Ferumoxytol enhances NF-κB-responsive GBA-targeted ferroptosis gene therapy in acute myeloid leukemia Authors: Tao Luo, Shijie Zhang, Meng Li, Kang Ding, Feng Gu Published: 2026 ABSTRACT AbstractAcute myeloid leukemia (AML) has a poor prognosis and is associated with aberrant NF-κB activation, providing a tumor-selective therapeutic window. Therefore, a combination therapy involving FDA-approved ferumoxytol (Feraheme, FeNPs) and NF-κB-responsive GBA interference was developed in this study. GBA is a key enzyme that maintains lysosomal lipid metabolism and membrane composition homeostasis. Its inhibition causes lysosomal storage disorders and functional impairment, leading to the release of labile iron that contributes mainly to ferroptosis in AML. Using a miR30-based shRNA system, we constructed and screened an AML-specific GBA-knockdown plasmid, pNM6-miGBA (miGBA). Low-dose FeNPs promote the Fenton reaction and amplify an NF-κB positive feedback loop, markedly enhancing miGBA-mediated cytotoxicity without activating NF-κB in normal cells. Detection of ferroptosis biomarkers, transmission electron microscopy, and RNA-seq confirmed that the synergistic mechanism of FeNPs and miGBA involves ferroptosis. The combination significantly prolonged survival in AML-bearing mice and produced no notable toxicity to major organs, hematologic parameters, or liver and kidney function tests. In summary, combination therapy with FeNPs and miGBA achieves potent anti-AML efficacy with favorable in vivo safety, and has potential for clinical development.Graphical abstract Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12951-026-04731-5.Keywords: GBA, NF-κB, Ferroptosis, Lysosome, AML Introduction Acute myeloid leukemia (AML) is a highly heterogeneous malignancy arising from bone marrow hematopoietic stem and progenitor cells. Clinically, it features abrupt onset, high relapse rates, and frequent treatment resistance [1]. Although targeted and immune therapies have advanced in some subtypes, conventional induction chemotherapy often fails to cure elderly patients [2, 3]. Cases that involve chemoresistant leukemia stem cells (LSCs) are particularly difficult to eradicate [4]. At the molecular level, the NF-κB signaling pathway is aberrantly activated in AML cells, where it regulates cell proliferation, antiapoptotic mechanisms, antioxidant defenses, and stress responses [5–7]. Sustained NF-κB activity, which is marked by phosphorylated p65 (p-p65), supports LSCs survival and chemotherapy resistance [8, 9]. Therefore, NF-κB is both a key node driving resistance and relapse and an attractive therapeutic target. However, it has been difficult to develop NF-κB inhibitors as drugs, mainly because of their lack of selectivity, and they can burden normal cells or the immune system [10]. How to exploit NF-κB hyperactivation in AML for therapeutic intervention remains a major challenge for current AML therapies. Since the concept of ferroptosis was proposed in 2012, its potential value for cancer therapy has drawn widespread attention [11, 12]. Studies have shown that ferroptosis can kill tumor stem cells and overcome tumor drug resistance; hence, this form of cell death is considered vulnerable to refractory tumors [13, 14]. Ferroptosis is iron dependent and driven by lipid peroxidation (lipid ROS, lipROS). The immune system may partly prevent tumorigenesis by inducing ferroptosis [15]. Ferroptosis exerts potent killing effects on adjacent cells and has the potential to propagate in a wave-like manner among cells [16]. Therefore, the antitumor effect of ferroptosis has been explored rapidly and widely across diverse cancers [14, 17]. Iron oxide nanoparticles (IONPs) induce cellular iron accumulation and are widely used in imaging and tumor therapy [18–21]. Ferumoxytol (Feraheme, FeNPs) is an FDA-approved intravenous iron supplement. Its core is composed of Fe3O4/γ-Fe2O3, and the particles localize primarily to lysosomes after cellular uptake [22]. Studies have shown that in AML cells with low expression of the iron exporter ferroportin (FPN), FeNPs kill AML cells in lysosomes via the Fenton reaction while protecting hematopoietic stem cells [22, 23]. Lysosomes contain numerous acidic hydrolases, which serve as hubs for IONPs degradation and iron release. Lysosomal membrane integrity is critical for cell fate [24]. When lysosomal membrane permeability increases or integrity is compromised, large amounts of toxic contents, such as free iron, can escape into the cytosol [24, 25]. This release triggers multiple cell death pathways, including ferroptosis, autophagy, apoptosis, necrosis, pyroptosis, and lysosome-dependent cell death [26]. GBA (glucocerebrosidase) hydrolyzes glucosylceramide (GlcCer) to ceramide (Cer) and glucose (Glu), and it is a key enzyme that maintains lysosomal lipid metabolism and membrane composition homeostasis [27]. GBA inactivation leads to GlcCer accumulation, causing lysosomal storage disorders and functional impairment and altering lipid homeostasis and membrane permeability [28, 29]. Therefore, GBA activity can directly affect lysosomal function, trigger multiple cell death pathways, and is an important target for tumor therapy combined with nanomedicines, especially IONPs. In this study, we developed an NF-κB-specific GBA that predominantly induces ferroptosis in AML (Supplementary Fig. 1). The therapy comprises two core components: an AML-specific miGBA and low-dose FeNPs. miGBA uses a miR30-shRNA scaffold to sense NF-κB activity in AML cells and to silence the GBA precisely and efficiently. GBA silencing induces lysosomal storage disorders and membrane dysfunction, leading to increased lysosomal permeability. Importantly, miGBA activity is confined to AML cells with over-activated NF-κB, which bypass normal cells. For in vivo safety, we selected low-dose FeNPs as ferroptosis inducers, minimizing off-target NF-κB activation. At this dose, FeNPs do not activate NF-κB in normal cells, and they synergize with miGBA to predominantly induce ferroptosis. The combination shows strong AML cytotoxicity in vitro and in vivo, with a favorable safety profile. In summary, the combination of clinically available FeNPs with NF-κB-responsive miGBA represents a promising, primarily ferroptosis-driven strategy for AML therapy. Materials and methods Materials Feraheme (FeNPs) was obtained from AMAG Pharmaceuticals. TNFɑ, bafilomycin A1 (BA1), ZVAD-FMK (ZVAD), necrostatin-1 s (Nec1s), liproxstatin-1 (Lipro1), and Ac-DMLD-CMK (Ac) were purchased from MedChemExpress. KG1a (human acute myeloid leukemia cells), WEHI-3 (mouse myeloid leukemia cells), SKNO01 (human acute myeloid leukemia cells), and C1498 (mouse acute myeloid leukemia cells) cells were obtained from the American Type Culture Collection (ATCC). THP-1 (human monocyte leukemia cells), U937 (human monocyte leukemia cells), HL60 (human acute promyelocytic leukemia cells), MCF12a (human normal mammary epithelial cells), BEAS-2B (human normal lung epithelial cells), HMEC-1 (human microvascular endothelial cells), NIH-3T3 (mouse embryonic fibroblasts), L929 (mouse fibroblasts), GES-1 (human gastric mucosal epithelial cells), MRC-5 (human embryonic lung cells), AML-12 (mouse normal hepatocytes), HL-1 (mouse cardiomyocytes), C2C12 (mouse myoblast), and HEK-293T (human embryonic kidney cells) cells were procured from the Cell Resource Center at the Shanghai Institute for Biological Sciences, Chinese Academy of Sciences. C1498-GFP and EL9611 (mouse erythroleukemia) cells were acquired from Nanjing Hospital of Chinese Medicine. NB4 (human acute myelocytic leukemia) cells and ML-2 (human acute myelomonocytic leukemia) cells were purchased from the German Collection of Microorganisms and Cell Cultures (DSMZ). All the cellswere verified for authenticity and screened for mycoplasma contamination. Plasmid construction We modified the DMP sequence in the pDMP-miR plasmid [30] via seamless cloning to create a series of plasmids containing 1–10 NF-κB motifs. We subsequently performed double digestion of pAAV-MCS (Stratagene) with Mlu I and Afe I. Then, we inserted the amplified fragment containing NF-κB motifs, a minimal promoter, and the miR30 sequence into the digested plasmid, yielding NF-κB-responsive pNM series plasmids. In addition, we constructed pNM-zsGreen plasmids to screen pNM variants suitable for AML-specific therapy. We designed three sets of human or mouse miRNAs targeting the GBA coding sequence (CDS) via the BLOCK-iT RNAi Designer tool. The specific target sequences and oligonucleotide sequences are listed in Supplementary Tables 1 and 2. We denatured and annealed the synthesized sense and antisense oligonucleotides (Sangon), obtaining double-stranded DNA (dsDNA). Then, we performed Golden Gate assembly with T4 DNA ligase and BsmBI restriction endonuclease and screened clones, yielding pNM6-mihGBA-1, -2, and -3 and pNM6-mimGBA-1, -2, and -3. The plasmid showing the highest GBA silencing efficiency in human or mouse cells was designated pNM6-miGBA (miGBA). As a negative control plasmid, the NT fragment derived from the pcDNATM6.2-GW/EmGFP-miR-Neg plasmid was cloned and inserted into the pNM6 plasmid and designated NT. Plasmid maps and functional sequences are shown in the Supplementary information. All the plasmids were verified via DNA sequencing. Cell culture KG1a, WEHI-3, SKNO01, C1498, THP-1, U937, HL60, NB4, ML-2, and EL9611 cells were maintained in RPMI-1640 (Gibco). MCF12a, NIH-3T3, L929, GES-1, MRC-5, C2C12, and HEK-293 cells were cultured in DMEM (Gibco). HL-1 cells were cultured in Claycomb medium (Gibco) supplemented with 0.1 mM norepinephrine (Sigma) and 2 mM L-glutamine (Sigma). HMEC-1 cells were maintained in MCDB131 medium (Gibco) supplemented with 10 ng/mL human epidermal growth factor (hEGF). AML-12 cells were maintained in DMEM/F12 medium supplemented with 1% ITS liquid medium (Sigma) and 40 ng/mL dexamethasone (Sigma). All cell cultures were supplemented with 10% fetal bovine serum (Gibco), 100 U/mL penicillin (Gibco), and 100 µg/mL streptomycin (Gibco) and maintained in a humidified incubator at 37 °C with 5% CO2. AO&EB staining assay Cells in the logarithmic growth phase (1 × 105) were inoculated into 24-well plates and cultured overnight. The plasmid (500 ng) was subsequently transfected via Lipofectamine 2000 (Thermo), which included the NT, pNM6-mihGBA-1, -2, and -3, and pNM6-mimGBA-1, -2, and -3 plasmids. Depending on the cell type, the cells were transfected with plasmids targeting either mouse or human miGBA at 24 h, 48 h, or 72 h post-transfection. At each specified time point, all the cells were subjected to staining with AO&EB (Sangon). For adherent cells, the medium was removed and replaced with PBS, followed by the addition of 10 µL of AO & EB per milliliter of PBS. The mixture was incubated at room temperature in the dark for 5 min. For the suspended cells, the mixture was initially centrifuged at 1,000 rpm for 5 min, the medium was removed, and the mixture was subsequently added to the adherent cells. Live cells show uniform green fluorescence, and dead cells stain orange. The cells were imaged under a fluorescence microscope to observe live and dead cells, and the data were quantified with ImageJ software. CCK-8 assays Cell viability was assessed and analyzed via the Cell Counting Kit-8 (CCK-8; Beyotime). The cells (5 × 103) were inoculated into 96-well plates and incubated overnight at 37 °C in a 5% CO2 incubator. The plasmids (200 ng/well) were subsequently introduced into the cells via transfection with Lipofectamine 2000. Induction with TNFα (10 ng/mL) was conducted one hour prior to transfection. After overnight transfection, FeNPs (10 µg/mL) were added as needed and incubated for 24 h, 48 h, or 72 h. To evaluate the impact of different inhibitors on cell viability, the cells were transfected overnight and then exposed to the inhibitors for 48 h. The inhibitors utilized included a pyroptosis inhibitor (Ac, 20 µM), an apoptosis inhibitor (ZVAD, 50 µM), a necrosis inhibitor (Nec1s, 10 µM), an autophagy inhibitor (BA-1, 1 nM), and a ferroptosis inhibitor (Lipro-1, 50 nM). To assess the effects of different concentrations of FeNPs, the cells were incubated with media containing 0 to 200 µg/mL FeNPs for 72 h. After incubation, CCK-8 reagent (10 µL/well) was added, the mixture was incubated for 1 h, and the absorbance at 450 nm was measured on a microplate reader. GBA enzyme activity assay The GBA activity assay was performed according to the kit protocol (Abcam, ab273339). The cell pellets (1 × 107) were collected, homogenized in lysis buffer, incubated on ice for 10 min, and centrifuged at 12,000×g at 4 °C for 10 min, after which the resulting supernatant was retained as the sample. Samples and fluorophore-labeled GBA substrate were added to a 96-well plate, mixed thoroughly, and incubated at 37 °C for 30 min, after which the fluorescence was measured (Ex = 360 nm, Em = 445 nm). GPX4 enzyme activity assay The GPX4 enzyme activity assay was performed according to the kit protocol (Sigma, 353919). The cells were pelleted via centrifugation at 1,000×g and 4 °C for 10 min. The cell pellet was subsequently lysed on ice in homogenization buffer (50 mM Tris-HCl, pH 7.5; 5 mM EDTA; 1 mM DTT). The lysate was centrifuged at 10,000×g and 4 °C for 15 min, and the resulting supernatant was retained as the sample. Twenty microliters of the supernatant was added to each well of a 96-well plate. Diluted assay buffer (50 µL), cosubstrate mixture (50 µL), and NADPH (50 µL) were added to each well, and cumene hydroperoxide (20 µL) was added to initiate the reaction. The plate was gently shaken, and the absorbance at 340 nm was recorded every minute via a microplate reader. GlcCer, Cer, and Glu assays Initially, a specified quantity of stable isotope internal standards [Glucose-13C6, d7-Cer (d18:1-d7/18:0), d5-GlcCer (18:1)] was introduced into the cell particles. The Folch extraction method (methanol: chloroform: water = 2:1:0.8) was subsequently used, and the mixture was vortexed for 10 min, followed by centrifugation at 1,6000×g for 10 min at 4 °C to sequentially isolate the organic and aqueous phases. Each phase was divided evenly into two aliquots. The mixture was vortexed for 10 min and then centrifuged at 1,6000×g for 10 min at 4 °C. The organic and aqueous phases were divided, and then each phase was partitioned into two equal aliquots. The organic phase was desiccated under a nitrogen atmosphere and subsequently redissolved in methanol. Cer and GluCer were quantified by employing a reverse-phase C18 column interfaced with a triple quadrupole mass spectrometer (Waters Xevo TQ-XS) in positive ion MRM mode, utilizing d7-Cer and d5-GlcCer as internal standards. The aqueous phase was evaporated under a nitrogen atmosphere and subsequently reconstituted in a mixture of acetonitrile and water (1:1). Glu was quantified via an HILIC column on the same UPLC-TQ platform in MRM mode, with Glu-13C6 serving as the internal standard. The entire procedure was completed with self-added internal standards to correct for recovery and matrix effects. Matrix-matched calibration curves were prepared in the corresponding matrix. Recovery, matrix effects, linearity, recovery-corrected precision, and stability were validated according to Matuszewski to ensure reliable discrimination and reproducible quantification of GlcCer, Cer, and Glu. Fe2+ assay Quantitative analysis was conducted with the Fe2+ detection reagent (Boxbio, AKIC004M). Under acidic conditions, Fe2+ forms a blue complex with tripyridinetriazine. This product exhibited a characteristic absorption peak at 593 nm, enabling the quantitative determination of the Fe2+ concentration through variations in absorbance. The procedure comprises sample pretreatment, including cell lysis and tissue homogenization conducted on ice. After pretreatment, color development reagent was added to initiate the chromogenic reaction. The absorbance at 593 nm was measured via a microplate reader. ELISA ELISA kits were used to quantify the expression of proteins, including p-p65 (Abcam, ab176647) and GBA (Aviva, mouse OKEH03366, human OKCD08164). Reagents were prepared according to the manufacturer’s instructions and were equilibrated to room temperature. Fifty microliters of test sample or control sample was added to each well, and 50 µL of antibody mixture was added. The mixture was incubated at room temperature for one hour, the supernatant was removed, and three washes were performed with wash buffer. A total of 100 µL of TMB substrate was added, and the mixture was incubated in the dark for 15 min. Finally, 100 µL of stop solution was added to terminate the reaction, and the optical density at 450 nm was immediately measured. MDA assay A total of 2 × 106 cells were collected and washed with ice-cold PBS. The cells were resuspended in 300 µL of lysis buffer, and 3 µL of BHT stock solution was added (BHT prevents further peroxidation during processing). Subsequently, homogenization and sonication were performed. After processing, the samples were centrifuged at 13,000×g for 10 min, and the supernatant was collected as the assay sample. The MDA standards and developer mixture were prepared according to the kit manual (Abcam, ab118970). A total of 600 µL of developer mixture was mixed with 200 µL of standard or sample and incubated at 95 °C for 60 min. The samples were then placed on ice for 10 min. Two hundred microliters of the reaction mixture was transferred into the wells of a 96-well plate, and the optical density at 532 nm was immediately measured. ROS assay Lipid peroxidation within cells was identified via a BODIPY®581/591 C11 lipid peroxidation assay kit (Invitrogen). Fluorescence imaging was performed with emission filters at 590 nm and 510 nm, and a 40× objective was used. Images were analyzed via ImageJ software. The fluorescence intensities in the 590 nm and 510 nm channels were quantified, and their ratio was calculated to assess cellular lipid peroxidation. Hydroxyl radicals (•OH) in cells were detected and quantified via a hydroxyl-radical detection kit (Biorab, HR8841). The assay was based on the •OH-specific green fluorescent probe O27. O27 was allowed to freely permeate live cell membranes and was oxidized by intracellular •OH to yield a green fluorescent product. The fluorescence was measured with a microplate reader (Ex = 488 nm, Em = 525 nm). miGBA assay First-strand miRNA cDNA was synthesized via the miRNA 1st Strand cDNA Synthesis Kit (Vazyme, MR101). Genomic DNA was removed with gDNA Wiper Mix prior to reverse transcription. Stem-loop primers (Supplementary Table 3) and reverse transcriptase were used for first-strand synthesis. The reverse-transcription program was 25 °C for 5 min, 50 °C for 15 min, and 85 °C for 5 min. The synthesized cDNA was used for miRNA quantification, and the primers used are listed in Supplementary Table 3. The reaction procedure was conducted as follows: 95 °C for 15 min, followed by 35 cycles of 95 °C for 5 s and 60 °C for 30 s. Transmission electron microscopy (TEM) imaging AML cells and normal cells were transfected overnight with NT or pNM6-miGBA, respectively. The group positive for the substance was administered erastin (10 µM, 8 h). For FeNPs incubation, FeNPs (10 µg/mL) were added after overnight transfection and incubated for 48 h. After drug treatment, the cells were fixed overnight at 4 °C in 0.1 M cacodylate buffer (pH 7.4) containing 2.5% glutaraldehyde. The cells were then fixed at room temperature for 1 h in a solution containing 2% osmium tetroxide, 0.1 M sodium cacodylate and 1.5% potassium ferrocyanide. The cells were dehydrated through a graded ethanol series and embedded in epoxy resin. The embedded samples were cut into ultrathin sections, stained with uranyl acetate and lead citrate, and finally examined and imaged via TEM. LysoSensor staining Lysosomal acidification was monitored via LysoSensor (Invitrogen). After completion of the NT, pNM6-miGBA and FeNPs treatments, LysoSensor (1 µM) was added, and the cells were incubated for 2 h. The dye-containing medium was then replaced with fresh culture medium, the cells were observed via fluorescence microscopy, and the data were quantified via ImageJ software. Lysosomal proton transport activity assay The cells were collected in DMEM containing 2 mg/mL FITC-dextran on ice for 5 min and then transferred to a 37 °C incubator for 30 min. The medium was replaced, and the cells were incubated for an additional 30 min to allow FITC-dextran trafficking to lysosomes. Lysosomes were collected and resuspended in assay buffer (125 mM KCl, 1 mM EDTA, 20 mM HEPES, pH 7.5), equilibrated on ice for 1 h, mixed with 5 µM Con A or DMSO and incubated at 37 °C for 10 min. FITC fluorescence was recorded via a plate reader. Finally, 5 mM Mg-ATP was added, and the initial slope of fluorescence quenching was measured to assess lysosomal proton transport activity. Galectin-3 puncta assay AML Cells (C1498 and KG1a) were fixed in 4% paraformaldehyde for 10–15 min at room temperature, washed three times with PBS, and permeabilized with 0.1% Triton X‑100 for 5–10 min. After blocking in 3% BSA/PBS for 30 min, cells were incubated with anti-Galectin‑3 primary antibody (abcam, ab76245, 1:1000) and fluorescent secondary antibody (abcam, ab150077, 1:500), and nuclei were counterstained with DAPI. Images were acquired on a confocal fluorescence microscope, and quantification were performed in ImageJ software. Virus preparation and titration HEK-293T cells were seeded at 5 × 106 cells per flask, cultured overnight, cotransfected with helper plasmids (pHelper and pAAV-RC) and pAAV plasmids (pNM6-NT or pNM6-miGBA) and incubated for 72 h. Among these, pAAV-RC, which was derived from the pXX2-187 plasmid, is capable of expressing the NQVGSWS peptide [31]. This modified AAV2 vector has been shown to efficiently transduce acute myeloid leukemia (AML) cells while exhibiting negligible transduction of normal hematopoietic stem/progenitor cells and peripheral blood mononuclear cells. The cells and culture media were collected and subjected to three freeze-thaw cycles, followed by the addition of chloroform with shaking. NaCl was then added, and the samples were centrifuged. The supernatant was mixed with PEG8000, fully dissolved, and centrifuged. The resulting pellet was dissolved in PBS, treated with DNase/RNase (1 µg/mL), and further purified via chloroform extraction to obtain purified virus. Viral titers were determined by qPCR using the primers listed in Supplementary Table 4. The obtained serotype capsid-modified AAV2 preparations were designated NT and rNM6-miGBA, respectively. RNA-seq The study included six groups of RNA‑seq samples (blank controls: C1498 and KG1a; FeNPs cotreated with pNM6-miGBA, and pNM6-miGBA transfection alone for 72 h, each for C1498 and KG1a), with three biological replicates per group. Total RNA was extracted with TRIzol. RNA degradation and contamination were assessed by 1% agarose gel electrophoresis, purity was evaluated by OD260/OD280 using a NanoDrop, concentration was measured with a Qubit, and integrity and insert size were determined via an Agilent 2100. After passing quality control, the libraries were sequenced on an Illumina platform (150 bp paired-end), and the clean reads were aligned to the reference genome via HISAT2. GO and KEGG analysis HTSeq was used to quantify the mRNA expression levels of each gene, and differentially expressed genes (DEGs) were identified via the DESeq R package (P < 0.05). DEGs with an FDR < 0.05 and an absolute fold change (|fold change|) ≥ 2 were subjected to further GO and KEGG analyses. GO functional enrichment of DEGs was performed via DAVID, and KEGG pathway mapping was used for pathway annotation. Quantitative PCR (qPCR) Total RNA was extracted with TRIzol reagent and reverse transcribed to cDNA via the HiScript II 1st Strand cDNA Synthesis Kit (Vazyme). Quantitative real-time PCR (qPCR) was then performed with HiScript II Q RT SuperMix for qPCR (Vazyme). The transcript levels were normalized to those of GAPDH (mRNA transcription: RQ = 2−ΔΔCt), and the viral abundance was normalized as RQ = 2−ΔCt. The primer sequences are listed in Supplementary Table 4. All the experiments were performed in triplicate. Treatment of acute myeloid leukemia in mice AML mouse models were established via the tail vein injection of 5 × 106 C1498 cells into male C57BL/6 mice (8 weeks). The mouse experiments were carried out in four cohorts. Cohort 1 comprised 12 groups (n = 10): healthy, PBS, NT, FeNPs, NT+FeNPs (0), miGBA, miGBA+FeNPs (0), miGBA+FeNPs (1), miGBA+FeNPs (2), miGBA+FeNPs (3), miGBA+FeNPs (5), and miGBA+FeNPs (7); the numbers in parentheses indicate the number of days between the administration of rNM6-miGBA and FeNPs. The rNM6-miGBA dose was 1 × 1011/mouse, and the FeNPs dose was 1 mg/kg. The in vivo FeNPs dose was calculated based on mouse blood volume multiplied by the in vitro FeNPs concentration, normalized to mouse body weight, which yielded an in vivo dose of 1 mg/kg. Body weights and survival curves were recorded. Cohort 2 included four groups (n = 6): healthy, NT+FeNPs, miGBA, and miGBA+FeNPs groups. On day 20 after AML model establishment, the mice were euthanized, and major organs (heart, liver, spleen, lung, and kidney) were weighed and processed for H&E staining. In cohort 3, mice (C57BL/6 male, 8 weeks) were injected via the tail vein with 5 × 106 C1498-GFP cells. The enrichment of AML cells within the bone marrow was observed across four groups (n = 6): healthy, NT+FeNPs, miGBA, and miGBA+FeNPs groups. In cohort 4, we used Lipro‑1 (10 mg/kg, i.p.) to perform the ferroptosis rescue assays. Cohort 4 comprised 8 groups (n = 10): Health, PBS, NT+FeNPs, miGBA, miGBA+FeNPs, NT+FeNPs+Lipro1, miGBA+Lipro1, miGBA+FeNPs+Lipro1. The rNM6-miGBA dose was 1 × 1011/mouse, and the FeNPs dose was 1 mg/kg. Biosafety analysis Healthy male C57BL/6 mice (8 weeks) were randomly allocated into six groups (n = 6): the control, NT, FeNPs, NT+FeNPs, miGBA, and miGBA+FeNPs groups. The rNM6-miGBA dose was 1 × 1011 vg/mouse, and the FeNPs dose was 1 mg/kg. The interval between the rNM6-miGBA and FeNPs injections via the tail vein was two days. Body weight, hematologic parameters, liver and kidney function tests, rAAV abundance in major tissues, tissue iron content, and histopathology (H&E) were assessed. Simultaneously, the metabolism of rNM6-miGBA and FeNPs was evaluated in mouse blood (0 min, 1 min, 5 min, 10 min, 30 min, 1 h, 2 h, 4 h, 8 h, 12 h, and 24 h) and in key tissues (12 h, 24 h, and 48 h). Statistical analysis All the data are presented as the means ± standard deviations (SDs). Statistical analyses and graphing were performed via GraphPad Prism 8.0. Data normality was assessed by the Shapiro-Wilk test, and homogeneity of variance between groups was evaluated by Levene’s test. When the data were normally distributed with equal variances, two‑tailed unpaired Student’s t tests were used for significance testing; when these conditions were not met, nonparametric tests were applied. Statistical significance was defined as *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. Results Construction of the AML-specific pNM6-miGBA plasmid In the canonical NF-κB activation pathway, the activation of the p65/p50 heterodimer is crucial, with the phosphorylation of p65 (p-p65) serving a critical function in regulating the transcriptional activity of NF-κB [32, 33]. Therefore, p-p65 activity was first assessed in AML and normal cells. High p-p65 activation was observed in AML cells. Relatively high expression was detected in THP-1, U937, and KG1a cells, and the lowest level was detected in C1498 cells (Fig. 1A). Conversely, p-p65 was nearly undetectable in normal cells (Fig. 1A). NF-κB-responsive pNM plasmids were designed to exploit NF-κB transcriptional activity and its high level of activation in AML cells. Each plasmid was constructed to contain a high-affinity NF-κB motif (NF-κB decoy), a minimal promoter, and downstream effector elements (genes or RNA interference) (Fig. 1A). Although NF-κB remains predominantly quiescent in normal cells, to ensure the in vivo safety of the pNM plasmid (which exhibits no response in normal cells) and maximizes effector expression in AML cells, we constructed a series of pNM variants containing 1–10 NF-κB motifs. Detection was carried out utilizing zsGreen as the reporter gene (Fig. 1B). The data indicated that as the number of 1–10 motifs increased, the zsGreen fluorescence intensity markedly increased in AML cells. However, in normal cells weak fluorescence was observed in MCF12a cells when the number of NF-κB motifs reached seven (Fig. 1B and C). Although pNM7-zsGreen exhibited very weak fluorescence in MCF12a cells but not in HL-1 cells, pNM6 was chosen as the subsequent therapeutic plasmid owing to its superior safety profile in vivo. Compared with normal cells, AML cells have higher p-p65 levels and reporter expression. (A) Levels of p-p65 in ten different AML cells (THP-1, U937, KG1a, HL60, WEHI-3, NB4, SKNO01, ML-2, EL9611, and C1498) and ten different normal cells (MCF12a, BEAS-2B, HMEC-1, NIH-3T3, L929, GES-1, MRC-5, AML-12, HL-1, and C2C12) (n = 3). (B) Representative fluorescence images of pNM-zsGreen containing varying numbers of NF-κB motifs transfected into two AML cells (KG1a and C1498) and two normal cells (MCF12a and HL-1) at 48 h. (C) Quantitative analysis of fluorescence intensity (n = 3) Lysosomes are essential cellular organelles responsible for maintaining metabolic homeostasis and functioning as an intracellular iron pool [25]. GBA mutation or dysfunction directly compromises lysosomal enzyme activity and overall metabolic capacity [27, 34]. Therefore, we incorporated the miR30 system to construct the pNM6-miR30-shGBA plasmid, referred to as pNM6-miGBA (abbreviated as miGBA). The miR30 gene-silencing technology was designed to exploit the miR30 structural scaffold, into which shRNA sequences were inserted, and to use tissue-specific promoters to silence target genes in defined tissues or cell types [35–37]. Direct expression of shGBA was avoided because shRNA transcription is controlled by RNA polymerase III and is not cell-specific [37]. NF-κB-driven expression is mediated by RNA polymerase II, allowing cell-specific expression of the miGBA (Fig. 2A). We integrated miR30-shGBA with NM6 to develop three unique miR30-shGBA variants for human and mouse sources, named mihGBA-1, -2, and -3 and mimGBA-1, -2, and -3, respectively. The results indicated that these plasmids progressively increased expression in AML cells (KG1a, C1498) from 24 h to 72 h (Fig. 2B), effectively silencing GBA mRNA (Supplementary Fig. 2A) and protein levels (Supplementary Fig. 3A and B) while markedly reducing GBA enzyme activity (Supplementary Fig. 3A and B). In contrast, they had little effect on normal cells (MCF12a, HL-1) (Fig. 2C, Supplementary Fig. 2B, Supplementary Fig. 3C and D). Furthermore, upon induction by TNFɑ (a well-known NF-κB activator), normal cells similarly expressed miGBA (Fig. 2C), exhibited suppressed GBA mRNA levels (Supplementary Fig. 2B) and protein expression (Supplementary Fig. 3E and F), and demonstrated a significant reduction in GBA enzyme activity (Supplementary Fig. 3E and F). These findings suggest that the miGBA is specific for NF-κB. We identified the most effective silencing variants, mihGBA-1 and mimGBA-2, which were selected for subsequent studies and collectively designated miGBA (Supplementary Figs. 2 and 3). Time course of miGBA activation in AML and normal cells. (A) Schematic of the pNM6‑miGBA plasmid construct, a plasmid specifically driven by NF-κB (Pol II). (B) Time course of miGBA expression in KG1a and C1498 cells at 24, 48 and 72 h after transfection (n = 3). (C) Expression levels of miGBA at different time points (24–72 h) in MCF12a and HL-1 cells with or without TNFα (10 ng/mL, 1 h) induction (n = 3). (D) Representative AO/EB fluorescence staining of AML cells and normal cells at 72 h after miGBA transfection; additional time points are shown in Supplementary Fig. 4–6 To examine the cytotoxic effects of miGBA on AML cells, we transfected miGBA into both AML cells and normal cells. The results demonstrated that miGBA increases the cytotoxicity against AML cells in a time-dependent manner but has no effect on normal cells (Fig. 2D and Supplementary Figs. 4 and 5). Furthermore, normal cells were also eliminated by miGBA following TNFα induction, demonstrating that miGBA functions as an NF-κB-dependent AML-specific cytotoxic agent (Fig. 2D and Supplementary Fig. 6). Considering that infection, tissue injury or elevated cytokines can produce marked inflammation and thereby increase NF‑κB transcriptional activity, an NF‑κB driven miGBA strategy could potentially damage non-malignant tissues. We therefore recommend following common clinical practice of inflammation control (e.g., with antibiotics, corticosteroids or nonsteroidal anti-inflammatory drugs) during surgery, radiotherapy or immunotherapy. Based on this concept, we treated TNFα-stimulated normal cells with aspirin (ASA, 1 mM). Under these conditions, miGBA did not reduce the viability of normal cells and produced only minimal changes in p-p65 and downstream GBA activity (Supplementary Fig. 7A-D). We also tested whether this treatment strategy affects miGBA activity against AML. ASA did not significantly alter p‑p65 or downstream GBA activity in AML cells (Supplementary Fig. 7E-H). This likely reflected the intrinsic behavior of AML cells, which had high stemness, and could rapidly restore p‑p65 activity within 72 h to support their proliferation (Supplementary Fig. 7E-G). miGBA induces lysosomal storage disorders and ferroptosis in AML To investigate the killing mechanism of miGBA in AML cells, GBA substrates (GlcCer) and downstream metabolites, Cer and Glu, were quantified. The results showed that GBA inhibition causes GlcCer accumulation in AML cells, accompanied by notable decreases in Cer and Glu levels, whereas normal cells remain unaffected (Fig. 3A and B). GlcCer is a vital constituent of lysosomal membranes, and its excessive accumulation was found to cause lysosomal storage disorders. The decrease in Glu was less pronounced than that in Cer (Fig. 3B) because GBA-mediated GlcCer metabolism is not the main pathway responsible for Glu synthesis [38]. Significant upregulation of the lysosomal markers Lamp1 and Lamp2 was observed, suggesting that miGBA induced lysosomal storage disorders in AML cells (Supplementary Fig. 8A). Lysosomal activity was further shown to be reduced by measuring LysoSensor probe fluorescence and lysosomal proton transport activity (Supplementary Fig. 8B and C). No effect on lysosomal function was detected in normal cells (Supplementary Fig. 8). We assessed p-p65 expression in cells transfected with miGBA and demonstrated that miGBA enhances the activation of p-p65 in AML cells (Fig. 3C). This activation was attributed to lysosomal storage disorders, which resulted in increased lysosomal membrane permeability and the release of lysosomal contents that activated NF-κB. p-p65 activation was found to establish a positive feedback loop, which further promoted miGBA expression and culminated in AML cell death due to lysosomal storage disorders (Fig. 3A). Conversely, p-p65, which is minimally activated in normal cells, is insufficient to initiate this loop (Fig. 3C). pNM6‑miGBA selectively induces lysosomal storage disorders in AML cells. (A) Schematic showing the mechanism by which miGBA selectively kills AML cells while bypassing normal cells. (B) Levels of the GBA substrate GlcCer and downstream metabolites Cer and Glu following miGBA treatment (n = 3). (C) Levels of p-p65 in AML cells (KG1a, C1498) and normal cells (MCF12a, HL-1) upon miGBA treatment (n = 3) To establish a direct, threshold-dependent relationship between miGBA-mediated GBA knockdown and the phenotype of lysosomal storage disorders, we transfected AML cells (KG1a and C1498, 5 × 104 cells) with a titration series of miGBA plasmid (0, 1.5625, 3.125, 6.25, 12.5, 25, 50, 100, 200, 400 ng). For each condition we measured GBA activity (Supplementary Fig. 9A), downstream GlcCer content (Supplementary Fig. 9B), lysosomal integrity assessed by the proportion of Galectin-3 positive cells using the galectin-3 puncta assay (Supplementary Fig. 9C), and cell viability (Supplementary Fig. 9D). GBA activity fell progressively with increasing miGBA plasmid dose (Supplementary Fig. 9A), while GlcCer levels and the proportion of Galectin-3 positive cells rose and cell viability declined (Supplementary Fig. 7B-D). At 12.5 ng miGBA plasmid, there was only modest GBA activity loss, 12.6% in KG1a and 13.4% in C1498 (Supplementary Fig. 9A), without significant changes in GlcCer (Supplementary Fig. 9B), Galectin-3 positivity (Supplementary Fig. 9C), or viability (Supplementary Fig. 9D). In contrast, 25 ng miGBA plasmid caused a significant decrease in GBA activity (KG1a: 26.1%; C1498: 25.4%; Supplementary Fig. 9A), accompanied by a 1.75-fold and 1.8-fold increase in GlcCer respectively (Supplementary Fig. 9B), and by a marked increase in Galectin-3 positive cells and cell death (Supplementary Fig. 9C and D). Despite different baseline GlcCer levels between the cell lines, a ~ 25% reduction in GBA activity raised GlcCer to 0.0027 nM in KG1a and to 0.0011 nM in C1498 (Supplementary Fig. 9B); these absolute increases were sufficient to induce lysosomal storage disorders in the AML cells (Supplementary Fig. 9C and D). Higher doses produced progressively stronger lysosomal storage disorders phenotype (Supplementary Fig. 9). Overall, these dose-response data demonstrated a positive correlation between the extent of GBA knockdown and the phenotype of lysosomal storage disorders and revealed a threshold effect, with roughly 25% loss of GBA activity being sufficient to trigger GlcCer accumulation, loss of lysosomal integrity, and reduced cell viability (Supplementary Fig. 9). To explore the precise mechanisms of miGBA-induced AML cell death, various cell death inhibitors have been used to rescue AML cells. Inhibitors of pyroptosis (Ac), apoptosis (ZVAD), necrosis (Nec1s), autophagy (BA-1), and ferroptosis (Lipro-1) each partially rescued miGBA-induced cell death, with Lipro-1 producing the strongest effect (Fig. 4A). This effect was attributed to lysosomes serving as the cellular iron pool and to the release of iron ions caused by lysosomal storage disorders, which then primarily triggered ferroptosis. To confirm this hypothesis, the ferroptosis marker malondialdehyde (MDA) was first measured. miGBA, similar to the ferroptosis-positive compound erastin, significantly upregulated MDA in AML cells, and this effect was reversed by Lipro-1 (Fig. 4B). TEM observations confirmed that miGBA induced lysosomal enlargement and storage disorders in AML cells (Fig. 4C). The mitochondria were observed to have shrunk, with increased membrane density and thickened cristae (Fig. 4C). Notably, the effects of miGBA on mitochondria were similar to those of erastin, whereas erastin did not induce lysosomal storage disorders (Fig. 4C). These findings suggest that miGBA-induced disruption of lysosomal storage is crucial for triggering ferroptosis in AML cells. LipROS, Fe2+, and GPX4 activities were further measured in AML cells. miGBA induction significantly increased lipROS and Fe2+ levels, and the activity of the critical ferroptosis antioxidant enzyme GPX4 was suppressed (Fig. 4D-F). Notably, miGBA did not affect mitochondrial or lysosomal morphology, nor did it modify lipROS, Fe2+, or GPX4 activity in normal cells. Conversely, erastin demonstrated notable ferroptotic activity without specificity for AML (Fig. 4D-F). These findings conclusively establish the specificity of the miGBA for AML and its safety for therapeutic application. Notably, compared with erastin, miGBA considerably reduced the activity of MDA, lipROS, Fe2+, and GPX4, especially that of Fe2+ (Fig. 4E). These findings suggested that miGBA primarily induced ferroptosis, which can be further enhanced by exogenous Fe2+ supplementation. pNM6‑miGBA induces AML cell death primarily via ferroptosis. (A) Rescue experiments. KG1a and C1498 cells were transfected with miGBA overnight and then incubated for 48 h with cell death inhibitors (n = 3). Inhibitors: pyroptosis inhibitor (Ac, 20 µM), apoptosis inhibitor (ZVAD, 50 µM), necroptosis inhibitor (Nec1s, 10 µM), autophagy inhibitor (BA-1, 1 nM), and ferroptosis inhibitor (Lipro-1, 50 nM). (B) MDA levels as a measure of lipROS (n = 3). (C) TEM images of AML cells and normal cells treated with NT, miGBA, or erastin (10 µM); red arrows indicate mitochondria, and white dashed circles indicated lysosomes. (D) Quantification of lipROS by C11-BODIPY (n = 3). (E) Intracellular Fe2+ levels (n = 3). (F) GPX4 enzyme activity (n = 3) miGBA synergistically induces ferroptosis in AML cells with ferumoxytol In fact, excessive intracellular Fe2+ can undergo the Fenton reaction to generate large amounts of toxic hydroxyl radicals (•OH), causing lipROS and triggering ferroptosis [39]. Moreover, •OH can further promote p-p65 activation [40, 41], which is expected to upregulate miGBA expression. Therefore, to enhance the therapeutic effect of miGBA, we introduced the FDA-approved intravenous iron supplement ferumoxytol (Feraheme, FeNPs). As nanomedicine, FeNPs have an iron oxide core (Fe3O4/γ-Fe2O3) containing abundant Fe2+ and a carboxydextran outer coating, with particle diameters of approximately 17–31 nm [22]. Previous studies have shown that FeNPs localize primarily to lysosomes and can induce iron overload and the Fenton reaction to selectively kill AML cells with low FPN expression [22, 23]. The biological functions of FeNPs are highly compatible with the AML-killing mechanism of miGBA, motivating the investigation of the ability of FeNPs and miGBA to cooperate with ferroptosis induction. The data demonstrated that FeNPs concentrations ranging from 0 to 60 µg/mL display outstanding biosafety for both AML cells and normal cells (Supplementary Fig. 10A). However, FeNPs concentrations exceeding 30 µg/mL triggered the activation of p-p65 in cells (Supplementary Fig. 10B). Crucially, an ultralow dose of FeNPs (10 µg/mL) synergized with miGBA to kill the vast majority of AML cells while bypassing normal cells (Supplementary Fig. 10C). Indeed, the concentration of FeNPs is a crucial factor in determining the efficacy of miGBA. Initially, excessive FeNPs generate abundant •OH that can damage normal cells. Second, the released •OH must not increase p-p65 levels in normal cells because p-p65 activation triggers miGBA expression, which is catastrophic for normal cells. To balance safety and efficacy, we therefore selected an ultralow FeNPs dose of 10 µg/mL for subsequent experiments. We hypothesize that after miGBA induction, lysosomal storage disorders in AML cells result in the accumulation of FeNPs within lysosomes. This induces the continuous generation of •OH, promoting p-p65 activation and establishing a positive feedback loop, which ultimately results in AML cell death through lysosomal damage and ferroptosis (Fig. 5A). To validate this hypothesis, we used TEM to directly observe AML and normal cells after miGBA and FeNPs treatments. TEM images revealed that AML cells treated with miGBA alone or with miGBA plus FeNPs exhibited significantly larger lysosome area and an increased number of lysosomes compared with normal cells (Fig. 5B, Supplementary Fig. 11). Furthermore, lysosome area in the miGBA+FeNPs group was significantly larger than in the miGBA group, consistent with •OH released from FeNPs further reinforcing the NF-κB positive-feedback loop (Fig. 5B and Supplementary Fig. 11A). Notably, abundant FeNPs accumulation was observed inside lysosomes only in the miGBA+FeNPs group, whereas FeNPs alone had no effect on any cells (Fig. 5B), suggesting the biosafety of low-dose FeNPs. We performed ferroptosis characterization and found that the miGBA+FeNPs group presented the greatest increases in lipROS and •OH, far exceeding those of the miGBA alone group (Fig. 5C-E). Single low-dose FeNPs did not affect intracellular lipROS or •OH because cells can metabolize low quantities of FeNPs (Fig. 5C-E, Supplementary Fig. 12A). Furthermore, intracellular Fe2+ measurements indicated that the miGBA+FeNPs group presented significantly increased Fe2+ levels in AML cells, surpassing the limited Fe2+ induction observed with miGBA alone (Fig. 5F). Similarly, the miGBA+FeNPs markedly suppressed GPX4 enzyme activity in AML cells but had little effect on normal cells (Fig. 5G). Ultimately, we assessed the synergistic cytotoxic impact of miGBA and FeNPs on AML cells. The results showed that miGBA+FeNPs eliminated the majority of AML cells, greatly improving single-agent efficacy while maintaining good biosafety for normal cells (Supplementary Fig. 12B). This enhanced killing was attributable to sustained p-p65 activation via the treatment-induced positive feedback loop (Supplementary Fig. 12C). In summary, the combination of miGBA and FeNPs induces devastating, synergistic AML cell death through lysosomal storage disorders and ferroptosis. FeNPs enhance miGBA-induced ferroptosis in AML cells. (A) Schematic illustration of the mechanism by which miGBA combined with FeNPs kills AML cells while bypassing normal cells. (B) TEM images of AML cells (KG1a, C1498) and normal cells (MCF12a, HL-1) after treatment with NT+FeNPs, miGBA+FeNPs, miGBA, or FeNPs (10 µg/mL); white dashed circles indicated lysosomes. (C) Merged fluorescence images of C11-BODIPY staining in AML cells and normal cells after 72 h of NT+FeNPs, miGBA+FeNPs, or miGBA treatment. Representative nonoxidized and oxidized channel images are shown in Supplementary Fig. 8A. (D) Quantification of lipROS levels (n = 3). (E) Quantification of the •OH fluorescent signal (n = 3). (F) Intracellular Fe2+ levels (n = 3). (G) GPX4 enzyme activity assay (n = 3) To further demonstrate that the effects of miGBA alone or the synergistic action of miGBA with FeNPs are attributable to ferroptosis driven by an NF-κB positive‑feedback loop, we performed in vitro rescue experiments using two ferroptosis inhibitors, NAC and Lipro1, which are also regarded as free‑radical scavengers. The results showed that in AML cells (C1498 and KG1a), intracellular •OH generated by miGBA treatment alone or by the combination of miGBA and FeNPs could be effectively eliminated by NAC and Lipro-1 (Supplementary Fig. 13A). Blocking •OH disrupted the NF‑κB positive‑feedback loop; consequently, p‑p65 could no longer be sustainably activated (Supplementary Fig. 13B), leading to insufficient expression of the effector component miGBA (Supplementary Fig. 13C) and ineffective suppression of GBA activity (Supplementary Fig. 13D). Ultimately, cell viability was rescued and ferroptosis was reversed (Supplementary Fig. 13E). Collectively, these findings provide strong evidence for the existence of the NF‑κB positive‑feedback loop and support a ferroptosis mechanism driven by this loop. Additionally, the classical NF‑κB inhibitor BAY 11-7082 (BAY, 10 µM) was used to test whether direct blockade of NF‑κB breaks the loop and reduces miGBA expression and cytotoxicity. The results showed that treatment with BAY decreased miGBA expression and rescued AML cell viability (Supplementary Fig. 14). In summary, both clearance of •OH by the ferroptosis inhibitors NAC and Lipro1 and inhibition of NF‑κB by BAY suppressed miGBA expression and restored AML cell viability, supporting the existence of the proposed NF‑κB positive‑feedback loop. To investigate the molecular mechanism of the synergy between miGBA and FeNPs, we performed RNA-seq on C1498 and KG1a cells. Venn and volcano plots revealed that GBA was significantly knocked down in both cellsafter miGBA or miGBA+FeNPs treatment (Fig. 6A and B). This result confirms prior experimental data (Supplementary Fig. 2A and Supplementary Figs. 3 A and B) and demonstrated effective GBA mRNA silencing by miGBA. Notably, the ferroptosis marker ferritin light chain (FTL) was significantly upregulated in all the treatment groups (Fig. 6A and B), indicating that both miGBA alone and the combination of miGBA-FeNPs substantially affected iron metabolism in AML. KEGG pathway analysis demonstrated that the ferroptosis pathway was the most substantially enriched pathway in KG1a cells after combined miGBA and FeNPs treatment (Fig. 6C). In contrast, single-agent miGBA treatment preferentially enriched neurodegenerative disease pathways, including Parkinson’s disease and Alzheimer’s disease pathways, in KG1a (Fig. 6D), which is consistent with the physiological roles of the GBA [28]. In C1498 cells, both miGBA alone and the combined treatment of miGBA with FeNPs markedly promoted ferroptosis and fatty acid metabolism pathways, suggesting the impairment of iron and lipid metabolic processes in AML cells (Supplementary Fig. 15). Analysis of the top 20 KEGG pathways and GO terms revealed that miGBA in combination with FeNPs promoted the disassembly of cellular components in AML cells, whereas miGBA alone elicited a response primarily to oxidative stress (Supplementary Fig. 16). Overall, the RNA-seq data robustly validated miGBA-mediated GBA mRNA interference and supported the conclusion that both miGBA treatment and combined miGBA+FeNPs treatment kill AML cells primarily via ferroptosis. Finally, we performed western blot assays for AML cell lines (KG1a, C1498) and normal cell lines (MCF12a and HL-1) treated with PBS, NT, FeNPs, NT+FeNPs, miGBA, and miGBA+FeNPs. The results showed that miGBA and miGBA+FeNPs treatments markedly downregulated GBA and significantly suppressed anti-ferroptotic proteins GPX4, SLC7A11, FTH1, and FSP1, indicating that miGBA and miGBA+FeNPs induce ferroptosis in AML cells (Supplementary Fig. 17A and B). Furthermore, due to the synergistic effect between miGBA and FeNPs, miGBA+FeNPs treatment leads to increased p-p65 levels and further inhibition of GBA, thereby exhibiting a more pronounced ability to induce ferroptosis (Supplementary Fig. 17A and B). Notably, neither miGBA nor miGBA+FeNPs treatment exerted any detectable effect on normal cells (Supplementary Fig. 17C and D). RNA-seq analysis upon miGBA and FeNPs treatment. (A) Venn diagram showing DEGs in KG1a and C1498 cells upon miGBA transfection (KG1a-miGBA, C1498-miGBA) and upon subsequent coincubation with FeNPs (10 µg/mL) for 48 h. (B) Volcano plot of DEGs. (C) KEGG pathway enrichment results for DEGs identified in KG1a-miGBA upon coincubation with FeNPs (10 µg/mL). (D) KEGG pathway enrichment results for DEGs in the KG1a-miGBA group In vivo anti-AML effects of the combined administration of miGBA and FeNPs Inspired by the potent AML‑cell killing observed with combined miGBA and FeNPs treatment in vitro, we packaged miGBA within the adeno-associated virus rNM6-miGBA (serotype 2) for in vivo therapy of AML-bearing mice. Importantly, the AAV2 vector used in our study is not wild-type but a capsid-modified variant. This modified platform, exemplified by the NQVGSWS peptide-expressing clone, is a validated approach for targeted gene delivery to AML cells [31]. AAV and FeNPs were chosen for combination delivery, primarily because both have well-established clinical safety profiles, which may facilitate translation. Treatment was delivered through the intravenous injection of miGBA and FeNPs at concentrations of 1 × 1011 miGBA and 1 mg/kg FeNPs, corresponding to a significantly reduced FeNPs dosage relative to earlier studies [22, 42–44], thereby substantially improving in vivo safety (Fig. 7A). Because miGBA requires 72 h in cell experiments to achieve significant inhibition of GBA, while FeNPs typically have a blood halftime of approximately 15 h [45], there is a marked difference in the onset time between the two agents. To establish the most effective dosing interval and ascertain the optimal timing for FeNPs administration, we organized groups with dosing intervals of 0, 1, 2, 3, 5, and 7 days. Day 0 indicates simultaneous administration of the combined miGBA and FeNPs. The data demonstrated that both the miGBA+FeNPs group and the miGBA group presented restored mouse body weight and significantly prolonged survival (Fig. 7B and C). The results indicated that a 2-day or 3-day interval between miGBA and FeNPs administration produced optimal efficacy, allowing 60% of the mice to survive beyond 150 days (Fig. 7C). The body weight and survival curve data conclusively demonstrated that the combined administration of miGBA and FeNPs has notable anti-AML efficacy and maintains favorable safety profiles in vivo (Fig. 7B and C). In vivo anti-AML efficacy of FeNPs and pNM6‑miGBA. (A) Schematic of the AML-bearing mouse treatment regimen. (B) Body weight changes over the treatment course (n = 10). (C) Kaplan–Meier survival curves (n = 10). (D) Weights of the spleen, liver, and lung at the endpoint (n = 6). (E) Representative H&E staining of major organs (heart, liver, spleen, lung, and kidney). Black dashed boxes indicated areas of AML cell infiltration in the liver and lungs. Black arrows pointed to regions in the spleen where AML cell infiltration has blurred the boundary between red and white pulp, indicating impaired splenic function. (F) Fluorescence images of C1498-GFP in BM. (G) Percentage of C1498-GFP-positive cells in the BM (n = 6) We measured miGBA expression, GBA activity, and intracellular iron content in bone marrow leukemic cells from AML mice at different time points (6 h, 12 h, 1, 2, 3, 4, and 7 days) following administration of rNM6-miGBA or FeNPs. After rNM6-miGBA treatment, miGBA expression peaked on day 3 and then gradually declined, stabilizing thereafter (Supplementary Fig. 18A). Suppression of GBA activity was slightly delayed, with maximal inhibition observed on days 3–4 (Supplementary Fig. 18B), likely reflecting the time required for miGBA-mediated degradation of GBA mRNA, subsequent reduction of GBA protein levels, and ultimate inhibition of enzymatic activity. In contrast, following FeNPs administration, iron accumulation in bone marrow leukemic cells peaked on day 1 and then rapidly decreased (Supplementary Fig. 18C). Collectively, these data indicated that rNM6-miGBA required approximately 3–4 days to achieve maximal GBA inhibition, whereas FeNPs reached peak intracellular iron accumulation about 1-day post-dose (Supplementary Fig. 18D). Accordingly, a 2-3-day interval between rNM6-miGBA and FeNPs was expected to maximize therapeutic efficacy (Supplementary Fig. 18D), consistent with our dosing-interval optimization experiments showing the best outcomes when FeNPs were administered 2–3 days after rNM6-miGBA (Fig. 7C). Considering the therapeutic efficacy, we chose a 2-day interval between miGBA and FeNPs administration for subsequent animal experiments. To investigate the in vivo therapeutic mechanism of the synergistic administration of miGBA and FeNPs, we analyzed the infiltration of AML cells in key mouse tissues. The results indicated that combined miGBA and FeNPs treatment markedly decreased the AML cell burden in the spleen, liver, and lung tissues of the mice (Fig. 7D). The weights of these organs were close to those of the healthy group, the effect of miGBA alone was intermediate, and FeNPs alone were ineffective (Fig. 7D). These findings were further supported by the H&E staining results (Fig. 7E). AML cells primarily infiltrate the bone marrow but can also involve extramedullary tissues, particularly the liver and spleen. Analysis of H&E-stained sections of major organs showed that the miGBA and miGBA+FeNPs treatment groups exhibited a significant reduction in AML cell infiltration in the liver, spleen, and lung (Fig. 7E). Moreover, AML infiltration blurred the boundary between red and white pulp in the spleen, indicating impaired splenic function; this abnormality was alleviated by treatment with miGBA+FeNPs (Fig. 7E). The immunohistochemistry experiments were conducted to assess AML cell infiltration in major mouse organs, using CD11b and CD33 as AML-positive markers [46]. The results showed that AML cell infiltration in the liver, spleen, and lung was markedly reduced after miGBA+FeNPs treatment, with miGBA alone showing a moderate effect (Supplementary Figs. 19 and 20). These findings indicated that miGBA+FeNPs significantly suppressed AML cell infiltration into tissues. Overall, the miGBA+FeNPs group was more effective than miGBA alone at suppressing AML infiltration in major organs. Finally, the minimal residual disease (MRD) was assessed in the BM of AML mice and found that MRD (CD34+CD45+CD19+) was negative (< 0.01%) after miGBA+FeNPs treatment (Supplementary Fig. 21). Together, these data demonstrated that miGBA+FeNPs eliminated MRD in treated AML mice. In addition to using AML mouse models for anti-AML experiments, human biopsy samples were also employed to evaluate the combined therapeutic efficacy of miGBA and FeNPs. Specifically, AML cells (CD34+CD38−CD123+) and hematopoietic stem cells (HSCs) (CD34+CD90+CD133+) were first enriched and isolated from human biopsy samples. These cells were then treated with NT, NT+FeNPs, miGBA, or miGBA+FeNPs. Subsequently, Electrophoretic Mobility Shift Assays (EMSA) were performed to measure the transcriptional activity of NF-κB, thereby confirming the existence of an NF-κB positive feedback loop. The results showed that nuclear proteins from the miGBA+FeNPs group exhibited the strongest binding to the NF-κB probe, followed by miGBA, whereas NT and NT+FeNPs had minimal effects (Supplementary Fig. 22A). Importantly, none of the treatments altered NF-κB activity in Hematopoietic Stem Cells (HSCs) (Supplementary Fig. 22B). These EMSA results further confirm that the synergistic interaction between miGBA and FeNPs enhanced NF-κB activation. Cytotoxicity tests comparing the treatment with FDA-approved anti-AML drugs were also conducted on human biopsy samples. The results showed that miGBA+FeNPs markedly reduced the number of AML cells, demonstrating superior therapeutic efficacy compared to CPX-351 or miGBA alone, whereas NT and NT+FeNPs exhibited minimal effects (Supplementary Fig. 23A). Importantly, CPX-351 showed cytotoxicity toward HSCs, whereas miGBA+FeNPs did not (Supplementary Fig. 23B). Furthermore, we examined the effects of miGBA+FeNPs and CPX-351 on protein expression in primary human AML cells and found that miGBA+FeNPs were more effective than CPX-351 in reversing AML cell stemness and inducing ferroptosis (Supplementary Fig. 24). Collectively, the EMSA, cytotoxicity, and western blot data supported both the efficacy of miGBA+FeNPs against human tumor samples and their favorable safety profile. In addition, stable C1498-GFP cells were introduced to track AML cell infiltration in mouse bone marrow (BM). The results indicated that the number of AML cells in the bone marrow was markedly reduced by combined miGBA and FeNPs administration, that the effect of miGBA alone was intermediate, and that of FeNPs alone was ineffective (Fig. 7F and G). We constructed and packaged the rNM6-miGBA-mCherry vector, in which miGBA and mCherry are driven by the same promoter. Therefore, expression of the downstream mCherry indicated that the upstream miGBA was also transcriptionally activated, and mCherry fluorescence can be used to assess both rNM6-miGBA infection efficiency and miGBA expression. We established an AML mouse model using C1498-GFP stable cells and administered PBS or rNM6-miGBA-mCherry via tail-vein injection. One week later, C1498-GFP cells were enriched from the bone marrow, and the proportion of mCherry-positive cells was quantified. The percentage of mCherry-positive cells was 82.2%, indicating an AAV transduction rate of 82.2% in the bone marrow (Supplementary Fig. 25A and B). Furthermore, we examined primary human AML cells and quantified viral genome copy numbers by qPCR, which yielded approximately 84.5% positivity. These results fully demonstrated that combined miGBA and FeNPs administration was associated with potent in vivo anti-AML effects and favorable biosafety. To investigate the in vivo mechanism of combined miGBA and FeNPs therapy, bone marrow AML cells from each group were enriched, and GBA expression was measured. Both the miGBA+FeNPs group and the miGBA group induced robust miGBA expression in AML cells, GBA mRNA and protein levels were downregulated, and GBA enzyme activity was inhibited (Fig. 8A). These data comprehensively confirm the in vivo inhibitory effect of miGBA on the GBA. Simultaneously, we quantified GlcCer and its metabolites, Cer and Glu. Consistent with the cellular findings, GlcCer exhibited significant accumulation, whereas Cer and Glu levels markedly declined in both the miGBA+FeNPs and miGBA groups (Fig. 8B), suggesting lysosomal storage disorders. Lysosomal function was further assessed, and the lysosomal markers Lamp1 and Lamp2 were markedly increased in the miGBA+FeNPs group. Lysorescent fluorescence intensity and the proton transport rate were strongly suppressed, intermediate effects were observed with miGBA, and FeNPs alone were ineffective (Fig. 8C). Lysosomal storage disorders and functional impairment in AML cells in vivo were indicated by these results. Markers of ferroptosis, including MDA, •OH, and Fe2+ contents and GPX4 enzyme activity, were measured. In vivo, AML cells in the miGBA+FeNPs group presented significantly increased MDA, •OH, and Fe2+ contents, and GPX4 enzyme activity was markedly suppressed; intermediate effects were observed with miGBA, and FeNPs alone were ineffective (Fig. 8D). These findings suggest that ferroptosis is the core mechanism underlying the in vivo anti-AML activity of the combination of miGBA and FeNPs. To further elucidate the in vivo anti-AML mechanism of miGBA alone or in combination with FeNPs, we performed ferroptosis rescue experiments using Lipro1. The results showed that Lipro‑1 significantly attenuated the therapeutic efficacy of miGBA alone or in combination with FeNPs, consistent with ferroptosis being a key contributor to the antitumor effect (Supplementary Fig. 26). In vivo, the corresponding western blot assays were performed. The AML cells were collected from BM, and the data were consistent with the in vitro findings, demonstrating that miGBA+FeNPs conferred stronger p65 activation, more potent inhibition of GBA, and a more robust capacity to induce ferroptosis (Supplementary Fig. 27). In addition, we examined AML stemness markers (CD34+CD38−) [47]. The results showed that miGBA+FeNPs dramatically suppressed CD34 expression and upregulated CD38, suggesting that miGBA+FeNPs markedly attenuated the stemness of residual AML cells and reduced the likelihood of relapse (Supplementary Fig. 27). In summary, the in vivo anti-AML effects of either agent alone were greatly enhanced by combined miGBA and FeNPs administration, lysosomal storage disorder was induced, and ferroptosis was triggered in AML cells in vivo by combination therapy. Mechanism of the in vivo anti-AML efficacy of FeNPs and pNM6‑miGBA. (A) Quantification of miGBA levels, GBA mRNA levels, GBA protein levels, and GBA enzyme activity in C1498-eGFP cells (n = 3). (B) Levels of GlcCer, Cer, and Glu and quantification of p-p65 (n = 3). (C) Lamp1 and Lamp2 mRNA expression, LysoSensor fluorescence intensity, and lysosomal proton transport activity measurements (n = 3). (D) MDA content, •OH fluorescence signal, Fe2+ levels, and GPX4 enzyme activity (n = 3). (E) Blood metabolism and biodistribution of FeNPs (n = 3). (F) Blood metabolism and biodistribution of rNM6-miGBA (n = 3) In vivo safety of the combined administration of miGBA and FeNPs The in vivo safety profile of miGBA and FeNPs is a critical factor of concern for AML therapy (Supplementary Fig. 28A). The in vivo safety of miGBA, FeNPs, and their combination was evaluated, and body weight data confirmed that they had no effect on mouse weight (Supplementary Fig. 28B). Furthermore, we evaluated viral abundance and iron levels in key tissues (heart, liver, spleen, lung, kidney, and bone marrow) after the administration of miGBA and FeNPs. The results demonstrated that both miGBA and FeNPs are dispersed across all the tissues, with the greatest accumulation observed in the liver and spleen, respectively (Supplementary Fig. 28C and D). This distribution is considered an important reason for the clearance of leukemic cell infiltrates in the liver, spleen, and lungs. H&E staining, hematologic parameters, or liver and kidney function tests were performed, and little toxicity to major organs or blood was detected (Supplementary Fig. 28E, and 29). Finally, the pharmacokinetics in the blood and tissues were measured. The blood half-life of the FeNPs was 8.18 h (Fig. 8E), and the blood half-life of the miGBA was 11.83 h (Fig. 8F). In tissues, both miGBA and FeNPs reached peak accumulation in the liver at 24 h (Fig. 8E and F). At 48 h, approximately 15% of the FeNPs and 18% of the miGBA accumulated in the bone marrow. Together, these findings illustrate the favorable in vivo safety profile of miGBA, FeNPs, and their combination therapy. In addition, during the progression of AML, complications such as inflammation and infections may occur. Therefore, in clinical practice, chemoradiotherapy or targeted therapy is often accompanied by adjunctive treatment with anti-inflammatory drugs (e.g., ASA) or antibiotics (e.g., amoxicillin) [48]. Notably, these complications may elevate NF-κB levels in certain normal tissues, posing a potential threat to the in vivo safety of combination therapy with miGBA and FeNPs. Given that survival curves are among the most critical indicators of both safety and therapeutic efficacy for rapidly progressing and refractory tumors like AML, we evaluated the survival outcomes of miGBA and FeNPs combination therapy in the context of AML with concurrent complications. To model inflammation and infection in AML mice, we administered (LPS, 1 mg/kg, i.p.) and Staphylococcus aureus (ATCC-29213, 106 CFU/mouse, i.v.), respectively (Supplementary Figs. 30 and 31). To better assess the therapeutic efficacy, we used doxorubicin (ADM, 3 mg/kg·d, i.p., 10 d) as a positive control. Throughout the treatment period for inflammatory and infectious complications, we provided supportive care with either ASA (50 mg/kg·d, p.o., 3 d) or amoxicillin (1.6 mg/kg·d, p.o., 7 d) to mimic clinical adjunctive strategies (Supplementary Fig. 30A, and 31 A). The results demonstrated that the survival outcomes of mice treated with FeNPs and miGBA were significantly superior to those in the positive control group, indicating that the combination therapy remained effective even in the presence of complications and did not lead to mortality due to elevated NF-κB in normal tissues (Supplementary Fig. 30B, and 31B). In contrast, ADM exhibited mild tissue toxicity (Supplementary Fig. 30C, and 31 C). Notably, after ASA or amoxicillin intervention, p-p65 levels in normal tissues were only slightly elevated (Supplementary Fig. 30D, and 31D), resulting in minimal miGBA expression and little suppression of GBA activity, thereby substantially reducing potential therapeutic risks (Supplementary Fig. 30E and F, and 31E and F). Therefore, the combination therapy of miGBA and FeNPs remains feasible in AML patients presenting with complications such as inflammation and bacterial infection. Discussions We developed a precision therapy for AML based on aberrant NF-κB activation by combining low-dose FeNPs with an NF-κB-responsive miGBA (Fig. 9). This combination induces ferroptosis to potently kill AML cells while maintaining a favorable safety profile. Through comprehensive in vitro, in vivo, and RNA-seq data, we elucidated the molecular mechanisms, efficacy, and safety of the combined therapy, providing a new direction for treating refractory AML. A novel ferroptosis-based therapy in which GBA expression is controlled by an NF-κB-responsive plasmid was developed. miGBA was engineered via a miR-30-based shRNA cassette to be highly responsive to NF-κB in AML cells while remaining quiescent in normal cells. NF-κB-dependent expression of miGBA selectively lowers the GBA level, causing GlcCer accumulation, depletion of Cer and Glu, and lysosomal storage disorders. Lysosomal disruption releases labile iron, which, together with low FPN in AML, favors feroptosis. FDA-approved FeNPs were coadministered to enhance efficacy, producing •OH in lysosomes that further activated NF-κB and created a positive feedback loop to amplify GBA knockdown and ferroptosis. Careful dose optimization of FeNPs is required to avoid NF-κB activation in normal cells while producing sufficient •OH to drive the feedback loop and primarily induce ferroptosis in AML cells Hyperactivation of NF-κB underlies AML chemoresistance and offers a targetable weakness The NF-κB pathway is hyperactivated in AML, promoting tumor cell survival and chemoresistance and posing a major barrier to curing AML [5, 7–9]. Leveraging this AML-specific feature, we used a miR30-shRNA platform to design and screen an NF-κB-responsive GBA-silencing plasmid, pNM6-miGBA, which efficiently suppresses GBA expression in AML cells while bypassing normal cells. This strategy leverages hyperactivated NF-κB in AML cells to increase therapeutic specificity and prevent potential harm to normal cells and tissues caused by broad-spectrum NF-κB inhibitors. However, infection, tissue injury, and other inflammatory states can also elevate systemic NF‑κB activity. To minimize potential treatment risks and adverse effects, subsequent preclinical and clinical study designs should incorporate inflammatory status into inclusion/exclusion criteria and safety monitoring, and should prespecify biomarker thresholds (e.g., CRP, TNFα, IL‑6, and p-p65) and assessment timepoints [49, 50]. Studies should also explore strategies such as adjusting dose and optimizing timing of administration to further reduce off‑target toxicity. It is important to acknowledge that every therapeutic approach has limitations. Given the high lethality of AML, careful risk assessment and management, strict screening, and targeted inflammation control can help select appropriate patient populations, thereby maximizing clinical benefit and extending patient survival. A pertinent example is arsenic trioxide [51]. Despite being a highly toxic substance capable of causing multi-organ failure and death, through strict measures such as dose control, avoiding granulocyte colony-stimulating factor, and using dexamethasone to prevent differentiation syndrome, it has become one of the most successful drugs for treating acute promyelocytic leukemia (APL). Dismissing its clinical application based on potential toxicity and management complexity would deprive numerous patients of a life-saving opportunity. In fact, no drug is perfect; correctly assessing and managing risks in a high-stakes context to strive for longer survival is paramount for patients. In the face of highly lethal diseases, survival must be the primary goal. Targeting iron metabolism in AML to induce ferroptosis Ferroptosis shows great therapeutic potential against drug-resistant tumor cells [11]. In this study, we exploited the iron metabolism traits of AML cells, such as low FPN expression [23], which impairs iron homeostasis, to potentiate combined therapy. Furthermore, on the basis of NF-κB transcriptional activity, we found that FeNPs promote •OH generation via the Fenton reaction and help sustain aberrant NF-κB activation. This establishes a positive feedback loop that facilitates miGBA-mediated suppression of the GBA and induces lysosomal stroge disorders and functional impairment. As lysosomes serve as the cellular iron pool [25], their dysfunction results in the release of large amounts of labile iron, ultimately achieving a more efficient antitumor effect through molecular synergy. Low-dose FeNPs induce ferroptosis and markedly impair AML viability FeNPs are intravenous iron supplements used to treat iron deficiency anemia in patients with chronic kidney disease [23]. In this study, we tested the safety and therapeutic effect of low-dose FeNPs in AML treatment. Low-dose FeNPs (10 µg/mL) were highly selective and cooperated with miGBA to kill AML cells while bypassing normal cells. Combined treatment, assessed by ferroptosis markers, lysosome function tests, and RNA-seq, revealed that therapy altered iron metabolism and lysosome function by inducing ferroptosis, leading to effective AML cell death. In animals, a therapeutic dose of 1 mg/kg FeNPs caused no obvious toxicity. This dose is much lower than those used in previous studies and has strong antitumor effects with fewer side effects [22, 42–44]. Because rNM6-miGBA and FeNPs act at different times in vivo, we optimized the dosing interval and found that administering both drugs every 2–3 days markedly prolonged the survival of AML-bearing mice, with 60% surviving beyond 150 days. Importantly, both AAV-based vectors and FeNPs are already used clinically, underscoring the translational potential of this regimen. In addition, although the relatively low FeNP dose used in this study represents a potential advantage for translational development, dose selection in humans will require careful evaluation to balance efficacy and safety. Notably, human tumors may display substantial inter-patient and intra-tumoral heterogeneity in basal iron availability, antioxidant buffering capacity (e.g., the GSH/GPX4 axis), hypoxia and nutrient gradients, and stromal interactions [52–54]; these factors could limit ROS accumulation and thereby weaken ferroptosis induction. In our mouse AML model, despite the low dosing regimen, we nonetheless observed clear intratumoral oxidative stress and ferroptosis-associated changes accompanied by therapeutic benefit. However, translation to humans will still require (i) dose-finding studies in larger animals informed by pharmacokinetic/pharmacodynamic (PK/PD) modeling, with quantitative characterization of FeNP biodistribution and clearance kinetics in the bone marrow and major organs, together with systematic measurement of pharmacodynamic endpoints (ROS, lipid ROS, and ferroptosis markers); and (ii) pre-treatment profiling of oxidative status and p-p65 expression in AML bone marrow samples to help identify patients who are more likely to respond. Targeting lysosomal storage disorders is a promising strategy for eradicating tumor cells This study specifically downregulated GBA in AML cells via NF-κB modulation, effectively inducing lysosomal storage disorders and causing AML cell death. Targeting and eliminating AML cells by intervening in lysosomal storage disorders offers a novel approach for lysosome-based cancer therapy. Lysosomal storage disorders can exert direct cytotoxic effects through ferroptosis. Additionally, they can synergize with other therapeutic mechanisms, such as autophagy, apoptosis, and necrosis, to further enhance treatment efficacy. In this study, leveraging the synergy between lysosomal storage disorders and ferroptosis was not limited to AML treatment and may provide innovative solutions for a variety of treatment-refractory malignancies. In addition, improving the in vivo delivery efficiency of miGBA will be an important focus of future research to ensure precise delivery to target cells during treatment and further reduce off-target toxicity. Challenges and strategies for AAV-based rNM6-miGBA therapy AAV is a safe gene-delivery platform and has been approved for clinical gene therapy in humans, owing to several advantages such as low immunogenicity, lack of pathogenicity, non-integration into the genome, and the ability to confer long-term stable expression [55]. At the doses used in this study, rNM6-miGBA showed no apparent toxicity and no loss of therapeutic efficacy in the AML mouse model. Nevertheless, the potential clinical application of rNM6-miGBA still faces challenges. Pre-existing or neutralizing antibodies against AAV can impede AAV administration and re-administration, representing a common obstacle for AAV-based therapies [56]. However, this issue may be addressed by emerging strategies, including elimination of pre-existing anti-AAV antibodies using endopeptidases (e.g., imlifidase (IdeS) [57] or IgG-degrading enzyme (IdeZ) [58], transient suppression of endogenous Myd88 via CRISPR [59], and the use of engineered AAV vectors incorporating TLR9-inhibitory sequences [60]. In addition, alternative serotypes or capsid engineering can be employed to evade neutralizing antibodies and enhance bone marrow tropism, such as AAV9 and engineered AAV9 capsid variants [61], as well as our capsid-modified AAV2 [31]. Conclusion In this study, we engineered and screened an NF-κB-responsive pNM6-miGBA plasmid (miGBA) that is robustly activated in AML cells but remains quiescent in normal cells. miGBA selectively knocked down GBA mRNA, protein, and enzyme activity and induced lysosomal storage disorders and ferroptosis in AML cells while bypassing normal cells. We potentiated miGBA-mediated ferroptosis with FDA-approved FeNPs, which sustain NF-κB activation through the Fenton reaction, establishing a positive feedback loop that amplifies AML cytotoxicity. Systematic dose-response studies revealed that low-dose FeNPs combined with miGBA eradicated most AML cells without activating NF-κB in normal cells. In vitro, in vivo, and RNA-seq results confirmed ferroptosis as the principal killing mechanism. To accommodate the distinct in vivo onset times of miGBA and FeNPs, we optimized the dosing schedule every 2–3 days, which resulted in 60% survival of treated AML mice in the past 150 days. Notably, the FeNPs dose (1 mg/kg) was markedly lower than the doses reported in prior studies. Safety assessments revealed little histopathological, hematological, or biochemical toxicity, supporting the potential for further preclinical development. Overall, this NF-κB-driven, GBA-targeted strategy primarily induces ferroptosis and offers a potent, selective, and safe therapeutic approach for AML that warrants futher clinical evaluation. Supplementary Information Supplementary Material 1 Acknowledgements Not applicable. Abbreviations Ferumoxytol Acute myeloid leukemia pNM6-miGBA Leukemia stem cells phosphorylated p65 lipid peroxidation Iron oxide nanoparticles Ferroportin Glucocerebrosidase Glucosylceramide Ceramide Glucose Bafilomycin A1 ZVAD-FMK Necrostatin-1s Liproxstatin-1 Ac-DMLD-CMK Coding sequence Hydroxyl radicals Ferritin light chain Bone marrow Author contributions Tao Luo contributed to writing-original drafts, visualization, and secured funding acquisition; Shijie Zhang and Meng Li assisted in writing-review & editing; Kang Ding and Feng Gu led conceptualization and provided supervision. All the authors have read and agreed to the published version of the manuscript. Funding This work was supported by the National Natural Science Foundation of China (82400186), the Natural Science Foundation of Jiangsu Province (BK20240269), the Jiangsu Province Young Scientific and Technological Talents Promotion Program (JSYJ-2025-728), and the Feihong Program of Nanjing Hospital of Chinese Medicine (FHJH202402). Data availability All data supporting the findings of this study are available within the paper and its Supplementary Information. Declarations Ethics approval and consent to participate This study complied with all the relevant national regulations and institutional policies, was in accordance with the Declaration of Helsinki and approved by the Ethics Committee of Nanjing Hospital of Chinese Medicine Affiliated to Nanjing University of Chinese Medicine (Approved Number: KY2024088). All animal experiments were performed in accordance with the protocols sanctioned by the Animal Ethical and Welfare Committee (AEWC) at the Shenzhen Glorybay Biomedical Laboratory Animal Center (Approved Number: RW-IACUC-24-0132). Consent for publication All authors have approved the manuscript and agree with the submission. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Tao Luo, Email: luotao@hnu.edu.cn. Kang Ding, Email: fsyy00237@njucm.edu.cn. Feng Gu, Email: 15852482679@126.com. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement All data supporting the findings of this study are available within the paper and its Supplementary Information.