Cytoprotective autophagy maintains leukemia-initiating cells in murine myeloid leukemia

The prognosis of acute myeloid leukemia (AML) remains dismal in the majority of patients even after recent advances in the pathophysiology in AML. One reason for this is the acquisition of resistance to key chemotherapeutic drugs such as cytarabine (AraC) and anthracyclines, resulting in relapse, which is mediated by a small subset of self-renewing cells called leukemia-initiating cells (LICs). Thus, further studies on the biological features of LICs and their mechanisms of drug resistance are essential to treat AML patients.

Macroautophagy (hereafter referred to as autophagy) is a process used for the degradation of cytosolic macromolecules by lysosomal enzymes.^1,2  Autophagy has various physiological roles through the maintenance of cellular homeostasis by degrading damaged or excess cytosolic components and supplying nutrients to starved cells.^3,4  Through autophagic processes, double membrane vesicles, called autophagosomes, engulf cytosolic components and transport their cargo to lysosomes. A group of proteins including autophagy–related 5 (ATG5) and ATG7 are indispensable for autophagy.^1,5,6 

Based on the complicated relationship between autophagy and tumors, it is unknown if inhibition of autophagy could be a beneficial approach to treat cancer.^7,8  In general, autophagy contributes to cell survival in established tumors.^9,10  Accordingly, inactivation of autophagy, through Atg3 deletion, in breakpoint cluster region–Abelson murine leukemia viral oncogene homolog 1 (BCR-ABL1)–transduced murine bone marrow (BM) cells diminishes their capacity to produce chronic myeloid leukemia (CML) in transplanted mice because of p53-dependent growth arrest.¹¹  As such, the efficacy of the autophagy inhibitor chloroquine and its derivatives is currently being tested in clinical trials for the treatment of various types of cancer.¹²  In contrast, it has also been reported that autophagy genes function as tumor suppressors in normal tissues.¹³  Additionally, even in established tumors, it has been reported that inhibition of autophagy promotes cancer progression in a murine TP53-deficient pancreatic cancer model and in EGFR-mutated lung cancer.^14,15  Considering these facts, the dependency on autophagy could be strongly affected by biological context such as gene mutations and physiological conditions of the tumor microenvironment.

In addition, its association with drug resistance makes autophagy an attractive drug target. Autophagy is recognized as a nonapoptotic mechanism of programmed cell death and is required for cell death after chemotherapy in some cases.^16,17  However, combination treatment using an autophagy inhibitor with chemotherapy, irradiation, or targeted therapy has demonstrated efficacy in various cancers including hematological malignancies.^8,18,19  Considering the fact that the function of autophagy is context-dependent, it is worthwhile to evaluate the impact of inhibition of autophagy in vivo in different types of tumors, using animal models that can recapitulate physiological conditions and tumor microenvironments.

The role of autophagy in AML progression, including LIC maintenance, is not yet fully understood. Moreover, the exact function of autophagy in drug resistance in vivo needs to be elucidated; however, some studies have revealed that autophagy is required for drug resistance.^20-24  In the present study, to investigate the role of autophagy in AML progression and drug resistance in vivo, we inhibited autophagy in murine mixed lineage leukemia–eleven nineteen leukemia (MLL-ENL) leukemia and CML blast crisis (CML-BC).

Atg5^flox/flox mice,²⁵ Atg7^flox/flox mice,⁵  and GFP-LC3 mice²⁶  were obtained from RIKEN BRC (Tsukuba, Japan). Cre-ER^T2 mice²⁷  were obtained from JAX (Bar Harbor, ME). Atg5^flox/flox or Atg7^flox/flox mice were crossed with Cre-ER^T2 mice. Genotyping was performed with primers shown in supplemental Table 1 (available on the Blood Web site). Wild-type C57BL/6J (8- to 12-week-old) mice were purchased from Sankyo Laboratory Service (Tokyo, Japan). These mice were kept at the Center for Disease Biology and Integrative Medicine at the University of Tokyo, according to institutional guidelines. All animal experiments were approved by the Institutional Animal Care and Use Committee of the University of Tokyo.

The procedure for developing murine bone marrow transplantation (BMT) leukemia models was derived from previous reports.^28,29  For producing retrovirus, MSCV-MLL-ENL-IRES-GFP, MSCV-MLL-ENL-IRES-Puro,²⁸  pGCDNsam-BCR-ABL1-IRES-GFP, or pGCDNsam-NUP98-HOXA9-IRES-KusabiraOrange were transfected into Plat-E packaging cells³⁰  using polyethylenimine.³¹  The detailed protocol is described in the supplemental Methods.

For excising the loxP-flanked region in vivo, 1 mg/mouse tamoxifen (TAM; Cayman Chemical Company, Ann Arbor, MI) dissolved in peanut oil (Sigma, St. Louis, MO) was intraperitoneally administrated once daily for 5 consecutive days. In leukemia mouse models, TAM treatment started from day 7 after BMT. AraC (Tokyo Chemical Industry, Tokyo, Japan) was dissolved in phosphate-buffered saline (PBS) and intraperitoneally administrated to leukemic mice at a dose of 1 mg/mouse once daily.

For staining, isolated mononulcear cells were incubated with 3% fetal calf serum/PBS containing diluted antibody as listed in supplemental Table 2. Stained cells were washed once with 3% fetal calf serum/PBS and analyzed. For apoptosis analysis, stained cells were washed once and subsequently incubated with annexin-V-allophycocyanin (BioLegend, San Diego, CA) and 4′,6-diamidino-2-phenylindole (DAPI; 1 µg/mL) in accordance with the manufacturers’ protocols. Mitochondrial activity was detected by MitoTrackerOrange CMTMRos (Life Technologies, Waltham, MA), and reactive oxygen species (ROS) levels were determined by CellRox DeepRed (Life Technologies). Cells (2 × 10⁵) that were stained with antibodies against surface markers were washed once and incubated with 1× Hanks balanced salt solution containing a 1:2000 dilution of MitoTracker or a 1:500 dilution of CellRox at 37°C for 15 minutes. Stained cells were analyzed using an LSRII (BD Biosciences, San Jose, CA) and sorted with a BD FACSAria (BD Biosciences). Data were analyzed by FlowJo software (TreeStar, Ashland, OR).

Sorted c-Kit⁻CD11b⁺ BM or peripheral blood (PB) cells from MLL-ENL leukemic mice were fixed with 2% paraformaldehyde and 2% glutaraldehyde in 50 mM phosphate buffer (pH 7.4) for 10 minutes at 4°C. The buffer was then replaced with 2% glutaraldehyde in 0.1 M phosphate buffer (pH 7.4) and incubated at 4°C overnight. Thereafter, sample preparation and imaging were conducted at Tokai Electron Microscopy Inc. (Nagoya, Japan).

P values, for comparison of 2 different groups, were calculated using an unpaired 2-tailed Student t test unless otherwise mentioned. LIC frequency was assessed using the Poisson distribution. To analyze the survival curves, a log-rank test was used. P values <.05 were considered significant. Data analysis was performed using R software (http://www.R-project.org).

To investigate the importance of autophagy in AML progression, we developed a leukemia mouse model through BMT with Atg5- or Atg7-floxed BM cells transduced with MLL-ENL. To delete the Atg5 or Atg7 gene in vivo, Atg5^flox/flox or Atg7^flox/flox mice were crossed with TAM-inducible Cre-ER^T2 mice.²⁷  Green fluorescent protein (GFP)-labeled MLL-ENL-driven AML cells, from primary recipients, were serially transplanted into secondary recipients, and secondary recipients received TAM treatment to delete Atg5 or Atg7 (supplemental Figure 1A). Genomic polymerase chain reaction and western blots of BM cells from TAM-treated, Atg5^flox/flox:Cre-ER^T2 or Atg7^flox/flox:Cre-ER^T2 MLL-ENL leukemic mice confirmed the in vivo deletion of Atg5 or Atg7 (supplemental Figure 2A; Figure 1A). Unconjugated microtubule-associated protein 1A/1B-light chain 3A (LC3A, or LC3A-I) covalently binds to phosphatidylethanolamine to form LC3A-II to induce autophagy through associations with several ATG molecules including ATG5 and ATG7.¹  Thus, LC3A-II is commonly used as an indicator of autophagy. We confirmed diminished conversion of LC3A-I to LC3A-II in TAM-treated Atg5^Δ/Δ and Atg7^Δ/Δ leukemic cells, revealing that genetic deletion of Atg5 or Atg7 can efficiently suppress autophagy in MLL-ENL leukemic cells (Figure 1A). In terms of disease progression in MLL-ENL mice, Atg5^Δ/Δ and Atg7^Δ/Δ animals displayed significantly longer survival than Atg5^flox/flox and Atg7^flox/flox controls (Atg5^Δ/Δ: median = 29 days vs Atg5^flox/flox: median = 25 days; Atg7^Δ/Δ: median = 23 days vs Atg7^flox/flox: median = 20 days) (Figure 1B-C). Atg7^Δ/Δ MLL-ENL mice showed lower BM cellularity than control mice (Figure 1D), although the frequency of GFP⁺ MLL-ENL cells in BM and spleen weight were not different between vehicle- and TAM-treated mice (supplemental Figure 2B-C).

To further examine the role of autophagy in MLL-ENL leukemic mice, LIC frequency was evaluated in Atg7-deficient leukemic mice; MLL-ENL LICs are enriched in lineage⁻c-Kit⁺Sca1⁻CD34⁺CD16/32⁺ (L-GMP) or lineage⁻c-Kit⁺ (LK) cells.^32,33  Flow cytometric analysis revealed that LK and L-GMP frequency were decreased in Atg7-deleted leukemic mice (Figure 2A; supplemental Figure 3A-B). Additionally, a similar reduction in LICs was observed in Atg5^Δ/Δ leukemic mice, suggesting that the decreased frequency of LICs was caused by autophagy-related mechanisms (Figure 2B; supplemental Figure 3C). Hereafter, we mainly focused on the Atg7 conditional knockout model to study the role of autophagy, as phenotypes were identical after knocking out Atg5 or Atg7. In normal hematopoietic stem cells (HSCs), an Atg7 deletion causes accumulation and increased activity of mitochondria, resulting in increased ROS production.^34,35  Similarly, Atg7^Δ/Δ MLL-ENL LK cells demonstrated enhanced mitochondrial activity and increased ROS production compared with those of their Atg7^flox/flox counterpart (Figure 2C-F). In lineage⁺ cells, a population rich in differentiated cells, the Atg7 deletion had no effect on mitochondrial activity or ROS production (Figure 2C-F). When we assessed the apoptotic status of Atg7^Δ/Δ and Atg7^flox/flox MLL-ENL mice, annexin-V⁺/DAPI⁺ dead cells increased only in the Atg7^Δ/Δ LK fraction (Figure 2G; supplemental Figure 3D). We confirmed that cell death in the LK fraction was also induced in TAM-treated Atg5^flox/flox MLL-ENL leukemic mice (Figure 2H).

To exclude the possibility that TAM treatment alone or activation of Cre-estrogen receptor fusion protein (Cre-ER) could cause artificial effects in hematopoietic cells,^36,37  we used Cre-ER negative (Cre⁻) Atg7^flox/flox leukemic mice and Cre-ER wild-type leukemic mice as controls. TAM treatment did not prolong survival in Atg5^flox/flox:Cre⁻ or Atg7^flox/flox:Cre⁻ leukemic mice and did not reduce the frequency of L-GMPs in Atg7^flox/flox:Cre⁻ leukemic mice (supplemental Figure 4A-C). Moreover, the frequency of L-GMP cells in Cre-ER wild-type leukemic mice did not change after TAM treatment (supplemental Figure 4D).

These results suggest that autophagy is required for LIC maintenance by preventing cell death associated with accumulation of mitochondria and induced ROS production.

To characterize the defects of LICs in autophagy-deficient leukemic mice, functional analyses of Atg7^Δ/Δ leukemic cells were performed. GFP⁺Atg7^Δ/Δ BM cells had decreased in vitro colony-forming capacity compared with GFP⁺Atg7^flox/flox BM cells (supplemental Figure 5A-B). When sorted GFP⁺ BM cells from Atg7^flox/flox or Atg7^Δ/Δ MLL-ENL leukemic mice were serially transplanted into tertiary recipient mice, it was confirmed that the Atg7 deletion had no impact on homing ability (Figure 3A-B; supplemental Figure 1B). Limiting dilution transplantation analysis revealed that the frequency of functional LICs decreased in autophagy-inactivated Atg7^Δ/Δ mice (Atg7^flox/flox vs Atg7^Δ/Δ, 1:46 vs 1:301, respectively) (Figure 3C-D). Taken together, Atg7 is required for maintenance of functional LICs in MLL-ENL leukemia.

Intriguingly, both Atg5^Δ/Δ and Atg7^Δ/Δ MLL-ENL leukemic mice had significantly lower peripheral white blood cell (WBC) counts compared with those of their respective counterparts (Figure 4A-B). This contrasted the minimal contribution of Atg5/Atg7 deletion in reducing AML burden in BM (Figure 1D; supplemental Figure 2B-C). These results led us to hypothesize that various extracellular factors in PB can result in autophagy dependence in peripheral leukemia cells. We thus evaluated the apoptotic status of PB Atg5^Δ/Δ and Atg7^Δ/Δ MLL-ENL leukemia cells. To minimize possible bias derived from differential frequencies of c-Kit⁺ cells between BM and PB, CD11b⁺ cells, in which MLL-ENL (GFP⁺) leukemia cells accumulate in both BM and PB, were subdivided into c-Kit⁻ and c-Kit⁺ fractions for further analysis (supplemental Figure 6A). Strikingly, a TAM-induced Atg5/Atg7 deletion caused increased apoptosis only in PB CD11b⁺ leukemia cells and not in similar cells of BM origin, irrespective of c-Kit expression (Figure 4B-C; supplemental Figure 6B).

To further analyze autophagy-related functions in PB MLL-ENL leukemia cells, in vitro colony formation assays and in vivo serial BMT assays with PB leukemia cells were performed. PB Atg7^Δ/Δ leukemia cells showed decreased colony forming capacity in vitro (Figure 4D), and recipients of PB Atg7^Δ/Δ leukemia cells displayed marginal but significantly prolonged survival in vivo (Figure 4E; supplemental Figure 1C). This demonstrated that PB MLL-ENL leukemia cells also had leukemia-initiating capacity, which was partly regulated by autophagy.

Additionally, we evaluated the apoptotic status of PB cells from Atg7^flox/flox:Cre⁻ or Cre-ER leukemic mice with wild-type Atg genes as controls. TAM treatment did not cause reduction in WBCs or induction of apoptosis in PB. Therefore, the possibility of a cytotoxic effect on PB leukemic cells mediated by the Cre-ER system was excluded (supplemental Figure 4E-I).

Taken together, autophagy protects PB leukemia cells from apoptosis irrespective of their immunophenotypic maturity.

We next investigated whether PB leukemia cells have enhanced autophagic activity. For this, BM and PB c-Kit⁻CD11b⁺ MLL-ENL leukemia cells were subjected to western blotting to quantify LC3A conversion. Protein analysis revealed that PB leukemia cells had a higher level of LC3A-II than BM leukemia cells, indicating that PB leukemia cells were rich in autophagosomes (Figure 5A-B). In addition, electron microscopic analysis confirmed that PB leukemia cells contained an increased number of autophagosomes compared with BM cells (Figure 5C-D). We next used GFP-LC3 transgenic mice to evaluate autophagy.²⁶  When autophagy is activated in GFP-LC3 mice, GFP-LC3 is degraded resulting in decreased GFP intensity.³⁸  We developed GFP-LC3 transgenic MLL-ENL leukemic mice (supplemental Figure 1D), and GFP expression in BM and PB leukemia cells was measured by flow cytometry (supplemental Figure 7A). In both c-Kit⁻CD11b⁺ and c-Kit⁺CD11b⁺ fractions, GFP expression in PB leukemia cells was decreased compared with that in BM leukemia cells (Figure 5E-F). To examine if the accumulation of autophagosomes was cell autonomous, we performed ex vivo culture of BM and PB leukemia cells. The enhanced LC3A-II expression in PB cells that was detected in vivo immediately disappeared when BM and PB c-Kit⁻CD11b⁺ cells from MLL-ENL leukemic mice were cultured (supplemental Figure 8A). Moreover, the difference in GFP-LC3 expression between BM and PB was maintained in leukemic mice after BMT of PB leukemic cells that expressed lower levels of GFP-LC3 (supplemental Figure 8B-C). These results suggest that the PB environment might enhance autophagy in PB leukemia cells.

The importance of autophagy in hematopoietic cells has been emphasized by several prior reports.^39-42  Considering the enhanced dependency of PB MLL-ENL leukemia cells on autophagy, we next elucidated the role of autophagy in normal PB myeloid cells. For this, PB cells obtained from vehicle- or TAM-treated MLL-ENL-free Atg7^flox/flox:Cre-ER^T2 mice were analyzed. Atg7^Δ/Δ mice showed no change in PB WBC counts even after TAM administration (supplemental Figure 9A-B). Additionally, an Atg7 deletion had no impact on apoptosis in PB c-Kit⁻CD11b⁺ or c-Kit⁺CD11b⁺ cells (supplemental Figure 9C). Moreover, normal c-Kit⁻CD11b⁺ cells of the PB exhibited lower expression of LC3A-II and displayed enhanced expression of GFP-LC3 compared with those of their counterparts in BM (supplemental Figure 9D-E). Therefore, normal BM or PB myeloid cells have no dependency on autophagy for at least short-term survival, suggesting that enhanced autophagy dependency in PB is specific for MLL-ENL leukemia cells.

To investigate the role of autophagy in other AML models, we used BCR-ABL1 and NUP98-HOXA9–driven murine CML-BC models (supplementary Figure 1E).⁴³  Similar to the MLL-ENL model, Atg7^Δ/Δ CML-BC mice survived longer with no reduction in tumor burden in the BM or spleen (Figure 6A; supplemental Figure 10A-B). Moreover, Atg7^Δ/Δ CML-BC mice showed a significantly diminished LK fraction, which contains LICs in this model⁴⁴  (Figure 6B; supplemental Figure 10C). In addition, deletion of Atg7 reduced the WBC count and induced apoptosis specifically in PB leukemia cells, which was similar to results observed for MLL-ENL leukemia (Figure 6C-D). The BM of deceased or moribund CML-BC mice was occupied by myeloblasts irrespective of Atg7 deletion, suggesting leukemia development in these mice (supplemental Figure 10D). These results suggest that the leukemia promoting and PB cell supportive roles of autophagy are not restricted to MLL-ENL leukemia but are also present in other types of myeloid leukemia.

Although the contribution of autophagy to drug resistance in AML has been investigated by in vitro culture, the exact role of autophagy in drug resistance in vivo remains elusive. Hence, the effect of autophagy inactivation in combination with AraC, one of the most common clinically used chemotherapeutic agents for AML, was investigated in MLL-ENL leukemic mice. First, the expression of GFP-LC3 in MLL-ENL leukemia cells after AraC treatment was measured. AraC treatment reduced GFP-LC3 expression in L-GMP cells but not in lineage⁺ cells (supplemental Figure 11A-C). The overall survival of Atg7^flox/flox:Cre-ER^T2 MLL-ENL mice was enhanced in the cohort treated with TAM and AraC, indicating that Atg7 deletion potentiated the cytotoxic effect of AraC (Figure 7A). Atg7 deletion showed an additive effect with AraC treatment on BM cell reduction (supplemental Figure 11D). Furthermore, Atg7 deletion enhanced the efficacy of AraC in reducing the frequency and the absolute number of LICs (Figure 7B; supplemental Figure 11E). When the effect of TAM plus AraC therapy was assessed in PB from Atg7^flox/flox:Cre-ER^T2 MLL-ENL mice, WBC counts in MLL-ENL mice treated with TAM and AraC were prominently reduced and were comparable to those of mice treated with AraC alone (Figure 7C). Of note, AraC treatment combined with Atg7 deletion resulted in increased apoptosis in PB MLL-ENL leukemia cells compared with AraC treatment or Atg7 deletion alone; however, the combinatorial effect on apoptosis in BM leukemia cells was not significant (Figure 7D; supplemental Figure 11F). In deceased or moribund TAM- and AraC-treated mice, the BM was occupied by myeloblasts similar to that observed in control mice, suggesting that these mice developed leukemia before death (supplemental Figure 11G). Altogether, Atg7 deletion potentiated the efficacy of AraC treatment, and especially the capacity to target LICs, which highlights the important role of autophagy on AraC resistance in MLL-ENL leukemia.

In this study, we evaluated the role of autophagy in leukemia progression, LIC maintenance, and drug resistance in vivo. Our results revealed that in the BM autophagy is specifically required for LICs, but not for differentiated leukemic blasts, to prevent oxidative stress; this should result in the maintenance of appropriate LIC frequency. In contrast, in the PB, autophagy supported survival of leukemia cells irrespective of their differentiation status. Furthermore, inactivation of autophagy potentiated the efficacy of AraC treatment in vivo.

The reason why in the BM LICs, but not other myeloblasts, required autophagy remains elusive. Our data support the hypothesis that accumulation of mitochondria and ROS production are possibly associated with the maintenance of LICs. In fact, normal HSCs have a fewer numbers of mitochondria than those of progenitor cells and use glycolytic metabolism in the hypoxic BM niche.⁴⁵  Although leukemia cells contain more numbers of mitochondria than those of normal counterparts,⁴⁶  fewer mitochondria might be required for maintenance of self-renewal capacity in LICs. These mechanisms are expected to be affected by the microenvironment, and this might account for inconsistencies in previous reports, specifically that pharmacological inhibition of autophagy did not inhibit proliferation in leukemic cell lines.^20,23  For comprehensive understanding of the mechanisms of autophagy in LIC maintenance, further investigation is warranted.

Our study could also help to characterize the impact of autophagy on AraC resistance in vivo. The relationship between autophagy and drug resistance has been vigorously investigated; histone deacetylase inhibitors, mammalian target of rapamycin inhibitors, and AraC were all affected by autophagy in AML cell lines.^20-24  However, considering that the dependency on autophagy in leukemia cells is expected to be highly affected the surroundings such as the BM or PB microenvironment, it is important to evaluate if autophagy has an essential role in vivo. As expected, we revealed that in vivo treatment of AraC activated autophagy specifically in LICs, and that Atg7 deletion potentiated the therapeutic efficacy of AraC.

An excess of leukemic blasts in the PB is known to be a sign of overflow from the BM. Our unique findings could suggest that PB leukemia cells also have leukemia-initiating capacity, which is partly regulated by autophagy. Possible explanations for the dependency on autophagy in PB leukemia cells can be derived from several environmental differences between BM and PB. These include oxygen partial pressure, cytokine concentrations, and adhesion to stromal cells, among others. These factors might render PB leukemia cells more dependent on autophagy. Taking into account the recent finding that granulocyte colony-stimulating factor–mobilized PB HSCs have enhanced autophagy and dependency on autophagy,⁴⁷  it is suggested that undifferentiated cells, which normally reside in BM, might require autophagy to manage the cellular stress of the PB environment.

In this study, we detected an enhanced number of autophagosomes in PB leukemia cells through both electron microscopy and biochemical methods. At the same time, we confirmed that the degradation of autophagic substrates was induced in PB leukemia cells through measuring GFP-LC3 expression. These results support the notion that autophagy is activated in PB leukemia cells, although further experiments by using lysosome inhibitors are still required to confirm autophagic flux.⁴⁸  Molecular mechanisms underlying the possible activation of autophagy in PB leukemia cells were not revealed by this study. Semicomprehensive reverse transcription polymerase chain reaction analysis revealed that a small number of genes were upregulated or downregulated in PB cells (greater than twofold), although expression of the majority of autophagy-associated genes was not altered (data not shown). More precise investigation of the mechanisms of altered autophagy would provide further understanding of the intratumor heterogeneity of AML.

Some issues remain before autophagy inhibitors can be applied to human AML. First, we revealed that survival in leukemic mice was slightly but significantly prolonged after Atg5/7 deletion. To determine the clinical importance of this finding, similar studies using models that have high clinical predictability, such as human AML-transplanted immunodeficient mice, are required. Second, the fact that normal HSCs also depend on autophagy (shown in previous reports) raises concern about possible toxicity of autophagy inhibition.^34,40,49  However, defects in HSCs and myeloproliferation, specifically caused by autophagy deficiency, seem to be observed 6 to 9 weeks after autophagy inactivation. In contrast, the deficiencies in LICs were detectable 1 week after TAM treatment in our study, implying that LICs might be more susceptible to autophagy inhibition than HSCs. As our study did not evaluate the effects of systemic inhibition of autophagy, further investigation is required. Third, it has recently been reported that heterozygous deletion of Atg5 results in increased leukemogenic potential in a murine BMT model with transduction of MLL-ENL.⁵⁰  In that study, it was also shown that homozygous deletion of Atg5 induces substantial cell death in a MLL-ENL induced leukemic cell line. These results suggest that complete ablation of autophagy could inhibit tumor survival or self-renewal activity; however, mild inactivation of autophagy might enhance the malignant phenotype of MLL-ENL leukemia. Thus, small-molecule inhibitors with high efficacy might be required for successful autophagy inhibition therapy.

Despite these concerns, autophagy inactivation by small molecules including chloroquine is a promising concept for leukemia therapy. As autophagy inhibition reduced LIC frequency, with little effect on total tumor burden (as observed in this study), autophagy inhibitors might be suitable for use in combination with other conventional drugs such as AraC or anthracyclines. Furthermore, based on the higher dependency on autophagy of PB leukemia cells, treatment with autophagy inhibitors combined with stem cell mobilizer drugs like plerixafor could be a promising for antileukemia therapy.

In summary, our findings demonstrate that leukemia cells use cytoprotective autophagy for LIC maintenance, survival in PB, and AraC resistance. Even considering that normal hematopoiesis requires autophagy, this function could be a possible therapeutic target in AML and other cancers.

The authors thank M. Komatsu for providing Atg7 conditional knockout mice; N. Mizushima for Atg5 conditional knockout mice and GFP-LC3 mice; T. Kitamura for Plat-E packaging cells; R. Ono and T. Nosaka for MLL-ENL cDNA; T. Nakamura for NUP98-HOXA9 cDNA; F. Ueki and R. Toyama for expert technical assistance; and Editage (http://www.editage.jp) for English-language editing.

This work was supported by the Japan Society for the Promotion of Science Grants-in-Aid for Scientific Research (KAKENHI) (24249055, 16H01196, and 15K09466).

Contributions: Y.S., J.K., K.N., K.K., T.S., and M.K. designed the research; Y.S., T.T.-K, and K.M. performed the experiments; and Y.S., J.K., K.N., T.S., and M.K. wrote the manuscript.

Conflict-of-interest disclosure: Y.S. is an employee of Kyowa Hakko Kirin Co. Ltd. The remaining authors declare no competing financial interests.

Correspondence: Mineo Kurokawa, Department of Hematology & Oncology, Graduate School of Medicine, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8655, Japan; e-mail: kurokawa-tky@umin.ac.jp.