N⁶-Methyladenosine Modulates Nonsense-Mediated mRNA Decay in Human Glioblastoma

Nonsense-mediated mRNA decay (NMD) contributes to mRNA surveillance pathways that affect a broad spectrum of cellular functions and maintain homeostasis. Although the primary function of NMD in reducing errors in gene expression by eliminating mRNA transcripts that contain premature termination codons (PTC) is well known, the mechanisms of target mRNA selection for NMD are still not well understood.

RNA methylation is a reversible modification of mRNA and has been linked to many types of cancer. N⁶-Methyladenosine (m⁶A) represents the most abundant methylation modification of mRNAs in eukaryotes (1–3), and it regulates almost every aspect of mRNA metabolism, including RNA processing (4, 5), transport from the nucleus to cytoplasm (6, 7), translation (8, 9), and decay (10, 11). The m⁶A methylation marks on mRNA are dynamically regulated in mammals through the methyltransferase complex, composed of the catalytic subunit METTL3, and demethylases (e.g., FTO and ALKBH5; refs. 7, 11–13) and are detected by “m⁶A readers.” YTH domain–containing proteins, including YTHDF1-3, YTHDC1, and YTHDC2, act directly as “m⁶A readers” and can interact with distinct subsets of m⁶A sites to produce different effects on RNA processing (9, 10, 14–16). The function of m⁶A RNA methylation is highly variable and context-dependent, and its underlying mechanisms in the recognition of NMD targets are not well understood.

Recent studies have revealed that m⁶A methylation of mRNA results in diverse regulatory functions in cancer initiation and progression. In addition, dysregulated m⁶A methylation is closely related to various types of cancers. It has been reported that the m⁶A methyltransferase METTL3 is required for the growth, survival, and invasion of cancer cells (17–19). The m⁶A demethylase FTO was found to play a critical oncogenic role in promoting acute myeloid leukemia (AML; ref. 20). Although evidence is emerging, linking m⁶A modulators and tumorigenesis, it remains to be determined whether m⁶A modifications on different regions of mRNA, recognized by distinct readers, will lead to different cell fates. METTL3 elevates m⁶A methylation modification to promote glioma stem cells (GSC) stemness by enhancing SOX2 stability in glioblastoma (GBM; ref. 21). Controversially, another research group found that knockdown (KD) of METTL3 dramatically promoted GSC self-renewal and tumorigenesis (22). Moreover, ALKBH5, which decreases m⁶A modification in GSCs, exerts an important tumorigenic role in the progression of GBM through regulation of FOXM1 expression (23). These findings have raised questions about whether m⁶A methylation modifications that affect GBM progression are dependent on the RNA sequence and are dynamically regulated.

Here, we observed preferential distribution of m⁶A peaks in GBM cells. Elevated METTL3 in clinical specimens correlated with higher grades of gliomas, increased tumor recurrence, and worse clinical outcomes. Moreover, we found that silencing of METTL3 led to reduced aggressive and tumorigenic capabilities, as well as diminished GSC phenotypes in GBM cells. Methylated RNA immunoprecipitation-seq (MeRIP-seq) and RNA-seq analyses revealed that KD of METTL3 led to downregulation of NMD-targeted splicing factor mRNA transcripts that was dependent on the m⁶A reader YTHDC1. Importantly, splice alterations of targeted mRNAs were critical for tumor growth inhibition and suppression of stemness due to METTL3 KD. Together, our study identifies m⁶A methyltransferase METTL3 as a modulator of NMD to sustain malignancy in GBM.

Both GBM and normal brain tissue surgical specimens were collected in The First People's Hospital of Changzhou and Xiangya Hospital of Central South University, in accordance with institution-approved protocols. Written informed consent was obtained from each study participant after a thorough explanation of the procedure and its risk, in compliance with the Declaration of Helsinki. Collected specimens were further split into two parts for RNA extraction and protein isolation. If only a limited amount of specimens was obtained, only an RNA extraction assay was performed. Three freshly obtained specimens were specifically used for primary cell establishment (see section below). All specimens were examined by neuropathologists to verify tumor types and grades.

The human GBM cell lines of U251 and U87MG were provided as a gift from Dr. Jun Cui's laboratory at Sun Yat-sen University and were grown in Gibco DMEM containing 10% FBS (Gibco) at 37°C in a humidified atmosphere containing 5% CO₂.

For the culture of primary GBM cells, surgically removed GBM specimens were washed with and minced in sterile PBS. Next, a single-cell suspension was obtained by pressing the minced tissues through 40-μm cell strainers (Falcon). Dissociated cells were cultured in DMEM supplemented with 15% FBS (Gibco), 1 × B27 (Invitrogen), 20 ng/mL epidermal growth factor (CantonBIO), and 20 ng/mL FGF (CantonBIO) at 37°C in a humidified atmosphere containing 5% CO₂. All primary cells were passaged every 7 days.

Five-week-old female Balb/c athymic nude mice were purchased from Model Animal Research Center of Nanjing University and housed in individually ventilated microisolator cages. Nude mice were divided into three groups of 6 mice each.

For subcutaneous tumor model, each mouse was injected subcutaneously in the right flank with 2 × 10⁶ U87MG cells (METTL3-KD or control) in 100-μL PBS. Tumor sizes were determined with calipers every 5 days by measuring the length and width. Tumor volumes were calculated according to the following formula: volume (mm³) = (length × width × width)/2. Fifty-eight days after the tumor cell injection, the mice were sacrificed and tumor xenografts were removed, weighted, fixed in formalin, and stored at 4°C.

For intracranial GBM Xenograft Model, each mouse was intracranially injected with 5 × 10⁵ luciferase-transduced U87MG cells (METTL3-KD or control) in 10-μL PBS solution as described previously (24). Tumor growth was monitored by using a Xenogen IVIS Spectrum system (Caliper Life Sciences) weekly.

Animal experiments were approved by the Animal Care and Use Committee of Sun Yat-sen University.

Total m⁶A content was measured in 200-ng aliquots of total RNA extracted from METTL3-KD or scrambled, control U87MG, or U251 GBM cells using an m⁶A RNA methylation quantification kit (EpiGentek), according to the manufacturer's instructions.

Total RNA was isolated from METTL3-KD or scrambled control U87MG GBM cells, as mentioned above, and the mRNA was further separated using Dynabeads mRNA Purification Kit (Invitrogen, 61006). After fragmentation, using RNA fragmentation reagent (Invitrogen, AM8740), the obtained mRNA was immunoprecipitated with anti-m⁶A antibody (Synaptic Systems, 202003), and then washed and eluted by competition with m⁶A sodium salt (Sigma-Aldrich, M2780). Both input samples and immunoprecipitation (IP) eluates were used for preparing the sequencing libraries using NEBNext Ultra RNA Library Prep Kit for Illumina and submitted for sequencing using Illumina HiSeq 2500. Reads, mapping and m⁶A peak calling, were performed as previously described (25). The m⁶A peaks of shMETTL3 U87MG cells were from the overlapped peaks of shMETTL3-1 and shMETTL3-2.

In YTHDC1 RNA immunoprecipitation (RIP)-qPCR experiments, U87MG cells were harvested and lysed in IP lysis buffer (150 mmol/L KCL, 0.5 mmol/L DTT, 5 mmol/L EDTA, 0.5% NP-40, 25 mmol/L Tris, pH 7.4). Each lysate was further divided into three groups for anti-YTHDC1, anti-IgG (negative control), and input (positive control). Either YTHDC1 antibody (Abcam) or IgG was added to each sample to enrich RNA-binding protein (RBP). Subsequently, the RBP of interest, together with the bound RNA, was collected using Dynabeads (Thermo Fisher Scientific). After washing off unbound material, the RBP was digested by proteinase K, and the RNA bound to immunoprecipitated RBP was purified and reverse transcribed into cDNA. Then, qPCR assay was performed to measure the %Input of SRSFs mRNAs in each group. The primer sequences used for RIP-qPCR analysis were provided in Supplementary Table S2.

All analyses were performed using GraphPad Prism version 5.0 (GraphPad Software). The survival curves for combined expression of METTL3 and splicing factors were plotted according to the Kaplan–Meier method, using PROGgene V2 software online (http://watson.compbio.iupui.edu/chirayu/proggene/). The association among the expression levels of METTL3 and splicing factors was analyzed using Spearman rank correlation. Data were presented as the mean ± SD, and the significance levels of all tests were two-sided. The P value of less than 0.05 was considered statistically significant and marked as *; a P value less than 0.01 or 0.001 was marked as ** and ***, respectively.

Previous studies have suggested that GSCs are derived from mutated neural progenitor cells (NPC), which are critical for GBM tumorigenesis (26). Using MeRIP-seq data from GSE87515 (23) and GSE54365 (27), we first compared the m⁶A peaks at each locus between GSCs and NPCs, respectively (Fig. 1A) and then divided the peaks into three categories comprising gene loci with m⁶A peak enrichment in (i) both GSCs and NPCs; (ii) GSCs, or (iii) NPCs (denoted as “shared,” “GSC,” and “NPC,” respectively; Fig. 1B; Supplementary Fig. S1A). We found that 9,627 loci in GSCs had elevated levels of m⁶A modifications that were initially unmodified in NPCs (Fig. 1B). We further analyzed the signaling pathways of the three categories and found that loci with elevated levels of m⁶A modifications were associated with metabolic pathways (Fig. 1C; Supplementary Fig. S1B). Specifically, the cancer metabolism–associated loci TGFB2, TGFB3, and TEAD2 were highly enriched with m⁶A modifications in GSCs (Fig. 1D).

To determine the expression of m⁶A modulators in GBM, we first queried the Repository for Molecular Brain Neoplasia Data (REMBRANDT) datasets. Compared with normal brain controls, GBM specimens displayed increased levels of METTL3, METTL14, YTHDF1, and YTHDF2 (Supplementary Fig. S2A). To confirm these findings, we examined the mRNA expression of m⁶A modulators in 36 GBM specimens from the First People's Hospital of Changzhou. METTL3 and YTHDF2 were markedly elevated in GBM patient samples as compared with normal brain tissues (Fig. 2A). Western blot analysis also showed higher METTL3 protein levels in GBMs compared with that in normal brain tissues (Supplementary Fig. S2B). Next, we examined METTL3 expression in a paraffin-embedded human glioma tissue array by IHC. As expected, there were remarkably more METTL3-positive cells in GBMs (grade 4) than in normal brain tissues or in lower grade gliomas (Fig. 2B). Intriguingly, significantly higher levels of METTL3 were found in recurrent tumors compared with paired, original tumors (Fig. 2C).

Next, we designed a screen based on a CRISPR/Cas9 library of sgRNAs to identify potential m⁶A modulators regulating GBM cell proliferation and GSC sphere formation (Supplementary Table S1; Supplementary Fig. S2C). We transduced U87MG cells with 16 sgRNA lentiviral supernatants (one sgRNA per well), containing a puromycin selection cassette to eliminate uninfected cells. Infected cells were grown in a two-dimensional monolayer and three-dimensional Matrigel for 7 days. After three rounds of screening, we found that METTL3 and YTHDF2 were involved in the maintenance of GBM cell proliferation (Supplementary Fig. S2D) and sphere-forming capacity (Fig. 2D). Notably, high METTL3 levels in GBMs predicted poorer patient survival [GBM patient data from REMBRANDT, GSE7696 (28), and GSE43378 (Fig. 2E; ref. 29)]. However, the expression of other modulators, except YTHDF2, did not significantly correlate with the time of GBM patient survival (Supplementary Fig. S2E). Collectively, METTL3 was upregulated in GBMs and may be critical for tumor growth.

To test whether METTL3 was essential for GBM cell growth, we suppressed the expression of METTL3 in GBM cells through shRNA-expressing lentiviruses. Both shMETTL3-1 and shMETTL3-2 could downregulate METTL3 expression in GBM cells (Supplementary Fig. S3A and S3B). As expected, depletion of METTL3 also led to significantly reduced m⁶A modification levels of mRNAs in both GBM cell lines (Supplementary Fig. S3C). Compared with cells expressing control shRNAs, both METTL3-KD GBM cell lines (U87MG and U251) showed significantly reduced cell proliferation (Fig. 3A). Similar results were also obtained in METTL3-KD primary GBM cells derived from 3 patients with GBM at Xiangya Hospital (Fig. 3A). Moreover, overexpression of the m⁶A catalytic inactive mutant METTL3 acted in a dominant-negative manner to suppress cell growth and m⁶A modification levels of mRNAs in U87MG and U251 cells (Fig. 3B; Supplementary Fig. S3D and S3E). Consistent with previous reports, METTL3 KD increased the proportion of apoptotic cells in GBMs (Fig. 3C; Supplementary Fig. S3F). Furthermore, METTL3 KD resulted in significantly decreased migration and invasiveness of GBM cells (Supplementary Fig. S3G and S3H). Conversely, overexpression of the METTL3 dominant-negative mutant in GBMs inhibited cell migration and invasion (Supplementary Fig. S3I). These data support an important role of the m⁶A catalytic domain of METTL3 in controlling GBM cell growth, survival, and invasion in vitro.

We further investigated whether METTL3 inhibits the tumorigenic capacity of GBM cells in vivo. When U87MG cells, transduced with shMETTL3 lentiviruses, were inoculated into nude mice, the cells produced much smaller xenograft tumors than cells expressing scrambled, control shRNAs (Supplementary Fig. S3J and S3K). Subsequently, the excised xenograft tumors were processed for IHC and terminal deoxynucleotidyl transferase–mediated dUTP nick end labeling assay (Supplementary Fig. S3L). Significantly reduced ratios of KI67-positive cells to apoptotic cells were observed in METTL3 KD cell-derived tumors (Supplementary Fig. S3L).

Compared with the control shRNA, both shMETTL3 sequences significantly decreased stem cell size (Supplementary Fig. S4A), number (Fig. 3D), and frequency (Fig. 3E) in GSC-derived tumor neurospheres. Whereas overexpression of WT METTL3 promoted U87MG and U251 neurosphere formation and stem cell frequency, the METTL3 mutant inhibited these phenotypes (Fig. 3F; Supplementary Fig. S4B and S4C). Moreover, shMETTL3 decreased the CD133-positive populations in U251- and U87MG-derived neurospheres (Supplementary Fig. S4D). Compared with the mice injected with control U87MG spheroid-derived GSCs, those injected with shMETTL3 spheroid-derived GSCs displayed impaired tumor growth and a lower rate of tumor formation (Fig. 3G).

To map m⁶A modification sites and unveil potential METTL3 functions in GBM, we performed MeRIP-seq on U87MG cells with silenced METTL3 (Supplementary Fig. S5A and S5B). Consistent with previous studies, we demonstrated that m⁶A peaks in GBMs were enriched in the RGACH motif (R = G/A; H = A/C/U; Fig. 4A), and abundant in coding sequences (CDS) and untranslated regions (UTR) of mRNAs (Fig. 4B). Compared with the control GBMs, the m⁶A peaks across entire gene bodies and 3′UTRs were markedly decreased in shMETTL3 GBMs (Supplementary Fig. S5C). Furthermore, 6,444 genes with significantly decreased m⁶A levels in shMETTL3 GBMs were identified as potential m⁶A-regulated genes (Fig. 4C). Moreover, carcinogenesis pathways were significantly enriched in these m⁶A-regulated genes (Fig. 4D), suggesting a role for METTL3-mediated m⁶A modifications in GBM tumorigenesis.

We further characterized the molecular signaling pathways regulated by METTL3 using RNA-seq (Fig. 4E; Supplementary Fig. S5D and S5E). Transcripts encoding apoptotic signaling pathways and glial cell differentiation genes were enriched in both METTL3-KD GBM cells (Fig. 4F and G). The upregulated expression of apoptotic and differentiation genes was confirmed by RT-qPCR (Fig. 4H). Interestingly, a large number of genes were also significantly downregulated in METTL3-KD cells. Gene ontology (GO) and gene set enrichment analysis analyses revealed that these downregulated genes were mainly involved in RNA processing and mRNA splicing (Fig. 4F and G). The downregulated expression of these splicing factors was also confirmed by RT-qPCR (Fig. 4H; Supplementary Fig. S5F). Next, we randomly selected several splicing factors regulated by METTL3 and found that KD of these genes impaired the proliferation of GBM cells (Fig. 4I; Supplementary Fig. S5G). Consistently, we found that METTL3 expression positively correlated with splicing factors, especially SRSF3, SRSF6, and SRSF11 (Fig. 4J; Supplementary Fig. S5H). GBM specimens displayed increased levels of SRSF3/6/11 compared with normal brain controls (Supplementary Fig. S5I). Kaplan–Meier survival analysis revealed that patients with elevated expression of METTL3 and SRSF3/6/11 had shorter overall survival time (Supplementary Fig. S5J). Together, these results identify SRSFs as downstream targets regulated by METTL3, which are dysregulated in GBM.

Expression of splicing factors is regulated via alternative splicing of the conserved regions to yield mRNAs, which are degraded by NMD. However, whether a significant proportion of splicing factor mRNA transcripts occur via reduced levels of m⁶A modifications and are ultimately degraded by NMD is not known. NMD analyses using the METTL3-KD RNA-seq data revealed that the mRNA transcripts that undergo NMD were significantly enriched in genes associated with RNA splicing (Supplementary Fig. S6A). By comparing the m⁶A modification of SRSFs in NPCs and GSCs, we found that m⁶A modifications around start codon of splicing factor transcripts (e.g., SRSFs) were elevated (Fig. 5A; Supplementary Fig. S6B). The specific m⁶A modification sites of the mRNAs of SRSFs were identified by miCLIP-seq analyses (Supplementary Fig. S6C; ref. 30). Importantly, on the basis of the markedly decreased m⁶A modification peaks, premature termination (i.e., stop) codons (PTCs) in the mRNAs of SRSFs were generated via exon inclusion or skipping upon METTL3 depletion (Fig. 5B; Supplementary Fig. S6D and S6E). Therefore, these mRNAs with PTCs were predicted to be subject to NMD, and the predicted protein products were not physiologically relevant. A similar phenotype was also found in other splicing factors (Supplementary Fig. S6E). Silencing METTL3 significantly reduced protein-coding mRNAs (mRNAs without PTCs) of SRSFs (Fig. 5C). In contrast, an inhibitory effect on mRNAs with PTCs was also observed upon overexpression of WT METTL3 (Fig. 5D). Moreover, protein-coding mRNAs of SRSFs were significantly decreased, upon overexpression of mutant METTL3 (Fig. 5E).

To investigate whether the mRNAs of SRSFs with PTCs were substrates for NMD, we analyzed steady-state levels of these mRNAs via inhibition of NMD. We used shRNA targeted at UPF1 (Supplementary Fig. S6F), which is the central component of the NMD pathway. Treatment with shRNA against UPF1 greatly increased the steady-state levels of the mRNAs with PTCs and total mRNAs of SRSFs (Fig. 5F). Similar increases in the levels of the mRNAs with PTCs were also observed after inhibition of NMD by treatment with cycloheximide (Fig. 5G). In addition, we found that the expression levels of SRSFs NMD variants were decreased in clinical GBM samples (Supplementary Fig. S6G).

To analyze m⁶A modification levels around start codon that modulate NMD signaling, we used a previously described (31) in vitro luciferase reporter systems (Fig. 6A). The reporter systems were derived from SRSF6, including the pre-mRNA sequence from transcription start site up to exon 3. The PTC in the intron 2 of SRSF6 is maintained. In addition, we mutated the A with G to inactivate the m⁶A modification–mediated exon inclusion. We observed that the SRSF6-Renilla–fused mRNA is m⁶A modified (Supplementary Fig. S7A). Consistent with previous results, KD METTL3 efficiently promoted the formation of the mRNA with PTC in the WT reporter, as indicated by reduced luciferase activity (Fig. 6B). However, no significant change of luciferase activity occurred in the reporter with the mutation in m⁶A modification (Fig. 6B). Similarly, a lack of luciferase activity increase has also been observed with addition of the METTL3 mutant in GBM cells, compared with addition of the WT METTL3 (Fig. 6C). RT-qPCR analyses showed higher NMD RNA (RNA with PTC) levels and lower protein-coding mRNA (mRNA without PTC) levels in mutated SRSF6 reporter compared with WT reporter, further suggesting that the m⁶A modification is critical for inhibition of NMD (Supplementary Fig. S7B). Inhibition of NMD with shUPF1 greatly increased the Renilla-SRSF6 NMD RNA and total mRNA of mutated reporter (Supplementary Fig. S7C). To further analyze the effects of m⁶A modifications on NMD, a reporter was constructed consisting of exon 2, exon 3, and flanking intron sequences of SRSF3 without m⁶A modification site (Supplementary Fig. S7D). In contrast to the m⁶A modification around start codon reporter system data, this reporter was greatly resistant to KD of METTL3-induced NMD (Supplementary Fig. S7E). Consistently, WT METTL3 cannot increase the luciferase activity of SRSF3 minigene reporter contracts without m⁶A modification site (Supplementary Fig. S7F).

Adenine base editing is a novel genome-editing approach to convert a target T•A to C•G without requiring homology-directed repair or introducing double-stranded DNA breaks (32). To generate m⁶A site mutation in the SRSFs gene of interest in U87MG cells, we transduced the base editor construct and guide RNA targeting the SRSF3 m⁶A modification site around start codon (Fig. 6D). A total of 14 cell clones were obtained by limited dilution and Sanger sequencing indicated that nine carried the expected mutation (Fig. 6E). We next found that the NMD RNA of SRSF3 significantly increased in m⁶A-mutant U87MG cells compared with WT control (Fig. 6F). These results support that m⁶A modifications around start codon mediate repression of NMD in GBM.

YTHDC1 has been reported as an m⁶A modification reader that mediates mRNA splicing, and mutating either W377 or W428 to alanine completely disrupts its binding to m⁶A RNA (15). KO of YTHDC1 reduced sphere number substantially in METTL3 overexpression cells but not in control cells (Supplementary Fig. S7G). Overexpressing mutant YTHDC1 (m⁶A binding activity loss) failed to enhance the sphere formation capacity of U87MG cells (Supplementary Fig. S7H), suggesting that YTHDC1 contributes to the GBM phenotype (e.g., sphere formation) dependently on its m⁶A-binding activity. Moreover, RIP-qPCR assay results showed that YTHDC1 binds to SRSF3, SRSF6, and SRSF11 mRNA when compared with negative control (Fig. 6G; Supplementary Fig. S7I). In addition, by analyzing PAR-CLIP-seq data from GSE74397 (7) and GSE58352 (2), we found that YTHDC1 binds near the start codon region of SRSFs mRNAs (Supplementary Fig. S7J). KD YTHDC1 led to the accumulation of NMD of SRSF3, SRSF6, and SRSF11 mRNAs in GBM cells (Fig. 6H). Moreover, KD YTHDC1 could affect the luciferase activity of the WT SRSF6 minigene reporter but not the SRSF6 reporter with the mutated m⁶A site (Fig. 6I). Together, these data suggest that YTHDC1 KD-mediated NMD is dependent on m⁶A around start codon of the mRNA in GBM cells.

We compared the alterative splicing events in METTL3 KD U87MG cells compared with U87MG control cells. Using the rMATS tool, a total number of alternative splicing events was identified with an obvious change of exon inclusion levels (ψ, ψ ≥ 0.1). We found that downregulation of splicing factors can affect various types of alternative splicing, including skipped exon (SE), alternative 5′ ss exon (A5SS), alternative 3′ ss exon (A3SS), retained intron (RI), and mutually exclusive exons (MXE), and, in particular, SE events in shMETTL3 U87MG cells were most affected (Fig. 7A). Subsequent analysis indicated that the SE type of alternative splicing was negatively regulated by METTL3-KD, whereas A3SS, MXE, and RI were positively regulated by METTL3-KD (estimated by changes of ψ after METTL3-KD; Fig. 7B). Next, we compared genes that were differentially spliced in METTL3-KD U87MG cells to cells with scrambled control shRNA. When analyzing the cellular functions of METTL3-regulated alternative events using GO, we found that METTL3 affected alternative splicing of a number of genes, such as BCL-X and NCOR2, with functions in cancer cell death and motility (Fig. 7C; Supplementary Fig. S8A).

BCL-X is a well-known example of genes critical for cancer that has splicing variants that can function as cancer biomarkers and therapeutic targets. The BCL-XL isoform is antiapoptotic in various cancer types, whereas the BCL-XS isoform is proapoptotic in cancer. Using semiquantitative RT-PCR and qPCR, we confirmed that KD of METTL3 significantly shifted the transcription of BCL-XL into BCL-XS (Fig. 7D and E; Supplementary Fig. S8B). The protein levels of BCL-XL were also reduced in METTL3-KD GBM cells (Supplementary Fig. S8C). To examine whether the splicing alteration of BCL-X was responsible for the METTL3-KD phenotypes, we designed shRNAs to specifically target BCL-XS to inhibit the expression of BCL-XS but not BCL-XL (Supplementary Fig. S8D). As shown in Fig. 7F and G, METTL3 and BCL-XS double-KD GBM cells grew significantly faster with reduced apoptosis than cells with the METTL3-KD alone. This phenotypic rescue suggests that METTL3 maintains the tumorigenicity of GBM cells, at least partially, through the splicing of BCL-X.

NCOR2 (also known as SMRT) exists in two major splicing isoforms, α and τ, which have different roles in preserving cellular identity and tissue homeostasis. Little is known about potential functional differences between these two isoforms in GBM. Using semiquantitative RT-PCR and qPCR, we revealed that KD of METTL3 significantly increased the transcription of isoform α of NCOR2 (Fig. 7H and I; Supplementary Fig. S8E). RT-PCR analysis showed that shRNAs targeting isoform α specifically inhibited its expression but not the τ isoform of NCOR2 in U87MG cells (Supplementary Fig. S8F). Furthermore, we demonstrated that KD of the NCOR2α isoform partially rescued METTL3 KD-induced inhibition of U87MG cell growth (Fig. 7J). In addition, our neurosphere formation data show that KD of the NCOR2α isoform significantly increased the neurosphere formation capacity and stem cell frequency in METTL3-KD U87MG cells (Fig. 7K and L), and together these data suggest that the NCOR2α isoform may play an important role in the regulation of GSC self-renewal.

In this study, we established a novel mechanism for m⁶A modifications around start codon of mRNA splicing factors in modulating NMD of these splicing factors. In addition, we found that METTL3 modulated alternative splicing of BCLX and NCOR2, which leads to GBM tumor outgrowth and self-renewal. Unlike previous studies, reporting that m⁶A modifications at 3′-terminal ends are destabilizers of mRNA, our study revealed that m⁶A modifications around start codon stabilize the mRNAs of SRSFs by preventing NMD. By preventing NMD and promoting mRNA degradation, m⁶A modifications act as a molecular rheostat to finely adjust the transcript levels of SRSFs to influence alternative splicing events.

The role that METTL3 plays in cancer is complex. Two research teams reported opposite conclusions on the role that METTL3 plays in the self-renewal and tumorigenesis of GSCs. The reasons for these opposite conclusions may depend on the patients from whom the GBM cells originated and other compensatory genetic mutations and epigenetic changes in GBM cells. In this study, we chose clinical GBM samples from different stages, knocked down METTL3 expression using both sgRNAs and shRNAs in primary GBM cells, and validated our findings using catalytic inactive mutants of METTL3. All of these results consistently demonstrated that METTL3 promotes proliferation and self-renewal of GBM cells. Notably, the oncogenic ability of METTL3 is dependent upon the m⁶A methyltransferase catalytic domain. Interestingly, although KD METTL14 expression reduced mRNA m⁶A levels in GSCs (22), KO of METTL14 has no effect on GBM oncogenicity. A recent study in AML cells showed that METTL3 bound the promoter regions of active genes (about 80 genes) independent of METTL14 (33). They also showed that CEBPZ is required for recruitment of METTL3 to the promoters. Their results indicate that specialized partner proteins might exist at splicing factor loci in GBM cells that give clues to decipher METTL14-independent METTL3′s functions in GBM.

M⁶A modifications may play different roles at different developmental stages of GMB tumors. In glioma, the mutation of isocitrate dehydrogenase 1 (IDH1) occurs frequently, which results in accumulation of the metabolic by-product 2-hydroxyglutarate (2-HG). The 2-HG could inhibit FTO activity, thereby increasing global m⁶A modifications and contributing to cancer initiation. In the late stage of glioma, high m⁶A modification levels may increase epigenetic reprogramming of non-GSCs into GSCs, whereas KD METTL3 may reduce the proportion of GSCs in GBM. This hypothesis was indirectly verified by the observation that KD of METTL3 in pluripotent stem cells, at naїve or primed states, resulted in different cell fate transitions (34). Thus, strategies designed to reduce levels of m⁶A modifications may provide a means to target malignant GBM and to develop more effective therapies.

The prevailing goal of understanding the regulatory roles of m⁶A modifications in RNA processing has been mainly focused on the regulation of mRNA translation or mRNA stability. Indeed, we found that METTL3 regulates the stability of a specific set of transcripts, such as apoptosis pathways and glial differentiation genes, in GBM. The m⁶A-binding protein YTHDF2 may recognize these methylated mRNAs, leading to their decay and subsequently to decreased cell apoptosis and differentiation, thereby promoting GBM tumor growth and dedifferentiation. Another m⁶A modification reader, YTHDC1, is involved in the process of alternative splicing through recruitment and modulation of splicing factors to their targeted RNAs (35). We have presented a number of findings supporting the notion that m⁶A modifications modulate the NMD of splicing factors. The NMD pathway protects eukaryotic cells by reducing the production of harmful truncated proteins translated from PTC-bearing mRNA transcripts. In our study, the reduced m⁶A modifications by KD METTL3 consequently led to the degradation of SRSFs transcripts via triggering NMD and thus control of GBM initiation and progression. It should be noted that KD METTL3 affected not only YTHDC1-mediated NMD but also the m⁶A-mediated mRNA degradation. Therefore, the effect of KO YTHDC1 on GBM phenotype was not as significant as that of KD/KO METTL3.

SRSFs proteins are known for their ability to promote exon inclusion and exon-skipping events, suggesting the regulatory role of SRSFs proteins in alternative events. Brain tissue has been found to have particularly high levels of alternative splicing (36, 37). Consistently, a large number of cancer-relevant genes have undergone splice alterations in GBM (38–40). The importance of alternative splicing in the development of GBM was further reinforced by the findings that a large number of splicing factors were overexpressed in GBM (41, 42), yet the mechanisms responsible for this upregulation and its clinical relevance remain to be fully addressed. Our results support a model where METTL3 controls the cancer-relevant phenotypes of GBM cells by promoting the expression of SRSFs. This subsequently results in the creation of cancer-specific alternative splicing patterns, such as the preferential expression of the antiapoptotic transcript variant of BCL-X and the GSC-promoting transcript variant of NCOR2 (Fig. 7M). It is worth noting that analysis of clinical outcomes revealed significant relationships between combined expression of METTL3 and splicing factors with GBM patient prognosis.

Taken together, this study provides an important mechanistic insight into how m⁶A methyltransferase METTL3 serves as an NMD regulator of splicing factors with potential clinical implications of alternative splicing events of BCL-XS and NCOR2. Our study also demonstrates that expression of METTL3 can be used to dissect the molecular differences between histologically similar GBM entities and to help predict GBM prognosis.

No potential conflicts of interest were disclosed.

Conception and design: F. Li, Y. Yi, Y. Miao, Q. Cao, W. Zhao

Development of methodology: F. Li, Y. Yi, Y. Miao, C. Zou, Y. Zheng, D. Chen, F. Zhi, Q. Cao, W. Zhao

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Miao, W. Long, C. Zou, K. Zhu, Q. Cao, W. Zhao

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Li, Y. Yi, Y. Miao, T. Long, Y. Zheng, X. Wu, J. Ding, Q. Cao, W. Zhao

Writing, review, and/or revision of the manuscript: F. Li, Y. Yi, Y. Miao, Q. Cao, W. Zhao

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): F. Li, Y. Miao, W. Long, T. Long, S. Chen, W. Cheng, Q. Xu, J. Wang, Q. Liu, F. Zhi, J. Ren, Q. Cao, W. Zhao

Study supervision: Q. Cao, W. Zhao

This work was supported by National Natural Science Foundation of China (81572766, 81702784, 81972651, 31771630, 81802974, 31771462, 81772614, 31471252, 31500813, and 31871009), Guangdong Innovative and Entrepreneurial Research Team Program (2016ZT06S029), the Natural Science Foundation of Guangdong Province (2017A030312009, 2017A030310228, 2014TQ01R387, 2017A030313134, and 2016A030313238), the Special funds for Dapeng New District Industry Development (KY20160309), and Natural Science Foundation of Jiangsu Province (BK20181156). Q. Cao is supported by U.S. Department of Defense (W81XWH-15-1-0639 and W81XWH-17-1-0357), American Cancer Society (TBE-128382), and NIH/NCI (R01CA208257).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.