In Vivo Formation of a Human b -Globin Locus Control Region Core Element Requires Binding Sites for Multiple Factors Including GATA-1, NF-E2, Erythroid Kruppel-like Factor, and Sp1*

The active elements of the b -globin locus control region (LCR) are located within domains of unique chromatin structure. These nuclease hypersensitive sites (HSs) are characterized by high DNase I sensitivity, erythroid specificity, similar nucleosomal structure, and evolutionarily conserved clusters of cis -acting elements that are required for the formation and function of the core elements. To determine the requirements for HS core formation in the setting of nuclear chromatin, we constructed a series of artificial HS cores containing binding sites for GATA-1, NF-E2, and Sp1. In contrast to the results of previous in vitro experiments, we found that when constructs were stably integrated in mouse erythroleukemia cells the binding sites for NF-E2, GATA-1, or Sp1 alone or in any combination were unable to form core HS structures. We subsequently identified two new cis -acting elements from the LCR HS4 core that, when combined with the NF-E2, Sp1, and tandem inverted GATA elements, result in core structure formation. Both new cis -acting elements bind Sp1, and one binds erythroid Kruppel-like factor (EKLF). We conclude that in vivo b -globin LCR HS core formation is more complex than previously thought and that several factors are required for this process to occur. The human b -globin

The human ␤-globin gene locus contains five expressed genes that code for the ␤ chains of the embryonic, fetal, and adult hemoglobin molecules. The genes of the locus are expressed in a tissue-specific and developmentally regulated fashion (reviewed in Ref. 1). The embryonic ⑀-globin gene is expressed in the fetal yolk sac, the ␥-globin genes are expressed in the fetal liver, and the adult ␦and ␤-globin genes are expressed in adult bone marrow. Because of this complicated pattern of transcriptional regulation, the ␤-like globin genes have served as an important model for understanding the regulation of complex, multi-gene, eukaryotic loci. Since the original description of the ␤-globin locus control region (LCR) 1 more than 10 years ago, this ϳ25-kb region has been a primary focus of efforts to un-derstand the mechanisms of regulated globin gene expression. Although significant advances have been made, the functions attributable to the LCR, its mechanisms of action, and its physical structure remain to be determined (2)(3)(4). Most of the transcriptional activity of the LCR has been localized to specific core elements that are a few hundred base pairs in size (5)(6)(7)(8). These LCR core elements share several common structural and functional properties. Each is closely associated with one of the five nuclease HSs currently considered to comprise the LCR (5Ј HS1-5). These HSs are developmentally stable, and LCR HSs 1-4 are erythroid-specific (9,10). In contrast to the promoterassociated HSs of the locus, the ␤-globin LCR HSs are ϳ10 -20-fold more sensitive to DNase I digestion (9,13,14). At the nucleosomal level, the LCR cores are characterized by regions of nucleosome disruption or displacement spanning ϳ200 bp (14,15). At the DNA level, each LCR HS core sequence is highly conserved between species (16). Structural similarities are also present between the different LCR cores within the same species. We have previously observed that the cores of LCR HSs 1-4 contain tandem, inverted GATA-binding elements located ϳ50 bp downstream from AP-1/NF-E2-binding elements (also known as maf response elements) (14,17). These binding sites are located within the nucleosome-free regions of the HS cores; when these sites are mutated, loss of the normal chromatin structure of the HS and of appropriate expression from linked transgenes in transgenic mice occurs (8,14,18). Recent experiments in ␤-globin yeast artificial chromosome (YAC) transgenic mice have emphasized the importance of the HS core elements to the function of the LCR, because deletion of a 280-bp element containing the HS4 core or a 225-bp element containing the HS3 core both result in dramatic changes in gene expression despite the presence of otherwise intact ␤-globin yeast artificial chromosomes (YACs) (19,20). One interpretation of this data is that the LCR core elements are required for the formation of an LCR holocomplex that incorporates each of the individual LCR elements.
In an effort to better understand the structure and function of the LCR core elements, we have studied the requirements for formation of the unique chromatin structures associated with these elements. In previous experiments we found that a 1.1-kb fragment from the region of LCR HS4 is able to autonomously form the characteristic core structure in fetal liver cells of transgenic mice (21). Similar results have been found for human LCR HS2 (22) and for the endogenous mouse LCR elements (23). Truncation experiments localized this activity to a 101-bp element that is both necessary and sufficient for the formation of LCR HS4 (21). We have termed this the HSforming element or HSFE (see Fig. 1A). Researchers in our laboratory (21) and Pruzina et al. (24) have used in vitro DNAprotein binding assays to show that several binding sites are present within this element. These include binding sites for the ubiquitous factors AP-1 and Sp1 and the hematopoietic factors NF-E2 and GATA-1 (see Fig. 1A). The GATA elements are present as tandem, inverted repeats and are located ϳ50 bp 3Ј of the AP-1/NF-E2 site. We have also shown that mutation of the NF-E2 site or the GATA binding sites reduces the sensitivity of HS4 so that it becomes equivalent to the promoterassociated HSs (14). Mutation of both NF-E2 and GATA sites within the same construct results in almost complete loss of the HS. Examination of DNA sequences from the other human LCR HS cores revealed that NF-E2 and tandem, inverted GATA binding sites, separated by ϳ50 bp, were common to each HS (14). These sites are conserved in the ␤-globin LCR HS cores of chickens, rabbits, goats, mice, and galagos (14,25,26). We have shown that mutation of these conserved NF-E2-and GATA-binding elements within HS2 and HS3 of the human LCR also results in the loss of normal HS formation (2). Boyes and Felsenfeld (27), using an experimental design that also involved plasmid integration, obtained similar results with the chicken ␤-globin enhancer. These investigators compared chromatin accessibility between single and multi-copy integrants and found little difference, except that some single copy clones exhibited position effects.
These findings have led us to hypothesize that there is a common core structure, located near the center of each of the LCR HSs, that is necessary for the formation and normal functioning of each of the elements. The data also lead us to propose that the NF-E2-binding element and tandem, inverted GATAbinding elements found within each LCR HS core are not only necessary, but also sufficient, to establish the active chromatin structures of the ␤-globin LCR HSs in vivo. Although previously published results studying LCR HS formation in vitro or in unintegrated vector systems have implicated NF-E2, GATA-1, and EKLF in globin locus DNase I HS formation (28 -31), the roles of these factors in native chromatin have not been determined. To test our hypothesis that GATA-1-and/or NF-E2-binding elements are necessary and sufficient for LCR HS core formation in the native chromatin of erythroid cells, we constructed a series of artificial HS-forming elements based on the LCR HS4 HSFE, in which consensus binding sites for NF-E2, GATA, Sp1, and two previously unidentified binding elements were incorporated into a portion of the firefly luciferase cDNA. We have determined that neither the NF-E2 or GATA binding sites, alone or in combination, are sufficient for the formation of an LCR core HS in nuclear chromatin and that additional factor binding sites, including those for Sp1 and/or EKLF, are also required.

MATERIALS AND METHODS
Plasmid Constructions-A 1.6-kb XbaI/XbaI fragment of the firefly luciferase cDNA was subcloned into the XbaI site of a previously described pUC-based plasmid containing a neomycin resistance gene and a 1.9-kb HindIII/HindIII fragment from the region of the ␤-globin LCR HS3 (see Fig. 1B) (14,21). The 5Ј XbaI site is internal to the cDNA, whereas the 3Ј XbaI site is derived from a polylinker sequence. Control HS4 sequences were isolated from previously described plasmids (21), blunted, and cloned into a unique HincII site within the luciferase cDNA (see Fig. 1B). Site-directed mutagenesis (Chameleon method; Stratagene, La Jolla, CA) was used to create consensus NF-E2-, GATA-, and Sp1-binding motifs within the luciferase cDNA. The following oligonucleotides, containing the factor binding sites flanked by 8 -16 bp of luciferase cDNA sequence, were used in the mutagenesis procedure (see Fig. 3) (specific factor-binding motifs are underlined): NF-E2 site oligonucleotide, 5Ј-ACAAAGGATATGCTGAGTCACCCGCTGAAT-3Ј; tandem, inverted GATA site oligonucleotide, 5Ј-ACCCCAACTGATAG-CACTATCTCGTGGCAG-3Ј; Sp1 site oligonucleotide, 5Ј-CCCGCTGAA-TTGGAATGGGCAGGGTTACAACACCCC-3Ј. The following oligonucleotides containing the 5Ј and 3Ј ends of the HS4 HSFE were also used in the mutagenesis procedure. Sequences derived from the HSFE are underlined. The 5Ј HSFE sequence is 5Ј-CATATGGCTTGCCCTGCCT-CTCTACTAGGCTCAGGTGGCCCCCGCTG-3Ј, and the 3Ј HSFE sequence is 5Ј-GCAGAGCCAGGGCCGGGACGATGACGCCGGTGAAC-3Ј.
Multiple rounds of mutagenesis were performed to obtain constructs with multiple factor binding sites. The HS4-derived matrix attachment region (MAR) is a 691-bp BamHI/AccI fragment from the 5Ј end of the 1.4-kb BamHI/SphI HS4 region (14). Test constructs were stably transfected into mouse erythroleukemia (MEL) cells and clones selected in G418 as previously described (21).
Nuclease Sensitivity Assays-In vivo DNase I sensitivity assays were performed on nuclei isolated from pools of 25 clones as previously described (14,21). Reactions were performed at 37°C for 10 min. Following DNase I treatment, genomic DNA was isolated and digested to completion with XhoI for Southern blotting (see Fig. 1B). A 547-bp XbaI/EcoRI fragment from the 5Ј end of the luciferase fragment served as an indirect end-labeling probe for Southern blotting (see Fig. 1B).
Gel Mobility Shift Assays-Nuclear extracts from MEL cells were prepared, and gel shift assays were performed as previously described (32,33). Double-stranded versions of the nucleotides used to form the artificial factor binding sites for NF-E2, GATA-1, and Sp1 were used as gel shift probes. Control HS4 GATA and porphobilinogen deaminase (PBGD) NF-E2 gel shift oligonucleotides have been previously described (21,34). The control Sp1 binding site oligonucleotide 5Ј-ATTC-GATCGGGGCGGGGCGAGC-3Ј was from the ␣-2(I) collagen gene promoter (35) (Santa Cruz Biotechnology, Santa Cruz, CA). The HS4 HSFE Sp1 binding site oligonucleotide was 5Ј-TCCAAGGCCCAGCAAT-GGGCAGGGCTCTGTCAGGGC-3Ј (14). Anti-Sp1 antibody used in super-shift and competition experiments was from Santa Cruz Biotechnology (catalog number sc-59-G). The oligonucleotides used for gel shifts with the 5Ј HSFE and 3Ј HSFE were 5Ј-ATGGCTTGCCTGC-CTCTCTACTAGGC-3Ј and 5Ј-GCAGAGCCAGGGCCGGACCGCTT-GAAG-3Ј, respectively. The oligonucleotide used for the EKLF binding site was 5Ј-AGCTAGCCACACCCTGAAGCT-3Ј, which is the consensus site from the ␤-globin promoter (36). The EKLF antibody used in the competition experiment was a kind gift from Dr. James Bieker (37).

RESULTS
To test the hypothesis that NF-E2 and tandem, inverted GATA sites, separated by ϳ50 bp, are sufficient for HS formation in nuclear chromatin, we constructed the test plasmid shown in Fig. 1B. When stably transfected into MEL cells, the construct is able to direct DNase I HS formation (14,21,38). This plasmid contains a neomycin resistance gene (neo R ) for the selection of stable transfectants and a 1.9-kb HS3 fragment, which serves as an internal control for LCR HS formation. The HS3 fragment forms two HSs, primary HS3 (1°HS3) and secondary HS3 (2°HS3). The 1°HS3 contains the functional core of HS3 and forms the typical "super" LCR HS, whereas the 2°HS3 is a much weaker HS with a DNase I sensitivity approximately equal to that of the globin promoterassociated HSs. The presence of these two internal control HSs allows for semiquantitative determination of the DNase sensitivity of artificial or mutant HSs. The neo R gene also serves as a control for HS formation, because a strong HS is formed in the region of the gene promoter. This element contains portions of the herpes simplex thymidine kinase gene promoter and an enhancer from the polyoma virus (39). The MEL cell line has an adult erythroid phenotype, because it expresses the murine adult ␤-globin gene. GATA-1 is the only GATA family member expressed in these cells (40,41), and NF-E2 and AP-1 are the two major NF-E2 consensus sequence-binding factors in MEL cells (32,42). The test construct also contains a portion of the firefly luciferase cDNA. This 1.6-kb XbaI/XbaI DNA fragment was chosen to serve as a "neutral" or non-HS-forming fragment of DNA because it contains no known regulatory elements and is non-mammalian in origin. We used site-directed mutagenesis to create consensus factor binding sites within this fragment of DNA ( Fig. 1) or subcloned portions of LCR HS4 into the HincII site of the luciferase cDNA to serve as positive control sequences for HS formation. Test plasmids were stably transfected into MEL cells, and following selection in G418, pools of 25 individual clones were analyzed for the ability of the inserted sequences to direct HS formation.
Control Experiments-For our assay system to work, the luciferase cDNA fragment should not form any DNase I HSs itself; nor should it inhibit HS formation. To determine whether our model met these criteria, we first tested the wildtype luciferase sequence alone. As shown in Fig. 2, this construct formed each of the expected internal control HSs including the neo R -associated HS and both 1°and 2°HS3. No DNase I HS is formed within the luciferase cDNA sequence. We next placed a full-length, 1.1-kb HS4 fragment into the HincII site of the luciferase cDNA. This resulted in the formation of a typical super DNase HS, which is much more sensitive than 2°HS3. Finally, we placed the previously described 101-bp HS4 HSFE into the HincII site. This construct also formed a strong DNase I HS. These results indicated that in the context of the luciferase cDNA all the cis-acting sequences necessary for forming a globin LCR HS are present within the HSFE and that the luciferase cDNA does not inhibit HS formation. NF-E2 and Tandem, Inverted GATA Elements Are Not Sufficient for the Formation of a DNase I HS-The next step in testing our hypothesis was to use site-directed mutagenesis to place consensus NF-E2 and tandem, inverted GATA sites into the luciferase cDNA. These sites were inserted near the ClaI restriction site (Figs. 1 and 3). We utilized the NF-E2-binding sequence from the PBGD promoter as our model NF-E2 site. NF-E2 binding to this sequence has been well characterized (32,43,44) and is an 11/11-bp match of the consensus-binding sequence (42). The tandem GATA sequences from the HS4 HSFE were used in the test construct. Factor binding to these sites has also been well characterized (21,24), and both elements are 6/6-bp matches of the consensus GATA-1-binding sequence (45). The spacing between the NF-E2-and GATAbinding elements was the same as in the HS4 HSFE (Fig. 3). As shown in Fig. 4A, the presence of either the NF-E2 site (artificial HS construct 1, AH1) or tandem, inverted GATA binding sites, alone (AH2) or in combination (AH3), was insufficient for the formation of a DNase I HS. The open arrows in Fig. 4A indicate the locations of the inserted factor-binding sequences. In each case the internal control HSs formed normally.
One possible explanation for these results was that our artificial binding sites were, due to surrounding nucleotide sequences, unable to interact with NF-E2 and/or GATA proteins. To assess this possibility we performed gel mobility shift assays. In each case, oligonucleotides containing the artificial binding sites and their flanking luciferase cDNA sequences were compared with previously characterized oligonucleotides containing the naturally occurring binding sites. Nuclear extracts from MEL cells were used for these experiments. These The HSFE is a 101-bp element that is necessary for the formation of LCR HS4 (shaded oval) and contains binding sites for several factors. B, test construct used to evaluate the ability of trans-factor binding sites to mediate HS formation in erythroid cells. Test HS-forming elements were created within a 1.6-kb portion of the luciferase cDNA. The construct also contains HS3 and the neo R gene promoter as internal controls for HS formation. The neo R gene allows selection of MEL cells stably transfected with the test constructs. The probe used in DNase I HS assays is shown. kbp, kilobase pair(s). are the same cells used in our DNase I HS assays and in previously published binding studies. The results of these experiments are shown in Fig. 4B. Using the artificial GATAbinding sequences we saw two bands (Fig. 4B). These bands comigrate with the two bands seen with the wild-type tandem GATA sites from HS4. These bands are due to single and double occupancy of the two sites (21,24). Both gel shift bands are competed by 200-fold excess unlabeled self-and wild-type HS4 GATA sequences. An unrelated (NF-E2) oligonucleotide did not compete. Similarly, the artificial GATA sites are able to compete for binding to the wild-type HS4 sites. Because GATA-1 is the only GATA protein present at appreciable levels in MEL cells (40,41), these results demonstrate that the artificial GATA binding sites are able to interact appropriately with GATA-1 in vitro.
In Fig. 4B the results of gel shift assays using the artificial NF-E2 site are shown. Here the control binding site is a previously characterized oligonucleotide derived from the PBGD promoter (43,44). These two oligonucleotides differ only in the sequences flanking the consensus NF-E2-binding element. Our results with the control binding site are equivalent to those published using the same oligonucleotide with extracts from MEL cells (32,43,44). As has been previously demonstrated, three distinct bands are present under these conditions. The lower band is due to NF-E2 binding, and the middle band is due to AP-1 binding. The upper band has not, to our knowledge, been specifically identified. The artificial binding site exhibits three gel shift bands that comigrate with the control bands. The one significant difference appears to be a relative increase in the upper band relative to the two lower bands. Once again, all bands are specifically competed by 200-fold excess unlabeled self-and control oligonucleotides. These results demonstrate that the artificial NF-E2 binding sites are able to interact with NF-E2 and AP-1 in vitro. From these findings we conclude that NF-E2-and GATA-binding elements, alone or in combination, are unable to direct DNase I HS formation.
Addition of an MAR Does Not Confer DNase Hypersensitivity-The above results suggest that a cis-acting element or elements are missing from our artificial HS constructs. We have previously shown that the 1.4-kb HS4-containing fragment exhibits in vitro matrix attachment activity (46). Additional unpublished results from our laboratory indicate that most of this activity resides in the region of the fragment 5Ј of the HSFE. To determine whether this MAR element would contribute to HS formation, we subcloned a 691-bp BamHI/AccI fragment from the 5Ј end of the HS4 fragment into the HincII site of constructs with (AH3) and without (luciferase wild type, LWT) the combined NF-E2 and tandem GATA sites (see Fig. 3). These constructs and experimental results are shown in Fig. 5. These experiments demonstrate that the MAR region alone does not form a DNase I HS and that the combination of the MAR, NF-E2-, and tandem GATA-binding elements is also unable to form an HS.
An Sp1-binding Element Induces the Formation of a Weak DNase I HS-Our results thus far indicated that we were still missing sequences necessary for HS formation, which must be present within the 101-bp HSFE. Previous work on HS4 from our own laboratory and the Grosveld laboratory and from the Felsenfeld laboratory on a homologous region of the chicken 3Ј ␤-globin enhancer have demonstrated that, in addition to the in vitro DNase I footprints associated with the AP-1/NF-E2 and GATA sites, there is also a centrally located Sp1 binding site (21,24,47). This suggested that the Sp1 site might be the missing element for HS formation. Each of the LCR HS core regions contains several Sp1 binding sites (6,7,21,24,26). Next we constructed artificial HSs containing the HS4 Sp1 element in the same location as it is found within the HSFE (Fig. 3). Results of these experiments are shown in Fig. 6A. When the Sp1 binding site is inserted alone (AH4), a set of very weak DNase I HSs is formed within the expected region. The most prominent HS is directly over the region of the Sp1 binding site (Fig. 6A, open arrowhead). At least two other HSs are also seen lower on the gel (Fig. 6A, small arrows). The intensity of these HSs is much less than that of the internal control HSs, including the 2°HS3. When the NF-E2, Sp1, and tandem GATA sites are all present within the test sequence (AH5), the same series of very weak HSs is seen again, with the most prominent being positioned over the inserted binding

FIG. 4. Neither NF-E2-nor GATAbinding elements are able to mediate HS formation. A, DNase I HS assays
were performed on pools of 25 MEL cell clones transfected with the indicated constructs. Arrows indicate increasing concentrations of DNase I. Black arrowheads indicate bands associated with the parental XhoI fragment (P) and internal control DNase I HSs. Open arrowheads indicate the location of the introduced factor binding site(s). N indicates the NF-E2-binding element. GG indicates the tandem, inverted GATA-binding elements. Molecular weight markers are as described in the legend to Fig. 2. B, gel mobility shift assays demonstrate appropriate factor binding to artificial (Artif.) GATA binding sites. MEL nuclear extracts were incubated with probes containing the HS4 HSFE tandem GATA-binding elements and with the artificial GATA-binding elements incorporated into the luciferase cDNA. NF-E2 competitor is from the PBGD gene promoter. Gel shift bands corresponding to occupancy of one GATA element (GATA ϫ 1) and both elements (GATA ϫ 2) are indicated. C, gel mobility shift assays demonstrate appropriate factor binding to the artificial NF-E2 binding site. MEL nuclear extracts were used with probes from the PBGD gene promoter containing the 11/11-bp NF-E2 consensus binding site (control) and the artificial NF-E2-binding element, which contained the same NF-E2 sequence mutated into the luciferase cDNA fragment. Bands associated with NF-E2 and AP-1 gel shifts are indicated. sites. Although both artificial elements are able to form HSs, the sensitivity of the HSs is clearly not equivalent to the control 1°HS3 or even the weaker 2°HS3. The addition of NF-E2 and GATA binding sites appears to add little to the nuclease sensitivity produced by the Sp1 binding site alone.
To verify that the artificial Sp1 binding site was actually able to bind Sp1, we performed gel mobility shift assays (Fig. 6B). Gel shift probes containing a previously well characterized Sp1-binding element (35), the Sp1 element from the HS4 HSFE, or the artificial Sp1 element with flanking luciferasederived sequences were tested using nuclear extracts from MEL cells. Each probe produced a similar pattern, forming a single major, specific band. To verify that Sp1 is able to bind each probe, super-shift experiments with an anti-Sp1 antibody were performed. As shown in Fig. 6B, each major band is super-shifted, confirming that the major in vitro DNA binding activity for each of the probes, including the artificial Sp1 site, is due to Sp1.
Identification of Two New Factor Binding Sites within the HS4 HSFE-Researchers from both our laboratory and the Grosveld laboratory previously evaluated the region of the HS4 HSFE for protein binding activity by in vitro DNase I footprint analysis (21,24). In each study only the NF-E2/AP-1-, Sp1-, and tandem GATA-1-binding elements were found. However, our above results suggested that additional binding elements were present and required for the formation of a normal LCR HS core structure. To evaluate this possibility gel mobility shift experiments were performed using MEL and HeLa nuclear extracts and oligonucleotide probes spanning the regions between and flanking the known factor binding sites of the HSFE. The results of these experiments are shown in Fig. 7. Probe 1, from the 5Ј end of the HSFE shows two specific gel shift bands with both MEL and HeLa extracts. Probes 2, 3, and 4 show no specific binding activity. Because an Sp1-like binding site was present at the 3Ј boundary of the HSFE, we also tested a probe that included several bp derived from flanking DNA. This probe also showed two specific bands with MEL and HeLa extracts.
All Identified Factor Binding Sites Are Required for LCR HS Formation-The next step in our analysis was to determine whether the newly identified 5Ј and 3Ј binding elements were important for HS formation. Each of these sequences, alone and in combination, was inserted into the wild-type luciferase construct. As shown in Fig. 8, when these constructs were evaluated for the ability to form nuclease HSs in pools of MEL cell clones, the addition of the 5Ј (AH6) or the 3Ј (AH7) sequence elements did not result in HS formation. We next mutated the 5Ј and 3Ј binding elements into the AH5 construct, which already contained the NF-E2/AP-1-, Sp1-, and tandem GATAbinding elements (AH8 -10, Fig. 8). The addition of either the 5Ј or 3Ј element alone did not lead to HS formation. Only when all identified factor-binding elements were included was a DNase I HS characteristic of the LCR HSs formed. These results indicate that a complex array of cis-acting elements and their associated DNA-binding proteins are required for the formation of the domains of altered chromatin structure that characterize the active elements of the ␤-globin LCR. Also of note is that, in comparison to the internal neo gene HS control, HS3 appears to be more intense in the presence of a functional artificial HS. This may indicate structural indications between the sites.
EKLF and Sp1 Binding Sites Are Present within the 5Ј and 3Ј HSFE Elements-To determine what nuclear proteins might bind to the 5Ј and 3Ј HSFE elements, we performed sequence analysis of these regions with the TRANSFAC program. This revealed relatively low homology Sp1-like binding sequences within both the 5Ј and 3Ј elements. This finding also suggested that the erythroid-specific transcription factor EKLF might bind these sites and assist in HS formation. This is particularly interesting because EKLF has been shown in an in vitro system to mediate HS formation at the ␤-globin promoter (31). To determine whether Sp1 or EKLF was able to bind the 5Ј or 3Ј sequences in vitro, we performed gel shift assays with anti-Sp1 and anti-EKLF (a gift of Dr. J. Bieker) antibodies. In Fig. 9A the results of experiments with MEL nuclear extracts, the 5Ј and 3Ј HSFE probes, and the Sp1 antibody are shown. These assays demonstrate similar banding patterns with the control Sp1 and EKLF probes and the 5Ј and 3Ј oligonucleotides. In each case the upper band is competed by the addition of Sp1 antibody, indicating that this factor is able to bind both of the HSFE elements. In Fig. 9B EKLF antibody is able to compete the lower bands of the EKLF and 5Ј probes but not the 3Ј HSFE probe. The upper band for the 5Ј and 3Ј shift probes is not seen as well in Fig. 9B, because the amount of nuclear extract protein used in this experiment was lessened to increase the sensitivity of the assay to detect inhibition of EKLF binding to the probes. This pattern could also be consistent with the upper band representing an Sp-1-containing complex that is preferentially formed at higher protein concentrations. These results indicate that EKLF is able to bind the 5Ј element but does not appear to bind the 3Ј sequence under the conditions of this assay. DISCUSSION The chromatin structure of the ␤-globin locus undergoes a dramatic reorganization in erythroid cells. This reorganization includes the formation of the nuclease HSs, which delineate the  Fig. 2. B, gel mobility shift assays demonstrate Sp1 binding to the artificial (Artif.) Sp1 site. MEL nuclear extracts are incubated with probes from the artificial Sp1 site, the Sp1 site of the HS4 HSFE, and a control Sp1 site from the ␣2(I)-collagen gene promoter. In the first panel a single major band is produced with each probe. The bands are specifically competed by excess unlabeled self-probe. In the second panel an anti-Sp1 antibody causes a super-shift of the major band for each probe. active elements of the LCR. To begin to understand the formation of the HS core elements and to determine their underlying structure, we have identified the minimal cis-acting elements required for core formation in nuclear chromatin. Although our previous work indicated that NF-E2 and tandem GATA elements are necessary for the formation of normal HS structures, and in vitro studies suggested that GATA-1 and NF-E2 might be sufficient for HS formation, our current studies indicate that reorganization of chromatin structure at the LCR HS cores is more complex than previously thought.
Roles of NF-E2, GATA-1, EKLF, and Sp1 in LCR HS Core Formation-Several studies have now been published on the role of NF-E2 and GATA-1 in LCR HS formation. Our own studies on HS4, HS3, and HS2, in which we have stably transfected mutant HS binding sites into MEL cells, indicate that loss of conserved NF-E2-and tandem, inverted GATA-binding elements significantly reduces or eliminates completely the DNase sensitivity of the HSs, particularly when both elements are mutated within the same construct (14,15). Boyes and Felsenfeld (27) performed similar experiments using chicken ␤-globin 3Ј enhancer sequences. Here too, mutation of NF-E2 and GATA elements resulted in decreased HS formation. However, in both cases, weak HSs were still formed. Our experiments on HS4 indicated that the underlying nucleosomal structure of these very weak sites is unchanged, and Boyes and Felsenfeld (27) concluded that the decrease in HS sensitivity was due to a shift in the probability that the nucleosomal structure of the HS would be reorganized. Gong et al. (29) have studied the role of the tandem NF-E2 binding site of HS2 in HS formation and linked ⑀-globin gene expression. In these exper- Open arrowheads indicate the location of the introduced factor binding sites and the associated DNase I HS. Molecular weight markers are as described in the legend to Fig. 2. A, HS assays performed on constructs containing only the 5Ј element or the 3Ј element inserted into the luciferase wild-type sequence. B, HS assays performed on constructs containing the trans-factor-binding elements NF-E2 (N), Sp1 (S), and tandem, inverted GATA (GG) as well as the 5Ј and 3Ј HSFE elements. Sequences for the constructs and the elements are in Fig. 3.

FIG. 7. Identification of factor binding sites in the 5 and 3 regions of the HSFE.
Gel mobility shift assays were performed with probes from the regions between and flanking the factor binding sites of the HSFE, as indicated by the schematic diagram. All probes were incubated with nuclear extracts from MEL (M) and HeLa (H) cells. Arrowheads indicate specific binding sites observed with the different extracts for probes 1 and 5. The shifted bands were competed with a 200-fold excess of the probe used for the gel shift assay.

␤-Globin LCR HS Core Formation
iments, which were performed in the context of an Epstein-Barr virus-derived minichromosome, mutation of the HS2 NF-E2 element resulted in the loss of both the HS2 and promoter HSs. Armstrong and Emerson (28) have used an in vitro nucleosome assembly assay to show that NF-E2 is able to disrupt an established nucleosomal array and to allow GATA-1 binding to adjacent sites within the HS2 core element. Although the results of each of these studies imply an important role for NF-E2 (or at least NF-E2-binding elements) in LCR HS formation, it is not clear what that role is. Our current study and the study of Armstrong and Emerson both address related questions. Is NF-E2 or an NF-E2-binding element able to independently direct the remodeling of local chromatin structure? Results from the in vitro system clearly demonstrate that NF-E2 is able to disrupt local nucleosomal structure and even promote GATA-1 binding to adjacent sites. A major advantage of this system is that all the molecular entities (DNA sequences, nucleosome composition, DNA-binding factors) are clearly defined and readily manipulated. A potential disadvantage of the system is that it lacks the complexity of in vivo systems, where chromatin structure is studied within its normal, nuclear context. Our results indicate that although NF-E2 binding is able to reorganize nucleosomal structure in vitro, it is not sufficient to do so in vivo.
Results from studies of the role of GATA-1 in LCR HS formation are less variable. Our current results agree with those of Armstrong and Emerson (28), who showed in their in vitro system that GATA-1 is unable to independently remodel local nucleosomal structure. However, we have recently performed experiments that indicate that expression of GATA-1 in HeLa cells results in the formation of ␤-globin LCR HSs (48). 2 Thus, whereas GATA-1 alone is not sufficient for HS core formation, it may be able to interact with non-erythroid factors to initiate LCR HS formation.
EKLF has been implicated in the formation of the DNase I HSs associated with the ␤-globin promoter both in vitro (31) and in vivo (49). This action has been shown to be mediated by interactions with an SWI/SNF-related complex termed EKLF coactivator-remodeling complex 1 (31). The fact that EKLF is able to bind to a portion of the HSFE in vitro that is required for LCR HS core formation raises the possibility that EKLF also recruits EKLF coactivator-remodeling complex 1 or another remodeling complex to the HS cores. EKLF has also been shown to interact with HS3 of the LCR (50), and in EKLF knockout mice HS3 (but not HS4) fails to form (51). The fact that HS4, on which our model HS is based, forms in these mice suggests that although the 5Ј site (which binds EKLF in vitro) is required for HS formation, the related factor, Sp1, is capable of forming the HS4 in the context of the full site. This factor has also been recently shown to recruit an SWI/SNF-related chromatin-remodeling complex to a promoter (52).
Model of LCR DNase I HS Formation-Several models of DNase I HS formation have been proposed (for reviews see Refs. [53][54][55][56][57]. Most of these models involve the binding of specific trans-acting factors to DNA-binding elements, with the resultant displacement, disruption, or specific repositioning of nucleosome cores. As reviewed, the formation of an active chromatin structure is often necessary for normal gene expression Step 1, Sp1 and/or EKLF binding to sites within the HS core region results in the formation of a phased nucleosomal array and a weak DNase HS. Nucleosome remodeling complexes are proposed to be involved in this step. Step 2, this positioning allows the binding of additional factors, including GATA-1 and NF-E2 (or related factors) to the core region, resulting in the stabilization of the nucleosome-free region, as indicated by the formation of a strong DNase I HS.
Step 3, the full complement of factors that comprise the HS core are assembled, resulting in a functional HS core element. Localized histone acetylation mediated by secondary core proteins is also likely to be an important part of the core structure. S, Sp1 binding site; M, maf-related factor or maf response element (MARE) binding site; GG, tandem GATA binding sites.
FIG. 9. Factor binding to the 5 and 3 HSFE elements. A, gel mobility shift assays were performed by incubating MEL nuclear extracts with probes for Sp1, EKLF, and the 5Ј HSFE-and 3Ј HSFE-binding elements. The black arrowhead indicates the specific Sp1 band that is competed by the Sp1 antibody. B, gel mobility shift assays to determine whether EKLF binds to the 5Ј or 3Ј element. MEL nuclear extracts were incubated with probes for EKLF and the 5Ј HSFE-and 3Ј HSFE-binding elements. The black arrowhead indicates the specific EKLF band that is competed by the EKLF antibody.
␤-Globin LCR HS Core Formation and, in some cases, appears to allow the binding of a full complement of trans-acting factors to a regulatory element. The DNase I HSs of the ␤-globin LCR are unique because they are erythroid-specific, developmentally stable, and more sensitive to DNase I than the promoter-associated HSs of the locus. These HSs offer an opportunity to study the mechanism by which a functionally important, complex domain of locally altered chromatin structure is formed. A model of LCR HS formation based on current data is presented in Fig. 10. We propose that the formation of the LCR HSs is a multistep process and that the first step is the binding of non-erythroid factors such as Sp1 to the core-forming element. Supporting data for this function of Sp1 come from in vitro studies from Workman and co-workers (58) showing that Sp1 is able to bind to nucleosome-associated DNA and from Jongstra et al. (59) and Widlak et al. (60), who demonstrated that Sp1 participates in the formation of the nuclease HSs of the SV40 early gene promoter and human immunodeficiency virus, type I LTR. Data from Pondel et al. (61) indicate that Sp1 is critical for the chromatin-dependent expression of the human ␣-globin gene, and work from Marin et al. (62) shows that embryonic globin genes were expressed at reduced levels in Sp1 homozygous knockout mice. The establishment of the core domains is likely to involve the action of nucleosome-and/or chromatin-remodeling factors such as SWI/SNF and NURF (reviewed in Refs. 57,63,64). Recently, Sp1 has been shown to interact with an SWI/SNF-related complex, providing a direct link between this factor and nucleosomal remodeling (50). Our own data showing weak HS formation with an Sp1 site alone (Fig. 6) are consistent with the idea that binding of Sp1 may be an important initial step in HS core formation. Although we did not specifically test the ability of all three Sp1-binding elements to direct HS formation, it may be that they can act to synergistically open the nucleosomal structure of the HS. EKLF may also participate in HS core formation at this level, as it seems to do at the ␤-globin promoter. The next step in our model is the binding of the hematopoietic factors NF-E2 (or other related maf response element-binding proteins) and GATA-1. Data indicate that these factors are unable to mediate core formation themselves but act to stabilize the structure. The final step involves the recruitment of non-DNA-binding core factors (i.e. FOG-1, histone deacetylases, etc.), which interact with the DNA-bound factors, resulting in the formation of the full multiprotein LCR core structure, which is able to functionally interact with other elements of the same HS, other LCR HS cores, the distant globin genes, and other elements of the locus. For example, the core may mediate local histone acetylation, allowing the many factors that bind to regions flanking the core access to their sites. Such acetylation in the region of the LCR HS2 has recently been demonstrated by Forsberg et al. (65). Although the exact physiologic roles of the ␤-globin LCR HS elements remain to be determined, further understanding of the higher order structures and individual components of these elements should provide insight into their functions.