Distinct Roles for Conformational Dynamics in Protein-ligand Interactions

Summary

Conformational dynamics has an established role in enzyme catalysis, but its contribution to ligand binding and specificity is largely unexplored. Here we used the Tiam1 PDZ domain and an engineered variant (QM PDZ) with broadened specificity to investigate the role of structure and conformational dynamics in molecular recognition. Crystal structures of the QM PDZ domain both free and bound to ligands showed structural features central to binding (enthalpy), while NMR-based methyl relaxation experiments and isothermal titration calorimetry revealed that conformational entropy contributes to affinity. In addition to motions relevant to thermodynamics, slower μs-ms switching was prevalent in the QM PDZ ligand binding site consistent with a role in ligand specificity. Our data indicate that conformational dynamics plays distinct and fundamental roles in tuning the affinity (conformational entropy) and specificity (excited-state conformations) of molecular interactions. More broadly, our results have important implications for the evolution, regulation and design of protein-ligand interactions.

Graphical Abstract

The role of conformational dynamics in molecular recognition remains controversial. In a model PDZ domain system, Liu et al. (2016) shows that conformational dynamics play distinct roles in protein-ligand interactions. Fast motions contribute to the entropy of binding and slow motions relate to binding specificity.

INTRODUCTION

Molecular recognition is critical for the function and regulation of signal transduction in cells. Over the last several decades tremendous effort has been expended to rigorously understand the structural, thermodynamic and dynamic bases for protein-protein interactions. Although many insights have been obtained from these studies, the role of conformational dynamics is poorly understood. In particular, the connection between conformational dynamics and binding specificity remains elusive. Seminal studies with calmodulin (Frederick et al., 2007; Marlow et al., 2010) and catabolite activator protein (CAP) (Tzeng and Kalodimos, 2012) have established a role for fast (ps-ns) dynamics in modulating conformational entropy and thus affinity, yet the generality of this relationship has not been established. Similarly, the contribution of slower, μs-ms motions to the binding mechanism has received attention. Most studies have focused on enzymes and relating μs-ms motions to kinetics (Henzler-Wildman and Kern, 2007; Lisi and Loria, 2016; Palmer, 2015). Recent efforts have begun to analyze the mechanisms of binding, particularly attempting to distinguish between conformational selection and induced fit models (Boehr et al., 2009; Chakrabarti et al., 2016; Clore, 2014). In spite of these studies, experimental evidence connecting slower motions, excited states and binding specificity is lacking. Thus, a major factor motivating the study presented here is to delineate the potential role(s) of fast and slow conformational dynamics in molecular recognition.

PDZ (PSD-95/Dlg/ZO-1) domains are small globular protein-protein interaction domains consisting of ~80–90 amino acids that predominantly recognize the C-terminus of their binding partners. These modular domains have been excellent model systems for investigating the details of molecular specificity (Lee and Zheng, 2010). Several large-scale proteomic studies have redefined the general classification of PDZ domain ligands and have shown that specificity is optimized over the entire ligand-binding site and tuned across the proteome to minimize cross-reactivity (Chen et al., 2008; Stiffler et al., 2007; Tonikian et al., 2008). Thus, PDZ/peptide interactions feature epistasis (i.e. context dependence) and therefore specificity results from a precise combination of interactions throughout the entire binding interface (Ernst et al., 2010; Ernst et al., 2009; Liu et al., 2013; Melero et al., 2014; Shepherd et al., 2011; Stiffler et al., 2007). Structural analyses of PDZ complexes provide ample support for this idea (Ernst et al., 2014; Lee and Zheng, 2010).

In contrast to the structural characterization of PDZ domains, relatively few studies on their conformational dynamics have been published. Computational and solution nuclear magnetic resonance (NMR) spin relaxation studies indicate that PDZ domains contain energetic and dynamic “hot spots” that are distributed throughout the domain at the ligand-binding site and distally (Fuentes et al., 2004; Law et al., 2009; Lockless and Ranganathan, 1999; Ota and Agard, 2005). These hot spot regions coincide with protein “sectors” that describe functionally correlated residues in tertiary structures of proteins and map to residues critical for specificity (Fuentes et al., 2004; Halabi et al., 2009; McLaughlin et al., 2012). Collectively, these studies provide evidence for a correlation between structure, dynamics and specificity.

T-cell lymphoma invasion and metastasis gene 1 (TIAM1) and 2 (TIAM2) encode for guanine nucleotide exchange factor (GEF) proteins specific for the Rac1 GTPase (Chiu et al., 1999). Tiam1 and 2 are large, modular proteins containing several protein-protein interaction domains, including a single PDZ domain. The Tiam1/2 PDZ domains share ~28% amino acid sequence identity and display distinct but overlapping specificity (Shepherd et al., 2011). For instance, the interaction between the Tiam1 PDZ domain and a C-terminal peptide from the adhesion receptor neurexin1 (NRXN1) is weak (K_d ~2.4 mM), whereas the affinity of the Tiam2 PDZ domain for NRXN1 is ~500-fold tighter (K_d ~5 μM). In contrast, both the Tiam1 and Tiam2 PDZ domains bind tightly to a C-terminal peptide from Caspr4 (related to the NRXN1 receptor family) with K_ds of ~18 and 3 μM, respectively (Shepherd et al., 2011). Previously, we determined the structure of Tiam1 PDZ domain in complex with a C-terminal peptide from syndecan1 (SDC1), which showed that ligand specificity was determined by the S₀ and S₋₂ binding pockets, specific for the ultimate (P₀) and antepenultimate (P₋₂) amino acids of the peptide (Figure 1A) (Liu et al., 2013; Shepherd et al., 2010). Further studies revealed that substitution of four non-conserved residues in the S₀ and S₋₂ binding pockets of the Tiam1 PDZ domain with the analogous residues found in the Tiam2 PDZ domain (i.e. Tiam1 quadruple mutant or QM PDZ) was sufficient to switch the ligand binding specificity from Tiam1 to that of the Tiam2 PDZ domain (Figure 1B) (Shepherd and Fuentes, 2011; Shepherd et al., 2011). As such, the affinity of the QM PDZ domain was ~50-fold tighter for NRXN1 and ~5-fold weaker for SDC1 compared to the Tiam1 PDZ domain. However, the structural and energetic mechanism(s) underlying this specificity switch and the role of conformational dynamics remain unknown.

Figure 1. Design of the QM PDZ Domain.

(A) Space filling model of Tiam1 PDZ in complex with SDC1 peptide ligand (PDB code 4GVD). Residues that form the S₀ and S₋₂ specificity pockets are labeled and colored in red. (B) Primary sequence alignment of the human Tiam1, QM and mouse Tiam2 PDZ domains. Secondary structure is indicated by rectangles (α-helix) and arrows (β-strand). The six residues labeled in (A) are highlighted in the sequence alignment. The four mutations in the QM are colored yellow.

In this work, we determined the structural, thermodynamic and dynamic origins of the engineered specificity of the QM PDZ domain. Using a panel of peptide ligands, we established that the QM PDZ domain has a broadened ligand specificity biased towards that of the Tiam2 PDZ domain. Crystal structures of the QM PDZ domain free and bound to C-terminal peptides derived from the proteins Caspr4 and NRXN1 revealed novel binding interactions, primarily through a mutated phenylalanine residue in the S₀ pocket and a glutamic acid residue in the S₋₂ pocket. NMR-based methyl relaxation analyses coupled with isothermal titration calorimetry provided evidence for increased dynamics that correlated with changes in conformational entropy. Finally, the QM PDZ domain had pronounced motions on the μs-ms timescale suggesting a relationship with binding specificity. Taken together, our data provide a near complete thermodynamic and structural model for the engineered binding specificity of the QM PDZ domain. Importantly, we find that conformational dynamics contributes significantly to both the thermodynamics and specificity of binding.

RESULTS

Structure and Stability Changes in the QM PDZ Domain

To probe the effects of mutations on the structure of the QM PDZ domain in solution, we collected a ¹H-¹⁵N-heteronuclear single quantum coherence (HSQC) spectrum (Figure 2). Overall, the QM PDZ domain had well-dispersed chemical shifts in the ¹H and ¹⁵N dimensions, suggesting an intact global fold. We were able to assign the majority of non-proline backbone amide resonances by standard triple-resonance experiments with the exception of four residues (Y858, E880, S908 and S909), whose resonances were broadened beyond detection. Comparison of the Tiam1 and QM PDZ domain spectra showed that the chemical shifts of most residues did not change, indicating that there was no gross structural alteration. However, several chemical shift differences between the Tiam1 and QM PDZ domains were apparent and clustered around the site of the mutations (L911M, K912E, L915F and L920V). In particular, residues in the α2 helix (residues F914, S916, Q917 and S919) and F860 located in the β2 strand had large perturbed chemical shifts (Figure 2). In addition, several minor peaks were visible at low contour levels. Although we were not able to assign these peaks, they likely result from alternate conformations or local unfolding as they disappear upon titration with peptide ligand (data not shown).

Figure 2. Chemical Shift Changes in the QM PDZ Domain.

Overlaid ¹H-¹⁵N HSQC spectra of QM (blue) and Tiam1 PDZ (red) domains. Residues with the largest changes in chemical shift are labeled and indicated by arrows. See also Figure S1 and S3.

To assess the thermodynamic stability of the QM PDZ domain, we performed thermal and chemical denaturation experiments. The change in molar ellipticity at 222 nm was monitored as a function of temperature or guanidinium hydrochloride concentration. Temperature denaturation data showed that the midpoint of denaturation (T_m) for the Tiam1 PDZ was 5 °C higher than the QM (T_m,WT = 47.8±0.5 °C and T_m,QM = 42.7±0.6 °C). Consistent with this, the chemical denaturation data showed that the QM mutant was destabilized by 1.78±0.09 kcal/mol (Figure S1). The stability of the QM was found to be surprisingly low (0.82 kcal/mol; Figure S1B), suggesting a large fraction (~20%) might be unfolded at 25 °C. However, no evidence for unfolding was seen in the ¹H-¹⁵N-HSQC spectrum, even at lower contours. Together, these data confirm that the four mutations destabilized the QM PDZ domain but did not cause gross structural changes.

The QM PDZ Domain Has Broadened Ligand Specificity Similar to the Tiam2 PDZ Domain

We previously determined the binding specificity of the QM PDZ domain using three representative peptide ligands (Shepherd et al., 2011). To thoroughly probe the specificity profile of the QM PDZ, we tested the binding of seven additional dansyl-labeled peptides by fluorescence anisotropy (Figure 3 and Table S1). Figure 3 shows representative binding curves and the determined affinities of the QM PDZ domain for ten peptides. These data show that the QM PDZ domain bound six peptides with improved affinity (lower K_d), while four peptides had poorer affinity (higher K_d) compared to the Tiam1 PDZ domain. Further comparison indicates that the changes in affinity of the QM PDZ tended towards those of Tiam2 PDZ domain for nine out of ten peptides (Table S1). The sole exception was CADM1, which did not bind the Tiam2 PDZ domain but showed ~14-fold improved affinity for the QM compared to the Tiam1 PDZ domain. Finally, if one uses 200 μM as an cutoff for very weak binding, the QM PDZ domain bound eight peptides, while Tiam1 and Tiam2 PDZ each bound six peptides with only four common to both. These data indicate the QM PDZ domain has a broader specificity than either the Tiam1 or Tiam2 PDZ domains with a clear bias towards the specificity of the Tiam2 PDZ domain.

Figure 3. The QM PDZ Domain has Switched Ligand Binding Specificity.

(A) Representative binding curves for the interaction between the QM PDZ domain and dansylated peptides derived from the syndecan family (SDC1-4), Caspr4, neurexin1 (NRXN1), and CADM1.

(B) Summary of QM PDZ/peptide binding data. The sequence of each peptide and the determined dissociation constant (K_d) is indicated. Fold change represents K_d(WT)/K_d(QM). The K_d values reflect the mean and standard deviation from at least three technical replicates. See also Figure S3 and Table S1.

We used isothermal titration calorimetry (ITC) to gain insight into the thermodynamics of the QM PDZ/Caspr4 interaction (Figure 4A). The K_d of the QM PDZ for the Caspr4 ligand was determined to be ~38 μM, which is approximately 2-fold weaker than the value obtained by fluorescence anisotropy measurements (K_d = 18.3±0.3μM). This is consistent with the differences found previously due to the influence of dansyl chloride moiety used in the fluorescence experiments (Liu et al., 2013). The free energy of binding (ΔG) for Tiam1 PDZ/Caspr4 (Liu et al., 2013) and QM PDZ/Caspr4 interactions was similar (~−6 kcal/mol); however, the enthalpy and entropy differed significantly (Figure 4B). The Caspr4 binding to the Tiam1 PDZ was an entropically driven process, whereas Caspr4 binding to the QM PDZ was enthalpically driven. These data indicate that the Tiam1 and QM PDZ domains use distinct thermodynamic strategies to achieve nearly identical affinity.

Figure 4. Thermodynamic Analysis of the QM PDZ/Caspr4 Domain Interaction by ITC.

(A) Thermogram and integrated titration curve for the Caspr4 ligand bound to the QM PDZ domain.

(B) Thermodynamic parameters for the Tiam1 PDZ/Caspr4 and QM PDZ/Caspr4 interactions at 298 K. Each parameter represents the mean and standard deviation from three technical replicates. See also Table S5.

Crystal Structures of the Free and Ligand-Bound Forms of the QM PDZ Domain

To investigate the structural origin of the QM PDZ binding specificity switch, we solved crystal structures of the PDZ domain free and in complex with two peptide ligands. The structure of the free QM PDZ domain was refined to a resolution of 2.3 Å (Table 1). The QM PDZ domain has a prototypical PDZ domain structure, containing five β-strands and two α-helices arranged in a β-sandwich fold (Figure 5A). Similar to the Tiam1 PDZ structure, the electron density of the β1-β2 loop (residues 851–856) was not visible, suggesting conformational heterogeneity in this loop. The structural overlay of C_α atoms found in secondary structure (excluding the α2 helix) between Tiam1 and QM PDZ domains indicated very little variation, having a root-mean-squared deviation (rmsd) of 0.22 Å. However, closer inspection revealed that the α2 helix had the largest rmsd (residues 908–920 with rmsd of 0.80 Å) caused by a displacement of the helix away from the binding pocket (Figure 5A and Table S2). Taken together, the overall fold of QM PDZ domain was intact but local structural perturbations occurred near the site of the mutations in the α2 helix.

Table 1.

Crystallographic Data Collection and Refinement Statistics

	QM PDZ	QM PDZ/Caspr4	QM PDZ/NRXN1
Data Collection Statistics
Temperature (K)	100	100	100
Wavelength (Å)	1.542	1.000	1.000
Space group	P3221	P21	P212121
Unit cell parameters
a, b, c (Å)	45.84, 45.84, 71.36	51.08, 50.82, 53.04	26.36, 50.32, 61.94
α, β, γ (°)	90.00, 90.00, 120.00	90.00, 92.29, 90.00	90.00, 90.00, 90.00
Molecules per asymmetric unit	1	3	1
Resolution range (Å)	19.85 – 2.30 (2.38 – 2.30) a	37.52 – 2.10 (2.21 – 2.10)	39.06 – 1.90 (1.94 – 1.90)
I/σ(I)	7.7 (2.1)	13.9 (3.9)	39.7 (11.0)
Completeness (%)	93.9 (97.3)	99.4 (100.0)	98.7 (87.1)
Rmerge (%)b	10.4 (42.1)	8.1 (38.3)	3.5 (14.1)
Redundancy	3.18 (3.17)	3.7 (3.8)	6.7 (4.4)
Refinement Details
Resolution (Å)	2.30	2.10	1.90
Rwork/Rfree (%)c	21.69/25.78	18.51/23.56	15.24/18.73
Number of atoms
Protein (peptide)	666 (0)	2092 (187)	714 (92)
Water	47	160	99
B-factor average (Å2)
Protein (main chain)	31.0 (30.2)	27.0 (25.7)	11.86 (10.49)
Peptide	0	36.8	13.77
Water	36.2	29.7	20.22
RMS Deviation from Ideal Geometry (Overall)
Bond lengths (Å)	0.0057	0.0082	0.0118
Bond angles (°)	1.0991	1.1662	1.3759
Dihedral angles (°)	16.205	12.501	14.061
Planarity (°)	0.0035	0.0060	0.0053
Chirality (°)	0.0661	0.0520	0.0449
Ramachandran Plot (% residues)
Most favored	97.56	99.30	98.98
Additionally allowed	2.44	0.70	1.02
Disallowed	0	0	0

rmsd, root-mean-square deviation.

^aValues in parentheses are for the highest resolution shell. One crystal was used for each data collection.

^bR_merge = Σ|I_i-〈I〉|/ΣI_i, where I_i is the intensity of the i^th observation, and 〈I〉 is the mean intensity of the reflections.

^cR = Σ|F_obs-F_calc|/Σ|F_obs|, crystallographic R-factor, where all reflections belong to a test set of randomly selected data.

Figure 5. Structures of the QM PDZ Domain Free and Bound to Caspr4 and NRXN1 Peptides.

(A) Ribbon representation of the QM PDZ (pink) and Tiam1 PDZ (PDB code 3KZD) (gray) domains. Side-chains of the four residues mutated in this study are labeled in red and shown as sticks. The dashed line represents residues without interpretable electron density in both structures.

(B) Structural model of the QM PDZ/Caspr4 complex showing backbone and side-chain interactions. PDZ domain residues involved in peptide binding are colored yellow and labeled, while the Caspr4 peptide is colored cyan. The zoomed in view on the right shows several unique interactions denoted by dotted lines. The conformation of the four mutated residues in the apo QM (pink) and Caspr4-bound (yellow) structures are shown for comparison.

(C) Structural model of the QM PDZ/NRXN1 complex showing backbone and side-chain interactions. PDZ domain residues involved in peptide binding are colored yellow and labeled, while the NRXN1 peptide is colored green. The zoomed in view on the right shows several interactions and the conformation of the four mutated residues in the apo QM (pink) and NRXN1-bound (yellow) structures. See also Figure S2 and Table S2.

To gain insight into how the QM PDZ domain was able to interact with ligands, we determined the structure of two QM PDZ/ligand complexes. We first solved the crystal structure of the QM PDZ bound to a C-terminal peptide from Caspr4 (Ac-ENQKEYFF_COO-). The asymmetric unit of the crystal contained three PDZ domain complexes and the structure was refined to a resolution of 2.1 Å (Table 1). Chain A contained the most complete electron density for the ligand (P₀ to P₋₆ residues) and thus we chose to present its structure, however all conclusions are supported by the three complexes in the unit cell. The QM PDZ/Caspr4 structure had an identical fold as the free QM PDZ domain (C_α rmsd was 0.40 Å) (Figure S2 and Table S2). The electron density for the β1-β2 loop was observable, with apparent hydrogen bonding between backbone amides of Y858-G859 (β1-β2 loop) and the C-terminal (P₀) carboxylate group in the ligand suggesting that the ligand binding stabilizes a single conformation in the β1-β2 loop region (Figure 5B). Additional backbone hydrogen-bonding interactions between the peptide and residues along the β2 strand of QM PDZ (residues 860–865) were observed. In particular, the side chains of mutated residues in the α2 helix facilitated recognition of the Caspr4 peptide by altering the S₀ and S₋₂ binding pockets and introducing new interactions. Notably, the benzene ring of the Phe at the Caspr4 P₀ position made a “parallel-displaced” Π-Π interaction (McGaughey et al., 1998) with the aromatic ring introduced by the L915F mutation in the S₀ binding pocket (Figure 5B). In addition to this Π-Π stacking, the structure suggests that an anion-Π interaction (Jackson et al., 2007) occurs between the E912 side chain oxygens and the edge of the P₀ phenylalanine benzene ring located ~4.0 Å away. Strikingly, the E912 side chain was flipped ~90° from its original orientation in the absence of ligand. Similarly, the side chain of M911 (from the L911M mutation) was flipped ~180° such that the sulfur atom approached the P₋₂ tyrosine ring allowing for a favorable S-Π interaction (Valley et al., 2012) in the S₋₂ pocket. These observations are consistent with the HSQC-based peptide titrations that showed chemical shift perturbations in residues in the α2 helix and β2 strand, particularly in F915, E912 and F860 (Figure S3A and C).

We also determined the structure of the QM PDZ/NRXN1 (Dan-NKDKEYYV_coo-) complex to gain insight into the 50-fold increase in binding affinity of the QM PDZ domain for the NRXN1 ligand. This structure was refined to a resolution of 1.90 Å and the asymmetric unit contained one copy of the QM PDZ/NRXN1 complex (Table 1). As expected, the entire NRXN1 peptide interacted with the β2 strand (residues 858–866) in the PDZ domain through hydrogen-bonding interactions along the backbone (Figure 5C). Compared to the Tiam1 PDZ/SDC1 structure (Liu et al., 2013), the S₀ and S₋₂ binding pockets in the QM PDZ domain underwent unique changes upon binding NRXN1. For instance, the S₀ pocket formed by the side chains of residues Y858, F860, L915 and L920 was enlarged by 20 Ų of accessible solvent area (ASA) providing additional room for the bulkier Val side chain at P₀ in the NRXN1 ligand. Moreover, the L911M and K912E mutations expanded the S₋₂ pocket by ~9Ų to accommodate Tyr at P₋₂ in NRXN1. In addition to these changes, a favorable electrostatic interaction between lysine at P₋₄ in NRXN1 and E912 in the QM PDZ was evident with a distance between lysine nitrogen (NZ) and glutamate oxygen (OE1) of ~5 Å. Collectively, the QM PDZ/NRXN1 structure showed that the four mutations caused an enlargement of the S₀ and S₋₂ hydrophobic pockets and a favorable electrostatic interaction. Moreover, solution NMR binding studies are consistent with the observations for NRXN1 binding (Figure 3SB and D). Importantly, NMR binding studies showed that the QM PDZ domain had distinct responses upon binding the Caspr4 and NRXN1 peptides, reflecting the unique structural interactions observed in the crystal structures (Figure S3E).

Changes in Fast Timescale Backbone Dynamics of the QM PDZ Domain

NMR-based investigations allow for the study of protein motions over a wide range of timescales (i.e. dynamics) to probe conformational fluctuations that are not readily observed in crystal structures. Here, we employed ¹⁵N-based spin relaxation methods to monitor the changes in backbone amide dynamics on the pico- to nanosecond (ps-ns) timescale of the QM PDZ domain in solution. Longitudinal and transverse ¹⁵N-relaxation times and {¹H}-¹⁵N nuclear Overhauser effects were measured at two magnetic field strengths followed by model-free analysis to yield an order parameter (S²) and timescale of motion (τ_e) for each backbone amide N-H bond vector. S² can range from 0 (completely unrestricted) to 1 (completely rigid), indicating the degree of motional restriction. S² for residues in the free QM PDZ ranged from 0.20 to 0.92 with an average of 0.80. A plot of S² as a function of protein sequence revealed several regions with depressed S² values indicating enhanced mobility (Figure 6A). These regions included the N-, C-termini, β1-β2 loop residues 852–857), β2-β3 loop (residues 869–871) and β4-α2 loop (residues 905–907). Also noteworthy, residues in the β2 strand displayed chemical exchange (R_ex) indicating conformational exchange in the micro- to millisecond (μs-ms) timescale (Figure 6B).

Figure 6. Fast (ps-ns) Timescale Backbone Dynamics of the QM PDZ Domain.

(A–B) The order parameter (S²), timescale of motion (τ_e, ○) and chemical exchange (R_ex, △) of the free QM PDZ domain is plotted against backbone amide residue. The secondary structure is shown on the top of the graph. Regions with enhanced dynamics are shaded in gray.

(C–D) The changes of order parameter (ΔS², ●) and chemical exchange (ΔR_ex, △) in the Tiam1 PDZ (WT) domain caused by the four mutations (C) and those in QM PDZ domain caused by Casrpr4 binding (D). Error bars for each parameter represent the propagated uncertainty determined from Monte Carlo simulations. Symbols for residues that experience significant changes in a particular parameter (> 2-fold the propagated error) are colored black in C and D. An asterisk indicates that the data for either the free or bound state was analyzed using a dynamic model that did not include a R_ex term. See also Figures S4 and S5.

Comparison of the order parameters of apo QM and Tiam1 PDZ domains (ΔS²_QM-WT = S²_QM,apo – S²_WT,apo) identified residues with changes in dynamics (Figure 6C). We deemed the change significant if ΔS²_QM-WT was 2-fold greater than the propagated error. Only a few residues met this criterion and they are located in the β1-β2 loop, the β3-α1 region (residues 877 and 884) and the β4-α2 loop (residues 902, 903 and 910). Interestingly, the ΔS²_QM-WT for many residues was found to be less than 0, signifying that QM PDZ had increased dynamic motions along the backbone in the ps-ns regime (Figure 6C). In addition, residues in the β2 strand from 860 to 865 had significant chemical exchange (ΔR_{ex, QM-WT} >0), indicative of ‘slow’ motions in QM that were absent in the Tiam1 PDZ domain. Finally, analysis of the relaxation data for the free QM PDZ domain gave an overall correlation time (τ_m) of 7.58±0.03 ns, significantly longer than that previously determined for the Tiam1 PDZ (τ_m = 6.38±0.02 ns) (Liu et al., 2013), indicating that the Tiam1 PDZ is more compact when compared to the QM PDZ domain.

Next, we examined the influence of the Caspr4 peptide binding on dynamics. Analysis of the QM PDZ/Caspr4 relaxation data resulted in a τ_m of 6.92±0.03 ns, implying the complex is more compact than the free QM PDZ domain. The plot of order parameter versus sequence upon Caspr4 binding (ΔS²_{QM, Caspr4-Apo} = S²_{QM, Caspr4-bound} – S²_{QM, Apo}) showed residues with changes in dynamics induced by ligand binding (Figure 6D). In particular, ΔS²_{QM, Caspr4-Apo} > 0 revealed QM residues whose dynamic motions were dampened upon Caspr4 binding. This is similar to the previously reported results for the WT PDZ upon Caspr4 binding (Liu et al., 2013). This dampening effect was also seen in the chemical exchange (ΔR_{ex, QM, Caspr4-Apo} < 0), indicating an overall quenching of both fast and slow timescale dynamics in the Caspr4-bound state. Again, this was similar to that observed for the WT PDZ/Caspr4 complex (Liu et al., 2013). Comparison of the backbone dynamics for the WT and QM PDZ/Caspr4 complexes indicates only small changes in both order parameters and R_ex (R_ex < 1 s⁻¹) (Figure S4). Thus, overall the ¹⁵N-spin relaxation data showed that the fast timescale (ps-ns) backbone dynamics of the WT and QM PDZ domains, either free or complexed with Caspr4, were similar having only subtle differences. The most noteworthy finding was that apo QM displayed conformational exchange on the μs-ms timescale, which was not present in either the apo WT or WT/Caspr4 and QM/Caspr4 complexes.

Global Changes in Fast Timescale Side-chain Dynamics of the QM PDZ Domain

Deuterium-based relaxation experiments provide access to fast timescale motions of methyl-bearing side chains. Many studies have shown that side-chain dynamics are more heterogeneous, potentially revealing distinct information not available through the analysis of backbone motions (Wand, 2013). Comparison of the ps-ns dynamic changes of methyl-bearing side chains between the Tiam1 and QM PDZ domains showed a global reduction in S²_axis with the average S²_axis of 0.45 and 0.51 for the for the QM and Tiam1 PDZ domains, respectively (Liu et al., 2013). Closer inspection revealed a near uniform reduction in order parameter (ΔS² _{axis, QM-WT} < 0) for the QM PDZ domain and several residues with modulation of τ_e (Figure 7A–B). Changes in dynamics parameters were deemed significant if they were 2-fold greater than the propagated errors. Residues with changes in S² are mapped onto the structure of the QM PDZ domain (Figure 7C and Table S3). The affected residues were distributed throughout the PDZ domain, both near the site of mutations (β2-α2 region, where the ligand binds) and distally (β3-α1 region and β1-α2 loop). The most pronounced changes in dynamics occurred at A854^β and T857^γ2 in the β1-β2 loop (or carboxylate-binding loop), T881^γ2 in the β3-α1 region, and L873^δ2 in the ligand binding groove.

Figure 7. Fast (ps-ns) Timescale Methyl-bearing Side-chain Dynamics in the QM PDZ Domain.

(A–C) The change in S²_axis and τ_e in the QM PDZ compared to the Tiam1 PDZ (WT).

(D–F) The change in S²_axis and τ_e in the QM PDZ upon binding Caspr4.

Black colored bars indicate residues that experience significant (> 2-fold the propagated error) changes in this parameter. The error bars represent propagated uncertainty as derived from Monte Carlo simulations. Methyl groups exhibiting changes in dynamics are mapped onto structural models (C and F) of the free and Caspr4-bound QM PDZ domains, respectively. The methyl groups (spheres) are colored in a continuous gradient from red to blue, with their intensity scaling to the magnitude of ΔS²_axis. The Caspr4 peptide is shown in cyan. Methyl groups that had a significant Δτ_e but no ΔS²_axis are shown as spheres and colored yellow. Residues Y858, F860, M911, E912, F915 and V920 are shown as sticks and colored yellow. See also Tables S3–S5.

Changes in side-chain dynamics due to Caspr4 binding were also probed. The average S²_axis value for methyl groups in the QM PDZ/Caspr4 complex was very similar to the Tiam1 PDZ/Caspr4 complex (0.53 compared to 0.55, respectively) (Table S3) (Liu et al., 2013). Comparison of the changes in order parameter (ΔS² _{axis, QM/Caspr4-free QM}) and τ_e showed that motions were quenched upon Caspr4 binding – i.e. Caspr4 binding had a global effect on the PDZ dynamics (Figure 7D–F). In addition, the response of the methyl groups to Caspr4 binding in the Tiam1 and QM PDZ domains was very similar. Analysis of a two-way contingency table comparing the number of methyl groups having significant or no change in ΔS²_axis indicated that the two PDZ domains had a similar pattern (Table S4). Taken together, the methyl side-chain dynamics data indicates that the QM PDZ domain is more dynamic than the Tiam1 PDZ domain, but upon binding the peptide ligand the dynamics of the two complexes are indistinguishable.

Slow Conformational Switching in the QM PDZ Domain

Analysis of the QM PDZ ¹⁵N-spin relaxation data indicated that several residues exhibited conformational exchange (R_ex), suggesting slower μs-ms motions (Figure 6B). We sought to rigorously quantify these motions using ¹⁵N-CPMG relaxation dispersion experiments (Loria et al., 1999; Mulder et al., 2001). No significant R_ex (R_ex > 2 s⁻¹) was found in the Tiam1 PDZ domain, free (Figure 8A) or when complexed with Caspr4 (Figure S5A). In contrast, 26 residues had R_ex in the free QM PDZ domain (Figure 8B, Table 2 and Figure S6) that was quenched upon binding Caspr4 (Figure S5B). The free QM PDZ relaxation curves were fit to the general Carver-Richards expression (Carver and Richards, 1972) resulting in a global exchange rate (k_ex = k₁+k₋₁), residue-specific chemical shift change (Δω), and the populations of major and minor states (p_A and p_B, respectively). The data fitting yielded a k_ex = 1082±29 s⁻¹ and p_A = 0.974±0.002 (Table 2). Residues with R_ex were mapped onto the structure of the QM PDZ (Figure 8C). R_ex was distributed mostly around the ligand binding pocket in the α2 helix (e.g. residues 911, 912, 915, 916 and 919) that contained three of the mutated residues (i.e. L911M, K912E and L915F) and the β2 strand (residues 860–865) as initially suggested by the backbone relaxation data. Residues containing R_ex were also found in loop regions, including residues 854 and 857 in the β1-β2 loop, residue 878 in the β3-α1 loop, residue 889 in the α1-β4 loop and residues 904, 906 and 907 in the β4-α2 loop.

Figure 8. Slow (μs-ms) Timescale Motions in the QM PDZ Domain.

(A–B) Representative CPMG relaxation dispersion curves are shown for the Tiam1 and QM PDZ domains, respectively. Individual curves for each residue with R_ex are shown in Figure S6, while their fitted parameters are indicated in Table 2. Data collected at 800 MHz (closed circle) and 500 MHz (open circle) are shown. Error bars were determined by the analysis of peak intensities from duplicate experiments.

(C) Residues with R_ex in the QM PDZ domain are labeled and colored in red. The six residues shown in Figure 1 have their side chains displayed. Those colored yellow did not have R_ex. See also Figures S4 and S5.

Table 2.

Fitted Parameters of ¹⁵N-CPMG Relaxation Dispersion Curves for the QM PDZ Domain^a

Residue	Δω (ppm)	$R_{2B}^0$ (s−1) (500MHz)	$R_{2B}^0$ (s−1) (800MHz)
843	0.841 ± 0.037	11.76 ± 0.07	14.05 ± 0.07
848	0.969 ± 0.045	12.76 ± 0.10	15.29 ± 0.10
851	1.624 ± 0.092	14.04 ± 0.15	18.92 ± 0.22
854	0.685 ± 0.045	12.76 ± 0.11	14.89 ± 0.10
857	1.367 ± 0.078	11.95 ± 0.16	15.05 ± 0.21
860	1.888 ± 0.194	20.08 ± 0.56	29.39 ± 0.86
861	2.110 ± 0.166	17.04 ± 0.32	25.79 ± 0.53
863	1.788 ± 0.105	13.63 ± 0.16	17.20 ± 0.21
865	1.559 ± 0.080	13.50 ± 0.12	16.78 ± 0.15
872	1.236 ± 0.059	12.89 ± 0.10	15.50 ± 0.11
875	1.026 ± 0.048	14.12 ± 0.12	16.27 ± 0.12
876	1.305 ± 0.069	13.82 ± 0.16	17.46 ± 0.19
878	2.899 ± 0.213	15.47 ± 0.28	24.91 ± 0.37
879	1.473 ± 0.073	13.60 ± 0.14	16.65 ± 0.18
883	1.517 ± 0.076	14.98 ± 0.13	20.32 ± 0.17
889	0.793 ± 0.037	13.02 ± 0.07	15.19 ± 0.07
898	0.692 ± 0.032	12.34 ± 0.07	14.63 ± 0.07
904	0.710 ± 0.028	11.20 ± 0.04	13.79 ± 0.05
906	0.709 ± 0.028	10.47 ± 0.05	12.48 ± 0.04
907	1.107 ± 0.043	10.63 ± 0.08	13.46 ± 0.09
911	0.841 ± 0.034	11.78 ± 0.06	14.25 ± 0.06
912	0.970 ± 0.044	11.65 ± 0.08	14.21 ± 0.11
915	1.657 ± 0.079	13.17 ± 0.13	18.46 ± 0.18
916	0.962 ± 0.043	11.73 ± 0.07	14.08 ± 0.07
919	0.865 ± 0.034	11.61 ± 0.05	14.31 ± 0.06
924	1.143 ± 0.052	13.45 ± 0.11	15.01 ± 0.11

^aResults from global fitting of data with k_ex = 1082 ± 29 s⁻¹ and p_A = 0.974 ± 0.002. The error for each parameter was estimated from Monte Carlo simulations.

DISCUSSION

Remodeled Binding Pockets and Novel Interactions Contribute to QM PDZ Specificity

Comparison of the QM PDZ domain structures complexed with the NRXN1- and Caspr4-bound peptides showed that the four mutations in QM combined to remodel the S₀ and S_-2 pockets to support novel interactions that contribute to specificity. The QM PDZ/NRXN1 structure (Figure 5C) showed an expansion of the S₀ pocket, consistent with biochemical data indicating a preference for amino acids with larger side chains at the P₀ position of the peptide (Figure 3). Similarly, the QM PDZ/Caspr4 structure showed that substitutions in the S₀ pocket supported new interactions not previously available. In particular, the L915F substitution was involved in Π-Π stacking and an anion-Π interaction (Figure 5B).

The QM PDZ structures also revealed new interactions mediated by the substitutions in the S₋₂ pocket. In the QM PDZ/NRXN1complex, an electrostatic interaction was identified between E912 and Lys at P₋₄ (Figure 5C). Consistent with the structure, previous binding data showed that this interaction is an important determinant for QM PDZ/NRXN1 interactions as the L911M/K912E variant in the S₋₂ pocket accounted for ~1.3 kcal/mol of favorable binding energy (Shepherd et al., 2011). This charge-pair interaction also played a critical role in determining the specificity of the QM PDZ for other peptides (Figure 3) as was previously seen for Tiam1 PDZ domain towards SDC peptides (Liu et al., 2013). Here, the QM PDZ preference for SDC peptides was reversed such that the SDC2 isoform bound tighter (with the K at P₋₄) than the SDC1 and SDC3 (with the E at P₋₄) peptides. These data clearly show that electrostatics play a key role in defining PDZ binding and that manipulation of residues in the S₋₂ pocket can modulate specificity.

The QM PDZ/Caspr4 structure provided insight into how substitutions in the S₀ and S₋₂ pockets can synergize to fine tune specificity. In particular, the structure revealed an anion-Π interaction between the E912 side chain and Phe at P₀ in the S₀ pocket and a sulphur-Π interaction mediated by the M911 side chain in the S₋₂ pocket. Importantly, this network of interactions connected the peptide ligand to both the S₀ and S₋₂ pockets, suggesting an important role for the anion-Π interaction in Caspr4 binding. Indeed, anion-Π interactions can stabilize binding interactions ~2 kcal/mol (Philip et al., 2011). Our binding data support the importance of this interaction, as the Caspr4 P₀ F→A mutant, which disrupts anion-Π and Π-Π stacking interactions, weakened the binding affinity ~9-fold (~0.6 kcal/mol) relative to Caspr4 (Figure 3). The QM PDZ/Caspr4 structure also indicated that a sulphur-Π interaction occurred between the methionine sulfur atom of M911 and the phenyl ring of Tyr at P₋₂ (Figure 5B). Methionine-aromatic pairs are prevalent in one-third of all known structures and often found to stabilize protein/ligand interactions (Valley et al., 2012). However, in the context of the QM PDZ/Caspr4 complex it appears that this interaction is of only modest importance, as the L911M mutant provided only ~0.2 kcal/mol of stability to the Caspr4 interaction (Shepherd et al., 2011).

The interactions described above provide a structural role for how the four substitutions in the QM PDZ domain cooperate to bind the Caspr4 peptide ligand. Our previous binding data are consistent with the structural data but indicate that the energetics of individual interactions alone is not sufficient to explain overall binding (Shepherd et al., 2011). For instance, the Π-Π stacking interactions mediated by F915 (in both the L915F and L915F/L920V substitutions) destabilize binding to Caspr4 by ~0.7 kcal/mol, while the K912E and the L911M/K912E substitutions were also destabilizing to Caspr4 binding by ~0.7 and ~0.2 kcal/mol, respectively. Yet, together the four mutations (QM) marginally stabilized binding (−0.02 kcal/mol) relative to the Tiam1 PDZ domain. Thus, Caspr4 binding to the QM PDZ domain results from structural cooperativity between the four mutations, a result that a priori would be very difficult to predict. Collectively, these results reinforce the role of epistasis in PDZ domain interactions and highlight the difficulty in predicting (and designing) protein-protein interactions.

Additional insight into the energetics of Caspr4 binding to the QM PDZ domain comes from ITC data (Figure 4). Consistent with the biochemical binding data and the structure, the ITC experiments indicated that the binding of QM PDZ to Caspr4 was an enthalpically driven process while binding to the Tiam1 PDZ was entropically driven. This classic enthalpy-entropy compensation is consistent with the new interactions revealed in the QM PDZ/Caspr4 structure but also indicates that entropy can play an important role in tuning affinity.

Changes in Fast Methyl Side-chain Dynamics Correlate with Changes in Conformational Entropy of Caspr4 Binding

Although the structures of the QM PDZ/NRXN1 and QM PDZ/Caspr4 help rationalize the switched specificity found in the QM, NMR-based studies suggested that fast timescale (ps-ns) dynamics also play a role. In particular, we found enhanced fast timescale motions in the backbone and methyl groups throughout the QM PDZ domain (Figures 6 and 7). Furthermore, the methyl side-chain motions in both apo Tiam1 and QM PDZ domains were dampened by Caspr4 binding (Figure 7) and essentially indistinguishable (Table S3). Thus, despite the differences in the Tiam1 and QM PDZ apo state dynamics there was a high degree of similarity in the dynamics once bound to Caspr4. Together, these results suggest that the increase in dynamics in the apo QM state might modulate conformational entropy, and thus the energetics of binding (Li et al., 1996; Yang and Kay, 1996). Indeed, multiple studies have shown a remarkable correlation between the entropy of binding obtained from ITC experiments and the conformational entropy derived from methyl side-chain order parameters (Kasinath et al., 2013; Marlow et al., 2010; Tzeng and Kalodimos, 2012). Thus, the pronounced changes in fast timescale dynamic motions found in the Tiam1 and QM PDZ domains may reflect different entropic contributions to the Caspr4 binding processes as seen in the ITC data (Figure 4).

To test this hypothesis, we sought to estimate the change in entropy between the Tiam1- and QM-bound Caspr4 complexes seen in the ITC data (Figure 4B). In principle, the entropy change could arise from a number of factors, including differences in solvent or conformational entropy. If the entropy from solvent release upon ligand binding is assumed to be similar in both PDZ/Caspr4 complexes, which is reasonable given that both the ligand and protein are nearly identical, then the change in entropy may be due to conformational entropy. Estimates of the entropy of solvation (ΔS_sol) calculated based on the surface area buried by the peptide (Hilser et al., 2006) using the QM PDZ/Caspr4 crystal structure and a homology model of the Tiam1 PDZ/Caspr4 complex were very similar, reinforcing this assumption (Table S5). Conformational entropy was estimated using the approach by Wand and colleagues that empirically relates ΔS²_axis to conformational entropy (ΔS_conf) (Kasinath et al., 2013). Using this approach, –TΔS_conf,WT and –TΔS_conf,QM were determined to be +3.34 and +7.26 kcal/mol, respectively (Tables S3 and S5). Remarkably, calculation of the change in –TΔS_tot,WT-QM from estimates of solvent and conformational entropy yielded a value of −4.54 kcal/mol, which is within 10% of value obtained by ITC (−4.13 kcal/mol) (Figure 4B and Table S5). Furthermore, the contribution from solvent was small, indicating that the change in binding entropy between the Tiam1 and QM PDZ Caspr4 complexes can be attributed almost entirely to a change in conformational entropy. Importantly, this result originates from the enhanced ps-ns motions in the apo QM, as the Tiam1- and QM-Caspr4 complexes have virtually identical dynamics. These results along with published data on calmodulin (Frederick et al., 2007; Marlow et al., 2010) and CAP (Tzeng and Kalodimos, 2012) provide compelling evidence in support of a general role for fast timescale dynamics in tuning the entropic contribution of ligand affinity. Moreover, we note that the origin of this effect is in side chain dynamics rather than the backbone as seen in several other studies (Marlow et al., 2010; Tzeng and Kalodimos, 2009). Together, our data suggest that fast timescale dynamics play a role in dictating affinity as the QM PDZ displayed enhanced ps-ns dynamics that correlated with changes in binding thermodynamics.

Slow Conformational Switching Correlates with Specificity Changes in the QM PDZ Domain

The four mutations introduced into the QM PDZ domain induced slow (μs-ms) timescale motions along the backbone. The 26 residues with μs-ms motions (R_ex) mapped primarily to the α2 helix and β1-β2 and β2-β3 loops (i.e. the ligand binding groove) with the only exception being T843, L889 and V924 located in β1, α2/β4 loop and β5, respectively (Figure 8C). Interestingly, these three residues connect to the binding site via sequential van der Waals interactions to residue F860 in the binding site. These results are similar to the slow timescale dynamics found in the ligand-free AF-6 and Par6 PDZ domains. In the AF-6 PDZ domain, there were fewer (~16) residues with conformational exchange and these sites were limited to the β2 and α2 regions of the binding groove suggesting a role in specificity (Niu et al., 2007). In contrast, both the Par6 PDZ and the CRIB-PDZ bi-domain showed extensive slow conformational exchange (32 and 47 residues, respectively). Moreover, this conformational exchange was linked to the local unfolding of the β1-β2 and β2-β3 loops to promote access of the so-called L/K switch to its high-affinity state for binding to target peptides (Whitney et al., 2013; Whitney et al., 2011). These authors went on to argue that this local unfolding event is a key aspect of Cdc42 GTPase regulation (through CRIB binding) of PDZ binding affinity. The QM PDZ domain shares similar features found in both AF-6 and the Par6, in particular, conformational exchange in the ligand-binding site near the β2-β3 loop and α2 helix. Similar to Par6, we propose that the four mutations in the QM PDZ domain induce a transient, local unfolding event at the ligand-binding site that occurs on the μs-ms timescale. In turn, this conformational flexibility provides the plasticity to assume multiple conformations leading to broadened ligand specificity. In principle, this notion is consistent with the conformational selection; however the induced fit model cannot be ruled out without additional kinetics experiments (Chakrabarti et al., 2016; Gianni et al., 2014; Hammes et al., 2009; Vogt et al., 2014; Weikl and Paul, 2014). More broadly, our data support a functional role for μs-ms conformational dynamics in modulating specificity.

It is now appreciated that both increased flexibility and lowered stability can contribute to the evolution of protein function (Bloom et al., 2006; Tokuriki and Tawfik, 2009). Here, we show that the QM PDZ domain variant has the hallmark of a protein with “evolved” function (i.e. novel specificity) – the QM has increased flexibility on both fast and slow timescales and lowered stability coincident with a broadened specificity towards peptide ligands. These data provide insight into the potential mechanism by which these two features couple to support protein evolution. In particular, lowered stability – whether global or local – supports access to a wide range of conformational substates that facilitate populating conformations relevant for novel specificity (Gonzalez et al., 2016). Thus, the QM variant might be regarded as a key intermediate in the evolutionary trajectory between the Tiam1 and Tiam2 PDZ domains. This notion is supported by our previous analysis of sequence conservation of Tiam-family PDZ domains that showed the identity of the four residues in the QM segregate into four distinct Tiam subfamilies with Tiam1 and Tiam2 at the extremes (Shepherd et al., 2011). The onset of large-scale evolutionary datasets (Aakre et al., 2015; Raman et al., 2016) should provide new and interesting opportunities to probe the role of conformational dynamics in molecular evolution.

Conclusions

Conformational dynamics has been established in enzyme function but its role in molecular recognition is controversial. We present a study describing the structural and dynamic origin for the engineered specificity of the Tiam1 PDZ domain harboring four mutations in the ligand binding pocket. Crystal structures of the QM PDZ alone and in complex with ligands provided insight into the enthalpic component of the interaction, revealing that specificity was obtained, in part, through expansion of the S₀ and S₋₂ binding pockets and the acquisition of several new interactions (e.g. Π-Π, anion-Π and salt bridges). Importantly, NMR relaxation experiments indicated that conformational dynamics also plays a fundamental role in molecular recognition. Fast (ps-ns) timescale dynamics contributed to the thermodynamics (conformational entropy) of the interaction, while slower (μs-ms) dynamics were critical for specificity (excited-state conformations). Together, our results provide an unprecedented view for how structure and conformational dynamics combine to govern the thermodynamics and specificity of molecular recognition. Finally, on a practical level, our data have implications for protein design where conformational dynamics should be an important factor to consider when designing novel protein-protein interactions or identifying cryptic sites to be targeted by small-molecule modulators.

EXPERIMENTAL PROCEDURES

Protein expression and purification

The Tiam1 and QM PDZ proteins were expressed and purified as previously described (Shepherd et al., 2011). ¹⁵N and ¹⁵N, ¹³C uniform isotopic labeling of the proteins was achieved by growing cells in M9 minimal media that contained ¹⁵NH₄Cl (99%) and D-glucose (U-¹³C-99%). Minimal media containing ¹⁵NH₄Cl (99%), D-glucose (U-¹³C-99%) and 60% ²H₂O was used to produce random, fractionally labeled ²H-methyl proteins.

In vitro binding measurements

Fluorescence anisotropy binding experiments were performed as previously reported (Shepherd and Fuentes, 2011). Additional details of these procedures can be found in the Supplemental Experimental Procedures.

Isothermal titration calorimetry

Isothermal titration experiments were performed as previously reported (Liu et al., 2013). Additional details of these procedures can be found in the Supplemental Experimental Procedures.

Crystallization and data collection

The details of crystallization and data collection of the apo QM PDZ domain and complexes bound to NRXN1 and Caspr4 peptides are provided in the Supplemental Experimental Procedures.

Structure determination and refinement

Detailed procedures for crystal structure determination and refinement are provided in the Supplemental Experimental Procedures.

NMR spectroscopy

NMR experiments were carried out at 298 K (calibrated with methanol) on Bruker Avance II 500 (RT probe), 800 MHz (cryoprobe) and Varian Inova 600 (RT probe) MHz spectrometers equipped with ¹H/¹⁵N/¹³C probes and z-axis pulsed-field gradients. Chemical shifts for non-proline backbone residues, side-chain methyl groups and prochiral methyl groups were assigned by 3D triple resonance experiments as previously described (Liu et al., 2013). NMR titration experiments were performed in phosphate buffer (20 mM NaPO₄, 50 mM NaCl, pH 6.8) containing 0.5 mM ¹⁵N-labeled QM PDZ. A series of 2D ¹H-¹⁵N-HSQC spectra were recorded with an increasing amount of concentrated Caspr4 or NRXN1 ligand added in eight steps until the final molar ratio of QM PDZ to ligand was 1:5. All NMR data was processed using NMRPipe (Delaglio et al., 1995) and analyzed using NMRView (Johnson and Blevins, 1994).

The QM PDZ/Caspr4 samples for fast timescale dynamics experiments were prepared by adding small amounts of concentrated peptide to 1 mM PDZ domain until saturation (the final molar ratio of PDZ to ligand was 1:5). The complex was lyophilized and resuspended in 90% H₂O/10% D₂O prior to NMR analysis. Backbone ¹⁵N T₁, T₂ and {¹H}-¹⁵N NOE data and side-chain ²H T₁ and T_1ρ data at 500 and 600 MHz for free and Caspr4-bound PDZ QM were collected using standard (¹⁵N) and ²H-methyl relaxation experiments as previously described (Liu et al., 2013). For each experiment, nine relaxation time points and three duplicates were collected. Relaxation rate constants (R₁ or R₂) were best fit to a single exponential function using in-house programs.

¹⁵N relaxation dispersion experiments were performed at 500 and 800 MHz using a relaxation-compensated CPMG experiment (Loria et al., 1999; Mulder et al., 2001). The total relaxation time was 60 ms in the CPMG train and the effective field strength was modulated by altering the delay between CPMG pulses. A total of twelve experiments and two duplicates were collected in an interleaved manner at each magnetic field with a delay time ranging from 0.68 to 15 ms. A reference experiment without relaxation delay was collected to calculate the R_2,eff values.

Relaxation analysis

The Lipari-Szabo model free approach was used to characterize the backbone dynamics in the pico- to nanosecond timescale (Lipari and Szabo, 1982a, b). Global correlation times (τ_m) were determined to be 7.58±0.03 and 6.92±0.03 ns for the free QM PDZ and the Caspr4-bound complexes. Backbone dynamic parameters were fit to the five standard models using FAST-Modelfree (Cole and Loria, 2003), assuming a ¹H-¹⁵N bond distance of 1.02 Å and ¹⁵N chemical shift anisotropy of −170 ppm. The Akaike’s information criterion (Chen et al., 2004) was used for model selection and gave similar results as FAST-Modelfree. In all, 81/88 and 82/88 non-proline amides were analyzed for the free and Caspr4-bound QM PDZ, respectively. The program Relxn2.2 (Lee et al., 1999) was used for the analysis of side chain methyl groups motions. Errors in the fitted parameters were estimated using Monte Carlo simulations. Out of 56 methyl groups, 54 and 48 were analyzed respectively for the free QM PDZ and the Caspr4-bound complex to obtain S²_axis and τ_e.

Peak intensities for relaxation dispersion experiments were obtained using the program SPARKY (Goddard and Kneller, 2007). Residues with resonance overlap or weak intensities were not analyzed. In all, 73/88 and 72/88 non-proline amides were analyzed for the Tiam1 and QM PDZ domains, respectively. R_2,eff values at each field (500 and 800 MHz) were calculated by equation(1) (Mulder et al., 2001),

$$R_{2,\mathit{eff}}=\frac{-1}{T}ln(\frac{I_{\mathit{cpmg}}}{I_0})$$ (1)

where T is the total relaxation time, I_cpmg and I₀ are peak intensities from spectra at each relaxation delay and reference, respectively. Only residues with a change in R_2,eff value greater than 2 Hz over the series of effective field strengths were further analyzed. The intensities were best-fit to both a simple two-state model and a model with no exchange. An F-test (α_critical = 0.01) was used to identify residues with statistically significant chemical exchange. Finally, exchange parameters were determined by fitting the data to the Carver-Richards expression (2) (Carver and Richards, 1972),

$$\begin{array}{c}{R}_{2,\mathit{eff}}(1/{\tau }_{cp})=\frac{1}{2}(R_{2A}^0+R_{2B}^0+k_{ex}-\frac{1}{{\tau }_{cp}}{\mathit{cosh}}^{-1}[D_{+}cosh({\eta }_{+})-D_{-}cos({\eta }_{-})])\\ D_{\pm }=\frac{1}{2}[\pm 1+\frac{\mathrm{\Psi }+2\mathrm{\Delta }{\omega }^2}{{({\mathrm{\Psi }}^2+{\zeta }^2)}^{1/2}}]\\ {\eta }_{\pm }=\left(\frac{{\tau }_{cp}}{\sqrt{2}}\right){[\pm \mathrm{\Psi }+{({\mathrm{\Psi }}^2+{\zeta }^2)}^{1/2}]}^{1/2}\\ \mathrm{\Psi }={(R_{2A}^0+R_{2A}^0+p_A{k}_{ex}+p_B{k}_{ex})}^2-\mathrm{\Delta }{\omega }^2+4p_A{p}_B{k}_{ex}^2\\ \zeta =2\mathrm{\Delta }\omega (R_{2A}^0+R_{2B}^0-p_A{k}_{ex}+p_B{k}_{ex})\end{array}$$ (2)

where p_A and p_B are the populations of the major and minor states,Δω and k_ex are the difference in chemical shift and the rate of exchange between the two states, respectively (Carver and Richards, 1972; Palmer et al., 2001). Fits were performed using the program exrate2.0 (Mauldin et al., 2009). Errors in the fitted parameters were determined by Monte Carlo simulations. Comparison of the Chi-squared (χ²) values of individual residue (local) and all residue (global) fits indicated that a global exchange process best fit the data. The fitted global (k_ex and p_A) and individual (Δω and $R_{2B}^0$) parameters are reported in Table 2.

Accession numbers

The atomic coordinates and structure factor amplitudes for the free QM PDZ, QM PDZ/Caspr4, and QM PDZ/NRXN1 structures have been deposited in the PDB with accession codes 4NXP, 4NXQ and 4NXR, respectively.

Highlights.

• The QM PDZ/Caspr4 structure reveals Π– Π and Π-anion interactions

• Remodeled binding pocket and electrostatics explain the QM PDZ/NRXN1 affinity

• Conformational entropy from NMR dynamics correlates with binding entropy from ITC

• Slow, μs-ms motions suggest a conformational selection model of ligand recognition