Coupling of binding and differential subdomain folding of the intrinsically disordered transcription factor CREB

The cyclic AMP response element binding protein (CREB) contains a basic leucine zipper motif (bZIP) that forms a coiled coil structure upon dimerization and specific DNA binding. Although this state is well characterized, key features of CREB bZIP binding and folding are not well understood. We used single‐molecule Förster resonance energy transfer (smFRET) to probe conformations of CREB bZIP subdomains. We found differential folding of the basic region and leucine zipper in response to different binding partners; a strong and previously unreported DNA‐independent dimerization affinity; folding upon binding to nonspecific DNA; and evidence of long‐range interdomain interactions in full‐length CREB that modulate DNA binding. These studies provide new insights into DNA binding and dimerization and have implications for CREB function.

The cyclic AMP response element binding protein (CREB) contains a basic leucine zipper motif (bZIP) that forms a coiled coil structure upon dimerization and specific DNA binding. Although this state is well characterized, key features of CREB bZIP binding and folding are not well understood. We used single-molecule Förster resonance energy transfer (smFRET) to probe conformations of CREB bZIP subdomains. We found differential folding of the basic region and leucine zipper in response to different binding partners; a strong and previously unreported DNA-independent dimerization affinity; folding upon binding to nonspecific DNA; and evidence of long-range interdomain interactions in full-length CREB that modulate DNA binding. These studies provide new insights into DNA binding and dimerization and have implications for CREB function.
Cyclic AMP response element binding protein binds to a DNA consensus sequence known as the cyclic AMP response element (CRE), which occurs in both a "full CRE" palindromic form (5 0 -TGACGTCA-3 0 ) and a "half CRE" form (5 0 -CGTCA-3 0 ) [1,12,22]. Like other members of the bZIP family, CREB binds DNA via a dimeric coiled coil motif [23,24] (Fig. 1C). During full CRE binding, each monomer is arranged with its helical basic region (BR) in the major groove to contact one half of the palindromic sequence. Contacts with the DNA backbone additionally lend the protein nonspecific binding potential [24][25][26]. The leucine zipper (LZ) forms the hydrophobic dimerization interface, without which the bZIP is unable to bind DNA [27,28]. Evidence from circular dichroism experiments suggests that the bZIP motif is disordered as a monomer and folds upon binding to its specific DNA site [23,26].
Unlike many bZIPs, CREB nearly exclusively forms homodimers. This specificity for homodimerization comes from (1) one of the shortest leucine zipper domains of any bZIP, and (2) specific intermolecular contacts between the monomers [24,29]. Previous reports suggested that CREB's dimerization affinity is weak (20 μM or more) or entirely DNA-dependent [26,[30][31][32]. However, the low-resolution or nonequilibrium nature of these experiments, combined with the lack of evidence for monomer-DNA binding, leaves open questions about the coupling of CREB bZIP folding, dimerization, and DNA binding.
In this study, we used single-molecule Förster resonance energy transfer (smFRET) to probe the conformation of the LZ and BR domains in individual CREB molecules during DNA binding and DNA-  (LZ). Basic residues in the BR are shown in blue; hydrophobic residues in the LZ are shown in orange. bZIP residues used for labeling (S270, C310, C337) are shown in red. (B) Protein constructs and label sites. bZIP constructs consisted of residues 270-341, labelled at 270 and 310 (BR) or 310 and 337 (LZ). Full-length (FL) constructs consisted of residues 1-341 with the same label sites, with an additional FL construct labelled at residues 90 and 310. Fluorophore labelling is stochastic, so either dye can be present at each cysteine. (C) Crystal structure of bZIP (residues 285-341) bound to full CRE DNA (PDB 1DH3). BR is shown in blue; LZ is shown in orange. Labelling residues 310 and 337 are highlighted in red; labelling residue 270 was not part of the construct in this crystal structure. A coordinated Mg 2+ ion is not shown for visual simplicity. (D, E) FRET efficiency (E FRET ) histograms of bZIP constructs bound to full CRE DNA. Each histogram is normalized such that total counts = 1 and fit with a Gaussian distribution with the mean shown in black. Fit details are available in Tables S1 and S2. (D) bZIP-BR + 50 nM unlabelled bZIP +500 nM full CRE DNA. The mean E FRET = 0.37 (black), matching the predicted value of 0.37 (red). (E) bZIP-LZ + 50 nM unlabelled bZIP +500 nM full CRE DNA. The mean E FRET = 0.57 (black), compared to the FPS prediction of 0.59 (red). independent dimerization. The single-molecule nature of our experiments allowed us to resolve and quantify distinct states in the heterogeneous conformational landscape of the bZIP. We investigated bZIP folding in the context of different binding partners; its dimerization and binding to full CRE, half CRE, and nonspecific DNA sequences; and bZIP binding and folding as a truncated construct versus in the context of full-length CREB.

Protein fluorescent dye labelling
Exogenous and endogenous cysteines of CREB constructs were labelled with Alexa Fluor 488 and Alexa Fluor 647 maleimide dyes (Thermo Fisher Scientific, Waltham, MA, USA). Purified CREB was labelled at 10-20 μM in labelling buffer (20 mM Tris pH 7.5, 500 mM NaCl, 1 mM TCEP) using a 5 mM dye stock. To avoid preferential labelling by one dye over the other, substoichiometric additions of the dye mixture were made to a purified protein construct over 3 h to a final threefold molar excess of each dye. (For instance, six additions of dye mixture at half the concentration of protein were made every 30 min) Labelled CREB was purified again by ion exchange to remove excess fluorophores. bZIP constructs were labelled before cleavage of the GB1 tag for improved solubility. Purity of the labelled product was assessed via SDS/PAGE and mass spectrometry.

DNA affinity measurements
Intramolecular smFRET measurements of DNA binding required sufficient unlabelled bZIP to ensure that dimerization between two labelled bZIP monomers did not significantly alter FRET histograms. 10 nM unlabelled bZIP was included in DNA binding measurements, representing 100× excess over labelled bZIP and corresponding to a maximum dimer concentration of 5 nM. DNA binding measurements were conducted only at DNA concentrations > 5 nM, at which DNA is in excess over bZIP dimer. At lower concentrations, the excess of protein would saturate high-affinity DNAs.

Single-molecule MFD-PIE FRET data analysis
The FRET efficiency (E FRET ) is inversely related to the sixth power of the ratio of the distance (R) and the Förster radius (R 0 ), a characteristic of the dye pair. For Alexa Fluor 488 and Alexa Fluor 647, the dyes used in this manuscript, R 0 ¼ 55:6Å [36].
Data were analysed using PIE Analysis with Matlab (PAM) software [37] via standard procedures for MFD-PIE smFRET burst analysis [38,39]. Signals from each TCSPC routing channel (corresponding to the individual detectors) were divided in time gates to discern 483-nm excited FRET photons from 635-nm excited acceptor photons. A twocolour MFD all-photon burst search algorithm using a 500μs sliding time window (minimum of 100 photons per burst, minimum of 5 photons per time window) was used to identify single donor-and/or acceptor-labelled molecules in the fluorescence traces. Double-labelled single molecules were selected from the raw burst data using a kernel density estimator (ALEX-2CDE ≤ 15) that also excluded other artefacts [40]. Sparse slow-diffusing aggregates were removed from the data by excluding bursts exhibiting a burst duration > 12 ms. By generating histograms of E versus measurement time, we corroborated that the distribution of E was invariant over the duration of the measurement. Data were corrected in this order to obtain the absolute stoichiometry parameter S and absolute FRET efficiency E: background subtraction, donor emission crosstalk correction, acceptor direct excitation correction and relative detection efficiency correction. To obtain the relative detection efficiency correction factor (γ), the centre values of the E-S data cloud for each protein were estimated manually, plotted in an E versus 1/S graph, and a straight line was fitted to the resulting data: where Ω is the intercept and Σ the slope of the linear fit. For bZIP-BR, bZIP-LZ, and FL-LZ, γ ¼ 0:7757. For FL-BR and FL 90-310, γ ¼ 0:53.
Static PDA was carried out to fit data to Gaussian distributions and quantify their relative areas [41]. For each FRET dataset, raw bursts were re-binned in 1 ms time bins, and histograms were constructed and analysed. Binned data were plotted in a raw (uncorrected) FRET efficiency (E PR ) versus uncorrected stoichiometry (S PR ) plot and only bins with 0.2 < S PR < 0.6 were included in the analysis to remove burst sections containing complex acceptor photophysics or photobleaching. Furthermore, only bins with at least 20 and maximally 200 photons (to reduce calculation time) were used for PDA analysis. A two-state model for a Gaussian distance distribution was used to generate a library of simulated E FRET values, which was fitted to the experimental E FRET histogram using a reduced χ 2 -guided simplex search algorithm to obtain the amplitude, mean distance R and width σ of all Gaussian distributed substates and, in the case of multiple states, their area fraction A (%). Across all samples, σ was required to be a constant fraction of R, which reduces the number of parameters to fit and is a reasonable assumption for samples without significant conformational heterogeneity [42,43]. A Jacobian was used to estimate confidence intervals (95%) of the fit parameters. Criteria for a good fit were a low reduced χ 2 value, as well as a weightedresidual plot free of trends.

Mass photometry
Mass photometry experiments were performed on a Refeyn TwoMP (Refeyn Ltd, Oxford, UK) calibrated with a mix of BSA (Sigma-Aldrich) and thyroglobulin (Sigma-Aldrich, St. Louis, MO). Coverslips (WillCo Wells, Amsterdam, NL) and gaskets (Grace Bio Labs, Bend, OR, USA) were prepared by washing with ddH 2 O followed by isopropanol and again with ddH 2 O, repeated three times, and then dried. 20 μL of buffer was added to each well to focus the instrument, then 15 μL of buffer was removed and replaced with 15 μL of sample and mixed by pipette for 3 s before frame acquisition. Frames were acquired over 120 s using AcquireMP (version 2022 R1, Refeyn Ltd, Oxford, UK) using standard settings. Data were processed and analysed by fitting a Gaussian distribution to the data using Dis-coverMP (version 2022 R1, Refeyn Ltd, Oxford, UK). We compared mass photometry measurements for each CREB construct after incubating the sample for 20, 30, and 45 min to assure that incubation times were long enough to reach equilibrium. All samples used to fit the dissociation constants were incubated for 1 h before the measurement.
Data fitting GRAPHPAD PRISM 9 (GraphPad Software, La Jolla, CA, USA) was used to plot graphs.
FRET histograms were fit either using the Photon Distribution Analysis (PDA) functionality of PAM or with Gaussian distributions using GRAPHPAD PRISM 9.
Dimerization affinity measurements were fit using GRAPH-PAD PRISM 9 with the equation: where Y is the fraction of protein in the folded/bound form, Y min is the minimum Y value, Y max is the maximum Y value (set equal to 1), M T is the total concentration of protein monomers, and K d is the dissociation constant of dimerization. The concentration of labelled protein was assumed to be negligible. Derivation of this equation can be found in the Supporting Equations.
To obtain apparent affinities, DNA-binding measurements were fit using GRAPHPAD PRISM 9 to the standard equation for a one-site binding model where ligand concentration cannot be neglected: where Y is the fraction of protein in the folded/bound form, Y min is the minimum Y value, Y max is the maximum Y value, K d is the dissociation constant between protein and DNA, P is the protein concentration expressed in terms of dimers (the DNA-binding form), and D is the DNA concentration. Derivation of equations and details of each fit can be found in the Supporting Information.

Results and Discussion
To generate FRET reporters, we created protein constructs with two cysteines flanking the region to be monitored. Across all our constructs, we introduced one cysteine residue (C270) and made use of three native cysteine residues (C90, C310, C337; Table 1).
For DNA-binding studies, we designed 14 base pair DNAs with full CRE, half CRE, and nonspecific sequences (Fig. 3I). As CREB has a DNA-binding footprint of at least eight base pairs [24], 14 base pair DNAs are short enough to exclude binding of multiple CREB molecules for simplicity of analysis, and have been used in previous binding studies [33].
smFRET observations of the DNA-bound CREB bZIP agree with the structure observed by X-ray crystallography We first tested whether our smFRET measurements matched the known conformation of the CREB bZIP bound to full CRE DNA. In the published crystal structure, the full CRE DNA-bound bZIP adopts a canonical α-helical coiled coil conformation (PDB 1DH3, Fig. 1C) [24]. To assess whether bZIP adopts the same conformation in solution, we used FRET Positioning and Screening (FPS) software to predict E FRET values for comparison to our measurements. FPS computes a mean E FRET from a structure by calculating the dye-accessible volume that results from steric clashes and the flexibility of the dyes and their linkers (Fig. S2) [48].
For bZIP-LZ in the presence of full CRE DNA, FPS predicts a mean value of E FRET = 0.59 (Table 2), which agrees well with our measurement of E FRET = 0.57 ( Fig. 1E; measured value in black, prediction in red).
In the crystal structure of DNA-bound bZIP, the protein is truncated at residue 285, preventing us from computing an expected distance for the BR FRET reporter. Instead, we generated an AlphaFold model for the bZIP dimer (residues 270-341), which predicts an α-helical conformation in BR extending to T276, matching the crystal structure reasonably well (Fig. S3). We used the AlphaFold model to obtain an expected E FRET for the BR in the putative helical conformation, resulting in a prediction of E FRET = 0.37. For bZIP-BR in the presence of 14 bp full CRE, we measured a mean E FRET = 0.37 ( Fig. 1D; measured value in black, prediction in red), suggesting that our extended 270-341 construct is fully helical while binding full CRE DNA.
The CREB bZIP dimerizes with nanomolar affinity in the absence of DNA, inducing folding of the LZ but not the BR In its monomeric state, the many hydrophobic residues of the LZ make it likely to adopt a compact disordered conformation. In contrast, an α-helix is a linear, relatively extended structure. Therefore, we expect fluorophores at opposite ends of a potential α-helix to exhibit a comparatively low E FRET if the helix is folded and a higher E FRET if the region is in a compact disordered conformation.
Both bZIP-BR and bZIP-LZ display relatively high E FRET (0.61 and 0.80, respectively) at the~100 pM concentration used for single molecule experiments ( Fig. 2A,C). These E FRET values are consistent with free monomeric bZIP adopting a compact conformational ensemble in both the BR and LZ regions, relative to the helical conformation.  We fit the areas of each state using PDA and calculated the overall dimer fraction, normalized and plotted here, taking the high-FRET state as disordered monomer and the medium-FRET state as the helical dimer. Each point represents two histograms, and error bars represent the 95% confidence intervals of the PDA fits. The dimerization curve was fit to Eqn (3), yielding K d = 33 nM (95% CI = 20-55 nM). The maximum dimer fraction (Y max ) was fixed = 1, while the minimum dimer fraction (Y min ) was fit, and the data were normalized such that fit Y min = 0, which did not change the fit K d (Fig. S4). The open point plotted at x = 0.11 represents x = 0. Fit details can be found in Fig. S4  To assess whether bZIP dimerization occurs in the absence of DNA, we added 1 μM of unlabelled bZIP to both FRET reporters. bZIP-LZ transitioned to a low-FRET state with E FRET = 0.47, consistent with folding of the LZ α-helix, which is required to form the coiled coil dimerization interface (Fig. 2D). The population of this low-FRET state was dependent on the concentration of unlabelled bZIP added, suggesting a binding interaction (Fig. 2E). This interpretation is supported by mass photometry results (see below). Meanwhile, bZIP-BR showed a slightly increased E FRET = 0.71 in the presence of excess unlabelled bZIP, suggesting greater compaction, likely resulting from the folding of the adjacent LZ (Fig. 2B). The simplest interpretation of these results is that DNAindependent dimerization is associated with folding of the LZ but not of the BR.
We note that bZIP-LZ bound to full CRE DNA exhibits a higher mean E FRET (0.57, Fig. 1E) than bZIP-LZ dimerized in its absence (0.47; Fig. 2D). Although the LZ is extended in both conditions, its conformation is influenced by the presence of DNA, presumably via the conformation of the BR. We suggest that, during DNA-independent dimerization, the LZ forms an α-helical structure with some fraying at its N-terminal end, adjacent to the disordered BR. This fraying would account for the greater extension of bZIP-LZ in the absence of DNA.
LZ helix folding is necessary to form the amphipathic dimerization interface in which hydrophobic residues are buried, balancing the entropic cost associated with folding. In a similar manner, BR folding forms the binding interface for DNA; however, without the DNA providing a counterion surface, formation of the highly charged BR helix is likely unfavourable. Our results are consistent with a model in which BR forms an α-helix only in the presence of DNA, remaining disordered in isolation or during DNA-independent dimerization. We next used the conformational change of bZIP-LZ to measure the dimerization affinity. At low concentrations of unlabelled bZIP, E FRET ≈ 0.8, whereas an increasing concentration induced a gradual transition to a population at E FRET ≈ 0.45, consistent with formation of an LZ α-helix (Fig. 2E). We used photon distribution analysis (PDA), which accounts for photon shot noise, to fit these E FRET histograms with two primary states to extract the area of each (Fig. S4) [41,49]. We used the relative populations of each state to calculate the overall dimer fraction for each dataset and construct a titration curve (Fig. 2F).
Our fit yielded a dimerization K d = 33 nM (95% CI 20-55 nM). We found that CREB has a strong tendency to adsorb to tubes and pipet tips, leading to effective concentrations that are lower than those we calculated. This suggests the true K d is likely lower than the value we obtained, which should be regarded as an upper bound. Despite these caveats, our dimerization assay provided evidence for a significantly stronger dimerization affinity than has previously been reported [26,[30][31][32].
Large concentrations of unlabelled FL CREB introduced high background signal to our measurements, so we were unable to perform an smFRET dimerization titration for FL CREB.
To corroborate evidence that the LZ conformational change reflects dimerization and to extend our affinity measurements to FL CREB, we performed mass photometry (MP) dimerization assays [50,51].
Mass photometry has a lower detection limit of about 30 kDa, significantly above the molecular weight of CREB bZIP (8.8 kDa). To make the bZIP amenable to mass photometry, we prepared a WT bZIP construct with an N-terminal MBP tag. FL CREB has a mass of 36 kDa, and we used both WT and GB1tagged forms in mass photometry. For each construct, we measured the numbers of monomer-and dimersized molecules in solutions between 5 and 50 nM and plotted the data (Fig. 2G,H; Fig. S5). We found no significant difference between the datasets (F test, P = 0.3887), allowing us to fit all three constructs with a single curve, yielding a dimerization K d = 15 nM (95% CI = 12-17 nM).
The dimerization affinity measured by MP is about half the value measured for bZIP dimerization by smFRET. This difference is most likely due to the adsorption error discussed above, although it is possible that the fluorescent labels or point mutations impacted dimerization. The lack of significant difference between the three MP constructs, two with solubility tags and one without, indicates that the solubility tags did not impact dimerization. The MP assay did not reveal a significant difference in dimerization affinity between bZIP and FL CREB. Finally, the dimerization affinities obtained through both approaches support our interpretation of bZIP as being fully monomeric in the absence of DNA at thẽ 100 pM concentration used for smFRET. However, we expect CREB to be at least partially dimerized at the higher concentrations used for most other techniques. Our smFRET and MP results show that increasing CREB concentrations drive dimerization and LZ folding in the absence of DNA.  Error bars reflect the 95% confidence intervals of the PDA fits (Fig. S7). The curves were fit to Eqn (4) with a protein concentration fixed at 5 nM (the concentration of potential bZIP dimers). The black open circle at x = 1.1 represents x = 0. The minimum dimer fraction (Y min ) was fit globally for all three titrations, while the maximum dimer fraction (Y max ) was fit separately for each. Curves were normalized to their fit Y min and Y max values for visual clarity, which did not change the K d,app s (Fig. S8). Nonspecific DNA (grey, x in circles) was fit with K d,app = 146 nM (95% CI = 93-228 nM). Half CRE DNA (pink, half circles) was fit with K d,app = 1.4 nM (95% CI = 0.33-3.2 nM). The K d,app for full CRE DNA (purple, solid circles) could not be reliably fit, but the upper 95% CI was < 1 nM. Fit details can be found in Figs S7 and S8 and Tables S8-S11. The CREB bZIP binds specific and nonspecific DNA sequences with extended, likely helical conformations Next, we compared bZIP conformations during binding to various DNA sequences with identical GC content. In each DNA-binding condition, we included 50 nM of unlabelled bZIP to ensure that labelled molecules did not dimerize with each other, which would introduce unintended intermolecular FRET.
Binding of bZIP-BR to full CRE and half CRE DNA yielded similar E FRET distributions, centred at 0.37 and 0.39 respectively (Fig. 3A,B). For bZIP-BR bound to nonspecific DNA, the population was centred at E FRET = 0.43, indicating a slightly less extended but still partially folded conformation (Fig. 3C). We also tested bZIP-BR binding to several nonspecific DNA sequences with varying GC content and saw no significant differences in the FRET histograms (Fig. S6). The binding conformation for bZIP bound to nonspecific DNA is more similar to the conformation of helical, full CRE-bound bZIP than to that of disordered unbound bZIP, in contrast to early reports using circular dichroism that suggested little or no bZIP folding in response to nonspecific DNA [26,32].
The bZIP-LZ conformation also varies slightly depending on the DNA sequence used. bZIP-LZ showed a trend of successively lower E FRET for full CRE binding, half CRE binding, nonspecific DNA binding, and DNA-independent dimerization (Fig. 3E-H). This decrease in the mean E FRET in the LZ is correlated with increasing E FRET in the BR, consistent with the idea that incomplete folding in the BR is related to fraying of the LZ helix.
CREB binds specific DNA sequences with at least 100-fold stronger affinity than nonspecific sequences We conducted DNA affinity assays with similar design to our smFRET dimerization assay. In the presence of 10 nM unlabelled bZIP, we added increasing quantities of DNA and quantified bZIP-LZ folding as a measure of DNA binding (Fig. 3J, Figs S7 and S8).
We initially fit these data with a standard one-site binding model to obtain apparent affinities. We measured a full binding curve for the interaction of bZIP with nonspecific DNA (Fig. 3J), fit to K d,app = 146 nM (95% CI = 93-228 nM). In the presence of half and full CRE DNA sequences, bZIP-LZ was mostly folded even at the lowest observed DNA concentrations. For the half CRE, we fit K d,app = 1.4 nM (95% CI = 0.33-3.2 nM), although given the poor sampling of the curve we do not place high confidence in the precision of this value. Binding to full CRE was too tight to fit by this assay (see Methods), but we were able to extract an upper 95% confidence value K d,app < 1 nM.
The agreement of our specific DNA binding measurements with published values (0.3-12 nM for full CRE and 2-69 nM for half CRE) supports the use of bZIP folding as a proxy for DNA binding [7,22,25,31,[33][34][35]52,53]. CREB binding to nonspecific DNA has been less thoroughly examined, and affinity may vary based on how similar the chosen sequence is  (Fig. S9). The curves were fit to Eqn (4) with a protein concentration fixed at 5 nM (the concentration of potential bZIP dimers). The black open circle at x = 1.2 represents x = 0. The minimum and maximum dimer fractions (Y min and Y max ) were fit globally for all three titrations. Curves were normalized to their fit Y min and Y max values for visual clarity, which did not change the K d,app s (Fig. S10). Nonspecific DNA (grey, x in circles) was fit with K d,app = 1.2 μM (95% CI = 0.6-2.2 μM). Half CRE DNA (pink, half circles) was fit with K d,app = 9.0 nM (95% CI = 3.6-20 nM). Full CRE DNA (purple, solid circles) was fit with K d,app = 1.7 nM (95% CI = 0.33-4.7 nM). Fit details can be found in Figs S9 and S10 and Tables S15-S18. (Q) Comparisons of fit K d,app s for bZIP and FL CREB binding to full CRE, half CRE, and nonspecific DNAs. to the CRE [25,31]. Our results indicate a K d,app in the hundreds of nanomolar for nonspecific DNA. The application of a simple binding model to this system fails to capture the linkage between the dimerization and DNA-binding steps, resulting in a K d,app that depends on the experimental design [54]. However, comparisons between K d,app values measured in identical conditions, as in this paper, are still quantitatively valid. We additionally analysed these data to determine stepwise K d s for dimerization and DNA binding (Supporting Equations; Fig. S8).

Conformations of the CREB bZIP are similar in full-length and truncated constructs
To test whether the Q1-KID-Q2 region of CREB influences bZIP conformation, dimerization, and interactions with DNA, we prepared full-length CREB constructs with identical label sites as described in Table 2.
Overall, the FL constructs produced similar results to the bZIP constructs. Both FL-BR and FL-LZ show high FRET in isolation (Fig. 4E,J), indicating collapsed, disordered conformational ensembles. When exposed to an excess of unlabelled bZIP, FL-LZ transitions to a lower-FRET state at E FRET = 0.47, suggesting folding of the LZ α-helix upon dimerization, whereas FL-BR does not undergo a similar folding event, showing that the BR does not fold upon formation of the LZ coiled coil (Fig. 4D,I). In the presence of DNA, FL-BR adopted similar conformations for full and half CRE sequences (E FRET = 0.37 for both) and a less extended conformation for nonspecific DNA (E FRET = 0.53; Fig. 4A-C). FL-LZ also showed higher E FRET upon DNA binding than upon DNA-independent dimerization with unlabelled bZIP, suggestive of helix fraying in the latter condition ( Fig. 4F-I).
We also noted some differences between FL and bZIP constructs. Whereas addition of 1 μM unlabelled bZIP led to a compaction in the bZIP-BR construct (E FRET increase from 0.61 to 0.71), the FL-BR was already compacted in absence of unlabelled bZIP (E FRET increase from 0.75 to 0.77).
In addition, while E FRET values were similar for FL-BR and bZIP-BR upon binding to full and half CRE DNAs, nonspecific DNA binding resulted in a higher E FRET for FL-BR than for bZIP-BR (0.53 versus 0.43; Fig. 4C). This suggests that the presence of Q1-KID-Q2 affects BR conformation differentially depending on the DNA sequence bound.
Previous work established an interaction between bZIP and KID that influenced DNA binding activity [33]. Therefore, we tested for such interactions between the Q1-KID-Q2 region of CREB and the bZIP using a full-length construct fluorescently labelled at native cysteines 90 and 310 (Table 1). These residues are sequentially distant from one another in a region that is predicted to be entirely disordered, so we expected significant FRET only if intramolecular interactions bring the labelled residues into proximity. We measured a low E FRET = 0.17 for this construct at 100 pM, far below the dimerization K d , indicating that the residues are over 70Å apart (Fig. 4K; Eqn 1). Upon addition of 50 nM FL CREB, the mean E FRET transitioned to 0.39 (Fig. 4L). Although this still reflects a substantial distance between the residuesnearly 60Åit indicates a conformational change related to dimerization that manifests outside of the leucine zipper domain (which is C-terminal to C310 and not monitored by the FL 90-310 FRET reporter). Strikingly, addition of DNA disrupts this higher-FRET conformation, driving transition to a low mean E FRET (between 0.10 and 0.16) regardless of the sequence of DNA (Fig. 4M-O). That DNA can relieve this interaction indicates crosstalk between bZIP and other regions of FL CREB.
These conformational changes are likely related to weak contacts between the bZIP and an acidic patch at the C-terminal end of the KID [33]. We hypothesized that the KID acidic patch competes with DNA for contacts in the basic region. A similar long-range interaction has been observed in p53, where weak intramolecular interactions disproportionately discourage binding to nonspecific DNA [55]. If an analogous mechanism exists in CREB, it might explain the reduced structure in the FL-BR upon binding to nonspecific DNA.
To test our hypothesis, we measured the apparent affinity of FL CREB for various DNA sequences in the same manner as we did for bZIP in Fig. 3J. Although FL-BR showed a significant conformational difference from bZIP-BR only upon binding to nonspecific DNA, we found that FL CREB had a weaker affinity for all three DNA sequences: the K d,app for full CRE was 1.7 nM (95% CI 0.33-4.7 nM); for half CRE, 9.0 nM (95% CI 3.6-20 nM); and for nonspecific DNA, 1.2 μM (95% CI 0.6-2.2 μM) (Fig. 4P,Q; Figs 9 and 10). The change in affinity was significant in each case, although we note that the full CRE affinity is at the lower limit of our measurement capabilities, and therefore less precise (F tests; full CRE, P = 0.0269; half CRE, P = 0.0080; nonspecific DNA, P = 0.0001). The stepwise affinity of nonspecific DNA binding was also significantly weaker than that measured with the bZIP (Fig. S10). Thus, although the basic region is the only domain to directly contact DNA, other domains of CREB influence DNA binding, likely through intramolecular competition for basic region contacts.

Conclusion
In this study, we showed that the CREB BR and LZ can fold separately in response to their respective binding partners, despite their canonical conformation as a continuous α-helix. Both subdomains are disordered in the monomer and fold upon binding, a common property of transcription factors that has several potential regulatory benefits [56][57][58][59].
Using two orthogonal methods, we measured DNAindependent dimerization affinities of 33 nM (bZIP dimerization by smFRET) and 15 nM (bZIP and FL dimerization by MP), both significantly stronger than has previously been reported. At nuclear concentrations between 100 and 400 nM, with even higher concentrations within CREB puncta, we would anticipate the majority of free CREB to exist as dimers within the cell [31,60,61].
The CREB bZIP undergoes a clear conformational change upon binding any sequence of DNA, in contrast to previous reports that no folding occurs in response to nonspecific DNA [26,32]. While the conformation we observed upon full and half CRE binding agrees well with the coiled coil structure observed via crystallography, binding to nonspecific DNA results in only partial BR extension. BR extension is further reduced in the full-length context, demonstrating that the presence of Q1-KID-Q2 affects the bZIP conformational landscape. bZIP binding to nonspecific DNA is strong (K d, app = 146 nM), but at least 100-fold weaker than binding to half or full CRE DNA (K d,app = 1.4 nM, and < 1 nM, respectively). The interaction between the CREB bZIP and nonspecific DNA has rarely been reported but is likely common in the context of the cell where non-cognate DNA is abundant. In vivo binding to both specific and nonspecific DNA is supported by single-molecule imaging of fluorescent CREB in cells, where two distinct residence times on DNA were observed and interpreted as specific (long residence time) versus nonspecific (short residence time) binding [61]. Binding to nonspecific DNA likely plays a key role in facilitating target site search, analogously to other eukaryotic transcription factors [62][63][64][65]. We found differences in basic region conformation between specific and nonspecific DNA binding, lending credence to this model. Finally, we compared the conformation of the truncated bZIP to its full-length counterpart. We found evidence for a KID-bZIP interaction that has the potential to modulate bZIP conformation in response to other cellular conditions like post-translational modification and cofactor binding. We found that FL CREB has reduced affinity for all DNA sequences, likely because this KID-bZIP interaction competes with DNA binding (K d,app = 1.2 μM, 9.0 nM, and 1.7 nM for nonspecific, half CRE, and full CRE DNA). Given previous examples of protein regions that do not contact DNA yet nonetheless affect DNA binding, we plan to continue probing these constructs for conformational variations [59,66,67].