The transcription factor HIF2α partakes in the differentiation block of acute myeloid leukemia

Abstract One of the defining features of acute myeloid leukemia (AML) is an arrest of myeloid differentiation whose molecular determinants are still poorly defined. Pharmacological removal of the differentiation block contributes to the cure of acute promyelocytic leukemia (APL) in the absence of cytotoxic chemotherapy, but this approach has not yet been translated to non‐APL AMLs. Here, by investigating the function of hypoxia‐inducible transcription factors HIF1α and HIF2α, we found that both genes exert oncogenic functions in AML and that HIF2α is a novel regulator of the AML differentiation block. Mechanistically, we found that HIF2α promotes the expression of transcriptional repressors that have been implicated in suppressing AML myeloid differentiation programs. Importantly, we positioned HIF2α under direct transcriptional control by the prodifferentiation agent all‐trans retinoic acid (ATRA) and demonstrated that HIF2α blockade cooperates with ATRA to trigger AML cell differentiation. In conclusion, we propose that HIF2α inhibition may open new therapeutic avenues for AML treatment by licensing blasts maturation and leukemia debulking.


Introduction
Acute myeloid leukemia (AML) is an aggressive disease characterized by uncontrolled proliferation and arrest of myeloid differentiation.AML is genetically heterogeneous, with different karyotypic aberrations, mutations, gene expression, and epigenetic profiles that define disease subsets and clonal populations within individual patients (Do ¨hner et al, 2017).The main therapeutic opportunity for AML patients consists of intensive chemotherapy and allogeneic hematopoietic cell transplantation for eligible candidates, with the recent introduction of novel targeted therapies for selected groups of patients.However, elderly patients cannot sustain overly toxic treatments, and younger patients who undergo remission upon standard therapies often relapse due to genetic plasticity of clonal AML populations and therapy-resistant leukemia stem cells (LSCs).For these reasons, AML survival is still discouragingly low, and new therapeutic options are urgently needed (Do ¨hner et al, 2017).
Blockade of myeloid differentiation is a common feature of AML, occurring at different stages of myeloid maturation, and generating morphological subsets that are only partly defined by genetic features.In some AML subsets, oncogenic drivers impose a block of differentiation by directly perturbing the expression of lineage commitment genes, as is the case of the PML-RARa fusion protein of acute promyelocytic leukemia (APL) (van Gils et al, 2017).In other instances, oncogenic transcriptional regulators affect expression of differentiation genes via epigenetic mechanisms (e.g., IDH or TET mutants; Figueroa et al, 2010).However, in most cases, the molecular underpinnings of the differentiation block remain to be elucidated, and it is not known if common regulatory mechanisms may exist across AML subsets.
Defining the details of halted differentiation is crucial not only to gain insights into AML pathogenesis but also to transform this feature into an actionable vulnerability.In this respect, the finding that all-trans retinoic acid (ATRA) triggers differentiation of APL blasts has been a turning point in AML therapy and has sparked considerable interest into translating ATRA treatment to other AML subsets.However, few non-APL AML subtypes undergo differentiation upon ATRA treatment (El Hajj et al, 2015;Ma et al, 2016;Verhagen et al, 2016;Mugoni et al, 2019), and it is hypothesized that most AMLs are resistant to ATRA-induced differentiation because of epigenetic silencing of myeloid differentiation genes (van Gils et al, 2017).
In this work, we identify the transcription factor HIF2a as a novel regulator of the AML differentiation block.Hypoxia-inducible factors (HIFs) are heterodimeric transcription factors composed of an inducible a and a constitutive b subunit.The two main a subunits, HIF1a and HIF2a, perform nonredundant functions and regulate different and cell type-specific target genes (Magliulo & Bernardi, 2018).The function of HIF factors has been widely studied in solid tumors, where they promote tumor progression by regulating cell metabolism, neo-angiogenesis, metastasis, and stem cell features (Wigerup et al, 2016).In AML, recent work has described HIF1a and HIF2a as either tumor promoters (Wang et al, 2011;Matsunaga et al, 2012;Rouault-Pierre et al, 2013;Coltella et al, 2014;Forristal et al, 2015;Gao et al, 2015;Migliavacca et al, 2016) or tumor suppressors (Velasco-Hernandez et al, 2014, 2019;Vukovic et al, 2015), a distinction that may be dictated by molecular specificities of leukemia subtypes, or different outputs of HIFs activity in normal hematopoietic progenitors versus leukemic cells (Magliulo & Bernardi, 2018).
Here, by comparing the activity of HIF1a and HIF2a in models of established AML, we confirmed that both play oncogenic functions, and uncovered a new role of HIF2a in hindering AML differentiation.Also, we found that HIF2a inhibition cooperates with ATRA to favor AML maturation.This finding has attractive therapeutic implications, as a small molecule inhibitor of HIF2a has been recently approved for patients with von-Hippel Lindau disease and is in clinical testing for other tumor types (Wallace et al, 2016;Courtney et al, 2018;Renfrow et al, 2018;Hasanov & Jonasch, 2021).Thus, we propose that HIF2a inhibition may be exploited as a new therapeutic avenue to treat AML.

HIF2a is a novel regulator of AML differentiation
To comparatively define the roles of HIF1a and HIF2a in AML, we interfered with their expression in cell lines representative of favorable (Kasumi1 and NB4 cells) or high-risk (Molm13 and THP1 cells) patients' categories, including cell lines conventionally used for differentiation studies (HL60 and NB4 cells).In accordance with previous observations (Kocabas et al, 2012;Schulz et al, 2012), silencing of either HIFa factor caused variable compensatory upregulation of the cognate gene (Figs 1A and EV1A).Phenotypically, we found that both HIFa factors promote proliferation and colony formation in all AML cell lines (Figs EV1B and 1B), confirming previous results that described their oncogenic function in established leukemia models (Wang et al, 2011;Rouault-Pierre et al, 2013;Coltella et al, 2014;Forristal et al, 2015;Gao et al, 2015;Migliavacca et al, 2016).Of note, efficacy of HIFa silencing was reduced upon cell passaging (Fig EV1C), suggesting that cells with higher shRNA were being counter selected in the population due to their decreased proliferation.For this reason, all experiments were performed in the first 10 passages upon shRNAs transduction.Thus, although EPAS1 (HIF2a gene) is expressed at lower levels than HIF1A in AML cell lines and patients (Fig EV1D and E), we confirmed that tampering HIF2a expression has important phenotypic consequences in AML, as previously observed by us and others (Rouault-Pierre et al, 2013;Coltella et al, 2014).
Intriguingly, for the first time, we observed that specific silencing of HIF2a promotes AML differentiation, as measured by surface expression of the myeloid differentiation marker CD11b (Fig 1C,Appendix Fig S1) and morphological changes of maturing myeloid cells such as nuclear multilobulation and reduced nucleus/cytoplasm ratio (Fig 1D).On note, the increase in CD11b + cells upon HIF2a silencing was variable and more modest in non-APL cell lines than in NB4 cells.
These results were confirmed with an additional set of short hairpin RNAs (shRNAs).Similar levels of HIF1a and HIF2a downregulation led to comparable inhibition of colony formation, while CD11b expression was induced only by HIF2a-specific silencing in both Kasumi1 and HL60 cells .
Taken together, these data confirm that both HIFa factors promote AML expansion and suggest that HIF2a is specifically involved in the AML differentiation block.
Interestingly, analysis of HIFa expression along normal hematopoiesis revealed that EPAS1 is predominantly expressed in hematopoietic stem and progenitor cells (HSPCs), with minimal expression in differentiated lineages, while HIF1A is also expressed in differentiated cells like monocytes and dendritic cells (Fig 1E), indicating that in hematopoiesis HIF2a functions are preferentially exerted in uncommitted progenitors.A Immunoblot analysis showing silencing efficiency of shRNAs against HIF1a, HIF2a, or a scrambled shRNA as control (shCTRL) in five AML cell lines at early passages upon retroviral infection (P5-10).a-tubulin was used as loading control.Densitometric analyses in the bottom boxes show relative levels of HIFa factors over control cells.Data are representative of one out of three independent experiments.B Colony forming capacity of indicated AML cell lines expressing shHIF1a, shHIF2a, or shCTRL.Shown is the average number of colonies/field in 20 fields (10× objective).Data represent mean AE SD of four biological replicates (Student's t-test).C Upper panel: Percentages of AML cells expressing the myeloid differentiation marker CD11b upon HIFa-specific silencing in the indicated cell lines.Lower panel: mean fluorescence intensity (MFI) of CD11b in the indicated cell lines.Data represent mean AE SD of four biological replicates (Student's t-test).D May-Grunwald Giemsa staining of shCTRL and shHIF2a Kasumi1 and NB4 cells.Scale bar, 20 lm (40× objective).Data are representative of one out of three independent experiments.Dot plots on the right indicate nucleus/cytoplasm ratio of shCTRL and shHIF2a Kasumi1 and NB4 cells, with each dot representing a single cell (n = 30, mean AE SD, Student's t-test).Areas of nucleus and cytoplasm were calculated using ImageJ software.E Hierarchical hematopoietic trees showing expression of HIF1A and EPAS1 genes in normal human hematopoiesis using HemaExplorer dataset (data obtained from BloodSpot; Bagger et al, 2016).
Source data are available online for this figure.) showed that HIF1a silencing predominantly perturbed the expression of glycolytic/metabolic pathways, thus confirming previously defined metabolic functions in hematopoiesis and AML (Wierenga et al, 2014(Wierenga et al, , 2019)).
Silencing of HIF2a caused larger gene expression perturbations (Appendix Fig S2B), with 20% of deregulated genes common to the two AML cell lines (Fig EV2B).To identify shared functions of HIF2a, we focused on gene sets concordantly regulated in HL60 and Kasumi1 cells.Functional enrichment analysis of 118 genes coupregulated after HIF2a depletion (Appendix Table S1) revealed that the most significant categories are centered on neutrophil maturation and activation (Fig 2A).Shared in these categories are integrins (ITGB2 and ITGAV) and other genes involved in myeloid differentiation (e.g., CD53, IFI16, MYD88, and CD4) ( In search of HIF2a target genes that may explain the phenotypic consequences of its inhibition, functional enrichment analysis was performed on the 74 genes co-downregulated upon HIF2a silencing (Appendix Table S2).Interestingly, the most significant gene categories are implicated in transcriptional regulation and include epigenetic regulators and chromatin organizers (Fig 2C).Regulation of these gene sets by HIF2a appears specific to AML cells as this was not observed in renal cancer, hematopoiesis, or a chronic myeloid leukemia (CML) cell line (Wierenga et al, 2014(Wierenga et al, , 2019;;Courtney et al, 2020).HIF2a-regulated genes comprehend known inducers of AML pathogenesis and proliferation (FLT3, CDK6, BCL11A, and RUNX2) (Gilliland & Griffin, 2002;Kuo et al, 2009;Scheicher et al, 2015;Sunami et al, 2022) and epigenetic regulators involved in cell fate determination and differentiation via heterochromatin formation (TRIM28 and UHRF1; Czerwi nska et al, 2017 ;Oleksiewicz et al, 2017;Zhao et al, 2017;Kim et al, 2018;Fig 2D).qPCR analysis confirmed HIF2a-mediated regulation of representative genes in additional AML cell lines, with the exception of NB4 cells where RUNX2, TRIM28, and UHRF1 were not regulated upon HIF2a silencing (Fig EV2C).
Enrichment of heterochromatin factors within genes induced by HIF2a is in line with a recently described function of HIF2a in modulating heterochromatin via EZH2 recruitment and H3K27me3mediated epigenetic silencing of specific target genes in macrophages (Li et al, 2021).Because it is generally assumed that the AML differentiation block is caused by epigenetic silencing of myeloid differentiation genes (Momparler et al, 2020)

I
Immunoblot analysis of HIF2a upon induction of exogenous HIF2a expression in Kasumi1 cells transduced with the X on system.UI: uninduced cells; I: induced cells.a-tubulin was used as a loading control.The blot represents one out of two independent experiments with similar results.J Proximity ligation assay with HIF2a and HIF1b antibodies in Kasumi1 cells transduced with the X on system.UI: uninduced cells; I: induced cells.Numbers of nuclear interaction foci/cell are represented (n = 80, three independent experiments, Student's t-test).K qPCR analysis of the indicated genes in Kasumi1 cells transduced with the X on system.UI: uninduced cells; I: induced cells.   .Thus, HIF2a silencing provokes a general deregulation of facultative heterochromatin that is linked to transcriptional regulation.Mechanistically, because genes belonging to the polycomb repressive complex 2 and histone demethylases were not in the HIF2a-regulated transcriptome, we hypothesize that the role of HIF2a in promoting H3K27me3 modifications in AML may be indirect.
Interestingly, analysis of chromatin accessibility revealed that HIF2a-regulated myeloid differentiation genes were not within the gene sets with increased chromatin accessibility upon HIF2a silencing (Appendix Fig S3G).This contrasts with transcriptional regulators including BCL11A and UHRF1, which are positively regulated by HIF2a and show coherent changes in chromatin accessibility (Appendix Fig S3H).Of note, the regulatory regions of representative myeloid differentiation genes revealed a state of open chromatin in control AML cells that was not modified by HIF2a silencing, a condition that was confirmed in primary AML cells (Gambacorta et al, 2022a) (Fig EV2F).Thus, these observations indicate that transcriptional repression of myeloid differentiation genes in AML cells is not always mediated by chromatin compaction at their regulatory regions.
In searching for HIF2a-regulated genes that may be directly implicated in blocking AML differentiation, we focused on Runx2 and BCL11A, which reportedly interfere with the expression of myeloid differentiation genes in AML via transcriptional repression or recruitment of co-repressor complexes (Kuo et al, 2009;Sunami et al, 2022).We confirmed that specific HIF2a silencing led to reduced Runx2 and BCL11A protein levels in Kasumi1 cells (Fig 2E).Hypoxia-responsive elements (HREs) were identified in the regulatory regions of both genes, and HIF2a was found associated to these genetic elements similarly to the bonafide HIF2a-target gene SCL7A5 (Elorza et al, 2012) (Fig 2F-H).Also, induction of exogenous HIF2a, which correlated with increased association with the obliged transcriptional partner HIF1b, confirmed increased HIF2a association to their regulatory regions and transcriptional upregulation (Fig 2I-L).
In conclusion, we found that in AML HIF2a regulates proleukemogenic factors that suppress myeloid differentiation via transcriptional repression.Concordantly, HIF2a inhibition unleashes expression of myeloid differentiation genes and directs AML cells towards a neutrophilic differentiation path.

HIF2a is required for leukemia progression in in vivo AML models
To validate and compare the functional consequences of HIF1a and HIF2a suppression in vivo, we utilized two AML patient-derived xenograft (PDX) models from leukemia samples collected at diagnosis (Toffalori et al, 2019).Leukemic cells expanded in immunodeficient mice were recovered from bone marrow, transduced ex vivo with lentiviral vectors containing specific shRNAs along with OFP (orange fluorescent protein) and reinoculated in recipient mice in a competitive assay between transduced and untransduced cells.AML-01 and AML-02 are representative of patients with adverse prognosis (i.e., patients carrying DNMT3A, NPM1, and FLT3 mutations; Appendix Fig S4A), but show different in vivo disease aggressiveness and lentiviral transduction efficiency (30% and 98%,respectively;Appendix Fig S4B).Overall, we observed that while HIF1a had minor effects on leukemia progression, reduction of HIF2a affected leukemia expansion in both PDX models, albeit in different compartments (Fig 3A).Importantly, analysis of OFP + cells revealed that cells with HIF2a suppression were at a competitive disadvantage compared to control-transduced cells, albeit for AML-02 the decrease in OFP + cells was not significant (Fig 3B).Surprisingly, analysis of myeloid differentiation at experimental endpoint did not reveal increased CD11b + cells upon HIF2a silencing (Appendix Fig S4C).However, we observed that OFP + cells from the bone marrow of leukemic mice had recovered HIFa expression when compared to gene silencing at preinoculation (Fig 3C), suggesting in vivo compensatory mechanisms of HIFa expression that are presently uncharacterized.
Relevance of HIF2a to AML pathogenesis was confirmed via HIF2a downregulation in AML cell lines.We selected Kasumi1 and Molm13 cells as representative of favorable and high-risk AML, respectively.Kasumi1 cells were transplanted subcutaneously (Li et al, 2017;Neldeborg et al, 2023), while Molm13 were injected intravenously (Migliavacca et al, 2016).HIF2a downregulation in Kasumi1 cells drastically reduced tumor progression (Fig 3D and E) and caused a modest and not-significant increase in CD11b expressing cells (Appendix Fig S4D).However, qPCR analysis revealed upregulation of representative myeloid differentiation genes (Fig 3F), indicating that a differentiation process was being triggered.Also, in line with in vitro data (Fig EV1C), HIF2a silencing was counter selected in vivo (Fig 3G).HIF2a downregulation also affected Molm13 expansion in vivo, particularly in spleen and peripheral blood (Fig 3H), which was accompanied by increased CD11b expressing cells in all compartments (Fig 3I) and upregulation of representative myeloid differentiation genes in bone marrow (Fig 3J).In addition, HIF2a silencing was counter selected also in this in vivo model (Fig 3K).
Taken together, these data indicate that HIF2a plays a prominent role in AML progression, and its inhibition exerts a significant antileukemic function.

Pharmacological inhibition of HIF2a induces AML differentiation
A specific small molecule inhibitor of HIF2a has been recently approved for von-Hippel Lindau disease and is being tested for renal cancer and glioblastoma (Wallace et al, 2016;Courtney et al, 2018;Renfrow et al, 2018;Hasanov & Jonasch, 2021).In all AML cell lines, 2-days treatment with increasing concentrations of PT2385 induced a dose-dependent reduction in cell counts (Figs 4A and EV3A) and a concordant increase in CD11b + cells (Figs 4B and EV3B)  Finally, an additional compound with reported HIF-inhibitory functions was tested in vivo.EZN-2208 is a polyethylene glycol conjugate of camptothecin (a topoisomerase inhibitor) that also inhibits HIFa factors (Rapisarda et al, 2004;Pastorino et al, 2010).Treatment of AML-01 and AML-02 engrafted mice with an established regimen of EZN-2208 that did not induce leukemia cell death (Fig EV4C ) affected leukemia progression and induced AML differentiation (Fig EV4D -F).Moreover, EZN-2208 inhibited HIF2a and not HIF1a, as measured by expression of HIFa factors and their regulated genes in AML cells recovered from treated mice (Fig EV4G).Therefore, although we cannot exclude that EZN-2208 exerts additional effects on other molecular targets, with these experiments we identified an additional anti-leukemic compound that triggers AML differentiation.
Taken together, our data demonstrate that HIF2a inhibitory molecule PT2385 recapitulates the induction of AML differentiation observed upon HIF2a knockdown and holds the potential of acting as a novel differentiation agent for AML treatment.

HIF2a is a direct target of ATRA receptors and its inhibition cooperates with ATRA towards AML differentiation
Ongoing efforts to enhance ATRA-induced differentiation and/or proliferation arrest in AML are aiming to combine ATRA with a ◀ Figure 3. HIF2a knockdown impairs leukemia progression and induces AML differentiation in vivo.broad repertoire of anti-leukemia drugs (Geoffroy et al, 2021).Based on the newly identified function of HIF2a, we speculated that HIF2a inhibition might cooperate with ATRA to promote AML differentiation.Accordingly, silencing of HIF2a significantly augmented ATRAinduced differentiation in all AML cell lines tested (Figs 5A and EV5A).This was confirmed by use of a low dose of PT2385, which in combination with ATRA provoked a dramatic increase in CD11b + cells and caused cell cycle arrest, as measured by accumulation of cells in G0/G1 and decreased S phase (Figs 5B and C,and EV5B and C).RNA sequencing upon HIF2a silencing and ATRA administration in Kasumi1 cells revealed upregulation of similar gene families, with an enrichment of myeloid maturation terms within the most significant gene ontologies (Fig 5D).Notably, combined ATRA and HIF2a silencing increased the number of genes within these families and the expression levels of concordantly regulated genes (Fig 5D and E), indicating that HIF2a inhibition and ATRA converge to stimulate the same pro-differentiation programs.Intriguingly, we observed that ATRA treatment caused a significant upregulation of HIF2a and not HIF1a in AML cell lines (Fig 6A and B).In investigating the molecular mechanism of this regulation, we found that both RARa and RARc bind the HIF2a promoter.In accordance with previous literature (Rochette-Egly & Germain, 2009), RARs binding occurred in the absence of ATRA stimulation (Fig 6C ) and was further increased upon ATRA treatment in a time-dependent manner (Fig 6D).These data show that HIF2a is a direct target of RAR transcription factors and is increasingly expressed upon ATRA administration.These results are in line with the reported induction of hypoxia/stress response genes upon ATRA treatment in a mouse model of AML1-ETO-driven AML (Chee et al, 2013).

A Percentages of leukemic cells (hCD45
Because it has been suggested that in AML ATRA may induce differentiation while also promoting self-renewal of leukemic blasts (Geoffroy et al, 2021), we wondered whether HIF2a induction may provide a self-renewal signal downstream ATRA, which is hampered by HIF2a inhibition.Along these lines, by overlapping genes that were induced by ATRA and reduced upon HIF2a silencing in Kasumi1 cells (Fig 6E), we found several genes implicated in selfrenewal mechanisms in HSCs, such as PML, GFI1, KDM6B and EPAS1 itself (Ito et al, 2008;Rouault-Pierre et al, 2013;Mallaney et al, 2019;Mo ¨ro ¨y & Khandanpour, 2019).Increased expression of these genes in Kasumi1 cells treated with ATRA was abolished by concomitant HIF2a inhibition (Fig 6F).
Taken together, our data suggest that targeting HIF2a cooperates with ATRA for differentiation induction and removes a negative feedback loop of HIF2a upregulation that may be implicated in promoting AML self-renewal, thus potentiating ATRA-based therapies in AML.

Discussion
In this work, we place the transcription factor HIF2a within the molecular circuitry of the AML differentiation block and propose that HIF2a inhibition may add therapeutic efficacy to differentiation therapy for AML treatment thus broadening the therapeutic horizon of ATRA beyond APL.
Recent work has implicated HIF transcription factors as either tumor promoters or tumor suppressors in AML, via several studies performed with different experimental approaches in distinct AML subsets (Wang et al, 2011;Matsunaga et al, 2012;Rouault-Pierre et al, 2013;Coltella et al, 2014;Velasco-Hernandez et al, 2014, 2019;Forristal et al, 2015;Gao et al, 2015;Vukovic et al, 2015;Migliavacca et al, 2016).To reconcile this apparent contradiction, we speculated that HIFa factors may exert different functions in specific AML subsets, or at different stages of leukemia development (Magliulo & Bernardi, 2018).Here, we found that both HIF1a and HIF2a exert leukemia-promoting functions in models of established AML (i.e., cell lines and PDX).Importantly, we described a new and specific involvement of HIF2a in blocking AML differentiation.Mechanistically, we observed that HIF2a promotes expression of transcriptional repressors/corepressors (i.e., RUNX2 and BCL11A) that suppress myeloid differentiation genes in AML (Kuo et al, 2009;Sunami et al, 2022), suggesting that differentiation blockade by HIF2a occurs via regulation of transcriptional repressive programs.
Of note, our in vitro experiments were performed in normoxia, thus indicating that HIF factors play important functions in experimental settings where they are not stabilized by low oxygen conditioning.Similar results were obtained by other investigators in hematopoietic progenitors, where HIF2a silencing impacted proliferation and colony formation in normoxic conditions (Rouault-Pierre et al, 2013).Notably, by comparing the functions of HIF1a and HIF2a, these investigators identified HIF2a as the main regulator of self-renewal in human long-term repopulating hematopoietic progenitors (Rouault-Pierre et al, 2013), an observation that is consistent with HIF2a being expressed specifically in hematopoietic stem and progenitor cells.Therefore, we speculate that the function of HIF2a in the AML differentiation arrest may reflect its physiological function in normal hematopoiesis.In this respect, it is notable that HIF2a is not mutated or overexpressed in AML, thus further 12 of 19 With our work, we suggest that HIF2a partakes to the mechanisms of ATRA-induced self-renewal, as we found that ATRA directly induces HIF2a expression and HIF2a in turn promotes the expression of ATRA-regulated genes implicated in self-renewal of hematopoietic stem cells (PML, KDM6B, and GFI1).Once again, we hypothesize that this molecular circuit may reflect a functional crosstalk that exists in normal hematopoiesis, where RAR transcription factors and HIF signaling may cooperate to promote selfrenewal.In addition, we observed that HIF2a inhibition converges onto ATRA transcriptional outputs by increasing expression of gene sets linked to myeloid differentiation.In conclusion, we propose that HIF2a inhibition may add therapeutic value to ATRA-based therapies via a dual mechanism that favors myeloid differentiation, whilst reducing self-renewal.

Lentiviral infection
Human AML cell lines were spinoculated in medium containing concentrated viral supernatant, 8 lg/ml polybrene, and Hepes 1 M pH 7.4 pH for 90 min at 1,200 g at 30°C.After 24 h, fresh medium was added, and cells were allowed to recover for 48 h before antibiotic selection.Optimized puromycin (Sigma, Cat# P8833) concentrations were: 2 lg/ml for Kasumi1, NB4 and Molm13 cells, 6 lg/ml for HL60 cells, and 7 lg/ml for THP1 cells.Experiments were conducted with bulk populations.
For infection of Kasumi1 cells with the X on -HIF2a expressing vector, cells were plated in 6-well plates at 2.5 × 10 5 cells/ml and transduced with concentrated viral supernatant.Mice were sacrificed at the end of treatment.For in vivo experiments with Kasumi1 shCTRL and shHIF2a transduced cells, 1.5 × 10 6 cells were injected subcutaneously into the flanks of 6-8 weeks old NSG recipient mice.Tumor progression was measured every 3/4 days using the caliper method and the formula V (mm 3 ) = (width × length) 2 × p/6.Mice where sacrificed at 20 days from injection.For in vivo experiments with Molm13 shCTRL and shHIF2a transduced cells, 5 × 10 6 cells were injected intravenously into 6-8 weeks old NSG recipient mice.Mice were sacrificed at 20 days from injection.
PT2385 was added at the indicated concentration and time points.

Cell proliferation
For HIFa-silenced AML cells, 1 × 10 4 cells were seeded in 24-well plates in technical triplicates, and their growth and viability was evaluated by trypan-blue exclusion assay.Cells were counted every 24 h for 4-5 consecutive days, and cell proliferation ratio was calculated as the mean value of triplicates compared to day 0. For PT2385 treatment, 3 × 10 5 cells were seeded in 24-well plates in triplicates, and their growth and viability was evaluated after 48 h by trypan-blue exclusion assay.
Methylcellulose colony-forming assay 5 × 10 3 cells were resuspended in human methylcellulose base media and cell resuspension solution (R&D Systems, Cat# HSC002) according to manufacturer's instructions and plated in technical duplicates in 6-wells with water supply in the inter-well chamber to prevent evaporation.After 5-7 days, colonies were counted blindly in 20 fields per condition using standard light microscopy (Zeiss Axiovert 40C, 10× objective).

May-Grunwald Giemsa (MGG) staining
1 × 10 5 cells were resuspended in PBS with 10% fetal bovine serum (FBS) and centrifuged on slides by cytospin at 500 rpm for 5 min.For MGG staining, cells were stained by May-Grunwald and Giemsa dyes.After drying and mounting, cellular morphology was examined with AxioImager M2m microscope, 40× objective (Carl Zeiss).To obtain nucleus/cytoplasm ratio, areas of cytoplasm and nucleus were calculated for 30 cells/condition, using ImageJ software (v1.53e,National Institutes of Health).

Cell cycle analysis
1 × 10 4 cells were fixed in 70% cold ethanol and stored at À20°C overnight.After fixation, cells were centrifuged at 3,100 g for 2 min and washed once with PBS.After centrifugation, cells where permeabilized with PBS Triton X-100 0.25% for 15 min on ice and then washed once with PBS.DNA was stained with 20 lg/mL PI (Merck, Cat# P4864) and RNA was digested with RNaseI 10 lg/ml (ThermoFisher, Cat# 12091021).DNA content was measured with the BD FACSCanto II (Becton Dickinson) and analysis was performed with FCS Express 7 Research software.

Quantitative PCR (qPCR)
Total RNA from AML cell lines was isolated using RNeasy mini Kit (Qiagen).Total RNA from PDX-derived cells was isolated using Relia-Prep RNA Cell Miniprep System (Promega).Equal amounts of RNA were reverse transcribed into cDNA with Advantage RT-for-PCR Kit (Clontech) and analyzed by qPCR using a 7900 Fast-Real Time PCR System (Applied Biosystem).Probes for TaqMan assays were purchased from Applied Biosystem (sequences are provided in Appendix Table S3).Each sample was evaluated in technical triplicates, and data were normalized to 18s gene.Relative expression was calculated using the comparative threshold cycle method (2 ÀDDCt ), except for assessing basal gene expression where the 2 ÀDCt was used.

Chromatin immunoprecipitation (ChIP) qPCR, ChIP sequencing and data analysis
ChIP experiments were performed as previously described (Cabianca et al, 2012).50-100 lg of chromatin were used for ChIP of HIF2a and FLAG, whereas 10 lg of chromatin were used for H3K27me3 ChIP-seq.For ChIP-qPCR experiments, GoTaq qPCR Master Mix (Promega) was used to amplify DNA fragments.To measure enrichment, qPCR values were normalized over input.

RNA sequencing and data analysis
For RNA sequencing analysis, specific silencing of HIF1a and HIF2a and absence of compensatory upregulation of HIFa subunits was evaluated by qPCR in HL60 and Kasumi1 cells stably expressing shCTRL, shHIF1a or shHIF2a.For RNA sequencing upon combination of HIFa inhibition and ATRA, Kasumi1 cells were treated with 1 lM ATRA for 24 h.RNA sequencing experiments are representative of two independent experiments performed upon different lentiviral infections.Each sample was processed as follows: (i) total RNA was isolated from 1 to 3 × 10 6 cells with QIAGEN RNeasy Plus Micro Kit, according to manufacturer's instructions.(ii) RNA was treated with DNAse I (Sigma, D5307), according to manufacturer's instructions.(iii) RNA quality was evaluated with a 2100 Bioanalyzer (Agilent) to select RNA with a RIN above 9.TruSeq stranded mRNA protocol was used for 5 0 /3 0 library preparation starting from 100 ng of total RNA.Libraries were barcoded, pooled and sequenced on an Illumina Nova-Seq 6000 sequencing system.For each run, RNA sequencing experiments were performed generating 30 M single-end reads, 100 nucleotide long.After trimming, sequences were aligned using the STAR aligner (Dobin et al, 2013) to human reference genome GRCh38, and counted with feature-Counts (Liao et al, 2014) on the last Gencode (Harrow et al, 2012) release for RNA sequencing.Differential gene expression was evaluated in R/BioConductor (Huber et al, 2015) using the DESeq2 package (Love et al, 2014).A significant threshold of 0.05, adjusting the P-value by FDR (False Discovery Rate) was established to identify differentially expressed genes.Functional enrichment analysis were performed using Enrichr (Kuleshov et al, 2016).

ATAC sequencing and data analysis
ATAC sequencing experiments are representative of three experimental replicates.A total of 6 × 10 5 cells were lysed with digitonin (Promega, Cat# G944A) and tagmented with an engineered Tn5 transposase (Illumina, Cat# 15027865) at 37°C for 30 min, following a protocol optimized for blood cells (Corces et al, 2016).Tagmented DNA was purified using the MinElute Reaction Cleanup kit (Qiagen) and then amplified with 10 cycles of PCR.Before sequencing, fragments with a 1-5 kb size range were removed by magnetic separation with AMPure XP beads (Beckman Coulter, Cat# A63881).DNA concentration was measured with the Qubit fluorometer (Life Technologies), and quality of samples' enrichment was assessed using Agilent TapeStation system.Sequencing was performed using Illumina High throughput Sequencing technology (NovaSeq 6000).Raw reads were trimmed using the software BBDuck.Reads were aligned to the human genome assembly (GRCh38) using the BWA software with standard parameters, and uniquely mapped reads were selected with MarkDuplicates from Picard Tools [http:// broadinstitute.github.io/picard/].Further filtering was done on reads mapping in regions present in the ENCODE hg38 blacklist (Amemiya et al, 2019).ChIP read counts were normalized to library size using the reads per genome coverage (RPGC) function in Deeptools v3.5.1 and mean among replicates was calculated using wiggletools v1.2.Bigwig files for normalized read counts were visualized using Integrative Genomics Viewer (Robinson et al, 2011).Peaks were called with MACS2 v2.2.7.1.Intersects and unique peaks were determined using BEDOPS v2.4.41 (https://github.com/bedops/bedops/releases/tag/v2.4.) and profile plots were computed with Deeptools (https://doi.org/10.1093/nar/gkw257).Gene annotation was performed with GREAT (PMID 20436461) with Two nearest genes association rule settings.

Statistical analysis
Animals were randomized into different treatment groups such that leukemia engraftment was similar between the groups.The experiments were conducted as non-blind tests and no mice were excluded from the experiments.One-way ANOVA was used for comparison of three or more groups, with the addition of post-hoc Tukey's multiple comparison test.Two-sided Student's t-test was used for comparison of two groups.All data are expressed as means AE standard deviations (SD), and significance is indicated with exact P-value in the figures.Data were processed using GraphPad Prism version 9.0.2 (GraphPad Software, San Diego,

The paper explained Problem
Acute myeloid leukemia (AML) is an aggressive disease affecting blood cells of the myeloid lineage.AML patients have a 5-year overall survival rate of less than 30%, and new therapeutic strategies are urgently needed to improve this grim prognosis.In AML, uncontrolled cell proliferation is intertwined with differentiation arrest, that is the inability of cells with a progenitor phenotype to undergo differentiation, to mature and self-exhausting myeloid cells.The molecular basis of this differentiation arrest is complex and a matter of ongoing investigation.

Results
We found that the transcription factor HIF2a, a gene that evolved to adapt cellular physiology to variations in oxygen tension, partakes to the AML differentiation block.We identify important transcriptional regulators and suppressors of myeloid differentiation in the HIF2aregulated transcriptome and demonstrate that inhibiting HIF2a via genetic or pharmacological manipulation prompts AML differentiation, induces leukemia debulking, and potentiates the effect of all-trans retinoic acid (ATRA), a compound that has revolutionized the treatment of acute promyelocytic leukemia.

Impact
This study adds new insights into the molecular mechanisms that suppress differentiation programs in AML and proposes a novel therapeutic strategy for leukemia debulking via HIF2a inhibition.Because a small molecule inhibitor of HIF2a has been recently generated and is entering the clinic for solid cancers, this work sets the basis for extending the use of this compound to another disease in need of additional therapies and with the therapeutic endpoint of cell exhaustion via differentiation, rather than the conventional cytotoxic or cytostatic activity of anticancer agents.A qPCR analysis of HIF1a and HIF2a in the indicated AML cell lines stably expressing shRNAs against HIF1a or HIF2a or a scrambled shRNA as control (shCTRL).Data are expressed as fold change in HIFa-silenced cells compared to shCTRL cells.Data represent mean AE SD of three biological replicates (Student's t-test).B Cell proliferation of the indicated AML cell lines carrying shHIF1a, shHIF2a, or shCTRL.Values represent cell numbers normalized over day 0. Data represent mean AE SD of three biological replicates (Student's t-test).C qPCR analysis of HIF1a and HIF2a in HL60 and Kasumi1 cells stably expressing shCTRL, shHIF1a or shHIF2a.Data are expressed as fold change in HIFa-silenced cells compared to shCTRL cells.Data represent mean AE SD of three biological replicates of cells at passages 4-10 (< P10) or passages 11-16 (> P10) (Student's t-test).D Analysis of HIF1A and EPAS1 (encoding HIF1a and HIF2a respectively) basal expression in the human AML cell lines utilized in this study.Data represent mean AE SD of three biological replicates.E mRNA expression of HIF1A and EPAS1 in 451 AML patients from the Oregon Health & Science University (OSHU) dataset (Tyner et al, 2018).Data are expressed as normalized RPKM (Reads Per Kilobase Million), and were obtained from the cBioportal database (Cerami et al, 2012).Data represent mean AE SD of 451 biological replicates.F Immunoblot analysis showing silencing efficiency of two independent shRNAs against HIF1a (shHIF1a#1 and shHIF1a#2) and HIF2a (shHIF2a#1 and shHIF2a#2), or a scrambled shRNA as control (shCTRL) in Kasumi1 cells.shHIF1a#1 and shHIF2a#2 are shRNAs utilized in the main figures.a-tubulin was used as loading control.G qPCR analysis of HIF1a and HIF2a upon HIFa-specific silencing and compared to shCTRL in Kasumi1 (left graph) and HL60 (right graph) cells.Data represent mean AE SD of three biological replicate (Student's t-test).H Colony forming capacity of Kasumi1 (left graph) and HL60 (right graph) cells expressing shHIF1a#1, shHIF1a#2, shHIF2a#1, shHIF2a#2, or shCTRL.Shown is the average number of colonies/field in 20 fields (10x objective).Data represent mean AE SD of three biological replicates (Student's t-test).I Percentages of CD11b + in cells described in (H).Data represent mean AE SD of three biological replicates (Student's t-test).Relative expression  CD11b (hCD11b + CD45 + ) in Kasumi1 cells carrying shCTRL (n=6) or shHIF2a (n=5) injected subcutaneously.

EMBO Molecular Medicine
Appendix Table S1
Fig 2B), which are induced upon HIF2a and not HIF1a silencing, thus confirming morphological and immunophenotypical differentiation features shown in Fig 1.

Figure 2 .
Figure 2. HIF2a suppresses expression of myeloid differentiation genes and promotes transcriptional repressors and leukemogenic factors.A-D (A, C) Gene set enrichment analysis of differentially expressed genes (DEGs; significant threshold of 0.05, adjusted P-value by False Discovery Rate) commonly upregulated (A) and downregulated (C) in HL60 and Kasumi1 cells upon HIF2a silencing.Indicated are the terms most significantly enriched in the following libraries: gene ontology (GO) biological process, GO molecular function, GO cellular component, Bioplanet, Reactome, and Hallmarks of cancer.Dot sizes represent the number of genes in each term, and colors indicate Enrichment Scores expressed as Àlog 10 (P-value).(B, D) Heatmaps of commonly upregulated (B) and downregulated (D) genes within the terms most significantly enriched.The red-blue color scale reflects normalized RPKM (Reads Per Kilobase Million), with red indicating genes with higher expression and blue indicating genes with lower expression.Asterisks indicate genes that are mentioned in the main text.Results for each cell line represent the average of two independent experiments.E Immunoblot analysis of BCL11A and Runx2 upon HIFa-specific silencing in Kasumi1 cells.Vinculin was used as a loading control.The blot represents one out of three independent experiments with similar results.F Schematic view of HREs location in the regulatory regions of BCL11A and RUNX2 genes.HREs positions are numbered relative to annotated promoters (green boxes).HRE consensus sequence was obtained from MotifMap (motif ID: M01249).G ChIP-qPCR for HIF2a with primer pairs amplifying the HREs of BCL11A and RUNX2 and the positive control SLC7A5 gene in Kasumi1 cells.IgG was used as negative control.Results are represented as percentage of enrichment over input and represent mean AE SD of three biological replicates (Student's t-test).H ChIP-qPCR for HIF2a with primer pairs amplifying the HREs of BCL11A and RUNX2 and the positive control SLC7A5 gene in shCTRL and shHIF2a Kasumi1 cells.Data were normalized over input and control IgG and presented as fold enrichment over control cells.ChIP-qPCR data represent mean AE SD of three biological replicates (Student's t-test).IImmunoblot analysis of HIF2a upon induction of exogenous HIF2a expression in Kasumi1 cells transduced with the X on system.UI: uninduced cells; I: induced cells.a-tubulin was used as a loading control.The blot represents one out of two independent experiments with similar results.J Proximity ligation assay with HIF2a and HIF1b antibodies in Kasumi1 cells transduced with the X on system.UI: uninduced cells; I: induced cells.Numbers of nuclear interaction foci/cell are represented (n = 80, three independent experiments, Student's t-test).K qPCR analysis of the indicated genes in Kasumi1 cells transduced with the X on system.UI: uninduced cells; I: induced cells.Values are represented as fold change in gene expression compared to uninduced cells.Data represent mean AE SD of three biological replicates (Student's t-test).L ChIP-qPCR for HIF2a with primer pairs amplifying the HREs of BCL11A and RUNX2 and the positive control SLC7A5 gene in Kasumi1 cells transduced with the X on system.UI: uninduced cells; I: induced cells.Data were normalized over input and control IgG and presented as fold enrichment over control cells.Data represent mean AE SD of three biological replicates (Student's t-test).
+ ) in the bone marrow (BM), spleen (SP), and peripheral blood (PB) of mice injected with cells derived from AML-01 (n = 6, upper panel) and AML-02 (n = 5, lower panel) PDX and carrying shCTRL, shHIF1a, or shHIF2a.Data are represented in box and whisker plots where the central band denotes the median value, box contains interquartile ranges, while whiskers mark minimum and maximum values.All biological replicates are shown (n = 5/6, Student's t-test).B Percentages of leukemic cells expressing the OFP marker (OFP + hCD45 + ) in mice described in (A).Data are represented in box and whisker plots where the central band denotes the median value, box contains interquartile ranges, while whiskers mark minimum and maximum values.All biological replicates are shown (n = 5/6, Student's t-test).C qPCR of HIF1a (upper panel) and HIF2a (lower panel) genes in pre-inoculated (Pre) AML-01 and AML-02 cells transduced with shCTRL, shHIF1a, and shHIF2a and in cells isolated from the bone marrow of transplanted mice at experimental endpoint (Post).Values indicate fold changes in gene expression compared to shCTRL cells.Data represent mean AE SD of three biological replicates (Student's t-test).D Tumor progression of Kasumi1 cells expressing shCTRL or shHIF2a injected subcutaneously.Tumor volumes were measured at indicated days (D) upon injection.Data represent mean AE SD of six biological replicates (Student's t-test).E Tumor weights of mice described in (D) at experimental endpoint (D20).Data are represented in box and whisker plots where the central band denotes the median value, box contains interquartile ranges, while whiskers mark minimum and maximum values.All biological replicates are shown (n = 6, Student's t-test).Lower panel, photograph of excised tumors.N/D: not detected.F qPCR analysis of the indicated representative myeloid differentiation genes in Kasumi1 shCTRL or shHIF2a tumors.Values are represented as fold change in gene expression compared to shCTRL.Data represent mean AE SD of three biological replicates (Student's t-test).G qPCR of HIF2a in pre-inoculated (Pre) Kasumi1 shCTRL or shHIF2a cells and in cells isolated from Kasumi1 shCTRL or shHIF2a tumors (Post).Values indicate fold changes in gene expression compared to shCTRL cells.Data represent mean AE SD of three biological replicates (Student's t-test).H Percentages of leukemic cells (hCD45 + ) in the bone marrow (BM), spleen (SP), and peripheral blood (PB) of mice injected intravenously with Molm13 cells with shCTRL or shHIF2a and sacrificed at day 20 post-injection.Data represent mean AE SD of five biological replicates (Student's t-test).I Percentages of Molm13 leukemic cells expressing CD11b (hCD11b + hCD45 + ) in mice described in (H).Data represent mean AE SD of five biological replicates (Student's t-test).J qPCR analysis of the indicated representative myeloid differentiation genes in leukemic Molm13 shCTRL or shHIF2a cells recovered from bone marrow.Values are represented as fold change in gene expression compared to shCTRL cells.Data represent mean AE SD of three biological replicates (Student's t-test).K qPCR of HIF2a in pre-inoculated (Pre) Molm13 shCTRL or shHIF2a cells and in cells isolated from bone marrow at experimental endpoint (Post).Values indicate fold changes in gene expression compared to shCTRL cells.Data represent mean AE SD of three biological replicates (Student's t-test).Source data are available online for this figure.

▸Figure 4 .
Figure 4. Specific inhibition of HIF2a by PT2385 promotes AML differentiation and impairs leukemia progression.A Cell numbers of HL60 and Kasumi1 cells 2 days after treatment with PT2385 at the indicated doses and compared to vehicle treated cells.Data represent mean AE SD of three biological replicates (one-way ANOVA followed by Tukey's multiple comparison test).B Percentages of CD11b + HL60 and Kasumi1 cells 2 days after treatment with PT2385 at the indicated doses and compared to vehicle treated cells.Data represent mean AE SD of three biological replicates (one-way ANOVA followed by Tukey's multiple comparison test).C qPCR analysis of the indicated genes in HL60 and Kasumi1 cells 2 days after treatment with 200 lM PT2385.Values are represented as fold change in gene expression compared to vehicle treated cells.Data represent mean AE SD of three biological replicates (Student's t-test).D Percentages of CD11b + HL60 and Kasumi1 cells at the indicated days (D2, D4, D6) after treatment with 50 lM PT2385 and compared to vehicle treated cells.Data represent mean AE SD of three biological replicates (Student's t-test).E Percentage of leukemic cells (hCD45 + ) in the bone marrow (BM) of mice injected with AML-01 cells and treated with 100 mg/kg PT2385 or vehicle.Data represent mean AE SD of four biological replicates (Student's t-test).F Spleen weights (left graph) and percentages of leukemic cells (hCD45 + , right graph) in the spleen (SP) of mice injected with AML-01 cells and treated with 100 mg/kg PT2385 or vehicle.Data represent mean AE SD of four biological replicates (Student's t-test).G Percentages of leukemic cells (hCD45 + ) in the peripheral blood (PB) of mice injected with AML-01 cells and treated with 100 mg/kg PT2385 or vehicle.Data represent mean AE SD of four biological replicates (Student's t-test).H-J Percentages of leukemic (hCD45 + ) cells expressing CD11b, CD15 and CD14 in the bone marrow (BM; H), spleen (SP; I) and peripheral blood (PB; J) of mice injected with AML-01 cells and treated with 100 mg/kg PT2385 or vehicle.Data represent mean AE SD of four biological replicates (Student's t-test).

▸Figure 5 .Figure 6 .
Figure 5. HIF2a inhibition cooperates with ATRA to promote AML differentiation.A Percentages of CD11b + HL60 and Kasumi1 cells with shRNAs against HIF1a, HIF2a, or a scrambled shRNA as control (shCTRL) treated with 1 lM ATRA for 2 days.Data represent mean AE SD of three biological replicates (one-way ANOVA followed by Tukey's multiple comparison test).B Percentages of CD11b + HL60 and Kasumi1 cells following treatment with 50 lM PT2385, 1 lM ATRA, or combination for 4 days.Data represent mean AE SD of three biological replicates (one-way ANOVA followed by Tukey's multiple comparison test).C Percentages of HL60 (left graph) and Kasumi1 (right graph) cells in the indicated phases of the cell cycle following treatment with 50 lM PT2385, 1 lM ATRA, or combination for 4 days.Data represent mean AE SD of three biological replicates (Student's t-test).D List of top common upregulated Gene Ontology (GO) terms in Kasumi1 cells upon shHIF2a, 1 lM ATRA treatment or combination with respect to shCTRL cells.Dot sizes represent the number of genes in each term, and colors indicate experimental conditions shown in legend.E Venn diagram indicating the overlap of commonly upregulated genes (38 genes), which are represented in terms shown in (D).Violin plot indicating fold induction of each of the 38 genes commonly upregulated in each condition.Values represent the Log 2 (FoldChange) with respect to shCTRL cells.Data indicate fold enrichment over control cells (Student's t-test).Source data are available online for this figure.

Figure EV1 .
Figure EV1.Expression and silencing of HIF1a and HIF2a in AML cell lines.
+ HL60 and Kasumi1 cells at the indicated days (D2, D4, D6) after treatment with 50 lM PT2385 and compared to vehicle treated cells.Data represent mean AE SD of three biological replicates (Student's t-test).E Percentage of leukemic cells (hCD45 + ) in the bone marrow (BM) of mice injected with AML-01 cells and treated with 100 mg/kg PT2385 or vehicle.Data represent mean AE SD of four biological replicates (Student's t-test).F Spleen weights (left graph) and percentages of leukemic cells (hCD45 + , right graph) in the spleen (SP) of mice injected with AML-01 cells and treated with 100 mg/kg PT2385 or vehicle.Data represent mean AE SD of four biological replicates (Student's t-test).G Percentages of leukemic cells (hCD45 + ) in the peripheral blood (PB) of mice injected with AML-01 cells and treated with 100 mg/kg PT2385 or vehicle.Data represent mean AE SD of four biological replicates (Student's t-test).H-J Percentages of leukemic (hCD45 + ) cells expressing CD11b, CD15 and CD14 in the bone marrow (BM; H), spleen (SP; I) and peripheral blood (PB; J) of mice injected with AML-01 cells and treated with 100 mg/kg PT2385 or vehicle.Data represent mean AE SD of four biological replicates (Student's t-test).Source data are available online for this figure.
(Chee et al, 2013)13;Geoffroy et al, 2021) is not the main contributor to ATRA efficacy in APL, where spontaneous reversal of the differentiated phenotype has been documented(McKenzie et al, 2019)and combined treatment with arsenic trioxide (ATO) is necessary to target the driver mutation PML-RARa and eliminate LSCs(Lo-Coco et al, 2013;Geoffroy et al, 2021); iii) beside promoting differentiation, ATRA increases self-renewal of stem cells in normal hematopoiesis by regulating RARa or RARc respectively, and a similar antagonistic function has been observed also in AML1-ETO transformed progenitors(Chee et al, 2013).Taken together, these observations suggest that further mechanistic investigations are needed to drive clinical application of ATRA-based therapies for AML treatment.