Imetelstat-mediated alterations in fatty acid metabolism to induce ferroptosis as a therapeutic strategy for acute myeloid leukemia

Telomerase enables replicative immortality in most cancers including acute myeloid leukemia (AML). Imetelstat is a first-in-class telomerase inhibitor with clinical efficacy in myelofibrosis and myelodysplastic syndromes. Here, we develop an AML patient-derived xenograft resource and perform integrated genomics, transcriptomics and lipidomics analyses combined with functional genetics to identify key mediators of imetelstat efficacy. In a randomized phase II-like preclinical trial in patient-derived xenografts, imetelstat effectively diminishes AML burden and preferentially targets subgroups containing mutant NRAS and oxidative stress-associated gene expression signatures. Unbiased, genome-wide CRISPR/Cas9 editing identifies ferroptosis regulators as key mediators of imetelstat efficacy. Imetelstat promotes the formation of polyunsaturated fatty acid-containing phospholipids, causing excessive levels of lipid peroxidation and oxidative stress. Pharmacological inhibition of ferroptosis diminishes imetelstat efficacy. We leverage these mechanistic insights to develop an optimized therapeutic strategy using oxidative stress-inducing chemotherapy to sensitize patient samples to imetelstat causing substantial disease control in AML.

Telomerase enables replicative immortality in most cancers including acute myeloid leukemia (AML).Imetelstat is a first-in-class telomerase inhibitor with clinical efficacy in myelofibrosis and myelodysplastic syndromes.Here, we develop an AML patient-derived xenograft resource and perform integrated genomics, transcriptomics and lipidomics analyses combined with functional genetics to identify key mediators of imetelstat efficacy.In a randomized phase II-like preclinical trial in patient-derived xenografts, imetelstat effectively diminishes AML burden and preferentially targets subgroups containing mutant NRAS and oxidative stress-associated gene expression signatures.Unbiased, genome-wide CRISPR/Cas9 editing identifies ferroptosis regulators as key mediators of imetelstat efficacy.Imetelstat promotes the formation of polyunsaturated fatty acid-containing phospholipids, causing excessive levels of lipid peroxidation and oxidative stress.Pharmacological inhibition of ferroptosis diminishes imetelstat efficacy.We leverage these mechanistic insights to develop an optimized therapeutic strategy using oxidative stress-inducing chemotherapy to sensitize patient samples to imetelstat causing substantial disease control in AML.
AML is an aggressive and lethal blood cancer with a 5-year overall survival rate of less than 45% for patients younger than 60 years of age and less than 10% for older patients, predominantly due to disease relapse after chemotherapy or targeted treatments.AML has been extensively classified based on biological features and advances in sequencing technologies have led to a comprehensive genetic classification strategy (European LeukemiaNet, ELN2017) 1,2 .Despite this improved understanding of the individual disease subtypes, targeted treatment algorithms have resulted in only modest clinical benefits to date 3 .The development of effective therapies to improve remission rates and prevent relapse remains a top priority for patients with AML.
Telomerase is an attractive target as it is highly expressed and reactivated in the majority of AML and absent in most cell types including healthy hematopoietic cells.We have previously shown that genetic  Fig. 1 | Integrative analysis of samples from patients with AML.a, Unsupervised hierarchical clustering analysis on the expression of 300 transcripts with the greatest variance-to-mean ratios among 30 individual AMLs from our repository that can successfully generate AML PDX.MLD, multi-lineage dysplasia; MDS, myelodysplastic syndromes; MLL, mixed-lineage leukemia; NOS, not otherwise specified.b, Key clinical characteristics of patients from whom AML samples were derived including age at diagnosis, sex, ELN2017 prognostic risk group and WHO class of disease.c, OncoPrint of the most frequently detected mutations in AMLs by targeted next-generation sequencing of 585 genes associated with hematological malignancies (the MSKCC HemePACT assay) 31 .

Article
https://doi.org/10.1038/s43018-023-00653-5 depletion of telomerase eradicates leukemia stem cells, particularly upon enforced replication 4 .Despite promising preclinical evidence, the development of effective and specific telomerase inhibitors has been challenging.Imetelstat is a first-in-class covalently lipidated 13-mer thiophosphoramidate oligonucleotide that can competitively inhibit telomerase activity by binding to the telomerase RNA component TERC 5 .Imetelstat has shown clinical efficacy in essential thrombocythemia 6 , myelofibrosis 7 and lower-risk myelodysplastic syndromes 8 .In myelodysplastic syndromes, clinical benefits are associated with reductions in telomerase activity and TERT expression 8 .
In addition to its canonical role as critical regulator of telomere length maintenance, telomerase fulfills important non-canonical roles contributing to stress elimination, regulation of Wnt/β-catenin, NF-κB and p65 signaling, as well as resistance to ionizing radiation 9 .Hence, the clinical activity of imetelstat may be driven by mechanisms independent of telomere shortening and potentially canonical telomerase activity.
Preclinical trials in patient-derived xenografts (PDXs) provide genetically diverse, tractable models to define the efficacy of drugs and to identify biomarkers of response and resistance in AML 10 .PDX-based trials also allow, within the same cohort, the evaluation of new combination therapies with agents that may enhance efficacy and also critically compare their additive value to current, established standard treatments.
In this study, we aimed to assess the preclinical efficacy of imetelstat in a large AML PDX resource that reflects the diversity of genetic abnormalities found in large patient cohorts.We utilized this AML PDX resource to identify biomarkers of resistance and response to imetelstat therapy and to test potentially synergistic combination therapies.To elucidate the mechanism of action of imetelstat in an unbiased manner, we performed genome-wide CRISPR/Cas9 editing allowing the identification of gene knockouts that confer resistance to imetelstat therapy.This study reveals that imetelstat is a potent inducer of ferroptosis that effectively diminishes AML burden and delays relapse following oxidative stress-inducing therapy.

Generation of a comprehensive AML PDX resource
To generate a representative AML PDX inventory, primary bone marrow or blood samples from 50 patients were tested for engraftment and development of AML in NOD/SCID/IL2gR−/−/hIL3,CSF2,KITLG (NSGS).The overall success rate for primary engraftment in NSGS was 70%, defined by bone marrow, spleen or peripheral blood donor chimerism of at least 20%, splenomegaly (spleen weight >70 mg), anemia (HCT < 35%) or thrombocytopenia (PLT < 400 × 10 6 ml −1 ), microscopically visible AML infiltration into the spleen or liver and peripheral blood blast morphology (Extended Data Fig. 1a-n).Successfully engrafted NSGS recipients developed AML with a median onset of 173 d post-transplant (Extended Data Fig. 1p).
From the individual samples from patients with AML that successfully engrafted in NSGS, 30 were randomly selected and characterized based on clinical parameters, including patient age, sex, ELN2017 risk, World Health Organization (WHO) disease classification and molecular profiles obtained by transcriptional and mutational sequencing (Fig. 1a-c).All ELN2017 prognostic risk (favorable, intermediate and adverse) and age categories were represented; 17 samples were from female and 13 samples from male AML patient donors (Fig. 1b).Oncogenic mutations were most frequently detected in NPM1, DNMT3A and FLT3 loci and overall, this AML PDX resource recapitulated the genetic abnormalities that are observed in large clinical AML cohorts 2 (Fig. 1c).

A phase II-like preclinical trial of imetelstat in AML PDX
To test the preclinical efficacy of imetelstat in AML, the characterized 30 individual samples from patients with AML were each transplanted into 12 NSGS recipients (n = 360 PDXs in total).Once AML burden was detected, PDXs were randomized and treated with imetelstat or vehicle control (PBS) until disease onset or a survival benefit of at least 30 d was reached.Median survival was significantly prolonged in imetelstat compared to PBS-treated PDXs (155 d versus 100 d after start of treatment, P < 0.0001; Fig. 2a).AML burden measured as peripheral blood donor chimerism per day was significantly lower in imetelstat compared to vehicle-treated recipients (Fig. 2b).Moreover, end point peripheral blood donor chimerism, bone-marrow cellularity and donor chimerism as well as the absolute number of AML patient-derived cells were significantly reduced in recipients treated with imetelstat when compared to vehicle control (Fig. 2c-f).Furthermore, imetelstat treatment significantly reduced splenic AML donor chimerism (Fig. 2g).We next assessed AML surface marker expression associated with leukemia-initiating activity [11][12][13] (Fig. 2h).Imetelstat significantly diminished the CD34 + CD38 − leukemic stem cell-enriched splenic AML cell population (Fig. 2i).In normal human hematopoiesis using two independent CD34-enriched cord blood xenografts in NSG recipients, the effects of imetelstat were predominantly seen in B lymphocytes with relative preservation of the myeloid and stem cell populations (Extended Data Fig. 2a-q).
We next aimed to compare imetelstat responses to those obtained with standard induction chemotherapy (cytarabine plus anthracycline) in AML PDX from 20 individual samples from patients with AML in an independent cohort using NOD.Rag1−/−Il2Rg−/−/ hIL3,CSF2,KITLG (NRGS) recipients 14 .Imetelstat matched the similar benefit conveyed by standard chemotherapy (139 d) comparative to 104 d in the vehicle control group and this was accompanied by significant reductions in peripheral blood AML burden (Extended Data Fig. 3a,b); however, the individual samples from patients with AML could be classified into either preferential imetelstat or preferential chemotherapy responders (Extended Data Fig. 3c).Preferential responses to imetelstat when compared to standard induction chemotherapy were associated with baseline mutations in NRAS, JAK2 or GLI1 (Extended Data Fig. 3d).

A CRISPR/Cas9 screen to identify key effectors of imetelstat
To investigate the mechanism of action of imetelstat in AML in an unbiased manner, we applied the Brunello guide RNA (gRNA) library 15 as a positive selection screen to identify gene knockouts that confer resistance to imetelstat.We used NB4 cells as these demonstrated highest sensitivity to imetelstat when compared to 13 other human hematopoietic cell lines (Extended Data Fig. 7a).Half-maximum inhibitory concentration (IC 50 ) values strongly depended on cell density, demonstrating the presence of an imetelstat inoculum effect (Extended Data Fig. 4a) 16 .Cas9-expressing NB4 cells transduced with the Brunello library or untransduced controls were cultured in the presence of imetelstat concentrations that resulted in substantial cell death (IC98) of the untransduced control cultures but allowed the enrichment of imetelstat-resistant cells in Brunello-transduced cultures over a time course of 45 d in culture (Extended Data Fig. 4b).Vehicle or mismatch control-treated NB4 cells grew exponentially throughout the course of treatment (Extended Data Fig. 4c).Specific guide RNAs were selectively enriched in Brunello-transduced imetelstat-resistant compared to vehicle-treated and input control cultures (Extended Data Fig. 4d-g).
We next aimed to validate the most significant hits identified (FADS2 and ACSL4) using single guide RNA (sgRNA)-mediated editing in the NRAS wild-type expressing NB4 and MV411 and the NRAS-mutant KO52 (p.G13R) and TF1 (p.Q61P) AML cell lines.Editing was confirmed by TIDE analysis 17 and reduced protein levels (Extended Data Fig. 4h-j).
We performed competition assays to confirm that loss-of-function editing of FADS2 or ACSL4 confers competitive growth advantage under imetelstat pressure in all AML cell lines analyzed (Fig. 3c-f).The observed effects were target-specific as a competitive outgrowth under imetelstat pressure was not observed when CD33 (predicted to have neutral effects on cell functions 18 ) knockouts or empty vector controls were used (Fig. 3c-f).
These results demonstrate that loss-of-function editing of FADS2 or ACSL4 confers competitive growth advantage under imetelstat pressure, identifying ACSL4 and FADS2 as mediators of imetelstat efficacy in AML.

Imetelstat is a potent inducer of ferroptosis
ACSL4 and FADS2 encode key enzymes regulating polyunsaturated fatty acid (PUFA)-containing phospholipid synthesis.FADS2 is a key enzyme in a lipid metabolic pathway that converts the essential fatty acids linoleate (18:2n6) and α-linolenate (C18:3n3) into long-chain PUFAs 19 .Targeted lipidomics analysis on 593 lipid species and their desaturation levels 20 demonstrated clear effects of imetelstat treatment and FADS2 editing on the cellular lipidome, with imetelstat-treated empty vector control AML cells showing greatest difference to vehicle-treated empty vector control and FADS2-edited cells.(Extended Data Fig. 5a).Moreover, we found a significant enrichment of phospholipids containing fatty acids with three unsaturated bonds in imetelstat-treated compared to vehicle control-treated NB4 cells and this enrichment of lipid desaturation was diminished by FADS2 editing (Fig. 4a and Extended Data Fig. 5b).Moreover, imetelstat increased the levels of phospholipids with triglycerides and reduced the levels of phospholipids containing cholesteryl esters and ceramides when compared to vehicle control in an FADS2-dependent manner (Extended Data Fig. 5c).Taken together, these data demonstrate imetelstat-induced PUFA phospholipid synthesis in an FADS2-dependent manner.
ACSL4 has been previously identified as key regulator of ferroptosis 21 .Ferroptosis is a form of cell death that is driven by an imbalance between the production of reactive oxygen species (ROS) during lipid peroxidation and the antioxidant system and may involve autophagic processes depending on the trigger 22 .A hallmark of ferroptosis is lipid peroxidation, the oxidation of PUFA-containing phospholipids that occurs via a free radical chain reaction mechanism 22 .Cancer therapies can enhance ferroptosis sensitivity via lipid remodeling that increases levels of peroxidation-susceptible PUFA-containing phospholipids 23 .
To test whether imetelstat induces lipid peroxidation, we treated various AML cell lines with C11-BODIPY, a fluorescent fatty acid probe that changes its emission spectrum from red to green upon oxidation.In all four AML cell lines tested, imetelstat treatment resulted in a significant increase in mean fluorescence intensity (MFI) of the oxidized fatty acid probe, demonstrating that imetelstat induces lipid peroxidation in AML cells in vitro (Fig. 5b).We next assessed whether also ROS levels were affected by imetelstat.Using CellROX Green to measure ROS production, we found that its MFI was increased by imetelstat and this increase was diminished when the lipid ROS scavenger ferrostatin-1 was added during the incubation step with CellROX Green, demonstrating that imetelstat increases predominantly lipid ROS levels in AML cell lines in vitro (Fig. 5a).Both lipid peroxidation and lipid ROS production were significantly diminished in ACSL4 or FADS2 loss-of function edited AML cell lines, demonstrating that imetelstat-induced lipid peroxidation and lipid ROS production are dependent on functional FADS2 and ACSL4 in vitro (Fig. 4b,c).Pharmacological inhibition of ferroptosis using the lipid ROS scavengers ferrostatin-1 and liproxstatin-1 diminished imetelstat efficacy in all AML cell lines tested (Extended Data Fig. 6).Moreover, the iron chelator deferoxamine mesylate, the 5-lipoxygenase inhibitor zileuton and menadione diminished imetelstat-induced cell death in a substantial proportion of AML cell lines tested (Extended Data Fig. 6).
In AML PDXs in vivo, imetelstat-induced lipid peroxidation was associated with increased ACSL4 expression (Fig. 5c).To investigate whether lipid ROS and lipid peroxidation are essential for imetelstat's mechanism of action in AML PDXs in vivo, we treated AML PDXs with either vehicle control, imetelstat (15 mg kg −1 three times per week), liproxstatin-1 (15 mg kg −1 twice daily) or the combination of both imetelstat and liproxstatin-1 for 2 weeks.Imetelstat-driven lipid peroxidation and ROS production were prevented by liproxstatin treatment (Fig. 5d,e).In vivo liproxstatin treatment diminished imetelstat efficacy in PDXs as measured by peripheral blood AML burden (Fig. 5f).
Taken together, these data provide evidence that imetelstat is a potent inducer of ferroptosis through ACSL4-and FADS2-mediated alterations in PUFA metabolism, excessive lipid peroxidation and oxidative stress.

Lipophagy precedes imetelstat-induced ferroptosis
By integrating transcriptomics and functional genetics, we aimed to investigate the mechanism by which imetelstat induces ferroptosis.We performed an overlay of the in vivo AML PDX RNA-seq datasets from imetelstat and vehicle-treated mice with the Brunello library CRISPR/Cas9 knockout screen data (cutoff criteria of RNA-seq adjusted  denote statistically significant comparisons based on distinct 95% CI on mCherry chimerism AUC between imetelstat and PBS-treated cultures.95% CI (lower limit, upper limit): NB4 FADS2 PBS (444.4,456.7) versus imetelstat (771.9, 795.9);ACSL4 PBS (320.
Recent independent work demonstrated the capacity of imetelstat to form G-quadruplex structures in vitro and this capacity is attributed to the presence of a triple G-repeat (GGG) in its sequence 25 .These insights prompted us to obtain an additional mismatch control harboring a similar triple G-repeat, but containing enough mismatches to prevent efficient binding to telomerase (Extended Data Fig. 7a).Using an antibody raised against (T4G4)2 intermolecular G-quadruplex DNA structures [26][27][28] , we found that imetelstat or GGG-containing mismatch but not mismatch 1 significantly interfered with endogenous DNA G-quadruplex structures (Extended Data Fig. 7b).In a panel of 14 human hematopoietic cell lines, GGG-containing mismatch control and imetelstat demonstrated similar efficacies in the majority of AML cell lines tested (Extended Data Fig. 7a).Moreover, GGG-containing mismatch was similarly effective as imetelstat in increasing ROS levels when compared to vehicle control (Extended Data Fig. 7c).Ferrostatinor deferoxamine mesylate-mediated inhibition of ferroptosis rescued both imetelstat as well as GGG-mismatch-induced cell death (Extended Data Fig. 7d).We next compared the preclinical efficacy of imetelstat with GGG-mismatch and mismatch 1 in an NRAS/KRAS-mutant AML PDX model (RCH-11).In this model, GGG-mismatch was also effective in reducing AML burden (Extended Data Fig. 7e).
In addition to binding telomeric G-quadruplexes, vimentin has long been established as structural component of lipid droplets regulating their biogenesis and stability.
Lipid droplets can undergo selective autophagy (lipophagy) that can result in the induction of ferroptosis 29 .We hypothesized that imetelstat-induced PUFA phospholipid synthesis, oxidation and ferroptosis can result from lipophagy.Vimentin was highly expressed at protein level in AML cells in vitro (Fig. 6b) and loss-of-function editing of vimentin resulted in a modest competitive growth advantage of AML cells under imetelstat pressure (Fig. 6c).We next assessed lipophagy using C12-BODIPY, a fluorescent fatty acid probe for lipid droplets, in conjunction with the late endosomal marker LAMP1 (ref.30).Imaging flow cytometry revealed significantly increased colocalization of lipid droplets with the late endosomal marker LAMP1, indicating increased lipophagy (Fig. 6d).To test whether pharmacological inhibition of     lipophagy can prevent imetelstat-induced ferroptosis, we cultured AML cells in the presence of imetelstat combined with chloroquine, which inhibits lysosomal hydrolases by increasing the pH and thus lipophagy.Notably, in all AML cell lines tested, chloroquine diminished imetelstat-induced cell death (Fig. 6e).These results provide evidence for a role of lipophagy-induced ferroptosis in imetelstat's mechanism of action in AML via impaired lipid droplet homeostasis due to G-quadruplex mediated interference with the structural components of lipid droplets.

Oxidative stress signatures distinguish sustained responders
We next aimed to identify biomarkers of imetelstat response and resistance.Improved survival in imetelstat-treated AML PDXs correlated with significantly reduced engraftment and disease burden; however, there were clear differences in the magnitude and duration of individual responses (Extended Data Fig. 8).To understand determinants of imetelstat response, we allocated each individual AML patient sample into either sustained, intermediate or poor imetelstat response categories based on the individual effect of imetelstat on AML burden measured in peripheral blood over time (Extended Data Figs. 9 and 10a).All ELN2017 prognostic risk categories were represented in each imetelstat response group, suggesting that the effects observed were not solely explained by favorable disease (Extended Data Fig. 10b).In addition, cytogenetics, sex, age, FLT3-ITD allelic ratio and TERT messenger RNA expression levels at baseline seemed similar among imetelstat response groups (Extended Data Fig. 10c-h).
We next aimed to identify genetic biomarkers of response and resistance to imetelstat therapy by analyzing the data from individual samples from patients with AML at baseline that were generated by genomic sequencing using a comprehensive panel of 585 genes frequently mutated in hematological malignancies 31 (Extended Data Fig. 9c).Oncogenic mutations in genes annotated in signaling or cell adhesion/metabolism were trend-wise more frequently observed in sustained compared to poor responders to imetelstat (Fig. 7a and Extended Data Fig. 9c).
Mutant NRAS was associated with enhanced responses to imetelstat therapy.This was evidenced by reduced AML burden and improvement in survival when compared to wild-type NRAS containing AML PDXs (Fig. 7b,c).Moreover, variant allelic frequencies of the relevant NRAS mutations inversely correlated with AML burden in imetelstat-treated AML PDXs (Extended Data Fig. 10i).Additionally, gene set enrichment analysis (GSEA) of the RNA-seq data obtained from the individual samples from patients with AML at baseline revealed that sustained responders had a positive enrichment of gene signatures associated with translation/viral infection and negative enrichment for gene signatures associated with cell cycle, antiviral immunity, transmembrane transport and heme scavenging compared to poor responders to imetelstat (Fig. 7d and Extended Data Fig. 9a).Hallmark signatures revealed significant enrichment of gene sets annotated as apoptosis, interferon-α response, DNA repair, TP53 pathway, peroxisome, fatty acid metabolism and ROS pathway in sustained compared to poor responders to imetelstat (Fig. 7e).
We next examined whether baseline telomere length could predict imetelstat response.Telomere length was determined by telomere restriction fragment (TRF) analyses and peak telomere lengths varied between 2.7 and 12 kb among individual AML patient samples (Fig. 7f and Extended Data Fig. 10j).Five out of the 30 AML patient samples contained multiple subclones with distinct telomere lengths (Fig. 7f and Extended Data Fig. 10j).Overall, there was no correlation between baseline telomere length and imetelstat response (Fig. 7f).
These data demonstrate that imetelstat is effective in a large proportion of AML PDXs.Furthermore, sustained responses to imetelstat are independent of baseline telomere length and are associated with marked improvements in survival, mutant NRAS and baseline molecular signatures annotated as oxidative stress.

Oxidative stress induction sensitizes AML PDX to imetelstat
The finding that responses to imetelstat are associated with baseline molecular signatures annotated as oxidative stress and that the mechanism of action of imetelstat features ROS-mediated ferroptosis led to the hypothesis that oxidative stress induction can sensitize to imetelstat therapy.
Standard induction chemotherapy composed of cytarabine and an anthracycline is a potent inducer of ROS 32 .To test whether oxidative stress-inducing therapy can sensitize AML cells to imetelstat treatment, we pretreated AML cell lines with oxidative stress-inducing standard induction chemotherapy (cytarabine in combination with doxorubicin) and subsequently switched to imetelstat treatment.Standard induction chemotherapy significantly increased ROS levels in a dose-dependent manner that led to augmented cell death in AML cell lines (Fig. 8a,b).
In a pilot study using an NRAS wild-type AML PDX model (poor responder to imetelstat monotherapy; RBWH-44), a single dose of standard induction chemotherapy followed by a single dose of imetelstat resulted in significantly increased ROS levels in AML patient-derived cells in PDXs in vivo (Fig. 8c).At this early time point, lipid peroxidation was not significantly different between the treatment groups (Fig. 8c); however, after a complete cycle of induction chemotherapy followed by prolonged treatment with imetelstat consolidation therapy, both lipid peroxidation and ROS levels were significantly increased (Fig. 8d).
Finally, as a proof of concept in vivo, we sequentially administered oxidative stress-inducing standard induction chemotherapy before imetelstat in a diverse PDX cohort from 20 distinct AML patient samples   8f).AML burden was significantly reduced in the combination therapy group when compared to either monotherapy or vehicle-treated control groups (Fig. 8g).
These data demonstrate that the rational sequencing of imetelstat and chemotherapy, using standard induction chemotherapy to induce oxidative stress and sensitize AML cells to imetelstat-induced lipid peroxidation and ferroptosis, results in significantly improved disease control of AML (Fig. 8h).

Discussion
Imetelstat is a first-in class telomerase inhibitor with clinical efficacies in hematological myeloid malignancies, including essential thrombocythemia, myelofibrosis and lower-risk myelodysplastic syndromes [6][7][8] .The efficacy of imetelstat in AML and its mode of action have remained    Ferroptosis is a recently discovered type of non-apoptotic regulatory cell death that relies on the balance of the production of ROS during lipid peroxidation and the antioxidant system and it is generally characterized by three hallmarks: (1) loss of peroxide repair capacity through GPX4; (2) availability of redox-active iron; and (3) oxidation of polyunsaturated fatty acid-containing phospholipids 22,33 .The experiments performed in this study have revealed evidence for imetelstat directly affecting the third hallmark of ferroptosis; the increased synthesis and subsequent oxidation of PUFA phospholipids.In AML PDXs in vivo, imetelstat-induced lipid peroxidation is associated with significantly increased ACSL4 expression.In human AML cell lines, imetelstat treatment significantly increased lipid ROS levels that preceded massive cell death.Treatment with the lipid ROS scavengers ferrostatin-1 or liproxstatin-1 rescued imetelstat-induced cell death in all AML cell lines tested.In addition, pharmacological iron chelation using deferoxamine mesylate, 5-lipoxygenase inhibition using zileuton or menadione supplementation were able to prevent imetelstat-induced cell death in a substantial proportion of AML cell lines tested.In contrast to ferrostatin-1 and liproxstatin-1, higher concentrations of deferoxamine mesylate were detrimental for AML cells, suggesting that iron availability is crucial for AML cell survival at a level specific for each cell line.Iron metabolism is altered in AML at the cellular and systemic level and elevated iron levels help to maintain the rapid growth rate of AML cells by activating ribonucleotide reductase that catalyzes DNA synthesis in an iron-dependent manner 34 .Notably, imetelstat has shown efficacy in patients with pathology featuring ringed sideroblasts 7,8,35 , a cellular morphological abnormality that is defined by iron-laden granules in mitochondria surrounding the nucleus, further supporting the role of iron-dependent cell death.
Our functional genetic experiments using Brunello library CRISPR/Cas9 editing have provided further evidence that imetelstat restricts leukemic progression via ferroptosis, revealing a closely related functional network of seven genes.One of the identified targets, ACSL4, has previously been identified as a key regulator of ferroptosis sensitivity through the shaping of the cellular lipid composition 21 .We have functionally validated the most significantly enriched targets, FADS2 and ACSL4.The canonical role of FADS2 in fatty acid metabolism is the catalysis of the desaturation of linoleic and α-linolenic acid to long-chain PUFAs 36,37 .Lipidomics analysis has revealed increased levels of phospholipids containing triglycerides and also increased levels of phospholipids containing PUFAs with three unsaturated bonds in imetelstat-treated AML cells in an FADS2-dependent manner.Using a fluorescent sensor, we have confirmed that imetelstat stimulates lipid peroxidation.These data demonstrate imetelstat-induced alterations in fatty acid metabolism that promote the formation of substrates for lipid peroxidation.Of note, in some lung cancer cell lines, FADS2 activation is associated with ferroptosis suppression 38 .This dichotomy may be explained by the fact that in some cancer cells, FADS2 enables the desaturation of palmitate to sapienate (cis-6-C16:1) as part of an alternative desaturation pathway, thus potentially reducing the levels of monounsaturated fatty acids and ultimately PUFA-containing phospholipids as substrates for lipid peroxidation 39 .
G-quadruplexes are recognized by and regulate the activity of many proteins involved in telomere maintenance, replication, transcription, translation, mutagenesis and DNA recombination 40 .The recognition of G-quadruplexes can be dictated by R-loops that show a close structural interplay and can modulate responses involving DNA damage induction, telomere maintenance and alterations in gene expression regulation 41 .Notably, G-quadruplex/R-loop hybrid structures were detected in vitro in the human NRAS promoter and at human telomeres [42][43][44][45] .R-loop binders and epigenetic R-loop readers have been recently linked to altered fatty acid metabolism and ferroptosis [46][47][48] .Moreover, constitutively activated RAS/MAPK signaling downstream of mutant NRAS is associated with enhanced sensitivity to ferroptosis 33,49,50 ; however, the activity of this pathway alone is unlikely to be the sole determinant of ferroptosis sensitivity 51,52 .Our integrative analysis of transcriptomics with functional genetics data has identified imetelstat target candidates that were recently discovered as G-quadruplex-binding proteins (VIM and LMNA) 29 .Moreover, VIM and LMNA have been characterized as proteins directly interacting with lipid droplets 53,54 .Recent independent work has provided evidence for a role of lipid droplets in ferroptosis.In hepatocytes, the degradation of intracellular lipid droplets via autophagy (lipophagy) promotes RSL3-induced ferroptosis by decreasing lipid storage that subsequently induces lipid peroxidation 29 .Our imaging flow cytometry analysis demonstrates significantly increased colocalization of markers for lipid droplets and late endosomes, proposing imetelstat-induced lipophagy as trigger for ferroptosis in AML.
Using a newly established, comprehensive AML patient-derived xenograft resource that reflects the overall genetic abnormalities found in large clinical cohorts, we demonstrated a proof of concept for the sequential administration of standard induction chemotherapy followed by imetelstat consolidation to induce oxidative stress and sensitize AML patient samples to imetelstat treatment.This approach was able to cause significant delay or prevention of AML relapse.The efficacy of sequential therapy suggests that imetelstat may be particularly useful in preventing relapse after chemotherapy, for example, as a maintenance therapy.Recently, maintenance therapy with oral CC486 has shown a survival benefit in AML; however, there is no survival plateau and therefore, most patients still relapse and die of their disease 55 .A substantial proportion of AML patient samples tested (14 out of 30 samples) were classified as sustained responders to imetelstat monotherapy and are characterized by genetic lesions in genes involved in cell adhesion, metabolism and signaling, with the most striking result obtained for NRAS.NRAS is the fourth most commonly observed gene with driver mutations in adult AML 2 .Moreover, AML cell clones harboring mutant NRAS arise in some patients relapsing on targeted therapies, particularly FLT3 inhibition (crenolanib 56 and gilteritinib 57 ) and BCL2 inhibition in some cases (venetoclax 58,59 ).The demonstrated sustained responses to imetelstat in NRAS-mutant AML patient samples raise the possibility that imetelstat may be used as salvage therapy or possibly in combination with FLT3 inhibitors or venetoclax to prolong remission and prevent relapse.
In conclusion, imetelstat is a potent inducer of ferroptosis that effectively diminishes AML burden and delays relapse following oxidative stress-inducing chemotherapy.
Clinical trials will address the efficacy of imetelstat in AML and may focus on this compound as a consolidation strategy for preventing relapse or potentially together with targeted therapies to improve outcomes in patients with AML.

Methods
Our research complies with all relevant ethical regulations, including QIMR Berghofer human research ethics committee protocol P1382 (HREC/14/QRBW/278) and QIMR Berghofer animal research ethics committee protocol A11605M.Animals were monitored daily and immediately killed based on the scoring criteria detailed below.

Xenograft transplantation experiments
AML samples were obtained from patients, after informed consent in accordance with the Declaration of Helsinki.Ficoll density gradient was used to recover viable mononuclear cells.Viably frozen AML cells were thawed and CD3-depleted with biotinylated anti-human CD3 (SK7) and biotin-binder Dynabeads (Invitrogen) and subsequently injected via the lateral tail vein into 2.8 Gy irradiated (24 h before transplant) female NSGS or NRGS recipients (6-8 weeks old).For normal hematopoiesis studies, viable mononuclear cells were isolated from cord blood samples (provided by the Wesley-St Andrew's Research Institute Tissue Bank with appropriate ethics approval) by Ficoll density gradient, CD3-depleted as above and subsequently enriched for CD34 + cells using the human CD34 MicroBead kit (130-046-702 MACS Miltenyi Biotec).A total of 56,000 cells (donor 1) or 212,500 cells (donor 2) were injected via the lateral tail vein per irradiated female NSG recipient (6-8 weeks old).

Drug treatment studies
NSG, NSGS or NRGS mice were treated with 15 mg kg −1 imetelstat (GRN163L), mismatch controls (mismatch 1 also referred to as MM1 or GRN140833; mismatch 2 also referred to as GGG-mismatch, MM2 or GRN142865) or vehicle control (PBS) via the i.p. route for the period of time specified in the respective experiment three times per week, at least every 72 h.For standard induction chemotherapy studies, cytarabine (AraC; 1 g in 10 ml isotonic water; Pfizer) and doxorubicin (Doxo; 50 mg in 25 ml saline; Pfizer) were freshly diluted with saline (sodium chloride 0.9% for irrigation; Baxter) to achieve a final concentration of 50 mg kg −1 body weight AraC or 1.5 mg kg −1 body weight Doxo in 200 μl total injection volume per recipient.Both AraC and Doxo were co-delivered i.v.(in the same syringe) on days 1 to 3, followed by i.v.injection of AraC alone on days 4 and 5, each in strict 24-h intervals.For chemotherapy plus imetelstat combination studies, the first imetelstat injection was administrated 1 d after the standard induction chemotherapy cycle was completed.For in vivo liproxstatin-1 treatment studies, liproxstatin-1 (SEL-S7699; Jomar Life Research) was dissolved in dimethylsulfoxide (7.9 mg in 400 μl) and then diluted with 2.44 ml 0.9% NaCl (saline).Liproxstatin-1 (15 mg kg −1 ) was administered by i.p. injection via a 27 G insulin needle twice daily for 2 weeks (200 μl per recipient).

Blood analysis
Blood was collected into EDTA-coated tubes and analyzed on a Hemavet 950 (Drew Scientific).PB smears were stained with Wright-Giemsa according to the manufacturer's protocol (BioScientific).

Histology
Tissues were fixed in 10% neutral buffered formalin, embedded in paraffin and stained with hematoxylin and eosin.Images of histological slides were obtained on a ScanScope FL (Aperio).

Terminal restriction fragment analysis
TRFs were obtained from genomic DNA by complete digestion with the restriction enzymes HinfI and RsaI.TRFs were separated by pulsed-field gel electrophoresis.Gels were dried, denatured and subjected to in-gel hybridization with a γ-[32P]-ATP-labeled (CCCTAA)4 oligonucleotide probe.Gels were washed and the telomeric signal visualized by phosphorimage analysis.TRFs were processed by ImageJ 1.52a analysis software to quantitate mean telomere length.

Telomere length qPCR
Samples were purified using the DNeasy Blood and Tissue kit (QIAGEN).DNA isolation was performed as described previously 61 , including degassing of buffers and supplementation with 50 μM of phenyl-tert-butyl nitrone to minimize oxidative damage.Telomere length was assessed using qPCR [62][63][64] .The Supplementary Note contains further details of the procedure.

Cell lines
The Supplementary Note contains purchasing details of the human cell lines used in this study.All cell lines were authenticated by STR profiling at an early passage before the first culture experiment, performed by the QIMR Berghofer Analytical Core facility.None of the cell lines used has been known as misidentified cell lines according to v.12 of the cross-contamination database maintained by the International Cell Line Authentication Committee (https://iclac.org/databases/cross-contaminations/).All cell lines tested negative for Mycoplasma during regular monthly testing using the biochemical MycoAlert™ Mycoplasma Detection kit (Lonza) by QIMR Berghofer core facility.

Imaging flow cytometry
Lipophagy was detected by assessing colocalization of C12-BODIPY and LAMP1 using a previously published method with modifications 65 .
In detail, 1 × 10 6 cells were collected and washed in warm PBS (without FCS).Cells were then resuspended in warm RPMI (without FCS) containing 200 ng C12 FL BODIPY (Thermo Fisher Scientific) per ml and incubated for 30 min at 37 °C.Cells were then washed in 9 ml wash buffer (PBS + 2% FCS + 1 mM EDTA) and subsequently fixed and permeabilized using a FIX & PERM Cell Permeabilization kit (GAS-004; Invitrogen) and incubated with Alexa Fluor 647 anti-human CD107a (LAMP1; BioLegend; dilution 1:100) according to the manufacturer's instructions.Cells were subsequently washed and resuspended in wash buffer with 0.2 mg ml −1 Hoechst 33342 (Invitrogen).The acquisition was performed using an Amnis ImageStream X Mark II Imaging Flow and data were analyzed with IDEAS (Image Data Exploration and Analysis Software).

Confocal microscopy
Cytospins were fixed and permeabilized with methanol:acetone (prechilled) at a ratio of 1:1 for 10 min at room temperature, then washed twice with cold PBS and once with room temperature PBS (5 min each).Cytospins were then incubated with 1% BSA-PBS at room temperature for 1 h, washed three times in PBS for 5 min each and then incubated in primary antibody (anti-human Vimentin XP rabbit monoclonal antibody Alexa Fluor 488 conjugate, Cell Signaling; cat.9854; clone D21H3) or isotype control (rabbit monoclonal antibody IgG XP Isotype Control Alexa Fluor 488 conjugate, Cell Signaling; cat.2975; clone DA1E) at a dilution of 1:400 in 1% BSA-PBS for 1 h at room temperature or O/N at 4 °C.Cytospins were then washed three times in PBS for 5 min each.Coverslips were mounted in pro-long DAPI Gold.Images were acquired on a Zeiss 780-NLO confocal microscope.

CRISPR/Cas9 editing
The Brunello genome-wide gRNA library contains 76,441 gRNAs targeting 19,114 genes and was obtained from Addgene (cat.73178) 66,67 .Streptococcus Pyogenes Cas9 and blasticidin resistance construct expressed from an EFS promoter (pFUGWb) was obtained from Addgene (lentiCas9-Blast, plasmid #52962) 68 .Lentivirus containing the Brunello library was generated and used to transduce NB4 cells.The Supplementary Note contains detailed descriptions of the CRISPR/ Cas9 screen as well as sgRNA-mediated editing approaches.

Mutational sequencing
Genomic alterations were profiled using the HemePACT assay (integrated mutation profiling of actionable cancer targets related to hematological malignancies) 31 .This assay uses solution phase hybridization-based exon capture and massively parallel DNA sequencing to capture all protein-coding exons and select introns of 585 actionable cancer related genes.Samples were molecularly barcoded, to allow optimal cost efficiency during the capture process as well as at the sequencing step.Then, 250 ng of genomic DNA was used for library construction.Pools of 12 samples equimolarly mixed were sequenced at the Genomics Core Laboratory at MSKCC, in one lane of a HiSeq 2500, using the SBS chemistry for paired-end 100/100 reads.The average coverage was greater than 400-fold, with a minimum of 99% of the targeted sequences covered 30-fold.Reads were aligned to the reference human genome (hg19) using the Burrows-Wheeler alignment tool 69 .Local realignment and quality score recalibration were conducted using the Genome Analysis Toolkit (GATK) according to GATK best practices 70 .Somatic alterations were identified (single-nucleotide variants, small insertions/deletions (indels) and copy number alterations).Single-nucleotide variants were identified using UnifiedGenotyper and mu Tect 71 .All samples were paired (AML/ healthy) and candidate genomic alteration were reviewed manually in the Integrative Genomics Viewer 72 .

RNA sequencing
RNA was isolated from a maximum of 0.5 × 10 6 cells using the QIAGEN RNeasy Micro kit according to the manufacturer's instructions.Total RNA (100 μg) was used for next-generation sequencing and prepared according to the NEBNext Ultra II RNA Library Prep kit for Illumina (New England Biolabs, NEB; cat.E7770S).The Supplementary Note contains Article https://doi.org/10.1038/s43018-023-00653-5further detailed descriptions of the method for library preparation.Libraries were sequenced using a high-output, single-end, 75 cycle (v.2) sequencing kit on the Illumina NextSeq 550 platform.Reads were trimmed for adaptor sequences using Cutadapt (v.1.11)and aligned using Spliced Transcripts Alignment to a Reference (STAR) (v.2.5.2a) 73 to the GrCH37 assembly using the gene, transcript and exon features of Ensembl (release 70) gene model.Expression was estimated using RNA-seq by Expectation Maximization (RSEM) (v.1.2.30).Transcripts with zero read counts across all samples were removed before analysis.Normalization of read counts was performed by dividing by 1 million reads mapped to generate counts per million, followed by the trimmed mean of M-values method from the edgeR package (v.3.14.0) 74 .For the differential expression analysis, reads were filtered but not normalized, as edgeR performs normalization (library size and RNA composition) internally.For the differential expression (DE) analyses, the glmFit function was used to fit a negative binomial generalized log-linear model to the read counts for each transcript.Using the glmLRT function, we conducted transcript-wise likelihood ratio tests for each genotype comparison.Principal-component analysis was also performed on all DE transcripts with FDR < 0.05.GSEA of transcriptomics data was performed using GSEA (v.4.1.0)from the Broad Institute 75 .P values were generated from 1,000 gene set permutations, excluding gene sets with more than 3,000 genes or fewer than five genes against custom made gene sets and the Broad Institute's Hallmark database.

Lipidomics
Targeted lipidomics was performed on a 1290 Infinity II UHPLC coupled to a 6470 QQQ mass spectrometer via AJS ESI source (Agilent) in positive ionization mode, using a scheduled multiple reaction monitoring (MRM) method 76 .The MRM transition list contained 20 lipid classes and 593 lipid species (excluding internal standards CUDA and SPLASH Lipidomix).Skyline-daily software was used for lipid species assignment 77 .A lipid set enrichment analysis was performed by ranking fold changes, calculating enrichment scores and estimating the significance of enrichment using a permutation algorithm 78 .The Supplementary Note contains detailed descriptions of the procedure.

Statistical analyses
Unless otherwise stated, statistical analyses were carried out using GraphPad Prism v.9.4.0.Microsoft Excel for Mac v.16.75.2 was used to re-calculate those P values < 0.0001 obtained from GraphPad Prism with higher precision.JMP Pro v.17 was used to calculate statistics related to the AML PDX trials, including PB donor chimerism AUC values and testing of normal distributions.

Statistics and reproducibility
Study design was based on sample sizes that proved to be adequate in previous experiments using similar approaches and thus no statistical methods were used to predetermine sample sizes for this study 4,10 .The PDX trials were subject to randomization after transplantation by independent technicians without the involvement of the researcher who had performed the transplantations.The investigators were not blinded to allocation during experiments and outcome assessments; however, unbiased PDX monitoring and scoring were performed by independent technicians not intellectually involved in the study.For cell culture experiments, samples were allocated equally to ensure that covariates were identical between the compared groups.The investigators performing cell culture experiments were not blinded during allocation and outcome assessment.Data distribution was assumed to be normal, but this was not formally tested in each experiment.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Article
https://doi.org/10.1038/s43018-023-00653-5 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material.If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.

c
Vehicle

Fig. 4 | 2 P
Fig.4| Imetelstat is a potent inducer of ferroptosis.a, Lipid desaturation analysis of FADS2-edited (FADS2-sg1 and FADS2-sg2) or non-edited (empty vector control) NB4 cells treated with imetelstat (4 μM at a seeding density of 2.5 × 10 5 cells per ml culture) or vehicle control (PBS) for 24 h.The graph depicts the median log 2 FC of the number of total unsaturated bonds in lipid species in the respective comparisons outlined in the legend.Shading represents the 95% CI. n = 3 replicates from distinct cell passages and independent experiments.b,c, CellROX Green (b) and C11-BODIPY (c) analysis in ACSL4edited (n = 4 independent gRNAs), FADS2-edited (n = 2 independent gRNAs) or non-edited (n = 2 independent replicates, Cas9, empty vector) NB4 or MV411 cell lines treated with imetelstat (4 μM) or PBS.Time points of analysis were 24 h (NB4) and day 4 (MV411).Three technical replicates per condition were pooled.Data are presented as mean ± s.e.m.One-way analysis of variance (ANOVA) was used and adjusted for multiple comparisons.NB4, P < 1× 10 −4

Extended Data Fig. 4 |
| Comparative analysis of AML PDX responses to imetelstat versus standard induction chemotherapy.a, Median survival was 104 (vehicle control (Saline / PBS) -treated PDX; black; n = 120 PDX) versus 139 (chemotherapy-treated PDX; Arac+Doxo; blue; P = 1.7 × 10 −3 ; n = 120 PDX) and 139 (imetelstat-treated PDX; Saline / Imetelstat; red; P = 3.3 × 10 −3 ; n = 120 PDX) according to Gehan-Breslow-Wilcoxon (two-sided Kaplan-Meier analysis).b, AML burden quantified as peripheral blood donor chimerism per day.Statistics based on ordinary One-way-ANOVA adjusted for multiple comparisons.N = 120 PDX per treatment group.P = 1.57× 10 −2 (vehicle versus imetelstat).c, Imetelstat versus chemotherapy response calculation as log2FC of peripheral blood chimerism area under the curve per day in AML PDX.The red arrow indicates samples defined as preferential imetelstat responders, and conversely, the blue arrow indicates PDX defined as preferential chemotherapy responders.d, Cytoscape visualization of genes with mutations identified exclusively in preferential imetelstat responders (red) or chemotherapy responders (blue) at baseline.Identification of key mediators of imetelstat efficacy using a genome-wide CRISPR/Cas9 screen.a, Identification of an imetelstat inoculum effect by quantification of IC50 values in NB4 cultures seeded at different densities: IC50 = 0.2 microM (25,000 cells per ml); IC50 = 1.5 microM (75,000 cells per ml); IC50 = 7.2 microM (250,000 cells / ml); IC50 = 31.3microM (1,000,000 cells / ml).Dots represent mean IC50 values ± SEM from n = 2 independent experiments.b,c, Viability (b) and cell growth analysis (c) of Brunello-library-transduced NB4 cells cultured in the presence of vehicle control (PBS), mismatch control (MM1), or imetelstat compared to non-transduced NB4 cells over a time course of 45 days.The respective concentrations are indicated in the panel.N = 3 independent biological replicates per genotype and treatment condition.d,e, Next generation sequencing of DNA isolated from Brunello library-transduced NB4 cells harvested at day 45 in culture treated with vehicle control (PBS; black bars), mismatch control (MM1; blue bars), or imetelstat (red bars), and compared to DNA isolated from Brunello library-transduced NB4 cells before treatment as input control (gray bars): d, The number of different guide RNAs present (left y-axis) and percentage of guide RNA coverage (right y-axis);
elusive to date.By developing and utilizing a comprehensive AML PDX resource and human cell lines for genomics, transcriptomics and lipidomics approaches combined with functional genetic and pharmacological validation experiments, we demonstrate that imetelstat is a potent inducer of ferroptosis that effectively diminishes AML burden and delays relapse following chemotherapy.