A Frustrated Binding Interface for Intrinsically Disordered Proteins*

Background: Protein-protein interactions often involve intrinsically disordered protein domains. Results: The binding interface between two disordered protein domains is suboptimal, or frustrated, with regard to the energetics. Conclusion: The frustration likely results from the promiscuous binding behavior of these disordered domains. Significance: Suboptimal binding interfaces may be common among intrinsically disordered proteins with multiple binding partners. Intrinsically disordered proteins are very common in the eukaryotic proteome, and many of them are associated with diseases. Disordered proteins usually undergo a coupled binding and folding reaction and often interact with many different binding partners. Using double mutant cycles, we mapped the energy landscape of the binding interface for two interacting disordered domains and found it to be largely suboptimal in terms of interaction free energies, despite relatively high affinity. These data depict a frustrated energy landscape for interactions involving intrinsically disordered proteins, which is likely a result of their functional promiscuity.

Intrinsically disordered proteins are very common in the eukaryotic proteome, and many of them are associated with diseases. Disordered proteins usually undergo a coupled binding and folding reaction and often interact with many different binding partners. Using double mutant cycles, we mapped the energy landscape of the binding interface for two interacting disordered domains and found it to be largely suboptimal in terms of interaction free energies, despite relatively high affinity. These data depict a frustrated energy landscape for interactions involving intrinsically disordered proteins, which is likely a result of their functional promiscuity.
Intrinsically disordered proteins (IDPs) 3 are abundant in the eukaryotic proteome and participate in various cellular processes such as signaling, cell cycle control, transcription, and translation (1). Due to their frequent occurrence and the fact that many IDPs are involved in diseases such as cancer and neurodegenerative disorders such as Parkinson and Alzheimer (2), the relationship between disorder and (dys)function has been the subject of intense research in the last decade. A recurrently cited reason for the prevalence of IDPs is that disorder enables a combination of low affinity and high specificity and that it also facilitates binding to several different targets (1). However, this binding malleability could also lead to interactions with targets for which IDPs were not intended to bind and hence lower their specificity. This apparent contradiction leads to the following question. How optimized are the interactions involving IDPs? To address this issue, we have analyzed in detail the largely hydrophobic interaction surface between the molten globule nuclear coactivator-binding domain (NCBD) of CREBbinding protein and the fully disordered activation domain from the p160 transcription coactivator for thyroid hormone and retinoid receptors (ACTR) (3)(4)(5)(6).
We have previously used ACTR/NCBD as a model system to investigate the mechanisms of coupled binding and folding (7)(8)(9). The association between NCBD and ACTR is fast, with a transition state that contains only a few native hydrophobic interactions, followed by a cooperative formation of native contacts after the rate-limiting barrier, resembling the nucleation condensation mechanism in protein folding (10).
In this work, we used protein engineering, namely the double mutant cycle (11), in combination with kinetic experiments to map in detail the binding interface between ACTR and NCBD. We found that the binding interface contains energetic frustration and propose that this is a necessary result of the promiscuity of IDPs such as ACTR and NCBD.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-Human NCBD and ACTR were expressed and purified as described previously (7,8).
Binding Kinetics-Kinetic measurements of the interactions between mutant NCBD and mutant ACTR were performed using an upgraded SX-17MV stopped-flow spectrometer (Applied Photophysics, Leatherhead, United Kingdom). Stoppedflow binding data for single mutant interactions with the wild type are from Dogan et al. (8). The change in Trp fluorescence was monitored, with excitation at 280 nm and a 320-nm longpass emission filter. Experiments were done at 277 K in 20 mM sodium phosphate (pH 7.4) and 150 mM NaCl. Apparent association rate constants of binding (k on app ) were obtained by measuring k obs values for binding at different concentrations of ACTR (0.5-25 M) and 1 M NCBD. The data (k obs versus [ACTR]) were fitted to the general equation for a reversible association of two molecules (12). Displacement experiments were performed to determine apparent dissociation rate constants of binding (k off app ) by mixing a preformed ACTR-NCBD complex with an excess of NCBD WT (7,8). K d values were obtained by taking the ratio of k off app to k on app . Calculation of ⌬⌬⌬G c was performed as described previously (13).
Isothermal Titration Calorimetry Experiments-The thermodynamics of ACTR/NCBD association was characterized using an iTC200 isothermal titration calorimeter (GE Healthcare). Proteins were dialyzed prior to isothermal titration calorimetry measurements, which were performed in 20 mM sodium phosphate (pH 7.4) and 150 mM NaCl and at 278 K. NCBD (150 -365 M) was titrated into an ACTR solution with concentrations ranging from 15 to 30 M. Each titration series consisted of an initial 0.4-or 0.5-l injection, followed by 18 subsequent 2-l injections. Data were corrected for the heats of dilution and fitted to a one-to-one binding model using the Origin software provided with the iTC200 instrument.

RESULTS
We mapped the energetics of the binding interface between ACTR and NCBD using site-directed mutagenesis and binding kinetics. In a previous study (8), we made 10 deletion mutations at hydrophobic positions in both ACTR and NCBD and performed a ⌽ value analysis for binding. Here, we used the same 20 mutants to measure the coupling free energies of binding (⌬⌬⌬G c ) between different positions in ACTR-NCBD using the double mutant cycle analysis (11). The side chains subjected to mutation are in the interface of ACTR and NCBD as defined by the NMR structure (4). Thus, our measured ⌬⌬⌬G c values reflect the energetic cross-talk in the binding interface.
Stopped-flow fluorometry was used to determine the binding dissociation constants (K d ) as the ratio between k off and k on for each pair of mutants. These K d values are identical within error to K d values determined by isothermal titration calorimetry ( Fig. 1). ⌬⌬⌬G c values were obtained from K d values as described in detail previously (13,14). Briefly, the K d values of four bimolecular ACTR-NCBD complexes were measured in each cycle (7,8): the wild-type complex, single mutants X and Y, and the double mutant (in which both X and Y mutations are present), as illustrated in Fig. 2. If the sum of the differences in the free energy of binding between the wild-type and single mutant complexes (⌬⌬G mutantϪX ϩ ⌬⌬G mutantϪY ) is identical to ⌬⌬G mutantϪXY when both mutations are present, then the FIGURE 1. Comparison of equilibrium dissociation binding constants (K d ) obtained by stopped-flow fluorometry (NCBD Y2108W /ACTR WT (a) and NCBD Y2108W /ACTR L1049A (b)) and isothermal titration calorimetry (NCBD Y2108W /ACTR WT (c) and NCBD Y2108W /ACTR L1049A (d)). Observed rate constants (k obs ) are plotted against the concentration of ACTR, and from the fitting (12), the apparent association rate constant of binding (k on app ) is obtained. The insets show a trace in a displacement experiment in which a preformed NCBD Y2108W /ACTR solution was mixed with an excess of NCBD WT to obtain the apparent dissociation rate constant (k off app ). Stopped-flow data shown in a are from Dogan et al. (7), and data shown in b are from Dogan et al. (8). The two different methods produced the same K d .
free energies are additive, and positions X and Y are not energetically coupled to each other. However, if ⌬⌬⌬G c does not equal zero, then a cross-talk in terms of free energy between X and Y exists.
If the wild-type protein-protein interaction is highly optimized in terms of all intra-and intermolecular interactions, it is likely that it becomes less optimized by the first mutation in the cycle. This means that the first mutation weakens the effect of the second mutation, and ⌬⌬⌬G c becomes positive. Conversely, if the first mutation boosts the effect of the second mutation, then ⌬⌬⌬G c becomes negative, which is a sign of a less optimized wild-type complex.
We were able to measure the interaction free energy for 63 pairs of amino acid side chains situated in the interface between ACTR and NCBD in the complex (Table 1). ⌬⌬⌬G c values for several other pairs were not possible to obtain by the stoppedflow technique due to elevated observed rate constants (k obs ), which in turn result from severe destabilization of the ACTR-NCBD complex upon double mutation. Nevertheless, the obtained values were well distributed, thus ensuring a good coverage of the whole ACTR-NCBD binding surface. The coupling free energies for a few positions are, to some extent, distance-dependent. However, when all of the coupling free energies are plotted against the distance between the positions, no clear correlation is observed (Fig. 3).

DISCUSSION
Despite a recent avalanche of articles on IDPs, there is still a paucity of quantitative data on the interactions in the interface of disordered proteins. Such knowledge is needed for a general understanding of IDP interactions but may also help drug design. We have here investigated in detail the interaction energetics in the interface formed in a coupled binding and folding reaction of two disordered protein domains, ACTR and NCBD, and found it to be rather distinct from those of ordered domains.
The coupling free energies determined for ACTR and NCBD reflect the network of noncovalent bonds holding the two domains together in the complex. Note that our interpretation of ⌬⌬⌬G c values is not dependent on mechanism. Neverthe-  less, for the following reasons, it is likely that they reflect effects in the binding interface: (i) the nature of the mutations (conservative deletion mutations of hydrophobic residues), (ii) their location (in the interface), and (iii) the effect on k off rather than k on (8), which means that the short-range interdomain hydrophobic interactions break before the rate-limiting step of dissociation.
To visualize the spatial distribution of the interaction energies, we mapped the ⌬⌬⌬G c values onto the structure of the ACTR-NCBD complex (Fig. 4). Surprisingly, ϳ80% of the ⌬⌬⌬G c values adopt negative values ( Fig. 3 and Table 1), which means that the effect of having both mutations present at the same time is more unfavorable compared with the theoretical combination of single mutations. One interpretation of these negative ⌬⌬⌬G c values is that the binding interface is not fully optimized for the ACTR/NCBD interaction. Previously determined coupling free energies in the context of folded and stable proteins generally displayed positive values (13,(15)(16)(17)(18)(19), although for some systems, both positive and negative ⌬⌬⌬G c values have been reported (20). A positive ⌬⌬⌬G c means that the first mutation diminishes the effect of the second mutation, and along the same line of reasoning as above, the interpretation is that the interactions of these ordered proteins are highly specific and optimized throughout the binding interface and even the whole protein domain (Fig. 5a) (13).
The large number of negative coupling free energies in the interaction surface between ACTR and NCBD may be understood in terms of frustration. The concept of minimal frustration was initially proposed for protein folding (21) but applies also to binding. In a hypothetical fully optimized protein-protein interaction, all possible bonds are satisfied in the complex, and there is no frustration. In the free ground states, however, the respective binding interface may contain frustration, which, upon binding, is minimized due to the favorable interactions between the two proteins. In a protein-protein interface with all interactions satisfied, a perturbation of the interface by mutation may affect neighboring residues and result in a positive coupling free energy in a double mutant cycle. The first mutation introduces strain in the protein-protein complex, which lowers the effect of the second mutation. Conversely, in a suboptimal strained binding interface, a mutation might even "relax," i.e. stabilize the protein-protein complex and make it more sensitive for the second mutation, leading to a negative coupling free energy (Fig. 5b).
Thus, protein-protein interactions are optimized by minimizing the frustration upon binding. However, IDPs with multiple binding partners, such as ACTR and NCBD, cannot minimize the frustration in the binding interface to the same degree as proteins involved in more specific interactions because their amino acid residues make different interactions with different partners and may even adopt distinct conformations. For example, NCBD adopts a completely different conformation when bound to IRF-3 compared with the ACTR-NCBD complex (22), and the C terminus of p53 has been shown to have at least four different bound conformations depending on the interacting ligand (23). We therefore suggest that, in general, protein-protein interactions involving functionally promiscu-  Table 1) between pairs of residues plotted against the shortest distance between them using the first model in the NMR structure of the ACTR-NCBD complex (Protein Data Bank code 1KBH) (4). Residues are color-coded by ⌬⌬⌬G c values between a certain position colored cyan, as indicated by the label next to each image, in NCBD to different residues in ACTR (a) and in ACTR to different residues in NCBD (b), with a gradient that ranged from Ϫ1 (red) to 1 (blue) kcal/mol. Images were generated using PyMOL (39). FEBRUARY 28, 2014 • VOLUME 289 • NUMBER 9 JOURNAL OF BIOLOGICAL CHEMISTRY 5531 ous IDPs display more frustration than those of well folded proteins with specific binding partners.

A Frustrated Interface between Disordered Domains
One important implication of this suggestion is that it challenges the prevalent low affinity-high specificity argument for binding of IDPs. It has been previously suggested that the interactions forming the highly hydrophobic binding interfaces by ␣-helical molecular recognition features (so-called ␣-MORFs, which undergo a disorder-to-order transition upon binding) are predominantly nonspecific (24). NCBD has been shown to bind to several widely different ligands (4,22,(25)(26)(27)(28)(29)(30). The ACTR-NCBD complex has a rather high affinity (93 nM at physiological salt) resulting from hydrophobic and electrostatic interactions but yet apparently combined with a frustrated interface. There are previous studies that have collected data from literature on the binding kinetics for stable proteins and IDPs (31,32), and they found that IDPs tend to have higher k off values than folded proteins. It is believed that such a difference would be one of the key advantages of being disordered (33) because it means that complexes formed by IDPs can dissociate rapidly, which is important for proteins acting as hubs in signaling pathways. Thus, it is tempting to speculate that the frustrated interface is related to a high k off (26 s Ϫ1 at 25°C for ACTR-NCBD). However, there are a couple of shortcomings with such tabulations of binding kinetics, such as data being collected at different ionic strengths, temperatures, and pH values and the use of different experimental methods. Therefore, at present moment, it is difficult to make such comparisons and draw any conclusions about the magnitudes of binding rate constants when comparing IDPs with stable and folded proteins. However, it is clear that IDPs interact with their targets with a wide range of binding affinities, from picomolar to micromolar (34 -37). The most likely general scenario for IDPs is, almost by necessity, a suboptimized binding site in combination with basically any affinity depending on the relevant intracellular concentration of ligand and biological function.
Interestingly, although the majority of coupling free energies were negative, most of the positive coupling free energies were obtained for two residues: Leu-2087 in NCBD and Leu-1049 in ACTR. The positions in NCBD that displayed positive interaction energies with Leu-1049 in ACTR may form a contiguous positive pathway (Fig. 4), although the coupling between NCB-D Leu-2096 and ACTR Leu-1049 , which seems to be important for this pathway, is missing because we were not able to determine an accurate value of ⌬⌬⌬G c between these two positions. Similarly, the residues in ACTR that displayed positive interaction energies with Leu-2087 in NCBD form a putative contiguous pathway (Fig. 4). Indeed, both of these hydrophobic side chains could interact with other proteins: they are partially solventexposed in the complex, with 50 and 30% of the surface area being accessible in the bound state for Leu-2087 in NCBD and Leu-1049 in ACTR, respectively. Such contiguous pathways were previously observed for PDZ domains (14), and they offer the possibility for a specific route for long-range signal transmission (38).