m⁶A-dependent glycolysis enhances colorectal cancer progression

We used three Cohorts of patients with colorectal cancer who underwent surgery between 2012 and 2019. Cohort 1 (fresh tissues and paraffin-embedded tissues) were from Xuzhou Central Hospital, Xuzhou Medical University; Cohort 2 (paraffin-embedded tissues) were from the Eastern Campus of Renji Hospital, Shanghai Jiao Tong University School of Medicine; Cohort 3 (fresh tissues) were from the Western Campus of Renji Hospital, Shanghai Jiao Tong University School of Medicine. The study protocol was approved by the ethics committee of Shanghai Jiao Tong University School of Medicine. Written informed consent was obtained from all participants in this study. All the research was carried out in accordance with the provisions of the Declaration of Helsinki of 1975. None of these patients had received radiotherapy or chemotherapy prior to surgery.

Polyadenylated mRNA was purified by GenElute™ mRNA Miniprep Kit (Sigma, St. Louis, MO) from previously isolated total RNA. The m⁶A dot blot assay was performed as previously described [25]. The global m⁶A levels in mRNA were measured with EpiQuik m⁶A RNA Methylation Quantification Kit (Colorimetric) (Epigentek, Farmingdale, NY) following the manufacturer’s protocol. The detailed m⁶A dot blot and m⁶A quantification protocols were described in the Supplementary Materials and Methods.

The m⁶A-immunoprecipitation and library preparation was performed according to a published protocol [10]. Real-time PCR was carried out following m⁶A-IP to quantify the changes to m⁶A methylation of a certain target gene. The detailed MeRIP and MeRIP-qPCR protocols were described in the Supplementary Materials and Methods.

Glucose Uptake, L-Lactate Colorimetric assay, Hexokinase Colorimetric assay, Seahorse XF Glycolysis Stress Test and Seahorse XF Cell Mito Stress Test were detailed described in the Supplementary Materials and Methods.

RNA Immunoprecipitation (RIP) assays were conducted using the Magna RIP Kit (Millipore, New Bedford, MA) and the detailed protocol was described in the Supplementary Materials and Methods.

Statistical analyses were carried out using the program R (www.​r-project.​org). Recurrence-free survival was evaluated by Kaplan-Meier survival curve and Log-rank tests. Statistical significance was assessed by unpaired two-tailed Student’s-tests. Single-sample gene set enrichment analysis (ssGSEA) was used to assess gene set activation scores in gene expression profiling data. ssGSEA calculates a sample level gene set score by comparing the distribution of gene expression ranks inside and outside the gene set. The ssGSEA score was calculated by Gene Set Variation Analysis (GSVA) R package. Data were examined to determine whether they were normally distributed with the One-Sample Kolmogorov-Smirnov test. If the data were normally distributed, comparisons of measurement data between two groups were performed using independent sample t test and the comparisons among three or more groups were first performed by one-way ANOVA test. If the results showed significant difference, when the data were skewed distribution, comparisons were performed by nonparametric test. Measurement data between two groups were performed using nonparametric Mann-Whitney test.

The raw sequencing data have been deposited in the Gene Expression Omnibus database under the accession number GSE130012. All the other data generated in this study are included in the article and the additional files.

More detailed materials and methods are in the Supplementary Methods.

To explore the correlation between m⁶A modifications with glycolysis metabolism in colorectal cancer (CRC), real-time PCR analysis was performed to compare the regulated-m⁶A gene expression profiles in 47 CRC patients who had been imaged for ¹⁸F-FDG PET (Cohort 1). We observed that there was a most significant correlation between FDG uptake and METTL3 expression in CRC patients (Fig. 1a-b), as measured by FDG maximal standardized uptake value (SUV_max). In addition, we observed that there was a significant correlation between FDG uptake and METTL3 expression in an independent CRC patients’ dataset (GSE110225), as measured by SUV_max (Fig. 1c). Further analysis revealed that a significant correlation between FDG uptake and METTL3 immunohistochemical staining existed in CRC patients of Cohort 1 (Fig. 1d-e). In line with METTL3 higher expression in CRC patients with vigorous glucose metabolism, LC-MS/MS (liquid chromatography-tandem mass spectrometry) (Fig. 1f) and m⁶A dot blot (Fig. 1g) experiments revealed that the m⁶A modification level was significantly increased in these CRC patients with higher FDG uptake as well. To elucidate whether METTL3 plays a role in CRC glycolysis and tumorigenesis, we then performed RNA-seq analysis to compare the gene expression profiles of METTL3-knockout and wild type (WT) HCT116 CRC cells, which exhibit higher METTL3 expression (Figure S1a). METTL3 knockout of CRISPR-cas9 generated cell was confirmed by sequencing and Western blot analysis (Figure S1b-c). A total of 2848 downregulated genes and 3046 upregulated genes (p < 0.05) were detected (raw data accessible via GEO number GSE130012) after knockout of METTL3 in HCT116 cells (Table S1). Gene set enrichment analysis (GSEA) of RNA-seq data showed that the gene signatures of “Glucose uptake” were enriched in HCT116 WT cells (Figure S1d). Single-sample GSEA (ssGSEA) [26] revealed that the gene sets related to Reactome_Glycolysis, Mootha_Glycolysis, Reactome_Glucose_Transport and KEGG_Glycolysis_Gluconeogenesis (glycolysis pathway) and Grade_Colon_And_Rectal_Cancer_Up (colorectal-carcinogenesis signature) were negatively correlated with METTL3-knockout CRC cells (Fig. 1h, Table S2). Besides, further mass spectrometric metabolomics analysis [27] was performed in HCT116 WT and METTL3-knockout cells. The high through put metabolic analysis showed that knockout of METTL3 significantly reduced the levels of key components of glycolysis pathway, such as glucose-6-phosphate, fructose 1,6-bisphosphate, pyruvate and lactate in HCT116 cells (Fig. 1i and Table S3). These data indicate that METTL3 may mediate glycolytic metabolism and carcinogenesis in CRC patients.

Further functional colorimetric validation showed that lactic acid production (a key metabolite of glycolysis) and glucose uptake were both significantly decreased after METTL3 knockout in HCT116 cells (Fig. 2a-b). Knockdown of METTL3 significantly reduced lactic acid production as well as glucose uptake in HCT116 cells (Fig. 2c-d, Figure S2a) and SW480 cells (Figure S2b-d). To figure out whether alteration of METTL3 directly influence glycolytic metabolism, we measured extracellular acidification rate (ECAR) and oxygen consumption rate (OCR) in CRC cells after manipulating METTL3. Knockout or knockdown of METTL3 significantly reduced ECAR levels in HCT116 (Fig. 2e-f) and SW480 cells (Figure S2e), compared with control cells. However, the deletion of METTL3 had no significant effect on OCR levels in HCT116 cells and SW480 cells (Figure S2f).

To elucidate whether METTL3-induced CRC glycolysis depends on its methyltransferase function, we constructed METTL3 wild-type (pcDNA3.1-METTL3) and mutant (pcDNA3.1-METTL3-mut, with MTase domain deletion) recombination plasmids (Figure S2g) [28]. Both of pcDNA3.1-METTL3 and pcDNA3.1-METTL3-mut were successfully overexpressed in DLD1 cells (Figure S2h-i). In gain of function assays, overexpression of METTL3 dramatically increased lactic acid production (Fig. 2g), glucose absorption (Fig. 2h) and ECAR levels (Fig. 2i) in DLD1 cells. Deletion of the MTase domain of METTL3 blocked METTL3-induced glycolysis in CRC cells (Fig. 2g-i). To further examine the effects of METTL3 on glycolysis, we used ¹⁸F-FDG PET ([¹⁸F]-fluoro-2-deoxyglucose positron emission tomography) to measure glucose uptake in vivo. Mouse PET-CT data showed that knockout of METTL3 significantly reduced glucose uptake in xenograft mouse tumor model (Fig. 2j). Overexpression of WT METTL3, but not METTL3-mut, dramatically increased glucose uptake in vivo (Fig. 2k). These data indicate that METTL3 regulates glycolytic metabolism in colorectal cancer via its methyltransferase domain.

METTL3 has been reported as a pivotal modulator in tumorigenesis [29]; however, the role of METTL3 in CRC is paradoxical [30, 31]. GSEA of RNA-seq data showed that the gene signatures of CRC and DNA replication were enriched in HCT116 WT cells, compared with METTL3-knockout cells (Figure S3a-b). The top-scoring genes in the gene sets included key carcinogenesis genes, CDK1, PCNA, and CDCA7. Real-time PCR confirmed that knockout or knockdown of METTL3 expression dramatically reduced the expression of the key genes of tumorigenesis (Figure S3c-e). Our functional validation data showed that knockout or knockdown of METTL3 abolished cell proliferation (Fig. 3a, Figure S3f) and colony formation (Fig. 3b, Figure S3g) in HCT116 cells and SW480 cells (Figure S3h-i). Knockout of METTL3 dramatically reduced HCT116 tumor growth (Fig. 3c-d) and tumor weight (Fig. 3e) in xenograft mouse models.

In gain of function assays, overexpression of METTL3 increased cell proliferation (Fig. 3f) and colony formation (Fig. 3g) of DLD1 cells in vitro. Furthermore, overexpression of METTL3 dramatically increased DLD1 tumor growth (Fig. 3h-i) and tumor weight (Fig. 3j) in xenograft mouse models. However, the same phenomenon was not observed both in vitro (Fig. 3f-g) and in vivo (Fig. 3h-j) after overexpression of METTL3-mut. In addition, 2-DG (an inhibitor of glycolysis pathway) treatment significantly blocked METTL3-induced cell proliferation (Fig. 3k), and colony formation (Fig. 3l) in vitro and in vivo (Fig. 3m-o). There was no difference in tumor growth (Figure S3j-k) and tumor weight (Figure S3l) among different control groups. The data strongly suggests that METTL3 may be a tumor-driver gene and promote CRC progression by regulating colorectal cancer glucose metabolism.

To identify potential mRNA targets of METTL3 whose m⁶A levels were increased by METTL3 in CRC cells, METTL3-knockout and WT HCT116 cells were selected for transcriptome-wide m⁶A-sequencing (m⁶A-seq, MeRIP-seq) and RNA-sequencing (RNA-seq) assays. A noticeable decrease in m⁶A levels was observed in polyadenylated RNAs (poly(A) RNAs) of the METTL3-knockout cells, compared to the WT HCT116 cells, as measured by m⁶A dot blot, EpiQuik™ m⁶A quantification assay and LC-MS/MS analysis (Fig. 4a-c). Consisted with the previous researches, the most common m⁶A motif ‘GGAC’ is significantly enriched in the m⁶A peaks (Fig. 4d). Furthermore, most of the METTL3-binding sites (> 80%) were located in protein-coding transcripts (CDS region) and were highly enriched in 5’UTR and 3’UTR, especially enriched in the vicinity of the stop codon, which was coincidence with the m⁶A distribution (Fig. 4e and Figure S4a).

We next compared the genes with altered-m⁶A modifications between METTL3-knockout cells and the WT cells. The analysis of m⁶A-seq revealed a global hypo-methylation of m⁶A in the transcription level after knockout of METTL3 in HCT116 cells (Fig. 4f-g). A total of 2632 and 584 peaks showed a significant decrease or increase in m⁶A modifications (p < 0.05, Table S4) in METTL3-knockout cells relative to WT cells respectively, and thereby they were termed as hypo-methylated and hyper-methylated m⁶A peaks (Figure S4b). Through analysis of the RNA-seq data, we identified 429 hypo-methylated m⁶A genes whose mRNA transcripts were down-regulated (p < 0.05, Hypo-down) and 595 hypo-methylated m⁶A genes whose mRNA transcripts were up-regulated (p < 0.05, Hypo-up) in METTL3-knockout cells, compared with WT cells (Fig. 4h). Considering the role of METTL3 in the m⁶A methyltransferase complex, mRNA transcripts carrying hypo-methylated m⁶A peaks in HCT116 METTL3 knockout cells were likely potential targets.

We further found that the hypo-down transcripts (p < 0.05) were significantly enriched with genes involved in ‘Metabolism of carbohydrates’ and ‘Glucose metabolism’, which included the glycolysis pathway through gene ontology analysis (Figure S4c). Nevertheless, those hypo-up transcripts were significantly enriched with “Post-translational protein modification”, “Chromatin organization”, which are not associated with glycolytic metabolism (Figure S4c). Therefore, the hypo-methylated m⁶A genes, with decreased expression in METTL3-knockout CRC cells (p < 0.05) (hypo-down), were selected for further validation (Fig. 4i). Eighteen genes (PYGB, FUT1, GALNS, HK2, IDUA, PC, ENO3, SGSH, ST3GAL2, SLC2A1, HS6ST1, HS3ST1, SLC26A1, NUP62, GLCE, CHST12, ADOGK and AGRN), which involved in ‘Carbohydrates metabolism’ and ‘Glucose metabolism’, were sorted as potential target genes (Fig. 4i). Real-time PCR and MeRIP-qPCR assays were performed to measure the mRNA and m⁶A levels of eighteen hypo-down candidates in different cells. Real-time PCR data showed that the expression of six target genes was significantly impaired both in METTL3-knockout HCT116 (Figure S4d) and METTL3-knockdown SW480 CRC cells (Figure S4e). METTL3 up-regulation increased the RNA level of these six transcripts in DLD1 cells (Figure S4f-g). The m⁶A peaks of these six transcripts were remarkably decreased in METTL3-knockout HCT116 cells, compared with WT cells (Fig. 4j and Figure S4h) according to the MeRIP-seq data. MeRIP-PCR data was consistent with the MeRIP-seq data (Fig. 4k and Figure S4i). Further MeRIP-qPCR [18, 32] analyses showed the m⁶A levels in five of the six target genes were dramatically reduced in HCT116 METTL3-knockout cells (Fig. 4l, Figure S4j). The rest five target genes were further validated by functional glycolysis assay. Lactate production was only significantly reduced in HCT116 cells with Hexokinase 2 (HK2) and SLC2A1 (GLUT1) downregulation, but not in those cells with the other three genes downregulation (Figure S4k). In addition, we performed MeRIP-seq in 5 CRC patients with METTL3 high expression and 5 CRC patients with METTL3 low expression. The Integrative Genomics View (IGV) data revealed that the m⁶A level of HK2 and SLC2A1 (GLUT1) was higher in 5 CRC patients with METTL3 high expression than those with METTL3 low expression (Figure S4l), indicating that higher m⁶A level of HK2 and SLC2A1 (GLUT1) depends on METTL3 modification in CRC patients. The MeRIP-seq data in CRC patients is consistent with the data in CRC cell lines. Therefore, we selected HK2 and SLC2A1 (GLUT1) as candidate targets of METTL3 for further investigation.

In biological validation assay, HK2 and SLC2A1 (GLUT1) expression were confirmed to be downregulated in METTL3-knockout cells (Fig. 5a) and METTL3-knockdown cells (Fig. 5b and Figure S5a). In addition, overexpression of METTL3, but not METTL3-mut dramatically increased HK2 and SLC2A1 (GLUT1) mRNA and protein levels (Fig. 5c). We further measured m⁶A levels of glycolysis genes HK2 and SLC2A1 (GLUT1) by MeRIP-qPCR after transfection of control, METTL3 and METTL3-mut plasmids in DLD1 cells. The MeRIP-qPCR data showed that overexpression of METTL3, but not METTL3-mut, dramatically increased the m⁶A levels of HK2 and SLC2A1 (GLUT1) in DLD1 cells (Figure S5b). It indicates that m⁶A modification of HK2 and SLC2A1 (GLUT1) directly affects its mRNA level. We detected the hexokinase activity in CRC cells as well. The hexokinase activity was noticeably decreased in METTL3-knockout cells and METTL3-knockdown cells (Fig. 5d-e and Figure S5c). Moreover, overexpression of METTL3, but not METTL3-mut, significantly up-regulated the hexokinase activity (Fig. 5f). To further address the effect of m⁶A modification on HK2 and SLC2A1 (GLUT1) gene, we constructed both wild-type and mutant HK2 and SLC2A1 (GLUT1) luciferase reporter plasmids, which contained the wild-type 5′ and 3’UTR of HK2, the 3’UTR of SLC2A1 (GLUT1) or mutations (m⁶A was replaced by T) in the m⁶A sites (Fig. 5g-i). As expected, compared with control siRNA, METTL3 siRNA substantially reduced luciferase activity of the individual reporter constructs bearing wild-type HK2 5′ and 3’UTR that have intact m⁶A sites, and wild-type 3’UTR of SLC2A1 (GLUT1). However, downregulation of METTL3 had no significant effect on the luciferase activity in these reporter plasmids with m⁶A sites mutations in HCT116 (Fig. 5g-i) and SW480 cells (Figure S5d-f). Furthermore, overexpression of METTL3, but not METTL3 mutant, significantly increased luciferase activity of the individual reporter constructs bearing wild-type HK2 5′ and 3’UTR, and wild-type SLC2A1 (GLUT1) 3’UTR in DLD1 cells (Fig. 5g-i). Overexpression of METTL3 and METTL3-mut had no significant effect on HK2 and SLC2A1 (GLUT1) luciferase reporter plasmids with m⁶A sites mutation (Fig. 5 g-i).

To analyze the effect of m⁶A modification on the stability of METTL3 target transcripts, we conducted RNA stability assays [8]. The RNA stability curves showed that knockdown of METTL3 reduced the half-life of HK2 and SLC2A1 (GLUT1) mRNA in HCT116 (Fig. 5j-k) and SW480 cells (Figure S5g-h). Thus, METTL3-induced upregulation of HK2 and SLC2A1 (GLUT1) level is at least in part due to the increased stability of HK2 and SLC2A1 mRNA transcript. Insulin-like growth factor 2 mRNA binding proteins (IGF2BP1, IGF2BP2 and IGF2BP3) are three well-established m⁶A readers in mammalian cells [33], and responsible for stabilizing mRNA. m⁶A readers YTHDF2, YTHDF3 and YTHDC2 are responsible for mRNA decay [34‐36]. Real-time PCR data showed that knockdown of IGF2BP2, but not IGF2BP1/3 significantly reduced the mRNA level of HK2 in HCT116 and SW480 cells (Figure S5i). Downregulation of IGF2BP2/3, but not IGF2BP1 noticeably reduced the mRNA level of SLC2A1 (GLUT1) (Figure S5j). Knockdown of YTHDF2, YTHDF3 or YTHDC2 did not influence the mRNA levels of HK2 or SLC2A1 (GLUT1) (Figure S5k-l). Furthermore, the half-life of HK2 mRNA was significantly decreased in HCT116 cells after transfection with IGF2BP2 siRNA (Fig. 5l), and the half-life of SLC2A1 (GLUT1) mRNA was significantly reduced in IGF2BP2/3 down-regulated HCT116 cells (Fig. 5m-n). RIP assay showed that IGF2BP2 directly bound to the 5′ and 3’UTR of HK2 mRNA (Fig. 5o) and IGF2BP2/3 directly bound to the 3’UTR of SLC2A1 (GLUT1) mRNA (Fig. 5o). Further RIP real-time PCR assay indicated that knockout of METTL3 significantly reduced the binding efficiency of IGF2BP2 to the 5′/3’UTR of HK2 mRNA and the binding efficiency of IGF2BP2/3 to the 3’UTR of SLC2A1 (GLUT1) mRNA (Fig. 5p). Taken together, our data demonstrate that METTL3-mediated m⁶A modification increases HK2 and SLC2A1 (GLUT1) expression through IGF2BP2-dependent and IGF2BP2/3-dependent mRNA stability regulation, respectively.

Hexokinase 2 (HK2) is a pivotal kinase in the glycolytic pathway [37]. Previous studies have demonstrated that HK2 activity is remarkably increased in various malignant neoplasms, as well as in CRC [38, 39]. Glucose transporter 1 (GLUT1), encoded by SLC2A1, is the predominant glucose transporter expressed on colonic epithelial cells [40]. Here we first revealed that METTL3 increased HK2 and SLC2A1 (GLUT1) expression via promoting their mRNA stability. Therefore, we next performed rescue experiments to investigate whether HK2 and SLC2A1 (GLUT1) participated in the biological function of METTL3 in CRC. HCT116 WT and METTL3-knockout cells were transfected with control, HK2 or SLC2A1 overexpression plasmids. The overexpression efficiency of HK2 and SLC2A1 were confirmed by real-time PCR and Western blot analysis (Figure S6a-b). Ectopic expression of HK2 or SLC2A1 partially restored the proliferation (Fig. 6a) and colony formation ability (Fig. 6b) of METTL3-knockout cells and tumor growth (Fig. 6c-e, Figure S6c-e). In addition, DLD1 cells were transfected with METTL3 overexpression plasmid and then treated with HK2 or SLC2A1 siRNA. The downregulation efficiency of HK2 and SLC2A1 siRNA were confirmed by real-time PCR and Western blot analysis (Figure S6f-g). Downregulation of HK2 or SLC2A1 dramatically impaired METTL3-induced cell proliferation (Figure S6h) and colony formation (Figure S6i) in vitro and tumor growth in vivo (Figure S6j-l), which further supported HK2 and SLC2A1 (GLUT1) as critical target genes of METTL3 in CRC.

In glycolytic assays, ectopic expression of HK2 or SLC2A1 (GLUT1) restored the decrease of lactate production in HCT116 METTL3-knockout cells (Fig. 6f). Meanwhile, overexpression of SLC2A1 (GLUT1) significantly restored the decrease of glucose uptake in HCT116 METTL3-knockout cells in vitro (Fig. 6g) and in vivo (Fig. 6h). Furthermore, the downregulation of HK2 or SLC2A1 (GLUT1) significantly reduced METTL3-mediated higher lactate production in DLD1 cells (Fig. 6i). Downregulation of SLC2A1 (GLUT1) dramatically impaired METTL3-induced higher glucose uptake in DLD1 cells in vitro (Fig. 6j) and in vivo (Fig. 6k). Thus, HK2 and SLC2A1 (GLUT1) mediate the regulatory function of METLL3 in CRC cells.

We next performed immunohistochemical staining in CRC patients’ tissues of Cohort 2. Interestingly, the samples with METTL3 higher expression displayed strongly staining for HK2 and GLUT1 (Fig. 7a, left panel). In addition, samples with low expression of METTL3 appeared low levels of HK2 and GLUT1 (Fig. 7a, right panel). Statistically, METTL3 expression was positively correlated with HK2 and GLUT1 expression in CRC tissues (Fig. 7b). There was no significant correlation between METTL3 and IGF2BP2/3 expression in CRC tissues of Cohort 2 (Fig. S7a). We next assessed the association between the intensity of METTL3, glycolysis components, and disease-free survival after tumor resection in this patients’ cohort. This analysis showed that elevated expression of METTL3, HK2 or GLUT1 in CRC tissues predicted robustly shorter disease-free intervals, either as a linear (Fig. 7c-e) or categorized variable (Fig. 7f-h). Furthermore, a linear association was found between the combined expression of METTL3, HK2 and GLUT1 and the risk of relapse after therapy in CRC (Fig. 7i). The shortest disease-free survival times were detected in those patients with three highly expressed markers METTL3/HK2/GLUT1 (Fig. 7j). From these observations, we conclude that elevated expression levels of METTL3 and its target genes may identify CRC patients with poor prognosis. We next analyzed the correlation of METTL3, HK2 or GLUT1 expression and different clinicopathological features in Cohort 2. METTL3 expression was positively correlated with pathological differentiation, AJCC stage, and recurrence in CRC patients (Figure S7b). HK2 and GLUT1 expression were positively correlated with AJCC stage, and recurrence (Figure S7b). In addition, METTL3 expression is gradually increased from normal colorectal epithelial tissues, adenoma to cancer tissues in Cohort 3 (Figure S7c).

So far, we have proved that METTL3 may function as an oncogene by stabilizing HK2 and SLC2A1 mRNA levels in CRC cells. Recently, several RNA methyltransferase inhibitors have been developed as anti-tumor drug candidates. We next hypothesized that those CRC patients, harboring higher METTL3 expression, will be more sensitive to the anti-tumor drug candidates targeting METTL3. To test this prediction, we treated the HCT116 CRC cells, which express a higher METTL3 level, with DAA (3-Deazaadenosine, an inhibitor of the internal N⁶-Methyladenosine) [41, 42]. DAA treatment resulted in 39% inhibition of cell proliferation in HCT116 cells. Conversely, treatment in DLD1, which has a lower expression of METTL3, with the same dose of DAA, resulted in only 20% inhibition of cellular proliferation (Fig. 7k). These inhibition results could also be recapitulated in vivo. DAA is more efficient to inhibit tumor growth (Fig. 7l-m) and tumor weight (Fig. 7n) in nude mice bearing HCT116 cells than those bearing DLD1 cells. Taken together, these data suggest that targeting the METTL3 may be more effective in CRC cells with higher METTL3 expression than in those CRC cells with lower METTL3 expression.