EWS/FLI is a Master Regulator of Metabolic Reprogramming in Ewing Sarcoma

This article is featured in Highlights of This Issue, p. 1467

Ewing sarcoma is an aggressive malignancy that most often emerges from bone and bone-associated tissues of children and adolescents. It is the second most common pediatric bone tumor; second only to osteosarcoma. Surgical strategies have achieved very good outcomes for many of these patients. But truly effective molecular therapies continue to evade discovery, translating to especially poor prognosis for patients who fall victim to metastasis or relapse. Indeed, the rate of relapse after surgery is approximately 35%, and the probability of 5-year survival after relapse is approximately 20% (1, 2). Continued investigation of the molecular underpinnings of this disease are thus crucial, and have the promise of providing additional candidate therapies to these patients.

Ewing sarcoma tumors are uniquely devoid of recurrent mutations in classical oncogenes or tumor suppressors (3). Instead, oncogenesis of Ewing sarcoma is almost entirely due to the far-reaching consequences of a single genetic lesion: a translocation event between chromosomes 11 and 22. This chromosomal rearrangement generates an in-frame fusion product merging the EWSR1 and FLI1 genes, giving rise to the oncogenic fusion protein known as EWS/FLI (EF). EF contains the N-terminal transcriptional regulatory domain of EWS and the DNA-binding domain of FLI1. Interestingly, while EF can affect known targets of FLI1 by binding to canonical FLI1-binding sites, it also influences expression of genes that are not targets of FLI1 or EWSR1 (4). Thus, EF is an aberrant transcription factor with neomorphic functions and tumorigenic consequences.

The transcriptional reprogramming driven by EF has been well profiled (5). In-depth study of EF targets has revealed that this master regulator exerts its oncogenic effects on numerous processes, contributing to many of the hallmark capabilities of cancer cells. These include: proliferation and growth (6), resistance to cell death (7), immortality (8), angiogenesis (9), invasion, and metastasis (7), and evading immune destruction (10). However, very little is known about how EF affects another characteristic feature of cancer cells—the pro-oncogenic reprogramming of cellular metabolism.

Cancer cells characteristically adopt programs that modify substrate utilization and alter metabolite abundance in ways that support the biosynthetic requirements of these highly proliferative and resilient cells (11). Evidence increasingly indicates that metabolic alterations observed in cancer are not mere spectator phenomena, but are active contributors to oncogenesis (11). Significant interrelationships exist linking cellular metabolic processes to nuclear control, and vice versa (11). As a result, defining the metabolic processes undergirding cancer cells provides promising new opportunities for discovery of potentially actionable therapeutic targets.

Currently, there is a paucity of data concerning the role of EF in altering metabolic processes in Ewing sarcoma. Given its relatively low mutation burden compared with other cancers (3), Ewing sarcoma provides a unique model to study metabolic alterations that are important to oncogenesis. Changes in cellular metabolism that result from EF transcriptional misregulation likely contribute directly to oncogenesis. Thus, studying cellular metabolism in Ewing sarcoma promises to increase our understanding of metabolic reprogramming in general, and inform development of novel therapeutic strategies for Ewing sarcoma specifically. With these goals in mind, we set out to investigate EF-driven metabolic reprogramming in Ewing sarcoma.

Cell lines expressing EWS/FLI (A673, UTES-14-01872; see Supplementary Fig. S2B and S2C) or EWS/ERG (TTC466) were grown in 10%FBS/DMEM or 10% FBS/RPMI (RPMI 1640 Glutamax; Thermo Fisher Scientific), respectively. DMEM (Corning) contained 25 mmol/L glucose, 4 mmol/L l-glutamine, and 1 mmol/L sodium pyruvate. Media were supplemented with 3 μg/mL puromycin for selection of shRNA-expressing cells. 293T cells were grown in 10% FBS/DMEM. Mycoplasma testing was performed as described previously (12). Population doublings were assessed using a 3T5 growth assay as described previously (13). A673 and TTC466 lines were validated prior to experimentation by STR analysis (Genetica DNA Laboratories). The patient-derived cell lines A673 and TTC466 were acquired from stocks maintained by S.L. Lessnick. UTES-14-01872 is a recently derived cell line of Ewing sarcoma established in 2014 by J. Schiffman.

EWS/FLI- and Luciferase-siRNA sequences were used previously for transcriptional profiling of Ewing sarcoma (14–16). siRNA sequences for PHGDH and PSPH were designed using the Whitehead Institute siRNA Selection Program (17). shRNA-encoding inserts containing these sequences were ligated into the pMKO.1p retroviral vector. siRNA sequences were: shEF, 5′-ATAGAGGTGGGAAGCTTAT-3′; shPHGDH, 5′-CCCTGTAGTACAGCAATAA-3′; shPSPH, 5′-GGCTGAAATTCTACTTTAA-3′. Retrovirus was produced by transfecting 293T cells with OptiMEM and Lipofectamine 2000, and viral supernatant was collected at 48 and 72 hours postinfection. Viral supernatant was filtered at 0.45 μm, aliquoted, and stored at −80° C. Infections were performed by adding viral supernatant (plus 8 μg/mL polybrene) to subconfluent cultures. Selection in puromycin-containing media was continued for 3 days, and puromycin efficacy was confirmed by observing cell death in uninfected cells. As EF is the primary oncogenic driver of Ewing sarcoma, most Ewing sarcoma cell lines do not tolerate its silencing well. A673 cells are known to best tolerate EF knockdown. Accordingly, we have had the most success using A673 cells, and limited success performing such experiments in other lines.

A673 cells expressing shLUC or shEF were seeded onto 1.5 borosilicate chambered coverglass slides (Thermo Fisher Scientific) and incubated overnight. Media were replaced with phenol red-free DMEM containing 25 mmol/L glucose, 1% Glutamax, 1% sodium pyruvate, 100 nmol/L Mitotracker Red FM, 200 nmol/L Mitotracker Green FM, and 5 μg/mL Hoechst stain (Thermo Fisher Scientific). Cells were incubated in staining medium for 10 minutes, after which staining medium was removed and cells were washed twice with phenol red-free medium. Cells were immediately imaged using a Zeiss Axio Observer Z1 microscope.

Images were quantified using ImageJ software. Three cells were selected in each of 5 high-power field images for a total of 15 quantified cells. Integrated density measurements were taken for Mitotracker Red and Mitotracker Green channels for each cell with identical areas, and corrected total cell fluorescence (CTCF) was calculated for each channel as described previously (18). CTCF calculations were done as: CTCF = integrated density – (area of selected cell × mean fluorescence of background readings).

Whole-cell extract (WCE) was prepared using RIPA buffer containing protease inhibitor cocktail (Sigma). Total protein content was determined via BCA assay, and WCE was prepared for SDS-PAGE by adding Laemmli buffer. Immunoblotting was done using the following antibodies: Fli1 (Abcam, 1:1,000), PHGDH (Cell Signaling Technology 1:1,000), PSAT (Abcam, 1:1,000), PSPH (Abcam, 1:1,000), SHMT1 (Cell Signaling Technology, 1:1,000), SHMT2 (Cell Signaling Technology, 1:1,000), TYMS (Cell Signaling Technology, 1:1,000), DHFR (Abcam, 1:1,000), HK1 (Abcam, 1:1,000), H3K9me3 (Abcam, 1:1,000), H3K27me3 (Cell Signaling Technology, 1:1,000), Histone H3 (Abcam, 1:5,000), α-Tubulin (Cell Signaling Technology, 1:5,000), β-actin (Sigma, 1:5,000). Densitometry was performed using ImageJ software as described previously (19).

Procedures for IHC are described previously (20–22). Briefly, paraffin was removed and tissue was subjected to high pressure antigen retrieval (Vector Unmasking Solution; Vector Laboratories). Samples were incubated in primary antibody (PHGDH, Cell Signaling Technology) overnight at 4°C. Secondary antibodies, raised in donkey (Jackson ImmunoResearch), were used at a dilution of 1:250 at room temperature for 1 hour. Vectastain reagents and diaminobenzidine (DAB; Vector Laboratories) were used to develop IHC. Images were subsequently taken on a Zeiss Axio Observer Z1 microscope. Tissue samples from human patients were obtained with patient consent and approval from the University of Utah Institutional Review Board.

The PHGDH inhibitors NCT502 and NCT503 (Cayman Chemical) and the inactive control molecule called PHGDH-inactive (Cayman Chemical) were reconstituted in DMSO. Final concentrations of inhibitors were 3.7 μmol/L, 18.5 μmol/L, and 37 μmol/L for PHGDH-inactive and NCT502; and 2.5 μmol/L, 12.5 μmol/L and 25 μmol/L for NCT503. These concentrations were chosen to achieve maximal levels of inhibition at the highest doses based on previously published IC₅₀ data (23).

Global 5-methylcytosine levels were assessed using the Global DNA Methylation Assay – LINE1 (Active Motif). D-2-hydroxyglutarate was quantified using a colorimetric D2HG assay kit (BioVision). All procedures were carried out as indicated by the manufacturer.

Cells were seeded into 96-well format Seahorse analyzer plates 24 hours prior to performing Seahorse experiments. We performed side-by-side Mito Stress Tests and a Glycolysis Stress Tests in 96-well format using the XF96e analyzer. The Mito Stress Test was performed in standard assay media (DMEM, 25 mmol/L Glucose, 1 mmol/L pyruvate, 2 mmol/L Glutamine, pH 7.4) with the final concentrations of drugs as follows [Oligomycin: 2 μmol/L; carbonyl cyanide-4 (trifluoromethoxy) phenylhydrazone (FCCP): 2 μmol/L; Rotenone: 0.5 μmol/L; Antimycin A: 0.5 μmol/L]. Assay protocol was standard (3 measurements per phase, acute injection followed by 3-minute mixing, 0-minute waiting, and 3-minute measuring). Data were normalized to cell number because morphologic changes that result from EF silencing potentially confound normalization to total cellular protein (Supplementary Fig. S1D; ref. 24). The Glycolysis Stress test was performed in standard assay media (DMEM, 2 mmol/L glutamine, pH 7.4) and the following final concentrations (glucose: 10 mmol/L; oligomycin: 2 μmol/L; 2-deoxy-glucose: 50 mmol/L). Results were normalized as above and analyzed in WAVE software and processed through the XF Mito Stress Test Report and Glycolysis Stress Test Generators. All statistics were generated through GraphPad Prism 7 using unpaired two-tailed Student t tests.

Gene ontology analysis was performed on lists of differentially expressed genes from RNA-seq data (Lg2Rto: 0.585; P_adj ≥ 10). Gene lists and data are available in Supplementary Data. Methodology of the PANTHER gene ontology platform and gene set enrichment analysis (GSEA) are described elsewhere (25, 26). Ewing sarcoma patient survival data were compared with gene expression by analyzing a publicly available dataset (GSE17679) with the SurvExpress online tool (27, 28). Risk groups were defined by prognostic index, with low-risk being defined as less than the median prognostic index, and high-risk being defined as greater than the median prognostic index.

RNA-Seq of A673 cells treated with shLUC or shEF was previously performed (16). The datasets we acquired are publicly available in the NCBI BioProject database, accession # PRJNA176544, and the Sequence Read Archive (SRA059329). We reanalyzed these data to determine differential expression analysis using USeq (useq.sourceforge.net) version 8.8.3. Lists of differentially expressed genes are provided in Supplementary Data.

All GC-MS analysis was performed with a Waters GCT Premier mass spectrometer fitted with an Agilent 6890 gas chromatograph and a Gerstel MPS2 autosampler. Dried samples were suspended in 40 μL of a 40 mg/mL O-methoxylamine hydrochloride (MOX) in pyridine and incubated for one hour at 30°C. To autosampler vials was added 25 μL of this solution. Forty microliters of N-methyl-N-trimethylsilyltrifluoracetamide (MSTFA) was added automatically via the autosampler and incubated for 60 minutes at 37°C with shaking. After incubation 3 μL of a fatty acid methyl ester standard (FAMES) solution was added via the autosampler then 1 μL of the prepared sample was injected to the gas chromatograph inlet in the split mode with the inlet temperature held at 250°C. A 10:1 split ratio was used for analysis. The gas chromatograph had an initial temperature of 95°C for one minute followed by a 40°C/minute ramp to 110°C and a hold time of 2 minutes. This was followed by a second 5°C/minute ramp to 250°C, a third ramp to 350°C, then a final hold time of 3 minutes. A 30-m Phenomex ZB5-5 MSi column with a 5-m long guard column was employed for chromatographic separation. Helium was used as the carrier gas at 1 mL/minute. Because of the high amounts of several metabolites, the samples were analyzed once more at a 10-fold dilution. Data were collected using MassLynx 4.1 software (Waters). Metabolites were identified and their peak area was recorded using QuanLynx. This data was transferred to an Excel spread sheet (Microsoft). Metabolite identity was established using a combination of an in-house metabolite library developed using pure purchased standards and the commercially available NIST library. Metabolomic data were further analyzed using MetaboAnalyst 3.0 (29).

Transcription profiling of Ewing sarcoma cells has been previously performed by microarray and RNA-seq experiments after silencing EWS/FLI (EF) in A673 cells (14, 16, 30). We analyzed data from a publicly available RNA-seq dataset wherein EF was silenced by shRNA (shEF) and compared with control knockdown (shLUC). Our analysis identified 7,094 genes that are differentially expressed in shEF versus control shLUC cells (gene lists found as Excel files in Supplementary Data). We defined significant changes in expression as a fold change of ≥1.5-fold and a P < 0.05. After EF knockdown, 4,140 genes exhibited increased expression, while 2,954 genes were less expressed. Gene ontology (GO) analysis was performed on the list of all EF-misregulated genes, revealing metabolic processes (GO: 0008152) as the second most enriched GO term (2,058 genes; 33.7%; Fig. 1A). These genes were further subdivided into primary metabolic processes (GO: 0044238), revealing that the majority of these genes were involved in nucleobase-containing compound metabolism (GO: 0006139) and protein metabolism (GO: 0019538; Fig. 1B).

To further assess whether EF significantly alters genes of metabolic importance, gene set enrichment analysis (GSEA) was performed comparing our dataset of differentially expressed genes in shEF cells to gene sets from the Molecular Signatures Database (MSigDB; ref. 25). Significant correlations were observed between differentially expressed genes and genes in the Hallmark-Glycolysis gene set and the KEGG-Pathways in Cancer gene set. (Fig. 1C and D). Significant correlations were also observed with gene sets of Hallmark-Reactive Oxygen Species, Hallmark-Epithelial-Mesenchymal Transition, but not in the Hallmark-Oxidative Phosphorylation gene set (Supplementary Fig. S1A).

To more directly study whether EF alters cellular metabolism, we performed stress tests on glycolysis and mitochondrial function using the Seahorse XF96 analysis system. During the glycolysis stress test, addition of glucose and oligomycin caused greater increases of extracellular acidification rate (ECAR) in shEF versus shLUC cells (Fig. 2A). Oxygen consumption rate was not different between shEF and shLUC cells throughout the glycolysis stress test (Fig. 2B). These data suggest that silencing of EF results in increased utilization of glucose by glycolysis (Table 1; Supplementary Fig. S1B). Basal ECAR was not different between shEF and shLUC cells at the beginning of the glycolysis stress test, reflecting the lower glucose (10 mmol/L) conditions required for a glycolysis stress test (Fig. 2B). However, in the mitochondrial stress test glucose is present at typical cell culture levels (25 mmol/L), allowing us to observe that basal ECAR is higher in shEF versus shLUC cells (Fig. 2C). The apparent increase of glycolytic rate caused by EF knockdown is surprising, and is possibly explained by derepression of hexokinase 1 (HK1) by shEF (Fig. 2E). Indeed, uptake of 2-deoxy-2-[(7-nitro-2,1,3-benzoxadiazol-4-yl)amino]-D-glucose (2-NBDG; a fluorescently labeled analogue of deoxyglucose) is elevated in shEF versus shLUC A673 cells (Fig. 2F).

Tests of mitochondrial respiration showed enhanced response to mitochondrial uncoupling by carbonyl cyanide-4 (trifluoromethoxy) phenylhydrazone (FCCP) in shEF versus shLUC cells (Fig. 2D). These data indicate that silencing EF results in increased maximal respiration and spare respiratory capacity, with no change in basal respiration, ATP production, proton leak, nonmitochondrial respiration, and coupling efficiency (Table 1; Supplementary Fig. S1B). To additionally assess mitochondrial respiratory function, cells were treated with the membrane potential-sensitive Mitotracker Red FM dye. Comparisons were made relative to a membrane potential-independent mitochondrial dye (Mitotracker Green). Consistent with increased maximal respiration and respiratory capacity, shEF cells also had elevated mitochondrial membrane potential compared with shLUC cells (Fig. 2G). Together, these data suggest that EF increases the proportion of glucose-derived carbon that is shunted away from oxidative metabolism consistent with a biosynthetic, pro-oncogenic metabolic program.

Global changes in metabolite abundance were assessed by performing steady-state metabolomics analyses comparing shEF with shLUC A673 cells. Examining metabolomic data by principal components analysis revealed that EF knockdown results in a distinct metabolic profile compared with control cells (Fig. 3A and B). Topologic analysis of metabolites within metabolic pathways highlighted several metabolic pathways that were more significantly impacted by silencing of EF (Fig. 3C; ref. 31). Pathways significantly affected by EF silencing included alanine, aspartate and glutamine metabolism, and glycine, serine, and threonine metabolism.

In addition, metabolomic profiling data were combined with transcription profiling data to perform analyses of pathways in which both metabolite abundance and gene expression were significantly altered in shEF versus shLUC cells. This allowed us to determine which pathways were impacted the most by shEF, potentially by altering expression of enzymes and metabolites in that pathway. Intriguingly, the enzymes of serine and glycine synthesis were found to be significantly downregulated by EF silencing in the RNA-seq dataset, and corresponding decreases in serine and glycine were observed in our metabolomic dataset (Fig. 3C; Supplementary Fig. S2A).

The de novo serine synthesis pathway is one way by which glycolytic intermediates can be shunted toward biosynthetic processes, and this has been implicated in many cancers (32, 33). Specifically, phosphoglycerate dehydrogenase (PHGDH) commits the glycolytic intermediate 3-phosphoglycerate to serine and glycine synthesis, providing important precursors for protein, lipid, and nucleic acid synthesis. Increased PHGDH has been linked to the oncogenic biology of other tumors (32, 34). In fact, PHGDH expression is higher in Ewing sarcoma than all other cancer cell lines in the Cancer Cell Line Encyclopedia (Supplementary Fig. S3A). In concordance with RNA-seq data, protein levels of PHGDH were decreased in shEF versus shLUC cells (Fig. 4A). Protein levels of phosphoserine aminotransferase (PSAT) and phosphoserine phosphatase (PSPH), the two enzymes downstream of PHGDH in the serine synthesis pathway, were also lower upon silencing of EF (Fig. 4A). Similar patterns of expression were observed for serine hyrdroxymethyltransferase (SHMT1 and SHMT2), which converts serine to glycine and feeds into 1-carbon folate metabolism cycles, as well as dihydrofolate reductase (DHFR) and thymidylate synthetase (TYMS)—important enzymes in 1-carbon folate metabolism cycles (Fig. 4B; Supplementary Fig. S2D and S2E).

To explore correlations between PHGDH expression and clinical outcome, we used the SurvExpress tool to perform Cox survival analysis of a Ewing sarcoma dataset (27, 28). Patients who were categorized as high-risk had significantly higher expression of PHGDH and SHMT1/2 (Fig. 4C and D). Similarly, higher expression of PSPH, DHFR, TYMS, MTHFD2, and SHMT2 was correlated with poorer survival in Ewing sarcoma patients (Supplementary Fig. S3B and S3C). Similar correlations were also observed between expression levels and event-free survival (data not shown).

Furthermore, PHGDH expression was assessed in histologic samples from Ewing sarcoma patients. PHGDH staining was much higher in tumor (defined by IHC for CD99) than in adjacent normal tissue (Fig. 5A). CD99 exhibited staining along cell surface boundaries, while PHGDH staining appeared diffusely cytoplasmic.

Inhibition of PHGDH function by NCT502 and NCT503 caused growth defects and cell death in Ewing sarcoma cell lines, including the EWS/ERG-harboring TTC466 line (Fig. 5B–E). This effect was less prominent in 293T or U2OS cells (Fig. 5F and G). This and previous data indicate that upregulated expression of serine synthesis and 1-carbon metabolism contributes importantly to the oncogenic phenotype in Ewing sarcoma (Fig. 5H).

While the major product of PHGDH is 3-phosphohydroxypyruvate (3-PP) from 3-phosphoglycerate (3-PG), the enzyme also produces relatively small amounts of 2-hydroxyglutarate (2-HG) from α-ketoglutarate (Fig. 6A; ref. 35). 2-HG has been shown to be oncogenic in high amounts, typically resulting from neomorphic mutations of isocitrate dehydrogenase enzymes (36). Interestingly, we observed that 2-HG was lower in shEF versus shLUC cells (Fig. 6B), suggesting that EF might be driving elevated production of 2-HG via increased expression of PHGDH.

To test whether elevated levels of 2-HG in Ewing sarcoma cells was due to PHGDH, we assessed 2-HG levels after inhibition of PHGDH function with NCT502. 2-HG was lower in cells treated by NCT502 versus its inactive analogue (Fig. 6C). Importantly, the levels of 2-HG in Ewing sarcoma cells was lower than previously observed in cells harboring mutations in isocitrate dehydrogenase (IDH) enzymes (37). To test whether 2-HG generated from PHGDH had effects on DNA methylation, we silenced PHGDH expression and determined 5-methylcytosine (5-mC) content of LINE-1 elements (a surrogate for bulk DNA methylation). However, knockdown of EF or PHGDH had no effect on 5-mC levels (Fig. 6D). PSPH was also silenced to distinguish between serine synthesis and 2-HG–producing functions of PHGDH, and disruption of serine synthesis downstream of PHGDH also had no effect on 5-mC levels.

To further assess whether modulation of PHGDH affects chromatin, we immunoblotted histone proteins. This was performed with A673 cells, as well as another more recently established patient-derived EF-harboring Ewing sarcoma cell line (UTES-14-01872; Supplementary Fig. S2C). Interestingly, methylation of histone H3 at lysine (K) 9 and K27 was increased by PHGDH knockdown in UTES-14-01872 and TTC466 cells, but not in A673 cells (Fig. 6E and F). In addition, total histone protein was increased by PHGDH knockdown in two of the three cell lines. These data raise the possibility of a role for PHGDH in chromatin regulation, possibly by affecting abundance of metabolites used as cofactors for histone-modifying enzymes.

EF has long been recognized as the primary oncogenic driver of Ewing sarcoma. Overexpression of EF in NIH3T3 murine fibroblasts results in oncogenic transformation, and silencing EF expression in patient-derived cell lines results in a loss of tumorigenesis (5, 38, 39). However, effective therapeutic strategies directly targeting transcription factors have been notoriously elusive. For this reason, studies of the biology of Ewing sarcoma have primarily focused on identifying important EF target genes, as well as the interacting partners EF requires to elicit its effects. Indeed, numerous EF targets have been shown to be necessary, but not sufficient, for oncogenesis in these tumors (6, 14, 40, 41). Unfortunately, many of these critical EF targets and partners lack enzymatic function and are themselves not susceptible to pharmacologic modulation. With this in mind, we sought to determine whether EF generates an oncogenic program of cellular metabolism by misregulating the transcription of genes encoding key metabolic enzymes.

By analyzing existing EF transcriptional datasets, we were able to identify significant patterns of altered metabolic gene expression (Fig. 1). Specifically, gene ontology analysis revealed significant alterations in the expression of genes involved in glycolysis, nucleobase-containing metabolic processes, and protein metabolism. Consistent with a cancerous phenotype, Ewing sarcoma cells are hyperproliferative and require increased macromolecule synthesis to sustain accelerated growth and proliferation. In addition, EF-driven alterations in nucleobase-containing metabolic processes such as (NAD salvage) could be sensitive to inhibition and therapeutic exploitation (42). The increased metabolic demands of rapid proliferation and growth could be exploited in similar ways.

We also observed that EF knockdown resulted in increased maximal respiration and mitochondrial membrane potential (Fig. 2D and G). This is in line with studies of other malignancies, wherein various oncogenic processes lead to a metabolic program of aerobic glycolysis, diverting metabolites toward pathways that are important for various kinds of macromolecule synthesis (11). As more carbon is diverted to biosynthetic pathways from glycolysis, flux through the TCA cycle may decrease, resulting in oncogene-induced reduction in respiratory capacity, as suggested by our data. These findings reinforce the concept that such a shift away from oxidative metabolism is a characteristic feature of oncogenesis, important in all cancers, and not a spectator phenomenon merely correlating with oncogenic processes. Furthermore, they demonstrate that EF is a driver of oncogenic metabolic reprogramming, in addition to and as a consequence of transcriptional reprogramming.

Surprisingly, we also observed an increase of glycolysis in shEF versus shLUC cells (Fig. 2A and C). Such observations appear to run contrary to the known glycolytic metabolism that predominates in the vast majority of cancer cells. These observations could be explained by transcriptional downregulation of genes important for glucose import or utilization. Indeed, we observed that hexokinase 1 (HK1) is increased in shEF versus shLUC cells (Fig. 2E). In such a scenario, EF may restrict glucose entry into glycolysis, effectively reducing the cell's ability to metabolize additional glucose (as seen in our Seahorse analysis data). However, if this restriction is not so severe to cause cell death by starvation, glucose uptake may remain sufficient to fulfill cellular energy requirements. Simultaneously, EF upregulates genes of biosynthetic pathways such as de novo serine synthesis (Fig. 4A), resulting in an increased proportion of glycolytic intermediates being shunted into biosynthetic pathways. Thus, it may be possible to generate the net effect of oncogenic and biosynthetic reprogramming of metabolism by increasing the proportion of glucose-derived carbon shunted toward biosynthetic pathways, even in the face of EF-induced downregulation of HK1. In addition, considering that the putative cell of origin for Ewing sarcoma is a mesenchymal stem cell, and that EF is known to repress mesenchymal features, the baseline glycolytic metabolism of these stem cells may provide ample flexibility to be permissive to decreased HK1 within the context of oncogenic metabolic reprogramming induced by EF (24, 43). Thus, any decrease in glycolysis caused by EF is not likely to be pro-oncogenic or proproliferative. Instead, we believe it is merely part of a constellation of changes, not all being programmatically consistent, that together contribute to transformation. Rather than increasing glucose utilization to provide metabolites for biosynthetic pathways, EF could increase the proportion of glycolytic intermediates being diverted to pathways such as de novo serine synthesis. These and other hypotheses are a subject of current and future study.

Our findings also suggest that EF drives global shifts in metabolite abundance in Ewing sarcoma cells (Fig. 3). Integrating metabolomic and transcriptomic data enabled us to identify EF-driven upregulation of de novo serine synthesis via PHGDH as an important candidate for further study. PHGDH is amplified in various tumors, and has been implicated previously in tumor biology (32, 35). PHGDH commits the glycolytic intermediate 3-PG to de novo serine synthesis, and EF-driven upregulation of PHGDH expression appears to be an important factor in diverting metabolites toward macromolecule synthesis (Fig. 4A). Importantly, elevated expression of PHGDH also correlates with a poor prognosis in Ewing sarcoma patients, suggesting that these phenomena have important clinical implications (Fig. 4C). Our findings are similar to previous findings demonstrating upregulation of serine synthesis by other oncogenes. PHGDH, PSPH, and PSAT are upregulated in some Myc-driven lymphomas (44). In addition, oncogene-driven transcriptional misregulation in other malignancies (e.g., HER2, SP1, NFY, NRF2, ATF4) has been shown to increase expression of enzymes of de novo serine synthesis (45–47). EF thus appears to generate a metabolic program of upregulated serine synthesis similar to other malignancies. Existing transcription profiling data suggest that SP1 is an upregulated target of EF, while NFY, NRF2, and ATF4 are not differentially expressed after EF knockdown; HER2 may be slightly downregulated by EF (data not shown). Further study is required to determine whether EF exerts its effects by coopting any of the aforementioned regulatory factors. Moreover, existing ChIP-seq data are inconsistent on whether EF binds directly at the promoter of PHGDH (48, 49). However, the fact that EF binding does appear to occur in some datasets suggests that it could be directly regulating expression of PHGDH, and further assessment of EF-directed regulation of PHGDH deserves more in-depth study.

Downstream of de novo serine synthesis, conversion of serine to glycine is carried out by serine hydroxymethyltransferase (SHMT). This reaction is coupled to the conversion of tetrahydrofolate to 5,10-methylenetetrahydrofolate, and is a major provider of 1-carbon units needed for multiple cellular functions (50). Our data suggest that EF is also driving upregulation of these processes, further establishing a pro-oncogenic, biosynthetic metabolic program in these cells (Fig. 5H). It is likely that this is also clinically relevant, as patients with high levels of SHMT also have worse overall survival (Fig. 4D). Together, these data also raise the interesting possibility of utilizing expression levels of metabolic enzymes such as PHGDH and SHMT to inform predictions of patient prognosis, or even using metabolic indices to define different patient populations.

The nature of the relationship between patient survival and expression levels of these genes deserves further consideration. Several hypotheses exist for why increased expression of these enzymes correlate with poorer prognosis. For instance, upregulation of de novo serine synthesis may contribute importantly to increased synthesis of glycine and cysteine. Such a process would contribute more amino acids to meet the biosynthetic requirements of proliferative cancer cells. In addition, as glycine and cysteine both contribute to production of glutathione, increased serine synthesis also potentially contributes to greater redox buffering. Indeed, serine metabolism has been shown to be activated under conditions of increased ROS production, perhaps contributing to increased capacity to confront redox stress (51). Furthermore, the conversion of serine to glycine contributes importantly to one-carbon metabolism and the folate cycle. The de novo synthesis of adenosine, guanosine, and thymidylate require one-carbon units derived from serine. Thus, increased serine synthesis could contribute to greater nucleotide synthesis in general. Upregulation of SHMT1/2 and TYMS is consistent with these hypotheses. Upregulation of such processes could permit cancer cells to become more proliferative or aggressive, thus contributing to poorer prognosis in patients. However, as these hypotheses have yet to be tested, more in-depth investigation is clearly needed.

Upregulation of de novo serine synthesis and 1-carbon metabolism has far-reaching consequences for cancer cells, as flux through these pathways is important for regulation of methyl donor (methionine) recycling and epigenetic states. In addition, previous work has suggested that amplified PHGDH may play a role in cancer through production of 2-HG (52). 2-HG is a known oncometabolite when overproduced by neomorphic mutations in IDH enzymes (36). Intriguingly, our data also suggest that important links exist between de novo serine synthesis (specifically PHGDH) and nuclear processes. We observed decreased amounts of 2-HG in shEF versus shLUC cells (Fig. 6B and C). However, the amounts of 2-HG generated by EF-induced upregulation of PHGDH are lower than levels seen in other malignancies with IDH mutations, and 2-HG likely has less of a role in Ewing sarcoma biology. In addition, IDH mutations are rare in Ewing sarcoma (53). Some work has shown that the small amounts of 2-HG produced by PHGDH may be sufficient to promote histone methylation, contributing to an oncogenic hypermethylation of histones and, possibly, DNA (52). However, our results suggest that this is not the case in Ewing sarcoma. Rather, knockdown of PHGDH did not affect DNA methylation (Fig. 6D), but did cause significant increases in histone methylation (Fig. 6E and F). Interestingly, these histone alterations were observed in EF-expressing cells (UTES-14-01872) as well as in TTC466 cells, which harbor the alternate translocation EWS/ERG (found in ∼10% of Ewing sarcomas). While the mechanisms responsible for these chromatin changes are unclear, we hypothesize that abundance of metabolites important for the function of other enzymes (either as substrates or cofactors) might play a role. Future efforts are aimed at testing these hypotheses and further elucidating the mechanisms linking metabolism and nuclear control.

Taken together, our findings demonstrate that EF drives metabolic misregulation resulting in diversion of metabolites toward biosynthetic pathways to meet the increased demands of oncogenesis. EF induces increased expression of PHGDH, PSAT, and PSPH, shunting glycolytic intermediates into serine and glycine synthesis, and contributing to 1-carbon metabolism through upregulated SHMT. While most research of Ewing sarcoma has previously focused on transcriptional regulation, epigenetic modification and other nuclear processes, little attention has been focused on metabolic pathways. Indeed, metabolic systems that are co-opted or altered by the presence of EWS/FLI might include targetable enzymes that could be modulated by new therapies. An EWS/FLI–specific metabolic program could provide a targetable surrogate for EWS/FLI itself, providing a way to specifically destroy EWS/FLI-harboring cells through pharmacologic modulation of metabolic or other enzymes. Such efforts aspire to inform the development of hitherto elusive molecular targeted therapies for Ewing sarcoma.

S.L. Lessnick is an acting chief medical officer at Salarius Pharmaceuticals and has ownership interest (including patents) in Salarius Pharmaceuticals in a patent on NKX2.2 and patent on GSTM4. No potential conflicts of interest were disclosed by the other authors.

Conception and design:J.M. Tanner, J. Rutter

Development of methodology:J.M. Tanner, P. Wei, S.L. Lessnick

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.):J.M. Tanner, C. Bensard, P. Wei, N.M. Krah, J.C. Schell, J.D. Gardiner

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis):J.M. Tanner, C. Bensard, P. Wei, N.M. Krah, S.L. Lessnick

Writing, review, and/or revision of the manuscript:J.M. Tanner, C. Bensard, J.D. Gardiner, J.D. Schiffman, J. Rutter

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases):J.M. Tanner

Study supervision:J. Rutter

The authors would like to thank Dr. Kevin Jones, Dr. Emily Theisen, Dr. Ranajeet Saund, and Dr. Kathleen Pishas, and all members of the Lessnick and Rutter labs for many useful and thought-provoking discussions throughout the course of these studies.

J. Rutter was supported by HHMI and NIH grant GM 110755. J. Schiffman was supported by St. Baldrick's Foundation, Sarcoma Alliance for Research through Collaboration (SARC) Career Development Award, Damon Runyon Clinical Investigator Award, and the Eunice Kennedy Shriver Children's Health Research Career Development Award NICHD 5K12HD001410. Metabolomics analysis was performed at the Metabolomics Core Facility at the University of Utah which is supported by 1 S10 OD016232-01, 1 S10 OD021505-01, and 1 U54 DK110858-01. The authors also acknowledge the support of the Huntsman Cancer Institute (P30CA042014) and the Huntsman Cancer Foundation.

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.