METTL3 promotes tumour development by decreasing APC expression mediated by APC mRNA N6-methyladenosine-dependent YTHDF binding

The adenomatous polyposis coli (APC) is a frequently mutated tumour suppressor gene in cancers. However, whether APC is regulated at the epitranscriptomic level remains elusive. In this study, we analysed TCGA data and separated 200 paired oesophageal squamous cell carcinoma (ESCC) specimens and their adjacent normal tissues and demonstrated that methyltransferase-like 3 (METTL3) is highly expressed in tumour tissues. m6A-RNA immunoprecipitation sequencing revealed that METTL3 upregulates the m6A modification of APC, which recruits YTHDF for APC mRNA degradation. Reduced APC expression increases the expression of β-catenin and β-catenin-mediated cyclin D1, c-Myc, and PKM2 expression, thereby leading to enhanced aerobic glycolysis, ESCC cell proliferation, and tumour formation in mice. In addition, downregulated APC expression correlates with upregulated METTL3 expression in human ESCC specimens and poor prognosis in ESCC patients. Our findings reveal a mechanism by which the Wnt/β-catenin pathway is upregulated in ESCC via METTL3/YTHDF-coupled epitranscriptomal downregulation of APC.

T he adenomatous polyposis coli (APC) is known as a key tumour suppressor gene. APC plays a crucial role in suppressing the canonical Wnt signalling pathway, which controls cell proliferation and differentiation [1][2][3] . Loss of APC function leads to abnormal stabilization of β-catenin. Transactivated β-catenin forms a complex with T-cell factor/lymphoid enhancer-binding factor family members and induces many instrumental downstream genes, such as CCND1 and MYC, which promote tumour development 4,5 . Mutation rates of APC in colorectal cancer can reach 80% [6][7][8] . In contrast, APC is rarely mutated in other types of cancer, such as oesophageal squamous cell carcinoma (ESCC), which has only~1.5% APC mutation 9,10 . Whether APC expression is regulated at the epitranscriptomic level, thereby contributing to tumour development, has not been determined. RNA modifications reveal a new level of posttranscriptional gene expression regulation [11][12][13][14][15] . N 6 -methyladenosine (m 6 A) modification is the most abundant RNA modification in eukaryotic mRNAs and is mainly mediated by m 6 A regulators, including 'writers', 'erasers' and 'readers', which add, remove and recognize methylation, respectively. m 6 A modification regulates RNA stability and translation efficiency, chromatin state, alternative polyadenylation and pre-mRNA splicing [16][17][18][19][20] . The identified writers include methyltransferase-like (METTL) 3/14, Wilms tumour 1-associated protein (WTAP), RNA-binding motif protein 15/15B and KIAA1429. Fat mass and obesity-associated protein (FTO) and alkB homologue 5 (ALKBH5) are erasers, whereas YT521-B homology (YTH) domain-containing proteins (YTH Domain Family [YTHDF] 1-3 and YTH Domain-Containing 1-2), heterogeneous nuclear ribonucleoprotein protein families and insulin growth factor-2 mRNA-binding proteins families are regarded as readers 21,22 . RNA m 6 A modification and its key m 6 A methyltransferase METTL3, which forms a heterodimer with METTL14, have been reported to be essential for tumour progression in several types of cancer [23][24][25] . However, the mechanism governing METTL3-promoted tumour progression has not been elucidated.
In this study, we demonstrate that METTL3 is upregulated in ESCC and enhances m 6 A of APC mRNA, thereby leading to recruitment of YTHDF for APC mRNA degradation and subsequent enhanced aerobic glycolysis, ESCC cell proliferation, tumour formation in mice and poor prognosis in patients.

Results
Upregulated METTL3 is correlated with poor survival in ESCC patients. Oesophageal cancer is the sixth most common tumour in the world, with only a 12-20% 5-year patient survival rate being reported 26,27 . To explore the expression profile of METTL3 in ESCC, which constitutes~90% of oesophageal cancer 28 , we analysed the Cancer Genome Atlas (TCGA) data and found that METTL3 expression was significantly upregulated in 95 ESCC specimens compared to that in 11 adjacent normal oesophageal epithelium tissues ( Fig. 1a and Supplementary Fig. 1a). In addition, among the direct regulators of m 6 A, including METTL3, METTL14, WTAP, FTO and ALKBH5, we found that only METTL3 was consistently upregulated in ECSS tissues compared to normal tissues basing on the analyses of TCGA (Supplementary Fig. 1a) and GSE53625 ( Supplementary Fig. 1b) dataset. Consistent with this result, microarray analysis of the gene expression profiles of 119 paired ESCC specimens and adjacent normal oesophageal epithelium tissues ( Supplementary Fig. 1c), and immunohistochemical (IHC) staining of tissue arrays containing 81 paired ESCC specimens and adjacent normal oesophageal epithelium tissues (Fig. 1b, c) showed that METTL3 expression levels were significantly higher in the ESCC tissues than in the paired adjacent normal tissues. Kaplan-Meier analysis showed that the patients with high METTL3 expression had shorter overall survival time than those with low METTL3 expression ( Fig. 1d and Supplementary Fig. 1d). In line with this finding, METTL3 mRNA ( Supplementary Fig. 1e) and protein (Fig. 1e) expression levels were higher in nine different ESCC cell lines than in Het-1a-immortalized normal oesophageal epithelial cells. These results indicate that METTL3 expression is upregulated in ESCC and inversely correlated with ESCC patient survival time.
METTL3 promotes ESCC cell proliferation and tumour growth in mice. To determine the role of METTL3 in cell proliferation, we depleted METTL3 in KYSE180 (Fig. 2a)  We next established tetracycline-inducible expression of METTL3 in KYSE450 cells. Tetracycline treatment increased METTL3 expression in a dosage-dependent manner, which elicited correspondingly increased levels of cell proliferation (Fig. 2e) and colony formation (Fig. 2f). Compared to overexpression of wild-type (WT) METTL3, overexpression of a METTL3-inactive mutant in ESCC cells failed to largely promote cell growth and proliferation (Fig. 2g, h and Supplementary  Fig. 2e, f). These results strongly suggested that METTL3 promotes tumour cell proliferation dependent on its expression levels and intact activity.
To determine the role of METTL3 in tumour growth in mice, we subcutaneously injected KYSE180 or KYSE450 cells with or without METTL3 depletion or overexpression of WT METTL3 or the METTL3-inactive mutant into athymic nude mice. We showed that METTL3 depletion reduced tumour size ( These results indicate that METTL3 expression is instrumental for tumour growth in mice. METTL3 mediates m 6 A upregulation on APC mRNA. To determine the mechanism underlying METTL3-promoted tumour cell proliferation, we examined the role of METTL3 in epitranscriptomal regulation in ESCC cells. METTL3 depletion reduced the total m 6 A abundance in KYSE180 cells (Fig. 3a). Methylated RNA immunoprecipitation (MeRIP) with an m 6 Aspecific antibody followed by RNA sequencing (MeRIP-seq) revealed that the identified m 6 A sites in the mRNAs of KYSE180 are consistent with the m 6 A consensus sequence RRACH (R = G or A; H = A, C or U) 15 , especially GRACU (p = 1.5e10 − 31) (Fig. 3b). In line with previous reports 15 , the m 6 A signal was enriched around the stop codon and 3′-untranslated region (UTR) of mRNAs (Fig. 3c).
METTL3 depletion reduced 1199 m 6 A peaks in mRNAs of 1101 genes (fold-change > 2) in MeRIP-seq, which included the transcripts that were well known to be modified by m 6 A, such as ACTB, EEF2 and MALAT1 long non-coding RNA (Supplementary Fig. 3a) 29 , and regulated expression of 2973 genes (foldchange > 2) in mRNA sequencing (Supplementary Data 1).
Among these genes, 93, including APC, were overlapped (Fig. 3d). Gene Ontology (GO) enrichment analyses of cellular component (CC) terms in MeRIP-seq showed that m 6 A levels were substantially decreased in the gene set of the β-catenin destruction complex ( Supplementary Fig. 3b), which included APC, in KYSE180 cells with METTL3 depletion.
APC is a negative Wnt regulator that promotes β-catenin degradation [1][2][3] . We showed that the m 6 A peak is in the last exon near the termination codon region of APC, and that METTL3 depletion reduced the m 6 A level of APC mRNA (Fig. 3e). These results are in line with previous reports showing that m 6 A level of APC mRNA were high in different types of cancer cells, such as HeLa cervical cancer cells, H1299 lung cancer cells and HepG2 liver cancer cells, within the top 10% of most methylated transcripts ( Supplementary Fig. 3c) 17,[30][31][32][33][34][35][36] . Analyses of available m 6 A individual-nucleotide-resolution cross-linking and immunoprecipitation (miCLIP) data [37][38][39][40][41] revealed that three adenosine bases were methylated and are within the decreased m 6 A peak of APC mRNA by METTL3 depletion (Fig. 3f). These results indicate that METTL3 regulates the m 6 A level of APC mRNA, which is high in many types of cancer cells. m 6 A RIP followed by real-time quantitative PCR assay showed that m 6 A levels in APC mRNA were notably decreased upon METTL3 depletion in KYSE180 (Fig. 3g) Fig. 3e) cells, and this increase was abrogated by METTL14 depletion (Fig. 3h and Supplementary Fig. 3e). In addition, RIP with an anti-METTL3 antibody showed that METTL3 bound to APC mRNA in KYSE180 and KYSE450 cells (Fig. 3i). These results indicate that METTL3 binds to APC mRNA and enhances its m 6 A levels in an METTL14-dependent manner.
METTL3-dependent m 6 A upregulation on APC mRNA suppresses APC expression. To determine the effect of METTL3-dependent m 6 A regulation on APC expression, we constructed a luciferase reporter gene with an integrated coding sequence (CDS) containing the WT or mutated m 6 A from the 3′-end of APC mRNA, in which the three adenosine bases within the m 6 A consensus sequences were mutated to thymine (T) (Supplementary Fig. 4a). Luciferase assays showed that METTL3 depletion largely increased the activity of luciferase with the WT APC, whereas the mutated APC elevated the luciferase activity and rendered this activity resistant to regulation by METTL3 depletion (Fig. 4a). In contrast, METTL3 overexpression suppressed the activity of luciferase with WT APC but not mutated APC (Fig. 4a). These results suggest that METTL3-mediated m 6 A level upregulation of APC mRNA suppresses APC expression.
Consistent with this finding, METTL3 depletion increased APC mRNA and protein expression ( Fig. 4b and Supplementary  Fig. 4b), and these increases were abrogated by reconstituted expression of rMETTL3 in both KYSE450 (Fig. 4c) and KYSE180 ( Supplementary Fig. 4c) cells. In addition, overexpression of METTL3 decreased APC expression in KYSE450 cells (Fig. 4d) and tetracycline dosage-dependently increased METTL3 expression levels were inversely correlated with APC levels (Fig. 4e). In contrast, METTL3-inactive mutant, unlike its WT counterpart, failed to modulate APC expression ( Fig. 4f and Supplementary  Fig. 4d). Of note, METTL14 depletion in the ESCC cells abrogated the METTL3 overexpression-decreased APC expression ( Fig. 4g and Supplementary Fig. 4e). These results indicate that APC mRNA and protein expression are suppressed by APC mRNA m 6 A upregulation mediated by METTL3 in concert with METTL14 in ESCC cells.
METTL3-enhanced m 6 A of APC mRNA and subsequent binding of YTHDF suppresses APC expression. m 6 A readers bind to m 6 A-modified mRNAs, leading to increased or decreased protein expression. YTHDF1-3 mediate mRNA degradation 30    . c ***p = 4.72E − 05. d ***p = 0.0007. Two-tailed t-test; ns, not significant. e, f KYSE450 cells expressing the METTL3 expression-inducible Tet-on lentiviral vector were treated with or without the indicated dosages of tetracycline, followed by immunoblotting analyses. The cell proliferation (e) and colony formation (f) assays were performed. Data represent means ± SD of triplicate samples. e ***p = 6.37E − 04 (top), ***p = 9.58E − 04 (bottom). f ***p = 9.24E − 04 (left), **p = 0.0095 (right). Two-tailed t-test. g, h KYSE450 cells with or without expressing WT METTL3 or an inactive METTL3 mutant were analysed by immunoblotting analyses. The cell proliferation (g) and colony formation (h) assays were performed. Data represent means ± SD of triplicate samples. ***p = 0.0004 (g), ***p = 0.0003 (h). Two-tailed t-test; ns, not significant. i-k KYSE180 cells with or without METTL3 depletion were subcutaneously injected into the mice (n = 6). Six weeks later, tumour sizes (i), volumes (j) and weight (k) were measured. Scale bar, 1 cm. Data represent means ± SD of six mice per group. ***p = 2.58E − 06 (j), *p = 0.0138 (k) (two-tailed t-test). l-n KYSE180 cells with or without expressing WT METTL3 or an inactive METTL3 mutant were subcutaneously injected into the mice (n = 6). Five weeks later, tumour sizes (l), volumes (m) and weight (n) were measured. Scale bar, 1 cm. Data represent means ± SD of six mice per group. **p = 0.0015 (m), 0.0013 (n) (two-tailed t-test). ns, not significant. Source data are provided as a Source Data file. Fig. 5a) 18,44,46,47 . RIP analyses with antibodies against YTHDF1, YTHDF2 or YTHDF3 followed by real-time quantitative PCR assay revealed that the amount of YTHDF2 that bound APC mRNA was much more than those of YTHDF1 and YTHDF3 in KYSE180 (Fig. 5a) and KYSE450 ( Supplementary Fig. 5b), and was reduced by METTL3 depletion (Fig. 5b and Supplementary Fig. 5c). These results suggest that METTL3-increased m 6 A of APC mRNA largely promotes the binding of YTHDF2 to APC mRNA.

YTHDF1-3 bound APC mRNA (Supplementary
To examine the role of the binding of YTHDF2 to APC mRNA in APC expression, we depleted YTHDF2, which increased the mRNA (Fig. 5c and Supplementary Fig. 5d) and protein ( Fig. 5d and Supplementary Fig. 5e) expression of APC in KYSE450 (Fig. 5c, d) and KYSE180 ( Supplementary Fig. 3d, e) cells. Combined depletion of YTHDF1-3 further increased the mRNA (Fig. 5e and Supplementary Fig. 5f) and protein ( Fig. 5f and Supplementary Fig. 5g) expression of APC in these cells. A similar increase was also observed in HeLa cells with depletion of YTHDF1-3 ( Supplementary Fig. 5h). In addition, we performed a luciferase reporter assay and showed that depletion of YTHDF2 enhanced the activity of luciferase driven by the WT APC CDS sequence containing m 6 6 A upregulation on APC mRNA. a m 6 A content of total RNA in KYSE180 cells with or without METTL3 depletion was determined. Data represent the means ± SD of triplicate samples. **p = 0.0094 based on two-tailed Student's t-test. b The most frequent m 6 A motifs detected by DREME software in MeRIP-seq of KYSE180 cells are shown. c MeRIP-seq of KYSE180 cells with or without METTL3 depletion was conducted. The proportions of m 6 A peak distribution in the 5′-untranslated region (5′-UTR), start codon, coding region (CDS), stop codon and the 3′-untranslated region (3′-UTR) across the entire set of mRNA transcripts were calculated. d The genes with diminished m 6 A peaks identified by MeRIP-seq (1101 genes, fold-change > 2) and the genes with upregulated or downregulated mRNA expression (2973 genes, fold-change > 2) identified by RNA-seq in KYSE180 with or without METTL3 depletion were analysed (93 genes overlapped). e The m 6 A peak visualization of m 6 A-seq in APC transcripts in KYSE180 cells with or without METTL3 depletion is shown. The m 6 A peaks are in the last exon of APC. f The MeRIP-Seq peak in KYSE180 regulated by METTL3 depletion and miCLIP peaks in other types of cancer cells from the Gene Expression Omnibus were aligned. The three methylated adenosine bases were denoted with red colour. g, h Methylated RNA in KYSE180 cells with or without METTL3 depletion (g), METTL3 expression or METTL3 and METTL14 shRNA expression (h) was immunoprecipitated with an m 6 A antibody followed by qPCR analyses with primers against APC mRNA. Data represent means ± SD of triplicate samples. g ***p = 3.84E − 05 (left), 8.99E − 05 (right); **p = 0.0012 (middle). h **p = 0.0019. Two-tailed t-test; ns, not significant. i RIP analyses of KYSE180 and KYSE450 cells were performed with an anti-METTL3 antibody followed by qPCR analyses with primers against APC mRNA. Data represent means ± SD of triplicate samples. ***p = 2.50E − 05 (left), 1.08E − 05 (right). Two-tailed t-test. Source data are provided as a Source Data file. mutated APC enhanced the luciferase activity and rendered this activity resistant to regulation by YTHDF2 depletion (Fig. 5g).
In addition, METTL3 overexpression-suppressed luciferase activity driven by the WT APC CDS sequence was restored by YTHDF2 depletion, which did not affect the mutated APCenhanced luciferase activity ( Supplementary Fig. 5i). These results indicate that METTL3-increased m 6 A of APC mRNA and subsequent binding of YTHDF suppress APC expression.  Fig. 6c) cells, and this METTL3 overexpression-induced effect was eliminated by METTL14 depletion (Supplementary Fig. 6b, c). In addition, METTL3 depletion decreased glucose uptake ( Fig. 6b and Supplementary Fig. 6d), lactate production ( Fig. 6c and Supplementary Fig. 6e), cell proliferation ( Fig. 6d and Supplementary Fig. 6f) and numbers of cell colonies ( Supplementary  Fig. 6g). Notably, this METTL3 depletion-elicited inhibition of gene expression (Fig. 6a), glycolysis (Fig. 6b, c), cell proliferation ( Supplementary Fig. 6f) and colony formation ( Supplementary  Fig. 6g) was largely abrogated by APC depletion in these cells. Conversely, METTL3 overexpression elicited an increase of glucose consumption and lactate production, which was abrogated by YTHDF2 depletion (Supplementary Fig. 6h-k). In addition, metabolic flux analysis of 13 C 6 glucose-labelled KYSE180 cell showed that METTL3 overexpression-enhanced glycolysis was reduced by c-Myc depletion ( Supplementary Fig. 6l). These results indicate that METTL3 reduces APC expression and promotes β-catenin-mediated downstream gene expression, aerobic glycolysis and ESCC cell proliferation.
To determine the role of METTL3-induced APC expression downregulation in tumour growth in mice, we subcutaneously injected KYSE180 and KYSE450 cells with or without METTL3 Immunoblotting analyses were performed with the indicated antibodies for three times with similar results. g KYSE180 cells expressing luciferase reporter genes fused with or without the wild-type (WT) or mutated m6A nucleotides from APC genes were transfected with or without a vector expressing YTHDF2 shRNA. The relative luciferase activity after normalization to the shControl group is shown. Data represent the means ± SD of triplicate samples. ***p = 3.94E − 06 based on two-tailed Student's t-test. ns, not significant. Source data are provided as a Source Data file.
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23501-5 ARTICLE depletion, or with or without combined METTL3 and APC depletion into athymic nude mice. We showed that METTL3 depletion reduced tumour size ( Fig. 6e and Supplementary Fig. 7a), volume ( Fig. 6f and Supplementary Fig. 7b) and weight ( Fig. 6g and Supplementary Fig. 7c), as well as the amount of lactate in the tumour tissue ( Supplementary Fig. 7d) were restored by APC depletion. In addition, IHC staining showed that METTL3 depletion increased APC expression with corresponding decreased expression of β-catenin, cyclin D1, c-Myc and PKM2 in tumour tissues (Fig. 6h, i). These METTL3 depletion-induced alteration of protein expression were abrogated by APC depletion. Consistent with the tumour-promoting effect of METTL3 overexpression (Fig. 2l-n), overexpression of WT, but not the inactive mutant, METTL3 increased the amount of lactate in tumour tissue ( Supplementary Fig. 7e). These results indicate that suppression of APC expression by METTL3 promotes tumour development.
Downregulated APC expression correlates with upregulated METTL3 expression in human ESCC specimens and poor prognosis of ESCC patients. To determine the clinical relevance of METTL3-suppressed APC expression, we performed TCGA data analyses and showed that APC mRNA expression was inversely correlated with METTL3 mRNA expression in ESCC (Fig. 7a). In addition, we analysed 119 paired ESCC specimens and their adjacent normal tissues by gene expression profiles, and showed that APC expression was significantly downregulated in ESCC specimens compared to that in normal tissues (Fig. 7b). Consistent with the results obtained from TCGA data analyses, an inverse correlation between METTL3 protein expression and APC protein expression (Fig. 7c, d), and a positive correlation between METTL3 protein expression and β-catenin protein expression (Fig. 7c, e)   ( Fig. 7f). These results suggest that downregulated APC expression correlates with upregulated METTL3 expression in human ESCC specimens and poor prognosis of ESCC patients.

Discussion
APC is a critical tumour suppressor gene that plays an instrumental role in tumour development. APC mutations have been frequently detected in colorectal cancers but with a considerably lower frequency in some other types of cancers [6][7][8][9][10]53 . In addition to regulation by the gene mutation, whether APC is regulated at epitranscriptomic levels is unclear. Analysis of TCGA data and 200 paired ESCC specimens and their adjacent normal tissues showed that mRNA of METTL3, an important methyltransferase for m 6 A modification, is highly expressed in ESCC tissues compared to normal tissues. MeRIP and transcriptomal RNA sequencing revealed that METTL3 upregulates m 6 A modification in a large number of genes, including β-catenin destruction complex genes, such as APC. METTL3-enhanced APC mRNA m 6 A modification recruited YTHDF and elicited suppression of APC expression and subsequent enhanced expression of β-catenin, cyclin D1, c-Myc and c-Myc-regulated glycolytic genes, including PKM2. The reprogrammed expression of metabolic and cell cycle-promoting genes enhanced aerobic glycolysis, ESCC cell proliferation and tumour formation in mice (Fig. 8). YTHDF mediate m 6 A-dependent RNA degradation 42,43 . We demonstrated that YTHDF bound METTL3-mediated m 6 A of APC mRNA and reduced APC expression, revealing a previously unknown mechanism by which the Wnt/β-catenin pathway is upregulated in cancer in an epitranscriptomal regulationdependent manner. The clinical relevance and significance of our findings are evidenced by the correlation of upregulated METTL3 expression in human ESCC specimens with APC downregulation and poor prognosis in ESCC patients. The discovery of the critical regulation of APC by METTL3/YTHDFcoupled m 6 A regulation may provide promising approaches for the therapy of ESCC patients.
Immunohistochemistry. After deparaffinization, rehydration and antigen retrieval, TMA slides were incubated with primary rabbit anti-human METTL3 (dilution 1 : 500; Abcam Antibody; ab181064) 20 , primary rabbit anti-human APC (dilution 1 : 500; Abcam Antibody; ab154906) 21 or nonspecific IgG (as a negative control) overnight at 4°C. The slides were then incubated with anti-rabbit secondary antibody followed by chromogen DAB staining and haematoxylin counterstaining, and mounted with xylene-based medium. We quantitatively scored the tissue slides under a microscope according to the percentage of positive cells and staining intensity. A H-score (maximum score, 300) was assigned using the following formula: 3 × percentage of strongly staining + 2 × percentage of moderately staining + percentage of weakly staining 54 . Two pathologists (X.F. and S.S.) who were blinded to the clinical information independently validated the reproducibility of the scoring system.
RNA extraction and quantitative RT-PCR analysis. Total RNA was isolated with TRIzol reagent (Invitrogen, USA) according to the manufacturer's instructions. RNA (1000 ng) was reverse-transcribed into cDNA with a RevertAid First Strand cDNA Synthesis kit (Thermo). SYBR Green-based quantitative reversetranscriptase PCR was performed using a 7900HT fast real-

Fig. 8 A mechanism underlying METTL3-promoted ESCC development.
Overexpressed METTL3 increases the levels of the APC mRNA N 6 -methyladenosine, which recruits YTHDF to APC mRNA, to downregulate APC expression. Downregulated APC stabilizes β-catenin expression and promotes β-catenin-dependent gene expression for aerobic glycolysis, and ESCC cell proliferation and tumour growth.
(Applied Biosystems/Life Technologies, Waltham, USA). The relative mRNA expression levels were calculated by the 2 −ΔΔCt method with normalization to ACTB and the PCR primers are listed in Supplementary Table 2.
RNA m 6 A quantification. The m 6 A content of 200 ng of RNA extracted from tissues was measured by an EpiQuik m 6 A RNA methylation quantification kit (P-9005; Epigentek) according to the manufacturer's instructions.
Lentivirus production and infection. Plasmids containing transgenes and packaging plasmids were cotransfected into HEK 293T cells using Lipofectamine 3000 (Invitrogen, USA). Viruses were collected and concentrated after 48 h 55 . When tumour cells reached 50-60% confluence, we infected the cells with concentrated virus and then selected them by antibiotic treatment. The shRNA sequences are listed in Supplementary Table 3.
Immunoblotting and immunoprecipitation analysis. Extraction of proteins with a modified buffer from cultured cells was followed by immunoprecipitation and immunoblotting with antibodies 56 .
Cell proliferation assay. A total of 2 × 10 4 cells were plated and counted 6 days (for KYSE180 cells) and 5 days (for KYSE450 cells) after seeding in RPMI-1640 medium with 10% bovine calf serum. KYSE180 and KYSE450 cell proliferation was measured using the Cell Counting Kit 8 (Dojindo) according to the manufacturer's instructions. Data are presented as the means ± SD from three independent experiments.
Colony formation assay. Cells were seeded in 60 mm plates (1000 cells/plate) and cultured for 10-12 days. The cells were then fixed with 4% formaldehyde, diluted in phosphate-buffered saline, stained with 2% crystal violet diluted in water and photographed.
Dual-luciferase reporter assays. For m 6 A reporter assays, the DNA fragments of APC-Last Exon containing the WT m 6 A motifs, as well as the mutated motifs (m 6 A was replaced by T), were inserted into the Xhol site of pMIR-REPORT luciferase reporter vector. Dual-luciferase reporter assays were performed in HEK 293T cells 57 .
APC-Last Exon with WT m 6 A sites: 5′-TGAACTCTATTTCAGGAACCAAACAAAGTAAAGAAAACCAAGTAT CCGCAAAAGGAACATGGAGAAAAATAAAAGAAAATGAATTTTCTCCCAC AAATAGTACTTCTCAGACCGTTTCCTCAGGTGCTACAAATGGTGCTGAAT CAAAGACT-3′ APC-Last Exon with the mutated m 6 A sites: 5′-TGATCTCTATTTCAGGAACCAAACAAAGTAAAGAAAACCAAGTATC CGCAAAAGGAACATGGAGAAAAATAAAAGAAAATGAATTTTCTCCCACA AATAGTACTTCTCAGTCCGTTTCCTCAGGTGCTACAAATGGTGCTGAATC AAAGTCT-3′ Measurements of glucose consumption and lactate production. Cells were seeded in culture dishes and the medium was changed when cells reached 50% confluence. After incubation for 12-24 h, the culture medium was collected. The glucose levels were detected by a glucose colorimetric assay kit (#K606, BioVision), the lactate levels were detected by a lactate colorimetric assay kit (#K627, BioVision) according to the manufacturer's instructions and values were calculated as previously described 51 .
Animal experiments. For the subcutaneous implantation model, 1 × 10 6 cells were injected subcutaneously into the flank regions of female BALB/c nude mice (4-5 weeks). The width (W) and length (L) of the tumours were measured every week, and the volume (V) of each tumour was calculated using the formula V = (W 2 × L/2). All animal experiments were approved by the Animal Care and Use Committee of the Cancer Hospital of the Chinese Academy of Medical Sciences.
RIP assays. A Magna RIP kit (17-700, Millipore, MA) was used to perform the RIP assays. Sufficient numbers of KYSE180 or KYSE450 cells were lysed by RIP lysis buffer and the supernatant of the RIP lysate was incubated with specific antibodies on beads overnight at 4°C. After washing, RNA was extracted and analysed by reverse-transcriptase quantitative PCR.  15 . Briefly, fragmented mRNA was incubated with an antim 6 A polyclonal antibody (Synaptic Systems, 202003) in IPP buffer (10 mM Tris-HCl, 150 mM NaCl, 0.1% NP40, pH 7.4) for 2 h at 4°C. The mixture was then immunoprecipitated by incubation with protein-A beads (Thermo Fisher) at 4°C for an additional 2 h. The bound RNA was eluted from the beads with m 6 A (Berry & Associates, PR3732) in IPP buffer and then the RNA was extracted with TRIzol reagent (Thermo Fisher) following the manufacturer's instructions. Purified RNA was used for RNA sequencing library generation with the NEBNext® Ultra™ RNA Library Prep kit (NEB). Both the input sample without immunoprecipitation and the m 6 A IP samples were subjected to 150 bp paired-end sequencing on an Illumina HiSeq 4000 sequencer.
Sequencing data analysis. The paired-end reads were quality controlled by Q30 followed by removing of 3′-adapters and low-quality reads by cutadapt software (v1.9.3). Then, clean reads were aligned to the reference genome (UCSC HG19) by Hisat2 software (v2.0.4). The m 6 A peaks were predicted by Model-based Analysis of ChIP-Seq (MACS) software 58 . Motif enrichment analysis of predicted m 6 A peaks was performed by HOMER 59 and distribution of m 6 A peaks was analysed by R package MetaPlotR 60 . Differentially methylated sites with a fold-change cut-off of ≥2 and false discovery rate cut-off of ≤0.00001 were identified by diffReps 61 . Raw counts of mRNA sequencing were got by HTSeq software (v0.9.1) and normalized by edgeR software. Then, the peaks were identified by MACS. GO analysis involving CC was performed with the database for annotation, visualization and integrated discovery. The p-value denotes the significance of GO term enrichment of the genes.
Analysis of publicly available MeRIP-Seq and CLIP data. To study m 6 A level of APC mRNA in different types of cancer cells, the published MeRIP-Seq and CLIP data involved in this study were downloaded from GSE134380 30 , GSE87190 31 , GSE76367 17 , GSE128443 32 , GSE93911 33 , GSE102336 34 , GSE112795 35 and GSE106122 36 . The percentage rank of APC m 6 A among all the gene transcripts with m 6 A in the indicated cancer cell types from published MeRIP-Seq and miCLIP data sets are shown in Supplementary Fig. 3c. To analyse the m 6 A peak of APC mRNA in available miCLIP data (Fig. 3f), the published miCLIP data were downloaded from GSE71154 37 , GSE98623 38 , GSE122948 39 , GSE63753 40 and GSE86336 41 . To study whether YTHDF1-3 bound APC mRNA, the published YTHDF1-3 CLIP data were downloaded from GSE63591 18 , GSE49339 44 , GSE86214 46 and GSE78030 47 . Reads were analysed and aligned to the reference genome (UCSC HG19) by Hisat2 software (v2.0.4). Visual files were shown by IGV v2.8.3 in Fig. 3f and Supplementary Fig. 5a. 13 C metabolic flux analysis and measurement of glycolytic intermediates. Steady-state labelling of glycolytic intermediates was accomplished by culturing KYSE450 cells (5 × 10 6 cells/sample) in RPMI-1640 medium (Gibco) containing 2 g/L of D-glucose (U-13 C 6 , 99%; Sigma-Aldrich) for 6 h. All treatments were conducted in quintuplicate. Intracellular metabolites were then extracted and measured. The Dionex Ultimate 3000 UPLC system was coupled to a TSQ Quantiva Ultra triple-quadrupole mass spectrometer (Thermo Fisher, CA), equipped with a heated electrospray ionization probe. The source parameters are as follows: capillary temperature: 350°C; heater temperature: 300°C; sheath gas flow rate: 35; auxiliary gas flow rate: 10. Tracefinder 3.2 (Thermo, USA) was applied for metabolite identification and peak integration.
Statistical analysis. We used unpaired or paired Student's t-tests to compare means between groups and all data are expressed as the mean ± SD. The survival analyses were performed using the Kaplan-Meier method to plot survival curves and the log-rank test to compare survival rates. The correlation between METTL3 and APC level was analysed by the Pearson correlation coefficient. p-values < 0.05 were considered to be significant. All statistical tests were two-sided.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data of Figs. 1a and 7a, and Supplementary Fig. 1a are available in a public repository from the TCGA website. The clinical records and RNAseqV2 level 3 gene level ESCC data were downloaded from TCGA [http://xena.ucsc.edu/welcome-to-ucscxena]. Analysis data in Fig. 1a is available in UALCAN [http://ualcan.path.uab.edu/cgibin/TCGAExResultNew2.pl?genenam=METTL3&ctype=ESCA] 62 . The clinical records and genome-wide gene expression profiles (Affymetrix GeneChip Human Exon 1.0 ST arrays) of a total of 119 paired ESCC data sets were downloaded from the Gene Expression Omnibus with accession number GSE53625 63 . Gene transcription estimates for each gene were analysed using Robust Multiarray Average (RMA) software. The MeRIP-seq and mRNA-seq data have been deposited into the Gene Expression Omnibus repository under accession number GSE154555. The remaining data are available within the article, Supplementary Information or available from the authors upon request. Source data are provided with this paper.