Activation of Eklf expression during hematopoiesis by Gata2 and Smad5 prior to erythroid commitment

During hematopoiesis, a small pool of uncommitted stem and progenitor cells gives rise to large numbers of terminally differentiated cells of the various lymphoid and myeloid lineages through a process of proliferation and gradual differentiation. This process is characterized by the hierarchical progression of cells from a state of high plasticity to successively more-restricted states of lineage commitment and maturation. The elucidation of the signaling pathways and transcription factor networks that regulate these cell fate decisions and establish lineage-specific differentiation programs is crucial to the understanding of blood cell development in particular and of stem cell biology in general (Orkin and Zon,2008).

In lower vertebrates and mammals alike, signaling by Bmp4, a member of the Tgfβ-like group of growth factors that signal via the Smad protein family of intracellular effectors, is essential for the establishment of hematopoiesis shortly after gastrulation(Gupta et al., 2006; Maeno et al., 1996; Winnier et al., 1995). In the mouse embryo, Bmp4 induces a mesodermal cell population within the primitive streak at embryonic day (E) 7.0 to give rise to a common progenitor for hematopoietic and endothelial lineages termed the hemangioblast. This progenitor subsequently migrates into the extraembryonic yolk sac where it differentiates into the hematopoietic, endothelial and vascular smooth muscle cell lineages that will form the cardiovascular system of the embryo(Huber et al., 2004).

Of particular importance to the understanding of the mechanisms that underlie these early events of hematopoietic commitment has been the application of embryonic stem (ES) cell in vitro differentiation technology(Keller, 2005). With regard to yolk sac hematopoiesis, ES cell differentiation studies have shown that Bmp4 signaling is required throughout embryoid body (EB) development, from the initial establishment of hematopoietic fate from mesoderm(Park et al., 2004) to its subsequent progression to erythroid commitment(Adelman et al., 2002). In addition to Bmp4 signaling, members of the GATA-motif-binding family of zinc-finger transcription factors play key regulatory roles in hematopoietic development. Gata2 expression occurs early in hematopoiesis and is crucial for the proliferation and expansion of hematopoietic stem cells and progenitors(Tsai et al., 1994), whereas Gata1 is activated subsequently and is essential for the establishment of erythroid commitment and differentiation(Pevny et al., 1991; Weiss et al., 1994). Importantly, Gata2 is a direct Bmp4 target gene(Lugus et al., 2007), and Smad-dependent Bmp4 signaling is necessary and sufficient to induce the expression of Gata1 alongside that of erythroid Krüppel-like factor(Eklf; Klf1) (Adelman et al.,2002), another key regulator of erythropoiesis.

Eklf was originally identified as an erythroid-specific transcription factor that functions as an essential regulator of β-like globin switching in red cells (Miller and Bieker,1993; Nuez et al.,1995; Perkins et al.,1995). However, in contrast to this well-defined role, Eklf must fulfill additional functions during hematopoiesis as it is expressed much earlier, prior to erythroid differentiation. In the developing mouse embryo, Eklf message is first detected in situ in the extraembryonic mesoderm of the yolk sac as early as at the neural plate stage, by E7.5(Southwood et al., 1996),which coincides with the presence of primitive erythroid progenitors(Palis et al., 1999). Similarly, during the differentiation of mouse ES cells in vitro, Eklf expression is activated prior to terminal erythroid differentiation(Adelman et al., 2002). These observations raise the possibility that Eklf plays a role prior to erythroid commitment and lead us to ask where exactly Eklf fits into the transcription factor hierarchy that establishes the hematopoietic program.

In order to address these issues, we have developed novel reporter gene and RNAi-based loss-of-function assays for use during ES cell in vitro differentiation, as well as a transgenic mouse model to characterize the onset of Eklf expression during hematopoiesis. We show that Eklf is activated in a progenitor population of erythroid-megakaryocytic potential prior to erythroid commitment and delineate the transcriptional mechanisms that govern its differential expression in the two compartments. Based on our findings, a model integrating Bmp4- and Gata factor-mediated transcriptional control of hematopoiesis is presented.

The Peklf-GFP reporter construct was generated by cloning the SacI to NcoI (-950 to the ATG) murine Eklf fragment(Chen et al., 1998) into the EcoRI site of the plox-plasmid(Kyba et al., 2002). GFP(pEGFP-C₁, Clontech) was inserted downstream of Peklf, replacing the Eklf with the GFP translation start codon (ATG), placing GFP upstream of the plox-plasmid NotI site, followed by poly(A). The Peklf-intron-GFP construct was generated by inserting murine Eklfexon 1 and intron 1 into Peklf-GFP upstream of GFP, mutating Eklfexon 1 ATG to GTA to make the GFP start codon the first ATG present in the resulting spliced mRNA. All mutations and deletions were introduced using QuikChange (Stratagene); primer sequences are available upon request.

The plox-GFP-Intron-miR plasmid was generated by inserting the miR-30 backbone (SalI to PmeI) of pSM2 (Open Biosystems) into theβ-globin intron of pSG5 (Stratagene) at MfeI. The resulting Intron-miR sequence was placed between the GFP and poly(A) of pEGFP-C₁, subsequently inserting GFP-Intron-miR-poly(A) into the plox-plasmid. shRNAs sequences were generated by PCR as described previously(Paddison et al., 2004) and cloned into the miR-30 backbone (XhoI to EcoRI) of plox-GFP-Intron-miR. shRNA sequences against NM_008541 (shRNA-1 is HP_248638;shRNA-2 is HP_257681) were obtained from RNAiCodex(http://codex.cshl.edu/scripts/newmain.pl). The non-silencing shRNA-control was purchased from Open Biosystems(RHS1703).

Ainv18 mouse ES cells (Kyba et al.,2002) were maintained and differentiated according to established protocols (Choi et al., 2005; Kennedy and Keller, 2003) and have been described previously (Manwani et al., 2007; Zafonte et al.,2007). Each ES cell clone was generated by targeting the respective plox construct to the Ainv18 homing site by co-electroporation with a Cre recombinase expression plasmid and subsequent antibiotic selection. Site-specific integration of the plox construct was checked by PCR on genomic DNA. Retention of hematopoietic differentiation capacity as measured by Flk1 expression levels on EB day 4 was verified for each clone. All reporter assay experiments were performed in the absence of doxycycline (dox) (Sigma); shRNA expression was induced by adding of 1 μg/ml dox.

Single-cell suspensions of murine fetal livers or EBs were prepared and stained using the following antibodies: CD71-PE, Ter119-APC (eBiosciences),Flk1-PE, Ter119-PE, CD41-APC, Gr1-APC, Mac1-APC, CD42d-APC (BD Biosciences). Flow cytometry data were analyzed using FlowJo (TreeStar). Sorted cells were cultured in MethoCult M3234 (StemCell Technologies) supplemented with cytokines according to manufacturer's instructions. Cytospins were fixed with May-Grünwald Solution (Fluka) and stained with Giemsa Solution (Sigma),and pictures were taken using an Axioplan2 microscope (Zeiss) with a 100× or 63× oil-immersion objective.

Genomic sequences of mammalian Eklf or Gata1 loci were obtained from the ENSEMBL Genome Browser between February and August, 2005. Each alignment and the subsequent identification of conserved blocks of sequence homology and putative transcription factor binding sites were performed according to Chapman et al.(Chapman et al., 2004). Smad binding motif consensus sequences: 5′-AGAC (and complement 5′-GTCT) (Shi et al.,1998; Zawel et al.,1998); 5′-GTCTT (Lee et al., 2004); 5′-CAGC(Korchynskyi and ten Dijke,2002); 5′-GCTG(Benchabane and Wrana, 2003);5′-TGGAGC (Hata et al.,2000); 5′-TGGACC (Oren et al., 2005). Sp1/E2F/Smad binding motif: 5′-CCGCCC(Chen et al., 2002; Frederick et al., 2004; Yagi et al., 2002).

ChIP experiments were performed as previously described for G1E-ER-Gata1 cells (Im et al., 2005) or EBs(Lugus et al., 2007). G1E-ER-Gata1 cells were treated with 0.1 μM 17β-estradiol (Sigma) for 14 hours to induce ER-Gata1 nuclear translocation prior to cross-linking. Anti-Gata2 and anti-Gata1 rabbit polyclonal antibodies were a gift from Emery H. Bresnick (University of Wisconsin Medical School, Madison, WI). Normal rabbit IgG (Upstate) was used as a negative control. Primers were designed using Primer Express software (Applied Biosystems). Primer specificity was verified by showing that each primer pair generated a single amplicon and dissociation curve; primer sequences are available upon request.

Purified total RNA (1 μg) was reverse-transcribed using the ImProm-II RT Kit (Promega). Eklf, Gata1, Gata2, GFP, globin βH1, Smad1,Smad5, Gapdh mRNA or 18S rRNA expression was detected using pre-made TaqMan Gene Expression Assays (Applied Biosystems) according to the manufacturer's protocol. Gene expression levels were calculated using a relative standard curve made from serially diluted cDNA samples, followed by normalization to 18S rRNA or Gapdh mRNA. For each clone and time point, three independent RNA samples derived from three independent EB differentiations performed in parallel were used to calculate the arithmetic mean (average) and s.d. P-values were computed using Student's t-test.

Primary antibodies: Eklf (Southwood et al., 1996), Gata1 (N-6, Santa Cruz Biotechnology), Gata2 (H-116,Santa Cruz Biotechnology), Hsp90 (H-114, Santa Cruz Biotechnology), GFP (JL-8,Clontech). Secondary antibodies: goat anti-rabbit IgG(H+L)-HRP (Southern Biotechnologies); goat anti-mouse IgG(H+L)-HRP and goat anti-rat IgG(H+L)-HRP(Pierce).

To characterize the expression pattern of Eklf during hematopoiesis, we developed a reporter expression cassette for use in a transgenic mouse model. To this end, a cDNA encoding green fluorescent protein (GFP) was placed under the control of the previously described 950 bp mouse Eklf promoter(Peklf) (Chen et al., 1998),which had been shown to direct erythroid-specific transgene expression in developing mouse embryos (Xue et al.,2004). However, to avoid the pitfalls usually associated with the random integration of a transgene, such as position-effect variegation, we took advantage of a site-specific integration mechanism featured in the previously engineered Ainv18 ES cell line(Kyba et al., 2002). Using Cre-mediated recombination, we targeted a single copy of the Peklf-GFP construct in a uni-directional fashion to a loxP homing site in a genomic locus of open chromatin conformation and selected transgenic ES cell clones based on antibiotic resistance (Fig. 1A). After establishing that the tet operon (tetOP)contained in Ainv18 ES cells does not interfere with the transcriptional control conferred by Peklf (see Fig. S1 in the supplementary material), we created a Peklf-GFP `reporter mouse' by injecting ES cells of one Peklf-GFP Ainv18 clone into mouse embryos at the blastocyst stage and breeding chimeric animals to homozygosity.

Next we investigated the pattern of Eklf expression in vivo by analyzing Peklf-GFP transgene activity during fetal liver hematopoiesis, as it is the site of blood cell production between E11 and E16 in the developing mouse embryo. Flow cytometric analysis of fetal livers from Peklf-GFP embryos revealed that ∼36% of all cells were GFP⁺ at E13.5(Fig. 1B). To correlate GFP expression with the status of erythroid lineage commitment in fetal liver cells, we determined the transferrin receptor (CD71; Tfrc - Mouse Genome Informatics) versus Ter119 (Ly76) cell-surface marker expression profile of fetal liver cells as an indicator of hematopoietic differentiation, as previously described (Zhang et al.,2003). At E13.5, we found ∼61% of all fetal liver cells in the committed erythroid compartment, as indicated by their CD71^hi/Ter119^- to CD71^lo/Ter119⁺expression profile (Fig. 1C, Table 1). Most GFP⁺cells were part of this population as they also expressed the Ter119 marker(∼34%, Fig. 1D) and were thus identified as erythroid cells. By contrast, ∼12% of fetal liver cells were characterized as progenitors by their low to medium CD71 levels in combination with a lack of Ter119 expression(CD71^lo/med/Ter119^-, Fig. 1C, Table 1). Interestingly, a small fraction of GFP⁺ cells (∼2-3% of total, Fig. 1D; 17% of progenitors, Table 1) were found in this progenitor compartment (CD71^lo/med/Ter119^-) of the murine fetal liver.

Therefore, we next tested directly the progenitor activity of the GFP⁺/Ter119^- cells in hematopoietic colony assays in culture to investigate whether the Peklf-GFP transgene is indeed active prior to erythroid commitment. Peklf-GFP fetal liver cells were obtained from mouse embryos at E13.5 and separated from differentiating erythroid cells(Ter119⁺). The remaining pool of fetal liver cells(Ter119^-) was sorted into Peklf-GFP-expressing(GFP⁺/Ter119^-, 2.56%) and non-expressing(GFP^-/Ter119^-, 27.3%) populations(Fig. 1D), followed by seeding in semisolid media in the presence of appropriate growth factors.

The sorted GFP⁺/Ter119^- population gave rise to a substantial enrichment of erythroid progenitor-derived colonies(colony-forming unit erythroid, CFU-E; burst-forming unit erythroid, BFU-E) in culture (Fig. 1E,I). Most interestingly, it was also enriched for megakaryocytic progenitor-derived colonies, which were present in significant numbers(Fig. 1F,I), but not for granulocytic (Fig. 1G,I) or macrophage colonies (that are readily detectable in assays seeded with GFP^-/Ter119^- cells, Fig. 1H). Our data clearly indicate that the Eklf promoter (Peklf) activates transcription in progenitor cells well before the onset of erythroid differentiation. More specifically, Peklf-GFP expression preferentially marks progenitors that will give rise to erythroid or megakaryocytic colonies.

According to current models of hematopoiesis, the erythroid and megakaryocytic lineages are derived from a common, bipotential progenitor(megakaryocytic-erythroid progenitor, MEP), which in turn stems from a progenitor common to all myeloid cells(Adolfsson et al., 2005; Forsberg et al., 2006). Therefore, our findings strongly suggest that Eklf is expressed at least as early as in a progenitor population of erythroid-megakaryocytic potential.

To better define the onset and pattern of Eklf expression during hematopoiesis in the context of stage- and lineage-specific cell-surface markers and genes, we next employed the differentiation of the Peklf-GFP Ainv18 ES cell clone to EBs as an in vitro assay of hematopoietic development. Throughout an 8-day ES cell differentiation time-course, GFPtranscription under control of Peklf closely mirrored that of the endogenous Eklf gene with regard to onset and levels of expression(Fig. 2A). Expression of Eklf and GFP was initiated at or near day 4 during the stage of EB development that is characterized by the presence of mesoderm-derived hematopoietic progenitor cells as indicated by a peak in Gata2 expression(Fig. 2A) and high levels of the receptor tyrosine kinase Flk1 (Kdr - Mouse Genome Informatics)(Chung et al., 2002)(Table 2). Interestingly, the activation of Eklf (and GFP) expression occurred simultaneously to that of Gata1 at day 4, and all three were expressed in virtually the same pattern thereafter (Fig. 2A). Following the initial activation of expression, Eklf and GFP levels began to rise rapidly around day 5 of EB development. By day 6, Eklf and GFP expression peaked, whereas the levels of the progenitor markers Gata2 and Flk1 decreased, which coincided with the appearance of committed erythroid cells in EBs as indicated by the expression of the red cell-restricted cell-surface marker Ter119 (Table 2) and embryonic β-like globin (globin βH1; Hbb-bh1 -Mouse Genome Informatics) (Fig. 2A). Subsequently, mature myeloid cells emerged by day 8 of EB development (see Fig. S2 in the supplementary material). At that point, GFP expression among these lineage-committed cells was restricted to erythroid cells and did not occur in cells of the granulocyte-macrophage or megakaryocytic lineages (see Fig. S2 in the supplementary material).

As these findings imply that Eklf is first expressed in a progenitor population during hematopoietic EB differentiation, analogous to its expression pattern during fetal liver hematopoiesis, we next isolated GFP⁺ cells from EBs at day 5.5 and subjected them to hematopoietic colony assays in culture (Fig. 2B). As expected, GFP⁺ cells gave rise to colonies of mixed erythroid-myeloid lineages as well as megakaryocytic-myeloid lineages in addition to erythroid colonies (Fig. 2C), demonstrating that Peklf-regulated transcription of GFP is activated prior to erythroid commitment in EBs.

Therefore, the expression of Eklf (GFP) during the hematopoietic differentiation of EBs can be divided into two phases: the onset of expression in a progenitor population between days 4 and 6 of EB development, followed by erythroid-restricted expression after day 6.

To elucidate the mechanism regulating Eklf expression in a progenitor population, we sought to identify genomic regulatory regions that control Eklf transcription at this early stage in hematopoiesis, in contrast to previous studies, which had been focused on the regulation of Eklf expression in differentiating erythroid cells.

In order to predict putative transcription factor binding sites based on cross-species sequence conservation(Chapman et al., 2004), we generated a comparative, multi-sequence alignment of the Eklf locus from different mammalian species. This approach has been demonstrated to be superior in identifying functional cis-regulatory elements to pairwise alignments, which had previously been performed for the Eklf locus(Chen et al., 1998).

Fig. 3A shows a five-species alignment across the entire genomic Eklf locus between the two neighboring genes. Importantly, aside from exons, only two domains of non-coding sequence display a significant level of conservation between all five species across the 17 kb of the alignment: a stretch of ∼1000 bp with three peaks of conservation directly upstream of the Eklftranscription start site, and a shorter peak region within the first intron. The former corresponds to the 950 bp region used in the Peklf-GFP construct described above, which harbors an upstream enhancer, two erythroid hypersensitive sites (EHS1 and EHS2) (Chen et al., 1998) and a proximal promoter(Crossley et al., 1994). The latter consists of a previously unreported sequence element just upstream of the splice branch site preceding the second exon, which we designated as a putative intronic enhancer.

Fig. 3B provides a more detailed view of the aligned mouse sequence from the first repeat upstream of the 950 bp mark to the beginning of exon 2. Sequence blocks of four or more nucleotides perfectly conserved between all five species cluster predominantly within the three alignment score peak regions. Most surprisingly, we found the majority of perfectly conserved sequence blocks within the Eklfalignment to represent consensus motifs for Smad binding sites. In total, we identified ten potential Smad binding sites across the entire length of the alignment (see Materials and methods for a list of binding motif consensus sequences and references). Notably, these conserved Smad sites are not found in a random distribution throughout the locus, but are confined to the upstream enhancer, the proximal promoter and the intronic enhancer of the Eklf gene. In addition, we found blocks of perfect conservation within each of the three cis-elements that do not match any known consensus motifs but which are mostly GC-rich in nature, a feature that has been described to serve Smad binding as well(Ishida et al., 2000).

Fig. 3C shows the three Eklf regulatory regions at the nucleotide level in an alignment expanded to seven mammalian species. Surprisingly, all conserved Smad sites,numbered Smad binding motif 1 (SBM1) through SBM10, are in the vicinity of previously reported transcription factor binding sites. These include: two conserved Gata factor binding sites surrounding a bHLH factor binding site[GATA/E-box/GATA (GEG) motif] as well as a putative Sp1 site within the upstream enhancer (Anderson et al.,1998); and a Gata factor site next to a Cp1 (Nfya - Mouse Genome Informatics) binding site within the proximal promoter(Crossley et al., 1994). In addition, we found two previously unreported WGATAR motifs in the putative intronic enhancer in an arrangement similar to that of the upstream enhancer,albeit with a lesser degree of homology as only one of the intronic GATA motifs is conserved between all seven species.

Most importantly, however, all three regulatory regions of the Eklf gene display a similar layout, in which Smad sites are clustered around one or two Gata factor binding motifs. The upstream enhancer contains four Smad sites and a putative Sp1/E2F/Smad site surrounding the GEG motif,whereas the proximal promoter contains one Smad site next to the Gata/Cp1 sites. Similarly, the five Smad sites in the intronic enhancer are arranged around the one perfectly conserved Gata site.

Taken together, the high number of conserved Smad binding sites located exclusively within the three cis-regulatory regions of the Eklf locus strongly suggest that Smad proteins play a functional role in the transcriptional control of Eklf expression in response to Bmp4 signaling. Interestingly, we found such a clustering of conserved Smad binding motifs around a GATA motif within a known enhancer of the Gata1 gene as well (for a phylogenetic alignment and detailed description of Gata1 cis-regulatory regions, see Fig. S3 in the supplementary material).

To examine the contribution of the identified cis-regulatory elements and individual transcription factor binding sites to the control of Eklftranscription during hematopoiesis, we expanded the ES cell in vitro differentiation reporter assay by generating a set of new GFP transgene reporter constructs that incorporated insertions, deletions or point mutations into Eklf cis-regulatory elements based on the alignment in Fig. 3. In total, we created four new Ainv18 ES cell clones as shown in Fig. 4A.

The introduction of a point mutation in each of the two Gata binding sites of the upstream Eklf enhancer (Peklf-2xG/A-GFP) completely abolished GFP transgene transcription between days 4 and 7 of EB development(Fig. 4B). By contrast, a small but significant rise in GFP transgene expression levels occurred between days 4 and 5.5 when a point mutation was introduced into the Gata site at the proximal Eklf promoter (Peklf-G/A-GFP; Fig. 4C). Only subsequently,upon erythroid commitment after day 5.5, did this Gata site mutation result in a failure to increase GFP transgene levels, whereas mRNA levels of the endogenous Eklf gene continued to grow.

Next we tested the potential contribution of the newly identified Smad binding motifs found within the three Eklf cis-regulatory regions to the control of GFP transgene expression. As the onset of Eklf expression at day 4 of EB differentiation requires functional Gata binding sites within the upstream Eklf enhancer, which also contains Smad binding motifs, we hypothesized that Smad-mediated control of Eklf transcription was most likely to occur at the upstream enhancer. To test this, we deleted the two Smad binding motifs in the upstream enhancer that directly surround the GEG motif (Peklf-2xΔSBM-GFP), without altering the GEG motif itself so as to avoid interfering with Gata factor binding. The deletion of the two Smad binding sites abolished the onset of GFP transgene transcription between days 4 and 5.5 of EB development (Fig. 4D). Thereafter however, transgene expression was activated as GFP levels began to rise between days 5.5 and 7. Importantly, the GFP expression pattern resulting from the deletion of Smad sites in the upstream enhancer differed from the pattern produced by the mutation of GATA motifs in the upstream enhancer, arguing that the deletion of Smad sites in the upstream enhancer does not simply impact Gata binding in this region.

As Smads and Gata factors act on the upstream enhancer, we reasoned that a similar mode of transcriptional control might occur at the intronic Eklf enhancer, given that both regulatory regions display a similar layout of Gata sites surrounded by Smad binding motifs. Inclusion of the intronic Eklf enhancer (Peklf-intron-GFP) increased the maximal level of GFP expression about threefold as compared with levels recorded with Peklf-GFP alone, without altering the overall pattern of transgene expression(Fig. 4E). Furthermore, the intronic Eklf enhancer extended the rise of transgene expression until day 8 of EB development, in contrast to expression of the endogenous Eklf, which plateaus at day 7. This argues that although the newly identified, highly conserved region within the first intron of Eklfis not required for transgene expression, it indeed acts as an enhancer of Eklf transcription.

In summary, our EB reporter assay results in the context of a transgene integrated at a homing site suggest that Gata factor and Smad binding at the Eklf upstream enhancer are required for the onset of Eklftranscription in a progenitor population. By contrast, upon erythroid commitment, Gata factor-mediated control of transcription at the upstream enhancer and the proximal promoter is sufficient for the maintenance of Eklf expression. In addition, the highly conserved region within the first Eklf intron that was identified through the phylogenetic alignment acts as an enhancer of Eklf transcription throughout hematopoiesis.

Having established that Gata factor binding to the upstream Eklfenhancer is required for the onset of transgene expression in EBs at day 4, we next asked which Gata factor regulates Eklf expression at the progenitor stage. Gata2 is an obvious candidate in this regard, as its functional role in hematopoietic progenitors is well documented and we find Eklftranscription initiated in EBs at a time when high levels of Gata2, but not of Gata1, are present. To test this, we performed quantitative chromatin immunoprecipitation (ChIP) assays in G1E-ER-Gata1 cells and in differentiating EBs.

The G1E-ER-Gata1 cell line is used as a tool to study erythroid differentiation in vitro as it has been engineered to progress from an undifferentiated to a more differentiated state in a estradiol-dependent manner mediated by Gata1 target gene regulation(Grass et al., 2003; Weiss et al., 1997; Welch et al., 2004), which coincides with a rise in Eklf expression levels (see Fig. S4 in the supplementary material) (see Im et al.,2005).

In undifferentiated G1E-ER-Gata1 cells, Gata2 occupied the Eklfupstream enhancer (Fig. 5A). Gata2 occupancy levels at the upstream enhancer were twofold enriched compared with those at the proximal promoter and the intronic enhancer, which did not display any significant Gata2 occupancy as compared with a non-conserved GATA motif in the 3′UTR of the murine Eklf locus that served as a negative control. Upon differentiation of G1E-ER-Gata1 cells, Gata2 occupancy levels at the Eklf upstream enhancer dropped, which, in turn,coincided with an increase of ER-Gata1 fusion protein binding to the Eklf upstream enhancer and to the proximal promoter(Fig. 5A). This switch in Gata factor occupancy argues that ER-Gata1 replaces Gata2 at the upstream enhancer while also binding to the proximal promoter during the induced erythroid differentiation of G1E-ER-Gata1 cells, which correlates with an increase in Eklf expression.

To assess Gata factor-mediated control of Eklf expression under conditions that more closely resemble the endogenous state than does a cell line, we next examined Gata occupancy patterns at the Eklf locus during hematopoiesis in differentiating EBs. At day 4 of EB development, an enrichment of Gata2 occupancy was detected at the upstream enhancer and the proximal promoter of the Eklf locus, and to a lesser degree at the intronic enhancer (Fig. 5B). Therefore, the onset of Eklf expression at the progenitor stage coincides with Gata2 occupancy at all three cis-regulatory sites of the Eklf gene.[Similar to Lugus et al., we detected an overall background signal with the anti-Gata2 ChIP in EBs that was higher than in G1E-ER-Gata1 cells, most likely owing to the cellular heterogeneity of EBs(Lugus et al., 2007).] At day 6 of EB differentiation, Gata2 occupancy levels decreased. By contrast, Gata1 binding occurred at all three conserved Eklf cis-regulatory regions,especially at the upstream enhancer (Fig. 5B). Thus, a Gata switch at the Eklf locus dependent on the progression of hematopoietic development is observed in EBs. This in vivo occupancy profile of Gata factors suggests a mechanism of transcriptional control by which Gata2 binding mediates the initiation of Eklf expression during hematopoiesis prior to erythroid commitment, whereas Gata1 replaces Gata2 at the Eklf locus during erythroid differentiation, which results in an increase of Eklf expression.

Importantly, a Gata factor switch upon erythroid commitment, similar to the one observed at the Eklf locus between days 4 and 6 of EB differentiation, was also detected at a cis-regulatory region of the Gata1 gene (Fig. 5B),analogous to results previously described in G1E-ER-Gata1 cells(Pal et al., 2004) and reproduced here (Fig. 5A). This regulatory region, termed hypersensitive site 1 (HS1) within the hematopoietic enhancer (HE) of the Gata1 (G1) gene (G1HE/HS1, see Fig. S3 in the supplementary material), is known to be required for Gata1 expression in a progenitor population (Vyas et al.,1999). Thus, the initial onset of Eklf and Gata1transcription does not only occur at the same time during the progenitor stage in EBs, but is also regulated in a similar fashion, dependent on Gata2 binding to specific enhancer elements.

As Bmp4 signaling via the Smad pathway is necessary and sufficient to induce Eklf and Gata1 expression at day 4 of EB development(Adelman et al., 2002), we hypothesized that the onset of Eklf expression in a progenitor population prior to erythroid commitment is regulated directly by Smad proteins, in light of the fact that Smad binding sites in the Eklf upstream enhancer are required for GFP transgene expression between days 4 and 5.5 of EB development. The most likely Bmp4 effector in this regard is Smad5, as it has been implicated in promoting the establishment of erythroid fate during hematopoiesis (Fuchs et al.,2002; Liu et al.,2003; McReynolds et al.,2007). By contrast, Smad1 acts earlier in hematopoiesis, at the hemangioblast stage (Zafonte et al.,2007).

As no anti-Smad5 antibodies functional in ChIP assays have been reported to date, we developed a novel loss-of-function assay for the Ainv18 ES cell line that allowed us to specifically knockdown Smad5 levels in an inducible,RNAi-mediated manner during EB development so as to test the impact of Smad5 activity on Eklf expression. Utilizing the tet operon of the Ainv18 ES cell line (Kyba et al.,2002), we coupled the transcription of a specific, short hairpin RNA (shRNA) within a microRNA (miR) backbone to that of a GFP cDNA in a doxycycline (dox)-dependent fashion (Fig. 6A). We generated three different expression cassettes, each of which was stably inserted into the plox homing site of the Ainv18 ES cell line via Cre-mediated recombination. Two of these encoded shRNAs targeting a unique sequence within the Smad5 mRNA (shRNA-1 and shRNA-2), whereas a third encoded a no-target control shRNA. Fig. 6B shows the tight transcriptional control imparted by the tet operon in response to dox treatment as exemplified by GFP expression levels in the case of the shRNA-1 Ainv18 ES cell clone.

As we hypothesized that Smad5-mediated transcriptional control occurs at the onset of Eklf expression prior to erythroid commitment, we induced the knockdown of Smad5 mRNA at day 4 of EB development by treating the shRNA-1 Ainv18 ES cell clone with dox for 24 hours. To quantify the maximal RNAi effect, we sorted EBs at day 5 into dox-treated,GFP-expressing cells (+dox/GFP⁺) or untreated, GFP-negative cells(-dox/GFP^-), and isolated RNA for qRT-PCR analysis. In the shRNA-1-expressing population (+dox/GFP⁺), the Smad5 mRNA level was knocked down to ∼40% of the level seen in the non-expressing population (-dox/GFP^-), demonstrating that the transgenic RNAi assay is functional (Fig. 6C). More importantly, the dox-induced knockdown of Smad5 coincided with a reduction of Eklf and Gata1 mRNA to ∼25% and ∼30%,respectively, of the levels observed in untreated cells(-dox/GFP^-), suggesting that Smad5 regulates both genes.

Given the strong knockdown effect observed in the shRNA-1 Ainv18 clone upon dox treatment, we reasoned that a cell-sorting step based on GFP expression should not be required to detect the substantial drop in Smad5, Eklf and Gata1 levels. Instead, we used unsorted, whole EBs for a more extended loss-of-function analysis. Analogous to the effect described above, a 24 hour dox treatment of developing EBs from day 4 to 5 resulted in a 50% reduction of Smad5 mRNA levels as compared with EBs not treated with dox(Fig. 6D). Moreover, the knockdown of Smad5 correlated once again with a significant drop in Eklf and Gata1 levels, to ∼65% and ∼70% of those observed in untreated EBs, respectively. Importantly, shRNA-1 targets Smad5 mRNA specifically, as the levels of Smad1 mRNA were unperturbed, despite the high degree of sequence homology between the two mRNAs. In addition, Gata2 mRNA levels were not affected by the dox treatment, demonstrating that the negative effect observed on Eklfand Gata1 transcription between EB day 4 and 5 is in fact a direct and specific consequence of the reduction of Smad5 mRNA levels.

To further corroborate this point, we next included the shRNA-2 and the shRNA-control clones in our experiments. A western blot analysis of whole-cell lysates made from EBs on day 6 of differentiation after 48 hours of dox treatment revealed that Eklf and Gata1 protein levels are strongly reduced in cells carrying either shRNA-1 or shRNA-2, but not in cells carrying the shRNA-control (Fig. 6E). By contrast, Gata2 protein levels did not display any significant variability between dox-treated and untreated cells in any of the three different clones. Importantly, the observed RNAi effect is due to the specific knockdown of Smad5, as two independent shRNAs complementary to different regions of the target mRNA produced the same result, in contrast to the no-target control shRNA that did not alter the expression of any of the genes tested. We conclude that Smad5 protein is required at day 4 of EB development for the correct onset of Eklf and Gata1 transcription.

In this report, we demonstrate that the initial activation of Eklfduring mammalian hematopoietic development occurs prior to erythroid commitment and is likely to be within a progenitor population of erythroid-megakaryocytic potential. Utilizing a bioinformatics-based phylogenetic footprinting approach, we identify three evolutionarily conserved cis-regulatory elements that are critical to the control of Eklf expression throughout hematopoiesis and which share a similar architecture of Smad binding consensus motifs clustered around Gata factor binding sites. Applying novel reporter assay and loss-of-function technology to the ES in vitro differentiation system in combination with in vivo binding studies, we show that the onset of Eklf expression at the progenitor stage is dependent on Gata2 and Smad5, whereas the maintenance of Eklf expression in committed erythroid cells is regulated by Gata1. As the activation of Eklfrequires Bmp4 signaling (Adelman et al.,2002), we propose a two-tiered, stage- and lineage-dependent model of Eklf regulation during hematopoiesis, as described below.

Early in hematopoiesis, Gata2 and Smad5 activate Eklf in a cooperative fashion. This integration of the Bmp4 signaling pathway with the Gata2 target gene network is achieved at the upstream enhancer of Eklf, where functional Gata and Smad binding sites are found in close proximity to one another. It results in the low-level expression of Eklf. Upon erythroid lineage commitment, Gata1 replaces Gata2 and binds to the Eklf upstream enhancer and proximal promoter, regulating Eklf expression at high levels throughout erythroid differentiation in a Smad-independent manner. At this stage, Gata1 nucleates an Scl (Tal1 -Mouse Genome Informatics)-containing protein complex occupying the GEG motif at the upstream enhancer (Rodriguez et al., 2005), similar to the GATA/E-box-bound complexes regulating the erythroid-specific expression of other Gata1 target genes(Anguita et al., 2004). In parallel, Gata1 binding occurs at the Eklf proximal promoter next to a CCAAT-box, possibly in a complex with Cp1 or a C/EBP family member(Crossley et al., 1994; Gordon et al., 2005), while occupancy changes from Gata2 to Gata1 at the intronic enhancer as well. Such`Gata-switches' at the same GATA motif have also been shown to occur during the regulation of Gata1 (Pal et al.,2004) (see also below), α-globin(Anguita et al., 2004), Scl(Lugus et al., 2007), Gata2(Grass et al., 2003; Grass et al., 2006; Kobayashi-Osaki et al., 2005)and Kit expression (Jing et al.,2008), resembling the two-tiered regulatory mechanism described here for Eklf. Thus, the layout of transcription factor binding sites contained in the Eklf cis-regulatory regions enables a response of dual specificity that is dependent on the stage of hematopoiesis, which in turn results in a transcriptional output of varying degree in a lineage-specific manner.

Parallels to the Gata2-Smad5 cooperation in our model of Eklfactivation can be found in the regulation of cardiac-specific genes during embryonic heart development (reviewed by Peterkin et al., 2005). Similar to the architecture observed in the cis-regulatory regions of Eklf, neighboring Smad and Gata factor binding sites are found in enhancers of Nkx2.5, which is the earliest known marker of cardiogenesis, and the presence of both types of consensus motif is required for conveying transcriptional control in response to BMP signaling(Brown et al., 2004; Lee et al., 2004).

Given the requirement of Bmp4 signaling for the establishment and progression of (yolk sac) hematopoiesis as well as the prominence of Gata2-controlled gene expression in hematopoietic progenitors, we propose that the functional cooperation of (BMP-activated) Smads and hematopoietic Gata factors underlies target gene regulation in hematopoietic development and lineage fate decisions, as exemplified here for the case of Eklf.

In support of the above notion, we show that the onset of Gata1 expression is dependent on Smad5 as well. We identify a cluster of conserved Smad binding motifs next to a known Gata site in the HE/HS1 region of the Gata1locus, which we find to be occupied by Gata2 at the progenitor stage during EB differentiation. This Gata1 enhancer region is required for transgene expression in both the erythroid and megakaryocytic lineage(Vyas et al., 1999). Thus, as the regulation of Gata1 expression is stage- and lineage-dependent, we speculate that the onset of Gata1 transcription in a progenitor population of erythroid-megakaryocytic potential is mediated via Gata2-Smad5 cooperation at the G1HE/HS1 element, followed by Gata2 replacement and Gata1 autoregulation upon erythroid commitment in a Smad-independent manner.

These models of Eklf and Gata1 activation argue that a layout featuring GATA motifs in combination with Smad binding sites denotes a cis-regulatory element that is utilized at the progenitor stage and could explain the block in erythroid differentiation that is observed upon perturbation or lack of Smad5 expression during hematopoiesis(Fuchs et al., 2002; Liu et al., 2003; McReynolds et al., 2007) or in response to stress erythropoiesis(Porayette and Paulson,2008).

The addition of Eklf to the growing list of Bmp4-regulated genes illustrates that the Bmp4 signaling pathway and Smad activity are required throughout hematopoiesis, as opposed to being necessary only for the initial establishment of hematopoietic fate from mesoderm during development(Lugus et al., 2007; Pimanda et al., 2007; Zafonte et al., 2007). As Gata factors regulate a multitude of genes central to the hematopoietic program,the delineation of mechanisms that integrate the cross-talk between Bmp4 signaling and the Gata factor-controlled gene network is essential to an understanding of hematopoiesis (Loose et al., 2007). Of particular interest in this regard (and for stem cell biology in general) are the mechanisms underlying lineage fate decisions and the accompanying progression from a highly proliferative progenitor state to that of lineage commitment and differentiation.

According to models of lineage fate decisions from a (bipotential)progenitor cell, commitment to one specific lineage over another is established through a cross-antagonistic mechanism of opposing transcription factors (Cantor and Orkin,2002). Quantitatively, the uncommitted progenitor state is characterized by the co-expression of such antagonistic transcription factors at a low level. However, this priming state is disrupted following a rise in the transcription levels and thereby dominance of one regulatory factor over the other, which subsequently leads to lineage commitment and ultimately lineage differentiation (Huang et al.,2007; Roeder and Glauche,2006).

Here, we describe how the stage- and lineage-dependent integration of Gata2 activity and Bmp4/Smad5 signaling versus Gata1-anchored complex binding to the same Eklf cis-regulatory element translates into such a two-tiered transcriptional profile between the megakaryocytic-erythroid progenitor and differentiating erythroid cells. As the expression of Eklf prior to erythroid commitment has recently been corroborated(Bottardi et al., 2006; Frontelo et al., 2007),studies conducted in parallel in our laboratory show that Eklf indeed plays a directive role in erythroid versus megakaryocytic development in accordance with the antagonistic model described above(Frontelo et al., 2007; Siatecka et al., 2007). Thus,a new role for Eklf as a regulator of lineage fate decisions during hematopoiesis is defined, the misregulation of which could potentially underlie disease mechanisms.

Expanding on the gain-of-function assay previously described for the Ainv18 ES cell line (Kyba et al.,2002; Manwani et al.,2007; Willey et al.,2006), we provide novel transgenic reporter and loss-of-function assays for use in ES cells and EBs. In combination with in vivo binding studies performed during the in vitro differentiation of ES cells, we are able to delineate the mechanisms of Eklf transcriptional regulation at a higher resolution than had previously been achieved, either in transgenic mouse models of Eklf expression (Anderson et al., 2000; Xue et al.,2004) or in gene ablation studies of Gata factors(Pevny et al., 1991; Tsai et al., 1994) and Smad5(Liu et al., 2003). Together with the rapidly expanding number of protocols available for the differentiation of ES cells in culture, these assays provide a powerful`toolbox' for the study of genetic interactions that govern early mouse development and lineage decision processes, similar to techniques established for the fish and frog model systems.

We thank Ian Donaldson for help with SynPlot; Michael Kyba for kindly providing the Ainv18 ES cell clone; Hogune Im, Jeffrey Grass and Emery Bresnick for training, advice and for the gift of Gata1 and Gata2 antibodies;Li Xue for animal husbandry; Paul Gadue, Jesse Lugus and Kyunghee Choi for reagents and advice; Saghi Ghaffari for help in the preparation of the manuscript; and Pilar Frontelo, Deepa Manwani, Miroslawa Siatecka and Tanushri Sengupta for advice throughout the project. This work was supported by NIH PHS grant R01 DK48721 to J.J.B. The Flow Cytometry, Microscopy and Quantitative PCR Shared Research Facilities are supported by MSSM. The Mouse Genetics Shared Research Facility is partially funded by NCI grant R24 CA88302.