BPGM as an intrinsic brake to constrain metastasis through phospho-epigenetic-mediated carnitine biosynthesis suppression

Highlights

• •BPGM is a novel intrinsic metabolic-epigenetic brake against tumor metastasis.
• •BPGM triggers a 2,3-BPG-dependent phospho-epigenetic relay to convert glycolytic flux into carnitine synthesis silencing.
• •Targeting BBOX1 with Meldonium reduces metastatic burden and offers a metastasis-preventive strategy.

Abstract

Metabolic adaptations that fuel metastatic dissemination are increasingly mapped, yet the existence of intrinsic metabolic "brakes" that actively restrain metastatic progression remains enigmatic. Here, we unveil bisphosphoglycerate mutase (BPGM) as a previously unrecognized metastasis suppressor that orchestrates a phospho-epigenetic relay linking glycolytic flux to carnitine-dependent fatty acid oxidation. Through high-resolution metabolomics, we discover that BPGM and its catalytic product 2,3-bisphosphoglycerate (2,3-BPG) constitute a metabolic checkpoint whose disruption predicts metastatic virulence in multiple cancers. Mechanistically, BPGM suppresses metastasis by triggering CDK1-T₁₄ phosphorylation-dependent assembly of an EZH2-H3K27me3 repressor complex that silences γ-butyrobetaine hydroxylase (BBOX1), the rate-limiting enzyme in carnitine biosynthesis. This phospho-switch mechanism converts glycolytic 2,3-BPG levels into epigenetic orchestrator, thereby starving metastatic cells of carnitine-required fatty acid oxidation. Hypoxia-mediated KDM4A-H3K9me3 cascade emerges as the upstream inactivator of this metabolic-epigenetic checkpoint, explaining how tumor microenvironmental stress liberates metastatic potential. Therapeutically, pharmacological BBOX1 inhibition with Meldonium recapitulates BPGM-mediated metastasis suppression in orthotopic models, reducing metastatic burden. These findings reveal BPGM as a metabolic gatekeeper that integrates bioenergetic sensing with chromatin remodeling to constrain metastatic competence, while hypoxia-mediated checkpoint failure unleashes carnitine-fueled metastatic progression. Targeting the hypoxia-BPGM-BBOX1 axis represents an innovative approach for metastasis-preventive therapy.

Introduction

Cancer remains the second leading cause of global mortality, with metastasis driving over 90 % of cancer-associated deaths through a multi-step cascade involving cell migration, invasion, dissemination, and colonization of distant organs [1,2]. Hypoxia-mediated metabolic reprogramming is a well-established hallmark of solid malignant tumors with high metastasis and mortality rates, including hepatocellular carcinoma (HCC), head and neck squamous cell carcinoma (HNSCC) and melanoma, yet conventional research has largely focused on cell-autonomous metabolic adaptations within the primary tumor, such as aerobic glycolysis and glutamine dependence [3]. While metabolic traits in tumors are increasingly characterized, the dynamic and stage-specific regulation of metabolism during the metastatic cascade remains poorly understood. Recent studies underscore metastasis not a passive outcome, but an actively metabolically programmed process. Metastasizing cells sequentially rewire their metabolism to meet the distinct demands of invasion, circulation, and colonization [2,4]. This plasticity converges into metabolite inflexibility at specific steps, creating targetable vulnerabilities that are unique to metastatic cells and opening new therapeutic avenues for preventing and treating tumor metastasis [5].

Accumulating evidence now delineates how dynamic metabolic adaptations specifically fuel each stage of the metastatic journey. These pro-metastatic metabolic mechanisms primarily operate through four interconnected dimensions: (1) Bioenergetic and biosynthetic reprogramming, where cancer cells enhance glycolysis (the Warburg effect), glutaminolysis, and fatty acid oxidation to generate ATP, reducing equivalents, and macromolecular precursors necessary for motility, membrane remodeling, and rapid proliferation at distant sites [[6], [7], [8]]; (2) Redox homeostasis maintenance, wherein circulating tumor cells upregulate pathways like the pentose phosphate pathway to regenerate NADPH, bolstering antioxidant defenses (e.g., glutathione system) to survive detachment-induced oxidative stress and anoikis [9,10]; (3) Metabolite-mediated signaling and epigenetic regulation, where oncometabolites such as lactate, succinate, fumarate, and acetyl-CoA directly modulate key signaling cascades (HIF, mTOR, NF-κB) and induce post-translational modifications (e.g., histone lactylation, acetylation) to drive epithelial-mesenchymal transition (EMT), stemness, and invasive gene programs [[11], [12], [13], [14], [15], [16]]; (4) Microenvironmental co-operation and remodeling, whereby cancer cells adapt to and modify the nutrient landscape of distant organs, utilizing locally abundant resources like fatty acids in lipid-rich omental or lymph node niches, or acetate in the brain, and secrete factors that alter stromal cell metabolism to forge a supportive pre-metastatic niche [[17], [18], [19], [20]]. However, while the pro-metastatic functions of metabolic rewiring are increasingly identified, a fundamental and underexplored dimension persists: the potential existence of intrinsic metabolic checkpoints or "brakes" that naturally constrain the metastatic phenotype remains unrecognized. Current paradigms are predominantly focused on how metabolism is co-opted to enable dissemination, leaving a critical gap in understanding whether specific metabolic configurations or enzymes might inherently suppress invasion and colonization. This conceptual gap raises several pivotal scientific questions: Do certain metabolic pathways or metabolites function as endogenous gatekeepers against metastasis? What are the precise molecular mechanisms, particularly at the metabolite-epigenetic-transcriptional interface, through which such restrictive metabolic signals impede pro-dissemination programs? Furthermore, how are these putative protective circuits disrupted during tumor evolution, especially under pervasive microenvironmental pressures such as hypoxia, which is a known driver of malignancy and therapy resistance?

This study aims to identify and characterize intrinsic metabolic checkpoints that constrain metastasis, shifting the paradigm beyond pro-metastatic metabolic adaptations. By elucidating how such suppressive circuits are dysregulated under tumor microenvironments, we seek to identify novel therapeutic targets to halt the metastatic progression.

Methods

Reagents

The following reagents were used: l-carnitine (541-15-1, Sigma-Aldrich, Saint Louis, MO, USA); 2,3-BPG (D5764, Sigma-Aldrich); CDK1 inhibitor, RO-3306 (S7747, Selleck, Shanghai, China). Chaetocin (S8068, Selleck); Meldonium (HY-B1836, MCE, Shanghai, China). Unless otherwise indicated, a final concentration of 1 mM l-carnitine, 1 mM 2,3-BPG or 5 μM RO-3306 was used to treat hepatoma cells. PBS or DMSO was used as a vehicle control.

Human HCC and HNSCC specimens

Human HCC tissues were obtained from 18 patients who underwent HCC resection and were pathologically confirmed as hepatocellular carcinoma at Xiangya Hospital of Central South University. Human HNSCC tissues were obtained from 15 patients who underwent HNSCC resection and were pathologically confirmed as head and neck squamous cell carcinoma at Hunan Cancer Hospital. Adjacent non-tumor liver tissues with a distance of 1.5 ∼ 3 centimeters from the tumor tissues were collected. The patients had not received any local or systemic anti-cancer treatments prior to the surgery, and no postoperative anti-cancer therapies were administered prior to relapse. All patients were followed postoperatively to assess survival rates and to monitor for recurrence and metastases. These patients were divided into two groups based on their clinical pathological status: those with microvascular invasion (+MVI, for HCC) or lymph node metastasis (NX, for HNSCC) and those without MVI (-MVI, for HCC) or lymph node metastasis (N0, for HNSCC) at the time of surgery. The metabolomic profiling was conducted on the primary tumor tissues and matched adjacent para-carcinoma tissues to investigate metabolic differences between primary tumors with acquired metastatic capability versus those without metastasis.

The relevant characteristics of the studied subjects are shown in Table S1 and Table S2. Informed consent was obtained from each patient, and the study was approved by the Ethics Committee of Medical Research of Xiangya Hospital Central South University (2024010087) and Hunan Cancer Hospital (KY2021109).

Cell lines

HEK293T cells, human hepatoma cell line SK-HEP-1, Human pharyngeal squamous cell carcinoma cell Fadu, mouse hepatoma cell line Hepa1-6 and mouse melanoma cell line B16-F10 were cultured in Dulbecco’s modified Eagle’s medium (DMEM, Gibco, ThermoFisher Scientific, Waltham, Massachusetts, USA) supplemented with 10 % fetal bovine serum (FBS, Gibco). Another human hepatoma cell line SNU-449 was maintained in RPMI 1640 medium (Gibco) supplemented with 10 % FBS (Gibco). All cells were cultured in a humidified atmosphere of 5 % CO₂ at 37°C.

Plasmid construction

Lentivirus expression vectors pCDH-BPGM, pCDH-Bpgm, pCDH-BBOX1, pCDH-shNC and pCDH-shBPGM were generated using pCDH—CMV-MCS-EF1-copGFP-T2A-Puro (System Biosciences, Palo Alto, CA, USA), which contained a copGFP expression cassette and was designated pCDH—Ctrl in this study. Firefly luciferase reporter vectors pGL3-basic-p(−2.0/+0.1k) and pGL3-basic-pBPGM-Mut were constructed based on pGL3-basic vector (Promega, Madison, WI, USA).

Lentivirus production and infection

For lentivirus production, HEK293T cells were co-transfected with the lentivirus expression vector that contained the target sequence and the packaging plasmid mix (Lenti-X HTX Packaging Mix, Clontech, Palo Alto, CA, USA) via calcium phosphate precipitation. The lentivirus supernatant was harvested and stored in aliquots at −80°C until use. Target cells, grown to 30 % confluence at 24-well plate, were incubated in 1 ml lentivirus supernatant supplemented with 10 μg/ml polybrene (Millipore, Billerica, MA, USA).

The stable cell lines were established by infecting SK-HEP-1, SNU-449, FaDu, Hepa1-6 or B16-F10 cells with lentivirus that expressed the target sequence. Sublines with stable expression of human BPGM and mouse Bpgm with full length sequence (SK-BPGM, SNU-BPGM, FaDu-BPGM, Hepa-BPGM and B16-F10-BPGM) and the control lines (SK-Ctrl, SNU-Ctrl, FaDu-Ctrl, Hepa-Ctrl and B16-F10-Ctrl), as well as SK-HEP-1, SNU-449 or FaDu cells with stable silencing of BPGM (SK-shBPGM, SNU-shBPGM, FaDu-shBPGM) and the control lines (SK-shCtrl, SNU-shCtrl, FaDu-shCtrl), were constructed.

In vitro migration assays

The migration of tumor cells was analyzed in 24-well Boyden chambers with 8-μm pore size polycarbonate membranes (Corning, NY, USA). Briefly, SK-HEP-1, SNU-449 or FaDu cells in serum-free DMEM or RPMI were placed into the upper chamber of 24-well Boyden chamber, while the lower chamber was filled with 600 μl 10 % FBS-containing DMEM/RPMI. After 10 hours of incubation, cells were fixed and stained with crystal violet. All the migrated cells were counted.

Cell counting assay

Cell counting assay was used to evaluate cell growth. Cells (6 × 10⁴ SK-HEP-1, 6 × 10⁴ SNU-449, 6 × 10⁴ FaDu) were seeded in a 12-well plate infected with pCDH—Ctrl/shNC or pCDH-BPGM/shBPGM lentivirus, and then cultured for 72 hours before cell counting by Countstar (ALIT Life Sciences, Shanghai, China).

Mouse tumor xenograft models

All procedures for animal experiments were performed in accordance with the Guide for the Care and Use of Laboratory Animals (National Institutes of Health publication no 80-23, revised 1996) and according to the Central South University Institutional Ethical Guidelines for animal experiments (CSU-2023-0467).

For Orthotopic Liver Tumor Model, Hepa-Ctrl and Hepa-BPGM cells (7 × 10⁵) were resuspended in Matrigel (R&D Systems) and then inoculated under the capsule of the left hepatic lobe of male C57BL6 mice at 5-6 weeks of age. Four weeks later, the xenografted mice were euthanized, and liver lobes (excluding the primary tumor site) as well as lung tissues were collected and then applied to evaluate the metastasis. The length (L) and width (W) of the dissected tumors were measured with calipers and the tumor volume (V) was calculated using the formula V = (L × W²) × 0.5. Aliquots of tumor tissues were freshly frozen in liquid nitrogen, or fixed in 10 % formalin, and embedded in paraffin. To evaluate the metastasis, serial sections from lungs and livers were stained with hematoxylin-eosin (HE) and screened for metastatic nodules.

For melanoma lung metastasis model, 1 × 10^6 B16-F10 melanoma cells are suspended in phosphate-buffered saline (PBS) and injected into C57BL/6 mice via tail vein. For treatment assay, three days after tumor inoculation, the mice were administered intraperitoneal injections at a dosage of 70 mg/kg once every two days and were euthanized 17 days later. Sacrifice mice at 17 days post-injection for optimal metastatic nodule counting. The melanoma nodules were counted under dissection microscope. To evaluate the metastasis, the lungs were stained with HE and screened for metastatic nodules.

Metabolomic profiling

Tissues were suspended with lysis buffer to a concentration of 15 mg/ml and were homogenized with small glass beads for 5 minutes by the Next Advance Bullet Blender. Plasma was lysed with lysis solution (methanol: acetonitrile: water 5:3:2 v/v/v) at 1:25 dilutions. Suspensions were then vortexed continuously for 30 minutes at 4°C and then centrifuged at 18,213 g for 10 minutes at 4°C. Supernatants were injected into the Ultra-High-Pressure Liquid Chromatography–Mass Spectrometry (UHPLC-MS) using a Vanquish UHPLC coupled to a Q Exactive MS (Thermo Fisher, Bremen, Germany). Samples were analyzed using a 5-minute gradient as previously described [[61], [62]]. In brief, metabolites were separated on a Kinetex C18 column (150 × 2.1 mm, 1.7 um, Phenomenex, 00F-4475-AN) by the following chromatography conditions: flow rate 0.45 ml/min, column temperature 45°C, and sample compartment temperature 7°C. Positive solvent gradient was as follows: 0–0.5 minute 5 % B, 0.5–1.1 minute 5 %–95 % B, 1.1–2.75 minutes hold at 95 % B, 2.75–3 minutes 95 %–5 % B, and 3–5 minutes hold at 5 % B (A: 0.1 % formic acid in water; B: 0.1 % formic acid in acetonitrile). Negative solvent gradient was as follows: 0–0.5minute 0 % B, 0.5–1.1 minute 0 %–100 % B, 1.1–2.75 minutes hold at 100 % B, 2.75–3 minutes 100 %–0 % B, and 3–5 minutes hold at 0 % B (A: 5 % acetonitrile/95 % water/1 mM ammonium acetate; B: 95 % acetonitrile/5 % water/1 mM ammonium acetate). Samples were randomized and run in positive and negative ion modes independently. The mass spectrometer was operated in full MS mode at resolution of 70,000, scan range 65–900 m/z, maximum injection time 200 ms, microscans 2, automatic gain control 3 × 106 ions, source voltage 4.0 kV (for both positive and negative ion modes), capillary temperature 320°C, and sheath gas 45, auxiliary gas 15, and sweep gas 0 (all nitrogen).

Raw data files were converted to mzXML format using RawConverter (Scripps Research Institute) and analyzed via Maven (Princeton University, Princeton, NJ). Instrument stability and quality control were assessed through replicate injections of a technical mixture every ten runs as described. The metabolomic data were normalized based on website MetaboAnalyst (https://www.metaboanalyst.ca/MetaboAnalyst/). Sample normalization was conducted by the sum. The different metabolites (fold change > 1.2) were listed in Table S3 and Table S4.

Isotopic-labeled glucose flux in vivo

To investigate glucose metabolism and flux dynamics in vivo, we employed an isotopic labeling approach using [U-¹³C6] glucose in C57BL/6 mice. Hepa-Ctrl and Hepa-BPGM-OE xenografts bearing mice were fasted for 6 hours prior to the experiment and then administered a bolus of 5 % (w/v) [U-¹³C6] glucose intravenously at a dose of 14 µmol/kg body weight. At selected time points (30 minutes), mice were sacrificed, blood and tumor tissues were rapidly harvested, snap-frozen in liquid nitrogen, and stored at −80°C. Metabolites were extracted as mentioned above and analyzed via Vanquish UHPLC coupled to a Q Exactive MS (Thermo Fisher, Bremen, Germany), which were listed in Table S5. Metabolite assignments and isotopologue distributions were performed through the Maven (Princeton, NJ).

Targeted quantification of l-carnitine in tumor tissues and plasma

Tumor tissues and plasma were suspended with a pre-chilled (−20°C) extraction solution containing stable isotope internal standards to a concentration of 15 mg/ml and were homogenized with small glass beads for 5 minutes by the Next Advance Bullet Blender. The subsequent steps followed established procedures. Quantification relied on the integrated peak areas of extracted ion chromatograms at the MS1 level. The concentration of l-carnitine was calculated.

Molecular docking

A molecular docking strategy was used to predict the binding affinity between 2,3-BPG and CDK1 or EZH2. Three-dimensional structures of CDK1 or EZH2 were obtained from UniPort database. The two-dimensinal structures of the 2,3-BPG were obtained from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/). All small molecules were processed into Pdb format using OpenBable for energy minimization in the 7.4 environment followed by hydrogenation with Pymol (https://www.pymol.org/) and converted into PDBQT format. Then, the affinity between 2,3-BPG and CDK1 or EZH2 were calculated via the AutoDock Vina docking mode (https://vina.scripps.edu/). Affinity ≤ −5.0 kcal/mol indicated a strong interaction. The PyMOL software was applied to visualize the results.

Immunofluorescence staining

Immunofluorescence staining assay was performed to examine the expression and localization of BPGM, cytokeratin 7 (CK7), BBOX1 and H3K9me3. Briefly, human HCC and HNSCC tissues paraffin sections (5 µm) were de-waxed and hydrated for H&E stain and immunofluorescence staining. The sections were incubated with antibodies against CK7 (1:500; ab181598, Abcam, Cambridge, United Kingdom) and/or BPGM (1:200; 17173-1-AP, Proteintech, Wuhan, China), BBOX1 (ab171959, Abcam) and H3K9me3 (61014, Active Motif, CA, USA) overnight at 4°C. Counterstaining was performed with antifade mounting medium with 4′,6-Diamidino-2′-phenylindole (DAPI) (P0131, Beyotime, Shanghai, China).

Analysis of gene expression

Real-time quantitative polymerase chain reaction (qPCR) assay was performed to evaluate the RNA levels. TRIzol reagent (Thermo Fisher Scientific, USA) was used to extract total RNAs from cells and the frozen tissues. 1 µg RNA was reverse transcribed to generate cDNA by using PrimeScript RT Master Mix Kit (TaKaRa Bio.Inc, Dalian, China). The primers used for qPCR were synthesized by Tsingke Biotechnology (Beijing, China) and listed in Table S6. Q-PCR reaction was carried out on MiniOpticon™ Real-Time PCR Detection System (BioRad, Hercules, CA, USA). The reaction solution consisted of 2.0 µL diluted cDNA, 0.2 µM of each paired primer and 1 × ChamQ Universal SYBR qPCR Master Mix (Vazyme Biotechnology, Nanjing, China). The housekeeping gene β-actin was used for as an internal control. The specificity of PCR products was examined by the melting curve at the end of the amplification and subsequent sequencing. To determine the relative quantitation of gene expression for both target and housekeeping genes, the comparative Ct (threshold cycle) method with arithmetic formulae (2 ^−∆∆Ct) was used.

Western blotting was performed to determine the protein levels. The antibodies used included mouse antibody against β-actin (BM0627, Boster, Wuhan, China), rabbit antibody against BPGM (17173-1-AP, Proteintech), EZH2 (F0281, Selleck), phospho-EZH2 (Thr345) (TA3584S, Abmart, Shanghai, China), phospho-CDK1 (Thr14) (AP1465, Abclonal, Wuhan, China), ubiquitin (10201-2-AP, Proteintech), HIF1α (36169, Cell Signaling Technology, CST, Beverly, MA, USA), H3K4me3 (91264, Active Motif), H3K79me3 (cat 49-1020, Thermos Fisher), H3K9me3 (61014, Active Motif), H3K27me3 (91168, Active Motif) and Histone 3 (F0057, Selleck).

Chromatin immunoprecipitation (ChIP) assay

SK-HEP-1 cells that overexpressing BPGM and the control cell lines were cross-linked by formaldehyde. The chromatin complexes were immunoprecipitated using anti-H3K27me3 antibody (91168, active motive), or isotype-matched IgG (negative control), then collected with Protein A/G MagBeads (P2108, Beyotime, Shanghai, China). The immunoprecipitated DNAs were analyzed by qPCR with primers listed in Table S6.

Dual-luciferase reporter assay

Potential hypoxia response elements (HREs) within the BPGM promoter region were predicted using the JASPAR database (https://jaspar.elixir.no/). Mutations were introduced into sequences fully matching the HRE motif, and wild-type and mutant sequences were cloned into pGL3 vectors carrying the Firefly luciferase gene. SK-HEP-1 cells were transfected with pGL3 constructs (expressing wild-type or mutant BPGM promoter region sequences) under normoxia or hypoxia condition. According to the dual-luciferase assay protocol, Renilla luciferase was used as an internal control, and luciferase activity was calculated based on different co-transfection conditions.

Statistical analysis

Data were expressed as the mean ± standard error of the mean (SEM) from at least three independent experiments. Unless indicated, student’s t-test was performed to compare the differences between two groups and one-way ANOVA was applied to compare more than two groups. All statistical tests were two-sided and P < 0.05 was considered to be statistically significant. All analyses were performed using GraphPad Prism version 8.0 (GraphPad Software, Inc., San Diego, CA, USA).

Results

Downregulated BPGM is common in high-metastatic tumors and correlated with high metastatic potential

In an effort to identify potential molecules and specific metabolic pathways that constrain the metastatic phenotype, we enrolled two independent cohorts of patients with highly metastatic and fatal malignancies: hepatocellular carcinoma (HCC) and head and neck squamous cell carcinoma (HNSCC). Each cohort was stratified into two groups based on disease-specific pathological criteria indicative of metastatic potential: the presence or absence of microvascular invasion (MVI) for HCC, and with or without lymph node metastasis (LNM) for HNSCC, at the time of surgery (Table S1 and S2). Primary tumor tissues and matched adjacent para-carcinoma tissues were collected from these groups. Subsequently, non-targeted metabolomic profiling was performed (Fig. 1A). To identify potential differential metabolic patterns, we first performed partial least squares discriminant analysis (PLS-DA). The PLS-DA model revealed distinct clustering between tumor and adjacent non-tumor tissues, and between the low- and high-metastatic groups, for both HCC and HNSCC (Fig. 1B-C). Using volcano plot analyses, we identified numerous differentially abundant metabolites (DAMs) in both comparisons (tumor vs. adjacent tissue and low- vs. high-metastatic groups) (Fig. S1A-B).

Fig. 1.

BPGM is downregulated in high metastatic tumors and its loss correlated with high metastatic potential. (A) Flowchart of Metabolomic Analysis. HCC, Hepatocellular Carcinoma, HNSCC, Head and Neck Squamous Cell Carcinoma. “-” indicates HCCs without microvascular invasion (MVI, n = 5); “+” indicates HCCs with MVI (n = 8). “N0” indicates HNSCCs without lymph node metastasis (n = 8); “NX” indicates HNSCCs with lymph node metastasis (n = 7). The metabolomic profiling was conducted on the primary tumor tissues and matched adjacent para-carcinoma tissues. (B) PLS-DA score plot of tumor and matched adjacent non-tumor tissues metabolites in both HCC and HNSCC. Pa, para-cancerous tissue; HCC (n = 13); HNSCC (n = 15). The plot displays the first two latent components. Ellipses denote the 95 % confidence region for each group. Data was log-transformed and Pareto-scaled prior to analysis. (C) PLS-DA score plot of low and high-metastatic tumors metabolites in both HCC and HNSCC. (D) The intersection of the DAMs from both HCC and HNSCC downregulated datasets revealed 32 shared metabolites. (E) A volcano plot illustrating the differential expression of the 32 downregulated metabolites in HCC and HNSCC tumor tissues with high versus low metastatic potential. 1/2,3‑BPG was labeled in red color. (F) 1/2,3-BPG was downregulated in HCC and HNSCC tumors with high metastatic potential. (G) Schematic diagram of the glycolytic bypass RLS pathway. RLS, Rapoport-Luebering Shunt. (H-I) Higher level of BPGM was significantly associated with a lower metastasis potential in HCC and HNSCC patients. The mRNA (n = 18) level of BPGM in HCC tissues were detected by qPCR analysis and the protein level of BPGM was detected by immunofluorescence staining, respectively. For H, “-”, n = 10; “+”, n = 8. For I, “-”, n = 5; “+”, n = 5; N0, n = 5; NX, n = 7. Scale bar,50 μm. (J) BPGM level was lower in advanced tumors. BPGM mRNA levels were analyzed according to the pathological staging by using TCGA database. (K) Schematic diagram of the negative correlation between BPGM levels and metastasis in both HCC and HNSCC patients. Error bar: mean ± SEM. One-way ANOVA test and unpaired t-test were used for statistical analysis. P-values are labeled above the bar chart.

Subsequently, to identify key metabolites with potential anti-metastatic functions, we screened for those that were consistently downregulated in highly metastatic HCC and HNSCC tumors. A comparative analysis revealed that a total of 32 metabolites were commonly and significantly downregulated in both HCC and HNSCC tissues compared to their adjacent non-cancerous counterparts (Fig. 1D and Table S3). Further volcano plot analysis revealed that 1/2,3‑bisphosphoglycerate (1/2,3‑BPG), an intermediate of the glycolytic bypass pathway known as the Rapoport-Luebering shunt (RLS), was downregulated in highly metastatic tumors of both types (Fig. 1E-F). This inverse correlation supports the hypothesis that 1/2,3‑BPG could function as a metabolic suppressor of tumor metastasis.

Given that bisphosphoglycerate mutase (BPGM) is the rate-limiting enzyme for the conversion of 1,3-BPG to 2,3-BPG[21](Fig. 1G), we investigated whether BPGM expression was also altered in tumors. Consistently, BPGM expression at both the mRNA and protein levels was reduced in HCC tumors exhibiting microvascular invasion (MVI) and in HNSCC tumors with lymph node metastasis (Fig. 1H-I). Further analysis of The Cancer Genome Atlas (TCGA) data confirmed that BPGM expression was progressively downregulated across tumor stages (Stage I to IV) in HNSCC (Fig. 1J). Together, these results suggest that BPGM and its catalyzed product, 2,3-BPG, were negatively correlated with tumor metastatic potential and may function as a metastasis suppressor (Fig. 1K).

BPGM is an intrinsic orchestrator of cancer cell migration and metastasis

To test this intriguing possibility, we first examined the expression of BPGM among a variety of tumor cell lines, including hepatoma cell lines (SK-HEP-1 and SNU-449), HNSCC cell line (FaDu) and CESC cell lines (Caski, C33A, Hela and S12) (Fig. S2A). The result showed that BPGM was broadly expressed across multiple tumor cell lines. Further gain- and loss- of function studies were conducted to assess the role of BPGM on tumor cells growth and migration in vitro. As shown, knockdown of BPGM significantly promoted the migration ability of tumor cells with high in vitro migratory capacity, such as SK-HEP-1, SNU-449 and FaDu (Fig. 2A and Fig. S2B), whereas overexpression of BPGM dramatically impaired the migration ability of these tumor cells (Fig. 2B and Fig. S2C). Surprisingly, the in vitro cell growth of tumor cells was not significantly affected when BPGM was silenced or overexpressed (Fig. S2D&E). Moreover, consistent with the anti-migration effect mediated by BPGM, the expression of pro-invasion genes including MMP2 and MMP9 were significantly downregulated in BPGM-overexpressing tumor cells (Fig. 2C). These evidences indicate that BPGM may inhibit tumor cells migration and invasion in vitro.

Fig. 2.

Overexpression of BPGM inhibits tumor metastasis in vitro and in vivo. (A) Silencing BPGM promoted migration of tumor cells. Tumor cells stably expressing shBPGM and its control cells (shCtrl) were examined. Scale bar, 200 μm. (B) Overexpressing BPGM suppressed migration of tumor cells. Tumor cells stably expressing BPGM and its control cells (Ctrl) were examined. Scale bar, 200 μm. (C) The mRNA levels of MMP2 and MMP9 reduced in BPGM-overexpressing tumor cells. SK-HEP-1 cells stably expressing BPGM (BPGM-OE) and its ctrl cells were employed to detect the mRNA levels of MMP2 and MMP9 by qPCR analysis. β-actin was used as an internal control. (D-E) Xenografts of stable Bpgm-overexpressing cells displayed a lower rate of liver and lung metastasis and less metastatic nodules in the liver. For (D), Hepa-Ctrl (n = 7) and Hepa-BPGM sublines (n = 6) were inoculated under the capsule of the left hepatic lobe of C57BL/6 mice. Upper panel, a schematic diagram of orthotopic hepatic implantation model. The number of metastatic rate and nodules is shown (D, lower panel). Scale bar, 1 cm. Hematoxylin-eosin staining was performed on serial sections of livers (E, left panel) and lungs (E, right panel) to detect the metastatic nodules. The red arrows indicated the metastatic nodules. Scale bar in left panel, 200 µm; Scale bar in right panel, 100 μm. (F-H) Overexpressing of BPGM inhibited lung metastasis of tumor xenografts. Scale bar in F, 1 cm. B16-F10 cells transfected with Ctrl (n = 6) or BPGM-OE (n = 6) was injected into the tail vein of C57BL/6 mice, respectively. Upper panel in F, a schematic diagram of lung metastasis model by tail vein injection. H&E staining of lung sections was performed to observe metastatic foci (G). For G, scale bar in left panel, 500 μm; scale bar in right panel, 200 μm. The number of melanoma nodules and lung metastasis is shown in H. (I) The model deciphers the inhibitory role of BPGM in tumor metastasis. Error bar: mean ± SEM. P-values are labeled above the bar chart.

Next, we prompted to investigate the anti-metastasis function of BPGM in vivo in multiple experimental mice model systems. First, the stable murine hepatoma Hepa1-6 cell line with the specific BPGM overexpression was successfully generated (Fig. S2F), followed by establishing a hepatoma orthotopic transplantation model. Consistently, compared with the control group, the xenografts derived from BPGM-overexpressing Hepa1-6 cells (BPGM-OE) displayed a lower rate of total metastasis (Ctrl vs. BPGM-OE: 8/8 vs. 4/7), liver metastasis (Ctrl vs. BPGM-OE: 6/8 vs. 2/7) and lung metastasis (Ctrl vs. BPGM-OE: 5/8 vs. 4/7) (Fig. 2D&E), while the tumor mass was slightly increased (Fig. S2G). Moreover, mice bearing BPGM-OE xenografts also showed fewer metastatic nodules within the liver (Ctrl vs. BPGM-OE: 82.25 ± 50.15 vs. 6.43 ± 4.23) and lung (Ctrl vs. BPGM-OE: 44.13 ± 24.96 vs. 3.57 ± 1.33) (Fig. 2D&E). These data provided the in vivo evidence that overexpression of BPGM may halt hepatoma cells metastasis without attenuating primary tumor growth.

To explore the broadly anti-tumor metastatic effect by BPGM overexpression, we employed another well-established melanoma-derived and metastatic pulmonary tumor model via tail vein injection. Similarly, overexpression of BPGM in melanoma B16-F10 cells (Fig. S2H) significantly attenuated its pulmonary metastasis, indicated by a notable decrease in the number of melanoma nodules and lung metastatic nodules, respectively (Fig. 2F-H). Altogether, our findings demonstrate that BPGM acts as a potent metastatic gatekeeper, specifically reprogramming tumor cells into a migration-deficient state, supporting its therapeutic potential as a molecular brake against cancer dissemination (Fig. 2I).

BPGM-mediated l-carnitine deprivation collapses the fuel supply chains for cellular migration

We then investigate the molecular mechanisms of BPGM-mediated tumor metastasis suppression. Given the vital role of BPGM in the RLS of glycolytic pathway [21], we initially conducted in vivo ¹³C₆-labelled glucose flux analyses to accurately compare and trace the dynamic changes of glucose metabolism between tumor tissues derived from BPGM-overexpressing (BPGM-OE) xenografts bearing mice and the control mice, respectively (Fig. 3A and Table S4). Supporting the glucose metabolic regulation mediated by BPGM, the proportion of ¹³C-labelled RLS products, such as 1/2,3-BPG and 2/3-PG, were increased in BPGM-OE tumor tissues, compared with that in the control group (Fig. 3B, as shown in the dashed box). Intriguingly, the proportion of ¹³C-labelled RLS upstream intermediates, including glucose, fructose-1,6-bisphosphatase (FBP) and glucose 6-phosphate (G6P), and the RLS downstream intermediates, such as pyruvate (PYR) and lactate (LAC), were both significantly reduced in BPGM-OE tumor tissues (Fig. 3B). Moreover, the proportion of ¹³C-labelled alanine (Ala) and p‑serine (p-Ser), which were biosynthesized de novo from glycolysis intermediates, were also diminished in BPGM-OE tumor tissues (Fig. 3B). Importantly, the proportion of ¹³C-labelled AMP and IMP were increased when BPGM was overexpressed (Fig. 3C). As shown, the ratio of LAC (M + 2)/LAC (M + 3), which reflects glucose metabolism through the pentose phosphate pathway (PPP) versus glycolysis was also increased in BPGM-OE tumor tissues (Fig. 3D). To our surprise, the ¹³C-labelled citrate, succinate, fumarate and malate were with no change between BPGM-OE and control groups (Fig. S3A). Altogether, these data indicate that BPGM is essential to rewire glucose flux toward RLS and PPP over glycolysis and TCA.

Fig. 3.

BPGM constrains cell migration by decreasing l-carnitine to paralyze β-oxidation. (A) Metabolic flux map from ¹³C₆-glucose tracing. (B-C) The levels of ¹³C₆-glucose derived metabolic intermediates in tumors. 5 % (w/v) of ¹³C₆ labeled-d-glucose was injected into mice bearing Hepa-BPGM-OE or Hepa-Ctrl tumors by tail vein for 0.5 hours. The tumor tissues were employed to conduct glucose metabolic flux analysis. The blue dash box indicated the metabolites of RLS. (D) The ratio of ¹³C₆-glucose derived two-carbon labelled lactate (M + 2) verse three-carbon labelled lactate (M + 3). (E) The untargeted metabolomics experimental protocol. (F) Volcano plot displayed differentially expressed metabolites in tumor tissue and plasma. (G) GO enrichment analysis showed significant enrichment of methionine synthesis, carnitine synthesis, and fatty acid oxidation pathways in BPGM-overexpressing tumor tissue and plasma. (H) Heatmap showed that l-carnitine and most of acyl-carnitines were downregulated in BPGM-OE group. Columns: individual samples; rows: metabolites. (I) l-carnitine was significantly reduced in BPGM-OE tumor tissues and plasmas. The concentration of l-carnitine was quantified by LC-MS/MS. (J-K) Replenishing l-carnitine attenuated the inhibitory effect of BPGM on the migration of tumor cells. Sublines with stably overexpression of BPGM were treated with 1 mM l-carnitine for 72 hours followed by transwell assays. The total number of migrated cells were stained by violet and counted under microscope. Scale bar, 200 μm. (L) The model deciphered l-carnitine mediated the inhibitory role of BPGM in cellular migration. Error bar: mean ± SEM. P-values are labeled above the bar chart.

To further define the comprehensive metabolic signature mediated by BPGM beyond glucose metabolism, untargeted metabolomics profiling was performed (Fig. 3E). Almost 400 metabolites were identified in tissues and plasma derived from BPGM-OE or Ctrl xenografts-bearing mice (Fig. S3B). Volcano plot analysis revealed that the differential metabolites between the control and BPGM-overexpression xenografts with fold change >1.2 (Fig. 3F), which were listed in Tables S5. GO enrichment analysis of differential metabolites demonstrated significant enrichment of methionine metabolism, carnitine synthesis, and β-oxidation of fatty acids in both tumor tissues and plasma (Fig. 3G). Since fatty acid oxidation has been documented in the literature as a critical energy source for tumor cells during metastasis, we further investigated the expression changes of key metabolites involved in this process, including l-carnitine and acyl-carnitines. The results revealed that l-carnitine and almost all of those acyl-carnitines were dramatically reduced in tumors tissues and plasma derived from BPGM-OE xenografts-bearing mice (Fig. 3H), compared to that of control mice. Importantly, targeted quantification of carnitine further confirmed the significant downregulation changes of l-carnitine in tumor tissues and plasma derived from BPGM-OE xenografts-bearing mice (Fig. 3I). Consistently, l-carnitine and almost all of those acyl-carnitines were dramatically increased when BPGM was stably knocked down in SK-HEP-1 cells (Fig. S4). Based on these data, we hypothesized that BPGM might strangle fatty acid oxidation and starve migration machinery in tumors through l-carnitine deprivation. To test this possibility, we conducted an in vitro assay to determine if l-carnitine supplements could directly enhance the migration of BPGM-OE tumor cells. Intriguingly, l-carnitine supplementation reversed the inhibitory effect of overexpression of BPGM on tumor cells migration (Fig. 3J&K). Thus, we revealed that BPGM executes a metabolic double strike diverting glucose flux through RLS-PPP while crippling fatty acid oxidation via l-carnitine restriction to dismantle the bioenergetic infrastructure required for tumor cell migration (Fig. 3L).

Impaired BPGM-BBOX1 axis depletes l-carnitine biosynthesis to paralyze β-oxidation in migrating cells

We next investigated how BPGM impacts l-carnitine levels in tumors. Intracellular l-carnitine level is regulated by its transporter OCTN2 and de novo biosynthesis [22,23]. To delineate whether downregulated l-carnitine levels-mediated by BPGM through its transporter and/or de novo biosynthesis, we first examined the expression level of l-carnitine transporter (OCTN2) following BPGM knockdown or overexpression. The results showed that knockdown of BPGM did not affect the mRNA level of OCTN2, while overexpression of BPGM significantly increased OCTN2 mRNA level in SK-HEP-1 cells (Fig. S4A), which upregulation was not consistent with the downregulation of l-carnitine in BPGM-OE tumor tissues. Moreover, given our findings that the level of l-carnitine was reduced in plasma from BPGM-OE xenograft-bearing mice (Fig. 3H, right panel and Fig. 3I, lower panel), we immediately hypothesized that BPGM might regulate l-carnitine de novo biosynthesis but not by OCTN2-mediated l-carnitine uptake.

It is known that lysine degradation and methionine metabolism are the core pathways controlling l-carnitine de novo biosynthesis [22,[24], [25], [26], [27]]. The methyl donor of l-carnitine originates from methionine-produced S-adenosylmethionine (SAM). During post-translational modification, lysine residues are trimethylated by methyltransferases to form N6-trimethyllysine (TML) residues, which is subsequently hydrolyzed into free TML via protein degradation. Then l-carnitine is synthesized from TML in four steps by the enzymes TML dioxygenase (TMLD or TMLHE), HTML aldolase (HTMLA), TMABA dehydrogenase (TMABADH) and γ-butyrobetaine hydroxylase (BBOX1) (Fig. 4A, upper panel). Notably, our previous GO enrichment analysis revealed that methionine metabolism was the top one pathway enriched in BPGM-OE tumor tissues (Fig. 3G, left panel). Intriguingly, further untargeted metabolomics analysis showed that the upstream of carnitine biosynthesis related metabolites including methionine, SAM, TML and γ-butyrobetaine were significantly increased when BPGM was overexpressed, while the final product l-carnitine was significantly diminished (Fig. 4A, lower panel). Conversely, silencing BPGM significantly increased intracellular l-carnitine levels, whereas methionine, SAM, TML, and γ-butyrobetaine were decreased in SK-HEP-1 cells (Fig. S5B). Thus, these findings raise a new but compelling hypothesis that the de novo biosynthesis of carnitine might be blocked at the last step converting γ-butyrobetaine to l-carnitine when BPGM is overexpressed.

Fig. 4.

BPGM limits cellular migration by inhibiting l-carnitine biosynthesis through reduced BBOX1 expression. (A) The change of l-carnitine biosynthesis-related metabolites in tumors. The de novo biosynthesis pathway of l-carnitine was indicated in the upper panel. Ctrl, n = 8; BPGM-OE tumor tissues, n = 5. (B-D) Overexpression of BPGM inhibited BBOX1 expression and silencing BPGM induced BBOX1 expression. Human hepatoma cell lines SK-HEP-1 stably expressing shBPGM/BPGM and its control cells (shCtrl/Ctrl) were used to detect BPGM level by qPCR analysis (B) and western blotting (C). The protein level of BBOX1 was quantified and summarized in D. (E) Overexpressing BBOX1 promoted the migration ability of SK-HEP-1 cells. Sublines with stably overexpression of BBOX1 and the ctrl cell lines were employed to transwell assays. (F) Inhibition of BBOX1 reduced the migration ability of tumor cells. SNU-449 cells were treated with 100 mM Meldonium or PBS (vehicle) for 48 hours and then were employed to transwell assays. (G-H) Meldonium treatment decreased the lung metastatic foci in mice. 1 × 10^6 B16-F10 melanoma cells were injected into C57BL/6 mice via tail vein. Three days after tumor inoculation, the mice were administered intraperitoneal injections at a dosage of 70 mg/kg meldonium once every two days and were euthanized 17 days later. The metastasis rate of kidney and lung was indicated. (I-J) The protein level of BBOX1 was upregulated in HCC tissues with MVI and HNSCC tissues with lymph node metastasis. “-” indicates patients without MVI (n = 5); “+” indicates patients with MVI (n = 5). The protein level of BBOX1 in tumors was detected by immunofluorescence. The fluorescence aera (right panel) was calculated. Scale bar, 50 μm. (K) The protein level of BBOX1 was negatively correlated with the protein level of BPGM in human HCC and HNSCC tissues. Each dot represents an individual sample. (L) The change of l-carnitine biosynthesis-related intermediate metabolites in HNSCC tumors. N0, n = 8; NX, n = 7. (M) The model deciphered BBOX1 mediated the inhibitory role of BPGM in cellular migration. Error bar: mean ± SEM. P-values are labeled above the bar chart.

To test above possibility, we conducted the gene expression profiling of key enzymes involved in l-carnitine biosynthesis including TML dioxygenase TMLHE, HTML aldolase SHMT1/2³¹, TMABA dehydrogenase ALDH9A1³² and butyrobetaine hydroxylase BBOX1. Among all of the genes we screened, silencing BPGM significantly induced both mRNA and protein levels of BBOX1 (Fig. 4B-D, left panel), the key rate-limiting enzyme in l-carnitine biosynthesis at the last step. Conversely, overexpressing BPGM dramatically reduced both mRNA and protein levels of BBOX1 in tumor cells (Fig. 4B-D, right panel). However, no significant effects on the mRNA levels of other enzymes involved in carnitine biosynthesis were observed with BPGM depletion or overexpression (Fig. S5C). These data demonstrate that BPGM is a negative regulator of BBOX1 gene expression.

Similar as carnitine supplements (Fig. 3J&K), overexpressing BBOX1 significantly enhanced tumor cells migration (Fig. 4E). Moreover, treatment of Meldonium, a clinically approved competitive inhibitor of BBOX1[[28], [29], [30]] significantly suppressed tumor cell migration in vitro (Fig. 4F), suggesting its potential as an anti-metastatic therapeutic agent. Therefore, we conducted preclinical studies in a well-established C57BL/6 mouse model of melanoma lung metastasis with Meldonium treatment (Fig. 4G, upper panel). Similarly, the Meldonium-treated group exhibited significant inhibition of melanoma lung and kidney metastasis, including a reduction in both the lung weight (Fig. 4G, right panel) and the number of metastatic lung foci (Fig. 4H). Thus, we concluded that BPGM inhibits tumor metastasis by suppressing BBOX1 expression to block l-carnitine biosynthesis and inhibiting BBOX1 by Meldonium is a potential safe and effective treatment to block tumor metastasis.

Lastly, to validate the significance regulatory relationship between BPGM and BBOX1 in humans, we compared their expression in tumor specimens isolated from patients by immunofluorescence staining. Intriguingly, the BBOX1 protein level was higher in HCC tumors with microvascular invasion and HNSCC tumors with lymph node metastasis (NX), than that in HCC tumors without microvascular invasion and HNSCC tumors without lymph node metastasis (N0), respectively (Fig. 4I-J). Moreover, an inverse correlation between BPGM and BBOX1 expression in HCC and HNSCC tumors was observed, respectively (Fig. 4K). Notably, untargeted metabolomics by using tumor tissues derived from HNSCC patients verified that the upstream of l-carnitine biosynthesis related metabolites including methionine, SAM, TML and γ-butyrobetaine were decreased in HNSCC tumors with lymph node metastasis, while l-carnitine itself was significantly upregulated (Fig. 4L). Collectively, we decoded BPGM as a critical migration brake by strangle holding on BBOX1-driven l-carnitine biosynthesis to halt tumor metastasis (Fig. 4M).

2,3-BPG-CDK1-EZH2-H3K27me3 Axis: BPGM’s epigenetic circuit breaker for invasion

We next to define how BPGM suppresses BBOX1 expression. Functional categorization of the differential metabolites revealed that BPGM-altered metabolites mainly clustered in methyl-donor pathways (e.g., SAM, serine and glycine etc. listed in Table S5) (Fig. 5A), we therefore hypothesized that methylation-dependent epigenetic modification is a potential molecular basis underlying BPGM-mediated downregulation of BBOX1 gene expression (Fig. 5B). Supporting this possibility, immunoblotting results shown that silencing BPGM specifically decreased the protein level of H3K27me3, while had no effect on other histone modifications including H3K79me3, H3K4me3 and H3K9me3 (Fig. 5C, upper panel; Fig. S6A, lane1 and lane2). Consistently, overexpressing BPGM increased H3K27me3 protein levels but did not affect the expression of H3K79me3, H3K4me3 and H3K9me3 (Fig. 5D, lower panel; Fig. S6A, lane3 and lane4). H3K27me3 is a repressive histone mark, which plays a critical role in epigenetic regulation by compacting chromatin structure and silencing gene expression [31,32]. Further chromatin immunoprecipitation (ChIP) assays disclosed that the promoter fragment of BBOX1 gene precipitated by anti-H3K27me3 antibody, were significantly increased in BPGM-overexpressed group compared to that in the control group (Fig. 5D, upper panel). In addition to the BBOX1 gene, the promoter fragment of MMP9, a gene that encode matrix metalloproteinases to degrade extracellular matrix, was also significantly increased in BPGM-overexpressed cells (Fig. 5D, lower panel). Thus, these data indicate that BPGM may repress BBOX1 and MMP9 transcription in a H3K27me3-dependent manner.

Fig. 5.

2,3-BPG-CDK1-EZH2-H3K27me3 Axis: BPGM’s epigenetic circuit breaker for cellular migration. (A) Integrated functional metabolomics analysis revealed BPGM-altered metabolites clustered in methyl donor group. Bubble size: metabolites count. (B) Hypothesis of molecular mechanism underlying BPGM regulated BBOX1 expression by post transcriptional modification (PTM). (C) Silencing BPGM significantly reduced the protein level of H3K27me3, while overexpressing BPGM increased its level. Cells stably expressing shBPGM/BPGM and its control cells (shCtrl/Ctrl) were used to detect protein level by western blotting. (D) ChIP assays disclosed that the fragments of BBOX1 and MMP9 promoter precipitated by anti-H3K27me3 antibody were increased upon overexpressing BPGM. SK-HEP-1 cells stably expressing BPGM and its control cells (Ctrl) were employed to ChIP assay. The antibody precipitated DNAs were amplified by qPCR. 5 % of the total DNAs were amplified to serve as the control for DNA content. Values shown are signal of α-H3K27me3-precipitated DNA relative to the input and the mean value of the control group was normalized as 1. (E) Overexpressing BPGM significantly increased the protein level of EZH2 but decreased the protein level of p-EZH2-T₃₄₅ in tumor cells. (F) The molecular docking of 2,3-BPG and CDK1. Predicted structure of 2,3-BPG binding with CDK1. Key contact residues: Thr14, Arg127, Arg170. (G) Overexpressing BPGM significantly increased the protein level of p-CDK1-T₁₄ in tumor cells. Cells stably expressing BPGM (BPGM-OE) and its control cells (Ctrl) were used to detect protein level by western blotting. (H) 2,3-BPG treatment enhanced the phosphorylation of CDK1 at thr14 in tumor cells. The indicated concentration of 2,3-BPG was incubated with the lysate of trophoblasts and tumor cells for 30 minutes followed by western blotting. (I-J) RO-3306 treatment enhanced the phosphorylation of CDK1 at thr14 and reduced the phosphorylation of EZH2 at thr345 in tumor cells. The tumor cells were treated with the indicated concentration of RO-3306 for 12 hours followed by western blotting. (K) The model deciphers the role of BPGM in regulating BBOX1 and MMP9 expression. Error bar: mean ± SEM. P-values are labeled above the bar chart.

H3K27me3 is primarily catalyzed by the Polycomb Repressive Complex 2 (PRC2), with EZH2 (Enhancer of Zeste Homolog 2) serving as its catalytic subunit, we therefore examine whether BPGM affects EZH2 expression. Western blot analysis demonstrated that BPGM overexpression significantly increased the protein level of EZH2 in tumor cells (Fig. 5E, left panel). Given that EZH2 stability is regulated by CDK1-mediated phosphorylation of EZH2 at site T₃₄₅, which promotes its ubiquitin-dependent degradation [33,34], we then assessed this regulatory axis in BPGM-overexpressed cells. Notably, BPGM overexpression concurrently suppressed EZH2 phosphorylation at T₃₄₅ (Fig. 5E, right panel). Consistently, ubiquitination-specific Co-Immunoprecipitation (Co-IP) assays revealed reduced EZH2 ubiquitination upon BPGM overexpression (Fig. S6B). These results suggest that BPGM stabilizes EZH2 by inhibiting its phosphorylation at T₃₄₅.

Given that 2,3-BPG produced by BPGM acts as a phosphate donor to phosphorylate and activate phosphoglycerate mutase 1 (PGAM1) [35], we hypothesized that BPGM might regulate EZH2 expression via 2,3-BPG-mediated phosphorylation signaling. Supporting this possibility, total phospho‑serine/threonine protein levels were significantly altered upon BPGM overexpression in Hepa1-6 cells (Fig. S6C). To our surprise, molecular docking indicated that 2,3-BPG showed a weak binding affinity to EZH2 but a strong binding potential to CDK1, which is the kinase responsible for EZH2 phosphorylation at T₃₄₅ (Fig. 5F and Fig. S6D). Further structure analysis showed that 2,3-BPG may interact with CDK1 at site 14 (Fig. 5F, lower panel), a residue whose phosphorylation is known to inhibit CDK1 activity [36,37], suggesting that BPGM may indirectly regulate EZH2-T₃₄₅ phosphorylation by modulating CDK1 phosphorylation.

Consistent with this prediction, western blot analysis revealed that BPGM overexpression significantly increased CDK1-T₁₄ phosphorylation without affecting total CDK1 levels (Fig. 5G). To further validate the direct effect of 2,3-BPG on CDK1, we incubated 2,3-BPG directly with cell lysates in vitro and observed a marked increase in CDK1-T₁₄ phosphorylation (Fig. 5H). Additionally, treatment with the CDK1 inhibitor RO3306, which mimicked 2,3-BPG binding to CDK1-T₁₄, led to a concentration-dependent elevation in p-CDK1-T₁₄ levels, while p-EZH2-T₃₄₅ levels progressively decreased with increasing RO3306 concentration (Fig. 5I&J).

Taken together, these results demonstrated that BPGM produces 2,3-BPG, which promotes its autoinhibitory phosphorylation of CDK1 at T₁₄, thereby suppressing EZH2-T₃₄₅ phosphorylation, stabilizing EZH2 protein levels, resulting in increased H3K27me3 binding to the promoters of BBOX1 and MMP9 genes, and ultimately repressing their transcription (Fig. 5K).

Hypoxia inhibits BPGM transcription via disruption of KDM4A/H3K9me3 axis

Finally, given the importance of induced BPGM in anti-tumor metastasis, we motivated to probe the molecular bases underlying BPGM downregulation in tumor development. Hypoxia is a hallmark in tumor niches, play an essential role for cell migration and invasion [[38], [39], [40]]. As shown in Fig. 6A, the more hypoxic tumor microenvironment is known linked to more advanced and metastatic tumor [[41], [42], [43], [44]]. Thus, based on these facts and in view of our findings that BPGM expression is decreased in high-metastasis potential tumor, we immediately hypothesized that hypoxia may be a negative factor involved in regulating BPGM gene expression. Supporting this possibility, exposure to hypoxia directly and significantly decreased both mRNA and protein level of BPGM in SK-HEP-1 and SNU-449 cells (Fig. 6B-D), suggesting BPGM is a hypoxia-repressed gene.

Fig. 6.

Hypoxia inhibits BPGM expression in a KDM4A/H3K9me3-dependent manner. (A) The oxygen concentration and its positive correlation with BPGM level in different stage of tumor development. (B) Hypoxia treatment decreased the mRNA level of BPGM. (C-D) Hypoxia treatment decreased the protein level of BPGM. For B-C, SK-HEP-1 and SNU-449 cells were incubated under normoxia or hypoxia for the indicated time. The BPGM level was detected by qPCR (B) or western blotting (C-D). (E) Hypoxia treatment decreased the luciferase activities of p(−2/+0.1k) promoter. Cells were co-transfected with BPGM promoter plasmid (p(−2/+0.1k)) and pRL-CML vectors for 36 hours, then untreated or treated with hypoxia for another 12 hours. The promoter activity was examined by luciferase activity assay. Basic, pGL3-basic vector. (F) Knockdown of hypoxia-inducible-factor 1α did not block the inhibitory effect of hypoxia on the luciferase activities of p(−2/+0.1k) promoter. (G-H) The histone modification features and binding sites of histone demethylase in the promoter region of BPGM gene. ChIP-seq profiles of H3K4me1, H3K4me3, H3K27ac and H3K9me3 are visualized using the UCSC genome browser (http://genome.ucsc.edu/). The transcription direction and the potential promoter region of BPGM are indicated. TSS, transcription start site. (I) The level of H3K9me3 was downregulated in HCC tissues with MVI. The protein levels of H3K9me3 in human HCC tissues (“-”, n = 5; “+”, n = 5) were detected by immunofluorescence. (J) The protein level of BPGM was negatively correlated with the level of H3K9me3 in human placentae and HCC tissues. (K) Hypoxia increased H3K9me3 level but decreased BPGM level in tumor cells. SK-HEP-1 cells were incubated under normoxia or hypoxia for the indicated time. (L) H3K9me3 inhibitor blocked the inhibitory role of hypoxia on BPGM expression. SK-HEP-1 cells were incubated without or with the indicated concentration of chaetocin (Chao) under normoxia or hypoxia condition for 12 h. Scale bar, 50 μm. (M) KDM4A inhibitor suppressed BPGM expression under normoxia condition. SK-HEP-1 cells were incubated without or with the indicated dose of ML324 (KDM4A inhibitor) under normoxia condition for 12 h. (N) The model deciphers the role of hypoxia in repressing BPGM expression. Error bar: mean ± SEM. P-values are labeled above the bar chart.

To further explore how hypoxia repressed BPGM gene expression, the promoter reporter plasmids p(−2/+0.1k), containing −2000 to +140-bp genomic region spanning the transcription start site (TSS) of BPGM, were constructed (Fig. S7A). As shown, under normoxia condition, the luciferase activity of p(−2/+0.1k) luciferase reporter constructs were significantly higher than that of pGL3-basic vector, whereas its activity was blocked upon hypoxia in tumor cells (Fig. 6E). Interestingly, the inhibitory effect of hypoxia on the luciferase activity of p(−2/+0.1k) luciferase reporter was not reversed by silencing hypoxia induced factor 1α (HIF1α) (Fig. 6F) or mutation of six hypoxia response elements (HREs) predicted within the 2-kb genomic region upstream of the TSS of BPGM (Fig. S7A&B). These findings indicate that hypoxia may suppress BPGM transcription in a HIF1α-independent manner.

Next, to identify HIF1α-independent mechanisms underlying hypoxia-mediated inhibition of BPGM transcription, chromatin immunoprecipitation (ChIP)-sequencing data from ENCODE was analyzed [45,46]. Specifically, we found that histone modifications associated with an active promoter, including H3K4me1, H3K4me3 and H3K27ac, were enriched within the 2-kb genomic region upstream of the TSS of BPGM, and 5 histone demethylase (HDMs) including KDM1A, KDM4A/B and KDM5A/B, were predicted binding to the promoter of BPGM (Fig. 6G). Because oxygen molecule (O₂) is an essential substrate for HDMs, their activity is suppressed under hypoxia condition, which increases histone methylation modifications and subsequently represses or induces gene expression [47,48]. Among these enzymes, KDM4A and KDM4B primarily mediate H3K9me3 modification [49]. Downregulation of their activity may cause H3K9me3 accumulation, thereby silencing gene expression [50]. Notably, we found that the H3K9me3 signal located in BPGM gene region existed in human hepatoma cell line HepG2 and normal liver cells, and the H3K9me3 signal in HepG2 was stonger than that in normal liver cell (Fig. 6H). Notably, immunofluorescent staining revealed that the H3K9me3 level was significantly lower in HCC tissues without MVI than that in HCC tissues with MVI (Fig. 6I). And H3K9me3 expression showed a significantly inverse correlation with BPGM expression in both placental and tumor tissues (Fig. 6J). Further hypoxia treatment revealed that hypoxia upregulated H3K9me3 levels in a time-dependent manner, while simultaneously suppressed BPGM expression in tumor cells (Fig. 6K), and this inhibitory effect of hypoxia was reversed by chaetocin co-treatment, which is a H3K9me3 specific inhibitor (Fig. 6L). Lastly, ML324 (a KDM4A inhibitor) treatment inhibited the expression of BPGM under normoxia condition (Fig. 6M). Altogether, these data establish hypoxia as an O₂-sensitive epigenetic switch that silencing BPGM transcription by inactivating KDM4A and thereby triggering a repressive H3K9me3 signature at the BPGM promoter, highlighing a previously unappreciated O₂–histone demethylase–metabolic gene axis that reprograms BPGM gene expression in a HIF1α-independent manner (Fig. 6N).

Discussion

This study establishes BPGM as a previously unrecognized metabolic and epigenetic gatekeeper that intrinsically constrains tumor metastasis through an integrated regulatory circuit. While prior research has predominantly focused on pro-metastatic metabolic adaptations—including bioenergetic reprogramming, redox maintenance, oncometabolite signaling, and microenvironmental remodeling—the potential for intrinsic metabolic pathways to actively suppress dissemination has remained underexplored. Our findings delineate a mechanism wherein BPGM, through its product 2,3-BPG, stabilizes EZH2 and elevates S-adenosylmethionine (SAM) levels to enforce H3K27me3-mediated transcriptional repression of a metastasis-promoting effector: BBOX1, essential for carnitine biosynthesis and subsequent fatty acid β-oxidation. This restrictive circuit is dynamically inactivated under hypoxic conditions via an O₂-KDM4A-H3K9me3 axis, illustrating how a prevalent tumor microenvironmental stressor overcomes metabolic suppression to license invasive progression (Fig. 7). Collectively, these results expand the functional paradigm of glycolytic enzymes beyond energy metabolism, positioning BPGM within a metabolic-epigenetic network that actively constrains malignant dissemination, and revealing a targetable axis for therapeutic intervention in metastatic disease.

Fig. 7.

Working model of BPGM-mediated metabolic-epigenetic regulation axis and its gatekeeper role in tumor metastasis. In low-metastatic tumors, higher oxygen levels activate KDM4A, which removes repressive H3K9me3 marks at the BPGM promoter, thereby promoting BPGM transcription. Elevated BPGM expression increases the production of 2,3-BPG, which stabilizes EZH2 and enhances SAM-dependent H3K27me3 deposition. This epigenetic remodeling leads to transcriptional silencing of BBOX1, a key gene involved in carnitine biosynthesis, consequently suppressing fatty acid oxidation and inhibiting tumor metastasis. In contrast, under hypoxic conditions commonly found in high-metastatic tumors, KDM4A activity is diminished, resulting in the accumulation of H3K9me3 at the BPGM promoter and subsequent downregulation of BPGM expression. This disruption of the BPGM-mediated regulatory axis abrogates its anti-metastatic function. Notably, preclinical studies revealed that pharmacological inhibition of BBOX1 with Meldonium restores the metabolic-epigenetic barrier, effectively impeding metastatic progression.

Although traditionally viewed as the erythroid “2,3-BPG factory” that fine-tunes hemoglobin–O₂ affinity, BPGM has been reported to be expressed in tumor cells [35,51]. Notably, early studies based on TCGA database hinted that BPGM expression may be related to the occurrence of cervical cancer [51]. However, to date, the spatiotemporal expression and function of BPGM in tumorigeneses remain ambiguous. Our study rewrites the textbook view of BPGM from a simple erythroid 2,3-BPG generator to a tumor invasion rheostat. By generating the first high-resolution spatiotemporal map of human BPGM, we reveal a precipitous drop in high-metastatic and advanced-stage tumors. Gain-/loss-of-function demonstrates that BPGM is sufficient to terminate cellular migration and arrest tumor dissemination in multiple experimental models. Crucially, this brake operates independently of maternal erythrocyte function, positioning BPGM as a O₂-sensitive metabolic checkpoint. Thus, we solve the puzzle of why BPGM is reduced in late-stage tumor and establish it as a dual-purpose diagnostic biomarker and therapeutic vulnerability for metastatic cancer.

While BPGM is classically recognized for regulating glycolytic flux and 2,3-BPG production in erythrocytes, its regulatory role in metabolism has garnered attention. For example, early studies report that inactivating mutations of BPGM in red blood cells inhibit intra-erythrocytic parasite replication by reducing ATP levels [52], while overexpressing BPGM inhibits glycolysis in astrocytes and distal nephron [53,54]. Knocking out BPGM in tumor cells promotes the serine synthesis by decreasing p-PGAM1 level without affecting glycolytic flux, while the proliferation of tumor cells remains unaffected [35]. However, majority of the work is limited to the cellular level and lacks comprehensive molecular and metabolic insights into BPGM-switching off the invasion during malignant tumor metastasis within intact animals. Here, employing state-of-art multi-omics including untargeted metabolomics, isotopically labelled glucose flux tracing analyses and comparative and integrative analyses in metastatic tumor experimental models, we define BPGM unprecedented role as a metabolic orchestrator in cancer cells, integrating glycolysis, one-carbon metabolism, carnitine biosynthesis and epigenetic regulation through a unified axis. Using in vivo models coupled with high-throughput metabolomics, we demonstrate that BPGM operates as a metabolic switch that BPGM is essential to rewire glucose flux toward RLS and PPP over glycolysis and TCA, and reciprocally regulates serine/glycine/methionine flux and carnitine biosynthesis, a mechanism distinct from its erythrocytic function. Specifically, BPGM knockout elevates serine synthesis, which is consistent with prior tumor cell studies [35] and enhances PPP, while its overexpression depletes carnitine and acyl-carnitines but enriches precursors including methionine, SAM, TML, firstly linking BPGM to fatty acid oxidation control. These findings may partially explain why BPGM overexpression suppresses tumor metastasis without affecting primary tumor growth, highlighting the metabolic adaptability and plasticity of tumors during their progression, as well as the importance of precision targeted therapy.

Fatty acids serve as a crucial fuel source preferentially utilized by metastatic cells. Fatty acids can be oxidized to generate acetyl-CoA. Major fatty acid oxidizing organelles are the mitochondria into which many long-chain fatty acids, such as palmitate, are transported depending on carnitine palmitoyltransferase 1 (CPT1) [55]. It is reported CPT1A-mediated fatty acid oxidation promoted colorectal cancer cell metastasis by inhibiting anoikis [10]. However, little is known about the alterations in carnitine, a critical carrier responsible for transporting fatty acids into mitochondria, during tumor metastasis. Our findings reveal for the first time that highly metastatic tumor cells possess the capability to synthesize carnitine autonomously. Overexpression of BPGM inhibits tumor metastasis by suppressing carnitine biosynthesis key enzyme BBOX1 expression and subsequently blocking carnitine biosynthesis. Importantly, Ye. et al. reports that BBOX1 is a metabolic checkpoint that mediates metastatic cancer cell evasion from immunosurveillance by NK Cells recently [56]. In line with this finding, we provided the both in vitro and in vivo preclinical evidence that pharmacological inhibition of BBOX1 by Meldonium (a clinical approved drug to treat myocardial ischemia (stable angina), heart failure, and cerebral circulatory disorders (e.g., post-stroke recovery)) [[28], [29], [30]] exerted an anti-tumor cell migration effect as well as a potent therapeutic effect against lung metastasis of melanoma cells, respectively. Thus, we defined that BPGM-mediated BBOX1 downregulation is a critical regulatory machinery for downregulation of carnitine biosynthesis and that BBOX1 inhibitor is a potential effective and safe treatment counteracting tumor metastasis

Metabolite-driven epigenetic reprogramming constitutes a pivotal mechanism underlying the regulation of tumor metastasis. Crucially, we identify a competitive substrate relay wherein BPGM redirects SAM from carnitine synthesis to fuel H3K27me3 deposition, mediated by a novel 2,3-BPG/CDK1-T₁₄/EZH2 signaling cascade: 2,3-BPG phosphorylates CDK1 at Thr14 residue and inhibits its activity, attenuating CDK1-mediated EZH2 phosphorylation and degradation, amplifying H3K27me3-mediated silencing of BBOX1 and MMP9 to halt carnitine synthesis and ECM remodeling, respectively. This dual metabolic-epigenetic gatekeeping explains BPGM’s anti-invasive function, as carnitine supplementation rescues migration defects in BPGM-overexpressing cells. While CDK1 is typically associated with cell cycle promotion in cancers [57,58], and EZH2 is frequently reported as overexpressed in metastatic tumors [31,59], we observed that their functional output is context-dependent. In BPGM-OE cells, CDK1-T₁₄ phosphorylation induced by 2,3-BPG promotes EZH2 stability with only a slight cell proliferation activation (Fig. S2E) but a dramatic migration decrease, suggesting CDK1-T₁₄’ specifical role in cellular migration. This metabolic-epigenetic coupling indicates that EZH2 inhibitors may show limited efficacy in BPGM-low tumors, and suggests CDK1-T₁₄ phosphorylation as a biomarker for BPGM-functional cancers. Overall, our work transcends prior cellular studies by revealing how BPGM spatially partitions SAM utilization to balance histone methylation and metabolic homeostasis, offering a new concept for how glycolytic enzymes coordinate cell fate through metabolite-directed chromatin remodeling. These findings position BPGM as a therapeutic node for metastatic cancer by targeting its downstream metabolic-epigenetic network.

The metabolic plasticity of disseminating tumor cells is intrinsically linked to the dynamic interplay within the tumor microenvironment. Hypoxia is traditionally viewed as the master driver of invasion in tumor [[38], [39], [40]]. Hypoxia intensifies with disease progression [[41], [42], [43], [44]], yet BPGM expression steadily falls. More recent studies revealed that hypoxia up-regulates placental BPGM in mice as an adaptive response, whereas human placentae from idiopathic fetal-growth-restriction (FGR) exhibit selective BPGM loss [60], hinting hypoxia might be a key regulator in BPGM expression. To date, the mechanism of how hypoxia regulates BPGM expression in tumors remain unclear. Our findings elucidate a novel mechanism underlying hypoxia-mediated repression of BPGM in a HIF1α-independent manner. We presented multi-faceted evidence demonstrating that hypoxia, a hallmark of tumor metastasis, induced time-dependent repression of BPGM transcription and loss of promoter activity. Remarkably, this repression can be effectively rescued by blocking H3K9me3 or inhibiting KDM4A, highlighting the critical role of the KDM4A-H3K9me3 axis in this process. Importantly, this mechanism operates independently of the canonical HIF-1α pathway, revealing a previously unrecognized layer of complexity in hypoxia-induced gene regulation. Severe hypoxia-induced KDM4A/H3K9me3-mediated BPGM shutdown is pathologically hijacked to fuel malignant metastasis, positioning this axis as an immediately actionable therapeutic target for aggressive cancers.

In conclusion, our work establishes a fundamental hypoxia-metabolite-epigenetic axis centered on BPGM that differentially controls physiological and pathological invasion through a novel three-tier regulatory paradigm. At the O₂-sensing tier, hypoxia induces KDM4A-mediated H3K9me3 deposition to epigenetically silence BPGM, creating an invasion-permissive state in metastatic cancers. At the metabolite signaling tier, BPGM-generated 2,3-BPG acts as a dual-function molecular switch: (1) directly binding CDK1 to inhibit Thr14 phosphorylation, which stabilizes EZH2, and (2) redirecting glycolytic flux to elevate SAM, synergistically amplifying H3K27me3 deposition. At the epigenetic execution tier, this results in simultaneous silencing of BBOX1 (disrupting carnitine-dependent β-oxidation) and MMP9 (blocking ECM remodeling), establishing an irreversible invasion barrier. Strikingly, this axis exhibits bidirectional therapeutic control - while hypoxia-induced BPGM suppression drives metastasis, BPGM overexpression or BBOX1 inhibition in hypoxic tumors restores invasion control, and carnitine supplementation bypasses the epigenetic blockade. Our work establishes BPGM as the first example of a glycolytic enzyme that: (1) converts O₂ tension into chromatin-based invasion memory through metabolite-guided phospho-signaling, and (2) creates a therapeutic vulnerability window for combined metabolic-epigenetic therapy in invasive disorders. These findings redefine our understanding of how metabolic enzymes orchestrate cellular plasticity across development and disease. This discovery highlights BPGM serve as a potential prognostic biomarker for metastatic cancers and opens new avenues for developing targeted interventions that can modulate BPGM activity, offering hope for improved outcomes in oncology.

Funding statement

This work was supported by grants from the National Natural Science Foundation of China (NSFC 82230023 to Y.X.; NSFC 32100573 to M.W.; NSFC 82401827 to W.L.; NSFC 82400873 to C.C.), Feifan Scholar Fund of Xiangya Hospital Central South University to Y.X., Hunan Provincial Natural Science Foundation of China (2023JJ40920 to M.W.; 2023JJ40919 to W.L.; 2025JJ60551 to C.C.), the China Postdoctoral Science Foundation (2023M743968 to W.L; GZB20240866 to C.C.) and MOE Key Laboratory of Gene Function and Regulation.

Data availability

All data supporting the findings of this study are available within the article and its supplementary information files.

CRediT authorship contribution statement

Meng-Zhi Wu: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Software, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Dou Feng: Visualization, Validation, Software, Project administration, Methodology, Investigation, Formal analysis, Data curation. Wu-Ping Liu: Visualization, Software, Methodology, Data curation. Wei-Lun Huang: Resources, Data curation. Qiang Wu: Resources, Data curation. Tian-Sheng Chou: Software, Methodology. Wen-Hao Xiao: Software, Methodology. Zhou-Zhou Yao: Data curation. Zhen-Jiang Li: Data curation. Ting-Ting Xie: Resources, Methodology. Chang-Han Chen: Resources, Methodology. Zhi-Yu Yang: Methodology. Rui-Wen Mao: Resources. Ci-Chun Wu: Resources. Jun-Cheng Wang: Resources, Methodology. Yu-Jin Zhang: Methodology. Rodney E Kellems: Writing – review & editing, Writing – original draft. Yang Xia: Writing – review & editing, Writing – original draft, Supervision, Project administration, Funding acquisition, Conceptualization.

Declaration of competing interest

The manuscript entitled “BPGM as an intrinsic brake to constrain metastasis through phospho-epigenetic-mediated carnitine biosynthesis suppression” have not been previously reported and are not under consideration for publication elsewhere. All the authors agree to the content of the paper and their being listed as an author in the paper. The authors declare that they have no conflict of interest. If accepted, the article will not be published elsewhere in the same form, in English or in any other language, including electronically, without the written consent of the copyright-holder.