Detailed molecular studies have shown individual TFs to play critical roles at various stages of MK maturation. For example, a hypomorph mouse model of Gata1
displays reduced platelet numbers associated with deregulated MK proliferation and severely impaired cytoplasmic maturation (Shivdasani et al., 1997
). The other TFs investigated in this study, GATA2, RUNX1, FLI1, and SCL, have also been previously shown to play important roles in MK differentiation (Growney et al., 2005; Hart et al., 2000; Huang et al., 2009b; Ichikawa et al., 2004; Mikkola et al., 2003; Song et al., 1999; Spyropoulos et al., 2000
). Of note, these four factors also represent key players in regulatory network circuits operating in HSPCs (Gottgens et al., 2002b; Landry et al., 2009; Pimanda et al., 2007
) consistent with the previously suggested notion of significant overlap between transcriptional control mechanisms in MKs and HSPCs (Huang and Cantor, 2009
The five-factor ChIP-Seq data set generated made it feasible for us to perform a global analysis of combinatorial transcriptional control in a human primary myeloid cell type. Previous studies, which only examined individual promoter fragments, had revealed transcriptional cooperativity in MKs between RUNX1 and GATA1 (Elagib et al., 2003; Xu et al., 2006
) as well as between FLI1 and GATA1 (Huang et al., 2009a
). The latter was somewhat surprising given the well characterized antagonism between GATA1 and ETS family TFs during erythroid differentiation. Our genome-wide data sets support frequent co-occupancy of GATA1 and FLI1 in MK cells with a total of 1335 regions bound simultaneously by both factors. However, the vast majority (1094) of these are bound by additional factors and binding of GATA1 and FLI1 without any of the other factors was in fact the most underrepresented occupancy pattern. These observations suggest that activatory interactions between GATA1 and FLI1 may at least in part be mediated through the assembly of larger multiprotein complexes. Of particular relevance may be the observation that >700 regions are bound by GATA1 and FLI1 together with RUNX1. RUNX1 has been shown previously to interact at the protein level with both GATA1 and FLI1 (Huang et al., 2009a; Elagib et al., 2003
). Moreover, RUNX1 expression is maintained in MKs whereas it is downregulated during erythroid differentiation (Elagib et al., 2003
). Our combinatorial binding data are therefore consistent with a model where additional factors such as RUNX1 mediate transcriptional cooperation between GATA1 and FLI1 in MKs whereas antagonistic interactions may prevail in their absence as previously reported for the erythroid lineage.
Genes next to regions bound by all five factors were highly enriched for known regulators of MK differentiation and/or function. This observation prompted us to investigate the possibility that genes of unknown hematopoietic function with binding peaks for all five factors may also be enriched for the same functional categories. To provide biological insight, genome-scale hypothesis-generating screens such as ChIP-Seq experiments need to be coupled with meaningful assays for downstream functional validation. Here we took advantage of the zebrafish model because hematopoietic control mechanisms are highly conserved between zebrafish and human/mouse, yet only the zebrafish allows relatively high-throughput knockdown analysis in vivo in the context of a whole animal. Coupling the multifactor ChIP-Seq screen with zebrafish in vivo validation allowed us to identify eight regulators of thrombopoiesis and/or erythropoiesis. Of note, none of the MOs caused a lack of blood circulation and at the concentration used, with the exception of two (smox and max), MO knockdown did not affect wider aspects of morphology, underscoring the hematopoietic specificity of the uncovered phenotypes. Given that seven MOs affected both erythrocytes and thrombocytes, it is possible that at least some of those genes may operate at the level of immature progenitors or blood stem cells. Given the overlap of key transcriptional regulators in MKs and HSCs, potential functions in blood stem cells might perhaps not be too surprising but will require further fine dissection of knockdown phenotypes.
Recent ChIP-Seq studies of individual transcription factors in two different hematopoietic lineages suggested that binding patterns are largely lineage-specific (Heinz et al., 2010
) with shared peaks in one study of Scl being as low as 6% (Palii et al., 2011
). Comparison of the megakaryocyte data generated here with a recently published hematopoietic progenitor data set (Wilson et al., 2010a
) confirmed the predominance of lineage-specific binding patterns, and in addition demonstrated that regions bound by multiple factors are also largely lineage-specific. This suggests that additional lineage-specific transcription factors may play a part in controlling cell-type specific accessibility of regulatory regions through interactions with chromatin modifying enzymes. The likely relevance of additional MK-specific transcription factors is further emphasized by our observation that the five factors studied here may only account for 30% of MK-specific expression. Additional important players within MK transcriptional programs are likely to include NF-E2, MEIS1, and E2A (Shivdasani, 1996; Hisa et al., 2004; Semerad et al., 2009
). Our demonstration that PDZK1IP1 shares transcriptional regulatory elements with the blood stem cell regulator SCL has implications reaching beyond a better understanding of this particular gene locus. The notion that a given regulatory region can control multiple genes is well established, but past experimental analysis has largely focused on clusters of related genes such as the β-globin locus (Tolhuis et al., 2002
) or tandem arrays of interleukin genes (Loots et al., 2000
). Coregulation within these gene loci is likely to have arisen in parallel with local gene duplication events resulting in coordinated control of evolutionary and functionally related genes. By contrast, SCL and PDZK1IP1 encode unrelated proteins and it is likely that their original juxtaposition within vertebrate genomes was accidental. Our identification of PDZK1IP1 as a hematopoietic regulator together with the demonstration that SCL and PDZK1IP1 share transcriptional enhancer elements provides a rationale for the tight linkage between these two genes throughout vertebrate evolution. Moreover, our data illustrate the potential pitfalls of mapping transcription factor binding events to a single target gene when reconstructing regulatory networks. The +19 enhancer is approximately equidistant to the SCL and PDZK1IP1 promoters and evidently can control both. Transgenic in vivo analysis as performed here will be too time consuming and costly for global mapping of regulatory interactions between distal enhancers and the promoters of neighboring genes. However, further reductions in DNA sequencing costs may mean that chromatin-capture based methods providing a genome-wide view of promoter/enhancer interactions (Fullwood et al., 2009
) will eventually perform this integral part of regulatory network reconstruction.
Previous multi-TF ChIP-Seq studies have revealed layers of information that can only be obtained from integrated analysis of multiple factors. For example, computational analysis of a 12-factor ChIP-Seq data set from mouse ESCs suggested that genome-scale analysis of combinatorial TF occupancy has the potential to predict absolute and differential gene expression (Ouyang et al., 2009
). A five-factor ChIP-Seq study in an erythroid model cell line revealed that multi-TF complex binding often marks sites of long-range genomic interactions (Soler et al., 2010
). Here we have shown that multifactor ChIP-Seq surveys coupled to high-throughput in vivo functional screening provide a powerful strategy toward isolating key regulators of cellular phenotypes. Future detailed functional analyses of the eight hematopoietic regulators identified here, has the potential to integrate new regulatory pathways into an emerging framework of blood development.