MiR-15a-5p Confers Chemoresistance in Acute Myeloid Leukemia by Inhibiting Autophagy Induced by Daunorubicin

Anthracyclines remain a cornerstone of induction chemotherapy for acute myeloid leukemia (AML). Refractory or relapsed disease due to chemotherapy resistance is a major obstacle in AML management. MicroRNAs (miRNAs) have been observed to be involved in chemoresistance. We previously observed that miR-15a-5p was overexpressed in a subgroup of chemoresistant cytogenetically normal AML patients compared with chemosensitive patients treated with daunorubicin and cytarabine. MiR-15a-5p overexpression in AML cells reduced apoptosis induced by both drugs in vitro. This study aimed to elucidate the mechanisms by which miR-15a-5p contributes to daunorubicin resistance. We showed that daunorubicin induced autophagy in myeloid cell lines. The inhibition of autophagy reduced cell sensitivity to daunorubicin. The overexpression of miR-15a-5p decreased daunorubicin-induced autophagy. Conversely, the downregulation of miR-15a-5p increased daunorubicin-induced autophagy. We found that miR-15a-5p targeted four genes involved in autophagy, namely ATG9a, ATG14, GABARAPL1 and SMPD1. Daunorubicin increased the expression of these four genes, and miR-15a-5p counteracted this regulation. Inhibition experiments with the four target genes showed the functional effect of miR-15a-5p on autophagy. In summary, our results indicated that miR-15a-5p induces chemoresistance in AML cells through the abrogation of daunorubicin-induced autophagy, suggesting that miR-15a-5p could be a promising therapeutic target for chemoresistant AML patients.


Introduction
Acute myeloid leukemia (AML) is a malignant clonal myeloid neoplasm resulting from the acquisition of several molecular, genetic, and epigenetic alterations in myeloid progenitors, affecting their differentiation, proliferation, and apoptosis [1,2]. The first-line treatment commonly used for de novo AML is the standard induction "7 + 3" chemotherapy regimen combining cytarabine with an anthracycline (daunorubicin or idarubicin). Twenty to thirty percent of young adult patients and fifty percent of older adult patients (≥60 years) with newly diagnosed AML fail to achieve a complete remission due to primary drug resistance or death [3]. In addition, a high percentage of patients who initially achieved a complete remission will relapse due to acquired secondary resistance. Chemotherapy resistance remains the main obstacle in the treatment of AML and critically contributes to short-term overall survival. The mechanisms of drug resistance in AML are poorly understood. A main research challenge in AML therapy is the identification of pathways involved in chemotherapy resistance that could provide new targets for future therapeutic interventions.
Daunorubicin (DNR) is the most widely used anthracycline combined with cytarabine in the standard "7 + 3" induction regimen. DNR dose intensification has been an important topic in the last decade. Clinical trials reported improvements in remission rates as well as in survival with higher doses of DNR in subgroups of AML patients following cytogenetic and molecular classification [4,5]. Patients younger than 65 years old or with an unfavorable cytogenetic risk seemed to benefit from high doses of DNR without increased toxicity [6]. A deeper understanding of the mechanisms of DNR resistance will help to choose the best therapy and anthracycline dose for each AML patient without additional toxicities.
Autophagy is an important homeostatic cellular recycling program. In cancer cells, autophagy constitutes a prosurvival and cytoprotective mechanism in response to stress conditions but can also, if inappropriately activated, promote apoptosis and/or autophagic programmed cell death. In response to a wide array of therapies, autophagy in human AML cells predominantly drives a cytodestructive cascade that induces the catabolism of oncogenic fusion proteins and leads to cell death. Arsenic trioxide, a potent inducer of autophagy in AML, induced anti-leukemic effects via p62/SQSTM1-mediated degradation of the PML/RARA oncoprotein in NB4 cells [7]. Similarly, autophagy induction by proteasome inhibitors was responsible for the degradation of FLT3-ITD and subsequent activation of cell death in AML cell lines [8]. Furthermore, several studies demonstrated a decrease in autophagy levels in primary human blast cells and that a loss of key autophagy genes leads to leukemia initiation and progression [9,10]. Multiple autophagy genes were localized within chromosomal regions commonly heterozygously deleted in AML, including ATG12, GABARAPL1, and GABARAPL2, whose expressions were particularly affected in AML blasts [9].
MicroRNAs (miRNAs) constitute a family of small non-coding RNAs (18-24 bp) that bind to the 3 untranslated region (UTR) of target transcripts to induce their degradation or block their translation. Accumulating evidence suggests that miRNAs play critical roles in AML by regulating the expression of a wide array of target genes involved in cell survival, proliferation, differentiation, and apoptosis [11,12]. In addition, miRNAs are important regulators of autophagy [13][14][15]. For instance, miR-130a targeted ATG2B to inhibit autophagy and promote chronic lymphocytic leukemia cell death [14]. In chronic myeloid leukemia cells, miR-30a increased imatinib sensitivity through the regulation of ATG5 and BECN1/Beclin 1 [16]. In AML cells, the overexpression of miR-106a prevented ULK1 induction by all-trans retinoic acid (ATRA) treatment [10].
A wide range of studies has demonstrated the role of multiple miRNAs in modulating the sensitivity to cytotoxic agents in AML cell lines [17][18][19][20]. However, little is known about the roles of miRNAs in chemoresistance in AML patients. In our previous study, we showed that miR-21-5p and miR-15a-5p were overexpressed in a subgroup of chemoresistant, cytogenetically normal AML patients compared to chemosensitive patients treated with the standard induction chemotherapy combining cytarabine and DNR [21]. In AML patients, a high expression of miR-15a-5p was shown to predict shorter survival and worse response to chemotherapy [22]. The mechanisms of chemoresistance induced by miR-15a-5p in AML are not completely understood.
In the current study, we assessed the role of miR-15a-5p in DNR-induced autophagy. We identified and validated four target genes involved in autophagy. Our data provide a new mechanism of DNR resistance in AML by which miR-15a-5p negatively regulates chemotherapy-induced cytotoxic autophagy.

Daunorubicin Induces Autophagy in Myeloid Cell Lines
The autophagy process begins with the formation of an isolation membrane known as the phagophore. The assembly of the phagophore is promoted by conditions of nutrient, energy, or growth factor deprivation, and requires the conjugation of microtubule-associated protein 1 light chain 3 (LC3). ULK1 turns on the subsequent vesicle nucleation by regulating the activation of a multiprotein complex composed in part of Beclin 1 and the class III phosphoinositide 3-kinase (PI3K) Vsp34. Portions of cytoplasm, protein aggregates, or organelles are sequestered within double-membraned vesicles called autophagosomes. The initiation and elongation of autophagosomes are mediated by multiple members of the autophagy gene (ATG) family. Subsequently, loaded autophagosomes fuse with lysosomes to form autolysosomes, enabling cellular cargo degradation and recycling ( Figure 1A) [13,23].
To evaluate the regulation of autophagy by daunorubicin (DNR), we first analyzed the conversion of the cytosolic form of LC3 (LC3-I) to the conjugated form (LC3-II), which is recruited to autophagosomal membranes ( Figure 1A), by Western blotting. High doses of DNR (2 µM) enabled the switch of LC3-I into LC3-II and were accompanied with a significant increased expression level of LC3 in DNR-treated myeloid leukemia cell lines K562 and KG1a ( Figure 1B and Figure S1A). Consistently, the expression of LAMP-2, a lysosomal membrane protein involved in chaperone-mediated autophagy ( Figure 1A), was up-regulated in AML cells treated with DNR ( Figure 1B and Figure S1A). Concurrently, p62/SQSTM1, an intra-autophagosomal component that is degraded by autophagy ( Figure 1A), was significantly downregulated by DNR treatment. The effects of DNR on these proteins were dose-dependent ( Figure 1B and Figure S1A).   antibodies. Expression of β-actin was also detected as a loading control. The Western blotting band intensities of three independent experiments were quantified using the ImageJ software. The untreated conditions were set to 1. The normalized means are shown with SEM. (*, p < 0.05, **, p < 0.01, ***, p < 0.001, in black: all the conditions compared with the untreated conditions, and in grey: CQ conditions compared with DNR and CQ conditions). (D) Double-labeling immunofluorescence and representative images of LC3 (in green) and LAMP-2 (in red). Nuclear staining (DAPI, blue) is also shown.
We next used chloroquine (CQ), a lysosomal inhibitor that impairs autophagosome fusion with lysosomes ( Figure 1A), to monitor autophagic flux in the presence of DNR. As expected, CQ treatment triggered the accumulation of LC3 and LAMP-2 autophagic markers, as shown by Western blotting and immunofluorescence ( Figure 1C,D). In combination with CQ, DNR significantly increased the expression of LC3-II and LAMP-2 and decreased p62/SQSTM1 compared with the vehicle condition, indicating that DNR activated an autophagic flux ( Figure 1C). Moreover, the treatment of K562 cells with DNR increased the immunofluorescence staining of LC3, which localizes to isolation membranes and autophagosomes, and the staining of LAMP-2, a marker for autophagosomes and lysosomes ( Figure 1D). Noticeably, without treatment, the LC3 + /LAMP-2 + staining highlighted autophagosomes, which appeared as perinuclear punctuations, whereas under DNR treatment, the staining distribution was more diffuse in the cells ( Figure 1D). Taken together, our results show that autophagy is promoted by DNR in myeloid leukemia cell lines, in accordance with the study of Han et al. [24].

Autophagy Inhibition Increases Myeloid Cell Growth and Prevents Downregulation by Daunorubicin
Next, we assessed the implication of autophagy regulation by DNR in myeloid leukemia cell growth. To prevent the induction of autophagy by DNR, we treated K562 cells with 3-methyladenine (3-MA), which blocks autophagosome formation via the inhibition of class III phosphatidylinositol-3 kinase (PI3K) ( Figure 1A). As expected, autophagy inhibition by 3-MA resulted in a decreased protein level of LC3-II and LAMP-2 and an increased expression of p62/SQSTM1 (Figure 2A). In addition, 3-MA treatment partially abolished the effects of DNR on autophagy, as shown by Western blotting (Figure 2A).
We analyzed the survival of K562 cells by counting the number of cells over four days in the presence of DNR and/or 3-MA treatment. First, we observed that DNR totally prevented K562 cell proliferation and that 3-MA increased cell growth ( Figure 2B). Second, the combination of 3-MA and DNR significantly increased K562 cell survival compared with the DNR alone condition ( Figure 2B), suggesting that DNR decreases myeloid leukemia cell growth partially by inducing autophagy.
abolished the effects of DNR on autophagy, as shown by Western blotting (Figure 2A).
We analyzed the survival of K562 cells by counting the number of cells over four days in the presence of DNR and/or 3-MA treatment. First, we observed that DNR totally prevented K562 cell proliferation and that 3-MA increased cell growth ( Figure 2B). Second, the combination of 3-MA and DNR significantly increased K562 cell survival compared with the DNR alone condition ( Figure 2B), suggesting that DNR decreases myeloid leukemia cell growth partially by inducing autophagy.  Proteins were extracted after 24 h, and the level of autophagy was analyzed by Western blotting with anti-LC3B, anti-LAMP-2, and anti-p62 antibodies. Expression of β-actin was also detected as a loading control. The Western blotting band intensities of three independent experiments were quantified using the ImageJ software. The untreated conditions were set to 1. The normalized means are shown with SEM. (*, p < 0.05, **, p < 0.01, in black: all the conditions compared with the untreated conditions, and in grey: DNR conditions compared with DNR and 3-MA conditions). (B) Viable cells were counted in the presence of Trypan Blue after 1, 2, 3, and 4 days of culture. The mean of three independent experiments is shown with SEM. The statistical analysis was by a two-way ANOVA followed by Bonferroni's test (***, p < 0.001).

Overexpression of miR-15a-5p Inhibits Autophagy Induced by Daunorubicin
In our previous study, we found that miR-15a-5p was significantly overexpressed in a subgroup of cytogenetically normal chemoresistant AML patients compared with chemosensitive patients [21]. In addition, we showed that miR-15a-5p contributed to chemoresistance by reducing apoptosis induced by cytarabine and DNR [21]. Interestingly, using microarray analysis on K562 cell line treated with DNR (1 µM), we found that miR-15a-5p could downregulate not only pro-apoptotic genes but also genes involved in autophagy ( Figure S2). This result suggested that the overexpression of miR-15a-5p in AML could also induce resistance to DNR treatment by regulating autophagy. To test whether miR-15a-5p was able to regulate autophagy, we overexpressed an miR-15a-5p mimic in myeloid leukemia K562 and KG1a cell lines treated with 2 µM of DNR or a vehicle and analyzed the expression of autophagy markers. The efficiency of transfection was confirmed by measuring the expression of mature miR-15a-5p using RT-qPCR. MiR-15a-5p expression was increased by about 50 times and 3 times upon transfection of K562 and KG1a cells, respectively ( Figure 3A and Figure S1B). Remarkably, the DNR treatment did not modify miR-15a-5p expression before or after transfection ( Figure 3A and Figure S1B). As shown in Figure 3B,C and Figure S1B, miR-15a-5p overexpression resulted in significantly decreased protein levels of LC3 and LAMP-2 and a concurrent increase of p62/SQSTM1 in cells treated with or without DNR, suggesting that miR-15a-5p might negatively regulate autophagy. To confirm the downregulation of autophagy by miR-15a-5p, we also assessed autophagosome formation by immunofluorescence. Cells transfected with miR-15a-5p demonstrated a decrease of both LC3 and LAMP-2 staining compared with the scrambled condition ( Figure 3D). Interestingly, the overexpression of miR-15a-5p also prevented in part the autophagosome formation induced by DNR and was able to limit the accumulation of autophagosomes under CQ treatment ( Figure 3D), indicating that miR-15a-5p expression reduced the autophagic flux. Collectively, our results showed that miR-15a-5p expression inhibits basal level and DNR-induced autophagy in myeloid leukemia cell lines.

Overexpression of miR-15a-5p Increases Myeloid Cell Growth and Prevents Downregulation by Daunorubicin
Then, we assessed the regulation of myeloid leukemia cell growth by miR-15a-5p under DNR treatment. We analyzed the survival of K562 cells transfected with a miR-15a-5p mimic by counting the number of cells over three days in the presence of 0.1 µM of DNR, 0.5 µM of DNR, or vehicle. We also performed a cell viability assay after 72 h of DNR treatment. First, we observed that the overexpression of miR-15a-5p significantly increased K562 cell growth compared with the scrambled condition ( Figure 4A,B). Second, miR-15a-5p overexpression significantly prevented the downregulation of cell survival upon DNR treatment ( Figure 4A,B), suggesting that miR-15a-5p decreases the DNR induction of autophagic cell death.

Inhibition of miR-15a-5p Induces Autophagy and Decreases Myeloid Cell Growth
To further evaluate the regulation of autophagy by miR-15a-5p, we tested the effect of miR-15a-5p inhibition. For this purpose, K562 and KG1a myeloid cell lines were trans-

Inhibition of miR-15a-5p Induces Autophagy and Decreases Myeloid Cell Growth
To further evaluate the regulation of autophagy by miR-15a-5p, we tested the effect of miR-15a-5p inhibition. For this purpose, K562 and KG1a myeloid cell lines were transfected with miR-15a-5p mirVana inhibitor versus a scrambled control, and cells were treated with or without 2 µM of DNR. Inhibition of miR-15a-5p was confirmed by RT-qPCR and resulted in a significant decrease of miR-15a-5p expression ( Figure 5A and Figure S1C). As shown in Figure 5B,C and Figure S1C, myeloid leukemia cells transfected with miR-15a-5p inhibitor revealed increased amounts of LC3-II and LAMP-2 detected by Western blotting and immunofluorescence and a decreased protein level of p62/SQSTM1 relative to the scrambled condition. This result confirmed the negative regulation of autophagy by miR-15a-5p. Moreover, the inhibition of miR-15a-5p recapitulated the autophagy regulation obtained with DNR.
We also analyzed the proliferation and viability of K562 cells transfected with miR-15a-5p inhibitor upon DNR treatment. We observed that the inhibition of miR-15a-5p significantly decreased K562 cell growth compared with the scrambled condition ( Figure 5D,E). In addition, miR-15a-5p inhibition enhanced the decrease of myeloid cell viability induced by DNR, as shown by cell counting and cell viability assay ( Figure 5D,E).

MiR-15a-5p Downregulates the Expression of Autophagy Target Genes
We next investigated potential target genes involved in the regulation of autophagy by miR-15a-5p and in chemoresistance. Our microarray analysis showed that, of the transcripts downregulated by miR-15a-5p in K562 cells treated with DNR, four were implicated in autophagy: ATG9A, ATG14, GABARAPL1, and SMPD1 ( Figure S2A-D). This list was crossed in advance with target mRNAs identified by the TargetScan prediction program [25] (Figure S2A), revealing that the 3 UTR region of these four transcripts contained one or two predicted binding sites for miR-15a-5p ( Figure S2E). ATG9a, which is an autophagy-related membrane protein localized in the phagophore/pre-autophagosomal structure, plays a crucial role in the formation of autophagosomes [26]. ATG14 forms a protein complex with Beclin 1, Vsp15, and Vsp34 (the catalytic subunit of class III PI3K), which initiates autophagy [27]. In addition, ATG14 promotes autophagosome-endolysosome fusion by interacting with the STX17-SNAP29 binary t-SNARE complex. GABARAPL1 is required for the late maturation of autophagosomes [28]. Finally, SMPD1 encodes a lysosomal acid sphingomyelinase also known as ASM and participates in the late autophagy stage associated with the lysosomes [29].
To validate the regulation of these four potential targets by miR-15a-5p, we measured their expression by Western blotting and RT-qPCR after the ectopic expression of miR-15a-5p in K562 cells. The overexpression of miR-15a-5p resulted in a downregulation of ATG9A, ATG14, and GABARAPL1 at protein and RNA levels with respect to the scrambled condition in untreated cells and in cells treated with DNR ( Figure 6A,B). SMPD1 expression was also reduced by miR-15a-5p. This was only validated at the RNA level because we were not able to detect SMPD1 expression by Western blotting due to a lack of functional antibodies ( Figure 6B). Interestingly, DNR treatment significantly increased the expression of these autophagy genes at the protein and/or RNA level, except for ATG9A, and the overexpression of miR-15a-5p counteracted this induction ( Figure 6A,B). To confirm the direct regulation of these four targets by miR-15a-5p, we cloned the 3 UTR regions of ATG9A, ATG14, GABARAPL1, and SMPD1, which were predicted by TargetScan to interact with miR-15a-5p into a pMIR-luciferase reporter ( Figure 6C and Figure S2E). There was one predicted interaction site for miR-15a-5p in the ATG9A and SMPD1 3 UTR and two interaction sites in the ATG14 and GABARAPL1 3 UTR ( Figure S2E). We co-transfected the luciferase reporter with the synthetic miR-15a-5p into HEK 293T cells. We observed a significant reduction (40% to 60%) in the luciferase activity for these four target constructs transfected with miR-15a-5p ( Figure 6C). This effect was abrogated when we co-transfected mutated luciferase reporter vectors carrying deletions of five bases in the seed sequences of miR-15a-5p ( Figure 6C). For ATG9A and SMPD1, the mutation of the only predicted site was sufficient to lose the direct regulation by miR-15a-5p. For ATG14 and GABARAPL1, which contain two predicted binding sites for miR-15a-5p, the mutation of one site had a partial effect. Nevertheless, the double mutant abrogated the decrease in luciferase activity, indicating that the two 3 UTR binding sites were necessary for the regulation of ATG14 and GABARAPL1 by miR-15a-5p. Altogether, these experiments showed that miR-15a-5p directly regulated the expression of four autophagy genes: ATG9A, ATG14, GABARAPL1, and SMPD1.

MiR-15a-5p Downregulates the Expression of Autophagy Target Genes
We next investigated potential target genes involved in the regulation of autophagy by miR-15a-5p and in chemoresistance. Our microarray analysis showed that, of the transcripts downregulated by miR-15a-5p in K562 cells treated with DNR, four were implicated in autophagy: ATG9A, ATG14, GABARAPL1, and SMPD1 ( Figure S2A-D). This list

Expression of ATG9A, ATG14, and GABARAPL1 Are Downregulated in Chemoresistant AML Patients
To assess the implication of autophagy regulation by miR-15a-5p in chemoresistance, we analyzed the gene expression profiles from the HOVON dataset collected with a cohort of AML patient samples, as previously described (GEO accession: GSE6891) [30]. Gene expression was measured in bone marrow samples from 38 AML patients at diagnosis before treatment with standard chemotherapy by using Affymetrix U133A GeneChips. We selected samples from patients younger than 65 years with cytogenetically normal AML and wild-type NPM1 without FLT3-ITD. We compared the four target genes ATG9A, ATG14, GABARAPL1, and SMPD1 expression in chemosensitive (n = 24) and chemoresistant (n = 14) patients. Figure 7A shows the downregulation of the four target genes in chemoresistant AML patients compared with chemosensitive patients, with a significant differential expression for ATG9A and GABARAPL1 between both groups of patients, confirming the direct roles of the autophagy targets in chemoresistance. The expression of miR-15a-5p was previously analyzed in this cohort of AML patients by small RNA-sequencing, and we reported a significant overexpression of miR-15a-5p in chemoresistant compared with chemosensitive AML patients [21]. Figure 7B shows the correlation plot of miR-15a-5p and autophagy gene expression. We observed a negative correlation between miR-15a-5p and ATG9A, ATG14, GABARAPL1, and SMPD1 gene expression

Expression of ATG9A, ATG14, and GABARAPL1 Are Downregulated in Chemoresistant AML Patients
To assess the implication of autophagy regulation by miR-15a-5p in chemoresistance, we analyzed the gene expression profiles from the HOVON dataset collected with a cohort of AML patient samples, as previously described (GEO accession: GSE6891) [30]. Gene expression was measured in bone marrow samples from 38 AML patients at diagnosis before treatment with standard chemotherapy by using Affymetrix U133A GeneChips. We selected samples from patients younger than 65 years with cytogenetically normal AML and wild-type NPM1 without FLT3-ITD. We compared the four target genes ATG9A, ATG14, GABARAPL1, and SMPD1 expression in chemosensitive (n = 24) and chemoresistant (n = 14) patients. Figure 7A shows the downregulation of the four target genes in chemoresistant AML patients compared with chemosensitive patients, with a significant differential expression for ATG9A and GABARAPL1 between both groups of patients, confirming the direct roles of the autophagy targets in chemoresistance. The expression of miR-15a-5p was previously analyzed in this cohort of AML patients by small RNA-sequencing, and we reported a significant overexpression of miR-15a-5p in chemoresistant compared with chemosensitive AML patients [21]. Figure 7B shows the correlation plot of miR-15a-5p and autophagy gene expression. We observed a negative correlation between miR-15a-5p and ATG9A, ATG14, GABARAPL1, and SMPD1 gene expression in chemosensitive and chemoresistant patients, suggesting that the overexpression of miR-15a-5p observed in chemoresistant AML patients contributes to chemoresistance by reducing the expression of autophagy target genes. in chemosensitive and chemoresistant patients, suggesting that the overexpression of miR-15a-5p observed in chemoresistant AML patients contributes to chemoresistance by reducing the expression of autophagy target genes.

Downregulation of miR-15a-5p Targets Decreases the Activation of Autophagy
We next determined whether the autophagy targets of miR-15a-5p participated in its regulation of autophagy using siRNAs against ATG9A, ATG14, GABARAPL1, and SMPD1 to downregulate their expression. To assess the impact on autophagy, we co-transfected K562 cells with a miR-15a-5p inhibitor to increase the autophagy process and with a combination of siRNAs against each target gene. The efficiency of siRNAs was evaluated by Western blotting and RT-qPCR, demonstrating a significant downregulation of ATG9A,

Downregulation of miR-15a-5p Targets Decreases the Activation of Autophagy
We next determined whether the autophagy targets of miR-15a-5p participated in its regulation of autophagy using siRNAs against ATG9A, ATG14, GABARAPL1, and SMPD1 to downregulate their expression. To assess the impact on autophagy, we cotransfected K562 cells with a miR-15a-5p inhibitor to increase the autophagy process and with a combination of siRNAs against each target gene. The efficiency of siRNAs was evaluated by Western blotting and RT-qPCR, demonstrating a significant downregulation of ATG9A, ATG14, GABARAPL1, and SMPD1 expression at protein and/or mRNA levels ( Figure 8A,B). Moreover, the inhibition of miR-15a-5p led to the overexpression of the four targets, confirming their direct regulation. As expected, the combined decreased expression of ATG9A, ATG14, GABARAPL1, and SMPD1 reduced basal autophagy, as shown by the reduced expression of LC3 and LAMP-2 protein and increased expression of p62/SQSTM1 protein ( Figure 8A). The downregulation of the four target genes indeed reproduced the functional effects of miR-15a-5p. Finally, we observed that the combined downregulation of ATG9A, ATG14, GABARAPL1, and SMPD1 using siRNAs blocked autophagy induced by inhibition of miR-15a-5p. Collectively, these results showed that miR-15a-5p regulated autophagy by directly targeting several genes involved in different steps of this process. ATG14, GABARAPL1, and SMPD1 expression at protein and/or mRNA levels ( Figure  8A,B). Moreover, the inhibition of miR-15a-5p led to the overexpression of the four targets, confirming their direct regulation. As expected, the combined decreased expression of ATG9A, ATG14, GABARAPL1, and SMPD1 reduced basal autophagy, as shown by the reduced expression of LC3 and LAMP-2 protein and increased expression of p62/SQSTM1 protein ( Figure 8A). The downregulation of the four target genes indeed reproduced the functional effects of miR-15a-5p. Finally, we observed that the combined downregulation of ATG9A, ATG14, GABARAPL1, and SMPD1 using siRNAs blocked autophagy induced by inhibition of miR-15a-5p. Collectively, these results showed that miR-15a-5p regulated autophagy by directly targeting several genes involved in different steps of this process.
In conclusion, we demonstrated that the direct regulation of ATG9A, ATG14, GABARAPL1, and SMPD1 contributes to the inhibition of autophagy by miR-15a-5p. Figure 8. miR-15a-5p regulates autophagy by targeting autophagy genes. K562 cells were transiently co-transfected with miR-15a-5p inhibitor or scrambled inhibitor as control and with either a combination of four siRNA-targets (ATG9A, ATG14, GABARAPL1, and SMPD1) or a siRNA-control for 24 h before protein and RNA extractions. (A) Level of autophagy was analyzed by Western blotting with anti-LC3B, anti-LAMP-2, and anti-p62 antibodies. Protein expression of ATG9A, ATG14, and GABARAPL1 was also analyzed, as well as the expression of β-actin as a loading control. (B) In all the conditions, the expression of miR-15a-5p, ATG9A, ATG14, GABARAPL1, and SMPD1 was measured by RT-qPCR and normalized with the expression of RNU44 for miR-15a-5p or with RPLP0 for the autophagy target genes. The scrambled inhibitor and siRNA condition was set to 1. The normalized means of three independent experiments are shown with SEM. (*, p < 0.05, **, p < 0.01, ***, p < 0.001, all the conditions compared with the scrambled condition).

Discussion
Acute myeloid leukemia still has in most cases a poor long-term survival rate with a high risk of relapse due to resistance to the standard induction chemotherapy combining cytarabine and DNR. This paper provides new evidence for the important role of miRNAs in DNR resistance in AML mediated by target genes implicated in autophagy.
The induction of autophagy in response to several chemotherapies can favor either death or survival, contributing to drug efficacy or resistance [31]. Most chemotherapies activate autophagy in cancer cells as a protective response to stress-induced damage [23]. Emerging studies show that chemotherapies can also induce autophagy cell death. In our study, we demonstrated that high doses of DNR induced autophagy in myeloid leukemia cell lines, as previously reported by Han and colleagues [24]. Treatment of K562 cells with DNR enables LC3-I/II conversion, punctate distribution of endogenous LC3, and p62 degradation. We also found that DNR-induced autophagy plays a pivotal role in leukemic cell survival. Indeed, the DNR chemosensitivity was partially prevented by pre-treating Figure 8. miR-15a-5p regulates autophagy by targeting autophagy genes. K562 cells were transiently co-transfected with miR-15a-5p inhibitor or scrambled inhibitor as control and with either a combination of four siRNA-targets (ATG9A, ATG14, GABARAPL1, and SMPD1) or a siRNA-control for 24 h before protein and RNA extractions. (A) Level of autophagy was analyzed by Western blotting with anti-LC3B, anti-LAMP-2, and anti-p62 antibodies. Protein expression of ATG9A, ATG14, and GABARAPL1 was also analyzed, as well as the expression of β-actin as a loading control. (B) In all the conditions, the expression of miR-15a-5p, ATG9A, ATG14, GABARAPL1, and SMPD1 was measured by RT-qPCR and normalized with the expression of RNU44 for miR-15a-5p or with RPLP0 for the autophagy target genes. The scrambled inhibitor and siRNA condition was set to 1. The normalized means of three independent experiments are shown with SEM. (*, p < 0.05, **, p < 0.01, ***, p < 0.001, all the conditions compared with the scrambled condition).
In conclusion, we demonstrated that the direct regulation of ATG9A, ATG14, GABARAPL1, and SMPD1 contributes to the inhibition of autophagy by miR-15a-5p.

Discussion
Acute myeloid leukemia still has in most cases a poor long-term survival rate with a high risk of relapse due to resistance to the standard induction chemotherapy combining cytarabine and DNR. This paper provides new evidence for the important role of miRNAs in DNR resistance in AML mediated by target genes implicated in autophagy.
The induction of autophagy in response to several chemotherapies can favor either death or survival, contributing to drug efficacy or resistance [31]. Most chemotherapies activate autophagy in cancer cells as a protective response to stress-induced damage [23]. Emerging studies show that chemotherapies can also induce autophagy cell death. In our study, we demonstrated that high doses of DNR induced autophagy in myeloid leukemia cell lines, as previously reported by Han and colleagues [24]. Treatment of K562 cells with DNR enables LC3-I/II conversion, punctate distribution of endogenous LC3, and p62 degradation. We also found that DNR-induced autophagy plays a pivotal role in leukemic cell survival. Indeed, the DNR chemosensitivity was partially prevented by pre-treating cells with 3-MA, an inhibitor of autophagy. Similarly, Ristic et al. demonstrated that idarubicin induces cytotoxic autophagy in K562 cell line [32]. Autophagy inhibitors reduced the cytotoxicity of idarubicin in K562 cells. Idarubicin-induced autophagy contributes to the proapoptotic action of chemotherapy [32]. We were not able to observe an induction of autophagy after cytarabine treatment of K562 cells at low or high doses (data not shown).
In agreement with our observations, Jin and colleagues showed that autophagy was reduced in AML blasts, and key autophagy genes such as ULK1, ATG3, ATG4D, and ATG5 were found to be downregulated in primary AML patient samples [10]. Watson et al. concluded that decreased autophagy may facilitate aberrant proliferation and contribute to AML development [9]. Accordingly, we observed an increase of K562 cell growth after treatment with the 3-MA autophagy inhibitor. We also found that ATG9A, ATG14, GABARAPL1, and SMPD1 gene expression was downregulated in chemoresistant AML patient samples compared with chemosensitive patients. Despite its dual role in cancer, autophagy has been robustly shown to kill acute myeloid leukemia cells, especially via the degradation of the oncogenic fusion protein that drives leukemogenesis [33]. For instance, the antileukemic role of autophagy has been reported in AML with FLT3-ITD alteration. Larrue et al. showed that FLT3-ITD molecules become detectable within the autophagosomes and are eventually degraded upon proteasome inhibitor-initiated autophagy [8].
Our previous studies investigated the important roles of miRNAs in mediating sensitivity and resistance to chemotherapeutic agents in AML. We previously demonstrated that miR-15a-5p decreased apoptosis induced by DNR and/or cytarabine in leukemia by downregulating three pro-apoptotic target genes-PDCD4, ARL2, and BTG2-validating the implication of miR-15a-5p in drug resistance [21]. Our results revealed that miR-15a-5p contributes to leukemia chemotherapy resistance by regulating, in addition to apoptosis, autophagy induced by DNR treatment. Accordingly, we observed an increase of K562 cell growth after the overexpression of miR-15a-5p and a decrease of cell growth after the inhibition of miR-15a-5p. Our gain-of-function experiments in AML cell lines showed that miR-15a-5p overexpression decreased the activation of autophagy and the downregulation of K562 cell viability after DNR treatment.
In the present study, we demonstrated that miR-15a-5p prevented basal and DNRinduced autophagy in AML by directly downregulating four genes involved in this catalytic process: ATG9A, GABARAPL1, ATG14, and SMPD1. Thereby, miR-15a-5p might prevent autophagy by inhibiting different phases: initiation, elongation and autophagosome formation, fusion, and autolysosome formation. Interestingly, treatment of K562 leukemic cells with DNR increased the expression of GABARAPL1, ATG14, and SMPD1 without changing miR-15a-5p expression, suggesting that DNR regulates autophagy through another mechanism that remains to be elucidated. We also observed a negative correlation between miR-15a-5p and autophagy gene expression in chemosensitive and chemoresistant AML patients, suggesting the implication of autophagy regulation by miR-15a-5p in chemoresistance.
Altogether, our studies revealed that miR-15a-5p might play a dual role in the emergence of drug resistance to DNR by regulating both apoptosis and autophagy. MiR-15a-5p could be considered as a poor predictive biomarker in AML. Further clinical studies would be required to analyze whether patients with a higher level of expression of miR-15a-5p would benefit from DNR dose intensification. In addition, our results suggested that inhibiting miR-15a-5p could be a promising adjuvant therapeutic option in drug-resistant AML patients. Indeed, the modulation of miR-15a-5p, using antagomiRs for instance, could restore key genes involved in autophagy and in apoptosis and improve remission rates mainly in chemoresistant AML patients [37].

RNA Extraction and RT-qPCR
Total RNA was extracted using TriPure Isolation Reagent (Sigma Aldrich, Saint-Louis, MO, USA). For the quantification of miRNA expression levels, 50 ng of total RNA was used for reverse transcription using the TaqMan MicroRNA Reverse Transcription Kit (Life Technologies), and real-time PCR was performed in triplicate using the TaqMan MicroRNA assay kit (hsa-miR-15a-5p: ID000389, RNA-control RNU44: ID001094, Life Technologies) according to the instructions of the manufacturer. Normalization was completed with small nucleolar RNU44, and the relative expression was calculated using the comparative cross threshold (Ct) method. For the quantification of mRNA levels, 1 µg of total RNA was subjected to reverse transcription using MMLV (Moloney Murine Leukemia Virus) reverse transcriptase enzyme (Invitrogen, Carlsbad, CA, USA). Quantitative PCR analysis was performed in triplicate using ABgene SYBR green-based kits (Thermo Fisher Scientific) using the oligonucleotides shown in Table 1 as described previously [38,39]. The ribosomal protein RPLP0 served as normalization control.

Immunofluorescence
Cells were transferred to slides by centrifugation at 250 rpm for 5 min according to the Cytospin method. Cells were then fixed using 4% paraformaldehyde (#P6148, Sigma Aldrich) in PBS at pH 7.4 at room temperature for 15 min, permeabilized by incubation with 0.05% saponin (#47036, Sigma Aldrich) in PBS pH 7.4 for 20 min, and blocked using Q-PBS solution (PBS + 1% BSA + 0.01% saponin) for 20 min. Cells were incubated overnight at 4 • C with the indicated primary antibodies in Q-PBS (dilution of 1:200) and then washed three times with Q-PBS before incubation at room temperature for 1 h with the secondary antibodies (Alexa Fluor ® 488 donkey anti-rabbit #A21206 or Alexa Fluor ® 568 goat antimouse #A11004 dilution 1:500 and DAPI dilution 1:5000, Invitrogen, Carlsbad, CA, USA). The stained cells were washed two times with Q-PBS and two times with PBS and then post-fixed using 4% paraformaldehyde in PBS at pH 7.4 at room temperature for 10 min. Cells were finally washed three times with PBS, and coverslips were added on top of Dako Fluorescence mounting medium (Agilent Technologies, Santa Clara, CA, USA).

Cell viability Assay
K562 cells were seeded in 96-well plates (10.000 cells/well) in 100 µL of serumsupplemented medium. After 72 h, 100 µL of the single reagent (CellTiter-Glo ® Reagent) was directly added and mixed into the cell culture. After 10 min, the luminescent signal was monitored using a GloMax ® instrument (Turner Biosystems, Sunnyvale, CA, USA).

Microarray Assay
Microarrays were performed according to the Affymetrix ® WT PLUS standard protocol. In total, 100 ng of total RNA was used as a starting material. The GeneChip WT PLUS Reagent Kit (Affymetrix, Santa Clara, CA, USA) was used for ss-cDNA preparation, fragmentation, and labeling. Human Transcriptome Arrays 2.0 (HTA2.0) chips were used for hybridization. The hybridization, wash, and scan were performed according to the Affymetrix kits and procedures specific to the HTA2.0 chips. After the scan, the quality controls of the hybridization were checked using the Gene Expression Console software (Affymetrix, Santa Clara, CA, USA). The RMA-Sketch procedure was used for data normalization. Fold change was determined for each comparison based on these normalized values. We found 520 genes differentially expressed after miR-15a-5p overexpression in K562 cells compared to the scrambled condition, of which 369 genes were downregulated ( Figure S1A) and 151 were upregulated. The list of the most downregulated mRNAs (with the highest negative fold change) that were predicted as target genes by TargetScan program is shown in Figure S2B. We used the DAVID web tool [40] for the functional annotation and pathway analysis of the selected gene list ( Figure S2C).

Cloning and Site-Directed Mutagenesis
The 3 UTR segments containing the target sites for miR-15a-5p (TGCTGCT) were amplified by PCR from cDNA and inserted into a pMirTarget vector (#PS100062, Ori-Gene) between the EcoRI and NotI restriction sites for ATG9A (ENST00000361242.9, +3387 to +3648 pb), SMPD1 (ENST00000342245.9, +2108 to +2410 pb) and GABARAPL1 (ENST00000266458.10, +731 to +1611 pb), and between the MluI and NotI restriction sites for ATG14 (ENST00000247178.6, +1842 pb to +4666 pb). All the mutants were generated by the deletion of 5 bp within the site of perfect complementarity using the QuickChange™ XL-II kit (Stratagene, La Jolla, CA, USA) according to the manufacturer's protocol. Mutagenesis primers were synthesized by Eurogentec (Ougrée, Belgium) and are listed in Table 2. All the constructs were verified by sequencing.

Gene Expression Profiles of Patient Samples
Frozen diagnostic bone marrow RNA samples were obtained from 38 adults who had a confirmed diagnosis of AML. Twenty-four of these patients were considered "chemosensitive" because they reached a complete response after receiving one cycle of induction chemotherapy. The 14 remaining patients were considered "chemoresistant", defined as having more than 5% blast cells in the bone marrow after induction chemotherapy. Cytogenetic analyses and patient clinical characteristics were described in Table S1 in our previous paper [21]. Blasts and mononuclear cells at diagnosis were purified by Ficoll-Hypaque (Nygaard) density gradient centrifugation and cryopreserved. RNA was isolated with either RNA-Bee or RLT following the protocols of the manufacturer (Bio-Connect BV, Huissen, The Netherlands). Expression levels of the four target genes-ATG9A (202492_at), ATG14 (204568_at, 233984_at), GABARAPL1 (211458_s_at, 208868_s_at, 208869_s_at), and SMPD1 (209420_s_at)-were analyzed in 38 AML patient samples by using Affymetrix U133A GeneChips HOVON dataset as previously described (GEO accession: GSE6891) [30].

Statistical Analysis
Experiments were repeated at least three times with identical results. In most figures, the average of multiple replicate experiments is shown with the standard error of the mean (SEM), unless otherwise stated. Statistical analysis was performed using a bilateral Student's t-test, Wilcoxon test, correlation test or a two-way ANOVA followed by Bonferroni's test (*, p < 0.05; **, p < 0.01; ***, p < 0.001) using GraphPadPrism7 software.