m6A modification: recent advances, anticancer targeted drug discovery and beyond

Abnormal N6-methyladenosine (m6A) modification is closely associated with the occurrence, development, progression and prognosis of cancer, and aberrant m6A regulators have been identified as novel anticancer drug targets. Both traditional medicine-related approaches and modern drug discovery platforms have been used in an attempt to develop m6A-targeted drugs. Here, we provide an update of the latest findings on m6A modification and the critical roles of m6A modification in cancer progression, and we summarize rational sources for the discovery of m6A-targeted anticancer agents from traditional medicines and computer-based chemosynthetic compounds. This review highlights the potential agents targeting m6A modification for cancer treatment and proposes the advantage of artificial intelligence (AI) in the discovery of m6A-targeting anticancer drugs. Three stages of m6A-targeting anticancer drug discovery: traditional medicine-based natural products, modern chemical modification or synthesis, and artificial intelligence (AI)-assisted approaches for the future.


Introduction
More than 170 types of posttranscriptional RNA modifications have been discovered since the 1960s [1]. Among them, N6-methyladenosine (m6A) is the most common modification in eukaryotic messenger RNAs (mRNAs) [1]. Accumulating evidence has demonstrated that m6A modification plays a critical role in regulating RNA processing, splicing, nucleation, translation, and stability, which is crucial for the development of multiple human diseases, such as cancer [2]. m6A modification is a dynamic and reversible process regulated by methylases ("writers") and demethylases ("erasers"). The "writers" consist of a complex including methyltransferase-like 3 (METTL3), METTL14, METTL16, RNA-binding motif protein 15 (RBM15) and its paralogue RBM15B, zinc finger CCCH-type containing 13 (ZC3H13), vir-like m6A methyltransferase-associated protein (VIRMA, also named KIAA1429), and Wilms tumor 1 associated protein (WTAP), which are responsible for transferring methyl groups from the donor S-adenosylmethionine (SAM) to adenine [3][4][5]. Among them, METTL3, METTL14, and WTAP are the core members of this complex [5]. Then, m6A methylation is recognized by binding protein "readers", such as YTH domain family proteins, insulin-like growth factor 2 mRNA-binding proteins (IGF2BPs), and heterogeneous nuclear ribonucleoprotein (HNRNP) family proteins [6]. The reversible process of m6A demethylation is also facilitated by "erasers", such as fat mass and obesity-related protein (FTO) and alkB homologue 5 (ALKBH5) [3]. An increasing body of evidence shows that the "writer" METTL3, "eraser" FTO, and "reader" YTH domain families are involved in various stages of many types of hematomas and solid tumors and could be promising targets for anticancer therapy.

Graphical abstract
Three stages of m6A-targeting anticancer drug discovery: traditional medicine-based natural products, modern chemical modification or synthesis, and artificial intelligence (AI)-assisted approaches for the future.
which are characterized by the integration of omics data, network pharmacology, natural resource-derived chemical databases, computer-aided design, and chemical modifications, have recently been applied to drug discovery [16][17][18][19]. Notably, this approach can effectively avoid the waste of experimental raw materials and laborious efforts, which will help to further evolve the discovery process of m6A-targeting drugs.
Although the physiological roles of m6A modulation in the development and progression of cancer have been wildly studied and some review articles related to m6A have been published [2,3,20], an update on the academic progress of m6A modulation is still necessary since many new related findings have been described recently. Here, we summarize the advances in m6A modulation and the core function of segments of m6A modulators in cancer.
We also summarize the discovery of m6A-targeting anticancer agents from traditional medicine-based natural products and the use of a combination of artificial intelligence (AI) and chemosynthesis for drug exploration.
Dynamic m6A methylation can be reversed by m6A demethylases in nucleus, also called "m6A erasers", including FTO and ALKBH5. The demethylation of m6A modification in nucleic acids by FTO relies on the oxidative function of FTO in an Fe(II)-and α-KGdependent manner [31]. ALKBH5 is another m6A demethylase that regulates the export and metabolism of mRNA by demethylating m6A modification [32]. In addition, members of the Alkb subfamily, such as ALKBH3, are responsible for removing m6A modification on tRNA [33].
The m6A recognition factor is known to regulate mRNA splicing, nuclear export, decay/degradation, translation, and stability and is also called an "m6A readers". The YTH domain-containing family includes YTHDF1, YTHDF2, and YTHDF3, which are cytosolic m6A readers that regulate m6A degradation and translation [34]. YTHDF1 is reported to play a vital role in promoting translation in the cytosol, whereas YTHDF2 regulates mRNA degradation by mediating the lifetime of target transcripts [35,36]. Nonetheless, gene expression, cell death, and survival are associated with the YTHDF2-mediated RNA process [37,38]. YTHDF3 cooperates with YTHDF1 and YTHDF2 to affect the translation and decay of m6A-decorated mRNA, and inversely regulates their function [39]. Another YTH domain-containing family YTHDC1 regulates RNA nuclear export [40] and splicing [41], while YTHDC2 modulates the translation and abundance of target genes [42]. IGF2BPs include IGF2BP1, IGF2BP2, and IGF2BP3, and they primarily promote the stability and translation of target mRNAs [43]. The RNAbinding protein fragile X mental retardation protein (FMRP), encoded by the fragile X mental retardation 1 gene (FMR1), can promote the nuclear export [44] and stability [45] of m6A-modified RNAs. Eukaryotic initiation factor 3 (eIF3) preferentially binds to m6Amakred mRNA rather than nonmethylated RNA and is associated with the process of mRNA translation [46]. HNRNP family contains HNRNPA2B1, HNRNPC, and HNRNPG [6]. HNRNPA2B1 recognizes m6A-marked primary miRNAs (pri-miRNAs) and stimulates miRNA processing [47], while HNRNPC recognizes m6A to induce splicing in mRNA secondary structures [48]. Proline rich coiled-coil 2A (PRRC2A) is a novel m6A reader that can bind to a consensus GGACU motif in the Olig2 coding sequence to stabilize Olig2 mRNA [49].

Aberrant m6A functions and tumor progression
m6A modification is aberrant in various types of cancer and is associated with patient prognosis. The dysregulation of m6A modification also critically regulates malignant behaviors, such as proliferation, metastasis, tumor stemness, and drug resistance [50][51][52][53]. It has been reported that m6A modification regulators can function as either tumor promoters or tumor suppressors in different tumor types. The roles of m6A modification regulators in different types of cancer are summarized in Fig. 2.
m6A modification regulators affect the pathogenesis and progression of tumors via various mechanisms. When a m6A modulator acts as a tumor promotor, it promotes tumor progression by upregulating oncogenes or downregulating tumor suppressor genes. In contrast, when a m6A modulator functions as a tumor suppressor, it inhibits tumor progression by suppressing the expression of oncogenes or upregulating tumor suppressor genes. The effects and mechanisms of m6A modification regulators on tumor progression are summarized in Table 1.

LILRB4
Maintains cancer stem cell self-renewal and contributes to immune evasion [59] ALKBH5

m6A modification regulators downregulate tumor suppressor genes
METTL3 inhibited the expression of SOCS2 via an m6A/YTHDF2 mechanism, which resulted in HCC tumorigenicity and metastasis [111]. METLL3 was downregulated in sorafenib-resistant HCC, and METTL3 inhibition conferred autophagy-related sorafenib resistance in HCC by decreasing the expression of FOXO3 in an m6A/YTHDF1 manner [168]. The METTL3/YTDHF2 axis has been found to induce β-catenin and PCNA upregulation by inhibiting the expression of YPEL5, which enhances tumorigenicity and metastasis in CRC [169]. METTL4 promoted the growth and metastasis of PAAD by decreasing the expression of PERP via m6A modification [152]. METTL14 inhibited skin tumorigenesis by promoting global genome repair through DDB2 m6A methylation and YTHDF1mediated translation [170]. WTAP promoted the posttranscriptional suppression of Ets proto-oncogene 1 (Ets-1), contributing to HCC progression though the HuR/Ets-1/p21/p27 pathway [113]. ALKBH5 promoted the development and maintenance of AML and enhanced the self-renewal of leukemia stem/initiating cells by downregulating TACC3 [60]. YTHDF2 promoted the decay of UBXN1 mRNA via METTL3-mediated m6A modification, which induced the activation of the NF-κB pathway and promoted glioma progression [171].

m6A modification regulators function as tumor suppressors m6A modification regulators downregulate oncogenes
METTL14 inhibited CRC proliferation and metastasis by downregulating the oncogenic lncRNA XIST in an m6A-YTHDF2 manner [51]. The METTL14/YTHDF2 axis decreased SOX4 mRNA expression and inhibited CRC epithelial to mesenchymal transition (EMT) and metastasis [84]. METLL14 was downregulated in RCC and led to the stability of bromodomain PHD finger transcription factor (BPTF) via m6A modification, which promoted metastasis and glycolytic reprogramming in RCC [162]. Downregulation of METTL3 and METTL14 increased the expression of TRIM7 via an m6A-YTHDF2 manner and thus promoted OS tumorigenesis and chemoresistance by the ubiquitination of breast cancer metastasis suppressor 1 (BRMS1) [172]. ALKBH5 impaired the stability of LYPD1 through an m6A-IGF2BP1 mechanism and inhibited the malignant behaviors of HCC [116]. FTO reduced the m6A modification and the stability of PDE1C and PDE4B that promoted cAMP hydrolysis, which inhibited cAMP signaling and suppressed the selfrenewal of ovarian cancer stem cells [144]. YTHDF2 can inhibit HCC proliferation and growth by binding to m6Amodified EGFR mRNA and degrading EGFR in HCC cells [173]. YTDHC2 promoted the decay of SLC7A11 in an m6A-dependent manner and consequently suppressed the tumorigenesis of lung adenocarcinoma by targeting SLC7A11-mediated antioxidant function [174].

Natural products from traditional medicines targeting m6A regulators
Natural products from traditional medicine could be used as a chemical library for m6A-targeting anticancer drug discovery. In this section, we focus on mechanistic insight into natural products derived from traditional medicines targeting m6A regulators and their current findings in cancer treatment.

Phenols
Curcumin (Fig. 3A), a natural phenolic compound, reduced the expression of ALKHB5 and induced higher m6A-modified TNF receptor-associated factor 4 (TRAF4) mRNA that was bound by YTHDF1, leading to enhanced translation of TRAF4 [176]. Resveratrol (Fig. 3A) is a natural polyphenol with antioxidant, antiinflammatory, heart-protective and anticancer properties [177]. Combining resveratrol with curcumin effectively improves growth performance and intestinal mucosal integrity by decreasing m6A as evidenced by enhanced YTHDF2 in the ileum [178].

Flavonoids
Quercetin (Fig. 3B) is commonly used as a dietary supplement and has many biological functions, including anticancer activities. Fluorescence quenching measurements indicated that among the 3 flavonoids (quercetin, apigenin, and naringenin), quercetin showed the strongest binding with FTO through hydrophobic interactions and hydrogen bonds [179]. In addition, quercetin has a synergistic effect with cisplatin on inhibiting the proliferation, migration and invasion of HeLa and SiHa cells by inhibiting the expression of METTL3 [180]. Baicalin (Fig. 3B) is widely found in the traditional Chinese medicine (TCM) Huang Qin, and it possesses significant antitumor effects in many cancers [181]. The baicalin hydrate inhibited tumor growth in NPC both in vivo and in vitro by influencing the genomic stability and affecting the splicing of Suv39H1 by upregulating m6A RNA methylation, as evidenced by increased METTL3 and METTL14 and decreased FTO and ALKBH5 [182]. Epigallocatechin gallate (EGCG, Fig. 3B) is a tea flavonoid with powerful antioxidant, anti-inflammatory and anticancer effects, which may be associated with the regulation of cyclin A2 and CDK2 in an m6A-dependent manner mediated by inhibiting the expression of FTO and increasing expression of YTHDF2 [183].

Alkaloids
Betaine (Fig. 3C) is rich in the roots of Beta Vulgaris and acts as a methyl donor in the transformation of homocysteine to methionine [184]. As methionine is a substrate for SAM, an essential methyl group donor for mRNA m6A modification, betaine is likely to play an important role in m6A methylation. Zhang et al. found that betaine suppressed the expression of the m6A methylases METTL3 and METTL14 but facilitated the expression of the demethylases FTO and ALKBH5 in HepG2 cells [185]. In addition, clausine E (Fig. 3C) and camptothecin and their analogs (Fig. 3C) [186,187].

Anthraquinone
Rhein (Fig. 3D), an anthraquinone rich in Rheum rhabarbarum [188], was identified as the first cell-active reversible and competitive inhibitor of FTO [189,190]. Molecular modeling combined with biophysical techniques revealed that the inhibition of FTO by rhein occurred through directly binding to nucleic acids, competitively binding to the 2-oxoglutarate (2-OG) cofactor at the active site, or both [189].

Terpenoids
Saikosaponin is a classical triterpenoid that is extracted from Radix Bupleuri (Chinese name: Chaihu) and possesses anti-inflammatory and anticancer activities [191]. Saikosaponin D (Fig. 3E) inhibited FTO to rescue m6A hypomethylation in MYC and RARA. These actions in turn disrupted the stability of MTHFR and BCL2, thus sensitizing MV4-11-or Kas-1-resistant human myeloid mononuclear leukemia cells to tyrosine kinase inhibitors [192].

Other natural products targeting m6A regulators
Apart from the natural products mentioned above, other active natural products have been shown to possess biological activities against m6A and to exert anticancer activity. Fusaric acid (Fig. 3F) is a mycotoxin produced by Fusarium species [193]. It caused a decrease in p53 expression in HepG2 cells by downregulating m6A methylation of p53 mRNA, as indicated by the decreased expression of METTL3 and METTL14. In addition, the translation of p53 was simultaneously blocked by downregulating YTHDF1, YTHDC2, and YTHDF3 [194]. Radicicol (Fig. 3F) was isolated from the fungus Monosporium bonorden [195].
The crystal structure showed that the 4-Cl-1,3-diol group was an essential structure in radicicol responsible for binding to the FTO protein with an IC 50 value of 16.04 μM [196]. Simvastatin (Fig. 3F) is a synthetic modification of a fermentation product derived from Aspergillus terreus [197], and it inhibited the migration and invasion of A549 cells by reducing m6A enrichment and its methyltransferase METTL3 in EZH2 mRNA, thus inhibiting the interaction between IGF2BP2 and EZH2 [198]. Sulforaphane (Fig. 3F) was identified as an epigenetic modulator by diminishing m6A methylation levels in BC cells to induce cell cycle arrest, autophagy and apoptosis [199].

Lead compounds targeting m6A regulators from integrating AI technology and chemosynthesis
Modern approaches that integrate AI technology and chemosynthesis into the field of drug discovery have advantages such as speed, ease of use, and cost saving.
Here, we summarize the anticancer m6A modulators that have been discovered with the help of modern technologies in recent years.

Targeting demethylases
Since FTO was the first recognized m6A modification demethylase, targeting FTO is currently the most popular direction in research on m6A regulation. Cai-Guang Yang and coworkers developed a series of FTO inhibitors by applying AI-based approaches. The crystal structure of FTO was used in docking studies to screen the inhibitor of FTO from the drug-like SPECS database that contains 100,000 compounds. The natural product rhein (1; Fig. 4A) was identified as the first cell-based FTO inhibitor, which also inhibited ALKBH2 activity with a IC 50 value on the same order of magnitude as FTO (IC 50 = 21 μM) in 2012 [189]. To avoid competition with the AlkB subfamily, a high-throughput fluorescence polarization (FP) assay was applied to screen selective inhibitors of FTO from an older drug library containing 900 drugs. The anti-inflammatory drug meclofenamic acid (2; Fig. 4A) was identified as an inhibitor of FTO  [200]. Furthermore, 8 fluorescein molecules, which have structures similar to those of 2, were designed and synthesized. The structure-activity relationships of these fluorescent FTO inhibitors are elucidated through the X-ray crystal structures of FTO/fluorescein complexes. Among these fluorescein derivatives, FL1 (3) and FL2 (4) (Fig. 4A) were selected as bifunctional molecules for selectively inhibiting and specifically photoaffinity labeling of FTO with IC 50 values of 6.65 and 1.72 μM in HeLa cells, respectively [201]. In 2019, Yang et al. employed a structure-based rational design and achieved a promising FTO inhibitor FB23 (5) and it inhibited FTO-mediated demethylation with an IC 50 value of 0.06 μM, which is 100-fold more active than that of 2. Due to the poor permeability of 5 against AML cells, FB23-2 (6; Fig. 4A) with significantly improved antiproliferative activity and cellular efficacy was synthesized. Both 5 and 6 display high selectivity toward FTO but no effect on ALKBH5 demethylation in vitro. Mechanistically, 6 directly bound to and inhibited FTO, upregulating the expression of RARA and ASB2 and downregulating the expression of MYC and CEBPA to exert antileukemia therapeutic effects on a series of AML cell lines, patient-derived primary leukemia cells and patient-derived xenograft (PDX) mouse models [202]. In 2021, Yang et al. published a new finding that tumors exploit FTO-mediated regulation of glycolytic metabolism to evade immune surveillance. They developed a new potent FTO inhibitor Dac51 (7; Fig. 4A) based on 5/6. It exerted promising inhibitory activity on FTO and inhibits the glycolytic capacity of B16-OVA and LLC cell lines via the FTO-m6A-Jun/Cebpb signaling pathway [203]. Moreover, 7 may exert antitumor effects mediated by T cells to prevent tumors from recurring through the memory T cell response in patient-derived organoids and a mouse model of diverse cancer types. Impressively, a combination treatment of 7 and anti-programmed death-ligand 1 (PD-L1) blockade could enhance therapeutic outcomes [203]. Inspired by the binding sites of FTO in 2, a combination of structure-based drug design and molecular docking with Schrödinger software was applied to screen FTO inhibitors, FTO-02 (IC 50 = 2.2 μM) and FTO-04 (8; IC 50 = 3.39 μM). However, only the anticancer ability of 8 was validated; it was found to inhibit neutrosphere formation in multiple GSC cell lines and significantly increased m6A levels [204]. This work represented an important step forward by combining structure-based drug design and a highthroughput in vitro inhibition assay system to identify a new chemical class of FTO inhibitors with tightly defined physicochemical properties. FTO is a 2-OG-dependent N-methyl nucleic acid demethylase, and approximately 150 2-OG analogs are screened by differential scanning fluorometry-and liquid chromatography-based assays [205], among which 9 (Fig. 4B) have been used in clinical studies and have also shown inhibitory activity against FTO [205]. Furthermore, 10 (Fig. 4B) was generated with distinct selectivity for FTO (IC 50 = 0.81 μM) against other AlkB subfamilies and 2-OG oxygenases [206]. A prototype example of AI-based approaches was applied in the discovery of the FTO inhibitor entacapone (11; Fig. 4B) from a library of FDA-approved drugs. Huang et al. combined multiple methodologies, including structure-based hierarchical virtual screening strategies, biochemical experiments, in vivo experiments, and transcriptome sequencing analyses, to identify entacapone as an FTO inhibitor with an IC 50 value of 3.5 μM [207]. In 2020, through structurebased virtual screening, Chen et al. found two potent FTO inhibitors, CS1 (12) and CS2 (13) (Fig. 4B). They shared similar key biological pathways with 6, which directly bound to FTO and efficiently suppressed its m6A demethylase activity, with IC 50 values of 142.6 nM and 712.8 nM, respectively. Nonetheless, 12 and 13 targeting FTO might exert antileukemic activity by suppressing AML stem cell maintenance, sensitizing cancer cells to T cell cytotoxicity, and overcoming immune evasion [59]. Diacerein (14; Fig. 4B) was another potent FTO inhibitor identified by a single quantum dot-based fluorescence resonance energy transfer (FRET) sensor. Rather than being a chelator of metal ions or a structural mimic of 2-oxyglutarate, diacerein directly bound to FTO (IC 50 = 1.51 μM) to inhibit the demethylation activity of FTO in HeLa cells [208].
The FTO inhibitors mentioned above provide small molecular tools for the study of m6A modification and the biological function of FTO. In addition, other potent m6A demethylase ALKBH5 inhibitors, such as 15 and 16, were identified from a library of 144 000 compounds, and most of them showed strong anticancer properties [209][210][211]. Among them, MV1035 (17; Fig. 4C), selected by 3D proteome-wide scale screening, was recently recognized as a ALKBH5 inhibitor that possessed potent anticancer activity against U87 glioblastoma cells [211]. Miao et al. found that the HSP90 inhibitor ganetespilb facilitates the translation of DNAJB4 by m6A modification at A114 site by increasing the expression of YTHDF3 in M14 cells [212].

Targeting methylases
As epitranscriptomic writers for m6A methylation, the m6A methylases METTL3 is involved in various stages of multiple hematoma and solid malignancies, including tumor stemness, immune microenvironment, drug resistance, metastasis and recurrence [85,96,132]. Therefore, METTL3 has been recognized as one of the most promising therapeutic targets for anticancer drug discovery. In an effort to explore a specific inhibitor of METTL3, a virtual screening of ZINC and DrugBank 4.0 databases identified 4 compounds (18-21; Fig. 4D) with piperidine or piperazine rings, and they functioned as METTL3/14-WTAP activators to facilitate RNA methylation, which increased the mRNA m6A levels to shift the cell cycle to G 0 and G 1 phase without cytotoxic effects at 100 μM [213]. Another METTL3 and WTAP inhibitor was chidamide (22; Fig. 4), which downregulates c-MET expression by suppressing m6A methylation, as evidenced by the downregulation of METTL3 and WTAP in NSCLC to increase the therapeutic efficacy of crizotinib [214]. Meanwhile, by screening a library of 4000 analogs and derivatives of the adenosine moiety of SAM accompanied by the high-throughput docking assay, two adenine derivatives, 23 and 24 (Fig. 4) were selected as METTL3 inhibitors. They have been validated to bind to METTL3 by X-ray crystallography [215]. In addition, a high-throughput screen of 250,000 diverse drug-like compounds was performed. Among these candidates, STM2457 (25; Fig. 4) was validated to bind to the METTL3-METTL14 heterodimer specifically and directly in the SAM site by X-ray crystallography. As the first METTL3 inhibitor, 25 has been validated to increase cell differentiation and apoptosis by reducing m6A enrichment in METTL3-dependent core leukemogenic m6A substrates, such as HOXA10 and MYC. As a result, 25 exerts a potent therapeutic effect on multiple AML mouse models by affecting the AML stem cell or leukemia propagating compartment [9]. The effects of STM2457 in other tumor types are now being investigated by STORM, which is aiming to put STM2457 in phase trials in 2022 [8]. By screening an adenine-based library with a homogenous time-resolved fluorescence (HTRF) enzyme inhibition assay, a potent and selective METTL3 inhibitor, UZH1a (26; Fig. 4) was identified. It was further validated to selectively bind to METTL3 by X-ray crystallography, and it slightly suppresses the expression of METTL3 but significantly reduces m6A levels in the mRNA fraction in the leukemia cell line MOLM-13 and human osteosarcoma U2OS cells [216].
In contrast to conventional inhibitors, photoactivated compounds have been creatively constructed. With the assistance of computational docking, Lan et al. identified a caged molecule activator of METTL3/14, photocaging substituent-linked MPCH (27 and 28) (Fig. 4). It activates METTL3/14 and results in considerable m6A hypermethylation after short UV light exposure in different cells. Owing to the rapid uncaging of MPCH by light radiation, the side effects are minimal and can be controlled. As the release of medicine could be swiftly initiated by short light irradiation, it might be suitable for utilization in living systems instead of depending on the addition or deletion of endogenous enzymes [217]. The above small molecules targeting METTL3 showed potent therapeutic effects in tumor treatment, indicating that METTL3 could be the most promising target. However, since METTL3 is widely involved in the expression of various genes, the in vivo side effects of METTL3targeted agents should be strictly tested. In addition to the inhibitors of METTL3 mentioned above, some other METTL3 inhibitors are now being investigated by Accent Therapeutics and Gotham Therapeutics. These companies are aiming to put their own METTL3 inhibitors into phase I trials in 2022 [8].

Conclusions and perspectives
Epigenetic regulation has become a hot topic in recent decades and RNA m6A modification in cancer research has been developed into one of the most popular fields in recent years. Epigenetic regulation inhibitors, such as azacytidine and decitabine, which are two inhibitors of DNA methylation, have shown great anticancer effects in clinical use. The dysregulation of m6A modification frequently occurs in many types of cancers and m6A modification regulates the malignant phenotypes and behaviors primarily by controlling the expression of oncogenes and tumor suppressor genes. Notably, aberrant m6A modification is critically associated with tumor progression and cancer patient prognosis. Therefore, targeting m6A modification regulators might also be a potential and promising therapeutic strategy for cancer treatment.
There are many advantages in using traditional medicines or natural products to screen the inhibitors and activators of m6A medication regulators. First, the efficacy and safety of traditional medicines and natural products have been validated by generations through the repeated experiences of countless rounds of trial and error over thousands of years. Moreover, many bioactive small molecules derived from traditional medicines and natural products have novel chemical structures and multiple biological activities, and more than 60% of anticancer drugs are natural origins or contain the pharmacophores of natural products [218]. These powerful advantages make traditional medicines and natural products reliable sources for the discovery of new therapeutic agents targeting m6A modification.
AI-assisted techniques have been widely used for the discovery and development of drug candidates [219,220] and several online databases related to traditional medicines or natural products have been developed. The TCM Systems Pharmacology Database and Analysis Platform (TCMSP), is a comprehensive phytochemical database for drug discovery from herbal medicines, and it includes 29,384 ingredients of approximately 500 Chinese herbal medicines, more than 3, 000 targets, and 837 related diseases [221]. Indian Medicinal Plants, Phytochemistry, And Therapeutics (IMPPAT) is a database containing 9596 phytochemicals, 1742 Indian medicinal plants, and 1124 therapeutic uses spanning 27,074 plantphytochemical associations and 11,514 plant-therapeutic associations [222]. Therefore, it would be more efficient to develop novel and effective therapeutic agents that inhibit m6A modification-mediated tumor progression by combining traditional medicines and natural product databases with AI-based drug discovery approaches. Additionally, we present a framework for m6A-targeting drug discovery through integration of AI and traditional medicines and natural products (Fig. 5). Specifically, we could collect and organize data from protein databases related to m6A modification regulators, and compound libraries (e.g., TCMSP, IMPPAT) that contain the natural origins, chemical structures, physicochemical properties, pharmacological activities, side effects and toxicities, and pharmacokinetic parameters of natural products and active small molecules derived from traditional medicines. Then, AI-based methods were used for high-throughput virtual screening of lead compounds through target protein structure-based approaches (e.g., molecular docking simulation), ligand-based approaches (e.g., quantitative structure-activity relationship (QSAR) models) and drug-target interaction data. With a range of AI techniques, we can better screen and predict potential compounds, develop the modification and optimization of chemical structures, and assess the druggability of lead compounds targeting m6A modification.
Although more than 20 m6A modification regulators have been identified, only a few of them have been confirmed to be druggable and could serve as therapeutic targets for cancer treatment. Meanwhile, many inhibitors and activators of m6A modification have been reported, but none of them have been approved for cancer treatment in the clinic. The currently developed m6A modification inhibitors and activators might have poor target specificity, therapeutic efficacy, safety, and pharmacokinetics [202,204]. It takes several years, or even decades, to develop an anticancer drug from the laboratory to the clinic and incurs high costs. The current m6A modification inhibitors and activators must be investigated thoroughly in a series of preclinical and clinical trials before approval for clinical use. These obstacles critically hinder the development of current m6A modification inhibitors and activators into drugs for clinic use. Currently, revolutionary AI-assisted approaches to drug discovery, design, and development have been developed, and we could take full advantages of AI to develop m6A inhibitors and activators with better specificity, efficacy, safety, and pharmacokinetics, which will reduce the cost and shorten the time of drug development related to m6A Fig. 5 A framework for the m6A-targeting drug discovery through integration of AI and traditional medicines and natural products. The data collected and organized from protein databases and compound libraries were performed with high-throughput virtual screening, followed by screening and predication, modification and optimization of chemical structures, and assessment of druggability of lead compounds that target m6A modification modification by AI. We therefore believe that an increasing number of novel, specific, and effective m6A modification inhibitors and activators will be developed and approved for clinical use in the near future.