GATA Zinc Finger Interactions Modulate DNA Binding and Transactivation

GATA-1 and other vertebrate GATA factors contain a DNA binding domain composed of two adjacent homologous zinc fingers. Whereas only the C-terminal finger of GATA-1 is capable of independent binding to the GATA recognition sequence, double GATA sites which require both fingers for high affinity interaction are found in several genes. We propose a mechanism whereby adjacent zinc fingers interact to influence the binding and transactivation properties of GATA-1 at a subset of DNA binding sites. Using two such double GATA sites we demonstrate that the N-terminal finger and adjacent linker region can alter the binding specificity of the C-terminal finger sufficiently to prevent it from recognizing some consensus GATA sequences. Therefore, the two zinc fingers form a composite binding domain having a different DNA-binding specificity from that shown by the constituent single C-terminal finger. Furthermore, we compare two of these double sites and show that high affinity binding of GATA-1 to a reporter gene does not necessarily induce transactivation, namely, the sequence of the DNA binding site can alter the ability of GATA-1 to stimulate transcription.


INTRODUCTION
GATA factors are widely distributed, and targeted gene inactivation experiments in mice have shown that five of the six vertebrate GATA family members are essential for embryonic viability (1 -6). While the vertebrate GATA factors contain two zinc fingers (Figure 1; 7, 8), GATA proteins with only a single finger, most similar to the vertebrate Cterminal finger, exist in fungi and worms (9 -11). GATA-1, the original isolate, is expressed in erythroid cells, megakaryocytes and mast cells and is required for correct regulation of virtually all erythroid-specific genes (1). These target genes have distinct temporal patterns of expression during erythropoiesis, and some genes are inhibited rather than activated by GATA-1. The basis for these dissimilar expression patterns, which is not yet fully understood, may involve differences in the DNA recognition sequences.
The GATA transcription factor family is defined by a zinc finger of the form CX 2 C-X 17 -CNAC and a basic linker region which is required by GATA-1 for DNA binding (7,8,12,13,14). The C-terminal finger (CF 1 ) of GATA-1 is both necessary and sufficient for DNA binding to the GATA recognition sequence (WGATAR, 15,16). The N-terminal finger (NF 2 ) shows no independent DNA binding. It does, however, stabilize GATA-1 interactions with some naturally occurring double GATA sites that are critical for gene expression (17)(18)(19)(20)(21). Although the GATA motifs within these high affinity sites are found in diverse orientations and spacing, GATA-1 binds as a monomer. This suggests site-specific variability in the position and orientation of the two GATA-1 zinc fingers on DNA.
Some important binding sites that require both GATA zinc fingers for strong interaction are conserved among species. An overlapping palindromic GATA sequence (GATApal) with an adjacent inverted single site is found in at least three vertebrate hematopoietic GATA-1 by guest on July 24, 2018 http://www.jbc.org/ Downloaded from promoters. Notably, this site is not found in most GATA regulated genes (21). A double site consisting of non-overlapping direct repeats is also conserved in the testis GATA-1 promoters of mice and rats (22). Both of these sequences are essential for promoter activity and their evolutionary conservation implies that they have a significant and specific function.
Here we identify two additional sites that require the N-terminal finger of GATA-1 for high affinity binding. Using derivatives of these sites we show that the N-terminal finger of GATA-1 can stabilize binding, disrupt binding, or modify the DNA binding specificity of the C-terminal finger. We compare the binding properties of one of these sites, a nonconsensus GATC recognition sequence, with binding at canonical (GATA) sites and demonstrate a distinct type of binding to some GATC containing probes by several criteria, in spite of similar binding affinities. We further show that GATA-1 activity can be influenced by its mode of binding: whereas GATA-1 is unable to induce transactivation through the GATC1 binding site, transactivation does occur through several GATA sites placed in the same position on a minimal promoter. Thus the ability of GATA-1 to mediate transactivation can be modified by its DNA binding sequence. These observations provide a new mechanism through which 7 acids 253 to 317 of hGATA-1, (24) into pET-11d (Novagen) digested with NcoI and BamHI. The PCR primers (5'   Expression and purification were carried out as previously described (16,21), except that the HPLC purification step was omitted for human GATA-1 CF peptide: active fractions from the S-column were pooled, adjusted to 10% glycerol and stored at -80 0 C. Peptide concentrations were determined by SDS-PAGE and coomassie staining, and confirmed by evaluation of the binding activity with saturating amounts of probe.
Nuclear extracts and whole cell extracts containing GATA-1 and GATA-2 were made by standard procedures (25,26).

Electrophoretic mobility shift assay (EMSA). DNA and peptides
were incubated in 10 µl of a solution containing 50 mM Tris (pH = 7), 0.0125% triton, 3.2% ficoll, 0.2 mM EDTA and 4 µg/ml of poly (dI . dC) for 15 minutes and electrophoresed in an 8% polyacrylamide gel in 10 mM Hepes, 10 mM Tris and 1 mM EDTA all at room temperature. Binding assays with nuclear extracts were carried out with standard procedures as previously described (21). Oligonucleotides were purchased from Operon Technologies and purified by polyacrylamide gel electrophoresis. performed according to manufacturer's instructions, and a liquid scintillation assay was used to determine CAT activity (27).

RESULTS
To study the contributions of the individual GATA zinc fingers to DNA binding, we have generated a series of single and double zinc finger peptides and N-terminal fusion proteins by expression in E. coli (Figure 1 Table   1.

A conformational change is induced when the N-finger of GATA-
double GATA recognition sequences migrate faster in electrophoretic mobility shift assays (EMSA 7 ) than hDF peptide complexes formed with probes containing only single GATA binding sites (21). These migration differences are exemplified by the binding of hDF peptide to the double (GATC1) and single (GATApal M1) site probes in Figure 2. Fast complex migration was observed with all of the double sites tested ( Table 1, data not shown), indicating that it is likely a general feature of hDF peptide complexes in which the GATA-1 N-finger participates in DNA binding. In an accompanying manuscript we show that this altered mobility is not due to differential DNA bending (23). The hDF peptide evidently adopts a different conformation as a result of the N-finger interaction with DNA ( Figure 2 and data not shown).
Binding interference between GATA zinc fingers. Two double GATA binding sites, the GATC1 site mentioned above and the ε-globin gene silencer site (εS) were characterized to demonstrate the involvement of both GATA-1 zinc fingers in binding. A series of mutations were generated for each of these binding sequences to determine the requirements for high affinity interaction with GATA-1. These experiments led to the surprising observation that the N-finger of GATA-1 can prevent the C-finger from binding to certain DNA sites. A description of these studies is presented below.
The -globin gene silencer. The human ε-globin gene silencer has three potential overlapping GATA recognition sequences (εS, Table 1,) which lie 258 bp upstream of the transcription initiation site (29,30). To determine whether both zinc fingers of GATA-1 interact with this site, dissociation constants for the single (CF) and double finger (DF) peptides were determined by EMSA on εS and derived mutant sites ( Table 1). All peptides, as well as the wild type protein, bound tightly to εS. Only single complexes were observed indicating that two protein molecules do not bind to εS simultaneously. In order to identify the DNA bases that participate in binding , a series of εS mutations, which removed the GATA sites, were tested. Removal of GATA site 1 (M1) had no effect on the binding of CF peptide, DF peptide or full-length GATA-1 (Table 1).
However, mutation of either site 2 (M2) or 3 (M3) reduced the binding affinity of full-length GATA-1 (data not shown) and of hDF peptide, while only mutation of site 3 (M3) altered CF peptide binding. Therefore, εS site 3 is the preferred site for CF peptide and the binding specificity of CF and DF peptides are distinct. Thus in the 1:1 complex formed, hDF peptide by guest on July 24, 2018 http://www.jbc.org/ Downloaded from interacts with both sites 2 and 3 concurrently, suggesting that both zinc fingers interact with the DNA to generate strong binding.
To confirm the involvement of both binding sites in the interaction of hDF peptide with εS, we tested double mutations (εSM1, 2 and εSM2, 3) which lead to single site probes. CF peptides bind well to both probes (Table 1), however the binding of wild type GATA-1 was abolished (Table   1). While a lower affinity of GATA-1 for these single site probes was anticipated, the complete lack of detectable binding was surprising.
Therefore, CF peptides bind with higher affinity to these single site probes than does full-length GATA-1, indicative of interference by another region of GATA-1. To localize the cause of this inhibition we also tested the binding of hDF peptide to these probes. Like the native protein, hDF peptide (which binds strongly to εS) fails to bind εSM1, 2 or εSM2, 3 ( Table 1). As shown in the accompanying manuscript, MBP fusions with the double fingers of chicken and human GATA-1 also bind well to εS (23), but fail to bind to εSM1, 2 and εSM2, 3 (data not shown). We conclude that the presence of the GATA-1 N-terminal finger (finger plus adjacent arm) is sufficient to prevent the C-finger from binding with high affinity to some DNAs. The CF peptide alone binds strongly to εSM1, 2 and εSM2, 3, the DF peptide, MBP-DF fusion proteins and wild type GATA-1, not at all.

GATA-1 binding to GATC sequences.
To ascertain whether a similar binding inhibition or modulation occurs with other DNA sites and GATA factors, we used a series of probes derived from (not identical to, see Table 1) those selected by GATA-2 and -3 in a site selection study (31).
These probes contain a non-canonical motif having an AGATCTTA consensus. Unlike GATA-1, the N-terminal fingers of GATA-2 and -3 can bind independently to various DNA sites that include this consensus (28).
In fact, the cGATA-2 NF, cGATA-3 NF and cGATA-1 CF peptides all bind with similar affinities to many such DNAs (for example GATC1, -3 and -4, Table 1; 28). The advantage of these probes over canonical binding sites is that they allow us to examine the reciprocal influence of the GATA-2 C-finger on N-finger binding (in addition to the effects of NF on CF binding).
A 26 base pair probe (GATC1, Table1) derived from a larger sequence that was selected by GATA-2 and -3 (31) was used. As shown in Figure 2 and Table 1, it contains adjacent GATA and GATC binding sites. To characterize GATA-1 binding to this DNA, mutational analyses were performed. Alteration of the GATA site 5' of the GATC motif did not modify the binding of cGATA-1, cGATA-2, or hDF peptide (GATC2, Table 1 and 31). However, removal of five bases from the 5' end of GATC2 did by guest on July 24, 2018 http://www.jbc.org/ Downloaded from abolish the binding of all three of these proteins (GATC3 and GATC4, Table   1). The four most 3' bases were dispensable for binding (GATC5 and GATC6, Table 1), but the cytosines within the palindromic GATC site were essential (GATC7, Table 1). In marked contrast to these results, GATA-1 CF peptides bind to all of these probes except GATC7 (Table 1).
We note that all GATC probes used here, except for GATC7, contain a second consensus sequence, T/GAAG, previously identified as a site for GATA-1 NF (AGATCTTA, 32). While this sequence may contribute to the binding of DF peptide and GATA-1 to these probes, it is not sufficient, as no complexes are formed with GATC3 and -4 (which contain it).
In summary, while a canonical single GATA site is only 6 bp in length, our mutational analysis has shown that binding of GATA-1 and hDF peptide to GATC2 requires sequences that are at least 14 base pairs apart.
Surprisingly, it does not require the canonical GATA sequence (present in GATC1).
We found it noteworthy that all of the double finger proteins tested above (i.e. GATA-1, GATA-2 and hDF peptide) failed to bind to GATC3 or GATC4, even though the GATA-1 CF and cGATA-2 NF peptides bind well.
MBP-cDF also does not bind to these probes (data not shown). Thus the presence of a second zinc finger is sufficient to inhibit binding, this time to a GATC sequence, even though the double finger proteins bind strongly to other GATC probes (GATC1, -2, -5, and -6, Table 1). Sequences flanking the consensus evidently contribute to the binding properties observed here and in (31). Analogously, the C-finger of chicken GATA-2 interferes with binding of the N-finger. While cGATA-2 NF binds to GATC3 and -4 (Table 1; 28), it is unable to do so when within the context of the wild type protein (Table 1), or cGATA-2 double finger peptide (data not shown, amino acids 277 to 396). Therefore, the native arrangement and linkage of the fingers of GATA-1 and GATA-2 alters their DNA binding specificity. Some single sites are structured such that the two-finger combination binds well (e.g. WGATAR) whereas other DNA sites (e.g. εSM1, 2 and εSM2, 3, GATC3 and -4) do not allow for a stable interaction with the double finger peptide (or proteins), GATA-1 or GATA-2.

Covalent linkage of the N-and C-fingers is required for their interaction.
To determine whether the GATA fingers must be linked to influence one another, we performed mixing experiments with GATA-1 CF and NF. MBP-hNF, hNF, MBP-cCF and CF (h or c) peptides were tested individually and together with the GATApal, GATC4, εSM1, 2 and εSM2, 3 probes ( Figure 3A and B, and data not shown). In all cases the presence of by guest on July 24, 2018 http://www.jbc.org/ Downloaded from MBP-hNF or hNF showed no influence on complex formation. We therefore conclude that the GATA-1 fingers must be covalently linked to influence one another positively (GATApal) or negatively (GATC4, εSM1, 2 and εSM2, 3).

GATA-1 DF peptide interactions with GATC probes are not equivalent to those with canonical binding sites.
As noted above, GATC1 is a site to which both GATA fingers bind, because it is fast migrating when complexed with hDF peptide. We initially assumed that this was due to the presence of both the GATA and the GATC consensus sequences on this probe. However, the GATC2 complex (which lacks the GATA site) is also fast migrating (data not shown), suggesting that the N-finger interacts with a non-canonical sequence within GATC2. To compare GATA protein interactions at GATC containing sites with those at canonical binding sites, C to A substitutions (CA) in GATC1 and GATC2 were tested with hDF peptide. With these mutations the fast migrating GATC1 complexes are partially converted to slower migrating forms, consistent with binding through a single finger and site (CA1 and CA2 probes, Figure 4A). Surprisingly, modifying these sequences to ones more similar to the WGATAR consensus reduced the overall binding can also lead to more subtle changes in specificity. A complex between cCF peptide and a palindromic GATA site (GATApal) is competed effectively by a single GATA sequence (ß/ε), but not as efficiently by GATC1 ( Figure 4B). In contrast, the DF peptide complex with the same probe displays the opposite competition pattern ( Figure 4B), namely the ß/ε single GATA site is a much less effective competitor than GATC1 when NF and CF are joined. Therefore the two zinc fingers act in concert and the presence of the N-finger modifies the inherent specificity of the C-finger.  Figures   5A, 5B), but the K d for this interaction is reported to be 0.78 nM (28). CF peptide may therefore require tighter binding affinities than DF for footprint formation, or may bind to GATC1 in a manner that still allows DNase I access to the DNA. In summary, cCF peptide binds to, but is unable to protect the GATC1 DNA, however the addition of the N-finger (DF peptide) results in protection from DNase I.
Taken together these results confirm that GATC1 contains an extended compound site on which both GATA-1 fingers participate in complex formation. Thus GATA-1 binding to GATC1 is demonstrably different than to the consensus (GATA) recognition sequence by several criteria, suggesting that the mode of binding to these two types of sequences is distinct, and results in different conformations of the bound protein. To test this premise we used a reporter construct containing a minimal promoter with a double GATA site (pαD3, Figure 6A), previously employed to show that GATA-1 is a transactivator (18). It was shown that co-expression of GATA-1 with this reporter stimulated its transcription approximately 100 fold, whereas the same reporter gene with mutated GATA sites (pαD4) led only to a 5-fold stimulation. This activation is not absolutely dependent on the position or orientation of the GATA sites. For example, a pαD4 derivative (pαD4/5, Figure 6A were all active in this assay (data not shown).
In this study GATC1 was compared with a canonical high affinity double GATA site, mGATApal. Both were placed in pαD4 at the distal position and tested for transactivation by cGATA-1. Reporter plasmids with one copy of each insert in either orientation and two or four adjacent copies were compared ( Figure 6B). Although these two binding sites do not differ substantially in affinity, significant differences in transactivation were observed. As expected, all constructs containing GATApal (K d = 8 nM) were activated by GATA-1. In contrast, all GATC1 containing plasmids (K d = 5 nM) yielded transactivation levels similar to the negative control, pαD4 ( Figure 6B). This lack of transactivation is not specific to fibroblasts as the GATC1 containing plasmids were much less active than GATApal containing plasmids in an erythroid precursor cell line that continues to rise ( Figure 6C). Thus, the orientation of the GATA sequences (direct repeat or palindrome) relative to one another influences transactivation, and the degree of activation does not correlate with binding affinity. In summary, strong binding of GATA-1 to a reporter gene does not necessarily lead to transactivation, and the GATA-1 protein is transcriptionally inactive in response to the GATC1 sequence. These data suggest that DNA can modulate GATA-1 activity in a sequence dependent manner, by changing its conformation.

Transactivation is restored through C to A substitution in GATC
sites. This model of GATA-1 action predicts that a GATA-1 protein lacking the N-terminal zinc finger might stimulate transcription of the GATC1 containing plasmids, even though the wild type protein does not.
Because the N-finger induces the conformational changes observed here, and is required for the distinct type of interaction seen between GATA-1 and the GATC1 sequence, inactivation of the N-finger might restore the "normal" interaction. Consequently we tested cGATA-1 mutants in which the first two cysteines of the N-terminal finger were replaced (NF -GATA-1): these mutants have previously been shown to bind to and activate GATA dependent reporter genes (13).
In our hands NF -GATA-1 did activate pαD3 (canonical binding site, Figure 6A), but the GATC1 reporter plasmids were not stimulated (data not shown). However EMSA analysis shows that the NF -GATA-1 mutants do not bind to GATC1 or εS oligonucleotides, even though they do bind to other canonical GATA sequences ( Figure 7A and data not shown). These binding studies confirm, once more, the involvement of both zinc fingers in GATA-1 interactions with GATC1 and εS.
We note that as with the DF peptide, binding of the N-finger of fulllength cGATA-1 to DNA also causes fast migration of the protein/DNA complex. Double site probes such as GATApal lead to faster migration than single site probes such as GATApal M1 with wild type cGATA-1, but NF -cGATA-1 (HB2) complexes with both probes migrate slowly and at identical positions ( Figure 7A). Therefore, the conformational change and altered migration induced by N-finger binding (Figure 2) is not specific to the GATA-1 DF peptide but also occurs with the native protein.
In an alternate approach, to try to restore transactivation to the GATC driven reporter genes, we used the C to A substitution binding sites, CA1 and CA2 (shown in Table 1 and Figure 4A), in the transactivation assay. We reasoned that if the lack of activation observed with the GATC1 site was due to the conformation of GATA-1 on the DNA, then restoring the "normal" conformation by converting GATC1 to a canonical binding site (GATA) might restore transactivation. This is in fact the case. Even though the C to A substitutions (CA1 and CA2) result in reduced binding affinity relative to GATC1, pαD4 plasmids containing these sequences are activated by GATA-1 ( Figure 7B A second new attribute of GATA-1, namely the inability to activate transcription at a high affinity-binding site, is also described here. In an accompanying manuscript we show that this DNA dependence of activation does not involve site specific DNA bending (23). DNA induced modulation of transactivation has also been seen with the steroid hormone receptors which are allosterically regulated by DNA. Activation, repression, or lack of a response can occur upon DNA and ligand binding to these receptors, all dependent on the binding sequence rather than the affinity of the interaction: the DNA induces conformational changes that specify the activity of these receptors (33). Recently a similar conclusion was reached in studies with nuclear factor-κB (NF-κB): it was suggested that DNA acts as an allosteric effector of this transcription factor (37). Here we show that GATA-1 activity can also be influenced by the sequence of its cognate DNA site. Alterations in binding specificity mediated by adjacent zinc fingers are somewhat unusual; many studies indicate that while tandem Cys 2 His 2 zinc fingers can cooperate with one another to stabilize DNA binding, they retain their individual specificities when separated by only seven amino acids (48,49). The N-terminal finger of GATA-1 participates in DNA binding (13,17,32,50), but previous studies did not distinguish between contributions mediated by each finger acting independently and the formation of a composite binding domain. We have shown that the GATA         Table 1. Arrows indicate full-length GATA-1 while the arrowhead indicates a degradation product. The wild type GATA-1 complex with GATApal is fast migrating relative to the single site complex (GATApalM1) and to the NF -GATA-1 (HB2) complexes.

(B)
The CA1 and CA2 oligonucleotides (Table 1 and Figure 4A) were inserted into the Spe I site of pαD4 in 1, 2 or 4 copies and tested in the transactivation assay as described in Figure 6. The number following the construct name indicates the copy number of the oligonucleotide insert in each construct.

GATA-2 N-finger (cGATA-2 NF)
This amino acid is a C in the native protein.