Consensus and Variant cAMP-regulated Enhancers Have Distinct CREB-binding Properties*

Recent determination of the cAMP response element-binding protein (CREB) basic leucine zipper (bZIP) consensus CRE crystal structure revealed key dimerization and DNA binding features that are conserved among members of the CREB/CREM/ATF-1 family of transcription factors. Dimerization appeared to be mediated by a Tyr307–Glu312interhelical hydrogen bond and a Glu319–Arg314electrostatic interaction. An unexpected hexahydrated Mg2+ion was centered above the CRE in the dimer cavity. In the present study, we related these features to CREB dimerization and DNA binding. A Y307F substitution reduced dimer stability and DNA binding affinity, whereas a Y307R mutation produced a stabilizing effect. Mutation of Glu319 to Ala or Lys attenuated dimerization and DNA binding. Mg2+ ions enhanced the binding affinity of wild-type CREB to the palindromic CRE by ∼20-fold but did not do so for divergent CREs. Similarly, mutation of Lys304, which mediates the CREB interaction with the hydrated Mg2+, blocked CREB binding to the palindromic but not the variant CRE sequences. The distinct binding characteristics of the K304A mutants to the consensus and variant CRE sequences indicate that CREB binding to these elements is differentially regulated by Mg2+ ions. We suggest that CREB binds the consensus and variant CRE sequences through fundamentally distinct mechanisms.

The cAMP response element-binding protein (CREB) 1 is a 43-kDa basic leucine zipper (bZIP) transcription factor that couples gene activation to a wide variety of cellular signals (for review, see Ref. 1). Initially identified as a mediator of the cAMP pathway, CREB is now recognized to respond to calcium/ calmodulin and growth factor pathways as well (2)(3)(4). The prototypical target sequence for CREB is the palindromic cAMP response element (CRE; 5Ј-TGACGTCA-3Ј), first identified in the neuropeptide somatostatin gene (5). Consensus CRE sequences, or slight variants of this sequence, have since been identified in hundreds of cellular and viral genes. In many instances, functional studies have indicated that these sequences are required for second messenger-directed transcriptional responses (for review, see Ref. 6). Whether variations of the CRE from the consensus sequence have functional consequences has not been determined.
Two general categories of factors can recognize the CRE (1,7,8). The first is composed of factors that can dimerize with CREB. This group includes the CRE modulator CREM␣ (and the multitude of splice variants that contain the CREM␣ DNA binding domain) and the activating transcription factor-1 (ATF-1). CREM␣ and ATF-1 recognize the CRE as homodimers and as heterodimers with CREB (9 -11). The second category of CRE binding proteins includes factors that cannot heterodimerize with CREB, such as c-Jun, some of the other ATFs, and members of the CAAT/enhancer-binding protein gene family (12,13). In addition to their inability to heterodimerize with CREB, these factors appear to be capable of recognizing DNA sequences that are distinct from the CRE (14). This promiscuity in DNA binding is not entirely unexpected, because some of these distinct elements, such as the AP-1 site, differ from the CRE by only a single nucleotide. CREB family members are notable in that they recognize the CRE exclusively, or nearly so. One caveat in this characterization, however, is that the defining features of the CRE have not been established unequivocally.
CREB binds as a dimer to the CRE with an affinity of ϳ1-2 nM (15). Dimerization and DNA binding are mediated by the adjacent basic and leucine zipper domains (7,8). At the extreme C-terminal end of the protein is the leucine zipper, characterized by a conserved heptad repeat of seven residues, denoted a-g. The residues at positions a and d typically form a hydrophobic interface, with conserved leucines at position d. As with other bZIP proteins, this configuration in CREB allows for the formation of a two-stranded parallel coiled-coil and, along with charged residues at positions e and g, presents a plane for dimerization specificity (12, 16 -19). In addition, dimerization of the CREB monomers apposes the basic regions in a parallel orientation to the DNA, allowing the dimer to bind at a right angle to the DNA helical axis. DNA binding elicits an ␣-helical conformation of the basic region that facilitates DNA recognition. The basic region of CREB abuts the amino terminus of the leucine zipper. This segment mediates DNA half-site sequence recognition and binding by passing through the major groove and forming several direct and indirect protein-DNA contacts.
Insights into the mechanisms underlying CREB dimerization and DNA binding were obtained from the crystal structure of the CREB bZIP⅐CRE complex (20). Two features of this structure were unexpected. First was a hexahydrated Mg 2ϩ ion centered on the CRE between the two bZIP monomers at the dimer fork. This hexahydrated Mg 2ϩ ion is not found in related bZIP-DNA complexes and participates in a water-mediated contact between Lys 304 in the CREB basic domain and the CRE. Structurally, the hexahydrated Mg 2ϩ ion was proposed to function by positioning and stabilizing the CREB basic region in the CRE major groove. It may also impart additional rigidity to the CREB basic domain, which may prevent binding to shorter or longer divergent CRE sequences. Fluorescence anisotropy assays revealed that CREB bZIP binding to the CRE was dramatically regulated by Mg 2ϩ and other divalent cations. In the absence of divalent cation, binding decreased by at least 20-fold. The divalent cation concentration required for half-maximal binding was 340 M, approximately the concentration of free Mg 2ϩ ions in many types of cells. Thus, it was proposed that the concentration of free Mg 2ϩ might regulate the level of CREB binding to the CRE.
The second unexpected feature of the complex was the mechanism underlying the specificity of CREB dimerization. In addition to predicted electrostatic interactions between Glu 319 on one monomer with Arg 314 on the other, a critical interhelical hydrogen bond was found between Tyr 307 and Glu 312 (Fig. 1). The combination of Glu 319 -Arg 314 and Tyr 307 -Glu 312 is unique to members of the CREB family and was proposed to be the primary determinant of CREB, CREM␣, and ATF-1 heterodimerization. Of note, although dimerization is generally attributed to leucine zipper sequences, the Tyr 307 is located at the C-terminal end of the basic region.
Studies in this paper were designed to test the major predictions of the CREB bZIP⅐CRE crystal structure. First, we wanted to test the specificity of the divalent cation effect by determining whether the interaction of Lys 304 with the hexahydrated Mg 2ϩ ion was essential for CREB binding to DNA. Next, we asked whether Mg 2ϩ ions regulated the binding of CREB to consensus and nonconsensus CREs. Finally, we tested the functional importance of the residues predicted to direct CREB dimerization specificity. Our studies confirm the predictions of the CREB-consensus CRE crystal structure and sug-gest that consensus and nonconsensus CREs interact with CREB through fundamentally distinct mechanisms.

EXPERIMENTAL PROCEDURES
Plasmid Construction and Protein Purification-Construction of the wild-type full-length and CREB bZIP expression vectors was described by Richards et al. (15). The three cysteine residues (Cys 300 , Cys 310 , and Cys 337 ) in the bZIP region of rat CREB (residues 285-341) were mutated to serine. These mutations improved protein solubility but did not alter DNA binding, as demonstrated by the similar relative affinities of CREB/Ser and wild-type CREB for the CRE (15). The following mutations were introduced into CREB bZIP/Ser by site-directed mutagenesis using oligonucleotide primers from Life Technologies, Inc.: Y307F, Y307R, E319A, E319K, and K304A. Purification of CREB bZIP/Ser proteins was described by Schumacher et al. (20). Purification of fulllength CREB protein was described by Richards et al. (15). Protein concentrations were determined by Bradford protein assay (Bio-Rad).
Thermal Stability Measurements-The stability of each bZIP protein was assessed in thermal melting assays by monitoring fluorescence as previously described (21). In brief, the fluorescence of the two tyrosine residues (positions 307 and 336) in wild-type CREB bZIP was monitored as a function of temperature. Measurements were taken on 1.3 ml of a sample containing 5 M CREB bZIP in HEPES buffer (pH 7.6; with 50 mM or 150 mM NaCl in the presence or absence of 10 mM Mg 2ϩ ) in a 6-mm cuvette. Temperature was monitored using a thermistor element/ controller device (Omega Engineering Inc.; model no. DP25-TH-A), with the thermistor element positioned in the solution inside the cuvette. Excitation was set at 280 nm, and emission was monitored at 300 nm using 3-nm band pass settings. The emission data were collected by ramping the temperature from 4 to 85°C at 2°C/min. The temperature was ramped back down to 4°C, and the process was repeated for each sample to confirm reversibility of protein folding. The fluorescence intensity data were fit using a nonlinear least squares regression analysis (Sigma Plot; Jandel Scientific). The thermodynamic relationship between the folded and unfolded states of the protein, represented by the melting temperature (T m ), was used to compare protein stability and is represented as follows, where ⌬H u is the enthalpy and ⌬S u is the entropy of unfolding (22,23). Tyrosine emission (300 nm; y axis) was plotted as a function of temper- The hydrated magnesium ion (gray) complexed with six water molecules occupies the dimer cavity. b, magnification of the dimer cavity with a view of the hexahydrated magnesium ion contacted by the Lys 304 side chains, centered above the CRE illustrated at the base of the dimer fork as a Cory-Pauling-Kolton space-filling model. ature (in degrees Centigrade). In the thermal melting experiments, the sharp upward slopes represent the transitions between protein folding and unfolding. Pretransition state fluorescence defines the fluorescence of the folded state of the protein (represented by the left tails of the melting curves). Post-transition state fluorescence defines the fluorescence characteristics of the unfolded state (represented by the right tails of the melting curves). By measuring free tyrosine in solution, we confirmed that the downward slopes observed for the pre-and posttransition states were due to the temperature sensitivity of tyrosine fluorescence (data not shown). Assays were performed on the wild-type CREB bZIP, and the mutants K304A, E319A, E319K, Y307F, and Y307R. Note that in the case of the Y307F and Y307R mutants, the emission of only one tyrosine was measured. Although the signal was reduced, the single tyrosine was adequate to measure the transition state of the mutant proteins. Selected proteins were also denatured with urea, yielding profiles that were similar to those produced by thermal melting (data not shown), indicating that heat is a satisfactory denaturing reagent for measuring tyrosine emission.
Fluorescence Polarization Measurements-The binding affinities of CREB bZIP proteins for DNA were determined in solution using a PanVera Beacon 2000 fluorescence polarization system. 5Ј-Fluoresceinated oligonucleotides corresponding to the somatostatin (5Ј-CCTG-ACGTCAGCCCCCTGACGTCAGG-3Ј), rat c-Fos (5Ј-AGTGACGTAG-GCCCCCCTACGTCACT-3Ј), phosphoenolpyruvate carboxykinase (PEPCK; 5Ј-CCTTACGTCAGCCCCCTGACGTAAGG-3Ј), and the tyrosine aminotransferase (TAT; 5Ј-TCTGCGTCAGCCCCCTGACGC-AGA-3Ј) CREs were purchased from Life Technologies, Inc. The CRE binding sites are indicated in boldface type. Each oligonucleotide was designed to form a hairpin with a 5-nucleotide loop and was selfannealed by heating to 95°C and snap-cooled on ice. Binding experiments with the somatostatin CRE were performed with wild-type and mutant CREB bZIP (residues 283-341) and wild-type full-length CREB (residues 1-341) proteins. Binding reactions with the PEPCK CRE were performed with wild-type and K304A bZIP and wild-type fulllength CREB, and those with the c-Fos and TAT CREs were performed with wild-type and K304A CREB bZIP. Protein was titrated into 0.99-ml solutions that contained 1 nM fluoresceinated oligonucleotide in 25 mM Tris-HCl, pH 7.6, 50 mM NaCl, 0.5 mM EDTA, 5% glycerol, 6 g of bovine serum albumin, 10 g of poly(dI-dC) in the presence or absence of 10 mM MgCl 2 . Binding reactions were performed at 25 or 37°C, and samples were incubated for 1 min to achieve equilibrium. Each sample was excited at 490 nm, and emission was measured at 530 nm. The binding curves were fit with a nonlinear least squares regression analysis using Sigma Plot (Jandel Scientific). In addition to the rectangular hyperbolic binding function, the equation utilized for curve fit determination included a nonspecific component to account for the linear increase in polarization at higher protein concentrations. For a curve that begins at y 0 (minimum polarization) and rises to y 1 (maximum polarization), the following equation was used,   Fig. 1 depicts two magnified views of a portion of the CREB bZIP-consensus CRE complex as reported by Schumacher et al. (20). Key residues in the basic and leucine zipper domains are highlighted. The continuous ␣-helical chains are symmetrically apposed at the C-terminal leucine zipper and are fastened by an electrostatic interaction between Arg 314 and Glu 319 as well as hydrogen bonds between Tyr 307 and Glu 312 . The helices pass across the DNA through the major groove, where several residues from the basic region make protein-DNA contacts. The most extraordinary feature of the complex is the hydrated Mg 2ϩ ion, which is juxtaposed between the two ␣-helices and appears to stabilize the dimer fork on the CRE sequence.
A lysine residue (Lys 304 ) in the basic region of each subunit provides a hydrogen bond to one of the coordination waters of the Mg 2ϩ ion (Fig. 1b). Thus, we hypothesized that Lys 304 was essential for Mg 2ϩ ion binding. To test this idea, we mutated Lys 304 to Ala and performed thermal stability and fluorescence polarization assays on the mutated and wild-type CREB proteins in the presence or absence of Mg 2ϩ ions. Alanine was chosen for this substitution because it is predicted to maintain the helical structure of CREB bZIP. Moreover, the related factor CAAT/enhancer-binding protein ␣ contains an Ala in the corresponding position. The thermal stability of each CREB protein was determined by measuring the intrinsic tyrosine fluorescence (contributed by Tyr 307 and Tyr 336 ) as a function of temperature. These studies indicated that the conformational stability of the wild-type protein was unaffected by the presence or absence of Mg 2ϩ ions, as indicated by the unchanged T m values ( Fig. 2a and Table I; T m ϭ 45.0 Ϯ 0.2°C with Mg 2ϩ , T m ϭ 43.2 Ϯ 0.4°C without Mg 2ϩ ). Similarly, the K304A mutation did not have an adverse effect on the conformational stability of the protein (Fig. 2a and Table I; T m ϭ 47.0 Ϯ 0.3°C with Mg 2ϩ , T m ϭ 46.7 Ϯ 0.3°C without Mg 2ϩ ), indicating that divalent ions are not necessary for maintaining structure in the absence of DNA. In contrast, fluorescence polarization measurements indicated that the absence of Mg 2ϩ ions markedly decreased the binding affinity of wild-type CREB for the CRE (Fig. 2b and Table II; K d ϭ 1.8 Ϯ 0.2 nM with Mg 2ϩ , K d Ͼ 20 nM without Mg 2ϩ ), as reported previously (20). Moreover, no binding of CREB bZIP K304A to the CRE was detected even in the presence of Mg 2ϩ ions (Fig. 2b). Thus, the enhancement of CREB binding to the consensus CRE mediated by Mg 2ϩ depends upon the integrity of Lys 304 .
With the exception of the tyrosine aminotransferase (TAT) CRE, the binding properties of CREB for nonconsensus CRE sequences have been poorly explored. We recently reported that CREB binds the nonconsensus PEPCK CRE with high affinity (24). This finding is not altogether surprising, because this sequence differs from the somatostatin CRE by only a single nucleotide (TTACGTCA; the single change in boldface type). However, these studies were performed under standard conditions, in the presence of Mg 2ϩ ions. To test whether CREB binding to nonconsensus CREs similarly required Mg 2ϩ , we reanalyzed binding to the PEPCK element in the presence and absence of divalent cation. High affinity binding was detected in both cases (Fig. 3, a and b,   CREB (data not shown). Consistent with these observations, the K304A mutant bound equally to the PEPCK CRE in the presence or absence of Mg 2ϩ ions ( Fig. 3c and Table II; K d ϭ 2.0 Ϯ 0.4 nM with Mg 2ϩ , K d ϭ 2.0 Ϯ 0.2 nM without Mg 2ϩ ). These studies indicated that the mechanisms of CREB binding to the PEPCK and consensus CREs must be fundamentally different.
Additional binding studies were performed using the c-Fos and TAT CREs. The rat c-Fos promoter contains a CRE element that differs from the somatostatin CRE by two nucleotides, in the seventh and eighth positions (TGACGTAG). The binding affinity of CREB for the c-Fos CRE had not been determined using equilibrium measurements in solution. The TAT promoter contains a CRE that differs from the consensus sequence by three nucleotides (CTGCGTCA) and has been characterized as a prototypical low affinity binding site (25). The binding affinity of CREB for the TAT CRE was determined to be 11 nM by Richards et al. (15). We analyzed the TAT and c-Fos CREs in the presence and absence of Mg 2ϩ ions and found that in both instances CREB bZIP binding did not require Mg 2ϩ (Figs. 4 and 5; K d ϭ ϳ5 nM for the c-Fos CRE and 10 nM for the TAT CRE in the presence or absence of Mg 2ϩ ions). This Mg 2ϩ independence of binding to the c-Fos and TAT elements was supported by the ability of CREB bZIP K304A to bind to both sequences irrespective of Mg 2ϩ (Figs. 4c and 5c).
Two key determinants of dimerization unique to CREB family members are interactions between Glu 319 -Arg 314 and Tyr 307 -Glu 312 . To investigate the contributions of these interactions to protein stability and DNA binding, we utilized CREB derivatives generated by mutating Tyr 307 to Phe or Arg and Glu 319 to Ala or Lys. The Y307F substitution was designed to test the importance of the Tyr 307 -Glu 312 hydrogen bond (Fig.  1a). The Y307R substitution was designed to replace the hydrogen bond with an electrostatic interaction, which was predicted to increase the stability of a potential dimer structure. As compared with wild-type CREB (Fig. 2a), the Y307F mutant shows a dramatic reduction in thermal stability in 50 mM NaCl ( Fig. 6a and Table III; T m ϭ 24.3 Ϯ 0.6°C), while the Y307R mutation had a stabilizing effect under this ionic condition ( Fig. 6b and Table III; T m ϭ 55.3 Ϯ 0.3°C). However, the Y307R mutant was irreversibly denatured at physiological ionic strength (150 mM NaCl), indicating that this concentration of NaCl can interfere with the formation of the Arg 307 -Glu 312 salt bridge (data not shown). Fig. 7b shows that the Y307F mutant also exhibited a temperature-dependent decrease in binding to the CRE, as determined by fluorescence polarization assays. Compared with wild-type CREB (Fig. 7a and Table IV; K d ϭ 2.0 Ϯ 0.1 nM at 25°C, K d ϭ 15 Ϯ 1.1 nM at 37°C), DNA binding of the Y307F protein was greatly diminished at 25°C and was abolished at 37°C (Fig. 7b and Table  IV). Thus, the interaction between Tyr 307 and Glu 312 is crucial for dimerization and subsequent DNA binding.
The importance of the Glu 319 -Arg 314 interaction to dimer stability and specificity was tested by measuring the thermal stability of proteins containing mutations of Glu 319 to Ala or Lys. As shown in Fig. 8, both mutations reduced CREB protein stability compared with the wild type CREB bZIP control (Fig.  2a). The E319A substitution reduced dimer stability in an ionic strength-independent manner (Fig. 8a and Table III; T m ϭ Measurements were plotted as in Fig. 2. The binding curves were repeated at least three times for each condition. 29.5 Ϯ 0.8°C; and data not shown), and the E319K mutant appeared completely unfolded at 4°C, regardless of the ionic strength of the buffer (Fig. 8b and data not shown).
Despite the lowered thermal stability, both derivatives were able to bind to the consensus CRE, in the presence of Mg 2ϩ ions, although with greatly reduced affinity at 25°C (Fig. 9a and Table IV, E319A K d ϭ 65.0 nM Ϯ 1.1; Fig. 9b, E319K K d Ͼ 250 nM). Consistent with their T m values, binding of both Glu 319 substitutions to the CRE was completely abolished at 37°C (Fig. 9, a and b, and Table IV).   Fig. 2. The binding curves were repeated at least three times for each condition.

DISCUSSION
In this study, we examined the contributions of key residues of the CREB bZIP to structural stability and DNA binding. For these determinations, we utilized thermal melting and fluorescence polarization assays. The melting assays measured the transitions of the mutant and wild-type CREB bZIP proteins from the folded to the unfolded states during thermal denaturation by tracking the emission spectra from tyrosine residues.
In the context of the folded protein, the tyrosine fluorescence is quenched, presumably by neighboring residues (26). Upon thermal unfolding of the protein, this quenching is removed, and the tyrosine fluorescence increases. Indirect measurements of bZIP dimer association (using circular dichroism in the absence of DNA) indicate that dimer formation occurs at micromolar concentrations of peptide (27,28). However, there has been some disagreement as to whether CREB exists as a dimer or monomer in solution in the absence of DNA. Previously, our laboratory used fluorescence resonance energy transfer to determine that the K d for CREB dimerization in the absence of DNA was 0.6 nM (29), but equilibrium sedimentation studies suggested that CREB exists primarily in a monomeric form (30). The melting curves in the current study were carried out using relatively high concentrations of CREB protein (5 M), and we have interpreted these results assuming that, under our experimental conditions, CREB exists in the dimeric form. Consistent with this interpretation, mutations that were expected to disrupt dimerization caused observable changes in T m and reductions in DNA binding, while the substitution in the basic region (K304A) did not alter the T m value from that obtained for wild-type CREB bZIP. Thus, the thermal melting analyses described herein can be interpreted reasonably as an indirect measure of dimer stability. The finding that the thermally unstable E319K mutant binds to the CRE with low affinity (K d ϭ 250 nM) suggests that if this form is monomeric, then a single bZIP helix interacts rather poorly with DNA. Alternately, it may be that the CRE can recruit and stabilize the very low concentration of the dimeric form of this mutant, or that it overcomes the interhelical repulsion between Lys 319 and Lys 314Ј at high protein concentrations.
Role of Mg 2ϩ Ions on CREB bZIP Stability and DNA Binding-We utilized fluorescence polarization assays to obtain equilibrium measurements of protein-DNA binding in solution. Precise quantitation of these interactions might not be achieved by using the more standard gel mobility shift assays, because the proteins can dissociate from the DNA within the gel matrix, and the ionic environment is difficult to control during electrophoresis.
We first tested whether Lys 304 , which interacts with the hexahydrated Mg 2ϩ ion, was necessary for CREB dimerization and DNA binding. The importance of Lys 304 in CREB function was tested previously by Dwarki et al. (31). Their study exam-   ined the properties of a CREB K304E mutant in an effort to determine whether the predicted electrostatic repulsion between the Glu and the negatively charged phosphate backbone of the CRE would be deleterious for CREB function. As predicted, this mutant failed to bind the CRE in vitro or to activate transcription in vivo. The present study took advantage of the crystal structure of the CREB bZIP⅐CRE complex to examine the role of Lys 304 specifically in Mg 2ϩ ion-mediated DNA sequence recognition. We mutated the Lys 304 to Ala, the residue that occupies this position in the related transcription factor CAAT/enhancer-binding protein ␣. The thermal stability of the CREB bZIP K304A mutant was the same as that of the wildtype CREB protein, irrespective of Mg 2ϩ (summarized in Table  I). Therefore, we suspect that the stability of the mutant dimer is not impaired. However, binding of wild-type CREB bZIP to the consensus CRE was highly dependent upon Mg 2ϩ (summarized in Table II). In the absence of Mg 2ϩ , CREB bound poorly to the consensus CRE, and no binding was detected with K304A, regardless of whether Mg 2ϩ was included in the buffer. It is possible for Lys 304 to interact with the phosphate backbone of the DNA in the absence of the hexahydrated Mg 2ϩ ion (determined by manipulation of the complex in RASMOL), which would be consistent with the ability of the wild-type CREB protein to bind to the CRE at low affinity in the absence of Mg 2ϩ . Elimination of the positive charges contributed by the Lys 304 side chain would diminish its interaction with DNA, regardless of the presence of Mg 2ϩ . Alternately, the presence of a Mg 2ϩ ion might physically obstruct the binding of the K304A mutant to the consensus CRE, if the Ala substitution effectively removed a Mg 2ϩ -dependent spacing requirement from the dimer fork. Nonetheless, our data indicate that both a Mg 2ϩ ion and Lys 304 are required for CREB to bind with high affinity to the consensus CRE but do not contribute to CREB stability.
Role of Mg 2ϩ Ions for CREB bZIP Binding Specificity-The next set of experiments tested whether the Mg 2ϩ ion requirement for CREB binding was unique to the consensus CRE (Table II). The PEPCK, TAT, and c-Fos genes are all regulated by CREB via nonconsensus CREs. The PEPCK CRE differs from the consensus CRE at the Ϫ3 base position (Ϫ4, Ϫ3, Ϫ2, Ϫ1/1, 2, 3, 4 corresponds to TGAC/GTCA), in which there is a pyrimidine for purine substitution (TTACGTCA), the TAT CRE contains substitutions at the Ϫ4, Ϫ3, and Ϫ2 base positions with a pyrimidine for purine substitution in the Ϫ3 base position (CTGCGTCA), and the rat c-Fos CRE contains substitutions in the 3 and 4 base positions (TGACGTAG), with a purine for pyrimidine substitution at the 3 base position. The binding affinities of CREB for the PEPCK and TAT elements were determined previously to be 2.0 and 11.0 nM, respectively (see Refs. 24 and 15, respectively). The binding affinity of CREB for the c-Fos element was determined in the present study to be 4.5-5.0 nM, approximately 3-fold lower than that for the consensus CRE. CREB bZIP binding to the lower affinity TAT CRE, the moderate affinity c-Fos CRE, and the high affinity PEPCK CRE were all independent of Mg 2ϩ (Table II). Similarly, the K304A mutant was capable of recognizing each of the nonconsensus CREs but not the consensus CRE. These studies indicate that deviation from the consensus CRE sequence by even one nucleotide can result in significant changes in CREB binding under specific ionic conditions. The CREB bZIP⅐CRE crystal structure (20) reveals that there are specific 2-fold symmetry-related contacts made by each CREB monomer to nucleotides of base pairs Ϫ2/2 and Ϫ4/4. Specifically, the C-␤ of residue Ala 297 makes van der Waals contacts with Ϫ2/2 thymine, and Ϫ4/4 thymine is engaged in van der Waals contacts with the C-␤ of Ala 296 and the C-␤ of Ser 300 . Loss of these van der Waals contacts could explain the lower affinity of CREB for the TAT CRE. The cytosine of base pair Ϫ3/3 makes a hydrogen bond to the side chain of Asn 293 . A cytosine to adenine substitution at this base pair position, such as with the PEPCK CRE, could, with slight rearrangement, permit the same interaction between Asn 293 and the N-6 of adenine. The moderate affinity of CREB bZIP for the c-Fos CRE might result from the poorer hydrogen bond that would form between N-293 and O-6 of guanine Ϫ4/4, which would require side chain rearrangement. Such a rearrangement could also explain the altered Mg 2ϩ requirement. The fact that the CREB K304A mutant also distinguished consensus from nonconsensus CREs indicates that the Mg 2ϩ effect does not occur through nonspecific changes in CRE structure. Rather, the palindromic nature of the consensus CRE must provide an interface that is dependent upon Mg 2ϩ for CREB recognition.
Role of Glu 319 -Arg 314 and Tyr 307 -Glu 312 on CREB bZIP Stability and DNA Binding-We next tested the prediction that the Glu 319 -Arg 314 and the Tyr 307 -Glu 312 interactions were critical for CREB dimerization by examining their affect on CREB bZIP stability (summarized in Table III). The interactions of these residues are unique to members of the CREB family. We found that the E319K protein was less stable than the E319A mutant and that both had significantly lower T m values than wild-type CREB bZIP. However, even the E319K mutant was capable of low affinity binding to the CRE, suggesting that it might be able to interact with DNA as a monomer or as a DNA-stabilized dimer. Interestingly, naturally occurring E319K mutants of CREB have been detected in Raji B lymphocytes and C6 glioma cells (32,33). It is possible that other members of the CREB gene family mediate responses to cAMP in these cells.
The importance of the interhelical hydrogen bond formed between Tyr 307 and Glu 312 was tested by mutating Tyr 307 to Phe and Arg. The Phe substitution decreased protein stability, and the Arg substitution increased protein stability (although only at the lower salt concentration), as predicted (Table III). Furthermore, the unstable CREB Y307F mutant had a reduced ability to bind to the CRE (Table IV). These data support a dual role for the Tyr 307 -Glu 312 interaction, one that serves to promote dimerization specificity and a second that stabilizes the dimer on the DNA.
Conclusions-Our studies support the principal predictions of the CREB bZIP⅐CRE crystal structure and confirm the functional roles of the key residues that define the unique features of transcription factors in the CREB gene family. The importance of Mg 2ϩ ions in the association of CREB with the consensus CRE suggests that this divalent cation may have a regulatory role for discrimination at the promoter level. How the difference in Mg 2ϩ sensitivity of consensus and nonconsensus CREs might affect the regulation of specific genes remains to be determined, however. In neurons, depolarization induces calcium fluxes that cause a 3-10-fold increase in the concentration of free Mg 2ϩ (34,35). It is possible that these Mg 2ϩ changes could regulate the binding of CREB to some cellular genes, but not others, depending on the presence of particular CRE sequences in the relevant transcriptional control regions. This complexity and the potential increase in regulatory control have not been appreciated previously.