Metabolic Reprogramming by MYCN Confers Dependence on the Serine-Glycine-One-Carbon Biosynthetic Pathway

MYCN is a member of the MYC family of oncogenic transcription factors that also include MYC and MYCL. A primary function of MYCN, as well as other MYC family proteins, is to promote cell growth and proliferation through transcriptional regulation (1, 2). Aberrant MYCN activation, mainly via genomic amplification, is commonly observed in neuronal and neuroendocrine cancers, including neuroblastoma (3), medulloblastoma (4, 5), Wilms tumor (6), retinoblastoma (7), neuroendocrine prostate cancer (8–10), glioblastoma (11, 12), and small-cell lung cancer (13–15). However, MYCN is a poor drug target (16). We reasoned that the cell proliferation program driven by MYCN activation might create a selective pressure on metabolic pathways to meet the biosynthetic demand of growth, thereby imposing metabolic dependences that can be exploited therapeutically. Our investigation reveals that MYCN and ATF4, a master regulator of amino acid metabolism and stress responses (17–19), form a positive feedback loop for transcriptional activation of the serine-glycine-one-carbon (SGOC) metabolic pathway. This metabolic reprogramming generates a SGOC pathway–dependent synthetic lethal effect in MYCN-amplified neuroblastoma cells. Thus, targeting the SGOC metabolic pathway may provide selective therapeutic benefits for patients with MYCN-amplified cancers.

Neuroblastoma cell lines BE(2)-C (CRL-2268), IMR32 (CCL-127), SH-SY5Y (CRL2266), SK-N-AS (CRL-2137), SK-N-DZ (CRL-2149), SK-N-FI (CRL-2141), and SK-N-SH (HTB-11) were obtained from ATCC, LA1-55n (06041203) from Sigma-Aldrich, and CHLA-90, LA-N-5, LA-N-6, SMS-KANR, and SMS-KCNR from Children's Oncology Group Cell Culture and Xenograft Repository. NBL-S was a gift from S.L. Cohn (University of Chicago, Chicago, IL), NLF from M.C. Simon (University of Pennsylvania, Philadelphia, PA), and SHEP1 from V.P. Opipari (University of Michigan, Ann Arbor, MI). All cell lines had been authenticated using short tandem repeat profiling (ATCC) and after authentication, large frozen stocks were made to ensure against contaminations by other cell lines. All cell lines were used within 10 passages after reviving from the frozen stocks and were free of Mycoplasma contamination as determined by a LookOut Mycoplasma PCR Kit (Sigma-Aldrich) and staining cells with DAPI every 3 months. IMR32, SHEP1, SH-SY5Y, SK-N-AS, and SK-N-SH were cultured in DMEM (HyClone; SH30022), BE(2)-C in DMEM/Ham's F-12 1:1 (HyClone; SH30023), and all others in RPMI1640 (HyClone; SH30027). All culture media were supplemented with 10% FBS (Atlanta Biologicals; S11050).

The Sequencing Quality Control (SEQC) dataset of MAQC-III was downloaded from the NCBI Gene Expression Omnibus (accession number GSE62564; ref. 20). Kaplan–Meier survival analysis based on the SGOC 6-gene signature (PHGDH, PSAT1, SHMT2, MTHFD1, MTHFD1L, and MTHFD2) was performed using the software R 3.3.2. Gene expression correlation analyses were conducted online using the R2 Platform (https://hgserver1.amc.nl/cgi-bin/r2/main.cgi), and the resulting figures and P values were downloaded. Gene Ontology (GO) analysis of the MYCN-amplification gene signature (21) was performed using the software DAVID (22).

Total RNA was isolated using TRIzol (Invitrogen). Reverse transcription was performed using an iScript Advanced cDNA Synthesis Kit (Bio-Rad). Quantitative reverse-transcription PCR (qRT-PCR) was performed using a 2× SYBR green qPCR master mix (Bimake) on an iQ5 real-time PCR system (Bio-Rad) with gene-specific primers (Supplementary Information). All samples were normalized to β2 microglobulin (B2M) mRNA levels.

The Retro-X Tet-Off Advanced Inducible System (Clontech) was used to generate neuroblastoma cell lines with inducible ATF4, FBXW7α, or MYCN expression in the absence of doxycycline (Doxy). Human myc-tagged ATF4 and myc-tagged FBXW7α were generated by PCR using pRK-ATF4 (Addgene; #26114) and pCR4-TOPO-FBXW7α (Open Biosystems; MHS4426-99239216) as templates, respectively. The sequences were verified by sequencing and subcloned into pRetroX-Tight-pur (Clontech) and pCDH-CMV-MCS-EF1-puro (SBI System Biosciences). Lentiviral shRNA constructs shATF4-73 (TRCN0000013573), shATF4-74 (TRCN0000013574), shMYCN-94 (TRCN0000020694), and shMYCN-95 (TRCN0000020695) were obtained from Sigma-Aldrich, shPHGDH (RHS3979-9595900) from Thermo Fisher Scientific, and shFBXW7 (RHS4533-NM_001013435) from Open Biosystems. Retroviruses for the expression of ATF4, FBXW7α, or MYCN were produced in 293FT cells using the packaging plasmids pHDM-G and pMD.MLVogp (gifts from R. Mulligan Massachusetts Institute of Technology, Cambridge, MA), and lentiviruses for the expression of shRNA, ATF4, or MYCN were produced in 293FT cells using the packaging plasmids pLP1, pLP2, and pLP/VSVG (Thermo Fisher Scientific). Retroviral and lentiviral infections of cells were conducted according to standard procedures.

Proteins (20–50 μg) were separated on SDS-PAGE, transferred to nitrocellulose membranes, and probed with antibodies (Supplementary Information). Proteins were visualized and quantified using the Odyssey system (LI-COR Biosciences) or a Clarity Western ECL Kit (Bio-Rad) and ImageJ (version 1.50e). Films were exposed for various times for quantification of target proteins within their linear range of detection.

Chromatin immunoprecipitation (ChIP) was performed as described (23) using the cell lines BE(2)-C, SMS-KCNR, SHEP1_ pCDH-MYCN, BE(2)-C_shATF4-73, and BE(2)-C_tetoff-ATF4 (± 2 μg/mL Doxy for 6 days). For each antibody, ∼4 × 10⁷ cells were used. Cross-linked chromatin was sheared through sonication (Fisher Scientific Model 150E) and immunoprecipitated using ChIP grade rabbit anti-ATF4 (sc-200), mouse anti-MYCN (B8.4.B, sc-53993), control mouse IgG (sc-2025) or rabbit IgG (sc-2027) from Santa Cruz Biotechnology. For qPCR, 2 independent ChIP samples were analyzed, and each sample was assayed in triplicate using ChIP and quantitative PCR (ChIP-qPCR) primers (Supplementary Information). The number associated with each primer set indicates the position of the forward primer relative to its target gene transcription start site (TSS, +1).

BE(2)-C_tetoff-ATF4 and SK-N-DZ_tetoff-ATF4 cells were cultured with or without 2 μg/mL Doxy for 5 days, followed by addition of cycloheximide to a final concentration of 10 μg/mL. Samples were then collected at various time points for immunoblot analysis of MYCN protein levels, which were quantified against β-actin using ImageJ.

The assay was performed as described (24). Briefly, BE(2)-C cells were cotransfected with pFlag-CMV-2-ubiquitin and pcDNA3.1-FBXW7α, along with or without pCDH-ATF4, and BE(2)-C cells expressing shGFP, shATF4-73, shFBXW7-55, or both shATF4-73 and shFBXW7-55 were transfected with pFlag-CMV-2-ubiquitin. Approximately 44 hours after transfection, cells were treated with 10 μmol/L MG-132 for 4 hours, washed with PBS, and lysed with a lysis buffer containing 2% SDS, 150 mmol/L NaCl, 10 mmol/L Tris-HCl, (pH 8.0), 2 mmol/L sodium orthovanadate, and 50 mmol/L sodium fluoride. The lysates were boiled for 10 minutes, sonicated (Fisher Scientific Model 150E, amplitude 40 and time 10 seconds), and diluted 1:10 with a dilution buffer (10 mmol/L Tris-HCl, pH 8.0, 120 mmol/L NaCl, and 1% Triton-X 100), followed by MYCN immunoprecipitation with mouse anti-MYCN (B8.4.B) and Protein G Dynabeads (Thermo Fisher Scientific; 10003D) and immunoblotting using rabbit anti-Flag for Flag-ubiquitin (Sigma-Aldrich; F-7425) and rabbit anti-MYCN (Cell Signaling Technology; #9405).

Small molecule inhibitors NCT-503 (Sigma-Aldrich; SML1659), NCT-503 inactive control (Sigma-Aldrich; SML1671), and CBR-5884 (Sigma-Aldrich; SML1656) were dissolved in DMSO, and stock solutions were aliquoted and stored at −80°C until use. Cells were treated with DMSO, NCT-503, or CBR-5884 for 2 to 3 days, and the number of viable cells was determined by trypan blue exclusion assay. To assess the effect of supplemental serine on NCT-503 action, cells were cultured in MEM (Gibco; 11095-080) supplemented with 10% dialyzed FBS (Gibco; 26400-044), 4.5 g/L d-(+)-glucose (Sigma-Aldrich; G8769), 1× MEN vitamins (Gibco; 11120-052), 1× sodium pyruvate (Gibco; 11360-070), and nonessential amino acids alanine, asparagine, glutamic acid, glycine, proline with or without 0.4 mmol/L serine.

[³H]-Glycine (60 Ci/mmol) and [³H]-Serine (11 Ci/mmol) were purchased from Moravek. Cells in 24-well plates were washed twice with the transport buffer (25 mmol/L HEPES/Tris, pH 7.5, 140 mmol/L NaCl, 5.4 mmol/L KCl, 1.8 mmol/L CaCl₂, 0.8 mmol/L MgSO₄, and 5 mmol/L glucose). Transport of 5 μmol/L glycine or serine (0.05 μmol/L [³H]-glycine or [³H]-serine plus 4.95 μmol/L unlabeled glycine or serine) was measured at 37°C for 2.5, 5.0, 10, and 15 minutes. Transport was terminated by aspiration of the transport buffer followed by 3 washes with 2 mL of ice-cold transport buffer. The cells were lysed with 0.5 mL of 1% SDS in 0.2 N NaOH and the radioactivity was determined by scintillation counting.

SHEP1-pCDH and SHEP1-pCDH-MYCN cells were washed with 5% mannitol and extracted with HPLC grade methanol (Thermo Fisher Scientific; A452-4) containing internal standards according to the instructions provided by Human Metabolome Technologies (HMT). Extracted metabolites were analyzed by capillary electrophoresis time-of-flight mass spectrometry (CE-TOFMS) at HMT. Peaks were annotated with putative metabolites from the HMT metabolite database based on their MTs/RTs and m/z values determined by TOFMS. The tolerance range for the peak annotation was configured at ±0.5 minutes for MT and ±10 ppm for m/z. Peak areas were normalized against the internal standards and then the cell number. Three biological replicate samples (∼10⁶ cells/sample) were analyzed for each cell type and P values were calculated by Welch t test.

BE(2)-C and SK-N-AS cells were cultured in glucose-deficient DMEM (11966-025; Invitrogen) supplemented with 10% dialyzed FBS and 20 mmol/L [U-¹³C]glucose (CLM-1396; Cambridge Isotope Laboratories) in the presence of DMSO or 10 μmol/L NCT-503 for 24 hours. Cells were collected by scraping and centrifugation, and cell pellets were washed once with ice-cold PBS, snap frozen in liquid nitrogen, and stored at −80°C until analysis. Four to six biological replicate samples (∼5 × 10⁶ cells/sample) were analyzed for each condition and cell type. Metabolite extraction and deprivation, instrumentation, and data processing are described in Supplementary Information.

Total RNA was isolated using TRIzol from 3 biological replicates of BE(2)-C cells treated with DMSO or 10 μmol/L NCT-503 for 48 hours. Affymetrix microarray was performed using the Human Gene 2.0 ST microarray chip. Data were normalized, significance determined by ANOVA, and fold change calculated with the Partek Genomics Suite. GO analysis was performed with DAVID (22) for all differentially expressed genes (≥±1.5 fold, P < 0.05), and GSEA was performed as described (25). The NCBI Gene Expression Omnibus (GEO) accession number for the microarray data is GSE120347.

BE(2)-C and SK-N-AS cells in 200 μL serum-free DMEM were injected subcutaneously into both flanks of 6-week-old male and female NOD/SCID mice at 2.5 × 10⁶ cells per injection site. Tumor volume was measured every other day using a digital caliper and estimated using the equation V = (L x W²)/2. Mice bearing tumors of ∼100 mm³ were randomly divided into 2 groups (5 per group), stratified by gender and tumor volume, and treated with vehicle (40 μLDMSO) or 64 mg/kg (body weight) NCT-503 delivered daily by intraperitoneal injection for 10 days. Animals were euthanized when their tumors reach ∼1.0 cm in diameter. The animal experiments were approved by the Institutional Animal Care and Use Committee of Medical College of Georgia, Augusta University.

Quantitative data are presented as mean ± SD and were analyzed for statistical significance by unpaired, 2-tailed Student t test or 2-way ANOVA. PHGDH inhibitor dose–response curves were fitted with the 4-parameter equation “log (inhibitor) vs. response – variable slope” and box plots were generated using the box and whiskers (Turkey) plot. For xenograft study, log-rank test was used to account for mouse survival. Unless otherwise stated, all statistical analyses were conducted using GraphPad Prism 7.0d for Mac.

Additional methods are provided in Supplementary Information.

High-risk neuroblastoma is a leading cause of pediatric cancer deaths, and MYCN amplification has a major role in its development (2, 26–28). To search for potential MYCN-dependent metabolic pathways, we performed GO analysis of the MYCN-amplification gene signature consisting of 143 genes upregulated in neuroblastoma tumors with MYCN amplification relative to those without (21). This analysis identified serine-glycine synthesis, one-carbon metabolism, and purine nucleotide biosynthesis as the most significantly upregulated metabolic processes in MYCN-amplified neuroblastoma (Fig. 1A). These metabolic processes are connected by the SGOC pathway (Fig. 1B; refs. 29, 30). Indeed, the MYCN-amplification gene signature contains several genes encoding enzymes within the SGOC pathway (21), including phosphoglycerate dehydrogenase (PHGDH), phosphoserine aminotransferase 1 (PSAT1), mitochondrial serine hydroxymethyltransferase 2 (SHMT2), cytosolic methylenetetrahydrofolate dehydrogenase, cyclohydrolase and formyltetrahydrofolate synthetase 1 (MTHFD1), and mitochondrial methylenetetrahydrofolate dehydrogenase 2 (MTHFD2). These enzymes link the glycolytic intermediate 3-phosphoglycerate (3PG) to the production of serine, glycine, and the one-carbon carriers 5,10-methylene-tetrahydrofolate (5,10-MTHF) and 10-formyl-THF (Fig. 1B). Both glycine and 10-formyl-THF donate carbon, and glycine also contributes nitrogen, moieties for purine nucleotide synthesis; 5,10-MTHF is essential for de novo thymidylate production as a coenzyme of thymidylate synthase.

We further assessed the correlation between MYCN amplification and transcriptional reprogramming of SGOC metabolism in neuroblastoma by analyzing the gene expression profiling data from a cohort of patients with neuroblastoma (n = 498, the SEQC dataset; ref. 20). The average expression of the 6 SGOC genes (PHGDH, PSAT1, SHMT2, MTHFD1, MTHFD1L, and MTHFD2) was significantly higher in high-risk tumors with MYCN amplification compared with low-risk tumors and to high-risk tumors without MYCN amplification (Fig. 1C). We confirmed the observation by analyzing the expression of individual SGOC genes in neuroblastoma tumors with or without MYCN amplification (Supplementary Fig. S1A). In agreement with the findings from tumors, MYCN-amplified neuroblastoma cell lines expressed markedly higher levels of PHGDH protein in comparison with non-MYCN–amplified cell lines (Fig. 1D). This PHGDH expression pattern is specifically associated with MYCN amplification but is not a general feature of rapidly proliferating neuroblastoma cell lines as the non-MYCN–amplified cell lines SK-N-AS, SK-N-FI, and SHEP1, despite their low PHGDH expression (Fig. 1D), proliferated at similar or faster rates in comparison with the MYCN-amplified cell lines BE(2)-C, SK-N-DZ, and SMS-KCNR as determined by cell population doubling time (Supplementary Fig. S1B). Moreover, inhibition of the proliferation of neuroblastoma cell lines by serum deprivation did not reduce PHGDH protein expression (Supplementary Fig. S1C and S1D). Overexpression of PHGDH in non-MYCN–amplified cells had no significant impact on cell proliferation (Supplementary Fig. S1E and S1F).

Further analysis of the SEQC dataset revealed that higher expression of the SGOC gene signature is significantly associated with advanced stages of neuroblastoma and poor prognosis in patients with neuroblastoma (Fig. 1E and F). Together, these findings provide genetic evidence for transcriptional activation of the SGOC metabolic pathway in high-risk neuroblastoma with MYCN amplification.

We next investigated whether MYCN is required for transcriptional activation of the SGOC genes. MYCN knockdown by shRNA in MYCN-amplified neuroblastoma cell lines reduced mRNA and protein expression of the SGOC genes (Fig. 2A and B; Supplementary S2A). Thus, MYCN has a major role in maintaining steady-state expression of the SGOC genes in these cell lines. In addition, MYCN depletion inhibited cell proliferation and induced differentiation characterized morphologically by neurite outgrowth (Supplementary Fig. S2B and S2C).

We next investigated whether an increase in MYCN level is sufficient to activate SGOC gene transcription. MYCN overexpression in non-MYCN–amplified neuroblastoma cell lines increased expression of the SGOC genes at the mRNA and protein levels (Fig. 2C–F) and enhanced cell proliferation (Supplementary Fig. S2D). Metabolomic profiling of non-MYCN–amplified SHEP1 cells with or without MYCN overexpression revealed that MYCN overexpression resulted in a significant increase in the intracellular levels of serine and glycine (Supplementary Fig. S2E) but had no effect on serine and glycine transport (Supplementary Fig. S2F), suggesting that MYCN overexpression enhanced biosynthetic activity of the SGOC metabolic pathway.

To determine whether MYCN directly targets the SGOC genes for transcriptional activation, we performed ChIP-qPCR assays. MYCN and other members of the MYC family bind to the consensus E-box sequence CANNTG. We focused on PHGDH, PSAT1, MTHFD1L, and MTHFD2, which contain the canonical E-box sequence CACGTG in their promoters or first introns (Fig. 2G). ChIP-qPCR analysis of the MYCN-amplified BE(2)-C cells showed significant levels of endogenous MYCN associated with the E-box sequences (Fig. 2H). We also detected significant levels of MYCN in the proximal promoter regions of PHGDH, PSAT1, MTHFD1L, and MTHFD2 (Fig. 2I). We obtained essentially the same results with the MYCN-amplified SMS-KCNR cells (Supplementary Fig. S2G and S2H) and the non-MYCN–amplified SHEP1 cells with MYCN overexpression (Supplementary Fig. S2I). Collectively, these findings suggest that the SGOC genes are direct transcriptional targets of MYCN.

ATF4 has a major role in transcriptional activation of enzymes within the serine-glycine synthesis pathway (31–35). Also, it has been shown recently that ATF4 acts downstream of mTORC1 and NRF2 in transcriptional activation of one-carbon flux and nucleotide biosynthesis (36, 37). Given the above findings on MYCN, we wondered if ATF4 has a significant role in transcriptional regulation of the SGOC genes in MYCN-amplified cells. Analysis of the SEQC dataset revealed a strong positive correlation in mRNA expression levels between ATF4 and the SGOC genes, except for PSPH and SHMT1 (Supplementary Fig. S3A). Overexpression of ATF4 in both MYCN-amplified and non-MYCN–amplified cell lines increased SGOC gene expression at the mRNA and protein levels (Fig. 3A and B; Supplementary Fig. S3B and S3C). Conversely, ATF4 depletion by shRNA in MYCN-amplified cell lines markedly decreased mRNA and protein expression of the SGOC genes (Fig. 3C and D) and inhibited cell proliferation (Supplementary Fig. S3D). Interestingly, ATF4 overexpression had no significant effect on the proliferation of non-MYCN–amplified cell lines (Supplementary Fig. S3E) but significantly enhanced the proliferation of MYCN-amplified cell lines (Supplementary Fig. S3F), suggesting that ATF4 might cooperate with MYCN in driving cell proliferation. Collectively, these findings indicate an essential role of ATF4 in maintaining SGOC gene expression and sustaining the proliferation of MYCN-amplified neuroblastoma cells.

To determine whether ATF4 directly targets the SGOC genes in MYCN-amplified cells, we performed ChIP-qPCR analysis of BE(2)-C_tetoff-ATF4 cells. In the absence of induction of exogenous ATF4, significant levels of endogenous ATF4 were found at the promoters of PHGDH, PSAT1, SHMT2, MTHFD1L, and MTHFD2, as well as those of ASNS and DDTI3, 2 classic ATF4 target genes (Fig. 3E; refs. 38, 39). Induction of ATF4 further increased the levels of ATF4 at these sites, except for the MTHFD2 promoter (Fig. 3E). In addition, ATF4 knockdown completely abolished the binding of endogenous ATF4 to the promoters of PSAT1, MTHFD2, and DDIT3 (Supplementary Fig. S3G). Thus, the SGOC genes are direct transcriptional targets of ATF4.

Because both MYCN and ATF4 are required for transcriptional activation of the SGOC genes in MYCN-amplified cells, we investigated how they coordinate their activities in this process. Analysis of the SEQC dataset revealed significantly higher ATF4 mRNA levels in tumors with MYCN amplification than those without (Supplementary Fig. S4A) and a positive correlation in mRNA expression levels between ATF4 and MYCN (Supplementary Fig. S4B), suggesting that one might regulate the expression of the other or that both are transcriptionally regulated by a common mechanism. To test these models, we examined the abilities of ATF4 and MYCN to regulate the expression of each other. ATF4 overexpression had no significant effect on MYCN mRNA expression (Fig. 4A), whereas MYCN overexpression significantly upregulated mRNA and protein expression of ATF4 (Fig. 4B and C). In addition, MYCN knockdown led to a significant reduction in ATF4 mRNA and protein levels (Fig. 4D and E). Thus, MYCN is a key transcriptional activator of ATF4 expression in neuroblastoma cells.

The ATF4 gene contains 2 canonical E-box sequences (CACGTG), one in the first intron (+767) and the other in the 3′ region (+3915). ChIP-qPCR analysis demonstrated significant levels of endogenous MYCN binding to these regions (Fig. 4F, ATF4p_+687 and ATF4p_+3818). In addition, we detected significant levels of MYCN at the ATF4 promoter 5′ region (Fig. 4F). Thus, ATF4 is a direct target gene of MYCN.

Collectively, these observations suggest that ATF4 functions downstream of MYCN in transcriptional activation of the SGOC genes. In support of this model, ATF4 knockdown significantly abrogated the ability of MYCN to induce the SGOC genes (Fig. 4G), indicating an essential role of ATF4 in mediating transcriptional activation of the SGOC genes by MYCN.

During our ChIP-qPCR studies with BE(2)-C_tetoff-ATF4 cells (Fig. 3E), we unexpectedly observed that following ATF4 induction, there was a significant increase in MYCN levels at the promoters of the SGOC genes (Fig. 5A) and ATF4 (Supplementary Fig. S4C). It has been shown previously that elevated levels of MYC lead to increased MYC occupancy at the promoters of active genes (40, 41). Because ATF4 had no effect MYCN mRNA expression (Fig. 4A), we asked whether it affects MYCN protein expression. Immunoblotting revealed that ATF4 overexpression increased MYCN protein levels (Fig. 5B), whereas ATF4 knockdown reduced MYCN protein expression (Fig. 5C) and increased MYCN protein ubiquitination (Fig. 5D), leading to a decrease in MYCN occupancy at the SGOC gene promoters (Supplementary Fig. S4D). Moreover, ATF4 overexpression resulted in an average of 2.5-fold increase in MYCN protein half-life (Fig. 5E and F). Collectively, these data suggest that ATF4 increases MYCN protein levels by blocking its degradation.

FBXW7 is a component of the SKP1/CUL1/FBXW7 E3 ubiquitin ligase complex that targets MYCN for proteasomal degradation (42, 43). As expected, we found that FBXW7 overexpression markedly reduced MYCN protein levels in MYCN-amplified cells (Fig. 5G; Supplementary Fig. S4E), leading to decreased protein levels of SGOC enzymes (Supplementary Fig. S4E). ATF4 overexpression was sufficient to neutralize the effect of FBXW7, leading to increased MYCN protein expression (Fig. 5G) and reduced MYCN protein ubiquitination (Fig. 5H). Conversely, FBXW7 depletion, which had no effect on MYCN mRNA expression (Supplementary Fig. S4F), significantly increased MYCN protein expression, leading to higher PHGDH expression (Fig. 5I). Moreover, when FBXW7 expression was silenced, ATF4 knockdown could no longer increase MYCN protein ubiquitination (Fig. 5J). Thus, both overexpression and knockdown studies suggest that ATF4 stabilizes MYCN protein by antagonizing FBXW7-mediated MYCN ubiquitination and subsequent degradation. These findings, along with those showing that MYCN transcriptionally activates ATF4 expression (Fig. 4B and C), provide evidence for a MYCN-ATF4 positive feedback loop for transcriptional activation of the SGOC metabolic pathway in MYCN-amplified cells.

The observation that the SGOC genes are transcriptionally activated in MYCN-amplified relative to non-MYCN–amplified neuroblastoma cells led us to hypothesize that inhibition of this metabolic pathway might be synthetically lethal with MYCN amplification. We tested this hypothesis in a panel of neuroblastoma cell lines with MYCN amplification [BE(2)-C, LA1-55n, SK-N-DZ, and SMS-KCNR] or without (NBL-S, SHEP1, SH-SY5Y, SK-N-AS, and SK-N-FI). Of note, BE(2)-C and SMS-KCNR cells were derived from relapsed tumors after chemotherapy (44) and, as expected, both cell lines were highly resistant to the chemotherapy drug doxorubicin in comparison with the non-MYCN–amplified cell lines SK-N-AS, SK-N-FI, and SH-SY5Y and the MYCN-amplified cell lines LA1-55n and SK-N-DZ (Fig. 6A).

We treated these cell lines with NCT-503, a small molecule inhibitor of PHGDH (45). NCT-503 (EC₅₀ = 8–16 μmol/L) at 10 to 20 μmol/L markedly inhibited the growth and survival of MYCN-amplified cell lines, including the chemoresistant BE(2)-C and SMS-KCNR cells, but had no significant or only a modest effect on non-MYCN–amplified cell lines (Fig. 6B and C; Supplementary Fig. S5A). This effect of NCT-503 was neither a result of nonspecific inhibition of MYCN expression (Supplementary Fig. S5B) nor dependent on the proliferation state since serum deprivation, which markedly inhibited the proliferation of both non-MYCN- and MYCN-amplified cell lines (Supplementary Fig. S1C), had no significant effect on the differential sensitivities of non-MYCN- and MYCN-amplified cell lines to NCT-503 (Supplementary Fig. S5C). Treatment with NCT-503 negative control compound showed no significant effect on all the cell lines examined.

Consistent with the previous report (45), we observed the growth-inhibiting effect of NCT-503 in the presence (Fig. 6B) or absence (Fig. 6C) of supplemental serine. In addition, non-MYCN–amplified (resistant) SK-N-AS cells and MYCN-amplified (sensitive) BE(2)-C cells showed no difference in their ability to uptake glycine, whereas BE(2)-C cells displayed a slightly higher activity in transporting serine (Supplementary Fig. S5D). Moreover, NCT-503 treatment modestly inhibited glycine and serine transport of SK-N-AS cells but had no effect on BE(2)-C cells (Supplementary Fig. S5E). Together, these findings suggest that serine and glycine transport is not a major factor in determining the sensitivity of non-MYCN- and MYCN-amplified cell lines to NCT-503.

We confirmed the sensitivity of MYCN-amplified neuroblastoma cells to PHGDH inhibition in parallel studies with another small molecule inhibitor of PHGDH, CBR-5884 (46). Compared with non-MYCN–amplified cell lines, MYCN-amplified cell lines were also significantly more sensitive to 15 to 30 μmol/L CBR-5884 (Fig. 6D). In addition, MYCN-amplified BE(2)-C and SMS-KCNR cells were significantly more sensitive to PHGDH knockdown than non-MYCN–amplified SHEP1 and SK-N-FI cells (Supplementary Fig. S5F and S5G). Importantly, MYCN overexpression was sufficient to sensitize non-MYCN–amplified SHEP1 and NBL-S cells to NCT-503 (Fig. 6E; Supplementary Fig. S5H).

We next evaluated NCT-503 antitumor activity in mouse xenograft models established with MYCN-amplified BE(2)-C cells and non-MYCN–amplified SK-N-AS cells. Tumor-bearing mice were treated with vehicle or 64 mg/kg (body weight) of NCT-503 by daily intraperitoneal injection for 10 days or until sacrifice. NCT-503 treatment significantly reduced the growth MYCN-amplified BE(2)-C xenografts [Fig. 6F, BE(2)-C], and prolonged the survival of tumor-bearing mice [Fig. 6G, BE(2)-C]. By contrast, NCT-503 treatment of mice bearing non-MYCN–amplified SK-N-AS xenografts showed no significant effect on tumor growth and mouse survival (Fig. 6F and G, SK-N-AS). In all cases, NCT-503 treatment at the indicated dose and time had no significant effect on mouse body weight (Supplementary Fig. S5I).

Collectively, these findings suggest that increased MYCN expression in neuroblastoma cells imposes a dependence on the SGOC metabolic pathway for survival and proliferation.

The SGOC pathway links glycolysis to the production of serine, glycine, and one-carbon units. To elucidate the metabolic basis for the differential sensitivities of MYCN-amplified and non-MYCN–amplified cells to NCT-503, we performed stable isotope flux analysis of glucose carbon into serine and glycine synthesis (Fig. 7A). NCT-503 treatment significantly reduced the production of m+3 serine and m+2 glycine from [U-¹³C]glucose in MYCN-amplified BE(2)-C cells (Fig. 7B and C, DMSO vs. NCT-503). By contrast, no such an effect was observed in non-MYCN–amplified SK-N-AS cells (Fig. 7B and C). Importantly, the proportions of serine and glycine derived from [U-¹³C]glucose (m+3/m+0 serine and m+2/m+0 glycine) were significantly higher in BE(2)-C cells relative to SK-N-AS cells (Fig. 7B and C, BE(2)-C vs. AS). Taken together, these data indicate that MYCN-amplified cells rely more on glucose-derived carbon for serine and glycine synthesis in comparison with non-MYCN–amplified cells, suggesting a metabolic basis for their differential sensitivities to PHGDH inhibition.

To gain a molecular understanding for the selective cytotoxic effect of NCT-503 on MYCN-amplified cells, we performed microarray gene expression profiling of BE(2)-C cells treated with DMSO or 10 μmol/L NCT-503. A total of 3,655 NCT-503–responsive genes (≥±1.50 fold, P < 0.05) were identified, with 1,223 genes being upregulated and 2,432 genes downregulated (Supplementary Table S1). GO analysis revealed that genes downregulated by NCT-503 treatment were highly enriched for GO terms associated with DNA replication and G₁- to S-phase transition (Supplementary Fig. S6A; Supplementary Table S2). We obtained essentially the same result with gene set enrichment analysis (GSEA), which showed downregulation of a large number of genes involved in DNA replication and G₁–S cell-cycle transition (Supplementary Fig. S6B). As reported previously (45), NCT-503 treatment induced G₁ arrest (Supplementary Fig. S6C), consistent with a key role of the SGOC pathway in supplying nucleotides for DNA replication.

For genes upregulated by NCT-503 treatment, GO analysis showed a significant enrichment for GO terms associated with various stress responses, including glucose starvation, endoplasmic reticulum stress, and the unfolded protein response, and oxidative stress (Fig. 7D; Supplementary Table S2). Similarly, GSEA revealed that genes involved in the amino acid deprivation response were most significantly enriched and upregulated (Fig. 7E; Supplementary Fig. S6D). These stress-responsive genes, including ASNS, ASS1, ATF3, CTH, DDIT3, TRIB3, and XBP1 (Supplementary Fig. S6D; Supplementary Tables S1 and S2), are induced by ATF4 in response to various stress signals (19). In addition, several genes within the SGOC pathway were induced by NCT-503, including PHGDH, PSAT1, SHMT2, and MTHFD2 (Supplementary Table S1). These data suggest that in MYCN-amplified cells, NCT-503 treatment elicited an adaptive response to metabolic stress induced by inhibition of SGOC synthesis.

We verified the microarray data by qRT-PCR and immunoblotting, which showed that NCT-503 treatment markedly induced mRNA and protein expression of ATF4, DDIT3 (also known as CHOP) and/or TRIB3 in MYCN-amplified, but not in non-MYCN–amplified, cell lines (Fig. 7F and G; Supplementary S6E). Overexpression of MYCN, which sensitized non-MYCN–amplified cells to NCT-503 (Fig. 6E) was sufficient to confer NCT-503 induction of ATF4, DDIT3, and TRIB3 in non-MYCN–amplified SHEP1 cells (Fig. 7F). Moreover, NCT-503 treatment induced autophagy in MYCN-amplified cells, but not in non-MYCN–amplified cells, as evidenced by increased LC3B-II production and the formation of LC3B-positive puncta (Fig. 7H and I; Supplementary Fig. S6F). Together, these findings suggest that PHGDH inhibition induces metabolic stress in MYCN-amplified neuroblastoma cells, leading to G₁ arrest and autophagy.

Our study reveals a MYCN-dependent vulnerability to suppression of the SGOC metabolic pathway. PHGDH inhibition triggered cell-cycle arrest and a metabolic stress response with autophagy in MYCN-amplified neuroblastoma cells, but not in non-MYCN–amplified cells. This increased sensitivity to PHGDH inhibition can be recapitulated by overexpression of MYCN in non-MYCN–amplified cells. Mechanistically, we show that in comparison with non-MYCN–amplified cells, MYCN-amplified cells, likely owing to their higher expression of PHGDH and other SGOC pathway enzymes, have an increased dependence on glucose-derived carbon for the synthesis of serine and glycine (and possibly also one-carbon units as suggested previously; ref. 45). In addition, a recent study showed that PHGDH inhibition in cells with high PHGDH expression could disrupt central carbon metabolism (47), which may help to explain the failure of supplemental serine to rescue the cytotoxic effect of NCT-503. Collectively, these findings suggest a metabolic basis for the differential sensitivities of MYCN-amplified and non-MYCN–amplified neuroblastoma cells to PHGDH inhibitors.

MYCN amplification has a major role in the development of high-risk neuroblastoma, a leading cause of pediatric cancer deaths (2, 26–28). However, targeting MYCN directly for therapy has proved to be challenging (16). An alternative strategy is to target metabolic pathways that are activated by MYCN for sustaining cancer cell growth and proliferation. In support of this idea, we found that blocking SGOC metabolism selectively inhibits the proliferation and survival of MYCN-amplified neuroblastoma cells. Also, a recent study reported that MYCN-amplified neuroblastomas display increased folate dependence and are more sensitive to methotrexate, an inhibitor of dihydrofolate reductase that functions to maintain the folate-one-carbon loop for thymidylate synthesis (48). Together, these findings suggest that targeting MYCN-activated metabolic pathways could be a selective therapeutic strategy against MYCN-amplified tumors. In addition, our findings have implications in treatment of patients with relapsed neuroblastomas carrying MYCN amplification. Although a majority of patients with high-risk neuroblastoma respond to initial therapy, approximately half of them will eventually have a relapse with resistant to further therapies (27). Currently, there are no effective treatment regimens for these patients. We found that PHGDH inhibitors are effective in inhibiting the growth of chemoresistant BE(2)-C and SMS-KCNR cells and of BE(2)-C xenograft tumors. Both BE(2)-C and SMS-KCNR cell lines were derived originally from relapsed tumors after chemotherapy (44). Thus, blocking SGOC metabolism might provide a treatment strategy for relapsed neuroblastomas with MYCN amplification.

ATF4 is a master regulator of amino acid metabolism and stress responses (17–19). We found that MYCN binds to the promoter of ATF4 to increase its mRNA expression. In addition, we show that MYCN-mediated transcriptional activation of SGOC pathway enzymes requires ATF4. Interestingly, a recent study reported that both MYCN and ATF4 can directly activate the transcription of ASCT2 (also known as SLC1A5), which encodes a glutamine transporter. It also reported a positive correlation in mRNA expression among MYCN, ATF4, and ASCT2 in high-risk neuroblastomas, suggesting a potential coordination between MYCN and ATF4 in regulation of ASCT2 expression (49). These findings, along with ours, provide evidence for MYCN and ATF4 to regulate a common set of genes in metabolic reprogramming in neuroblastoma. Our study further uncovers a MYCN–ATF4 positive feedback loop: MYCN transcriptionally activates ATF4 expression, and ATF4 stabilizes MYCN protein by antagonizing FBXW7-mediated MYCN ubiquitination and degradation. This positive feedback loop may be hardwired into MYCN-amplified cancer cells to reinforce the hyperactive state of SGOC gene transcription and to boost production outputs of the metabolic pathway required by the MYCN-driven growth program. The ability of ATF4 to stabilize MYCN protein also suggests a molecular mechanism for the observation that ATF4 overexpression significantly enhanced the proliferation of MYCN-amplified neuroblastoma cells but had no effect on the proliferation of non-MYCN–amplified neuroblastoma cells.

Given the key role of ATF4 as a master regulator of stress responses, our identification of the MYCN–ATF4 positive feedback loop suggests a molecular mechanism for the stress-sensitive phenotype of MYCN-amplified cancer cells. The feedback loop transcriptionally amplifies SGOC biosynthesis to sustain cell growth and proliferation, but, at the same time, confers to these cells an increased dependence on the SGOC pathway and an intensified stress response to PHGDH inhibition. This is reminiscent of the earlier findings that some oncogenes, such as MYC and E2F1, simultaneously promote cell proliferation and sensitize cells to apoptosis (50). Thus, there might be a coupled relationship between metabolic reprogramming and sensitization to metabolic stress. The notion that cancer cells reprogram their metabolism to meet the biosynthetic challenge of growth and proliferation but, as a result, are inherently sensitive to metabolic stress may offer numerous opportunities for manipulating cell metabolism in the direction of cancer therapy.

No potential conflicts of interest were disclosed.

Conception and design: Y. Xia, J. Ding, P.D. Prasad, Y. Zha, H.-F. Ding

Development of methodology: J. Ding, P.D. Prasad, Z.-C. Ding

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Xia, B. Ye, J. Ding, Y. Yu, A. Alptekin, B. Gao, O. Fiehn, C. Yan, H.-F. Ding

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Thangaraju, P.D. Prasad, E.J. Park, J.-H. Choi, B. Gao, Z. Dong, Y. Zha, H.-F. Ding

Writing, review, and/or revision of the manuscript: J. Ding, Z. Dong, Y. Zha, H.-F. Ding

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B. Ye, Z.-C. Ding

Study supervision: Y. Zha, H.-F. Ding

We thank Susan L. Cohn (University of Chicago), M. Celeste Simon (University of Pennsylvania), and C. Patrick Reynolds (Children's Oncology Group, Texas Tech University) for providing cell lines, and Rhea-Beth Markowitz for manuscript editing. Z. Dong is a VA senior Research Career scientist. The work was supported by a grant from NIH to H.-F. Ding (Grant No. R01 CA190429) and by grants from the National Natural Science Foundation of China to Y. Xia (Grant No. 81702486) and to Y. Zha (Grant Nos. 81201981 and 81550031).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.