Ligand-induced changes in dynamics in the RT loop of the C-terminal SH3 domain of Sem-5 indicate cooperative conformational coupling

Abstract

We report the effects of peptide binding on the ¹⁵N relaxation rates and chemical shifts of the C-SH3 of Sem-5. ¹⁵N spin-lattice relaxation time (T₁), spin-spin relaxation time (T₂), and {¹H}-¹⁵N NOE were obtained from heteronuclear 2D NMR experiments. These parameters were then analyzed using the Lipari-Szabo model free formalism to obtain parameters that describe the internal motions of the protein. High-order parameters (S² > 0.8) are found in elements of regular secondary structure, whereas some residues in the loop regions show relatively low-order parameters, notably the RT loop. Peptide binding is characterized by a significant decrease in the ¹⁵N relaxation in the RT loop. Concomitant with the change in dynamics is a cooperative change in chemical shifts. The agreement between the binding constants calculated from chemical shift differences and that obtained from ITC indicates that the binding of Sem-5 C-SH3 to its putative peptide ligand is coupled to a cooperative conformational change in which a portion of the binding site undergoes a significant reduction in conformational heterogeneity.

Molecular recognition remains one of the most challenging problems in structural biology, and a complete understanding of this phenomenon can only come by elucidating the complex interplay between structure, dynamics, and energetics (Weber 1992). Recently we solved the solution structure of the unliganded C-SH3 of Sem-5 (Ferreon et al. 2003), which when compared with the previously solved liganded structure (Lim et al. 1994b) revealed several significant differences in the RT loop, a region that makes direct contact with the ligand. In addition, as with other SH3 domains (Wang et al. 2001), ¹⁵N relaxation and hydrogen deuterium exchange reveal that the RT loop possesses significant conformational heterogeneity, indicating that consideration of the static structural differences alone may not be sufficient to provide a detailed accounting of the energetic contributions to molecular recognition.

In order to determine the quantitative role of dynamics in the recognition process, it is essential to determine not only the average structure of the bound and free forms of the protein, but also the magnitude and extent of the conformational fluctuations for each state. In ensemble terms, a quantitative characterization of the binding process requires knowledge of the structure and energetics of each of the microscopic states within the macroscopic bound and unbound ensembles.

Motions on a variety of timescales from femtoseconds to minutes have been shown to be important in enzyme catalysis, binding specificity, and regulation (Collins et al. 1995; Nicholson et al. 1995; Wagner 1995; Stivers et al. 1996; Gagné et al. 1998; Kay et al. 1998; Feher and Cavanagh 1999; Stock 1999), and NMR spectroscopy has emerged as a powerful tool to characterize these motions. Of particular importance is the use of ¹⁵N relaxation rate measurements to obtain information about picosecond to nanosecond dynamics (Kay et al. 1989; Clore et al. 1990; Palmer III et al. 1991). On application of the appropriate motional models (Lipari and Szabo 1982a,b; Mandel et al. 1995), relaxation measurements provide access to residue-specific order parameters, which can in turn be related to the entropy changes in the system (Akke et al. 1993; Yang and Kay 1996).

It has been shown that ligand-induced changes in backbone dynamics can be used to obtain unique insights into the role of dynamics in binding affinity and specificity (Nicholson et al. 1995; Kay et al. 1996, 1998; Spyracopoulos et al. 1998; Zidek et al. 1999; Wang et al. 2001). Several studies have shown an increase in backbone rigidity on ligand binding (Akke et al. 1993; Hodsdon and Cistola 1997; Olejniczak et al. 1997; Alexandrescu et al. 1998; Kristensen et al. 2000), whereas others have shown an increase in disorder on binding (Farrow et al. 1994; Stivers et al. 1996; Zidek et al. 1999). Here we investigate the changes in residue-specific ¹⁵N relaxation rates in Sem-5 C-SH3 domain on ligand binding. We correlate these changes with the observed cooperativity of the binding process. We show that consideration of the location, magnitude, and cooperativity of the dynamics changes provides unique insights into the thermodynamic origins of the dynamic contributions.

The model system used in these studies is the C-terminal SH3 of the protein Sem-5. SH3 domains are adaptor proteins that recognize proline-rich sequences and are found in many signal transduction and cytoskeletal proteins (Pawson 1995; Dalgarno et al. 1997; Buday 1999). In particular, the C-SH3 of Sem-5 from Caenorhabditis elegans and its mammalian and Drosophila homologs, Grb2 and Drk, are involved in linking protein tyrosine kinase activity to Ras. Sem-5 activates Ras signaling by binding to the PPPXPPR motifs found in the guanine nucleotide exchange factor Sos (Clark et al. 1992; Lowenstein et al. 1992; Egan et al. 1993; Olivier et al. 1993).

Results

Identification of the residues affected by ligand binding

Previously, we have determined the chemical shifts for the unliganded Sem-5 C-SH3 domain (Ferreon et al. 2003). Backbone amide ¹⁵N and ¹H were assigned using various homonuclear and heteronuclear NMR techniques. In the past, changes in chemical shifts observed on ligand binding have been used to identify residues that are involved in the binding interface for different SH3 domains (Booker et al. 1993; Wittekind et al. 1994; Horita et al. 1998). For Sem-5 C-SH3, chemical shifts in the ¹H-¹⁵N HSQC spectra were obtained in various saturating and subsaturating concentrations (15 concentrations) of the putative peptide ligand. Figure 1 ▶ shows the ¹H-¹⁵N HSQC spectra of both unbound (green) and bound Sem-5 C-SH3 domain (red). Although many resonances are not affected, residues near the binding interface (Yu et al. 1992; Booker et al. 1993; Lim et al. 1994b) show significant deviation (>0.1 ppm for ¹H and >0.5 ppm for the ¹⁵N dimension; Fig. 2A,B ▶). These residues are found primarily in the conserved 3₁₀ helix and the n-Src and RT loops. The line widths of the amide resonances do not change significantly throughout the titration, indicating that the complex is in fast exchange on the chemical shift NMR timescale (Lian and Roberts 1993). Such conditions enable calculation of the dissociation constant (K_d) assuming two states, unbound and bound (Lian and Roberts 1993; Horita et al. 1998). A nonlinear regression analysis (see Materials and Methods) of chemical shift changes as a function of peptide concentration yields K_d values for all residues showing significant chemical shifts (Fig. 2C,D ▶). The residues considered are F163, N166, E169, S170, L173, F175, W191, W192, F203, S205, and N206. The K_d values range from 286–430 μM with the average and standard deviation of 377 ± 38 μM. The associated error from the regression analysis is 5%–13%. The similarity in calculated equilibrium constants obtained from different chemical shift perturbations indicates that each of the perturbed residues is monitoring the same equilibrium and that the binding can be approximated as a two-state process. It should be noted and will be discussed below that several residues in the RT loop show significant chemical shift differences despite the fact that they are not directly involved in binding.

Figure 1.

Assigned ¹H-¹⁵N HSQC spectra of Sem-5 C-SH3 without peptide (green) and with peptide (red), taken using a 750-MHz Varian spectrometer at 25°C.

Figure 2.

(A) The magnitude of differences in ¹H (top graph) and ¹⁵N (bottom) chemical shifts for each residue between the free and complexed state. (B) Color mapping of the ¹H chemical shift difference in the NMR minimized average structure with the peptide overlayed in the minus orientation and the same position as seen in the X-ray structure. (C and D) Binding curves of chemical shift changes for sample of residues (Ser170, 362 ± 46 μM and Asn206H₂N^e, 427 ± 53 μM in ¹⁵N dimension; Ser205, 397 ± 36 μM and Leu173, 355 ± 34 μM in ¹H dimension) as a function of peptide concentration. Lines are drawn as the best-fitted lines as calculated (see text). T156 and Q160 are residues not involved in binding and do not show chemical shift changes during the titration of the ligand.

Energetics of binding

The binding affinity of Sem-5 C-SH3 domain to polyproline peptide was also assessed using ITC under similar conditions as those used for the chemical shift experiments (Fig. 3 ▶). Analysis of the fitted parameters reveals the binding between Sem-5 C-SH3 and the Sos peptide occurs in a 1:1 stoichiometry with a favorable enthalpy of interaction (ΔH°~−4.9 kcal mole⁻¹), a slightly unfavorable entropy of interaction (ΔS°~−0.33 cal deg⁻¹ mole⁻¹), and a dissociation constant of 337 μM at pH 4.8. This value is in agreement with that obtained from NMR analysis of the chemical shifts (Fig. 2 ▶) and is approximately 10-fold less than the binding constant obtained at pH 7.3 (Lim et al. 1994a) and 7.5 (data not shown). The agreement (within experimental error) between the calorimetrically obtained binding affinity and that obtained from the chemical shifts data indicates that both techniques are monitoring the same cooperative two-state process.

Figure 3.

Calorimetric titration of Sem-5 C-SH3 domain with the Sos peptide (Ac-VPPPVPPRRR-NH₂) in 50 mM NaOAc and 100 mM NaCl, and 10 mM CaCl₂ (pH 4.8) at 25°C.

Relaxation data analysis

¹⁵N relaxation data were obtained at 25°C and 400 MHz field strength, both for the free and complex C-SH3 domain. All of ¹⁵N−T₁ and ¹⁵N−T₂ were adequately fit to single exponential two-parameter decays. Figure 4 ▶ shows the representative best and worst ¹⁵N−T₁ and T₂ fits for the free C-SH3 domain. Best fits have standard errors <1% whereas worst fits have standard errors <2%. Forty-seven and 52 resonances (of 58 with 4 prolines) were determined for the unliganded and liganded forms, respectively. Data were unavailable for the remaining residues because resonances were either highly overlapped or too weak for reliable analysis. The average value of R₁ is 2.84 ± 0.29 s⁻¹, R₂ is 6.30 ± 0.86 s⁻¹, and NOE is 0.53 ± 0.39 for the free C-SH3, and for the complex C-SH3 the average value of R₁ is 2.85 ± 0.27 s⁻¹, R₂ is 6.50 ± 0.84 s⁻¹, and NOE is 0.65 ± 0.30. Figure 5 ▶ shows the relaxation rates R₁ and R₂, and NOE profiles for both the free and the bound C-SH3 domain. In both cases, the R₁ profiles show similar distributions, with significantly lower than average R₁ values in the loop regions (RT loop, n-Src, and distal loop), as well as residues in the termini. Slight increases in R₁ are observed for most residues on binding. Although several residues show a slight decrease in R₁, these changes are generally within error. The R₂ and NOE distribution also show similar profiles, with most residues in the loop regions exhibiting lower values. Significant increases occur for almost all residues in R₂ and NOE on ligand binding.

Figure 4.

Representatives of the worst (L162, E169, E155, G201) and the best (N206, E172, W191, D187) fits of ¹⁵N−T₁ (A) and T₂ (B). Fitting errors are <2% and <1% for the worst and best fits, respectively. Duplicate points are taken for each time delay and shown in the same graph.

Figure 5.

¹⁵N-relaxation data for Sem-5 C-SH3 domain both in the unliganded (open circles) and liganded state (filled circles). Elements of secondary structure are the same for both states. β-Sheets are represented as arrows, 3₁₀ helix by a cylinder, loops by straight black lines, and turns by bent lines.

Isotropic diffusion tensor

The degree of anisotropy is important in choosing the appropriate spectral density functions, as a slight degree of anisotropy could introduce potential errors in describing the appropriate motional model for each amide (Tjandra et al. 1995). Using the average NMR solution structure of the unbound state (Ferreon et al. 2003), the principal axes of the inertia tensor were found to be 1.00:0.91:0.67 using the program pdbinertia (Dr. Arthur Palmer, Columbia University). For the bound state, the crystal structure was used and the relative moments are 1:00:0.94:0.75. The r2r1_diffusion v.1.11 program (Dr. Arthur Palmer, Columbia University) was used to analyze the degree of rotational diffusion anisotropy.

Table 1 shows the summary of the rotational diffusion anisotropy. The diffusion tensor was also fitted to two models, isotropic and axially symmetric. Under the axial symmetric model, D_|/D_⊥ (D_| and D_⊥ are the principal components of the diffusion tensor along and perpendicular to, respectively, the long axis of rotation) is 0.89 and 0.96 (oblate) for free and bound states, respectively. A statistical test shows that Sem-5 C-SH3 domain can be approximated by an isotropic diffusion tensor (Tjandra et al. 1995). Initial estimates of the overall rotational correlation time (τ_m) were obtained by two methods (see Materials and Methods). Both gave the same result of ~5 nsec. Initial τ_m obtained from tmest program was chosen to be 5.091 ns for the unbound and 5.198 ns for the bound SH3 domain based on the procedure by Kay et al. (1989), wherein the initial estimate was obtained from the trimmed R₁/R₂ ratios. Residues that show conformational exchange contributions to R₂, and which have internal motions faster than 100 psec, are not included in obtaining the initial estimate of τ_m.

Table 1.

Rotational diffusion anisotropy of Sem-5 C-SH3 domain

	C-SH3 (free)a		C-SH3: Sos complex (bound)b
Parameter	Isotropic	Axially symmetric	Isotropic	Axially symmetric
θc		1.62		−0.526
φc		6.53		2.10
	1	0.891	1	0.957
χ2e	102.70	64.61	426.50	398.05
χ2 (red)f	3.67	2.58	10.66	10.76
Fg		4.91		0.88

^a NMR minimized average structure.

^b Crystal structure (PDB entry 1 SEM).

^c Orientation angles of the diffusion tensor.

^d Defined as 2 D_z/(D_x + D_y).

^e χ-square value (χ² = Σ[R_1e/R_2e − R_1c/R_2c]²/σ²_R1/R2).

^f Reduced χ-square value (χ² [red] = χ²/[N-m]).

^gF-test to test the significance of addition of extra variables to the fit.

A recent study has shown that removing the extreme values can bias the results toward isotropic tumbling (Pawley et al. 2001). Using the filtering procedure presented in that work resulted in the elimination of similar residues from the analysis to those eliminated as described earlier, confirming that our data can be approximated by an isotropic tumbling analysis. Final optimization yielded τ_m of 5.019 and 5.091 nsec for the free and bound states, respectively. It is noteworthy that the difference in τ_m observed here (~0.1 nsec) is less than the value determined for other SH3 domains (0.4 nsec; Pawley et al. 2001). Although we have no unequivocal explanation for this difference, it is likely that the spherical approximation of the model is a contributing factor. Nonetheless, iterative optimization using different values for τ_m did not qualitatively change the results.

Model-free analysis

Using the Lipari-Szabo model-free formalism, backbone amides were fitted to either one of the five models using the modeling strategy described in Materials and Methods. Table 2 shows the summary of the number of residues that fit to each of the five spectral density models for both the free and bound states. For both forms of the protein, the majority of residues were satisfactorily fit with the simplest model, either with S² (model 1) or S² and τ_e (model 2). Figure 6A ▶ illustrates the generalized order parameters for both ligation states of the protein. The average S² for the free C-SH3 domain is 0.79 ± 0.13 and 0.79 ± 0.12 for the bound state. These values are only slightly lower than the average found for the 20 proteins from the Indiana Dynamics database (0.84 ± 0.11; Ye et al. 1999).

Table 2.

Summary of spectral density models used to fit R1, R2, and NOE data

Parameters optimized	Uncomplexed C-SH3	Complexed C-SH3
Model 1: S2, τm	12	21
Model 2: S2, τm, τe	20	6
Model 3: S2, τm, Rex	4	13
Model 4: S2, τm, τe, Rex	6	4
Model 5: , , τm, τs	3	4
Not fit	2	4
Proline residues	4	4
Overlap	6	0
Weak peak	1	2

Figure 6.

(A) Order parameters (S₂) profile both for the free (open circles) and complexed (filled circles) C-SH3 domain with Sos peptide. (B) Difference between the order parameters of the free and complexed C-SH3 domain.

The S² profiles are similar for both ligation states, with higher-than-average S² (>0.8) found in β strands I-V and the 3₁₀ helix and lower-than-average S² values found in the terminal residues and residues in the loop regions (i.e., RT, n-Src, and distal loop). Of particular note is the fact that the S² values in the RT loop (N166–S171) are lowest in both the bound and free forms. Residue E169 was found to be the most mobile in both states (S² of 0.52 and 0.59 for free and bound state, respectively). Interestingly, the backbone amide of this residue forms H-bond interactions with the carboxylate group of E172 in the bound state. Despite the high mobility observed in many RT-loop residues, the conserved diverging type II β-turn of all SH3 domains shows S² of >0.83 in both states, a result that is consistent with other SH3 domains.

Most terminal residues were fitted with a distinct two-timescale motion, which includes slower motions (τ_s<2 nsec), with τ_e < τ_s < τ_m. Clore et al. (1990) described this two-timescale model as involving fast motions or local oscillations such as diffusion in a cone and slow motions originating from transition jumps between two distinct states. Hydrogen-deuterium exchange data (Ferreon et al. 2003) also show that amides in the secondary structural elements have slower exchange, whereas residues in the loops have faster exchange rates (i.e, beyond the detection limit). In the case of Sem-5 C-SH3 domain, there is a strong correlation between the high-order parameters and the location of secondary structure elements.

Although the S² profile is similar in both ligation states, significant differences in magnitude exist (Fig. 6B ▶). For example, residues in the RT loop show significant increases in rigidity on complex formation, the trend persisting over the entire length of the loop. Residues W191 and R199 of the distal loop, and residues comprising the 3₁₀ helix, also showed an increase in order parameters on binding. All residues exhibiting significant increases in order parameters (aside from R199 in the distal loop) are either in direct contact with the peptide or are proximal to the binding interface in the RT loop. This observation, coupled with the quality of the fits of the various motional models to the data, indicates that the observed differences are real and reflect decreases in disorder on binding. It is also important to note that the observed increase in S² is consistent with changes in the relaxation data.

Additionally, as shown in Table 3, most of the significant changes in S² arise for residues that are fitted using the same model in the bound and free states. For example, the relaxation data for E169 in the RT loop, which showed the largest change (ΔS² = 0.06), was fit to model 4, both in the unliganded and the liganded states. Moreover, data analyses showed that, regardless of models used, the S² values for the unliganded state are consistently lower than those in the liganded state for most residues in the RT loop.

Table 3.

Differential effects on the Sem-5 C-SH3 residues coupled by ligand binding

Res. No.		ASA changea	Chemical shiftb	Structural changec	S2d	Model freee	Model bounde	ΔS2i
Leu 162			+
Phe 163		+	+		+	2	1	+
Asp 164	R	+	++		+	n.a.f	1	n.a.f
Phe 165	T	+	++	+	+	n.a.f	2	n.a.f
Asn 166		+	+	+	++	4	n.a.f	n.a.f
Pro 167			N.A.g	+	N.A.g	N.A.g	N.A.g	N.A.g
Gln 168		+	++	+	+++	n.a.f	4	n.a.f
Glu 169	L	+	++	+	+++	4	4	+
Ser 170	O		++	+	++	4	3	+
Gly 171	O		++		+	4	3	+
Glu 172	P	+	+		+	3	3
Leu 173			+
Asn 190		+	++ (s.ch)
Trp 191		+	++		+	2	2	+
Pro 204		+	N.A.
Asn 206		+	++ (s.c.h)
Tyr 207		+	+

^a Accessible surface area (ASA) buried in the binding interface, (+) >0Ų.

^b (+) <0.5 ppm; (++) >0.5 ppm in ¹⁵N dimension.

^c (+) >1.0 Å difference in Cα (bound vs. unbound).

^d (+) <0.82; (++) <0.75; (+++) <0.65 in S².

^e Models used in fitting (see Materials and Methods), either for the free or the bound states.

^f Overlap either in the free or the complexed state.

^g Proline residue.

^h Effect monitored in the side-chain amide.

ⁱ (+) 70.025.

In fitting the data, 10 residues required the conformational exchange term R_ex in the fitting model for the unbound state and 17 residues for the bound state. These residues are found in the RT loop, diverging type II turn (175–177), β_III strand, and include residues F203 and N206. Most of these residues are involved in binding. All the values of the R_ex term, except for S170 (1.097 ± 0.07 and 1.358 ± 0.079 in the bound and unbound states, respectively) are <1 Hz. (Fig. 7B ▶). Three residues (N-terminal residues, 155–157) required a two-timescale motion model in the unbound state as well as four residues (N-terminal 155 and 157 and C-terminal 211 and 212) in the bound state. As shown in other cases (Tjandra et al. 1996), most of the terminal residues experience a two-timescale motion. The timescale of the slow motions is ~2 nsec.

Figure 7.

(A) Internal correlation times (τ_e) profile both for the free and complexed C-SH3 domain with Sos peptide. (B) Residues, which exhibit conformational exchange (R_ex), both for the free and complexed C-SH3 domain.

Tryptophan side-chain dynamics

Using the same procedure described for the backbone amides, dynamics of the side-chain amide (HN^ɛ) of tryptophan was also determined. There are two tryptophans in the Sem-5 C-SH3, which are consecutive in sequence. One is buried (W192) and the other (W191) is solvent-exposed and directly intercalates with the polyproline peptide in the bound state (Lim et al. 1994b). W192 HN^ɛ cannot be fit to any model, whereas W191 HN^ɛ was satisfactorily fit to model 1 with S² equal to 0.77 ± 0.002 and 0.85 ± 0.006 (in free and bound states, respectively).

Comparison with other SH3 domains

The general pattern of low-order parameters in the loop regions (RT, n-Src, and distal loop) has been observed in other SH3 domains (Hansson et al. 1998; Horita et al. 2000; Wang et al. 2001); however, the relative magnitude of disorder in each loop is different between different SH3 domains. In the Hck SH3 domain (Horita et al. 2000), the n-Src loop shows a higher degree of disorder compared with the other loops. In the Btk SH3 domain (Hansson et al. 1998), the RT and n-Src loop show similar disorder, which are both higher than in the distal loop. In the Sem-5 SH3 domain, the RT loop shows more disorder than the n-Src and the distal loop. Although we have no explanation for the relative differences, we note that the Hck n-Src loop, which has more disorder than the n-Src loop in Sem-5 SH3 domain, contains two additional residues (Horita et al. 2000).

The results presented here can be compared with a recent study focused on the binding of the peptide (RALPPLPRY) to the c-Src SH3 domain (Wang et al. 2001). As is the case with Sem-5, most of the regions of c-Src, which experience an increase in S² on binding, are located in loop regions. Interestingly, residues E36 and Q49 of c-Src, corresponding to residues D188 and R199 in Sem-5, showed a decrease in motion even though these residues are >9 Å from the ligand interface. Although ΔS² for D188 of Sem-5 could not be determined due to overlap, R199 showed a substantial decrease in S², similar to the case in c-Src. Conversely, residues V32 and T35 in c-Src (as well as the corresponding residues I184 and D187 in Sem-5) showed an increase in motion on binding. Despite the numerous similarities, differences between the two SH3 domains are also observed. For instance, in c-Src residues T17 and D20 show an increase in motion on binding, whereas the corresponding residues in Sem-5, E169 and E172, show a decrease on ligand binding.

Discussion

Dual character of peptide binding site on Sem-5

Accessible surface area calculations on the crystal structure of the Sem-5:Sos peptide complex reveal that the following residues are in direct contact with the peptide: F163, D164, F165, N166, Q168, E169, E172, N190, W191, P204, N206, Y207 (Table 3). Residues that also show a significant change in chemical shift in the ¹H-¹⁵N HSQC spectra on ligand titration (difference of 0.1 and 0.5 ppm in ¹H and ¹⁵N dimension, respectively) include those stated earlier, as well as residues S170, G171, L173, F175, W192, S205. In addition, of the total number of residues that show chemical shift differences, only a subset show structural (Ferreon et al. 2003) and dynamic changes on binding of ligand (Table 3). A significant observation can be made from these results. Namely, the binding site of Sem-5 can be viewed as having a dual character (Fig. 8 ▶).

Figure 8.

Space-filling model representation of the crystal structure determined (Lim et al. 1994b) for Sem-5 C-SH3 (gray) with and without the Sos peptide (yellow). Residues belonging to the binding interface but showing no dynamic changes in the picosecond timescale on binding are shown in green, and residues in red are those showing dynamic changes. Residues shown in blue (S170 and G171) are those that do not interact with the peptide in the complex, but which nonetheless experience significant changes in chemical shift and dynamics on binding.

Residues E172, N190, P204, N206, and Y207, although burying significant surface area on binding, do not undergo a change in either the structure or the dynamics. In essence, the binding in this part of the molecule can be considered a rigid body association, the structural determinants of the binding being well approximated by the high-resolution structure. Residues F163, D164, F165, N166, Q168, E169, and W191, on the other hand, show significant changes in chemical shifts and dynamics. These differences occur in the regions of Sem-5 C-SH3 that show the highest structural differences between the free and complexed protein (Ferreon et al. 2003). These results indicate that binding is coupled to a conformational change in the RT loop. The fact that residues 170 and 171, which do not directly interact with ligand, also show both chemical shift and dynamic differences between bound and unbound states, further supports this conclusion.

Two key features of the observed conformational change in the RT loop are important. First, as determined from chemical shift perturbation and ITC experiments, the binding reaction is two state, and thus the conformational change is cooperative. Second, the dynamics measurements clearly show that the binding reaction is coupled to the decrease in the flexibility in the RT loop, as discussed earlier. In ensemble terms, the results indicate that the unbound state consists of a conformational manifold, wherein the RT loop has considerably greater conformational variation than the rest of the Sem-5 C-SH3 molecule. On binding of the ligand, that conformational manifold is significantly reduced in a cooperative fashion, with the average structure of the bound and unbound states differing considerably. For this region of the molecule, it is likely that the thermodynamics of binding cannot be deduced simply by considering the differences in structure between the canonical bound and unbound conformation. Instead, elucidation of the thermodynamics requires insight into the thermodynamic consequences of redistributing (i.e., reducing the size of) the ensemble.

The spatial partitioning of the two portions of the binding site, as seen in Figure 8 ▶, is intriguing, as both regions are generally contiguous groupings of residues. In essence, the binding site has a rigid structural scaffold on one side and a flexible surface on the other. It should be noted that the dynamic character of the RT loop, although decreased on binding, does not acquire a level of rigidity consistent with the remainder of Sem-5 C-SH3 (Fig. 6 ▶). Instead, the bound form also maintains conformational flexibility in this region. This leads to the conclusion that the bound state, as well as the unbound, is more accurately represented as an ensemble of states rather than as a single discrete conformation. Although the current studies do not provide information about the determinants of affinity and specificity in Sem-5 C-SH3, the observed dual character of the binding site offers interesting possibilities in this regard. Conformation variability of the RT loop in the bound state may indicate that specificity for different peptide ligands may be a function of the number of accessible protein conformations that can also facilitate low-energy contacts with the peptide. The role of dynamics in determining the thermodynamics of binding is currently under investigation.

Conformational entropy contributions

The results reported here have important thermodynamic consequences pertaining to the relationship between order parameters and conformational entropy. As shown by Yang and Kay (1996), NMR-derived order parameters are related to the conformational entropy, as both terms contain the same description of the orientational probability distributions of the amide vectors. Assuming diffusion-in-a-cone for the motions of the amide vectors, the conformational entropy has been expressed as:

(1)

where S_conf,OP is the conformational entropy determined from order parameters, S is the order parameter from picosecond and nanosecond contributions, and k_B is the Boltzmann constant. Although the precise relationship between the conformational entropy determined by equation 1 and the experimentally obtained entropy remains ambiguous, the qualitative implications are straightforward. Specifically, the increase in order parameters in the RT loop of Sem-5 C-SH3 on ligand binding is indicative of a decrease in conformational entropy.

The observation that Sem-5 C-SH3 undergoes a cooperative conformational transition on binding and that the transition results in a decrease in conformational entropy means that the two processes are linked functions (Wyman 1964). This observation has several noteworthy implications. First, the ΔS_conf,OP values determined for each residue from equation 1 are monitoring the same process. As such, the overall conformational entropy for the binding process cannot be represented as a sum of the individual residue-specific contributions.

Second, the conformational entropy changes observed on binding do not provide direct access to the magnitude of the conformational variations of the various states in the ensemble. This is evident from the binding constant expression of a system that undergoes a cooperative decrease in the conformational manifold on binding:

(2)

where K₀ is the intrinsic association constant in the absence of conformational fluctuations, and ΣK_i is the sum of statistical weights of all binding-incompetent states (i.e., the conformational variants). In essence, the denominator is the conformational partition function for the protein. As indicated by equation 2, the magnitude of the effect of conformational redistribution on the observed binding energetics will depend on the statistical weight of the binding-incompetent states (i.e., if ΣK_i ≫ 1, the effect will be large; if ΣK_i ≪ 1, the effect will be small). In other words, the magnitude of the change in conformational entropy on binding, as determined from equation 1, does not provide details of the conformational entropy of the individual states, even if the relationship between order parameters and conformational entropy are known precisely. Instead, the magnitude of the observed conformational entropy change (or any other observable sensitive to the equilibrium) will depend on where the equilibrium is poised, the precise effect being determined by the linkage relationship in equation 2.

Third, the cooperative redistribution of the ensemble on binding will likely have enthalpic as well as entropic consequences, as the free energy of redistributing the ensemble, ΔG_i (=−RTln ΣK_i) will, under most circumstances, have both enthalpic and entropic contributions (ΔG_i = ΔH_i − T * ΔS_i). Equation 2 indicates that if the binding-incompetent states differ enthalpically from the binding-competent species, redistributing the ensemble will reflect these differences in the apparent enthalpy (Eftink et al. 1983). For Sem-5 C-SH3, it is possible, indeed probable, that the observed dynamics changes are coupled to enthalpy changes, although direct experimental validation awaits further study. Nevertheless, it is clear that enthalpy changes coupled to dynamics would be difficult to rationalize and impossible to predict in the context of a static structural model of the protein.

The results presented here show that Sem-5 C-SH3 undergoes a cooperative two-state binding process, wherein binding is coupled to a conformational transition. This cooperative transition is associated with a decrease in conformational entropy in the protein, and that decrease is restricted to a relatively contiguous group of binding site residue. Despite these experimental insights, deciphering the energetic consequences of dynamics in the Sem-5:Sos binding process is not straightforward. Because it is not possible to resolve the structural and energetic attributes of the component states in the ensemble from relaxation data (using the current experimental protocol), it is not possible to elucidate the structural determinants of the observed energetics of binding. As equation 2 reveals that such uncertainties will likely affect the enthalpy as well as the entropy, it is clear that an accurate molecular-level description of binding is predicated on resolving the structure and energy of the relevant states in both the bound and unbound ensembles.

Materials and methods

Sample preparation

The C terminus of SH3 domain Sem-5 was cloned, overexpressed, and purified as described (Lim et al. 1994a,b). This SH3 domain has a mutation on residue 209, from cysteine to alanine, to prevent possible oxidation and intermolecular crosslinking. The purity of the eluted protein (>95%) was checked by electrospray ionization mass spectrometry, reverse-phase high-performance liquid chromatography (HPLC), and SDS-PAGE electrophoresis. ¹⁵N-labeled protein was obtained from cells grown in M9 minimal medium containing ¹⁵NH₄Cl (Isotec Inc.). The purified protein was concentrated to ~0.7 mM in 50 mM NaOAc, 100 mM NaCl, and 10 mM CaCl₂ (pH 4.8; 90% H₂0/10% D₂0). Because of cloning, there are two extra residues both at the N and C terminus of the SH3 domain that are not present in the crystal structure. These resonances are also weak and were also excluded from the relaxation analysis.

The H^N and ¹⁵N chemical shifts for the unliganded form were assigned using various 3D and 2D heteronuclear NMR experiments (Ferreon et al. 2003). For the bound SH3 domain, unlabeled peptide (Ac-PPPVPPRRR-amide) was titrated to ¹⁵N-labeled SH3 domain and chemical shift perturbations were monitored via ¹H-¹⁵N HSQC spectra. The peptide used in the studies was synthesized and purified by the Peptide Synthesis Laboratory (UTMB). The purity of the peptide (>95%) was checked by mass spectrometry and reverse-phase HPLC.

NMR spectroscopy

All NMR experiments were collected at 25°C on a Varian UnityPlus 400-MHz spectrometer using a triple resonance probe equipped with a pulsed field z gradient. ¹⁵N−T₁, T₂, and {¹H}-¹⁵N NOE NMR experiments were done for both the unliganded and liganded forms of C-SH3 using the sensitivity-enhanced pulse field gradient sequences derived from the Varian ProteinPack, which are based on the pulse sequences described previously (Farrow et al. 1994). R₁ (T₁⁻¹) data were acquired with the following 10 relaxation delays: 10, 40, 80, 100, 150, 230, 320, 430, 590, 840 msec, and R₂ (T₂⁻¹) data were obtained with the following 11 relaxation delays: 10, 30, 50, 70, 90, 110, 150, 170, 190, 210, 250 msec. Both R₁ and R₂ experiments use 1-sec recycle delays. The {¹H}-¹⁵N NOEs were measured by recording HSQC spectra with and without proton saturation. The former spectrum uses 3-sec proton saturation and 2-sec delays, whereas the latter only has the 5-sec delay. Duplicates were done for all spectra acquired.

Data processing and analysis

All spectra were processed using the FELIX v.98 software (Biosym Technologies, Inc.) on Silicon Graphics Indy workstation. The spectra were applied with a 90° sine bell shift window function and zerofilled to a 256 × 1024 matrix. Peak height intensities were measured by the automated routine in the software. The ¹⁵N−T₁ and T₂ time constants were obtained by fitting the data to a two-parameter single exponential decay function using software obtained from Neil A. Farrow. The uncertainties of T₁ and T₂ values were estimated from the nonlinear least squares fit. The relaxation rates R₁ and R₂ were obtained from the inverse of the T₁ and T₂ and the uncertainties were obtained by using the formula σ/T_1,2² where σ is the uncertainty in the T₁ or T₂ value. We have also tested this by fitting the data to the R₁ or R₂ parameter and we obtain the same results both for the rates and the uncertainties. The NOE values were obtained from the ratio of the volumes with and without proton saturation and averaged for the duplicate spectra. Uncertainties in the NOEs were estimated from the base plane noise of the spectrum (Farrow et al. 1994).

Model-free formalism

The theoretical basis of the ¹⁵N relaxation has been discussed in a number of papers. Briefly, ¹⁵N relaxation is primarily dominated by dipolar and chemical shift anisotropy and is related to the spectral density functions at the ¹H and ¹⁵N frequency combinations (ω; Abragam 1961). The transverse relaxation rate (R₂) can also have, aside from contributions from dipolar and chemical shift anisotropy, conformational exchange contributions in the microsecond-to-millisecond timescale (R_ex) by the following relation:

The spectral density function is then approximated by the following equation, which is the model-free formalism developed by Lipari and Szabo (1982a^,b):

This model-free approach characterizes the spectral density function as composed of the overall tumbling of the molecule (τ_m) as well as the internal motions (τ_e) of a given ¹⁵N−¹H vector. This equation simplifies to the following spectral density function:

if all internal motions are very fast (τ_e < 10 psec). Using this Lipari-Szabo approach, relaxation parameters (R₁, R₂, and NOE) were fitted to the different models with an isotropic diffusion tensor using Modelfree 4.01 (Mandel et al. 1995,Mandel et al. 1996) to obtain parameters that describe the internal motions of the molecule. The five models are summarized as follows:

• Model 1: S²

• Model 2: S² and τ_e

• Model 3: S² and R_ex

• Model 4: S², τ_e, and R_ex

• Model 5: , , and τ_s

For models 1–4 → S² = where S² = * and = 1

The last model uses a two-timescale model to characterize both the fast motions () and slow motions (, τ_s; Clore et al. 1990). As described elsewhere (Mandel et al. 1995,Mandel et al. 1996), model fitting involves three steps: initial estimation of the rotational correlation time (τ_m) or the diffusion tensor, model selection, and final optimization.

Initial estimate of τ_m

Programs (which include the Modelfree 4.01) obtained from Dr. A.G. Palmer III (Columbia University, http://cpmcnet.columbia.edu/dept/gsas/biochem/labs/palmer) were used for obtaining the inertia tensor, statistical tests of anisotropy, and initial estimate of τ_m. The program pdbinertia was used to calculate the principal moments of the inertia tensor using the recently solved NMR solution structure for the unbound state and the crystal structure (wherein hydrogens are added) for the complexed state. MOLMOL v. 2K.1 (Koradi et al. 1996) was used to generate the pdb coordinates of the crystal structure with the hydrogens. The program r2r1_diffusion determines the significance of fitting to an isotropic versus an axially symmetric model. This uses the approach of Tjandra et al. (1995) to determine the diffusion tensors for spherical and axially symmetric motional models from the experimental relaxation data. An initial estimate of τ_m was obtained by two methods. The first is through the r2r1_diffusion, which also gives the τ_m for the isotropic model from the structural coordinates. The second uses the program tmest and is based on the approach of Kay et al. (1989). Residues having an R₁/R₂ ratio above or below one standard deviation from the mean and NOE <0.55 are not included in the analysis (Kay et al. 1989; Clore et al. 1990; Tjandra et al. 1995). Model selection is based on the procedure described in Mandel et al. (1995).

Model selection and optimization

Model-free parameters were calculated from the measured relaxation parameters using the program Modelfree 4.01 (Palmer III et al. 1991; Mandel et al. 1995). The process of analyzing the relaxation data starts with an estimate of the overall correlation time. The selection of the appropriate model for amide vector follows the same strategy outlined by Palmer III and coworkers (Mandel et al. 1995). Five hundred Monte Carlo simulations were generated to determine the cumulative probability distribution and to compare the χ² value with the critical value at α = 0.05 or 0.10. Models 2 and 3 were preferred over model 1 by evaluating the F-statistic at α = 0.20. Because we only have three relaxation parameters, model 4 and model 5 were resorted to only if the sum of the squares error was zero and if the residues were not modeled adequately by models 1–3. Model optimization was then done by Powell minimization to obtain the final τ_m and the optimized parameters. Uncertainties in the dynamics parameters were determined using the Monte Carlo simulations carried out by the Modelfree program, as described earlier. Residues were considered adequately fit if they followed the criteria for Farrow et al. (1994): (a) all three experimental relaxation parameters are reproduced within 95% confidence limits, i.e., 1.96 times the experimental uncertainty and (b) fitting parameters must be greater than their error. The tryptophan side chains were fitted in a similar procedure, except the value used for the difference between the parallel and perpendicular components of the chemical shift tensor for the indole amides was taken to be −89 ppm (Cross and Opella 1983; Stone et al. 1993), as opposed to the value for backbone amide groups (−160 ppm; Hiyama et al. 1988).

Chemical shift perturbation

The peptide Ac-PPPVPPRRR-amide was synthesized and purified by the Peptide Synthesis Laboratory (UTMB). The purity of the peptide (>95%) was checked by mass spectrometry (Louisiana State University) and reverse-phase HPLC. Peptide concentration was determined from quantitative amino acid analysis. The ¹⁵N-labeled sample (0.66 mM) was prepared in 50 mM NaOAc, 100 mM NaCl, and 10 mM CaCl₂ (pH 4.8; 90% H₂0/10% D₂0). ¹H-¹⁵N HSQC spectra were collected for each titration point using a 750-MHz Varian UnityPlus spectrometer. A total of 15 titration points were collected and, at the final point, the peptide concentration was approximately twice that of Sem-5 C-SH3 protein. The binding curves were analyzed using a nonlinear regression analysis following the procedure of Lian and Roberts (1993), using the following equation to monitor the chemical shifts in the labeled protein resonance:

where P_t, L_t, and δ_obs are the total protein, ligand concentrations, and the observed change in chemical shift on ligand titration, respectively, which are known, and δ_PL and K_d are to be determined by the regression analysis, where K_d is the dissociation constant and δ_PL is the chemical shift of the complex.

Isothermal titration calorimetry (ITC)

Titration experiments were performed using the VP-ITC system at 25°C as described elsewhere (Wiseman et al. 1989; Gómez and Freire 1995; Renzoni et al. 1996). The unlabeled protein was extensively dialyzed against 50 mM NaOAc, 100 mM NaCl, and 10 mM CaCl₂ (pH 4.8). The peptide (Ac-PPPVPPRRR-amide) was lyophilized from pure water and dissolved with the buffer from the final dialysis. Protein and peptide were then centrifuged to remove particulates and degassed, and the pH was adjusted to the final pH of 4.8. The protein (0.269 mM) was loaded into the 1.3-mL sample cell and the peptide (4.22 mM) into the injector. An initial 2-μL injection and 31 × 4μ-L injections were made, with a 6-min equilibration period between injections. A control experiment (peptide titrated into buffer alone) was performed to evaluate the heats of dilution, which were subtracted from the experimental titration results. Data were then fitted using a nonlinear least squares program in Origin for ITC v. 5.0 (Microcal), varying the stoichiometry (N), binding constant (K_b), and binding enthalpy (ΔH°). The K_b determined from the best-fit curve was used to determine the free energy (ΔG°) and the entropy (ΔS°) involved in binding using the following thermodynamic relation:

where R is the gas constant; T is the absolute temperature; and ΔG°, ΔH°, and ΔS° are the standard free energy, enthalpy, and entropy for the binding.

Abbreviations

• SH3, src-homology domain 3

• C-SH3, C-terminal SH3 domain

• Sos, Son of Sevenless

• NMR, nuclear magnetic resonance

• 2D, two-dimensional

• NOE, nuclear Overhauser effect

• ITC, isothermal titration calorimetry

• HSQC, heteronuclear single quantum coherence

• ppm, parts per million

• PDB, Protein Data Bank