Very Long Chain Fatty Acid Synthesis is a Targetable Dependency in RAS Pathway Mutant Leukemia

Abstract

Oncogenic RAS pathway mutations are associated with therapeutic resistance in acute myeloid leukemia, and identification of therapeutic vulnerabilities has been hindered by a lack of clinically relevant models and tractable ex vivo platforms. We utilize a bone marrow endothelial cell co-culture system to perform CRISPR screens on wild-type hematopoietic cells and isogenic leukemias with and without mutant Nras. We credentialed Elovl1, a very long chain fatty acid elongase, as a dependency in RAS pathway mutant leukemia using genetic and pharmacologic approaches. Metabolic and genetic studies in primary leukemias revealed that the fitness defect from Elovl1 loss reflects a mutant-specific dependency on de novo sphingolipid biosynthesis, specifically sphingomyelin production, as Sgms1 deletion phenocopies Elovl1 loss. Sphingomyelin-mediated generation of lipid rafts, key scaffolds for multiple signaling pathways, is essential to the survival of Nras-mutant AML cells. Our work leverages a new leukemia model to identify a targetable dependency in this treatment-refractory leukemia.

INTRODUCTION:

Acute myeloid leukemia (AML) is the most common acute leukemia seen in adults. Despite major advances in AML biology and therapeutics, overall survival remains low (5-year survival of 33%) as most patients ultimately relapse and succumb to disease(1). Oncogenic NRAS mutations, present in 10–30% of AML patients(2, 3), are one of the most consistent predictors of resistance to standard-of-care AML targeted therapies like venetoclax, IDH1/2 inhibitors, and FLT3 inhibitors(4–7). Ras-selective therapeutics also have limited utility in NRAS-mutant leukemias(8). Thus, there is an urgent need to characterize the role of oncogenic NRAS in leukemogenesis and identify new therapeutic vulnerabilities in this AML mutational subtype.

IDH mutations are common in AML(9, 10) and NRAS mutations are associated with resistance to IDH-targeted therapies along with other targeted therapies.(4, 5, 11, 12) We focused on the IDH1/2-mutant AML context as a useful system to investigate oncogenic NRAS and generated isogenic Idh-mutant leukemias to isolate the effects of Ras pathway activation. Because oncogenic Nras is particularly sensitive to gene dosage, available cell line or ectopic over-expression models fail to recapitulate critical aspects of Nras-driven neoplasms(13, 14). We therefore established a murine model of Idh1^R132H/Npm1^c/Nras^G12D (INR) co-mutant AML where each allele is expressed at endogenous levels. We compared our findings to Idh1^R132H/Npm1^c (IN) mutant AML which lacks additional signaling mutations. Employing these isogenic leukemia models enables a focused characterization of oncogenic Nras in clinically relevant genotypes, providing a foundation for further validation primary patient samples.

Functional genomic screens using CRISPR technology provide a powerful approach for uncovering mutant-selective dependencies. Previous screens in AML have employed cell lines or oncogene overexpression systems, as primary murine models were limited by the ability to maintain leukemogenesis ex vivo and insufficient cell divisions to identify depleted sgRNAs(15). We utilized an ex vivo bone marrow endothelial cell (BMEC) co-culture system previously developed by our lab and others(16–18) to perform focused CRISPR screens on our leukemia models. We leveraged this system to credential Elovl1 as a selective therapeutic vulnerability in oncogenic Nras mutant AML. Our work illustrates the power of using functional genomics in primary leukemia cells to uncover genotype-specific dependencies with therapeutic importance.

RESULTS:

Development of a Multigenic Murine AML Model to Evaluate Oncogenic Nras

We first aimed to establish Idh-mutant inducible genetically engineered mouse models (GEMMs) with and without oncogenic Nras. We generated Idh1^R132H/Npm1^c heterozygous co-mutant leukemia (IN) since this combination occurs frequently in patients,(19, 20) confers favorable prognosis,(21, 22) and lacks additional signaling mutations, allowing us to isolate the effects of oncogenic Nras. NPM1 is the most frequent co-occurring mutation in IDH1/2-mutant AML,(20) and although NRAS mutations are less common in this background, they nonetheless arise in a clinically significant subset of patients and confer resistance to IDH inhibitors and other mechanism-based AML therapeutics.(4, 5, 11, 12) We crossed the ubiquitously expressed tamoxifen (TAM) inducible Ubc:CreERT2 transgene(23) to mice containing a Cre inducible Lox-stop-Lox Idh1^R132H and a Cre-inducible Npm1^c frameshift mutant causing cytoplasmic localization.(24–26) Since oncogenic Nras is sensitive to gene dosage,(13) we generated Idh1^R132H/Npm1^c/Nras^G12D heterozygous triple mutant leukemia (INR, where R represents Nras^G12D) where Nras^G12D is induced at the endogenous locus. IN mice were crossed to Cre inducible Lox-stop-Lox Nras^G12D mice.(27) Bone marrow chimeras were established by transplanting IN or INR cells from donor mice into lethally irradiated primary recipient mice. After engraftment, alleles are activated with TAM to generate de novo leukemias. GEMM cells can be tracked with CD45.2 allelic markers to distinguish them from CD45.1 wild-type (WT) recipient mice. When recipient mice become moribund from leukemia, bone marrow (BM) was passaged to generate secondary recipients (Fig. 1A).

Figure 1: Oncogenic Nras promotes a distinct, differentiated leukemia.

A. Schematic representation of GEMM model development. CD45.2+ donor cells with indicated genotypes (IN=Idh1^R132H/Npm1^c and INR= Idh1^R132H/Npm1^c/Nras^G12D) are tail vein injected into lethally irradiated CD45.1+ recipient mice and dosed tamoxifen for allele activation (primary transplants). Bone marrow is passaged through additional CD45.1 recipients (secondary transplants). B. Kaplan-Meier survival analysis of primary transplant mice engrafted with cells from indicated genotypes (n=5–31 mice per genotype); logrank test p<0.0001. C. Kaplan-Meier survival analysis of secondary transplant mice (n=16–17 mice per arm); logrank test p=0.003. D. Peripheral blood white blood cell (WBC), hemoglobin (Hg), and platelet (Plt) counts of moribund IN and INR primary transplant mice and 20 week WT controls (WT n=21, IN n=22, INR n=20 mice, One-way ANOVA with Dunnett’s multiple test correction). E. H&E staining of bone marrow and spleen of IN and INR leukemic mice (Representative of n = 5). All images represented at total 400X magnification. Scale bar 20μm. F. IN and INR moribund mouse spleen weight (mg) with WT controls (n=5–20, One-way ANOVA with Dunnett’s multiple test correction). G. Bone marrow LSK (Lineage-Sca1⁺Kit⁺) cells as a percentage of Lineage- cells in IN and INR mice at survival end point (n=10–12 mice). H. Bone marrow LT-HSCs (Lineage⁻Sca1⁺Kit⁺CD48⁻CD150⁺) as a percentage of LSKs (n=10–12 mice). All data are mean ± SEM. *p<0.05, **p < 0.01, ***p < 0.001, ****p<0.0001

Oncogenic Nras Promotes a Distinct, Differentiated Type Leukemia

In the primary transplant setting, INR mice had significantly decreased overall survival (OS) compared to IN leukemias and WT mice (Median OS WT: 68.00wks, IN: 33.29wks, INR 5.86wks, p<0.0001; Fig. 1B). Both IN and INR GEMMs developed fully penetrant leukemia, with INR leukemias promoting significantly decreased survival in secondary recipients (Median OS IN: 22.84wks, INR: 15.14wks, p=0.003; Fig. 1C). INR mice uniformly developed leukocytosis, whereas some IN mice exhibited leukocytosis at end point and others had leukopenia. Mice engrafted with IN or INR cells consistently developed progressive anemia and thrombocytopenia (Fig. 1D). Histological analyses revealed morphologically immature sheets of blasts in IN and INR mice, consistent with leukemic infiltration (Fig. 1E). Both IN and to a larger extent, INR mice exhibited splenomegaly at end point (Fig. 1F). In the hematopoietic stem and progenitor cell (HSPC) compartment, INR leukemias had a significantly decreased percentage of LSKs (Lineage⁻Sca1⁺Kit⁺) compared to IN (Fig. 1G). In the LSK compartment, INR cells have relative expansion of MPPs (Lineage⁻Sca1⁺Kit⁺CD48⁺CD150⁻) at the expense of ST-HSCs (Lineage⁻Sca1⁺Kit⁺CD48⁻CD150⁻) and LT-HSCs (Lineage⁻Sca1⁺Kit⁺CD48⁻CD150⁺) (Fig. 1H, Supplementary Fig. 1, Supplementary Fig. 2A–2D). In the mature compartment, INR leukemias were characterized by elevated percentages of bone marrow (BM) Ly6C^hi and Ly6C^lo monocytes, decreased neutrophils, and decreased T cells relative to WT (Supplementary Fig. 2E–2J). IN leukemias were relatively more hypercellular at endpoint (Supplementary Fig. 2K) so only minor overall differences in absolute cell count were noted (Supplementary Fig. 2L–2V). INR leukemias were characterized by increased mature myeloid cell output in the periphery compared to WT (Supplementary Fig. 2W). Thus consistent with clinical data(28) and previous reports(6), expression of oncogenic Nras led to myeloid leukemias with a more differentiated monocytic phenotype. Both IN and INR leukemic mice had significantly elevated serum 2-hydroxyglutarate (2-HG), consistent with the known neomorphic activity of mutant Idh1 (Supplementary Fig. 2X). We isolated Kit⁺ and Kit⁻ cells from IN and INR leukemias and assessed their ability to engraft and give rise to disease. Kit⁺, but not Kit⁻ cells derived from IN and INR leukemias could propagate leukemias in secondary recipients, revealing the leukemia stem cell/propagating cell resides in this population (Supplementary Fig. 2Y, 2Z). These data show that both IN and INR can induce AML in vivo, and that INR leukemias had a distinct, differentiated myeloid phenotype similar to the phenotype observed in the clinical context in patients with NRAS-mutant AML.

CRISPR Screens in HSPCs Reveal Potential Therapeutic Targets

We next sought to use a functional genomic approach to identify mutant-specific dependencies in Nras-mutant AML. We established an ex vivo CRISPR screening platform using hematopoietic cells co-cultured with bone marrow endothelial cells (BMECs), previously employed by our lab to identify mutant specific dependencies in clonal hematopoiesis(16). We first evaluated whether AML cells co-cultured with BMECs maintain leukemogenicity. Both IN and INR cells engrafted into lethally irradiated mice after 7 days of BMEC co-culture and gave rise to fully penetrant leukemic disease. Consistent with the primary GEMM models, INR transplanted cells derived from BMEC co-cultures gave rise to disease with faster kinetics compared to IN (Supplementary Fig. 3A). Secondary transplant studies demonstrated engraftment and disease propagating capacity, confirming that the co-culture system allows for the growth and persistence of leukemia stem/propagating cells (Supplementary Fig. 3B).

To identify targets unique to Nras co-mutant AML, we developed a custom CRISPR sgRNA library curated to include Nras related gene targets(29, 30) which were expressed in AML transcriptome datasets(31, 32). Our final library consisted of 2532 sgRNAs targeting 561 genes (Supplementary Table 1); we optimized conditions for infection of HSPCs to achieve a multiplicity of infection (MOI) of 0.1–0.3 while maximizing cell viability and expansion. INR leukemias lacking Cas9 were infected with lentivirus containing our CRISPR library and co-cultured with BMECs. Genomic DNA was isolated at a Day 1 baseline timepoint and compared to Day 7; we observed robust representation of sgRNAs with minimal stochastic dropout or enrichment following culture (Supplementary Fig. 3C). We then crossed our IN and INR models to Rosa:26/Cas9(33) and re-generated primary Cas9 expressing IN and INR leukemias for screening (Supplementary Fig. 3D).

We performed CRISPR screens on Kit⁺ bone marrow cells from Cas9 expressing IN, INR, and wild-type Ubc:CreERT2 (WT) control mice infected with a lentivirus containing our sgRNA library (Fig. 2A). At baseline, sgRNAs were evenly represented across biological conditions (Supplementary Fig. 3E). After applying a broad FDR cutoff of 0.25, there were 32 genes depleted in WT HSPCs (Supplementary Fig. 4A). Known dependencies across multiple cellular contexts (Rpa3, Clp1, Plk, Dtymk, Bub1) scored highly as well as known WT HSPC specific dependencies (Mdm2, Kit, Men1)(34–36). Wee1 also scored highly, consistent with reports of myelosuppression in patients treated with Wee1 small molecule inhibition(37). These data show that our BMEC co-culture system is a robust ex vivo platform for interrogating genetic dependencies in AML.

Figure 2: Ex Vivo CRISPR screens reveal Elovl1 is a selective vulnerability in Nras mutant AML.

A. Schematic representation of CRISPR screens with BMEC platform. Kit+ cells are isolated from Cas9-expressing mice, infected with a BFP+ lentivirus encoding a custom sgRNA library, and co-cultured with BMECs. DNA is extracted pre- and post-culture for sgRNA library amplification and sequencing. MAGeCK is used to assess depleted and enriched sgRNAs. Each screen is repeated twice per genotype. B. KEGG Pathway analysis of INR vs WT Maximum Likelihood Estimation CRISPR hits; performed using MAGeCKFlute. C. Maximum Likelihood Estimation analysis of Top 40 CRISPR screen hits comparing bulk INR to WT with Wald FDR <0.02 (red symbols), calculated with MaGeCK. Top 40 gene hits labeled. D. Schematic of very long chain fatty acid processing; black arrows = elongation reactions; grey arrows = desaturation reactions; E=Elov) E. ELOVL1 expression in primary AML samples downloaded from TCGA, stratified by oncogenic mutant NRAS status. F. Kaplan-Meier survival analysis of NRAS pathway mutant primary AML samples downloaded from TCGA, stratified by top 50% versus bottom 50% ELOVL1 expression. G. Schematic for metabolic tracing experiments with uniformly ¹³C labeled oleate. H. After pulsing in ¹²C and ¹³C universally labeled oleate, M+18 percent labeled fraction in C20:1 VLCFA pool across genotypes at 4 hours and 24 hours (n=4 mice per arm) I. M+18 percent labeled fraction in C24:1 VLCFA pool across genotypes at 4 hours and 24 hours (n=4 mice per arm). All data are mean ± SEM. ns=not significant, *p<0.05, **p < 0.01, ***p < 0.001, ****p<0.0001

Elovl1 is a Therapeutic Target in Nras Mutant Leukemia

We next performed a screen in primary INR leukemia cells, using WT and IN cells as a counter-screen such that we could identify dependencies specific to INR leukemias. We calculated a weighted dependency score(38) comparing targets in INR leukemias to WT cells. Nras itself was a mutant-specific dependency in INR cells as were established Nras specific targets (Icmt, Mtor)(39, 40). We performed unbiased KEGG pathway analyses and found enrichment in pathways related to fatty acid elongation and downstream processes (Fig. 2B). The strongest dependency in the fatty acid elongation pathway specific to Ras-mutant leukemia cells was Elovl1 (Fig. 2C). Utilizing the Robust Rank Aggregation dependency scoring method(38), Elovl1 remained a selective dependency in INR AML compared to IN AML cells and WT HSPCs (Supplementary Fig. 4B). Elovl1 was both an Nras selective dependency in the bulk cell fraction (Supplementary Fig. 4C, 4D) and within the lineage negative cell fraction (Supplementary Fig. 4E, 4F).

Elovl1 is a metabolic protein that catalyzes the elongation of saturated and, specifically, mono-unsaturated C20–C22 acyl-CoAs to C22–C24 very long chain fatty acid species (VLCFAs)(41–43) (Fig. 2D). Elovl1 has been characterized in adrenoleukodystrophy and its elevated expression in certain tumor types is a negative prognostic marker(44–48). While its deletion in lymphoid cells promotes anti-tumor immunity in solid tumors(49), its role in cell-autonomous cancer biology including in myeloid malignancies is largely undefined. ELOVL1 is expressed at a relatively low level in normal human bone marrow gene expression datasets(44, 50). In primary AML samples from the TCGA AML cohort, ELOVL1 expression is significantly elevated in NRAS mutant AML cells compared to NRAS wild-type leukemias (Fig. 2E). When stratifying TCGA AML samples with NRAS pathway mutations by ELOVL1 expression, high ELOVL1 expression was associated with significantly worse overall survival (OS Low ELOVL1: 46.06mo, High ELOVL1: 8.51mo, p=0.0058 Fig. 2F). These data suggest heightened ELOVL1 activity is a feature of human NRAS-mutant AML and may be an appropriate therapeutic target. To assess whether there was increased utilization of this pathway in AML cells with Ras mutations cells, we performed ¹³C-oleate tracing (Fig. 2G). At 24 hours, INR cells showed increased incorporation of the M+18 uniformly labeled isotopologue, reflecting direct elongation of intact oleate, into C20:1 and C24:1 VLCFA species relative to IN cells (Fig. 2H, 2I), consistent with increased flux through the VLCFA elongation pathway.

We next generated Elovl1 and Rosa targeting sgRNAs distinct from those in the initial screening library. We performed CRISPR cut sequencing at the Elovl1 target locus and found high DNA editing efficiency (Supplemental Fig. 5A, 5B). We also confirmed decreased Elovl1 mRNA expression with RNA sequencing of sorted INR cells expressing targeted sgRNAs (Supplemental Fig. 5C) We then infected primary WT HSPCs with an MCherry+ lentivirus encoding these sgRNAs, sorted and expanded infected cells over BMECs, then performed targeted VLCFA metabolomic profiling. Consistent with its known role in VLCFA metabolism, Elovl1 knockout led to a build-up of its substrate C20:1(41) and a corresponding decrease in downstream metabolites (Fig. 3A).

Figure 3: Nras mutant leukemias require catalytically active Elovl1.

A. WT Cas9 HSPCs infected with MCherry+ lentivirus encoding indicated sgRNA, expanded for 6 days and FACS sorted. Very long chain metabolites measured by mass spectrometry after 7 days post sort. n=3 mice, each with 3 technical replicates. B. INR cells infected with lentivirus encoding sgRNAs targeting Rosa or Elovl1 co-cultured over BMECs. Day 14 measurements normalized to baseline Day 2 (n=7 mice sgRosa and sgElovl1#1, n=3 mice sgElovl1#2, One-way ANOVA with Dunnett’s multiple test correction). C. IN cells with indicated sgRNA co-cultured with BMECs. Day 7 and Day 14 measurements normalized to baseline Day 2 (n=5 mice). D. INR Kit⁺ cells infected with indicated MCherry⁺ lentivirus encoding sgRNA, cultured overnight, and transplanted into lethally irradiated recipients. Percentage of peripheral blood MCherry⁺ cells in engrafted CD45.2⁺ normalized to baseline Week 3 measurement (n=10 mice transplanted per sgRNA, One-way ANOVA with Dunnett’s multiple test correction, significance stars refer to both Elovl1 sgRNAs individually compared to Rosa). E. As in D but WT Kit⁺ cells infected with indicated MCherry⁺ lentivirus. Percentage of peripheral blood WT CD45.2⁺ cells that are MCherry⁺; normalized to baseline Week 3 measurement (n=5 mice transplanted per sgRNA, ns at all timepoints). F. IN Kit⁺ cells infected with indicated MCherry⁺ lentivirus encoding sgRNA, cultured overnight, and transplanted into lethally irradiated recipients. Percentage of peripheral blood IN CD45.2⁺ cells that are MCherry⁺; normalized to baseline Week 3 measurement (n=12–13 mice transplanted per sgRNA, ns at all timepoints). G. INR cells infected with lentivirus encoding sgRNAs targeting Rosa or Elovl1 co-cultured over BMECs for 9 days, sorted for MCherry+ cells, then expanded for another 6 days before harvesting, normalizing cell number, and injecting into primary irradiated recipients (n=5 mice transplanted for sgRosa, n=4 for sgElovl1). Percentage of edited cells in moribund bone marrow (BM). H. As in G, but spleen (SPL). I. Non TAM’d Cas9 INR cells infected with lentivirus encoding sgRNA targeting Rosa or Elovl1, transplanted into lethally irradiated recipients, then tamoxifen dosed to activate Idh1, Npm1, and Nras alleles and initiate leukemic formation. Edited cells tracked in peripheral blood over time; initiation window defined by abnormal changes in peripheral blood CBCs and flow cytometry prior to murine leukemic death (n=5 transplanted mice/arm). J. INR Kit⁺ cells infected with indicated MCherry⁺ lentivirus encoding sgRNA, co-cultured with BMECs for 14 days. Percentage of MCherry⁺ in CD45.2⁺ cells on Day 7 & 14 normalized to Day 2 baseline. (n=2 replicates, One-way ANOVA with Dunnett’s correction, representative of 2 independent experiments). K. INR Kit⁺ cells infected with indicated MCherry⁺ lentivirus sgRNA/cDNA, co-cultured with BMECs for 14 days. Percentage of Edited MCherry⁺ in CD45.2⁺ cells normalized to Day 2 baseline (n=3 replicates, One-way ANOVA with Dunnett’s correction, representative of 2 independent experiments). L. INR Kit⁺ cells infected with indicated MCherry⁺ lentivirus sgRNA + vehicle or metabolite, co-cultured with BMECs for 7 days. Percentage of Edited MCherry⁺ in CD45.2⁺ cells normalized to Day 2 baseline (n=3 mice, One-way ANOVA). All data are mean ± SEM. *p<0.05, **p < 0.01, ***p < 0.001, ****p<0.0001

In ex vivo culture, INR cells lacking Elovl1 were strongly depleted, both in the bulk and lineage negative cell fraction, compared to Rosa controls (Fig. 3B, Supplemental Fig. 5D). Consistent with the CRISPR screen data, IN cells lacking oncogenic Nras had no fitness defect upon infection with an Elovl1 targeting sgRNA (Fig. 3C, Supplemental Fig. 5E). To evaluate if Elovl1 dependency in INR cells is Idh1^R132H dependent, we generated Cas9 expressing Npm1^c/Nras^G12D (NR) leukemias(24) and found Elovl1 was also a dependency in Nras-mutant cells lacking a neomorphic Idh mutation (Supplemental Fig. 5F, 5G). Elovl1 deletion continued to impair INR cell fitness in the presence of mutant IDH inhibition with ivosidenib, with no effect on IN cells. (Supplemental Fig. 5H, 5I). Oncogenic NRAS and downstream signaling can also be activated in AMLs driven by other genetic mutations such as FLT3 ITD and mutant KRAS. We thus tested the requirement for Elovl1 in these other leukemia subtypes. We found Elovl1 was indeed a dependency in the Kras mutant SKM1 and CCRF-CEM cells as well as in Flt3 mutant Molm-13s (Supplemental Figure 5J–5L). Taken together, our results show Elovl1 is a selective dependency in Ras pathway mutant leukemia.

We next assessed whether Elovl1 is required for normal HSPC survival. We transduced Kit⁺ HSPCs with Rosa or Elovl1 targeting sgRNAs. Elovl1 knock out (KO) WT Lineage Negative cells showed no negative selection compared to Rosa targeted controls (Supplementary Fig. 6A). We then assessed the ability of WT stem cells to tolerate Elovl1 perturbation ex vivo. We infected Kit⁺ WT HSPCs with Rosa or Elovl1 targeting sgRNAs, sorted and expanded edited cells, and performed flow cytometry and viability analyses (Supplementary Fig. 6B). After 14 days, both Rosa and Elovl1 KO cells had similar cell viability and hematopoietic cell numbers (Supplementary Fig. 6C, 6D). We found similar distributions of mature myeloid cells (Supplementary Fig. 6E–6G) as well as stem and progenitor cells (Supplementary Fig. 6H–6J). Taken together, Nras-mutant leukemias exhibit marked sensitivity to Elovl1 perturbation ex vivo, whereas wild-type HSPCs remain unaffected.

Nras mutant Leukemias Require Elovl1 In Vivo

We next infected INR leukemias with an MCherry expressing lentivirus encoding sgRNAs targeting Rosa or Elovl1 and transplanted these into recipient mice. We then tracked the percentage of edited cells over time. Consistent with the ex vivo data, Elovl1 knock out cells were robustly depleted in INR AML across both bulk and mature myeloid compartments in the blood (Fig. 3D, Supplementary Fig. 7A, 7B) and bone marrow (Supplementary Fig. 7C, 7D). Notably, Elovl1 loss led to a marked reduction of Kit⁺ cells in the peripheral blood and bone marrow (Supplementary Fig. 7E, 7F).

We then evaluated the impact of Elovl1 loss in WT HSPCs and in IN AML cells. Consistent with the ex vivo data, WT Cas9 HSPCs infected with an Elovl1 targeting sgRNA showed no loss of fitness over 16 weeks, both within the CD45.2+ fraction and in the myeloid compartment through serial assessment of the peripheral blood (Fig. 3E, Supplementary Fig. 7G, 7H). To confirm that MCherry⁺ sgRNA-containing cells did not escape editing, we harvested bone marrow from mice 11 weeks post-transplant and performed CRISPR cut sequencing. At 11 weeks, we observed high editing efficiency at the Elovl1 locus in Elovl1 sgRNA transduced WT HSPCs (Supplementary Fig. 7I). Importantly there was no selective depletion of Elovl1 targeted cells compared to Rosa in the bulk CD45.2, mature myeloid, or HSPC compartments in the bone marrow when assessed at 16 weeks (Supplementary Fig. 7J–7L). Elovl1 loss also did not cause selective depletion of Elovl1 targeted cells in IN AMLs engrafted in vivo (Fig. 3F, Supplementary Fig. 7M–7O). These data show that Elovl1 is an Nras selective dependency in vivo.

To further assess the requirement for Elovl1 in INR cells, equal numbers of sorted Rosa control and Elovl1 KO INR cells were transplanted into separate cohorts of irradiated recipients. Despite equivalent starting purity, Elovl1 KO cells were rapidly outcompeted by residual unedited cells by 3 weeks (while Rosa controls remained >95% mCherry+) (Supplementary Fig. 8A). At endpoint, Elovl1 KO bone marrow and spleen from moribund mice showed significantly decreased leukemic burden across most cell compartments (Fig. 3G. 3H, Supplementary Fig. 8B–O) and KO recipients showed a modest survival benefit over Rosa controls (Suppl. Fig. 8P). We then assessed the role of Elovl1 during the leukemia initiation window in INR cells. We transduced Cas9 containing Idh1^R132H/Npm1^c/Nras^G12D with a virus expressing either a Rosa or Elovl1 targeted sgRNA, transplanted these transduced cells into irradiated recipients, and induced activation of leukemia alleles with tamoxifen after engraftment. Elovl1 targeted cells were already negatively selected against at 4 weeks post tamoxifen induction (Fig 3I), during the period of CD45.2+ cell expansion (Supplemental Fig. 8Q). Because several mice had already succumbed to disease by 5.5 weeks after tamoxifen induction (Supplemental Fig. 8R), this time point captures the initiation phase before fulminant leukemia. The early depletion of Elovl1-KO cells reveals that Elovl1 confers a competitive fitness advantage during the leukemic initiation window.

Nras Mutant AML Requires Catalytically Active Elovl1 for Leukemia Cell Fitness

Elovl1 is one of seven mammalian Elovs that elongate VLCFAs of differing lengths.(41) To assess whether the requirement for Elovl1 was seen with other members of this enzyme family, we perturbed Elovl2–7 using single guide RNAs. Notably, only Elovl1 disruption had a strong fitness disadvantage in INR cells (Fig. 3J). To assess whether the catalytic activity of Elovl1 is required for INR cell fitness, we expressed a Elovl1 CRISPR resistant cDNA or a Elovl1 CRISPR resistant catalytically inactive cDNA in Elovl1 KO cells. For the catalytic dead controls, we mutated 3 conserved histidine residues within the HXXHH core motif to alanine or glutamine, a modification known to abolish Elov activity.(51–53) Only Elovl1 KO cells containing full Elovl1 CRISPR resistant cDNA rescued the INR fitness defect, with Elovl1 catalytically dead mutants failing to reverse the fitness defect seen with Elovl1 deletion (Fig. 3K). We next asked if adding the key Elovl1 metabolite product could rescue the genetic dependency. Add-back of C24:1 mono-unsaturated VLCFA, but not C24:0 saturated species and vehicle controls, was specifically able to rescue the fitness defect induced by Elovl1 knock-out (Fig. 3L). This confirms the metabolic products of Elovl1 are specifically required for INR cell fitness.

Peroxisomal Fatty Acid Processing is Dispensable for Nras mutant Leukemia

We then investigated the mechanism by which Elovl1 loss disrupts Nras mutant leukemia cell fitness. Very long chain fatty acids (VLCFAs) are either processed in the peroxisome for energy via peroxisomal beta oxidation or are converted into sphingolipid ceramide species (Fig. 4A)(41). To test the role of peroxisomal processing in INR fitness, we designed sgRNAs targeting VLCFA transporters Abcd1 and Abcd2 as well as Acox1, the rate limiting step in peroxisomal β-oxidation (Supplemental Fig. 9A)(45, 54–56). Infection of INR cells with these sgRNAs did not decrease cell fitness compared to Rosa controls in either the bulk or lineage negative cell compartment (Supplemental Fig 9B, 9C). To assess the requirement for peroxisomal processing in vivo, we infected INR Kit⁺ cells with sgRNAs targeting Pex13, an essential protein for peroxisomal biogenesis(57, 58) and transplanted into lethally irradiated recipient mice. INR cells lacking Pex13 did not have decreased relative fitness compared to Rosa targeted controls (Supplemental Fig. 9D, 9E), indicating that the deleterious effect of Elovl1 loss on INR cell fitness is not due to a dependency on peroxisomal VLCFA processing.

Figure 4. Nras mutant leukemias require Elovl1 derived VLCFAs to make lipid-raft associated sphingolipids.

A. Schematic detailing VLCFA processing through peroxisome or sphingolipid biosynthesis. B. Steady state sphingolipid metabolomics from murine bone marrow (n=4 per genotype with 3 replicates each). C. After pulsing in ¹²C and ¹³C universally labeled oleate, M+18 percent labeled fraction in d18:1/24:1 ceramide pool across genotypes at 4 hours and 24 hours (n=4 mice per arm). D. M+18 percent labeled fraction in d18:1/24:1 sphingomyelin pool across genotypes at 4 hours and 24 hours (n=4 mice per arm). E. INR Kit⁺ cells infected with indicated MCherry⁺ lentivirus encoding sgRNA, co-cultured with BMECs for 14 days normalized to Day 2 baseline (n=3 mice, One-way ANOVA with Dunnett’s correction). F. INR Kit⁺ cells infected with MCherry⁺ sgRNA encoding lentivirus; percentage of edited MCherry⁺ cells of CD45.2+ at Week 3 normalized to day of transplant baseline (n=10 transplanted mice per sgRNA). G. As in F but Week 5 normalized to Week 3 (n=5 transplanted mice per sgRNA). Unpaired T test performed. H. INR Kit⁺ cells infected with MCherry⁺ lentivirus encoding indicated sgRNA, co-cultured with BMECs for 14 days. Percentage of edited MCherry⁺ cells in CD45.2⁺ cells on Day 7 and Day 14 normalized to Day 2 baseline. (n=6 mice). All data are mean ± SEM. *p<0.05, **p < 0.01, ***p < 0.001, ****p<0.0001

Nras-Mutant Leukemias Require Ceramide and Sphingomyelin Biosynthesis

We next evaluated the contribution of sphingolipid biosynthesis, the alternative pathway of VLCFA metabolism, to INR cell fitness. Steady state lipidomic profiling of IN and INR cells revealed depletion of Elovl1-derived sphingomyelins with a trend towards depletion of mono-unsaturated VLCFAs and ceramides in INR versus IN cells (Fig. 4B, Supplementary Fig. 10A–C). C22:1 ceramides, an Elovl1 intermediate, were also elevated in RAS pathway mutant patient samples(59) (Supplementary Fig. 10D). To assess whether there was increased utilization of this pathway in INR cells, we performed ¹³C-oleate tracing. At 24 hours, INR cells showed increased incorporation of the M+18 isotopologue (reflecting direct incorporation of intact oleate) into C24:1 ceramide and sphingomyelin relative to IN cells (Fig. 4C, 4D), consistent with increased flux through VLCFA elongation and ceramide/sphingomyelin biosynthesis. Incorporation into other lipid species at 24 hours was unchanged (Supplementary Fig. 10E–J).

We then utilized genetic epistasis experiments to confirm our metabolic tracing results. Ceramides are generated in the de novo sphingolipid pathway by condensation of VLCFAs and serine into 3-keto-sphinganine via SPTLC proteins. A single enzyme, KDSR, then converts these intermediates to sphinganine, which is further processed into ceramides.(60) We therefore generated sgRNAs targeting Sptlc1, Sptlc2, and Kdsr and evaluated their effect on INR cell fitness. We found Sptlc1 and Kdsr perturbation phenocopied Elovl1 loss in INR cells ex vivo (Fig.4E). This result is consistent with evidence that Sptlc1 is absolutely required for de novo sphingolipid synthesis, whereas Sptlc2 loss does not have a non-redundant role in this pathway.(61) To validate our findings in vivo, we infected Kit⁺ INR cells with a lentivirus encoding sgRNAs to Rosa or Kdsr, transplanted transduced AML cells into lethally irradiated recipients, and assessed the engraftment potential of targeted cells. We found markedly suppressed engraftment of Kdsr targeted cells compared to Rosa targeted cells in vivo (Fig. 4F). When assessing the selection of Kdsr targeted cells in the peripheral blood over time, we found robust negative selection of Kdsr edited cells in mice engrafted with INR leukemias (Fig. 4G).

To further isolate downstream ceramide pathways governing INR fitness, we individually perturbed the rate limiting steps in sphingomyelin synthesis (Sgms1), glycosphingolipid synthesis (Ugcg), and S1P production (Sphk1, Sphk2) in primary murine leukemias. We found that loss of Sgms1, the rate limiting step in sphingomyelin biosynthesis, negatively affected INR cell fitness (Fig. 4H). INR cells lacking Ugcg, Sphk1, or Sphk2 had no differences in cell fitness compared to Rosa controls (Supplemental Fig. 11A–11F), showing that the dependency in Ras-mutant AML cells was specific to sphingomyelin synthesis downstream of Elovl1. We additionally confirmed all sgRNAs had high on target activity (Supplemental Fig. 12A–U). Taken together, metabolic tracing and genetic knock out experiments validate the VLCFA/Ceramide/Sphingomyelin axis as essential for INR cell fitness.

Elovl1-Derived Very Long Chain Fatty Acids Maintains Membrane Lipid Rafts

Thus far, we have demonstrated that (1) Elovl1 is a synthetic lethal target in oncogenic Nras mutant AML; (2) that Elovl1 is the singular enzyme within the VLCFA elongase class that induces a relative fitness disadvantage; and (3) that this fitness effect includes a specific requirement for sphingomyelin production. Sphingomyelin is an essential sphingolipid necessary for the generation of lipid rafts, with C22-C24 sphingolipids being the most enriched species.(62, 63) Of note, all C22-C24 mono-unsaturated sphingolipids are thought to derive uniquely from Elovl1 enzymatic activity(41), so we next investigated if Elovl1 loss disrupts membrane lipid rafts in Nras mutant AML and if lipid raft integrity was required for the survival of Ras-mutant AML cells.

We first examined lipid raft abundance using a fluorophore tagged Cholera toxin B (CtxB) antibody that binds ganglioside GM1 in lipid rafts(64, 65). In INR cells, we found Elovl1 perturbation decreased CtxB cell surface binding, indicating fewer ganglioside-rich lipid rafts at the plasma membrane (Fig. 5A). We next tested whether lipid raft disruption selectively affects Nras-mutant cells by treating IN and INR bone marrow cells ex vivo with 2 mM methyl-β-cyclodextrin (MβCD), a cholesterol depleting agent known to disrupt lipid rafts(66). INR leukemia viability substantially decreased following 24 hours of MβCD treatment (Fig. 5B), whereas the viability of IN leukemia cells was unaffected (Fig. 5C). We then investigated potential additive effects between Elovl1 depletion and lipid raft disruption by infecting INR cells with Rosa or Elovl1 targeting sgRNAs and culturing in the presence of MβCD. We found at an early timepoint, the fitness defect from Elovl1 loss was exacerbated with MβCD treatment in INR cells (Fig. 5D), but not IN cells (Fig. 5E), suggesting additive targeting of a shared pathway.

Figure 5: Elovl1 mediated very long chain fatty acid synthesis maintains membrane lipid rafts that scaffold key signaling in Nras mutant AML.

A. Flow cytometry of Lineage- sgRosa versus sgElovl1 INR cells; lipid rafts quantified by Cholera Toxin B (CtxB) mean fluorescence intensity (n=3 mice). B. INR cells treated for 24 hours with Vehicle or 2mM MβCD; cell viability assessed via Trypan blue exclusion (n=3 replicates, representative of 2 independent experiments). C. IN cells treated for 24 hours with Vehicle or 2mM MβCD (n=3 replicates, representative of 3 independent experiments). D. INR cells infected with sgRNA targeting Rosa or Elovl1, co-cultured with BMECs in presence of Vehicle or 2mM MβCD. MCherry positivity in CD45.2⁺ cells assessed at Day 7, normalized to Day 2 baseline (n= 7 mice, Brown-Forsythe and Welch One Way ANOVA tests). E. As in D, but IN cells (n= 3 mice, Brown-Forsythe and Welch One Way ANOVA tests). F. Uniform manifold approximation and projection (UMAP) visualization of INR cells (n = 3 mice, total 32,239 sgRosa cells and 35,465 sgElovl1 cells) colored by cell types and genotype. G. PROGENy Pathway activity (sgElovl1 – sgRosa) across cell types. Blue indicates decreased activity in sgElovl1; *FDR <0.05. All data are mean ± SEM. *p<0.05, **p < 0.01, ***p < 0.001 unless otherwise noted. H. Unbiased GO Pathway analysis of proteome from Vehicle or 2mM MβCD treated INR cells at 24 hours in indicated subcellular fraction (n=4–5 mice pooled per biological replicate, 3 biological replicates per arm). Only terms with q value < 0.05 are displayed. I. INR cells were treated with Vehicle or Methyl beta cyclodextrin for 24 hours. Significantly differentially expressed Ras related proteins in membrane fraction at 24 hours and their expression in cytosolic fraction at 24 hours. Differentially expressed proteins were defined as |Log2FC| >0.25 and q value <0.05.

Elovl1 Loss Perturbs Key Membrane Ras Signaling Complexes

To capture the landscape of early signaling changes induced by Elovl1 loss, we performed Cellular Indexing of Transcriptomes and Epitopes by sequencing (CITE-Seq) on INR cells co-cultured with BMECs. As expected, Elovl1 gene expression was significantly suppressed in cells transduced with an Elovl1 targeted sgRNA (Supplementary Fig. 13A). We annotated cell clusters based on their transcriptional profiles and cell surface protein expression (Fig 5F, Supplementary Table 2, Supplementary Fig. 13B–13D). In the Elovl1 KO condition, we observed an increased proportion of monocyte/macrophage-like and non-classical monocyte-like AML cells at the expense of neutrophils and classical monocytes (Supplemental Fig. 13B). We next applied Pathway RespOnsive GENes (PROGENy) to estimate signaling pathway activity from pathway-responsive gene footprints in leukemia clusters.(67) In Elovl1 KO cells, we find decreased PI3K, MAPK, and TGF-β pathway activity in all cell-types except the dendritic-cell like cluster. We also observed decreased TNF-α signaling in neutrophil-, monocyte/macrophages-, and classical monocyte-like cells (Fig. 5G). We next performed unbiased gene set enrichment analyses in the main blast clusters: neutrophil-like cells and monocyte/macrophage-like cells. Elovl1 KO neutrophil-like cells exhibited significantly decreased Ras signaling as well as reduced TGF-β and TNF-α signaling with reduced NF-κB pathway activity. Similarly, we observed reduced Ras and TGF-β signaling in the monocyte/macrophage-like cluster. (Supplementary Fig. 13E). Focusing on the predominant neutrophil-like cluster, the expression of Ras dependent genes (including Ank1, Plaur, Plau(68, 69)) were reduced in Elovl1 KO cells as were the expression of key TNF-α pathway components (Tnf, Ets2) and TGF-β signaling pathway members (Tgfb1, Thbs1(70)) (Supplementary Fig. 13F).

To further explore a mechanistic link between lipid raft disruption and Ras driven signaling, we performed quantitative proteomics on membrane and cytoplasmic fractions of lipid raft disrupted cells. In INR cells treated with MβCD compared to vehicle, unbiased pathway analysis identified the MAPK signaling cascade as the only signaling-pathway term specifically enriched among downregulated membrane proteins, distinct from the cytoskeletal and vesicle-trafficking subsets (Fig. 5H). Consistent with this, quantitative proteomics revealed broad, pathway-wide depletion of Ras effectors and downstream signaling components from the membrane fraction, including core MAPK cascade members (Sos1, Sos2, Raf1, Map2k2, Map2k4, Map2k5, Map2k7, Mapk8, Mapk9), Ras-related GTPases and their regulators (Rap1a, Rap1b, Rap2b, Rin2, Rin3, Rgl2, Tiam1), and Pi3k/Akt pathway components (Pik3r2, Akt3) (Fig. 5I). Taken together, our data reveals Elovl1 perturbation impairs the ability of Nras mutant leukemias to utilize sphingolipid containing lipid rafts, culminating in attenuation of key signaling pathways.

Elovl1 Inhibition Reduces Ras Mutant Leukemic Burden in Murine and Human Cells

To evaluate the therapeutic tractability of these findings, we tested Compound 27, a first in class, highly specific ELOVL1 inhibitor which was developed for adrenoleukodystrophy and not tested in any malignant context.(71) We performed the same membrane and cytoplasmic fractionation proteomics as in the MβCD experiments, and consistent with that data, found that small molecule ELOVL1 inhibition promoted broad, pathway wide depletion of Ras related proteins in the membrane fraction, including core Ras GTPases, RTK/adaptor signaling components (Gab2, Shcbp1, Sos1, Sos2), MAPK kinase cascade members (Araf, Braf, Ksr1, Map2k1/2/5/7, Map3k3/4/6/7, Map4k2/5, Mapk12/14), Pi3k/Akt components (Pik3cd, Pik3r2, Akt3), and RAS associated regulators (Rassf1/2/3, Rgl2, Rin1/3, Dab2ip) (Fig. 6A). Small molecule ELOVL1 inhibition induced membrane specific depletion of Nras itself. Indeed, similar to MβCD treatment, pathway analysis showed that ELOVL1 inhibition induced downregulation of MAPK cascade signaling specifically in the membrane fraction (Supplemental Fig. 14A).

Figure 6: Small molecule Elovl1 inhibition decreases leukemic burden of Ras pathway mutant disease.

A. Normalized protein abundances of Ras related proteins in INR cells cultured for 5 days over BMECs with Vehicle or Compound 27 Elovl1 inhibitor (Elovl1i) treatment. Left panel is significantly differentially expressed proteins in membrane fraction and right panel is their abundance in cytoplasmic fraction. Significantly differentially expressed membrane proteins were defined as |Log2FC| >0.25 and q value <0.05. B. IN and INR cells plated over BMECs and treated ex vivo with vehicle (DMSO), 1uM Compound 27 Elovl1 inhibitor (hereafter referred to as Elovl1i), or 10uM Elovl1i (n=4 mice per genotype). Number of CD45.2+ cells after 7 days of treatment. C. As in A but number of LSK cells (Lineage-Sca1+Kit+). D. Survival of mice transplanted with vehicle or 10uM Elovl1i treated INR cells. Cells were treated for 1 week ex vivo over BMECs before transplant. (n=4 transplanted mice/arm). E. NRAS mutant patient derived xenograft (PDX) samples treated ex vivo over HUVECs with Vehicle, 1uM, or 10uM Elovl1i. Percentage of human CD45+ cells in the culture at 7 days (n=3–4 replicates per condition). F. As in E but NRAS wild-type PDX samples (n=4 replicates per condition G. Patient samples with indicated RAS pathway mutation treated ex vivo over HUVECs with Vehicle or 10uM Elovl1i for 7 days. Percentage of human CD34+ leukemic stem cells in the culture assessed (n=6 replicates per condition). H. Bone marrow engraftment of myeloid cells following ex vivo treatment of NRAS-mutant PDX (“PDX 5”) cells with vehicle or 10 μM Elovl1i for 1 week in co-culture with HUVECs prior to transplantation into sublethally irradiated NSGW41 mice (n = 4 mice transplanted /arm). Engraftment assessed from bone marrow aspirate at 3 weeks. I. Ex vivo treatment of NRAS-mutant PDX (“PDX 5”) cells with vehicle (DMSO), 1uM ivosidenib, 10 μM Elovl1i, or combination ivosidenib and Elovl1i for 1 week in co-culture with HUVECs. Number of human CD45+ cells assessed (n=4 replicates per condition).

We next assessed the impact of small molecule ELOVL1 inhibition (Elovl1i) in our murine AML models and in primary human AML cells. Consistent with our sgRNA knockout data, Elovl1 inhibition in our GEMM models selectively decreased leukemic burden in Nras mutant disease, especially in the HSPC fraction, without effect in Nras wild type leukemias (Fig. 6B, 6C). Notably, 1 week of ex vivo Elovl1 inhibitor treatment of INR cells significantly extended survival of transplanted mice, (Fig. 6D). White blood cell counts at 3 weeks post-transplant were lower in mice receiving Elovl1i treated cells compared to vehicle (Supplemental Fig. 14B), with the effects of Elovl1 inhibition on INR cell number seen as early as 3 days of treatment (Supplemental Fig. 14C).

We next explored ELOVL1 as a dependency in human AML. We first analyzed published single cell Genotyping of Transcriptomes data from a patient with complex karyotype AML containing IDH1, DNMT3A, and NRAS mutations.(72) We found that ELOVL1 expression is markedly elevated in NRAS mutant clones, and we found that gene sets related to very long chain fatty acid metabolism are significantly upregulated in NRAS-mutant cells compared to NRAS-wildtype cells from the same patient (Supplemental Fig. 14D). To further enhance the clinical translational relevance of these findings, we generated patient derived xenograft (PDX) models of IDH1-mutant leukemia with and without a co-occurring NRAS mutation (3 NRAS-WT, 2 NRAS-mutant; patient information in Supplemental Table 3). Treatment with Compound 27 selectively decreased the human CD45+ fraction in NRAS-mutant but not NRAS-WT PDX cells ex vivo (Fig. 6E–F, Supplemental Fig. 14E–G). Inhibition also significantly decreased the CD34+ HSPC fraction across multiple RAS pathway-mutant patient samples ex vivo (Fig. 6G). Similar to the GEMM models, 7-day ex vivo treatment of NRAS-mutant PDX cells with Elovl1 inhibition reduced myeloid engraftment in vivo (Fig. 6H). Lastly, we found combining Elovl1 inhibition with ivosidenib significantly reduced leukemic burden beyond either monotherapy alone in an NRAS-mutant PDX model, suggesting a potential therapeutically tractable combination therapy (Fig. 6I).

DISCUSSION:

NRAS mutant AML is a molecularly and clinically important AML subtype, as NRAS mutations are seen in a significant subset of AML patients. Notably, NRAS mutations, and mutations in other MAPK pathway members are associated with adverse outcome(73), particularly in the setting of mechanism-based therapies including IDH inhibition and venetoclax-based regimens, which have become the standard-of-care first line therapy for most adults with AML(4–7). As such, there is a pressing need to identify dependencies in Nras mutant AML that can inform the development of therapeutics that target Nras mutant cells in the dominant AML clone or in relapse-driving subclones.

Many studies have used chemical and genomic approaches to identify dependencies in AML cells; however, most approaches have either focused on AML cell lines or have used models which can be easily cultured or propagated but do not represent the genetic repertoire of somatic mutations in AML (e.g. ectopic expression of oncogenic MLL fusions). We therefore set out to perform functional genomic studies in primary AMLs with defined somatic mutational genotypes similar to those observed in patients to delineate dependencies linked to specific somatic alterations. We adopted and optimized the BMEC co-culture platform(16) to culture primary AML cells with different genotypes, ensuring that cells cultured retained in vivo leukemogenicity and could be functionally perturbed ex vivo. Using this approach, we identified and credentialed Elovl1 as a dependency in Nras mutant AML.

We show that Elovl1 catalytic activity is required for Nras mutant AML, suggesting that Elovl1 small molecule inhibition or PROTAC mediated degradation could represent therapeutically tractable approaches to target Nras mutant leukemia cells. Prior efforts to develop Elovl1 inhibitors for adrenoleukodystrophy required blood-brain barrier penetrance for efficacy but caused on-target CNS toxicity in primate models, halting their development. This liability is far less relevant in AML, where CNS penetrance is not required for therapeutic efficacy. Thus non-CNS-penetrant compounds previously deprioritized for adrenoleukodystrophy may be well suited for repurposing in NRAS-mutant leukemia. Our data shows that small molecule Elovl1 inhibition reduced leukemic burden across mouse and patient samples. Most importantly, we show that Elovl1 is a dependency in leukemia cells but not in normal HSPCs, further underscoring Elovl1 as a leukemia-specific target. This also illustrates the importance of functional genomic screens that systematically test for dependency in different malignant genotypes compared to wild-type primary cells of the same lineage. These findings thus support a therapeutic window for metabolic targeting in this genetically defined patient population. Additionally, Elovl1 remains a strong dependency under the pressure of ivosidenib treatment and combination Elovl1 inhibitor and ivosidenib treatment strongly depletes leukemic cells, suggesting a potential role for combinatorial therapy.

While Elovl1 has not been extensively studied in hematopoiesis or leukemogenesis, we found that Elovl1 loss markedly depleted Idh1^R132H/Npm1^c/Nras^G12D leukemic cells in both ex vivo and in vivo models. Our mechanistic data reveal that this effect is via the unique spectrum of Elovl1 activity, which ultimately generates the C22-C24 sphingolipids necessary for lipid raft production and, subsequently, essential cell signaling in Nras mutant leukemia. Our metabolic tracing experiments validate that Nras mutant leukemias have increased metabolic flux through very long chain fatty acids, ceramide, and ultimately sphingomyelin production. Coupled with our steady state metabolic measurements revealing decreased sphingomyelin and a trend towards decreased VLCFA and ceramide species, we find INR cells exist in a precarious metabolic position highly sensitive to Elovl1 perturbation. We further validate this with genetic perturbations of metabolic proteins in primary leukemias to dissect the mechanism of Elovl1 dependency. To complement our genetic data, we used a membrane cholesterol disrupting small molecule known to destabilize rafts and found treatment selectively depletes Nras mutant cells and is additive with Elovl1 loss. Taken together, our results provide an initial link between Elovl1 and lipid raft biology. Due to the constraints of our ex vivo primary cell co-culture system, which precludes microscopy-based imaging and centrifugation-based lipid raft isolation, we are currently unable to assess raft stability, composition, or subcellular localization, including at the endoplasmic reticulum and Golgi, important sites of raft signaling.(74, 75) While our findings suggest that Elovl1 modulates lipid raft organization, a comprehensive understanding of its impact on raft dynamics and compartment-specific signaling will require further careful investigation.

NRAS-driven leukemias exhibit a hyperactive signaling state, with enhanced downstream PI3K and MAPK pathway activity resulting from the constitutively GTP-bound state of mutant NRAS, and from the upregulation of additional signaling pathways within this oncogenic context.(76–78) Importantly, lipid rafts are key concentrating scaffolds for multiple signaling pathways, including many that are required in Nras mutant AML like inflammatory pathways mediated by TNF Receptors and TLRs(79, 80) as well as cytokine receptors like GM-CSF(81). Nras^GTP itself is known to preferentially associate with sphingomyelin containing lipid rafts.(82–88) One straightforward interpretation of our data is that Elovl1 perturbation of rafts disrupts Nras signaling output directly. Consistent with this hypothesis, quantitative proteomics of membrane and cytoplasmic fractions from Elovl1 inhibited INR cells showed selective depletion of Nras and of Ras/MAPK signaling effectors from the membrane compartment, supporting a loss of the raft microenvironment required for its proper membrane localization and productive Ras signaling. Given the broad array of pathways affected in Elovl1-deficient cells, many of which rely on lipid rafts, it remains plausible that perturbing rafts produces pleiotropic effects on multiple survival pathways beyond the Ras pathway. This is consistent with our single cell transcriptional data, showing Elovl1 KO INR blasts have decreased Ras transcriptional output and attenuated TNF alpha and TGF beta signaling output. Delineating the relative contributions of NRAS versus other raft-dependent pathways represents an important avenue for future investigation.

Our data illustrate the power of functional genomic studies in primary AML cells to identify mutant-specific dependencies and to molecularly dissect the key pathway effectors which mediate this dependency. As genetic perturbation systems in primary human AML cells improve in efficiency and scale, it will be important to perform similar functional genomic studies in a broader spectrum of somatic genotypes in other cancer contexts. We expect that similar approaches will lead to the identification of a spectrum of new cancer-specific therapeutic targets, which can be integrated with the increasing compendium of somatic cancer data to inform and accelerate the development of mechanism-based cancer therapies.

MATERIALS AND METHODS

Experimental Animals

All animal studies were performed in accordance with institutional guidelines established by Memorial Sloan Kettering Cancer Center (MSKCC) under the Institutional Animal Care and Use Committee-approved animal protocol (#07–10–016) and the Guide for the Care and Use of Laboratory Animals (National Academy of Sciences 1996). All experimental animals were maintained on a 12-hour light-dark cycle with access to water and standard chow ad libitum. Veterinary staff provided regular monitoring and husbandry care. For GEMM models, all mice had intact immune systems, were drug and test naïve, and had not been involved in previous procedures. Animals were monitored 1–2 times daily for signs of disease or morbidity, bleeding, failure to thrive, infection, or fatigue and sacrificed immediately if they exhibited any signs of the above. Idh1^R132H, Npm1^c, Nras ^G12D, Ubc:CreERT2, and Cas9 mice were described previously(23, 24, 26, 27, 33). Mutant alleles were maintained on a standard C57BL/6 CD45.2 background whereas transplant recipient mice were purchased from Jackson Labs on C57BL/6 CD45.1 background (JAX stock #002014). Mice with Ubc:CreERT2 were administered 100 mg/kg tamoxifen two times (two day drug holiday in between dosing) via oral gavage to induce Cre recombinase activity. Age-and sex- matched 6–16-week-old mice were used in all experiments and only female mice were used as transplant recipients.

Bone Marrow Endothelial Cell (BMEC) Co-Culture with Primary Leukemias

AKT1-activated bone marrow endothelial cells (BMECs) were cultured as previously described(17, 18). BMECs were seeded 1–3 days prior to HSPC plating at a density of 30,000–100,000 BMECs/well in a fibronectin-coated 24-well plate. To obtain HSPCs, bone marrow cells were isolated from limb bones into RBC lysis buffer (Thermo) via centrifugation (8000 rpm for 1 minute) and resuspended. RBC lysis proceeded for 15 minutes and was quenched with FACS buffer (phosphate buffered saline [PBS] + 2% fetal bovine serum). Kit⁺ cells were isolated using the CD117 MicroBead Enrichment Kit (Miltenyi) according to the manufacturers’ protocol. Cells were seeded in 1 mL StemSpan SFEM with 20 ng/mL recombinant murine SCF (PeproTech) and/or additional cytokines as indicated over BMECs. Co-cultures were maintained for a total of 7 days at 37°C and 5% CO2. Media changes were performed every 2–3 days. On day 7, cells were harvested and used in downstream applications. For assays with day 14 readouts, cells were harvested and replated onto fresh BMEC plates. For assays with infected cells, after harvesting Kit⁺ bone marrow, cells were incubated for 1–3 hours at 37°C with 5% CO2 in IMDM media containing 6 ng/mL IL-3 (Peprotech), 10 ng/mL IL-6 (Peprotech), and 20 ng/mL SCF (Peprotech). Then Kit+ cells, lentivirus, and 12 ug/mL polybrene (Millipore Sigma) were incubated in 1 mL StemSpan SFEM (StemCell Technologies) containing 6 ng/mL IL-3 (Peprotech), 10 ng/mL IL-6 (Peprotech), and 20 ng/mL SCF (Peprotech) for 10 minutes at room temperature. 30uL of 1M HEPES (Thermo) was added to each infection and each infection was transferred to a single well of a 6-well plate. Cells were centrifuged at 800 rcf for 1 hour at 32°C, following which an additional 2 mL of StemSpan SFEM containing 6 ng/mL IL-3, 10 ng/mL IL-6, and 20 ng/mL SCF was added to each well. 16 hours after infection, cells were harvested and plated over BMECs. For drug treatments, cells were treated with 500 nM ivosidenib (MedChem Express), or 2mM methyl-β-cyclodextrin (Med Chem Express) with their corresponding vehicles. For Elovl1 inhibitor experiments, Compound-27, a potent and selective Elovl1 inhibitor(71) was added to StemSpan SFEM media supplemented with 20ng/mL SCF at 1uM or 10uM with DMSO as a vehicle control (Cayman Chemical, ELOVL1–27, Item No. 35323). Readouts were performed at Day 7 via flow cytometry and cell counting (ViCell Blu) or cells were transplanted into lethally irradiated CD45.1 recipient mice and assessed for engraftment and survival.

Transplantation Assays and In Vivo Experiments

For cell of propagation studies, Kit⁺ cells were isolated with CD117 microbeads (Miltenyi Biotech) twice following manufacturer’s protocol. Kit⁻ cells were depleted twice using CD117 microbeads. 75000 Kit⁺ or Kit⁻ cells were injected into lethally irradiated CD45.1 recipient mice and monitored for disease onset. For transplantation of cells from BMEC cultures, CD45 cells were isolated with CD45 microbeads (Miltenyi Biotech) following manufacturer’s protocol. 450,000 CD45+ cells were combined with 1×10⁶ CD45.1 (Jackson Laboratories, JAX stock #002014, RRID:IMSR_JAX:002014) whole bone marrow cells. Cells were washed, resuspended in PBS, and transplanted via lateral tail vein injection into recipient mice. Recipient mice were lethally irradiated (900 cGy) 6–8-week-old female CD45.1 mice 16–24 hours before tail vein injection.

For in vivo single sgRNA validation studies, Kit⁺ HSPCs were isolated from donor mice using the CD117 MicroBead Enrichment Kit (Miltenyi) according to the manufacturers’ protocol and infected with lentivirus as described above. A list of all sgRNAs can be found in Supplemental Table 1. Following overnight culture, cells were washed with PBS and 50,000–500,000 infected Kit⁺ cells were transplanted with 200,000–500,000 CD45.1 whole bone marrow cells into lethally irradiated recipient CD45.1 mice. For all secondary transplants described, 0.5 × 10⁶ whole bone marrow cells harvested from primary transplants were transplanted with 0.5 × 10⁶ CD45.1 support marrow into lethally irradiated secondary recipient CD45.1 mice. For all transplants, peripheral blood was isolated at the indicated timepoints by submandibular bleeds and complete blood counts determined using a ProCyte Dx (IDEXX Laboratories) or Element HT5 (Heska) according to manufacturer’s protocol. For terminal tissue isolation, mice were euthanized with CO2 asphyxiation and tissues were dissected and fixed with 4% paraformaldehyde for histopathological analysis. For whole bone marrow isolation, limb bones and/or spine & spleen were dissected and whole bone marrow isolated using centrifugation followed by RBC lysis (BioLegend) as described above.

Ex Vivo CRISPR Screen

Whole bone marrow was isolated from donor mice as described above and Kit⁺ HSPCs were isolated using Cd117 microbeads (Miltenyi Biotech) according to manufacturer’s protocol. Cells were prepped for lentiviral infection as described above. Cells were infected at an efficiency of 10–30% so that most cells received only one sgRNA (MOI of 0.1–0.3). After lentiviral infection, cells were incubated overnight at 37°C with 5% CO2. 12–16 hours after infection, cells were harvested, and cell numbers were determined using an automated cell counter (ViCell Blue, Beckman Coulter). A minimum of 8 × 10⁶ cells were collected for baseline sequencing, sufficient for a representation of at least 500 cells per sgRNA. Screens were performed in primary transplants for wild-type mice and in secondary leukemias for IN and INR. Bone marrow from 2–3 animals was pooled to generate 1 biological replicate of 8 million CD117+ cells for the screen. Each screen was performed twice for a total of two biological replicates. A minimum of 8 × 10⁶ cells were plated over BMEC monolayers in 24-well plates at a cell density of 30,000 HSPCs/well in StemSpan SFEM with 10 ng/mL SCF (Peprotech). Media was changed every 2–3 days after plating. On day 7, co-cultures were harvested with Accutase (Biolegend) and cells were collected for sequencing. For identifying cell-type specific dependencies, co-cultures were harvested and Lineage+ and Lineage− cells were isolated with the Lineage Cell Depletion Kit (StemCell Technologies). A bulk, Lineage+ and Lineage- sample was collected for each screen replicate. Cell pellets for sequencing were lysed, genomic DNA was extracted using the Purogene kit (Qiagen), and DNA quantified using Qubit (Thermo). Genomic DNA was PCR amplified to add Illumina adapters and multiplexing barcodes. Amplicons were quantified by Qubit and Bioanalyzer (Agilent) and sequenced on an Illumina HiSeq 2500.

We followed the exact data analysis pipeline developed by Wang et al (entitled “MAGECK”, RRID:SCR_025016)(38). Day 7 data was normalized to Day 1 baseline data for all analyses (RRA and MLE). MAGeCK RRA and MAGeCK MLE are the two pipelines for identifying CRISPR screen hits. MAGeCK RRA uses robust rank aggregation (RRA) and MAGeCK MLE utilizes a maximum-likelihood estimation (MLE) for robust identification of CRISPR screen hits. MAGeCK RRA allows comparison of two experimental conditions (ie Day 7 to Day 1 baseline). It can identify sgRNAs and corresponding genes that are substantially selected between the two conditions. MAGeCK RRA ranks sgRNAs on the basis of their P values calculated from the negative binomial model and uses a modified RRA algorithm named α-RRA to identify positively or negatively selected genes. MAGeCK RRA uses the RRA enrichment score to indicate the essentiality of a gene. On the other hand, MAGeCK MLE analyzes data from screens with multiple conditions, in our case comparing leukemias to wild-type.MAGeCK MLE calculates a ‘beta score’ for each targeted gene to measure the degree of selection upon gene perturbation, similar to the ‘log fold-change’ measurement in differential expression analysis. In both analysis pipelines, we find Elovl1 was a dependency only in INR cells. In all screens, a gene labeled a significant hit had to have all 4 sgRNAs score. For the MLE analysis in Fig. 3C used a WALD FDR cutoff of 0.02. In Fig. 3C, only the top 30 genes are labeled due to space constraints. For RRA analysis in Supplemental Fig. 4, we used a very broad FDR cutoff of 0.25 in wild-type cells to call a hit to avoid any potential leukemic dependency having an effect in wild-type cells. For IN and INR RRA analysis, hits were called with an FDR cutoff of <0.0001. All FDR cutoffs and analysis specifics are detailed in the figure legend.

Cell Line Culture

HEK293T cells (RRID:CVCL_006) for virus production were cultured in high glucose DMEM and 10% FBS, supplemented with sodium pyruvate (Thermo) and L-glutamine (Thermo). CCRF-CEM (ATCC, CCL-119, RRID:CVCL_0207) cells were newly purchased from ATCC. SKM-1 (DMSZ, ACC 547, RRID:CVCL_0098) and MOLM-13 (DMSZ, ACC 554, RRID:CVCL_2119) were newly purchased from DMSZ. All cell lines used in experiments were confirmed to be negative for Mycoplasma. MOLM-13 and CCRF-CEM cells were cultured in RPMI 1640 medium with 10% FBS. SKM-1 cells were cultured in RPMI 1640 with 20% FBS. All cell lines were cultured in medium containing 1% penicillin/streptomycin at 37°C with 5% CO2. CCRF-CEM, MOLM-13, and SKM-1 cell lines were transduced with a lentivirus containing an MCherry fluorophore, an sgRNA (either non targeting or targeting human Elovl1), and Cas9 in the lenticrisprV2 backbone (Addgene, Plasmid #99154, RRID:Addgene_99154). Cells were cultured for up to 14 days. At harvest, flow cytometry and cell counting (ViCell Blu) was performed.

Cloning

CRISPR libraries were cloned into Lentiguide-BFP backbones as previously described(16). For cloning individual sgRNAs, protospacer sequences for each target gene were identified using the CRISPick algorithm (Broad Institute) and complementary oligonucleotides containing BsmBI-overhang sequences were synthesized (IDT). For single murine sgRNA validation experiments, sgRNA sequences were unique from those used in the original screen. Complementary oligonucleotides were annealed and ligated into BsmBI-digested LentiGuide-MCherry (Addgene plasmid # 104375, RRID: Addgene104375) or LentiGuide-BFP where noted. For human cell line sgRNA validation experiments, sgRNAs were designed as described above and ligated into BsmBI-digested LentiCrisprV2-MCherry (Addgene # 99154) or LentiCrisprV2-BFP where noted. The sgRNA expression backbone LenticrisprV2-BFP was cloned with a single Gibson reaction (New England Biolabs) replacing the MCherry in LentiGuide-MCherry with BFP. For Elovl1 add-back constructs, a single Gibson reaction was performed with digested Lentiguide-MCherry constructs containing Elovl1 sgRNA and gene blocks with a codon optimized murine Elovl1 cDNA containing a 3’ P2A sequence. Codon optimization was performed to eliminate sgRNA homology and protospacer sequences when possible without modifying final amino acid sequence. The complete addback construct contains a U6 promoter driven Elovl1 sgRNA and an EF-1 alpha driven Elovl1 cDNA separated from MCherry with a P2A sequence. For catalytically dead add-back constructs, gene blocks were ordered with codon optimized Elovl1 cDNA as described above with catalytic core HXXHH motif mutated to AXXAA or QXXQQ. For individual sgRNAs and cDNA constructs, DNA was transformed into Stabl3 chemically competent bacteria (Thermo), bacterial DNA was extracted from colonies using Miniprep or Midiprep Kits (Qiagen), and plasmids were sequence confirmed using oxford nanopore sequencing (Plasmidsaurus).

Lentivirus Production

For lentivirus production, plasmids were co-transfected into 293Ts with the packaging vector psPAX2 (RRID:Addgene_12260) and the envelope vector pMD2.G (RRID:Addgene_12259) using the jetPRIME transfection reagent (Polyplus) following manufacturer’s protocol. Media was changed 24 hours post transfection. Viral-containing supernatant was collected 72 hours after transfection and passed through a 0.45-micron filter. Virus was ultracentrifuged at 24,000 rpm for 2 hours at 4°C using a Sorvall Ultracentrifuge (Thermo) or at 1500 x g at 4°C for 40 minutes using Amicon Ultra 15 tubes (Millipore Sigma). Viral pellets were resuspended in remaining media and rotated for 4–24 hours at 4°C, following which virus was aliquoted and stored at −80°C for long-term storage.

Flow Cytometry and Sorting

Freshly isolated cells were washed with FACS buffer. For cell surface staining, cells were then resuspended in Fc (CD16/32, 2.4G2) block for 15 minutes at 4°C and then subsequently incubated with antibody cocktails for an additional 25 minutes at 4°C. For peripheral blood analysis model without fluorophore tagged sgRNAs, antibodies targeted CD3 (17A2, RRID:AB_1595492), CD45R/B220 (RA3–6B2, RRID: AB_312990), Gr-1 (RB6–8C5, RRID:AB_2137485), CD11b (M1/70, RRID:AB_2721166), CD45.2 (104, RRID:AB_313443), Kit (2B8, RRID:AB_2629799), and CD45.1 (A20, RRID:AB_313503). For analysis of BMEC co-cultures, antibodies targeted Gr-1 (RB6–8C5), CD11b (M1/70), CD45.2 (104), Kit (2B8), and CD34 (RAM34, RRID:AB_10596826) were used. For BMEC co-cultures with BFP tagged sgRNAs additional antibodies targeting B220 were used. For bone marrow analysis, additional antibodies targeted Sca-1 (D7, RRID:AB_493596), FcγRII/III (93, RRID: AB_2563039), CD34 (RAM34), CD150 (TC15–12F12.2, RRID: AB_493461), and CD48 (HM48–1, RRID: AB_493489) were used as well as Ly6C (HK1.4, RRID:AB_1659241) and Ly6G (1A8, RRID: AB_2616732). For bone marrow flow characterization in Fig. 1, additional antibodies targeting Cd11c (N418, RRID: AB_287343), Nk1.1 (PK136, RRID: AB_2871088), Ter119 (Ter-119, RRID: AB_11204244), Fcer1a (MAR-1, RRID: AB_2574632) were used. All antibodies were purchased from BD, BioLegend or eBioscience. Following antibody incubation, cells were washed with FACS buffer and resuspended in FACS buffer containing a live/dead cell stain, either DAPI (1:1000 in FACS buffer, Thermo) or Helix NP NIR (1:10,000 in FACS buffer, Biolegend). Flow cytometry was performed on a BD LSRFortessa (BD Biosciences) or a using FACSDiva (RRID:SCR_001456) and FlowJo (v10.9, RRID:SCR_008520).

For assays with sorted cells, cells were resuspended in Fc (CD16/32) block (BD Biosciences) for 15 minutes and then subsequently stained for CD45.2 (Clone 104) for 30 minutes. All FACS antibodies were purchased from BD, BioLegend, or eBioscience. Following antibody incubation, cells were washed with FACS buffer and resuspended in a live/dead cell stain solution, either DAPI (Thermo, 1:1000 in FACS) or Helix NP NIR (Biolegend, 1:10,000 in FACS). Cells were then sorted on a SONY SH800S.

For assays with human cells, cells were resuspended in Fc block (Human TruStain FcX, Biolegend) for 15 minutes then stained with human CD45 (HI30, RRID:AB_2564154), human CD117 (104D2, RRID: AB_2629837), human CD34 (561, RRID: AB_2074356), CD11b (M1/70, RRID:AB_2721166), CD15 (W6D3, RRID: AB_2561669), CD19 (H1B19, RRID: AB_314248) for 30 minutes. All FACS antibodies were purchased from BD, BioLegend, or eBiosciences. Following antibody incubation, cells were washed with FACS buffer and resuspended in a live/dead cell stain solution containing DAPI (Thermo, 1:1000 in FACS). Flow cytometry was performed on a BD LSRFortessa (BD Biosciences) using FACSDiva and FlowJo (v10.9).

2-HG Measurements

Serum samples (100 μL) were extracted by adding 400 μL of ice-cold methanol containing D/L-2-hydroxyglutaric acid-d3 (D/L-2-HG-d3 disodium salt, Toronto Research chemical, H942578) as the internal standard. After overnight incubation at −80° C, extracts were centrifuged at 20,000 g for 20 minutes at 4°C to precipitate protein and 800 μL of the supernatant was dried in an evaporator (Genevac EZ-2 Elite). Samples were then derivatized with 100 μL of freshly prepared 50 mg/mL (+)-diacetyl-L-tartaric anhydride (DATAN, Sigma) in dichloromethane-acetic acid (v/v=4:1) at 75° C for 30 minutes. After cooling to room temperature, derivatized samples were dried under nitrogen at room temperature and resuspended in 100 μL of water for LC-MS/MS analysis. LC-MS/MS analysis was performed using a Thermo Vantage triple-quadrupole mass spectrometer operating in MRM and negative ionization modes. LC separation was achieved using an Acquity UPLC HSS T3 analytical column (2.1 × 100 mm, 1.8 μm, Waters) and an Agilent 1260 infinity binary pump. Mobile phase A was 125 mg/L ammonium formate in water, adjusted to pH 3.5 with formic acid, and mobile phase B was methanol; flow rate was 0.3 mL/min. Initial conditions were 3% B for 5 min, increasing to 97% B at 5.5 min and held for a further 2.5 min. 7 min of re-equilibration time was used to ensure retention time stability. The column temperature was held at 40° C. Samples were kept at 4° C and the injection volume was 10 μL. MS source parameters were: spray voltage 2500 V; capillary temperature 300° C; vaporizer temperature 250° C; sheath gas pressure 50; aux gas pressure 40; S-lens 37 V. Individual reactions monitored and collision energies (CE) were: D/L-2-HG m/z 363.0 → 147.1 (CE: 12 V)*, 129.1 (CE: 27 V); D3–2HG m/z 366.0 → 150.1 (CE: 12 V)*, 132.0 (CE: 27 V), with * indicating the primary transition used to quantify each metabolite. Pure standards of L-2-HG and D-2-HG were derivatized and analyzed in parallel and the identity of the 2-HG enantiomers was determined by comparing them to the retention times of the derivatized pure standards. Data was acquired and processed with TraceFinder 4.1 software (ThermoFisher, RRID:SCR_023045).

Very Long Chain Metabolomics Measurements

1 million cells per mouse replicate per condition were utilized. Frozen cell pellets were thawed and extracted using an acidified Folch extraction. To the cell pellets, 700 μL of chloroform was added. After brief vortexing, 300 μL of methanol containing Palmitic Acid-d31 (Cayman Chemical) as the internal standard was added, followed by the addition of 25 μL of 1 N HCl. Thereafter, 180 μL of water was added to induce phase separation. After vortexing, the extract was centrifuged at 3,000 g for 5 min at 4°C. Afterwards, the lower, chloroform layer was obtained and transferred to a glass vial. Samples were then re-extracted with 700 μL of chloroform:methanol:water (2:1 v/v). After brief vortexing, the lower, chloroform layer was collected and pooled with the previous chloroform layer in the same glass vial. The combined chloroform layer was dried under nitrogen and reconstituted in 100 μL of acetonitrile:isopropanol (1:1 v/v) for LC–MS analysis.

Fatty acid profiling was performed using an Agilent 6546 Q-TOF mass spectrometer operated in negative ion mode, coupled to an ACQUITY UPLC BEH C18 column (100 mm × 2.1 mm, 1.7 μm; Waters). Mobile phase A consisted of 10 mM ammonium acetate in 60:40 acetonitrile:water, and mobile phase B consisted of 10 mM ammonium acetate in 85:10:5 isopropanol:acetonitrile:water. The LC gradient was as follows: 0 min, 10% B; 2.0 min, 10% B; 16 min, 60% B; 21 min, 100% B; 24.1 min, 10% B; 27.6 min, 10% B. Other LC parameters included a flow rate of 0.3 mL/min, a column temperature of 45 °C, and an injection volume of 5 μL. MS source parameters included: gas temperature: 200°C; gas flow: 11 L/min; nebulizer pressure: 50 psig; sheath gas temperature: 300°C; sheath gas flow: 12 L/min; VCap: 3000 V; nozzle voltage: 0 V; fragmentor: 150V. Data were acquired from m/z 50–1700 with active reference mass correction (m/z 301.9981 and 980.0163). Data processing, including peak identification and integration, was carried out using Skyline(89) (RRID:SCR_014080). Raw peak areas were baselined to mean of control followed by log 2 transformation.

Stable Isotope Tracing to Fatty Acids and Sphingolipids using LC-MS

Bone marrow from IN and INR mice (n=4 per arm) was freshly harvested and 36 million cells per mouse were plated across two non tissue culture treated 12 well plates; cells were allowed to recover overnight. The next morning, uniformly ¹³C Oleate (Cambridge Isotope Laboratories CLM-460–0.1) or unlabeled ¹²C Oleate (Cambridge Isotope Laboratories ULM-10649–0.1) conjugated to fatty acid free BSA (6:1 molar ratio of VLCFA: BSA) was added at a final concentration of 100uM tracer. At 4 hours and 24 hours, cell pellets were harvested, cell counts were normalized, and pellets were snap frozen on dry ice.

Frozen cell pellets were thawed and extracted using a Folch extraction. To the cell pellets, 300 μL of methanol was added, briefly vortexed, followed by the addition of 600 μL of chloroform. Thereafter, 180 μL of water was added to induce phase separation and incubated at room temperature for 10 min. After vortexing, the extract was centrifuged at 16,000 g for 5 min at 4°C. Afterwards, the lower, chloroform layer was obtained and transferred to a glass vial. Samples were then re-extracted with 700 μL of chloroform:methanol:water (2:1 v/v). After brief vortexing, the lower, chloroform layer was collected and pooled with the previous chloroform layer in the same glass vial. The combined chloroform layer was split into equal volumes (500 μL) for fatty acid and sphingolipid analysis respectively. The organic extracts were then dried under nitrogen and each reconstituted in 100 μL of acetonitrile:isopropanol (1:1 v/v) for LC–MS analysis.

Stable isotope tracing to fatty acids was performed using an Agilent 6546 Q-TOF mass spectrometer operated in negative ion mode, coupled to an ACQUITY UPLC BEH C18 column (100 mm × 2.1 mm, 1.7 μm; Waters). Mobile phase A consisted of 10 mM ammonium acetate in 60:40 acetonitrile:water, and mobile phase B consisted of 10 mM ammonium acetate in 85:10:5 isopropanol:acetonitrile:water. The LC gradient was as follows: 0 min, 10% B; 2.0 min, 10% B; 16 min, 60% B; 21 min, 100% B; 24.1 min, 10% B; 27.6 min, 10% B. Other LC parameters included a flow rate of 0.3 mL/min, a column temperature of 45 °C, and an injection volume of 5 μL. MS source parameters included: gas temperature: 200°C; gas flow: 11 L/min; nebulizer pressure: 50 psig; sheath gas temperature: 300°C; sheath gas flow: 12 L/min; VCap: 3000 V; nozzle voltage: 0 V; fragmentor: 150V. Data were acquired from m/z 50–1700 with active reference mass correction (m/z 301.9981 and 980.0163).

Stable isotope tracing to sphingolipids was performed using an Agilent 6546 Q-TOF mass spectrometer in both positive and negative ionization modes, coupled to a ZORBAX Eclipse Plus C18 column (100 mm × 2.1 mm, 1.8 μm particle size; Agilent Technologies). Mobile Phase A consisted of 10 mM ammonium formate in 50:30:20 water:acetonitrile:isopropanol. Mobile Phase B consisted of 10 mM ammonium formate in 1:9:90 water:acetonitrile:isopropanol. The LC gradient was as follows: 0 min, 0% B; 2.7 min, 45% B; 2.8 min, 53% B; 9 min, 65% B; 9.1 min, 89% B; 11 min, 92% B; 11.1 min, 100% B; 12 min, 10% B; 15 min, 10% B. The flow rate was 0.4 mL/min; column temperature was maintained at 60 °C, and the injection volume was 5 μL. MS source parameters were as follows: gas temperature, 200 °C; gas flow, 10 L/min; nebulizer pressure, 50 psig; sheath gas temperature, 300 °C; sheath gas flow, 12 L/min; VCap, 3,000 V; nozzle voltage, 0 V; fragmentor, 150 V. Data were acquired from m/z 50–1700 with active reference mass correction (positive mode: m/z 322.0481 and 922.0097; negative mode: m/z 301.9981 and 966.0007). Five iterative MS/MS acquisitions were performed on pooled lipid extracts in both polarities for lipid annotation. Lipid identification performed using Agilent Lipid Annotator (v1.0.1).(90) Peak picking and integration was performed using Skyline (MacCoss Lab),(89) and natural abundance correction was achieved using IsoCorrectoR.(91) For data analysis, percent labeled was calculated by dividing the labeled pool for a given sample by the total unlabeled pool. Unpaired T test was performed to assess significance.

Metabolite Add-Back Experiments

Whole bone marrow was freshly harvested from INR mice, transduced with lentivirus containing MCherry tag and indicated sgRNA, and plated over BMECs in Stemspan media with 20ng/mL mouse SCF as described above. At time of plating, either Vehicle (DMSO), 20uM C24:0 VLCFA (MedChemExpress, Catalogue #: HY-121883), or 20uM C24:1 VLCFA (MedChemExpress Catalogue #: HY-N2526) was added to media after Fatty Acid Free BSA conjugation. Briefly for BSA conjugation, 10% Fatty Acid Free BSA and media was pre-warmed and vehicle, C24:0, or C24:1 was added. Mixture was warmed at 37 degrees Celsius for at least 30 minutes before diluting 1:10 in media and adding dropwise to cells. The final VLCFA:BSA molar ratio was ~6:1. Flow cytometry analysis was performed after 7 days in culture.

Fractionation Proteomics Experiments

Experimental Set Up: Whole bone marrow was freshly harvested from INR mice (bone marrow from 4–5 mice were pooled per replicate) and plated in 10cm plates. 60 million cells / replicate were plated and cells were allowed to recover overnight. The next morning, Vehicle (DMSO) or 2mM Methyl Beta Cyclodextrin (MedChemExpress, Catalogue # HY-101461) was added to the cells. Cells were harvested at 24 hours, cell numbers were normalized, and 12 million cells were utilized for fractionation. For Elovl1 inhibitor experiments, whole bone marrow was freshly harvested from INR mice (bone marrow from 4–5 mice were pooled per replicate) and plated over BMECs overnight and allowed to recover. They were then treated with Vehicle versus 10uM Compound 27 Elovl1 inhibitor for 5 days. 20 million cells were then utilized for fractionation. Membrane and cytoplasmic fractions were isolated according to manufacturer’s instructions (Thermo Fisher Scientific, Catalog No. PI78840) and fractions were snap frozen on dry ice. Protein Digestion: Protein concentrations of membrane and cytoplasm fractions were determined using the Piece Protein Assay Kit (Thermo Fisher Scientific). A total of 50 mg of protein per sample was reduced with tris (2-carboxyethyl) phosphine (TCEP) at a final concentration of 10 mM for 15 minutes at room temperature (RT) with shaking at 1,000 rpm on a Thermomixer. Free cysteine residues were alkylated with iodoacetamide (IAA) at a final concentration of 10 mM for 20 minutes in the dark at RT with shaking at 1,000 rpm. SP3 Sera-Mag Carboxylate SpeedBeads beads (Cytiva, cat#: 65152105050250; 45152105050250, 50 mg/ml) were prepared and added to the samples according to the manufacture’s instructions. Beads were then resuspended in 50 mL of 50 mM triethylammonium bicarbonate (TEAB), and proteins were digested overnight at 37°C with shaking at 1,200 rpm using trypsin (Promega) at an enzyme-to-protein ratio of 1:50. After digestion, 1.2 mL of 100% acetonitrile (ACN) was added, followed by incubation for 10 minutes at RT with shaking at 1,200 rpm. Samples were washed three times with 100% ACN, and peptides were eluted by resuspending the beads in 50 mL of 2% DMSO at 37°C for 30 minutes with shaking at 1,500 rpm. The eluted peptides were dried under vacuum, reconstituted in water with 0.1% trifluoroacetic acid (TFA), and sonicated in a water bath sonicator. Peptides were desalted using C18 stage tips, sequentially conditioned with i) methanol), ii) 70% ACN/0.1% TFA, iii) 0.1% TFA twice. Acidified peptide samples were loaded onto the conditioned stage tips, washed twice with 0.1% formic acid (FA), and eluted twice with 70% ACN/0.1% FA. Eluates were dried under vacuum using a SpeedVac centrifuge. Dried peptides were reconstituted in 12 mL of 0.1% FA, sonicated, and transferred to autosampler vials. Peptide yield was estimated using a NanoDrop spectrophotometer (BioTek Synergy H1, Agilent). Mass spectrometry analyses: Peptides were separated using a PepSep analytical column (25 cm x 75 mm inner diameter). The peptides were eluted using a gradient 3%−26% buffer B over 80 minutes, followed by 26%−34% buffer B over 10 minutes, an increase to 95% buffer B over 0.1 minutes and a final hold at 95% buffer B for 19.9 minutes. Buffer A consisted of 0.1% FA in HPLC-grade water, and buffer B consisted of 99.9% ACN with 0.1% FA. Separation was performed at a flow rate of 300 nL/min using a NanoElute2 system (Bruker). MS data were acquired on a timsTOF HT instrument (Bruker) equipped with a CaptiveSpray source using a data-independent acquisition PASEF (dia-PASEF) method. The mass range was set from 100 to 1700 m/z, with an ion mobility range of 0.70 V.s/cm²-1.3 V.s/cm² and collision energies 20–59 eV. Ramp time and accumulation time were set to 100 ms. The dia-PASEF method covered a m/z range of 299.5–1200.5 with an estimated cycle time of 0.85 s. The isolation windows were defined with a width of 36 Da, a 1 Da overlap, and 25 windows per cycle. Data Analysis: Raw data were processed using Spectronaut version 20.5 (Biognosys) and searched with the PULSAR search engine against a Mus musculus UniProt protein database (downloaded on June 18, 2025; 25,666 entries, RRID:SCR_002380). Cysteine carbamidomethylation was set as fixed modifications, while methionine oxidation, protein N-terminal acetylation, and deamidation (N/Q) were specified as variable modifications. A maximum of two missed cleavages was allowed. A reverse decoy database strategy was used to control the false discovery rate (FDR), which was set at 1% at the peptide level. Statistical analysis was performed using an unpaired t-test. Volcano plots were generated based on log2 fold change and q-value (Benjamini-Hochberg adjusted p-values). A q-value ≤0.05 was considered statistically significant. For pathway analysis and differential protein expression, targets with |logFC| > 0.25 and q value < 0.05 are included.

Histology and Pathology Imaging and Staining

Spleen, liver, and tibia samples were fixed (4% paraformaldehyde) for >24 hours and embedded in paraffin. Sections were cut using microtome (Mikrom International AG), mounted on slides, and dried at 37°C overnight. Hematoxylin and eosin (H&E) staining was performed on a COT20 stainer (Medite). All sample handling and preparation was performed by the Tri-Institutional Laboratory of Comparative Pathology (LCP) facility. Pictures were taken at a total of 400X magnification (10x magnification eyepiece and 40x objective) using an Olympus microscope. Tissue sections were formally evaluated by a hematopathologist (W. Xiao).

CRISPR Cut-Site Sequencing

To assess sgRNA editing efficiency, CRISPR Cut-Site Sequencing was performed on all sgRNAs (Supplemental Fig. 12A–U). DNA was extracted from sorted BFP+ or MCherry+ targeted Cas9 containing wild-type bone marrow cells cultured over BMECs 3–7 days post viral transduction. PCR amplicons of the targeted site was generated. Sequencing libraries were prepared from amplicons using the KAPA Hyper Library Preparation Kit (Kapa Biosystems) according to the manufacturer’s instructions using 8 cycles of PCR. Barcoded libraries were pooled equivolume and run on a MiSeq or NovaSeq 6000 (Illumina). The CRISPResso2 package was used to filter low quality reads, trim adapters, align reads to the reference amplicon, and quantify editing frequency in the amplicon(92).

RNA Extraction, Sequencing and Data Analysis

RNA was isolated from up to 125,000 FACS purified live cells sorted into Trizol LS (Thermo Fisher, Cat# 10296010). N=3 biological replicates were used per condition. Phase separation in cells lysed in RLT Buffer (Qiagen catalog # 79216) was induced with chloroform. RNA was extracted from the aqueous phase using the miRNeasy Mini Kit (Qiagen catalog # 217004) on the QIAcube Connect (Qiagen) according to the manufacturer’s protocol. Samples were eluted in 13 μL RNase-free water. After RiboGreen quantification and quality control by Agilent BioAnalyzer, 2 ng total RNA with RNA integrity numbers of 10 underwent poly(A) selection and amplification using the SMART-Seq v4 Ultra Low Input RNA Kit (Clonetech catalog # 63488), with 12 cycles of amplification. Subsequently, 10 ng of amplified cDNA was used to prepare libraries with the KAPA EvoPrep Kit (Roche 10212250702) using 8 cycles of PCR. Samples were barcoded and run on a NovaSeq X in a PE100 run, using the NovaSeq X 25B Reagent Kit (Illumina). An average of 55 million paired reads were generated per sample and the percent of mRNA bases per sample ranged from 92% to 94% and ribosomal reads averaged 0.28%.

RNA sequencing reads were 3’ trimmed for base quality >15 and adapter sequences using version 0.4.5 of TrimGalore (RRID:SCR_011847) and aligned to mouse assembly mm10 with STAR v2.4 (RRID:SCR_004463) using default parameters. Data quality and transcript coverage were assessed using the Picard tool CollectRNASeqMetrics. Read count tables were generated with HTSeq v0.9.1 (RRID:SCR_005514). Normalization and differential expression were evaluated with DESeq2 (RRID:SCR_000154) using an FDR-adjusted p-value threshold of 0.1 and magnitude of fold change >= 1.5. Gene set enrichment analysis (GSEA, http://software.broadinstitute.org/gsea, v3.0) was run against MSigDB v6 using the pre-ranked option and log2 fold change for pairwise comparisons. Network analysis was performed using enrichplot::cnetplot in R with default parameters.

Cellular Indexing of Transcriptomes Sequencing (CITE-Seq) and Data Analysis

Bone marrow cells were infected with BFP expressing lentivirus containing sgRNAs targeting Rosa and Elovl1. After 5 days in BMEC culture, live BFP+ hematopoietic cells were FACS sorted. Cells were stained with Trypan blue and Countess II Automated Cell Counter (ThermoFisher) was used to assess both cell number and viability. Following QC, the single cell suspension was loaded onto Chromium GEM-X Single Cell 5’ v3 Chip (10x Genomics PN 1000698) and GEM generation, cDNA synthesis, cDNA amplification, and library preparation of 26–33,000 cells proceeded using the GEM-X Single Cell 5’ Kit v3 (10x Genomics PN 1000699) according to the manufacturer’s protocol. Amplification of cDNA proceeded for 12 cycles, after which 10 μL of the resulting material was used for sequencing library construction with 10 cycles of PCR. Libraries were indexed, pooled at equimolar ratios, and sequenced using a NovaSeq X (PE28/88 run format, NovaSeq X 25B Reagent Kit, Illumina), yielding an average of 23 thousand reads per cell. Feature barcode analysis for cell surface proteins was performed on amplification products—generated as described above—containing both cDNA and feature barcodes labeled with cell barcodes and unique molecular identifiers (UMIs). Longer amplified cDNA fragments were removed from smaller feature barcode fragments via a 0.6X AMPure XP bead cleanup (Beckman Coulter, catalog # A63882). Library construction was carried out per the manufacturer’s protocol using the 5’ Feature Barcode Kit (10X Genomics, PN 1000256), with 12 cycles of PCR. As with the cDNA libraries, feature barcode libraries were indexed, pooled equimolar, and sequenced on a NovaSeq X (PE28/88, NovaSeq X 25B Reagent Kit, Illumina), generating an average of 156 million reads per sample.

Raw FASTQ files were processed using Cell Ranger (v9.0.1; 10x Genomics, RRID:SCR_017344)(93) with the multi-modal workflow (`cellranger multi`), which simultaneously performs read alignment to mm10, UMI deduplication, cell-barcode calling, and HTO-based sample demultiplexing. The resulting quality-filtered feature-barcode matrices generated by Cell Ranger were used for all subsequent analyses. Low-quality or dying cells were identified and removed using an iterative clustering and filtering strategy. To identify low-quality cells, we first combined count matrices from all samples and clustered cells using PhenoGraph(94) (k = 30) implementation in Scanpy (v1.9.5, RRID:SCR_018139)(95). For each cluster, we evaluated multiple indicators of poor cell quality, including low total transcript counts (library size), low library complexity (unique genes per UMI), and elevated mitochondrial RNA content (>5%). Clusters enriched for cells exhibiting these characteristics were removed, and the remaining cells were reclustered. This procedure was repeated until no discrete clusters of low-quality cells remained. Raw counts were retained in a separate layer for highly variable gene selection. Counts were normalized per cell to the median library size and log-transformed (natural log, pseudocount = 1). Highly variable genes were identified from the raw-count layer using the Scanpy `highly_variable_genes` function with `flavour = seurat_v3`. Dimensionality reduction and downstream analyses were performed using the log-normalized expression matrix. Specifically, 60 principal components (PCs) were computed using the highly variable genes, explaining approximately 42% of the variance. A k-nearest-neighbor graph (k = 20) was constructed in PCA space using Euclidean distance with the Scanpy (v1.9.5) ǹeighbors` function. Cells were embedded using UMAP (min_dist = 0.5). Clusters were identified from the PCA representation using the PhenoGraph implementation (RRID:SCR_016919) in Scanpy with the Leiden community detection (k = 20).

Differentially expressed genes (DEGs) for each cluster versus all other clusters were identified using the `rank_gene_groups` function in Scanpy with the Wilcoxon rank-sum test and Benjamini-Hochberg correction. Genes with adjusted p value < 0.05 and log₂ fold change > 0.25 were considered significantly differentially expressed (Supplementary Table 2). Cluster-specific DEGs were further used for downstream cell-type annotation. To identify transcriptional differences between sgElovl1 and sgRosa genotypes within each cluster, we again applied the `rank_gene_groups` function in Scanpy, using sgRosa as the reference group. Gene Set Enrichment Analysis (GSEA, RRID:SCR_003199) was performed using gseapy(96) (v1.0.6) with 10,000 permutations and default parameters. Enrichment was assessed against the Hallmark gene sets from MSigDb(97). For integration with R-based workflows, the processed AnnData object was converted to a SingleCellExperiment object using zellkonverter (v1.12.1) and then to a Seurat object (v5.2.1) for analysis in R (v4.3.1). Antibody-derived tag (ADT) counts were stored in a separate assay within the Seurat object and normalized using centered log-ratio (CLR) transformation with per-cell scaling (`margin = 2`).

Cell type annotation combined automated and manual approaches. Leiden clusters were first annotated using SingleR(98) (v2.4.0) against the ImmGen reference dataset with fine-grained labels, and predictions were subsequently refined through manual inspection of cluster-specific marker genes. Differential neighborhood abundance between sgElovl1 and sgRosa genotypes was assessed using MiloR(99) (v1.10.0) with k = 30 for neighborhood detection. Neighborhoods with FDR < 0.05 were considered significantly differentially abundant.

Pathway activity inference was performed using PROGENy(67) (v1.24.0). For each cell, PROGENy z-scores were computed from normalized gene expression values. Differences in pathway activity between sgElovl1 and sgRosa were quantified using a linear model for each pathway, with the z-score as the response and genotype as the predictor. Model coefficients represented pathway activity differences, and significance was assessed using Benjamini-Hochberg-adjusted p values.

Processing of TCGA and Publicly Available Single Cell Datasets

We used TCGA RNA-sequencing data (NEJM 2013) patients from the cBioPortal for cancer genomics(100). Data for the presence of mutations in NRAS or NRAS Pathway mutations (defined as NRAS, PTPN11, CKIT, CBL(4, 101)) and expression of ELOVL1 was downloaded. For each individual dataset, we stratified patients based on the presence of mutations in NRAS (or NRAS pathway as indicated) plotted expression scores across these subsets. TCGA samples were split into top and bottom 50% ELOVL1 expression groups for survival analysis. ELOVL1 expression data and GSEA analysis derived from Genotype of Transcriptomes data in NRAS mutant clones was described previously(72).

Patient Samples and Patient Derived Xenograft Experiments

Seven viably frozen patient samples were obtained from the MSKCC Hematologic Oncology Tissue Bank (HOTB). An additional 8^th patient sample (“PDX AML9”) was obtained from the BEAT-AML cohort.(31) Written informed consent was obtained from patients according to protocols approved by the institutional IRBs and in accordance with the Declaration of Helsinki. This study was approved by MSKCC Institutional Review Board (Protocol #16–171). Bone marrow or PBMCs from patients were collected and processed by the MSKCC Human Oncology Tissue Bank (HOTB). Mononuclear cells were obtained by centrifugation on Ficoll from peripheral blood or bone marrow and viably frozen (stored in −80 or liquid nitrogen until used). Samples were included if they were from an AML patient sample with a known mutational background as assessed by targeted sequencing. Additional sample information can be found in Supplemental Table 3. For PDX generation, up to 10 million bone marrow or PBMCs were injected to NSGS mice (The Jackson Laboratory, Catalogue # 013062, RRID:IMSR_JAX:013062). When mice were moribund or >85% hCD45+ on bone marrow aspirate, bone marrow was passaged into sublethally irradiated (200 cGy) NSG mice (The Jackson Laboratory, Catalogue # 005557, RRID:IMSR_JAX:005557). Experiments were performed on bone marrow from secondary or tertiary transplants. For Elovl1 inhibitor experiments, cells were plated over primary Human Umbilical Vein Endothelial Cells (HUVECs, RRID:CVCL_2959) and treated with vehicle (DMSO) or 10uM Compound 27, a potent and selective Elovl1 inhibitor(71) (Cayman Chemical, ELOVL1–27, Item No. 35323) for 7 days then analyzed or transplanted into sublethally irradiated male NSGW41 (Jackson Labs, #026497, RRID:IMSR_JAX:026497) mice to assess engraftment potential. Readout was performed using flow cytometry and cell counting (Vicell Blu). In vivo readout was performed on bone marrow aspirates. For non PDX patient samples with high CD34 expression, cells were enriched initially with human CD34 beads (Miltenyi, Catalogue # 130-046-702) before plating and treatment. Ex Vivo culture media consisted of StemSpan SFEM (StemCell Technologies) supplemented with 10% knock out serum replacement (Gibco), 1% P/S, Non-Essential Amino Acids (Gibco), L-Glutamine (Gibco), HEPES (Corning), and Heparin Sodium (Sigma Aldrich) as well as 50ng/mL human SCF (MedChemExpress), TPO (Peprotech), FLT3-L (Peprotech) and 10ng/mL FGF (Biolegend). Additional patient sample information is available in supplemental table 3.

Statistical Analyses

Statistical analyses were performed using Student’s t-test using GraphPad Prism (GraphPad Software, San Diego, CA, RRID:SCR_002798). For comparing more than 2 groups, an ordinary One Way Anova was performed with Dunnett’s multiple comparison test unless otherwise noted. P < 0.05 was considered statistically significant. The number of animals, technical replicates, and experimental replications can be found in the respective figure legends.

STATEMENT OF SIGNIFICANCE.

Oncogenic NRAS mutations promote therapeutic resistance in AML. We utilize novel isogenic models to perform CRISPR dependency screens in primary cells and identify Elovl1 as an Nras-selective target. We show Elovl1-derived very long chain sphingolipid species are utilized in lipid rafts to sustain Nras-mutant signaling.

DATA AVAILABILITY STATEMENT

The data generated in this study are publicly available in Gene Expression Omnibus (GEO) at GSE310778 and GSE310553. Proteomics data is publicly available on the PRIDE database (accession # PXD081526).