ABSTRACT
The ability of a transcription factor to locate and bind its cognate DNA site in the presence of closely related sites and a vast array of non-specific DNA is crucial for cell survival. The CREB/ATF family of transcription factors is an important group of basic region leucine zipper (bZIP) proteins that display high affinity for the CRE site and low affinity for the closely related AP-1 site. Members of the CREB/ATF family share in common a cluster of basic amino acids at the N-terminus of their bZIP element. This basic cluster is necessary and sufficient to cause the CRE site to bend upon binding of a CREB/ATF protein. The possibility that DNA bending and CRE/AP-1 specificity were linked in CREB/ATF proteins was investigated using chimeric peptides derived from human CRE-BP1 (a member of the CREB/ATF family) and yeast GCN4, which lacks both a basic cluster and CRE/AP-1 specificity. Gain of function and loss of function experiments demonstrated that the basic cluster was not responsible for the CRE/AP-1 specificity displayed by all characterized CREB/ATF proteins. The basic cluster was, however, responsible for inducing very high affinity for non- specific DNA. It was further shown that basic cluster-containing peptides bind non-specific DNA in a random coil conformation. We postulate that the high non- specific DNA affinities of basic cluster-containing peptides result from cooperative electrostatic interactions with the phosphate backbone that do not require peptide organization.
Eukaryotic transcription factors in the basic region-leucine zipper (1 ) (bZIP) family employ an extremely simple structural motif to accomplish the sequence-specific recognition of duplex DNA (2 ). The DNA binding activity of a bZIP protein is localized within 60 contiguous amino acids termed the bZIP element (3 ; Fig. 1 ). Each bZIP element contains three segments that each play a specific role in DNA recognition. A basic segment of ~20 amino acids near the N-terminus of the element contains residues that contact DNA directly (4 -7 ), while a zipper segment of ~25 amino acids orchestrates the formation of a coiled coil dimer from two protein monomers (4 -6 ,8 ). The basic and zipper segments are connected by a six residue spacer segment whose sequence is not conserved amongst members of the bZIP family (9 ).
Although bZIP proteins are structurally simple, they recognize a set of inverted half-sites of widely varying sequence. In addition to their ability to discriminate between target sites that differ in half-site sequence, bZIP proteins also discriminate between target sites of identical half-site sequence but different half-site spacing (half-site spacing specificity) (10 ). For example, bZIP proteins related to the oncogene products Fos and Jun (AP-1 family) prefer the pseudosymmetric 9 bp AP-1 target site (ATGACTCAT) (11 ,12 ), whereas those related to CREB and ATF-2 (CREB/ATF family) prefer the symmetric CRE target site (ATGACGTCAT) (13 ) in which the same inverted pair of half-sites is separated by 2 bp. The yeast bZIP protein GCN4 binds both sites with comparable affinity (14 ). In spite of a large body of research on bZIP-DNA interactions, the determinants of CRE/AP-1 selectivity displayed by CREB/ATF proteins are not well understood.
Many, if not all, CREB/ATF family members share in common the ability to bend their specific CRE target site (15 -18 ). Because the CRE site bends intrinsically toward the major groove, interaction with a CREB/ATF protein results in an overall straightening of the DNA. It has been demonstrated that bending of the CRE site by the CREB/ATF protein CRE-BP1 requires two basic residues (the basic cluster) located at the N-terminus of the basic segment (18 ). Bending is proposed to result from neutralization by these two basic residues of two symmetry-related phosphate linkages on a single face of the DNA helix. The resulting asymmetric charge distribution bends the DNA spontaneously in the direction of the neutralized phosphates, as predicted by Mirzabekov and Rich (19 ) and demonstrated in model systems by Strauss and Maher (20 ). Because all members of the CREB/ATF bZIP family studied discriminate effectively between the CRE and AP-1 target sites and also bend DNA, we wondered whether CRE/AP-1 specificity and DNA bending were linked, i.e. whether DNA bending resulted in high CRE/AP-1 specificity. We demonstrate here that it does not. The presence of the basic cluster in native and chimeric bZIP peptides does not result in selective binding to the CRE site. The basic cluster does, however, have a dramatic effect on the ability of the bZIP peptide to distinguish specific DNA in the presence of competing non- specific sequences.
The chimeric bZIP peptides used in these experiments (Fig. 1 ) were reported previously (10 ,18 ) and, with the exception of g5c, were purified as described. g5c was subjected to additional purification by reverse phase HPLC using a semipreparative Vydac C-18 column and a gradient of 98-20% water in acetonitrile containing 0.05% TFA over 40 min at a flow rate of 4 ml/min. The identities of all peptides were confirmed by amino acid analysis and either MALDI or electrospray mass spectrometry.
The CRE and AP-1 affinities of peptides g5c, c5g and c10g were determined by use of an electrophoretic mobility shift assay employing binding buffer (1.4 mM KH2PO4, 4.3 mM Na2HPO4, 2.7 mM KCl, 137 mM NaCl, 1 mM EDTA, 1 mM DTT, 0.1% NP-40, 400 [mu]g/ml BSA and 5% glycerol, pH 7.4)and oligonucleotides, at a final concentration of <50 pM, containing the CRE or AP-1 site (CRE24 or AP-123) reported previously (10 ). The conditions and data analysis were as used previously to determine the specificity of the peptides ccc and ggg (10 ). c5g and c10g bound well to both CRE24 and AP-123 at 25oC, whereas g5c bound to both only at 4oC. Complexes of peptides with non-specific DNA were not stable under the described gel electrophoresis conditions; neither specific bands nor well shifts were detected. The affinities of the chimeric peptides (Fig. 1 ) for non-specific DNA were therefore measured by use of a competition electrophoretic mobility shift assay in which the fraction of [32P]CRE24 bound to a given peptide was monitored as a function of the concentration of either Non24 (AGTGGAGTAAGGCCTATCTCGTGC), calf thymus DNA (CTD) (Gibco BRL) or poly(dI-dC)[middot]poly(dI-dC) (Pharmacia). Peptide concentrations were chosen such that maximal binding was achieved in the absence of competitor DNA. The concentrations used were as follows: ggg, 9 nM; ggc, 20 nM; gcc, 15 nM; c5g, 30 nM; c10g, 60 nM; ccc, 90 nM; ccg, 60 nM; cgg, 120 nM; g5c, 60 nM. Reactions were incubated at 25oC in binding buffer. Addition of 10 mM MgCl2 to the binding buffer did not affect the affinity of c5g for Non24 (data not shown). Hence, there was no effect on specificity when a divalent cation was present.
Direct electrophoretic mobility shift data were analyzed as described previously (10 ). This assay monitors the equilibrium shown in Scheme 1 in which two unfolded bZIP peptide monomers (U) are converted into a single, dimeric DNA complex (A2O).
Circular dichroism and fluorescence experiments were performed in CD buffer (1.4 mM KH2PO4, 4.3 mM Na2HPO4, 2.7 mM KCl, 137 mM NaCl, 1 mM EDTA and 1 mM DTT, pH 7.4), which contains the same salt concentration as binding buffer. Circular dichroism experiments were conducted at 25oC on an Aviv 62DS spectrometer using a 1 mm path length cell. Fluorescence experiments were conducted at 25oC on a SLM Aminco 4800s spectrofluorimeter using a 3 * 3 mm cell. The intrinsic fluorescence of the unique tryptophan residue in the spacer region of ccc was monitored between 300 and 400 nm upon excitation at 292 nm, a wavelength chosen to reduce inner filter effects from the DNA.
We made use of a series of chimeric bZIP element peptides containing sequence from CRE-BP1 and GCN4 (Fig. 1 ) to examine the role of the CRE-BP1 basic cluster in CRE/AP-1 selectivity. Peptide c5g contained the N-terminal basic cluster (KRRKF) of CRE-BP1 in the context of sequence from GCN4 (21 ); g5c contained the corresponding five residues from GCN4 (SAALK) in the context of sequence from CRE-BP1 (22 ). The GCN4 and CRE-BP1 bZIP element peptides are referred to as ggg and ccc respectively.
First we compared the CRE/AP-1 specificities of peptides ggg and c5g to determine whether the basic cluster of CRE-BP1 was sufficient to induce preferential recognition of the CRE site. The free energy of the c5g[middot]CRE24 complex, [Delta]Gcre = -22.6 kcal/mol, was essentially equivalent to the free energy of the c5g[middot]AP-123 complex, [Delta]Gap-1 = -22.5 kcal/mol. The free energy of the ggg[middot]CRE24 complex, [Delta]Gcre = -23.7 kcal/mol, was approximately equal to the free energy of the ggg[middot]AP-123 complex, [Delta]Gap-1 = -24.1 kcal/mol (10 ). Thus both ggg and c5g bound equally well to oligonucleotides containing the CRE and AP-1 sites CRE24 and AP-123 (Fig. 2 ). The peptide c10g, in which 10 N-terminal residues of ggg were replaced with the corresponding sequence from CRE-BP1 (KRRKFLERNR), also exhibited little selectivity for one target site over the other; the free energy of the c10g[middot]CRE24 complex, [Delta]Gcre = -20.6 kcal/mol, was almost identical to the free energy of the c10g[middot]AP-123 complex, [Delta]Gap-1 = -20.5 kcal/mol. These results indicate that the CRE-BP1 basic cluster is not sufficient to encode for preferential recognition of the CRE site.
Table 1
Next we asked whether the presence of the CRE-BP1 basic cluster influenced affinity for non-specific DNA. To measure non-specific DNA affinity, we made use of a competition assay in which the fraction of [32P]CRE24 bound to a given peptide was monitored as a function of the concentration of non-specific DNA added to the reaction (23 ,24 ). The non-specific DNA used for this purpose, Non24, contained a sequence identical to CRE24 except that the 10 bp of the CRE site were scrambled.
Both ccc and c5g displayed surprisingly high affinities for Non24 (Table 1 ). The stability of the ccc[middot]Non24 complex, [Delta]Gnon = -20.4 kcal/mol,was almost identical to the stability of the ccc[middot]CRE24 complex, [Delta]Gcre = -20.5 kcal/mol, and the stability of the c5g[middot]Non24 complex, [Delta]Gnon = -21.8 kcal/mol, was only 0.6kcal/mol less than the stability of the c5g[middot]CRE24 complex. Thus, there is virtually no free energy difference between the specific and non-specific complexes for ccc and c5g. Although there are potentially multiple non-specific binding sites on Non24 and a commensurate statistical advantage for the composite site over a single specific site, we compare overall affinities of the peptides for two oligonucleotide sequences of the same length, one which contains a specific site and one which does not.
In contrast to results obtained with peptides ccc and c5g, the analogous peptides ggg and g5c, which do not contain the basic cluster, bound more poorly to non-specific DNA, Non24, than to specific DNA, CRE24 (Table 1 ). The free energy of the ggg[middot]Non24 complex, -20.1 kcal/mol,was 3.5 kcal/mol less favorable than the free energy of the ggg[middot]CRE24 complex and the free energy of the g5c[middot]Non24 complex, -19.6 kcal/mol,was 2.8 kcal/mol less favorable than that of the g5c[middot]CRE24 complex. Thus, the presence of the basic cluster in ccc decreases the specificity for CRE24 over Non24 by 2.9 kcal/mol relative to g5c and the absence of the basic cluster in ggg increases the specificity for CRE24 over Non24 by 2.9 kcal/mol relative to c5g. This dramatic change in the specificity of these peptides for their target site relative to non-specific DNA requires only the five residues of the CRE-BP1 basic cluster.
The CRE24 and Non24 affinities of chimeric peptides in which whole segments of the GCN4 and CRE-BP1 bZIP elements were interchanged (Fig. 1 ) followed, with two exceptions, the same trend (Table 1 ). When the GCN4 basic region, which lacks a basic cluster, was fused to the CRE-BP1 spacer and zipper segments, the result was a peptide (gcc) that preferred the CRE24 sequence to the Non24 sequence by >3 kcal/mol. Conversely, when the CRE-BP1 basic segment was fused to the GCN4 spacer and zipper segments, the result was a peptide (cgg) that preferred the CRE24 sequence to the Non24 sequence by only [Delta][Delta]Gnon/cre = 0.3 kcal/mol. Peptide ccg, which contains the basic and spacer segments of CRE-BP1 fused to the zipper segment of GCN4, as expected, displayed a high degree of non-specific binding ([Delta][Delta]Gnon/cre = 1.1 kcal/mol). Peptides c10g and ggc behaved anomalously. While one contained a basic cluster and one did not, they displayed [Delta][Delta]Gnon/cre = 2.0 and 1.8 kcal/mol respectively, directly between the two extreme cases.
To test the abilities of both large pieces of random DNA and also a simple repeating DNA polymer to compete with CRE24 for specific binding, we measured the affinities of ggg, ccc and c5g for both calf thymus DNA (CTD) and poly(dI-dC)[middot]poly(dI-dC) (Fig. 3 ). The abilities of these large DNAs to compete with [32P]CRE24 was compared with the ability of unlabeled CRE24 to compete with itself. The data show that ccc and c5g bound CTD and poly(dI-dC)[middot]poly(dI-dC) approximately as well as they bound CRE24, whereas ggg did not. There was a 3.3 kcal/mol difference between the stabilities of both the ggg[middot]CTD and ggg[middot]poly(dI[middot]dC) complexes and the stability of the ggg[middot]CRE24 complex. The corresponding complexes of c5g differed by 1.5 and -0.1 kcal/mol respectively from the stability of the c5g[middot]CRE24 complex. The simple repeating polymer poly(dI[middot]dC) competed as effectively for c5g as DNA containing a specific binding site. A similar difference between specific and non-specific binding was seen in the case of ccc, which, as expected from the earlier results, displayed low specificity. While the difference in stability between the ccc[middot]CTD complex and the ccc[middot]CRE24 complex was a moderate 2.0 kcal/mol, there was no difference between the stabilities of ccc[middot]poly(dI[middot]dC) and ccc[middot]CRE24 complexes. The high affinity of a peptide for both a non-specific, heterogeneous DNA sequence, CTD, as well as a simple repeating polymer, poly(dI[middot]dC), results directly from the presence of the basic cluster in the peptide sequence.
Although peptides ccc and c5g displayed affinities for Non24 that were comparable with those for CRE24, the complexes with Non24 were not stable during gel electrophoresis (data not shown). Fluorescence spectroscopy was therefore used to verify the existence of a direct peptide-Non24 interaction. Experiments were performed with peptide ccc which contains a unique tryptophan within the spacer segment. The fluorescence of a 20 [mu]M solution of ccc was monitored in the presence or absence of CRE24 or Non24 (Fig. 4 A). Fluorescence intensity was greatest in the absence of DNA. Strong quenching of the fluorescence was observed upon addition of CRE24 (48%) or Non24 (63%). The fluorescence data indicate that ccc interacts with both DNA sequences and suggests that the interaction with the specific CRE24 sequence differs from that with the non-specific Non24 sequence.
We tested whether the basic cluster of CREB/ATF proteins could control CRE/AP-1 specificity. The basic cluster was removed from a CREB/ATF peptide or added to a GCN4 peptide and the CRE/AP-1 selectivities of the resulting chimeric peptides were determined. The results of both the loss and gain of function experiments were the same: basic cluster residues do not control CRE/AP-1 selectivity and therefore there is no causal relationship between DNA bending and CRE/AP-1 selectivity. In agreement with a study of the specificity of the ATF-1 bZIP element (30 ), our results point to residues in the C-terminal half of the basic segment and in the spacer segment as being important for target site selection: c10g, which contains sequence from CRE-BP1 in the N-terminal half of the basic segment, fails to discriminate between the CRE and AP-1 sites, while cgg, in which all 20 basic segment residues are from CRE-BP1, discriminates between the two sites almost as well as does ccc (10 ).
While the basic cluster and its accompanying ability to bend DNA have no influence on the relative affinity of a peptide for the closely related CRE and AP-1 target sites, it has a dramatic influence on the relative affinity for specific and non-specific DNA. Removal of the basic cluster from peptide ccc increases [Delta][Delta]Gnon/cre by 3 kcal/mol and addition of the basic cluster to peptide ggg decreases [Delta][Delta]Gnon/cre by 3 kcal/mol. The fact that CTD and poly(dI-dC)[middot]poly(dI-dC) compete well for peptides containing a basic cluster strengthens the assertion that non-specific binding is enhanced and that the basic cluster increases DNA affinity regardless of sequence.
Although our data reveal that the ccc[middot]CRE24 and ccc[middot]Non24 complexes possess similar stabilities, only the specific ccc[middot]CRE24 complex contains significant [alpha]-helical structure. The ccc[middot]Non24 complex, like the uncomplexed peptide, remains in a predominantly random coil conformation. Thus, formation of the ccc[middot]Non24 complex requires little conformational reorganization of the peptide and, hence, neglecting solvation, the energy gained through DNA interactions translates directly into binding energy. In contrast, part of the energy gained by additional interactions with DNA in the specific ccc[middot]CRE24 complex must be expended in driving the coil-to-helix transition of the basic region. Part of the effectiveness of the basic cluster in binding to random DNA could be its ability to form multiple, simultaneous salt bridges to backbone phosphates in a conformation (or set of conformations) that does not require protein folding and decreases the access of solute cations to the phosphate backbone. A similar situation has been noted in the interaction between arginine-rich peptides from HIV Rev and the Rev response element (RRE) RNA (31 ). In the specific Rev peptide[middot]RRE complex, the peptide is helical, whereas in non-specific complexes the peptide is not. Increasing the helical propensity of the peptide enhances the specific but not the non-specific binding.
Several groups (18 ,20 ,34 ) have provided evidence that asymmetric neutralization of the anionic phosphodiester backbone via salt bridges caused DNA to bend. It has been suggested that appending a cationic surface to an adjacent sequence-specific DNA recognition motif should bend the DNA in a controlled, predictable manner (32 ). The peptide c5g exhibits this design; it bends DNA toward the neutralized surface. The result in terms of bending is as predicted. However, the large increase in non-specific binding was unexpected and urges caution in the design of similar molecules.
The 109 bp that comprise the human genome provide a vast background of non-specific DNA through which a transcription factor must sort to find its cognate site (33 ). A large fraction of this DNA differs considerably from the cognate sequence, but other sequences may differ by just 1 or 2 bp. Transcription factors must not only avoid activating from sequences that are similar to their target site, they must also remain free from the massive excess of completely non-specific DNA to bind their target sites at reasonable concentrations. Hence the ability of a transcription factor to avoid non-specific binding is a critical factor in its function of specific regulation. Here we have shown that a cluster of basic amino acids present at the N-terminus of all CREB/ATF bZIP elements is not responsible for their preference for the CRE site over the closely related AP-1 site. The presence of the basic cluster does, however, greatly increase the affinity of a peptide for non-specific versus specific DNA. It appears as though the function of the basic cluster, DNA bending, is paid for at the expense of specific binding. We surmise that the cost must be regained from some transcriptional advantage conferred on the complex containing an altered DNA architecture.
We are grateful to the National Institutes of Health (GM52544) for their financial support of this research. We also thank the Arthur Wayland Dox Foundation and the Organic Chemistry Division of the American Chemical Society for graduate fellowships to S.J.M.
*To whom correspondence should be addressed. Tel: +1 203 432 5094; Fax: +1 203 432 6144; Email: alanna@milan.chem.yale.edu
Peptide
Kcre * 1018 (M2)
Knon * 1018 (M2)
[Delta][Delta]Gcre/non (kcal/mol)
Induced bend?
ggg
5 +- 1.2
1986 +- 662
3.5
No
ccc
1128 +- 341
940 +- 112
-0.1
Yes
g5c
37 +- 11
4416 +- 1535
2.8
No
c5g
90 +- 23
274 +- 2
0.6
Yes
cgg
645 +- 328
1104 +- 61
0.3
Yes
gcc
6 +- 0.4
2128 +- 642
3.5
No
ccg
80 +- 39
552 +- 5
1.1
Yes
c10g
3 +- 0.8
83 +- 24
2.0
Yes
ggc
29 +- 12
623 +- 260
1.8
No
REFERENCES