Hif-1α and Hif-2α synergize to suppress AML development but are dispensable for disease maintenance

To investigate the requirement for Hif-2α in leukemogenesis, we used a well-characterized mouse model of AML in which the development and maintenance of LSCs is driven by coexpression of Meis1 and Hoxa9 oncogenes (Lessard and Sauvageau, 2003; Wang et al., 2010; Lehnertz et al., 2014). These two oncogenes are frequently overexpressed in several human AML subtypes (Lawrence et al., 1999; Drabkin et al., 2002) and their overexpression in mouse hematopoietic stem and progenitor cells (HSPCs) promotes self-renewal and perturbs their differentiation, resulting in the generation of self-renewing LSCs (Kroon et al., 1998). In the Meis1/Hoxa9 model used in this study, the BM c-Kit⁺ HSPC population, which contains all of the cellular targets for leukemic transformation, is transduced with retroviruses expressing Meis1 and Hoxa9. After three rounds of replating, it gives rise to the population of preleukemic cells that, upon transplantation to primary recipients, generates LSCs, thereby causing AML with a long latency (Fig. 1 A). LSCs are defined by their capacity to propagate AML with short latency in secondary recipients (Fig. 1 A). To establish the functional significance of Hif-2α in leukemogenesis, we used mice with Vav-iCre–mediated hematopoiesis-specific deletion of Hif-2α (Hif-2α^fl/fl;Vav-iCre mice), which we previously demonstrated to have normal numbers of hematopoietic stem cells (HSCs) and progenitor cells and display no hematopoietic defects (Guitart et al., 2013). We transduced c-Kit⁺ cells from Hif-2α^fl/fl;Vav-iCre and control (Hif-2α^fl/fl without Vav-iCre) mice with Meis1/Hoxa9 retroviruses and serially replated them under normoxic and hypoxic conditions. Hif-2α–deficient and control cells displayed similar replating capacity and generated comparable numbers of compact colonies at each passage (Fig. 1, B and C). Operetta high-content automated microscope analyses revealed that the colonies of both genotypes had comparable sizes (unpublished data). Flow cytometry analyses of the CFC3 colonies using c-Kit, Mac-1, and Gr-1 markers revealed no differences in immunophenotypic composition of colonies of the two genotypes (unpublished data).

Proliferation assays and cell cycle analyses on Hif-2α–deficient preleukemic cells (i.e., cells obtained after the third round of replating) cultured under hypoxia and normoxia revealed that Hif-2α–deficient cells had increased proliferative capacity (Fig. 1 D) and that a larger proportion of Hif-2α–deficient cells were in S and G2/M phases of cell cycle compared with control cells (Fig. 1 E). Thus, Hif-2α is not required for the generation of preleukemic cells but restricts their proliferative capacity.

To investigate the impact of Hif-2α deletion on LSC generation, we transplanted Hif-2α–deficient and control preleukemic cells generated under normoxic conditions into primary recipient mice (Fig. 1 A). Blood sampling revealed that a significantly smaller proportion of recipients of Hif-2α–deficient cells remained leukemia-free compared with recipients of control preleukemic cells (Fig. 1 F). Consequently, recipients of Hif-2α–deficient cells had decreased survival compared with recipients of control preleukemic cells (Fig. 1 G). Analyses of BM from leukemic primary recipient mice revealed that Meis1/Hoxa9-transduced control and Hif-2α^fl/fl;Vav-iCre preleukemic cells generate leukemia with the same immunophenotypic characteristics (unpublished data). In conclusion, Hif-2α suppresses the establishment of LSCs and delays the onset of Meis1/Hoxa9-induced AML.

To investigate the ability of Hif-2α–deficient LSCs to propagate leukemia, we sorted the CD45.2⁺ donor-derived c-Kit⁺ cell population (that contains LSCs; Somervaille et al., 2009) from the two cohorts of leukemic primary recipients and transplanted equal numbers of c-Kit⁺ cells into secondary recipients. LSCs of both genotypes generated aggressive AML with similar latency (Fig. 1 H). Therefore, Hif-2α is not required for the maintenance of LSCs and their ability to propagate Meis1/Hoxa9–induced AML.

We next set out to confirm the ability of Hif-2α to suppress the development of AML in the Mll-AF9 knock-in (Mll-AF9^KI/+) mouse model of human AML (Dobson et al., 1999; Chen et al., 2008; Kumar et al., 2009). This model is thought to closely mimic the human disease, as each cell contains a single copy of the Mll-AF9 fusion oncogene, physiologically expressed from the endogenous Mll promoter. In this model, Mll-AF9^KI/+ Lin⁻Sca-1⁺c-Kit⁺ (LSK) stem and primitive progenitor cells are sufficient to initiate leukemia with long latency upon transplantation. We generated Mll-AF9^KI/+;Hif-2α^fl/fl;Vav-iCre mice and controls, i.e., Mll-AF9^KI/+;Hif-2α^fl/fl (without Vav-iCre), Hif-2α^fl/fl;Vav-iCre, and Hif-2α^fl/fl mice and transplanted LSK cells from these mice into primary recipients (together with 200,000 WT CD45.1⁺ unfractionated BM cells). We found that recipients of Mll-AF9^KI/+;Hif-2α^fl/fl;Vav-iCre and Mll-AF9^KI/+;Hif-2α^fl/fl LSK cells had similar percentages of CD45.2⁺ donor-derived cells indicating that the overall reconstitution capacity of LSK cells of both genotypes was not different. Furthermore, LSK cells of both Mll-AF9^KI/+;Hif-2α^fl/fl;Vav-iCre and Mll-AF9^KI/+;Hif-2α^fl/fl genotypes had multilineage differentiation potential that was sustained for at least 19 wk after transplantation (not depicted). However, recipients of Mll-AF9^KI/+;Hif-2α^fl/fl;Vav-iCre LSK cells had increased percentages of donor-derived myeloid cells in the peripheral blood compared with recipients of Mll-AF9^KI/+;Hif-2α^fl/fl LSK cells (Fig. 2 A). Consequently, recipients of Mll-AF9^KI/+;Hif-2α^fl/fl;Vav-iCre LSK cells became cachectic and succumbed to AML (confirmed by flow cytometric analyses of the BM; not depicted) faster compared with recipients of Mll-AF9^KI/+;Hif-2α^fl/fl LSK cells (Fig. 2, B and C). This was in contrast to recipients of Hif-2α^fl/fl;Vav-iCre and Hif-2α^fl/fl LSK cells which, as expected (Guitart et al., 2013), did not develop leukemia (Fig. 2, B and C). To investigate whether Hif-2α deletion compromises LSC maintenance in the Mll-AF9^KI/+ model, we sorted the CD45.2⁺ donor-derived Lin⁻c-Kit⁺Sca-1⁻ (LK) cells from primary recipients and transplanted them into secondary recipients (together with 200,000 WT CD45.1⁺ unfractionated BM cells). LK cells of both Mll-AF9^KI/+;Hif-2α^fl/fl and Mll-AF9^KI/+;Hif-2α^fl/fl;Vav-iCre genotypes equally efficiently engrafted (Fig. 2 D), and propagated full-blown disease (Fig. 2 E) in secondary recipients. We also tested the LSC frequency in primary recipients of Mll-AF9^KI/+;Hif-2α^fl/fl or Mll-AF9^KI/+;Hif-2α^fl/fl;Vav-iCre LSK cells that developed leukemia by performing a secondary transplantation of donor-derived LK cells sorted from primary leukemic recipients using an extreme limiting dilution assay (Hu and Smyth, 2009) and found no difference in LSC frequency between the genotypes (Table 1 and Table 2). We conclude that Hif-2α delays the development of LSCs in the Mll-AF9 model of AML, but once LSCs are established Hif-2α is dispensable for their maintenance.

Given our surprising findings, which indicated that Hif-2α is not required for mouse AML propagation, we next determined the impact of HIF-2α ablation in human leukemic cells using the CRISPR-Cas9 genome editing approach. We transduced human AML (M5) THP-1 cells harboring MLL-AF9 translocation with lentiviruses expressing two independent single guide RNAs (sgRNAs) targeting exon 12 of the HIF-2α (EPAS1) gene (Fig. 3, A and B) and coexpressing a mammalian codon-optimized Cas9 nuclease (Sanjana et al., 2014). These lentiviruses will be referred to as HIF-2α sgRNA1-Cas9 and HIF-2α sgRNA2-Cas9. We isolated two independent clones, one harboring HIF-2α sgRNA1-Cas9 lentivirus and the second harboring sgRNA2-Cas9 lentivirus (Fig. 3 A). Both clones had frameshifting mutations (Fig. 3 B), resulting in the loss of HIF-2α protein (Fig. 3 D). Two clones of THP-1 cells transduced with Cas9 lentivirus lacking sgRNA were used as control. Our analyses performed under normoxic and hypoxic conditions revealed that the loss of HIF-2α expression had no impact on the cell cycle (Fig. 3 E), proliferation rate (Fig. 3 F), and colony formation capacity (Fig. 3, G–I) of THP-1 cells under normoxic and hypoxic conditions. Thus, consistent with our findings in the mouse models of AML, HIF-2α is not required for propagation of human leukemic cells under normoxic and hypoxic conditions.

The tumor suppressor function of Hif-2α in AML development, taken together with the recent study indicating that Hif-1α can also suppress leukemic transformation (Velasco-Hernandez et al., 2014), suggested that Hif-1α and Hif-2α may collaborate to suppress leukemogenesis. To test this, we transduced BM-derived c-Kit⁺ cells from Hif-1α^fl/fl;Vav-iCre, Hif-2α^fl/fl;Vav-iCre, Hif-1α^fl/fl;Hif-2α^fl/fl;Vav-iCre, and control mice with Meis1/Hoxa9 retroviruses. Although cells of each genotype cultured under normoxia and hypoxia self-renewed equally in serial replating assays (Fig. 4 A), preleukemic cells lacking Hif-1α, Hif-2α, or both had increased proliferative capacity compared with control cells (Fig. 4 B). After transplantation, the cell cycle profile of Hif-1α–deficient cells was not different from that of control cells but a slightly larger proportion of Hif-2α–deficient and Hif-1α/Hif-2α–deficient cells were in S/G2/M phases of the cell cycle compared with Hif-1α–deficient cells (notably there was no difference in the cell cycle profile between Hif-2α–deficient and Hif-1α/Hif-2α–deficient cells; unpublished data). Remarkably, upon transplantation, preleukemic cells lacking both Hif-1α and Hif-2α generated dramatically accelerated AML compared with those lacking either Hif-1α or Hif-2α (Fig. 4, C and D). Therefore, Hif-1α and Hif-2α synergize to suppress the development of LSCs. Finally, control, Hif-2α- and Hif-1α/Hif-2α–deficient BM c-Kit⁺ cells sorted from primary leukemic recipients equally efficiently generated aggressive leukemia in secondary recipients (Fig. 4 E). Thus, despite the hypoxic nature of the BM, at least in AML involving Meis1/Hoxa9-mediated transformation the disease is propagated through Hif-independent mechanisms.

To test the impact of inhibiting the HIF pathway in human AML cells, we incubated (M5) THP-1 cells and NOMO-1 cells, both harboring the MLL-AF9 fusion, with BAY 87–2243, which inhibits HIF-1α and HIF-2α protein accumulation under hypoxia and decreases HIF target gene expression in tumor cells (Ellinghaus et al., 2013; Helbig et al., 2014; Chang et al., 2015). As expected, BAY 87–2243 treatment inhibited the expression of HIF target genes (ALDOA, PDK1, BNIP3, and EGLN3) in both THP-1 (Fig. 5 A) and NOMO-1 (Fig. 5 B) cells under hypoxia. Under these conditions, BAY 87–2243 had no effect on the proliferative capacity and survival of THP-1 (Fig. 5, C and D) and NOMO-1 (Fig. 5, E and F) cells. Thus, inhibition of the HIF pathway in two independent, established human AML cells with MLL-AF9 translocation has no impact on their survival and growth.

To understand the molecular signatures associated with accelerated leukemia development upon Hif-1α and Hif-2α deletion, we performed global gene expression profiling of Meis1/Hoxa9-transduced Hif-1α^fl/fl;Vav-iCre, Hif-2α^fl/fl;Vav-iCre, Hif-1α^fl/fl;Hif-2α^fl/fl;Vav-iCre, and control preleukemic cells incubated under hypoxic conditions (1% O₂) for 8 d. Although deletion of Hif-1α or Hif-2α had a minor effect on gene expression (i.e., 87 and 6 genes were differentially expressed in Hif-1α^fl/fl;Vav-iCre and Hif-2α^fl/fl;Vav-iCre cells, respectively, compared with control cells), deletion of both Hif-1α and Hif-2α deregulated the expression of 1,013 genes (Fig. 6, A and B; FDR< 0.05), indicating the synergism between Hif-1α and Hif-2α in the control of gene expression. Indeed, multiple known Hif target genes identified previously in nonhematopoietic cells (Mole et al., 2009; Schödel et al., 2011) were deregulated in Hif-1α^fl/fl;Hif-2α^fl/fl;Vav-iCre preleukemic cells (Fig. 6 C). Strikingly, pathways analyses revealed that the most significantly affected group of genes deregulated upon Hif-1α and Hif-2α deletion were genes involved in energy metabolism. Whereas genes involved in glycolysis were significantly down-regulated in Hif-1α^fl/fl;Hif-2α^fl/fl;Vav-iCre cells, those controlling fatty acid degradation, the TCA cycle, and mitochondrial oxidative phosphorylation pathways were up-regulated in preleukemic cells lacking Hif-1α and Hif-2α (Fig. 6, D and E). Thus, the ability of Hif-1α^fl/fl;Hif-2α^fl/fl;Vav-iCre cells to generate accelerated leukemia is associated with the gene signature indicative of the metabolic switch from glycolysis to fatty acid oxidation and mitochondrial oxidative phosphorylation.

In addition to the pathways analyses, our data revealed that a subset of modulators of AML leukemogenesis was deregulated upon Hif-1α and Hif-2α deletion. The expression of several negative regulators of leukemogenesis was decreased, and the expression of distinct genes that promote AML was increased (Fig. 6 F). For example, the expression of a direct Hif target Kdm5b (Jarid1b) which functions as a H3K4-specific demethylase that negatively regulates leukemogenesis in murine and human MLL-rearranged AML cells (Wong et al., 2015), was decreased in Hif-1α^fl/fl;Hif-2α^fl/fl;Vav-iCre cells (notably, Kdm5b knockdown in MLL-rearranged cells decreases the latency of AML development; Wong et al., 2015). Levels of Tspan3 (encoding tetraspanin 3), which is required for the development and propagation of AML (Kwon et al., 2015), were increased (Fig. 6 F). A hypoxia-inducible gene Bnip3 (Sowter et al., 2001), encoding a proapoptotic protein that is silenced in hematopoietic malignancies, including 17% of AML cases (Murai et al., 2005), was down-regulated in Hif-1α^fl/fl;Hif-2α^fl/fl;Vav-iCre cells (Fig. 6 F). Furthermore, one of the most significantly up-regulated pathways in Hif-1α^fl/fl;Hif-2α^fl/fl;Vav-iCre cells was the proteasome pathway (Fig. 6, D and E), whose expression and activity is increased in malignancies, including leukemia (Kumatori et al., 1990; Ma et al., 2009), and inhibition of which reduces survival of human and mouse AML cells (Servida et al., 2005; Bernot et al., 2013; van der Helm et al., 2015). Finally, consistent with their increased proliferation, numerous genes involved in cell cycle progression or tumorigenesis were differentially expressed in Hif-1α/Hif-2α–deficient preleukemic cells compared with control (unpublished data). Therefore, Hif-1α and Hif-2α deletion promotes a molecular signature that facilitates survival and proliferation of preleukemic cells.

The role of Hif-1α and Hif-2α in leukemic transformation remains a subject of intense debate (Losman and Kaelin, 2013; Gezer et al., 2014; Vyas, 2014). Here, we used a conditional gene deletion approach to investigate the requirement for Hif-1α and Hif-2α in leukemogenesis. Neither Hif-1α nor Hif-2α was essential for the generation of preleukemic cells in vitro under normoxic and hypoxic conditions. Notably, however, loss of Hif-2α accelerated the conversion of preleukemic cells to LSCs and shortened AML latency, and this effect was potentiated by Hif-1α codeletion. Our data, taken together with the tumor suppressor functions of Hif-1α (Velasco-Hernandez et al., 2014), imply that Hif-1α and Hif-2α synergize within the hypoxic bone marrow to suppress LSC generation and AML development. Although deletions in HIF-1α or HIF-2α have not been thus far identified in any subtype of human AML (Ley et al., 2008; Cancer Genome Atlas Research Network, 2013; Andersson et al., 2015; Lavallée et al., 2015), it will be of interest to explore the possibility that HIF-1α or HIF-2α are mutated or silenced in leukemias with MLL translocations or in those expressing high levels of MEIS1 or HOXA9.

Gene knockdown studies suggested the requirement for HIF-1α or HIF-2α in human LSC functions, indicating that HIFs are potential therapeutic targets for AML LSC elimination (Wang et al., 2011; Rouault-Pierre et al., 2013). However, the interpretation of these knockdown studies is confounded by the lack of information regarding the subtype and cytogenetics of AML samples. Here, we report that, surprisingly, Hif-2α is dispensable for the ability of established LSCs to sustain murine AML induced by Mll-AF9 or Meis1/Hoxa9. Remarkably, established LSCs lacking both Hif-1α and Hif-2α efficiently propagate AML. Furthermore, CRISPR-Cas9–mediated knockout of HIF-2α or pharmacological inhibition of the HIF pathway in human AML cells with MLL-AF9 translocation had no impact on their survival and proliferation under hypoxic conditions. Thus, our study in mouse and human cells, taken together with the recent demonstration that Hif-1α is dispensable for disease propagation in MLL-ENL– and Meis1/Hoxa9-driven mouse AML (Velasco-Hernandez et al., 2014), indicates that HIF antagonists are unlikely to have a major beneficial therapeutic potential in AML subsets resulting from the activation of the Mll–AF9–Meis1/Hoxa9 pathway. Finally, given the heterogeneity of AML, we propose that the systematic dissection of the roles of HIF-1α and HIF-2α in different AML subsets is required to identify those that are sensitive to the inhibition of HIF-1 and/or HIF-2.

Collectively, we provide genetic evidence that Hif-2α acts as a tumor suppressor in the development of AML, but is dispensable for LSC maintenance, at least within Mll–AF9–Meis1/HoxA9-driven leukemia. In concordance, the HIF pathway is not essential in human AML cells harboring the MLL–AF9 translocation. Finally, at least in the Meis1/Hoxa9 AML model, Hif-1α and Hif-2α are not required for LSC maintenance and AML propagation, but they act synergistically to suppress LSC development. Considering that LSCs serve as a paradigm for other cancer stem cells, it will be interesting to investigate the synergistic tumor suppressor functions of Hif-1 and Hif-2 in other cancers.

Hif-1α^fl/fl;Vav-iCre, Hif-2α^fl/fl;Vav-iCre, and Hif-1α^fl/fl;Hif-2α^fl/fl;Vav-iCre mice were described previously (Guitart et al., 2013). Mll-AF9^KI/+ mice (Chen et al., 2008) were obtained from The Jackson Laboratory. 8–10-wk-old sex-matched mice were used as donors of c-Kit⁺ cells. Animal experiments were approved by the Animal Welfare and Ethical Review Body of the University of Edinburgh and authorized by the UK Home Office.

BM and blood samples were stained and analyzed as described previously (Kranc et al., 2009; Guitart et al., 2013).

Total RNA was isolated using the RNeasy Mini kit (QIAGEN) and cDNA synthesis was performed with the High-Capacity cDNA Reverse Transcription kit (Applied Biosystems). qRT-PCR was performed in triplicate with the Light Cycler 480-II (Roche) using the human TaqMan Gene Expression Assay probe and primer sets for HPRT (Hs02800695_m1), BNIP3 (Hs00969291_m1), PDK1 (Hs01561850_m1), EGLN3 (Hs00222966_m1), and ALDOA (Hs00605108_g1; Life Technologies). Relative gene expression was analyzed by the ΔΔC_t method using HPRT as reference control and an assigned calibrator.

Preleukemic cells cultured in hypoxia (1% O₂) were collected and resuspended in TRIzol (Life Technologies). RNA was isolated by standard phenol/chloroform extraction. RNA samples (500 ng) were processed and labeled for array hybridization using the Ambion WT Expression kit (Life Technologies). Labeled, fragmented cDNA (GeneChip WT Terminal Labeling and Controls kit; Affymetrix) was hybridized to Mouse Gene 2.0 arrays for 16 h at 45°C (at 60 rpm; GeneChip Hybridization, Wash, and Stain kit; Affymetrix). Arrays were washed and stained using the Affymetrix Fluidics Station 450, and scanned using the Hewlett-Packard GeneArray Scanner 3000 7G.

For bioinformatic analyses, a total of 18 arrays (n = 5 for control, n = 4 for Hif-1α^fl/fl;Vav-iCre, n = 5 for Hif-2α^fl/fl;Vav-iCre, n = 4 for Hif-1α^fl/fl;Hif-2α^fl/fl;Vav-iCre) were QC (quality control) analyzed using arrayQualityMetrics in Bioconductor. Raw data were quantile normalized across all arrays. Pairwise group comparisons were undertaken using linear modeling, using the limma package in Bioconductor. Subsequently, empirical Bayesian analysis was applied, including vertical (within a given comparison) p-value adjustment for multiple testing, which controls for false-discovery rate. Functional-enrichment hypergeometric test analyses were performed for KEGG pathways using the appropriate packages. Focused genes of interest lists were assembled from the literature and other publically available resources.

THP-1 cells were incubated with 150 µM CoCl₂ (Sigma-Aldrich) for 16 h, washed in ice-cold phosphate-buffered saline (PBS), and prepared in lysis buffer containing 6.7 M urea, 10 mM Tris-Cl, pH 6.8, 10% glycerol, 1% SDS, 1 mM dithiothreitol, supplemented with Complete Protease Inhibitor and PhosStop (Roche). 20 µg of proteins were separated by electrophoresis on a 10% Tris-Glycine gel and transferred to Immun-Blot polyvinylidene difluoride membranes (Bio-Rad Laboratories). Membranes were blocked in 3% milk 3% BSA in Tris-buffered saline – 0.1% Tween-20 (TBST), and then incubated with HIF-2α primary antibody (NB100-122; Novus) in 3% milk 3% BSA TBST overnight, washed in TBST, and incubated with horseradish peroxidase (HRP)­–labeled secondary antibodies (R&D Systems) in 3% milk TBST. Immunodetection was performed using the Supersignal West Dura Extended Duration Substrate (Thermo Fisher Scientific) and Odyssey Fc (Licor Biosciences).

BM cells were extracted from 8–10-wk-old mice. After c-Kit enrichment using MACS LS columns (Miltenyi Biotec), cells were transduced with MSCV-Meis1a-puro and MSCV-Hoxa9-neo retroviruses and serially replated every 6 d in MethoCult M3231 (STEMCELL Technologies) supplemented with SCF, IL-3, IL-6, and GM-CSF in normoxia (20% O₂) and hypoxia (1% O₂).

Live cells were counted with Accuri C6 Flow Cytometer (BD) at the time points indicated in the figures. For cell cycle analyses, cells were resuspended in the DAPI/NP-40 solution and acquired on FACSFortessa V (BD).

Guide RNAs to target the human HIF-2α gene were designed using the Zhang laboratory web resource (www.genome-engineering.org). Two individual sgRNAs were designed to target exon 12 of HIF-2α (sgRNA1, 5′-AGAAGACAGAGCCCGAGCAC-3′; sgRNA2, 5′-GCTTCCGGCATCAAAGAAGA-3′), which is the most 5′ exon common to all UCSC-annotated transcript variants. sgRNA-encoding oligonucleotides were cloned into pLentiCRISPRv2 vector (Sanjana et al., 2014) using standard procedures (www.genome-engineering.org). Lentiviral production was performed as described previously (Kranc et al., 2009). After lentiviral transduction and selection, THP-1 cells were subjected to single-cell sorting (FACSAria Fusion; BD), and individual clones were expanded and screened for frameshift mutations. In brief, a region spanning the target site was amplified from genomic DNA isolated from individual clones. PCR products were subsequently cloned into TOPO TA Cloning Sequencing vector (Invitrogen). 15 sequences were analyzed per clone by aligning them to the WT HIF-2α sequences using BLAST. Only the clones with frameshifting indel mutations and no detectable WT allele were used for subsequent validation and in vitro assays.

THP-1 and NOMO-1 cells were cultured at 400,000 cells/ml in RPMI-1640 GlutaMAX containing 10% FBS, 25 mM HEPES, 100 U/ml penicillin, and 100 µg/ml streptomycin and incubated with 10 nM BAY 87–2243 (Selleckchem) or vehicle control (0.001% EtOH [vol/vol]) for 6 d under hypoxic conditions (1% O₂). At days 3 and 6, the cells were incubated with 7-AAD, and their proliferative capacity (number of 7-AAD⁻ cells) and percentage of dead (7-AAD⁺) cells was measured by flow cytometry using the Accuri C6 Flow Cytometer (BD).

Mll-AF9^KI/+, Hif-1α^fl/fl, Hif-2α^fl/fl, and Vav-iCre mice were CD45.2⁺. For transplantations with Meis1/Hoxa9-expressing preleukemic cells, 100,000 c-Kit⁺ preleukemic cells were transplanted together with 200,000 WT CD45.1⁺ unfractionated BM cells into lethally irradiated (10 Gy) CD45.1⁺/CD45.2⁺ recipients. The mice were monitored for AML development. For secondary transplants with Meis1/Hoxa9-expressing LSCs, 10,000 CD45.2⁺c-Kit⁺ cells were sorted from primary recipients and transplanted into secondary recipients together with 200,000 WT CD45.1⁺ unfractionated BM cells. For Mll-AF9^KI/+ cell transplantations, lethally irradiated CD45.1⁺/CD45.2⁺ syngeneic recipients were transplanted with 2,000 LSK or 5,000 LK cells (together with 200,000 CD45.1⁺ BM cells per recipient) as indicated in Fig. 2.

Kaplan-Meier survival curve statistics were determined using the Mantel-Cox test. LSC frequencies were determined using ELDA statistics software (Hu and Smyth, 2009). All other statistical significance was determined using the Mann-Whitney test.

We are grateful to Dr. Tim Somervaille for Meis1 and Hoxa9 constructs. We thank Dr. Vladimir Benes and Jelena Pistolic from the Genomics Core facility of the European Molecular Biology Laboratory (EMBL, Heidelberg) for performing the gene expression profiling using Affymetrix GeneChip arrays. We thank Fiona Rossi and Dr. Claire Cryer for their help with flow cytometry.

K.R. Kranc is a Cancer Research UK Senior Fellow. K.R. Kranc's laboratory is supported by grants from Cancer Research UK, Bloodwise (formerly called Leukaemia and Lymphoma Research), Edinburgh Cancer Research UK Centre Development Fund, The Wellcome Trust ISSF award, Medical Research Council, and the Kay Kendall Leukaemia Fund. T.L. Holyoake was supported by Cancer Research UK (C11074/A11008).

The authors declare no competing financial interests.