Roles of Phosphorylation and Helix Propensity in the Binding of the KIX Domain of CREB-binding Protein by Constitutive (c-Myb) and Inducible (CREB) Activators*

cAMP-response element-binding protein (CREB)-binding protein (CBP) is a general transcriptional co-activator that mediates interactions between transcription factors and the basal transcription machinery. To obtain insights into the mechanism by which the KIX domain of CBP can recognize the transactivation domains of many different transcription factors, we have used NMR and biochemical analyses to study the interactions of KIX with the transactivation domain from the constitutive activator c-Myb and with the kinase-inducible transactivation domain (KID) from CREB. NMR chemical shift mapping shows that both activation domains bind to the same surface of KIX. In the unbound state, both the phosphorylated KID and c-Myb activation domains are only partly structured, and binding to KIX is coupled with folding to form an amphipathic helix. Helix-destabilizing mutations significantly impair binding, whereas mutations that increase the intrinsic secondary structure content of the free phosphorylated KID peptide have only a small influence on binding affinity. Low affinity but specific binding of unphosphorylated KID to KIX was measured by ITC and was also observed in Western blot assays and by a fluorescence resonance energy transfer experiment in living cells. The large increase in the affinity for phosphorylated KID is due to favorable intermolecular interactions involving the phosphate moiety. After induction by phosphorylation, CREB is able to compete effectively with other transcriptional activators for binding to CBP.

The general mechanism of transcriptional activation in eukaryotes involves interactions between DNA-bound activators, co-activators, and components of the basal transcription complex. Cyclic AMP response-element binding protein (CREB) 1 -binding protein (CBP) and its paralog p300 are general coactivators that are essential for the function of numerous transcriptional regulators. They appear to act as scaffolds, interacting simultaneously with the activation domains of DNA-bound transcription factors and other components of the transcriptional machinery (reviewed in Ref. 1). CBP was originally identified as a cofactor for the cAMP-regulated transcription factor CREB. Phosphorylation of the kinase-inducible activation domain (KID) of CREB at Ser-133 was shown to be critical for binding the KIX domain of CBP and for target gene expression (2,3). The NMR structure of the phosphorylated KID (pKID)-KIX complex (4) showed that the phosphoserine is located between the two helices (␣ A and ␣ B ) of the KIX-bound pKID. The phosphate group participates in intermolecular interactions with residues Tyr-658 and Lys-662 of KIX, which have been shown to be critical for binding (4). The position and negative charge of the phosphate group could also provide additional stabilization of the ␣ B helix dipole.
Although phosphorylated Ser-133 has been shown to be important for KIX binding to pKID, most of the interactions at the interface are hydrophobic, between the ␣ B helix of KID and the shallow hydrophobic groove formed by the ␣ 1 and ␣ 3 helices of KIX. Mutagenesis studies of KIX implicated the same hydrophobic groove as the docking site for c-Myb (5, 6), a constitutive transcriptional activator regulating cell growth and differentiation, mainly in hematopoiesis (reviewed in Ref. 7). Although pKID and c-Myb appear to share a common binding site on KIX, no obvious sequence similarity can be detected. However, the spacing of critical hydrophobic residues in the binding region is similar. In free c-Myb, these hydrophobic residues cluster on one face of a partially formed amphipathic helix (6). Although pKID is largely unfolded in the free state (8), it also adopts an amphipathic helical conformation upon complex formation (4).
The binding affinity of KIX for pKID is 20 -50-fold higher than for c-Myb (6). This striking difference in affinity for KIX is consistent with the different kinetic profiles for the two activators. Whereas CREB is activated by phosphorylation within the KID domain, the activity of c-Myb is believed to be constitutive. Complex formation between pKID and KIX is enthalpy-driven, whereas c-Myb-KIX binding is driven by both enthalpy and entropy (6).
The apparent differences in the mode of binding of KIX to pKID and c-Myb prompted us to investigate the structural features of these complexes in order to obtain insights into the mechanisms by which the KIX domain of CBP recognizes nu-merous transcriptional activation domains. In particular, we were interested in evaluating the roles of phosphorylation and secondary structure formation in determining binding affinity for KIX and hence in the regulation of transcription. Our results confirm that a common hydrophobic binding interface of KIX is used to bind either pKID or c-Myb. In the absence of a phosphate group, KID gives rise to a low affinity complex with KIX that appears to be specific and is also formed in vivo. The intermolecular interactions of the phosphoserine in pKID provide substantial binding enthalpy, which, since it is not present in the unphosphorylated form of the protein, constitutes the factor that makes this interaction inducible. Our results also demonstrate that the binding region of the transactivation domains must be helical for tight binding to CBP and that mutations that destabilize the helix significantly impair binding.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-The wild-type KIX domain (residues 586 -672) of mouse CBP (identical amino acid sequence to human) was expressed in either unlabeled or uniformly 15 N-or 13 C, 15 Nenriched forms in BL21(DE3) Escherichia coli and purified to homogeneity as previously described (4). KIX 586 -672 mutants Y658F, K662A, and Y650A were expressed and purified as glutathione S-transferase fusion proteins as described elsewhere (6). Glutathione S-transferase fusion proteins were then cleaved by thrombin, and KIX mutants (with an additional Gly-Ser peptide resulting from cleavage) were purified to homogeneity following the procedure used for wild-type KIX. The wildtype KID60 domain (residues 101-160 plus an N-terminal methionine) of mouse CREB (identical amino acid sequence to human) was expressed, purified, and phosphorylated as previously described (4). Shorter pKID/KID peptides (residues 119 -147 (pKID29) or residues 129 -149 (pKID21) of mouse CREB) and unlabeled Myb25 (residues 291-315 from mouse) were chemically synthesized using a Perseptive Biosystems peptide synthesizer (PerkinElmer Life Sciences) and purified to homogeneity by reverse-phase HPLC. Uniformly 15 N-or 13 C, 15 Nlabeled Myb25 291-315 were overexpressed in E. coli using a ubiquitin fusion protein system, which was a generous gift from Dr. Toshiyuki Kohno. The c-Myb 291-315 sequence was ligated to the ubiquitin sequence using a similar protocol to that described by Kohno et al. (9). Briefly, the decahistidine-tagged ubiquitin coding sequence was subcloned from plasmid pUBK19 (9) into a pET-21d plasmid, digested by NsiI, blunted with T4 DNA polymerase, and digested by BamHI. The chemically synthesized oligonucleotide encoding c-Myb, followed by two stop codons and the BamHI site, was amplified by PCR and digested by BamHI, and the 5Ј-end was phosphorylated by T4 polynucleotide kinase. Successful blunt-end ligation of the Myb25 insert into the ubiquitin vector was confirmed by DNA sequencing. The ubiquitin fusion protein (ubiquitin-Myb25) and yeast ubiquitin hydrolase were expressed in E. coli and purified on Ni 2ϩ -nitrilotriacetic acid-agarose as previously described (9). Both ubiquitin-Myb25 and yeast ubiquitin hydrolase were dialyzed against the cleavage buffer (10 mM Tris, pH 8.0, 0.1 M NaCl, and 1 mM ␤-mercaptoethanol). Cleavage of ubiquitin-Myb25 by yeast ubiquitin hydrolase at 37°C was monitored by SDS-PAGE. The cleaved Myb25 peptide was separated from His-tagged proteins by affinity chromatography using Ni 2ϩ -nitrilotriacetic acidagarose. Final purification was done by reverse-phase HPLC. The identity and integrity of all proteins and peptides were confirmed by mass spectrometry.
Construction of Fusion Proteins-Plasmids encoding fusion proteins of mouse KIX 553-679 with enhanced cyan fluorescent protein (KIX-ECFP) and wild-type or S133A rat KID 100 -160 with enhanced yellow fluorescent protein (wild-type or S133A KID-EYFP) were prepared as previously described (10). Expression vectors for KID-KIX fusion proteins were constructed by inserting a PCR-generated KIX 553-679 fragment at the C terminus of the KID 100 -160 vector used above.
NMR Spectroscopy-NMR samples were prepared in the concentration range of 0.5-1.0 mM in 90% H 2 O, 10% D 2 O buffer (20 mM tris-d 11acetate-d 4 (pH 5.5), 50 mM NaCl, 2 mM NaN 3 ). NMR spectra were recorded at 27°C on Bruker AMX500, DRX600, and DMX750 spectrometers, equipped with triple axis gradient probes. Titration of labeled KIX with unlabeled Myb25 or labeled Myb25 with unlabeled KIX was monitored by two-dimensional 1 H-15 N HSQC spectra. The unlabeled component was added in excess of 10 -20% after saturation as indicated by the HSQC spectra. NMR data processing and analysis were per-formed using Felix97 (Molecular Simulations Inc., San Diego, CA) or NMRPipe (11) and NMRView (12). Nearly complete backbone assignments for Myb25-bound KIX were accomplished using three-dimensional HNCACB, CBCA(CO)NH, HNCO, HCACO, 15 N-edited NOESY-HSQC, and 15 N-edited TOCSY-HSQC spectra. Only the backbone amides of Arg-623 and Lys-667 were not assignable due to exchange broadening. Complete backbone assignments for both free and KIXbound Myb25 were accomplished using three-dimensional HNCACB and CBCA(CO)NH spectra. Secondary chemical shifts were calculated using random coil values for 13 C␣ shifts (13), incorporating a correction for proline residues (14). The 13 C chemical shifts were referenced relative to 2,2-dimethyl-2-silapentane-5-sulfonate (DSS). Chemical shift deviations between the bound and free states of KIX were calculated by residue using the formula, where ⌬␦ HN and ⌬␦ N correspond to the differences between the bound and free states of KIX in the amide proton and nitrogen chemical shifts, respectively (15).
Circular Dichroism-Spectra were collected at 27°C on the Aviv model 202 CD spectrometer using a path length of 2 mm. Samples contained 40 -50 M KID or Myb25 peptide in 5 mM potassium phosphate buffer (pH 7.0) and 0 -40% (v/v) trifluoroethanol (TFE). The signal at 222 nm was normalized to give the molar ellipticity ( 222 ) using peptide concentrations determined by optical density at 280 nm for KID peptides (containing one Tyr) or by amino acid analysis for Myb25 peptides. The isodichroic point observed at 202 nm was used to correct for minor concentration differences (16). Helical content was calculated according to the published formulas for 100% helix (17) and 100% coil (18).
Isothermal Titration Calorimetry (ITC)-The titration of a KID or Myb25 peptide into a KIX protein was performed at 27°C using an MCS titration calorimeter from MicroCal Inc. Simultaneous dialysis of all peptides and proteins against the ITC buffer (50 mM Tris (pH 7.0), 50 mM NaCl), followed by filtration and degassing were done to minimize background noise. KIX and KID concentrations were determined by absorbance measured at 280 nm. The stoichiometric ratio obtained from the curve fit was consistently 1:1, within 5% error. The concentration of Myb25 was determined according to the stoichiometric ratio obtained from the curve fit. The concentration of KIX in the ITC cell was 50 -165 M with higher concentrations used for the lower affinity complexes, whereas the concentration of the peptide in the syringe was 12-fold over that of KIX. Typically, two injections of 5 l were followed by 28 injections of 10 l until a molar ratio of 2.5 was obtained. Integration of the thermogram and subtraction of the blanks yielded a binding isotherm that was fit to a model of one-site interaction (ITC data analysis software in Origin 2.3 of MicroCal Inc.). Only K d values are reported (together with S.D. from duplicate experiments), since ⌬H values for the low affinity complexes have high uncertainties and were therefore not interpreted.
Western Blotting-HEK293 cells were transfected with expression plasmids for KID, KIX, KID-KIX fusion protein, or S133A KID-KIX fusion protein. Western blotting from whole cell extracts using antisera against KID, KIX, and the KID-KIX complex was performed as previously described (19,20). Phosphorylation of wild-type KID (but not S133A KID) on Ser-133 was confirmed with an anti-pKID antiserum.
Spectrofluorometric Fluorescence Resonance Energy Transfer (FRET) Measurement-HEK293 cells (10 6 ) were transfected for 24 h with plasmids encoding KID-EYFP and KIX-ECFP fusion proteins, as previously described (10). The cells were resuspended in Hanks' balanced salt solution (Invitrogen) and placed in a glass cuvette with a stir bar. Emission spectra (460 -540 nm) were recorded following excitation at 430 nm on a PTI spectrofluorometer.

RESULTS
Binding of c-Myb to KIX-To map the site of c-Myb interaction on KIX, triple resonance NMR experiments were used to assign the backbone chemical shifts for 13 C, 15 N KIX complexed with a 25-mer peptide derived from the c-Myb activation domain (residues 291-315, hereafter called Myb25). Backbone amide chemical shift deviations of KIX upon complex formation (Fig. 1A) show that both Myb25 and pKID bind to the ␣ 1 -␣ 3 interface of KIX, consistent with previous mutagenesis studies (6). The difference between experimental 13 C␣ chemical shifts and random coil values (termed secondary chemical shift) was used as a measure of backbone secondary structure (reviewed in Ref. 13) to evaluate the presence and extent of structural changes in KIX upon binding of c-Myb. As for pKID, Myb25 does not perturb the overall structure of KIX, since the C␣ secondary shifts are indistinguishable from those previously determined for free KIX (21).
The KIX residues Tyr-658 and Lys-662, which interact with the phosphoserine in pKID, show larger chemical shift deviations for the complex with pKID (residues 119 -147, hereafter called pKID29) than for the Myb25-KIX complex (Fig. 1A), suggesting that these residues contribute less to binding of Myb25, which lacks the phosphate. This is supported by biochemical studies showing that mutations of residues Tyr-658 or Lys-662 had a drastic effect on the affinity of pKID-KIX but only a minor effect on the affinity of c-Myb (6). Larger chemical shift perturbations are observed in the Myb25 complex for other KIX residues such as Tyr-650 and Leu-653 in the ␣ 3 helix and Leu-603, His-605, and Leu-607 residues in the ␣ 1 helix. The NMR data (Fig. 1A) are in good agreement with biochemical studies, in which mutations of residues Tyr-650 or Leu-653 had a larger effect on c-Myb binding than on pKID binding (6). Whether these differences result from direct contacts unique to one complex or subtle conformational changes in KIX remains to be explored.
The NMR data show unambiguously that Myb25 binds to the same ␣ 1 -␣ 3 face of KIX as the amphipathic ␣ B helix of pKID (Fig. 2). In order to determine whether the structure of bound Myb25 is similar to that of pKID, the backbone resonances of 13 C, 15 N Myb25, in the free and KIX-bound states, were assigned using triple resonance NMR experiments. As shown in Fig. 1B, the secondary chemical shifts of free Myb25 indicate that residues 295-309 populate a partially helical conformation. The helical content of this region, estimated by the magnitude of the secondary shifts, is 25-30%, which is consistent with CD results (6). Binding of KIX to Myb25 leads to a significant stabilization of the c-Myb helix, as evident in Fig. 1B. In particular, residues 295-306 in the center of Myb25 are completely helical in the bound state, whereas residues 292-294 and 307-311 show fraying of the helix. A similar fraying of the bound helix was observed in the interaction of pKID with KIX (8). Binding of both Myb25 and pKID to KIX involves a coupled folding event; the intermolecular interactions with KIX serve to stabilize the helical conformation of the two transcriptional activation domains.
Binding of Unphosphorylated KID to KIX-The binding of CREB to CBP in the absence of phosphorylation on Ser-133 has not been detected so far by in vitro techniques, including fluorescence polarization (22), glutathione S-transferase pull-down (23), and gel shift assays (3,24,25). This is puzzling in view of the important role of hydrophobic interactions of the ␣ B helix of pKID in KIX binding (4) and of the structural similarities between binding of Myb25 and pKID. It is possible, however, that low affinity binding of unphosphorylated KID could escape the limited sensitivity of these assays. Isothermal titration calorimetry (ITC) was therefore used to measure the binding of unphosphorylated KID 101-160 (KID60), unphosphorylated KID 119 -147 (KID29), and Myb25 to KIX. The results for KID60 and Myb25 are shown in Fig. 3A, and the K d values calculated for all of the peptides are reported in Table I (top). Our results demonstrate that unphosphorylated KID, either KID60 or KID29, will bind to KIX but with a 2-order of magnitude lower affinity than that of the phosphorylated state and a 7-fold lower affinity than that of the constitutive Myb25. The contribution of the phosphate to binding, calculated from these results to be a ⌬⌬G of Ϫ3.0 kcal/mol, agrees well with an NMR study that estimated the contribution of a dianionic phosphate group relative to a monoanionic one to be Ϫ1.5 kcal/mol (26). To examine the specificity of complex formation between unphosphorylated KID and KIX, we employed an antiserum that has been shown to bind specifically to the pKID-KIX complex (20). A fusion protein joining KID and KIX on a single polypeptide chain was prepared in order to achieve a high local concentration that would lead to detectable complex formation despite the low affinity of unphosphorylated KID for KIX. Following transfection of HEK293 cells with an expression plasmid encoding the fusion protein, Western blot analyses using the anti-complex-specific antibody recognized the KID-KIX complex when KID was phosphorylated on Ser-133 (Fig. 3B,  lane 2). No signal was obtained when the cells were transfected with either KID or KIX alone (Fig. 3B, lanes 1 and 4, respectively), although these domains were expressed at comparable levels (Fig. 3B, lanes 9 and 8, respectively). Most interestingly, the anti-complex antibody also bound to a mutant KID-KIX fusion protein that lacks the phospho-acceptor Ser-133 (Fig. 3B, lane 3). The reduced intensity, relative to wild-type KID-KIX, is consistent with the lower affinity of KIX for unphosphorylated KID. The mutant S133A KID-KIX fusion protein was expressed at levels comparable with wild type as detected by both anti-KIX and anti-KID antibodies (Fig. 3B,  lanes 7 and 11, respectively).
It has been shown previously that the anti-complex antibody recognizes specifically the ␣ B helix of KIX-bound KID (20). Our present results therefore indicate that the coil to helix transition in the ␣ B region of KID that accompanies binding to KIX is phosphorylation-independent. Thus, low affinity binding of unphosphorylated KID to KIX also results in folding of the ␣ B region into a helical structure.
Significance of Helix Propensity in Binding to KIX-The dependence of binding affinity on the propensity for helix in the free and bound states of the pKID and c-Myb peptides was examined by characterizing pKID29 peptides (Fig. 4A) with mutations at solvent-exposed positions in the ␣ B helix (residues 133-144). These mutations were designed, with the aid of the algorithm AGADIR (27), to increase (pKID29␣ B ϩ5 and pKID29␣ B ϩ6) or decrease (pKID29␣ B -1 and pKID29␣ B -2) the intrinsic helical content of the pKID29 peptide. It is important to note that the pKID29 peptides used in this study are derived from the minimal binding sequence 119 -147, which has 4-fold lower affinity than the longer pKID 101-160 (Table I), presumably due to less stable interactions of KIX with the shorter ␣ A helix.
CD spectra were recorded for the mutant pKID29 peptides in potassium phosphate buffer (5 mM, pH 7.0) to obtain insights into their relative helical propensities (i.e. their propensity to spontaneously form helical structure in the absence of stabilizing intermolecular interactions). The peptides were also titrated with TFE, a co-solvent that is known to induce and stabilize helical structure and that is often used to mimic helix formation upon binding to a receptor (28), to probe the relative stability of the helix formed upon binding of each peptide to the KIX domain. Previous CD and NMR studies of wild type pKID peptides showed that the ␣ A region spontaneously forms helical structure (50 -60% population of helix) in aqueous solutions of the free peptide, whereas the intrinsic population of helix in the ␣ B region is very small (4,6,29). Thus, the helical contribution to the ellipticity at 222 nm ( 222 ϭ Ϫ4000 degrees cm 2 dmol Ϫ1 ) (Fig. 4B) comes predominantly from the ␣ A region. The addition of TFE, up to 40%, to the wild-type peptide results in a decrease in 222 to Ϫ18,000 degrees cm 2 dmol Ϫ1 , corresponding to a change in helix content from ϳ10 to ϳ53%. This large change in ellipticity is due to stabilization of helical structure   (Fig. 4B) are in accord with the AGADIR predictions. The mutations in pKID29␣ B ϩ5 and pKID29␣ B ϩ6 result in more negative values of 222 , attributed to an increase in the intrinsic helicity of the ␣ B region. Assuming that the contribution of ␣ B helix to the CD spectrum of wild type pKID29 is negligible, an estimate of the helical population in the ␣ B region in the mutant peptides can be derived from the changes in ellipticity (ϳ10% helix in pKID29␣ B ϩ5 and ϳ18% in pKID29␣ B ϩ6). The maximal helical content in pKID29␣ B ϩ6 at 40% TFE is the same as wild type peptide and is slightly lowered in pKID29␣ B ϩ5. At 0% TFE, 222 for pKID29␣ B -1 and pKID29␣ B -2 differs only slightly from that of wild type pKID29, consistent with our assumption that the ␣ B helix contributes little to the ellipticity, but there is a significant difference in the population of helical structure inducible by TFE (Fig. 4B). Mutation of even one solvent-exposed residue to Gly results in significant destabilization of the ␣ B helix in 40% TFE; the final ellipticity at this TFE concentration represents ϳ34% helix, which arises largely if not entirely from the ␣ A helix. These results imply that pKID29␣ B -1 and pKID29␣ B -2 should be impaired in forming the ␣ B helix in the complex with KIX and should therefore have a significantly lower binding energy than the wild-type pKID29 peptide. The ellipticity of the Myb25 peptide (data not shown) is higher than that of the wild type KID29 peptides in aqueous solution, corresponding to the small but significant increase in intrinsic helicity of this peptide (6). As TFE is added, the ellipticity decreases to a minimum of about Ϫ32,000 degrees cm 2 dmol Ϫ1 , corresponding to over 90% helix formation. This is consistent with the C␣ chemical shift plot in Fig. 1B, which shows that the KIX-bound Myb25 peptide is completely helical between residues 295 and 307, without the kink between ␣ A and ␣ B that is seen for pKID (4,8). This result implies that, although pKID and c-Myb bind in a similar site on KIX, the structure of the Myb peptide in the complex may have interesting differences from that of pKID.
Dissociation constants for binding of the mutant peptides to KIX were measured by ITC (Fig. 4C and Table I). The introduction of helix-destabilizing glycine residues in the ␣ B helix in pKID29␣ B -1 and pKID29␣ B -2 leads to a very significant loss of binding affinity. The loss of affinity is 1 order of magnitude greater for the double mutant, although the CD data indicate that the two mutants have similar helical content in 40% TFE. These data show unequivocally that the stability of the ␣ B helix of pKID in the bound state is a major determinant of high affinity binding. In contrast, a double mutation to glycine in the ␣ A helix of the peptide pKID29␣ A -2 reduced the affinity for KIX only 2-fold (Table I). These results suggest that formation of helical structure in the ␣ A region contributes little to the overall binding affinity. However, N-terminal truncation of the peptide to completely eliminate the ␣ A helix, in pKID21, results in a ϳ25-fold decrease in binding affinity. This can be mainly attributed to loss of Leu-128, which forms part of the hydrophobic interface between pKID and KIX in the NMR structure (4). Thus, our data clearly indicate that a helical conformation is necessary for ␣ B residues to interact favorably with KIX but that helical structure is not essential in the ␣ A region of pKID. The importance of helix propensity for c-Myb interactions with KIX is also indicated by the inability of helix-disrupting c-Myb mutants L301P or E299P to bind KIX and activate transcription (6). It is likely that the loss of c-Myb activity observed in this functional assay reflects an inability to fold onto the KIX template.
The wild type pKID29, pKID29␣ B ϩ5, and pKID29␣ B ϩ6 peptides have comparable helicity in 40% TFE, and all bind KIX with high affinity (Fig. 3C, Table I). The substitution of five residues in pKID29␣ B ϩ5 results in a slight decrease in binding affinity relative to wild type pKID29, although all of the sites of mutation are on the solvent-exposed surface of ␣ B . The molecular basis for this is unknown, although we note that substitution of Lys-136 and Asn-139 with other helix-favoring side chains has been shown previously to make KIX binding less favorable (20). Substitution of Asp-140 in pKID29␣ B ϩ5 by Glu, in the peptide pKID29␣ B ϩ6, leads to a 3-fold increase in binding affinity ( Table I). The increased binding affinity of pKID29␣ B ϩ6 might reflect the increased helical population in the ␣ B region of the free peptide, although at this stage we cannot rule out potential contributions from more favorable interactions between the Glu-140 side chain and Lys-606 in helix ␣ 1 of KIX. What is clear, however, is that the dominant factor that determines the relative binding affinities of the various phosphorylated KID peptides is the stability of the ␣ B helix in the KIX-bound state.
Taken together, these results indicate that although the propensity for helix formation in the free state of the transcriptional activator domain may have a small influence on its affinity for KIX, the stability of the helix formed upon binding (particularly the ␣ B helix of pKID and, for c-Myb, the entire peptide) plays a major role in determining the overall affinity.
Significance of Phosphorylation in Binding of KID to KIX-Phosphorylation of KID on Ser-133 is required for binding to KIX and subsequent transactivation (2, 3). The contribution of the phosphate to the binding affinity of the inducible transcrip-  Table I. tional activator domain is clearly very large (Fig. 3A), but the question remains to what extent the other, mainly hydrophobic, interactions between the remainder of the peptide and KIX can compensate for the absence of the phosphate. These contributions have been examined by determining the binding affinity of pKID29 with mutants of KIX where critical residues that interact with the phosphate and other groups have been changed. The NMR structure of the pKID-KIX complex (4) showed that the phosphate group, located at the N terminus of the KID ␣ B helix, forms intermolecular contacts with the critical Tyr-658 and Lys-662 of KIX. Fig. 5 shows that changing Tyr-658 to Phe, which removes the possibility of the hydrogen bond between the phosphate and the ring hydroxyl of the Tyr, greatly impairs binding to pKID, giving the same binding constant as that for wild-type KIX with unphosphorylated KID (Table I). The K662A mutant also caused a decrease in binding affinity but to a lesser extent, in agreement with previous studies (6) that suggested a secondary role for this residue in interaction with the phosphate group. These results strongly suggest that the induction of transcriptional activity in CREB by phosphorylation is primarily due to the contribution of the phosphoserine to the intermolecular interactions in the CREB-CBP complex, primarily through hydrogen bond formation with the side chain of Tyr-658.
The behavior of the Y650A mutant protein illustrates the dependence of the binding of pKID to KIX on both the phosphate moiety and hydrophobic groove interactions. Tyr-650 of KIX forms one wall of the deep hydrophobic pocket that accommodates Leu-141 from the ␣ B helix of pKID. Mutation of this critical hydrophobic residue resulted in a 12-fold decrease of affinity, demonstrating the importance of hydrophobic groove interactions for complex formation (Fig. 5).
KID-KIX Complex Formation in Vivo-In the absence of stimuli that result in the phosphorylation of Ser-133, CREB has very low transcriptional activity in vivo; any residual activity has been attributed to a small degree of Ser-133 phosphorylation (19). Our findings, however, raise the possibility that unphosphorylated KID might be able to bind to CBP in vivo. To address this question, we used a recently described FRET system, in which KIX was fused to the enhanced cyan fluorescent protein (ECFP) and KID was fused to the enhanced yellow fluorescent protein (EYFP) (10). Expression of KIX-ECFP and KID-EYFP in HEK293 cells, followed by treatment with a cAMP agonist, showed clear FRET from KIX-ECFP to KID-EYFP, indicating in vivo complex formation of pKID and KIX (Fig. 6). Substitution of KID-EYFP by the mutant S133A KID-EYFP, in which the PKA phosphorylation site is absent, resulted in a small but significant difference in the FRET signal from the control (Fig. 6). Integration of fluorescence loss and gain in the relevant wavelength ranges leads to estimates of basal binding to be ϳ15% of the inducible binding. Thus, binding of unphosphorylated KID to KIX can be detected in living cells as well as in vitro. Since these experiments require overexpression of KID and KIX in the cell, the functional significance of this interaction remains to be determined.

Determinants of Binding to the KIX Hydrophobic Groove-
The results presented here highlight the significance of both secondary structure and phosphorylation for recognition of an inducible transcriptional activator by the KIX domain of the coactivator CBP and provide insights into the differences that are mandated in a constitutive transcriptional activator by the absence of the phosphate group. The combination of a relatively low binding constant for the constitutive activator and an even lower binding constant for the uninduced CREB suggests that competition between these factors (and others) for CBP may constitute a physiological control mechanism. According to our findings (schematically illustrated in Fig. 7A), the ability of the activator to form an amphipathic helix is necessary and sufficient to form a constitutive low affinity complex with KIX. Both the inducible pKID and constitutive c-Myb activation domains bind to the same face of KIX, formed by the ␣ 1 and ␣ 3 helices. Hydrophobic interactions comprise the major driving force for low affinity binding. The higher affinity of KIX for Myb25 relative to unphosphorylated KID suggests that the interactions at the common binding surface may be more extensive for Myb25. This is supported by the changes observed in KIX chemical shifts upon binding of Myb25, which, as shown in Fig. 2, extend significantly beyond the pKID contact surface.
The observed differences in the molecular interactions with pKID and c-Myb may be crucial to enable constitutive binding of c-Myb to KIX and at the same time to achieve low level basal interactions between CREB and KIX (Fig. 7). In accordance with this suggestion, Shaywitz et al. (30) showed that the magnitude of CREB-dependent transcriptional activity is determined by the strength of KID-KIX interaction. In particular, the L607F KIX mutant interacted more efficiently than wild type KIX with KID, leading to a significant increase in both basal and induced transcriptional activity of CREB but not c-Myb. Since Leu-607 is a core residue (4), these results imply that this KIX mutation may rearrange the ␣ 1 -␣ 3 interface in a way that specifically enhances binding of KID but not c-Myb. Forced overexpression of unphosphorylated KID led to detect-  Table I. FIG. 6. Measurement of KID-KIX complex formation in vivo by FRET. Emission spectra (excitation at 430 nm) of living HEK293 cells transfected with KID-EYFP together with KIX-ECFP (black line) or with S133A KID-EYFP together with KIX-ECFP (gray line) or with ECFP together with ECFP (control, dashed line). Cells were treated with forskolin to achieve PKA-mediated phosphorylation of KID on Ser-133 when present. According to calculation of fluorescence loss at 460 -500 nm and gain at 510 -540 nm, the net FRET signal of S133A KID-EYFP is 14.6 Ϯ 0.3% of that of wild-type KID-EYFP. able binding to KIX in living eukaryotic cells (Fig. 6) as well as in an E. coli two-hybrid system (30), indicating that under physiological conditions, low basal activity of CREB may be ensured by competition with constitutive activators such as c-Myb. An extracellular signal leading to the phosphorylation of KID on Ser-133 by PKA would increase the affinity of CREB for KIX to ϳ20-fold higher than that of c-Myb and therefore allow CREB to efficiently compete with c-Myb and other constitutive activators. A model for such a mechanism is shown in Fig. 7B.
Possible Roles for Induced Fit in CREB Signaling-Our data show that the stability of the ␣ B helix of pKID in the KIX-bound state is a critical determinant of binding affinity; destabilization of this helix through glycine mutagenesis diminishes binding affinity by 1-2 orders of magnitude. In contrast, changes in the intrinsic population of the ␣ B helix in the unbound state have only a modest influence on the thermodynamics of complex formation. The evolutionary conservation of a sequence that is able to form stable helical structure only upon binding to its target raises interesting questions regarding a possible physiological role for the unfolded state of unbound KID. It is possible that this intrinsic lack of structure may confer a functional advantage for CREB by allowing it to serve as a ligand for several different proteins. In particular, the catalytic subunit of PKA has been shown to bind a peptide inhibitor in an extended conformation (31). Therefore, the absence of strong secondary structural propensities in the unbound state of KID may facilitate interaction with and phosphorylation by PKA, which in turn promotes high affinity binding to CBP though the favorable hydrogen bonding interactions formed by the phosphoryl group (4). In addition, a theoretical study (32) suggested that the speed of molecular recognition can be enhanced by having folding (necessary for the required specificity) coupled to binding rather than occurring before. Bienkiewicz et al. (33) have recently demonstrated for the p27 Kip1 -cyclin A-Cdk2 system that binding-coupled folding has an advantage over a preformed structure by acceleration of molecular recognition. Therefore, apart from the thermodynamic differences, kinetic discrimination between the constitutive activator c-Myb and kinase-inducible activation domain of CREB is possible due to the larger degree of intrinsic disorder in the ␣ B helix region of unbound pKID.
Relationship between Affinity for KIX and Transcriptional Activation Mechanism-The 20-fold lower affinity of Myb25 relative to pKID is in apparent contradiction to their comparable in vivo activities (6). However, this discrepancy may be explained by the differing kinetic profiles for the two transcription factors. The rate-limiting step for initiation of CREBmediated transcription is transport of the catalytic subunit of PKA to the nucleus (19), whereas dephosphorylation by the Ser/Thr phosphatase PP-1 rapidly turns CREB off (34). In contrast, constitutive c-Myb activity is necessary to maintain the proliferative state of immature hematopoietic cells (35,36). The activity of c-Myb is mainly regulated at the level of its expression and degradation (reviewed in Refs. 7 and 37). Therefore, activation of the inducible pathway mediated by CREB may require high affinity binding to CBP in order to compete effectively with constitutive pathways converging at CBP. The low binding affinity of the c-Myb activation domain is consistent with the constitutive activity of c-Myb, which has a cumulative effect over time. The fact that c-Myb binds CBP with a lower relative affinity might also be an advantage for regulation by other transcription factors, to either increase (CCAAT/enhancer-binding protein ␤, Ets-1, and AML-1) or attenuate (GATA-1) its activity in a synergistic fashion (reviewed in Refs. 1, 38, and 39) Conclusion-Our results provide new insights into the mechanism of recognition of various transcriptional activator domains by the highly conserved KIX domain of CBP and p300 and illustrate the structural and thermodynamic basis for constitutive activation by c-Myb and inducible activation by CREB. We suggest that the minimal requirement for interaction with the hydrophobic groove of KIX is binding-coupled stabilization of an amphipathic helix, which we have shown to be sufficient for the generation of a low affinity complex. Modulation of that complementarity enables the KIX domain to distinguish between the constitutive c-Myb that activates transcription and the unphosphorylated form of CREB that should not activate transcription. Induction of CREB by phosphorylation converts the low affinity complex into a specific high affinity complex mainly via formation of additional intermolecular interactions. aration and Drs. Eduardo Zaborowski, Gerard Kroon, and John Chung for expert advice regarding NMR experiments. We are particularly grateful to Drs. Natalie Goto and Roberto De Guzman for critical reading of the manuscript, and we thank Drs. Ishwar Radhakrishnan and Gabriela Perez-Alvarado for helpful discussions.