miR‐15a‐5p and miR‐21‐5p contribute to chemoresistance in cytogenetically normal acute myeloid leukaemia by targeting PDCD4, ARL2 and BTG2

Cytarabine and daunorubicin are old drugs commonly used in the treatment of acute myeloid leukaemia (AML). Refractory or relapsed disease because of chemotherapy resistance is a major issue. microRNAs (miRNAs) were incriminated in resistance. This study aimed to identify miRNAs involved in chemoresistance in AML patients and to define their target genes. We focused on cytogenetically normal AML patients with wild‐type NPM1 without FLT3‐ITD as the treatment of this subset of patients with intermediate‐risk cytogenetics is not well established. We analysed baseline AML samples by small RNA sequencing and compared the profile of chemoresistant to chemosensitive AML patients. Among the miRNAs significantly overexpressed in chemoresistant patients, we revealed miR‐15a‐5p and miR‐21‐5p as miRNAs with a major role in chemoresistance in AML. We showed that miR‐15a‐5p and miR‐21‐5p overexpression decreased apoptosis induced by cytarabine and/or daunorubicin. PDCD4, ARL2 and BTG2 genes were found to be targeted by miR‐15a‐5p, as well as PDCD4 and BTG2 by miR‐21‐5p. Inhibition experiments of the three target genes reproduced the functional effect of both miRNAs on chemosensitivity. Our study demonstrates that miR‐15a‐5p and miR‐21‐5p are overexpressed in a subgroup of chemoresistant AML patients. Both miRNAs induce chemoresistance by targeting three pro‐apoptotic genes PDCD4, ARL2 and BTG2.


| INTRODUC TI ON
Acute myeloid leukaemia (AML) is a heterogeneous group of clonal myeloid neoplasms with still a poor prognosis. 1 For several decades, the standard of care for AML patients has been the classical "7 + 3" regimen combining 7 days of cytarabine and 3 days of an anthracycline. 2,3 During these years, the incremental improvements in AML patient overall survival that were observed were mainly because of better supportive care. Unfortunately, up to 30%-40% of patients will develop a chemorefractory disease, defined as failure to achieve morphological complete response (CR) after one to two cycles of induction chemotherapy. This subgroup of AML patients faces a dismal prognosis with a median survival of <1 year. 4,5 To decipher underlying mechanisms of resistance to induction chemotherapy in AML, a better understanding of the heterogeneity of refractory patients at the cellular and molecular levels is critical. The development of novel therapies is very much needed to improve outcome of chemoresistant AML patients.
Acute myeloid leukaemia is the result of a clonal multistep evolutionary process starting with the acquisition of founder mutations in the haematopoietic stem cell followed by driver mutations ultimately leading to the development of AML. 2 The cause of drug resistance can be intrinsic in patients who fail the initial chemotherapy because of the presence of multiples subclones or acquired after chemotherapy through the selection or acquisition of mutations in resistant subclones.
Several pathways were altered in resistance including DNA damage and repair, cell cycling, cell death, drug targets, drug metabolism and drug trafficking. 6,7 Targeted therapies directed to the chemotherapy resistance mechanisms may improve the outcome for these patients. microRNAs (miRNAs) constitute a class of short, evolutionary conserved, non-coding RNAs with major regulatory functions. miR-NAs can regulate up to 60% of protein-coding genes. By binding to the 3′ untranslated region of target mRNAs, a single miRNA can negatively regulate the expression of a wide variety of target genes (up to 200 mRNAs). miRNA expression was found deregulated in human cancers and in particular in AML. 8,9 Each AML subtype exhibits a unique miRNA signature. 8 Several studies showed that miRNA can function as tumour suppressor or onco-miRNAs according to the roles of their target genes. 10 miRNAs can affect crucial leukaemic processes such as proliferation, survival, differentiation, self-renewal, epigenetic regulation, in vivo disease progression and chemotherapy resistance. Modifying miRNA expression is an exciting potential therapeutic approach in AML. Because miRNA can repress several targets involved in different pathways, miRNA-based therapy would be an efficient tool against chemoresistance. microRNAs were reported to regulate sensitivity and resistance to cytotoxic agents in AML cell lines. 7,11 Increased expression of some miRNAs resulted in chemosensitivity, whereas others were linked to chemoresistance. The mechanisms whereby miRNAs cause chemoresistance in AML patients are not well understood. Equally unclear are the molecular bases for differential expression of miRNAs in histologically and genetically similar AMLs. Understanding how miRNAs are differently expressed, their targets and effects on the AML clone might also help to better understand the leukemogenesis process.
In the current study, we assessed whether miRNAs play a role in chemoresistance in cytogenetically normal AML patients treated with standard induction chemotherapy. We analysed miRNA expression by high-throughput small RNA sequencing in two cohorts of AML patients. Next, we performed functional tests on cell lines and identified target genes involved in chemoresistance. These results pave the way for therapeutic interventions to circumvent chemotherapy resistance.

| Patient samples
Frozen diagnostic bone marrow RNA samples were obtained from 47 adults who had a confirmed diagnosis of AML. Twenty-seven of these patients were considered "chemosensitive" because they reached CR after receiving one cycle of induction chemotherapy.
The twenty remaining patients were considered "chemoresistant", defined as more than 5% blast cells in the bone marrow after induction chemotherapy. Cytogenetic analyses were performed at diagnosis. FLT3-ITD and NPM1 mutation analysis was performed by PCR amplification and sequencing. Patients with a cytogenetically normal AML (CN-AML) and wild-type NPM1 without FLT3-ITD were included in the study (Table S1). TP53, ASXL1 and RUNX1 mutational status was obtained for 37 patients by next-generation sequencing (Illumina TruSight Myeloid Panel) as previously described. 12 Patient clinical characteristics are shown in Table S1.
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). Frozen diagnostic bone marrow RNA samples were analysed by small RNA sequencing. See supplemental methods (Appendix S1) for details.
K562 cells were treated with daunorubicin 0.5 µmol/L and/or cytarabine 5 µmol/L for 24 hours. OCI-AML3 cells were treated with daunorubicin 0.1 µmol/L and/or cytarabine 1 µmol/L for 24 hours as these cells are more sensitive to chemotherapy.
The efficiency of the infection was checked by measuring the green fluorescent protein (GFP) expression by flow cytometry and was around 88% and 72% in K562 and OCI-AML3, respectively, as shown in Figure S1A. miRNA mature expression was measured by qRT-PCR ( Figure S1B). An empty vector was used as control.

| Microarray studies on cell lines
Microarrays were performed according to the Affymetrix ® WT PLUS standard protocol following the manufacturer's instructions.
See supplemental methods (Appendix S1) for details.

| Quantitative reverse transcription polymerase chain reaction
For the quantification of miRNA levels after lentiviral infections, reverse transcription was carried out using the QuantiMir Reverse Transcription Kit (System Biosciences, #RA420A-1) and real-time PCR was performed using the designed forward primer, the universal re-

| Luciferase reporter experiments
The 3′UTR segments containing the target sites for miR-15a-5p and miR-21-5p were amplified by PCR from cDNA and inserted into pGL3 control vector (Promega), using the Xba1 site.
See supplemental methods (Appendix S1) for details.

| PCA and correlation studies
A principal component analysis (PCA) using gene and miRNA expression data for the selected miRNAs (miR-15a-5p and miR-21-5p) and probesets of PDCD4, ARL2 and BTG2 genes was performed in R program using the mixOmics package. 15 Chemoresistant patient numbers 59, 61, 62 and 63 were removed from the PCA because we did not have enough RNA left to perform gene expression profiles. To study the correlation between genes and miRNAs that were the basis of this PCA, correlation plot was generated from the PCA. Normal regression analysis was performed between the clinical data and the patient loadings on the three first PCs. Each clinical phenotype (bone marrow blast percentage, white blood cell count, platelet count) was evaluated individually against the patient loading on the three PCs.

| miR-15a-5p and miR-21-5p are overexpressed in chemoresistant AML patients
To identify miRNAs that are involved in chemoresistance, we first analysed miRNA expression in bone marrow samples from 47 AML patients at diagnosis, before treatment with cytarabine and anthracycline, using Illumina small RNA sequencing. We selected samples from patients younger than 65 years with cytogenetically normal AML (CN-AML) and wild-type NPM1 without FLT3-ITD. This subgroup of AML patients was included in the intermediate-risk category by the 2010 ELN genetic risk stratification. 16 Table S3 showed the list of miRNAs differentially expressed between both groups of patients with a highly significant P-value (≤10E-08) and a frequency of selection of 100%.
We selected the four miRNAs of this list with a linear fold change above 7 (Table 1). Figure 1A shows the distribution of the expression of miR-21-5p, miR-340-5p, miR-181c-5p and miR-15a-5p. The distribution confirmed the difference between both groups of patients. We validated the significant differential expression of two of these miRNAs-miR-15a-5p and miR-21-5p-by qRT-PCR in patient samples ( Figure 1B).
We did not find a significant differential expression of miR-340-5p and miR-181c-5p between the two groups of patients by qRT-PCR (data not shown).
The more recent ELN2017 genetic risk stratification specified that this subgroup of AML patients with wild-type NPM1 and no FLT3-ITD (or with FLT3-ITD low) was classified in the intermediate-risk category if they did not have any adverse-risk genetic lesion.
We collected the mutational status for RUNX1, ASXL1 and TP53 in eighty percent of our patients and excluded the patients with RUNX1 or ASXL1 mutation. TP53 mutation was not found in our patients. AML patients regarding chemosensitive patients was confirmed in our smaller cohort of patients classified in the intermediate-risk category following the ELN2017 genetic risk stratification ( Figure S4).
We compared patient clinical characteristics at diagnosis (bone marrow blast percentage, white blood cell count, platelet count) in chemoresistant and chemosensitive AML patients (Table S1). The regression analysis did not show any significant correlation between the clinical characteristics and miR-21a-5p or miR-15a-5p expression level or the chemoresistance (data not shown).
To evaluate the impact of miR-15a-5p and miR-21-5p on chemosensitivity, we stably overexpressed both miRNAs in the cell lines (K562 and OCI-AML3) using a lentiviral construct containing the pre-miR-15a-5p or pre-miR-21-5p. The empty lentivirus was used as control. The efficiency of the infection was measured by analysing the GFP expression by flow cytometry and miRNA mature expression by qRT-PCR ( Figure S1). Infected cells were treated with cytarabine and/or daunorubicin for 24 hours. As shown in Figure 2, the overexpression of miR-15a-5p or miR-21-5p significantly decreased apoptosis induced by the combination of cytarabine and daunorubicin in K562. The overexpression of miR-15a-5p or miR-21-5p also decreased apoptosis induced by cytarabine alone or by daunorubicin alone ( Figure S6). Like in K562, in OCI-AML3, the overexpression of either miRNAs decreased apoptosis induced by cytarabine and daunorubicin alone or combined ( Figure S7).
As a note, note that the overexpression of miR-15a-5p or miR-21-5p did not lead to a significant reduction in apoptosis in the absence of chemotherapy in both cell lines ( Figure S8).
Remarkably  (Table S4). After overexpression of miR-15a-5p, we found 141 down-regulated genes that were predicted as target genes by both programmes (Table S4). The most down-regulated genes were BTG2 and PDCD4 after overexpression of both miRNAs, and ARL2 after overexpression of miR-15a-5p only ( Table 2).
To validate these three potential miR-15a-5p and miR-21-5p target genes, we measured their expression by quantitative RT-PCR and by western blotting after ectopic expression of miR-15a-5p and miR-21-5p. Increased level of both miRNAs upon transfection was confirmed by qRT-PCR ( Figure S2A).

miR-15a-5p and miR-21-5p overexpression significantly reduced
ARL2, PDCD4 and BTG2 mRNA expression in K562 cells treated with cytarabine and daunorubicin ( Figure 3A). The overexpression of miR-15a-5p or miR-21-5p in K562 cells treated with cytarabine and daunorubicin resulted in down-regulation of PDCD4 at the protein level with respect to the scrambled control ( Figure 3B,C). ARL2 expression was also decreased after overexpression of miR-15a-5p only ( Figure 3B,C). Noticeably, we could not detect BTG2 expression by western blotting by processing with the available commercial antibodies. The same results were obtained in K562 cell lines treated with cytarabine alone or daunorubicin alone versus vehicle ( Figure S9).
We cloned the 3′UTR of ARL2, PDCD4 and BTG2 containing the seed sequences that were predicted to interact with miR-15a-5p and miR-21-5p into a luciferase reporter vector pGL3 control ( Figure S10). We cotransfected the luciferase reporter with the synthetic miRNA of interest into K562 cell line. There are two predicted interaction sites for miR-15a-5p in the PDCD4 3′UTR, five in the ARL2 3′UTR and two in the BTG2 3′UTR ( Figure S10).
The miR-21-5p has one predicted interaction site in the PDCD4 3′UTR and one in the BTG2 3′UTR ( Figure S10). A significant reduction in the luciferase/βgal ratio was observed for ARL2, PDCD4 and BTG2 constructs transfected with synthetic miRNA compared with the control ( Figure 3D). This effect was abrogated when we cotransfected mutated 3′UTR luciferase reporter vectors and the synthetic miRNA. The mutations were created by the deletion of 4-6 bases in the seed sequence of the miRNAs ( Figure S10). These results confirmed the direct regulation of the three target genes by miR-15a-5p and/or miR-21-5p.

| Target gene validation in patient samples
To validate the three target genes in our patient cohort, we analysed the data from the gene expression profiles performed on patient samples. Figure 4A shows the down-regulation of the three target F I G U R E 2 miR-15a-5p and miR-21-5p reduce apoptosis induced by the combination of cytarabine and daunorubicin. K562 cells were infected with lenti-miR virus to overexpress miR-15a-5p, miR-21-5p or EV. Annexin V assay was performed after 24 h of treatment with cytarabine (Ara-C) and daunorubicin (DNR). The results are presented as percentage of apoptotic cells. One representative experiment is shown in (A) and the average of th independent experiments ± SD in (B). P values were obtained using a t test. * Indicates a significant difference P ≤ .05, and ** indicates a significant difference P ≤ .01 We performed a PCA based on selected gene and miRNA expression data. The first three principal components explained eighty-four per cent of the variance. Figure 4B shows the PCA plot of PC2 to PC1 (the two principal components). Individuals such as "54r", "50", "47", "46" and "48" are chemoresistant and clearly separated from the rest. Individuals "20", "31", "30", "40", "36" and "17"

| D ISCUSS I ON
Acute myeloid leukaemia is only cured in 35%-40% of young adults. The main problem remains resistance to the old commonly used chemotherapy combining cytarabine and an anthracycline, that is daunorubicin. This paper provides new evidence of the important roles of miRNAs in chemoresistance in cytogenetically normal AML patients mediated by target genes implicated in early apoptosis.
In the present study, we demonstrated that miR-15a-5p and miR-21-5p are significantly overexpressed in a cohort of chemoresistant compared with chemosensitive cytogenetically normal AML patients with wild-type NPM1 without FLT3-ITD. Our gain-of-function experiments in AML cell lines showed that miR-15a-5p or miR-21-5p overexpression decreases apoptosis after daunorubicin and/or cytarabine treatment to almost reach the level of untreated cells. We found three interesting genes with a pro-apoptotic function. We confirmed that PDCD4, BTG2 and ARL2 are responsible for miR-15a-5p and miR-21-5p modulating effect on chemosensitivity of leukaemic cells to cytarabine and daunorubicin. The effect on chemosensitivity was stronger when the cells were treated with the combination of cytarabine and daunorubicin such as the treatment of the patients TA B L E 2 List of most down-regulated mRNAs after miR-15a-5p or miR-21-5p overexpression in K562 cells than with each drug alone. The most differentially down-regulated genes after miR-15a-5p or miR-21-5p overexpression in our AML cell lines treated with daunorubicin are PDCD4, BTG2 and ARL2. It seems that the expression of both miRNAs was positively correlated in our patient series supporting a synergistic role on the same target genes and on chemoresistance. No common mechanism of deregulation has been described for both miRNAs. Moreover, a negative correlation between the expression of both miRNAs and the three target genes was confirmed in patient samples. Our data suggest that several miRNAs can modulate chemosensitivity of myeloid leukaemia cells to cytarabine and/or daunorubicin by targeting the same genes.
Based on PCA performed on miRNA and gene expression, some patients could be phenotypically separated in chemoresistant or chemosensitive. Although we observed a tendency for phenotype separation, other patients could not be segregated as their expression profile for the selected genes and miRNAs was intermediate.
Mechanisms affecting their expression need to be further explored.
We cannot exclude additional mechanisms of chemoresistance in this group of unsegregated patients.
These three target genes have been previously linked to chemoresistance. PDCD4 (programmed cell death 4) is a tumour suppressor gene that not only inhibits proliferation, migration and invasion but also promotes apoptosis in tumours. 17 21-5p or SCR. The protein loading control was performed using β-actin. C, Bands were quantified by densitometry using Image J software. Data represent the average of three independent experiments ± SD. P values were obtained using a t test. * Indicates a significant difference P ≤ .05, and ** indicates a significant difference P ≤ .01. D, Luciferase activity in HEK293 cells cotransfected with synthetic miR-15a-5p, miR-21-5p or SCR and luciferase reporter constructs containing wild-type (WT) or mutated (miRNA_Δ) PDCD4, ARL2 and BTG2 3′UTR. Luciferase activities were determined at 24 h and were normalized using β-galactosidase activity. The mutant plasmids were generated by deleting the miR-15a-5p or miR-21-5p binding site(s)  miR-15a-5p and miR-21-5p could be used as future targets for therapy. Synthetic miRNAs are now involved in phase 1 and phase 2 clinical trials. For example, MRG-106 is an LNA antagomir that targets miR-155 and is now tested in the treatment of selected types of lymphoma or leukaemia. 34 Our results suggested a promising further therapeutic option that will combine anti-miR-15a-5p and/or

ACK N OWLED G EM ENTS
We thank Dr Nicolas Dauguet for expert flow cytometry support. Action#29.

CO N FLI C T O F I NTE R E S T
The authors confirm that there are no conflicts of interest.