Mitochondrial Uncoupling Induces Epigenome Remodeling and Promotes Differentiation in Neuroblastoma

One major metabolic cancer hallmark is the diversion of glucose flux from mitochondrial oxidation to fermentation and lactate production, known as the Warburg effect (1, 2). Warburg reported that the consequence of this metabolic reprogramming is the conversion of normal differentiated cells to dedifferentiated (cancer) cells (3, 4). Growing evidence has pointed out that metabolic reprogramming causes cell dedifferentiation through epigenetic remodeling (4–6).

The cancer epigenome generally displays global hypomethylation and CpG island hypermethylation compared with the epigenomes of corresponding healthy tissues (7). Particularly, CpG island hypermethylation in promoter regions leads to the silencing of tumor suppressors and cell differentiation markers and is associated with therapeutic resistance and poor prognosis in patients with cancer (8, 9). Unlike genetic mutations, epigenetic changes are reversible. Therefore, CpG island hypermethylation has become an attractive target for cancer therapy.

Recent reports have shown that mitochondria play a key role in epigenetic regulation and cell fate decisions by regulating the availability of certain metabolites (4–6). Ten-eleven translocation (TET) DNA demethylase uses α-ketoglutarate (α-KG) as a substrate to perform sequential oxidation of 5-methylcytosine, eventually removing the methylation (10). α-KG supplementation promotes pancreatic and colorectal cancer cell differentiation by reducing DNA methylation (11, 12). However, in tumors, α-KG can be converted to the TET inhibitor 2-hydroxyglutarate (2-HG). While D-enantiomer (D-2-HG) is produced through gain-of-function point mutations in isocitrate dehydrogenases 1/2 (IDH1/2; refs. 13, 14), in hypoxic cancer cells such as neuroblastoma cells, lower NAD⁺/NADH ratios favor the conversion of α-KG to L-enantiomer (L-2-HG; refs. 15, 16). Notably, lower NAD⁺/NADH ratios also favor the conversion of pyruvate to lactate, promoting the Warburg effect. Unlike D-2-HG, L-2-HG is produced via promiscuous substrate usage by lactate dehydrogenase (LDH), with additional contributions from malate dehydrogenase. While there are mutant IDH inhibitors to reduce D-2-HG levels, no effective therapeutic strategy inhibiting L-2-HG production is available. We hypothesized that CpG island hypermethylation could be reversed by increasing the intracellular NAD⁺/NADH and α-KG/2-HG ratios, correcting aberrant metabolic and epigenetic cancer cell phenotypes.

The electron transport chain (ETC) is the primary site for regenerating NAD⁺ from NADH in cells. Mitochondrial uncouplers facilitate proton influx across the mitochondrial inner membrane without generating ATP, leading to ETC activation. We hypothesized that ETC activation by mitochondrial uncouplers would increase the NAD⁺/NADH ratio, leading to an increased α-KG/2-HG ratio. In this study, we used a mitochondrial uncoupler, niclosamide ethanolamine (NEN), to reprogram the metabolism and epigenetic landscape of cancer cells. We show that NEN treatment increased intracellular NAD⁺/NADH, pyruvate/lactate, and α-KG/2-HG ratios. This metabolic reprogramming reduced DNA methylation in promoter regions and activated a neuronal differentiation program. Intriguingly, the differential methylation in gene bodies was highly associated with neuron differentiation-related genes. N-Myc, an oncogenic transcription factor often amplified in neuroblastoma and associated with poorly differentiated histologic features and worse patient survival, was downregulated by NEN treatment. Importantly, NEN treatment reduced 2-HG production, hypoxia-inducible factor (HIF) α subunit levels, and hypoxia-induced HIF target gene expression. Consistent with the in vitro data, NEN treatment reduced 2-HG levels and slowed tumor growth in an orthotopic neuroblastoma mouse model. Altogether, these results uncover a new role of mitochondrial uncoupling as an intervention that reverses the Warburg effect and promotes neuroblastoma differentiation.

MYCN-amplified neuroblastoma cell lines SK-N-BE(2) (RRID:CVCL_0528), NB16 (RRID: CVCL_B323), and CHP134 (RRID:CVCL_1124) were obtained from Dr. John M. Maris's laboratory (Children's Hospital of Philadelphia) in 2016. A certificate of analysis is available from Dr. Maris's group. Ovarian cancer cell line OVCAR3 was obtained from Dr. Erinn Rankin's laboratory (Stanford University) in 2020. Lung cancer cell lines H29 and H82 were from Dr. Julien Sage's laboratory (Stanford University) in 2020. Every 2 months, all cell lines were tested and found negative for Mycoplasma (MycoAlert Mycoplasma Detection Kit; Lonza). Last testing was performed on September 2022. Cell lines within 5 to 15 passages were used for experiments. The authors did not authenticate these cell lines. All cell lines were cultured in DMEM/F12 medium (Caisson Labs; DFL15) supplemented with 1% penicillin–streptomycin (Gibco; 15140122), 10% FBS (Sigma-Aldrich), and extra 1 mmol/L glutamine (Gibco; 25030081). Dimethyl 2-oxoglutarate (DMKG; #349631) and 5-aza-2′-deoxycytidine (5-AZA; #A3656) were obtained from Sigma-Aldrich. NEN (#17118) was obtained from Cayman. All cell proliferation assays were performed with the indicated treatments and times in triplicate. Anti–N-Myc antibody (#sc-53993, RRID:AB_831602) and anti-p53 antibody (#sc-126, RRID:AB_628082) were obtained from Santa Cruz Biotechnology. Anti–β-tubulin III antibody (#5568S) and anti-histone H3 antibody (#3638S) were obtained from Cell Signaling Technology. Horseradish peroxidase secondary antibody and goat anti-rabbit immunoglobulin G (H+L) cross-adsorbed secondary antibody Alexa Fluor 594 (#A-11012) were obtained from Invitrogen.

Cells were washed with ice-cold PBS buffer (Invitrogen; 20012050) and lysed with harvest buffer [10 mmol/L 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES; pH 7.9), 50 mmol/L sodium chloride (NaCl), 500 mmol/L sucrose, 0.1 mmol/L ethylenediaminetetraacetic acid (EDTA), and 0.5% Triton X-100] supplemented with 1% Halt inhibitors (Thermo Fisher Scientific; 78443) for 10 minutes. The cell lysate was centrifuged at 3,000 rpm for 3 minutes at 4°C, and the supernatant containing cytosolic proteins was transferred to a new tube. The pellet containing nuclear proteins was dissolved with nuclear lysis buffer [10 mmol/L HEPES (pH 7.9), 0.5 mol/L NaCl, 0.1 mmol/L EDTA, 0.1 mmol/L egtazic acid, and 0.1% NP40] supplemented with 1% Halt inhibitors (Thermo Fisher Scientific; 78443) for 5 minutes at 4°C. Then, the solution was sonicated at 4°C for six cycles [1 cycle = 30 s sonication (high intensity) and 30 s cooldown) using a Bioruptor Plus sonication (Diagenode; UCD-300). After sonication, the solution was centrifuged at 15,000 rpm for 10 minutes, and the supernatant was retained for nuclear extraction. Protein concentration was determined with the BCA assay (Thermo Fisher Scientific, 23227). Immunoblotting was performed as previously described (4). All the blots are representative of at least two independent experiments.

The procedure was performed as previously described (17). Briefly, total RNA was isolated from 60 mm tissue culture plates according to the TRIzol Reagent (Invitrogen; 15596026) protocol. Next, 3 μg of total RNA was used in a reverse transcription reaction with the iScript cDNA Synthesis Kit (Bio-Rad). Real-time quantitative polymerase chain reaction (RT-qPCR) amplification was performed using the Prism 7900 Sequence Detection System (Applied Biosystems) and TaqMan Gene Expression Assays (Applied Biosystems). Gene expression levels were normalized to the 18S ribosomal RNA. Data are presented as mean ± SD across three RT-qPCR reactions, and representative values from three independent experiments are shown.

The total RNA from treatment groups (control and NEN-treated; n =  3) was extracted using Trizol Reagent (Invitrogen; 15596026) according to the manufacturer's instructions. The library was constructed and subjected to 150 bp paired-end sequencing on an Illumina sequencing platform at Novogene. RNA sequencing (RNA-seq) analysis was performed using the kallisto and sleuth analytical pipelines. Briefly, a transcript index was generated on the basis of Ensembl version 67 of build hg19 of the human reference genome sequence. Paired-end RNA-seq reads were pseudo-aligned using kallisto (v0.42.4) using this transcript index with 100 bootstraps (-b 100) to estimate the variance of estimated transcript abundances. Transcript-level estimates were aggregated to transcripts per million (TPM) estimates for each gene, with Ensembl gene names assigned using biomaRt. Differential gene expression analysis was performed using the sleuth R package across pairwise groups using Wald tests, with significant hits identified on the basis of a sleuth q-value < 0.05 and absolute log₂ fold change > 0.693.

Gene set enrichment analysis (GSEA) was used to identify significantly enriched pathways by comparing the normalized data of the entire RNA-seq TPM dataset between groups against the molecular signatures database (MSigDB v7.4). All annotated transcripts (∼21,427 total features) with expression values were uploaded to a locally installed GSEA instance (v.4.1.0) and compared against the 50 Hallmark gene sets. Favorable or unfavorable prognosis gene sets were generated using a Kaplan–Meier assay (overall survival) for all genes independently and all 10 available neuroblastoma patient databases from R2 (18). Each gene that passed the Kaplan–Meier assay filter showed a significant correlation with prognosis (P < 0.05) in the relevant databases. Differentially expressed genes were functionally annotated by submitting them to the online Database for Annotation, Visualization and Integrated Discovery (DAVID) website (19). The DAVID analyses were performed using default parameters.

The SK-N-BE(2) cells were treated with DMSO or 1 μmol/L NEN for 24 hours under normoxia and hypoxia (0.5%) before DNA extraction (n = 3). DNA extraction was performed using the PureLink Genomic DNA Mini Kit (Gibco; K182001) according to the manufacturer's protocol. DNA was quantified with a Nanodrop, and 3 μg of total DNA was used for bisulfite modification using the EZ DNA Methylation Kit (Zymo Research; D5001) according to the manufacturer's protocol with the modification step recommended for array sample processing. Control PCR reactions were performed before array analysis to confirm successful DNA modification. Bisulfite-modified DNA (500 ng) was applied to the Infinium MethylationEPIC array (Illumina). According to the manufacturer's protocol, after bisulfite DNA treatment, cytosines in the CpG sites genotype as C/T polymorphisms according to the manufacturer's protocol. Raw array data was collected with ∼866,000 probes for 12 samples across four groups: control and NEN-treated under normoxia or hypoxia. The fluorescence signals were measured from the BeadArrays using an Illumina BeadStation GX Scanner. The data analysis was performed in the R environment. A detection P value for every probe in every sample, representing the signal quality, was generated using the minfi method (20). Poor probes with all samples’ P values ≥ 0.01 were removed, leaving ∼864,000 probes. Analyses used methylation level estimations based on M values (log₂-based methylation over unmethylation signal). Differentially methylated probes were identified by first standardizing M values using a Student two-tailed t test. Significant hits were identified on the basis of a Student two-tailed t test P < 0.05 and absolute M value difference > 0.5 between groups. Differentially methylated probes were annotated using the R CHAMP package.

Mutational analysis for exons 5 to 9 of the TP53 gene was performed using direct sequencing. DNA extraction was performed using the PureLink Genomic DNA Mini Kit (Gibco; K182001) according to the manufacturer's protocol. Each PCR was performed with an initial denaturation step of 3 minutes at 95°C, followed by 35 cycles of 95°C for 10 s, 58°C for 15 s, 72°C for 15 s, and a final extension step of 72°C for 5 minutes using the Phusion High-Fidelity PCR Kit (Thermo Fisher Scientific; #F553L). The PCR products were recovered after agarose gel electrophoresis. Both forward and reverse strands were sequenced to confirm TP53 mutations. The primers used are listed in Supplementary Table S4.

The quantification of neurite outgrowth was performed as previously described (4). Briefly, 1 × 10⁴ SK-N-BE(2) and NB16 cells per well were inoculated into a 12-well plate. After overnight incubation, cells were treated with the indicated conditions for 72 hours. Then, images were taken by a Leica Florescent Microscope Dmi8 in phase contrast mode (20 × magnification). Neurite lengths were traced and quantified using the ImageJ plugin NeuronJ (21). For each sample, total neurite length was measured and normalized to the number of cell bodies. Data are presented as mean ± SD across three biological replicates.

SK-N-BE(2) and NB16 cells were seeded into 8-champer slides (Thermo Fisher Scientific; 154534) at 6 × 10³ cells/well overnight and subjected to the indicated treatment for 72 hours. The procedure was performed as previously described (4). Cells were fixed with 4% paraformaldehyde in PBS with 0.1% Tween-20 at room temperature for 30 minutes, followed by permeabilization with 0.1%Triton X-100 in PBS at room temperature for 10  minutes. Next, the cells were washed with PBS twice and blocked with 2.5% horse serum in PBS at room temperature for 1 hour. Then, cells were subjected to immunofluorescence staining with primary antibody β3-Tubulin (CST; 5568) at 4°C overnight. Finally, after two washes with PBS, cells were incubated with Alexa Fluor 594-conjugated anti-rabbit secondary antibody (Life Technologies) at room temperature for 1 hour, followed by staining with 4′,6-diamidino-2-phenylindole (DAPI; Vector Laboratories, H-1800–2). Images were acquired with a Leica Dmi8 microscope. The data presented are representative of two independent experiments.

The metabolomics and isotope tracing analyses were performed on an Agilent 1290 Infinity Liquid Chromatography (LC) System (Agilent) coupled to a Q-TOF 6545 mass spectrometer (MS; Agilent). Targeted analysis, isotopologues extraction, and natural isotope abundance correction were performed with Agilent Profinder B.10.00 (Agilent) as previously described (4). Data are presented as mean ± SD across three biological replicates.

Glutamine isotope tracing under normoxia or hypoxia involved SK-N-BE(2) and NB16 cells being pretreated with DMSO or 1 μmol/L NEN for 3 hours. Next, the culture medium was changed to DMEM/F-12 without glutamine (Gibco; 21331020) or supplemented with 4 mmol/L U-¹³C-glutamine (Cambridge Isotope Laboratories; CLM-1822-H) or 10% dialyzed FBS (Gibco; 26400044) under normoxia or hypoxia for 2 hours. Data are presented as mean ± SD across three biological replicates.

All mouse procedures were performed in accordance with Stanford University guidelines for the care and use of animals and were maintained and handled under protocols approved by the Institutional Animal Care and Use Committee. All procedures were performed with a 1:1 sex ratio of 7-week-old nude mice (Taconic). Procedures and ultrasound measurements (see below) were performed under general anesthesia using isoflurane inhalation. Orthotopic tumors within the adrenal gland were created as previously described (22). Briefly, a transverse incision was made on the left flank to locate the left adrenal gland, and 1 μL of PBS containing 105 SK-N-BE(2) or NB16 cells was injected into the adrenal gland using a 30G needle. Fascia and skin were closed in separate layers. Tumor formation was monitored by noninvasive ultrasound measurements, and diet intervention was provided once the tumor volume reached 5 to 10 mm³. A control diet (#D11112201) and a diet with 2,000 ppm NEN (#D20101601) were obtained from Research Diets, Inc. The animals were euthanized when the tumor volume exceeded 1,500 mm³.

After securing the mouse in a prone position, a Vevo 2100 sonographic probe (Visual Sonics) was applied to the left flank to locate the left adrenal gland and the tumor. The monitoring of tumors by ultrasound was performed twice weekly. Serial cross-sectional images (0.076 mm between images) were taken. The tumor volume was measured using a 3D reconstruction tool (Vevo Software v1.6.0).

Hematoxylin and eosin (H&E) and immunohistochemically stained sections were prepared from formalin-fixed and paraffin-embedded blocks of mouse orthotopic tumors. Immunostaining involved heating unstained sections for 30 minutes in Bond Epitope Retrieval Solution 2 (No. AR9640; Leica Biosystems Newcastle Ltd.) and incubating them with anti-human N-Myc rabbit monoclonal antibody (1:300 dilution; Cell Signaling Technology, #51705). Then, counter-staining with hematoxylin was performed.

Cell proliferation and LC/MS experiments used three biological repeats for data analysis. Results are presented as mean ± SD. Mouse orthotopic neuroblastoma experiment data are presented as mean ± SEM. Student t tests and one- or two-way ANOVA tests were performed to assess the significance of differences between groups (two-tailed; unequal variance).

The sequencing data (GSE209801) and methylation array data (GSE211558) were submitted to the NCBI Gene Expression Omnibus database.

NEN is a salt form of the FDA-approved mitochondrial uncoupler drug with an excellent safety profile (23, 24). SK-N-BE(2) and NB16 cells were treated with NEN to investigate whether mitochondrial uncoupling could induce cell differentiation. Notably, NEN treatment induced neuron differentiation morphologic changes in both SK-NE-BE(2) and NB16 cells based on neurite length measurements and immunofluorescence staining against the neuron differentiation marker β-tubulin III (Fig. 1A and B). We confirmed that mitochondrial uncoupling could induce neuroblastoma differentiation by treating neuroblastoma cells with another mitochondrial uncoupler, BAM15 (25). Similarly, BAM15 treatment induced neuron differentiation morphologic changes in both SK-NE-BE(2) and NB16 cells (Supplementary Fig. S1A). In addition, NEN and BAM15 treatment induced cell proliferation arrest, which is usually associated with cell differentiation (Fig. 1C; Supplementary Fig. S1C). The IC₅₀ for NEN in inhibiting cell proliferation was ∼0.31 μmol/L in SK-N-BE(2) cells and ∼0.70 μmol/L in NB16 cells. The IC₅₀ of BAM15 in inhibiting cell proliferation was ∼2.76 μmol/L in SK-N-BE(2) cells and ∼4.53 μmol/L in NB16 cells. We determined whether proliferation arrest was the cause of differentiation by treating cells with an antimetabolite, 5-fluorouracil, and a cyclin-dependent kinase 4/6 inhibitor, palbociclib. Both drugs arrested proliferation but did not cause neuron differentiation morphologic changes (Supplementary Fig. S2C and S2D), indicating that proliferation arrest was not the cause of cell differentiation.

Because cell differentiation requires transcriptome changes, we performed RNA-seq analysis in SK-N-BE(2) cells treated with DMSO or NEN for 16 hours. We identified 495 significantly upregulated genes and 224 significantly downregulated genes with NEN treatment (Supplementary Table S1). The distribution of gene expression in the RNA-seq data is presented as a volcano plot (Fig. 1D). The upregulated genes were enriched in multiple pathways, including neurogenesis, nervous system development, and neuron differentiation (Fig. 1D and E; Supplementary Table S1), consistent with the neuron differentiation morphologic changes. In contrast, the NEN-downregulated genes were enriched in pathways involved in DNA replication and cell-cycle progression (Fig. 1D; Supplementary Fig. S2A; Supplementary Table S1), consistent with the inhibition of cell proliferation. In addition, GSEA of the RNA-seq gene sets showed that NEN treatment significantly reduced the expression of genes in two important cell division hallmarks, “E2F-TARGETS” and “G2M_CHECKPOINT” (Supplementary Fig. S2B). MYCN, a key oncogenic transcription factor, is often amplified in neuroblastoma, promoting pluripotency and unfavorable patient outcomes (26). Intriguingly, NEN treatment significantly reduced N-Myc mRNA and protein levels (Fig. 1F). Consistent with this, GSEA results showed that NEN treatment significantly reduced the expression of genes in the hallmarks “MYC_TARGETS_V1” and “MYC_TARGETS_V2” (Fig. 1G).

Next, we examined whether mitochondrial uncoupling could induce cell differentiation through metabolic reprogramming. Mitochondrial uncoupling is a process that dissipates the proton gradient across the inner mitochondrial membrane, inhibiting ATP synthesis and activating the ETC to promote NADH oxidation (27, 28). As expected, NEN treatment increased ADP/ATP and AMP/ATP ratios (Fig. 2A; Supplementary Fig. S3A), likely due to increased AMP and ADP levels caused by ATP synthesis inhibition (Supplementary Fig. S4A and S4B). Importantly, NEN treatment increased the intracellular NAD⁺/NADH ratio in both SK-N-BE(2) and NB16 cells (Fig. 2A; Supplementary Fig. S3A). Also, in line with the observation that cell-cycle arrest did not cause neuron differentiation morphologic changes (Supplementary Fig. S2C and S2D), 5-fluorouracil and palbociclib treatment did not alter the NAD⁺/NADH ratio (Supplementary Fig. S2E). In addition, the pyruvate/lactate ratio, determined by the NAD⁺/NADH ratio (29), also increased with NEN treatment, indicating inhibition of the Warburg effect (Fig. 2A; Supplementary Fig. S3A).

We also examined NEN's metabolic reprogramming effect on other cancer cell lines, including an ovarian cancer cell line, OVCAR3, and two lung cancer cell lines, H29 and H82. All cell lines showed similar metabolic reprogramming effects, characterized by increased ADP/ATP and α-KG/2-HG ratios, indicating that NEN-induced metabolic reprogramming is not limited to neuroblastoma cells (Supplementary Fig. S6A–S6C). Similar to NEN, BAM15 treatment also increased NAD⁺/NADH and α-KG/2-HG ratios (Supplementary Fig. S1B). Furthermore, NEN treatment did not appear to cause oxidative stress in our experimental setup because it did not reduce the glutathione/glutathione disulfide ratio in either SK-N-BE(2) or NB16 cells (Supplementary Fig. S4A and S4B).

Because multiple key reactions in the tricarboxylic acid (TCA) cycle use NAD⁺ as an electron acceptor, a high NAD⁺/NADH ratio is the major driving force for the TCA cycle (Fig. 2B). We next examined how mitochondrial uncoupling regulates TCA cycle metabolite levels. The levels of citrate, the first product of the TCA cycle, were unaffected by NEN treatment (Supplementary Fig. S4A, B). Surprisingly, intracellular α-KG levels increased significantly in both cell lines after NEN treatment (Fig. 2A; Supplementary Fig. S3A). Succinate also accumulated under NEN treatment, while the α-KG/succinate ratio remained unchanged (Supplementary Fig. S4A and S4B). No significant increases in fumarate or malate were observed under NEN treatment (Supplementary Fig. S4A and S4B). Aspartate, derived from oxaloacetate, also accumulated under NEN treatment (Supplementary Fig. S4A and S4B).

Because glutamate, derived from glutamine, is the major carbon source for generating α-KG in cultured cancer cells (30), we hypothesized that increased α-KG originated from glutamine. NEN treatment significantly reduced intracellular glutamine and glutamate levels, suggesting glutaminolysis acceleration via mitochondrial uncoupling (Fig. 2C; Supplementary Fig. S3B). We used the glutaminase inhibitor CB-839 to investigate whether glutaminolysis is necessary for NEN-induced differentiation. CB-839 treatment significantly reduced glutamate and downstream TCA cycle metabolite levels and the NAD⁺/NADH and α-KG/2-HG ratios (Supplementary Fig. S5B), suggesting that glutaminolysis is essential for maintaining ETC activity in these cells. CB-839 treatment reduced neuron outgrowth length induced by NEN, indicating that glutaminolysis is necessary for NEN-induced neuroblastoma cell differentiation (Supplementary Fig. S5C).

We investigated how mitochondrial uncoupling alters glutamine flux into the TCA cycle using an isotope tracing assay with uniformly ¹³C-labeled glutamine [(U-¹³C)-glutamine; Fig. 2B]. Surprisingly, in SK-N-BE(2) cells, no significant increase in m+5 α-KG was observed. However, the labeling percentage of m+3 (2nd cycle) and m+1/m+2 (3rd cycle) α-KG significantly increased under NEN treatment (Fig. 2D). In addition, the labeling fraction of m+2 (2nd cycle) and m+1 (3rd cycle) succinate and aspartate (from oxaloacetate) significantly increased under NEN treatment (Supplementary Fig. S3C), confirming that NEN treatment accelerates TCA cycle flux in the oxidative direction. Importantly, the m+5 labeling abundance of 2-HG was significantly decreased by NEN treatment in SK-N-BE(2) cells (Fig. 2D). The ratio of m+5 α-KG/m+5 2-HG increased under NEN treatment (Fig. 2D), indicating that NEN treatment reduced the conversion of α-KG to 2-HG. No significant reduction of m+5 2-HG was observed in NB16 cells, possibly due to lower 2-HG generation in this cell line (Supplementary Fig. S3D), while the m+5 α-KG/m+5 2-HG ratio still increased. The increased m+3/2/1 labeling in SK-N-BE(2) cells or m+5/3/1 labeling in NB16 cells reflected the α-KG increase associated with increased corresponding labeling in succinate and aspartate (Supplementary Fig. S3C and S3D). Altogether, these data suggest that NEN treatment increased cellular α-KG through two potential mechanisms: accelerated glutaminolysis and blocked α-KG to 2-HG conversion.

We assessed whether increased intracellular α-KG levels were sufficient to induce neuroblastoma cell differentiation by adding cell permeable α-KG (DMKG) to the cells and monitoring their morphologic changes. Indeed, α-KG supplementation, but not two other dimethyl ethers (dimethyl carbonate and dimethyl glutarate), was able to induce a neuron differentiation morphologic change in both SK-NE-BE(2) and NB16 cells, indicated by neurite length measurements and immunofluorescence staining for neuron differentiation marker β-tubulin III (Fig. 2E and F; Supplementary Fig. S5A), suggesting that increased intracellular α-KG levels can induce neuroblastoma differentiation. Furthermore, the α-KG treatment also reduced cell proliferation in neuroblastoma cells (Fig. 2G).

Because NEN treatment increased α-KG, the TET substrate, and the α-KG/2-HG ratio, we hypothesized that NEN treatment alters the DNA methylation landscape to induce neuronal differentiation gene expression. We used an Illumina human Infinium MethylationEPIC array to study loci-specific DNA methylation changes. Genomic annotations of differentially methylated sites under NEN treatment were categorized according to their genomic location. NEN treatment generated more hypomethylated sites than hypermethylated sites in CpG islands (Fig. 3A; Supplementary Table S2). In addition, NEN treatment generated more hypomethylated sites than hypermethylated sites in regions close to transcription start sites (TSS), including TSS1500, TSS200, 5′-untranslated region, and first exon (Fig. 3B; Supplementary Table S2), where DNA hypermethylation is correlated with transcription suppression (7). To investigate the biological functions of this epigenetic reprogramming, we performed a gene ontology (GO) analysis of genes harboring differential methylation under NEN treatment. Under NEN treatment, genes containing hypomethylated CpG islands in the promoter region (TSS1500 and TSS200), multiple neuron differentiation, and cell death–related pathways were enriched. However, they were not within the top 10 enrichments (Fig. 3C; Supplementary Table S2). In contrast, genes containing differentially methylated probes in the gene body CpG islands and with functions in pathways involved in nervous development, neuron projection morphogenesis, neuron differentiation, and axonogenesis were among the top 10 enrichments. These findings indicate that NEN treatment induced dramatic methylome reprogramming in the bodies of neuron differentiation-related genes (Fig. 3D; Supplementary Table S2). We determined whether DNA demethylation was sufficient to induce cell differentiation in neuroblastoma cells by treating cells with the DNA methyltransferase inhibitor 5-AZA. We found that 5-AZA treatment induced neuron differentiation morphologic changes, suggesting the neuron differentiation genes could be silenced through DNA methylation, consistent with previous observations (31) (Fig. 3E).

It has been reported that α-KG treatment activates p53 and inhibits the Wnt/β-catenin pathway while promoting cell differentiation (11, 32). Therefore, we hypothesized that NEN treatment also modulates these two pathways. Indeed, NEN treatment reduced the expression of canonical Wnt/β-catenin target genes and upregulated the expression of Wnt/β-catenin pathway inhibitors, including TP53 (33) and catenin beta interacting protein 1 (CTNNBIP1; ref. 34), in our RNA-seq data (Supplementary Fig. S7A). The downregulation of Wnt/β-catenin pathway genes transcription factor 7 (TCF7), baculoviral IAP repeat containing 5 (BIRC5), and cyclin D1 (CCND1) and upregulation of TP53 and CTNNBIP1 were validated using RT-qPCR (Supplementary Fig. S7B and S7C). In addition, GSEA results showed that NEN treatment significantly increased the expression of genes in the hallmark “p53_pathway” (Supplementary Fig. S7D). Consistent with gene expression changes, NEN treatment reduced β-catenin and increased p53 protein levels (Supplementary Fig. S7E and S7F). Notably, 5-AZA also increased p53 protein levels, indicating that TP53 upregulation by NEN was due to DNA demethylation (Supplementary Fig. S7E and S7F). However, 5-AZA treatment did not reduce β-catenin expression, suggesting that NEN treatment can regulate β-catenin expression at multiple levels (Supplementary Fig. S7E and S7F).

We determined the TP53 mutation status by performing genomic sequencing of exons 5–9 in SK-N-BE(2), NB16, and another p53 wild-type (WT) cell line, CHP134. We found that SK-N-BE(2) and NB16 harbor TP53 mutations C135F and R248W, respectively (Supplementary Fig. S7G), which could disrupt normal p53 function (35), suggesting that WT p53 is not necessary for NEN-induced cell differentiation. In addition, inhibition of cell proliferation and induction of neuron length outgrowth was also observed in the p53 WT neuroblastoma cell line CHP134 under NEN treatment (Supplementary Fig. S7H–S7J). Altogether, these data suggested that TP53 mutation status (at least for C135F and R248W) is not a key determinant in NEN-induced neuroblastoma differentiation.

Clinically, tumor hypoxia is a significant obstacle to treatment because hypoxic tumor cells are more resistant to radiation (36) and chemotherapy (37). Under hypoxia, α-KG is reduced and converted to L-2-HG due to a reduced NAD⁺/NADH ratio (15, 16), repressing α-KG–dependent dioxygenases (38), including DNA demethylase TET, and prolyl hydroxylases that hydroxylate the HIFα subunits. We found that NEN treatment could partially restore NAD⁺/NADH and pyruvate/lactate ratios under hypoxia (Fig. 4A). Importantly, NEN treatment partially restored α-KG levels and significantly reduced 2-HG levels under hypoxia (Fig. 4A). Surprisingly, NEN treatment induced dramatic global DNA demethylation changes under hypoxia, including methylation sites located in “ocean,” “shelf,” and “shore,” and especially in CpG islands (Fig. 4B). NEN treatment consistently induced neuron differentiation morphologic changes (Fig. 4C). Importantly, 4 days of hypoxia pretreatment increased proliferation when cells were replated under normoxia conditions (without treatment). When cells were pretreated under hypoxia with NEN, this gain of proliferation advantage from hypoxia pretreatment was blocked (Fig. 4D). In addition, NEN treatment reduced the accumulation of HIFα subunits (Fig. 4E) and the expression of HIF target genes such as pyruvate dehydrogenase kinase 1 (PDK1), pyruvate dehydrogenase kinase 3 (PDK3), phosphoglycerate kinase 1 (PGK1), and LDHA (Supplementary Fig. S8A). In contrast, α-KG supplementation increased HIF1α levels in SK-N-BE(2) cells (Fig. 4E), possibly because supplemented α-KG was converted to 2-HG (refs. 15, 39; Supplementary Fig. S8B and S8C). In NB16 cells, which have much lower 2-HG levels than SK-N-BE(2) cells (Supplementary Fig. S8B and S8C), α-KG supplementation did not significantly upregulate HIF1α (Fig. 4E). Altogether, these data suggested that NEN treatment effectively promotes DNA demethylation and inhibits HIF signaling, even under hypoxic conditions.

Next, we examined whether NEN supplementation inhibits tumor growth in vivo using an orthotopic neuroblastoma xenograft model (Fig. 5A). Because intraperitoneal injection of NEN shows poor serum pharmacokinetics (Supplementary Fig. S9A), we used a diet containing 2,000 ppm NEN for treatment delivery (40). LC/MS analysis indicated that dietary NEN supplementation led to abundant niclosamide accumulation in mouse serum (∼1 μmol/L after 24 hours and ∼3 μmol/L at the endpoint; Supplementary Fig. S9B and S9C). A previous study showed that the liver and kidney are the two organs with the highest niclosamide accumulation (40). NEN accumulation in orthotopic neuroblastoma xenograft tumors was comparable to levels found in the kidney but lower than in the liver (Supplementary Fig. S9C). NEN supplementation significantly decreased SK-N-BE(2) and NB16 cell-derived tumor growth compared with the control group (Fig. 5C and D; Supplementary Fig. S9E). Importantly, NEN treatment significantly prolonged mouse survival (Fig. 5C and D) without changing body weight or serum glucose level (Fig. 5C and D; Supplementary Fig. S9D). In addition, NEN treatment increased the NAD⁺/NADH ratio in the tumor (Fig. 5B). Importantly, 2-HG levels were reduced, α-KG levels did not change, and together the α-KG/2-HG ratio was increased in the tumors in the NEN-treated group (Fig. 5B; Supplementary Fig. S9F). Furthermore, tumor cells from the NEN-treated group had much fewer enlarged prominent nucleoli (Fig. 5E), an indicator of active ribosome biogenesis and worse prognosis in patients with neuroblastoma (26). Importantly, N-Myc and β-catenin protein levels were significantly reduced in NEN-treated xenografts (Fig. 5E), consistent with our observations from in vitro assays (Fig. 1F; Supplementary Fig. S7E and S7F).

We further explored the clinical relevance of NEN-induced gene expression profile changes with an integration analysis using the gene expression profile from the NEN treatment experiment and 10 neuroblastoma patient gene expression profiles from the R2 database (Supplementary Table S3A and S3B; ref. 18). We generated gene sets indicating favorable or unfavorable prognosis from the 10 patient datasets in the R2 database (the Kaplan–Meier analysis of each gene had P < 0.05). Then, we constructed sets of overlapping genes from seven gene sets containing > 1,000 genes: an overlapped “favorable prognosis gene set” with 306 genes and an overlapped “unfavorable prognosis gene set” with 289 genes (Fig. 6A and B; Supplementary Table S3C and S3D). These two gene sets were subjected to pathway enrichment analysis. The overlapped favorable prognosis gene set was enriched for multiple neuron differentiation terms, while the overlapped unfavorable prognosis gene set was enriched for cell cycle–related terms (Fig. 6A and B; Supplementary Table S3D and S3E).

Next, favorable prognosis and unfavorable prognosis gene signatures from each study were defined as pathways for GSEA analysis. Surprisingly, NEN-upregulated genes were more enriched in all the favorable prognosis gene pathways except one (Fig. 6C and D; Supplementary Table S3F). In contrast, NEN-downregulated genes were more enriched in unfavorable prognosis gene pathways (Fig. 6C and D; Supplementary Table S3G). The only exception was that NEN-upregulated genes were enriched in an unfavorable prognosis gene set derived from MYCN nonamplification neuroblastoma. A potential explanation for this discrepancy is that our RNA-seq data was generated from SK-N-BE(2) cell line, which is an MYCN-amplified cell line (Fig. 1F). We also compared our RNA-seq data with a previous study that identified niclosamide as an effective drug to target neuroblastoma with a “high-risk gene signature” comprising 99 genes (41). Most of the 94 genes upregulated in the high-risk tumor signature were downregulated under NEN treatment. Among the five genes downregulated in the high-risk tumor signature, three were upregulated under NEN treatment (Supplementary Fig. S10A and S10B). GO analysis of the “high-risk gene signature” found that multiple cell-cycle processes and DNA replication pathways were enriched (Supplementary Fig. S10C), consistent with the GO analysis of the NEN-downregulated genes (Supplementary Fig. S2A). Altogether, these data suggest that NEN treatment reprograms the transcriptome to a gene expression profile associated with a favorable prognosis in patients with MYCN-amplified neuroblastoma.

Since Otto H. Warburg discovered that cancer cells have a high glucose consumption rate and produce large amounts of lactate in the 1920s (1, 2), oncologists have been interested in how cancer cells alter metabolic pathways to gain advantages during cancer progression. In one of his milestone reviews, Warburg proposed that the cause of the Warburg effect was respiration injury and, consequently, cell dedifferentiation (3). At that time, he was aware that respiration inhibition led to metabolic reprogramming and insightfully proposed a connection between metabolic reprogramming and cell dedifferentiation. However, the underlying mechanism remained unclear due to the field's limited understanding of the metabolic control of epigenetics.

Respiration inhibition is essentially ETC inhibition, which oxidizes cellular NADH to NAD⁺. ETC inhibition leads to a decrease in the NAD⁺/NADH ratio. A previous study examining NAD⁺/NADH ratios using the SoNar fluorescence sensor in cell lines showed that most cancer cell lines have lower NAD⁺/NADH ratios than primary cell lines (42). A low NAD⁺/NADH ratio not only drives lactate production from pyruvate to enhance the Warburg effect but also disrupts TCA cycle flux, promoting the chemical reduction of α-KG to form L-2-HG (15, 16). Similar to D-2-HG produced by tumors carrying IDH1/2 mutations (13, 14), L-2-HG also inhibits α-KG–dependent dioxygenases, including TET and Jumonji C domain-containing enzymes (38), leading to DNA and histone hypermethylation (15, 43). This hypermethylation phenotype has been shown to block cell differentiation, promoting tumor progression (44, 45). Here, we showed that treatment using mitochondrial uncouplers NEN and BAM15 effectively increased cellular α-KG levels in neuroblastoma cells to promote differentiation. NEN treatment increased the NAD⁺/NADH ratio, accelerating glutaminolysis to upregulate α-KG and blocking the conversion of α-KG to 2-HG. Surprisingly, even under hypoxia, NEN treatment was still effective in restoring α-KG and reducing 2-HG levels, suggesting that NEN supplementation may be effective in vivo.

The cancer epigenome generally shows global DNA hypomethylation and promoter CpG island DNA hypermethylation. DNA hypermethylation in the promoter region leads to tumor suppressor and differentiation gene silencing. NEN treatment induced DNA hypomethylation in the promoter region and activated neuron differentiation-related gene expression. Intriguingly, NEN-induced differential methylation sites are enriched in the body of neuron differentiation genes, implying that gene body methylation status may also play a critical role in cell fate commitment. A previous study showed that DNA methylation in the gene body is critical in regulating its expression (46). Unlike pan-demethylating agents such as 5-AZA, NEN redistributed the DNA methylation landscape, decreasing methylation in the promoter while increasing methylation in the gene body, making it a more effective and precise epigenetic intervention for cancer treatment. Altogether, these data indicate that Warburg's original hypothesis was correct: ETC activity is essential for cell differentiation. Further studies are needed to examine whether DNA demethylation by NEN treatment depends on restoration of TET activity, because there is no direct evidence provided in this study.

It has been reported that niclosamide treatment can induce TP53 expression, and p53-deficient cancer cells are more sensitive to niclosamide treatment (47). Unlike in adult cancers, where TP53 is often mutated, pediatric cancers such as neuroblastoma exhibit a very low TP53 mutation rate (48). Epigenetic silencing through promoter methylation is one major mechanism of p53 inhibition in neuroblastoma (48). Both NEN and 5-AZA treatment increased p53 protein levels in NB16 and SK-N-BE(2) cells, suggesting that p53 is upregulated in neuroblastoma cells through DNA demethylation.

MYCN is the major oncogene amplified in neuroblastoma. High MYCN expression is one key prognosis marker indicating poor survival in patients with neuroblastoma. MYCN is downregulated upon induction of differentiation in neuroblastoma cells (49), suggesting a negative correlation between MYCN transcription and neuroblastoma cell differentiation. Here, we showed that mitochondrial uncoupler treatment could diminish the “undruggable” MYCN expression, highlighting its therapeutic potential in patients with high-grade neuroblastoma. A previous study suggested that niclosamide can be repurposed to treat high-risk neuroblastoma by inducing apoptosis and cell-cycle arrest (41). Our data suggest that inducing differentiation is an alternative mechanism of niclosamide-induced tumor suppression.

In addition to α-KG upregulation, mitochondrial uncoupler treatment caused global metabolic reprogramming, including increasing NAD⁺/NADH and AMP/ATP ratios. These metabolic changes may further alter other signaling pathways dependent on NAD⁺ or ATP to remodel epigenetics and change cell fate. For example, it was well established that NEN treatment activates AMP-activated protein kinase (27), which phosphorylates and stabilizes TET (50). ETC inhibition and induction of the Warburg effect alter metabolism, causing global epigenetic reprogramming and cell dedifferentiation rather than affecting a single gene or pathway. Therefore, activating the ETC with mitochondrial uncouplers is expected to be an effective approach for reversing the Warburg effect and restoring the global metabolome and epigenome, redirecting tumor cells into a differentiated state.

H. Jiang reports a patent for US20220175704A1 pending. Y. Li reports a patent for US20220175704A1 pending. J.J. Gruber reports grants from Hummingbird Bioscience and personal fees from Guidepoint and Sharma Therapeutics outside the submitted work. B. Chiu reports grants from NIH during the conduct of the study. J. Ye reports grants from Stanford Maternal and Child Health Research Institute and American Cancer Society during the conduct of the study; in addition, J. Ye has a patent for US20220175704A1, pending. No disclosures were reported by the other authors.

H. Jiang: Investigation, writing–original draft, writing–review and editing. R.L. Greathouse: Investigation. S.J. Tiche: Investigation. M. Zhao: Investigation. B. He: Investigation. Y. Li: Investigation. A.M. Li: Investigation. B. Forgo: Investigation. M. Yip: Investigation. A. Li: Investigation. M. Shih: Investigation. S. Banuelos: Investigation. M.N. Zhou: Investigation. J.J. Gruber: Resources. E.B. Rankin: Writing–review and editing. Z. Hu: Software. H. Shimada: Resources. B. Chiu: Investigation. J. Ye: Conceptualization, resources, supervision, funding acquisition, investigation, writing–review and editing.

This work was supported by a Stanford Maternal and Child Health Research Institute Research Scholar Award (2020) and an American Cancer Society Research Scholar Grant (RSG-20–036–01 to J. Ye) and NIH funding (R01NS094218 to B. Chiu). The authors thank Dr. Yuqin Dai (Stanford ChEM-H Institute Metabolomics Knowledge Center) for support and use of facilities, Vanita S. Natu (Stanford Genomics) for providing Infinium MethylationEPIC analysis service, Drs. Maximilian Diehn, Beverly S. Mitchell, Laura Attardi, and Roeland Nusse (Stanford University) for valuable advice and suggestions, and John M. Snyder (Radiation Oncology) for support on statistical analysis.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.