Epigenetic down-regulation of the HIST1 locus predicts better prognosis in acute myeloid leukemia with NPM1 mutation

The epigenetic machinery is frequently altered in acute myeloid leukemia. Focusing on cytogenetically normal (CN) AML, we previously described an abnormal H3K27me3 enrichment covering 70 kb on the HIST1 cluster (6.p22) in CN-AML patient blasts. Here, we further investigate the molecular, functional, and prognosis significance of this epigenetic alteration named H3K27me3 HIST1 in NPM1-mutated (NPM1mut) CN-AML. We found that three quarter of the NPM1mut CN-AML patients were H3K27me3 HIST1high. H3K27me3 HIST1high group of patients was associated with a favorable outcome independently of known molecular risk factors. In gene expression profiling, the H3K27me3 HIST1high mark was associated with lower expression of the histone genes HIST1H1D, HIST1H2BG, HIST1H2AE, and HIST1H3F and an upregulation of genes involved in myelomonocytic differentiation. Mass spectrometry analyses confirmed that the linker histone protein H1d, but not the other histone H1 subtypes, was downregulated in the H3K27me3 HIST1high group of patients. H1d knockdown primed ATRA-mediated differentiation of OCI-AML3 and U937 AML cell lines, as assessed on CD11b/CD11c markers, morphological and gene expression analyses. Our data suggest that NPM1mut AML prognosis depends on the epigenetic silencing of the HIST1 cluster and that, among the H3K27me3 silenced histone genes, HIST1H1D plays a role in AML blast differentiation.


Background
Acute myeloid leukemias (AMLs) are a heterogeneous group of severe hematological malignancies that arise through the acquisition of oncogenic mutations by hematopoietic progenitor cells. Patient prognosis mainly relies on the given treatment, the outcome of which depends on cytogenetics and molecular alterations. Cytogenetically normal (CN) AML patients are usually assigned to an intermediate prognosis group that can be further subdivided through the detection of mutations in a growing number of genes [1]. Mutations in the nucleophosmin 1 (NPM1) gene are the commonest molecular lesions occurring in ≥ 50% of cases with cytogenetically normal acute myeloid leukemia (CN-AML). NPM1 mutations result in the generation of a nuclear export signal causing the delocalization of the protein from the nucleoli to the cytoplasm [2]. Analyses of large numbers of patients have shown that NPM1 mutations are associated with a relatively favorable prognosis, which can be mitigated by internal tandem duplications (ITD) of the tyrosine kinase receptor Fms-like tyrosine kinase 3 (FLT3) and mutations in DNA-methyl transferase 3A (DNMT3A). Indeed, in NPM1mut CN-AML, FLT3ITD mutation, and/ or mutation in DNMT3A predict an increased risk of relapse and poorer outcome [3], [4]. Recent reports shed light on the importance of epigenetic deregulations that affect the epigenome and gene transcription in AML pathogenesis [5]. These deregulations are the consequence of numerous alterations found in genes encoding multiple classes of epigenetic proteins as DNA methylation and histone modification enzymes [6].
In addition to these mutations, recent reports underlined the importance of histone genes themselves in cancer onset. For example, the major histone cluster 1 (HIST1), encoding the vast majority of the redundant core and linker histones [7], has been found partially deleted in ALL [8] and deregulation of its expression is associated with breast cancer invasiveness [9]. Moreover, we previously described a new alteration that affects the HIST1 cluster in AML. This alteration consists in a marked histone H3 lysine 27 tri-methylation (H3K27me3) enrichment encompassing 70 kb of the HIST1 cluster, affecting histone genes and associates with the NPM1 mutation and a better leukemia-free survival [10]. In this study, we reported the clinical importance of this newly identified epigenetic alteration, called the H3K27me3 HIST1 mark, in relation to other known mutations and its functional consequences on the biology of CN-AML leukemic cells. We revealed that H3K27me3 HIST1 status and histone mRNA and protein levels define clinically and biologically different subgroups of NPM1mut CN-AML suggesting their importance in AML pathogenesis.

H3K27me3 level on HIST1 locus is associated with better survival
To characterize the H3K27me3 HIST1 mark, we performed H3K27me3 ChIP-qPCR on samples obtained from 46 de novo CN-AML patients included in GOELAMS multicenter clinical trials LAM2006IR (NCT00860639) or LAM2007SA (NCT00590837). H3K27me3 level was determined at five HIST1 genomic locations that are representative of the H3K27me3 HIST1 signature, as described previously [10]. Heatmap showing H3K27me3 HIST1 gene enrichment confirmed the variation of H3K27me3 HIST1 level among CN-AML patients (Fig. 1a). The average of the five normalized H3K27me3 HIST1 values was calculated and this index showed a clear segregation of the H3K27me3 HIST1 low and H3K27me3 HIST1 high patients. With a cut-off value at 15, approximately 55% of CN-AML samples displayed an H3K27me3 HIST1 enrichment mark (Fig. 1b).
H3K27me3 level on HIST1 locus is an independent biomarker predicting survival of NPM1mut CN-AML NPM1-mutated AMLs represent a distinct clinical and biological entity in the World Health Organization (WHO) classification, commonly associated with a better prognosis [3]. We used the biological material obtained from 25 NPM1mut samples previously analyzed [10], and 78 samples either provided by the GOELAMSTHEQUE (n = 33), or the IPC/ CRCM tumor bank (n = 45) (Fig. 2a). H3K27me3 status of the 78 new samples was analyzed (Additional file 2: Figure  S1) and revealed that 75% of NPM1mut CN-AMLs were H3K27me3 HIST1 high . The NPM1mut H3K27me3 HIST1 high subgroup of patients was not enriched with DNMT3A or FLT3ITD mutations, the most frequently NPM1mut cooccurring alterations [11] (Table 1 and Additional file 2: Figure S1), but IDH2R140 was significantly overrepresented in this subgroup in comparison with the NPM1mut H3K27me3 HIST1 low group (27.6% vs. 7.6%, P = .05). Interestingly, H3K27me3 HIST1 high leukemic cells had a significantly lower CD34 expression than their H3K27me3 HIST1 low counterparts (CD34 mean expression, 10.3% vs. 35%, P = 0.005) ( Table 1). To further explore a potential association between CD34 expression and the presence of H3K27me3 HIST1 mark, we selected four patient samples (two in each H3K27me3 HIST1 group), containing both CD34low and CD34high blast populations. We next analyzed the H3K27me3 HIST1 level in CD34low and CD34high fluorescence-activated cell sorted (FACS) blasts. As shown in Fig. 2b and Additional file 2: Figure S2, H3K27me3 HIST1 status (high or low), in both CD34 low and CD34 high blast fractions, was similar to the bulk population, indicating that H3K27me3 HIST1 high occurs independently of CD34 level of expression.
Our results suggest that H3K27me3 HIST1 status is an independent epigenetic marker that identifies patients with a poor outcome within the NPM1/FLT3wt group of patients.
Histone mRNA expression is anti-correlated to H3K27me3 HIST1 level and predicts NPM1mut CN-AML patient outcome To analyze the anti-correlation of histone mRNA level and the presence of the repressive H3K27me3 mark, we selected three histone genes; HIST1H1D, HIST1H2BG and HIST1H2BH, spread over the H3K27me3 HIST1 islet and associated with clinical outcome in public data (see below), and measured their mRNA levels. Expression of these three genes was lower in H3K27me3 HIST1 high patients (n = 34) than in H3K27me3 HIST1 low patients (n = 13) (Fig. 3a).
We next asked whether expression of these genes, as a consequence of H3K27me3 mark, was associated with patient survival. Given the small size of our cohort, we analyzed HIST1 gene expression in two published cohorts with publicly accessible clinical and mRNA expression data: TCGA [13] and Metzeler [14]. NPM1mut CN-AML patients, within these two cohorts, were identified by using a published gene expression signature that predicts the NPM1 mutational status [15] (see Additional file 3: Supplemental methods). Association of histone expression with survival was first tested for each of the 11 histone genes covered by the H3K27me3 HIST1 mark. This highlighted three histone genes, HIST1H1D, HIST1H2BG, and HIST1H2BH, for which a high level of expression was associated with a poor outcome (P = .004, .015 and .044 respectively, Fig. 3b). Then, we tested this 3-HIST1-mRNA signature in univariate analysis; 3-HIST1-mRNA low patients had a favorable OS with a median OS of 17.7 months versus 9.6 months (HR = 1.66, range, 1.13-2.42, P = .009) (Fig. 3c). Multivariate analyses showed that the 3-HIST1-mRNA low status was associated with a better prognosis (HR = 1.60, range 1.60-2.31, P = .01), independently of ChIP-qPCR on the indicated HIST1 genes. Enrichment was calculated as the percentage of bound/input and double normalized with HOXD4 and GAPDH. Each column represents a patient sample (n = 46) sorted by unsupervised hierarchical clustering. The separation between H3K27me3 HIST1 low and H3K27me3 HIST1 high groups is indicated by a dotted lane. Yellow square: mutation for the indicated gene. b Bimodal distribution of the patients according to H3K27me3 HIST1 mean level, obtained from the H3K27me3 enrichment values of the 5 HIST1 regions analyzed. In blue, patients with an H3K27me3 HIST1 mean value below 15 (H3K27me3 HIST1 low , n = 21), in red, patients with a mean value above 15 (H3K27me3 HIST1 high , n = 25). c OS and EFS in CN-AML patients according to H3K27me3 HIST1 low and H3K27me3 HIST1 high status other prognosis markers including age, FAB classification and FLT3 status (Additional file 1: Table S3). These results show that H3K27me3 HIST1 high is associated with a lower expression of histone genes, and that 3-HIST1-mRNA low signature defines a NPM1mut AML patient group with a better outcome.
Gene expression profiling associated with H3K27me3 HIST1 high identifies a "mature like" phenotype We next characterized the gene expression profile (GEP) of H3K27me3 HIST1 high samples (n = 16) in comparison to H3K27me3 HIST1 low samples (n = 11) from the IPC cohort (see Additional File 1: Table S4 for patient clinical characteristics). Eighty-one genes were differentially expressed (p < .05, fold-change > 1.5) between the two groups, 58 being up-and 23 being downregulated in the H3K27me HIST1 high group ( Fig. 4a and Additional file 4: Table S5). Analysis of enhancer of zeste homolog 2 (EZH2) and suppressor of zeste 12 (SUZ12) expression revealed that the two groups, H3K27me3 HIST1 high and H3K27me3 HIST1 low , equally expressed PRC2 components (Fig. 4b). GSEA identified, in H3K27me3 HIST1 high patients, genes associated with myelomonocytic differentiation such as immune or inflammatory responses ( Fig. 4c and Additional file 4: Table S5). Downregulated genes in these patients belong to cell cycle and chromatin regulation categories, including histone genes from the HIST1 cluster ( Fig. 4c and Additional file 4: Table S5). Using qPCR, we confirmed higher expression of three genes involved, in mature granulocyte functions, CYBB, FCN1, and CLEC4A [16], [17], [18] in H3K27me3 HIST1 high patients (Fig. 4d).
H3K27me3 level at the promoter of these genes was identical between H3K27me3 HIST1 low and H3K27me3 HIST1 high patients (Additional file 2: Figure S3) suggesting that the variation of expression was indirectly affected by H3K27me3 HIST1 high .
To further validate the relation between low mRNA level of HIST1 genes and the expression of granulocytic markers, we tested mRNA expression of myelomonocytic maturation genes (CYBB, FCN1, CLEC4, and ITGAM) in the TCGA and Metzeler cohorts of patients stratified with the previously defined 3-HIST1-mRNA signature. The 3-HIST1-mRNA low patient group overexpressed the differentiation genes in comparison to the 3-HIST1-mRNA high one (Fig. 4e), thus corroborating our previous observation (Fig. 4d). Reciprocally, genes such as SOCS2, CDK6, LAPTM4B, and NGFRAP1, which were described as associated with a leukemic stem cell signature [19], were less expressed in the 3-HIST1-mRNA low patient group (Fig. 4f).
Taken together, these results suggest that HIST1 mRNA downregulation by the H3K27me3 HIST1 high mark is associated with a more differentiated phenotype related to a committed state of leukemic cells.  Only histone genes covered by the H3K27me3 HIST1 mark described in [10] were analyzed. c Survival analyzed according to the expression of HIST1H1D, HIST1H2BG, and HIST1H2BH mRNA on independent cohorts of patients (TGCA and Metzeler). Patients were split into two groups according to the expression of HIST1H1D, HIST1H2BH and HIST1H2BG genes using a Cox regression prognosis model. Statistical significance was estimated using the log-rank test The histone linker H1-3 is poorly expressed in H3K27me3 HIST1 high CN-AML patients To study the role of histones on AML clinical and biological features, we looked at the effect of H3K27me3 HIST1 epigenetic silencing on histone protein level. First, we looked at the proportions of total histones and of each histone subtype (Additional file 2: Figure S4) in chromatin-bound fractions extracted from a series of 12 patient samples (six in each group) using intensity-based absolute quantification (iBAQ) approach. Normalized quantities of total linker histone H1 and core histones H2A, H2B, H3, and H4 were similar in both H3K27me3 HIST1 high and H3K27me3 HIST1 low patients (Additional file 2: Figure S5), suggesting that H3K27me3 HIST1 high status did not globally modify histone protein abundance. Then, we decided to analyze specifically the histone linker H1-3, encoded by HIST1H1D, because its mRNA level is affected by H3K27me3 HIST1 status (Fig. 3a) and it is the leading gene for the mRNA signature (Fig. 3b). In addition, contrary to other histone subtypes (i.e., H2A, H2B, H3, and H4), H1 histone subtypes are heterogeneous in amino acid composition [20], which probably reflects a subtype-specific function. Indeed, when looking at the H1 subtype abundance, we observed that the H1-3 subtype was decreased in the H3K27me3 HIST1 high group (normalized iBAQ value (Log2) = 6.09 vs 4.74; P = .04), whereas the other H1 subtypes, H1-5, H1-2, H1-4, H1-F0, and H1-FX, were unaffected (Fig. 5a). These results are consistent with the specific HIST1H1D mRNA expression decrease observed in AML samples harboring the H3K27me3 HIST1 mark. We confirmed the lower expression of H1-3 observed in H3K27me3 HIST1 high group in comparison to H3K27me3 (lysosome-associated protein transmembrane-4β), and NGFRAP1 (nerve growth factor receptor-associated protein 1) gene expression was analyzed. Statistical significance was estimated using Mann Whitney test * p < 0.05; **p < 0.005 ***p < 0.0001. NS non-significant HIST1 low group of patients by Western blot using pan H1 and specific H1-3 antibodies (Fig. 5b).
In conclusion, as a consequence of the presence of an H3K27me3 islet, NPM1mut CN-AML H3K27me3 HIST1 high patients express low level of H1-3.

H1-3 knockdown confers a more mature phenotype in AML cell lines
We hypothesized that the better prognosis of H1-3 low AML may be related to a gain of differentiation and a loss of stem cell features of the AML blasts. Thus, we tested the consequences of H1-3 knockdown (KD) on differentiation potential of the NPM1-mutated OCI-AML3 cell line, which also harbor the DNMT3AR882C mutation [21]. Efficiency and specificity of our KD were assessed by testing mRNA levels of the different H1 subtypes ( Fig. 6a and Additional file 2: Figure S6a and S6b) and by measuring H1-3 protein level, after H1-3 KD induction ( Fig. 6b and Additional file 2: Figure S6c and S6d). Consequences of H1-3 KD on differentiation were evaluated upon all-trans retinoic acid (ATRA)-treatment. Albeit no increase in CD11b was observed upon H1-3 KD alone, addition of ATRA (0.5 μM and 1 μM) induced a significant increase in CD11b expression, with a marked increase at 0.5 μM (22.6 ± 2.5% vs. 41 ± 4.3%; P = .008) (Fig. 6c and Additional file 2: Figure S7a and S7b) and a significant increase in the proportion of the double positive CD11b/CD11c population (29.8 ± 1.3% vs. 42.5 ± 2.1%; P = .003) (Fig. 6d). In addition, morphological and quantification analyses showed that cytoplasmic granules, which reflect the beginning of a maturation process appeared upon H1-3 KD after 96 h of ATRA-treatment (0.5 μM), (Fig. 6e, f and Additional file 2: Figure S8). Finally, mRNA expression levels of two ATRA-induced genes, CYBB and ITGAM, were tested in H1-3 KD condition under ATRA-treatment; H1-3 downregulation increased the amplitude of ATRAinduced upregulation of these two genes (Fig. 6g). To test whether this ATRA-sensitization was dependent of the presence of NPM1mut or DNMT3AR882C, we performed H1-3 KD in the NPM1wt and DNMT3Awt myeloid U937 cell line. Interestingly, H1-3 KD increased the proportion of the double positive CD11b/CD11c in 0.1 μM ATRA-treated U937 cells and induced morphological changes (Fig. 6h). Altogether, these results suggest that downregulation of histone H1-3 induces ATRA-sensitization independently of NPM1 and DNMT3A mutations.

Discussion
Aberrant epigenetic patterns in leukemia have been described but determination of their importance in leukemia onset and progression is still lacking [22]. We have studied here the clinical and biological consequences of an epigenetic alteration of the HIST1 locus, the H3K27me3 HIST1 signature, which we previously described in CN-AML [10]. We first confirmed the presence of H3K27me3 HIST1 high on 55% of CN-AML and 75% of NPM1mut CN-AML in an independent CN-AML cohort. H3K27me3 HIST1 high impacts on the survival of CN-AML and NPM1mut CN-AML patients, and confers a better survival independently of other molecular markers. While H3K27me3 HIST1 high is not associated with FLT3ITD or DNMT3A mutations,  IDH2R140Q is overrepresented in H3K27me3 HIST1 high patients (28% vs 7.6%); this is consistent with data highlighting the functional relationship between IDH2R140Q mutant expression and histone hypermethylation [23]. Although we do not know the mechanisms underlying the H3K27me3 HIST1 high signature, this focal high level of H3K27me3 may reflect a maintained activity of EZH2 previously shown to be associated with favorable outcome in AML [24]. Our data provide the first evidence that detecting an aberrant epigenetic pattern in addition to gene mutations has clinical interest and could guide therapeutic choices. Detection of our epigenetic biomarker in AML demonstrates that H3K27me3 HIST1 high status is invariably associated with a favorable outcome and may reveal a less aggressive disease. The less aggressive form of AML associated with H3K27me3 HIST1 high is supported by its anti-correlated stem cell signature, shown to be associated with an adverse prognosis [19]. It is also supported by its inflammatory and immune function signature, which has been associated with a better clinical response to dexamethasone in NPM1 CN-AML [25]. However, although CD34low AMLs, characterized by a leukemic stem cell arrested at a precursor-like stage [26], are enriched in the H3K27me3 HIST1 high group, we could not find H3K27me3 HIST1 high exclusively on CD34low blasts. This suggests that H3K27me3 HIST1 high marks a more mature leukemia independently of the level of CD34 expression.
The main consequence of the H3K27me3 accumulation at the HIST1 locus may be the downregulation of the histone genes affected by the repressive epigenetic mark. Levels of replication-dependent histone gene expression may reflect aggressiveness of the disease and may have a survival impact. Indeed, the major histone gene cluster has been described as one of the most upregulated across breast cancer progression [9]. Our study highlighted a peculiar pattern of a lower histone gene expression and defined a three-HIST1 mRNA signature, containing three histone genes (HIST1H1D, HIST1H2BH, and HIST1H2BG) that are directly affected by the H3K27me3 enrichment and can predict survival on CN-AML patients. Interestingly, one of the most affected histone genes in our AML data encodes the histone linker H1-3. High level of H1.3 has been previously linked to cancer aggressiveness, as its overexpression is associated with malignant ovarian adenocarcinoma [27] and with a poor pancreatic ductal adenocarcinoma survival [28]. Given its tight link with differentiation blockage in cancer cells [29] and given that H1 subtype individual KDs induce changes, notably in genes involved in cell cycle and chromatin regulation [30], diminution of H1.3 level appears to be a valid effector of the less aggressive H3K27me3 HIST1 high AML phenotype. If histone linker H1 subtypes have a well-described redundant roles in maintaining nucleosome architecture and regulating transcription [31], it becomes evident that H1 subtypes have also subtype-specific functions reflected by selective genomic binding and regulation of chromatin organization [20,32]. H1 subtypes would influence chromatin compaction at definite loci affecting specific gene expression in a tissue specific manner. Thus, it is tempting to speculate that in our H3K27me3 HIST1 high model, low H1-3 level would influence the differentiation state of the cells due to a specific change in chromatin targeting. In our cellular models, KD of H1-3 does not induce differentiation but sensitizes the cells to ATRA-treatment. Interestingly, this sensitivity is independent of DNMT3A activity, as the two cell line models (OCI-AML3 DNMT3Amut and U937 DNMT3Awt) are both sensitive to ATRA-treatment after H1-3 KD. In the light of previous reports, which documented an interaction between H1 proteins and DNMT3B [33,34], one possibility is that DNMT3B, although not found mutated in AML, (See figure on previous page.) Fig. 6 H1-3 KD promotes granulocytic differentiation in ATRA-treated AML cell lines. a qPCR expression analysis of the indicated histone H1 genes in shCtrl and shH1-3 (KD#1) without (Dox-) or with 6-days-induction of doxycycline (Dox+). Data represent three independent Dox inductions. Results are normalized on HPRT and expressed in fold change (FC) between Dox+ and Dox− conditions. b Immunoblot of H1-3, H1, and H3 in chromatin-bound extracts from shCtrl and KD#1 upon (+) or not (−) doxycycline (Dox). c, d Cell surface analysis of granulocytic markers consecutive to H1-3 KD in OCI-AML3 clones. c Percentage of CD11b positive cells in shCtrl and KD#1 upon doxycycline induction (Dox+) or not (Dox−) without ATRA (wo ATRA) or with 72 h of ATRA-treatment (0.5 μM or 1 μM). d Percentage of CD11b-CD11c cells upon 72 h of ATRAtreatment (0.5 μM). Data represent three independent experiments. Statistical significance was estimated using Mann Whitney test *p < 0.05; **p < 0.005. NS non-significant. e, f Cytological analysis of shCtrl and KD#1 clones upon doxycycline and after 96 h of ATRA treatment (0.5 μM). e May-Grünwald Giemsa coloration. Black arrows are pointing cytoplasmic azurophilic granules. f Quantification of cells with cytoplasmic granules. Results are presented as a percentage of positive cells (more than two granules). Statistical significance was estimated using Mann-Whitney test *p < 0.05; **p < 0.005. g Expression analysis of CYBB and ITGAM in untreated OCI-AML3 and in ATRA-treated (1 μM) KD#1 without or with doxycycline induction. Gene expression was normalized to two housekeeping genes (PGK1 and PPIA). Data represent three independent experiments. Statistical significance was estimated using T-test (one-tailed p value). h U937 cells were stably infected by a doxycycline inducible shCtrl (U-shCtrl) or shH1-3 (U-KD#1 and U-KD#2). Left panel: percentage of CD11b-CD11c cells upon 72 h of ATRA-treatment (0.1 μM). Level of H1d in the different conditions was measured by immunoblot. Right panel: May-Grünwald Giemsa staining of KD#1 clone with or without doxycycline upon 72 h hours of ATRA-treatment. Data represent two independent experiments. Statistical significance was estimated using unpaired Two-tailed t test * p < 0.05; ***p < 0.001 may be an influential epigenetic partner in AML and in ATRA sensitivity together with H1-3.
ATRA is successfully used for the treatment of acute promyelocytic leukemia (APL) by inducing terminal granulocytic differentiation of APL blasts [35]. To a lesser extent, ATRA can also induce differentiation in non-APL cells, in particular in AML with IDH1/IDH2 mutations [36] or induce cell death in NPM1mut AML when used in combination with Arsenic trioxide [37], [38]. Interestingly, ATRA sensitization of non-APL cells has been induced by targeting epigenetic enzymes: inhibition of histone deacetylase (HDAC) with valproic acid [39] or inhibition of the histone demethylase LSD1/ KDM1A [40] or by targeting SUMOylation [41]. In the same line, our study suggests that epigenetic modifications induced by H1-3 targeting could prime AML cells toward differentiation, revealed by ATRA-sensitivity, which may explains the more mature phenotype found in H3K27me3 HIST1 high leukemia and suggests that ATRA could be an efficient differentiating agent in AML with low H1-3 expression.

Conclusions
We showed that epigenetic silencing of a part of the HIST1 locus by the H3K27me3 mark is associated with a better outcome and a mature gene expression profile in NPM1mut CN-AML and we observed an important role of histone linker H1-3 expression in AML blast cell differentiation. Our study pinpoints the H3K27me3 HIST1 mark and HIST1H1D gene as two biomarkers potentially useful to stratify patient prognosis and defines targets that can be considered when developing epidrugs.

Patient samples
Blast cells were separated from blood or marrow samples through density-gradient (ficoll) separation, and stored in liquid nitrogen. Cryopreserved samples, with at least 70% of blasts, were collected from AML samples stored at Institut Paoli-Calmettes (IPC) Tumor Bank or at the Groupe Ouest Est d'Etude des Leucémies Aiguës et autres Maladies du Sang repository (GOELAMSTHE-QUE). GOELAMS samples were extracted from multicenter clinical trials LAM2006IR (NCT00860639) or LAM2007SA (NCT00590837). All patients received conventional induction chemotherapy consisting in daunorubicine (DNR) and aracytine (ARAC) with or without Mylotarg in the LAM2006IR trial for patients < 60 years [42] and idarubicine (IDA) and ARAC with or without lomustine in the LAM2007SA trial for patients ≥ 60 years [43]. Informed consent was provided by all patients according to the Declaration of Helsinki and subjected to ethical institutional review board approval.

ChIP-qPCR
Chromatin immunoprecipitation (ChIP) was performed as previously described [10]. Briefly, frozen samples were thawed, washed twice in phosphate-buffered saline (PBS), and chromatin was extracted with TRIS buffer pH 8, 0.25% Triton. Samples were sonicated to obtain DNA fragments of 300-600 base pair (Bioruptor PICO) and chromatin was immunoprecipated with an anti-H3K27me3 antibody (Abcam #6002). After immunoprecipitation, DNA was purified with the I-Pure kit (Diagenode). Quantification of ChIPed DNA was performed by real-time PCR using the SsoADV Univer SYBR Green Supermix (Bio-Rad) and detected with a CFX96 Real-Time PCR Detection System (Bio-Rad). IgG control "cycle over the threshold" Ct values were subtracted to input or IP Ct values and converted into bound value by 2 (−(IP Ct or input Ct-IgG IP Ct)) . Data are expressed as percent of bound/input. For each HIST1 region, H3K27me3 ChIP signal was double normalized with ChIP signal obtained at a genomic location invariably enriched with H3K27me3 (HOXD4) and a genomic region depleted of H3K27me3 (GAPDH). Heatmaps were performed with gene-e software (Broad institute). Hierarchical clustering was done using Euclidian distance.

RT-qPCR analysis
Total RNA was isolated from patient samples using the RNeasy mini Kit (Qiagen). RNA was treated with RNase-free DNase set (Qiagen) to remove contaminating genomic DNA. The cDNA was synthesized using the Transcriptor High Fidelity cDNA Synthesis Kit (Roche Applied Science), quantified by Power SYBR Green (Roche Applied Science) or TaqMan qPCR. For SYBR green analyses, signals were detected with a CFX96 Real-Time PCR Detection System (Bio-Rad). Primers used for gene amplification were designed using Primer3 software and are listed in the table below. Relative expression levels were determined by the delta Ct method and expression level of HPRT was used for normalization. For TaqMan method, PCR Master Mix (Thermo Fisher) was used and signal detected with 7500 Fast Real-Time PCR System (Applied Biosystem). Probes used for TaqMan analyses were HIST1H1D: Hs00271187_s1; HIST1H2BG: Hs00374317_s1; HIST1 H2BH: Hs00374322_s1; PGK1: Hs00943178_g1; PPIA: Hs04194521_s1 (Thermo Fisher). Relative expression levels were determined by the delta Ct method, taking the mean expression level of PKG1 and PPIA for normalization [44].

Statistical analyses
Statistical analyses were done using R software (version 2.15.2) (The Comprehensive R Archive Network. http://www.cran.r-project.org/) and Graph Pad Prism (Graph Pad Software, San Diego, CA, USA) and the significance of the differences between groups was determined via unpaired T test, Mann-Whitney test, or exact Fisher test. Data were presented as the median ± SEM. Overall survival (OS) and event-free survival (EFS) were calculated from the date of diagnosis to the date of death or to the date of relapse, death or the time to no response to intensive induction, respectively. Follow-up was measured from the date of diagnosis to the date of last news for living patients. Survivals were calculated using the Kaplan-Meier method and were compared with the log-rank test. Uni-and multivariate survival analyses were done using Cox regression analysis (Wald test). Variables with a p value < 0.05 were tested in multivariate analysis. All statistical tests were two-sided at the 5% level of significance, except when indicated otherwise. data, the Biopathology Department at IPC for support with MGG coloration, and the TrGET platform for cell culture advices. The authors are grateful to Jihane Pakradouni Demestre for helping with collecting patient data.

Availability of the data and materials
The datasets generated and analyzed during the current study are available on reasonable request. In addition, the Microarray data are accessible under the accession number E-MTAB-6997.