Functional interplay between protein domains in a supramodular structure involving the postsynaptic density protein PSD-95

Cell scaffolding and signaling are governed by protein–protein interactions. Although a particular interaction is often defined by two specific domains binding to each other, this interaction often occurs in the context of other domains in multidomain proteins. How such adjacent domains form supertertiary structures and modulate protein–protein interactions has only recently been addressed and is incompletely understood. The postsynaptic density protein PSD-95 contains a three-domain supramodule, denoted PSG, which consists of PDZ, Src homology 3 (SH3), and guanylate kinase-like domains. The PDZ domain binds to the C terminus of its proposed natural ligand, CXXC repeat–containing interactor of PDZ3 domain (CRIPT), and results from previous experiments using only the isolated PDZ domain are consistent with the simplest scenario for a protein–protein interaction; namely, a two-state mechanism. Here we analyzed the binding kinetics of the PSG supramodule with CRIPT. We show that PSG binds CRIPT via a more complex mechanism involving two conformational states interconverting on the second timescale. Both conformational states bound a CRIPT peptide with similar affinities but with different rates, and the distribution of the two conformational states was slightly shifted upon CRIPT binding. Our results are consistent with recent structural findings of conformational changes in PSD-95 and demonstrate how conformational transitions in supertertiary structures can shape the ligand-binding energy landscape and modulate protein–protein interactions.

Detailed characterization of protein-mediated interactions is a critical task for understanding cellular processes as well as for identification of novel drug targets. From an experimental perspective, a quantitative description of protein-mediated networks demands detection of specific interactions, followed by an attempt to understand larger systems. It should be noted, however, that describing protein interaction networks is a daunting task complicated by several factors that could be hard to estimate within the cellular environment, such as concentrations of individual proteins and compartmentalization. Additionally, proteins may experience conformational changes that may be difficult to capture but nevertheless modulate their function. Thus, a combination of structural, functional, and computational approaches is generally necessary to understand the role of conformational changes.
Postsynaptic density protein-95 (PSD-95) is a multidomain protein abundant in the postsynaptic density of neurons, where it acts as a scaffolding protein (1). Starting from the N terminus, this protein contains three PDZ domains followed by an Src homology 3 (SH3) 3 domain and a guanylate kinase-like (GK) domain (2). The PDZ domain is one of the most common protein interaction domains in our proteome (3). It presents a binding groove on its surface in which protein ligands can bind as a ␤-strand to form an extended ␤-sheet with the PDZ domain. PSD-95 contains a supertertiary structure (4); the first two PDZ domains form one supramodule (5), whereas the third PDZ domain (PDZ3) forms another supramodule together with the SH3 and GK domains (denoted PSG) (6 -9) (Fig. 1). This feature is shared by other membrane-associated GK family proteins (10), to which PSD-95 belongs. In general, PDZ domains display similar affinity for a range of protein ligands (11) so that specificity must rely on spatial and temporal control in the cell. Because of this, it is often difficult to pinpoint natural ligands of PDZ domains, but a likely native ligand of PDZ3 from PSD-95 and of PDZ3 from its closely related paralog SAP97 is CXXC repeat-containing interactor of PDZ3 domain (CRIPT) (12,13). PDZ3 binds the C terminus of CRIPT with a K d of around 0.3 M, and this affinity is achieved mainly via interactions between the six C-terminal residues of CRIPT (YKQTSV) and the binding groove of PDZ3 (14 -16). However, several recent studies have highlighted the role of context dependence and more or less well-defined interactions outside of the binding motif for modular protein-protein interactions (17)(18)(19)(20), mechanisms for which conformational transitions are likely important.
In this study, we investigated the binding kinetics of the PSG supramodule from human PSD-95 with a CRIPT peptide and found that the simplest mechanistic model consistent with the data involves two distinct conformational states of PSG. The two interconverting states can both bind a short CRIPT peptide with a similar affinity, but only one of the states can bind a longer CRIPT peptide. Moreover, the rate constants differ so that one conformer binds CRIPT faster than the other one, reflecting changes around the binding groove associated with the conformational transition. Previous work demonstrated conformational transitions within the PSG using structural approaches (8,9), and more recently, functional oligomerization has been shown to be dependent on the PSG supramodule (17,(21)(22)(23). The binding kinetics observed in this study could be related to these phenomena.

The binding kinetics of PSG and a six-amino-acid CRIPT peptide displayed biphasic behavior
We monitored binding of CRIPT to PSG using an N-terminal dansyl group attached to the six C-terminal residues of CRIPT (D-CRIPT 6 ). The dansyl quenches the fluorescence of an engineered Trp in PSG. The Trp was situated in the same position in PDZ3 (residue 337) as in previous studies of the isolated PDZ3 domain, where it did not affect affinity or kinetics of binding (16). Quenching of the Trp upon binding resulted in kinetic traces with high signal to noise ( Fig. 2A). We previously performed extensive experiments with isolated PDZ3 and D-CRIPT 6 that yielded single exponential traces as pseudofirst-order conditions were approached, consistent with an apparent two-state binding mechanism (16,24). However, fitting of the kinetic binding data of the PSG supramodule and D-CRIPT 6 to a single exponential function consistently yielded residuals displaying a clear nonrandom distribution around zero, even at D-CRIPT 6 concentrations 10-fold higher than those of PSG (Fig. 2), suggesting a more complex mechanism than two-state. A double-exponential equation fitted well to the experimental traces recorded over 1 s, and the observed rate constants, k obs1 and k obs2 , obtained from the fit were determined over a range of D-CRIPT 6 concentrations (Fig.  2C). The apparently linear dependence of both rate constants on D-CRIPT 6 concentration suggests that the PSG sample contained two distinct species, both of which could bind D-CRIPT 6 , one with an association rate constant, k on1 , of 5.8 ϫ 10 6 M Ϫ1 s Ϫ1 (from k obs1 ) and the other with a k on2 of ϳ0.56 ϫ 10 6 M Ϫ1 s Ϫ1 (from k obs2 ). In terms of amplitudes of the kinetic traces, k obs1 had a relatively large and k obs2 a relatively small amplitude (Fig. 2D). The k off values could not be determined directly from these binding data because they were too low to allow accurate extrapolation of k obs to zero D-CRIPT 6 concentration.

Displacement experiments corroborated the presence of two distinct molecular species
To investigate the dissociation kinetics of the PSG:CRIPT complex, we performed displacement experiments in which a preformed complex of PSG:D-CRIPT 6 (1 and 5 M final concentrations) was rapidly mixed with an excess (100 -200 M) of unlabeled CRIPT peptide. In contrast to what we observed previously for the isolated PDZ3 domain (16,24), the dissociation kinetics of PSG:D-CRIPT 6 were also biphasic ( Fig. 3A), which strongly supports the presence of two distinct binding-competent PSG conformers. In fact, in a displacement experiment, any sequential mechanism, such as induced fit, usually yield kinetic dissociation transients that are well-described by single-exponential kinetics. The observed rate constants, which approximate the overall apparent dissociation rate constants (k off app ), were approximately k off1 app ϭ 1.35 s Ϫ1 and k off2 app ϭ 0.17 s Ϫ1 , respectively (Fig.  3B). The errors in the amplitudes were large because of covariation of the parameters, a common phenomenon in curve fitting of complex models (25). Nevertheless, based on six independent experiments, we found that the fast phase (described by k off1 app ) was associated with a relatively smaller amplitude and the slow phase (described by k off2 app ) with a relatively higher amplitude (Fig. 3C).

Affinity between PSG and CRIPT, determined by isothermal titration calorimetry, confirmed an affinity in the low micromolar range
From the kinetic data, it is not clear which k off app belongs to which k on . We therefore performed ITC experiments to obtain an independent estimate of K d . ITC experiments at the temperature of the stopped flow (10°C) resulted in poor thermograms, but experiments at 20°C, 25°C, 30°C, and 35°C yielded good data (Fig. 4). The K d values at 20°C-30°C were all around 0.2 M, and we deem it likely that the affinity is similar at 10°C. The two K d values obtained from the ratios of rate constants, k off1 app /k on1 (0.23 M) and k off2 app /k on2 (0.30 M), respectively, are in very good agreement with the ITC data. This K d is also close to that obtained for the isolated PDZ3 domain (15,16). On the other hand k off1 app / k on2 and k off2 app /k on1 would give K d values of 29 nM and 2.4 M, respectively. Thus, the equilibrium binding data were most consistent with a scenario where k off1 app and k on1 are associated with the binding to one conformer (denoted PSG A ) and k off2 app and k on2 to the other conformer (PSG B ). However, a model with two noninterconverting conformers predicts that the larger association amplitude should be con-

Dynamics in a supramodular structure
nected to the larger dissociation amplitude. This is inconsistent with the relative amplitudes of the observed kinetic phases. Although the large variation in dissociation amplitudes precludes a detailed quantitative analysis, it is qualitatively clear that the kinetic amplitude of k off2 app ( Fig. 3C) is more prominent than that associated with k obs2 in the binding experiment (Fig. 2D). The most likely explanation for this discrepancy is that there is an equilibrium between PSG A and PSG B and that this equilibrium is shifted so that relatively more PSG B is present upon binding of D-CRIPT 6 ( Fig. 5).

Direct detection of the conformational transition in a doublejump interrupted binding experiment
In an attempt to directly monitor a transition between PSG A and PSG B in their bound conformation according to the scheme in Fig. 5, we performed interrupted binding experiments in a double-jump setup in the stopped flow instrument (Fig. 6A). 2 M PSG was mixed with 4 M D-CRIPT 6 in the "first jump" (concentrations after 1:1 mixing). After a defined delay (or aging) ranging between 10 ms and 200 s, this complex was dissociated by mixing with a large excess of unlabeled CRIPT in the "second jump." Following this second jump, the apparently irreversible dissociation kinetics of PSG:D-CRIPT 6 were monitored as described for a regular dissociation experiment, which, in practice, corresponds to an interrupted binding experiment with a very long delay time. We collected 14 kinetic traces from these double-jump experiments, with delay times between 10 ms and 200 s. The trace at 10 ms delay time was excluded because the mismatch between expected and actual delay time was deemed too large. Next, all remaining dissociation traces (delay times between 24 ms and 200 s) were fitted simultaneously to a double-exponential function with shared values for k off1 app and k off2 app , either locked to the average values reported in Fig. 3B or free-fitted (Fig. 6B). This is possible because the final conditions under which kinetics were measured were identical for each double-jump experiment. However, it is also necessary to perform a simultaneous fit because the kinetic amplitudes at short delay times are too small for accurate free-fitting of k off1 app and k off2 app . Next, the kinetic amplitudes associated with k off1 app and k off2 app were plotted as a function of delay time (Fig. 6, C and D). The change in the kinetic amplitudes over time followed biphasic behavior, as clearly seen for the amplitude related to k off1 app . This biphasic behavior reflects the initial build-up of the two complexes, PSG A :D-CRIPT 6 and PSG B :D-CRIPT 6 , respectively, followed by their equilibration. The change in the amplitudes over time thus follows exponential kinetics. We could fit a double-exponential function to the amplitude data and obtain two observed rate constants. However, the errors were very large, and this experiment is best interpreted qualitatively or semiquantitatively; k obs1 DJ (DJ, double jump) was around 20 -40 s Ϫ1 and k obs2 DJ in the range of 0.1-0.8 s Ϫ1 , depending on which set of amplitudes was used. Thus, at short delay times, both amplitudes increased with increasing delay time with an apparent rate constant, which corresponds to the binding of D-CRIPT 6 . k obs values of more than 150 s Ϫ1 were omitted because they very slightly but progressively deviated from a straight line because of instrumental limitations in the mixing. D, kinetic amplitudes associated with the respective k obs value.

Dynamics in a supramodular structure
to PSG and which, at least for k obs1 DJ , is in excellent agreement with the parameters obtained from single-jump binding exper- At longer delay times, the kinetic phase associated with k off1 app decreases, whereas the phase associated with k off2 app should increase (the latter is not as clear as the former because of scatter in the data). This reflects the redistribution of conformations in the bound complex occurring over 1-10 s with a net flow from PSG A : D-CRIPT 6 to PSG B :D-CRIPT 6 (Fig. 5).

Global fitting of a minimal model for the interaction between PSG and D-CRIPT 6
To test the robustness of the kinetic scheme in Fig. 5, we attempted a global fit of kinetic transients from binding and dissociation experiments using the Kintek software (26,27). Global fitting with numerical integration takes information from both rate constants and kinetic amplitudes into consideration. The information from kinetic amplitudes is often overlooked but can, in favorable cases, be used to distinguish between kinetic models. However, in the case of fluorescencemonitored measurements, this requires knowledge about the fluorescence yields of all molecular species, and those of intermediates are usually unknown. Here, the information from the kinetic amplitudes was used to restrain the parameters in the model. Thus, a global fit of a dataset, including seven binding curves from Fig. 2C and all dissociation curves from the doublejump experiment (Fig. 6), was performed, and it resulted in reasonable kinetic parameters (Fig. 7).
A comparison of the globally fitted rate constants with those obtained from fitting binding and dissociation transients individually shows that, although k on1 is similar, k on2 from the global fit is 4-fold larger than in Fig. 2C. The main reason is that we could not include the data points at high concentrations of D-CRIPT 6 in the global curve fitting because the binding step described by k on1 and k off1 app gets too fast. It is clear that k on2 is closer to 0.6 M Ϫ1 s Ϫ1 (Fig. 2C) than to 2.2 M Ϫ1 s Ϫ1 obtained from the global fit. k off1 and k off2 are also more similar to each other in the global fit than k off1 app and k off2 app in the individual fits presented in Fig. 3. However, the values of k off1 and k off2 in the global fit are dependent on the relative values of k 3 and k Ϫ3 , which are poorly defined. The rate constants k Ϫ3 and k 3 agree

Dynamics in a supramodular structure
well with the fit of the double-jump data in Fig. 6 insofar as their sum is 0.53 s Ϫ1 . Finally, k Ϫ4 and k 4 are fitted as 0.007-0.018 s Ϫ1 , but they are obviously very poorly defined, except that this step must be relatively slow relative to the other ones. In summary, despite low accuracy for several of the fitted rate constants, global fitting (Fig. 7) shows that the experimental data are overall consistent with the proposed minimal model depicted in Fig.  5 for the interaction between PSG and D-CRIPT 6 .

Binding kinetics of a 15-mer CRIPT peptide are monophasic
To further investigate the conformational transition in the PSG supramodule, we performed binding experiments with an N-dansylated peptide corresponding to the 15 C-terminal residues of CRIPT, D-CRIPT 15 . In contrast to D-CRIPT 6 , we only observed single exponential kinetics with D-CRIPT 15 in both binding and displacement experiments (Fig. 8). The apparent rate constants (k on and k off ) for D-CRIPT 15 were more similar to k on1 and k off1 app , associated with the fast phase for D-CRIPT 6 , and giving a similar K d (0.42 M) as D-CRIPT 6 . The K d determined by ITC for D-CRIPT 15 was 1.6 Ϯ 0.2 M at a higher temperature (25°C), in fair agreement with K d from kinetics. One explanation, considering the close match of the k on values, would be that D-CRIPT 15 can only bind to PSG A but not to PSG B . However, even with D-CRIPT 6 , the kinetic amplitude of the slow phase is low, and for the longer D-CRIPT 15 it may not be visible. Regardless of whether D-CRIPT 15 binds to PSG B , this conformation is present in equilibrium with PSG A , as monitored by the smaller D-CRIPT 6 , which can likely access a more narrow binding pocket present in PSG B .

Discussion
Much of our current knowledge regarding protein-mediated interactions is based on study of isolated domains in solution.
Nevertheless, it is clear that, because proteins are dynamic entities, the interaction between contiguous domains may play a role in their structure and function. A hallmark of these interactions is represented by hemoglobin, a system where it was established how the intricate interplay between the four subunits results in very complex binding schemes that ultimately mediate allosteric regulation and fine-tune oxygen (un)binding to optimize its transport from the lungs to tissues. Although the dynamics of allosteric multisubunit proteins have a long history (28), recent work has been focused on multidomain proteins in which several distinct protein domains occur in one single polypeptide as beads on a string. It is becoming clear that such proteins often display supertertiary structures (4) where supramodules could form from two or several domains, which may then cooperate positively or negatively in ligand binding or other functions. Similarly to hemoglobin, simplified models must be used to account for the behavior of such multidomain proteins.
The MAGUK family of proteins, to which PSD-95 belongs, is one class of multidomain proteins that has been well-studied, in particular from the point of view of single domains but also from the perspective of its two supramodules. We previously determined binding rate constants for the isolated PDZ3 domain and established that its interaction with CRIPT is a one-step process (16,24). Here we extended these studies to the PSG supramodule, which consists of PDZ3 and two more domains, SH3 and GK, and found that a model with two conformations, denoted PSG A and PSG B , can satisfactorily describe the data. Although PSG A binds CRIPT with similar rate constants as PDZ3, PSG B appears to display a more restricted binding pocket.
Previous work on PSG from the PSD-95 family or other homologs has not assessed the mechanism with direct kinetic methods. However, multiple complementary methods, including crystallography, NMR, small-angle X-ray scattering, singlemolecule FRET, binding experiments, and modeling, have demonstrated that the three domains form a well-defined and dynamic supertertiary structure (8 -10, 29, 30). For example, single-molecule FRET in PSD-95 showed that PDZ3 displays motional averaging relative to the SH3 domain (8). Furthermore, NMR, small-angle X-ray scattering, and simulations were consistent with distinct structural ensembles in free PSG and CRIPT-bound PSG, where the ensemble of bound conformations was shifted toward extended rather than compact species (9). In more detail, the peptide-binding groove of PDZ3 has been shown to interact weakly with the SH3 domain (or the linker between them) so that, in the ligand-free state, the SH3 domain restricts ligand binding to PDZ3. Binding of a ligand to PDZ3 would thus promote a more open conformation of PSG where SH3 is released from PDZ3. These conformational transitions could be the basis of the biphasic kinetics observed in this study. From a functional perspective, PSD-95 forms oligomers in the postsynaptic density, and one way is via the SH3 and GK domains (6). This oligomerization is mediated by ligand binding to PDZ3 (21). Furthermore, ligand binding to PDZ3 promotes binding of GK to other proteins, such as the G protein subunit Gnb5 (23) but restricts binding of another protein, GukHolder (29). These previous studies are consistent with our

Dynamics in a supramodular structure
kinetic data insofar as the binding kinetics suggest that the PSG B conformation is less accessible than PSG A , as mirrored by the differences in k on1 (PSG A ) and k on2 (PSG B ) upon binding to D-CRIPT 6 , and that D-CRIPT 15 apparently only binds to PSG A . A more restricted binding site provides fewer possibilities of productive initial encounters, reducing the apparent association rate constant. It is possible that the more open PSG A is the conformation allowing SH3-GK-mediated oligomerization and binding between GK and the G protein subunit Gnb5 (17,(21)(22)(23).
In conclusion, there is a firm structural basis for a dynamic supertertiary structure of the PSG from PSD-95 (8,9). This dynamic behavior is characterized by some distinctive kinetic signatures, as highlighted by our kinetic analysis. Furthermore, our experiments estimate a timescale of these transitions of about 0.5 s Ϫ1 at 10°C. We speculate that such conformational The best fit curves in the figure correspond to the latter fit. C and D, the kinetic amplitudes from the respective fits in B were plotted against delay time. This reflects the build up of PSG A :D-CRIPT 6 and PSG B :D-CRIPT 6 , respectively, with time. These amplitude data, in turn, were fitted to a double-exponential function to obtain two "observed double-jump rate constants" for the build up of PSG A and PSG B , respectively. From the fit, k obs DJ values of around 20 -40 s Ϫ1 and 0.1-0.8 s Ϫ1 , respectively, were obtained. Although the parameters are underdetermined, it is clear that there is an initial increase in the respective population, followed by a decrease in PSG A :D-CRIPT 6 . The expected concomitant increase in PSG B :D-CRIPT 6 is lost in the experimental noise.

Dynamics in a supramodular structure
transitions in the PSG are related to the previously reported conformational changes and are most likely connected to its function, as demonstrated previously in several studies (17,(21)(22)(23)29). Future work based on site-directed variants will further explore the functional role of these transitions.

Protein purification
The PSD-95 PDZ3-SH3-GK (PSG) supramodule (residues 308 -724) contained an engineered Trp in the same position (F337W) as in previous studies of the isolated PDZ3 domain (16,24). PSG was expressed from a modified pRSET plasmid (Invitrogen), which was transformed into Escherichia coli BL21(DE3) pLys cells. The cells were grown in lysogeny broth (LB) medium at 37°C until A 600 of the culture reached 0.6.
Then overexpression of protein was induced with 1 mM isopropyl ␤-D-thiogalactopyranoside, and the bottles were incubated overnight in a rotary shaker at 18°C. Cells were harvested by centrifugation at 4°C, and the pellet was resuspended in 50 mM Tris (pH 7.8), 100 mM NaCl, and 10% glycerol and stored at Ϫ20°C until purification. The pellet was thawed and sonicated twice for 4 min each time, followed by centrifugation. The supernatant was filtered and loaded onto a nickel-Sepharose Fast Flow column (GE Healthcare) pre-equilibrated with 50 mM Tris, 100 mM NaCl, 10% glycerol, and 0.5 mM DTT. The column was washed with equilibration buffer, and bound protein was It should be noted that the reported standard errors for all parameters are likely underestimated. We used the global fit to qualitatively test our suggested minimal model. Thus, less emphasis should be put on the estimates of the microscopic rate constants, which are, in some cases, poorly constrained. k on2 is likely overestimated because we could only use lower D-CRIPT 6 concentrations in the global fit to capture the fast phase (Fig. 2). B, fit of interrupted binding stopped flow data to the minimal model in A. Five kinetic traces are shown for clarity. The residuals of the fits are shown below the curve. C, fit of the single-jump stopped-flow data to the model, with residuals shown below the curves. The fits to kinetic traces of three different concentrations of D-CRIPT 6 are shown for clarity.

Dynamics in a supramodular structure
eluted with 250 mM imidazole. The protein sample was dialyzed overnight into 50 mM Tris, 2 mM DTT, 100 mM NaCl, and 10% glycerol. The PSG sample was concentrated and further purified using size exclusion chromatography (S-100, GE Healthcare). Protein purity was quantified by SDS-PAGE and identity by MALDI-TOF MS. Concentrations of protein samples were measured by absorbance at 280 nm and theoretical extinction coefficients based on the amino acid composition. Far-UV CD spectra (200 -260 nm) of the protein (10 M) were recorded to ensure that it was folded. CD experiments were performed in 50 mM sodium phosphate (pH 7.45) and 21 mM KCl (I ϭ 150) at 10°C on a JASCO J-1500 spectropolarimeter using the average of five scans.

Kinetic methods
All kinetic experiments were performed in 50 mM sodium phosphate (pH 7.45), 21 mM KCl (I ϭ 150 mM), 0.5 mM tris(2carboxyethyl)phosphine at 10°C. Kinetic binding and dissociation experiments were performed in an upgraded SX-17 MV stopped-flow spectrophotometer (Applied Photophysics, Leatherhead, UK) and carried out as described previously (16). Briefly, k on was obtained from binding experiments approaching pseudofirst-order conditions ([PSG] Ͻ [D-CRIPT]). PSG (1 M) was rapidly mixed with different concentrations of N-terminally dansyllabeled CRIPT. The shorter peptide, dansyl-YKQTSV, was denoted D-CRIPT 6 , and the longer one, dansyl-GKKVLDT-KNYKQTSV, was denoted D-CRIPT 15 . The binding reaction was monitored by fluorescence, where the Trp at position 337 in PDZ3 was quenched by the dansyl group of the CRIPT peptide. Trp 337 was excited at 280 nm, and emission was followed at 330 nm (using a 330 Ϯ 25 nm interference filter). Kinetic traces were recorded and averaged (a minimum of five individual traces were used at each concentration) and fitted to single-or double-exponential functions to obtain the k obs for each CRIPT concentration. To estimate k on values, k obs1 and k obs2 (for PSG with D-CRIPT 6 ) or k obs (for PSG with D-CRIPT 15 ) were plotted versus CRIPT concentration and fitted to a linear function, where the slope corresponds to k on . Because of slow dissociation rates, k off could not be accurately determined by extrapolation of k obs to zero CRIPT in the binding experiments. Thus, displacement experiments were performed to obtain more accurate estimates of k off . In these experiments, PSG (2 M) was preincubated with D-CRIPT 6 (10 M). D-CRIPT 6 was then displaced from PSG by rapidly mixing the complex with an excess of unlabeled CRIPT 6 (100, 150, and 200 M). The high excess of unlabeled peptide ensured an irreversible dissociation reaction, allowing k off to be estimated from the average of three experiments in which k obs (Ϸ k off ) did not change with CRIPT 6 concentration. Similar to binding, all displacement experiments performed with PSG and D-CRIPT 6 were fitted to a double-exponential function, yielding two distinct observed dissociation rate constants.
Double-jump mixing experiments were performed on an Applied Photophysics SX18-MV stopped-flow instrument according to the manufacturer's instructions with regard to mixing volumes. In particular, a solution containing 4 M PSG in the presence of 50 mM phosphate buffer (pH 7.5) was first mixed 1:1 with D-CRIPT 6 peptide at an initial concentration of 8 M. Then, after a controlled delay time (between 10 ms and 200 s), the solution was challenged with a high concentration of unlabeled CRIPT 6 in a second mixing event (the initial concentration of CRIPT 6 was 100 M. Thus, the final concentrations in the displacement reaction were 1 M PSG, 2 M D-CRIPT 6 , and 50 M unlabeled CRIPT. In analogy to single-mix stopped-flow experiments, the doublejump experiment was carried out at 10°C and analyzed as described under "Results."

Global fitting of kinetic data
Kinetic transients from single-and double-jump stoppedflow experiments were fitted globally to a minimal model (Fig. 5) using KinTek Explorer (26,27). This software utilizes numerical integration and allows fitting of raw kinetic data directly to a mechanistic model. The interrupted binding experiments, with delay times varying from 24 ms to 200 s, were included in the global fit together with binding experiments in which [D-CRIPT 6 ] varied between 2-14 M. For practical reasons, the lower boundary of k 4 and k Ϫ4 was set to 10 Ϫ5 s Ϫ1 in the fitting procedure. k on for the CRIPT 6 peptide used in dissociation experiments (with no dansyl group) was locked to be the same as for D-CRIPT 6 , and k off was locked to 0 (irreversible binding is practically achieved at concentrations where CRIPT 6 Ͼ Ͼ D-CRIPT 6 ). The remaining parameters in the model were fitted without constrains. To account for baseline shifts in the double-jump kinetic experiments, constants were added to the output expressions. A scaling factor was used to account for the difference in total fluorescence output between the single-jump and doublejump stopped-flow experiments.

Isothermal titration calorimetry
Isothermal titration calorimetry (ITC) experiments were performed in 50 mM sodium phosphate (pH 7.45), 21 mM KCl (I ϭ 150), and 0.5 mM tris(2-carboxyethyl)phosphine at 20°C, 25°C, 30°C, and 35°C in an iTC200 instrument (Malvern). D-CRIPT 6 and PSG were dialyzed overnight in the same buffer to minimize artifacts from buffer mismatch. D-CRIPT 6 (220 M) was titrated (16 -19 injections) into a cell containing 22 M PSG, and saturation was obtained by having approximately a two times excess of D-CRIPT 6 . Baseline correction was performed manually to minimize the value in the curve fitting using the software provided with the iTC200 instrument.