Actinomycin D Targets NPM1c-Primed Mitochondria to Restore PML-Driven Senescence in AML Therapy

The nucleolar chaperone nucleophosmin 1 (NPM1) exerts a wide range of activities, from ribosome biogenesis to control of MYC or TP53 signaling. In 50% of acute myeloid leukemia (AML) with normal karyotypes, highly clustered monoallelic NPM1 mutations cause frameshifts that delete tryptophan residues involved in nucleolar targeting and create de novo nuclear export signals (1). While occurrence of cytoplasmic NPM1 (NPM1c) oncoprotein is preceded by mutations sustaining clonal hematopoiesis (2), its continuous presence is required for maintaining the leukemic state of established AMLs, where NPM1c disables TP53 signaling and sustains high expression of HOX genes (3).

Promyelocytic leukemia (PML) nuclear bodies (NB) are reactive oxygen species (ROS)–responsive domains (4, 5) that exert prosenescent and tumor-suppressive functions by enhancing TP53 and Rb activities. Many protein degradation pathways associated with tumor progression target PML, presumably to impede its prosenescent function. PML NBs have directly been implicated in the therapeutic response of acute promyelocytic leukemia (APL) where the PML/RARA fusion oncoprotein disrupts PML bodies, but therapy-induced PML/RARA degradation restores them to drive cure (6–8). PML has also been implicated in tumor robustness through the control of mitochondrial fitness and stem cell metabolism (9–11).

Mitochondrial dysfunctions can induce the integrated stress response and ROS production, important features of normal or AML stem cells (12). Conversely, mitochondrial fitness and high oxidative phosphorylation (OXPHOS) metabolism have been implicated in resistance to BCL2 inhibitors or conventional chemotherapy (13–18). Accordingly, inhibitors of mitochondrial respiration were proposed to constitute a novel type of cancer drugs (19). In addition to their well-studied effects on nuclear DNA replication or transcription, anticancer antibiotics may exert ill-understood mitochondrial toxicities, implicated in long-term doxorubicin-induced cardiomyocyte loss (20).

Here, we demonstrate that NPM1c blunts PML NB biogenesis and weakens mitochondrial fitness. Actinomycin D (ActD), a drug with established clinical activity in therapy-resistant NPM1c-AML (21–23), induces acute mitochondrial stress, ROS production, restoration of PML NBs, and senescence, contributing to its therapeutic efficacy.

Primary leukemic cells or cell lines from patients with AML expressing NPM1c exhibit some defects in NB formation, resembling APL micro-speckles (refs. 24, 25; Supplementary Fig. S1A and S1B). As AMLs harbor multiple gene mutations that may contribute to NB alteration, we explored nonleukemic primary hematopoietic stem cells (HSC) from a Flp-inducible humanized Npm1c (variant A) knock-in mouse model (Npm1^frt-cA/+, R26^FlpoER; ref. 26). Four weeks after tamoxifen exposure, these HSCs exhibited a dramatic reduction of NB numbers (Fig. 1A). Similarly, hematopoietic progenitors that were differentiated from a mouse embryonic stem cell (mESC) NPM1c knock-in model (Supplementary Fig. S1C) displayed significantly fewer PML NBs.

A previous study suggested that overexpressed PML and NPM1c may interact (27). Transient expression of NPM1c (but not NPM1) disrupted PML NBs in mouse embryonic fibroblasts (MEF) stably expressing GFP-PML-III (Fig. 1B). Cysteine 288 in the de novo C-terminal sequence of NPM1c was implicated in nucleolar export and oxidative stress control (28). Mutation of this residue in NPM1c abolished its ability to oppose NB formation (Fig. 1B). Conversely, an exquisitely redox-sensitive cysteine residue of PML (C389; ref. 29) was required for NPM1c to target NBs. Immunoprecipitation experiments demonstrated that ectopically expressed NPM1c interacts with PML, a process that requires NPM1c C288 and PML C389, but not a PML cysteine residue implicated in arsenic binding (ref. 4; Fig. 1C and D). We determined whether a disulfide bond might govern interaction of NPM1c with PML. For this, we transiently transfected PML and NPM1 variants and purified His-PML proteins under denaturing conditions, followed by NPM1 Western blot analyses, in the absence or presence of the protein-reducing agent TCEP. High molecular weight complexes reactive with both PML and NPM1c antibodies were detected only when NPM1c C288 and PML C389 were both present and TCEP was omitted (Fig. 1E; Supplementary Fig. S1D). Thus, two reactive cysteines in NPM1c and PML interact, yielding impairment of PML NB assembly. However, while NPM1c binding is required for PML NB disruption, cytoplasmic localization of NPM1c did not require PML, as observed in NPM1c knock-in Pml^−/− mESCs (Supplementary Fig. S1E).

To explore functional consequences of NPM1c–PML interactions, we generated an isogenic system by stably expressing GFP-NPM1c, GFP-NPM1c^C288S, or GFP-NPM1 fusions in OCI-AML2 leukemic cells (hereafter named AML2) in which NPM1 gene is wild-type (WT). Expression of NPM1c blunted NB formation, whereas NPM1c^C288S failed to do so (Fig. 1F). Functionally, expression of NPM1c, but not NPM1c^C288S, increased clonogenic activity in methylcellulose cultures (Fig. 1G) and activated E2F and MYC target genes (Fig. 1H; Supplementary Table S1), possibly reflecting decreased basal level of TP53 and ARF (Fig. 1I). NPM1c expression also diminished PML protein levels (but not PML mRNA; Supplementary Fig. S1F). Low PML protein expression was confirmed in primary AML patient samples (Fig. 1J) and could independently amplify NPM1c-driven defects of PML NB formation. Finally, NPM1c reduced expression of WT NPM1 (Fig. 1I), which, in NPM1c AML blasts, may further aggravate NPM1 haploinsufficiency (30).

Because NPM1c impairs NB biogenesis and PML NBs control mitochondrial fitness (9, 10), we assessed the impact of NPM1c expression on mitochondrial status. In tamoxifen-treated NPM1c knock-in mice (26), we found an increase in mitochondrion number (but not volume) in phenotypic HSCs, but not in Lin-negative cells (Fig. 2A; Supplementary Fig. S2A). Similarly, in NPM1c-expressing isogenic AML2 cells, the number of mitochondria increased, while the branching pattern decreased (Fig. 2B) and cristae were lost, as determined by transmission electron microscopy (Fig. 2C). Transcriptomic and proteomic analyses revealed dysregulation of mitochondria-related pathways (Supplementary Fig. S2B). These mitochondrial defects were associated with enhanced production of ROS, mitochondrial superoxides, and membrane potential, with no alteration of mitochondrial mass (Fig. 2D; Supplementary Fig. S2C). Mitochondrial impairment was also substantiated by leakage of mitochondrial DNA (mtDNA) into the cytoplasm in NPM1c-expressing AML2 or mESCs (Fig. 2E; Supplementary Fig. S2D) and reduced electron transport chain (ETC) proteins, notably complex II, reflecting transcriptional downregulation of SDH genes (Supplementary Fig. S2E–S2G).

Release of mtDNA into the cytoplasm was associated with the ignition of intracellular stress pathways, particularly cyclic GMP-AMP (cGAMP) synthase (cGAS) activation, and cGAMP production, resulting in the downstream transcriptional activation of IFN and NFκB pathways (Fig. 2F–H; refs. 31, 32). Other stress pathways, such as oxidative stress and unfolded protein response (UPR), were also activated (Fig. 2H; Supplementary Fig. S2B; Supplementary Tables S2 and S3), explaining the decrease in global protein synthesis (Fig. 2I). Western blot analyses validated activation of these stress pathways in AML2-NPM1c cells, including the upregulation of key TP53-repressed metabolic enzymes (Supplementary Fig. S2H). As expected, these mitochondrial alterations decreased the ATP/ADP ratio, reflecting metabolic stress (Fig. 2J). Functionally, activation of these stress pathways could favor AML cell fitness (12). Indeed, inhibition of cGAS activity by G140 selectively suppressed growth of AML2-NPM1c cells (Supplementary Fig. S2I).

Mechanistically, two key regulators of mitochondrial protein expression, PPARγ coactivator 1-alpha (PGC1A) and transcription factor A mitochondrial (TFAM), showed decreased protein abundance in AML2-NPM1c (Fig. 2K) and OCI-AML3 cells (hereafter referred as AML3, which are NPM1c mutant) when compared with AML2 cells (Supplementary Fig. S2J). TFAM reduction may reflect transcriptional downregulation of PGC1A by IFN signaling (33). Importantly, stable NPM1c expression in AML2 cells drove PGC1α hyperacetylation resulting in its functional inactivation, a known consequence of PML NB downregulation (Fig. 2L; refs. 9, 10). In principle, NPM1c could alter mitochondrial functions independently from its ability to interfere with PML NB formation. Exploring mitochondrial status in NPM1c-knockins Pml^+/+ or Pml^−/− mESCs revealed that mitochondrial morphology was so profoundly altered in Pml^−/− mESCs that it precluded further detection of NPM1c-driven abnormalities (Supplementary Fig. S2K).

Recent clinical studies have shown that the rDNA transcriptional inhibitor ActD can induce complete remissions and even cure some patients with refractory/relapsed NPM1c-AML as a single agent (21–23). Transcriptional inhibition of rDNA genes by ActD induces nucleolar segregation and activates TP53 through ribosomal checkpoint activation upon MDM2–L5/L11 complex formation (34). When used at low doses (1–10 nmol/L), ActD triggers a transient cell-cycle arrest in the absence of TP53 stabilization, DNA damage, or apoptosis (35). We found that 5 nmol/L ActD (a concentration in the range of those found in patients; ref. 23) triggered TP53 activation in AML3, but not in AML2 cells (Supplementary Fig. S3A). Nonetheless, we observed only a small increase in the L11–HDM2 interaction in patient in vivo (see below), suggesting that ribosomal checkpoint activation may not solely account for TP53 activation (Supplementary Fig. S3B). Unexpectedly, we found that ActD treatment rapidly fragmented the mitochondrial network in AML3 or NPM1c-mESCs, with no or minor effects in AML2 or mESCs (Fig. 3A; Supplementary Fig. S3C and S3D). Moreover, ActD initiated rapid production of ROS and induction of a superoxide transcriptional response, loss of mitochondrial membrane potential, and loss of mtDNA through its leakage to the cytoplasm, all preferentially in AML3 and/or AML2-NPM1c cells (Fig. 3B–F), with little immediate effect on mitochondrial DNA expression (Supplementary Fig. S3E).

ActD-triggered mitochondrial toxicity drove stress responses, as indicated by increased cGAMP production, phosphorylation of eIF2α (Fig. 3G and H), transcriptional stress signatures (UPR, ROS signaling), activation of innate immunity (IFN, TNFA, and TGFB; ref. 36), and apoptosis/senescence (TP53; Supplementary Fig. S3F). While responses were often shared between the two isogenic cell lines, IFN and TP53 activation were particularly pronounced in AML2-NPM1c cells. We found a remarkable similarity between the basal expression of genes induced by NPM1c and those activated by ActD in AML2 cells (Supplementary Fig. S3G), likely reflecting their shared ability to impair mitochondrial function. Consequently, ActD-triggered induction of stress pathways may be functionally more significant in NPM1c-expressing cells where basal activation levels are already high. Finally, ActD exacerbated preexisting NPM1c-driven metabolic stress, as revealed by decreased ATP levels and increased AMP-activated protein kinase (AMPK) phosphorylation (Fig. 3I and J). Collectively, these observations suggest that ActD targets mitochondria, particularly those primed by NPM1c expression.

We then explored the cellular impacts of mitochondrial poisoning by ActD. Unexpectedly, ActD rapidly restored PML NBs in NPM1c-transfected MEFs, AML2-NPM1c or AML3 cells (Fig. 4A; Supplementary Fig. S4A). PML NB restoration was also observed in primary NPM1c-AML blasts treated ex vivo with ActD (Fig. 4B). In transiently transfected cells, NB restoration by ActD was associated with (and likely caused by) disruption of NPM1c-PML complexes (Fig. 4C). Similarly, in AML3 cells, ActD (as TCEP) restored the normal size of high molecular weight NPM1 or PML species (Supplementary Fig. S4B). Apart from disrupting NPM1c–PML adducts, ActD-induced ROS may also directly enforce PML NB biogenesis (4, 5). Accordingly, NB restoration in AML2-NPM1c cells was blocked by the ROS scavengers N-acetyl cysteine (NAC) and glutathione (Supplementary Fig. S4C).

PML NBs are tightly linked to senescence induction. In AML3 cells, ActD-driven PML NB reformation was associated to multiple signs of senescence, including TP53 activation, increase in p21 or Serpine-1 expression, activation of senescence-associated β-galactosidase (SA-β-Gal) activity, and loss of clonogenic potential (Fig. 4D–F; Supplementary Fig. S4D). Importantly, these features were abolished or attenuated by PML or TP53 deletion. Moreover, blocking PML NB reformation with antioxidants blunted ActD-driven TP53 and P21 activation, as well as loss of clonogenic activity (Fig. 4G and H). Collectively, these experiments establish a central role of a ROS/PML/TP53 cascade in driving senescence and ActD response ex vivo.

To substantiate the mitochondrial origin of ActD-induced ROS in driving TP53 activation, we cultured AML cells with low-dose antimycin A (Fig. 4I) to promote progressive mitochondrial clearance through mitophagy (37). Mitochondrial depletion abolished early ActD-triggered TP53 activation (Fig. 4I), suggestive for a key role of mitochondrial ROS in PML-dependent TP53 stabilization. Conversely, we used Thenoyl-trifluoroacetone (TTFA), a well-characterized inhibitor of mitochondrial complex II and major inducer of ROS (Supplementary Fig. S4E). In AML3 cells, TTFA decreased clonogenic activity in a PML-dependent manner (Supplementary Fig. S4F). In stable GFP-PML-III MEFs transiently expressing NPM1c, TTFA restored NB formation (Supplementary Fig. S4G). Collectively, these experiments establish the key role of mitochondrial ROS in ActD-driven activation of the PML NB-activated senescence checkpoint ex vivo.

We first assessed ActD response in primary cells. We found no significant ability of low-dose ActD to promote TP53 stabilization in normal bone marrow progenitors in vivo or ex vivo, or in human or mouse primary fibroblasts (Supplementary Fig. S5A; ref. 35). In immunodeficient mice xenografted with primary human NPM1c-AML blasts, a 4-day course of ActD therapy significantly stabilized human TP53, but not mouse Trp53, highlighting tumor-specific targeting and implying that NPM1c sensitizes cells to ActD (Fig. 5A, top). ActD therapy inhibited leukemic growth and induced blast differentiation in that model (Fig. 5A, bottom). Similarly, in a murine NPM1c-driven AML model (38), ActD treatment again induced leukemia-selective Trp53 activation, AML regression, and blast differentiation (Fig. 5B). Finally, in AML3 xenografts, PML was required for TP53 activation and antileukemic effects of ActD (Fig. 5C). In contrast, ActD failed to cause regression of MLL/ENL–driven murine AMLs after 7 days (Supplementary Fig. S5B). Collectively, these in vivo observations suggest that ActD favors PML-dependent TP53 activation and regression of NPM1c-AMLs, but (at least initially) spares normal cells.

We then explored a NPM1c-AML patient who exhibited a 10⁴-fold blast decrease 3 weeks after a 5-day course of single-agent ActD treatment, using the previously reported therapeutic scheme (refs. 21–23; Supplementary Fig. S5C). In leukemic blasts sequentially sampled from the peripheral blood of this patient during ActD treatment, the reformation of PML NBs was complete by 12 hours (Fig. 5D). Such ActD-induced NB reformation was accompanied by TP53 stabilization and target gene activation, including P21 (Fig. 5E and F). However, we failed to detect a significant increase in the L11/HDM2 interaction (Supplementary Fig. S3B) or γH2AX foci before 48 hours (Supplementary Fig. S5D, arrow), arguing against potent early activation of ribosomal and DNA-damaging checkpoints in vivo. Thus, PML NB reformation is an immediate response of AML cells to ActD therapy in vivo.

Pathway analyses of transcriptomes from AML-rich peripheral blood revealed activation of acute stress responses (expression of HSP1A, FOS, EGR1) and activation of PML NBs as early as 6 hours following initiation of therapy (Fig. 5G and H). We also observed immediate shutoff of NFκB targets, including extinction of IL8 expression, a distinct feature of mitochondrial dysfunction–driven senescence (ref. 37; Fig. 5H). We detected cGAS aggregation 6 hours after ActD treatment (Supplementary Fig. S5D). Despite daily ActD injections, these transcriptional changes were transient, pointing to the existence of adaptive control mechanisms (Fig. 5G). Nevertheless, ActD treatment triggered features of senescence (PML NBs, TP53 and P21 up, E2F down; Fig. 5E–H). Terminal deoxynucleotidyl transferase–mediated dUTP nick end labeling (TUNEL)–positive apoptotic cells were also detected 48 hours posttreatment (data not shown), indicating the cooccurrence of apoptosis.

A consolidation course of ActD treatment given when the patient had no residual leukemic blasts was not accompanied by major TP53 target activation or modulation of stress responses in normal blood cells (Fig. 5E; Supplementary Fig. S5E). Collectively, our findings support the idea that ActD-driven acute mitochondrial stress initiates PML NB–activated senescence/apoptosis in NPM1c AML cells in vivo.

In AML and myeloma cells, low basal complex II activity predicts venetoclax sensitivity (15, 39). Having demonstrated NPM1c-driven complex II impairment (Supplementary Fig. S2E–S2G) and ActD-mediated mitochondrial toxicity (Fig. 3), we hypothesized that ActD might enhance the established anti-AML clinical activity of venetoclax (40, 41). Venetoclax induced mitochondrial fragmentation, reduction of mitochondrial membrane potential, production of ROS, and leakage of mtDNA to the cytoplasm of AML3 cells (Fig. 6A–E). Critically, all of these features were strongly potentiated by cotreatment with ActD. In AML2-NPM1c (but not AML2-NPM1), venetoclax and ActD strongly synergized to activate cGAS activity (Fig. 6F) and PML NB reformation (Supplementary Fig. S6A). A brief pretreatment with both drugs abolished clonogenic growth of NPM1c-expressing cells, only in the presence of PML (Fig. 6G). Accordingly, in AML3 xenograft models, venetoclax and ActD strongly synergized to extend survival, an effect that also requires PML (Fig. 6H).

We then explored the ActD–venetoclax interaction in three other NPM1c-driven AML models in vivo. First, in murine NPM1c-driven AMLs (38), the combination treatment dramatically decreased leukemic burden and enhanced blast differentiation (Fig. 6I). Second, in a double conditional knock-in of NPM1c plus IDH1^R132H (ref. 42; to be fully described elsewhere), only the combined treatment eliminated AML cells from the bone marrow and yielded a significant survival advantage (Fig. 6J). Finally, in xenografted primary AML blasts, ActD and venetoclax (but neither agent alone) cleared AML cells from the bone marrow (Fig. 6K). Collectively, the ActD–venetoclax combination synergizes for mitochondrial targeting, yields PML-dependent growth arrest, and clears leukemia in several distinct in vivo AML models.

NPM1c expression alters PML NB formation in multiple cell types, including normal HSCs in vivo. NPM1c is an abundant protein that may impede NB biogenesis through direct PML binding and sequestration, degradation, or indirect control of PML oxidation, which drives NB formation (4, 43). PML is critical for mitochondrial fitness (9–11, 44). NPM1c-induced mitochondrial defects reflect, at least in part, NB disruption, because NPM1c expression in Pml^−/− mESCs did not further aggravate the prominent mitochondrial morphologic alterations resulting from PML loss (Supplementary Fig. S2K). Functionally, NPM1c-mediated NB impairment and mitochondrial defects, which drive the integrated stress response, may all contribute to the transformation of DNMT3A- or IDH1/2-immortalized stem cells into full-blown leukemia (2, 12, 26), through cell-autonomous mechanisms (including ROS signaling and impaired senescence) and/or inflammation-mediated remodeling of the microenvironment (45, 46).

Targeting mitochondrial function by antibiotics was proposed as a therapeutic option for cancer (19). Low-dose ActD can induce reversible cell-cycle arrest, without DNA damage or apoptosis (35) and accordingly was well tolerated by nonleukemic primary cells ex vivo or in vivo (Fig. 5). In contrast, ActD triggers rapid, but transient, alterations in NPM1c-primed mitochondria (Fig. 3A; Supplementary Fig. S3D). ActD also induces mtDNA leakage to the cytoplasm (Fig. 3E) possibly caused by ActD intercalation within the mitochondrial genome. Such mtDNA release, followed by activation of cGAS, and its downstream IFN pathway could activate senescence, as well as antileukemic immune responses (31, 36, 47, 48). Besides cGAS activation, ActD induces ROS, which are required for ActD-induced senescence ex vivo through PML NB reformation and TP53 activation (Figs. 4 and 6L; refs. 4, 5, 43). The actual origin of these ROS remains poorly characterized. Upon mitochondrial stress, complex II, at the junction between the TCA cycle and the ETC, which is highly expressed in Lin-ScaI+cKit+ (LSK) progenitors (49), regulated by PML (50) and by NPM1c (Supplementary Fig. S2F and S2G), could play a role in ROS production, as the complex II inhibitor TTFA mimics ActD in NPM1c-expressing cells ex vivo (Supplementary Fig. S4E–S4G). Our ex vivo observations suggest that a significant part of ActD activity is mediated through mitochondrial ROS, PML NB reformation, and TP53 activation. In the ActD-treated patient with AML, we also found some evidence for mitochondria dysfunction–induced senescence (37). Later, ActD intercalation into nuclear DNA, inhibition of RNA Pol I activity, and activation of the ribosomal checkpoint (34) could also contribute to ActD therapeutic activity. NPM1c-driven PML NB disruption and their ActD-induced reformation (Figs. 5D and 6K) evoke a striking, but unexpected, similarity with APL treatment by retinoic acid and arsenic (6). PML NBs may thus emerge as key hubs inactivated by oncogenes (NPM1c, PML/RARA), but reactivated by antitumor drugs (retinoic acid, arsenic, ActD, or IFNα; ref. 51). Actually, in an APL mouse model, an ETC inhibitor drove prolonged remissions (52), possibly reflecting mitochondrial ROS-induced PML NB restoration. This ROS/PML/TP53 cascade may also promote ActD responses in non-NPM1c-AMLs harboring dysfunctional mitochondria. Doxorubicin similarly fragments mitochondria of NPM1c-expressing cells to drive ROS-induced TP53 activation (Supplementary Fig. S6B–S6D), suggesting that dual targeting of nucleus and mitochondria may be shared by other antitumor antibiotics.

High mitochondrial ETC or OXPHOS activities were repeatedly linked to venetoclax or chemotherapy resistance in AML (13–15, 17) or other cancers (18). Conversely, inhibition of mitochondrial translation sensitizes AML cells to venetoclax (16). Thus, the NPM1c-associated defects in mitochondrial function unraveled here likely underlie the favorable prognosis of NPM1c-AMLs. Finally, ActD and venetoclax exert synergistic effects to clear NPM1c-AMLs in multiple in vivo models (Fig. 6). Clinically supporting our observations, ActD was very recently proposed to circumvent venetoclax resistance in salvage therapy of multi-relapsed patients with AML (53). Thus, mechanism-based prospective trials associating ActD and venetoclax could be envisioned in patients with AML.

Human AML cell lines were maintained in minimum essential medium-alpha plus GlutaMAX (Thermo Fisher Scientific; catalog no. 32561094) supplemented with 20% FBS in the presence of 100 U/mL of penicillin, 100 μg/mL of streptomycin, and were incubated at 37°C with 5% CO₂ (see key resource table for all details). To establish isogenic NPM1c-expressing AML2 cells, cells were transduced with a lentivirus expressing FLAG-GFP–tagged NPM1, NPM1c, or NPM1c-C288S and selected for GFP expression by FACS. WI38 cells were maintained in Eagle MEM (ATCC; catalog no. 30–2003) supplemented with 10% FBS in the presence of 100 U/mL of penicillin and 100 μg/mL of streptomycin. Primary MEFs were obtained from E18 mouse embryos and maintained in DMEM plus GlutaMAX (Thermo Fisher Scientific; catalog no. 41965062) supplemented with 10% FBS, 1× nonessential amino acids solution, 100 U/mL of penicillin, and 100 μg/mL of streptomycin. Normal CD34-positive cells were purified by using Dynabeads CD34 Positive Isolation Kit (Thermo Fisher Scientific; catalog no. 11301D). mESC line E14 was maintained on gelatin-coated dishes and cultured in DMEM supplemented with 10% FBS, 1,000 U/mL leukemia inhibitory factor, 50 U/mL penicillin, 50 μg/mL streptomycin, 2 mmol/L GlutaMAX, 0.1 mmol/L nonessential amino acids, 1 mmol/L sodium pyruvate, and 0.1 mmol/L 2-mercaptoethanol. The medium was changed every other day. All cell lines were repeatedly shown to be Mycoplasma negative by PCR testing (Eurofins). OCI-AML2 and OCI-AML3 were obtained from DSMZ, regularly authenticated by NPM1c expression, and genetically modified in house. Details on PML and TP53 inactivation by CRISPR in AML3 cells or NPM1c CRISPR in mESCs can be found in the Supplementary Materials and Methods. mESCs were obtained from P. Thérizol (INSERM U944), primary and SV40-immortalized MEFs were derived in house. AML2-NPM1c and AML2-NPM1 cells were regularly flow-sorted to ensure continuous GFP expression. ActD, NAC, glutathione monoethyl ester (GSH-MEE), antimycin A, and doxorubicin were purchased from Sigma (catalog nos. 01815, A7250, G1404, A8674, D1515). Human cGAS inhibitor G140 was purchased from InvivoGen (catalog no. inh-g140).

The study was conducted according to the Declaration of Helsinki and approved by our institutional ethics committee. Primary AML cells were obtained upon written informed consent from the patient during the off-label study. Human primary AML cells were isolated by density gradient centrifugation of either peripheral blood or bone marrow. AML was defined as NPM1-mutated on the basis of targeted sequencing analysis.

For the NPM1c leukemia mouse model, C57BI6-CD45.1 mice were irradiated with 4.5 Gy and then engrafted with 10⁶ CD45.2-NPM1c leukemic cells, a kind gift from P.G. Pelicci of the Instituto Europeo di Oncologia (38). In the experiment shown in Fig. 5B, mice were intraperitoneally injected with ActD (60 μg/kg/day) 19 days after engraftment and sacrificed 24 hours later after an additional ActD injection 1 hour before euthanasia. CD45.2⁺ bone marrow cells were sorted by FACS (ARIA2, Becton Dickinson) and lysed directly by Laemmli buffer containing 20 mmol/L NEM and 10% 2-mercaptoethanol, followed by Western blot analysis. For the spleen weight or cytometric analyses, mice were intraperitoneally injected daily with ActD (60 μg/kg/day) for 7 or 10 days after engraftment. CD45.2 and differentiation markers Mac-1/Gr-1 triple-positive cells were examined by FACS. In the experiment shown in Fig. 6H, 10 days after engraftment, mice were intraperitoneally injected with ActD (60 μg/kg/day) and/or orally fed with venetoclax (INTERCHIM; catalog no. 3 × 552A) daily for 5 days. Vehicle (60% Phosal 50 PG, 30% PEG400, 10% ethanol) was administered as control. Mice were sacrificed after an additional day without treatment. Spleen weights were measured and percentages of CD45.2 and differentiation markers Mac-1/Gr-1 triple-positive cells were analyzed using FACS.

For the patient-derived xenografts, shown in Fig. 5A (top), 10 to 12 week-old NOD/SCID gamma SGM3 (NSGS) mice were treated with busulfan (Busilvex, 20 mg/kg) followed 24 hours later by engraftment of 9 × 10⁴ NPM1c-AML patient mononuclear cells (NPM1c/FLT3 wt). Seventy-two days later, mice were treated daily with ActD (60 μg/kg/day) for 1 or 4 days and sacrificed for bone marrow cells.In the experiment shown in Fig. 5A (bottom), NSGS mice were irradiated with 1.5 Gy 24 hours before engraftment of 2.85 × 10⁵ AML patient mononuclear cells. Eighty-seven days later, mice received the first intravenous injection with PBS or ActD (120 μg/kg/day) and then were treated daily for 4 more days by intraperitoneal injection, followed by 2 days without any further treatment prior to sacrifice. Percentage of human CD33⁺ and/or CD45⁺ and KIT⁺ bone marrow cells were analyzed using FACS. In the experiment shown in Fig. 6J, 5 to 10-week-old NSGS mice were preconditioned by 1.25 Gy irradiation, followed by engraftment of 1 × 10⁶ NPM1c-AML patient mononuclear cells. Thirty-five days later, mice were ActD-treated (60 μg/kg/day) and/or venetoclax-treated (100 mg/kg/day) daily 5 days per week for 2 weeks. Percentage of human CD33⁺ and/or CD45⁺ bone marrow cells was analyzed using FACS by intrabone aspiration 5 days before and 15 days after starting the treatment.

For the NPM1c/IDH1^R132H GFP⁺ leukemia mouse model, C57BL/6 mice were irradiated with 4.5 Gy and then engrafted with 10⁶ AML cells, a kind gift from T.W. Mak. Thirteen days after engraftment, mice were intraperitoneally injected with ActD (60 μg/kg/day) and/or fed by venetoclax (100 mg/kg/day) daily, 4 or 5 days per week. Percentage of GFP⁺ bone marrow cells was analyzed using FACS by intrabone aspiration 3 days before and 16 days after starting the treatment. Mice were then followed until they reach humane endpoints and then euthanized.

For the experiment shown in Figs. 5C and 6H, NSGS mice were treated with Busulfan (Busilvex; 20 mg/kg) for 24 hours, followed by engraftment of 10⁶ AML3 derivatives for 7 days. Mice were intraperitoneally injected with ActD (60 μg/kg/day) for 4 weeks (5 days per week; Fig. 5C) or 3 weeks (Fig. 6H). Human CD45⁺ bone marrow cells were examined by FACS and Western blot analysis.

For the experiments shown in Figs. 1A, 2A, and Supplementary Fig. S2A, LSK fraction was sorted and stained as described previously (49) from NPM1c^A/+ and control mice (26) 4 weeks after tamoxifen injection. Briefly, committed hematopoietic cells were depleted through MACS LS column (Miltenyi Biotec) and FLSK cells were sorted by FACS using the following antibodies: CD3e-biotin, CD8-biotin, IgM-biotin (Life Technologies; catalog no. 553728, catalog nos. 13–0081–85 and 13–5790–85), CD4-biotin, B220-biotin, Mac1-biotin, Tert119-biotin, CD19-biotin, Nk1.1-biotin (BD Biosciences; catalog nos. 553728, 553086, 553309, 553672, 553784, 553163), Gr1-biotin, CD150-perCp/Cy5.5, c-kit-APC/Cy7 (BioLegend; catalog nos. 108404, 115922, 105826), Il7Ra-biotin, CD135-biotin, streptavidin-Pacific Blue, Sca1-PE/Cy7, CD48-APC (eBioscience; catalog nos. 13–1271–85, 13–1351–82, 48–4317–82, 25–5981–81, 17–0481–82). FLSK cells were resuspended in 20 μL of StemSPAN SFEM (StemCell Technologies; catalog no. 9600), then seeded on Lab-Tek II Chamber Slide (Thermo Fisher Scientific; catalog no. 154453) coated with Retronectin (ClonTech; catalog no. T202). Rabbit polyclonal anti-TOM20 (Santa Cruz Biotechnology; catalog no. FL-145, dilution 1:100), mouse monoclonal anti-B23 (Sigma-Aldrich; catalog no. B0556, dilution 1:100), and mouse monoclonal anti-PML (Millipore Sigma; catalog no. MAB3738, clone 36.1–104, dilution 1:100) were used for the detection. Z-stack values were acquired on a Leica SP5, then deconvolved using the Richardson–Lucy algorithm using measured point spread function. Representative iso-surface images and the analysis of number and volume of objects were obtained by Imaris 7 (Bitplane).

All experiments in mice were performed in accordance with protocols approved by the Comités d'éthique en Expérimentation Animale Paris-Nord no. 121 (project no. 20710–2019071616338948 v2). For the mouse NPM1c leukemia model, C57BL/6-CD45.1, mice were purchased from Charles River Laboratories (agreement no. C 69 208 1301) at 6 weeks old and kept under specific pathogen–free (SPF) conditions. For OCI-AML3 or primary patient blasts xenograft experiment, NOD/Shi-scid IL2rγ^−/− (NSG) mice were obtained from Jackson Laboratories and kept in SPF conditions.

For immunoprecipitation analysis, cell extraction was performed with RIPA lysis buffer containing 50 mmol/L Tris (pH 8.0), 0.15 mol/L NaCl, 1% NP40, 1% sodium deoxycholate, 0.1% SDS, 1 mmol/L phenylmethylsulfonylfluoride, 1 μg/mL aprotinin, and 1 μg/mL leupeptin, 10 mmol/L NEM. Equal amounts of protein lysates were used for immunoprecipitation with indicated antibodies. For protein expression analysis, cells were collected, counted, and lysed in XT sample buffer (Bio-Rad) or 2x Laemmli buffer both containing 20 mmol/L NEM and 10% 2-mercaptoethanol. Samples were heated at 95°C for 10 minutes. For disulfide-mediated PML–NPM1c complexes, protein samples were divided into two equal aliquots with XT sample buffer containing 20 mmol/L NEM. One of the aliquots was freshly added disulfide bond breaker tris-(2-carboxyethyl) phosphine (TCEP; Thermo Fisher Scientific) and heated at 95°C for 5 minutes. Details on immunodetection in precipitation sample and cell lysate can be found in Supplementary Materials and Methods.

Mitochondrial superoxide, membrane potential, and mass were measured by using MitoSOX, MitoTracker CMXRos, and MitoTracker red FM probes (Thermo Fisher Scientific; catalog nos. M36008, M7512, M22425), respectively. Live-cell labeling was done according to manufacturer's instructions and analysis made by FACS (FACSCalibur, BD Biosciences).

To measure relative mitochondrial gene expression by reverse transcriptase PCR, whole-cell RNA was isolated using RNeasy Mini Kit (Qiagen; catalog no. 74104). Reverse transcription was performed using iSCRIPT cDNA Synthesis Kit following the manufacturer's instructions (Bio-Rad). Quantitative real-time PCR analysis was performed using Roche LightCycler 480 system with SYBR Green PCR Master Kit (Roche). Primers are listed in Supplementary Material Methods.

Cytosolic/mitochondria resident mtDNA extraction was carried out as described previously (36). Briefly, 15 × 10⁶ mESCs or 10 × 10⁶ AML cells were washed with ice-cold PBS and permeabilized by digitonin buffer containing 150 mmol/L NaCl, 50 mmol/L HEPES (pH 7.4), 1 mol/L Hexylene glycol, and 25 μg/mL digitonin for 10 minutes at 4°C, then centrifuged at 2,000 × g for 15 minutes. The cytosolic supernatants were transferred to fresh tubes and spun at 12,000 rpm for 10 minutes to remove any remaining cellular debris, yielding cytosolic preparations free of nuclear, mitochondrial, and any other organelles. Cytosolic DNA was purified by using QIAquick Nucleotide Removal Kit (QIAGEN; catalog no. 28304). The intact cell pellets were resuspended in 50 μmol/L NaOH and boiled for 30 minutes to solubilize DNA. Fifty microliters of 1 mol/L Tris-HCl pH 8.0 was added to neutralize the pH. Mitochondria and nuclear DNA were purified by using QIAamp DNA Mini Kit (QIAGEN; catalog no. 51304). Primers are listed in in Supplementary Materials and Methods. mtDNA was analyzed by Southern blotting. Briefly, 1 μg of mitochondria DNA was separated over a 0.5% agarose gel overnight in TAE, blotted by capillary transfer onto Hybond-N+ membranes (GE Healthcare). The mtDNA was visualized by Southern blotting using a probe against nucleotide positions 168–604 of human mtDNA labeled using digoxigenin-labeled nucleotide (PCR DIG Probe Synthesis Kit; Sigma-Aldrich; catalog no. 11636090910) and later revealed using anti-digoxigenin antibodies.

To measure 2′3-cGAMP, 40 × 10⁶ AML cells were washed with ice-cold PBS and lysed by M-PER extraction reagent (Thermo Fisher Scientific; catalog no. 78501). 2′3′-cGAMP were quantified by an ELISA Kit (Cayman; catalog no. 501700), which is based on the competition between 2′3′-cGAMP and a 2′3′-cGAMP–horseradish peroxidase conjugate for a limited amount of 2′3′-cGAMP polyclonal antiserum.

Whole-cell RNA from one million cells of AML derivates or peripheral blood mononuclear cells of patient were purified using RNeasy Mini Kit (QIAGEN). Microarray were performed at the core facility of Institut de Recherche Saint Louis of Université de Paris (Paris, France). Details can be found in Supplementary Materials and Methods. Primary data are accessible in ArrayExpress (accession ID: E-MTAB-10397). The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD025507.

AML2-derived cell lines were washed and fixed in 1.6% glutaraldehyde in 0.1 mol/L phosphate buffer. Cell pellets were secondary fixed with 2% osmium tetroxide and dehydrated using ethanol. Cells were embedded in Epon 812. Ultrathin sections and then stained with standard uranyl acetate and lead citrate solutions. Mitochondrial morphology was analyzed using FEI Tecnai 12 electron microscopy equipped with SIS Mega view III CCD camera.

For the experiment shown in Figs. 1H and 5G, gene set enrichment analysis (GSEA) was performed using Hallmarks (h.all.v7.1), Gene ontology (c5.all.v7.1), or Curated (c2.cp.KEGG) gene set. 371 genes were selected on the basis of GSEA Hallmarks gene set (h.all.v7.1). For the experiment shown in Supplementary Fig. S2B, GSEA was performed using Gene ontology (c5.all.v7.1) or Curated (c2.cp.KEGG) gene set.

GraphPad Prism software was used to perform multiple unpaired two-tailed t test to determine the P values and look for significant changes in the data generated from the isogenic cell lines and different drug treatment in vivo or ex vivo. All data are expressed as mean ± SD of technical replicates from a representative experiment of the performed ones, as indicated. For all graphs, *, P = 0.01–0.05; **, P = 0.001–0.01; ***, P < 0.001. For the experiment shown in Supplementary Fig. S3F, ROS and inflammation-related gene changes from AML2-expressing NPM1c versus NPM1 or AML2-expressing NPM1 treated with ActD versus DMSO were selected to perform linear regression analysis using GraphPad Prism software.

L. Adès reports consulting or advisory roles for BMS/Celgene and AbbVie and research funding from BMS/Celgene. R. Itzykson has received honoraria from AbbVie, Amgen, Astellas, BMS-Celgene, Daiichi-Sankyo, Jazz Pharmaceuticals, Karyopharm, Servier, and Stemline, and research funding from Janssen, Novartis, and Oncoethix (now Merck SD). A. Bazarbachi is on the speaker bureau or advisory board of Novartis, Roche, Sanofi, Jazz, Adienne, Astellas, Takeda, Hikma, Celgene, Janssen, MSD, AbbVie, Pfizer, and Amgen. B. Falini licensed a patent on NPM1 mutants (n. 102004901256449). B. Falini and M.P. Martelli declare honoraria from Rasna Therapeutics, Inc for scientific advisor activities. M.P. Martelli also declares consultancy at scientific advisory boards for AbbVie, Amgen, Celgene, Janssen, Novartis, Pfizer, and Jazz Pharmaceuticals, and honoraria from Amgen, Celgene, Janssen, and Novartis. L. Brunetti declares consultancy at scientific advisory boards for AbbVie. Other authors have nothing to disclose.

H. Wu: Data curation, formal analysis, investigation. D. Rérolle: Data curation, investigation. C. Berthier: Investigation. R. Hleihel: Investigation. T. Sakamoto: Resources. S. Quentin: Investigation. S. Benhenda: Resources. C. Morganti: Resources, investigation. C. Wu: Investigation. L. Conte: Investigation. S. Rimsky: Methodology. M. Sebert: Resources. E. Clappier: Resources. S. Souquere: Investigation. S. Gachet: Resources, investigation. J. Soulier: Resources. S. Durand: Formal analysis. J.J. Trowbridge: Resources. P. Benit: Investigation. P. Rustin: Investigation. H. El Hajj: Validation. E. Raffoux: Resources. L. Ades: Resources. R. Itzykson: Resources. H. Dombret: Resources. P. Fenaux: Resources. O. Espeli: Methodology. G. Kroemer: Methodology. L. Brunetti: Resources. T.W. Mak: Resources. V. Lallemand-Breitenbach: Conceptualization. A. Bazarbachi: Conceptualization. B. Falini: Conceptualization. K. Ito: Resources, investigation. M. Martelli: Conceptualization. H. de Thé: Conceptualization, resources, data curation, formal analysis, supervision, writing–original draft.

We warmly thank P.G. Pelicci for sharing mouse leukemia models, V. Montcuquet for the animal husbandry, N. Setterblad and S. Duchez for the imaging and cytometry platforms, C. Vallot for bioinformatic advice, C. Bally for help with the patient samples, the 3P5 proteomic facility of Institut Cochin, the imaging ORION platform of Collège de France for decisive help in PML NBs and mitochondrial alteration quantification, A. Mourier for discussions and advice, A. Rötig for DNA of mitochondria-depleted cells, M.H. Verlhac, E. Gilson, U. Sahin, J. Ablain, R. Rodriguez, C. Esnault, and other lab members for advice and comments on the manuscript. This work is dedicated to the first author's late father Jung-Mao Wu and all those loved ones lost to COVID-19. Work in the Paris laboratories is supported by Collège de France, INSERM, CNRS, University de Paris and University PSL, INCA [CAMELIA project, TRANSCAN (DRAMA project) and PLBIO19-198], Fondation du Collège de France and Fonds Saint Michel, and the Sjöberg Award (to H. de Thé). This project has received funding from the European Research Council (ERC) under the European Union Horizon 2020 research and innovation program (Grant agreement n° 785917, PML-THERAPY). C. Morganti is supported by NYSTEM Einstein Training Program in Stem Cell Research. K. Ito and J.J. Trowbridge are supported by NIH grants, and K. Ito is a Leukemia Lymphoma Society Scholar. Work in Beyrouth and Perugia is supported by ERC grants (785917, 740230, 725725) and the ARC Foundation for Cancer Research (Leopold Griffuel Prize to B. Falini).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.