Combinatorial regulation of tissue specification by GATA and FOG factors

Throughout evolution, increases in organismal complexity have been associated with the development of intricate transcriptional regulatory mechanisms that control the differentiation of diverse tissues. These regulatory networks comprise diverse families of transcription factors, non-DNA-binding partners, and enzymes that modify their activity and the bound DNA. The GATA family of transcription factors is a group of zinc finger proteins that are conserved in organisms from worms to mammals. GATA factors are notable for controlling the development of hematopoietic, neural and cardiac tissues in vertebrates and related tissues in invertebrates. In all metazoan species studied with the exception of C. elegans, dedicated GATA co-factors termed Friend of GATA (FOG) proteins have been identified that function exclusively with GATA factors to control cell differentiation and function through diverse modes of gene regulation.

The founding member of the GATA family was discovered in the late 1980s as an erythrocyte-specific protein that binds to enhancers of the chicken globin genes (Evans et al., 1988; Tsai et al., 1989; Martin and Orkin, 1990). Named GATA1 (or GATA-1) after the WGATAA DNA sequence to which it binds, this protein was found to participate in both the activation of adult globin genes and the repression of embryonic globin genes in erythrocytes (Martin and Orkin, 1990; Talbot et al., 1990; Raich et al., 1995). GATA1 is a 50 kDa protein that is highly conserved in mouse and humans and contains two zinc finger domains and an N-terminal activation domain (Tsai et al., 1989; Martin and Orkin, 1990). The C-finger is required for DNA binding, while the N-finger participates in DNA binding at a subset of GATA motifs termed palindromic sites (Trainor et al., 1996). In addition, the N-finger is required for recruitment of its essential non-DNA-binding co-factor Friend of GATA1 (FOG1) (Tsang et al., 1997). Mice with targeted deletion of Gata1 or Fog1 (Zfpm1) die from severe anemia at E10.5 due to ineffective red blood cell differentiation, revealing that both of these proteins are required for erythropoiesis (Fujiwara et al., 1996; Tsang et al., 1998).

Since the initial discovery of GATA1 and FOG1, several homologs of both proteins have been extensively characterized. These studies have revealed intricate mechanisms of transcriptional regulation by which these proteins control cell fate decisions and drive cellular differentiation. In this review, we will discuss the diverse mechanisms by which GATA and FOG proteins collaborate to regulate transcription and control lineage specification. We begin with an overview of the FOG and GATA families. Next, we review the functions of these factors in mammalian hematopoiesis, as well as their contributions to the development of other tissues from flies to mammals. We then discuss recent mechanistic insights into how these factors cooperate and antagonize one another in the course of mammalian hematopoiesis. Finally, we provide an outlook of the open questions and future challenges of the field.

In mammals, there are six GATA proteins and two FOG proteins (Evans et al., 1988; Yamamoto et al., 1990; Lee et al., 1991; Arceci et al., 1993; Laverriere et al., 1994; Tsang et al., 1997; Tevosian et al., 1999) (Fig. 1). All six GATA factors contain both the N and C zinc finger domains and all are capable of interacting with FOG proteins (Tevosian et al., 1999) (Fig. 1A). The zinc finger domains of these proteins are highly conserved between factors and species (Lowry and Atchley, 2000). Additionally, there is a highly conserved region immediately C-terminal to the C zinc finger that contains the nuclear localization signal (NLS) as well as a region of basic amino acids containing sites for post-translational modifications that are required for DNA binding and transcriptional activity (Crossley and Orkin, 1994; Morrisey et al., 1997; Hung et al., 1999; Molkentin, 2000). Although all six GATA proteins also possess an N-terminal activation domain, this diverges considerably between factors (Martin and Orkin, 1990; Arceci et al., 1993; Morrisey et al., 1997; Molkentin, 2000). The N-terminus of GATA1 and GATA2 is moderately conserved, whereas the activation domain of GATA3 shares no sequence similarity with GATA1/2 (Yang and Evans, 1992; Yang et al., 1994). By contrast, GATA4 contains two well-defined activation domains at its N-terminus that are distinct from GATA1/2/3 but are moderately conserved in GATA5/6 (Morrisey et al., 1997). Thus, although the location of the activation domain of GATA proteins is conserved, there is significant divergence in the sequence of these domains. In addition to FOGs, GATA factors are capable of interacting with other transcriptional co-regulators, mostly through the C-finger and the neighboring basic domain; these are discussed further below (Lowry and Mackay, 2006).

Each GATA factor shows a specific and regulated expression pattern during development, with GATA1/2/3 being prominently expressed in hematopoietic cells (Weiss and Orkin, 1995). GATA2 and GATA3 also show functionally relevant expression in endothelia and the central nervous system (Lee et al., 1991; Nardelli et al., 1999). GATA4/5/6 are not expressed in hematopoietic cells, but play crucial roles in heart, lung, gastrointestinal, muscle and neuronal development (Molkentin, 2000).

FOG1 is a 110 kDa protein with nine zinc finger domains, none of which directly contacts DNA in in-vitro studies (Fig. 1B) (Tsang et al., 1997). The zinc fingers comprise two types: four C2H2 type fingers (2, 3, 4, 8) and five rare C2HC type fingers (1, 5, 6, 7, 9). The C2HC type fingers 1, 5, 6 and 9 are capable of mediating the interaction with the GATA N-finger (Fox et al., 1999). FOG1 also contains two domains that are known to recruit transcriptional co-repressors. These are located at the extreme N-terminus and between zinc fingers 6 and 7, and are known to recruit the nucleosome remodeling and histone deacetylase (NuRD) complex and the CtBP co-repressor, respectively (Fox et al., 1999; Hong et al., 2005). FOG1 expression largely overlaps spatiotemporally with that of GATA1 (Tsang et al., 1997). It functions predominantly in the hematopoietic system through interactions with GATA1/2/3, but it also has functions in other tissues, including the gonads, intestine and heart.

FOG2 (ZFPM2) is the only mammalian homolog of FOG1 (Svensson et al., 1999; Tevosian et al., 1999). It is slightly larger than FOG1, yet has only eight zinc finger domains, three of the C2H2 type (2, 3, 4) and five of the C2HC type (1, 5, 6, 7, 8) (Fig. 1B). FOG1 and FOG2 share 31% identity over their entire length, but have a much higher degree of conservation in the zinc finger domains (Cantor and Orkin, 2005). The NuRD and CtBP interaction domains are also conserved in FOG2. Like FOG1, FOG2 binds the N-fingers of GATA proteins via its C2HC type zinc fingers (Fox et al., 1999). However, FOG2 is not expressed in the hematopoietic system, but is instead broadly expressed at high levels in the heart, brain, liver and lungs (Lu et al., 1999; Tevosian et al., 1999). This expression pattern mirrors that of GATA4/5/6, implicating FOG2 as a co-factor for these non-hematopoietic GATA factors. Whereas the functional interaction between FOG2 and GATA4 in several developmental systems is well documented, little is known about the interaction between FOG2 and GATA5/6. Thus, we have focused our discussion of FOG2 function predominantly on its GATA4-dependent activity.

In the hematopoietic system, GATA1 is required for the differentiation of erythrocytes, megakaryocytes, eosinophils and mast cells (Pevny et al., 1991; Weiss and Orkin, 1995; Hirasawa et al., 2002; Migliaccio et al., 2003), whereas GATA2 maintains stem and progenitor cell populations and thus is required for proper development of the entire hematopoietic compartment (Tsai et al., 1994). Although not required for terminal differentiation of erythroid cells or megakaryocytes, GATA2 is required for mast cell maturation (Tsai and Orkin, 1997). GATA3 is expressed in thymocytes, where it is required for the development and function of Th2 T-cells (Ting et al., 1996; Zheng and Flavell, 1997). As mentioned above, expression of FOG1 largely overlaps with that of GATA1, including erythroid and megakaryocytic lineages, where it collaborates with GATA1 to promote differentiation. FOG1 is also expressed in T-lymphocytes, where it represses the development of Th2 helper T-cells through an association with GATA3 (Zhou et al., 2001) (Fig. 2; Table 1).

Several lines of evidence have revealed that the physical interaction between GATA1 and FOG1 is required for normal blood development. First, yeast two-hybrid screening identified a set of GATA1 N-finger mutants that could not interact with FOG1 and were unable to rescue maturation of a GATA1-deficient erythroid cell line (Crispino et al., 1999). Co-expression of a FOG compensatory mutant that can interact with the V205G GATA1 mutant rescued red blood cell development, confirming that the interaction is required (Crispino et al., 1999). Second, mice that harbor the V205G mutant died in utero with a phenotype that closely resembles that of GATA1 and FOG1 deficiencies (Chang et al., 2002). Finally, a set of closely related inherited human blood diseases has been associated with GATA1 mutations, including the GATA1^V205M mutation (Nichols et al., 2000; Ciovacco et al., 2008). Together, these studies demonstrated that GATA1 and FOG1 are critically dependent on one another for their functions in blood development. Notably, mouse embryos that bear FOG non-interacting mutants of both GATA1 and GATA2 fail to produce normal erythrocytes and megakaryocytes, just as with Fog1^−/− embryos, indicating that the entire function of FOG1 in these lineages is mediated through interaction with GATA proteins (Chang et al., 2002).

Although GATA1 and FOG1 cooperate in red blood cells and megakaryocytes, several studies have revealed a role of FOG1 in antagonizing GATA1 in order to inhibit mast cell and eosinophil fate in favor of erythro-megakaryocytic fate (Table 1) (Martin et al., 1990; Zon et al., 1993; Tsang et al., 1997; Querfurth et al., 2000; Yu et al., 2002; Migliaccio et al., 2003). First, Querforth and colleagues found that exogenous expression of FOG1 in eosinophils led to loss of eosinophil markers and acquisition of a multipotent phenotype (Querfurth et al., 2000). Second, careful study of the hematopoietic compartment of Gata1^V205G knock-in mouse embryos revealed increased mast cell potential at the expense of erythro-megakaryocytic cells and aberrant expression of mast cell genes in erythroid and megakaryocytic cells (Cantor et al., 2008). Third, two mouse models carrying the Fog1^R3K5A knock-in mutation, which abrogates the FOG1-NuRD interaction, display a similar phenotype: defective hematopoiesis characterized by extramedullary hematopoiesis, anemia, thrombocytopenia and granulocytosis (Gao et al., 2010; Gregory et al., 2010). Finally, exogenous expression of FOG1 in mast cell progenitors induces erythro-megakaryocytic features (Cantor et al., 2008). Together, these data indicate that FOG1 antagonizes mast cell and eosinophil differentiation, presumably by repressing GATA1-mediated activation of genes crucial for these lineages.

The roles of GATA4 and FOG2 in cardiac tissues have been extensively studied. Two separate Fog2 knockout mouse models have been shown to display severe malformation of the heart that caused embryonic lethality between E12.5 and E15.5 (Svensson et al., 2000b; Tevosian et al., 2000). Careful examination of these hearts revealed a common atrioventricular valve, an overriding aorta, subpulmonic stenosis, a subaortic ventricular septal defect and absence of the coronary vasculature.

Gata4 null mice show an embryonic lethal phenotype at E7.0-9.5 due to failed ventral morphogenesis and defective heart tube formation (Kuo et al., 1997; Molkentin et al., 1997). The considerably earlier lethality of Gata4^−/− mice compared with Fog2^−/− mice indicates that GATA4 has FOG2-independent roles during embryonic heart development. Early studies showed that GATA4 is expressed in the precardiac splanchnic mesoderm, where it activates the promoters of cardiomyocyte-specific genes (Charron et al., 1999). FOG2 represses GATA4 activation of these promoters in transcriptional reporter assays, indicating that FOG2 may function to antagonize GATA4 in cardiomyocytes (Svensson et al., 1999; Lin et al., 2004). In order to fully assess the role of FOGs in GATA4 function in vivo, a knock-in mutation of residue V217 of GATA4, which corresponds to V205 of GATA1, was engineered in mice. Gata4^V217G/V217G embryos also died in utero, but survived to E11.5-13.5, similar to the Fog2^−/− mice (Crispino et al., 2001). Consistent with this, the hearts of Gata4^V217G/V217G embryos revealed a phenotype strikingly similar to that of Fog2^−/− mice. Two shared characteristics of the malformed hearts of these mice are worth noting: (1) the coronary vasculature was completely absent; and (2) the endocardial cushion cells of the pulmonary and aortic outflow were more numerous and hyperproliferative. The former observation is indicative of FOG2 serving as an essential co-factor to GATA4 in the initiation of the coronary vasculature. Re-expression of FOG2 specifically in cardiomyocytes of Fog2^−/− mice recovered the formation of coronary vasculature, indicating that GATA4 and FOG2 function together in cardiomyocytes, probably in a paracrine fashion, to promote angiogenesis (Tevosian et al., 2000). The latter finding suggests a role for FOG2 in controlling the proliferation of endocardial cushion cells.

The role of GATA4 and FOG proteins in the formation of the endocardial cushions is complex and has required multiple conditional mouse knockout models to unravel – particularly in order to bypass the early lethality of Gata4^−/− embryos.

Cardiomyocyte-specific deletion of Gata4 caused a loss of mesenchymal cells from the endocardial cushions, suggesting a non-autonomous activity of GATA4 from the cardiomyocytes on cushion cells (Zeisberg et al., 2005). This function appears to be antagonized by FOG2, as cardiomyocyte-specific FOG2 ablation results in hyperplastic cushions, much like in Fog2^−/− mice (Zhou et al., 2009). The mechanistic basis of this non-autonomous effect remains unknown.

Conditional ablation of GATA4 in endothelial cells also resulted in hypocellular endocardial cushions (Rivera-Feliciano et al., 2006). In this case, the defect can likely be attributed to failure of the endothelial cells to undergo epithelial-to-mesenchymal transition (EMT), which is required for derivation of the mesenchymal cells of the cushions. Consistent with this interpretation, collagen invasion assays with Gata4 mutant endothelial cells revealed reduced EMT. Investigations of the role of FOG2 in this context have yielded conflicting results. Zhou et al. found no endocardial cushion defect upon endothelial ablation of FOG2 (Zhou et al., 2009). By contrast, another study found significant increases in the rate of EMT of Fog2 knockout endothelial cells, suggesting that FOG2 represses EMT required to derive mesenchymal endocardial cushion cells (Flagg et al., 2007). Epicardial cells can also contribute to the mesenchyme of endocardial cushions (Gittenberger-de Groot et al., 1998), and epicardial-specific knockout of Fog2 resulted in hyperplastic cushions (Zhou et al., 2009), again suggesting that FOG2 might act to inhibit the generation of mesenchymal cushion cells via EMT. Similar epicardial-specific knockouts of Gata4 or other GATA factors are needed to elucidate the mechanism by which they regulate the contribution of epicardial cells to endocardial cushion formation.

Finally, specific ablation of FOG1 in endothelial cells revealed hypoplastic endocardial cushions, similar to the endothelial-specific knockout of Gata4 (Katz et al., 2003). Here, though, EMT was normal but the resulting mesenchymal cells failed to proliferate. Thus, FOG1 is likely to collaborate with GATA4 to promote the proliferation of the mesenchymal cells once they have been derived. These complex activities are summarized in Table 1.

In addition to the endocardial cushion defect, early fetal cardiomyocyte-specific knockouts of Gata4 and Fog2 resulted in loss of the coronary vasculature and severely reduced proliferation of cardiomyocytes, causing myocardial thinning (Zeisberg et al., 2005). Loss of expression of either gene in later cardiomyocytes caused reduced ventricular function and early mortality due to heart failure (Bisping et al., 2006; Oka et al., 2006; Zhou et al., 2009). These mice also had less coronary vasculature than controls. Furthermore, re-expression of GATA4^V217G in the place of the ablated GATA4 failed to rescue the ventricular function or vasculogenesis (Zhou et al., 2009). Together, these studies reveal that GATA4 and FOG2 collaborate to control the proliferation, survival and function of cardiomyocytes and to promote angiogenesis in the developing and adult heart (Table 1).

The above analyses have revealed diverse mechanisms and modes of interplay between FOGs and GATAs. This principle also applies in other, less well studied, tissues where these factors are important. Although the following examples are not comprehensive, they illustrate the complexity of interactions between these factors in different tissues.

The reproductive tract provides a particularly elaborate case. In the testis, GATA4 is expressed in Sertoli cells, where it drives the production of MIS (Müllerian-inhibiting substance, or anti-Mullerian hormone) to inhibit female reproductive development (Anttonen et al., 2003). FOG2 expression decreases in Sertoli cells as testicular differentiation proceeds, but MIS expression remains high. In the female gonad, FOG2 is expressed in the ovary and inhibits the ability of GATA4 to activate transcription of the MIS gene, preventing production of MIS and allowing female development to occur (Table 1). Thus, FOG2 appears to repress MIS expression to control male versus female gonadal development. Additionally, GATA4 and FOG2 function early in male gonadal development to induce expression of Sry, which directs Sertoli cell formation (Tevosian et al., 2002). Thus, GATA4 and FOG2 collaborate to promote formation of a male-specific cell type in the early male gonad, but in the female gonad FOG2 antagonizes GATA4-mediated activation of a male-specifying hormone (Table 1). GATA1 is also expressed in Sertoli cells (Yomogida et al., 1994), but cell type-specific ablation of GATA1 did not alter mouse testicular function (Lindeboom et al., 2003).

In the small intestine, GATA4 and FOG1 are expressed in the proximal intestine, but not in the ileum. GATA4 has been shown to activate genes crucial for proximal intestine function (Bosse et al., 2006). To determine the contribution of FOG1, Beuling and colleagues used an intestine-specific Gata4 knock-in model in which wild-type GATA4 expression is ablated but GATA4^V217G continues to be expressed. They found that ileal-specific genes were induced in the proximal intestine but proximal-specific genes were unaffected (Beuling et al., 2008). Together, these data suggest that GATA4 and FOG1 function together to repress ileal fate but that GATA4 functions independently of FOG1 to activate proximal intestine-specific genes (Table 1).

Both GATA4 and FOG2 are expressed in the mesenchymal cells of the lung and diaphragm and are required for normal lung development and diaphragm function. Mice bearing a mutation that causes production of a truncated FOG2 protein completely lack the accessory lobe of the lung and have a demuscularized diaphragm (Ackerman et al., 2005). Gata4^−/− mice have similar defects (Jay et al., 2007), as do Gata4^V217G/V217G mice, in which the FOG interaction is disrupted (Ackerman et al., 2007). Mice bearing the analogous mutation in GATA6 show no phenotype (Ackerman et al., 2007). Together, these studies indicate that GATA4 and FOG2 collaborate to control the function of the mesenchymal cells of the lung and diaphragm (Table 1).

The six GATA proteins in mammals are highly conserved in all vertebrate species studied, including Danio rerio and Xenopus (Patient and McGhee, 2002). The genome of Danio rerio has three FOG genes, including one ortholog of Fog1 and two of Fog2 (Fig. 1B) (Walton et al., 2006). Each zebrafish FOG protein has the same zinc finger structure as its mammalian ortholog, except that Fog1 contains an additional zinc finger between fingers 4 and 5 that appears to be marginally conserved in mouse FOG1. All three zebrafish FOG proteins possess the conserved NuRD and CtBP interaction domains and are capable of repressing a GATA4-dependent reporter (Walton et al., 2006). In developing zebrafish embryos, fog1 (zfpm1) is expressed in the heart, the nervous system and the inner cell mass (ICM), which is the site of hematopoiesis. fog2a and fog2b (zfpm2a and zfpm2b) expression is restricted to the developing brain (Walton et al., 2006), although nothing is known about their functions. Xenopus, however, has only one FOG homolog, xFOG, which is most similar to murine FOG1 (Fig. 1B). xFOG is expressed in the spleen, liver, testis, heart and brain, and thus is thought to encompass the combined expression of mammalian FOG1 and FOG2 (Deconinck et al., 2000).

In the hematopoietic compartment, zebrafish fog1 depletion by morpholino injection caused a dramatic decrease in the number of mature erythroid cells and platelets and a concomitant increase in myeloid cells (Amigo et al., 2009). Nonsense mutations in the zebrafish homolog of mammalian GATA1 (gata1) resulted in a similar phenotype (Lyons et al., 2002), suggesting that zebrafish Fog1 and Gata1 collaborate to drive erythroid and megakaryocyte differentiation and inhibit myeloid fate (Table 1).

Several studies have revealed that GATA1/2 function similarly in Xenopus hematopoiesis as in other vertebrates (Zon et al., 1991; Kelley et al., 1994; Mead et al., 2001). The initial report of xFOG function in Xenopus hematopoiesis relied on an overexpression system in embryos (Deconinck et al., 2000). This study surprisingly revealed that xFOG repressed primitive red blood cell formation in a CtBP-dependent manner – in contrast to the known activity in mammals and zebrafish. A more recent study, however, has challenged these conclusions and argued that xFOG instead promotes red blood cell formation in Xenopus (Mimoto and Christian, 2012). In this latter study, morpholino knockdown of xFOG was found to cause decreased red blood cell numbers and hemoglobin levels. By repeating the overexpression experiments performed in the initial study, Mimoto and Christian found that overexpression of xFOG did indeed block red blood cell formation (suggesting a dominant-negative effect), but in their hands the effect was CtBP independent. Elegant experiments to rescue the defect caused by the anti-FOG morpholino revealed that xFOG is required for red blood cell maturation and survival in Xenopus and that the interaction with NuRD, but not CtBP, contributes to this activity (Table 1). In addition, the pro-apoptotic protein Bim, the gene for which (Bcl2l11) is known to be repressed by GATA1 in mouse erythrocytes, was identified as a crucial target of FOG-NuRD-mediated repression in Xenopus erythrocytes, which is likely to explain the increased apoptosis of these cells upon depletion of xFOG (Gregory et al., 1999; Maeda et al., 2009; Mimoto and Christian, 2012).

In the zebrafish heart, reduced fog1 expression caused a large pericardial effusion and an unlooped heart tube (Walton et al., 2006). This defect could be rescued by microinjection of wild-type murine Fog1 mRNA but not an mRNA encoding a mutant of FOG1 that lacks the NuRD interaction domain (Walton et al., 2006). Thus, in both the heart and the hematopoietic system, the activities of anamniote FOGs and GATAs strongly resemble those seen in the mammalian system (Table 1).

The Drosophila genome encodes two GATA factors, Pannier (Pnr) and Serpent (Srp), and a single FOG protein, U-shaped (Ush) (Abel et al., 1993; Ramain et al., 1993; Rehorn et al., 1996; Cubadda et al., 1997; Haenlin et al., 1997). Ush shares 20% homology with mouse FOG1 (Fig. 1). It contains nine zinc fingers in a slightly altered orientation compared with mammalian FOGs and does not contain the NuRD interaction motif. Early studies showed that Ush bound to the N-finger of Pnr and inhibited its activity in transcriptional assays (Haenlin et al., 1997). Loss-of-function mutants revealed a role for Pnr in promoting the formation of sensory mother cells, which control the production of imaginal bristles; this activity is inhibited by Ush (Haenlin et al., 1997). Thus, in the absence of Ush, there is an increase in bristle formation. Pnr is also required for cardial cell formation, and loss of Ush expression results in expansion of this lineage (Fossett et al., 2000). This phenotype is reminiscent of the expansion of endocardial cushion cells in the hearts of Fog2^−/− and Gata4^V217G/V217G mice (Table 1).

The second Drosophila GATA protein, Srp, was identified in studies on the fly equivalent of the hematopoietic system (Rehorn et al., 1996). There are three types of blood cells (or hemocytes) in Drosophila: plasmocytes, crystal cells and lamellocytes. Gain- and loss-of-function studies revealed that the specification of these lineages is intricately controlled by GATA and FOG factors. Srp promotes the formation of crystal cells (Rehorn et al., 1996), whereas enforced expression of Ush abolishes crystal cell fate (Fossett et al., 2001). However, Srp and Ush collaborate to produce plasmocytes (Fossett et al., 2001). Thus, a precursor hemocyte cell expressing both Ush and Srp will make a plasmocyte, but one expressing only Srp, or an isoform of Srp lacking the N-finger (SrpC), will make a crystal cell (Waltzer et al., 2002; Fossett et al., 2003) (Table 1). This mechanism is strikingly similar to the control of mammalian erythroid/megakaryocytic and mast cell fate by GATA1 and FOG1 (see Fig. 2) in that GATA and FOG collaborate to promote one lineage while FOG antagonizes GATA function in the other. More recently, Ush has also been shown to suppress lamellocyte fate and to serve as a tumor suppressor in flies with a leukemia-like overproliferation of lamellocytes (Sorrentino et al., 2007). Thus, Ush is a crucial regulator of hemocyte proliferation and differentiation.

Together, these studies reveal evolutionary conservation of a GATA/FOG network in cell fate determination through multiple species and tissues. However, this network is complex, with FOGs having diverse and context-dependent effects on GATA-regulated transcription. In the next section, we discuss the mechanisms underlying this diversity in transcriptional response.

Given the ease of accessing hematopoietic cells and the availability of targeted murine mutants of GATA and FOG factors, studies of mammalian hematopoiesis – particularly the red blood cell lineage – have provided a wealth of insights into the mechanisms by which these proteins control transcription. The G1E cell line model of erythroid differentiation, derived from mouse Gata1⁻ embryonic stem cells, has been especially useful. G1E cells are stalled at the proerythroblast stage of erythroid differentiation due to the lack of GATA1 (Weiss et al., 1994). Upon restoration of GATA1 expression, they faithfully recapitulate erythroid differentiation up until the late orthochromatic stage prior to enucleation (Weiss et al., 1997). By stable expression of a GATA1-ER fusion protein (in G1ER cells), differentiation of these cells can be achieved by the addition of β-estradiol to the medium (Gregory et al., 1999). A parallel cell line expressing FOG-binding mutant GATA1 (GATA1^V205M) as an ER fusion protein was generated, and these cells were deficient for markers of erythroid maturation, including expression of hemoglobin, cell cycle arrest, and nuclear condensation (Crispino et al., 1999; Nichols et al., 2000). Over the last decade, many studies have used these cell lines to elucidate the FOG-dependent and -independent activities of GATA1 during erythropoiesis.

Analysis of the expression level of known GATA1 target genes in reconstituted G1ER cells revealed that GATA1 both activates and represses gene expression through FOG-dependent and FOG-independent mechanisms. In the initial study of the GATA1^V205G mutation, it was found that GATA1^V205G failed to activate the globin genes and the Band3 (Slc4a1) gene, which encodes an essential erythrocyte membrane protein (Crispino et al., 1999) (Fig. 3A). Conversely, GATA1^V205G failed to repress Gata2, which is normally strongly repressed by GATA1 during erythroid differentiation (Fig. 3B). Fog1 was found to be a FOG-independent GATA1 target gene, as GATA1^V205G activated its expression as efficiently as GATA1 (Fig. 3C). Later studies revealed that GATA1 represses a small subset of target genes in a FOG-independent manner (Johnson et al., 2007) (Fig. 3D). Together, these data indicate that FOG1 is required for both activation and repression of subsets of GATA1 target genes, but GATA1 also regulates some genes independently of FOG1.

To closely examine the role of GATA1 and FOG1 in regulating β-globin gene expression, Vakoc and colleagues measured the proximity of the β-globin locus control region (LCR) to the β-globin promoter using the chromatin conformation capture (3C) method in G1E cells expressing GATA1 or GATA1^V205M (Vakoc et al., 2005). They found that wild-type GATA1, but not GATA1^V205M, induces a chromatin loop that brings the LCR and the promoter into close proximity (Fig. 3A). Thus, interaction between GATA1 and FOG1 is required for chromatin looping of some long-range enhancers to promoters (Vakoc et al., 2005), although the protein-protein interactions underlying chromatin looping are still not well understood. Even though only a few genes have been studied in such detail as the β-globin locus, it is likely that a similar mechanism applies to the FOG/GATA-mediated regulation of other targets via distal enhancers.

As the first known target of GATA1-mediated transcriptional repression, regulation of the Gata2 gene has been extensively studied. Using the G1E system, Grass and colleagues were the first to reveal that GATA1 directly represses Gata2 through enhancer elements (Grass et al., 2003; Grass et al., 2006). These elements are bound by GATA2 and confer positive autoregulation in proliferating erythroid progenitor cells. Upon differentiation, GATA2 is displaced from the enhancers by GATA1, the histone acetylase CBP (CREB binding protein) is removed, and locus-wide histone acetylation decreases (Fig. 3B). This mechanism of replacement of GATA2 by GATA1 as a means to alter transcriptional regulation of a shared target gene has been termed the ‘GATA switch’.

A second gene shown to be a target of the GATA switch is the cell surface receptor of stem cell factor, c-Kit (Rylski et al., 2003). Kit expression is high in hematopoietic progenitor cells, where it is thought to be necessary for maintenance of the undifferentiated, proliferative state. Upon induction of erythroid differentiation, Kit expression is sharply downregulated. Meticulous studies of the regulation of Kit using chromatin immunoprecipitation (ChIP) and 3C experiments in G1E cells revealed that a GATA2-bound distal enhancer element is in close proximity to the active Kit gene promoter as a result of chromatin looping (Jing et al., 2008). Upon GATA1 activation, GATA2 is displaced from the enhancer, the looping conformation is altered, and the gene is repressed. GATA1^V205M fails to alter the chromatin loops or repress the gene, indicating that recruitment of FOG1 by GATA1 is required for this activity (Fig. 3B). Together, these studies indicate that FOG1 is crucial for GATA1 to alter chromatin structure and transcriptional output at GATA switch target genes.

Although it is clear that the replacement of GATA2 by GATA1 on chromatin causes a drastic change in expression at many genes, the mechanism by which these highly related GATA factors confer such drastically different transcriptional outcomes is not well understood. Clearly, FOG1 plays a role in the repressive activity of GATA1, but GATA2 has also been shown to recruit FOG1 to these genes, indicating that the recruitment of FOG1 is unlikely to be the crucial difference. Perhaps GATA1-FOG1 complexes interact with a different set of co-regulators than GATA2-FOG1 complexes, creating different chromatin conformations that alter the transcriptional activity at the locus.

Initial analysis of FOG1 using transcriptional reporter assays indicated that it repressed GATA1-mediated activation of a transcriptional reporter (Fox et al., 1999). This activity was mapped to the newly discovered CtBP interaction domain between zinc fingers 6 and 7. However, later studies found that this domain was not essential for in vivo function, as mice carrying a mutation within the domain were born at the expected Mendelian ratios, developed normally, and had normal red blood cells (Katz et al., 2002).

Independently, Svensson and colleagues mapped the repression of GATA4 activity at cardiomyocyte promoters to the extreme N-terminus of FOG2 and not to the conserved CtBP-binding domain (Svensson et al., 2000a). In search of the co-regulators responsible for this regulation, conservation analysis revealed that the first 12 amino acids of FOG1 and FOG2 differ by a single residue, suggesting that this region might include an important domain. Blobel and colleagues performed GST pulldown experiments with N-terminal constructs of FOG1 as bait (Hong et al., 2005) and found that components of the NuRD complex bound to this region. NuRD is a 2 MDa complex composed of at least eight subunits, including the ATPase Mi2α/β, histone deacetylases HDAC1 and HDAC2, retinoblastoma-binding proteins RbAp46 and RbAp48, metastasis-associated proteins (MTA) 1, 2 or 3, methyl-CpG binding domain protein 3 (MBD3) and p66 (Tong et al., 1998; Wade et al., 1998; Zhang et al., 1998). The combination of the ATPase activity of Mi2α/β and the histone deacetylase activity of the HDACs allows this complex to alter both histone location and acetylation (Tong et al., 1998). GST pulldown and co-immunoprecipitation experiments revealed that FOG1 and FOG2 interact with a highly conserved domain at the C-terminus of MTA1 (Roche et al., 2008). A recent study revealed through a crystal structure that FOG1 can also interact directly with RbAp48, which can bind both proteins simultaneously via separate domains (Lejon et al., 2011).

Having established that FOG1 could physically interact with the NuRD complex, Blobel and colleagues went on to show by ChIP that FOG1 could recruit NuRD to regulatory elements of GATA1/FOG1-repressed genes (Hong et al., 2005). Point mutations within the conserved NuRD interaction domain abrogated the FOG-NuRD interaction and blocked the GATA1/FOG1-mediated gene repression. Notably, the mutant FOG1 was incapable of repressing Gata2 expression. Analysis of mice carrying a FOG1 mutation that abrogates the NuRD interaction, as discussed above, further confirmed the importance of this interaction for FOG1 function (Gao et al., 2010; Gregory et al., 2010). Importantly, these studies revealed that FOG1 recruits NuRD not only to GATA1/FOG1-repressed genes but also to GATA1/FOG1-activated genes, including the LCR of the β-globin locus (Miccio et al., 2010). Further studies indicated that recruitment of NuRD is indeed required for maximal expression of the adult globin genes (Miccio and Blobel, 2010). This was surprising because NuRD had not previously been associated with gene activation. The mechanism by which NuRD participates in gene activation remains poorly understood.

Analysis of the megakaryocyte-erythrocyte progenitors (MEPs) from non-NuRD-interacting FOG1 mutants showed dramatically reduced erythroid and megakaryocytic potential (Gregory et al., 2010). Instead, the mutant MEPs produced myeloid and mast cell colonies at much higher rates than MEPs from control mice. Global gene expression analysis of the mutant MEPs revealed increased expression of many mast cell-specific genes. This finding indicates that, without the FOG-NuRD interaction, GATA1 is allowed to activate mast cell genes despite the presence of FOG1, which normally inhibits mast cell fate. Thus, FOG1 requires interaction with the NuRD repressor for its intricate regulation of cell fate in the mammalian hematopoietic system and likely in many other tissues and species.

In order to determine whether FOG1 regulates GATA1 chromatin occupancy, Pal and colleagues expressed the GATA1-ER fusion protein in a hematopoietic cell line derived from murine Fog1^−/− cells and found that GATA1 failed to occupy a subset of known GATA1 binding sites in this context (Pal et al., 2004). Simultaneously, Letting and colleagues performed ChIP studies of the GATA1^V205M-ER fusion protein at known GATA1 binding sites in G1E cells (Letting et al., 2004). Their analysis revealed that the GATA1^V205M mutant was deficient for binding at a similar subset of sites, including the β-globin promoter and DNase hypersensitive site 2 (HS2) of the LCR, a well-characterized essential regulatory region. By contrast, HS3 of the LCR was bound normally by GATA1^V205M. These data indicated that FOG1 is likely to facilitate GATA1 chromatin occupancy in a context-dependent manner and that this regulation is required for activation of some erythroid genes (Fig. 3A). To study this activity genome-wide, we recently completed ChIP-Seq experiments for GATA1 and GATA1^V205G in a GATA1-deficient megakaryocyte/erythroid cell line termed G1ME (Chlon et al., 2012). Our analysis revealed that not only does FOG1 facilitate GATA1 occupancy at some sites, but that it also prohibits occupancy at other sites. Interestingly, a subset of the FOG-prohibited GATA1 binding sites lie at mast cell-specific genes, which FOG is known to repress (Maeda et al., 2006). Thus, FOG1 appears to repress GATA1-mediated activation of mast cell genes by prohibiting its occupancy at mast cell-specific gene regulatory elements (Fig. 3E). Collectively, these studies reveal that FOG1 intricately regulates GATA1 chromatin occupancy, although the molecular mechanism by which it accomplishes this regulation remains unclear.

It is important to note that several GATA1/2 ChIP-Seq studies have recently been completed in erythroid and megakaryocytic cells (Cheng et al., 2009; Fujiwara et al., 2009; Yu et al., 2009; Tijssen et al., 2011; Wu et al., 2011; Doré et al., 2012). These studies have revealed that GATA1/2 predominantly function from enhancer elements and bind relatively few promoters. Furthermore, they imply that GATA1 and GATA2 share many target genes and binding sites, suggesting that the GATA switch is a common transcriptional mechanism that is employed at hundreds of developmentally regulated target genes.

In summary, FOG proteins regulate GATA factor transcriptional activity in diverse ways. Extensive work using the GATA1^V205G mutant revealed that FOG1 is required for both activation and repression of distinct GATA1 target genes, and that FOG1 accomplishes this regulation through control of chromatin looping, GATA1 chromatin occupancy and the recruitment of co-regulator complexes.

In addition to FOG1, GATA1 and GATA2 have been found to interact with many different transcriptional regulators. For example, GATA1 interacts with the CBP/p300 acetylase complex, and this interaction is required for erythroid differentiation (Blobel et al., 1998). CBP/p300 acetylates GATA1 at lysine-rich regions C-terminal to both zinc fingers (Hung et al., 1999), and this acetylation is required for GATA1 chromatin occupancy (Lamonica et al., 2006). Recently, the bromodomain protein Brd3 was found to associate with GATA1 in an acetylation-dependent manner and to facilitate GATA1 chromatin occupancy at both activated and repressed target genes (Lamonica et al., 2011).

Another partner of GATA1 is SCL (Tal1), a bHLH transcription factor that is required for normal hematopoietic stem cell production and for terminal differentiation of erythrocytes, megakaryocytes and mast cells (Hall et al., 2003; Mikkola et al., 2003; McCormack et al., 2006; Salmon et al., 2007). In erythrocytes, a pentameric complex composed of SCL, E2A (TCF3), the LIM domain protein LMO2, LDB1 and GATA1 assembles on tandem GATA E-box motifs to cooperatively regulate transcription of many crucial erythroid genes (Wadman et al., 1997) (Fig. 3A,C). SCL complexes can function as both activators and repressors, the repressive function being provided at least in part by the co-repressor ETO2 (CBFA2T3) (Schuh et al., 2005; Meier et al., 2006). A recent study of the cooperativity between GATA1 and SCL in erythropoiesis found that the entire pentameric complex assembles on all sites where GATA1 activates gene expression, even sites without an E-box (Tripic et al., 2009). By contrast, SCL is not recruited at sites where GATA1 acts as a repressor. The role of FOG1 in the interaction between GATA1 and the SCL complex is not well understood. One study reported that GATA1 cannot associate with FOG1 and SCL simultaneously (Rodriguez et al., 2005). However, other analyses have found that FOG1 and SCL can occupy the same sites on chromatin and that GATA1 is physically capable of simultaneously interacting with both LMO2 and FOG1 via separate sites within its N-finger (Letting et al., 2004; Wilkinson-White et al., 2011). Since FOG1 and SCL are required for activation of many of the same GATA1 target genes, it is likely that FOG1 and SCL communicate in some way to alter GATA1 transcriptional output.

In this review, we have highlighted the diverse mechanisms by which FOG and GATA family proteins collaborate to control the development of tissues throughout metazoans. FOG proteins from Drosophila to humans are highly conserved in sequence, structure and function. In all species, FOG proteins both assist and antagonize GATA factor function in a context-dependent manner. Transcriptional studies in mammalian blood cells have revealed intricate mechanisms by which FOG proteins regulate GATA factor function to accomplish such diverse effects on transcriptional output and cell fate. It is likely that many of these transcriptional mechanisms are conserved in other tissues and species.

The development of hemocytes in Drosophila and blood in mammals provide an illustrative example of conservation of FOG protein function. In Drosophila, Ush and Srp collaborate in the formation of plasmocytes, but Ush inhibits Srp in the crystal cell lineage (Evans et al., 2003). Similarly, GATA1 and FOG1 collaborate in the formation of red blood cells and megakaryocytes, but FOG1 inhibits GATA1 in mast cells. The inhibition of mast cell fate in mammals has been mapped to the NuRD interaction domain (Gregory et al., 2010). However, the NuRD interaction domain has not evolved in the Drosophila FOG proteins. Instead, the inhibition of crystal cell fate by Ush has been mapped to the CtBP interaction domain, which has no known in vivo function in mammalian blood (Fossett et al., 2001; Fossett et al., 2003). Thus, it appears that FOG proteins in mammals have separately evolved a similar regulatory function to Ush in Drosophila through a newly evolved domain that recruits a different co-repressor complex.

Although the role of GATA and FOG proteins in mammalian blood and heart development has been extensively studied, there are many other tissues in which FOG proteins are expressed but comparatively little known about their role. For example, FOG2 is highly expressed in the central nervous system in late stage mouse embryos, but little is known about its function there (Tevosian et al., 1999). The generation of conditional knockout mouse models should allow detailed study of the role of FOG2 in this tissue at this particular stage of development. Moreover, since zebrafish contain two homologs of FOG2 that are expressed predominantly in the brain (Walton et al., 2006), a putative neuronal function is likely to be conserved.

Mechanistically, little is known about how FOG proteins achieve context-dependent effects on GATA factor function. It is likely that GATA and FOG proteins interact with other transcription factors and co-regulators which, when present in a particular cell type or at a specific genomic locus, can alter the behavior of GATA-FOG pairs. An example of such regulation might be provided by the SCL pentameric complex, which is present only at GATA1-activated genes. Perhaps the pentameric complex alters the interaction between GATA and FOG or among FOG and other co-regulators such that a transcriptional activation signal is induced and repressive regulation is inhibited.

Several studies have also revealed a cooperative relationship between GATA1/FOG1 and several Ets family transcription factors in megakaryocyte-specific gene expression (Wang et al., 2002; Eisbacher et al., 2003; Pang et al., 2006; Doré et al., 2012). The presence of an Ets protein at gene regulatory elements is required for GATA1/2 and FOG1 to properly regulate genes of this lineage, and thus the Ets proteins are likely to contribute to GATA/FOG function in some way. Moreover, the regulation by FOG1 of GATA1 chromatin occupancy might be controlled by the presence of various chromatin-bound factors, such as Ets proteins. In cases in which FOG1 facilitates chromatin occupancy, it is likely that FOG1 interacts with a co-regulator that stabilizes GATA-FOG occupancy. By contrast, cases in which FOG1 prohibits GATA1 occupancy are likely to involve the presence of a co-regulator that repels GATA-FOG pairs but allows ‘FOG-less’ GATA proteins to bind. The identity or even the existence of these co-regulators is yet to be determined, and thus we lack a full understanding of the transcriptional mechanisms by which these proteins control development.

With the recent advent of next-generation sequencing technology, new insights into the way that transcription factors regulate gene expression in the context of the cell have been rapidly realized. For example, ChIP-Seq studies have defined the chromatin occupancy patterns of transcription factors and modified histones genome-wide for a wide variety of cell types and at multiple stages of differentiation. These data will undoubtedly shed light on the altered regulation of genomic loci in different cell types and provide clues as to how GATA and FOG factors can accomplish context-dependent regulation. Gene expression and chromatin occupancy analyses at the single-cell level might be needed to fully understand how these complex transcriptional mechanisms are differentially realized, even in highly similar cell types.

FOG proteins are unique among transcriptional regulators in that they serve as dedicated co-factors for a small family of transcription factors and do not bind DNA. The conservation of the GATA-FOG families across metazoan evolution emphasizes the power of this mode of gene regulation. The complex interactions that occur between GATA and FOG proteins, which we have illustrated here, provide powerful yet adaptable modes of gene regulation that allow these proteins to control the diversity of gene expression programs required of complex organisms.

We thank Louis Doré for helpful comments on the manuscript.