Extracellular ATP and CD39 Activate cAMP-Mediated Mitochondrial Stress Response to Promote Cytarabine Resistance in Acute Myeloid Leukemia

Chemotherapy resistance is the major therapeutic barrier in acute myeloid leukemia (AML), the most common acute leukemia in adults. AML is characterized by clonal expansion of immature myeloblasts and initiates from rare leukemic stem cells (LSC). Despite a high rate of complete remission after conventional first-line induction chemotherapy [e.g., daunorubicin or idarubicin plus cytarabine (AraC)], the long-term prognosis is very poor for patients with AML. To date, the 5-year overall survival is still about 30% to 40% in patients younger than 60 years old and less than 20% in patients older than 60. This results from the high frequency of distant relapses (50% and 85% for patients younger and older than 60 years of age, respectively) caused by tumor regrowth initiated by chemoresistant leukemic clones (RLC) and characterized by a refractory phase during which no other treatment has shown any efficacy thus far (1, 2). Even with recent efficient targeted therapies that are FDA-approved or under clinical development, therapy resistance remains the major therapeutic barrier in AML. Therefore, understanding the molecular and cellular mechanisms driving chemoresistance is crucial for the development of new treatments eradicating RLCs and to improve the clinical outcome of these patients.

The biological basis of therapeutic resistance (drug efflux, detoxification enzymes, inaccessibility of the drug to the leukemic niche) currently represents an active area of research. However, the molecular mechanisms underlying AML chemoresistance are still poorly understood, especially in vivo. It is nevertheless increasingly recognized that the causes of chemoresistance and relapse reside within a small cell subpopulation within the bulk of leukemic cells. Supporting this view, clinical studies have shown that the presence of high levels of CD34⁺CD38^lo/−CD123⁺ cells at diagnosis correlates with adverse outcome in patients with AML in terms of response to therapy and overall survival (3, 4). Consistent with these data, Ishikawa and colleagues (5) have observed that this population is also the most resistant to cytarabine treatment in vivo. As a first step toward successful therapeutic eradication of these RLCs, it is now necessary to comprehensively profile the intrinsic and acquired characteristics of this cell population. We have recently established a powerful preclinical model to screen in vivo responses to conventional genotoxics and to mimic the chemoresistance and minimal residual disease observed in patients with AML after chemotherapy (6). Accordingly, we have fully analyzed the response to chemotherapy of leukemic cells in AraC-treated AML patient-derived xenograft (PDX) mouse models. Surprisingly, we have found that cytarabine treatment equally kills both cycling and quiescent cells and does not necessarily lead to LSC enrichment in vivo. However, we observed that cytarabine chemoresistant leukemic cells exhibit elevated oxidative phosphorylation (OxPHOS) activity and that targeting mitochondrial oxidative metabolism with OxPHOS inhibitors sensitizes resistant AML cells to AraC (6, 7). Consistent with our findings, several groups have also demonstrated that essential mitochondrial functions contribute to resistance to multiple treatments in other cancer types (8–11).

Hyperleukocytosis is a clinical condition observed in patients with AML, which may lead to life-treatening complications such as leukostasis and is associated with a higher risk of relapse. Importantly, this condition is sustained by several mediators of inflammation, which were also reported to contribute to chemoresistance in AML (12–14). Supporting this idea, a recent study reported that inhibition of the inflammatory chemoresistance pathway with dexamethasone improved outcome of patients with AML (15). In line with these observations, recent work from our group (6) has highlighted a gene signature associated with the immune and inflammatory response after AraC treatment of PDX models in vivo. Among immune response mechanisms, the adenosine signaling pathway is one of the most prominent in cancer. CD39/ENTPD1 (ectonucleoside triphosphate diphosphohydrolase-1) is a member of the family of ectonuclotidases present on the outer surface of cells and a key component of the adenosine pathway. Together with CD73, CD39 catalyzes hydrolysis of extracellular adenosine triphosphate (eATP) and adenosine diphosphate (ADP) to produce adenosine, a recognized immunosuppressive molecule (16, 17). Therefore, CD39 has a critical role in tumor immunosurveillance and inflammatory response. Furthermore, although other nucleoside triphosphate diphosphohydrolases (NTPDases) exist, CD39 appears to be the main NTPDase in T lymphocytes and regulatory T cells (18). Recent lines of evidence have revealed high expression and activity of CD39 in several blood and solid tumors (such as head and neck cancer, thyroid cancer, colon cancer, pancreatic cancer, kidney cancer, testis cancer, and ovarian cancer), implicating this enzyme in promoting tumor growth and infiltration (19), and CD39 blockade was recently shown to enhance anticancer combination therapies in preclinical mouse models of solid tumors (20). Furthermore, CD39 is frequently detected in primary tumor cells, including AML blasts, cancer-derived exosomes, and tumor-associated endothelial cells. Notably, CD39 was reported to contribute to the immunosuppressive microenvironment in AML (21), while extracellular nucleotides (ATP, UTP) can inhibit AML homing and engraftment in NSG mice (22).

In this study, we employed computational analysis of transcriptomic datasets obtained from PDX models treated with cytarabine and primary patient samples to identify new druggable and relevant cell-surface proteins specifically expressed by RLCs. Among these genes, we uncovered CD39/ENTPD1 and confirmed that CD39 expression and activity are increased in residual AML cells post-chemotherapy in vitro, in vivo, and in the clinical setting. Herein, we have also shown that high CD39-expressing resistant AML cells rely on an enhanced mitochondrial metabolism and are strongly dependent on the coordinated induction of cAMP-PKA and ROS/ATF4 signaling downstream of CD39 to survive to chemotherapy. Accordingly, targeting CD39 markedly enhanced cytarabine cytotoxicity in AML cell lines and primary patient samples in vitro and in vivo through the inhibition of mitochondrial OxPHOS function, and this effect could be mimicked by inhibition of the PKA pathway. Overall, this work shows that a key mechanism of resistance to AraC involves CD39-dependent cross-talk between the energetic niche and leukemic mitochondrial functions through the CD39–P2RY13–cAMP–PKA axis in AML.

To identify new potential therapeutic targets involved in the onset of cytarabine resistance in vivo, we analyzed a previously identified signature of 68 genes that are significantly upregulated in residual AML cells from PDXs upon cytarabine treatment in vivo (ref. 6; GSE97393). Bioinformatic analysis of this specific gene signature showed an enrichment in several key cancer and immune response signaling pathways, including 8 genes involved in the inflammatory response (Supplementary Fig. S1A). As inflammation has been previously shown to play a critical role in the development of chemoresistance and to be linked to a poor prognosis in AML (15), we focused on this latter group of genes. Within this subset, we identified five genes encoding plasma membrane proteins sensitive to existing inhibitors, thus representing relevant druggable targets of RLCs in vivo. Importantly, two of these genes, ENTPD1 (CD39) and the fatty-acid translocase CD36, were specifically overexpressed in AML cells compared with normal hematopoietic stem cells, highlighting their potential as therapeutic targets (Supplementary Fig. S1B and S1C). As CD36 has already been shown by our group and other groups to be a prognostic marker in myeloid leukemia (6, 23–25), we focused on CD39. The ecto-ATPase (eATPase) CD39 is well known for its immunosuppressive and proangiogenic function in multiple cancer types (16, 17). However, its role in AML cells and its contribution to AML chemoresistance are currently unknown.

Our gene expression data indicated that CD39 expression was upregulated in residual AML cells upon cytarabine treatment. To confirm that enhanced transcription correlated with increased surface protein levels, we studied CD39 cell-surface expression in residual viable AML cells from the bone marrow of 36 PDXs (Supplementary Table S1) following treatment with cytarabine (representative flow plot in Supplementary Fig. S2A). As expected, we observed a significant cytoreduction of the total cell tumor burden in the bone marrow and spleen (26 PDXs treated with 60 mg/kg/day of cytarabine; Fig. 1A; Supplementary Fig. S2B and S2C; 10 PDXs treated with 30 mg/kg of cytarabine; Supplementary Fig. S2D and S2E) of these different PDX models upon cytarabine treatment in vivo. In line with our gene expression data, despite a certain level of interpatient variability, we observed in the majority of our PDXs an increase both in the percentage of CD39-positive cells and in the intensity of CD39 expression in the bulk residual AML population as well as in the immature CD34⁺CD38⁻ residual cell subpopulation in cytarabine-treated compared with vehicle-treated xenografted mice (Fig. 1B and C; Supplementary Fig. S2B–S2E). We then investigated the expression of CD39 in our panel of cell line–derived xenografted (CLDX) models characterized by different levels of sensitivity to AraC in vivo (6) and in vitro (representative gating strategy for in vivo and in vitro experiments, respectively, in Supplementary Fig. S3A and S3B). Although AraC treatment resulted in the induction of cell death in the four AML cell lines tested (HL60, MOLM14, U937, and KG1a; Supplementary Fig. S3C), changes in the transcript and the cell-surface expression of CD39 were significant but variable across these AML cell lines in vitro (Supplementary Fig. S3D–S3F). While MOLM14 and OCI-AML3 CLDX models are highly resistant to AraC chemotherapy in vivo, the U937 model is more sensitive and initially responds well to the treatment (total cell tumor burden fold reduction greater than 10 in cytarabine- vs. PBS-treated mice; Fig. 1D). The majority (∼70%) of the intrinsically resistant MOLM14 and OCI-AML3 cells expressed CD39 in vivo. In contrast, only a small fraction (∼30%) of U937 cells expressed CD39. Interestingly, we observed a significant increase in the percentage of CD39-positive cells as well as the intensity of CD39 cell-surface expression (Fig. 1E and F; Supplementary Fig. S3G), associated with an increase in CD39 eATPase activity (Fig. 1G) in residual U937 cells surviving post chemotherapy, whereas no or minimal change in CD39 expression and activity was detected in MOLM14 and OCI-AML3 cells (Fig. 1E–G; Supplementary Fig. S3G).

Next, we studied the kinetics of upregulation of CD39 in RLCs in vivo, during cytarabine treatment (day+3), immediately after the last dose of cytarabine treatment (day+5), and at day+8 in AML-xenografted NSG mice. Starting from day+5, we observed the appearance of RLCs with an increased CD39 expression (Fig. 1H and I). Of note, CD39-positive cells were not decreased at day 3 (Fig. 1H and I), and selection for CD39-positive cells occurred without genetic and mutational changes over time, as major founder mutations were present at diagnosis in patients and in the PDX throughout the same time course (Supplementary Fig. S4A). Altogether, these data strongly suggest that the CD39-positive phenotype may preexist before xenotransplantation and chemotherapy, and be selected and enhanced by cytarabine treatment in vivo. To test this hypothesis, we assessed whether sorted CD39^hi and CD39^lo cell subpopulations had a differential sensitivity to cytarabine treatment. We first sorted CD39^hi and CD39^lo fractions from therapy-naïve AML cells from MOLM14 and U937 AML cell lines in vitro (Supplementary Fig. S4B). Although the proliferation rate and cell-cycle status were similar in the two cellular subsets (Supplementary Fig. S4C and S4D), both MOLM14 and U937 CD39^hi fractions showed a significantly higher IC₅₀ for cytarabine compared with their respective CD39^lo counterparts (Supplementary Fig. S4E and S4F). This diminished sensitivity to cytarabine of the CD39^hi cell subset was confirmed on FACS-purified CD39^hi and CD39^lo AML fractions obtained ex vivo from both CLDXs and PDXs (Supplementary Fig. S4G). Interestingly, this differential sensitivity of the CD39^hi and CD39^lo cells was specific to cytarabine because we did not observe any significant difference in the induction of cell death in the two cell subsets upon treatment with idarubicin or venetoclax (ABT-199; Supplementary Fig. S4H).

Overall, our data indicate that a CD39^hi phenotype characterizes in vitro and in vivo a subset of AML cells with a higher intrinsic resistance to cytarabine treatment. Importantly, this phenotype preexists and is amplified upon cytarabine chemotherapy in vivo.

To evaluate the clinical relevance of our findings, we analyzed the expression of CD39 in patients with AML. Analysis of a cohort of 162 patients with AML at diagnosis indicated heterogeneous expression of CD39 (Fig. 2A). The expression of CD39 was not associated with the presence of specific recurrent mutations in AML (Supplementary Fig. S5A). However, we observed a correlation between CD39 cell-surface expression and FAB classification with a lower level of expression associated with the most undifferentiated AML subtypes (Supplementary Fig. S5B). This observation was also supported by the analysis of publicly available gene expression datasets from patients with AML (Supplementary Fig. S5C and S5D).

We then followed 98 of these patients comparing CD39 cell-surface expression at diagnosis and at day 35 (D35) after intensive chemotherapy. In accordance with our preclinical model, we showed a significant tumor reduction or complete remission in most of the patients after treatment (Fig. 2B), and demonstrated an overall increase in the percentage of CD39-positive cells in the residual blasts from those patients at day 35 post–intensive chemotherapy (Fig. 2C). We then stratified these patients with AML based on their fold enrichment in CD39-expressing cells upon chemotherapy in respect to diagnosis (CD39 was considered enriched with D35 vs. diagnosis fold change > 1.25, cutoff higher than the technical coefficient of variation for FACS). This defined a group of “high CD39 ratio” (n = 74) and “low CD39 ratio” patients (n = 24; Fig. 2D). Strikingly, the “high CD39 ratio” patients displayed a significantly worse disease-free survival compared with the “low CD39 ratio” group (Fig. 2E). This survival disadvantage was even more evident when focusing on the group of patients younger than 60 years of age (Fig. 2F). Finally, we investigated whether CD39-positive cell expansion upon chemotherapy could further stratify patients classified in favorable, intermediate, and high cytogenetic risk groups (Supplementary Fig. S6A). The increase in CD39-positive cells did not significantly improve the prognostic classification of intermediate and high cytogenetic risk patients (Supplementary Fig. S6B and S6C). However, this analysis revealed that patients with AML from the favorable cytogenetic risk subgroup, but characterized by a marked increase in CD39-expressing cells upon chemotherapy, displayed a significantly higher rate of short-term relapse and poorer clinical outcome (Fig. 2G).

Overall, these findings highlight the clinical relevance of our results obtained in PDXs and CLDXs and define CD39 as a marker of minimal residual disease and adverse prognosis in patients with AML.

As previous studies have demonstrated that drug-resistant AML cells exhibit high OxPHOS function and gene signatures in vivo (6, 9, 10), we investigated whether an OxPHOS gene signature was enriched in the transcriptomes of AML cells with high CD39 expression. We confirmed a positive correlation between CD39 RNA expression and our previously defined “high OxPHOS” gene signature (6), making use of two independent transcriptomic databases from patients with AML that we stratified as CD39^lo and CD39^hi [GSE97393: NES = −1.84, FDRq < 0.001 and The Cancer Genome Atlas (TCGA): NES = −1.50, FDRq = 0.004, respectively; Fig. 3A; Supplementary Fig. S7A]. We then compared the metabolic status and mitochondrial activity of primary AML cells from patients with high or low levels of CD39 expression (Supplementary Fig. S7B and S7C). Primary cells from patients with CD39^hi AML displayed on average increased eATPase activity compared with CD39^lo patients (Fig. 3B), and this was associated with a modest increase in mitochondrial membrane potential (MMP) and a larger increase in basal oxygen consumption rate (OCR; Fig. 3C and D). We then sorted the low and high CD39 cell fractions from one of our PDXs (TUH10) after treatment in vivo with cytarabine or PBS (Supplementary Fig. S7D). In line with our previous results on the primary patient samples, the ex vivo analysis of the metabolic status and OCR of the two cell subsets showed increased basal, maximal uncoupler-stimulated respiration and ATP-linked respiration in CD39^hi AML cellular subsets (Fig. 3E; Supplementary Fig. S7E–S7H). This was coupled with an augmented extracellular acidification rate (ECAR) linked to glycolysis, suggesting a global enhancement of cellular energetic capacity (Supplementary Fig. S7I and S7J) in CD39^hi fractions compared with CD39^lo fractions from PBS-treated mice. In accordance with our previously published data (6), cytarabine treatment resulted in the selection of residual viable AML cells with substantially increased basal and maximal uncoupler-stimulated OCR as well as ATP-linked OCR (Fig. 3E; Supplementary Fig. S7E–S7H). Overall, this indicated that increased levels of CD39 were associated with an enhanced mitochondrial activity and OxPHOS function in AML cells, which we previously identified as a feature of cytarabine-resistant AML cells.

To specifically study the direct effect of modulating CD39 expression on AML cell metabolism, we transduced the AML cell line MOLM14 with viral vectors expressing two different shRNAs targeting CD39. Transduction of MOLM14 with the shCD39-expressing lentiviral vectors resulted in efficient silencing of the ectonucleotidase both at the mRNA level and at the protein level (Supplementary Fig. S8A and S8B), leading to a significant downregulation of the expression of this marker at the cell surface (Supplementary Fig. S8C). Silencing of CD39 resulted in a dramatic decrease in both basal and ATP-linked OCR in MOLM14 (Fig. 3F–H), which translated into a reduced generation of mitochondrial-derived ATP (Fig. 3I). This decreased mitochondrial OxPHOS activity was associated with a reduced expression of subunits of the electron transport chain (ETC) complexes and of well-known effectors of mitochondrial biogenesis (i.e., NRF1, PGC1α; Fig. 3J). Overall, these results indicate that CD39 positively controls mitochondrial function and OxPHOS, at least in part, by controlling the expression of the key transcriptional activators NRF1 and PGC1α promoting mitochondrial biogenesis.

Next, we sought to determine whether inhibition of CD39 activity by polyoxometalate 1 (POM1), a pharmacologic inhibitor of nucleoside triphosphate diphosphohydrolase activity (26), could inhibit the metabolic reprogramming triggered by cytarabine and sensitize AML cells to the chemotherapeutic treatment in vitro (Fig. 4A). As expected, POM1 inhibited the increase of the CD39 eATPase activity upon cytarabine treatment in all AML cell lines tested in vitro (MOLM14, OCI-AML3, MV4-11; Fig. 4B; Supplementary Fig. S8D), leading to an accumulation of eATP in the medium (MOLM14; Supplementary Fig. S8E). Accordingly, inhibition of CD39 by POM1 in MOLM14 abrogated the expansion of the extracellular ADP and AMP pools triggered by cytarabine (Supplementary Fig. S8E). Furthermore, POM1 treatment repressed the cytarabine-induced increase in basal OCR, mitochondrial mass, mitochondrial DNA (mtDNA) level, and the protein level of ETC subunits (Fig. 4C–F; Supplementary Fig. S8F–S8I). Importantly, POM1 treatment significantly enhanced the loss of mitochondrial membrane potential (Fig. 4G; Supplementary Fig. S8J) and the induction of apoptosis (Fig. 4H; Supplementary Fig. S8K) triggered by cytarabine treatment in vitro in all three AML cell lines tested.

Altogether, our results strongly suggest that CD39 activity directly affects AML cell sensitivity to cytarabine through the regulation of mitochondrial function.

Because our results strongly support the assertion that CD39 expression influences mitochondrial OxPHOS, we sought to explore signaling pathways downstream of CD39 that may promote OxPHOS metabolism and chemoresistance. Therefore, we performed targeted experiments to characterize global transcriptional changes induced by silencing CD39 (shCD39 vs. shControl) in the therapy-resistant MOLM14 AML cell line. A total of 152 genes were significantly and differentially expressed in MOLM14 upon silencing of CD39 [42 upregulated, 110 downregulated; log₁₀ (FDR) < –1.25, log₂ (fold-change) > 1.0; Fig. 5A; Supplementary Table S2]. Consistent with our metabolic assays, gene set enrichment analysis (GSEA) indicated that CD39 loss in MOLM14 cells was negatively correlated with a gene set representing High OxPHOS function (ref. 6; NES = −1.38, FDR q = 0.005; Supplementary Fig. S9A). Furthermore, the shCD39 downregulated gene signature was significantly enriched in the transcriptomes of AML patient samples characterized by poor response to cytarabine in vivo in NSG (low vs. high responders, GSE97393; Fig. 5B) as well as in the transcriptomes of cytarabine-resistant AML cells from three AML PDXs (GSE97631, Fig. 5C). Gene ontology analysis of the 110 genes downregulated upon CD39 silencing indicated an enrichment in biological processes involved in cell-cycle control, DNA repair, responses to stress/stimuli, metabolism, and signaling (Supplementary Fig. S9B; Supplementary Table S3). GSEA of the shCD39-downregulated gene signature revealed a strong enrichment in genes involved in oxidative metabolism and reactive oxygen species (ROS) balance (FOXO3 and mTORC1 targets, among others), unfolded protein response (UPR), ATF/CREB pathway, and cAMP/PKA signaling (P < 0.01; Fig. 5D; Supplementary Table S4). Interestingly, all these pathways are well known to be associated with oxidative stress response and mitochondrial homeostasis. In line with this observation, transcription factor enrichment analysis identified signatures of several transcription factors playing key roles in mitochondrial function and stress response (such as ATF4/6, PARP1, and E2F1; Supplementary Fig. S9C). Next, we compared the transcriptomes of AML patient cells characterized by high CD39 expression to AML patient cells with low CD39 expression (Fig. 5E; Supplementary Fig. S7A). We confirmed a strong link between CD39 expression and the activation of mitochondrial stress response–associated pathways. On the basis of these results, we formulated the hypothesis that CD39 upregulation upon cytarabine treatment could drive an adaptative mitochondrial stress response in AML cells by inducing the coordinated activation of multiple signaling pathways regulating mitochondrial biogenesis and activity as well as ROS production and antioxidant defenses. Dynamic transcriptomic analysis of residual AML cells ex vivo after treatment with cytarabine revealed that the CD39 gene signature and the associated cAMP, ATF4, and FOXO3 pathways are transiently upregulated at early timepoints after chemotherapy. This simultaneously occurred when the disease reached its nadir and was subsequently downregulated between one and two weeks after the end of the treatment (Supplementary Fig. S9D–S9F). This suggests that the induction of CD39 is likely playing a key role in the initial phase of metabolic adaptation in residual AML cells.

We therefore decided to investigate the induction of oxidative stress upon cytarabine treatment and the role of CD39 in the adaptative metabolic response of AML cells to this cytotoxic stimulus. Analysis of total ROS content in CD39-high and CD39-low cellular subsets obtained from AML PDXs pretreated or not with cytarabine showed that, while the chemotherapy tended to trigger an increase in ROS, the level of CD39 expression did not affect the accumulation of ROS in AML cells (Supplementary Fig. S10A). On the contrary, the CD39-high subset displayed a higher basal level of mitochondrial ROS in some samples, and this was maintained upon cytarabine treatment (Supplementary Fig. S10B). Next, we stratified our primary AML patient samples into ROS low and high (based on their DCF-DA staining) and we observed that ROS-high AML cells had a higher level of CD39 expression (Supplementary Fig. S10C and S10D). This suggested that, although the total level of intracellular ROS was not dependent on CD39, total ROS could affect and potentially be involved in CD39 induction in AML cells. We then investigated the role of CD39 in the regulation of ROS homeostasis as well as in the modulation of the stress response–associated ATF4 pathway in MOLM14 cells upon cytarabine treatment. Confirming previous observations in our PDXs, inhibition of CD39 did not significantly affect the accumulation of total cellular ROS triggered by the chemotherapy (Supplementary Fig. S11A). However, cotreatment with POM1 efficiently repressed the cytarabine-associated increase in mitochondrial ROS (Supplementary Fig. S11B). At the same time, cytarabine enhanced the expression of the transcription factor FOXO3a, an important regulator of mitochondrial function and antioxidant defenses, and its target mitochondrial superoxide dismutase SOD2 (but not cytosolic SOD1), as well as of the key mitochondrial stress response transcription factor ATF4. Importantly, concomitant inhibition of CD39 efficiently reversed all these effects (Supplementary Fig. S11C–S11F). We next investigated the modulation of the c-AMP–dependent PKA signaling pathway upon cytarabine treatment of MOLM14 and inhibition of CD39. Analysis of the expression and activation of key targets of the cAMP-dependent signaling pathway showed an increase in the phosphorylation of RRXS*/T*-PKA substrates upon cytarabine treatment (Supplementary Fig. S11G). These results were similar to those observed upon activation of cAMP–PKA in MOLM14 cells treated with four diverse PKA agonists (FSK, IBMX, extraATP, 8-BrcAMP), whereas treatment with PKA antagonist (H89 and PKA inhibitor 14–22 Amide PKAi) inactivated the cAMP–PKA pathway (Supplementary Fig. S11G). Finally, pharmacologic inhibition of CD39 decreased cytarabine-induced phosphorylation of RRXS*/T*-PKA substrates (Supplementary Fig. S11H). This result confirmed that CD39 activation was involved in the control of mitochondrial ROS homeostasis and redox balance, and in the activation of the mitochondrial stress–associated pathways cAMP–PKA and ATF4 upon cytarabine treatment.

To gain insight into the key pathways downstream of CD39 mediating its metabolic effects and supporting the onset of cytarabine resistance, we assessed the transcriptional and biological consequences of silencing ATF4 in MOLM14 AML cells. Notably, the siATF4 downregulated gene signature (100 genes) was significantly depleted in CD39-silenced AML cells and enriched in residual AML cells after cytarabine treatment in vivo (Supplementary Fig. S12A–S12C; Supplementary Table S5). Five transcription factors, including ATF4 itself, were commonly downregulated upon both CD39 and ATF4 silencing in AML cells (Supplementary Fig. S12D). Of note, ATF4 depletion did not significantly affect CD39 expression, supporting a role of ATF4 downstream of CD39 (Supplementary Fig. S12E and S12F). Supporting its important role in CD39-mediated acquisition of resistance to cytarabine, ATF4 genetic inactivation enhanced cytarabine-induced reduction in cell proliferation, loss of mitochondrial membrane potential, and induction of apoptosis (Supplementary Fig. S12G–S12I). However, ATF4 depletion did not have a significant impact on cytarabine-mediated increase in mitochondrial mass, the expression of ETC complex subunits and mitochondrial membrane potential (Supplementary Fig. S12J–S12L). Thus, ATF4 depletion was partially phenocopying cytarabine effects observed upon CD39 inhibition in AML cells. However, our results suggested that ATF4 did not directly contribute to the changes in mitochondrial biogenesis and function induced by CD39 in the leukemic cells exposed to chemotherapy stress.

We next investigated whether inactivation of cAMP–PKA pathway by H89, a well-known pharmacologic agent that inhibits PKA activity, could affect AML mitochondrial functions and enhance cytarabine treatment cytotoxicity similar to CD39 inhibition. As expected, H89-treated MOLM14 cells exhibited decreased levels of pRRXS*/T*-PKA substrates, including in the cytarabine setting (Fig. 5F). Importantly, the PKA-specific inhibitor H89 completely phenocopied the effect of the CD39 inhibitor POM1 on mitochondrial activity by counteracting the increase in mtDNA level (Fig. 5G), mitochondrial mass (Fig. 5H), the expression of ETC complex subunits and PGC1α (Fig. 5I), and basal OCR (Fig. 5J) induced by cytarabine treatment. Finally, H89-treated MOLM14 cells exhibited a significant loss of MMP (Fig. 5K) and decrease in cell viability (Fig. 5L). Interestingly, we noticed that the activation of PKA signaling was detectable as early as 1 hour after exposure to cytarabine in MOLM14 AML cells and peaked between 3 and 6 hours after the start of the treatment (Supplementary Fig. S13A), whereas CD39 induction was detectable only at later timepoints. Therefore, we hypothesized that the cAMP–PKA pathway could be involved in the upregulation of CD39 expression upon cytarabine exposure. In line with this, the PKA inhibitor H89 strongly repressed the increase of CD39 expression triggered by cytarabine treatment in MOLM14 cells (Supplementary Fig. S13B and S13C).

Overall, these results support a model in which cAMP–PKA signaling and the ROS–ATF4 axis act in a coordinated fashion downstream of CD39 as intrinsic pathways of resistance to cytarabine. We propose that the parallel activation of these molecular pathways orchestrates an adaptive mitochondrial stress response by regulating mitochondrial ROS production, OxPHOS biogenesis and activity, and antioxidant defenses. Moreover, our data suggest that the cAMP–PKA pathway is a key mediator of CD39-mediated effects on mitochondrial biogenesis and activity. Our results indicate that the cAMP–PKA axis acts upstream and downstream of CD39, creating a positive feedback loop, which controls and amplifies CD39-driven noncanonical intrinsic resistance pathways of adaptive stress response in AML.

Having established the key role of CD39–cAMP–PKA signaling in determining the metabolic rewiring of chemotherapy-resistant AML cells, we investigated the mechanisms responsible for the increase in intracellular cAMP upon cytarabine-linked CD39 induction. Hydrolysis of extracellular ATP into ADP and AMP by CD39 and the subsequent conversion of AMP to adenosine (ADO) by the ecto-5′-nucleotidase CD73 results in the stimulation of P1 (responsive to ADO) and P2 (responsive to ATP, ADP) purinergic X/Y receptors. This leads to modulation of intracellular cAMP and associated cAMP signaling (Fig. 6A). We therefore formulated the hypothesis that the activation of the cAMP–PKA axis was dependent on CD39-mediated increase of extracellular nucleotide levels and the associated activation of purinergic receptor signaling.

As expected, cytarabine treatment induced an increase in the production of extracellular AMP but surprisingly not ADP or adenosine in MOLM14 AML cells (Supplementary Fig. S14A). In line with our hypothesis, supplementation of exogenous ADP and AMP rescued the increase of cytarabine cytotoxic effect caused by cotreatment of AML cells with POM1 (Supplementary Fig. S14B). Interestingly, treatment of AML cell lines with the eADPase inhibitor ARL67156 (which inhibits only the CD39-mediated enzymatic conversion of ADP to AMP but not the conversion of ATP to ADP; refs. 27, 28) had only mild effects on intracellular cAMP content and did not significantly affect mitochondrial mass or activity. Furthermore, this inhibitor failed to synergistically enhance the cytotoxic effect of cytarabine compared with POM1 (Supplementary Fig. S15A–S15F). This suggested that CD39-derived eADP level was required for the activation of downstream signaling pathways that support AML metabolic rewiring and chemoresistance.

In parallel, analysis of transcriptomes from residual AML cells from PDXs with cytarabine in vivo highlighted a pronounced increase in the expression of the ADP-responsive G protein–coupled P2RY13 purinergic receptor concomitant with the increase of ENTPD1/CD39 expression (Fig. 6B). The increase of P2RY13 expression at the surface of residual cells upon cytarabine treatment was confirmed by flow cytometry analysis in vitro and in vivo on human AML cell lines and PDXs (Supplementary Fig. S16A and S16B). Of note, no other purinergic receptor gene was found differentially expressed in cytarabine residual AML cells compared with vehicle-treated controls (Fig. 6B). Moreover, although we occasionally observed an increase in cell surface CD73 expression upon cytarabine treatment in vitro in our AML cell lines, we did not observe any induction of CD73 expression in vivo in any of the PDXs tested after chemotherapy (Supplementary Fig. S16C and S16D). Intriguingly, we found the expression of P2RY13 and ENTPD1 to be strongly correlated in multiple AML patient datasets tested (Fig. 6C and D). Moreover, similar to the CD39 gene signature, we found the P2RY13-associated gene signature (which we generated from the transcriptomes of patients with AML) to be highly enriched in residual AML cells upon cytarabine treatment of PDXs in vivo (Supplementary Fig. S16E–S16H; Supplementary Table S6). Finally, we found that CD39 and P2RY13 were concomitantly upregulated not only after cytarabine treatment, but also upon exposure to hypoxia-induced oxidative stress (Supplementary Fig. S16I). Altogether, these data indicate that these two genes might control common pathways that are critical for oxidative stress response and potentially underpin the onset of resistance to cytarabine in AML.

Indeed, cotreatment with the selective P2RY13 inhibitor MRS2211 led to a decrease in intracellular cAMP of MOLM14 AML cells in vitro (Fig. 6E), inhibited mitochondrial activity (measured as an increase in mitochondrial ATP production; Fig. 6F), and significantly enhanced cytarabine-dependent cytotoxic effects (Fig. 6G–I; Supplementary Fig. S16J). This partially phenocopied CD39 inhibition upon cytarabine treatment. In comparison, inhibiting adenosine purinergic signaling using an A2a adenosine receptor antagonist (ZM 241385) failed to enhance the induction of apoptosis in MOLM14 when combined with cytarabine (Supplementary Fig. S17A). Similar results were obtained by combining cytarabine treatment with an inhibitor of the 5′-ectonucleotidase CD73 (M3463, Supplementary Fig. S17B) responsible for the conversion of AMP to adenosine. Overall, these results confirm that CD39-mediated increase of extracellular nucleotide levels and the associated activation of purinergic receptor signaling underpin the activation of cAMP signaling and promote the survival of resistant AML cells upon chemotherapy treatment. In particular, our data reveal the particularly important role of CD39-mediated eADP generation and the consequent stimulation of the G protein–coupled P2RY13 receptor.

Our data suggest that inhibiting CD39 activity may be a promising therapeutic strategy to enhance chemotherapy response of AML cells in vivo. To test this hypothesis, we generated NSG mice-based CLDX and PDX models from AML cell lines and primary patient cells, respectively. We then tested the consequences of CD39 repression on the response of our preclinical models of AML to cytarabine using as alternative experimental strategies genetic invalidation and pharmacologic inhibition of our target. Genetic invalidation of CD39 was achieved in the MOLM14 CLDX model in vivo using two different doxycycline-inducible shRNAs specifically targeting the ENTPD1 gene. Remarkably, CD39 depletion in combination with cytarabine treatment resulted in a significant reduction of total cell tumor burden in the bone marrow of the mice three days after the end of the chemotherapy cycle (day 18) compared with the both vehicle-treated counterparts and the shCTL-xenografted mice (Fig. 7A and B). Moreover, although no change in AML viability and loss of MMP was detectable in control MOLM14 upon cytarabine treatment, concomitant repression of CD39 triggered a significant decrease in viability and loss of MMP in AML cells (Fig. 7C and D). Altogether, this led to an enhanced cytarabine sensitivity in vivo, as further demonstrated by a significant increase in the overall survival of cytarabine-treated shCD39-xenografted mice compared with both the vehicle-treated shCD39-xenografted mice cohort and the cytarabine-treated shCTL-xenografted mice (Fig. 7E).

Pharmacologic inhibition of CD39 activity was also achieved by administering the eATPase inhibitor POM1 for 6 days at the dose of 25 mg/kg/day alone and in combination with cytarabine at 60 mg/kg/day for 5 consecutive days in two different and independent PDX models (TUH06, TUH10). Response to single and combinatory treatments and various characteristics of RLCs were specifically monitored at day 15 (3 days after the last administration of the combinatorial treatment; Fig. 7F). Similar to our previous results, we observed an enhancement of the cytarabine cytotoxic effect in combination with POM1 administration in all the PDXs tested with a significant reduction of the total cell tumor burden and of AML cell viability in the bone marrow in vivo in the combination treatment group compared with both the single treatment groups (Fig. 7G and H; Supplementary Fig. S18A and S18B). We confirmed that POM1 efficiently blocked in vivo the CD39 eATPase activity induced upon cytarabine treatment in two different PDX models (TUH06, TUH10; Supplementary Fig. S18C) and induced a significant reduction of the mitochondrial mass (Fig. 7I), basal and maximal mitochondrial respiration, and global energetic capacity upon the combination treatment compared with cytarabine alone (Fig. 7J; Supplementary Fig. S18D–S18G).

Notably, we did not observe an enhancement of the cytotoxic effect of the chemotherapy on healthy hematopoietic cells and progenitors in the presence of the CD39 inhibitor POM1. Indeed, the mouse CD45-positive population in the bone marrow of animals xenografted with human AML treated with the POM1/cytarabine combination did not display any decrease in viability compared with the single treatment groups (Fig. 7F; Supplementary Fig. S18H). Similarly, the combination of the CD39 inhibitor POM1 with cytarabine did not increase the sensitivity ex vivo of normal human CD34⁺ hematopoietic progenitors to the chemotherapy treatment (Supplementary Fig. S18I).

Overall, our results show that inhibition of CD39 expression or activity substantially improves the response to cytarabine treatment in preclinical models of AML in vivo without significantly affecting the drug sensitivity of normal hematopoietic progenitors. Therefore, our findings strongly support the rationale for targeting CD39 as a valuable therapeutic strategy to enhance response to cytarabine in therapy-resistant AML. This should be assessed in a clinical study for AML treatment combining anti-CD39 small molecules with chemotherapy.

Poor overall survival is mainly due to frequent relapse caused by RLCs in patients with AML (1, 2). Although recent studies have highlighted new mechanisms of drug resistance in AML, especially in vivo (6, 29, 30), their clinical applications are still unresolved or under assessment. New therapies that specifically target and effectively eradicate RLCs represent an urgent medical need. In this work, we have identified the cell surface eATPase ENTPD1/CD39 and its downstream signaling pathway as a new critical and druggable target involved in the resistance to cytarabine in AML. We showed that CD39 was overexpressed in residual AML cells post chemotherapy from both 36 PDX models and 98 patients in the clinical setting. CD39 is also highly expressed in several human solid tumors, in which it was shown to actively contribute to cancer cell proliferation, dissemination, and metastatic process (31, 32). In the context of AML, our data support a model in which cytarabine treatment induces the selection and amplification of preexisting and intrinsically resistant leukemic cells that express and upregulate CD39 and its associated gene signature. We have furthermore observed that a drug-induced increase in CD39 expression is associated with a poor response to cytarabine in vivo, persistence of residual disease, and poor overall survival in patients with AML, especially in the younger subgroup of patients. Interestingly, our work indicates that the CD39 resistance–associated response is triggered in AML cells upon stress stimuli, such as chemotherapy, and this pathway is not necessarily constitutively activated in patients before treatment. Altogether, CD39 upregulation and downstream signaling activation play a critical role in the initial phase of responsive metabolic adaptation and regeneration of the leukemia upon chemotherapic stress. Thus, this new marker is relevant at the nadir of the disease after chemotherapy and may be useful for assessing and monitoring minimal residual disease in the clinics.

Many prosurvival and antiapoptotic signals are activated in AML by the stroma (33, 34). Nucleotides and nucleosides have emerged as important modulators of tumor biology. In particular, eATP and adenosine are major signaling molecules present in the tumor microenvironment. A growing body of evidence shows that when these molecules are released by cancer cells or surrounding tissues, they act as prometastatic factors, favoring tumor cell migration and tissue colonization. Interestingly, eATP elicits distinct responses in different tumor types, ranging from promoting cancer cell proliferation (35, 36) and metastasis (36, 37) to induction of cell death (38–43). In AML, eATP was reported to reduce human leukemia growth in vivo and enhance the antileukemic activity of cytarabine (22). The increase of CD39 expression occurred at early timepoints of the chemotherapeutic response and residual disease processes in PDXs and patients, when its activity is likely to be potentiated by the enrichment in eATP released from apoptotic and dying cells. Bone marrow microenvironment is a key regulator of leukemia growth and can provide a chemoprotective niche for AML cells (23, 44, 45). We and others have shown that mitochondrial OxPHOS is a crucial contributing factor of AML chemoresistance, and its inhibition sensitizes cells to cytarabine treatment (6, 7, 46). This is supported by an increase in respiratory substrate availability and in mitochondrial machinery transfer from the bone marrow–derived mesenchymal stem cells (47, 48). Here we unveil an additional mechanism of resistance supporting high mitochondrial OxPHOS activity in reponse to cytarabine treatment through the enhancement of mitochondrial biogenesis and function triggered by CD39 eATPase activity.

Our work indicates that this dynamic metabolic adaptation occurs in the context of a mitochondrial stress response that is orchestrated by CD39 and involves the downstream coordinated activation of cAMP–PKA and ATF4 signaling pathways. Altogether this leads not only to the regulation of mitochondrial OxPHOS activity and ROS production, but also to the concomitant enhancement of antioxidant defenses, sustaining a prosurvival response to stress in AML cells. Indeed, CD39 inhibition strongly impaired the cytarabine-triggered increase of mitochondrial mass and mtDNA, and the upregulation of two central transcriptional regulators of mitochondrial biogenesis, PGC1α and NRF1 (49, 50). Accordingly, cotreatment with the CD39 inhibitor POM1 abrogated the increase in OxPHOS activity and mitochondrial ROS production associated with cytarabine resistance. In parallel, our results show that CD39 contributes to the upregulation of key antioxidant genes that fine-tune ROS balance and buffer the increase of mitochondrial oxidants. As evidenced by our transcriptomic analysis and confirmed by experiments of reciprocal genetic or pharmacologic inactivation, cAMP–PKA signaling and ATF4 pathway have emerged to be key downstream effectors of this CD39-driven mitochondrial stress response to cytarabine. In accordance with our findings, ATF4 has recently been reported to be a master regulator of mitochondrial homeostasis and stress response in mammalian cells (50). However, our results suggest that the activation of cAMP–PKA downstream of CD39 may be the main pathway involved in the elevation of mitochondrial biogenesis and OxPHOS activity observed in the context of AML response to chemotherapy (51, 52). In addition, the early induction of PKA signaling in AML cells upon cytarabine exposure, as well as the downregulation of CD39 upon PKA inhibition, suggest that the cAMP–PKA pathway might be acting both upstream and downstream of CD39, creating a positive feedback loop, which controls and amplifies the CD39-driven intrinsic resistance pathway through adaptive stress responses in AML.

Previous studies have reported the pleiotropic roles of cAMP signaling and its major downstream effector PKA in different cancers, including AML. Perez and colleagues showed that cAMP efflux from the cytoplasm protects AML cells from apoptosis (53). Similarly, others reported cAMP-mediated protection of acute promyelocytic leukemia against anthracycline (54) or against arsenic trioxide-induced apoptosis (55). PKA, whose activation initiates an array of transcriptional cascades involved in the immune response, cell metabolism, and mitochondrial biogenesis, is one of the main and canonical downstream effectors of cAMP signaling. Intriguingly, cAMP–PKA signaling can be localized not only at the plasma membrane or nucleus, but also at the outer mitochondrial membrane or matrix (56). Mitochondrial cAMP signaling was shown to regulate cytosol–mitochondrial cross-talk, mitochondrial biogenesis and morphology, mitochondrial dynamics, mitochondrial membrane potential, TCA cycle activities, and ETC complexes in basal and stress conditions, such as starvation or hypoxia (57). Of note, cAMP signaling is activated as an integral part of the mitochondrial stress response that allows the rewiring of cellular metabolism in the presence of cellular damage and oxidative stress conditions. In this context, PGC1α and NRF1/2 activation leads to an increase in expression and assembly of respiratory chain supercomplexes and an increase in OxPHOS activity, allowing dynamic adaptation of mitochondrial functions to survive adverse conditions (58). These studies establish a mechanistic link between cAMP, PKA, and PGC1α in the regulation of mitochondrial biogenesis/function through the activation of mitochondrial stress response.

Our data indicate that the CD39-mediated generation of extracellular ADP and AMP plays an essential role in limiting cytarabine cytotoxic effects in AML cells. In particular, eADP and its stimulation of the ADP-responsive G protein–coupled P2RY13 purinergic receptor plays an important role in the intrinsic chemoresistance function of CD39 in AML. Indeed, AML cells concomitantly increase the expression of CD39 and P2RY13 upon both cytarabine-induced cytotoxic stress and hypoxic stress. Moreover, the gene signatures associated with these two markers strongly correlate in multiple AML patient datasets tested. This suggests these two genes might control overlapping pathways underpinning the adaptation of AML cells to stress and the onset of therapy resistance. Accordingly, treatment with P2RY13 inhibitor partially phenocopied the effect of CD39 inhibition in AML cells cotreated with cytarabine. Interestingly, P2RY13 was previously reported to be involved in the protection against oxidative stress in neurons, to protect cancer cells from genotoxic stress (59, 60), and to be a potential target in AML (61). Of note, high doses of ADP induce the coupling of this purinergic receptor with G_S proteins, leading to the stimulation of adenylate cyclase and cAMP signaling (53). Treatment with P2RY13 inhibitor of AML cells resulted in a significant reduction of intracellular cAMP and inhibition of mitochondrial OxPHOS activity. This suggests that P2RY13 may be involved in the activation of the cAMP–PKA signaling downstream CD39 in AML cells upon cytarabine treatment.

Collectively, our data support a model in which CD39 eATPase activity sustains an intrinsic mitochondrial stress response that, through the coordinated downstream activation of the P2RY13–cAMP–PKA pathway and the ATF4 axis, rewires the energetic metabolism of RLCs toward OxPHOS, and perserves their ROS homeostasis (Fig. 7K). We propose that eATP and CD39 are key actors in a novel AML cell-intrinsic mechanism of resistance to cytarabine and that targeting CD39 would be a promising therapeutic strategy to sensitize AML cells to cytarabine. In light of the recently recognized “immune checkpoint mediator” function of CD39 that interferes with antitumor immune responses, our data further suggest the existence of a critical cross-talk between AML cells and their immune and stromal microenvironment mediated by extracellular nucleotides and/or CD39 in the response to therapy of AML cells. In this context, blocking CD39 activity could have a double therapeutic benefit by both dampening the stress response and metabolic reprogramming supporting cytarabine cell-autonomous resistance and disrupting the immune escape mechanisms. In conclusion, our study uncovers a noncanonical role of CD39 on AML resistance (Fig. 7K) and provides a strong scientific rationale for testing CD39 blockade strategies in combination with cytarabine treatment in clinical trials for patients with AML. Because CD39-blocking mAbs (20, 62, 63) are already in clinical trials as a single agent and in combination with an approved anti–PD-1 immunotherapy or standard chemotherapies for patients with lymphoma or solid-tumor malignancies, we expect that these findings have the potential for rapid translation into the clinic of our proposed combination therapy targeting CD39 for patients with chemotherapy-resistant AML.

Primary AML patient specimens are from Toulouse University Hospital (TUH; Toulouse, France). Frozen samples were obtained from patients diagnosed with AML at TUH after written informed consent in accordance with the Declaration of Helsinki, and stored at the HIMIP collection (BB-0033-00060). According to the French law, HIMIP biobank collections have been declared to the Ministry of Higher Education and Research (DC 2008-307 collection 1) and obtained a transfer agreement (AC 2008-129) after approval by the “Comité de Protection des Personnes Sud-Ouest et Outremer II” (ethical committee). Clinical and biological annotations of the samples have been declared to the CNIL (“Comité National Informatique et Libertés”; i.e., “Data processing and Liberties National Committee”). Peripheral blood and bone marrow samples were frozen in FCS with 10% DMSO and stored in liquid nitrogen. The percentage of blasts was determined by flow cytometry and morphologic characteristics before purification.

CD34-positive cells were obtained after sorting from umbilical cord blood samples (Etablissement Français du Sang, Besançon, France). Clinical and biological annotations of the samples have been declared to the CNIL. Briefly, mononuclear cells were separated by Ficoll–Hypaque density gradient centrifugation and incubated in RBC lysis buffer (ammonium chloride solution) to remove red blood cells. CD34⁺ cell isolation was performed using manufacturer's instructions (EasySepTM, STEMCELL). CD34⁺ samples were maintained in Iscove's Modified Dulbecco's Medium supplemented with 20% FCS.

Animals were used in accordance with a protocol reviewed and approved by the Institutional Animal Care and Use Committee of Région Midi-Pyrénées (France). NOD/LtSz-scid/IL-2Rγchain^null (NSG) mice were produced at the Genotoul Anexplo platform at Toulouse (France) using breeders obtained from Charles River Laboratory. Mice were housed and human primary AML cells were transplanted as reported previously (64–66). Briefly, mice were housed in sterile conditions using HEPA filtered microisolators and fed with irradiated food and sterile water. Transplanted mice were treated with antibiotic (baytril) for the duration of the experiment. Mice (6–9 weeks old) were sublethally treated with busulfan (20 mg/kg/day) 24 hours before injection of leukemic cells. Leukemia samples were thawed at room temperature, washed twice in PBS, and suspended in Hank's Balanced Salt Solution at a final concentration of 1–10 million cells per 200 μL of Hank's Balanced Salt Solution per mouse for tail-vein injection. Daily monitoring of mice for symptoms of disease (ruffled coat, hunched back, weakness, and reduced mobility) determined the time of killing for injected animals with signs of distress. If no signs of distress were seen, mice were initially analyzed for engraftment 8 weeks after injection except where otherwise noted.

Eight to 18 weeks after AML cell transplantation and successful engraftment in the mice (tested by flow cytometry on peripheral blood or bone marrow aspirates), NSG mice were treated by daily intraperitoneal injection for 5 days of 30 (for CLDX models) and 60 (for PDX models) mg/kg cytarabine, kindly provided by the Pharmacy of the TUH. For control, NSG mice were treated daily with intraperitoneal injection of vehicle, PBS 1×. Mice were monitored for toxicity and provided nutritional supplements as needed.

POM1 or ARL67156 was administered to xenografted mice by intraperitoneal injection every other day for two weeks. The time of dissection was fifteen days after the last dose of POM1 (or ARL67156) or 8 days for cytarabine, two days after the last dose of each treatment.

NSG mice were humanely killed in accordance with European ethics protocols. Bone marrow (mixed from tibias and femurs) and spleen were dissected in a sterile environment and flushed in Hank's Balanced Salt Solution with 1% FBS, washed in PBS, and dissociated into single-cell suspensions for analysis by flow cytometry of human leukemic cell engraftment and bone marrow cell tumor burden. Mononuclear cells from peripheral blood, bone marrow, and spleen were labeled with FITC-conjugated anti-hCD3, PE-conjugated anti-hCD33, PerCP-Cy5.5–conjugated anti-mCD45.1, APCH7-conjugated anti-hCD45, and PeCy7-conjugated anti-hCD44 (all antibodies from Becton Dickinson, except FITC-conjugated anti-hCD3 from Ozyme Biolegend) to determine the fraction of human blasts (hCD45⁺mCD45.1⁻hCD33⁺hCD44⁺ cells) using flow cytometry. Analyses were performed on a Life Science Research II (LSR II) flow cytometer with DIVA software (BD Biosciences) or Cytoflex flow cytometer with CytoExpert software (Beckman Coulter). The number of AML cells/μL peripheral blood and the cumulative number of AML cells in bone marrow and spleen (total tumor cell burden) were determined using CountBright beads (Invitrogen) following previously described protocols (6, 66).

Human AML cell lines were maintained in RPMI 1640 supplemented with 10% FBS (Invitrogen) in the presence of 100 U/mL of penicillin and 100 μg/mL of streptomycin, and were incubated at 37°C with 5% CO₂. The cultured cells were split every 2 to 3 days and maintained in an exponential growth phase. U937 was obtained from the DSMZ in February 2012 and from the ATCC in January 2014. MV4-11 and HL-60 were obtained from the DSMZ in February 2012 and February 2016. KG1 was obtained from the DSMZ in February 2012 and from the ATCC in March 2013. KG1a was obtained from the DSMZ in February 2016. MOLM14 was obtained from Pr. Martin Carroll (University of Pennsylvania, Philadelphia, PA) in 2011 and from the DSMZ in June 2015. DSMZ and ATCC cell banks provided authenticated cell lines by cytochrome C oxidase I gene (COI) analysis and short tandem repeat (STR) profiling. Furthermore, the mutation status was also verified by targeted resequencing of a panel of 40 genes frequently mutated in AML as described in the Supplementary Methods. Clinical and mutational features of our AML cell lines are described in Supplementary Table S1. These cell lines have been routinely tested for Mycoplasma contamination in the laboratory.

Cytarabine was provided by the Pharmacy of TUH. 3-Isobutyl-1-methylxanthine (IBMX), rotenone (Rot), antimycin A (AntiA), oligomycin (oligo), carbonyl cyanide 3-chlorophenylhydrazone (CCCP), sodium iodoacetate (iodo), ATP, ADP, AMP, M3763, ZM241385, and busulfan were obtained from Sigma-Aldrich. H89 and PKA Inhibitor 14–22 Amide (PKAi) were provided by Medchem Express and Calbiochem, respectively. ARL67156 (also referred to as ARL), POM1, and MRS2211 were shipped from Santa Cruz Biotechnology. 8-Bromo-cAMP and Forskolin were provided by Tocris Bioscience.

FITC-conjugated CD64 was provided by Beckman Coulter, and FITC-conjugated CD19, FITC-conjugated CD3, PE-Cy7-conjugated anti-hCD38, Alexa Fluor 700-conjugated anti-hCD34, APC-H7-conjugated anti-hCD45, PE-conjugated anti-hCD33, APC-conjugated anti-hCD39, PERCP-Cy5.5-conjugated anti-mCD45.1, V450-conjugated anti-hCD45 FITC-conjugated Annexin V, and BV421-conjugated Annexin V were shipped from BD Biosciences.

Cells were stained during 30 minutes with anti-P2RY13 antibody (LSBIO, C120934), washed in PBS, and stained for 20 minutes with a secondary antibody–conjugated Alexa Fluor 488.

5 × 10⁴ cells treated 48 hours with different conditions (ARL 100 μmol/L or POM1 100 μmol/L) were washed and plated on 96-well plates. Plates were incubated 30 minutes at 37°C with or without ATP (10 mmol/L). The concentration of "unhydrolyzed" extracellular ATP was determined using the ATPlite luminescence ATP DetectionAssaySystem (PerkinElmer) according to the manufacturer's instructions.

Supernatants from culture were prepared as follows: 5 × 10⁶ cells treated 24 hours with different conditions (AraC 2 μmol/L, POM1 100 μmol/L) were washed and plated on 24-well plates. Plates were incubated 1 hour at 37°C with or without ATP (10 μmol/L). The plate was centrifuged (1,500 × g, −4°C, 3 minutes) and the supernatant was collected and stored at −80°C until analysis. To get highly accurate quantification, the Isotope-Dilution Mass Spectrometry (IDMS) method was applied (67). For this purpose, the day of the analysis 50 μL of IDMS standard was added to 50 μL of each sample previously diluted 2 or 20 times with water.

Analysis of ATP, ADP, and AMP supernatants was performed by high-performance anion exchange chromatography (Dionex ICS 5000+ system) coupled to a LTQ Orbitrap Velos mass spectrometer (Thermo Fisher Scientific) equipped with a heated ESI probe. All samples were analyzed in FTMS mode at a resolution of 60,000 (at m/z 400) with the following source parameters: capillary temperature was 350°C, source heater temperature was 350 °C, sheath gas flow rate was 50, auxiliary gas flow rate was 5, S-Lens RF level was 60%, and source voltage was 2.7 kV. The injection volume was 15 μL. Samples were injected on a Dionex IonPac AS11 column (250 × 2 mm) equipped with a Dionex AG11 guard column (50 × 2 mm). Mobile phase was composed of a KOH gradient which varied as follows: 0 min 0.5; 1 min 0.5; 9.5 min 4.1; 14.6 min 4.1; 24 min 9.65; 36 min 60; 36.1 min 90; 43 min 90; 43.1 min 0.5; 45 min 0.5. Data treatment was done with TraceFinder software (Thermo Fisher Scientific).

AML cell lines were treated with medium alone or treated with ARL (100 μmol/L) or POM1 (100 μmol/L) for 6 hours; also MOLM14 shCD39 cells were treated with doxycycline in vitro. The concentration of cAMP was determined using the cAMP-Glo Assay (Promega) according to the manufacturer's instructions.

Quantification of cAMP levels was also performed using the homogeneous time-resolved fluorescence assay (HTRF) and dynamic2 cAMP kit (Cisbio). This is based on a competitive immunoassay using cryptate-labeled antibodies anti-cAMP and d2-labeled cAMP. For that purpose, M14 cells were washed and resuspended in SB1X with 0.5 mmol/L IBMX. Then, 10,000 cells/well (384-well plate) were plated and stimulated for 3 hours at room temperature with 10 μmol/L MRS2211 and 1 μmol/L cytarabine. Cells were then lysed using 5 μL of lysis buffer containing d2-labeled cAMP and 5 μL of cryptate-labeled anti-cAMP. The signal was measured after 1 hour at room temperature incubation using Artemis HTRF microplate reader. The RET signal was calculated by the ratio of d2-cAMP/cAMP Cryptate (665 nm/620 nm), the specific signal being inversely proportional to the concentration of cAMP in the sample. For each experiment, a calibration curve was established with cAMP standards allowing the quantification of cAMP levels by linear regression.

We assessed the statistical analysis of the difference between two sets of data using nonparametric Mann–Whitney test one-way or two-way (GraphPad Prism). The Mantel–Cox log-rank test was used for statistical assessment of survival. P values of less than 0.05 were considered to be significant (*, P < 0.05; **, P < 0.01 and ***, P < 0.001).

For in vitro and in vivo analyses of cytarabine residual disease and CD39 studies, see Supplementary Methods. RNA-sequencing data are available at the Gene Expression Omnibus under the accession number GSE136551.

N. Bonnefoy reports personal fees from OREGA Biotech (shareholder) outside the submitted work; in addition, N. Bonnefoy has a patent for WO2009/095478 issued and licensed to OREGA Biotech. C. Recher reports grants and personal fees from AbbVie, Astellas, Amgen, Celgene, Jazz Pharmaceuticals, Agios, Novartis, Roche, and Daiichi-Sankyo; grants from Maat Pharma and Chugai; and personal fees from Macrogenics and Pfizer outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

N. Aroua: Formal analysis, validation, methodology. E. Boet: Formal analysis, validation, methodology. M. Ghisi: Conceptualization, formal analysis, validation, writing-review and editing. M.-L. Nicolau-Travers: Formal analysis, methodology. E. Saland: Formal analysis, validation, methodology. R. Gwilliam: Formal analysis, validation, methodology. F. de Toni: Formal analysis, validation, methodology. M. Hosseini: Formal analysis, methodology. P.-L. Mouchel: Formal analysis, validation, methodology. T. Farge: Formal analysis, validation, methodology. C. Bosc: Formal analysis, validation, methodology. L. Stuani: Formal analysis, validation, methodology. M. Sabatier: Formal analysis, validation, methodology. F. Mazed: Formal analysis, methodology. C. Larrue: Formal analysis, validation, methodology. L. Jarrou: Resources. S. Gandarillas: Resources. M. Bardotti: Resources. M. Picard: Resources. C. Syrykh: Formal analysis, validation, methodology. C. Laurent: Formal analysis, validation, methodology. M. Gotanegre: Resources, formal analysis, validation, methodology. N. Bonnefoy: Resources, formal analysis, validation, methodology. F. Bellvert: Resources. J.-C. Portais: Resources. N. Nicot: Resources, formal analysis. F. Azuaje: Resources. T. Kaoma: Software, formal analysis, validation, methodology. C. Joffre: Formal analysis, validation. J. Tamburini: Resources, formal analysis, validation. C. Recher: Resources, investigation. F. Vergez: resources, data curation, formal analysis, supervision, visualization. J.-E. Sarry: conceptualization, supervision, funding acquisition, investigation, writing-original draft, project administration, writing-review and editing.

We thank all members of mice core facilities (UMS006, ANEXPLO, Inserm, Toulouse), in particular Marie Lulka, Christine Campi, Pauline Challies, Pauline Debas, and Yara Bareira for their support and technical assistance, Cécile Déjou (Institut de Recherche en Cancérologie de Montpellier) for her help with CD39 assays, Dr. Véronique Pons and Céline Galès (Institut des Maladies Métaboliques et Cardiovasculaires de Toulouse) and Prof. Véronique De Mas and Eric Delabesse for the management of the Biobank BRC-HIMIP (Biological Resources Centres-INSERM Midi-Pyrénées “Cytothèque des hémopathies malignes”) that is supported by CAPTOR (Cancer Pharmacology of Toulouse-Oncopole and Région). We thank Anne-Marie Benot, Muriel Serthelon, and Stéphanie Nevouet for their daily help with the administrative and financial management of our team. We also thank the patients and the Association GAEL for their generous support. The authors also thank Dr. Mary Selak and Laetitia Linares (Institut de Recherche en Cancérologie de Montpellier) for discussions and critical reading of the manuscript. This work was also supported by grants from the Cancéropole GSO (Projet Emergence 2014-E07, to J.-E. Sarry), Région Midi-Pyrénées/Occitanie (to J.-E. Sarry), the Programme “Investissement d'Avenir” PSPC (IMODI; to J.-E. Sarry), the Laboratoire d'Excellence Toulouse Cancer (TOUCAN and TOUCAN2.0; contract ANR11-LABEX, to J.-E. Sarry), the Programme Hospitalo-Universitaire en Cancérologie (CAPTOR; contract ANR11-PHUC0001, to C. Récher), Fondation Toulouse Cancer Santé (to J.-E. Sarry), Plan Cancer 2014-BioSys (FLEXAML; to J.-E. Sarry). N. Aroua, E. Boet, and M. Ghisi have a fellowship from the Fondation ARC, Fondation Toulouse Cancer Santé, and Fondation de France, respectively.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.