To Promote Leukemogenesis Through Dnmt3a Transactivation

ORIGINAL ARTICLE AML1/ETO cooperates with HIF1α to promote leukemogenesis through DNMT3a transactivation

XN Gao1,2,6,FYan2,6, J Lin3,6, L Gao1,6,XLLu1,SCWei2, N Shen2, JX Pang2, QY Ning1, Y Komeno4, AL Deng1,YHXu1, JL Shi1,YHLi1, DE Zhang4, C Nervi5, SJ Liu2,7 and L Yu1,7

The mechanisms by which AML1/ETO (A/E) fusion protein induces leukemogenesis in acute myeloid leukemia (AML) without mutagenic events remain elusive. Here we show that interactions between A/E and hypoxia-inducible factor 1α (HIF1α) are sufﬁcient to prime leukemia cells for subsequent aggressive growth. In agreement with this, HIF1α is highly expressed in A/E- positive AML patients and strongly predicts inferior outcomes, regardless of gene mutations. Co-expression of A/E and HIF1α in leukemia cells causes a higher cell proliferation rate in vitro and more serious leukemic status in mice. Mechanistically, A/E and HIF1α form a positive regulatory circuit and cooperate to transactivate DNMT3a gene leading to DNA hypermethylation. Pharmacological or genetic interventions in the A/E–HIF1α loop results in DNA hypomethylation, a re-expression of hypermethylated tumor-suppressor p15INK4b and the blockage of leukemia growth. Thus high HIF1α expression serves as a reliable marker, which identiﬁes patients with a poor prognosis in an otherwise prognostically favorable AML group and represents an innovative therapeutic target in high-risk A/E-driven leukemia.

Leukemia (2015) 29, 1730–1740; doi:10.1038/leu.2015.56

INTRODUCTION genes.17–22 However, it remains unclear whether and how A/E, The t(8;21)(q22;q22) translocation, resulting in a chimeric protein as an oncogenic transcriptional factor, directly regulates DNMT AML1/ETO (A/E), is one of the most common chromosomal expression and, in turn, genome-wide DNA methylation pattern- abnormalities in acute myeloid leukemia (AML).1 A/E is not ing. In the current study, we demonstrate that A/E and HIF1α sufficient in itself to induce leukemogenesis,2 thus the additional cooperatively promote leukemogenesis. This cooperation takes mutations conferring enhanced proliferation and survival properties, place via a reciprocal regulatory circuit, in which A/E binds to such as KIT, FLT3 or JAK2, are required for A/E to be a leukemic HIF1α promoter and transactivates HIF1α or vice versa, and driver.3–5 However, the incidence of these mutations in A/E- through their cooperation in controlling DNMT3a expression and positive AML patients (A/E-patients) varies between 3% and subsequent DNA methylation in AML cells. 22%,6–9 strengthening the notion that unconventional and more universal cooperating events are inevitable for A/E to drive MATERIALS AND METHODS leukemogenesis. Hypoxia-inducible factor 1α (HIF1α) is a transcription factor Patient samples and cell lines mediating the cellular response to hypoxia in malignant cells.10 Details are in Supplementary Materials and Methods. Bone marrow Although a hypoxic environment stabilizes and induces HIF1α,itis samples from patients were obtained with signed informed consent. This also upregulated in normoxia in response to cytokines, interferon- study was carried out in accordance with principles of Declaration of α, hormones or genetic alterations.11–13 HIF1α–notch interaction is Helsinki and was approved by the Human Subject Ethics Committee in Chinese PLA General Hospital. essential for survival maintenance of leukemia stem cells,14 and HIF1α upregulation predicts residual disease in AML patients after chemotherapy.15 However, the role of HIF1α in A/E-driven Gene knockdown, overexpression, quantitative real-time PCR and leukemia is unknown as yet. western blotting The presence of chromosomal rearrangements, such as t(8;21), Details, including small interfering RNAs (siRNAs), expression plasmids, t(15;17) and inv(16), is associated with specific DNA methylation primers and antibodies, are in Supplementary Materials and Methods. profiles that predict distinct patient outcomes.16 The underlying mechanisms likely involve the recruitment of DNA methyltrans- Chromatin immunoprecipitation (ChIP) and reporter assays ferases (DNMTs) by A/E to its target promoters, as others and we ChIP assays were performed as described previously using EZ-ChIP Assay have shown that A/E forms a complex with DNMTs and certain Kit (Millipore, Billerica, MA, USA).23 Details are in Supplementary Materials transcriptional factors to repress target tumor-suppressor and Methods.

1Department of Hematology, Chinese PLA General Hospital, Medical School of Chinese PLA, Beijing, China; 2The Hormel Institute, University of Minnesota, Austin, MN, USA; 3Institute of Basic Medicine, Chinese PLA General Hospital, Medical School of Chinese PLA, Beijing, China; 4Department of Pathology and Division of Biological Sciences, Moores UCSD Cancer Center, University of California San Diego, La Jolla, CA, USA and 5Department of Medical Surgical Sciences and Biotechnologies, University of Rome ‘La Sapienza’, Latina, Italy. Correspondence: Dr SJ Liu, The Hormel Institute, University of Minnesota, 801 16th Avenue NE, Austin, MN 55912, USA or Dr L Yu, Department of Hematology, Chinese PLA General Hospital, Medical School of Chinese PLA, 28 Fuxing Road, Beijing 100853, China. E-mail: [email protected] or [email protected] 6These authors contributed equally to this work. 7These senior authors contributed equally to this work. Received 14 August 2014; revised 16 February 2015; accepted 23 February 2015; accepted article preview online 2 March 2015; advance online publication, 17 March 2015 AML1/ETO–HIF1α–DNMT3a axis in leukemogenesis XN Gao et al 1731 Dot blot and bisulfite sequencing analysis analysis (Supplementary Table S3). In the overall cohort, HIF1α Dot blot was performed as described previously using 5-mC antibody.24 expression was an independent prognostic parameter for EFS but Bisulfite modification of genomic DNA was performed using EpiTect not for OS. However, in the A/E-positive cohort, it emerged as an Bisulfite Kit (QIAGEN, Valencia, CA, USA). Details are in Supplementary independent prognostic parameter for both OS and EFS. Materials and Methods. Remarkably, in the A/E-patients carrying wtKIT, high HIF1α level was the only parameter with prognostic significance for OS and Clonogenic assays, flow cytometry analysis, hematoxylin and eosin EFS. The stability of the classical multivariate Cox models was staining and immunohistochemistry further tested by a bootstrap resampling procedure. No significant Details are in Supplementary Materials and Methods. differences were found in P-values of regression coefficients between the classical and the bootstrap Cox regression models (Supplementary Table S4). Thus high HIF1α expression identifies Animal studies A/E-patients with inferior OS and EFS, supporting an important Animal experiments were performed in accordance with humane role for HIF1α signaling in A/E-associated leukemia. practices, national and international regulations and with the approval of the University of Minnesota Institutional Animal Care and Use Committee. Details for cell injection, immunostaining and tissue preparation along with A/E and HIF1α form a positive regulatory circuit through direct antibodies used are described in Supplementary Materials and Methods. promoter binding The positive correlation between A/E and HIF1α prompted us to Statistical analysis examine their reciprocal regulation. Bioinformatic analysis using 27 Details are in Supplementary Materials and Methods. TFSEARCH revealed three putative A/E-binding sites in 5′- flanking region 2-kb upstream of the transcription start site of HIF1α gene (Figure 2a). ChIP demonstrated approximately twofold RESULTS enrichment of A/E on HIF1α promoter in Zn2+-inducible A/E- HIF1α is highly expressed in A/E-patients and strongly predicts expressing U937A/E cells after ZnSO4 treatment (Figure 2a), which inferior prognosis was supported by a previous ChIP-ChIP assay showing that HIF1α 28 HIF1α mRNA levels were initially analyzed in AML cell lines is one of the A/E-binding targets. The functional interaction (Supplementary Figure S1a) and a microarray data set GSE689125 between A/E binding and HIF1a promoter was tested by luciferase (Supplementary Figure S1b and Supplementary Table S1). We assays using the reporter constructs containing different portions α found that HIF1α was significantly upregulated in t(8;21) AML of HIF1 promoter region surrounding the A/E-binding sites α compared with the non-t(8;21) group. To verify these results, (pGL3-HIF1 , H1-H4, Figure 2b). The H3 reporter exhibited the HIF1α levels were measured in another cohort of 132 AML patients greatest increase, while its mutant (H3mut) showed a marked and 15 healthy donors. This analysis confirmed that HIF1α was decrease, in luciferase activity upon A/E co-transfection. Deletion fi of all A/E sites led to the complete abrogation of activity, signi cantly elevated in A/E-positive patients compared with α fi A/E-negative and healthy controls (Figure 1a). suggesting that transactivity of HIF1 promoter was speci cto A/E binding. Further, ZnSO4-induced A/E overexpression resulted For statistical analysis, the 132 patients were divided into two fi α groups, high and low, based on the HIF1α expression levels in signi cant increases in HIF1 mRNA and protein levels in U937A/E cells (Supplementary Figure S2a). In contrast, shRNA/ (Figure 1b), which are detailed in Supplementary Statistical fi fi analysis. Among clinical features, A/E-positive and KIT mutation siRNA-mediated speci c A/E silencing led to signi cant decreases fi α in HIF1α expression in SKNO-1 and Kasumi-1 cells (Figures 2c and d). (mutKIT) was signi cantly associated with high HIF1 expression α (Supplementary Table S2). Of note, mutKIT is a known adverse Thus A/E functions as a transactivator for HIF1 gene. 9,23 Interestingly, we found three putative hypoxia-response prognostic factor in A/E-patients. Although mutKIT can 29,30 α 26 elements (HREs) in the 5′-flanking region 2-kb upstream of enhance HIF1 expression through phosphoinositide 3 kinase, α elevated HIF1α expression was detected in 15 of the 53 (28.3%) the transcription start site of A/E gene (Figure 2e). Hence, HIF1 A/E-patients carrying wild-type KIT (wtKIT) (Figure 1c), indicating probably binds to A/E promoter and, in turn, enhances A/E transcription. ChIP assays in SKNO-1 cells validated this that other factors, except for mutKIT, might contribute to the high α α hypothesis (Figure 2e). The CoCl2-mediated stabilization of HIF1 HIF1 expression in A/E-positive leukemia. Regardless, the positive α correlation between A/E and HIF1α expression (Figure 1d), protein led to 20-fold HIF1 enrichment on A/E promoter especially in A/E-patients carrying wtKIT (Figure 1e), suggests their compared with control, which was supported by a previous ChIP-ChIP assay showing that A/E promoter contains HREs.31 To functional relationships. fi α Prognostic implications for HIF1α overexpression were first address the biological signi cance of HIF1 binding, luciferase validated in the overall cohort (Figure 1f). High HIF1α expressers assays using the reporter constructs containing different portions showed a marginally worse overall survival (OS, median = 19.0 vs of A/E promoter region encompassing the HREs (pGL3-A/E, A1-A3, fi Figure 2f) were conducted. The activity of A1 reporter was 39.0 months, P=0.053) but signi cantly worse event-free survival α (EFS, median = 9.0 vs 13.4 months, P=0.010) than low HIF1α increased upon HIF1 co-transfection. Mutation (A1mut) or expressers. Stratification of patient groups based on A/E deletion (A2) of HRE1 fully abrogated the functional interaction between HIF1α and A/E promoter, indicating that HRE1 was status improved the predictive capability of HIF1α for both OS essential for the promoter activity. Moreover, siRNA-mediated and EFS (Figure 1g). In the A/E-positive cohort, both OS HIF1α knockdown resulted in significant decreases in (median = 19.0 months vs not reached, P=0.015) and EFS A/E expression levels in SKNO-1 and Kasumi-1 cells (Figures 2g (median = 9.0 vs 24.0 months, P=0.001) were significantly shorter and h). Together, A/E and HIF1α shape a positive regulatory circuit in high HIF1α than low HIF1α expressers (Figure 1g). Notably, in via promoter binding. the A/E-patients carrying wtKIT, higher HIF1α expression was significantly associated with shorter OS (median = 23.1 months vs not reached, P=0.008) and EFS (median = 9.4 months vs not HIF1α functionally cooperates with A/E in accelerating reached, P=0.001) (Figure 1h), indicating that HIF1α-associated leukemogenesis inferior prognosis is mutKIT independent. To characterize the roles of A/E and HIF1α in leukemic cell growth, According to the prognostic value of HIF1α expression in AML the clonogenic potential was determined in A/E- or HIF1α- patients, it was entered into a multivariate model in addition to manipulated leukemia cells. The siRNA-triggered A/E or HIF1α factors significantly associated with prognosis in univariate downregulation significantly decreased the colony number and

© 2015 Macmillan Publishers Limited Leukemia (2015) 1730 – 1740 AML1/ETO–HIF1α–DNMT3a axis in leukemogenesis XN Gao et al 1732 P<0.001 Total patients A/Epos patients P<0.001 n=33 15.0 n=73 15.0 15.0 n=15 10.0 10.0 n=11 HIF1ahigh 10.0 n=59 5.0 5.0 5.0 HIF1alow

mRNA n mRNA mRNA =99 a a a 1.7 1.7 n= 15 n n HIF1 HIF1 0.0 HIF1 0.0 0.0 =9 =38

pos neg high low HD a a KIT KIT A/E A/E wt HIF1 HIF1 mut Patients

A/Epos KIT A/Epos patients & wt patients 2.5

2.5 2.0 2.0 1.5 1.5 1.0 n=73 1.0 n=53 mRNA mRNA 0.5 ρ=0.468 0.5 ρ=0.502 A/E P<0.001 A/E P<0.001 0.0 0.0

0.0 5.0 0.0 5.0 10.015.0 10.015.0 HIF1a mRNA HIF1a mRNA

AML patients A/Epos patients A/Epos & wtKIT patients high high high HIF1a (n=33) HIF1a (n=26) HIF1a (n=15) HIF1alow (n=99) HIF1alow (n=47) HIF1alow (n=38) 1.0 1.0 1.0 P = 0.053 P = 0.015 P = 0.008 0.8 0.8 0.8 0.6 0.6 0.6

OS 0.4 OS 0.4 OS 0.4 0.2 0.2 0.2 0.0 0.0 0.0 06020 40 06020 40 06020 40

1.0 1.0 1.0 P = 0.010 P = 0.001 P = 0.001 0.8 0.8 0.8 0.6 0.6 0.6 EFS EFS EFS 0.4 0.4 0.4 0.2 0.2 0.2 0.0 0.0 0.0 06020 40 06020 40 06020 40 Time (months) Time (months) Time (months) Figure 1. Selectively high expression of HIF1α in A/E-positive AML predicts inferior prognosis. (a) Quantitative PCR showing HIF1α mRNA levels in AML patients and healthy donors (HD). Median values are depicted by the horizontal lines. (b) The patients were divided into the high and low HIF1α expression groups as described in Supplementary Statistical analysis. (c) Stratiﬁcation of A/E-patients with high and low HIF1α expression according to KIT mutation status. (d, e) Correlation analysis between HIF1α and A/E mRNA levels in the indicated patient subgroups. (f–h) Kaplan–Meier estimate for OS and EFS in the indicated patient subgroups.

size of SKNO-1 and Kasumi-1 cells (Figures 3a and b and marow, lung, spleen and liver of the recipients and considerably Supplementary Figures S3a and b), which motivated us to seek damaged the structure of organs (Figure 3d). We further verified the potential cooperation between A/E and HIF1α in controlling the role of A/E–HIF1α cooperation in A/E9a transgenic leukemia leukemia growth. To this end, Kasumi-1 cells were transfected with mouse model32 by showing that HIF1α overexpression remarkably A/E or HIF1α siRNA or both, and colony assays showed that enhanced, whereas HIF1α knockdown significantly suppressed, co-knockdown of A/E and HIF1α significantly disrupted the colony- the leukemia cell proliferation in vitro and leukemic disease in vivo forming capability of Kasumi-1 cells (Figure 3c). To recapitulate (Figures 3e and f and Supplementary Figures S3e and g). α this in vivo, A/E and HIF1 were co-expressed in murine C1498 Collectively, these data support the functional cooperation leukemia cells and co-overexpression led to the highest numbers between A/E and HIF1α in enhancing leukemogenesis. of colony-forming units compared with control (Supplementary Figures S3c and d). In agreement with the cooperative growth stimulating effect in vitro, injection of C1498 cells co- A/E and HIF1α promote DNMT3a transcriptional activation overexpressing A/E and HIF1α into syngeneic C57BL/6 J mice, to To gain insight into the molecular basis of A/E–HIF1α functional the greatest extent, increased leukemia infiltration into the bone cooperation in leukemia, the expression of DNMTs was compared

Predicted A/E binding sites -1563 -538-84 1 2 3 Mock H1 A/E 10 ng H2 A/E 50 ng A/E 100 ng H3 3.0 P < 0.05 H3mut 2.0 U937A/E H4 1.0 U937A/E+Zn2+ 0.0 0.0 2.0 4.0 6.0 8.0 Relative luciferase activity

1.5 1.5 * * * * *P < 0.05 siA/E Scramble Kasumi-1+Scramble siA/E 1.0 1.0 Kasumi-1+siA/E A/E 84 kD 0.5 0.5 SKNO-1PGK 120 kD β SKNO-1siA/E -actin 43 kD 0.0 0.0 Kasumi-1

-1359 -1204 -6 1 2 3 A1

A1 mut P< 0.05 Mock 30 A2 20 A3 10 0.0 1.0 2.0 3.0 0 Relative luciferase activity

1.5 * * 1.5 * * *P < 0.05 Kasumi-1+Scramble Scramble 1.0 1.0 Scramble Kasumi-1+siHIF1α A/E 84 kD 0.5 0.5 120 kD SKNO-1siHIF1α β-actin 43 kD 0.0 0.0 Kasumi-1 Figure 2. A/E and HIF1α form a positive regulatory circuit through direct promoter binding. (a) Top: schema showing 5′ upstream region of HIF1α gene with two consensus (TGT/CGGT) and one variant (TGGGGT) A/E-binding sites; bottom: ChIP showing binding of A/E to HIF1α promoter in A/E-inducible U937 cells. (b) Luciferase reporter assays demonstrating the transactivation of HIF1α on A/E promoter. (c, d) Quantitative PCR (qPCR) (c) and western blotting (d) showing A/E and HIF1α levels in A/E knockdown cells. (e) Top: schema showing 5′ upstream region of AML1 gene with three consensus HREs (A/GCGTG); bottom: ChIP with anti-HIF1α antibody showing the binding of HIF1α to A/E promoter in HIF1α-stabilizing SKNO-1 cells. (f) Luciferase reporter assays demonstrating the transactivation of A/E on HIF1α promoter. (g, h) qPCR (g) and western blotting (h) showing A/E and HIF1α levels in HIF1α knockdown cells. Bars in panels a and e indicate mean ± s.e.m. of duplicate samples from two independent experiments with four replicates in total. Bars in panels b, c, f and g indicate mean ± s.e.m from three independent experiments. between A/E-positive and -negative patients. The results showed TFSEARCH. Four A/E-binding sites and two HREs were identified that DNMT3a, but not DNMT1 and DNMT3b, was highly expressed (Figures 4d and f). ChIP assays revealed that A/E enrich- in t(8;21) AML (GSE6891, Supplementary Figures S4a and c). The ment on the DNMT3a promoter was decreased at least 4-fold high DNMT3a expression in A/E-patients was confirmed in another in SKNO-1 siA/E cells, but increased 13-fold in U937A/E cells, patient cohort described in Figure 1a (Figure 4a and compared with respective control. No such changes were noted Supplementary Figures S4d and e). Further, DNMT3a level was within the remote region of DNMT3a promoter without positively correlated with A/E and HIF1α expression (Figures 4b A/E-binding elements (P1) (Figure 4d), indicating the specificity and c), suggesting a functional relationship between A/E/HIF1α of A/E binding. To demonstrate the biological significance of and DNMT3a. A/E binding, luciferase assays were performed using the reporter To test this hypothesis, we searched the binding sites for A/E constructs containing the 5′ upstream region of DNMT3a with four and HIF1α within a 2-kb region spanning DNMT3a promoter using A/E sites and truncated derivatives (pGL3-DNMT3a, D1-D4,

P 200 * * * < 0.05 C1498 C57BL/6J Mouse Kasumi-1+Scramble Vector ++ + 150 A/E + + Kasumi-1+siA/E 100 HIF1α + + SKNO-1PGK 50

/ 750 cells SKNO-1siA/E 0 BM Colony number

*P < 0.05 300 * * Kasumi-1+Scramble 200 Kasumi-1+siHIF1α

100 SKNO-1+Scramble

/ 750 cells SKNO-1+siHIF1α

Colony number 0

P Kasumi-1 Spleen 150 < 0.001 Scramble 100 siA/E siHIF1α 50 α Liver Lung

/ 750 cells siA/E+siHIF1

Colony number 0

A/E9a C57BL/6J Mouse A/E9a C57BL/6J Mouse α P < 0.05 P Vector HIF1 Scramble siHIF1α < 0.05 200 160 160 120 BM

120 BM 80 80 α 40 40 Vector HIF1 Scramble siHIF1α 0 0 Spleen weight (mg) Spleen weight (mg) α α Spleen VectorHIF1 Spleen siHIF1 Scramble α Vector HIF1 Scramble siHIF1α Lung Lung Liver Liver Figure 3. HIF1α functionally cooperates with A/E in promoting leukemia growth. (a–c) Colony-forming assays showing growth inhibition in leukemia cells with knockdown of A/E (a), HIF1α (b) or both (c). Error bars indicate mean ± s.e.m. of duplicate samples from two independent experiments. (d–f) Histopathological examination showing inﬁltration of C1498 cells (d) or A/E9a cells (e, f) into bone marrow (Giemsa staining), spleen, lung and liver (hematoxylin and eosin staining) of C57BL/6 J mice. High magniﬁcation, the same area indicated by red box, shows the morphology of leukemia cells. Scale bars represent 50 μm (red), 100 μm (black) or 500 μm (blue).

Figure 4e). The activity of D1 reporter was increased dose- The binding affinity of HIF1α was also demonstrated by ChIP dependently upon A/E co-transfection (Figure 4e). Deletion of the assays showing that HIF1α enrichment on the DNMT3a promoter first and second A/E sites (D2) showed a significant loss in was increased approximately fourfold in CoCl2-treated SKNO-1 luciferase activity in response to A/E, but further deletion of the cells compared with control (Figure 4f). To determine whether third one (D3) led to no additional decrease in activity, indicating HIF1α binding could impact DNMT3a promoter activity, luciferase that sites 1 and 2 might be relatively important for maximal assays were performed using reporter constructs containing the promoter activity. To ascertain this, the A/E sites 1 and 2 (D1mut-a 5′ upstream region of DNMT3a with two HREs (D1) or a smaller and -b) were mutated one by one. Compared with the wild-type fragment with one HRE (D2). The activity of D2 reporter was sequence (D1), mutating each of them caused decreases in enhanced dose-dependently upon HIF1α co-transfection luciferase activity, indicating their individual functional relevance. (Figure 4g), which was severely impaired by deletion (D4) or Together, these findings suggest that multiple A/E sites are mutation of the HRE2 (D2mut), indicating that HRE2 was required for an optimal promoter activity. The regulatory role of essential for the promoter activity. The specificmodulationof A/E in DNMT3a expression was further verified by showing that DNMT3a by HIF1α was further confirmed by findings that DNMT3a levels were increased by enforced A/E expression, DNMT3a expression were upregulated upon HIF1α overexpression, whereas decreased by A/E knockdown (Figures 4h and i and whereas downregulated upon siRNA transfection (Figures 4h and i Supplementary Figures S5a and d). and Supplementary Figures S5e and h). Together, these results

2.5 1.5 n=23 2.0 4.0 1.5 mRNA 1.0 n=28 3.0 1.0 mRNA 2.0 mRNA n a n 0.5 =43 =43 0.5 ρ 1.0 ρ n= 10 =0.807 =0.596 A/E

P<0.001 HIF1 P<0.001 0.0 0.0 0.0 DNMT3a

pos neg HD 0.0 1.0 2.0 3.0 4.0 0.0 1.0 2.0 3.0 4.0 A/E A/E DNMT3a mRNA DNMT3a mRNA Patients

Predicted A/E binding sites pGL3-DNMT3a

-1890 -1821 -1498 -544 123 4 DNMT3a TSS D1 LUC mut D1 -a LUC CHIP primer P1 P2 mut D1 -b LUC Mock P 1.5 * 20 * * < 0.05 D2 LUC A/E10ng 15 SKNO-1PGK D3 LUC A/E50ng 1.0 A/E100ng 10 SKNO-1siA/E D4 LUC 0.5 5 U937A/E A/E CHIP 0.0 2.0 4.0 6.0 8.0 0.0 0 U937A/E+Zn2+

relative enrichment P1 P2 P1 P2 Relative luciferase activity

Predicted HREs pGL3-DNMT3a 12 -1879 -1394 DNMT3a TSS D1 LUC

CHIP primer D2 LUC P Mock < 0.05 mut 6 D2 LUC HIF1a10ng a 4 HIF1 50ng CHIP SKNO-1 D4 LUC a a HIF1 100ng 2 SKNO-1+CoCl2

HIF1 0.0 2.0 4.0 6.0 0

relative enrichment Relative luciferase activity a 1.5 * * *P < 0.05 SKNO-1PGK mRNA siA/E PGK 1.0 Scramble siHIF1 SKNO-1siA/E 0.5 SKNO-1+Scramble DNMT3a 102 kD b SKNO-1+siHIF1a -actin 43 kD 0.0 DNMT3a Figure 4. A/E and HIF1α induces DNMT3a high expression through direct promoter binding. (a) Quantitative PCR (qPCR) showing DNMT3a mRNA levels in AML patients and healthy donors. (b, c) Correlation analysis of the mRNA levels between A/E/HIF1α and DNMT3a.(d) Top: schema showing 5′ upstream region of DNMT3a gene with one consensus (TGTGGT) and three variant (TGGGGT) A/E-binding sites; bottom: ChIP with anti-ETO antibody showing the binding of A/E to DNMT3a promoter in A/E-manipulated cells. (e) Luciferase reporter assays demonstrating the transactivation of A/E on DNMT3a promoter. (f) Top: schema showing 5′ upstream region of DNMT3a gene with two consensus HREs (GCGTG); bottom: ChIP with anti-HIF1α antibody showing the binding of HIF1α to DNMT3a promoter in HIF1α-stabilizing SKNO-1 cells. (g) Luciferase reporter assays demonstrating the transactivation of HIF1α on DNMT3a promoter. (h, i) qPCR (h) and western blotting (i) showing DNMT3a levels in A/E or HIF1α knockdown cells. Bars in panels d and f indicate mean ± s.e.m. of duplicate samples from two independent experiments with four replicates in total. Bars in panels e, g and h indicate mean ± s.e.m. from three independent experiments. support that A/E or HIF1α directly promote DNMT3a transcrip- effect on DNMT3a upregulation, while their co-depletion induced tional activation. a more signiﬁcant DNMT3a downregulation, compared with the individual plasmid (Figures 5a and b and Supplementary Figures S6a and b). Dot blot analysis revealed that A/E overexpression in A/E and HIF1α cooperatively transactivate DNMT3a and alter DNA U937 or 293 T cells enhanced, whereas A/E knockdown in Kasumi-1 methylation proﬁle or SKNO-1 cells decreased, the global DNA methylation (Figures 5c We further investigated whether A/E cooperates with HIF1α to and d). Similarly, HIF1α overexpression in 293T cells promoted, regulate DNMT3a expression and, thereby, DNA methylation. As whereas HIF1α knockdown in Kasumi-1 or SKNO-1 cells reduced, expected, A/E and HIF1α co-expression had a more pronounced the global DNA methylation (Figures 5e and f). Together, these

© 2015 Macmillan Publishers Limited Leukemia (2015) 1730 – 1740 AML1/ETO–HIF1α–DNMT3a axis in leukemogenesis XN Gao et al 1736 293T Kasumi-1 Vector + ++ Scramble + ++ A/E + + siA/E + + HIF1α ++ siHIF1α ++ A/E 84 kD A/E 84 kD HIF1α 120 kD HIF1α 120 kD DNMT3a 102 kD DNMT3a 102 kD β-actin 43 kD β-actin 43 kD

P P 3.0 * * * < 0.051.5 * * * < 0.05 293T+Vector Kasumi-1+Scramble 2.0 1.0 293T+A/E Kasumi-1+siA/E 1.0 U937MT+Zn2+ 0.5 SKNO-1PGK 0.0 U937A/E+Zn2+ 0.0 SKNO-1siA/E Relative level Relative level Dot blot: 5-mC 5-mC Dot blot: 5-mC

P P 3.0 < 0.051.5 * * * < 0.05 Kasumi-1+Scramble 2.0 1.0 α 293T+Vector Kasumi-1+siHIF1 1.0 0.5 SKNO-1+Scramble 293T+HIF1α 0.0 0.0 SKNO-1+siHIF1α Relative level Relative level Dot blot: 5-mC Dot blot: 5-mC

P < 0.05 293TP < 0.05 Kasumi-1 3.0 1.5 Vector Scramble 2.0 A/E+Vector 1.0 siA/E+Scramble 1.0 HIF1α+Vector 0.5 siHIF1α+Scramble α 0.0 A/E+HIF1a 0.0 siA/E+siHIF1 Relative level Relative level Dot blot: 5-mC Dot blot: 5-mC

C1498 C1498 160 200 120 160 120 80 80

40 / 750 cells

/ 750 cells 40

0 Colony number 0 Colony number Vector ++ Vector ++ A/E ++ HIF1α ++ Scramble + + Scramble + + siDNMT3a + + siDNMT3a + + Figure 5. A/E and HIF1α cooperatively modulates DNMT3a expression and DNA methylation. (a, b) Western blotting showing the indicated protein levels in cell lines with A/E and HIF1α co-overexpression (a) or co-knockdown (b). (c–h) Dot blot showing 5-mC levels in genomic DNA extracted from A/E- or/and HIF1α-manipulated cells. Bars indicate mean ± s.e.m. from three independent experiments. (i, j) Colony-forming assays showing the growth-stimulating effect of A/E (i) or HIF1α (j) on C1498 cells, which was attenuated by DNMT3a knockdown.

results indicated that A/E and HIF1α, independently, modulate the but not DNMT3b (Supplementary Figures S6f and g); In contrast, levels of global DNA methylation. Notably, in a cooperative specific knockdown of A/E or HIF1α significantly reduced the levels manner, A/E and HIF1α co-expression promoted, whereas their of both DNMT1 and DNMT3b (Supplementary Figures S6h and i). co-knockdown inhibited, the global DNA methylation (Figures 5g and h). Together, these results suggest that the interplay between A/E Because only DNMT3a, but not DNMT1 or DNMT3b, positively and HIF1α contributes to DNMT3a upregulation and subsequent correlates with A/E and HIF1α in A/E-patients, the observations, DNA hypermethylation in A/E-positive leukemia. Finally, to which DNMT3a overexpression enhanced, whereas DNMT3a determine the biological significance of DNMT3a in A/E and ablation decreased, the global DNA methylation (Supplementary HIF1α cooperation, we knocked down DNMT3a in C1498 cells with Figures S6c and e), supported that DNMT3a is involved in A/E- enforced expression of A/E or HIF1α and found that HIF1α-mediated DNA hypermethylation. However, we could not DNMT3a depletion attenuated the effects of A/E and HIF1α exclude the role of DNMT1 or/and DNMT3b in regulating A/E– overexpression on cell colony formation (Figures 5i and j), HIF1α-associated DNA methylation in cell line models, as A/E or suggesting that A/E and HIF1α cooperativity takes places HIF1α overexpression significantly increased the levels of DNMT1 through DNMT3a.

Leukemia (2015) 1730 – 1740 © 2015 Macmillan Publishers Limited AML1/ETO–HIF1α–DNMT3a axis in leukemogenesis XN Gao et al 1737 Pharmacological inhibition of HIF1α disrupts A/E–HIF1α loop, siRNA-mediated HIF1α depletion (Figure 6c, right). Further, the induces DNA demethylation and suppresses leukemia cell growth DNA methylation frequency of p15INK4b promoter was decreased both in vitro and in vivo in SKNO-1 cells upon echinomycin exposure or HIF1α siRNA Given the central role of HIF1α in the A/E–HIF1α–DNMT3a cascade, transfection (Figure 6d), suggesting that HIF1α dysfunction we chose echinomycin, a reported HIF1α inhibitor,33 as a potential induces p15INK4b re-expression via promoter demethylation. therapeutic agent in A/E-positive AML. The echinomycin treatment The biological significance of echinomycin-mediated HIF1α induced dramatic downregulation of A/E, HIF1α and DNMT3a inactivation was supported by the observations that echinomycin levels (Figure 6a and Supplementary Figure S7a) and decrease of treatment induced a dose-dependent increase in apoptosis global DNA methylation (Figure 6b) in Kasumi-1 and SKNO-1 cells. (Figure 6e and Supplementary Figure S7b) and caspase activation Consistent with the reduced DNA methylation, the mRNA (Figure 6f) in Kasumi-1 and SKNO-1 cells. Notably, echinomycin expression of p15INK4b, a tumor-suppressor gene silenced by induced more apoptosis in SKNO-1PGK than SKNO-1siA/E cells DNA hypermethylation in leukemia,34 was dose-dependently (Figure 6e and Supplementary Figure S7b), supporting the relative restored by echinomycin treatment (Figure 6c, left). The specific specificity of echinomycin on A/E-positive cells, which could be effect of echinomycin on p15INK4b upregulation was verified by due to the higher HIF1α expression. Further, the clonogenic

Kasumi-1 Dot blot Kasumi-1 m + + 5- C 1 5 10 20 30 1 5102030 nM P<0.05 P<0.05 Echinomycin 1.2 A/E 84 kD 0.8 1nM 5nM 120 kD 0.4 10nM 102 kD 0.0 20 nM β Relative level 30nM -actin 43 kD Kasumi-1

ink4b 10nM 20nM P15 *P < 0.05 1nM 5nM 30nM CpG -12 +260 30 30 * P<0.05 P<0.05 31.9% 20 20 Scramble mRNA mRNA * α 10 10 siHIF1 Echinomycin 0.2% ink4b ink4b 0 0 p15 p15 Scramble 34.6% Kasumi-1 3.7% Kasumi-1

10nM 20nM Kasumi-1 1nM 5nM 30nM + + 15 P P Echinomycin 1 5 10 20 30 1 5 10 20 30 nM <0.05 <0.05 10 17kD Cleaved cas3 19kD Cleaved cas8 43kD 5 41kD Full length cas9 0 47kD Fold change in apoptosis Cleaved cas9 35kD Kasumi-1 PGK siA/E β-actin 43kD

P P * 300 <0.05 <0.05 400 300 * 200 1nM * *P < 0.05 5nM cells 200 100 10nM 4 20 nM 100 10 nM / 10 / 750 cells 0 30nM 0 Colony number Colony number #1 #2 #3 Kasumi-1 A/E-positive patients Figure 6. Pharmacological inhibition of HIF1α disrupts A/E/HIF1α-induced DNA hypermethylation and suppresses leukemic proliferation in vitro.(a, b) Western blotting of A/E, HIF1α and DNMT3a (a) and dot blot of 5-mC(b) in the indicated cell lines treated with echinomycin or vehicle. (c) Quantitative PCR of p15ink4b in the indicated cell lines upon echinomycin treatment or siRNA transfection. (d) Top: schema of location of the regions analyzed by bisulfite sequencing in p15ink4b promoter; bottom: bisulfite sequencing of the p15ink4b CpG sites in SKNO-1 cells upon echinomycin treatment (30 nM) or siRNA transfection. Open and filled circles, unmethylated and methylated CpG sites. (e, f) FACS analysis of apoptosis (e) and western blotting of cleaved caspases (f) in the indicated cell lines treated with echinomycin or vehicle. Histogram shows fold change in the percentage of apoptotic cells. Bars in panels b, c and e indicate the mean ± s.e.m. from three independent experiments. (g, h) Colony-forming assays in A/E-positive cell lines (g) or patient-derived blasts (h) upon echinomycin treatment or vehicle. Bars indicate mean ± s.e.m. of duplicate samples from two independent experiments with four replicates in total.

© 2015 Macmillan Publishers Limited Leukemia (2015) 1730 – 1740 AML1/ETO–HIF1α–DNMT3a axis in leukemogenesis XN Gao et al 1738 potential of A/E-positive cell lines and patient-derived blasts was sections. Downregulation of A/E, HIF1α and DNMT3a levels and impaired in the presence of echinomycin (Figures 6g and h and decrease of DNA methylation were responsible for such tumor Supplementary Figures S7c and d). Importantly, echinomycin- regression (Figures 7c and d). Thus the A/E–HIF1α–DNMT3a axis mediated growth inhibition in vitro was recapitulated in vivo,as represents a novel therapeutic avenue in A/E-positive AML. echinomycin administration in SKNO-1 xenograft tumor-bearing nude mice significantly slowed down the tumor growth (Figures 7a and b). No drug side effects were observed, because DISCUSSION echinomycin at the tested dosages induced no significant body As a leukemia-initiating transcriptional factor, A/E requires co- weight loss in the mice (Supplementary Figure S8). Hematoxylin operating gene mutations to cause leukemia.2,35 The fact that and eosin staining revealed that the tumor formation was most A/E-patients do not carry the known mutagenic ‘hits’ argues suppressed by echinomycin administration, which was verified for the existence of more ‘popular’ cooperating events. Our by the higher caspase-3 activity/lower Ki-67-index of tumor findings reveal HIF1α as an unconventional event for A/E to be

Vehicle Vehicle Echinomycin 300 ** 250 Echinomycin

) 200 ** 3 150

(mm 100 ** * 50 * Vehicle Tumor volume 0 Echinomycin 0 101214171921 Days

Vehicle Echinomycin Vehicle Echinomycin ** ** ** ** ** ** 1.2E+09

8.0E+08 H & E

density 4.0E+08

Integrated optical 1.0E+00 Ki-67 α C m A/E 5- Ki-67 HIF1 DNMT3a Caspase-3

Caspase-3 HIF1α promoter HIF1α TSS TSS A/E HIF1α A/E promoter A/E

α TSS A/E HIF1α DNMT3a

HIF1 DNMT3a promoter

DNMT3a DNMT3a DNMT3a CpG island

DNMT3a A/E Global DNA methylation

C Promoter methylation m

- TSS 5 TSG turn off Leukemia growth Figure 7. Pharmacological inhibition of HIF1α suppresses leukemic proliferation in vivo.(a, b) Visual analysis (a) and measurement of tumor size (b) in SKNO-1 xenograft tumor-bearing nude mice (n=3 mice/group) upon echinomycin or vehicle treatment. Data represent means ± s.e.m. (n=6 tumors/group). *Po0.05; **Po0.001. (c) Hematoxylin and eosin (H&E) and immunohistochemistry staining in SKNO-1 xenograft tumors. Representative images for each antibody are shown. Scale bars represent 100 μm. Insets show high-magniﬁcation images of the boxed areas. (d) Quantiﬁcation of immunohistochemical signal is shown in the histogram. Data represent means ± s.e.m. (n=6tumors/group). **Po0.001. (e) Proposed model for the A/E–HIF1α–DNMT3a network in regulating the leukemogenesis. Through the reciprocal promoter binding, A/E and HIF1α form a positive feedback loop in AML cells, which in turn cooperatively upregulate DNMT3a transcription and enhance DNA methylation, hence allowing the epigenetic tumor-suppressor gene silencing and conferring the leukemia cells with survival advantage.

Leukemia (2015) 1730 – 1740 © 2015 Macmillan Publishers Limited AML1/ETO–HIF1α–DNMT3a axis in leukemogenesis XN Gao et al 1739 leukemic driver and establish HIF1α as a reliable molecular marker, coactivators (p300 and PRMT1) to activate targets in cooperation which identify patients with a poor prognosis in an otherwise with H4 arginine 3 methylation and H3 Lys9/14 acetylation.44,45 prognostically favorable subgroup of AML. A/E and HIF1α shape a Given that HIF1α physically interacts with both A/E and p300,46,47 positive regulatory loop and cooperate to upregulate DNMT3a and A/E has direct physical interaction with p300,44 it is likely that expression leading to a DNA hypermethylation profile in AML, A/E could transcriptionally activate DNMT3a via the recruitment of which is different from existing views emphasizing the recruit- HIF1α/p300 complex on DNMT3a promoter, which requires further ment of DNMTs by A/E to target promoters (Figure 7e). validation. HIF1α overexpression strongly associates with unfavorable Recent studies found that HIF1α is upregulated in normoxic prognosis, drug resistance and tumor aggressiveness in multiple tissues,14,48 suggesting that hypoxia is not the only factor that – tumor types.36 38 Here we present evidence that high HIF1α levels modulates HIF1α levels, which is supported by our results that correlated with high A/E levels and predicted inferior survivals in HIF1α was transactivated by A/E in AML cells. Thus HIF1α functions A/E-patients. This cooperative feature of A/E and HIF1α in patients as a transcriptional regulator in a hypoxia-dependent and was experimentally demonstrated in cell lines and mouse models, -independent manner. Further, HIF1α induced and cooperated by showing that A/E and HIF1α co-expression, to the greatest with A/E to regulate DNMT3a-dependent DNA methylation. Thus extent, enhanced leukemia cell growth as compared with we conclude that HIF1α could be an epigenetic target in A/E- individual agent, and HIF1α was directly involved in A/E9a-driven positive AML. Of note, certain reports showed that HIF1α increases leukemia in mice. It is worth emphasizing that the correlation global and promoter methylation,49,50 but others argue that HIF1α between HIF1α expression and patient’s outcomes is KIT mutation- induces DNA hypomethylation.51,52 Although these controversies independent and, importantly, appears in the majority of patients. exist, our findings support that HIF1α mediates DNA hypermethy- Thus we believe that HIF1α serves as a cooperating non-genetic lation in A/E-positive AML. The differences here are probably due event in A/E-driven leukemogenesis. to the subtype of cancer. Further studies are needed to verify Although having a higher complete remission rate,39 up to this point. 70% of A/E-patients relapse for unknown reasons.9,40 Studies In conclusion, our discoveries that A/E–HIF1α cooperation have shown that the copy number of A/E transcripts serves as an determines leukemia cell fate add a new layer to the complexity indicator for relapse.41 However, how A/E is transcriptionally of mechanisms regulating A/E-driven leukemogenesis. The findings regulated remains poorly understood. Our results identified that A/E cooperates with HIF1α in monitoring DNMT3a transcription an A/E–HIF1α regulatory circuit, in which HIF1α and A/E provide an alternative explanation for why chimeric factors transactivated each other by direct promoter binding, thus produce distinctive epigenetic signature. Understanding the establishing the role of HIF1α in controlling A/E expression in underlying mechanisms responsible for the biological and clinical AML. These findings, together with the observation that HIF1α heterogeneous of A/E-driven AML will help to predict leukemia overexpression predicted poor outcomes in A/E-patients, support prognosis and develop personalized treatments. the notion that HIF1α could be a biomarker and therapeutic target being specifictoA/E-positive AML. Actually, our data revealed that treatment with HIF1α inhibitor echinomycin decreased the CONFLICT OF INTEREST expression of A/E and HIF1α, leading to the disruption of leukemic The authors declare no conflict of interest. clonogenic potential in vitro and the attenuation of leukemogenesis in vivo. The chimeric proteins, such as A/E, CBFβ-MYH11 and PML/RARα, ACKNOWLEDGEMENTS have been found to be associated with a unique DNA methylation This work was partially supported by grants from the National Institutes of Health/ patterning in leukemia.16 It is likely that A/E recruits DNMTs to National Cancer Institute (R01CA149623, R01CA104509 and R21CA155915), the target genes,20,22,42 thus partially making contribution to A/E- Hormel Foundation, the National Natural Science Foundation of China (90919044, associated promoter DNA methylation. However, how the DNA 30971297, 81170518, 81000221, 81370010, 81171820 and 81370635), the National Public Health Grand Research Foundation (201202017), the Capital Public Health methylation of non-A/E targets is regulated remains unknown. In Project (Z111107067311070), the Capital Medical Development Scientific Research this study, the positive correlation between A/E and DNMT3a Fund (2007–2040), the Beijing Natural Science Foundation (7122169 and 7112126), raises another possibility that A/E may regulate DNA methylation the Beijing New Stars Program of Science and Technology (2010B075) and the Italian signature through DNMT3a upregulation, specifically for the non- Association for Cancer Research (IG-11949). A/E targets. Indeed, a recent study identified a genome-wide set of target sites for A/E in t(8;21) AML cells by ChIP sequencing.28 Several high-confidence A/E peaks distribute over transcription REFERENCES start site-near region of DNMT3a gene, which supports our ChIP 1 Peterson LF, Zhang DE. The 8;21 translocation in leukemogenesis. Oncogene 2004; assays indicating that A/E was enriched on DNMT3a promoter. In 23: 4255–4262. our study, the regulatory role of A/E in DNMT3a transcription 2 Yuan Y, Zhou L, Miyamoto T, Iwasaki H, Harakawa N, Hetherington CJ et al. AML1- support an innovative notion that chimeric factors (that is, A/E) ETO expression is directly involved in the development of acute myeloid leukemia in 98 – could drive a unique epigenetic signature in AML through (1) the the presence of additional mutations. Proc Natl Acad Sci USA 2001; : 10398 10403. recruitment of DNMTs to target gene resulting in promoter DNA 3 Wang YY, Zhou GB, Yin T, Chen B, Shi JY, Liang WX et al. AML1-ETO and C-KIT mutation/overexpression in t(8;21) leukemia: implication in stepwise leukemo- hypermethylation; (2) the direct transactivation of DNMTs (that is, genesis and response to Gleevec. Proc Natl Acad Sci USA 2005; 102: 1104–1109. DNMT3a) to regulate the methylation of non-targets, including 4 Schessl C, Rawat VP, Cusan M, Deshpande A, Kohl TM, Rosten PM et al. The AML1- non-promoter regions (Figure 7e). It is unclear which pathway is ETO fusion gene and the FLT3 length mutation collaborate in inducing acute dominant in AML, but the observations that majority of genes do leukemia in mice. J Clin Invest 2005; 115: 2159–2168. not contain A/E-binding sites43 suggest that A/E-driven DNMT3a 5 Schneider F, Bohlander SK, Schneider S, Papadaki C, Kakadyia P, Dufour A et al. upregulation could be more essential for A/E-specific DNA AML1-ETO meets JAK2: clinical evidence for the two hit model of leukemogenesis methylation patterning. Its worth noting that A/E is generally from a myeloproliferative syndrome progressing to acute myeloid leukemia. 21 – considered to act as a transcriptional repressor through the Leukemia 2007; :2199 2201. physical interaction and recruitment of co-repressors (histone 6 Peterson LF, Boyapati A, Ahn EY, Biggs JR, Okumura AJ, Lo MC et al. Acute myeloid leukemia with the 8q22;21q22 translocation: secondary mutational events and deacetylases, DNMTs, nuclear receptor co-repressor) to target 110 – 17,19,22 alternative t(8;21) transcripts. Blood 2007; :799 805. genes, particularly tumor suppressors. Although it is not as 7 Kuchenbauer F, Schnittger S, Look T, Gilliland G, Tenen D, Haferlach T et al. Iden- common as repressive activities, reports did demonstrate that A/E tification of additional cytogenetic and molecular genetic abnormalities in acute could be a transcriptional activator by the recruitment of myeloid leukaemia with t(8;21)/AML1-ETO. Br J Haematol 2006; 134:616–619.

© 2015 Macmillan Publishers Limited Leukemia (2015) 1730 – 1740 AML1/ETO–HIF1α–DNMT3a axis in leukemogenesis XN Gao et al 1740 8 Dohner K, Du J, Corbacioglu A, Scholl C, Schlenk RF, Dohner H. JAK2V617F 31 Mole DR, Blancher C, Copley RR, Pollard PJ, Gleadle JM, Ragoussis J et al. Genome- mutations as cooperative genetic lesions in t(8;21)-positive acute myeloid wide association of hypoxia-inducible factor (HIF)-1alpha and HIF-2alpha DNA leukemia. Haematologica 2006; 91: 1569–1570. binding with expression profiling of hypoxia-inducible transcripts. J Biol Chem 9 Paschka P, Marcucci G, Ruppert AS, Mrozek K, Chen H, Kittles RA et al. Adverse 2009; 284: 16767–16775. prognostic significance of KIT mutations in adult acute myeloid leukemia with inv(16) 32 Yan M, Kanbe E, Peterson LF, Boyapati A, Miao Y, Wang Y et al. A previously and t(8;21): a Cancer and Leukemia Group B Study. JClinOncol2006; 24:3904–3911. unidentified alternatively spliced isoform of t(8;21) transcript promotes leuke- 10 Keith B, Simon MC. Hypoxia-inducible factors, stem cells, and cancer. Cell 2007; mogenesis. Nat Med 2006; 12:945–949. 129: 465–472. 33 Kong D, Park EJ, Stephen AG, Calvani M, Cardellina JH, Monks A et al. Echino- 11 Stroka DM, Burkhardt T, Desbaillets I, Wenger RH, Neil DA, Bauer C et al. HIF-1 is mycin, a small-molecule inhibitor of hypoxia-inducible factor-1 DNA-binding expressed in normoxic tissue and displays an organ-specific regulation under activity. Cancer Res 2005; 65: 9047–9055. systemic hypoxia. FASEB J 2001; 15: 2445–2453. 34 Herman JG, Jen J, Merlo A, Baylin SB. Hypermethylation-associated inactivation 12 Kuschel A, Simon P, Tug S. Functional regulation of HIF-1alpha under normoxia--is indicates a tumor suppressor role for p15INK4B. Cancer Res 1996; 56:722–727. there more than post-translational regulation? J Cell Physiol 2012; 227:514–524. 35 Higuchi M, O'Brien D, Kumaravelu P, Lenny N, Yeoh EJ, Downing JR. Expression of 13 Gerber SA, Pober JS. IFN-alpha induces transcription of hypoxia-inducible factor- a conditional AML1-ETO oncogene bypasses embryonic lethality and establishes a 1alpha to inhibit proliferation of human endothelial cells. J Immunol 2008; 181: murine model of human t(8; 21) acute myeloid leukemia. Cancer cell 2002; 1: 1052–1062. 63–74. 14 Wang Y, Liu Y, Malek SN, Zheng P. Targeting HIF1alpha eliminates cancer stem 36 Vaupel P. The role of hypoxia-induced factors in tumor progression. Oncologist cells in hematological malignancies. Cell Stem Cell 2011; 8: 399–411. 2004; 9(Suppl 5): 10–17. 15 Matsunaga T, Imataki O, Torii E, Kameda T, Shide K, Shimoda H et al. Elevated 37 Unruh A, Ressel A, Mohamed HG, Johnson RS, Nadrowitz R, Richter E et al. The HIF-1alpha expression of acute myelogenous leukemia stem cells in the endosteal hypoxia-inducible factor-1 alpha is a negative factor for tumor therapy. Oncogene hypoxic zone may be a cause of minimal residual disease in bone marrow after 2003; 22: 3213–3220. chemotherapy. Leuk Res 2012; 36: e122–e124. 38 Baba Y, Nosho K, Shima K, Irahara N, Chan AT, Meyerhardt JA et al. HIF1A over- 16 Figueroa ME, Lugthart S, Li Y, Erpelinck-Verschueren C, Deng X, Christos PJ et al. expression is associated with poor prognosis in a cohort of 731 colorectal cancers. DNA methylation signatures identify biologically distinct subtypes in acute Am J Pathol 2010; 176: 2292–2301. myeloid leukemia. Cancer cell 2010; 17:13–27. 39 Grimwade D, Walker H, Oliver F, Wheatley K, Harrison C, Harrison G et al. The 17 Linggi B, Muller-Tidow C, van de Locht L, Hu M, Nip J, Serve H et al. The t(8;21) importance of diagnostic cytogenetics on outcome in AML: analysis of 1,612 fusion protein, AML1 ETO, specifically represses the transcription of the p14(ARF) patients entered into the MRC AML 10 trial. The Medical Research Council Adult tumor suppressor in acute myeloid leukemia. Nat Med 2002; 8:743–750. and Children's Leukaemia Working Parties. Blood 1998; 92: 2322–2333. 18 Kitabayashi I, Ida K, Morohoshi F, Yokoyama A, Mitsuhashi N, Shimizu K et al. The 40 Cairoli R, Beghini A, Grillo G, Nadali G, Elice F, Ripamonti CB et al. Prognostic AML1-MTG8 leukemic fusion protein forms a complex with a novel member of impact of c-KIT mutations in core binding factor leukemias: an Italian the MTG8(ETO/CDR) family, MTGR1. Mol Cell Biol 1998; 18:846–858. retrospective study. Blood 2006; 107: 3463–3468. 19 Wang J, Hoshino T, Redner RL, Kajigaya S, Liu JM. ETO, fusion partner in t(8;21) 41 Yoo SJ, Chi HS, Jang S, Seo EJ, Seo JJ, Lee JH et al. Quantification of AML1-ETO acute myeloid leukemia, represses transcription by interaction with the human fusion transcript as a prognostic indicator in acute myeloid leukemia. Haemato- N-CoR/mSin3/HDAC1 complex. Proc Natl Acad Sci USA 1998; 95: 10860–10865. logica 2005; 90:1493–1501. 20 Liu S, Shen T, Huynh L, Klisovic MI, Rush LJ, Ford JL et al. Interplay of RUNX1/MTG8 and 42 Li Y, Gao L, Luo X, Wang L, Gao X, Wang W et al. Epigenetic silencing of DNA methyltransferase 1 in acute myeloid leukemia. Cancer Res 2005; 65: 1277–1284. microRNA-193a contributes to leukemogenesis in t(8; 21) acute myeloid leukemia 21 Liu Y, Cheney MD, Gaudet JJ, Chruszcz M, Lukasik SM, Sugiyama D et al. The by activating the PTEN/PI3K signal pathway. Blood 2013; 121:499–509. tetramer structure of the Nervy homology two domain, NHR2, is critical for AML1/ 43 Maiques-Diaz A, Chou FS, Wunderlich M, Gomez-Lopez G, Jacinto FV, Rodriguez- ETO's activity. Cancer cell 2006; 9:249–260. Perales S et al. Chromatin modifications induced by the AML1-ETO fusion protein 22 Fazi F, Racanicchi S, Zardo G, Starnes LM, Mancini M, Travaglini L et al. Epigenetic reversibly silence its genomic targets through AML1 and Sp1 binding motifs. silencing of the myelopoiesis regulator microRNA-223 by the AML1/ETO onco- Leukemia 2012; 26: 1329–1337. protein. Cancer cell 2007; 12:457–466. 44 Wang L, Gural A, Sun XJ, Zhao X, Perna F, Huang G et al. The leukemogenicity of 23 Liu S, Wu LC, Pang J, Santhanam R, Schwind S, Wu YZ et al. Sp1/NFkappaB/HDAC/ AML1-ETO is dependent on site-specific lysine acetylation. Science 2011; 333: miR-29b regulatory network in KIT-driven myeloid leukemia. Cancer cell 2010; 17: 765–769. 333–347. 45 Shia WJ, Okumura AJ, Yan M, Sarkeshik A, Lo MC, Matsuura S et al. PRMT1 24 Shen N, Yan F, Pang J, Wu LC, Al-Kali A, Litzow MR et al. A nucleolin-DNMT1 interacts with AML1-ETO to promote its transcriptional activation and progenitor regulatory axis in acute myeloid leukemogenesis. Oncotarget 2014; 5: 5494–5509. cell proliferative potential. Blood 2012; 119:4953–4962. 25 deJonge HJ, Valk PJ, Veeger NJ, ter Elst A, den Boer M L, Cloos J et al. High VEGFC 46 Peng ZG, Zhou MY, Huang Y, Qiu JH, Wang LS, Liao SH et al. Physical and func- expression is associated with unique gene expression profiles and predicts tional interaction of Runt-related protein 1 with hypoxia-inducible factor-1alpha. adverse prognosis in pediatric and adult acute myeloid leukemia. Blood 2010; Oncogene 2008; 27:839–847. 116: 1747–1754. 47 Freedman SJ, Sun ZY, Poy F, Kung AL, Livingston DM, Wagner G et al. Structural 26 Litz J, Krystal GW. Imatinib inhibits c-Kit-induced hypoxia-inducible factor-1alpha basis for recruitment of CBP/p300 by hypoxia-inducible factor-1 alpha. Proc Natl activity and vascular endothelial growth factor expression in small cell lung Acad Sci USA 2002; 99: 5367–5372. cancer cells. Mol Cancer Ther 2006; 5: 1415–1422. 48 Fahling M, Persson AB, Klinger B, Benko E, Steege A, Kasim M et al. Multilevel 27 Gardini A, Cesaroni M, Luzi L, Okumura AJ, Biggs JR, Minardi SP et al. AML1/ETO regulation of HIF-1 signaling by TTP. Mol Biol Cell 2012; 23: 4129–4141. oncoprotein is directed to AML1 binding regions and co-localizes with AML1 and 49 Lee JS, Kim Y, Bhin J, Shin HJ, Nam HJ, Lee SH et al. Hypoxia-induced methylation HEB on its targets. PLoS Genet 2008; 4: e1000275. of a pontin chromatin remodeling factor. Proc Natl Acad Sci USA 2011; 108: 28 Ptasinska A, Assi SA, Mannari D, James SR, Williamson D, Dunne J et al. 13510–13515. Depletion of RUNX1/ETO in t(8;21) AML cells leads to genome-wide changes in 50 Robinson CM, Neary R, Levendale A, Watson CJ, Baugh JA. Hypoxia-induced DNA chromatin structure and transcription factor binding. Leukemia 2012; 26:1829–1841. hypermethylation in human pulmonary fibroblasts is associated with Thy-1 pro- 29 Ortiz-Barahona A, Villar D, Pescador N, Amigo J, del Peso L. Genome-wide iden- moter methylation and the development of a pro-fibrotic phenotype. Respir Res tification of hypoxia-inducible factor binding sites and target genes by a prob- 2012; 13: 74. abilistic model integrating transcription-profiling data and in silico binding site 51 Shahrzad S, Bertrand K, Minhas K, Coomber BL. Induction of DNA hypomethyla- prediction. Nucleic Acids Res 2010; 38: 2332–2345. tion by tumor hypoxia. Epigenetics 2007; 2: 119–125. 30 Xia X, Lemieux ME, Li W, Carroll JS, Brown M, Liu XS et al. Integrative analysis of 52 Liu Q, Liu L, Zhao Y, Zhang J, Wang D, Chen J et al. Hypoxia induces genomic DNA HIF binding and transactivation reveals its role in maintaining histone methyla- demethylation through the activation of HIF-1alpha and transcriptional upregu- tion homeostasis. Proc Natl Acad Sci USA 2009; 106:4260–4265. lation of MAT2A in hepatoma cells. Mol Cancer Ther 2011; 10: 1113–1123.

Supplementary Information accompanies this paper on the Leukemia website (http://www.nature.com/leu)