FOG‐1 recruits the NuRD repressor complex to mediate transcriptional repression by GATA‐1

Prior work indicated that FOG‐1 assists GATA‐1 in associating with select GATA elements in vivo (Letting et al, 2004; Pal et al, 2004). Using quantitative ChIP, we examined whether FOG‐1 is required for GATA‐1 occupancy and histone deacetylation at two genes that are repressed by GATA‐1, namely GATA‐2 (Grass et al, 2003) and c‐kit (R Kapur and M Weiss, unpublished). To this end, we used the GATA‐1‐deficient cell line G1E stably expressing an estradiol‐dependent form of wild‐type GATA‐1 (GATA‐1‐ER) or GATA‐1‐ER with a mutation that impairs FOG‐1 binding (GATA‐1(V205M)‐ER). Upon treatment with estradiol, GATA‐1‐ER‐expressing cells displayed a robust reduction in the expression of both GATA‐2 and c‐kit mRNA as measured by real‐time RT–PCR (Figure 1A). In contrast, cells expressing GATA‐1(V205M)‐ER were impaired in their ability to inhibit expression of GATA‐2 and c‐kit in response to estradiol treatment (Figure 1A). Therefore, the repression of both genes by GATA‐1 requires FOG‐1 binding.

Lack of transcriptional repression by GATA‐1(V205M)‐ER was accompanied by failure to substantially lower acetylation of histones H3 and H4 at the c‐kit and GATA‐2 genes (Figure 1B, and data not shown). Quantitative ChIP analysis with anti‐GATA‐1 antibodies shows that GATA‐1‐ER binds inducibly to a regulatory element at the GATA‐2 gene 2.8 kb upstream of the 1S promoter (Figure 1C), in accord with previous results (Grass et al, 2003). Moreover, GATA‐1‐ER bound well to a region at the c‐kit locus 5 kb downstream of the transcription start site that is conserved and contains three GATA consensus binding sites (R Kapur and M Weiss, unpublished) but not to a control region (Figure 1C, and data not shown). These results confirm that both GATA‐2 and c‐kit are direct GATA‐1 targets. Notably, histone deacetylation at the GATA‐2 gene occurred also at sites not occupied by GATA‐1, such as the 1S promoter (Supplementary Figure 1A), suggesting that histone deacetylation once initiated might spread across the GATA‐2 locus independent of GATA‐1.

We next examined whether FOG‐1 binding by GATA‐1 is required for GATA‐1 occupancy at the GATA‐2 and c‐kit genes. As measured by ChIP, levels of GATA‐1(V205M)‐ER at the GATA‐2 and c‐kit loci were comparable if not slightly higher than that observed with GATA‐1‐ER (Figure 1C), suggesting that FOG‐1 is not required for GATA‐1 occupancy at these sites. Similar results were obtained with a second independently derived GATA‐1(V205M)‐ER‐expressing cell line (data not shown). As an additional control, GATA‐1 occupancy was determined at the β‐major globin gene promoter. Consistent with our previous findings (Letting et al, 2004), high levels of GATA‐1‐ER were detected at the β‐globin promoter whereas GATA‐1(V205M)‐ER was virtually absent (Supplementary Figure 1B). ChIP experiments were repeated with antibodies against the ER portion of GATA‐ER and yielded essentially the same results (data not shown). We next examined FOG‐1 occupancy by ChIP. Treatment of G1E cells expressing GATA‐1‐ER augmented the amounts of FOG‐1 present at the c‐kit locus and, as control, at the β‐globin gene promoter (Figure 1D). In contrast, in G1E cells expressing GATA‐1(V205M)‐ER, FOG‐1 recruitment to the c‐kit and β‐globin loci was diminished (Figure 1D).

Together, these results confirm that FOG‐1 is essential for GATA‐1 occupancy at some sites but appears largely dispensable at others such as the c‐kit and GATA‐2 loci. Since GATA‐1(V205M)‐ER can associate with regulatory elements at the GATA‐2 and c‐kit genes but fails to inhibit their expression, this suggests that at these sites FOG‐1 functions by recruiting corepressor molecules. Given the substantial decreases in histone acetylation, these corepressors are expected to harbor one or more histone deacetylases.

Previous work identified the corepressor CtBP‐2 as FOG‐1 binding partner (Fox et al, 1999). Mutations that abrogate CtBP‐2 binding reduce FOG‐1's inhibitory activity in transfection assays. Moreover, the ability of FOG‐2 to inhibit erythropoiesis in Xenopus embryos depends on its ability to bind CtBP (Deconinck et al, 2000). Therefore, it was surprising that engineered mice bearing a FOG‐1 mutant defective for CtBP binding displayed normal erythropoiesis even under conditions of hematopoietic stress (Katz et al, 2002). This suggested that other domains in FOG‐1 might provide interactions with corepressors, thus compensating for the loss of CtBP binding. The N‐terminus of FOG‐1 contains a stretch of 14 amino acids that is identical between mouse and human FOG‐1 and differs by only one amino acid between FOG‐1 and FOG‐2 proteins (Svensson et al, 2000). Deletion of this domain impaired the ability of FOG‐1 to repress GATA‐4 activity in transient reporter assays. When fused to the DNA‐binding domain of GAL4, the first 45 amino acids of FOG‐2 (Svensson et al, 2000) and FOG‐1 (see below) are potent inhibitors of transcription, indicating that this domain can function as independent repressor module.

To identify corepressors that mediate repression by FOG‐1, we generated a fusion protein consisting of amino acids 1–45 of FOG‐1 fused to GST. GST‐FOG‐1(45) was bound to glutathione‐agarose beads and incubated with nuclear extracts of the erythroid cell line MEL. Bound proteins were washed extensively and analyzed for histone deacetylase activity by incubation with [³H]acetate‐labeled cellular histone molecules and measuring the release of radioactive acetate. The results show that GST‐FOG‐1(45) but not GST associates with histone deacetylase activity (Supplementary Figure 2). This activity was inhibited by the addition of the deacetylase inhibitor sodium butyrate, indicating that the reduced levels of histone acetylation were not simply the consequence of protein degradation.

GST‐FOG‐1(45)‐bound proteins were fractionated by SDS–PAGE and stained with colloidal blue. Several bands were identified that bound to GST‐FOG‐1(45) but not to GST alone (Figure 2). All proteins were retained following stringent washes with NaCl concentrations up to 650 mM. Bands were excised and proteins identified by tandem mass spectrometry. All proteins identified corresponded to known members of the NuRD complex, including Mi‐2β (also called CHD4), RbAp46, RbAp48, MTA‐1, MTA‐2, p66, HDAC1 and HDAC2 (Tong et al, 1998; Wade et al, 1998; Xue et al, 1998; Zhang et al, 1998; for review see Bowen et al, 2004). Among 33 peptides identified as Mi‐2β, 26 were specific for Mi‐2β and seven were in a region shared between Mi‐2β and Mi‐2α (CHD3). We did not detect any peptides unique to Mi‐2α. However, we cannot entirely rule out the presence of low amounts of Mi‐2α in our purification. Previous reports showed that Mi‐2α and Mi‐2β can coexist within the same complex (Tong et al, 1998; Xue et al, 1998). While 18 and 13 peptides were found to represent MTA‐1 and MTA‐2, respectively, only a single peptide corresponding to MTA‐3 was identified. This suggests that while MTA‐3 is present in small amounts, it is likely substoichiometric with regard to the other NuRD components. In accord with this interpretation, Western blotting with anti‐MTA‐3 antibodies failed to detect appreciable amounts of MTA‐3 (data not shown). One NuRD component, MBD3, was not identified by mass spectroscopy since it comigrates with GST‐FOG‐1(45), but its presence was confirmed by Western blotting (see below). Components of other deacetylase‐containing complexes such as Sin3, N‐CoR and SMRT were not detected (data not shown). Moreover, the methylated DNA‐binding protein MBD‐2 was not detected by Western blot, suggesting that FOG‐1 recruits NuRD but not the MeCP‐1 complex, which contains MBD2 in addition to NuRD (Ng et al, 1999; Zhang et al, 1999; Feng and Zhang, 2001). These data show that a single‐step high‐stringency purification yielded all of the previously described components of NuRD. The NuRD component Mi‐2β has ATPase activity that is required for nucleosome remodeling. Together with the presence within NuRD of the histone deacetylases HDAC1 and HDAC2, this suggests that FOG‐1 mediates GATA‐1‐induced gene repression by histone deacetylation and nucleosome remodeling.

To determine whether endogenous FOG‐1 associates stably with NuRD, we performed conventional protein purification using MEL cell nuclear extracts. The purification strategy and FOG‐1 elution profiles are summarized in Figure 3A with details described in Materials and methods. Fractions from the final purification over a Superose6 column were analyzed by Western blot (Figure 3B). Peak levels of FOG‐1 were found in fractions 22 and 24, which corresponds to a molecular weight of ∼1.2 to ∼1.3 MDa. This size is similar to that described previously for NuRD (Xue et al, 1998; Zhang et al, 1998). Importantly, FOG‐1 and members of the NuRD complex displayed virtually identical elution profiles (Figure 3B), indicating that the FOG‐1–NuRD association is stable over multiple steps of purification.

To examine whether FOG‐1 associates with NuRD in vivo, MEL cell nuclear extracts were immunoprecipitated with antibodies against the NuRD component MTA‐2 followed by Western blotting with antibodies against FOG‐1 and other NuRD proteins. Significant amounts of FOG‐1 co‐precipitated with anti‐MTA‐2 but not control antibodies (Figure 4A). As expected, other NuRD components but not Sin3A co‐precipitated efficiently. Anti‐HDAC2 and anti‐RbAp48 but not control antibodies precipitated comparable amounts of FOG‐1 (Figure 4B), suggesting that FOG‐1 associates with the intact NuRD complex. Notably, FOG‐1 migrated as a doublet, and it appeared that the upper band preferentially associated with NuRD. At present, it is unclear what accounts for the difference in mobility between these two forms of FOG‐1.

To examine whether NuRD binding depends on the N‐terminal repression domain of FOG‐1, plasmids expressing HA‐tagged forms of full‐length FOG‐1 and FOG‐1 lacking the N‐terminal 45 amino acids (HA‐FOG‐1(Δ45)) were transfected into COS cells. Nuclear extracts were immunoprecipitated with anti‐HA antibodies followed by Western blotting against endogenous HDAC2. Full‐length HA‐FOG‐1 but not HA‐FOG‐1(Δ45) immunoprecipitated the NuRD component HDAC2 (Figure 4C, compare lanes 3 and 8, lower panel) although both proteins were expressed at similar levels (lanes 1 and 6). In the converse experiment, the same extracts were precipitated with antibodies against MTA‐2. Consistent with the above results, only full‐length HA‐FOG‐1 but not HA‐FOG‐1(Δ45) associated with MTA‐2 (upper panel, lanes 5 and 10). Thus, NuRD binding is dependent on the N‐terminus of FOG‐1.

The components of NuRD that confer binding to FOG‐1 were determined by conducting GST pulldown experiments. GST‐FOG‐1(45) was incubated with in vitro‐translated ³⁵S‐labeled NuRD proteins and retained material analyzed by SDS–PAGE and autoradiography. MTA‐1 and RbAp48 showed the strongest binding to GST‐FOG‐1(45) relative to the input signal (Figure 4D). In addition, MTA‐2 and RbAp46 displayed significant binding albeit at reduced levels when compared to MTA‐1 and RbAp48. This indicates that the association between NuRD and FOG‐1 can be mediated by at least two classes of molecules. However, it remains possible that other NuRD subunits contribute to the stability of this complex.

A series of point mutations was generated by substituting conserved, charged amino acids with alanines or glycines up to residue 29 in the context of GST‐FOG‐1(45) (Figure 5A). GST‐FOG‐1 proteins were incubated with in vitro‐translated ³⁵S‐labeled MTA‐1, and bound protein analyzed by autoradiography. Two mutations (R4G and K5A) substantially impaired MTA‐1 binding (Figure 5B). Substitution of residues 3 and 10 showed a moderate reduction in MTA‐1 binding, whereas the remaining mutations had no effect. Similar binding profiles were observed for MTA‐2, RbAp48 and RbAp46 (Supplementary Figure 3). However, it appeared that binding of RbAp48 and RbAp46 was more sensitive to the R10G mutations of FOG‐1. Finally, comparable results were obtained when GST‐FOG‐1 proteins were incubated with crude MEL cell nuclear extracts and analyzed for NuRD binding by Western blot with antibodies against several NuRD proteins (data not shown). Together, these results show that NuRD binding by FOG‐1 depends on a small, defined motif at the N‐terminus of FOG‐1.

To determine whether NuRD binding correlates with transcriptional repression by FOG‐1, mutants of FOG‐1(45) were fused to the DNA‐binding domain of GAL4. Transcriptional repression was determined in NIH 3T3 cells by coexpressing these constructs with a luciferase reporter gene driven by the thymidine kinase promoter and five GAL4‐binding sites. A Renilla luciferase construct was cotransfected to monitor transfection efficiency. Wild‐type GAL4‐FOG‐1(45) repressed transcription in a dose‐dependent manner (Figure 5C). In contrast, R4G and K5A substitutions strongly reduced repressive activity (Figure 5C). Mutation at residue 3 partly impaired repression, while all other constructs were unimpaired (Figure 5C). The tight correlation between NuRD binding and transcriptional repression further supports that NuRD serves as a FOG‐1 corepressor.

Interrogation of the NCBI database for motifs similar to the N‐terminal 16 amino acids of murine FOG‐1 identified numerous proteins including EBFAZ, Evi3, Bcl11b, SALL‐1 (Figure 5D), and other members of the SALL family (not shown). All of these are transcriptional repressors (Tsai and Reed, 1997; Avram et al, 2000; Netzer et al, 2001; Kiefer et al, 2002) and harbor this domain at their N‐terminus. Thus, this work predicts that diverse transcriptional repressors recruit NuRD through a common motif.

To examine whether NuRD is present at GATA‐1‐repressed genes in vivo, ChIP assays were performed with antibodies against the NuRD protein Mi‐2β in GATA‐1‐ER‐expressing cells. Treatment with estradiol led to increased association of Mi‐2β with the GATA‐2 and c‐kit genes but not the GAPDH gene (Figure 6A and B). We also observed the presence of MTA‐2 and HDAC1 at the GATA‐2 locus (Supplementary Figure 4). Notably, MTA‐2 and HDAC1 occupied broader regions than GATA‐1 and FOG‐1 (Supplementary Figure 4), suggesting that NuRD and consequently histone deacetylation might spread along the GATA‐2 gene following the initial recruitment by GATA‐1 and FOG‐1.

To examine the generality of our findings, we examined NuRD occupancy at three additional genes regulated by GATA‐1. The c‐myc and c‐myb genes are repressed by GATA‐1 in a FOG‐1‐dependent manner (Supplementary Figure 5A; Crispino et al, 1999). At both genes we observed by ChIP substantial levels of MTA‐2 and HDAC1 (Supplementary Figure 5B and C). Signals were specific since they were not detected at control regions located in cis. As additional control, we also examined the eosinophil‐specific gene MBP, which is controlled by GATA‐1 (Du et al, 2002) but is not expressed at significant levels in G1E cells before and after differentiation (Welch et al, 2004). By ChIP, we detected only small amounts of MTA‐2 and HDAC1 at this gene, the significance of which remains uncertain. While our results generally agree with those reported in the accompanying paper by Rodriguez et al, there appear to be differences with regard to the relative levels of NuRD proteins found at the c‐myc, c‐myb and MBP genes. These differences might reflect the use of antibodies against distinct components of NuRD, or distinct epitope access during gene repression. Nevertheless, together our results demonstrate the presence of NuRD proteins at all four genes that are actively repressed by GATA‐1 and FOG‐1, suggesting that NuRD is a critical corepressor for GATA‐1/FOG‐1‐mediated gene repression.

To analyze the activity of FOG‐1 mutants without the confounding effects of endogenous FOG‐1, we used G1E cells expressing GATA‐1(V205M)‐ER. Expression of FOG‐1(S706R), which can bind to GATA‐1(V205M), can partially restore erythroid differentiation in these cells (Crispino et al, 1999) (Figure 7A). HA‐tagged FOG‐1(S706R), a derivative lacking the N‐terminal 45 amino acids (HA‐FOG‐1(S706R)Δ45) and one containing the K5A mutation (FOG‐1(S706R)K5A) were introduced into a retroviral vector upstream of an internal ribosomal entry site and GFP. GATA‐1(V205M)‐ER cells were infected and sorted for GFP expression by FACS analysis. Pools of cells were analyzed by anti‐HA Western blotting to confirm comparable expression levels (Figure 7D). Repressive activity of GATA‐1(V205M)‐ER was determined by measuring c‐kit levels via flow cytometry. In the absence of estradiol, approximately 3% of HA‐FOG‐1(S706R)‐expressing cells had low c‐kit levels (Figure 7B and C). Upon estradiol treatment, this number increased to ∼13% (Figure 7B and C). This result is significant since only ∼10–15% of cells expressing HA‐FOG‐1(S706R) differentiate as determined by staining for hemoglobin (data not shown), consistent with previous observations (Crispino et al, 1999). In contrast, 2.1% of untreated HA‐FOG‐1(S706R)Δ45 cells displayed low c‐kit levels. Estradiol treatment raised this number to ∼3.8%, which is comparable to estradiol‐treated cells lacking FOG‐1(S706R) (data not shown).

The FOG‐1(S706R)K5A construct produced an intermediate response, with levels of c‐kit‐low cells increasing from 3.9 to 7.4% upon estradiol treatment (Figure 7B and C). While the K5A mutation clearly impairs FOG‐1 function, remaining repressive activity is likely due to residual interaction with NuRD. This is consistent with detectable binding of MTA‐2 to GST‐FOG‐K5A (Supplementary Figure 3).

It is important to note that the rescue efficiency of GATA‐1(V205M) by FOG‐1(S706R) is moderate in part because the affinity between them is not as high as that between wild‐type GATA‐1 and FOG‐1. Moreover, while GATA‐1(V205M) is substantially impaired for FOG‐1 binding, a residual interaction remains (Nichols et al, 2000) that might account for the repressive activity of GATA‐1(V205M)‐ER after prolonged estradiol treatment (data not shown). Thus, the ‘window’ for FOG‐1(S706R)‐mediated rescue is somewhat narrow. Nevertheless, the number of cells showing c‐kit repression by HA‐FOG‐1(S706R) parallels that of globin induction, indicating that gene repression by GATA‐1(V205M)‐ER was partially restored. Taken together, these findings show that gene repression by GATA‐1 requires the FOG‐1–NuRD interaction.

GST‐FOG‐1(45) and GAL4‐FOG‐1(45) were generated by PCR and inserted into pGEX‐2TK (Amersham Biosciences) and pcDNA3‐Gal4 (gift from M Poncz), respectively. pcDNA3‐GAL4 contains amino acids 1–147 of GAL4. pGEM7Zf(+)‐rMTA1 and pcDNA3CF‐mMTA2 were gifts from W‐M Yang (Yao and Yang, 2003). pCMV5‐MTA‐3 and pcDNA3β‐Mi‐2β were gifts from P Wade and Y Zhang, respectively (Feng and Zhang, 2001; Fujita et al, 2004). pcDNA3‐HDAC1 and CMV‐RbAp48 were gifts from M Lazar and N Barlev, respectively. pcDNA3‐RbAp46 and pcDNA3‐p66 were generated by RT–PCR from murine mRNA and inserted between the BamH1 and EcoR1 sites of pcDNA3 (Invitrogen). pBluescript‐HDAC2 and pSPORT‐MBD3 were from American Type Culture Collection (ATCC). pcDNA3‐HA‐FOG‐1 was generated by inserting the FOG‐1 cDNA between the BamH1 and EcoR1 sites of pcDNA3 with a PCR‐generated HA tag at the N‐terminus. pcDNA3‐HA‐FOG‐1(Δ45) was generated by replacing the 5′ BamH1 and Kpn1 fragment with the corresponding fragment lacking the first 45 amino acids. Reporter vector G5‐TK‐luciferase was a gift from W‐M Yang (Yao and Yang, 2003).

A 20 μg portion GST‐FOG‐1 was incubated with 3 mg of MEL nuclear extracts overnight in buffer containing 150 mM NaCl, 50 mM Tris–HCl (pH 7.5), 0.5% Igepal (Sigma), protease inhibitor cocktail (P 8340, Sigma, 1:500) and 1 mM DTT. Bound proteins were washed twice in the above buffer with 350 mM NaCl, three times with NaCl raised to 650 mM, followed by an additional wash in 350 mM NaCl, separated by SDS–PAGE and stained with colloidal blue (Invitrogen). Proteins were identified by tandem mass spectroscopy analysis on an LCQ Deca XP (ThermoFinnigan, CA) mass spectrometer coupled online with a mini‐C18 reversed‐phase capillary high‐performance liquid chromatography column using electrospray ionization (ESI) interphase at the Proteomics Core Facility of the Genomics Institute, University of Pennsylvania.

G1E cells were maintained as described (Weiss et al, 1997). 3T3 or COS cells were transfected with Lipofectamine (Invitrogen). Immunoprecipitations and Western blots were performed with antibodies against Mi‐2β (PHD‐P, gift from W Wang), MTA‐1 (sc‐9445), MTA‐2 (sc‐9447), RbAp46 (sc‐8272), HDAC1 (sc‐7872), HDAC2 (sc‐7899), MBD3 (sc‐9402), FOG‐1 (sc‐9361) and Gal4‐DBD (sc‐510) (Santa Cruz Biotechnology Inc.), p66 (07‐365) and RbAp48 (05‐524) (Upstate Biotechnology Inc.) and MTA‐3 (gift from P Wade).

ChIP assays were performed as described (Letting et al, 2003). Antibodies were against GATA‐1 (sc‐265, Santa Cruz), FOG‐1 (sc‐9361, Santa Cruz), acetyl‐H3 and acetyl‐H4 (Upstate Biotechnology Inc.). Mi‐2β antibodies were a gift from K Georgopoulos. MTA‐2 antibodies used for ChIP were from Santa Cruz (sc‐9447). All primers are listed in Supplementary data.

³H‐labeled acetylated histones from HeLa cells were incubated with GST‐FOG‐1(45)‐bound proteins in 400 μl HDAC buffer (150 mM NaCl, 20 mM Tris–HCl (pH 8.0), 10% glycerol). Where indicated, 125 mM sodium butyrate was added 30 min prior to adding ³H‐labeled acetylated histones. Reactions were stopped with 100 μl 1 M HCl and 0.16 M acetic acid. Liberated [³H]acetate was extracted with 600 μl ethyl acetate and measured by scintillation counting.

Vectors with HA‐FOG‐1, HA‐FOG‐1(S706R), HA‐FOG‐1Δ45 or HA‐FOG‐1(S706R)Δ45 were generated by inserting HA‐tagged FOG‐1 constructs into MSCV MIGR1‐GFP (gift from W Pear). Virus was prepared using a transient transfection system (Pear et al, 1993). At 48 h after infection, GFP‐positive cells were isolated by FACS. A total of 10⁶ cells treated with 0.1 μM estradiol for 24 h were reacted with 0.5 μl phycoerythrin‐conjugated anti‐mouse CD117 (c‐kit) monoclonal antibodies (553869, BD Biosciences) and analyzed by two‐color flow cytometry.

See Supplementary data.

Supplementary data are available at The EMBO Journal Online.