Unanticipated Repression Function Linked to Erythroid Krüppel-Like Factor

Abstract

The erythroid cell-specific transcription factor erythroid Krüppel-like factor (EKLF) is an important activator of β-globin gene expression. It achieves this by binding to the CACCC element at the β-globin promoter via its zinc finger domain. The coactivators CBP and P300 interact with, acetylate, and enhance its activity, helping to explain its role as a transcription activator. Here we show that EKLF can also interact with the corepressors mSin3A and HDAC1 (histone deacetylase 1) through its zinc finger domain. When linked to a GAL4 DNA binding domain, full-length EKLF or its zinc finger domain alone can repress transcription in vivo. This repressive activity can be relieved by the HDAC inhibitor trichostatin A. Although recruitment of EKLF to a promoter is required to show repression, its zinc finger domain cannot bind directly to DNA and repress transcription simultaneously. In addition, the target promoter configuration is important for enabling EKLF to exhibit any repressive activity. These results suggest that EKLF may function in vivo as a transcription repressor and play a previously unsuspected additional role in regulating erythroid gene expression and differentiation.

One of the most important ways for cells to control their biological function is by regulating gene expression. In eukaryotic cells, DNA associates closely with histones and is folded into chromatin in a structure that makes genes inaccessible to the transcription machinery. This has led to the idea that transcriptionally active genes are in an “open” chromatin structure and transcriptionally inactive genes are in a “closed” chromatin structure. Histone acetylation and deacetylation can alter the interaction between DNA and histones and thus can regulate gene expression. Histone acetylases interact with a variety of enhancer-binding proteins and are associated with gene activation, whereas histone deacetylases (HDACs) interact with DNA-binding repressors or transcriptional corepressors and are associated with gene repression (36).

In vertebrates, erythropoiesis is regulated temporally and spatially during development. Each globin gene in the β-like globin locus is expressed at different times and locations as development proceeds (35), a phenomenon called “switching.” In humans, the β-like globin locus contains five genes (5′-ɛ-Gγ-Aγ-δ-β-3′). The earliest gene expressed is ɛ-globin in the yolk sac, followed by a switch in expression to γ-globin (embryonic to fetal) in the fetal liver. The second switch is from γ- to β-globin (fetal to adult) within the bone marrow. Besides their own promoters, these globin genes are also controlled by a far upstream region called the locus control region (LCR), which ensures the high level expression of the β-globin locus (35).

Erythroid Krüppel-like factor (EKLF) is an erythroid-specific transcription factor that is critically required for activating β-globin expression by binding to the CACCC element in the promoter (18, 24, 28). Until now, evidence for its function has been limited to its role in activating β-globin gene expression. However, a number of observations suggest this may be painting an incomplete picture. For example, EKLF expression arises early in development in the yolk sac on day 7.5 (34), which does not parallel the onset of adult β-globin expression. This opens a possibility that EKLF may have another function in the yolk sac. Transgenic studies indicate that EKLF is functional in these primitive cells (10, 37). It has also been shown that correction of the globin chain imbalance that results from the absence of EKLF cannot completely rescue EKLF^−/− animals, implying that EKLF could act on some target genes other than β-globin (27). Lastly, EKLF function in γ- to β-globin switching may be more involved, as EKLF-null embryonic stem cells have a higher βh1 globin level (18) and human globin transgenic mice have a higher γ-globin level after crossing with EKLF-null mice (8, 26). Conversely, overexpression of EKLF results in a premature decrease of transgenic γ-globin (37).

Recently it has been shown that transfected EKLF can be recruited to β-LCR 5′HS3 only in β-globin-expressing MEL cells and only upon linking it to the β-globin promoter. However, recruitment of transfected EKLF to 5′HS2 occurred in both MEL and γ-globin-expressing K562 cells after HS2 linkage to both γ-globin and β-globin promoters (16). These observations indicate that EKLF functional interactions may differ in different cellular environments.

Many transcription factors have been shown to possess both transcription activator and repressor abilities for different genes and in different environments, such as GATA1 (29), P53 (23), Ikaros (14), c-myc (8, 40), and TAL1/SCL (12). It is possible that EKLF may also function as both a transcription activator and a transcription repressor. Indeed, by using coimmunoprecipitation and reporter gene transfection assays, we now show that EKLF can interact with corepressors mSin3A and HDAC1 and function as a transcription repressor.

MATERIALS AND METHODS

Cell lines and plasmid constructs.

K562 cells were grown in RPMI media with 10% fetal bovine serum (42). Cos7 and NIH 3T3 cells were grown in Dulbecco modified Eagle media with 10% fetal bovine serum (6, 42).

Plasmid constructs pSG5/EKLF, pSG5/Zn (pSG5/EKLFΔpro), pSG5/D [pSG5/EKLF (Δ60–195)], pSG5/P (pSG5/EKLFΔZn), pGAL/P (pGAL1-147/EKLFΔZn), pGAL1-147, pC1G3tkCAT, pαLCR-GAL/βglob-CAT, and pμLCR-GAL/βglob-CAT have been previously described (4, 5, 21). Construct pG5tkCAT, in which five GAL4 DNA binding sites are placed in front of the thymidine kinase promoter, was a gift from Jonathan Licht (Mount Sinai School of Medicine). Constructs pCS2/Sin3A (15) and pBJ5/HDAC1 (11) have been described. Construct pBOS/EKLF, in which full-length EKLF is cloned into pEF-BOS (22), was kindly provided by Merlin Crossley (University of Sydney, Australia). Constructs pGAL/EKLF and pGAL/Zn were made by replacing the EKLF proline-rich domain in pGAL/P with full-length EKLF from pSG5/EKLF and the EKLF zinc finger domain from pSG5/Zn, respectively. Construct pSG5/fZn is similar to pSG5/Zn except a Flag tag was placed in frame at the N terminus.

Antibodies.

Anti-EKLF monoclonal antibody 6B3 was made in this laboratory and used for immunoprecipitation assay and Western blot analysis (42). Anti-Flag monoclonal antibodies M5 and M2 were purchased from Kodak-IBI. Anti-myc monoclonal antibody was purchased from the Hybridoma Core Facility at Mt. Sinai School of Medicine. Anti-mSin3A rabbit polyclonal antibody and anti-HDAC1 goat polyclonal antibody were purchased from Santa Cruz Biotechnology, Inc.

Transfection, immunoprecipitation, and Western blot analysis.

For Cos7 cells, 50 to 60% confluent cells in 100-mm dishes were transfected with 10 μg of pCS2/Sin3A or 10 μg of pBJ5/HDAC1 plus 10 μg of pSG5/EKLF, pSG5/P, pSG5/D, or pSG5/fZn by using DMRIE-C (GIBCO/BRL). For K562 cells, 6 × 10⁶ cells were transfected with 6 μg of pCS2/Sin3A and 6 μg of pBOS/EKLF by using DMRIE-C. Forty hours after transfection, cells were harvested and washed twice with phosphate-buffered saline. The cells were incubated on ice with 2 volumes of NE-A buffer (10 mM HEPES [pH 7.9], 10 mM KCl, 1.5 mM MgCl₂, 0.1% NP-40) supplemented with proteinase inhibitors for 10 min. Then a 1/5-volume mixture of 2.5 M NaCl and 50% glycerol was added and incubated for another 30 min. After centrifugation (Beckman TL-100) at 245,000 × g and 4°C for 15 min, the supernatants were mixed with the same volume of HEPES buffer (10 mM, pH 7.9) to dilute the salt concentration. The cell extracts were then subjected to immunoprecipitation with specific antibodies, and the precipitated protein complexes were washed, resolved, blotted, and detected as described previously (42).

Membranes were stripped by incubating in 62.5 mM Tris (pH 6.9)–2% sodium dodecyl sulfate–100 mM β-mercaptoethanol at 65°C for 1 h and then rinsing thoroughly with phosphate-buffered saline.

Cotransfection and CAT assay.

For K562 cells, 2 μg of reporter constructs (pG5tkCAT, pC1G3tkCAT, or pμLCR-GAL/βglob-CAT) and 2 μg of test constructs (pGAL/EKLF, pGAL/P, pGAL/Zn, pSG5/EKLF, pSG5/P, or pSG5/Zn) together with 0.5 μg of growth hormone construct pXGH5 were cotransfected into 2 × 10⁶ cells. For NIH 3T3 cells, 50 to 60% confluent cells in 100-mm dishes were cotransfected with 6 μg of reporter pG5tkCAT and 6 μg of test construct pGAL/EKLF, pGAL/P, or pGAL/Zn. After 40 h, cell lysates were prepared and chloramphenicol acetyltransferase (CAT) activity was assayed using the phase extraction method (32). The CAT assay was performed for 2 h, and the activity presented in relevant figures is the average of multiple assays after normalization to growth hormone levels (5).

RESULTS

EKLF interacts with corepressor mSin3A in vivo.

One of the most important mechanisms by which a transcription repressor exerts its function is to interact with corepressors and recruit HDACs to the site. To pursue the possibility that EKLF may also function as a transcription repressor, a coimmunoprecipitation assay was performed to determine whether EKLF could interact with corepressors and recruit HDACs. To begin, the interaction of EKLF with corepressor mSin3A was examined by cotransfection of EKLF (pSG5/EKLF) (5) and myc-tagged mSin3A (pCS2/Sin3A) (15) (or their empty vector controls) into Cos7 cells. Nuclear extracts were prepared and immunoprecipitated, and the resulting supernatants and pellets were analyzed by Western blotting. As expected, transfected EKLF (Fig. 1A, lanes 5 and 6), but not mSin3A alone (Fig. 1B, lane 4), can be precipitated efficiently by the anti-EKLF antibody. However, in the presence of EKLF protein, the anti-EKLF antibody (Fig. 1B, lane 5) can also precipitate mSin3A. This suggests that EKLF can specifically interact with mSin3A in vivo. To confirm this interaction, the reverse coimmunoprecipitation was performed. Again, transfected myc-tagged mSin3A proteins can be precipitated efficiently by anti-myc antibody (Fig. 1C, lanes 4 and 5). However, EKLF could be precipitated by anti-myc antibody only in the presence of mSin3A protein (Fig. 1D, lane 5). This result indicates again that EKLF specifically interacts with corepressor mSin3A in vivo.

FIG. 1.

Coimmunoprecipitation of EKLF and mSin3A in Cos7 cells. Cells were cotransfected with pSG5 and mSin3A (lanes 1 and 4), EKLF and mSin3A (lanes 2 and 5), or EKLF and pCS2 (lanes 3 and 6). Analyses of supernatants (lanes 1 to 3) or pellets (lanes 4 to 6) from each immunoprecipitation are shown. Samples were immunoprecipitated with an anti-EKLF antibody (A and B) and then probed with an anti-EKLF antibody to monitor EKLF (A) or probed with an anti-myc antibody to monitor mSin3A (B). Samples were immunoprecipitated with an anti-myc antibody (C and D) and then probed with an anti-myc antibody to monitor mSin3A (C) or probed with an anti-EKLF antibody to monitor EKLF (D). The top bands in panel D are antibody heavy chains.

Since EKLF is an erythroid-specific transcription factor, the interaction of EKLF with mSin3A was then determined in K562 cells, a human erythroleukemic cell line that expresses the fetal but not the adult globin program. As these cells do not contain any endogenous EKLF, the previous coimmunoprecipitation assay was repeated by cotransfection of EKLF (pBOS/EKLF) and mSin3A (pCS2/mSin3A) (or their empty vector controls). The results show that transfected EKLF proteins can be precipitated efficiently by anti-EKLF antibody (Fig. 2A, lanes 5 and 6) and that the association of EKLF with mSin3A also occurs in these erythroid cells (Fig. 2B, lane 5).

FIG. 2.

Coimmunoprecipitation of EKLF and mSin3A in K562 and MEL cells. (A and B) K562 cells were cotransfected with BOS and mSin3A (lanes 1 and 4), EKLF and mSin3A (lanes 2 and 5), or EKLF and pCS2 (lanes 3 and 6). Analyses of supernatants (lanes 1 to 3) or pellets (lanes 4 to 6) from each immunoprecipitation are shown. Samples were immunoprecipitated with an anti-EKLF antibody (A and B) and then probed with an anti-EKLF antibody to monitor EKLF (A) or probed with an anti-myc antibody to monitor mSin3A (B). (C) Nuclear extracts from untransfected MEL cells were prepared and subjected to immunoprecipitation with an anti-EKLF antibody (lane 2) or an M2 antibody (negative control in lane 1). Pellets were probed with an anti-mSin3A antibody (top panel; mSin3A protein is shown) or an anti-EKLF antibody (bottom panel; arrow indicates EKLF protein). Upper bands are antibody heavy chains.

As a final stringent test of EKLF/mSin3A associations, the interaction between endogenous EKLF and mSin3A was examined in MEL cells, a murine erythroid cell line that expresses high levels of EKLF and can be induced to differentiate along the adult globin pathway (21). MEL nuclear extracts were immunoprecipitated with an anti-EKLF antibody (or the M2 antibody as a negative control) and then were blotted and probed with anti-EKLF and anti-mSin3A antibody. The results show that endogenous EKLF can be precipitated efficiently by an anti-EKLF antibody (Fig. 2C, lane 2, bottom panel) but not by an M2 antibody (Fig. 2C, lane 1, bottom panel) and that mSin3A is coprecipitated by an anti-EKLF antibody (Fig. 2C, lane 2, top panel) but not by an M2 antibody (Fig. 2C, lane 1, top panel). This indicates that the interaction between EKLF and mSin3A also occurs endogenously in EKLF-expressing MEL cells.

The zinc finger domain of EKLF is responsible for interacting with mSin3A.

In order to map the subdomain of EKLF that is responsible for interacting with mSin3A, three deletion constructs (Fig. 3A) were made from full-length EKLF (21). The first construct, pSG5/fZn (Zn), contains only the zinc finger domain of EKLF (amino acids 287 to 376). A Flag tag was also introduced into this construct at the N terminus. The second construct, pSG5/P (P), contains only the proline-rich domain of EKLF (amino acids 19 to 291). The third construct, pSG5/D (D), has an internal deletion in which amino acids 60 to 195 of EKLF were removed. Since the anti-EKLF antibody recognizes the N terminus, an anti-Flag monoclonal antibody (M5) was used for detecting protein derived from the pSG5/fZn construct.

FIG. 3.

Coimmunoprecipitation of EKLF deletion mutants and mSin3A in Cos7 cells. Cells were cotransfected with mSin3A and full-length EKLF, its deleted derivatives, or vector control (pSG5). Analyses of supernatants (lanes 1 to 6) or pellets (lanes 7 to 12) from each immunoprecipitation are shown. (A) Schematic diagram of full-length EKLF and the deletion constructs used in this experiment (P, D, and Zn). (B) Samples were immunoprecipitated with an anti-EKLF antibody (for EKLF, P, and D; lanes 1 to 4 and 7 to 10) or an M5 anti-Flag antibody (for Zn; lanes 5, 6, 11, and 12) and then probed with an anti-myc antibody to monitor mSin3A protein. (C) To monitor expression of full-length EKLF and its deletions, the blot in panel B was stripped and reprobed with an anti-EKLF antibody. Full-length EKLF, proline-rich domain P, and internal deletion D proteins are shown (lanes 8 to 10).

The results show that mSin3A can be coprecipitated with EKLF proteins that encode the zinc finger domain (full-length, D, and Zn; Fig. 3B, lanes 8, 10, and 12) but not the proline-rich domain (P; Fig. 3B, lane 9). To eliminate the possibility that the differences among the levels of precipitated mSin3A are due to different expression of various constructs, the blot was stripped and reprobed with anti-EKLF antibody. Although the cells transfected with construct D expressed a lower amount of protein, they could still associate with a significant amount of mSin3A, unlike cells transfected with construct P (Fig. 3C). These data suggest that the zinc finger domain of EKLF is responsible for its interaction with mSin3A.

EKLF also associates in vivo with HDAC1 through its zinc finger domain.

The corepressor Sin3A exerts its function by recruiting HDACs (11, 15, 36). As a result, we monitored the association (direct or indirect) of EKLF with HDACs by cotransfection of pSG5/EKLF and HDAC1 constructs (in pBJ5 vector) (11) into Cos7 cells, followed by coimmunoprecipitation and Western blot analysis. The results show that EKLF can be precipitated by an anti-HDAC1 polyclonal antibody only upon cotransfection with HDAC1 (Fig. 4A, lane 5). The observation of a faint EKLF band in the sample which was transfected with pSG5/EKLF and the pBJ5 vector control (Fig. 4A, lane 6) is likely due to the coimmunoprecipitation of endogenous HDAC1 and exogenous EKLF. This result shows that EKLF can associate with HDAC1 in vivo.

FIG. 4.

Coimmunoprecipitation of EKLF and HDAC1 in Cos7 cells. (A) Cells were cotransfected with pSG5 and HDAC1 (lanes 1 and 4), EKLF and HDAC1 (lanes 2 and 5), or EKLF and pBJ5 (lanes 3 and 6). Anti-EKLF analyses of supernatants (lanes 1 to 3) or pellets (lanes 4 to 6) from immunoprecipitation with an anti-HDAC1 polyclonal antibody are shown. (B) Cells were cotransfected with EKLF and pBJ5 (lane 1), EKLF and HDAC1 (lane 2), P and HDAC1 (lane 3), Zn and pBJ5 (lane 4), or Zn and HDAC1 (lane 5). Samples were immunoprecipitated with an anti-HDAC1 antibody, and supernatants or pellets (as indicated) were probed with either an anti-EKLF antibody (for EKLF and P; lanes 1 to 3) or anti-Flag M5 antibody (for Zn; lanes 4 and 5). Constructs were as in Fig. 3.

The subdomain of EKLF that is responsible for the coprecipitation of EKLF and HDAC1 was determined by using the pSG5/EKLF, pSG5/fZn, and pSG5/P constructs. After cotransfection with HDAC1, samples were immunoprecipitated with an anti-HDAC1 antibody and then probed with an anti-EKLF antibody (for full-length EKLF and proline-rich domain P) or an anti-Flag M5 antibody (for Flag-tagged zinc finger domain Zn). The results show that, similar to mSin3A, only full-length EKLF and zinc finger domain Zn can be coprecipitated with HDAC1 (Fig. 4B, lanes 2 and 5), unlike the proline-rich domain P (Fig. 4B, lane 3). The faint bands in negative controls, which were transfected with pSG5/EKLF and pBJ5 (i.e., EKLF only) or pSG5/fZn and pBJ5 (i.e., zinc finger domain only) likely result from the interaction of transfected EKLF or zinc finger domain with endogenous HDAC1 (Fig. 4B, lanes 1 and 4). In this case the better negative control might be the sample transfected with both the EKLF proline-rich domain and HDAC1 (Fig. 4B, lane 3).

Again, to rule out the possibility that different expression levels of full-length EKLF, the proline-rich domain P, and the zinc finger domain Zn accounted for the differences among their coprecipitated protein levels, the amounts of various EKLF proteins in the samples were determined by analyzing the supernatants. These protein levels in the samples were very similar (Fig. 4B [left], lanes 1 to 5). This suggests that the zinc finger domain of EKLF is responsible for the interaction of EKLF with HDAC1. Although it cannot be determined from the data whether HDAC1 interacts with EKLF directly or indirectly (e.g., through endogenous mSin3A), the data from Fig. 1 to 4 clearly show that EKLF can form a complex with either mSin3A or HDAC1 in vivo.

EKLF can function as a repressor through its zinc finger domain, and this repressive function is associated with HDACs.

Since EKLF can form protein-protein complexes with mSin3A and/or HDAC1 in vivo, its ability to function as a transcription repressor was examined next. Full-length EKLF, its proline-rich domain (P), and its zinc finger domain (Zn) were each linked to the GAL4 DNA binding domain (GAL1-147) for correct targeting to the GAL4 binding sites present in the pG5tkCAT reporter (Fig. 5A, top). The GAL4 DNA binding domain (GAL1-147) is sufficient to target large heterologous proteins to the nucleus (33) and thus ensures the nuclear localization of each different EKLF domain. Since EKLF interacts with mSin3A and HDAC1 within its zinc finger domain, this design also avoids any influence of zinc finger DNA binding on its ability to interact with corepressors and HDACs.

FIG. 5.

Transcriptional activities of GAL-EKLF fusion proteins on pG5tkCAT. (A) K562 cells were transfected with the pG5tkCAT reporter (schematic shown on top) and pGAL/EKLF, pGAL/Zn, pGAL/P, or pGAL1-147. The CAT activity in the control transfection (with pGAL1-147) was given a value of 1 and was used to calculate the relative CAT activities from the other transfections. The data, normalized for transfection efficiency, are the averages of three individual experiments. (B) Effect of TSA on the transcriptional activities of GAL-EKLF fusion proteins on pG5tkCAT. K562 cells were transfected with the pG5tkCAT reporter and pGAL/EKLF, pGAL/Zn, and pGAL1-147. The CAT activity in the untreated control (transfected with pGAL1-147) was given a value of 1 and used to calculate the relative CAT activities from the other transfections. +, samples treated with TSA; −, samples not treated with TSA. The data, normalized for transfection efficiency, are the averages of three individual experiments.

After cotransfection of the pG5tkCAT reporter with pGAL/EKLF, pGAL/P, or pGAL/Zn into erythroid K562 cells, the ensuing CAT activities were measured. Since the reporter contains the thymidine kinase promoter, it will yield a significant level of expression by itself, making it easy to monitor its repression or activation upon cotransfection. As a result, levels of expression after cotransfection with the GAL4 DNA binding domain alone were used as the control and normalized to a level of 1 (Fig. 5A, lane 1). Transfection of full-length EKLF or its zinc finger domain resulted in an about threefold reduction in CAT activity (Fig. 5A, lanes 2 and 3). However, without the zinc finger domain, the proline-rich domain elevated the CAT activity by three- to fourfold (Fig. 5A, lane 4). As expected from previously published data, these results show that the EKLF proline-rich region is a transactivation domain (4, 5). However, they also show that EKLF can function as a transcription repressor through its zinc finger domain, a result that correlates well with the coimmunoprecipitation data.

To further study the relationship between the transcription repression function of EKLF and HDACs, we examined the effect of trichostatin A (TSA), a specific HDAC inhibitor, on the transcription repression function of EKLF. After cotransfection of the pG5tkCAT reporter and the GAL-EKLF fusion constructs as before, cells were grown in the presence or absence of 2 μM TSA for 24 h. Without TSA, as already shown, full-length EKLF and its zinc finger domain function as repressors, with CAT activities lowered to about 30% of control (Fig. 5B, lanes 3 and 5), and the proline-rich domain activates transcription (Fig. 5B, lane 7). In the presence of TSA the CAT activity of the control transfection did not show much change (Fig. 5B, lanes 1 and 2). However, the full-length EKLF was reversed to an activator; its CAT activity was elevated seven- to eightfold above the control (Fig. 5B, lane 4). The repression activity of the zinc finger domain was also reduced; its CAT activity was increased from 30% to about 80% of control (Fig. 5B, lane 6). The transactivation activity of the proline-rich domain was further elevated in the presence of TSA (Fig. 5B, lane 8). This last result is not surprising, as acetylation of the EKLF proline-rich domain upregulates its transcriptional activation function (42). As a result, inclusion of a general inhibitor (such as TSA) might be expected to prevent its deacetylation, possibly by different deacetylase complexes (e.g., as seen with p53 [19, 23]), and augment its activity. In sum, the results shown in Fig. 5 imply that HDAC activity may be involved in EKLF-dependent repression. Based on these results combined with our previous data, it is likely that EKLF exerts its repressive activity in vivo by formation of a protein-protein complex that effectively recruits corepressors and/or HDACs to a target promoter.

Both the promoter context and the way in which EKLF is targeted to a promoter are critical parameters for EKLF's ability to repress transcription.

We next examined whether the way EKLF is targeted to the promoter plays any role in its ability to function as transcription repressor, particularly because mSin3A and HDAC1 interact with the very same EKLF domain that is directly involved in DNA binding and promoter recognition. To test this possibility, a different reporter, pC1G3tkCAT, was used. In pG5tkCAT five GAL4 binding sites are located in front of tkCAT; in pC1G3tkCAT four EKLF binding sites are located in front of tkCAT (4). Instead of recruiting EKLF to the promoter by means of the GAL4 DNA binding domain in the GAL-EKLF chimera, the pC1G3tkCAT reporter enables wild-type EKLF to directly bind the same promoter by means of its zinc fingers. CAT activities were measured after cotransfection of pC1G3tkCAT with full-length EKLF (pSG5/EKLF), its zinc finger domain (pSG5/Zn), its proline-rich domain (pSG5/P), or a vector control (pSG5). The results show that when directly bound to DNA, full-length EKLF functions as an activator and its zinc finger domain has no repressive activity (Fig. 6, lanes 2 and 3). As expected, the proline-rich domain does not have any effect on CAT activity since it cannot bind to the promoter (Fig. 6, lane 4). These data suggest that to function as a repressor, EKLF has to be recruited to the promoter by another protein(s) instead of directly binding to the promoter through its zinc finger domain. This could be an important mechanism to enable EKLF to expose its zinc finger domain to corepressors and HDACs.

FIG. 6.

Transcriptional activities of EKLF and its deletion mutants on pC1G3tkCAT. K562 cells were transfected with reporter pC1G3tkCAT (schematic shown on top) and pSG5/EKLF, pSG5/Zn, pSG5/P, and pSG5. The CAT activity in the control transfection (with pSG5) was given a value of 1 and used to calculate the relative CAT activities from the other transfections. The data, normalized for transfection efficiency, are the averages of three individual experiments.

Next, we asked whether the target promoter context itself is also important for the EKLF transcription function. EKLF normally activates transcription of the β-globin gene by binding to the CACCC element located in the proximal promoter (4, 28). In the present experiment, we used a significantly less artificial reporter construct than previously used that contains the CAT gene under the control of the μLCR and the natural β-globin promoter (pμLCR-GAL/βglob-CAT). The use of the βLCR microcassette (μLCR) ensures high-level expression of the reporter downstream from the β-globin promoter in erythroid cells (4, 35). However, the EKLF binding site (CACCC) at −90 in the β-globin promoter was changed to a GAL4 binding site, enabling EKLF to be targeted to this β promoter through a linked GAL4 DNA binding domain instead of through its own zinc finger domain (Fig. 7A, top).

FIG. 7.

Transcriptional activities of GAL-EKLF fusion proteins on pμLCR-GAL/βglob-CAT and pαLCR-GAL/βglob-CAT. K562 cells were transfected with pGAL/EKLF, pGAL/Zn, pGAL1-147, and either the pμLCR-GAL/βglob-CAT reporter (A) or the pαLCR-GAL/βglob-CAT reporter (B). The CAT activity in the control (transfected with pGAL1-147) was given a value of 1 and used to calculate the relative CAT activities from the other transfections. The data, normalized for transfection efficiency, are the averages of three individual experiments.

This reporter was cotransfected with pGAL/EKLF, pGAL/Zn, pGAL/P, and control pGAL1-147 into K562 cells. The results show that although EKLF was targeted to the β promoter by the GAL4 DNA binding domain, it still activates the β-globin promoter six- to sevenfold (Fig. 7A, lane 2). The zinc finger domain, pGAL/Zn, did not show any significant repressive activity (Fig. 7A, lane 3). Once again the proline-rich domain of EKLF is a strong transcriptional activator (Fig. 7A, lane 4). These data are in significant contrast to the analogous experiment performed with the pG5tkCAT reporter depicted in Fig. 5A and show that in addition to the way that EKLF is targeted to the promoter (i.e., via GAL4 or zinc finger DNA motifs), the target promoter context is important for regulating EKLF function. Although the exposure of the zinc finger domain is similar in this experiment to that shown in Fig. 5A, the promoter configuration may prevent the domain from forming a productive complex with corepressors and HDACs.

The reporter construct used in Fig. 7A contains the μLCR, which also contains EKLF binding sites. Although the CACCC site in the β-globin promoter was changed to the GAL4 DNA binding site in the reporter construct pμLCR-GAL/βglob-CAT, the CACCC sites in the μLCR were not changed. This is important to consider, as EKLF also plays a role in activating gene expression that is mediated through the βLCR (7, 16, 37). To rule out the possibility that EKLF binding to the CACCC sites at the μLCR enhanced CAT expression in the previous experiment, we replaced the μLCR with the αLCR from the α-like globin gene cluster to form a new reporter construct, pαLCR-GAL/βglob-CAT. The αLCR can also elevate downstream gene expression in erythroid cells (4, 35) but it does not contain any EKLF binding sites. Using this new reporter construct, we repeated the previous experiment and found a similar result: GAL-EKLF and GAL-P fusion proteins are activators and GAL-Zn shows no repression of the β-globin promoter (Fig. 7B). Again, this further strengthens the suggestion that the promoter configuration is important for EKLF to function as a repressor or activator.

DISCUSSION

In these experiments, coimmunoprecipitation and reporter gene assays have been used to determine whether EKLF can function as a transcriptional repressor. Our data show that the zinc finger domain of EKLF interacts with mSin3A and HDAC1 and possesses transcriptional repression activity. The observation that TSA can relieve EKLF-mediated transcriptional repression also indicates that this activity is dependent on HDACs. Together these data suggest that EKLF can function as a transcription repressor and exert repressive activity by recruiting corepressors and HDACs via its zinc finger domain. The recruitment of HDAC1 by EKLF could be direct as observed for YY1 (41) or indirect through Sin3 corepressors as observed for the Mad protein, which interacts with both mSin3A and mSin3B (3, 31). Both HDAC1 and HDAC2 are associated with mSin3A (15). Although we have tested only the ability of mSin3A and HDAC1 to individually form a complex with EKLF, other corepressors or HDACs could also be involved in EKLF-mediated transcription repression.

As an activator, EKLF binds to DNA directly through its zinc finger domain, yet our data show that the same zinc finger domain is also responsible for interacting with mSin3A and HDAC1. It was formally possible that EKLF could be capable of binding both DNA and corepressors simultaneously. However, our data suggest that this is not the case: DNA binding by EKLF appears to block its repressive activity, as demonstrated by comparing the transfection results described in Fig. 5A and 6. In one case, cotransfection of GAL/EKLF or GAL/Zn with pG5tkCAT leads to repression, while in the other case, cotransfection of EKLF or Zn with pC1G3tkCAT leads to activation or has no effect. It is possible that a DNA-bound EKLF zinc finger is not accessible for interaction with corepressors and HDACs. The lack of effect by the control pGAL1-147 construct eliminates the possibility that repression is due to the GAL4 DNA binding module present in the GAL/EKLF and GAL/Zn fusion proteins. In addition, GAL/P functions as an activator. This also rules out the concern as to whether the conformation of the fusion protein inherently enables repression to occur.

This leads to the hypothesis that to function as a repressor, EKLF needs to be recruited by other DNA binding protein(s) in such a way that its DNA binding domain remains accessible for interactions with corepressors and/or HDACs. EKLF may thus be able to influence transcription through DNA-binding-dependent and -independent mechanisms similar to those of the glucocorticoid receptor (30). In addition, EKLF repression might be targeted to a set of genes that do not contain any CACCC site. It has been shown recently that a lacZ reporter, linked to a β-globin promoter with or without a CACCC box, was expressed to a higher level in EKLF^−/− fetuses than in wild-type animals (9). These authors suggested that EKLF might be able to inhibit transcription in certain contexts.

Protein-protein interactions are critical for EKLF function. Our previous work has shown that EKLF interacts with a positive-acting cellular factor through its proline-rich domain (5). Further, this domain has been shown to be required for coactivator E-RC1-dependent, β-globin promoter activation (1). Recently, it has also been shown that EKLF can interact with the CBP and P300 coactivators (42), and we have here shown that EKLF also interacts with the mSin3A and HDAC1 corepressors. How the erythroid cell regulates EKLF interaction with both positive and negative factors is not clear.

The Smad2-Smad4 complex can interact with coactivators to form a transcriptional activation complex or with corepressors to form a transcriptional repressor complex. The determining factor for forming one of these complexes is the relative level of Smad corepressors and coactivators within the cell (39). The case of EKLF is more complex, as it also functions differently under different target promoter configurations; i.e., EKLF behaves as a repressor at the thymidine kinase promoter and as an activator at the β-globin promoter even when EKLF has been recruited to them the same way (by a GAL4 DNA binding domain). Consistent with this idea, EKLF can activate the β-globin gene by binding to the CACCC site on the β-globin promoter but it cannot activate a γ-globin gene whose promoter contains a similar CACCC site, even when the γ CACCC site in the γ-globin promoter is changed to a β CACCC site (2). As a result, to repress a promoter it may not be sufficient to simply recruit EKLF, as the interaction of EKLF with corepressors and HDACs may be augmented or prevented by the overall promoter architecture. A sequence-specific binding protein or another factor may change this configuration to hinder or to facilitate EKLF's repressive activity at different promoters or in alternate cellular environments.

Our data suggest that EKLF's transcriptional function may be regulated at two levels that depend on how its critical domains are presented to other cellular proteins at a particular location (i.e., with its zinc finger domain unoccupied or sequestered at a CACCC element) and on what the specific configuration is at its target promoter site.

At the same time, EKLF interaction with repressors does not automatically exclude its accessibility for positive coactivators. For example, not only is GAL-EKLF repression at pG5tkCAT relieved by TSA, but it is also converted to an activator (Fig. 5B), suggesting that positive factors are still able to interact with EKLF when its repressive activity is blocked.

In addition to activating β-globin gene expression, EKLF is also implicated in consolidating the switch from γ- to β-globin. EKLF-null erythroid cells maintain βh1 globin and transgenic γ-globin levels that are higher than those of wild-type cells prior to silencing (18, 26, 38). Conversely, overexpression of EKLF leads to an earlier decrease of transgenic γ-globin (37). Based on the present data, one might postulate that besides competition, the higher level of γ-globin in EKLF-null mice might be due to the loss of EKLF repressive activity at the γ gene promoter (subsequent silencing would occur by an EKLF-independent mechanism). The observation that proliferation of EKLF^−/− cell lines is significantly reduced upon activation of a reintroduced conditional form of EKLF (25) is also consistent with the possibility that EKLF can function as a repressor within certain contexts.

In summary, although EKLF's role as an activator is well established, it may also be playing an unanticipated role in downregulating the expression of a subset of genes that are important for erythroid differentiation by recruiting corepressors and/or HDACs to target promoters. A detailed study of EKLF-protein interactions will reveal the mechanism by which EKLF function is regulated between these two states and will provide important information about the role of EKLF in the control of erythroid gene expression and differentiation.