New genome-wide experimental approaches are widely reported as the ‘next generation’ of technologies that will revolutionise our understanding of cancers such as acute leukaemia. However, this goal is substantially complicated by the fact that acute leukaemias are caused by heterogeneous genetic and epigenetic events, which can occur at multiple levels during the dynamic progression from a premalignant state to frank leukaemia. For example, while genome-wide mapping of MLL-AF4 binding events identified stem cell regulators as prevalent target genes 
, little is known about the dynamic nature of global transcriptional reprogramming required for initiation and subsequent progression to the full leukaemic phenotype. Importantly, a focus on transcriptional reprogramming has the potential to identify shared aspects of transcriptional dysregulation and thus reduce the complexity and heterogeneity of diverse oncogenic events to a small number of specific pathways for exploration as novel disease classifiers and/or potential drug targets.
Studying the initiation or early progression of diseases such as AML is not feasible in human patients. However, mouse models of AML not only represent a valuable model for end-stage disease but also possess several qualities that make them amenable for studying disease initiation and progression. Firstly, the time point of the primary hit responsible for initiation of the disease can be strictly controlled and subsequent changes monitored. Secondly, monitoring changes within AML progression models has the potential advantage to identify new effectors/collaborators, which usually remain undetectable by approaches focussing only on frank leukaemia. Thirdly, different AML mouse model systems are available, which rely on transduction of distinct oncogenes with varying latency periods, and hence provide a suitable platform to identify shared pathways of general relevance for leukaemogenesis. Mouse models therefore represent powerful tools to study mechanisms of leukaemogenesis. Nevertheless, molecular targets identified in mouse model systems need to be further validated in human cells and xenotransplantation models to demonstrate potential clinical applicability 
In this study, we specifically focussed on two broadly accepted AML mouse models based on transplantation of stem/progenitor cells retrovirally transduced with the oncogenic fusion-proteins MLL-ENL and MOZ-TIF2, respectively 
. The cell of origin for leukaemic transformation is still a matter of ongoing debate. Our selection of lin-
as control sample is in line with current evidence where transduction of either self-renewing HSC or short-lived CMP or GMP generated leukaemic cells with a similar immunophenotype 
. All three of these starting populations are highly enriched in our lin-
control population. Moreover, expression analysis of HSPC TFs (such as HoxA9, Tal1, Scl, Lmo2
) and myelomonocytic markers (such as GM-Csfr, M-Csfr and G-Csfr
) paralleled those described by Cozzio et al 
(see Figure S10 in Supporting Information S1
). This observation underlines the similarity of our model system with previously published results and further validates the suitability of the lin-
We and others have previously reported that promoters and regulatory elements are enriched for H3K9ac marks 
. Distinct histone marks, sometimes combined with DNA binding proteins, have been used to identify potential regulatory regions 
. However, as long as regulatory elements are incompletely identified and characterized, there is no single epigenetic mark ‘of choice’ that can be considered the most appropriate to highlight regulatory regions 
. In both AML models, we observed upregulation of abdominal HoxA
-gene expression as expected. Moreover, this was paralleled by substantially increased levels of histone acetylation and therefore validated our experimental approach of using regions of altered histone acetylation to interrogate transcriptional control mechanisms that may underlie leukaemia development. However, downstream effectors and/or or parallel contributors other than a deregulated Hox-program are required for leukaemia development 
. The integrated genome-wide approach used here provided a potential strategy to identify some of these additional transcriptional events. Inference of transcriptional control mechanisms from microarray expression profiling data alone has been attempted in the past using bioinformatic tools that typically search promoter regions or collections of evolutionarily conserved sequence blocks for overrepresented sequence motifs. When we examined our list of differentially expressed genes using two of the most widely used tools for this type of in silico
analysis, oPOSSUM 
only identified the binding site for one of the 19 factors shown in (i.e. Cebpa) and Webmotifs 
failed to identify any significant motifs at all (see Figure S11 in Supporting Information S1
). By contrast, our direct measurements of differential levels of histone acetylation allowed us to select an experimentally informed set of candidate regulatory regions that showed differential distribution of the binding sites for 3 of the 4 differentially expressed factors including the binding site for Gata2, the potential relevance of which we went on to validate experimentally. Consistent with arguments put forward in a recent review 
, integrated genomic approaches, such as expression and ChIP-Seq profiling used here, provide a much less arbitrary way to select regions for sequence motif analysis than pure in silico
approaches and should therefore be widely applicable to study transcriptional dysregulation in disease.
In all current AML models, leukaemic cells display a somewhat more mature surface marker phenotype compared with the cells originally transduced, suggesting that expression changes alone can not be taken as evidence for functional significance but need to be followed up with specific experimentation. Integrated expression and ChIP-Seq profiling implicated downregulation of Gata2 as an important step during MLL-ENL and MOZ-TIF2 driven leukaemogenesis. Gata2 is a major regulator of haematopoietic stem cells during ontogeny 
and a key component of wider HSC regulatory networks 
. High expression levels of Gata2 are found in HSPC and down-regulation is required during normal haematopoietic differentiation 
. Interestingly, enforced expression of Gata2 blocks normal haematopoiesis 
and induces quiescence in HSCs 
. Gata2 therefore represents a major regulator of HSPC homeostasis. Our finding that downregulation of Gata2 is a common feature for two AML mouse model systems suggests that abrogation of a Gata2-dependent transcriptional program may contribute to leukaemogenic transformation which is supported further by low GATA2 expression found in large-scale human AML datasets. Acute lymphoid leukaemia patient samples show overlapping, yet on average even lower levels of GATA2 than AML (data not shown) suggesting that the incompatibility of high GATA2 expression with leukaemia development may extend beyond AML. Additionally, two GATA-2 mutations (Δ341–346 and L359V), affecting the second zinc-finger domain (see Figure S8B in Supporting Information S1
) and conferring either reduced or increased transcriptional activity, respectively, have recently been described in patients progressing from chronic to acute myeloid leukaemia.
. These findings are in line with our conclusion that deregulated Gata2 expression is involved in leukaemogenic progression, although, the exact mechanisms remain still to be elucidated. Of note, high-level ectopic expression of Gata2 in the murine myeloid cell line 416B did not inhibit proliferation, but instead allowed the derivation of clonal cell lines with a propensity for megakaryocytic differentiation 
, thus suggesting that Gata2-mediated inhibition of proliferation is context dependent. Importantly, our observation that MLL-ENL transduced cells remain susceptible to Gata2 mediated induction of cell-cycle arrest reveals a potential Achilles heel that could be characteristic for MLL-ENL leukaemic cells and that warrants further investigation.
Taken together, gene-expression and ChIP-Seq profiling coupled with bioinformatic analysis and functional validation allowed us to identify a potentially important, yet previously unknown regulatory process, that operates during transcriptional reprogramming in two mouse models of AML. Our results suggest that down-regulation of Gata2 may contribute to leukaemogenic transformation and that reactivation of Gata2 could be a novel treatment strategy in patients with acute leukaemias. Moreover, further analysis of the transcriptional repression program associated with the leukaemic phenotype may provide new mechanistic insights into the efficacy of de-repressive epigenetic therapies such as inhibition of histone deacetylases and DNA-methyltransferases 
. Finally, the integrated genome-wide approach employed in this study should be readily adaptable to study transcriptional reprogramming in other leukaemias as well as many solid tumours.