Conservation of coactivator engagement mechanism enables small-molecule allosteric modulators

ATF6α and ERM Bind to Opposite Faces of Med25 AcID.

Separate NMR studies of AcID in complex with the transcriptional activation domains of VP16 and ERM suggest that the two activators both contact the H1 binding surface, with the significantly larger VP16 also interacting with the H2 surface (2⇓–4). While several lines of evidence indicate that ATF6α interacts with Med25 AcID as part of its function (5), the binding site within the protein has not been established. We first measured the dissociation constants for each of the activators by fluorescence anisotropy experiments using fluorescein-tagged variants of VP16 (438–490), ERM (38–68), and ATF6α (40–66), and this revealed that ERM and ATF6α interact with comparable affinities (Fig. 1B) (2⇓⇓–5). To provide a direct comparison of the binding modes of the three activators and identify the binding site of ATF6α, we measured the chemical shift changes in each activator–AcID complex via ¹H, ¹⁵N-HSQC NMR titration experiments, with VP16 (438–490), ERM (38–68), and ATF6α (40–66) in the presence of ¹⁵N-labeled Med25 AcID (see SI Appendix for details of the design, synthesis, and characterization of each activator peptide).

The amide proton perturbation patterns measured for the activator–AcID complexes suggest a different binding mode for each of the three activators (Fig. 2). VP16 induced changes at both AcID binding surfaces, consistent with the tandem transcriptional activation domains within its sequence (SI Appendix, Figs. S13–S16). ERM binding predominantly lead to perturbations at residues on the H1 surface of AcID, in agreement with the model in which it preferentially interacts at that site (Fig. 2A and SI Appendix, Figs. S17 and S18) (4, 6, 7). Key changes at residues K411, R538, and Q451, for example, were seen with both VP16 and ERM (Fig. 2C). In contrast, interaction with ATF6α lead to significant chemical shift changes on the H2 binding surface (Fig. 2B and SI Appendix, Figs. S19 and S20). ATF6α induced shifts of residues Q456, M470, and H474, which were also affected to varying degrees by VP16 and largely unaltered by ERM. Consistent with ATF6α and ERM interacting on opposing sides of AcID, mutations introduced on one or the other of the binding surfaces produced distinct effects (Fig. 2C). H1 mutations R538E, K411E, and Q451E inhibit ERM binding while ATF6α is largely unaffected. In contrast, H2 mutations R466D and M523E significantly inhibit ATF6α with minimal impact on ERM binding. Taken together, these data indicate that ATF6α binds on the H2 binding surface of Med25 AcID, opposite the site of ERM. Further, the distinct but overlapping chemical shift patterns observed upon binding of each of the activators to Med25 suggest several unique binding modes accommodated within AcID. This is analogous to helical activator binding domains such as GACKIX of CBP/p300, a three-helix bundle that contains at least two activator binding sites (8).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

ATF6α binds to the H2 surface of Med25 AcID. (A) Results of chemical shift perturbation experiments superimposed upon the Med25 AcID structure (PDB ID code 2XNF). Residues displaying chemical shift perturbation greater than 2 SD upon ATF6α binding are depicted in rust spheres. (B) Scatter plot illustrating correlations between the chemical shift perturbations (CSPs) of individual Med25 AcID residues from HSQC experiments with ERM, ATF6α, and VP16. The position of each maize square represents the CSP of an individual residue in Med25 AcID upon binding to ERM (y axis) and VP16 (x axis). Thus, squares along the dotted diagonal are residues that shift similarly in both ERM–AcID and VP16–AcID complexes. The same analysis for ATF6α is shown in rust circles. Specifically labeled are the positions of three residues that are on the H1 face of AcID (T542) and H2 face of AcID (R466, Q456), highlighting the distinct pattern of correlated CSPs for ERM and ATF6α, consistent with the model in which the two activators do not interact with the same binding site. Full CSP datasets for each of the three activator–Med25 AcID complexes are shown in SI Appendix. (C) Results of direct binding experiments with fluorescein-labeled activators and the indicated mutants of Med25 AcID as measured by fluorescence polarization expressed the fold change relative to the dissociation constant of each activator for the WT AcID. The indicated error is propagated from three independent dissociation constant measurements.

Activator–Med25 Complexes Are Conformationally Dynamic.

Next, the underlying mechanistic features of activator–AcID complex formation were examined by determining association mechanisms of AcID with the TADs of VP16, ERM, and ATF6α using stopped-flow fluorescence spectroscopy. These kinetic experiments allow calculation of microscopic rate constants for association and dissociation, as well as forward and reverse rate constants for any conformational changes that happen during the binding process (9, 10). In practice, however, conformational changes involved in activator–coactivator interactions are often difficult to detect and quantify, and we thus chose the environmentally sensitive fluorophore 4-N,N-dimethylamino-1,8-napththalimide (4-DMN) as a fluorescence probe (Fig. 3A) (11, 12). This fluorophore was synthesized as a conjugate with β-alanine and incorporated at the amino terminus of these activators for subsequent experiments (synthetic protocols and characterization in SI Appendix).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Transient kinetic experiments define minimal mechanism of activator–AcID complexation. (A) Structure of the fluorophore used in these experiments, 4-N,N-dimethylamino-1,8-napththalimide (4-DMN). (B) Representative kinetic trace of association experiment with DMN-VP16 and AcID. The red line is the fit to a two-step binding model (see SI Appendix, Figs. S2 and S3 for additional data). (C) Dependence of the two observed rate constants for the fast (white circles) and slow (gray circles) kinetic phases on the concentration of AcID for association experiments of VP16 with AcID. (D) Sample kinetic trace of a dissociation experiment in which unlabeled VP16 was added in excess to preformed DMN-VP16–AcID. (E) General kinetic mechanism for TAD–AcID complex formation as determined by these experiments for all activators. Microscopic equilibrium constants (K_C,n) are defined as the ratio of the respective forward and reverse rate constants. (F) Representation of equilibrium population distributions of bound states, calculated from equilibrium constants in 3G. Transparency of each state is scaled according to the indicated percentage population. When one equilibrium constant is too small to measure, the values are given as ranges. (G) Measured kinetic and equilibrium constants for all of the activators. Kinetic constants k_F,2 and k_R,2 are unable to be reliably calculated. a, The conformational change equilibrium constant is too small to be measured with precision. b, The overall equilibrium constant from Med25–ATF6a to Med25–ATF6a** is estimated to be ≤0.1 based on the limits of precision of our experiments, thus given that K_C,1 ≤ 0.1, K_C,2 must be ≤1.

Previous kinetic studies of helical coactivators revealed that complex formation with activators proceeds by fast association (k_on) and dissociation (k_off) rate constants (10, 13⇓⇓⇓–17). Consistent with this observation, we found that activator–AcID complexes form with elevated k_on and k_off values, with k_on ranging between 300 and 1,100 µM⁻¹·s⁻¹ and k_off ranging between 100 and 400 s⁻¹ (Fig. 3). This behavior allows for activators to form tight interactions (K_D values 50–500 nM) with Med25 AcID that are short-lived, with activator residence times less than 10 ms. The k_on values are 1–2 orders of magnitude faster than most other activator-coactivator systems, likely a result of significant electrostatic contributions to binding (SI Appendix, Fig. S10), which can elevate association rate constants by several orders of magnitude (14, 15, 18).

At least one conformational change during the binding process was observed in all cases, with similar observed rate constants (k_obs = 10–40 s⁻¹) for each activator. The conformational change was determined to occur after the initial binding event by a combination of global fitting in Kintek Explorer and an “inverted” association experiment (see SI Appendix for additional details, including SI Appendix, Figs. S3 and S5F) (19, 20). Importantly, the equilibrium constant of this conformational change (K_C,1) varied significantly across activators (Fig. 3F). For the activator VP16, which binds both sites simultaneously, the initial bound state was somewhat more favorable (K_C,1 = 0.35 ± 0.13). In contrast, the activators ERM and ATF6α, which bind to opposite sites of AcID, had K_C,1 values that were significantly larger and smaller than VP16 (438–490), respectively (K_C,1 = 2.6 ± 0.2 and <0.1). This range of K_C,1 values is not unusual for activator-coactivator association mechanisms; we have previously observed this behavior with activator–Med15 complexes, where K_C,1 values positively correlated with transcriptional output (10, 21, 22). A second conformational change for ERM and ATF6α that was slower (k_obs = 2–6 s⁻¹) was also detected in these experiments and had equilibrium constants that were too small to quantify reliably. Global fitting of the kinetic traces suggested this conformational change happens after the primary conformational change.

Some structural insight regarding the primary conformational change of the ERM–AcID complex was gained from the ¹H,¹⁵N HSQC titration of ERM with AcID, as many AcID residues display slow and intermediate exchange behavior in this complex (SI Appendix, Fig. S25). Specifically, the intermediate exchange behavior points to conformational exchange on the millisecond timescale (∆ν = 30–180 s⁻¹), consistent with the value of the exchange rate constant measured in the kinetics experiments (45 s⁻¹). Thus, inspecting the residues that display signal broadening shows that this conformational change involves residues at both sites (SI Appendix, Fig. S25), including many residues on the flexible loops and all three helices. Taken together, the NMR and kinetic data point to a mode of interaction where Med25 AcID forms conformationally unique complexes with its activator interaction partners. While these activators bind to Med25 with similar affinities, they each have different kinetic signatures characterized by a conformational change after binding that extends throughout the AcID structure. Finally, all of these complexes can be described as “fuzzy”; at equilibrium, each complex contains multiple populated conformational substates that are separated by low energy barriers (23).

The H1 and H2 Binding Sites Are in Allosteric Communication.

The presence of two binding sites that engage with distinct activators raises the question whether AcID contains an allosteric network linking the two sites to permit cooperative formation of specific activator–AcID ternary complexes. The prototypical case of allostery in ABD–activator complex formation is the GACKIX motif in CBP/p300 (10, 13, 15, 24, 25). In this example, the “signature” of allosteric communication is reduction of the k_off of an activator when another activator is bound at the allosteric site, with the k_on largely unaffected. Thus, for AcID, we measured dissociation rate constants as a primary method to detect allosteric communication.

We first looked to use the VP16 TAD as a model system; the two halves represent individual interaction motifs that bind separate sites, and when employed in trans, it would be expected that they could be used to dissect communication between AcID sites. However, when separated, the two VP16 halves lose 10- to 15-fold affinity and display poor selectivity for one binding site over another (SI Appendix, Fig. S12), which is not surprising given the topological similarity between the two binding sites of AcID and the high sequence homology between the two VP16 activator motifs. This reduced selectivity significantly complicates data interpretation as the presence of multiple distinct ternary complexes can mask allosteric effects. To address this, we took advantage of a distinguishing feature between the AcID binding sites, the presence of two solvent-exposed cysteine residues (C497 and C506) within the H1 site, which were then employed to tether the relevant VP16 fragment to that site via a disulfide. Tethering experiments with a library of disulfide-containing point mutations of the H1-targeting portion of VP16, VP16 (438–454), were carried out with AcID, and the G450C mutation led to 100% formation of a disulfide bond with the C506 residue of AcID, even under stringent conditions. Biophysical characterization of the covalent complex (SI Appendix, Figs. S9, S11, and S21) indicates that it recapitulates the features of the noncovalent complex (26). Consistent with the hypothesis of allosteric communication, multiple chemical shifts corresponding to residues in the H2 site are perturbed in the ¹H,¹⁵N HSQC spectrum (SI Appendix, Figs. S21 and S22). In contrast, the introduction of cysteine to the H2 surface at a variety of positions produced mutants with a high aggregation propensity, rendering them unsuitable for Tethering and binding studies.

With this tool in hand, two separate 4-DMN–labeled probes that interact with the H2 binding site, VP16 (467–488) and ATF6α, were used to assess changes in ternary complex formation upon occupancy of the H1 site of AcID (Fig. 4A). As a complement to the covalent system, ERM was also employed as a noncovalent H1 partner in a separate set of experiments due to its greater apparent selectivity for the H1 binding surface. With VP16 (438–454)_G450C tethered at the H1 site, the k_off value of the VP16 (467–488) ligand was reduced by 20% (Fig. 4B). The corresponding value for the ATF6α probe was reduced by ∼10%, although this did not reach statistical significance (P = 0.1). Consistent with our hypothesized model, k_on for ATF6α was unchanged; we were unable to measure VP16 (467–488) k_on due to a very high k_off value. When ERM was bound to the H1 site, k_off for VP16 (467–488) displayed a similar 20% reduction, while the ATF6α k_off value was reduced by 25%.* Taken together, the data demonstrates that the two binding sites of Med25 AcID are allosterically linked, and the mechanism of allosteric communication (reduction of k_off) is analogous to that of the GACKIX motif (13, 15). These measured cooperativity values (∼1.3) are similar to values previously measured for GACKIX, which are in the range of 1.4–2.2 for most ternary complexes, but in certain cases, are as high as 18. We expect as more Med25 binding partners are reported and characterized these cooperativity factors will also vary significantly for different ternary complexes (27).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Dissociation experiments reveal allosteric communication between two binding sites. (A) Schematic of the experiment. (B) Comparison of k_off for VP16 (467–488) (blue bars) for Med25 AcID, Med25 AcID with VP16 (438–454)_G450C covalently Tethered, and Med25 AcID with ERM prebound; the red bars summarize data from analogous experiments with ATF6α. The values shown are the average of 2–3 independent experiments with the indicated errors (SD). All changes from the binary complex were statistically significant (P < 0.01), except for the Tethered complex bound to ATF6α (transparent bar).

A Covalent Cochaperone Recapitulates Allosteric Changes.

We have previously demonstrated that prototypical conformationally dynamic coactivators such as GACKIX can be allosterically modulated by covalent cochaperones (13, 28); these can be rapidly identified with the covalent fragment discovery method of Tethering. In the GACKIX case, engagement with the most dynamic sites within the coactivator lead to the most effective cochaperones. To identify such regions within Med25 AcID suitable for chemical cochaperone discovery, all-atom molecular dynamics simulations were carried out using implicit solvent models (GBSW) and with temperature replica exchange in CHARMM (29⇓⇓–32). Simulations were performed on both unbound Med25 AcID and a model of this protein in which the VP16 (438–454)_G450C is tethered at C506. To identify the substructures most stabilized upon binding, the root-mean-square fluctuations (RMSF) for each residue were calculated from the resulting trajectories (Fig. 5 A and B). In these figures, the line color reflects the range of motion of each residue. In the unbound structure (Fig. 5A), the β-barrel core is relatively static whereas the loops and helices framing the two binding sites show particular mobility. The presence of VP16 (438–454) significantly alters the extent of motion (Fig. 5B). Particularly notable is that the upper loop on the H1 binding surface (residues 409–424 of Med25) appears to strongly interact with VP16. Supporting this model is the effect of mutations within this region on another H1 binding activator, ERM; a K411E mutation, for example, resulted in fourfold weaker ERM binding (Fig. 2C). The loop at the lower portion of this binding interface (residues 435–446) is also significantly altered upon interaction with VP16 and, again, interaction with H1 face-targeting ligands such as ERM is altered upon mutation at this site. The helices flanking the H1 binding surface also undergo significant stabilization upon binding, suggesting that they also play an important role in the defining the binding site. An analysis of RMSFs of residues in Med25 AcID unbound to any ligand reveals that the most dynamical regions of the protein are indeed the loops, with significant motion in the flanking helices as well, consistent with the preliminary structural model (2, 3).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

Emerging structural model for AcID–activator complex formation. (A and B) The NMR coordinates for Med25 AcID (PDB ID code 2XNF) (2) were used to construct the initial structure of Med25 in CHARMM using the Multiscale Modeling Tools for Structural Biology. For B, VP16 (438–454) G450C was constructed in CHARMM as a helical peptide, which was then patched in CHARMM to Med25 C506 through the formation of a disulfide bond at C506 (transparent blue helix). Using GBSW implicit solvent, temperature replica exchange was implemented using the CHARMM22 force field (32). The RMSFs were calculated for each Med25 AcID residue by overlaying Cα atoms for all of the coordinate files produced from the simulations. The coloring correlates with the degree of dynamical behavior of each region. (C) Structure of chemical cochaperone 22 obtained from a Tethering screen. The bar graph is a comparison of k_off for VP16 (467–488) (blue bars) for Med25 AcID or Med25 AcID with 22 covalently Tethered; the red bars summarize data from analogous experiments with ATF6α. The values shown are the average of 2–3 independent experiments with the indicated errors (SD).

The two solvent-accessible cysteines (C497 and C506) in Med25 AcID are adjacent to regions that are predicted to be the most mobile in the preliminary structural model outlined above and that are most affected by activator binding. Thus, a Tethering screen of Med25 AcID utilizing a 1,600-member library was carried out using standard methods (33, 34). This experiment identified cochaperone 22 as a molecule that covalently labels C506 in Med25 AcID with high efficiency (SI Appendix, Fig. S23). We tested the allosteric effects of the Tethered compound 22 using transient kinetics analogous to the experiments of Fig. 4. Similar to the effects of natural activator ligands, the values of k_off for both labeled activators were reduced by 25% in the presence of 22 while k_on was unaffected. Thus, even a fragment molecule can recapitulate the key binding features of a natural activator despite considerable differences in size. This suggests that Med25 AcID will be druggable through the targeting of its most dynamic regions, despite its large binding surfaces. Particularly given our prior success with a similarly dynamic but structurally distinct ABD from CBP/p300, this appears to be a general strategy for the discovery of small molecule modulators of transcriptional coactivators.