The BMP/BMPR/Smad pathway directs expression of the erythroid-specific EKLF and GATA1 transcription factors during embryoid body differentiation in serum-free media

Erythroid cells are one of eight distinct blood cell lineages derived from a small population of pluripotent hematopoietic stem cells that are first formed during early embryogenesis (Metcalf, 1988). Understanding how erythroid cell-specific gene expression is accomplished has relied partly on isolation of the transcription factors that play a role in inducing expression of the β-like globin genes (Orkin, 1995; Baron, 1997). This has led to identification of founders of transcription factor families (e.g. GATA1, EKLF (KLF1)) whose related members also play important roles in other lineages (Simon, 1995; Turner and Crossley, 1999; Dang et al., 2000; Tsang et al., 2000; Bieker, 2001). Because expression of many of these genes is restricted to particular hematopoietic cell lineages, the question quickly arises as to how these intracellular regulatory molecules are themselves induced and regulated. Addressing this is necessarily an issue of developmental control, as erythroid cell production and gene expression is sequentially established at multiple sites at specific times of embryogenesis (Dzierzak and Medvinsky, 1995; Orkin and Zon, 1997; Stamatoyannopoulos and Grosveld, 2001). As erythroid tissue is mesodermal in origin, the role of adjacent germ layers in providing an instructive or permissive signal upon a naive cell, and the nature of that signal, are questions that are being addressed using genetic, cellular and biochemical means in a wide range of organisms (Choi, 1998; Beddington and Robertson, 1999; Ray and Wharton, 2001; Zon, 2001).

EKLF and GATA1 are two such transcription factors that play critical roles in erythroid cell differentiation. EKLF is a C2H2 zinc-finger protein, the presence of which is crucial for consolidating the switch from fetal γ-globin to adult β-globin expression during development (Perkins, 1999; Bieker, 2000). Interaction with its high-affinity site (CACCC element) at the proximal β-globin promoter (Miller and Bieker, 1993) helps establish the correct local chromatin structure that leads to high level β-globin transcription (Armstrong et al., 1998; Zhang et al., 2001). EKLF-null mice die from a profound β-thalassemia at the time of the switch to adult β-globin (Nuez et al., 1995; Perkins et al., 1995). EKLF expression is tightly erythroid specific during development, with its onset at E7.5 (neural plate stage) being strictly localized to the extra-embryonic blood islands of the yolk sac followed by expression in the hepatic primordia by E9.5 (Southwood et al., 1996). Paradoxically, it is also transcribed early during hematopoietic differentiation long before globin is expressed (Ziegler et al., 1999). Analysis of its own promoter has defined a conserved distal enhancer element (∼–700) that, in conjunction with the EKLF proximal promoter (∼–100), leads to high level, erythroid-specific expression in transient transfection assays as well as in transgenic mice (Crossley et al., 1994; Anderson et al., 1998; Chen et al., 1998).

GATA1 is a C4 zinc-finger protein whose presence is crucial for expression of numerous erythroid genes, for both primitive and definitive erythroid cell as well as megakaryocyte cell maturation, and for red cell viability (Weiss and Orkin, 1995; Tsang et al., 2000). GATA1-null cells are stalled at the proerythroblast stage, after which they readily undergo apoptosis (Weiss et al., 1994). GATA1 is not only expressed within the hematopoietic (erythroid, mast, megakaryocytic and eosinophilic) lineage, but also in the testes via the use of an alternative promoter element (Ito et al., 1993). Erythroid-restricted expression of GATA1 in both primitive and definitive cells requires sequences within its first intron together with an element located approx. –2.5 to –4.0 kb (Bieker, 1998). GATA1 is expressed in the E7.5 extra-embryonic blood islands during development (Whitelaw et al., 1990; Silver and Palis, 1997) and early in hematopoietic differentiation (Ziegler et al., 1999).

The early cellular expression patterns of EKLF and GATA1 have made it difficult to directly address the mechanism by which they are initially induced in development and/or during hematopoiesis, as many multipotential cell lines already express these mRNAs (Hu et al., 1997; Reese et al., 1997). However, we felt that use of the differentiating embryonic stem (ES) cell system might provide an alternate approach to this problem. ES cells are derived from the inner cell mass of E3.5 blastocysts and are the cell line of choice for genetic ablation studies because they are able to mix with host blastocysts, efficiently form chimeric mice and contribute to all adult tissues (Nagy et al., 1993). In addition, ES cells can be induced to differentiate and form embryoid bodies (EBs) that recapitulate hematopoiesis (along with other lineages) in a sequential pattern that mimics that seen during normal murine development (Keller, 1995). The morphological characteristics and expression patterns of cells in colonies formed during EB formation have been elegantly analyzed, and have led to the discovery of novel colony forming cells (transitional- and blast-CFCs) (Faloon et al., 2000; Robertson et al., 2000), as well as the most direct evidence to date in favor of the existence of the hemangioblast (Choi et al., 1998). Dissociated EBs can be replated, and specific cellular progeny (erythroid, myeloid, lymphoid) can be assayed and their numbers can be quantified (Wiles and Keller, 1991; Keller et al., 1993).

Although ES cells have been extensively used for cellular studies and as an in vitro culture model for early development, they have not been used as tools to identify the inducers and to investigate the mechanism by which inducers promote expression of specific target genes. Of interest for the present studies, neither EKLF or GATA1 are expressed in ES cells, but their expression arises during EB formation before globin expression (Simon et al., 1992; Southwood et al., 1996). This property fulfills a requirement that has been missing in attempts to investigate the onset of EKLF and GATA1 expression in hematopoietic cell lines, thus making ES cells very appealing for such studies. Although the addition of cytokines can stimulate production of red cells in developing EBs, differentiation will occur in serum alone. As a result, we first established conditions where EB formation could efficiently proceed in the absence of serum. We then used this system to identify extracellular inducers of EKLF and GATA1 erythroid genes.

R1 ES cells were maintained on mitotically inactivated primary fibroblast feeder cells in DMEM+15% FCS (Southwood et al., 1996). Culture of ES cells after removal from feeder cells and EB differentiation in methylcellulose that contained IMDM+15% FCS precisely followed established protocols (Keller et al., 1993; Weiss et al., 1994; Kennedy et al., 1997). Alternatively, serum was replaced by inclusion of KnockOut SR (Life Technologies) or BIT 9500 (Stem Cell Technologies) at 15%. Typically, 2000-5000 ES cells were plated in 1.5 ml in a 35 mm dish. Cytokines were included as needed, usually at day 0 of differentiation: BMP4 (5 ng/ml from R&D Systems or 37.5 ng/ml Genetics Institute), 100 ng/ml SCF (R&D), 5 ng/ml VEGF (R&D), 2 U/ml erythropoietin (Amgen) and 1 μM T3 (Sigma).

EBs formed in individual dishes were harvested at day 8 and total RNA was isolated after homogenization in TRI Reagent (Sigma). Typically, one tenth of this material was used for cDNA synthesis using oligo-dT (Pharmacia) and Sensiscript reverse transcriptase (Qiagen) in a volume of 20 μl. One microliter of this material was used for semi-quantitative PCR analysis with Taq polymerase (Qiagen) in a volume of 50 μl that also contained [³²P]dCTP as tracer. Cycles for each primer pair were empirically determined so as to yield product within the early exponential phase of synthesis to assure comparative analyses in the linear range. These were between 18-23 cycles. Ten microliters of product was analyzed on a 5% polyacrylamide gel, and quantitation of the dried gel was performed using a Phosphorimager and analyzed with ImageQuant software (Molecular Dynamics). Under these conditions, at least a 30-fold linear range was attained (data not shown).

PCR primers for EKLF, HPRT, bh1, βmaj, Bra, BMP4, GATA2 and GATA1 have been previously described (Weiss et al., 1994; Johansson and Wiles, 1995; Southwood et al., 1996; Schuh et al., 1999). Other primer pairs were as follows: PECAM, 5′ TGCGATGGTGTATAACGTCA and 5′ GCTTGGCAGCGAAACACTAA (382 bp); FLK1, 5′ CCATACCGCCTCTGTGACTT and 5′ ACACGATGCCATGCTGGTCA (503 bp); SCL, 5′ TATGAGATGGAGATTTCTGATG and 5′ GCTCCTCTGTGTAACTGTCC (395 bp); Smad1, 5′ TTACCTGCCTCCTGAAGACC and 5′ TGAAACCATCCACCAGCACG (220 bp); Smad5, 5′ TATCCCAACTCCCCAGCAAG and 5′ CCCAGGCAGAATCTACTTTTG (331 bp); Smad8, 5′ TATGCACCCCAGCACCCC and 5′ CATGGAGACTGCGGAAACAC (606 bp). Annealing temperature in all cases was set at 2°C below the calculated denaturation temperature.

Wild-type and dominant negative (K231R) murine BMPR1B clones were obtained from Drs L. Niswander and P. ten Dijke (Zou and Niswander, 1996). Murine Smad6 was obtained from Dr Xu Cao (Bai et al., 2000). The coding sequences were subcloned downstream of the ∼3 kb BamHI/StuI EKLF promoter in a vector that also contains a puromycin selection marker (L. Ouyang and J. J. B., unpublished). This promoter has been shown to drive high-level erythroid specific expression in tissue culture cells (Chen et al., 1998) and in transgenic mice (J. J. B., unpublished).

ES cells were electroporated under standard conditions (BioRad Gene Pulser, 400 V/125 μF), and selection in 2 μg/ml puromycin began after a 48 hour recovery. Individual colonies were selected and expanded. As transcription from the clones are predicted to yield a 5′-untranslated region that is EKLF-derived, expression of electroporated BMPR1B was monitored by RT/PCR analysis of differentiated EBs with the following primers (the first primer of the pair is unique to expression of this clone): 5′-GGTAGGATTCACCATGGTC and 5′-CTCAGTCTCTCGGAACCAG.

Our first question was to determine whether EB differentiation could be established in methylcellulose in the absence of serum. Although hematopoietic cytokines can stimulate erythroid production within EBs as they differentiate in serum-containing media, their presence is not absolutely required (Wiles and Keller, 1991). At the same time, EB development is unlike that of Xenopus, which develops faithfully in a simple buffer (Nieuwkoop and Faber, 1967). One published report established a chemically defined medium for EB formation (Johansson and Wiles, 1995); however, this was performed under suspension conditions, where EBs can form from aggregates of ES cells. As our earlier studies (Southwood et al., 1996) followed protocols that establish EB formation in semi-solid media (Keller et al., 1993), we wished to monitor serum-free EB differentiation under those conditions. In addition to providing a stringent test for any putative inducers, this protocol demands that single, physically separated cells form individual EBs, such that all cells within the developing EB are clonal. In addition, reproducibility of temporal development occurs more synchronously than in suspension. Finally, expansion of hematopoietic cells occurs in close proximity to the original colony, minimizing dispersion and making scoring and secondary plating significantly easier (Wiles, 1993). Two sources of commercial serum-free substitutes (Knockout SR from Life Technologies and BIT 9500 from Stem Cell Technologies) were tested for this purpose and compared with EB formation in 15% FBS.

R1 ES cells were removed from feeder cell culture and allowed to differentiate in methylcellulose (Southwood et al., 1996), aiming for ∼50-100 EBs/35mm dish. Inspection of the resultant EBs (data not shown) indicated that: first, the efficiency of EB formation was comparable in FBS or the knockout SR (SR1) mix (both ∼6%), but was considerably lower in the BIT 9500 (SR2) mix (≤1%). Second, the morphology of EBs were comparable, although the size of the SR2-derived EBs was variably smaller than those derived from FBS or SR1. Third, EBs grown in FBS attained a robust redness after ten days differentiation (Fig. 1A); those grown in SR1 or SR2 were very pale at the equivalent time. These data indicate that although EBs can be formed in serum-substituted conditions, hemoglobinization is considerably reduced qualitatively in the absence of serum.

To obtain a more precise idea what occurs at a molecular level under these conditions, we monitored expression of several genes by semi-quantitative RT/PCR analysis. Expression of erythroid-specific markers (EKLF, GATA1, βh1 and βmaj) from single-dish pools of day 8 EBs indicate that these genes are expressed in EBs that have been differentiated in FBS, but not in EBs differentiated in either SR1 or SR2 (Fig. 1B). These data show that, although morphologically normal EBs can be formed in the absence of FBS, neither EKLF nor GATA1, in addition to the β-like globin genes, is expressed. As the efficiency of EB formation was more robust in SR1, we generated EBs with this serum substitute for the rest of the experiments.

Although these data were encouraging, the issue remained of how lack of EKLF expression caused by non-induction in a committed erythroid cell could be distinguished from lack of EKLF expression caused by a simple absence of blood cell formation. In the extreme, a lack of EKLF could arise as a trivial consequence of deficient mesoderm formation, such that the induction system would essentially become an assay of mesoderm inducers.

To help formulate a way to address this issue, Fig. 2A lays out a scheme of erythroid commitment (Orkin and Zon, 1997) and molecular expression markers that can be used to follow the presence/absence of particular cell types within the pathway. Expression of each gene by RT/PCR analysis was used as a means to determine how far along the path the differentiating EBs have proceeded in the absence of serum (Fig. 2B). Not surprisingly, brachyury is expressed, indicating that mesoderm is formed in the serum-free EBs. More surprising was the extent to which erythroid commitment proceeded in the absence of serum, as hemangioblast (FLK1, PECAM) and hematopoietic progenitor (SCL, GATA2) markers were also expressed. In combination with the observation that GATA1, EKLF and the β-like globins are not present, we conclude that EBs, differentiated in the absence of FBS, provide a suitable assay system to screen for inducers of EKLF and GATA1 expression in hematopoietic cells.

Using the serum-free EB system, we tested whether selected cytokines could reconstitute EKLF and GATA1 expression. Serum-free culture conditions for primary hematopoietic stem cells contain a mix of cytokines that typically include IL3, GM-CSF, IL6, SCF and erythropoietin (EPO). However, our choice was directed by three considerations. First, as hematopoietic progenitors were already formed, we excluded IL3, GM-CSF and IL6 from consideration. Second, the compelling data that indicate the importance of VEGF (Kennedy et al., 1997) and the BMP family (Hogan, 1996) for erythroid differentiation directed us to include these in our tests. Finally, thyroid hormone was also included (Bauer et al., 1998). As a result, we formed EBs in the presence or absence of a cytokine ‘cocktail’ that included BMP4, SCF, VEGF, EPO and T3, and found that these were sufficient to enable EBs to express EKLF and GATA1 (Fig. 3).

We next split this cocktail into two subgroups. Inclusion of VEGF, EPO and T3 were not sufficient for EKLF or GATA1 induction (Fig. 4A). However, inclusion of BMP4, SCF and VEGF yielded a weak but detectable signal for EKLF and GATA1 (Fig. 4B). Of particular interest was that BMP4 alone appeared to be sufficient to induce this level of expression. Quantitation of this data (after normalization to HPRT) revealed that the level of EKLF expression in EBs formed with BMP4 alone was ∼2% that seen in EBs formed in FBS. It is of note that expression of β-globin was observed only in the combined presence of BMP4, VEGF and SCF.

To optimize the level of BMP4 needed to generate reasonable levels of EKLF and GATA1 expression, and to assess the contribution of SCF and EPO to the results of Fig. 4, we titrated BMP4 that was obtained from two suppliers. We found that the level of EKLF and GATA1 expression is proportional to the amount of BMP4 added, and that this reaches a plateau whose optimum concentration varies depending on the BMP4 supplier (data not shown). We then formed EBs using this optimal concentration of BMP4, and also tested whether EB formation with only SCF and EPO could direct EKLF and/or GATA1 expression. The quantitated and averaged results of three sets of experiments shown in Fig. 4C demonstrate that BMP4 alone can yield EKLF and GATA1 levels that are up to 15-20% the level seen with serum. This level can be boosted to ∼50% by the inclusion of the other four cytokines. The data of Fig. 4 thus demonstrate that BMP4 is necessary and sufficient to induce EKLF and GATA1, and that any combination of SCF, VEGF, EPO and T3 are not able to substitute for this requirement.

BMP4 expression increases endogenously in EBs that are formed in the presence of serum (Faloon et al., 2000). Our data imply that this level must be significantly lower when EBs are formed in serum-free conditions. This was directly tested by monitoring BMP4 expression in EBs that were differentiated in the various combinations of cytokines shown in Fig. 5. Quantitation of these data demonstrate that levels of BMP4 are approximately 10-fold less in EBs differentiated in the knockout SR mix versus that seen in the presence of FBS. Importantly, inclusion of exogenous BMP4 alone is sufficient to recover to ∼100% the level of endogenous BMP4 expression seen in FBS.

We next addressed whether the timing of BMP4 addition was crucial for EKLF and GATA1 induction by adding BMP4 at varying times after initiating EB formation, and then harvesting all samples at day 8. We found that the ability of BMP4 to induce EKLF expression is lost if it is added after day 3, even though it is present for four days in these samples (Fig. 6A). As GATA1 (Robertson et al., 2000) and EKLF (Fig. 6B) normally comes on by day 4 in this system, whether grown in serum or BMP4 alone, these data indicate that the competence of cells present in the EB to respond to BMP4 and express EKLF and GATA1 is transient, and that its timing coincides within the same temporal frame as when transitional and blast colony forming cells arise in developing EBs (Choi, 1998; Faloon et al., 2000; Robertson et al., 2000).

Our data implicate BMP molecules, particularly BMP4, as being important inducers of EKLF. BMP4 interacts with the BMP receptor, enabling the interaction between its two subunits (BMPR-IA or IB and II) which then leads to phosphorylation of Smad1, Smad5 and/or Smad8 prior to their association with Smad4 and translocation to the nucleus (Dijke et al., 2000; Massague and Chen, 2000). We therefore checked three predictions based on this scheme.

First, we monitored whether the appropriate Smads are present and/or induced in our serum-free EBs that were grown in BMP4. Fig. 7A shows that all are present, and that inclusion of BMP4 results in a two-fold increase in the Smad1 level.

Second, we tested the involvement of the BMP receptor in EKLF activation by disrupting its activity. Expression of a BMPR-IB point mutant that has lost its ability to bind ATP yields a powerful dominant negative protein that effectively disrupts BMP receptor function (Zou and Niswander, 1996). However, we did not wish to express such a derivative constitutively in the developing EB. As a result, we used the EKLF promoter, which contains its erythroid specificity elements within the 950 base pairs proximal to its start site of transcription (Anderson et al., 1998; Chen et al., 1998), to drive expression of the BMP receptor K231R mutant (BMPR-DN) only in the erythroid cell (see Discussion). This promoter drives expression of a linked lacZ reporter specifically to yolk sac erythroblasts and to the developing fetal liver in transgenic mice (J. J. B., unpublished). Stable ES lines were established that express either the wild-type (648-3) or dominant negative (649-6) BMP receptor constructs, and these were differentiated in the presence of serum. The results (Fig. 7B) show that EKLF is not expressed and GATA1 is barely detectable in the 649-6 line. We next revisited our previous concern and addressed how far along the hematopoietic pathway these EBs had proceeded. The results (Fig. 7B) show that mesodermal (Bra), hemangioblast (PECAM) and hematopoietic precursor (SCL, GATA2) markers are all expressed. Intriguingly, FLK1 is not detectable.

Third, we tested whether interference with the downstream signalers of the BMP pathway would also alter EKLF and GATA1 expression. By a similar design to that discussed above, we established stable ES lines that expressed the inhibitory Smad6 protein in the erythroid cell under control of the EKLF promoter and allowed these to differentiate in the presence of serum. The results (Fig. 7C) from two stable lines (Smad6-4 and Smad6-5) show that EKLF and GATA1 levels are virtually nil in these lines, but that all the other markers (Bra, PECAM, SCL, GATA2 and FLK1) are expressed.

In toto, the data of Fig. 7 enable us to conclude that EKLF and GATA1 expression are dependent upon an intact BMP receptor function, and that the likely downstream molecules are not only in place to transmit this signal but play a necessary role in this process.

We next addressed whether the BMP/Smad pathway directly induces EKLF and GATA1 expression. Initially, we examined the kinetics of its induction by isolating differentiating EBs at daily intervals from d2 through d6, leaving them intact or dispersing them into single-cell suspensions (Kanatsu and Nishikawa, 1996), and incubating them with BMP4 for 24 hours. In no case did we see induction of EKLF or GATA1, even though the dispersion protocol left the cells biologically viable (data not shown). Based on the induction kinetics of Fig. 6, we therefore alternatively focused on isolating d2 or d3 EBs and incubating them for varying lengths of time with BMP4. In each case (Fig. 8) we found that at least 2 days (and optimally 3 days) was required for significant induction of EKLF and GATA1. These results demonstrate that EKLF induction by BMP4 is not an immediate-early response, but rather uses a less direct mechanism. In combination with Fig. 7, the results suggest that successful EKLF and GATA1 induction probably requires the synthesis and/or activation of an additional factor in the erythroid cell.

Induction of lineage-specific genes that are already expressed early during differentiation has been difficult to analyze owing to limitations in finding appropriate cell lines in which to address their onset in expression. To circumvent this problem, we have taken the novel approach of using the EB differentiation system as a means to identify potential extracellular mediators of EKLF and GATA1 expression. We find that EBs can differentiate, in the absence of serum, significantly down the hematopoietic pathway, and express hemangioblast and hematopoietic markers. However, erythroid markers are not present, thus providing a suitable assay for extracellular factors needed to establish their expression. Using this system we find that the BMP/BMP receptor/Smad pathway plays a crucial role in induction of EKLF and GATA1 expression. Our approach with the EB system has enabled us to show this in two ways by showing that (1) addition of BMP4 in the absence of serum was sufficient to induce EKLF and GATA1; and (2) erythroid-specific interference of the BMP transduction pathway, even in the presence of serum, abrogated EKLF and GATA1 expression. These experiments extend the (already remarkable) range of analyses that can be obtained from differentiating EBs.

Although our effects are manifested by BMP4, our studies have not necessarily differentiated between different, yet functionally very similar, BMP molecules. BMP4, BMP2 and BMP7 are closely-related TGFβ-family ligands that interact with a distinct pair of BMP receptor membrane proteins (of type I and II), which then dimerize and induce the serine/threonine kinase activity of the type I receptor (Hogan, 1996; Dijke et al., 2000). This enables the dimeric receptor to bind and phosphorylate the intracellular mediators Smad1, Smad5 or Smad8, which then interact with Smad4 before shuttling to the nucleus (Kretzschmar and Massague, 1998). Our studies implicate the BMP/BMP receptor/Smad pathway in specific erythroid gene induction. But which of the many components in this pathway are likely players? Genetic and developmental studies enable us to parse this list.

The role of BMP4 in blood formation was initially suggested because of its strong ventralizing activity in the Xenopus animal cap assay (Harland, 1994) and further verified by dominant-negative studies (Xu et al., 1999). Its importance for hematopoietic differentiation in the mouse has been established from effects of its genetic ablation (Winnier et al., 1995), which disrupts mesoderm and blood cell formation in the yolk sac. Similarly, disruption of one of the BMP receptor molecules (BMPR1A) disrupts mesoderm formation (Mishina et al., 1995). BMP2 and BMP7 have also been implicated in mesoderm related induction and activation of Smad1 and Smad5 (Massague et al., 2000). However, disruption of BMP7 leads to kidney, eye and skeletal problems, but does not affect hematopoiesis (Dudley et al., 1995; Luo et al., 1995). Ablation of BMP2 leads to malformation of the amnion and chorion and defects in cardiac development (Zhang and Bradley, 1996). Smad4 is essential for mesoderm induction (Yang et al., 1998), but Smad5 deficiency leaves hematopoietic precursors and blood cell formation unaffected, even though vasculogenesis and angiogenesis are disrupted (Chang et al., 1999; Yang et al., 1999). Disruption of murine Smad1 or Smad8 have not yet been reported; however, the role of Smad8 in mesodermal patterning may be complex, as there is evidence in Xenopus that it negatively modulates signaling by BMP (Nakayama et al., 1998).

In terms of developmental profile, BMP4 is expressed before gastrulation in the extra-embryonic ectoderm adjacent to the epiblast, in position to influence the adjacent mesoderm that emerges from the posterior primitive streak (Waldrip et al., 1998). Slightly later, the posterior mesoderm itself is expressing BMP4 (Winnier et al., 1995). In addition, a hedgehog/BMP4 pathway has been implicated in the ability of visceral endoderm to respecify ectodermal cells to a posterior mesodermal fate (Belaoussoff et al., 1998; Bhardwaj et al., 2001; Dyer et al., 2001). Intriguingly, Smad1 is also induced after gastrulation within the mesodermal cell region of the primitive streak (Waldrip et al., 1998). Our data demonstrate that both endogenous BMP4 and Smad1 levels increase in developing EBs in the presence of exogenous BMP4. As a result, from these genetic and developmental studies, our tentative model is that the BMP4/BMP receptor/Smad1 pathway is the crucial one for EKLF and GATA1 expression.

Such a scenario would also be consistent with studies in lower vertebrates. For example, BMP4 (Dale et al., 1992; Jones et al., 1992) and Smad1 (Wilson et al., 1997) are potent ventralizing agents in injected Xenopus embryos. In addition, induction of Xenopus blood cell gene expression in ectoderm by ectopic GATA1 requires an intact BMP pathway (Huber et al., 1998). Studies in the chick (Connolly et al., 2000) and in zebrafish (Hammerschmidt et al., 1996) also support a conserved role for BMP-mediated signaling in directing a ventral fate as part of a dorsal/ventral patterning process.

Consistent with this idea, the downstream effects of BMP4 on patterning are known to be crucially sensitive to its concentration, with gradients of effective BMP4 levels leading to dorsal (low level) or ventral (high level) fates within the responding mesoderm (Dale, 2000; Zon, 2001), and ventral mesoderm differentiating into blood. Our data are concordant with this observation in two ways. First, the highest level of EKLF and GATA1 expression is attained with the highest levels of input BMP4. Second, endogenous BMP4 levels in serum-free EBs must not be sufficient to induce EKLF and GATA1; however, an increase of 8- to 10-fold (owing to positive autoregulation by BMP4) leads to successful induction.

The inability to detect FLK1 expression in the BMPR-DN EBs was initially surprising, as FLK1 is expressed during serum-free EB differentiation. However, one explanation is that the low level of endogenous BMP4 present in serum-free EBs is sufficient for FLK1 transcript accumulation, while expression of the BMPR-DN construct depletes BMP4 below this threshold and thus prevents any detectable FLK1 expression. As a BMPR-DN construct has been shown to neuralize ventral tissue in Xenopus (Graff et al., 1994), it is also possible that cells within these EBs are not able to fully differentiate into endothelial cells.

Using the EKLF promoter to drive expression of dominant negative BMPR1B and Smad6 proteins circumscribes their effects in two ways. First, the impact of these molecules can only occur after induction of the EKLF gene. As a result, these negative effectors will not be expressed until after hematopoiesis has begun. Limiting their presence in this way assures that a wide-range, and thus less directed, effect is avoided. Second, even when their expression is eventually downregulated (in the same way that EKLF is downregulated), it will be beyond the window of opportunity to turn on EKLF and GATA1.

Implication of the BMP/BMP receptor/Smad pathway in EKLF regulation begs the question of whether there are any appropriate Smad binding sites in the EKLF promoter. A search for Smad consensus 5′CAGAC sites in both the murine and human EKLF and GATA1 promoters reveals their presence; however, Smad proteins bind this site with relatively low affinity, and usually require interaction with another DNA binding co-factor to effect a high-affinity interaction with DNA (Dijke et al., 2000; Massague and Wotton, 2000; Wrana, 2000). Of particular interest is the recent identification of OAZ as a Smad co-factor for BMP2 signaling (Hata et al., 2000). OAZ is a large protein with 30 zinc fingers that associates with activated Smad1/Smad 4 and mediates induction of the Xenopus homeobox Xvent-2 gene by binding to a DNA site adjacent to the Smad-binding site. However, not all BMP2 responsive genes have these sites and use OAZ, and the lack of other examples precludes using an established OAZ consensus element to search for such a site in the EKLF and GATA1 promoters. Also relevant is the ability of Smads to interact with members of the AML family of transcription factors (Pardali et al., 2000). AML (Runx1) is a transcription factor that is commonly rearranged in acute myeloid and lymphocytic leukemias and plays a crucial role in hematopoiesis (Speck and Dzierzak, 2000). Although the Smad/AML study focused on the TGFβ/Smad3 pathway induction of IgA, the other interactions observed may be relevant to EKLF expression, particularly that of Smad1 with AML1b (Pardali et al., 2000).

However, we have not yet proven that the effect of BMP4 on EKLF and GATA1 expression is conveyed directly by Smad protein. Although the ability of erythroid-driven dominant negative constructs to interfere with EKLF expression is consistent with this idea, it is possible that an intermediate step, such as transcriptional activation of another factor, may occur first. Indeed, our kinetic data imply that such a scenario may very well be operant.

In any case, the inability to use hematopoietic cell lines for analysis of EKLF and GATA1 promoters means that analyses designed to delimit the BMP responsive element will have to rely on stable reporter genes, which contain selected promoter deletions, that are integrated into ES cells and then tested after differentiation into EBs. As the boundaries of these elements have been localized in both the EKLF and GATA1 promoters, use of suitable promoter/reporter constructs that are cloned into the same chromosomal location (either by chosen-site integration into the HPRT locus (Bronson et al., 1996) or by recombination-mediated cassette exchange approaches (Bouhassira et al., 1997)) will provide a suitable forum within which to begin efforts at more precise localization.

Although we have used EB differentiation strictly as a tool to decipher inductive mechanisms of EKLF and GATA1 expression, it is of interest to compare the colonies formed in the absence of serum with other EB-derived cells. Secondary plating of very early differentiating EBs gives rise to three types of colonies: secondary EBs, transitional colonies and blast colonies (Faloon et al., 2000; Robertson et al., 2000). The expression profiles of markers within each of these colonies have been analyzed, and none of these resembles serum-free EBs. In particular, serum-free EBs are Bra+/FLK1+/SCL+/GATA1-/βmaj-, secondary EBs are Bra+/FLK1^+/–/SCL-/GATA1-/βmaj-, transitional colonies are Bra+/FLK1+/SCL+/GATA1+/βmaj^+/–, and blast colonies are Bra-/FLK1+/ SCL+/GATA1+/βmaj+. Instead, serum-free EBs most resemble primary colonies at day 3.5-3.75 of differentiation, which are Bra+/FLK1+/SCL+/GATA1-/βmaj- (Robertson et al., 2000). As a result, it would appear that serum-free EBs are ‘stalled’ at a stage where an appropriate input signal BMP is required (presumably by day 4) for further progression and differentiation. Consistent with this idea is our observation that the competence of EBs to respond to BMP4 and induce EKLF is lost after this time.

Two studies have demonstrated that EBs can form in the absence of serum. One used suspension EB cultures to follow formation of mesoderm and hematopoiesis after inclusion of TGFβ family members, and found that activin A and BMP4 were particularly important (Johansson and Wiles, 1995; Wiles and Johansson, 1999). A second study, published after we began the present analyses, used a similar system to ours to demonstrate that BMP4 is essential for formation of lymphoid, myeloid and erythroid cell lineages, as monitored by an extensive series of cell-surface markers (Nakayama et al., 2000). Interestingly, VEGF synergized with BMP4 to enhance its effect, with BMP4 being absolutely required during the first four days of differentiation. Consistent with this, our data show that EKLF is no longer induced if BMP4 is added after day 3. Nakayama et al. stated that EBs grown in BMP4 were white, but when VEGF (and SCF) were added they were red. Our data provide a molecular explanation for this, as BMP4 alone induced EKLF and GATA1, but β-globin expression was seen only when SCF and VEGF were also included.

Tremblay et al. recently demonstrated that Smad1-deficient mice, although embryonic lethal, are able to form blood cells (Tremblay et al., 2001). As a result, the molecular details of how the Smad pathway may be interfacing with EKLF and GATA1 expression remains a complex issue.

We thank Drs Liaohan Ouyang, Lee Niswander, Peter ten Dijke and Xu Cao for plasmids, Dr Gordon Keller for differentiation protocols, and Genetics Institute for BMP4. This work was supported by PHS grant DK48721 to J. J. B.