N ⁶-methyladenosine promotes temozolomide resistance through mRNA stabilization in glioblastoma cells

Abstract

N ⁶-methyladenosine (m⁶A) regulates nearly every aspect of messenger RNA (mRNA) processing and function, impacting downstream gene expression programs. Changes in m⁶A have been implicated in many different types of cancer, and inhibition of m⁶A installation is emerging as a cancer therapeutic strategy. However, chemoresistance remains a significant clinical challenge in the treatment of glioblastoma (GBM). We established GBM cell culture models of acquired temozolomide (TMZ) resistance, analyzed the role of m⁶A in controlling resistance-associated pathways, and assessed the effects of METTL3 inhibition. We show that m⁶A stabilizes key genes and pathways promoting TMZ resistance, and that METTL3 inhibition can reverse this and restore TMZ sensitivity. These findings highlight that TMZ resistance can occur independent of a glioma stem cell population and that changes in m⁶A need not be driven by changes in METTL3 expression. Collectively, our results suggest that genes associated with TMZ resistance are stabilized by m⁶A methylation even as the majority of the transcriptome remains subject to m⁶A-mediated mRNA decay. Moreover, these data highlight METTL3 inhibition as a promising therapeutic approach to overcoming TMZ resistance in GBM.

Graphical Abstract

Graphical Abstract.

Impact statement.

Temozolomide (TMZ) is a mainstay of glioblastoma (GBM) treatment, yet resistance often limits its effectiveness. This work shows how an RNA modification—the acquisition of a methyl group at the N6-position of adenosine—stabilizes TMZ resistance-associated messenger RNAs in GBM cells. Small-molecule inhibition of the enzyme that installs this modification, METTL3, reduces expression of these genes and suppresses GBM cell growth. Because METTL3 inhibitors are already being tested in other cancers, these findings motivate further exploration of this strategy as a promising, potentially translatable approach to overcoming treatment resistance.

Introduction

Chemical modifications are critical regulators of the structure and function of diverse RNA species [1, 2]. Recent work has highlighted the importance of modifications in messenger RNA (mRNA) processing and function, revealing their roles in mRNA capping, splicing, polyadenylation, export, translation, and decay [3–5]. N⁶-methyladenosine (m⁶A) is the most prevalent internal mRNA modification, occurring at ∼0.5% of adenosines [6–8]. Transcriptome-wide mapping of m⁶A sites [6, 9] and identification of m⁶A regulatory machinery [5, 7, 10–18] have spurred rapid progress in our understanding of the m⁶A function. For example, m⁶A can facilitate mRNA decay via the binding of YTHDF2 and recruitment of deadenylation and decay machinery [12, 13], and this mechanism has regulatory roles in many biological contexts including embryonic development [19–21], stem cell differentiation [22, 23], and cancer [24–27].

As additional m⁶A-binding proteins and regulatory mechanisms have been discovered, a more complicated picture of m⁶A-mediated gene expression regulation has emerged [5, 10]. For example, the IGF2BP proteins have been reported to stabilize specific mRNAs in an m⁶A-dependent manner [16], but how or why a specific m⁶A site is bound by one binding protein versus another remains an open question. Recent work has also suggested that direct interactions between m⁶A and the ribosome can cause ribosome stalling, inducing mRNA decay via non-YTHDF2 mediated mechanisms [28–30]. Adding another layer of complexity, m⁶A methylation can also alter RNA structure, which can occlude or expose regulatory sequences depending on local RNA folding interactions [31, 32]. All of these mechanisms of m⁶A action—specific interactions with binding proteins, interactions with translation and other cellular machinery, changes in RNA structure, and perhaps others not yet identified—could be operating simultaneously in cells to shape the transcriptome. However, how these mechanisms coexist and how groups of transcripts might be differentially regulated by different mechanisms within the same cells remains unclear. Moreover, a downstream consequence of these multiple m⁶A regulatory modes is the fact that the functional outcomes of m⁶A methylation on a given mRNA can change depending on the specific gene expression profile of the cell [33].

Given such wide ranging impacts of m⁶A on gene expression regulation, it is not surprising that changes in mRNA m⁶A status and in expression levels of m⁶A-associated regulatory factors have been implicated in many human cancers [1, 34]. A common theme is that changes in m⁶A occupancy on key oncogene or tumor suppressor transcripts regulates their expression, which in turn influences tumor progression, treatment response, and overall patient survival. It is also clear, however, that m⁶A studies focused on the same disease can come to different conclusions as to the tumor suppressive or oncogenic effects of m⁶A in that context, which could be a consequence of different combinations of mechanisms highlighted above [35]. Nevertheless, the role of m⁶A in oncogenic transformation motivated the development of small molecule inhibitors of the m⁶A mRNA methyltransferase METTL3, such as STM2457 [36]. STC-15, an orally bioavailable derivative of STM2457, is in clinical trials for a variety of solid tumors [37]. METTL3 inhibitors have also been explored in the context of adjuvant therapy to enhance the effects of current standard of care chemotherapies [38–41].

Acquired resistance to frontline cancer therapies is a significant challenge in the clinic, and remains a particularly formidable hurdle in the treatment of glioblastoma (GBM) [42, 43]. After surgical resection, patients are treated with radiation therapy and the alkylating agent temozolomide (TMZ) [44]. While many patients respond initially, the overall 5-year survival rate is estimated to be <10%, in large part due to frequent tumor recurrence, initial or acquired resistance to TMZ, and limited alternative treatment options [43]. O⁶-methylguanine–DNA methyltransferase (MGMT) is an enzyme that repairs cytotoxic DNA lesions caused by TMZ, reversing TMZ-induced DNA damage. DNA methylation of the MGMT promoter silences its expression and is associated with improved response to TMZ in GBM patients [45, 46]. While efforts have been made to sensitize GBM to TMZ with MGMT small molecule inhibitors, these drugs demonstrated high toxicity in early clinical trials [47].

Despite rapid progress in our understanding of RNA modifications in cancer, including GBM [26, 48, 49], less is known about their roles in drug resistance. Given the wide ranging impacts of m⁶A methylation on gene expression programs in cancer and the significant challenges confronting GBM patients in the clinic, we were motivated to explore the functions of mRNA m⁶A methylation in TMZ resistance. We hypothesized that the large-scale cellular changes that occur during acquired drug resistance could be regulated by m⁶A, and that better understanding these processes could expose much needed novel therapeutic strategies to combat resistance.

Here, we develop cell culture models of acquired TMZ resistance in GBM to investigate whether key resistance-associated genes and pathways are directly regulated by m⁶A. In GBM cell lines, prolonged TMZ exposure induced durable resistance accompanied by massive upregulation of MGMT expression, possibly driven by demethylation of its promoter. Treatment with the METTL3 inhibitor STM2457 is lethal to both parental and TMZ-resistant (TMZ-R) cell lines, consistent with what has been observed in other cancer cell lines. We also find that MGMT mRNA is m⁶A-methylated and remarkably stable, helping to sustain high MGMT expression in TMZ-R cells. STM2457 treatment destabilizes MGMT mRNA, reducing its expression and re-sensitizing cells to TMZ. We also find that other genes that are upregulated in our TMZ-R cells tend to be regulated in a similar manner, suggesting that m⁶A-mediated mRNA stabilization of TMZ-resistance associated genes may represent a broader mechanism that helps to maintain stable TMZ resistance. Taken together, our findings provide a molecular foundation supporting the possibility that m⁶A inhibition may be a viable therapeutic avenue for drug-resistant GBM.

Materials and methods

Cell culture

U-87 MG, LN-229, T98G, and 293T cells were obtained from ATCC and U-251 MG cells were obtained from Sigma–Aldrich. LN-229 cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 5% fetal bovine serum, 1% penicillin/streptomycin, and 1 mM sodium pyruvate (Gibco). 293T cells were cultured in DMEM supplemented with 10% fetal bovine serum, 1% penicillin/streptomycin, and 1 mM sodium pyruvate. U-87 MG, U-251 MG, and T98G were cultured in Eagle’s Minimum Essential Medium (ATCC) supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin. All cell lines were cultured at 37°C and 5% CO₂ in a humidified atmosphere. TMZ-R cell lines were generated by continuous maintenance in media supplemented with 50 μM TMZ (Sigma–Aldrich) for at least 4 weeks.

Constructs

To generate pLVX-MGMT, the insert was synthesized as a double-stranded DNA fragment (gBlock; IDT) (Supplementary Table S6) and cloned into pLVX-M-puro (AddGene #125839) [50] at the BamHI/EcoRI sites using Gibson Assembly Master Mix (NEB) according to the manufacturer’s instructions. pLVX-M-puro was a gift from Boyi Gan (Addgene plasmid #125839; http://n2t.net/addgene:125839; RRID:Addgene_125839).

Generation of stable cell lines

To package lentivirus, 293T cells were seeded 2 × 10⁶ cells/plate in 10-cm plates, one plate per construct. Cells were transfected the following day with 2 μg of pLVX-M-puro or pLVX-MGMT along with 2 μg of psPAX2 and 1 μg of pMD2.G using Lipofectamine 3000 (Invitrogen) as directed by the manufacturer. Sixteen hours post-transfection, the media was changed and replaced with fresh media. Media containing packaged lentivirus was harvested 48 h post-transfection and was centrifuged at 300 × g and filtered using 0.45 μm polyvinylidene fluoride (PVDF) syringe filters (Millex) to remove contaminating cells. Virus titer was determined by serial dilution and a multiplicity of infection of <1 was used for transduction.

For viral transduction, U87 MG cells were seeded 3 × 10⁵ cells/well in a six-well plate and transduced the next day by replacing the media with media containing virus. Twenty-four hours post-transduction, media containing virus was removed and replaced with fresh media and cells were allowed to recover for 24 h prior to selection with 1 μg/ml of puromycin (Gibco). pMD2.G and psPAX2 were gifts from Didier Trono (Addgene plasmid # 12259; http://n2t.net/addgene:12259; RRID:Addgene_12259; Addgene plasmid # 12260; http://n2t.net/addgene:12260; RRID:Addgene_12260).

Targeted bisulfite sequencing

Genomic DNA from U87 MG, U87-R, LN-229, LN229-R, and T98G was extracted from 3 × 10⁶ cells per sample using the DNeasy Blood & Tissue Kit (QIAGEN) in biological triplicate. For bisulfite sequencing of the MGMT CpG island (chr10:129466635–129467496), genomic DNA samples were processed and analyzed using the Zymo Research Targeted Bisulfite Sequencing Service according to their standard workflow.

Cell viability assays

To measure the effects of TMZ or STM2457 on cell viability, the CyQUANT Cell Proliferation Assay (Invitrogen) was used according to the manufacturer’s specifications and fluorescence was measured using a SpectraMax iD3 microplate reader (Molecular Devices). Cells were seeded 2500 cells per well in 96-well culture plates and treated the following day by replacing the culture media with media containing drug. For TMZ, cells were treated with 50 μM for 72 h. For STM2457, cells were treated for 96 h with concentrations ranging from 0 to 200 μM. Dose-response curves were fit using nonlinear regression with a four-parameter logistic (variable-slope) model. IC₅₀ values and their 95% confidence intervals (CI) were calculated from the fitted curves. Statistical comparisons of IC₅₀ values between parental and resistant cell lines were performed using an extra sum-of-squared F test. Dimethyl sulfoxide (DMSO) concentrations were normalized across all treatment conditions. Individual experiments were performed with six technical replicates per condition.

Colony formation assays

For all cell lines, 1000 cells per well were seeded in six-well plates and cultured with media containing TMZ (0 or 50 μM) or STM2457 (0, 10, 25, and 50 μM) for 10 days (T98G) or 14 days (U-87 MG, LN-229, and U-251 MG lines). DMSO concentrations were normalized across all treatment conditions. At the end of observation, cells washed with phosphate buffered saline (PBS), fixed and stained in 1 ml per well 0.5% crystal violet; 25% methanol for 20 min with light shaking, then rinsed with MilliQ water.

Growth curve

U-87 MG parental and TMZ-R cells were cultured in the presence of 50 μM TMZ, 50 μM STM2457, 50 μM TMZ and 50 μM STM2457 combined, or DMSO alone (control). DMSO concentrations were normalized across all treatment conditions. Cells were counted and 1 × 10⁶ cells were replated in 10-cm dishes with fresh drug every 3 days and population doublings over the indicated time course were plotted.

RNA extraction, reverse transcription, and quantitative polymerase chain reaction

For reverse transcription quantitative PCR (RT-qPCR) and RNA sequencing (RNA-seq) experiments, total RNA was extracted from cells using Trizol (Invitrogen) according to the manufacturer’s instructions. RNA concentration was determined using the NanoDrop 8000 Spectrophotometer (Thermo Scientific). 0.5 μg of total RNA was reverse transcribed using iScript Reverse Transcription Supermix for RT-qPCR (Bio-Rad) according to the manufacturer’s instructions. Real time quantitative PCR (qPCR) was performed using the Bio-Rad CFX96 Real-Time PCR System (Bio-Rad) in 96-well PCR plates using 10 μl reactions composed of 5 μl SsoAdvanced Universal SYBR Green Supermix (Bio-Rad), 1 μl of four-fold diluted complementary DNA (cDNA) per reaction, and 200 nM of each forward/reverse primer. All qPCR reactions were performed in triplicate for each sample and analyzed using the 2^−ΔΔCT method and normalized to GAPDH or ACTB, as specified. All steps were performed using qPCR grade water (Fisher Scientific). Primer sequences are provided in Supplementary Table S6.

RNA sequencing

For RNA-seq from U-87 MG cells, RNA was first polyA selected using the Dynabeads mRNA DIRECT Purification kit (Invitrogen), followed by fragmentation using RNA Fragmentation Reagents (Invitrogen) and cDNA library preparation performed using TruSeq Stranded mRNA Library Prep (Illumina). Paired-end 100 bp sequencing was performed on a NovaSeq X Plus instrument (Illumina) with 50 million read pairs. For RNA-seq from LN229 cells, RNA was polyA selected and libraries prepared using the Watchmaker mRNA Library Prep kit (Watchmaker Genomics). Paired-end 100 bp sequencing was performed on a NovaSeq X plus instrument (Illumina) with 25 million read pairs. For all RNA-seq analysis, reads were trimmed using Trimmomatic (version 0.39) [51], aligned to GRCh38 using STAR (version 2.7.11a) [52], and counts determined using Salmon (version 1.4.0) [53]. Differential expression analysis was performed using DESeq2 [54]. An adjusted P-value of 0.05 was considered significant. All RNA-seq experiments were performed with three biological replicates for each condition. Pathway analysis was performed using ShinyGO 0.82 [55] and Enrichr [56–58], and the functional protein association network was generated using STRING v12.0 [59].

Quantification of DNA methylation

DNA methylation was detected and quantified using the OneStep PLUS qMethyl PCR Kit (Zymo) according to the manufacturer’s instructions. Genomic DNA was extracted from 3 × 10⁶ cells using the DNeasy Blood & Tissue Kit (QIAGEN). Twenty nanograms of genomic DNA was used for each reaction, and real-time PCR parameters matched those specified by the manufacturer except for the annealing temperature, which was 62.5°C. Three biological replicates were used, each real-time PCR reaction was performed in technical triplicate, and statistical analysis was performed as recommended by the manufacturer.

m⁶A immunoprecipitation

For m⁶A immunoprecipitation (IP) and quantitative PCR (m⁶A-IP/qPCR), total RNA was extracted from cells using Trizol (Introgen) and then polyA selected using the Dynabeads mRNA DIRECT Purification kit (Invitrogen). Then, 4 μg of polyA-selected RNA was subjected to m⁶A-IP using the EpiMark N⁶-Methyladenosine Enrichment Kit (NEB) and RT-qPCR was performed as described above. For each sample, prior to immunoprecipitation, 4.05 μg of RNA was diluted to 50 ng/μl. and 1 μl of the provided m6A(−) control RNA diluted 1:1000 was spiked into the RNA samples. After that, 50 ng of each sample was set aside and used as input. RT-qPCR results were plotted as fold enrichment in IP over input relative to the m6A(−) control RNA spike-in.

m⁶A-eCLIP

For m⁶A enhanced crosslinking immunoprecipitation (m⁶A-eCLIP), three biological replicates of U-87 MG parental and TMZ-R cells and two biological replicates of U87-R cells treated with 50 μM STM2457 for 96 h were collected. RNA extraction was performed using the QIAGEN RNeasy Midi Kit (QIAGEN) and m⁶A-eCLIP and bioinformatics analysis was performed by EclipseBio, using their proprietary procedures [60]. Unique molecular identifiers (UMI) were trimmed (umi_tools, cutadapt) [61, 62], filtered for repetitive elements and aligned to hg38 (STAR) [52], and PCR duplicates removed (umi_tools) [61]. Peaks and single nucleotide sites were called using CLIPper [63, 64] and PureCLIP [65], respectively. Motif enrichment analysis was performed using STREME (v.5.5.9) from the MEME Suite [66].

Immunoblotting

Cells were collected by scraping into ice-cold PBS and pelleted by centrifugation at 1000 × g for 5 min at 4°C. Cell lysis was performed by resuspending cell pellets in three pellet volumes of lysis buffer (50 mM Tris, pH 7.4, 300 mM NaCl, 1% NP-40, 0.25% deoxycholate, 1 mM dithiothreitol (DTT), and 1× SigmaFast protease inhibitor [Sigma–Aldrich]), incubating on ice for 30 min, and clarifying by centrifugation at 20 000 × g for 20 min at 4°C. Protein concentrations determined by Pierce bicinchoninic (BCA) protein assay. Then, 10 μg of each sample were loaded onto 4%–12% NuPAGE Bis–Tris gradient gels (Invitrogen) and run at 200 V for 50 min in 1× 3-(N-Morpholino)propanesulfonic acid (MOPS) buffer and then transferred to nitrocellulose membranes (Bio-Rad) by semidry transfer at 17V for 45 min with 2× NuPage Transfer Buffer with 10% methanol. Membranes were blocked for 1 h in 5% nonfat dry milk powder resuspended in Tris-Buffered Saline with 0.1% Tween-20 (TBST). Primary antibodies were diluted in 5% nonfat dry milk in TBST and incubated with rocking overnight at 4°C. Antibodies for MGMT (Cell Signalling Technology, 2739), METTL3 (Abcam, ab195352), METTL14 (Abcam, ab309096), γH2AX (Cell Signalling Technology, 9718), IGF2BP2 (Abcam, ab124930), and IGF2BP3 (Abcam, ab177477) were used at a 1:1000 dilution. Anti-GAPDH (Cell Signalling Technology, 5174) served as a loading control and was used at a 1:10 000 dilution. Following incubation with primary antibody, membranes were incubated in donkey anti-rabbit IgG secondary antibody DyLight 800 (Invitrogen) at a 1:5000 dilution at room temperature for 2 h with rocking and then imaged using the ChemiDoc MP Imaging System (Bio-Rad).d

Liquid chromatography coupled to tandem mass spectrometry

Total RNA was extracted from cells using Trizol (Invitrogen) according to the manufacturer’s instructions. mRNA was enriched by polyA selection using the Dynabeads mRNA DIRECT Purification kit (Invitrogen), followed by ribosomal RNA (rRNA) depletion with the NEBNext rRNA Depletion Kit v2 (NEB) and size selection for transcripts >200 nucleotides long with the RNA Clean & Concentrator-5 kit (Zymo). mRNA enrichment was validated by electrophoresis on a Agilent 2100 Bioanalyzer using the RNA 6000 Pico kit (Agilent). The RNA was then digested and analyzed using minor modifications to previously established procedures [67, 68]. Specifically, 100 ng of RNA was digested into nucleosides in digestion buffer [50 mM Tris–HCl, pH 8 (Sigma), 1 mM MgCl₂ (Sigma), 0.2 U/μl bezonase (Thomas Scientific), 0.002 U/μl phosphodiesterase I (Sigma), 0.02 U/μl alkaline phosphatase (Sigma)] at 37°C for 3 h. Samples and standard mixes prepared from pure nucleoside standards were filtered by centrifugation through Millex PVDF sterile syringe filters (0.22 μm, Millipore Sigma) at 10 000 × g for 1 min, transferred to autosampler vials (Thermo Scientific), and injected into a Shim-pack GIST C18 (2 μm, 2.1 × 50 mm) reverse phase high performance liquid chromatography column coupled to a Shimadzu 8050 NX Triple Quadrupole mass spectrophotometer. Liquid chromatography was performed using a linear gradient with 0.1% aqueous formic acid to 50% methanol with 0.1% formic acid over 8 min at 400 μl/min. Nucleosides were quantified using retention time and nucleoside to base ion mass transitions, compared to the standard curve obtained from pure nucleoside standards (282.3 > 150.15 for m⁶A; 268.3 > 136.25 for adenosine). A blank containing only digestion buffer was used to control for nucleoside contamination. The m⁶A level was calculated as the percent of m⁶A over adenosine averaged across three injections.

RNA stability assays

RNA decay over time was measured using the Click-iT Nascent RNA Capture Kit (Invitrogen) according to the manufacturer’s instructions. A total of 2 × 10⁵ U87-R cells were plated in 35-mm dishes and treated the next day by replacing the media with fresh media containing 50 μM STM2457 or DMSO. After 96 h of STM2457 treatment, cells were incubated with 0.5 mM 5-ethynyl uridine (EU) for 1 h. Media containing EU was then aspirated, cells were rinsed with fresh media, then grown in fresh media until being collected in Trizol at the indicated time points following EU removal. RNA was extracted using Trizol and subsequent capture of EU-labeled RNA and RT-qPCR was performed according to the manufacturer’s recommendations. Biotinylation of EU-labeled RNA by click reaction was performed using 1 μg EU-labeled RNA and 0.5 mM biotin azide. Then, 0.5 μg biotinylated RNA was captured following RNA precipitation using 50 μl of provided Dynabeads MyOne Streptavidin T1 magnetic beads. For RT-qPCR, cDNA was synthesized using the SuperScript VILO cDNA synthesis kit (Invitrogen) using the RNA captured on the beads as a template and 1 μl of undiluted cDNA was used for qPCR, which was performed as described above. RNA stability assay using Actinomycin D (ActD) was performed by seeding 6 × 10⁵ U87-R cells in 35-mm dishes and treating the following day with 5 μg/ml ActD.

SiRNA transfection

A total of 3.5 × 10⁵ U87-R cells were reverse transfected in six-well plates with 25 pmol of ON-TARGETplus human IGF2BP2 SMARTpool siRNA (L-017705-00-0005) or ON-TARGETplus nontargeting control pool siRNA (Horizon Discovery) using Lipofectamine RNAiMAX Transfection Reagent (Invitrogen) according to manufacturer instructions. After 72 h, cells were collected and the effect of IGF2BP2 knockdown on MGMT expression was assessed by immunoblotting and RT-qPCR.

Results

Generation of TMZ-R cultured glioma cell lines

To explore whether RNA modifications regulate TMZ resistance, we generated paired TMZ-sensitive and TMZ-R cultured glioma cells. We chose the parental cell lines based on (i) ease of manipulation and scalability in culture, and (ii) low basal MGMT expression, an indicator of baseline sensitivity to TMZ [45]. We chose U87 MG, LN229, and U251 MG cell lines as they exhibit both properties (Fig. 1A) [69]. In addition, we used T98G cells as a control that has high basal MGMT expression and is resistant to TMZ. To generate our paired TMZ-sensitive and TMZ-R cell lines, we continually cultured U87 MG, LN229, and U251 MG cells with 50 μM TMZ (Fig. 1B). Following an initial phase of high cell death, a small subpopulation of cells survived and achieved stable TMZ-resistance after ∼4 weeks. Thereafter, these resistant cells could be continually cultured in the presence of TMZ. We refer to these TMZ-R cell lines as U87-R, LN229-R, and U251-R, and their parental TMZ-sensitive counterparts by their standard names. To test for stable TMZ resistance, we assessed the effect of acute TMZ treatment on cell viability, clonogenicity, and cell growth of these cells and their TMZ-sensitive counterparts. While parental U87 MG, LN229, and U251 MG cells showed reduced viability, clonogenicity, and growth upon TMZ treatment, U87-R, LN229-R, and U251-R cells were unaffected (Figs 1C and D, and 2D).

Figure 1.

Modelling acquired TMZ resistance in GBM. (A) Cancer Cell Line Encyclopedia [69] RNA-seq gene expression data for MGMT (transcripts per million) in cell lines used in this study. (B) Strategy for generating TMZ-R cell lines. Cells were continuously maintained with 50 μM TMZ for >4 weeks. (C) Representative images of colony formation assays for parental and TMZ-R cells treated with 50 μM TMZ or DMSO. (D) Treatment with 50 μM TMZ for 96 h reduced cell viability of U87 MG parental, but not TMZ-R, cells (n = 6 technical replicates). (E) Volcano plots showing differentially expressed genes (DEGs) in U87-R or LN229-R compared to respective parental lines. Genes with adjusted P-values (padj) < 0.05 are considered differentially expressed. (F) MGMT mRNA fold change in TMZ-R cells relative to parental lines using GAPDH as an internal control (n = 3 biological replicates). (G) Western blot analysis for MGMT in parental and TMZ-R cells. A light and dark exposure of the same blot are shown to more easily observe weaker signals. GAPDH served as a loading control. (H) Population doublings in U87 MG cells with (pLVX-MGMT) and without (pLVX) MGMT overexpression, treated with DMSO alone or 50 μM TMZ. Data are mean ± standard deviation (SD). Two-tailed Student’s t-test; ***P <.001; ^****P <.0001; ns, not significant.

Figure 2.

METTL3 inhibition reduces cell growth and improves TMZ response in glioma cells. (A) Liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS) quantification of m⁶A levels expressed as percent m⁶A over unmodified adenosine in total RNA and mRNA from U87 MG cells treated with STM2457 for 72 h. mRNA was purified by poly(A) selection, rRNA depletion, and size selection (>200 nt). Data are mean ± SD, n = 3 injections (***P <.001, ns, not significant, two-tailed Student’s t-test). (B) Dose-response curves for 96 h STM2457 treatment in U87 MG (IC₅₀ = 93.6 μM; 95% CI = 88.8–98.7 μM), U87-R (IC₅₀ = 81.7 μM; 95% CI = 74.6–88.9 μM; P-value versus parental = 0.0059), LN229 (IC₅₀ = 75.1 μM; 95% CI = 68.8–81.0 μM), and LN229-R (IC₅₀ = 45.2 μM; 95% CI = 38.3–52.4 μM; P-value versus parental = 3.50 × 10^-8). Data are mean ± SD, n = 6 technical replicates. Statistical comparisons of IC₅₀ values between parental and resistant cell lines were performed using an extra sum-of-squared F test. (C) Representative images of colony formation assays of glioma cells treated with STM2457 (P = Parental; R = TMZ-R). (D) Population doublings in U87 MG and U87-R cells treated with DMSO alone, 50 μM TMZ, 50 μM STM2457, or a combined dose of 50 μM TMZ and 50 μM STM2457.

To interrogate mechanisms driving TMZ resistance, we performed RNA-seq in U87 MG, U87-R, LN229, and LN229-R cells to profile TMZ resistance-related gene expression changes transcriptome-wide (Fig. 1E and Supplementary Table S1). Overall, differentially expressed genes (DEGs) between TMZ-R and parental cells were enriched for Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways known to be associated with TMZ resistance, including PI3K-Akt, MAPK, and P53 signaling pathways, as well as extracellular matrix-receptor interactions (Supplementary Fig. S1A) [70–72]. DEGs were also significantly enriched for genes associated with TMZ in the Comparative Toxicogenomic Database (Supplementary Fig. S1B) [73]. Although U87 MG and LN229 cells have distinct gene expression profiles, in both cases the most highly upregulated transcript in the respective TMZ-R cell lines is the GBM prognostic marker MGMT (Fig. 1E). We validated these results in additional biological replicates from the U87 MG and LN229 lines and in U251 MG cells using quantitative PCR (RT-qPCR) (Fig. 1F). Critically, we also observe a correlated increase in MGMT protein by immunoblotting (Fig. 1G), suggestive of upregulated MGMT enzyme activity in the TMZ-R cells. This substantial MGMT upregulation reflects what is seen clinically in GBM patients who are resistant to TMZ, because high MGMT activity is critical for repairing TMZ-induced DNA damage [45, 46].

To determine whether these RNA and protein level changes in MGMT expression were driven by previously reported epigenetic mechanisms [74], we measured the DNA methylation status of its promoter by qPCR and bisulfite sequencing. Specifically, we focused on two genomic regions whose methylation statuses are associated with MGMT transcriptional activation in patients, demethylated regions 1 and 2 (DMR1 and DMR2) [74]. Interestingly, both DMR1 and DMR2 were demethylated in U87-R cells but not in LN229-R cells relative to their parental counterpart cell lines (Supplementary Fig. S1C), suggesting that multiple mechanisms can drive MGMT expression. We corroborated these findings by quantifying DNA methylation at CpGs throughout this region by bisulfite sequencing (Supplementary Fig. S1D and Supplementary Table S2). In light of this unexpected result, to further validate that MGMT expression is sufficient to drive TMZ resistance we stably expressed MGMT in parental U87 MG cells and grew them with and without TMZ over multiple passages (Supplementary Fig. S1E). Cells transduced with pLVX-MGMT grew equally well in the presence and absence of TMZ, while cells transduced with the empty vector remained susceptible to TMZ treatment (Fig. 1H). Taken together, these data gave us confidence that we can use our TMZ-R cells as a cellular model to reveal molecular mechanisms that regulate TMZ resistance.

TMZ-R cell lines are sensitive to METTL3 inhibition

METTL3-mediated m⁶A methylation of mRNA has been implicated in a variety of cancers (reviewed in [1]), spurring development of small molecule inhibitors of METTL3 such as STM2457 [36]. Inhibition of METTL3 can also improve sensitivity to other therapies [38–41], and previous work has shown that loss of METTL3 function can be lethal [22, 23]. We were therefore curious to see how METTL3 inhibition and loss of m⁶A would impact our glioma cell lines, particularly the TMZ-R lines. To test STM2457 efficacy and tune dosing for subsequent experiments, we treated U87 MG cells with 10 and 50 μM STM2457 for 72 h. We analyzed bulk level m⁶A changes in total RNA and mRNA by LC-MS/MS. To isolate highly purified mRNA, we performed poly(A) selection, rRNA depletion, and then size selection to remove small RNAs (Fig. 2A); we validated successful purification by automated electrophoresis. Both the 10 μM and 50 μM doses of STM2457 resulted in a significant reduction in m⁶A level in mRNA (Fig. 2A) with minimal effects on cell viability in this time frame (Fig. 2B and Supplementary Fig. S2A). In contrast, METTL3 inhibition does not affect the m⁶A level in total RNA, which contains m⁶A from abundant rRNAs that is installed by other methyltransferases [75–78]. This is expected given the small fraction of total RNA that mRNA represents, and is consistent with the previously reported specificity of this inhibitor for the mRNA m⁶A methyltransferase METTL3 [36].

Because METTL3 inhibition has been previously shown to have anti-cancer effects [36, 40, 79, 80], we further investigated the effects of different STM2457 doses and treatment times on our glioma cell lines. We also included T98G cells, which express high levels of MGMT (Fig. 1A) and are resistant to TMZ at baseline. Despite their dramatically different responses to TMZ (Figs 1C and D, and 2D), STM2457 treatment reduces cell viability, clonogenicity, and growth in both the parental and TMZ-R lines (Fig. 2B–D and Supplementary Fig. S2A). We also noted a small but statistically significant increase in sensitivity to STM2457 in the TMZ-R lines relative to their TMZ-sensitive counterparts, indicated by the respective IC₅₀s in DNA-dye based cell viability assays (Fig. 2B and Supplementary Fig. S2A). We confirmed these results with a second, more potent, METTL3 inhibitor STM3006, which had similar effects at lower doses (Supplementary Fig. S2B). Cell viability in the presence of STM2457 highly depends not only on the dose, but also on treatment time and density of cells at the time of treatment. The sparse cell density and 14-day treatment time used for colony formation assays increased sensitivity to STM2457, and under these conditions the 50 μM dose severely inhibited colony formation in parental and TMZ-R cells (Fig. 2C). We also measured the ability of cells to grow over multiple passages in TMZ, STM2457, or both drugs together. Though U87-R cells are insensitive to the presence of TMZ alone, their doubling rate drops in the presence of STM2457 (Fig. 2D). This effect is even more pronounced when TMZ is combined with STM2457 (Fig. 2D), suggesting that METTL3 inhibition can resensitize cells to TMZ.

Bulk level m⁶A does not change with chronic or acute TMZ treatment

Inspired by observations in other cancers, we suspected that critical transcripts in our glioma cells are m⁶A-methylated [26, 35, 48, 49]. We first measured bulk levels of m⁶A in parental and TMZ-R cells, and did not detect significant differences in m⁶A levels in total RNA or mRNA between U87 MG and U87-R cells, nor the levels of METTL3 and its obligate binding partner METTL14 (Supplementary Fig. S2C–E). We next wondered whether acute TMZ treatment could result in bulk level m⁶A changes, and performed LC-MS/MS on RNA from U87 MG cells acutely treated with 100 μM TMZ for 72 h. To get a detailed picture of possible changes in different RNA sub-populations, we used a more comprehensive series of purifications that resulted in five different subpopulations of RNA: total RNA, short RNA (transfer RNA and small RNA species shorter than 200 nucleotides), long RNA (mRNA, rRNA, and long noncoding RNA [lncRNA] >200 nucleotides long), long/polyA + RNA (enriched for polyadenylated RNAs including mRNA, but with possible remaining rRNA contamination), and long/rRNA − RNA (depleted for rRNA species, enriched for mRNAs and lncRNAs regardless of polyA status) (Supplementary Fig. S2F). Once again, there were minimal detectable differences, although we did observe a small but statistically significant increase in m⁶A levels in the long/rRNA-fraction, perhaps suggesting that acute TMZ treatment may result in elevated m⁶A methylation of some mRNA or lncRNA transcripts (Supplementary Fig. S2F).

METTL3 Inhibition alters expression of key regulatory pathways in glioma cells

We next considered whether a subset of transcripts critical to glioma cell growth and TMZ resistance are affected by METTL3 inhibition. We performed RNA-seq in U87 MG, U87-R, LN229, and LN229-R cells treated with or without 50 μM STM2457 for 96 h (Supplementary Table S3, Fig. 2A and B, and Supplementary Fig. S2A). Overall, the majority of DEGs were upregulated in response to STM2457 treatment (Fig. 3A and Supplementary Fig. S3A), consistent with previous observations and the established role of m⁶A in destabilizing mRNA [12, 13]. KEGG pathway and Gene Ontology Biological Pathway (GOBP) analysis revealed that genes upregulated by STM2457 in TMZ-R cells were enriched for pathways related to autophagy, apoptosis, cell death and catabolism, whereas downregulated genes were enriched for pathways related to cell division and development (Fig. 3B, and Supplementary Fig. S3B and C). Genes downregulated by STM2457 treatment were also enriched for pathways important for promoting TMZ resistance, including DNA repair pathways, hypoxia response/HIF-1 signaling, PI3K-Akt signaling, TNF signaling, and glycolysis. This observation is consistent with our findings that STM2457 treatment decreased growth of glioma cells and improved TMZ sensitivity of resistant cells (Fig. 2B–D).

Figure 3.

Transcriptome-wide regulation by m⁶A in TMZ-R GBM. (A) Volcano plots showing differentially expressed genes (padj < 0.05) in U87-R or LN229-R treated with 50 μM STM2457 for 96 h. (B) KEGG pathways enriched for genes differentially expressed in response to STM2457 treatment in U87-R cells. (C) Total number of single-nucleotide m⁶A-eCLIP sites. Bars represent the number of reproducible sites in each condition while points represent values for individual replicates with mean ± SD (^****P <.0001; ns, not significant; two-tailed Student’s t-test). (D, E) Genomic distribution of m⁶A. (D) Percent of reproducible m⁶A sites in each condition at each feature. (E) Representative metagene analysis showing distribution of reproducible m⁶A-eCLIP peaks in U87 MG parental (magenta) and U87-R (turquoise). (F) Volcano plot showing differential enrichment of m⁶A-eCLIP sites in U87-R cells treated with STM2457 compared to untreated. (G) Volcano plot showing the effect of STM2457 treatment on the expression of transcripts which have reproducible m⁶A sites at STM2457-sensitive m⁶A-eCLIP peaks (reduced enrichment of log₂ fold change ≥ 1; P-value <.05) in U87-R cells.

To gain further insight into the direct mechanisms by which m⁶A regulates GBM tumorigenesis and TMZ response, we performed m⁶A enhanced crosslinking and immunoprecipitation (m⁶A-eCLIP) in U87 MG, U87-R, and U87-R treated with STM2457 to reveal transcript- and site-specific differences in m⁶A status (Supplementary Tables S4 and S5). In brief, RNA was fragmented and incubated with anti-m⁶A antibody before being crosslinked by ultraviolet light, and antibody-bound RNAs were then identified by high throughput sequencing [81, 82]. Using this method, we were able to map m⁶A in the transcriptome at high resolution. To increase the rigor of our experiment, we used a sub-lethal STM2457 treatment as a specificity control to help filter out non-METTL3-mediated m⁶A sites in these anti-m⁶A immunoprecipitations.

Consistent with our LC-MS/MS findings that bulk m⁶A levels are similar between U87 MG and U87-R cells (Supplementary Fig. S2C), m⁶A-eCLIP revealed no significant difference in the number of called m⁶A sites between the two cell lines (Fig. 3C). Additionally, there were no discernible differences in m⁶A deposition patterns between U87 MG and U87-R cells. In both cell lines, the majority of m⁶A sites were found at canonical DRACH motifs (where D = A, G, or U; R = A or G; and H = A, C, or U) located at the end of coding sequences and early in 3′UTRs near stop codons (Fig. 3D and E, and Supplementary Fig. S3D), which is typical of METTL3-mediated m⁶A methylation [6, 9]. STM2457 treatment significantly reduced the number and enrichment of m⁶A sites (Fig. 3C and F, and Supplementary Table S5). Of the m⁶A sites that remain in STM2457-treated U87-R cells, a smaller proportion are in coding sequences, 3′UTRs, and at DRACH motifs compared to untreated U87-R cells (Fig. 3D and E, and Supplementary Fig. S3D). Intersecting our m⁶A-eCLIP and RNA-seq datasets, we defined a set of high confidence m⁶A sites based on reduced enrichment in STM2457 treated samples relative to controls, and examined the effect of METTL3 inhibition on their expression level (Fig. 3G). m⁶A-methylated transcripts were both upregulated and downregulated by METTL3 inhibition, though the majority of transcripts were upregulated, as expected (Fig. 3G). Taken together, these results validate that STM2457 treatment effectively disrupts METTL3-mediated m⁶A deposition in glioma cells, and reveals that key cellular processes central to TMZ resistance are regulated by m⁶A.

MGMT mRNA is m⁶A-methylated and stabilized via binding of IGF2BP2

We next investigated whether m⁶A plays a direct role in TMZ resistance. We examined m⁶A-eCLIP single nucleotide sites which were differentially enriched between U87 MG and U87-R cells (Supplementary Table S4). Strikingly, two of the most highly significantly enriched m⁶A sites in U87-R cells were located on MGMT (Fig. 4A). At the transcript level, MGMT was also significantly enriched in U87-R m⁶A-eCLIP but not in STM2457-treated U87-R m⁶A-eCLIP (Fig. 4B). Both called m⁶A single-nucleotide sites lie within an m⁶A-eCLIP enrichment peak, are at DRACH motifs, and are positioned within the coding sequence proximal to the stop codon. To further validate MGMT m⁶A-methylation, we immunoprecipitated (IP) polyA-enriched RNA isolated from U87-R, LN229-R, and T98G cells with an anti-m⁶A antibody and quantified enrichment by RT-qPCR. MGMT mRNA was significantly and reproducibly enriched in the anti-m⁶A IP samples in all cell lines, at a level similar to that of other known m⁶A-methylated RNAs such as MALAT1 (Fig. 4C and Supplementary Fig. S4A–D). STM2457 treatment resulted in no significant enrichment of MGMT mRNA by m⁶A-IP (Supplementary Fig. S4D).

Figure 4.

MGMT is post-transcriptionally regulated by m⁶A. (A) Volcano plot of m⁶A-eCLIP sites differentially enriched in U87-R cells compared to U87 MG. m⁶A sites in MGMT are labeled. (B) Integrative Genomics Viewer (IGV) snapshot of input and m⁶A-eCLIP reads at the 3′ end of the MGMT coding sequence in untreated and STM2457-treated U87-R samples. Reproducible U87-R m⁶A-eCLIP sites and DRACH motifs are labeled. (C) m⁶A-IP/qPCR for MGMT in U87-R and LN229-R cells. Fold enrichment over input normalized to an unmodified spike-in control RNA is shown. (D, E) The effect of 96 h, 50 μM STM2457 treatment on MGMT expression determined by RT-qPCR [normalized to ACTB; panel (D)] and western blotting [panel (E)]. Data are mean ± SD of n = 3 biological replicates. (F) Percent of remaining EU-labeled MGMT mRNA at increasing times since EU removal in U87-R cells pre-treated for 96 h with 50 μM STM2457 or DMSO, quantified by RT-qPCR (n = 3 technical replicates). (G, H) Effect of 72 h IGF2BP2 siRNA knockdown in U87-R cells on MGMT mRNA (G) and protein (H) level. (G) Fold change is relative to matched siControl replicate and normalized to GAPDH (n = 6 biological replicates). Two-tailed Student’s t-test; *P <.05; **P <.01; ***P <.001; ^****P <.0001; ns, not significant.

The central role of MGMT in mediating GBM TMZ resistance, its confirmed m⁶A modification in resistant cells, and the canonical role of m⁶A as repressing gene expression presented a conundrum, leading us to consider an alternative role for m⁶A in enhancing its expression. To explore this idea, we tested the effects of STM2457 on MGMT expression. We tested multiple STM2457 doses (Supplementary Fig. S4E) and treatment times (Supplementary Fig. S4F), and found that treatment for 96 h with 50 μM STM2457 reduced MGMT expression at the RNA (Fig. 4D) and protein (Fig. 4E) levels in U87-R, LN229-R, U251-R cells, and T98G cells. We observed the same trend with STM3006, and found similarly dose-responsive reductions in MGMT RNA expression (Supplementary Fig. S4G). In other words, m⁶A methylation enhances MGMT expression, rather than repressing it as is typical of the YTHDF2-associated mRNA decay mechanism [12, 13]. The consistent loss of MGMT mRNA and protein with STM2457 treatment suggests that m⁶A has a stabilizing effect, which we confirmed using RNA stability assays. Transcription inhibition-based measurements of RNA decay revealed that MGMT is a highly stable mRNA (Supplementary Fig. S4H). To avoid the toxicity of long-term transcription inhibition, we used EU labeling in a pulse-chase format and observed that MGMT mRNA is destabilized upon STM2457 pre-treatment compared to DMSO control, with an observed half life of ∼8 h versus ∼20 h, respectively (Fig. 4F and Supplementary Fig. S4I). These results are partially consistent with recent work in a glioma stem cell (GSC) model, which found that MGMT is m⁶A methylated, but did not map the site [49]. METTL3 knockdown in this GSC model also reduced MGMT levels, but interestingly the high stability of MGMT was not observed in this system. Shi et al.also attributed MGMT m⁶A methylation to increasing levels of METTL3 enzyme over the course of GSC differentiation, while here we observe highly methylated MGMT in the absence of any significant expression changes in m⁶A regulatory enzymes.

While numerous examples of m⁶A-mediated decay have been reported, studies in acute myeloid leukemia were the first to demonstrate that a subset of m⁶A-methylated mRNAs can be stabilized by a second family of m⁶A-binding proteins, the IGF2BP family [10, 16]. Interestingly, we find predicted binding sites for IGF2BP2 and IGF2BP3 in the MGMT 3′UTR [83]. Our RNA-seq analyses suggest that both IGF2BP2 and IGF2BP3 are expressed in our TMZ-R cell lines while IGF2BP1 is not (Fig. 3A and Supplementary Table S1). IGF2BP3 protein levels varied more dramatically across cell lines, and dropped significantly in LN229-R cells relative to parental LN229 cells (Supplementary Fig. S4J), so we focused on IGF2BP2. We knocked down IGF2BP2 with siRNA and found that its depletion reduced MGMT mRNA and protein levels (Fig. 4G and H). Like MGMT, IGF2BP2 mRNA has m⁶A sites with decreased enrichment in STM2457-treatment U87-R (Supplementary Fig. S4K and Supplementary Table S4) and decreased expression in response to STM2457 treatment (Supplementary Fig. S4L), suggesting both transcripts are regulated similarly by m⁶A in TMZ-R glioma cells. Taken together, our findings suggest that MGMT stabilization by m⁶A is mediated at least in part by IGF2BP2 either via direct binding or an indirect interaction.

A broader mechanism of m⁶A-mediated stabilization of TMZ-R associated transcripts

Given the complex and interconnected signaling mechanisms that have been previously associated with TMZ resistance, we were curious whether other TMZ resistance-associated transcripts beyond MGMT are also stabilized by m⁶A. Overall, gene expression changes caused by STM2457 treatment in parental and TMZ-R cells were strongly correlated (Supplementary Fig. S5A), which is consistent with the similar m⁶A profiles in parental and resistant cells (Fig. 3C–E). However, we noticed that STM2457-induced gene expression changes in TMZ-R cells had a slight negative correlation with resistance-induced expression changes in TMZ-R cells relative to their parental counterpart (Supplementary Fig. S5B), and that this is not the case with STM2457-induced gene expression changes in the parental cell lines (Supplementary Fig. S5C). Indeed, genes significantly upregulated in TMZ-R cells were disproportionately downregulated by METTL3 inhibition (Fig. 5A), suggesting that genes that were highly upregulated in TMZ-R cells were also stabilized by m⁶A, similar to MGMT. In contrast, genes that were significantly downregulated in TMZ-R cells were largely upregulated by METTL3 inhibition (Supplementary Fig. S5D), possibly reflecting m⁶A-mediated mRNA decay. Turning back to our m⁶A-eCLIP dataset (Fig. 4), we looked more closely at the distinct subsets of transcripts that are m⁶A methylated and whose levels are elevated or reduced upon STM2457 treatment. We did not observe differences in sequence motif (Supplementary Fig. S5E), suggesting that additional factors may be involved in the different fates of these transcripts. There were also no major differences in deposition patterns between these groups, although STM2457-upregulated transcripts were slightly more likely to have m⁶A sites in coding sequences, while STM2457-downregulated genes were more likely to have sites in 3'UTRs (Supplementary Fig. S5F).

Figure 5.

Regulation of TMZ resistance-associated genes by m⁶A. (A) Volcano plots depicting the effect of STM2457 treatment on expression of genes which are significantly upregulated (adjusted P-value <.05) in TMZ-R cells. (B) KEGG pathways enriched for genes which are upregulated in U87-R cells compared to parental U87 MG cells and downregulated in response to STM2457 treatment in U87-R cells. (C, D) IGV snapshot of input and m⁶A-eCLIP reads at the 3′ ends of PIK3R3 (C) and HIF1A (D) in untreated and STM2457-treated U87-R samples. Loci of single nucleotide m⁶A-eCLIP sites in U87-R samples are labeled.

To explore whether genes that behave like MGMT (expression up in TMZ-R relative to the parental line and down with STM2457 treatment) are also associated with TMZ resistance, we looked more closely at the functional annotations associated with this gene set. KEGG pathway analysis revealed that this gene set was enriched for pathways known to be involved in TMZ resistance, including PI3K-Akt signaling, Ras and Rap signaling, extracellular matrix-receptor interactions, and HIF-1 signaling (Fig. 5B and Supplementary Fig. S5G). PI3K-Akt signaling is frequently amplified in GBM and contributes to TMZ resistance by promoting cell survival, proliferation, and angiogenesis [84]. PI3K-Akt signaling is tightly interconnected with upstream Ras/Rap signaling. While Ras directly activates PI3K in response to growth factors, Rap proteins also modulate PI3K activity via regulation of focal adhesion and integrin signaling. Additionally, the PI3K-Akt pathway positively regulates hypoxia-inducible factor-1 (HIF-1) signaling and MGMT expression, both of which promote cell survival in response to chemotherapy [85–88]. Furthermore, like MGMT, many of these key regulatory genes are regulated by m⁶A directly, including PIK3R3 (a regulatory subunit of PI3K) (Fig. 5C). Interestingly, m⁶A was previously shown to promote PIK3R3 mRNA stability via IGF2BP1 and IGF2BP2 in renal cancer, suggesting MGMT and PIK3R3 may be regulated by similar mechanisms [89]. HIF1A, the alpha subunit of the HIF-1 transcription factor, TGFA, a growth factor which activates PI3K signaling, and SLC2A1 (GLUT1), a critical HIF-1 target and proposed GBM therapeutic target, were also among transcripts upregulated in TMZ-R cells and directly stabilized by m⁶A (Fig. 5D and Supplementary Fig. S5H). Multiple previous studies have also linked m⁶A-mediated regulation to DNA damage response and repair pathways and our RNA-seq analysis also suggested that some of these pathways are downregulated with STM2457 treatment (Supplementary Fig. S3B and C) [90–92]. Consistent with this, we observe elevated γH2AX levels upon STM2457 treatment in LN229 cells, suggesting elevated levels of DNA damage upon m⁶A inhibition (Supplementary Fig. S5I). These results suggest that a broader network beyond MGMT, including PI3K-Akt signaling, DNA damage responses, and other associated pathways, are stabilized by m⁶A methylation to maintain TMZ resistance. Overall, this supports a model in which m⁶A methylation regulates key pathways that are critical for maintaining TMZ resistance, and METTL3 inhibition resensitizes cells to TMZ not only by destabilizing MGMT, but by negatively regulating this cohort of mRNAs.

Discussion

Here we show that m⁶A stabilizes a subset of transcripts associated with acquired TMZ resistance in glioma cell culture models. In particular, we show that MGMT, the DNA repair enzyme responsible for repairing TMZ-induced DNA damage, is destabilized by inhibition of the mRNA m⁶A methyltransferase METTL3 by STM2457. While most STM2457-responsive transcripts in these cells demonstrate canonical m⁶A-mediated regulation and are stabilized by METTL3 inhibition, TMZ-R-associated transcripts, like MGMT, are destabilized by METTL3 inhibition. Intriguingly, this suggests that while m⁶A-mediated regulation is not dramatically re-wired in these cells, there may be TMZ-R specific mechanisms that alter the functional impact of m⁶A. The identity of TMZ-R specific genes differs across TMZ-R lines, likely reflecting their different basal gene expression patterns. However, MGMT is the most highly upregulated gene in all cases and the functions associated with TMZ-R DEGs are shared, including PI3K-Akt signaling and related cellular pathways. While the differential m⁶A methylation of MGMT is likely driven by its strong transcriptional activation in TMZ-R cells compared to its near undetectable levels in the TMZ-sensitive cells, it is possible that some TMZ-R associated mRNAs are selectively methylated in one setting and not another, as has been demonstrated in other systems [91, 93, 94]. While further mechanistic work is needed to reveal the molecular mechanisms driving these changes, our data suggest that there may be specific features of these genes or feedback mechanisms involving m⁶A reader proteins that underlie this coordinated behavior.

Supporting this idea, we found that transcripts encoding m⁶A regulatory enzymes and binding proteins responded similarly to acquired TMZ resistance and STM2457 treatment compared to their parental line (Supplementary Fig. S4L). Previous work has proposed that METTL3 is upregulated in TMZ-R cells, but this is inconsistent with our data showing stable METTL3 expression and bulk level m⁶A methylation [49]. Instead, this points to a small subset of transcripts that contain altered m⁶A sites, which is also consistent with our RNA-seq data. One possible explanation for this is that the stabilizing effect of m⁶A increases the proportion of methylated transcripts of these TMZ-R mRNAs. But the question remains, why are TMZ-R transcripts stabilized while the majority of mRNA is destabilized by m⁶A? The majority of m⁶A binding proteins are also unchanged between parental and TMZ-R cell lines. Intriguingly, the relative expression of m⁶A-binding proteins does seem to change in response to STM2457 treatment at the RNA level, with the levels of YTHDF1 and YTHDF2 increasing, and IGF2BP2 and IGF2BP3 decreasing (Supplementary Fig. S4L), suggesting possible m⁶A-mediated feedback mechanisms. Paradoxically, however, changes at the protein level are not apparent (e.g. for IGF2BP2 and IGF2BP3, Supplementary Fig. S4I), suggesting that there are additional factors involved. Overall, these STM2457-induced changes would result not only in altered m⁶A status of TMZ resistance-related mRNAs such as MGMT, but could also change the functional consequences of remaining m⁶A sites via the changes in relative levels of m⁶A-binding proteins.

Beyond the effects of m⁶A-binding proteins, recent work has revealed that interactions between mRNA m⁶A sites and translating ribosomes can also regulate mRNA stability [28–30]. Importantly, these studies revealed that the precise location of the m⁶A site near a stop codon, specifically whether it is in the coding sequence or in the 3′UTR, may be a factor in whether m⁶A stabilizes or destabilizes mRNAs. We noted that m⁶A-stabilized transcripts tended to have a slightly higher proportion of sites in 3′UTRs, while m⁶A-destabilized transcripts had more CDS sites (Supplementary Fig. S5F), consistent with this model [28]. MGMT m⁶A-eCLIP peaks span both the coding sequence and 3′UTR, but the two MGMT m⁶A sites were called in the 3′ end of the coding sequence, with one occurring just upstream of the stop codon. This location of m⁶A adjacent to the stop codon strongly suggests that active translation of m⁶A-methylated MGMT may also influence whether m⁶A has a stabilizing or destabilizing effect, but the involvement of this mechanism requires more detailed future study. Future studies will also need to consider known post-translational mechanisms of MGMT regulation and the fact that it operates by a suicide mechanism, so MGMT is itself inactivated and turned over more quickly when high levels of alkylation damage are present [95, 96].

Our finding that m⁶A-mediated feedback mechanisms regulate the development and maintenance of TMZ resistance could also have clinical implications. The transcriptional activation of key TMZ-R associated genes, including MGMT, is likely primarily driven by epigenetic regulatory mechanisms. MGMT promoter methylation is used as a proxy for MGMT expression levels, and is a prognostic indicator for patient outcome in response to TMZ [45, 46]. Lack of DNA mismatch repair activity has broadly been shown to render patients more resistant to treatment with alkylating agents. Despite this, little is known about MGMT expression regulation in response to chronic TMZ treatment. In fact, in our own measurements MGMT promoter methylation status did not always correlate with MGMT expression (Supplementary Fig. S1C and D), suggesting that MGMT activation can occur by multiple mechanisms. Longitudinal studies in patients have attempted to explore this question but are impeded by limited availability of patient samples at recurrence [97]. A predominant current model is that GSCs drive recurrence because they are more chemoresistant [98], and previous work has compared MGMT m⁶A methylation in GSCs relative to differentiated cells [49]. Here we show that MGMT status changes in direct response to chronic TMZ treatment, that high levels of MGMT are sufficient to induce TMZ resistance, and that the high levels of m⁶A methylation on upregulated MGMT transcripts occurs in the background of stable levels of METTL3 expression.

Though alkylation agents are now less commonly used in the clinic, TMZ remains standard of care treatment for GBM patients. Because MGMT is central to reversing TMZ-induced alkylation damage, efforts have been made to target MGMT activity and expression as a potential path to re-sensitizing tumors that have become TMZ resistant [43]. O⁶-benzylguanine, an irreversible inhibitor of MGMT activity, was tested in Phase I and II clinical trials for its ability to re-sensitize tumors to TMZ but demonstrated high levels of toxicity [47, 99]. While alternative approaches have been proposed to re-methylate the MGMT promoter and shut down its transcription [100], this is not yet feasible in patients. The data we present here provide a molecular basis for pharmacological inhibition of METTL3 as a potential strategy to sensitize GBM tumors to TMZ; however, we note that GBM is currently excluded from ongoing clinical trials of METTL3 inhibitors [37].

Our work also opens many other interesting avenues for future investigation. Among the pathways upregulated in STM2457-responsive DEGs, we noted the NF-kappa B, Toll-like receptor, and NOD-like receptor signaling pathways, all key immune signaling pathways (Supplementary Fig. S3B and C). How m⁶A-targeted therapeutics influence immune responses could significantly impact their effectiveness and applicability across different cancers, so this represents an important question for future study. Furthermore, all of the cell lines used in this study have wild type isocitrate dehydrogenase (IDH) status, but a small proportion of GBM tumors have mutations in these key metabolic enzymes. Specific IDH mutations are known to have neomorphic activity that results in production of the reported oncometabolite 2-hydroxyglutarate [101]. 2-hydroxyglutarate is a known inhibitor of ɑ-ketoglutarate-dependent enzymes, including DNA and RNA demethylases, which could add an additional layer of feedback within the m⁶A regulatory network and influence the effects of METTL3 inhibition. Additional tricarboxylic acid (TCA) cycle intermediates have been shown to have similar effects and could influence the m⁶A landscape in gliomas and other cancers [102, 103]. It will also be interesting to dissect at a molecular level how MGMT m⁶A sites interact with m⁶A binding proteins and translation machinery, as these events would presumably be mutually exclusive. More broadly, it is not clear how different m⁶A-binding proteins cooperate or compete for binding to a specific m⁶A site, and MGMT could provide an interesting case study to investigate this. Finally, our work underscores the value of detailed molecular dissection of m⁶A function on specific transcripts, which can reveal diverse modes of m⁶A-mediated gene expression regulation with meaningful clinical implications.

Supplementary data

Supplementary data is available at NAR Cancer online.

Conflict of interest

S.N. holds equity and is a member of the scientific advisory board of RNA Connect, Inc.

Funding

This work was supported by a Distinguished Scientist Award from the Sontag Foundation, an NIGMS MIRA award (R35GM146919), and NHGRI R01HG011868 to S.N.. E.D. was previously supported by training grants T32GM007499 and T32GM145469, and is currently supported by an NIH NRSA F31 fellowship (F31CA306101). R.P. was supported by an award from the American Cancer Society Center for Innovation in Cancer Research Training. Research reported in this publication was also supported by the National Institute of General Medical Sciences of the National Institutes of Health under award number 1S10OD030363-01A1 (Yale Center for Genome Analysis).

Data availability

Raw and processed data files from all high throughput sequencing experiments have been deposited in the NCBI Gene Expression Omnibus (GSE306559, GSE306560).