Glutamine-rich regions of the disordered CREB transactivation domain mediate dynamic intra- and intermolecular interactions Authors: Maria A. Martinez-Yamout, Irem Nasir, Sergey Shnitkind, Jamie P. Ellis, Rebecca B. Berlow, Gerard Kroon, Ashok A. Deniz, H. Jane Dyson, Peter E. Wright Published: 2023 ABSTRACT SignificanceThe CREB (cyclic AMP response element binding protein) transcription factor contains a large N-terminal transactivation domain comprising a kinase inducible activation domain and two glutamine-rich regions. Here, we use complementary biophysical methods to show that the transactivation domain is intrinsically disordered and largely extended but exhibits local preferences for secondary structure. The glutamine-rich regions have a propensity to form transient β-hairpin structures that function as binding sites for other glutamine-rich sequences. These motifs mediate both intramolecular interactions within the CREB transactivation domain and intermolecular interactions with a glutamine-rich region of subunit 4 of transcription factor IID (TFIID). This work provides insights into the molecular basis by which glutamine-rich transactivation domains support dynamic and transient interactions between transcription factors and the basal transcription machinery. Intrinsically disordered regions (IDRs) are highly abundant in eukaryotic transcription factors (1, 2) where they perform a critical role in chromatin binding, recruitment of cofactors, regulation of transcription, and assembly of transcriptional hubs (3–8). The lack of stable structure and the intrinsic dynamic flexibility of IDRs is central to their functions in transcriptional regulation. Transcriptional activation domains are intrinsically disordered, contain low-complexity regions, and interact with other components of the transcriptional machinery through multiple mechanisms. Many contain short amphipathic regions (termed short linear interaction motifs) embedded in the IDR that fold upon binding and impart specificity to their interactions (9). Other activation domains undergo ill-defined, fuzzy interactions with their binding partners or mediate condensate formation through weak and transient multivalent interactions (10–12). An understanding of the fundamental molecular mechanisms by which activation domains perform their diverse functions remains limited, and there is an urgent need for detailed characterization of their conformational propensities and elucidation of the molecular determinants of their interactions. In the present work, we apply integrated biophysical methods to characterize the intrinsically disordered transcriptional activation domain of the cyclic AMP response element (CRE)–binding protein (CREB) to derive structural insights into the mechanism by which it regulates CREB transcriptional activity. CREB is a member of the CREB-ATF family of basic leucine zipper (bZip) transcription factors. It is ubiquitously expressed and regulates numerous biological processes including growth factor and calcium signaling, hematopoiesis, gluconeogenesis, spermatogenesis, neurodevelopment, neuronal plasticity, long-term memory, and cell survival (13–16). Dysregulation of CREB signaling promotes oncogenesis and hematological malignancies and contributes to neuropathological disorders (17–19). CREB regulates ~4,000 target genes in the human genome, binding with differential affinities to conserved palindromic (TGACGTCA) and half-site (CGTCA/TGACG) CREs and to some variant and clustered CREs (20, 21). CREB is phosphorylated in a tissue-specific manner by more than 20 different kinases, in response to numerous cellular and extracellular stimuli (22). Binding of CREB to cellular CREs is not constitutive (23), and the CREB transcriptional response depends upon the pattern of phosphorylation and interactions with coactivator proteins such as the CREB-binding protein CBP and its paralog p300 and the CRTC family of coactivators (16). CREB is overexpressed and constitutively phosphorylated in several human cancers; is involved in tumor initiation, progression, and metastasis; and is a potential target for cancer therapy (17, 24). CREB (Fig. 1) is typical of eukaryotic transcription factors in having a long IDR that plays an essential functional role (1, 2). Apart from the bZip domain, which folds upon dimerization and binding to the CRE (25, 26), the Q1, KID (kinase inducible activation domain), and Q2 activation domains of CREB (residues 1 to 280) are predicted to be intrinsically disordered. Schematic diagram showing the domains of CREB, the two glutamine-rich domains Q1 and Q2, the kinase-inducible domain KID, and the bZip. All domains are intrinsically disordered in isolation. Phosphorylated KID (pKID) folds into two helices, αA and αB, upon binding to the KIX domain of CBP (27). The structure of the complex (1KDX) is shown at the Bottom Left. The bZip domain dimerizes and binds to DNA (25), and the structure 1DH3 is shown at the Bottom Right. Interactions between CREB and CBP/p300 play a central role in cyclic-AMP (cAMP) signaling (28). In response to increased cAMP levels, the KID of CREB is phosphorylated at Ser133 by protein kinase A (PKA), resulting in recruitment of CBP through interaction with the KIX domain and activation of gene expression from the CRE (13, 14, 16). A constitutive glutamine-rich (Q-rich) transactivation domain, Q2, interacts with the Q1 domain of the TAF4 subunit of TFIID and mediates promoter occupancy and residence time on chromatin (29–32). Long Q-rich activation domains are found in several transcription factors (33) but the structural and molecular determinants by which Q-rich regions mediate transcriptional activation remain poorly understood (34). Current molecular-level understanding of CREB is derived from a reductionist approach, from structural analysis of the bZip:CRE and pKID:KIX complexes (25, 27). However, numerous biochemical assays reveal enhanced binding of coactivators by full-length CREB and point to shape and conformational changes upon phosphorylation and CRE binding and to thermodynamic coupling between the KID and bZip domains (35–39). Understanding the molecular basis of these processes requires a holistic approach to analyze the conformational ensemble of full-length CREB and its response to CRE and coactivator binding and post-translational modification. We posit that the intrinsic disorder in CREB is central to its regulatory function, rendering the key activation domains accessible to the post-translational modifications [phosphorylation (22), O-glycosylation (40, 41)] that modulate interactions with target proteins and promoting dynamic coupling between cellular signaling and coactivator and CRE binding events. The prevalence of intrinsic disorder in CREB presents challenges to the characterization of its structure and interactions. Crystallographic analysis is virtually impossible, and we rely on solution-state methods such as fluorescence and NMR spectroscopies. The extent of disorder in CREB is extremely high, and NMR spectra of the full-length protein are poorly resolved (42), making the task of resonance assignment particularly challenging. In addition, resonances of the bZip dimer are broadened beyond detection. In the present work, we have used NMR, single-molecule Förster resonance energy transfer (sm-FRET), and small-angle X-ray scattering (SAXS) to characterize the conformational propensities and interdomain interactions in the transcriptional activation domain of CREB (residues 1 to 280, termed the CREBTAD). We have employed a strategy of characterizing the isolated Q1, KID, and Q2 domains in detail, and then mapping these results onto the NMR data obtained for the full-length CREBTAD. This strategy is well suited to the characterization of disordered domains, and differences in the behavior of local regions of the protein when present in isolated domains or larger constructs can provide important insights into intramolecular interactions that mediate biological function. This approach has previously been used to characterize the interdomain interactions in the tumor suppressor p53 (43, 44). Here, we employ NMR chemical shift perturbation and paramagnetic relaxation enhancement (PRE), together with SAXS and sm-FRET to characterize the conformational propensities of the CREB transcriptional activation domain, as a step toward characterization of the full-length CREB dimer. Results Conformational Ensemble of the CREB Transactivation Domain. SAXS experiments were performed to characterize the global dimensions of the full-length transactivation domain of CREB (CREBTAD, residues 1 to 280). The dimensionless Kratky plot (Fig. 2, additional plots in SI Appendix, Fig. S1), which lacks the bell-shaped curve characteristic of a globular protein, is consistent with an expanded intrinsically disordered protein. Dimensionless Kratky plot of small-angle scattering data for CREBTAD (black) and pCREBTAD (red), using merged data. An equivalent plot using raw data is shown in SI Appendix, Fig. S1, together with a plot of log I(q) vs. q for CREBTAD and pCREBTAD. Fitting of the individual CREBTAD SAXS profiles using the heteropolymer molecular form factor (MFF) (45) gave an average Rg = 53 ± 1 Å and Flory exponent ν = 0.542 ± 0.007. For reference, a fully denatured protein and a compact, globular protein of the same number of amino acids would have Rg values of ~56 Å (with ν = 0.598) and 25 Å, respectively (46, 47). The SAXS data indicate that the CREBTAD conformational ensemble is expanded yet is slightly more compact than a chemically denatured polypeptide of the same length. Changes were observed in the SAXS profile and dimensionless Kratky plot upon phosphorylation of S98 and S133 by PKA (Fig. 2). Phosphorylation appears to have little effect on the global dimensions of the conformational ensemble of pCREBTAD, which has an average Rg of 52 ± 1 Å from MFF fits of the individual SAXS profiles. While the Rg values are similar for CREBTAD and pCREBTAD, the Flory exponents differ significantly (ν = 0.542 ± 0.007 for CREBTAD vs. ν = 0.60 ± 0.02 for pCREBTAD). The increase in ν suggests that upon phosphorylation, intrachain interactions become less favorable relative to interactions with the solvent (45). Since Rg = R0Nν, where N is the number of residues, the observed increase in ν without substantial change in Rg implies a decrease in the pre-exponential factor R0 (45). R0 is a function of the persistence length of the polypeptide chain (46) and the decrease upon phosphorylation is consistent with changes in persistence length due to loss of local intrachain interactions. Conformational Propensities of CREBTAD Subdomains. To obtain insights into local conformational propensities, we devised a strategy that begins with NMR investigations of the isolated Q1, KID, and Q2 domains and progresses by mapping their properties to the full-length transactivation domain of CREB (residues 1 to 280). We first expressed and purified 15N-labeled constructs spanning residues 1 to 100 (Q1), 85 to 160 [corresponding to KID, and including the α-region (residues 87 to 100) that is deleted from the alternately spliced Δ isoform (14, 48)], and 160 to 280 (Q2) and recorded 1H-15N heteronuclear single-quantum coherence (HSQC) spectra for each (SI Appendix, Fig. S2). All three spectra show little dispersion in the 1H dimension, indicating that the individual domains are intrinsically disordered. Resonances of the 15N, 13C-labeled domains were assigned using standard triple resonance 3D methods. The chemical shifts for most residues are close to random coil values predicted using POTENCI (49). The per-residue secondary chemical shift (observed chemical shift minus sequence-corrected random-coil chemical shift) for the three isolated domains is plotted in SI Appendix, Fig. S3. For the individual domains, Cα and Cβ chemical shifts were used to calculate the secondary structure propensities using SSP (50). Positive and negative SSP scores indicate propensity for the helical or β regions, respectively, of φ, ψ dihedral angle space, and the magnitude reflects the population. The SSP scores for the isolated CREB domains are shown in Fig. 3. Plot of secondary structure propensities calculated from the secondary chemical shifts of Cα, Cβ, and CO for the three isolated domains, using the program SSP (50). The boundaries of each construct are indicated by colored rectangles at the top of the figure. Red arrows within the graph show the location of putative β-turns identified using the MICS server (http://spin.niddk.nih.gov/bax/nmrserver/mics) (51). In several regions of the Q1 sequence (residues 29 to 32, 42 to 45, 51 to 57, and 62 to 73), there is a ~15 to 30% propensity for dihedral angles in the β-region of conformational space. Analysis of the chemical shifts using the MICS server (http://spin.niddk.nih.gov/bax/nmrserver/mics) (51) also indicates a propensity for β-structure between residues 52 to 57 and 62 to 67 and formation of a transiently populated β-turn between residues L58PNG61, a sequence with high β-turn propensity (52). Toward the C-terminal end of Q1, residues 89 to 94 sample a small (~14%) population of helix. It is interesting that this region is not present in the ΔCREB isoform, where residues 87 to 100 are deleted by alternate splicing (27). In the isolated KID peptide, SSP analysis of the chemical shifts indicates that residues 117 to 129 transiently sample an ~35% population of helix, in agreement with previous reports (53). Residues toward the C-terminus of KID have a propensity to populate the β-region of φ, ψ space (Fig. 3). Finally, we also observed propensities (average SSP scores of 0.25 to 0.3) for population of the β region of φ, ψ space for residues 172 to 177, 181 to 185, 190 to 194, 222 to 226, and 231 to 234 within Q2. MICS analysis of the chemical shifts indicates a propensity for β-turn formation by residues 177 to 180 and 227 to 230, consistent with transient formation of β-hairpins by residues 172 to 185 and 222 to 235. In summary, while the Q1, KID, and Q2 domain peptides are clearly disordered, they do not behave as random coils but have distinct propensities to sample transient elements of secondary structure in localized regions. NMR Characterization of CREBTAD. The 1H-15N HSQC spectrum of the CREBTAD clearly shows that the Q1, KID, and Q2 domains remain disordered, as in the isolated domains (Fig. 4 and SI Appendix, Fig. S4). Except for residues at domain boundaries, cross-peaks for most residues have very similar chemical shifts [Δδ(1HN) < 0.01 ppm, Δδ(15N) < 0.05 ppm] in the 1H-15N HSQC spectra of the CREBTAD and the isolated domains, so that assignments can be readily transferred (Fig. 4 and SI Appendix, Fig. S4). Region of the 1H-15N HSQC spectrum of CREBTAD (residues 1 to 280) of CREB, processed with iterative soft thresholding (IST) (black), showing assignments mapped from the spectra of the individual domains Q1 (green), KID (red), and Q2 (blue). Overlays of the complete CREBTAD spectrum and those of the isolated domains are shown in SI Appendix, Fig. S4. Resonances that could not be sequentially assigned in the spectra of the individual domains are shown with a star. Despite the overall similarity between the spectra of the isolated domains and CREBTAD, small differences in chemical shift are observed for several residues (Fig. 4), and some cross-peaks are of decreased intensity or are missing from the spectrum of CREBTAD. These changes are plotted in SI Appendix, Fig. S5A. Small chemical shift changes are observed for T47 and for several residues between T54 and the C-terminal end of Q1. In Q2, small shift changes are observed for several residues in the regions from 176 to 191 and 219 to 235. We emphasize that these changes are small (<0.06 ppm for 1HN and <0.2 ppm for 15N) and for the most part do not complicate the transfer of assignments between the isolated domains and CREBTAD. The 1H-15N HSQC spectrum of CREBTAD shows changes with increased solution concentration. Spectra were recorded at 10 μM, 50 μM, and 200 μM concentrations of CREB. The intensity of several cross-peaks is reduced as the concentration is increased, shown plotted in SI Appendix, Fig. S5B. Lowered cross-peak intensity is observed for residues 54 to 67 in Q1 and 174 to 194 and for several residues between 220 and 260 in Q2. PRE Experiments. Shift and broadening of cross-peaks upon incorporation of the isolated domains into longer constructs, together with concentration-dependent changes in peak position and intensity, are signs that interactions are occurring between the various regions of the intact CREBTAD. These interactions could be intramolecular or intermolecular. Information on the locations of the intramolecular interactions was obtained from PRE and sm-FRET experiments, and potential intermolecular interactions were probed using PRE experiments. Three sites within the CREBTAD were selected for attachment of the (1-oxyl-2,2,5,5-tetramethylpyrroline-3-methyl)methanethio- sulfonate (MTSL) spin label. CREBTAD contains a single cysteine at C90, and this site was used for one set of PRE data. The second site is in the Q1 domain; the construct for these experiments was a double mutant where the endogenous C90 was mutated to alanine and A50 was mutated to cysteine (A50C/C90A). The third site was in the Q2 domain; the construct used was C90A/T172C. It would be expected that the mutations themselves might cause changes in the NMR spectrum of CREBTAD. Resonance assignments for the mutant proteins were mapped from the CREBTAD spectrum to those of the mutants, as described above for the mapping of assignments to CREBTAD from those of the individual domains. Except for residues at the site of mutation, the majority of the cross-peaks are at nearly identical chemical shifts in spectra of the wild type and mutant CREBTAD constructs (SI Appendix, Fig. S6). The C90A substitution alone led to small 15N chemical shift changes for residues 88 to 94, likely due to a slight increase in helical propensity in this region, as predicted by AGADIR (54). The A50C substitution (in the construct A50C/C90A) caused only minimal chemical shift changes relative to the C90A spectrum except at the site of mutation. In contrast, the T172C substitution (in C90A/T172C) resulted in chemical shift changes (SI Appendix, Fig. S6) both near the site of mutation (169 to 180) and in distant regions of the sequence, including a cluster of residues (54–67) in Q1 and several clusters in Q2 (189 to 193, and 227 to 230). The observation of long-range chemical shift perturbations induced by the T172C mutation is consistent with transient intramolecular contacts between noncontiguous regions of CREBTAD, as also suggested by chemical shift differences between the individual domains and CREBTAD (SI Appendix, Fig. S5A). Attachment of the MTSL spin label and the subsequent reduction step, to give the paramagnetic and diamagnetic spectra respectively, caused few additional chemical shift changes in the 1H-15N HSQC spectra, compared to the mutations alone, but the intensity of many cross-peaks was decreased by the paramagnetic spin label. The HSQC spectra of CREBTAD with the spin label in all three positions (Fig. 5 and complete spectra in SI Appendix, Fig. S7) showed PRE, indicated by the loss of signal intensity between the sample with the paramagnetic spin label (red spectrum) and the same sample where the spin label had been reduced (black spectrum). To estimate the extent and location of any intermolecular interactions in CREBTAD, a PRE experiment was conducted with 15N-labeled C90A without spin label in the presence of an equimolar amount of unlabeled C90A/T172C with MTSL attached at C172. The result is shown in Fig. 5 and SI Appendix, Fig. S7D. Region of an overlay of the 1H-15N HSQC spectra of CREBTAD in the diamagnetic (reduced) form (black) and the paramagnetic (active spin label) form (red). (A) CREBTAD mutant A50C/C90A with spin label at C50. (B) Wild-type CREBTAD with spin label at C90. (C) CREBTAD mutant C90A/T172C with spin label at C172. (D) 15N-labeled CREBTAD mutant C90A in the presence of unlabeled CREBTAD mutant C90A/T172C with spin label at C172. The cross-peaks with decreased intensity due to the presence of the paramagnetic spin label differ between the panels of Fig. 5 and SI Appendix, Fig. S7, as expected for the three different positions of the spin labels within the sequence: Cross-peaks of residues in the immediate vicinity of the spin label should experience the greatest broadening from the unpaired electron. Plots of paramagnetic/diamagnetic cross-peak intensity ratios are shown in Fig. 6. Contacts are clearly occurring between Q1 and Q2, while KID appears to be affected to a lesser extent. Plots of the signal intensity ratio (Ipara/Idia) for each of the spin-labeled constructs of CREBTAD, as a function of residue number. (A) CREBTAD mutant A50C/C90A with spin label at C50. (B) Wild-type CREBTAD with spin label at C90. (C) CREBTAD mutant C90A/T172C with spin label at C172. The intensity ratio for the intramolecular PRE (spin label incorporated into the 15N-labeled protein) is shown in black, and the ratio for the intermolecular PRE (spin label incorporated into an unlabeled C90A/T172C protein, mixed with 15N-labeled C90A without spin label) is shown in blue. Small bars below each panel indicate missing data points. Green bars represent prolines, red bars show residues with cross-peaks in the spectra of the individual domains but not in that of CREBTAD, and purple bars show residues whose cross-peaks are missing from both the isolated domain and CREBTAD spectra. Blue bars denote residues with resonance overlaps that preclude evaluation of the cross-peak intensity. Incorporation of a paramagnetic spin label at residue 172 (in the construct C90A/T172C) leads to extreme broadening of cross-peaks of residues throughout Q2 (Fig. 6C, black), in marked contrast to the behavior expected for an extended, fully disordered polypeptide chain, where broadening is observed only for ±10 residues from the site of the spin label (55). The observed PREs suggest that the Q2 domain undergoes rapid averaging over a highly heterogeneous conformational ensemble in which the spin label makes transient nonlocal contacts with residues throughout the domain. Transient long-range contacts in a disordered or unfolded protein contribute strongly to the PRE but do not perturb the global properties (Rg) of the polypeptide chain (55). Extensive cross-peak broadening is also observed for residues 126 to 143 in the KID domain and for residues 28 to 107 in Q1 and in the N-terminal region of KID (Fig. 6C). The extent of broadening by the paramagnetic T172C spin label is remarkable and shows that this region of Q2 makes transient intramolecular contacts throughout CREBTAD; the only regions that are not affected are residues 1 to 24 at the N-terminus of Q1 and residues 109 to 123 and 155 to 158 of KID, as well as the extreme C-terminus of Q2. It is notable that each of the unaffected regions has an overall negative charge, which likely impairs interactions with the predominantly hydrophobic residues surrounding T172C. All regions that exhibit strong PRE effects from the T172C spin label are either uncharged or, in the case of residues 123 to 139, overall electrically neutral. A control experiment, in which MTSL-labeled 14N-C90A/T172C CREBTAD was mixed in 1:1 stoichiometry with 15N- C90A CREBTAD, confirmed that the PRE effects described above arise predominantly from intramolecular interactions. Weak intermolecular PREs were observed (Fig. 6C, blue) due to transient intermolecular interactions between residues 67 to 74 in Q1 and the paramagnetic spin label at T172C in Q2. However, these interactions are far weaker than the intramolecular interactions between these regions (average Ipara/Idia = 0.41 and 0.02 for intermolecular and intramolecular interactions, respectively). To confirm the presence of intramolecular contacts between Q1 and Q2, an MTSL spin label was coupled at A50C. PRE effects are observed for residues throughout the Q2 domain (average Ipara/Idia = 0.34 for residues 164 to 270), confirming the presence of transient intramolecular interactions between Q1 and Q2 (Fig. 6A). Interactions with the amphipathic αB region of KID (residues 133 to 139) are also evident. There is an interesting pattern of PREs in Q1. Strong PREs are observed in the vicinity of the spin label, as expected, and also in the region between residues 68 and 82, with weaker PREs for the intervening residues 58 to 63. This pattern strongly suggests formation of a well-populated hairpin structure, consistent with the β-turn between residues 58 and 61 predicted from chemical shifts (Fig. 3). With the spin label coupled at C90, PREs are observed in the αB region of KID and in several regions of Q2, with the largest effects in the regions between residues 188 to 197 and 233 to 245 (Fig. 6B). Resonances of residues 67 to 82 in Q1 are strongly broadened, with average Ipara/Idia = 0.07. Interactions with Glutamine-Rich Polypeptides. To obtain further insights into the intra- and intermolecular interactions mediated by the Q1 and Q2 domains, we recorded 1H-15N HSQC spectra of the isolated 15N-labeled Q2 domain in the presence of a 10-fold excess of an unlabeled His6-GB1-Q1 peptide (Fig. 7A). Cross-peaks associated with residues 174 to 185, 189 to 194, and 219 to 236 were reduced in intensity by more than 60% upon addition of His6-GB1-Q1. A control experiment, in which 10-fold excess of His6-GB1 was added to the Q2 domain, showed little change in cross-peak intensity and confirmed that the spectral changes are due to interactions between Q1 and local regions of Q2. Plots of the 1H-15N cross-peak intensity ratio (intensity of resonance with maximum titrant concentration/intensity of free protein resonance) (red bars) and average chemical shift difference for 1H-15N cross-peaks {Δδ = √ [(δH, titrant – δH, free)2 + ((δN, titrant – δN, free)2)/5]} (blue) for (A) CREB Q2 (residues 160 to 280) with the addition of a 10-fold molar excess of CREB Q1. (B) CREB Q2 (residues 160 to 280) with the addition of a fourfold molar excess of TAF Q1. (C) CREB Q1 (residues 1 to 98) with the addition of a fourfold molar excess of TAF Q1. Green bars represent prolines, purple bars show residues whose cross-peaks are missing from the spectra, and blue bars denote residues with resonance overlaps that precluded evaluation of the cross-peak intensity. It should be noted that several residues for which the Δδ (blue) is highest also correspond to cross-peaks of extremely low intensity, which may register as zero in the intensity ratio plot (red). We also recorded HSQC spectra of the 15N-labeled Q1 and Q2 domains of CREB in the presence of increasing amounts (1× and 4×) of the Q1 domain of TAF4 (Fig. 7 B and C). Large decreases in cross-peak intensity and small chemical shift changes were observed; the magnitude of the perturbations increased with increasing amounts of the TAF4 Q1 domain. The largest perturbations in the CREB Q1 spectrum involved the cross-peaks of residues 53 to 72. Addition of a fourfold excess of the TAF4 Q1 domain to CREB Q2 caused severe intensity loss (>80%) for residues 172 to 183, 186 to 194, 205 to 210, 219 to 246, and 251 to 256. sm-FRET Measurements. sm-FRET was used with donor-acceptor dye-labeled CREBTAD to analyze the effect of changing solution conditions on the conformational properties of individual CREB domains within the context of CREBTAD. Importantly, the low (sub-nM) concentrations used in single-molecule detection allow the intrinsic protein conformational properties to be decoupled from the effects of self-association. Four donor-acceptor dye-labeled constructs were prepared, to allow us to monitor the behavior of each individual domain within CREBTAD and between the Q1 and Q2 domains. Alexa 488 and Alexa 594 dyes were coupled to cysteine residues introduced at various sites in CREBTAD in addition to or instead of the endogenous cysteine at position 90: T2C/C90 (referred to as Q1FRET-CREBTAD or Q1), C90/T160C (KIDFRET-CREBTAD or KID), C90A/T172C/T259C (Q2FRET-CREBTAD or Q2), and T2C/C90A/T198C (Q1-Q2FRET-CREBTAD or Q1Q2). FRET efficiencies (EFRET) were measured for individual CREBTAD molecules diffusing in solution, to minimize potential perturbations due to surface interactions. EFRET measured for hundreds to thousands of single molecules were used to generate EFRET-histograms, shown in Fig. 8. FRET histograms for the four constructs of CREB in sm-FRET buffer (20 mM TRIS, and 2 mM DTT, pH 7.5). Locations of fluorescent tags are indicated by stars on the schematic diagrams of the sequence, shown at the Top of the figure. The presence of transient structure in disordered proteins can be identified by their behavior in the presence of chemical denaturing agents. The EFRET-histograms of the CREBTAD constructs all show a single peak that moves continuously to lower EFRET (longer interdye distance) with increasing guanidinium hydrochloride concentration (SI Appendix, Fig. S8). These observations indicate that the intrinsically disordered Q1, Q2, and KID domains within the CREBTAD populate a rapidly fluctuating conformational ensemble (within the ~500 μs time-resolution of the experiments) that includes relatively compact forms. The EFRET values for the Q1FRET-CREBTAD (0.54) and Q2FRET-CREBTAD (0.65) constructs, where the fluorophores have similar spacing (87 and 86 intervening residues, respectively), show that the Q2 domain samples compact forms more frequently than Q1. Both the Q1FRET-CREBTAD and Q2FRET-CREBTAD constructs experience substantial expansion upon addition of denaturant and exhibit similar EFRET values at high guanidinium hydrochloride concentrations (SI Appendix, Fig. S8). The EFRET value (0.8) for the C90/T160C KIDFRET-CREBTAD construct indicates substantial compaction, which is likely driven by positive and negative charge clusters in the KID sequence. PKA phosphorylation of S133 (located within a positive charge cluster) and S98 leads to expansion of the KID region (EFRET decreases from 0.8 to 0.68) (SI Appendix, Fig. S9). The behavior of the CREBTAD over a longer distance scale between the Q1 and Q2 domains was probed by attaching fluorophores at T2C and T198C. The observed EFRET (0.35) is larger than expected given the number of residues between the probes and indicates some compaction of the polypeptide chain. The rms distance between the fluorophores is estimated to be 80 to 100 Å (using equations in ref. 56 and ν values for IDPs and denatured proteins from ref. 57), a distance at which EFRET would be close to zero for this dye pair. Phosphorylation of S98 and S133 shifts the EFRET histogram to smaller values, and this effect is attenuated by the addition of NaCl (SI Appendix, Fig. S9). However, the change in EFRET is small, suggesting that the phosphorylation-induced expansion of the KID domain has little effect on the overall dimensions of the CREBTAD within the time scale of the observations. Phosphorylation causes little change in the PRE of the CREBTAD with the spin label at C90 (SI Appendix, Fig. S10) except for small increases in Ipara/Idia for residues near pS133 and an acidic region (residues 151 to 159) in the C-terminal part of KID that likely reflect electrostatic repulsion of pS98. Increasing the concentration of NaCl from 0 to 1 M had a pronounced effect on the dimensions of the CREBTAD domains (Fig. 9). Increasing salt results in a roughly monotonic increase in EFRET for Q2FRET-CREBTAD, indicating increased compaction of the Q2 domain at high salt. This behavior is attributed to the salting-out effect, which leads to compaction of intrinsically disordered proteins at high salt concentration (58). Increasing NaCl concentration has relatively little effect on the EFRET of Q1 (Q1FRET-CREBTAD, with fluorophores at T2C and C90). Both the Q1 and Q2 domains are comprised predominantly of polar and hydrophobic amino acids and contain very few charged residues. In contrast, the KID region is a polyampholyte with positive and negative charge clusters distributed along its sequence (SI Appendix, Fig. S11). The EFRET value for KIDFRET-CREBTAD, with dyes at C90 and T160C, initially decreases with increasing NaCl concentration up to ~0.4 M and then increases monotonically with further increase in salt concentration. This biphasic behavior arises from electrostatic screening at low salt concentrations, leading to expansion of the KID region, coupled with salting-out effects that result in compaction at NaCl concentrations > 0.4 M. The expansion due to electrostatic screening is abrogated by phosphorylation at S98 and S133 by PKA. Phosphorylation at these sites adds approximately –4 net negative charge, changing the net charge per residue and the charge patterning in the KID region. The KID domain expands upon phosphorylation (EFRET decreases from 0.8 to 0.68), and the biphasic salt dependence is lost; EFRET for the pKIDFRET-CREBTAD increases almost monotonically up to 1 M NaCl (Fig. 9). EFRET for the Q1-Q2FRET-CREBTAD construct also increases monotonically with increasing NaCl concentration (Fig. 9C), showing overall compaction of the CREBTAD due to salting-out effects. Effect of phosphorylation and increasing salt concentration on FRET efficiency. (A) Effect of increasing salt concentration on the FRET efficiency for the four fluorescently labeled CREBTAD samples: T2C/C90 (Q1) (red), C90/T160C (KID) (green), C90A/T172C/T259C (Q2) (blue), and T2C/C90A/T198C (Q1Q2) (black). (B) Effect of increasing salt concentration on the FRET efficiency for C90/T160C (KID) (green) and for C90/T160C phosphorylated at S98 and S133 by PKA (ppKID) (purple). (C) Effect of increasing salt concentration on the FRET efficiency for T2C/C90A/T198C (Q1Q2) (green) and for T2C/C90A/T198C phosphorylated at S98 and S133 by PKA (ppQ1Q2) (purple). Discussion Overall Chain Dimensions. The current NMR, sm-FRET, and SAXS data consistently indicate that the conformational ensemble of the isolated CREBTAD is largely extended and dynamically disordered. For the nonphosphorylated CREBTAD, the shape of the dimensionless Kratky plot is characteristic of an extended, intrinsically disordered protein. The Flory exponent derived from MFF (45) fits of the SAXS profiles, ν = 0.542, is slightly smaller than expected for a fully unfolded protein (ν = 3/5) (59), suggesting some degree of compaction of the polypeptide chain, as confirmed by the sm-FRET histograms of CREBTAD with fluorophores coupled at residues 2 and 198. It is of note that the values of ν determined from MFF fits to SAXS data for three intrinsically disordered proteins (ν ~ 0.54) by Riback et al. (45) are very similar to ν for the nonphosphorylated CREBTAD. The sm-FRET data provide insights into the relative dimensions of the individual domains within the CREBTAD construct and their sensitivity to salt concentration and PKA phosphorylation. The KID and Q2 domains become more compact at high NaCl concentrations while the dimensions of the Q1 domain are largely independent of salt. Phosphorylation at S98 and S133 by PKA results in decreased EFRET for fluorophores at C90 and T160C, suggesting an increase in the average distance between them and expansion of KID. Phosphorylation results in an increase in the Flory exponent to ν = 0.60, consistent with more favorable solvation of the polypeptide and expansion of the KID region, which likely makes residues that are helical in the complex (Fig. 1) more accessible and thereby facilitates binding to KIX. The intrinsic disorder and flexibility of the CREBTAD are central to CREB regulation and function, rendering the activation domains accessible to the post-translational modifications that modulate interactions with target proteins and promoting dynamic coupling between cellular signaling and coactivator and DNA binding events. Biochemical assays point to shape and conformational changes upon phosphorylation and DNA binding and to thermodynamic coupling between the KID and bZip domains (36–38). Multisite phosphorylation in the N-terminal region of KID functions as a rheostat to regulate CREB transcriptional activity by promoting dynamic long-range interactions, which would be facilitated by the flexibility of the intervening Q2 domain, between KID and the basic region of the bZip that compete for binding to DNA (42, 60, 61). Local Conformational Propensities. The limited chemical shift dispersion in the 1H dimension of the 1H-15N HSQC spectra of the Q1, KID, and Q2 domains shows that these are dynamically disordered both as isolated domains and in the full-length CREBTAD. However, careful analysis of differences between predicted random coil chemical shifts and the experimental data reveals that several regions of the CREBTAD exhibit a propensity to populate transient elements of secondary structure. The KID domain contains two subregions, αA (residues 120 to 129) and αB (134 to 144), which form helices upon binding to KIX following S133 phosphorylation (Fig. 1) (27, 53). The αA region adopts an ~35% population of helix between residues 117 to 129 in the isolated, non pKID peptide; the helical structure is further stabilized and the αA helix becomes fully folded upon binding to the KIX domain of CBP (53). In contrast, the αB region is unstructured in the unbound KID peptide and folds into its helical structure only in complex with KIX (27). Analysis of the chemical shifts using SSP (50) and MICS (51) suggests formation of a small (~20%) population of localized β-hairpins in both Q1 (residues 52 to 67) and Q2 (residues 172 to 185 and 222 to 235) that exchange rapidly with unfolded states. PRE data with the spin label coupled at A50C provide supporting evidence for the residue 52 to 67 β-hairpin in Q1. A schematic representation of each of the β-hairpins (Fig. 10) suggests that they form a common motif in which glutamine and hydrophobic amino acids form clusters that could potentially mediate molecular interactions. A notable feature of the hairpins is that all contain a QXXXGQ motif that would position a pair of glutamine side chains in close spatial proximity. We further note the presence of Tyr residues in the Q2 hairpins; it has been shown previously that aromatic residues play a critical role in promoting interactions between glutamine-rich regions (62).Fig. 10.Cartoon representation of the predicted β-hairpin structures in the Q1 and Q2 domains of the CREBTAD. Transient population of these structures is supported by analysis of chemical shifts and PRE intensities and AlphaFold structure predictions. Regions with a propensity for β-structure (green arrows) and β-turns (black brackets) are indicated. Cartoon representation of the predicted β-hairpin structures in the Q1 and Q2 domains of the CREBTAD. Transient population of these structures is supported by analysis of chemical shifts and PRE intensities and AlphaFold structure predictions. Regions with a propensity for β-structure (green arrows) and β-turns (black brackets) are indicated. It is instructive to compare the conformational propensities identified by NMR in solution with predicted structures for the rat CREBTAD in the AlphaFold2 database (https://alphafold.ebi.ac.uk) (63). AlphaFold2 has the ability to predict conditionally folded structures within intrinsically disordered proteins, i.e., structures of folded states formed upon interactions with binding partners (64). For KID, both the αA (119 to 130) and αB (132 to 143) helices are predicted with very high confidence [predicted local difference distance test (pLDDT) score > 90] by AlphaFold2. The boundaries of the helices correspond closely to the αA and αB helices in the pKID:KIX complex (27), although only αA populates helix in an isolated KID peptide. The Q1 and Q2 β-hairpins are also predicted by AlphaFold2 with varying degrees of confidence (residues 52 to 67, pLDDT < 50; residues 172 to 185, pLDDT 70 to 90; residues 222 to 235, pLDDT 50 to 70). AlphaFold2 also predicts an additional β-strand between residues 191 to 194 that adds to the 172 to 185 β-hairpin to form a small β-sheet (SI Appendix, Fig. S12). SSP analysis suggests that residues 190 to 194 have a weak propensity to populate β-structure in the Q2 peptide in solution. We posit that the structures predicted by AlphaFold2 are likely conditional structures that may be stabilized by binding interactions but are only weakly and transiently populated in the unbound CREBTAD. We stress that the β-hairpins observed by NMR in the isolated Q1 and Q2 domains in solution are transient, weakly populated structures that exchange rapidly with other states in a dynamic conformational ensemble. Dynamic Intramolecular Interactions in the CREBTAD. Chemical shift and PRE measurements indicate transient intramolecular interactions within Q2 and between the Q1 and Q2 domains in the full-length CREBTAD. The T172C mutation (Q2) causes small changes in the chemical shift of residues 54 to 67 (SI Appendix, Fig. S5A), a site of transient β-hairpin formation in Q1. Small chemical shift perturbations for residues 189 to 193 and 227 to 230 are also induced by the T172C mutation, consistent with transient interactions between noncontiguous regions of Q2. Both the intra-Q2 domain interactions and the Q1/Q2 interactions were confirmed in PRE experiments (Fig. 6), and additional interactions between the T172C spin label and the αB region of KID were observed. Addition of excess H6-GB1-Q1 peptide to 15N-labeled Q2 results in decreased intensity and chemical shift changes for Q2 HSQC cross-peaks, helping to localize the sites of interaction (Fig. 7A). The regions that contain weak or missing cross-peaks are also those that are most highly broadened in the PRE experiments. Interestingly, the largest perturbations are localized to the residues that form the transient β-hairpins shown in Fig. 10, plus residues 189 to 194 that correspond to the 3rd β-strand of a small β-sheet predicted by AlphaFold2. Taken together, the NMR data strongly suggest that interactions between Q1 and Q2 are mediated by the regions that exhibit a propensity for β-hairpin formation. Intermolecular Interactions Mediated by the β-Hairpin Motifs. The Q2 domain functions as a constitutive transcriptional activator by mediating interactions with the TAF4 subunit of TFIID and stabilizing CREB binding to chromatin (31, 32, 65). The residence time on chromatin is greatly shortened by deletion of Q2 but is fully restored by the glutamine-rich QB activation domain of Sp1 (31), which binds the Q1 domain of TAF4 (29, 34, 66). The Q1 domain of TAF4 is necessary and sufficient for binding and activation of a reporter gene by CREB, and the interaction of TAF4 with both CREB and Sp1 is largely abrogated by an F449S mutation in the TAF4 Q1 domain (29, 66). To probe the interaction between the CREB activation domain and TAF4, we acquired HSQC spectra of the 15N-labeled Q1 and Q2 peptides in the presence and absence of an unlabeled TAF4 Q1 domain peptide (Fig. 7). Changes in chemical shift and decreases in cross-peak intensity were observed for residues located in or near the β-hairpin motifs in the Q1 and Q2 domains of the CREBTAD, implicating these as the primary interaction sites for binding to TAF4. Thus, the β-hairpin motifs mediate both intramolecular interactions within CREB and intermolecular interactions with glutamine-rich domains in CREB binding partners. Because of the resonance broadening, we are unable to determine whether the interactions lead to formation of a stabilized secondary structure or whether the complexes remain highly dynamic. Mutation or deletion of residues in the Q2 β-hairpin motifs of CREB modulates its interactions with TAF4 and its residence time on chromatin. Deletion of residues 162 to 179 or substitution of residues 182 to 185 (YIAI to AAAA) greatly reduces the chromatin residence time and reporter gene activity (31). These mutations would disrupt the first and second β-strands, respectively, of the residue 172 to 185 β-hairpin motif (Fig. 10). TAF4 binding and transcriptional activity is impaired (30, 31, 65) by mutation of CREB residues G189, I191, and L193, which form the 3rd strand of a small β-sheet in the AlphaFold model (SI Appendix, Fig. S12). Amide cross-peaks of these residues are strongly broadened in the presence of the TAF4 Q1 domain peptide, providing confirmation of a direct interaction. Similarly, single or multisite mutations of an LQTL motif (residues 204 to 207) abrogate TAF4 binding and transcriptional activity (30, 31, 65), consistent with our NMR data that show interactions with the TAF4 Q1 domain in this region. Our NMR study also reveals TAF4 interactions with the residue 222 to 235 β-hairpin motif, but this does not appear to have been probed by mutagenesis. Finally, we observe interactions of the TAF4 Q1 domain with the residue 52 to 67 β-hairpin motif in CREB Q1 (Fig. 7), but these appear to be secondary to the Q2 domain interactions since deletion of the CREB Q1 domain has little effect on chromatin occupancy (31). The picture that emerges is that the weakly populated β-hairpin motifs in the CREBTAD play a critical role in mediating interactions with glutamine-rich regions, binding to each other through intramolecular interactions and to the Q1 domain of the TAF4 subunit of TFIID. The CREB Q2 domain contains multiple binding sites for TAF4, likely increasing the avidity. The Q2 domain alone, when fused to the DNA binding domain of Gal4, functions as a potent transcriptional activator in vitro and dominates the interactions between CREB and TAF4 (32, 67, 68). In contrast, the CREB Q1 domain, which contains only a single binding site for TAF Q1, is ineffective as a transcriptional activator although it is required for full basal activity (67). We postulate that efficient recruitment of CREB to TFIID requires dynamic multivalent interactions between the glutamine-rich region of TAF4 and multiple sites on the disordered Q1 and Q2 domains of CREB. The transient nature of these interactions is underscored by the short residence time of CREB (~2.5 s) (69) at promoters in the nucleus of living cells. Materials and Methods A complete version of the Materials and Methods can be found in SI Appendix. Protein Expression and Purification. Rattus norvegicus CREB1 (Uniprot ID: P15337) was used to prepare constructs for the full-length CREBTAD (residues 1 to 280), Q1 (residues 1 to 100), KID (85 to 160), and Q2 (residues 160 to 280) domains for expression in Escherichia coli. All CREB constructs and the Q1 domain (residues 428 to 492) of human TAF4 (Uniprot ID: O00268) were expressed as His6-GB1-[TEV]-GSHM fusions. NMR spectra were acquired in NMR buffer (20 mM 2-(N-morpholino)ethanesulfonic acid (MES), pH 6.0, 100 mM NaCl, and 10% D2O), with 0.5 mM tris-(2-carboxyethyl)phosphine (TCEP) or 2 mM dithiothreitol (DTT) for constructs containing cysteine residues with unreacted sulfhydryl groups. Spin Labeling for PRE. PRE experiments were performed with CREBTAD constructs containing a single cysteine: endogenous C90, C90A/T172C, and A50C/C90A coupled to the nitroxide spin label MTSL [(1-oxyl-2,2,5,5-tetramethylpyrroline-3-methyl)methanethiosulfonate]. The spin label was reduced by ascorbate to provide the diamagnetic control. Fluorophore Labeling for sm-FRET Experiments. CREBTAD double-cysteine variants (T2C/C90, C90/T160C, C90A/T172C/T259C and T2C/C90A/T198C) were fully reduced and purified from excess DTT and immediately mixed with donor (Alexa488 C5-maleimide derivative, Molecular Probes) and acceptor (Alexa594 C5-maleimide derivative, Molecular Probes) dyes in 1:1 and 1:5 molar excess to protein, respectively. By design, the most abundant species in the mixture was always acceptor-acceptor labeled CREB-TAD, followed by donor-acceptor labeled pair, while the amounts of singly labeled or donor-donor labeled proteins were negligible. Acceptor-acceptor labeled proteins are “invisible” to sm-FRET measurements; therefore, the labeled CREB-TAD was used without further purification. NMR Spectroscopy. NMR spectra were recorded at 298 K on Bruker Avance 600, 700, 800, and 900 MHz spectrometers. Spectra were processed with NMRPipe (70) and analyzed using Sparky (71) or NMRView (72) within the NMRbox resource (73). Resolution of transformed spectra was greatly enhanced by the application of iterative soft thresholding (IST), which can be used for uniformly sampled data as well as for nonuniformly sampled data (74). SAXS. All SAXS data were collected at the SIBYLS beamline at the Advanced Light Source in Lawrence Berkeley National Lab. sm-FRET Experiments. Intramolecular EFRET distributions were determined in experiments conducted in 20 mM TRIS and 2 mM DTT, pH 7.5 (sm-FRET buffer) at room temperature. All sm-FRET data were collected using a custom-built confocal microscopy–based setup as previously described (75). Full details are given in SI Appendix. Supplementary Material Appendix 01 (PDF) Click here for additional data file.