglutarate dependent dioxygenase; ALKBH5, alkB homolog 5, RNA demethylase; IGF2BP, insulin-like growth factor 2 mRNA-binding protein; YTHDF1/2, YTH N6-methyladenosine RNA binding protein 1/2; 3PG, 3-phosphoglycerate; 3PHP, 3-phosphohydroxypyruvate; 3PS, 3-phosphoserine; PHGDH, phosphoglycerate dehydrogenase; PSAT1, phosphoserine aminotransferase 1; SLC1A5, solute carrier family 1 member 5; BCAT1/2, branched chain amino acid transaminase 1/2; BCKA, branched-chain keto acids.

One hallmark of cancer metabolism is that cancer cells preferably metabolize glucose through glycolysis and subsequent lactate fermentation instead of oxidative phosphorylation (OXPHOS) even with an ample supply of oxygen, a phenomenon termed the Warburg effect or aerobic glycolysis. METTL3 was reported to methylate and stabilize SLC2A1/HK2 mRNAs in an IGF2BP-dependent manner in colorectal cancer¹⁵⁷ and activate the HDGF/SLC2A4/ENO2 pathway in GC⁵⁹ to promote glycolysis. In esophageal cancer, METTL3 suppresses expression of APC via YTHDF2-mediated decay and triggers the downstream β-catenin pathway to increase intracellular MYC/PKM2 levels to augment glycolysis¹⁵⁸. Additionally, the METTL3/YTHDF1 axis elevates HK2 expression to boost glycolysis in cervical cancer¹⁵⁹. In contrast, effects of METTL14 on tumour growth and glycolytic activity are inconsistent and context dependent^131,160,161. FTO supports aerobic glycolysis in AML by promoting expression of PFKP and LDHB⁸⁹ and in solid tumours by enhancing expression of c-Jun, JunB, and C/EBPβ¹⁶². ALKBH5 was reported to promote glycolysis in breast cancer but shift cancer metabolism towards OXPHOS in liver and cervical cancers^163,164.

Apart from glucose, amino acids serve as alternative fuels for tumour growth, rendering many cancer cells addicted to certain amino acids such as glutamine and serine. We reported recently that METTL3- and METTL14-mediated m⁶A promotes glutamine metabolism in an IGF2BP2-dependent manner by upregulating expression of MYC, GPT2, and SLC1A5 in AML, which in turn promotes AML development/maintenance and LSC/LIC self-renewal²². FTO deficiency in VHL^−/− RCC cells causes hypermethylation and consequential downregulation of the glutamine transporter SLC1A5, dampening exogenous glutamine uptake and eventually compromising cancer cell growth and survival¹⁶⁵. Transcripts of two enzymes implicated in serine biosynthesis, PHGDH and PSAT1, are stabilized by the ALKBH5/m⁶A/YTHDF2 pathway, which contributes to the leukemogenic activity of ALKBH5 in AML⁹⁷. Lately, reprogramming of branched chain amino acid (BCAA) metabolism by METTL16 is also unveiled to be critical for AML development/maintenance and LSC self-renewal⁸⁴.

Cancer cells also have a higher demand for lipids to sustain their metabolism and proliferation, but lipid availability is often limited in the TME. Thus, de novo lipid synthesis pathways are frequently reactivated in cancer. Emerging evidence supports the critical roles of m⁶A modifiers in rewiring cancer lipid metabolism. For instance, METTL3 and METTL14 are assumed to provoke lipid metabolism via enhancing their respective lncRNA targets LINC00958 and lncDBET, although direct corroboration of the regulation of lipid metabolism by the METTL14/lncDBET pathway is still missing^166,167. A potential link between METTL5 and lipid biosynthesis has also been demonstrated in human cervical cancer cells¹⁶⁸. Nevertheless, the relationship between FTO (the first identified obesity-associated gene) and cancer lipid metabolism remains largely unexplored, despite some recent efforts in delineating the function of FTO in physiological adipogenesis¹⁶⁹.

Genetic depletion of FTO in cancer cells restrains tumour glycolytic flux and releases the metabolic brake for CD8⁺ T cell activation, substantiating that tumour-intrinsic m⁶A machinery fine-tunes cancer cell metabolism to indirectly affect immune cell metabolism and anti-tumour immunity¹⁶² (Figure 3). We found that FTO can up-regulate expression of PDL1/2 and especially LILRB4 in AML cells and PD1 in T cells, which contributes to tumour immune evasion⁸⁷. FTO suppresses the YTHDF2-mediated degradation of LILRB4 to increase its expression. Similarly, ALKBH5 assists cancer cells to create an immunosuppressive TME by recruiting suppressive immune cells in melanoma¹⁷⁰, colorectal cancer¹⁷⁰, and glioblastoma multiforme⁹⁸; ALKBH5 in intrahepatic cholangiocarcinoma additionally maintains PD-L1 level in a YTHDF2-dependent manner to allow tumour immune escape¹⁷¹. On the other hand, the IGF2BP1 reader protein promotes PD-L1 expression in HCC cells and restricts tumour-infiltration of anti-tumour immune cells¹⁷², and the YTHDF1 reader protein enhances GC cell proliferation and blocks dendritic cell-mediated anti-tumour response¹⁷³. The fact that both m⁶A erasers (i.e., FTO and ALKBH5) and the stabilizing reader (i.e., IGF2BP1) could positively regulate the PD-L1 level in various types of cancer reiterates that distinct expression profiles of readers across different cellular contexts dictate dissimilar functions of m⁶A in modulating tumor immunity. Moreover, the difference in expression of m⁶A modifiers between tumor cells and stromal/immune cells may offer a therapeutic window to specifically target m⁶A-related pathways in tumor cells. The factors determining how m⁶A on the same transcript may be differentially recognized by distinct readers under different contexts warrant further investigation. It is possible that different readers are responsible for recognizing m⁶A marks on distinct regions of the same transcript, and the primary reader for that transcript is determined by the reader expression profiles in the specific cancer type.

Furthermore, the m⁶A machinery in immune cells (e.g., T cells) shapes TME by directly modulating their function and tumour infiltration. METTL3 governs physiological T cell homeostasis and development via the m⁶A/SOCS/IL-7/STAT5 signalling¹⁷⁴ and sustains immunosuppressive activity of regulatory T cells (Tregs) by methylating the same set of targets, SOCS mRNAs¹⁷⁵. METTL3 also guides differentiation of T follicular helper (Tfh) cells, a subset of T cells with B cell helper function, implying the concomitant engagement of METTL3 in programming B cell-mediated immunity¹⁷⁶. Loss of ALKBH5, but not FTO, reduces IFN-γ/CXCL2 abundance to incapacitate CD4⁺ T cells during autoimmunity, but whether this is also true in the scenario of anti-tumour immunity has yet to be elucidated¹⁷⁷.

Additional populations of immune cells that are essential components in TME are also under the control of the m⁶A machinery. METTL3 promotes tumouricidal activity of tumour-infiltrating NK cells¹⁷⁸ and facilitates maturation and function of dendritic cells (DCs) under physiological conditions to allow effective priming of T cells¹⁷⁹. Both exacerbation⁶³ and amelioration^180,181 of macrophage-related immunosuppressive TME by METTL3 have been reported, either by skewing macrophage polarization or via altering macrophage activation. METTL14 in tumour-infiltrating macrophages is believed to reduce EBI3 expression to impair CD8⁺ T cell infiltration and cytotoxicity through a macrophage-T cell crosstalk mechanism¹⁸². YTHDF1 in DCs is shown to diminish their capability to present tumour antigens to CD8⁺ T cells, impeding the cross-priming and anti-tumour response of CD8⁺ T cells¹⁸³. YTHDF2, on the other hand, contributes to the feedback control of DC migration during inflammation¹⁸⁴, and is indispensable for the survival/proliferation of NK cells and their anti-tumour function¹⁸⁵.

Mechanisms underlying the dysregulation of m⁶A in cancer

Referring to Feinberg’s classification system in cancer epigenome¹⁸⁶, we divide m⁶A-related genes into three categories: m⁶A modifiers (or regulators) that are enzymes directly adding or removing m⁶A on RNAs or reader proteins interpretating m⁶A in cancer; m⁶A mediators that are oncogenes or tumour suppressors regulated by m⁶A modifiers; m⁶A modulators that are upstream regulators of m⁶A modifiers in cancer. Genetic mutations and environmental factors usually contribute to the dysregulation of m⁶A modulators, modifiers and mediators in cancer, in which several underlying mechanisms have been proposed (Figure 4): (1) Genetic mutations of m⁶A modulators that are usually tumour driver genes in cancer lead to reprograming of m⁶A through regulating m⁶A modifiers. For instance, MLL-fusions lead to the increased expression of FTO, METTL14 and IGF2BP2 in MLL-rearranged AML^22,70,88, and NPM1 mutants and FLT3-ITD can also upregulate expression of FTO in AML^88,187; (2) Genetic mutations of m⁶A modifiers lead to abnormally stimulation or suppression of its activity. Crystal structure of the METTL3-METTL14 complex revealed that the R298 residue lies in proximity of the putative RNA-binding groove and is critical for target recognition of MTC¹⁸⁸. A hotspot R298P mutation of METTL14 is found in 70% of endometrial tumour, which significantly impairs the methylation activity of MTC and activates downstream AKT pathway to promote cancer growth¹³⁰; (3) Genetic mutation of m⁶A mediators on motif sequence that results in gain of de novo m⁶A or loss of m⁶A sites. Analysis of m⁶A sites inferred from disease-associated genetic mutations revealed that more than 5000 mutations that can destroy m⁶A forming motifs are associated with pathogenesis¹⁸⁹. In contrast, guanine to adenosine G-to-A mutations in transcripts of tumour suppressors TP53 and ANKLE2 lead to gain of m⁶A in these transcripts, and therefor increase protein expression of TP53-R273H and ANKLE1 in CRC^190,191; (4) Multiple layers of regulation, including transcription regulation by transcription factors (e.g., SPI1-METTL14⁷⁰, CEBPA-FTO⁹⁰ and HIF1A-YTHDF1¹⁹²) and DNA/histone modifications (e.g., KDM4C-H3K9me3-ALKBH5⁹⁷ and KDM5C-H3K4me3-METTL14¹³⁴), post-transcription regulation by ncRNAs (e.g., lncRNAs ARHGAP5-AS1-METTL3¹⁹³ and FOXM1-AS-ALKBH5⁹⁵, and miRNAs let-7g-METTL3¹⁹⁴ and miR-145-YTHDF2¹⁹⁵) and post-translation regulation by post-translational modifications (PTMs) (e.g., phosphorylation of MTCs¹⁹⁶, SUMOylation of METTL3, FTO and ALKBH5^64,147,197, and lactylation of METTL3⁶³), act alone or cooperatively to manipulate the expression or activity of m⁶A modifiers and thus affect m⁶A signalling; (5) m⁶A modulators transduce signals from environment factors, such as mutagens, infectious virus, oxidative stress and ligands, to alter the m⁶A pattern in cancer cells^73,198,199; and (6) Epigenetic metabolites such as methyl donor SAM and FTO inhibitor R-2-hydroxyglutarat (R-2HG), directly participate in the methylation and demethylation processes. Recently, it was reported that dietary methionine restriction reduced tumour growth and enhanced anti-tumour immunity by reducing SAM level and global m⁶A methylation²⁰⁰. Of note, the modulators and epigenetic factors are usually affected by m⁶A modification itself to comprise a feedback loop. Collectively, m⁶A regulation in cancer represents an additional layer of complexity in carcinogenesis, shedding light for development of new targets and new strategies to combat cancer.

Figure 4. Mechanisms underlying the dysregulation of m⁶A in cancer.

Genetic mutations and environmental factors act in different manners to affect three classes of m⁶A-related genes, m⁶A modulators, m⁶A modifiers and m⁶A mediators, and finally influence the abundance or function of m⁶A in carcinogenesis.

The clinical implications of m⁶A modification in cancer: As biomarkers

Biomarkers are widely used for early detection/diagnosis, assessment of prognosis and therapy response, which assist precision medicine to improve patient outcomes. RNA m⁶A and its machinery could serve as biomarkers for clinical applications given their critical roles in every stage of cancer. Here, we summarize and discuss recent advances on the identification of m⁶A-related biomarkers (see Figure 5).

Figure 5. Perspective schematics of m⁶A modification as assisting diagnostic, prognostic and drug predictive biomarkers in cancer patients.

m⁶A profile/signature as diagnostic markers for early detection and screening of cancer patients; as prognostic markers for assessment of patient prognosis (overall survival); as drug predictive marker to selection of cancer patients benefit from radio/chemo therapy, targeted therapy and immunotherapy. m6A modifiers in therapeutic resistance are listed. m6A modifiers promoting cancer treatment sensitivity are marked with green plus signs, indicating patients with certain level of indicated m⁶A modifier genes may benefit from indicated therapies; while m6A modifiers antagonizing cancer treatment are marked with red crosses, indicating patients with aberrant expressed indicated m⁶A modifier genes has low responds on indicated therapies. m⁶A profile/signature prediction model could be established by the global abundance of m⁶A, the gene mutation and expression signature of m⁶A associated modifiers (writers, erasers and readers), as well as gene signatures of a set of m⁶A modified non-coding RNAs, such as LncRNAs and miRNAs. METTL, methyltransferase; FTO, FTO alpha-ketoglutarate dependent dioxygenase; ALKBH5, alkB homolog 5, RNA demethylase. 5FU, 5 Fluorouracil.

Prognostic biomarker predicts patient outcomes (e.g., overall survival) regardless of treatment. RNA m⁶A regulatory enzymes are considered as promising prognostic biomarkers, as their aberrant expression levels and corresponding changes in global m⁶A level are often associated with prognosis. For example, high expression levels of METTL3 and METTL14 have been reported to be independent prognostic biomarkers for overall survival in some types of cancer patients. METTL3 is overexpressed in GC⁵⁹and CRC²⁰¹ patients, in which a higher METTL3 expression predicts a poorer survival. Oppositely, METTL14 is downregulated in CRC and rectal cancer patients, and a lower level of METTL14 predicts a shorter survival^134,202. METTL3 level in GC tissues was significantly higher in advanced-stage than in early stage cancers²⁰³. Furthermore, METTL3 level combined with the tumour, node, metastasis (TNM) stage enhances the prediction ability in GC patients⁵⁹. Besides assessing the prognostic impact of individual m⁶A modifiers, models established by integrative analysis of expression profiles with multiple m⁶A modifiers have been established in different cancer types such as HCC²⁰⁴, LUAD²⁰⁵, glioma²⁰⁶, GC²⁰⁷, etc, based on the public databases (e.g., TCGA, GEO, ICGC, UHB CC, TMA and GTEX). Besides m⁶A modifiers, some key m⁶A mediators such as some lncRNAs are also identified as m⁶A related prognostic markers in cancers such as colon cancer²⁰⁸, glioma²⁰⁹ and breast cancer²¹⁰.

Ever since liquid biopsy (peripheral blood or cell-free DNA/RNA) as a non-invasive way enters the clinical application, m⁶A profiling/signatures acquired by advanced m⁶A-seq could potentially serve as diagnostic markers for cancer early detection and as real-time monitors of metastasis/relapse. For instance, Ge et al analyzed the global m⁶A level in total RNA from peripheral blood samples of 100 GC patients, and found the level of global m⁶A are significantly higher in GC patients than in patients with benign gastric disease or healthy controls, suggesting that m⁶A in blood samples could be a promising non-invasive early diagnostic biomarkers in GC²¹¹. By analysis of 14,965 serum samples with 12 cancer types, another study found that expression signatures of m⁶A-modified miRNAs in cell-free RNA samples could serve as cancer screening tool and 92% of patients with a certain cancer type could be identified with such signatures, showing a good sensitivity in cancer diagnosis²¹².

Predictive markers are helpful in separating patients into groups that are sensitive or resistant to a therapy. Therapeutic resistance has always been a major challenge in cancer therapy. Recent studies reveal the participation of m⁶A in the development of therapy resistance, which are summarized below and in Figure 5, highlighting the potential of m⁶A-associated molecules as predictive markers for therapy resistance.

Radio- and chemo-therapy are the most widely used cancer treatments that utilize high-energy radiation or cytotoxic chemicals to kill cancer cells. Evidence indicates m⁶A modifications on certain key transcripts confer radio/chemo-resistance in cancer cells via promoting cell growth and CSC stemness or evading cell death. For instance, METTL3-mediated m⁶A modification on H2AX, SOX2 and circCUX1 contributes to the resistance of radiotherapy^213–215. FTO and ALKBH5 also increase radio-resistance in cervical squamous cell carcinoma (CSCC) and GBM by regulating excision repair and homologous recombination, respectively^216,217.

Acquired resistance of cancer cells to chemotherapeutic drugs is also associated with m⁶A modification. METTL3-mediated hypermethylation of LDHA and other transcripts participates in the induction of 5 Fluorouracil (5FU) resistance in CRC²¹⁸. In contrast, downregulation of FTO and ALKBH5 is observed in cisplatin resistant bladder cancer cells and protects bladder cancer cells from cisplatin-induced cytotoxicity^143,219. Due to the importance of m⁶A modifiers in radio/chemo-resistance, many predictive models based on gene signature of m⁶A modifiers were established. For example, a chemotherapy benefit prediction system (m⁶A score) was established based on expression signatures of seven modifiers (ZCCHC4, IGF2BP3, ALKBH5, YTHDF3, METTL5, G3BP1, and RBMX), which shows a better prediction accuracy than other pathological factors in small cell lung cancer (SCLC); among patients who underwent chemotherapy, patients with high m⁶A scores had significantly shorter overall survival²²⁰.

Albeit targeted therapy can specifically “target” molecules that are essential for tumourigenesis, the rapidly acquired resistance is still a major obstacle for the cure of cancer. Imatinib and nilotinib are both tyrosine kinase inhibitors (TKIs) against the BCR-ABL oncofusion protein for treating chronic myelogenous leukemia (CML). Characterization of m⁶A-containing transcripts in nilotinib- and imatinib-resistant leukemia cells revealed that FTO-induced reprograming of m⁶A modification facilitates TKI tolerance and promotes cell growth, linking dynamic m⁶A configurations with acquired TKI resistance²²¹. Gefitinib and Osimertinib are another class of TKIs frequently used to treat EGFR-mutant LUAD, whose sensitivity in LUAD cells is also associated with alteration of m⁶A modification in critical transcripts such as circASK1 and Let-7^222,223. Another study reported that METTL3 induces BRAF kinase inhibitor PLX4032 resistance in melanoma cells by promoting m⁶A-dependent EGFR translation²²⁴. Additional research revealed that RNA m⁶A methylation confers sorafenib resistance in liver cancer^225–227. The m⁶A hypermethylation of mRNA FZD10 contributes to PARPi resistance by upregulating the canonical Wnt/β-catenin signalling in BRCA1/2-mutated ovarian cancer cells²²⁸. METTL16 can antagonize MRE11-mediated homologous recombination repair and confer synthetic lethality to PARPi treatment in pancreatic ductal adenocarcinoma (PDAC)⁸⁰.

Selection of patients who can benefit from immunotherapy is urgently needed. As aforementioned, due to the importance of RNA m⁶A modification in regulating immune response, comprehensive evaluation of the gene signatures associated with m⁶A has been conducted in various cancers leading to the development of risk scoring systems to predict immunotherapy response^229–231. For example, expression signatures of 8 m⁶A modifier genes were integrated for construction of an m⁶A regulator prognostic risk score, which is able to predict survival and immunotherapy response in ccRCC²³². Yang et al. found that FTO regulates response of melanoma cells to anti-PD1 blockade through demethylating m⁶A on PD-1, CXCR4, and SOX10⁹³. Similarly, genetic knockout or pharmacologically inhibition of ALKBH5 enhances efficacy of immunotherapy and prolongs mice survival¹⁷⁰. In addition, Wang et al. showed that depletion of Mettl3 and Mettl14 enhanced response to anti-PD-1 treatment in CRC and melanoma¹⁷⁰.

Currently studies of m⁶A-related biomarkers are booming and show promising prediction ability. However, due to the limitations of the routinely used m⁶A-seq methods, no large-scale transcriptome-wide m⁶A profiles have been generated from clinical samples yet. In addition, reliable biomarkers need to be established and validated on large cohorts of clinical samples classified with multiple clinical risk factors (cancer stage, subtypes, pathological factors, etc.).

The clinical implications of m⁶A modification in cancer: As therapeutic targets

There are three major factors make oncogenic m⁶A modifiers as ideal therapeutic targets. One factor is that many oncogenic m⁶A modifiers are especially overexpressed in CSCs, which are responsible for cancer metastasis/recurrence and immune evasion, and thereby pharmacologically targeting them could eradicate CSCs and thereby cure cancers. The second factor is that many oncogenic m⁶A modifiers are lowly expressed in adult normal tissues/cells and thus pharmacologically targeting them could be safe. The third factor is that many oncogenic m⁶A modifiers play critical roles in mediating the resistance of cancer cells to various types of therapeutics, and thus pharmacologically targeting them can synergize with other therapeutics (e.g., standard chemotherapy, irradiation therapy, other targeted therapeutics and immunotherapy) to cure cancer. A number of inhibitors targeting the m⁶A machinery have been developed by virtual screening, rational design and structure-activity relationship study, some of which show great potency, selectivity and efficacy. Due to limited space of our review, we summarized the developed inhibitors that have proof of concept evidence with preclinical data in cancer treatment (Figure 6).

Figure 6. Small molecule inhibitors targeting m⁶A modifiers.

Small molecule inhibitors targeting FTO, METTL3 and IGF2BP2 are listed by timeline with chemical structures, half-maximal inhibitory concentration (IC₅₀) of enzymatic activity, and validated cancer types. FTOi, FTO inhibitors; METTL3i, METTL3 inhibitors; IGF2BP2, IGF2BP2 inhibitors. BC, Breast cancer; AML, Acute myeloid leukemia; GBM, Glioblastoma; PC, Pancreatic cancer; CC, Colon cancer; MN, Melanoma; GC, Gastric cancer; T-ALL, T-cell acute lymphoblastic leukemia.

Inhibitors targeting m⁶A “erasers”

A series of small-molecule inhibitors targeting FTO have been developed, which show promising therapeutic efficacy in treating multiple types of cancers. Since the crystal structure of FTO was reported in 2010²³³, researchers have been inspired to screen and design inhibitors based on its substrate binding sites²³⁴. By structure-based virtual screening, Rhein, a natural compound was identified as the first inhibitor of FTO in 2012²³⁵, which increases cellular m⁶A in mRNA, but with little selectivity on FTO over ALKBH5. By using high-throughput fluorescence polarization assay, Yang`s group identified meclofenamic acid (MA) and MA2 (ethyl ester form of MA), which selectively inhibits FTO<sup>236</sup> over ALKBH5, and MA2 was later reported to inhibit the growth and self-renewal of glioblastoma stem cells (GSCs)<sup>237</sup>. Additionally, by targeting FTO, Rhein and MA increased the sensitivity of TKI treatment in leukemia cells<sup>221</sup>, and MA2 enhances the effect of chemotherapy drug temozolomide on suppression of proliferation of glioma cells<sup>238</sup>. With the superior selectivity for FTO, MA was used as a basic structure for pursuing more potent inhibitors. Yang`s group further developed a series of new FTO inhibitors (FB23 and FB23–2⁹¹, Dac51¹⁶²) based on MA by structure-based study and chemical synthesis to improve the anti-FTO enzymatic activity and antitumour efficacy. FB23 shows a 140-fold higher efficacy than MA on inhibition of FTO enzymatic activity, while in comparison to FB23, FB23–2 inhibits AML cell proliferation with a better cell permeability⁹¹, which later was found an inhibition effect in pancreatic cancer²³⁹. Dac51 was reported to exhibit promising antitumour efficacy in melanoma through reprograming TME¹⁶². Similarly, another group reported FTO-04, as a derivative of FB-23, FTO-04 shows inhibition on the survival/proliferation of GSCs but spares healthy neural cells²⁴⁰. Later on, based on the structure of FTO-04, FTO-43²⁴¹ was reported as a compound of oxetanyl class, shows inhibitory effect on cancer cell proliferation in GBM, AML and GC.

Analogs of α-KG are another type of FTO inhibitors, which competitively locate at the binding pocket of FTO instead of α-KG, and chelating Fe to inhibit the catalytic ability of FTO. Our group reported that R-2HG is an α-KG competitive inhibitor of FTO, exerting an intrinsic and broad anti-tumour activity in multiple cancer types including AML and GBM, and mechanically attenuating cancer cell proliferation and aerobic glycolysis by suppression of FTO mediated m⁶A mRNA signalling pathway involving MYC, CEBPA, PFKP and LDHB^89,90. In addition to the analogs of α-KG, an analog of ascorbic acid, MO-I-500 was also reported to exhibit anti-tumour effect by inhibiting the enzymatic activity of FTO in breast cancer^242,243. Through virtual screening of a compound library composed of 260,000 compounds, followed by validation/mechanistic studies, we identified two highly potent FTO inhibitors (namely CS1 and CS2)⁸⁷. Both CS1 and CS2 exhibit potent therapeutic efficacy in treating various types of cancers such as AML (including relapsed AML, as shown in treating two relapsed AML patient-derived xenotransplantation (PDX) models), GBM, breast cancer and pancreatic cancer⁸⁷. Inhibition of FTO could attenuate LSC self-renewal and overcome immune evasion⁸⁷.

Unlike FTO inhibitors, inhibitors targeting ALKBH5 are less developed, and most of which lack in vivo data. Recently, 20m, as a 1-aryl-1H-pyrazole derivative was developed and showed a global m⁶A increase in cells²⁴⁴. Two ALKBH5 inhibitors were identified by virtual screening of a library containing 144,000 compounds and confirmed enzymatic inhibition effect by ELISA, which shows inhibitory effect on three leukemia cell lines²⁴⁵. Another ALKBH5 inhibitor, MV1035 was identified by 3D proteome-wide screening and showed inhibitory effect on cell survival/proliferation of GBM cells²⁴⁶.

Inhibitors targeting m⁶A writers

The majority of METTL3 inhibitors are SAM-structure related, which competitively bind to METTL3. Two METTL3 inhibitors were developed as derivatives of the adenosine moiety, which are SAM competitive inhibitors of METTL3²⁴⁷. Later on, the same group reported an enantiomeric compound UZH1a, which also screened from an adenine-based library, shows inhibitory effect on the viability of leukemia cells with an IC50 of 9 μM. In 2021, Kouzarides and colleagues reported a highly potent and selective first-in-class catalytic inhibitor of METTL3, namely STM2457, with a crystal structure showing the binding of STM2457 in complex with METTL3-METTL14; STM2457 treatment can significantly suppress the growth/proliferation and promote apoptosis/differentiation of AML cells⁶⁶. STM2457 also exhibited potent therapeutic efficacy in treating AML in vivo and could target key stem cell subpopulations of AML in multiple AML mouse models⁶⁶.

Inhibitors targeting m⁶A readers

BTYNB was identified as an inhibitor targeting IGF2BP1 (or IMP1), showing anti-tumour effect in IMP1-expressing ovarian cancer and melanoma cells but not in IMP1-negative cells²⁴⁸. Very recently, we identified a potent IGF2BP2 inhibitor, namely CWI1–2, which can selectively inhibit the binding of IGF2BP2 with its m⁶A-modified target RNAs; CWI1–2 also showed a promise anti-leukemia efficacy that greatly suppressed self-renewal of LSCs/LICs and significantly inhibited AML progression and prolonged survival in the treated AML mice²². Another IGF2BP2 inhibitor, JX5 shows anti-leukemia efficacy in T-ALL²⁴⁹.

Conclusions and perspectives

It is clear now that the m⁶A machinery plays essential roles in all types of bioprocesses and in tumourigenesis. The m⁶A modifiers, including writers, erasers and readers, are often dysregulated in various types of cancers, in which they play critical roles in cancer initiation, progression, metastasis, maintenance, drug response/resistance and immune response, as well as in self-renewal of CSCs and cancer metabolism and tumour microenvironment reprogramming, etc. Moreover, many oncogenic m⁶A modifiers appear to be promising therapeutic targets for cancer treatment. Nevertheless, although the roles and underlying molecular mechanisms of the m⁶A machinery in various types of cancers have been extensively studied in recent years, there are still many unanswered questions in this field. For instance, why do m⁶A writers and erasers (as well as readers) play similar oncogenic or tumour-suppressor roles in a same cancer type (e.g., m⁶A writers (METTL3/14/16), erasers (FTO and ALKBH5) and readers (YTHDF2, YTHDC1 and IGF2BP2) are all overexpressed and play oncogenic roles in AML^{22,32,65,70,87,88,90,91,96,97,99–101})? How do these m⁶A modifiers orchestrate the regulation of cancer epitranscriptome in a given cancer type, especially at the single-cell level? Why does a given m⁶A modifier play opposite roles in different types of cancers? How does the m⁶A machinery fine tune the crosstalk amongst cancer cells, immune cells and tumour microenvironment? Systematic studies are warranted to address such questions, which will further substantially advance our understanding of the roles and underlying molecular mechanisms of m⁶A modification in cancer and form a better basis for the development of m⁶A-based therapeutics. In addition, evidence is emerging that the m⁶A machinery also targets chromatin-associated regulatory RNAs (carRNAs) and thereby regulates chromatin state and transcription^250,251. Thus, it is also critical to investigate whether and how m⁶A-modification-mediated regulation of carRNAs plays a role in cancer. Moreover, the crosstalks between m⁶A modifications and DNA/histone modifications or other types of RNA modifications in cancers have yet to be systematically investigated.

As m⁶A writers and erasers could be similarly dysregulated and play similar pathological roles in the same types of cancers, the global m⁶A levels of cancer patient samples might not be useful biomarkers. Thus, specific m⁶A signatures from individual RNA transcripts and loci would be more valuable biomarkers. However, due to the limitations (e.g., a lot of RNA material required, no single-base resolution, lacking quantitative information for m⁶A peaks) of currently widely used conventional m⁶A technologies, by far no single-base-resolution m⁶A profiles from primary cancer patient specimens have been generated yet. Now, with the most advanced m⁶A-seq methods (e.g., m⁶A-SAC-seq³⁵, GLORI³⁶ and eTAM-seq³⁷) available, generating quantitative transcriptome-wide m⁶A maps with single-base resolution from clinical samples is feasible. With such m⁶A profile data available, we can identify m⁶A signatures from specific mRNA loci as biomarkers for cancer early diagnosis, classification, prognosis prediction and risk stratification. Additionally, besides mRNAs, noncoding RNAs such as carRNAs, lncRNAs and miRNAs are also regulated by m⁶A. Thus, development of single-base-resolution m⁶A-seq methods for mapping m⁶A in such non-coding RNAs in primary patient samples is also required, which will lead to the identification of novel class of biomarkers. The development of advanced m⁶A-seq should enable us sequence primary patient samples from large cohorts, which can accurately provide transcriptome-wide quantitative information of m⁶A profiles at single-base resolution with limited RNA input. Meanwhile, coupling m⁶A-seq with single-cell sequencing and spatial RNA sequencing technologies will allow us to reveal the complexity and spatial location of m⁶A modifications in disparate cell subpopulations. Moreover, if m⁶A-seq technologies with very high resolution and accuracy and much less sample required are available in the future, mapping m⁶A modification in circulating cell-free RNAs from primary patients could also be feasible, and the generated m⁶A signature-based biomarkers could serve as non-invasive tools for cancer early diagnosis, classification, prognosis, and therapeutics selections, as well as monitoring drug response.

Despite several proof-of-concept studies showing that pharmacological inhibition of dysregulated m⁶A modifiers represents an effective novel therapeutic strategy, currently no inhibitors targeting dysregulated m⁶A modifiers enter clinical trials yet. It is understandable that it takes time to develop highly effective and selective inhibitors with low toxicity. In addition, better understanding the roles and underlying mechanisms of the entire m⁶A machinery in cancers would also help us better select therapeutic targets and develop more effective therapeutics with minimal side effects. Besides development of small-molecule compound inhibitors to inhibit the activity of dysregulated m⁶A modifiers, development of proteolysis targeting chimera (PROTAC)-based inhibitors would be another option that can be used to degrade the targeted proteins, especially for m⁶A reader proteins. As m⁶A writers, erasers and readers could serve as oncogenes in the same cancer types (e.g., METTL3/14/16, FTO/ALKBH5 and YTHDF2/YTHDC1/IGF2BP2 all play oncogenic roles in AML), it would be also interesting to develop and test combinational therapeutics to target multiple oncogenic m⁶A modifiers in a cancer type to obtain the optimal therapeutic outcome. Targeting both oncogenic writer(s) and eraser(s) simultaneously in a cancer type is likely also feasible and promise, since the writer(s) and eraser(s) contribute to tumorigenesis usually through different mechanisms and thus targeting both may lead to synergistic therapeutic effects. The synergistic therapeutic effect of targeting multiple m⁶A modifiers by small-molecule inhibitors should be validated by functional study models with genetic depletion of these modifiers. As the roles of m⁶A modifiers are highly context-dependent, different combinational therapeutics maybe suitable for different cancer types. Thus, the strategy of targeting oncogenic m⁶A modifiers for cancer therapy should be cancer-(sub-)type specific and understanding of the underlying molecular mechanism(s) is important. Moreover, once some locus-specific m⁶A are proven to be essential for cancer, it is also possible to develop CRISPR-based tools to manipulate the specific m⁶A modification(s) for cancer therapy in the future. Finally, given the complex and interactive nature of cancer and tumour microenvironment, we expect that combining targeted therapeutics against dysregulated oncogenic m⁶A modifiers with other therapeutics will likely be needed to cure cancer. In sum, pharmacologically targeting dysregulated m⁶A machinery could be applied alone (by targeting one or multiple oncogenic m⁶A modifiers) and especially in combination with other therapeutics (e.g., chemotherapy, irradiation therapy, immunotherapy, and/or other types of targeted therapy) in the clinic to cure cancer in the future.

Key points:

• Epitranscriptomics represents a regulatory layer of gene expression, by posttranscriptional regulation of RNA with over 170 chemical modifications. m⁶A as a major modification in this layer dynamically involves in every bioprocess.

• m⁶A is reversibly added by writers or removed by erasers, and recognized by readers, deciding the fate of mRNA by RNA splicing, stability, translation, etc., thus tightly controls gene expression.

• Advanced global characterizing and mapping techniques of m⁶A, providing information on locus-specific m⁶A changes is a key to the door of deeply understanding the wide-ranging roles of m⁶A in gene expression.

• Dysregulated m⁶A modifiers (writers, erasers and readers) are widely found in multi-types of cancer, playing essential roles during cancer initiation, progression, metastasis, metabolism, drug resistance, immune evasion, and tumour microenvironment.

• Aberrantly expressed m⁶A modifiers in cancers have been associated with poor prognosis, therapy resistance and immune response, which might be independent biomarkers assisting patients diagnosis, prognosis and selection of benefited therapy.

• Preclinical data indicated that development of small molecule inhibitors targeting oncogenic m⁶A modifiers appears to be promising as either single treatment, or combination with conventional chemotherapy or immunotherapy.