m⁶A-mediated regulation of PBX1-GCH1 axis promotes gastric cancer proliferation and metastasis by elevating tetrahydrobiopterin levels

2.1 Patients and tissue specimens

Tumor and paired para-carcinoma tissues were collected from 100 patients with GC, diagnosed at the First Affiliated Hospital of Sun Yat-sen University (Guangzhou, Guangdong, China) between January 2011 and December 2016. All patients were diagnosed with GC preoperatively and underwent radical excision. Patients who died within three months after surgery or had more than one primary tumor been excluded. All tissue specimens were separated and frozen at -80℃ or formalin-fixed and paraffin-embedded. Experienced pathologists from the First Affiliated Hospital of Sun Yat-sen University reviewed the tissue sections. Ethical approval for research on human subjects was obtained from the Institutional Review Board of the First Affiliated Hospital of Sun Yat-sen University. Each participant signed an informed consent form before participating in the study.

2.2 Cell lines, cell culture, and transfection

Human gastric cancer cell lines (AGS, MKN1, SGC7901, and MGC803) were purchased from the Chinese Academy of Sciences Shanghai Branch Cell Bank (Shanghai, China). AGS cells were cultured in DMEM/F12 (Gibco, ThermoFisher, Waltham, MA, USA) and MKN1, SGC7901, and MGC803 cells were cultured in RPMI-1640 (Gibco) supplemented with 10% fetal bovine serum (FBS; Biological Industries, Israel) and 1% antibiotics (penicillin/streptomycin, Gibco) at 37°C with 5% CO₂. The cell lines were tested for potential mycoplasma contamination by PCR and confirmed mycoplasma-negative.

Drug treatments, such as STM2457 (MedChemExpress, New Jersey, USA), were performed as previously described [25]. Briefly, the cells were grown on 6 well culture plates with 10% FBS medium 24 h before STM2457 treatment. Then, STM2457 was added to the medium at a concentration of 10 μmol/L.

According to the manufacturer's instructions, siRNA transfection was performed using Lipofectamine RNAi Max (Invitrogen, Thermo Fisher Scientific, Inc., Waltham, MA, USA). Briefly, siRNAs were diluted in the 100 μL Opti-MEM (Gibco) at 5 nmol/L and mixed with 5 μL Lipofectamine RNAi Max. After incubation at room temperature for 20 min, the mixture was added to the cell culture medium, and the cells were collected after 48 h. The sequences of specific siRNAs are listed in Supplementary Table S1. Overexpression plasmids were constructed using pEZ-Lv201 (iGene Biotechnology, Guangzhou, Guangdong, China), and the knockdown plasmids were constructed using LVRU6GP (iGene Biotechnology). All plasmids were constructed by iGene Biotechnology Co., Ltd. Transfection of plasmids was performed using Lipofectamine 3000 (Invitrogen). Briefly, 2 μg plasmids and 5 μL Lipofectamine 3000 were diluted in 150 μL Opti-MEM and 5 μL Lipofectamine 3000 were diluted in 150 μL Opti-MEM. Then, the plasmids and Lipofectamine 3000 solution were mixed and incubated at room temperature for 15 min. Finally, the mixture was added onto 6 well plates, and cells were collected after 48-72 hours for further studies.

2.3 Animals and animal models

BALB/c male nude mice (5-6 weeks old) were purchased from Specific Pathogen Free (Beijing) Biotechnology Co., Ltd. (Beijing, China) and maintained in specific pathogen-free facilities. The Institutional Ethical Board approved animal studies of the First Affiliated Hospital of Sun Yat-sen University.

For tumor xenograft models, MGC803 cells (5×10⁶) with METTL3 knockdown and negative control were subcutaneously injected into the right axilla of nude mice (n = 6 per group). Tumor volume was monitored every other day (volume = length × width² × 1/2). When the largest tumor diameter reached 1 cm, the mice were sacrificed via a cervical dislocation manner. The tumors were weighed, imaged, and fixed in 4% paraformaldehyde or frozen for further analyses. To investigate the function of SPRi3, xenograft models were first established using MGC803 cells overexpressing METTL3 and a blank control. After one week, SPRi3 (300 mg/kg in 10 μL DMSO; Solarbio, Beijing, China) was injected into the tumors, and an equal volume of DMSO was injected as a negative control. The injection was performed weekly until the mice were sacrificed after 8 weeks. The tumors were resected, photographed, and fixed in 4% paraformaldehyde for further analysis.

For the lung metastasis model, MGC803 cells (luciferase labeled, 1×10⁶) with METTL3 knockdown and negative control were injected intravenously into the tail vein of nude mice. Six weeks after injection, fluorescein sodium salt (Solarbio) was injected into the tail vein of the mice. The bioluminescence images were captured using an IVIS Lumina II (Xenogen Corp., Alameda, California, USA). After 8 weeks, the mice were sacrificed by cervical dislocation, and the lungs were resected, photographed, and fixed in 4% paraformaldehyde for further analyses.

For the popliteal lymph node metastasis model, MGC803 cells (Luciferase labeled, 1×10⁶ in 20 μL PBS) were injected into the left footpad of mice (twice in three days). Six weeks after injection, fluorescein sodium salt (Solarbio) was injected into the tail vein of the mice. The bioluminescence images were captured using an IVIS Lumina II (Xenogen Corp.). After 8 weeks, the mice were sacrificed by cervical dislocation. The left legs' footpad and popliteal lymph nodes were resected, photographed, and fixed in 4% paraformaldehyde for further analysis.

2.4 RNA extraction and real-time PCR (qPCR) analysis

According to the manufacturer's instructions, total RNA was extracted from cells and tissues using RNAiso (Takara, Kusatsu, Japan). Briefly, the cells in 6 well plates were lysed in 500 μL RNAiso on ice and then mixed with 200 μL chloroform(Baishi, Guangzhou, Guangdong, China). After incubating for 10 min on ice, the mixture was centrifuged at 12000 rpm for 15 min, and the chloroform was collected. Next, 500 μL isopropanol (Baishi) was added to the mixture and incubated for 10 minutes. After centrifugation at 12000 rpm for 10 min, the RNAs were placed at the bottom of the tubes. Then, the RNA precipitate was washed with 75% ethyl alcohol (Baishi) and diluted in diethyl pyrocarbonate (DEPC) water (Takara).

As instructed, quantifications of specific RNA were used SYBR® Green I (Accurate Biotechnology, Changsha, Hunan, China). Briefly, 5 μL SYBR Green, 3 μL DEPC water, 1 μl primers and 1 μl cDNA were prepared for each sample, and the qPCR protocol was run using a LightCycler® 480 II (Roche, Basel, Switzerland). The fold change was calculated using the relative quantification method (2^−ΔΔCt). The qPCR primers used are listed in Supplementary Table S2.

2.5 Cell proliferation, colony formation, migration, and invasion assays

The cells were seeded in 96-well plates at 1,000 cells per well for the cell proliferation assays. Cell proliferation was evaluated using 10% CCK-8 (SolarBio) diluted in standard culture media for 2 h. Proliferation rates were determined at 0, 24, 48, 72, and 96 h after seeding, and quantification was performed on a microtiter plate reader (ALLSHENG, Hangzhou, Zhejiang, China) measured at UV wavelength (λ) = 450 nm.

For colony formation assay, cells were seeded in 6-well plates at a density of 500 cells/well. The cells were then cultured in complete media for 4-6 weeks until the size of a single colon reached 2-5 mm. Then the colonies were fixed by 4% paraformaldehyde and stained by 1% leucocrystal violet for 5 minutes.

For the cell migration and invasion transwell assays, 50,000 cells in 400 μL media with 10% fetal bovine serum were plated on the top chambers of Transwell Clear Polyester Membrane Inserts (for the migration assay, Corning Costar, New York, USA) and BioCoat Matrigel Invasion Chambers (for the invasion assay, Corning Costar), while culture media with 10% FBS was applied to the bottom. After 16-72 hours, migrated or invaded cells were stained with crystal violet and counted under a microscope (Olympus, Tokyo, Japan).

2.6 m⁶A dot blot assay

The dot blot assay was performed according to an online protocol (https://en.bio-protocol.org/e2095). Briefly, RNA was isolated from cells as previously described and denatured at 95°C for 2 min. Then, 2000 ng of mRNA was spotted onto a Hybond-N⁺ membrane (Cytiva, Shanghai, China) and crosslinked by UV (4000 × 100 μJ/cm², twice; Spectronics, USA). The membrane was then washed and incubated in blocking buffer (5% skim milk, EpiZyme, Guangzhou, Guangdong, China) and then gently shaken overnight with an anti-m⁶A antibody (synaptic system, 202003; Goettingen, Germany) at 4°C. The membrane was then washed, incubated with an anti-mouse antibody (EpiZyme), and washed again. Finally, the membrane was exposed to Hyperfilm ECL (Thermo Fisher, Waltham, MA, USA), and images were acquired using a ChemiDoc™ imaging system (Bio-Rad Laboratories, Inc., Hercules, CA, USA). Methylene blue (MB Solarbio) was used to interact with mRNA and as a loading control.

2.7 m⁶A-modified RNA immunoprecipitation (Me-RIP) and RNA immunoprecipitation (RIP)

Cells in 150 mm dishes were crosslinked by UV (4000 × 100 μJ/cm², twice; Spectronics) and collected in PBS. The RNA-m⁶A complexes were isolated using 0.5% NP-40 (Sigma-Aldrich, Merck Millipore, Darmstadt, Germany). After pelleting DNA via centrifugation, the suspension was added to 5 μg anti-m⁶A (for Me-RIP, synaptic system), anti-METTL3 (for RIP, Proteintech), or anti-IgG (EpiZyme) and incubated overnight at 4°C. Then, 40 μL protein A/G magnetic beads (Invitrogen, Thermo Fisher Scientific, Inc.) were added to the mixture for 1 h, and the beads were collected and washed twice with wash buffer (0.1% SDS and 0.5% NP-40 in PBS, Gibco). The RNA-m⁶A complex was de-crosslinked by heating at 45°C for 1 h. Finally, RNAiso (Takara) was added to the beads for RNA isolation. qPCR measured the quantification of m⁶A-enriched PBX1. The enrichment percentage was calculated using the relative quantification method: 2^ (CTIP-CTinput).

2.8 m6A-modified RNA immunoprecipitation-sequencing (MeRIP-seq)

For MeRIP-seq, the RNAs were purified by Spin Column RNA Cleanup & Concentration Kit (Sangon Biotech, Shanghai, China). Total RNA was greater than 120 μg, and the integrity and quantity of each RNA sample were assessed using agarose gel electrophoresis and a NanoDropTM instrument (ThermoFisher). According to the manufacturer's protocol, intact mRNA was first isolated from total RNA samples using an Arraystar Seq-StarTM poly(A) mRNA Isolation Kit (Aksomics, Shanghai, China). The isolated mRNA was chemically fragmented into 100-nucleotide-long fragments by incubation in fragmentation buffer (10 mmol/L Zn²⁺ and 10 mmol/L Tris-HCl, pH 7.0), and the size of the fragmented mRNA was confirmed using agarose gel electrophoresis. Then, m⁶A-methylated mRNAs were immunoprecipitated with an anti-N6-methyladenosine antibody (synaptic system), as previously described, and an aliquot of the fragmented mRNAs was collected. The major procedures included immunoprecipitation, washing, and elution. The eluted m⁶A mRNA fragments were then concentrated for RNA-seq library construction. A KAPA Stranded mRNA-seq Kit (F. Hoffmann-La Roche Ltd, Grenzacherstrasse, Basel, Switzerland) was used to create RNA-seq libraries for m⁶A antibody-enriched and input mRNAs (Illumina, San Diego, CA, USA). The prepared libraries were diluted to a final concentration of 8 pmol/L, and clusters were generated on an Illumina cBot using a HiSeq 3000/4000 PE Cluster Kit (#PE-410-1001, Illumina), followed by sequencing on an Illumina HiSeq 4000. Finally, all reads were mapped to the human genome 19 using BWA tools and annotated using miRBase version 21 (https://www.mirbase.org) and miRDeep2 (https://www.mdc-berlin.de/8551903/en/). The data were analyzed using R software (Free Software Foundation, Boston, MA, USA).

2.9 Chromatin immunoprecipitation (ChIP)

The ChIP assay was performed using a ChIP assay kit (SimpleChIP^® Plus Sonication Chromatin IP Kit (#56383, Cell Signaling Technology, MA, USA). Briefly, 5×10⁶ GC cells were fixed with 1% formaldehyde (Baishi), quenched with glycine (#G8200, Solarbio) at room temperature, following which they were collected, washed, and resuspended in lysis buffer (Cell Signaling Technology). The sonicated chromatin solution was immunoprecipitated with anti-IgG (DIA-AN, Q6004; Wuhan, Hubei, China) and anti-PBX1 (Abnova, H00005087-M01; Tsukiji, Tokyo, Japan). Immunoprecipitated DNA was purified by column collection and analyzed by qPCR. The enrichment percentage was calculated using the relative quantification method: 2^{CTChIP-CTinput}.

2.10 The bioinformatics analysis of online database

We downloaded the clinical data from the database of The Cancer Genome Atlas (TCGA) (https://cancergenome.nih.gov/) and Genotype-Tissue Expression (GTEx) (https://gtexportal.org/home/). The differentially expressed genes were calculated in R version 4.1.2 with EdgeR package [26]. CHIPBbase v2.0 [27] and JASPAR database [28] were used to identify potential downstream genes of transcription factors. For CHIPBase analysis, target transcription factors were imported and the binding region was chosen as “up/down 1000bp”. All the genes output from the database were downloaded for further analysis. For JASPR database, target transcription factors were selected and the sequence of 1000bp upstream of potential downstream genes were imported in “Scan” box. Relative profile score threshold was 80%.

2.11 Untargeted sequence of metabolism

Novogene Co., Ltd. (Beijing, China) performed the untargeted metabolomics sequencing. Briefly, the samples were placed in EP tubes and resuspended with prechilled 80% methanol (Baishi) and 0.1% formic acid (Rhawn, Shanghai, China) by vortexing. The samples were then melted on ice and washed for 30 s. After sonification for 6 min, the samples were centrifuged at 5,000 rpm at 4°C for 1 min. The supernatant was freeze-dried and dissolved in 10% methanol. Finally, the solution was injected into the Liquid Chromatography with tandem mass spectrometry (LC-MS-MS) system analysis (SCIEX Headquarters, Framingham, MA, USA).

2.12 Enzyme-linked immunosorbent assay (ELISA)

The NO levels in cells were detected by colorimetric ELISA-like assay via the Nitric Oxide (NO) assay kit (BioAssay Systems; Hayward, CA, USA) according to the manufacturer's instructions. Briefly, the cell lysis was added in 96-well plates and incubated with antibody from the kit. After washing the antibody away, reaction reagents were added and the absorbancy at λ = 405 mm was read.

2.13 Statistical Analysis

Overall survival (OS) is defined as the interval between the first radical resection to the date of death from any cause or to the date of the last follow-up visit. And the Progression-Free-Survival (PFS) is defined as the interval between the first radical resection to the first disease progression or death. Kaplan-Meier analysis was used to analyze the survival fraction.

All statistical values were calculated using Statistical Product and Service Solutions (SPSS) 22.0 (Chicago, IL, USA). Experimental studies were analyzed by independent sample t-test to compare two groups and by one-way ANOVA followed by post hoc Tukey's test to compare multiple groups. Kaplan-Meier analysis and log-rank tests were used to evaluate differences in patient survival. The Spearman rank correlation test was used to determine statistical correlations. Statistical significance was set at P < 0.05.

3.1 METTL3-mediated m⁶A modification promotes GC malignant progression

It has been conclusively demonstrated that METTL3 expression level was higher in GC tissues than in the adjacent normal gastric mucosa in the TCGA database and clinical patient samples. Elevated METTL3 levels were associated with worse long-term outcomes in GC patients [17]. To examine whether the elevated METTL3 proteins caused the outcome deterioration, we first detected the basic METTL3 protein levels in GC cell lines (Supplementary Figure S1A) and selected the AGS and SGC7901 cell lines with low METTL3 expression to construct METTL3 overexpressing cell models (Supplementary Figure S1B). The proliferation, colony formation, migration, and invasion abilities were significantly improved by METTL3 overexpression (Supplementary Figure S1C-E). In contrast, knockdown of METTL3 (Supplementary Figure S2A) significantly suppressed the proliferation, colony formation, migration, and invasion abilities of AGS, MKN1, and MGC803 cells (Supplementary Figure S2B-D). Considering the m⁶A catalytic ability and translation enhancement effect of METTL3 protein, we used plasmids containing catalytically inhibited METTL3 (aa395-398, DPPW→APPA), as previously reported [17], to transfected GC cells (Supplementary Figure S3A and B). The m⁶A dot blot results showed that the catalytic mutant did not affect the total RNA m⁶A modification level. At the same time, overexpression of wild type (WT) or knockdown of METTL3 significantly influenced RNA m⁶A levels (Supplementary Figure S3C). Moreover, catalytically inhibited METTL3 abolished METTL3 promoting proliferation, colony formation, migration, and invasion abilities (Supplementary Figure S3D-F). STM2457, a specific inhibitor of METTL3 transmethylation function, also impaired METTL3-mediated proliferation, colony formation, migration, and invasion (Supplementary Figure S4). Collectively, these results suggested that METTL3 promoted GC proliferation, colony formation, migration, and invasion in an m⁶A-dependent manner.

Furthermore, the effect of METTL3 on proliferation and metastasis in vivo was assessed. MGC803 cells treated with METTL3 shRNAs (shM3#1 and shM3#2) or control shRNAs (shNC) were employed. Compared to the negative control, in xenograft models, cells with METTL3 knockdown displayed significantly suppressed tumor growth, both in tumor volume and weight (Figure 1A-C, Supplementary Figure S5). IHC images showed that the proliferation biomarker Ki67 was decreased upon METTL3 knockdown (Figure 1D). Additionally, knockdown of METTL3 significantly inhibited lung metastasis in vivo, as reflected by bioluminescence imaging and H&E staining (Figure 1E-G). As lymph node metastasis is the most common metastasis in GC, we established a footpad-popliteal lymph node metastasis model to investigate the function of METTL3 in lymph node metastasis (Figure 1H). Post harvesting the popliteal lymph nodes after 8 weeks, the lymph nodes from the mice with knockdown of METTL3 cells were smaller than those from negative controls (Figure 1I and J). H&E staining revealed that lymph nodes of METTL3 knockdown groups had a lower percentage of GC metastasis (Figure 1K and L). These results suggested that METTL3 promoted GC proliferation and invasion both in vitro and in vivo.

3.2 METTL3 regulates PBX1 mRNA stability through m⁶A modification

To explore the molecular mechanism of METTL3 in GC progression, we conducted MeRIP-seq combined with RNA-seq. The MeRIP-seq showed that almost 50% of m⁶A modification was located in the three prime untranslated regions (3′-UTR) region and 25% in the five prime untranslated regions (5'-UTR; Supplementary Figure S6A and B). Most of these peaks were read both in METTL3 knockdown and negative groups (negative control: 35133; METTL3 knockdown: 33134). And the peaks were mapped in 11755 and 11382 genes respectively (Supplementary Figure S6C). Further analysis revealed that 604 m⁶A transcripts were lost, and 123 peaks were gained significantly after METTL3 knockdown (Supplementary Figure S6D). Since METTL3 is an m⁶A writer, we focused on analyzing the genes with lost peaks on METTL3 knockdown. On the other hand, RNA-seq revealed that the expression of 969 genes changed more than 2-fold after the knockdown of METTL3 in AGS cells (Supplementary Figure S6E). We then combined the results of the two sequencing analyses and found that 7 genes overlapped (ZNF264, SMUG1, ZNF668, ZBTB43, ZNF445, PBX1, and ZNF324B; Figure 2A, Supplementary Table S3).

To validate the downstream gene candidates, we used qPCR to measure the mRNA levels of the 7 genes mentioned above in METTL3 (both WT and mutant) overexpressed cells, METTL3 knockdown cells, and corresponding control cells. The results indicated that only ZNF668 and PBX1 transcription levels were significantly downregulated upon METTL3 knockdown and upregulated by elevated METTL3 levels but not influenced by mutant METTL3 levels (Figure 2B, Supplementary Figure S7A). Consistently, the protein level of PBX1 was positively correlated with METTL3 levels in GC cell lines (Figure 2C, Supplementary Figure S7B), whereas mutant METTL3 did not affect PBX1 levels (Supplementary Figure S7C). Moreover, IHC staining of our previous xenograft tumor samples showed that PBX1 expression was inhibited by METTL3 knockdown (Figure 2D). However, protein levels of ZNF668 decreased only in MKN1 cells upon METTL3 overexpression and increased only in SGC7901 cells upon METTL3 knockdown (Supplementary Figure S7D and E). TCGA analysis also confirmed a positive correlation between METTL3 and PBX1 in GC tissues (Supplementary Figure S7F). These data suggested that PBX1, but not ZNF668, may be directly regulated by METTL3.

MeRIP-seq revealed two m⁶A peaks on the 5'UTR and 3'UTR of PBX1 mRNA that decreased upon METTL3 knockdown (Supplementary Figure S7G). Furthermore, MeRIP analysis showed that PBX1 mRNA was significantly enriched by METTL3 antibody, and the enrichment was increased upon METTL3 overexpression (Figure 2E). The MeRIP-qPCR analysis indicated that the m⁶A-specific antibody significantly enriched the PBX1 mRNA, and the m⁶A-specific enrichment was decreased significantly in METTL3-knockdown AGS cells (Figure 2F). We employed actinomycin D to block the RNA transcript of AGS cells as m⁶A could regulate the stability of RNA targets. The PBX1 mRNA degraded more rapidly upon METTL3 knockdown and was highly stable in METTL3^wt-overexpressing AGS cells than those expressing METTL3^mut (Figure 2G and H). The IF images of GC tumor and paired adjuvant normal samples confirmed that METTL3 and PBX1 had higher levels in GC tissues than those in normal adjuvant tissues (Figure 2I). We also investigated whether other m⁶A enzymes affected PBX1 levels. However, there was no variation in PBX1 protein levels in ALKBH5 or METTL14 knockdown (Supplementary Figure S7H and I). These findings indicated that METTL3-mediated m⁶A modification regulated PBX1 levels by affecting PBX1 mRNA stability.

3.3 METTL3 promotes GC malignant process by elevating PBX1 expression

Both PBX1 and ZNF668 are characteristic oncogenes in some cancer types, but the biological functions of these two genes have not been determined in GC. We explored the RNA profile data of GC patients in the TCGA database. The statistical results showed that patients with higher expression levels of PBX1 or ZNF668 had worse overall and progression-free survival (Figure 3A, Supplementary Figure S8A and B). However, we combined the RNA profile data in TCGA and GTEx databases and found no differences between tumor and normal tissues on PBX1 and ZNF668 mRNA levels (Supplementary Figure S8C). Next, we used two specific siRNAs to inhibit PBX1 and ZNF668 in AGS cell lines. On PBX1 knockdown (Supplementary Figure S8D and E), proliferation viability was significantly suppressed (Figure 3B), and migration and invasion abilities were severely impaired (Figure 3C, Supplementary Figure S8F). However, the knockdown of ZNF668 had the opposite effect on the proliferation, migration, and invasion ability of AGS cells (Supplementary Figure S8G-I). These results suggested that PBX1, but not ZNF668, was a potential oncogene in GC and may be a downstream gene of METTL3 in GC progression.

Moreover, we used siRNAs to target PBX1 in AGS and MKN1 cells overexpressing METTL3 and quantified PBX1 protein levels by western blotting (Figure 3D, Supplementary Figure S9A). Knockdown of PBX1 dramatically inhibited the elevation of proliferation, migration, and invasion ability promoted by METTL3 overexpression (Figure 3E and F, Supplementary Figure S9B-D). Correspondingly, overexpression of PBX1 rescued the proliferation, migration, and invasion ability of METTL3-knockdown AGS cells (Figure 3G-I, Supplementary Figure S9E). These results indicated that PBX1 was a target gene of METTL3 that regulated the malignant process of GC.

3.4 PBX1 acts as a TF to induce GCH1 expression in GC

To comprehensively understand the function of PBX1 in GC development, we performed RNA-seq analysis in PBX1-overexpressing and negative control AGS cells, to explore the different genes. The data revealed that a series of pathways related to proliferation and metastasis, such as cell cycle, microtubule-based process, and DNA replication, were remarkably altered upon PBX1 overexpression (Supplementary Figure S10A and B). The molecular mechanism by which PBX1 promoted GC development was then investigated. Given that PBX1 acts as a TF in some cancer types, such as colorectal cancer [29], breast cancer [30], and acute lymphoblastic leukemia [31], we reviewed the CHIP-seq data of cells from colorectal and breast cancer in the CHIPBase database (http://rna.sysu.edu.cn/chipbase3/index.php) to explore potential downstream genes of PBX1. The detected DNA fragments were mapped onto 103 gene promoter regions in colorectal cancer and 3907 genes in breast cancer (Supplementary Table S4). The Venn diagram showed that 35 genes were enriched in both cancer types (Supplementary Figure S10C).

Furthermore, we compared the expression levels of the 35 genes mentioned above in GC and normal tissues in the TCGA database and found that only 23 genes had higher expression levels in GC (Supplementary Figure S10D). Among these genes, GCH1 was reported to be the first rate-limiting enzyme of BH₄ biosynthesis. BH₄ controls cell proliferation in liver [32] and colon cancer [33]. Thus, we wondered whether the METTL3-PBX1 axis could regulate GCH1 expression.

Accordingly, we then collected 10 tumor tissues and adjacent normal tissues of patients with GC. Western blotting showed that METTL3, PBX1, and GCH1 protein levels were increased in most tumor tissues (Figure 4A). Next, we determined the relationship between the expression levels of METTL3, PBX1, and GCH1 in GC cell lines. The results showed that METTL3 dramatically induced the mRNA and protein levels of GCH1 in an m⁶A-dependent manner (Figure 4B, Supplementary Figure S10E). In contrast, knockdown of METTL3 in AGS and MKN1 cells markedly suppressed GCH1 levels (Figure 4C and D). After treatment with specific siRNA of PBX1, GCH1 was significantly decreased (Supplementary Figure S10F). Moreover, the catalytic inhibition of METTL3 by STM2457 significantly decreased PBX1 and GCH1 protein levels in a time-dependent manner (Figure 4E). Clinical specimens demonstrated higher protein levels of METTL3 and GCH1 in GC than those in normal adjuvant tissues (Figure 4F). These results suggested that the METTL3-PBX1 axis induced GCH1 expression in GC.

To test whether GCH1 was regulated by PBX1 directly, we used specific siRNAs to block PBX1 expression in AGS cells and found that knockdown of PBX1 eliminated the promoting effect of METTL3 on GCH1 expression (Figure 4G, Supplementary Figure S10G). To explore the transcription mechanism of PBX1, JASPR online tools were used to predict the direct interaction sites between PBX1 and the GCH1 promoter region, which exhibited the likely domain with which the PBX1 protein may combined (Supplementary Figure S10H). Comparing the results from JASPR and CHIPBase, the binding site was suggested to be located approximately 500 bp upstream of the transcription start site (TSS) of GCH1. Next, the ChIP assay indicated that the PBX1 antibody significantly enriched the GCH1 promoter region, while the negative IgG had no such enrichment effect (Figure 4H). These results suggested that PBX1 acted as a TF to induce GCH1 expression, and therefore, GCH1 was the direct downstream gene of the METTL3-PBX1 axis.

3.5 BH₄ is the effector molecule of the METTL3-PBX1-GCH1 pathway

As GCH1 was the first rate-limiting enzyme of BH₄ biosynthesis [24] (Figure 5A), we measured the BH₄ levels in the lysates of cells with overexpression or knockdown of METTL3. BH₄ expression was downregulated by METTL3 knockdown and upregulated by METTL3^WT overexpression, similar to GCH1 expression; however, it did not change in the cells with METTL3^mut transfection (Figure 5B and C). In addition, the knockdown of PBX1 downregulated the intracellular BH₄ levels (Figure 5D). The results suggested that METTL3-PBX1 mediated GCH1 protein-regulated BH₄ biosynthesis in GC cells.

To explore the potential mechanism by which BH₄ promotes GC progression, artificial BH₄ analogues (sapropterin dihydrochloride) were added to the AGS cell medium. LC-MS showed that the levels of derivatives of BH₄, such as biopterin and 7,8-dihydropterin, increased markedly (Supplementary Figure S11A), which conformed to the cell membrane permeability of artificial BH₄. RNA-seq revealed that the expression profiles were completely discrepant in BH₄-treated cells (Supplementary Figure S11B). GO analysis also showed that the cell cycle and metabolic processes were significantly enriched in BH₄-treated cells (Figure 5E, Supplementary Figure S11C). These data suggested that BH₄ had a remarkable biological effect on GC cells.

BH₄ is a cofactor of nitric oxide synthase (NOS) for NO production. We measured the expression levels of NOS1-3 and NO levels in METTL3 overexpressed or knockdown cells. RT-qPCR revealed that only NOS2 was elevated upon METTL3 overexpression (Figure 5F), whereas NOS1 could not be detected in GC cells (data not shown). However, ELISA of NO production showed that cellular NO levels were unchanged (Figure 5G). Knockdown of METTL3 led to a decrease in NOS2-3, while NO levels slightly increased (Figure 5H and I) and BH₄ did not affect NOS or NO levels (Figure 5J and K). These results suggested that BH₄ could not regulate GC progression through NO synthases. Previous studies have shown that BH₄ regulates cellular metabolic processes depending on its antioxidant characteristics [24]. Therefore, untargeted metabolomic sequencing was performed to verify the metabolism variance in BH₄-treated AGS cells. A total of 1097 metabolites were annotated and catalyzed into different categories, such as benzenoids, nucleosides, nucleotides, analogues, lipids, and lipid-like molecules (Supplementary Figure S11D). It was shown that 22 of the mapped metabolites were downregulated and 49 were upregulated upon BH₄ treatment (Figure 5L). Consistent with the GO analysis of RNA-seq, the different metabolites included some types of nucleic acids and aromatic amino acids. However, the identification of metabolite functions is still incomplete, and our group will continue to investigate the significance of these metabolites in the future. Collectively, these data suggested that BH₄ was the effector molecule of the METTL3-PBX1-GCH1 axis.

3.6 METTL3-PBX1-GCH1 axis increases BH₄-induced GC progression

Finally, we investigated whether BH₄ promoted GC progression in vitro and in vivo. According to the CCK-8 and transwell analysis, the proliferation, migration, and invasion abilities of AGS cells were dramatically enhanced by BH₄ supplementation and restored the proliferation and metastasis defects caused by METTL3 knockdown (Figure 6A and B, Supplementary Figure S12A). In contrast, SPRi3, a small-molecule inhibitor of BH₄ synthesis, severely impaired cell proliferation, migration, and invasion abilities (Supplementary Figure S12B and C) and abolished the promoting effect of proliferation, migration, and invasion caused by METTL3 overexpression (Figure 6C and D, Supplementary Figure S12D). A xenograft model was used to verify the suppressive effect of SPRi3 on proliferation. The results showed that SPRi3 abolished the promoting effect of METTL3 on tumor growth (Figure 6E and F, Supplementary Figure S12E). In summary, BH₄ biosynthesis, induced via the METTL3-PBX1-GCH1 axis, promoted GC progression in vitro and in vivo.

To determine the mechanism in real GC patients, we measured BH₄ levels and found the levels to be significantly higher in GC than in normal adjuvant tissues (Figure 6G). We also collected tumor specimens from 117 GC patients at our clinical center with complete follow-up for 5 years. IHC staining was performed on METTL3, PBX1, and GCH1 proteins, and Kaplan-Meier analysis confirmed that these genes indicated a poorer OS of the patients (Figure 6H). Pearson tests revealed that there were positive correlations between METTL3, PBX1, and GCH1 (METTL3 vs. PBX1, r = 0.278, P = 0.0024; METTL3 vs. GCH1, r = 0.175, P = 0.05; PBX1 vs. GCH1, r = 0.38, P <0.001). These data suggested that the METTL3-PBX1-GCH1 axis had worse outcomes in GC patients.