Dynamically driven protein allostery

cAMP binding to CAP^N yields distinct conformational states

We have recorded 2D ¹H-¹⁵N Heteronuclear Single Quantum Coherence (HSQC) spectra of CAP^N in three states: apo-CAP^N, cAMP₁-CAP^N, and cAMP₂-CAP^N (Fig. 1). In the apo state, CAP^N exists as a symmetric dimer with identical ligand binding sites, as evidenced by the appearance of only one set of peaks for both subunits (Fig. 1a). Binding of the first cAMP molecule to CAP^N gives rise to the intermediate state complex, cAMP₁-CAP^N; because of the strong negative cooperativity, only one of the two subunits is occupied. As expected for an asymmetric dimer, the chemical shift degeneracy is lifted and two sets of signals are observed: one set remains largely unchanged and corresponds to the unliganded subunit and the other one, markedly affected by cAMP binding, represents the liganded subunit (Fig. 1b and Supplementary Fig. 2 online). Binding of the second cAMP molecule to yield the cAMP₂-CAP^N complex, restores the symmetry of the protein dimer to a conformation identical to that of the liganded subunit in the cAMP₁-CAP^N complex (Fig. 1c). Both cAMP binding reactions are very slow on the NMR time scale.

Further analysis of the chemical shift behavior of CAP^N during its sequential interaction with cAMP (Supplementary Fig. 2) provides further unambiguous evidence of the strong negative cooperativity, in agreement with previous biochemical data²². The key to discriminating between negative and positive cooperativity is to study the complex at the stoichiometric ratio of cAMP: CAP^N (dimer) 1:1. If cAMP bound to CAP^N with positive cooperativity, two species would be present at this ratio; the apo-CAP^N and the cAMP₂-CAP^N. Since both cAMP binding steps are slow on the NMR time scale, 50% of the CAP^N would remain in the apo state, that is, the resonances would not be affected by the presence of the cAMP. However, this is not what we observe experimentally. The unfilled subunit at the ratio of cAMP: CAP^N 1:1 is clearly affected, as many residues at the interface with the liganded subunit change substantially their chemical shift (Supplementary Fig. 2). These results demonstrate that there is only one protein complex present in solution at the cAMP: CAP^N 1:1 ratio corresponding to the intermediate cAMP₁-CAP^N complex. Therefore, cAMP binds to CAP with strong negative cooperativity.

Effect of sequential cAMP binding on the structure of CAP^N

To assess the effect on protein structure of the sequential binding of the first and second cAMP molecules to CAP^N, we have used chemical shift mapping. The exquisite sensitivity of the chemical shift to environment ensures that considerable shifts would take place if changes in the mean protein conformation occurred on ligand binding. The effect of first cAMP binding to the liganded subunit is pronounced throughout the whole monomer as evidenced by the observed widespread chemical shift changes (Fig. 2). In contrast, the unliganded subunit in the cAMP₁-CAP^N complex is slightly perturbed only at its long C helix, which forms the dimer interface. None of the residues in the β-roll or the phosphate binding cassette (PBC) motif, the most conserved and important region in the protein for its interaction with cAMP, is affected. Further examination of the chemical shift behavior of CAP^N during its sequential interaction with cAMP demonstrates that the effect of cAMP binding on protein structure is confined, in each step, only to the interacting subunit (Supplementary Fig. 2). Thus, the mean conformation of the unfilled cAMP site at the unliganded subunit of the cAMP₁-CAP^N complex is not at all affected by the presence of the first cAMP, suggesting that the contribution of the “structural” component to the observed negative cooperativity is negligible.

Effect of sequential cAMP binding on slow protein motions

To understand how protein dynamics adjust along the cooperative reaction coordinate, we characterized the backbone motions of CAP^N as a function of the cAMP ligation state over a wide range of functionally relevant time scales by measuring relaxation rates (Supplementary Fig. 3 online). Slow domain motions on the micro- to millisecond (μs-ms) time regime are biologically very important because they are close to the time scales on which functional processes take place, such as docking, protein folding and allosteric transitions^23-26. Upon binding of the first cAMP to CAP^N, resonances for backbone residues of both subunits undergo extensive peak broadening (Supplementary Fig. 4 online). This indicates the presence of substantial conformational exchange, which is identified by the high magnitude of the term R_ex (Fig. 3). The contribution to R_ex from cAMP binding and dissociation is negligible (see Supplementary Discussion). Thus, the pronounced increase of R_ex for almost all residues of both subunits means that the protein in the cAMP₁-CAP^N complex populates an ensemble of alternate conformations that interconvert on the μs-ms timescale. These slow motions are activated by the first cAMP binding.

Notably, despite the fact that binding of the first cAMP molecule cannot induce long-range structural effects to the other subunit (Fig. 2), the dynamics of the system respond to the binding in a highly cooperative way emphasizing their allosteric character. Residues of the unliganded subunit in the cAMP₁-CAP^N complex that are located as far as 35 Å from the bound cAMP exhibit enhanced dynamics (e.g., Tyr23, Leu29, Ile30). Strikingly, the prominent conformational dynamics on the μs-ms time scale exhibited by almost all residues in cAMP₁-CAP^N are completely quenched upon binding of the second cAMP molecule to yield the cAMP₂-CAP^N complex (Fig. 3).

Both subunits undergo the same conformational exchange

To better understand the conformational process that results in such a dramatic change in slow motions of the singly liganded cAMP₁-CAP^N complex, we performed a series of relaxation dispersion experiments (Fig. 4)²⁷. The additional line-broadening of NMR signals (R_ex) caused by conformational exchange between the ground and an excited state depends on the sum of forward and reverse rates of interconversion (k_ex), the relative populations of the exchanging species (p_A and p_B) and the chemical shifts between the exchanging species (Δω)²⁷. Because the conformational exchange process is in the fast exchange limit, only k_ex and the product p_Ap_B(Δω)² can be measured. The results show that k_ex appears to be similar for most of the residues in the liganded subunit as the majority of the residues with appreciable R_ex can be fitted with k_ex values ranging from 900 to 1600 s^-1 (Fig. 4a). Therefore, it appears that most of the residues in the liganded subunit report on the same exchange process. Assuming a population of 10% (p_B∼0.1) for the excited state, an estimation of Δω can then be made. The ¹⁵N chemical shift difference (Δω) between the two exchanging conformations may be as high as 4 ppm for residues with large R_ex values (∼10-15 Hz), implying major structural rearrangements. If the population of the excited state (p_B) is lower than 10%, then even larger Δω values would be present. As in the case of the liganded subunit, k_ex for most residues in the unliganded subunit can be fitted using k_ex values ranging from 1000 to 1500 s^-1. Another smaller group of residues in the unliganded subunit (e.g. Ile106; Fig. 4b) appear to have k_ex values close to 2,000 s^-1. It is possible that these few residues are affected by another, faster process. However, considering the uncertainty, almost all residues can be considered to exchange between the sites with similar k_ex. Therefore, it is very likely that all residues of the cAMP₁-CAP^N complex, both from the liganded and the unliganded subunits, report on the same conformational exchange process. This is corroborated by the observation that residues located at the dimer interface have very similar k_ex.

Effect of sequential cAMP binding on fast protein motions

Next, we measured changes in the pico- to nanosecond (ps-ns) timescale motions of the backbone of CAP^N as a function of the ligation state (Fig. 5). Motions on this fast time regime are biologically very important because of their strong effect on the entropy of the system^28,29. Even small changes of these motions have been shown to result in significant entropic changes capable of driving binding processes^30-32. The conformational flexibility of CAP^N on the ps-ns time scale, as indicated by the spectral density function J(0.87ωh), is illustrated in Figure 5. Our data show that binding of the first cAMP molecule to yield the cAMP₁-CAP^N complex does not affect these motions appreciably (Fig. 5). The effect is local and is confined within the cAMP binding site, where only few residues rigidify their backbone. Residues 80-84 at the PBC motif of the liganded subunit and 126-133 at the C terminus of the unliganded subunit are the only ones that slightly dampen their fast motions in the singly liganded complex. In contrast to the slow motions (Fig. 3), the ps-ns dynamics of the core of the unliganded subunit are not perturbed. Notably, binding of the second cAMP molecule causes widespread stiffening of the backbone fluctuations in both subunits (Fig. 5). The extensive suppression of protein motions throughout the whole molecule attests, once again, to the allosteric nature of protein motions and their exceptional capacity to serve as a carrier of allosteric energy

Effect of cAMP binding on the native-state ensemble

To directly probe the native-state conformational ensemble of CAP^N, we measured NMR-detected hydrogen/deuterium (H/D) exchange rates of the backbone amides of CAP^N as a function of the cAMP ligation state. The native-state ensemble of a protein consists of an equilibrium of conformational states and its redistribution upon ligand binding has been suggested to impinge on cooperativity^11,12,33-35. Fluctuations of the structures allow the amide sites to sample a manifold of conformations, some of which permit exchange with the solvent. Therefore, amide exchange rates can reflect dynamic motions over a wide range of time scales that are not accessible by relaxation methods. Binding of the first cAMP molecule to CAP^N results in a slight decrease in exchange rates of the residues in the liganded subunit (Fig. 6). However, the unliganded subunit remains largely unaffected. Strikingly, binding of the second cAMP molecule results in substantial and global decrease in exchange rates of residues in both subunits. Residues at the dimer interface show the most pronounced increase in protection from solvent, indicating optimization of the packing interactions at the interface only when both subunits are filled with cAMP. We conclude that a considerable reduction in the number of conformational states occurs in the cAMP₂-CAP^N complex, but not in the intermediate cAMP₁-CAP^N complex. It is again noteworthy the distinct effect that the sequential cAMP binding has on the stability of the protein, with only the second cAMP being capable of eliciting long-range effects.

Thermodynamic basis of the negatively cooperative binding

To determine the thermodynamic basis for the observed anti-cooperative binding of cAMP to CAP^N, we used high-sensitivity ITC to measure the association free energy ΔG and its enthalpic and entropic components, ΔH and ΔS. The biphasic nature of the isotherm (Fig. 7) clearly points to the existence of two non-equivalent sites in CAP^N with a large difference in the affinity for cAMP. The difference in the free energy change between the two binding steps, ΔG₁-ΔG₂, is -2.8 kcal mol^-1 resulting in the second cAMP molecule binding to CAP^N with a 100-fold lower affinity than the first cAMP. The second cAMP binding is accompanied by a more favorable enthalpy change, which, in line with the chemical shift mapping results, argues against its site being distorted as a result of the first cAMP binding. In fact, the measured enthalpy change between the two binding steps (ΔH₁-ΔH₁=+1.1 kcal mol^-1) opposes the observed allosteric phenomenology and favors a positive, rather than a negative cooperativity. Instead, the second cAMP binds to CAP^N with a drastically less favorable entropy change than the first cAMP, amounting to an entropic penalty of Δ(-TΔS)=4 kcal mol^-1. Thus, the thermodynamic basis for the observed negative cooperativity is entirely of entropic nature.

Conformational entropy change upon cAMP binding

The present findings suggest that the observed extensive changes in protein motions is the most probable source of the large difference in entropy change between the two cAMP binding steps. To assess the contribution of protein motions to the conformational entropy of the system^36,37, we calculated changes in order parameters for the first and second cAMP binding to CAP^N (Fig. 8 and Supplementary Figure 5 online). The order parameter, S², is a measure of the amplitude of the internal motions on the ps-ns time scale and may vary from S²=1, for a bond vector having no internal motion, to S²=0, for a bond vector rapidly sampling multiple orientations²⁸. Binding of the first cAMP to CAP^N results in small increases in S², confined primarily to the binding site (Fig. 8). In contrast, binding of the second cAMP molecule causes a widespread increase in S² (Fig. 8) indicating a global rigidification of the protein, in agreement with the J(0.87ωh) data (Fig. 5). By converting order parameters to entropy (see Methods) we estimate that binding of the first cAMP molecule to CAP^N is accompanied by a -TΔS_conf1=3.2 kcal mol^-1 conformational entropic penalty, whereas binding of the second cAMP molecule incurs a much higher entropic penalty, -TΔS_conf2, amounting to 18.1 kcal mol^-1 (Fig. 8). We should note here that conformational entropy estimated through order parameters should be taken as an upper-bound estimate, because the approach assumes that motions are uncorrelated^28,32. Thus, the extensive difference in S² is translated to an enormous decrease in conformational entropy of the system, amounting to -TΔS_conf∼15 kcal mol^-1, in going from the cAMP₁-CAP^N to the cAMP₂-CAP^N state. We conclude that the calorimetrically measured pronounced entropic penalty that accompanies the anti-cooperative cAMP binding is indeed dominated by the unfavorable conformational entropy change incurred by the second cAMP binding to CAP^N.

Sample preparation

The His₆-tagged 1-138 domain (CAP^N) was amplified with the polymerase chain reaction (PCR), using as a template the pAKCRP plasmid that encodes for wild-type E. coli CAP (residues 1-209), and transformed in BL21(DE3) cells. Isotopically labelled samples for NMR studies were prepared by growing the cells in M9 media in the presence of ampicilin. For the backbone assignment a U-²H/¹⁵N/¹³C-labeled sample was prepared by supplementing the growing media with 1 g l^-1 of ¹⁵NH₄Cl, 2 g l^{-1 2}H7/¹³C₆-glucose in 99.9%-²H₂O. For the relaxation experiments a U-²H/¹⁵N-labeled sample was prepared using 1 g l^-1 of ¹⁵NH₄Cl, 2 g l^{-1 2}H7/¹²C₆-glucose in 99.9%-²H₂O. Two purification steps were used for all protein samples. The first one involved a nickel-chelated sepharose fast flow resin (Amersham) and the second one a Superdex-200 size exclusion column (Amersham). Protein concentration was determined spectrophotometrically at 278 nm using an extinction coefficient of 40,800 M^-1 cm^-1. cAMP concentration was determined spectrophotometrically using a molecular extinction coefficient of 14,650 M^-1 cm^-1 at 259 nm. Protein concentration for samples used for NMR studies was typically 0.4 mM. The singly (cAMP1-CAP^N) and doubly (cAMP₂-CAP^N) liganded complexes were prepared by using cAMP: CAP^N molar ratios of 1:1 and 2:1, respectively. The strong binding of cAMP to CAP^N for both steps (K_d1= 0.04 μM and K_d2=4 μM, for the first and second binding steps, respectively) and the strong negative cooperativity ensures that at cAMP: CAP^N ratio of 1:1 one subunit of CAP^N is completely saturated and the other one completely empty. Similarly, at cAMP: CAP^N ratio of 2:1 both subunits are saturated, in the absence of excess of cAMP.

NMR assignment

All NMR spectra were recorded at 30 °C on a Varian 600 MHz spectrometer. Sequential assignment of the ¹H, ¹³C, and ¹⁵N protein chemical shifts in the apo, (cAMP)₁-CAP^N, and (cAMP)₂-CAP^N was achieved by means of through-bond heteronuclear scalar correlations along the backbone using a set of 3D HNCA, HN(CO)CA, HNCACB, and HN(CO)CACB experiments.

Relaxation measurement and analysis

Four relaxation parameters were measured for all backbone amides of CAP^N in the apo, singly liganded and doubly liganded complexes with cAMP: the {¹H}-¹⁵N nuclear Overhauser effect (NOE), the longitudinal relaxation time T₁, the transverse relaxation time T₂, and the rotating-frame longitudinal relaxation time T_1ρ. ¹⁵N relaxation dispersion T₂-CPMG data were recorded for the cAMP₁-CAP^N complex. Data sets were acquired as 256 × 1024 complex points in the t₁ × t₂ dimensions. Data points were fitted as a function of the length of the parametric relaxation delay to two-parameter decay curves of the form I(t)=I₀ exp (-Rt). The error in the rate constant was assessed from Monte Carlo simulations. {¹H}-¹⁵N NOE data were obtained by recording, in an interleaved manner, one spectrum with a 5 s recycle delay followed by 5 s saturation and another spectrum with no saturation and a 10 s recycle delay. Representative relaxation data were recorded at three different protein concentrations (0.05, 0.4, and 1.05 mM) to exclude contributions from protein aggregation. The final relaxation data were recorded using a protein sample of 0.4 mM concentration.

To characterize and quantify exchange contributions, R_ex, both the rotating frame⁴⁴ and T₂-CPMG²⁷ relaxation dispersion experiments were recorded. In both cases R_ex is measured independently of the presence of anisotropy and any change that ligand binding may elicit to the overall tumbling of the protein. These NMR experiments measure the excess transverse relaxation rate, R_ex, caused by the exchange of nuclei between different conformations/states with different chemical shifts. The similarity of exchange rates extracted from fits of dispersion profiles on a per-residue basis (Fig. 4) supports a two-state model of exchange.

To probe fast motions that take place on ps-ns time-scale we have used the reduced spectral density mapping approach⁴⁵, which makes no assumptions a priori regarding the molecular anisotropy or internal motions. Briefly, the relation between the relaxation rates and the spectral density function, J(ω), is used to extract values of the latter with the use of T₁, T₂, and cross-relaxation rate, σ. At high frequencies (>400 MHz), J(ω) varies very little and the three spectra densities J(ωh+ ωn), J(ωh), J(ωh- ωn) may be averaged to one average spectra density <J(ωh)>, in which ωh and ωn is the ¹H and ¹⁵N Larmor frequency, respectively. The frequency in <J(ωh)> may be taken equal to 0.87ωh. The value of the spectral density function at 0.87ωh frequency is the most diagnostic for ps-ns motions.

Assessment of conformational entropy (-TΔS_conf) from order parameters

The contribution to the conformational entropy of protein motions to the cooperative binding of cAMP to CAP^N was estimated by measuring changes in the order parameter, S², as a function of the ligation state. Because of the complex dynamics observed in the CAP system, reduced-spectral density mapping was employed for calculating S², as described before^46,47. This approach was successfully applied before in calculating the conformational entropy contribution to the association entropy of a protein-DNA complex⁴⁷. A local overall rotational correlation time, τm, is obtained from the spectral density function as:

Once a value of τm has been determined, the amplitude of intramolecular motion of the N-H bond vector in a molecular reference frame is given by:

As shown previously, the parameter of S² of equation (2) is a very good approximation of the square of the generalized order parameter extracted by the model-free spectral density function^46-48. In the case of ribonuclease H, for example, S² values calculated from Eq. 2 and using the full model-free formalism have a linear correlation coefficient of 0.97 (ref. 47). The conformational entropy of the CAP^N backbone is then calculated by³⁶:

in which k_B is Boltzmann's constant, and S_a and S_b are the order parameters for state a (e.g. apo-CAP^N) and state b (e.g., cAMP₁-CAP^N). Accurate extraction of order parameters is critically dependent on identifying and separating the chemical-exchange contribution. In our study, this has been done by recording rotating frame and T₂-CPMG relaxation dispersion, so the T₂ values used for obtaining the J(0) have been corrected for the exchange term R_ex.

Determination of exchange rates

Amide proton exchange rates were determined from the time course of the peak intensities in a series of ¹H-¹⁵N HSQC spectra after dissolving lyophilized samples in D₂O. The final pD was adjusted by DCl or NaOD solutions and was measured with a glass electrode at the temperature of the exchange experiment, taking into account the isotope effect: pD = pH_read + 0.4. Exchange rates were determined by plotting the intensities of each residue against time and fit by a first-order rate expression.

ITC experiments

Calorimetric titrations of CAP^N with cAMP were performed on a VP-ITC microcalorimeter (MicroCal). Protein samples were extensively dialyzed against the ITC buffer containing 50 mM KPi (pH 6.0), 500 mM KCl, and 1 mM TCEP. All solutions were filtered using membrane filters (pore size 0.45 μm) and thoroughly degassed for 20 min by gentle stirring under vacuum. The 1.35 sample cell was filled with a 30 μM solution of protein and the 250-μl injection syringe with 0.6 mM of the titrating cAMP. The ligand solution was prepared by dissolving cAMP in the flow through of the last buffer exchange. Each titration typically consisted of a preliminary 2-μl injection followed by 30 subsequent 8-μl injections. Data for the preliminary injection, which are affected by diffusion of the solution from and into the injection syringe during the initial equilibration period, were discarded. Binding isotherms were generated by plotting heats of reaction normalized by the modes of injectant versus the ratio of total injectant to total protein per injection. The data were fitted using the sequential binding site model embedded in Origin 7.0 (Microcal).