MYCN mediates cysteine addiction and sensitizes neuroblastoma to ferroptosis

Aberrant expression of MYC transcription factor family members predicts poor clinical outcome in many human cancers. Oncogenic MYC profoundly alters metabolism and mediates an antioxidant response to maintain redox balance. Here we show that MYCN induces massive lipid peroxidation on depletion of cysteine, the rate-limiting amino acid for glutathione (GSH) biosynthesis, and sensitizes cells to ferroptosis, an oxidative, non-apoptotic and iron-dependent type of cell death. The high cysteine demand of MYCN-amplified childhood neuroblastoma is met by uptake and transsulfuration. When uptake is limited, cysteine usage for protein synthesis is maintained at the expense of GSH triggering ferroptosis and potentially contributing to spontaneous tumor regression in low-risk neuroblastomas. Pharmacological inhibition of both cystine uptake and transsulfuration combined with GPX4 inactivation resulted in tumor remission in an orthotopic MYCN-amplified neuroblastoma model. These findings provide a proof of concept of combining multiple ferroptosis targets as a promising therapeutic strategy for aggressive MYCN-amplified tumors.

model for MYC-driven cancers 4 . Amplified MYCN identifies a highly aggressive subtype associated with malignant progression and poor outcome despite intensive multimodal treatments. However, a substantial proportion of low-risk neuroblastomas with elevated MYCN expressed from a normal MYCN locus, particularly those arising in children younger than 18 months, regress spontaneously (stage 4S disease) by unknown mechanisms even when the disease is metastatic 5 . High-risk neuroblastomas lacking amplified MYCN harbor rearrangements of other MYC gene family members, TERT or alternative mechanisms of telomere lengthening (ALT) often associated with ATRX gene mutations, the latter subtype being incompatible with high MYCN or MYC expression 6,7 (later referred to as MYC(N)). Beside these alterations linked to telomere maintenance mechanisms, mutations in ALK or other developmental genes lead to stalled differentiation and tumors composed of heterogeneous cell types resembling different states of the normal neuroendocrine differentiation trajectories. The spectrum of cell types ranges from differentiated over undifferentiated adrenergic-to-mesenchymal cell types (triggered by adrenergic-to-mesenchymal transition), where malignant progression, therapy resistance and disease relapse are strongly associated with undifferentiated cell types. 8, 9 MYCN-amplified neuroblastoma cells, like other MYC-driven cancer cells, have been found to be addicted to the amino acid glutamine (Gln), the absence of which causes growth arrest or apoptosis 10 . More recently, reports showed that neuroblastoma cells are also addicted to iron and are sensitized to ferroptosis 11,12 , a new iron-dependent oxidative form of cell death associated with lipid peroxidation and insufficient capacity to eliminate lipid peroxides 13 . Whereas the apoptosis pathway of regulated cell death is often genetically or epigenetically impaired in primary high-risk and relapsed neuroblastomas 14 , there is only limited knowledge of how metabolic rewiring as a consequence of aberrant MYCN controls the liabilities of fast-proliferating malignant cells that are confronted with accumulating reactive oxygen species (ROS).
In this study, by performing single amino acid deprivations in high MYCN and low MYCN neuroblastoma cells, we discovered strong dependency of high MYCN cells on the amino acid cysteine. Using functional MYCN synthetic lethal metabolic and genetic screens, we further identified cyst(e)ine deprivation and glutathione peroxidase 4 (GPX4) inhibition as selective liabilities in MYCN-amplified neuroblastomas. Combined targeting of cystine uptake, cysteine synthesis via transsulfuration and GPX4 in an orthotopic neuroblastoma model strongly reduced tumor growth in vivo. Multi-omics profiling identified multiple cell type-specific and MYCN-regulated mechanisms inhibiting ferroptosis in adrenergic and mesenchymal neuroblastoma cells. Taken together, our study uncovered mechanisms crucial to ferroptosis escape in MYCN-amplified neuroblastomas; simultaneous inhibition of those mechanisms led to tumor regression in vivo.

Results
Cystine deprivation induces MYCN-dependent ferroptosis. First, we analyzed the interplay of oncogenic MYCN activity with amino acid metabolism. Downregulating MYCN in the MYCN-amplified IMR5/75 neuroblastoma cell model 15 (approximately 65% reduction; Fig. 1a) slowed cell proliferation without inducing cell death (Extended Data Fig. 1a,b) and reduced the intracellular pools of all amino acids (Fig. 1b). Most prominently, cysteine was reduced nearly tenfold (Extended Data Fig. 1c). Inhibiting MYCN binding to Myc-associated factor X (MAX) using 10058-F4 (ref. 16 ) yielded similar results ( Fig. 1b and Extended Data Fig. 1d). These data show that high MYCN levels are associated with high levels of cellular cysteine, probably mediated by increased synthesis and/or uptake from the microenvironment. Systematic depletion of individual amino acids from the growth medium impaired cell viability in both high MYCN and low MYCN cells in most cases (Fig. 1c).
However, in line with recent reports 11,12 , cells with high MYCN expression exhibited stronger dependency on cystine imported by cystine/glutamate-exchange transporter x c − and readily reduced to two cysteine molecules intracellularly. Cystine deprivation caused robust cell death in high MYCN cells, which was largely prevented by downregulation of MYCN expression (Fig. 1c) or inhibition of MYCN-MAX binding (Fig. 1d). Overexpressing MYCN in MYCN diploid cells (Tet21N neuroblastoma cell model 17 ), rendered these cells highly vulnerable to cystine deprivation (Fig. 1e). Neuroblastoma cell lines with intermediate MYCN or MYC levels caused by gene translocations are known to exist. We inferred MYC(N) activity by target gene expression score 5 , which agrees with transcript and protein MYC(N) levels (Extended Data Fig. 1e,f). Cell death after cystine deprivation increased with MYC(N) activity score, being virtually absent in cell lines immortalized by alternative telomere lengthening and lacking MYC(N) aberrations and peaking in MYCN-amplified cell lines with the highest activity scores (Fig. 1f). These data demonstrate that oncogenic MYC(N) expression is associated with cysteine addiction, with cysteine reduction resulting in massive cell death in an MYC(N)-dependent manner.
Cell death in the cystine-deprived high MYCN cells was not abrogated by the inhibition of (1) caspases to prevent apoptosis, (2) lysosomal function to downregulate autophagy or (3) receptor-interacting serine/threonine-protein kinase 1 (RIPK1) to prevent necroptosis (Extended Data Fig. 1g). Availability of cystine/cysteine is the rate-limiting step in GSH synthesis counteracting ROS 18 , which prevents the execution of ferroptosis, an oxidative, non-apoptotic and iron-dependent form of regulated cell death caused by ROS-mediated lipid peroxidation 13,19,20 . Indeed, ferrostatin-1 (Fer-1), a specific inhibitor of ferroptosis, or a lipophilic antioxidant or an intracellular iron chelator such as ciclopirox olamine (CPX) averted death in cystine-deprived high MYCN neuroblastoma cells ( Fig. 1g and Extended Data Fig. 1h). Monitoring ROS formation by flow cytometry using the lipid peroxidation sensor, C11-BODIPY, showed that cystine deprivation dramatically increased cellular lipid peroxidation selectively in high MYCN cells (Fig. 1h). Inhibiting ferroptosis with lipophilic antioxidants or an intracellular iron chelator inhibited lipid peroxidation in cystine-deprived high MYCN cells (Fig. 1i). Hence, when deprived of cystine, high MYCN neuroblastoma cells exhibited lipid peroxidation and died via ferroptosis.
Addition of GSH to the cystine-free medium also prevented lipid peroxidation in MYCN-amplified cells similar to Fer-1 (Fig. 1i) and rescued cell viability similar to Fer-1 ( Fig. 1j and Extended Data Fig. 1i), suggesting that high GSH levels protect against ferroptosis. Consistent with their higher sensitivity to ferroptosis, MYCN-amplified neuroblastoma cells had lower baseline GSH and cysteine levels compared to cell lines with lower MYC(N) activity scores (Extended Data Fig. 1j,k). However, on downregulation of MYCN in MYCN-amplified cells, intracellular GSH levels were reduced threefold, the reduced-to-oxidized GSH ratio was halved and intracellular ROS levels increased (Fig. 1k), suggesting that high MYCN expression increases GSH synthesis and ROS clearance. Overall, these results indicate that although oncogenic MYCN activates the production of GSH, it is maintained at a low steady state level due to its rapid consumption in fast-proliferating cells.
The only other amino acid showing a selective dependency on MYCN was Gln, confirming previous reports 10,21 , although this was significantly less pronounced (Fig. 1c). Interestingly, cysteine and Gln (when converted to glutamate) are two important GSH precursors. Depleting Gln in addition to cystine partially restored GSH levels probably because of less lipid peroxidation 22 and rescued cell viability (Fig. 2a-c). Gln/cystine deprivation prevented the induction of ferroptosis-promoting genes (CHAC1, HMOX1) and reduced transcriptional activation of genes protecting against cellular stress (ATF4, ATF3) shown to be activated on cystine depletion 23     ( Fig. 2d). Besides Gln depletion, inhibiting glutaminolysis should prevent ferroptosis 24 . Indeed, knockdown of either glutaminase isoform, GLS KGA/GAC , averted ferroptosis confirming that glutaminolysis is required to induce ferroptosis in cystine-deprived high MYCN cells (Fig. 2e). However, since GLS KGA is repressed by MYCN (Fig. 2f), GLS GAC appears to be the main glutaminase isoform involved in these cells.

Identification of ferroptosis genes in MYCN-amplified cells.
To systematically identify the cellular vulnerabilities associated with high MYCN expression, we performed an unbiased synthetic lethal small interfering (siRNA) screen using high/low MYCN IMR5/75 cells (Fig. 3a). Among the top hits that reduced the survival of high MYCN cells were enzymes involved in GSH metabolism that detoxify lipid peroxides (Fig. 3b One of the screening hits, GPX4, is an anti-ferroptotic selenoprotein that protects against lipid peroxidation using GSH 25,26 . siRNA-mediated GPX4 knockdown induced ferroptosis in high MYCN cells, which was rescued by Fer-1 ( Fig. 3d and Extended Data Fig. 2b). Similarly, we observed reduction of viability using the GPX4 inhibitor RSL3, enhanced by supplementing iron and rescued by Fer-1 or GSH (Fig. 3e,f) or simultaneous MYC(N) activity inhibition using 10058-F4 (Extended Data Fig. 2c). The dependency of cell viability on GPX4 activity increased with MYC(N) activity across our broader cell line panel (Fig. 3g,h and Extended Data Fig. 2d,e) unlike GCLC or GSR, which exhibited no dependency (Extended Data Fig. 2f). Next, we studied whether the effect of GPX4 knockdown was linked to cystine uptake and metabolic gene expression in 348 cancer cell lines 27 . Cells dependent on GPX4 showed low expression levels of SLC7A11 of the x c − system, importing cystine in exchange for glutamate (Extended Data Fig. 2g). Neuroblastoma cell lines were, after ovarian cancer cell lines, the second most dependent on GPX4 (Extended Data Fig. 2h).
To therapeutically investigate the role of GPX4 inhibition in vivo, we used an inducible CRISPR-Cas9 system to eliminate GPX4 (GPX4 knockout). In the MYCN-amplified SK-N-DZ three-dimensional (3D) cell culture model, GPX4 knockout induced ferroptotic cell death. This was rescued by simultaneously treating cells with the 10058-F4 MYC-MAX inhibitor (Fig. 3i). We then generated an orthotopic mouse model for human neuroblastoma by transplanting these SK-N-DZ neuroblastoma cells orthotopically into the adrenal gland and allow the cells expressing native GPX4 levels to develop into small tumors (Fig. 3j). Subsequently, the CRISPR-Cas9 GPX4 knockout was induced by doxycycline (Dox) treatment 7 d after transplantation. Although this resulted in a significant reduction in tumor weight compared to controls (Fig. 3k), it was not sufficient for tumor eradication as also shown by others 28 . Ferroptosis markers such as CHAC1 and TFRC 29 increased despite only partial reduction of GPX4 transcripts, suggesting activation of ferroptosis in vivo (Fig. 3l). In summary, our data show that GPX4 partially protects high MYCN neuroblastoma cells from ferroptosis in vitro and in an orthotopic neuroblastoma mouse model. Both neuroblastoma cell lines and primary neuroblastomas had the lowest SLC7A11 expression compared to other tumor entities (Extended Data Fig. 2i,j). This suggests that the sensitivity of neuroblastoma to cystine depletion and GPX4 inhibition is due, at least in part, to reduced uptake of cystine 18 . SLC7A11 knockdown or inhibition by sulfasalazine moderately reduced viability regardless of MYCN level ( Fig. 4a-c) as shown recently 12 . Erastin, a more potent SLC7A11 inhibitor 23 , selectively induced ferroptosis in the high MYCN state (Fig. 4d), which was partially rescued by additional inhibition of MYC(N) activity (Fig. 4e). Erastin-induced cell death increased with MYC(N) activity score in neuroblastoma cell lines (Fig. 4f,g and Extended Data Fig. 2k), was associated with decreased GSH and reduction of the GSH reduced-to-oxidized ratio ( Fig. 4h) and could be prevented by providing either GSH or Fer-1 ( Fig. 4i). Taken together, cystine uptake by SLC7A11 maintains part of the necessary cysteine for GSH production to protect against ferroptosis.

MYCN induces transsulfuration and prevents ferroptosis.
Cellular cysteine can also be produced in some cell types by transsulfuration 30,31 . In this process, homocysteine (Hcy), an intermediate of the methionine cycle, and serine are combined to form cystathionine (Cysta), which is further converted to cysteine 32 (Fig. 5a). GPX4 dependency in cancer cell lines was associated with enhanced expression of cystathionine beta-synthase (CBS), the rate-limiting enzyme for transsulfuration (Extended Data Fig. 2g), with neuroblastoma being among the cancer entities with the highest CBS expression (Extended Data Fig. 2m,l). We hypothesized that transsulfuration provides a cysteine source for neuroblastoma cells preventing ferroptosis in cystine deprivation conditions. Cystathionine gamma-lyase (CTH), converting Cysta to cysteine, and S-adenosyl-L-homocysteine hydrolase (AHCY), synthesizing Hcy for transsulfuration, show synthetic lethality with high MYCN (Fig. 5b,c and Extended Data Fig. 3a), as are two methyltransferases that feed into Hcy production (Extended Data Fig. 3b). Supplementing cystine-deprived cells with either Hcy or Cysta prevented ferroptosis in all adrenergic neuroblastoma cell lines tested with high or intermediate oncogenic MYC(N) expression, but not in the less common mesenchymal neuroblastoma lines ( Fig. 5d and Extended Data Fig. 3c). Pharmacologically inhibiting CTH using propargylglycine (PPG) 33 sensitized adrenergic, but not mesenchymal, high MYCN cell lines to either erastin-or imidazole ketone erastin (IKE)-induced cell death (Fig. 5e). Knockdown of AHCY in adrenergic high MYCN but not mesenchymal neuroblastoma cells impaired colony formation, which was associated with reduced GSH levels and reduction of GSH reduced-to-oxidized ratios ( Fig. 5f, g). In summary, these data indicate that transsulfuration provides an internal cysteine source for GSH biosynthesis protecting high MYCN adrenergic neuroblastoma cells from ferroptosis. Intracellular cysteine is required for two rate-limiting cellular processes: GSH-mediated ROS clearance and production of building blocks in the synthesis of proteins, nucleotides and lipids 34 (Fig. 5a). In line with this, protein synthesis inhibition with cycloheximide increased cysteine and total GSH levels (Extended Data Fig. 3d). Notably, reduction/deprivation of cystine (Cys 2 ) in the medium drastically reduced intracellular cysteine and GSH in the high MYCN state before ferroptosis (Fig. 2a,b) but did not affect cell cycle progression and only moderately reduced protein synthesis (Extended Data Fig. 3e,f). MYCN protein levels were not affected on cystine deprivation compared to methionine-or Gln-deprived cells (Extended Data Fig. 3g). These data suggest robust cysteine channeling into protein synthesis under limited cystine supply when GSH production is already diminished.
To investigate this further, we performed transcriptomic analysis of IMR5/75 cells cultured in cystine-free medium or inhibiting the x c − system with erastin. This revealed activation of a stress response 35 before ferroptosis that channeled cysteine into protein synthesis by inducing CARS 31 and cysteine recycling from GSH ( Fig. 2d and Extended Data Fig. 3h-j). HMOX1 was also activated in these cells, suggesting that the free iron pool is increased before inducing ferroptosis 36 (Fig. 2d and Extended Data Fig. 3h-j). Inhibiting CARS prevented ferroptosis in the cystine-deprived high MYCN state, highlighting the competition for intracellularly synthesized cysteine between protein synthesis and redox balance (Fig. 5h). Taken together, transsulfuration supplies cysteine for both protein and GSH synthesis in adrenergic MYCN-amplified neuroblastoma cells but prioritizes cysteine for protein synthesis at the expense of GSH and redox balance when cystine uptake is limited, thus triggering ferroptosis.
Next, we asked how cysteine metabolism and redox homeostasis are affected by oncogenic MYC(N) activity in the adrenergic (active transsulfuration) or mesenchymal (inactive transsulfuration) neuroblastoma subtypes 8 . Adrenergic cells upregulated three key enzymes in transsulfuration, CBS, AHCY and D-3-phosphoglycerate dehydrogenase (PHGDH), in the high MYCN state, while the x c − system (SLC7A11) controlling cystine uptake was unaffected by changes in MYCN (Extended Data Fig. 4a) 37 . In a panel of 32 neuroblastoma cell lines, MYCN or MYC amplification or translocation in adrenergic subtypes was accompanied by upregulated CBS expression, while MYC translocation/activation in mesenchymal cell lines was not (Fig. 5i). Ectopically expressing MYCN in mesenchymal Tet21N cells left transsulfuration via its rate-limiting enzyme CBS unchanged but induced SLC7A11, GSR, GCLC and thioredoxin reductase 1 (TXNRD1), the last one being significantly expressed in MYCN-amplified primary neuroblastomas (Extended Data Fig. 4b,c) 37 . At the gene regulatory level, we found that the CBS locus harbored both activating (H3K27ac, H3K4me3) and silencing (H3K27me3) histone modifications, with the former being increased and the latter decreased in the presence of amplified MYCN in adrenergic cells compared to non-amplified or mesenchymal cells (Fig. 5j and Extended Data Fig. 5). In primary neuroblastomas, differences in CBS expression correlated with histone modifications and methylation of intragenic CpGs dependent on genomic MYCN status ( Fig. 5k and Extended Data Fig. 6). Together, this suggests that transsulfuration is active in the adrenergic state and regulated at the epigenetic level in high MYCN cells.
We found that the levels of CBS and AHCY were associated with poor patient survival (Fig. 6a). Global gene expression profiles from 498 primary neuroblastomas 38 confirmed elevated AHCY and CBS in MYCN-amplified neuroblastomas (Fig. 6b,c). Higher HMOX1 and lower SLC3A2 and TXNRD1 expression were found in stage 4S tumors (Fig. 6d), which tend to regress spontaneously. CBS, AHCY and PHGDH expression was also elevated in mass spectrometry-based global proteomes from MYCN-amplified neuroblastoma tumors (Fig. 6e). SLC7A11 antiporter expression did not correlate with MYCN amplification or other risk factors for poor patient outcomes (Fig. 6c). In addition, genes involved in Gln (that is, SLC38A5, SLC1A5), methionine (that is, SLC7A5) and iron uptake (TFRC) had higher expression in MYCN-amplified or MYCN-overexpressed neuroblastomas; TFRC was recently associated with ferroptosis in neuroblastomas 12 (Fig. 6c,d and Extended Data Fig. 4c). Genes associated with GSH synthesis/metabolism (that is, GCLC, GSR) were also upregulated both at the transcript and protein level (Fig. 6c,e,f). Higher expression of the mitochondrial glutaminase (GLS GAC ) 24,39 and lower expression of the cytoplasmatic glutaminase (GLS KGA ) isoforms were observed in MYCN-amplified tumor transcriptomes and proteomes (Fig. 6c,e), suggesting dependence on mitochondrial glutaminolysis in these neuroblastomas. Together, our results show that MYCN-amplified neuroblastomas increase transsulfuration activity, iron import, glutaminolysis and GSH production through coordinated changes in gene expression thereby increasing the susceptibility to ferroptosis.

Cyst(e)ine and GPX4 inhibition as therapeutic targets.
To exploit our findings in vivo, we first tested simultaneous inhibition of cysteine uptake and transsulfuration in vitro using IKE, an erastin analog with acceptable pharmacokinetic properties 40,41 , and PPG 33 (Fig. 7a). We observed a strong synergistic effect with the two drugs in only MYCN-amplified adrenergic cells (Fig. 5e). We then injected mice orthotopically with SK-N-DZ neuroblastoma cells and treated them for two 5-d cycles with a combination of IKE 45 mg kg d −1 and PPG 45 mg kg d −1 (Fig. 7b). When combining these two drugs, we observed a 60% reduction in tumor growth in MYCN-amplified SK-N-DZ-driven tumors (Fig. 7c). These data suggest that MYCN-amplified neuroblastoma cells with very high levels of MYCN are sensitive to a reduction of the intracellular cysteine pool by simultaneous inhibition of cystine import and cysteine synthesis via transsulfuration.
Next, we combined this protocol for reducing cellular cysteine with genetic targeting of GPX4 activity (Fig. 7b). We observed a robust effect with complete remission in most animals (Fig. 7d,e). Transcriptional profiling of residual small tumors revealed induction of ferroptosis markers after combined inhibition of cystine uptake/cysteine synthesis and GPX4 compared to tumors treated with vehicle control (Fig. 7f). These data provide strong in vivo evidence that concomitant reduction of cystine uptake, transsulfuration and GPX4 activity can be utilized as a new therapeutic strategy for high-risk, MYCN-amplified neuroblastomas.

Discussion
We demonstrated that oncogenic MYCN sensitizes neuroblastoma cells to ferroptosis when intracellular cysteine availability for GSH synthesis and the cystine/cysteine redox cycle are limited. A high MYCN state in neuroblastoma cells sensitizes them to lipid peroxidation, which in combination with acute intracellular cysteine reduction triggers massive ferroptotic cell death. Our study shows transsulfuration and GSH redox activity to be crucial to escape ferroptosis, whereas Gln import and glutaminolysis are required for ferroptosis in neuroblastoma cells with oncogenic MYCN (Fig. 7a). Expression levels of key genes for these processes are correlated with MYCN expression in high-risk tumors. The recently identified adrenergic and mesenchymal subtypes 8 also appear to determine how cysteine is maintained in neuroblastoma cells, with transsulfuration only being activated by oncogenic MYC(N) in adrenergic cells, where CBS is accessible and not epigenetically repressed as in mesenchymal cells.
In this study, we describe metabolic rewiring in MYCN-amplified adrenergic neuroblastoma cells, where high consumption of cysteine used for the synthesis of cellular building blocks at the expense of GSH synthesis and ROS clearance creates a new MYCN-dependent liability. To prove that this liability can be exploited as a new therapeutic concept, we established and optimized an orthotopic model for neuroblastoma using intra-adrenal gland tumor cell transplantation. This model allowed robust in vivo testing of     Recent studies 11,12 highlighted the role of iron in MYCN-dependent neuroblastoma and ferroptosis. The TFRC gene, involved in iron uptake, is activated by MYC in several cell types enhancing cellular proliferation 42 . In line with these studies, we showed a synergistic effect of iron in drug-induced ferroptosis in MYCN-amplified cells. In addition, we demonstrated that the iron chelator CPX rescues ferroptotic cell death under cystine deprivation conditions (Extended Data Fig. 1h). We further observed high expression of TFRC in MYCN-amplified cell lines and tumors (Fig. 6c and Extended Data Figs. 3j and 4a) reflected on the protein level (Fig. 6e). Similarly, we observed HMOX1 upregulation, involved in increasing the labile iron pool, in cystine-deprived neuroblastoma cells before ferroptosis induction. Together, this highlights a prominent role for iron metabolism in promoting ferroptosis in neuroblastoma cells.
We have described a detailed mechanism that explains how cystine deprivation triggers ferroptosis in MYCN-amplified tumors, which depends on the (epi)genetic context of adrenergic or mesenchymal subtypes, the later having transsulfuration silenced. These results differ from Floros et al. 11 , who suggested that high MYCN mediates upregulation of SLC7A11 in neuroblastoma cells. In contrast, our results indicate that in most, if not all, adrenergic neuroblastoma cells (>95% of neuroblastomas) high MYCN fails to boost cystine uptake via x c − . Instead, MYCN strongly activates transsulfuration, the intracellular conversion of methionine to cysteine. In general, neuroblastoma cells have very low SLC7A11 expression compared to other cancer types, which may explain the remarkable sensitivity of neuroblastoma cells to ferroptosis stimuli. Our results demonstrate that the uptake of cystine via x c − is not regulated on tumor progression in adrenergic MYCN-amplified cells. Upregulation of SLC7A11 by MYC oncoproteins was only found in cells of the rare mesenchymal subtype, having epigenetically silenced transsulfuration. Therefore, upregulation of SLC7A11 may become relevant in the relapse scenario, which is associated with elevated proportions of mesenchymal cells.
Intriguingly, our data may also explain spontaneous neuroblastoma regression, the mechanisms of which have remained elusive. In low-risk, metastatic neuroblastomas (stage 4S), which are likely to regress spontaneously 5 , we identified multiple markers predisposing them to ferroptosis (Fig. 7a), including: (1) failure to upregulate enzymes involved in GSH biosynthesis (GCLC) and redox activity (GSR) and transsulfuration (AHCY, CBS) with increasing MYCN activity (Fig. 6f); (2) CBS silencing by intragenic CpG methylation (Extended Data Fig. 6a); (3) HMOX1 upregulation to increase free iron pools; and (4) SLC3A2 and TXNRD1 downregulation (Fig.  6d). Spontaneous regression may be the physiological resolution of this cellular state sensitive to ferroptosis. Unlike low-risk tumors, MYCN-amplified neuroblastomas appear to metabolically adapt to survive events that deplete intracellular cysteine, such as high systemic glutamate present in the first two years of life 43 . The cysteine requirement of cancers dependent on oncogenic MYCN activity creates a previously unknown Achilles' heel that could be exploited to selectively induce ferroptosis for treatment. Our findings identify cysteine uptake, transsulfuration and the lipid peroxidation-specific scavenging system as vulnerabilities in cancer cells driven by oncogenic MYC(N) activity such as MYCN-amplified neuroblastomas. were calculated using two different methods: (1) impedance measurements were assessed using the RTCA system (Roche) by seeding at different cell densities and registering impedance signals every 20 min; (2) standard growth curves were generated by counting cells by FACS at the indicated time points and excluding propidium iodide-positive cells as necrotic. In both methods, exponential curves were fitted and doubling times calculated.

Ethics
Inducible stable AHCY knockdown cell cultures. Stable IMR5/75 and SH-EP cell lines expressing shRNA against AHCY under control of the Tet repressor were generated stepwise as described previously 15 using the following oligonucleotide sequence to target AHCY: f or wa rd: g at cc cc GG AT CA CT AC CG CT AC TG At tc aa gaga TC AG TA GC GGTAGTGATCCtttttggaaa; reverse: g cttttccaaaaaGGATCACTACC GCTACTGAtctcttgaaTCAGTAGCGGTAGTGATCCggg. IMR5/75 (5,000 cells per well) and SH-EP-AHCYsh (1,000 cells per well) were seeded in 6-well plates and simultaneously treated with Dox (1 µg ml −1 ) to induce the AHCY-targeting shRNA. Cells were fixed (11% glutaraldehyde; Sigma-Aldrich) and Giemsa-stained five (SH-EP-AHCYsh) or 7 d (IMR5/75-AHCYsh) later. Colony counting was performed using a GelDoc Documentation System and Quantity One software version 4.6.6 (Bio-Rad Laboratories) and quantification using Microsoft Excel 2016.
Quantification of amino acids. Pellets of 2 × 10 6 cells were extracted with 0.1 ml ice-cold 0.1 M HCl. Non-thiol-containing amino acids were quantified after specific fluorescent labeling with AccQ-TagTM (Waters) as described previously 46 . Cysteine and GSH levels were determined after labeling with monobromobimane (Calbiochem) as described previously 47 .
Western blot analysis of proteins in cell extracts. Whole cell lysates were prepared and protein expression was visualized as described previously 48 Table 2.
Generation of inducible Cas9 neuroblastoma cell lines. Molecular cloning. The CRISPR-Cas9-mediated inducible knockout experiments were performed using the lentiviral pCW-Cas9-EGFP plasmid. The plasmid was obtained as follows: the lentiviral pCW-Cas9 plasmid (plasmid no. 50661; Addgene) was subcloned by cutting it with the restriction enzymes HincII and XbaI (New England Biolabs). Next, a gBlock DNA fragment (Integrated DNA Technologies) encoding enhanced green fluorescent protein (eGFP) with complementary overhangs was cloned into the plasmid backbone thus replacing the puromycin resistance gene and generating the pCW-Cas9-EGFP plasmid. Guide RNAs (gRNAs) for the CRISPR knockout experiments were designed with Benchling (https://www.benchling. com) using the human reference genome GRCh38. Oligonucleotides were ordered (Sigma-Aldrich) with complementary overhangs to the lentiviral delivery plasmid backbone pLKO5.sgRNA.EFS.tRFP6572 and listed in Supplementary Table 2.
Lentivirus production. Lentivirus production was performed using a second-generation lentiviral system and a Calcium Phosphate Transfection Kit (Invitrogen) in HEK 293T cells. Briefly, early passaged HEK 293T cells were cotransfected with the lentiviral transfer plasmid, a packaging plasmid (psPAX2; plasmid no. 12260; Addgene), and with a plasmid coding for the VSV-G envelope (pMD2.G; plasmid no. 12259; Addgene). All experimental procedures for lentivirus production were performed in a biosafety level 2 laboratory.
The SK-N-DZ cell line was previously transduced with lentiviral particles carrying the luciferase reporter pLX-puroR-Luc at a multiplicity of infection (MOI) of 0.3. After recovery, resistant cells were transduced with pCW-Cas9-EGFP lentiviral particles at an MOI of 0.3. Polyclonal cell lines were maintained in DMEM supplemented with 10% tetracycline-free FCS (Clontech Laboratories). Next, cells expressing eGFP and the luciferase reporter were transduced with lentiviral particles carrying the gRNA targeting GPX4 at an MOI of 0.3. To generate monoclonal CRISPR cell lines, cells were individualized based on eGFP and RFP657 expression using FACS. In vitro Cas9 expression was induced supplementing the culture medium with 1 µg ml −1 Dox (Sigma-Aldrich).
In vivo orthotopic mouse experiments. All studies involving mice and experimental protocols were conducted in compliance with German Cancer Research Center guidelines and approved by the governmental review board of the state of Baden-Württemberg, Karlsruhe District Council, under authorization no. G-176/19, according to German legal regulations. The mouse strains used in the study were NOD.Cg-Prkdc scid Il2rgtm1 Wjl /SzJ (stock no. 005557; The Jackson Laboratory). Female mice (3-4 months old) were used for the experiments. Mice were housed in individually ventilated cages under temperature and humidity control. Cages contained an enriched environment with bedding material. To generate orthotopic mouse models for neuroblastoma, 2 × 10 5 SK-N-DZ cells were transplanted into the right adrenal gland after the surgical site was prepared 49 . Cells were resuspended in a 1:1 (vol/vol) mix of growth factor-reduced Matrigel (Corning) and PBS. Overall, 20 µl of this cell suspension was injected into the right adrenal gland of anesthetized mice. After tumor cell transplantation, we monitored mice for evidence of tumor development by bioluminescent signal using an IVIS Spectrum Xenogen device (Caliper Life Sciences). We observed a clear signal from the tumors 1 week after the injection of 2 × 10 5 SK-N-DZ cells. IKE and PPG were used at a concentration of 45 mg kg d −1 through intraperitoneal injection. Animal health was monitored daily and mice were euthanized as soon as they reached the termination criteria defined in the procedure. Sample size was calculated with the help of a biostatistician using R v.3.4.0. Assumptions for the power analysis were as follows: Alpha error, 5%; Beta error, 20%. Mice were randomized into treatment groups before treatment. In case animals had to be euthanized before the predefined end point (due to weight loss or other termination criteria), they were excluded from any downstream analyses. All animal experiments (apart from animal treatment) were blinded during the experiments and follow-up assessment.
Transcript profiling of tumors and cell lines. Microarray analysis. RNA isolated from the remaining tumors (five vehicle-treated versus five triple combination of GPX4 knockout + IKE + PPG) was analyzed using the Affymetrix GeneChip Clariom S Assay. Microarray data (GSE192976) were processed using a modification of a pipeline described previously 50 . Briefly, raw CEL files were robust multichip average-normalized and two-group comparisons were performed using the limma package v.3.46 (Bioconductor) with an empirical Bayes test for differential expression. An FDR < 0.1 was regarded as statistically significant.
RNA sequencing. Total RNA was isolated using the miRNeasy Mini Kit (QIAGEN) according to the manufacturer's protocols. RNA libraries were prepared using the NEBNext Ultra Directional RNA Library Prep Kit for Illumina (New England Biolabs). Raw RNA sequencing (RNA-seq) sequences of neuroblastoma cell lines and model systems were mapped to the UCSC hg19 genome using STAR 51 v.2.5.3a with default parameters. RNA-seq sequences from the IMR5/75 high and low MYCN cells were processed as described previously 37 . Differentially expressed genes were identified using the edgeR v.3.20.9 generalized linear model (glmFit) approach in R v3.4.3. The expression profiles of the high MYCN depletion/rescue experiments were visualized in R v4.3.0 using the gplots v.3.1.1 function heatmap.2, after filtering out lowly expressed genes (count per million ≤ 1) and normalizing for library size using trimmed mean of M-values normalization as implemented in edgeR 52 v.3.20.9. RNA-seq expression profiles from 498 primary neuroblastomas 38 (GSE49711) were analyzed. The Wilcoxon rank-sum test was used to test the association between candidate gene expression and amplified MYCN oncogene. Maximally selected log-rank statistics were used to describe the relationship between patient survival and candidate gene expression; the resulting expression cutoff points were used for dichotomization. Survival curves were estimated using the Kaplan-Meier method. Pearson correlation coefficients were calculated to estimate linear dependence between the expression values of candidate genes.
Quantitative PCR with reverse transcription. Complementary DNA was synthesized using the SuperScript IV Reverse Transcriptase Kit (catalog no. 18090-200; Invitrogen) according to the manufacturer's instructions. quantitative PCR with reverse transcription (RT-qPCR) was performed for the genes of interest and two housekeeping genes using the Platinum SYBR Green qPCR Superix-UDG kit (catalog nos. 11733038/11733-046; Invitrogen). Primers are listed in Supplementary Table 2.
Epigenetic characterization of tumors and cell lines. Chromatin immunoprecipitation followed by sequencing analysis of histone modifications in neuroblastoma primary tumors and cell lines. Formaldehyde cross-linking of cells, cell lysis, sonication, chromatin immunoprecipitation and library preparation were performed as described previously 53 , starting with approximately 4 × 10 6 cells (1 × 10 6 cells per individual immunoprecipitation). Direct cell lysis for each sample was achieved by 30 min incubation on ice in 950 µl radioimmunoprecipitation assay I buffer using approximately 30 mg of fresh-frozen tumor tissue per individual chromatin immunoprecipitation followed by sequencing (ChIP-seq) experiment. Library preparation was performed using the NEBNext Ultra DNA Library Prep Kit (New England Biolabs) according to the manufacturer's protocol. Samples were mixed in equal molar ratios and sequenced on an Illumina sequencing platform.

ChIPmentation of MYCN transcription factor in neuroblastoma cell lines.
Formaldehyde cross-linking, cell lysis, sonication and chromatin immunoprecipitation were performed as described previously 6 , adding the ChIPmentation module by Schmidl et al. 54 with the following changes: a Bioruptor Plus with automated cooling (4 °C) was used for high-intensity sonication (20-30 min each with 30 s on and 30 s off intervals) and 10 µg MYCN antibody and 10 6 cells for ChIP. The tagmentation reaction (Illumina Nextera DNA library Prep Kit) was performed at 37 °C for 1 min with the bead-bound chromatin sample or 5 ng purified input DNA for normalization. After de-cross-linking, purified samples were amplified using the dual index barcodes of the Nextera Index Kit and 13 PCR cycles. Enriched libraries were purified and pooled. ChIPmentation libraries were sequenced (50 single-end bases) on the Illumina sequencing platform (German Cancer Research Center Core Facility).

Data analysis of ChIP-seq and ChIPmentation.
Single-end reads were aligned to the hg19 genome using Bowtie2 v.2.3.0, keeping uniquely aligned reads only. The BAM files of aligned reads were further processed using the deepTools suite v.3.0 (ref. 55 ). Input files were subtracted from the treatment files using the bamCompare tool v.3.0, applying the simple exponential smoothing method to normalize signal to noise. The resulting signals were normalized to an average 1× coverage to produce signal (bigWig) files. Peaks were called using the MACS2 v.2.1 tool using default parameters. Data are available at the Gene Expression Omnibus (GEO) under accession no. GSE189174.
DNA methylation analysis. DNA methylation and gene expression data from 105 primary neuroblastomas assessed by Infinium HumanMethylation450 BeadChips and 44K Agilent oligonucleotide microarrays 56 (GEO accession no. GSE73518) were analyzed for candidate loci. The R2: Genomics Analysis and Visualization Platform (http://r2.amc.nl) was used to visualize the expression/methylation of selected gene-CpG pairs.
Tumor proteome analysis. Tumor proteome data were generated as part of a previous study 7 and was reanalyzed for this study. Briefly, tumor samples were lysed in SDS, homogenized, split into replicates, reduced, alkylated and purified by Wessel-Flügge precipitation. Samples were then digested by LysC and trypsin and fractionated by strong cation exchange before being measured by reversed phase liquid chromatography-mass spectrometry (LC-MS) on Q Exactive Plus instruments (Thermo Fisher Scientific). Proteins quantified in less than 50% of MYCN high-or low-risk cases were excluded. Data were imputed by random draw from a normal distribution with default parameters: 0.3 width and 1.8 downshift. A two-sided Welch's t-test was used to calculate the P values for the differential protein expression analysis; multiple testing correction by the Benjamini-Hochberg method was applied.

Statistics and reproducibility.
No statistical method was used to predetermine sample size. No data were explicitly excluded from the analyses unless they were of poor quality as determined by standard sequencing quality control metrics. The experiments were not randomized and the investigators were not blinded to allocation during the experiments and outcome assessment. Data are presented as the mean ± s.e.m. Statistical analyses were performed using Prism 7 (GraphPad Software). A two-tailed unpaired or paired Student's t-test was used for comparisons between two groups. For the animal experiments, sample size was calculated with the help of a biostatistician using R v.3.4.0. Assumptions for the power analysis were as follows: Alpha error, 5%; Beta error, 20%. Mice were randomized into treatment groups before treatment. In case animals had to be euthanized before the predefined end point (due to weight loss or other termination criteria), they were excluded from any downstream analyses. All animal experiments (apart from animal treatment) were blinded during the experiments and follow-up assessment.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Proteome data of neuroblastoma tumors were previously published by Hartlieb et al. 7 . All data were deposited with the European Genome-phenome Archive (as dataset no. EGAD00001006737) as part of study no. EGAS00001004349. Data are available upon request by contacting F.W. The RNA-seq data of 498 primary neuroblastoma patients were previously published 38 and are available at the GEO under the accession no. GSE49711. The DNA methylation data of primary neuroblastoma tumors were previously published 18 and are available at the GEO under the accession no. GSE73518. Time course RNA-seq profiling of IMR5/75 high MYCN and low MYCN cells was previously published 11 and can be accessed at the GEO under the accession no. GSE97774. ChIP-seq data were deposited at the GEO under the accession no. GSE189174. The aligned BAM files of the RNA expression profiles of the high MYCN depletion/rescue experiments were submitted to the European Nucleotide Archive and can be found under accession no. PRJEB25184. The data of the MYCN synthetic lethal siRNA screen has been added to this article as supplementary data. The microarray data from tumor samples were deposited with the GEO under accession no. GSE192976. The remaining data are available within the article, supplementary information or from the corresponding authors upon reasonable request. Source data are provided with this paper.

Code availability
The custom scripts used to analyze the RNA-seq, microarray and ChIP-seq data are available upon request.