Many protein functions are known to depend drastically on the structure and dynamics of the solvent.[1–5] For example, Fenimore et al[5]. have distinguished between “slaved” and “non-slaved” protein processes, depending on whether they are coupled to the solvent fluctuations or not. Because the coupling of the protein to the surrounding solvent bath is controlled by the properties of the protein–solvent interface, the characterization of protein hydration is essential for understanding these phenomena. This does not only require an understanding of the solvent effects upon the properties of the protein, but, vice versa, also of the effect of the protein, for instance the topology of the protein surface and its interfacial dynamics, on the solvent.[6–8]
The present paper aims at contributing to a more detailed understanding of protein hydration by investigating effects of H2O–D2O substitution. Isotopic substitution is used in many experiments on biomolecular systems to exploit properties of the deuterium nucleus not shown by protons, or to generate solute/solvent contrast. For example, the properties of the deuterium nucleus render D2O to be the solvent of choice in neutron scattering studies of protein and hydration[3,9] and in NMR studies of biomolecular and hydration dynamics.[10–13] Deuterium is also used as a tracer, for instance, in kinetic studies of enzymatic reactions.[14] Thereby, it is commonly assumed that the biomolecular interactions are not modified by the solvent isotopic substitution. In contrast, we focus here on the structural and dynamical changes associated with protein–water interactions in H2O and D2O, respectively, and their consequences for the thermodynamic properties of the system. We consider the enzyme ribonuclease A (RNase A, 124 amino acids, molecular mass M = 13.7 kDa) as a prototypical protein used in many studies of protein folding. RNase A is a single-domain pancreatic enzyme protein which catalyses the cleavage of single-stranded RNA. Its crystal[15] and solution[16] structure comprise three helices and a large β-sheet region. It is known since a long time that H2O–D2O substitution increases the thermal unfolding temperature of RNase A.[17] Such temperature shifts are found with other proteins as well.[18–21] Some additional measurements were also performed, for comparison, for the small globular protein ubiquitin (76 amino acids, M = 8565 Da).
At the molecular level, H2O–D2O isotope effects result from the fact that deuterium bonds in water are stronger than hydrogen bonds, because the larger mass of the deuteron lowers the zero-point vibrational energy of the intermolecular modes.[22,23] The resulting increase in the solvent structure of D2O and in D2O–D2O affinity reveals itself in distinct isotope effects on the thermodynamic and dynamical properties of pure water and aqueous solutions of simple solutes.[24] The rationale is that the increase in solvent structure causes a stronger solvation of hydrophilic and a less efficient solvation of hydrophobic species. In protein solutions, such effects might cause polypeptides to reduce their solvent exposure by adopting a more compact shape or by associating into larger aggregates.[19,25–28] This conjecture does not only explain the observed temperature shifts of unfolding transitions, but also the promotion of protein aggregation in D2O.[27,28] However, such an interpretation is not unambiguous, and by a detailed analysis of entropy and enthalpy contributions, some authors have come to the opposite conclusion that the native protein should be destabilized by D2O.[20,21]
Important information on properties of the protein–water interface can also be deduced from volumetric properties. For example, it has been shown that the apparent thermal expansion coefficient, α, of the protein and its temperature coefficient, dα/dT, are drastically influenced by protein–solvent interactions. By a rather new and sensitive technique called pressure perturbation calorimetry (PPC), α(T) can now be measured with very high precision.[29–31] In PPC, the coefficient of thermal expansion of the protein is deduced from the heat consumed or produced after small isothermal pressure jumps. Using this technique, we have recently studied the solvation properties RNase A in its native and unfolded state in the presence of several chaotropic and kosmotropic co-solvents.[30] In the course of these studies we have found that the substitution of H2O by D2O has drastic effects on the volumetric properties, indicating a stabilization of the native form in D2O. In the present work, we reconsider these isotope effects in more detail, and probe local isotope effects at the protein–water surface by dielectric relaxation experiments. Dielectric relaxation in the MHz region is a standard technique for studying the structure and dynamics of protein solutions.[32,33] Recent extensions over a broader frequency band up to the GHz regime[34,35] and data evaluation assisted by molecular dynamics (MD) simulations[36,37] considerably increase its power.
Solutions of bovine pancreatic RNase A (Sigma Chemical Co.) were prepared with 10 mM Na2HPO4 buffer. The experiments in H2O were performed at pH = 5.5, at which DSC traces indicated the highest stability of the native form. Solutions in D2O (Aldrich, D-content > 99.9 atom-%) were adjusted to the same proton activity as in H2O. This implies that the value of pD, e.g. recorded by a conventional glass electrode was 5.9 (i.e. pH + 0.4). Lyophilized and essentially salt-free ubiquitin from bovine red blood cells (Fluka, protein content > 90%) was dissolved in buffer-free H2O and D2O solutions.
The thermal unfolding of RNase A in H2O and D2O was measured by means of a high precision VP DSC micro-calorimeter (MicroCal, Northamption, MA, USA). The same instrument, supplemented by the MicroCal PPC accessory, was used in the PPC experiments. Experimental details can be found in [refs. 29 and 30]. As the PPC technique is comparatively new, a brief description is given here. PPC measures the heat consumed or released by a sample after small isothermal pressure jumps. In the differential PPC experiment two cells of equal volume (here 0.5 mL), containing the protein solution and the buffer, are subject to the same pressure jump. In a decompression step, a pressure of 5 bar (here 5 bar by using nitrogen) is applied and is then released to ambient pressure. After equilibration, the gas is used to initiate a compression step. During the pressure jumps, constant temperature is achieved by active compensation of the heat changes. Integration of the supplied power yields the heat released or consumed. The heat peaks in the compression and decompression steps should be equal in value, but are of opposite sign. Thermodynamics relates the pressure coefficient (∂Qrev/∂p)T of the heat Qrev exchanged in a reversible process to the coefficient of thermal expansion, α = (1/V)(∂V/∂T)p, of the sample volume. For a sufficiently dilute solution containing mS grams of solute and m0 grams of solvent, the volume is given by V = m0V0 + mSV̄S, where V0 is the specific volume of the solvent and V̄S the partial specific volume of the solute. One then finds (∂Qrev/∂p)T = −TVα = −T[m0V0α0 + mSV̄SS].α0 = (1/V0)(∂V0/∂T)p and S = (1/V̄S)(∂V̄S/∂T)p are the coefficients of thermal expansion associated with the solvent volume and the solute partial volume, respectively. In a differential PPC experiment, the volume occupied by the solute in the sample cell, mSV̄S, is replaced by the same volume of solvent in the reference cell. Within small pressure intervals, the pressure dependence of V and α can be neglected, and eqn. (1) can be integrated to yield the working equation ΔQrev = −T[mSV̄SS − mSV̄Sα0]Δp Then, S can be determined, if α0 is known. In practice, it is a good approximation, to replace the partial specific quantities S and V̄S by the corresponding apparent quantities, in the following denoted by α and V, respectively. Thus, all volumetric properties refer to apparent molar volumes and apparent expansibilities of the protein.
Dielectric spectroscopy monitors the real part (dielectric dispersion ε′(ω)) and imaginary part (dielectric loss ε″(ω)) of the complex dielectric permittivity of a sample as a function of the angular frequency ω = 2πν. Accounting processes in the optical and vibrational regime by an effective limiting value ε∞ of the real part, the complex dielectric permittivity is given by[32,33,38]ε*(ω) = ε′(ω) − iε″(ω) = ε∞ + Δε′(ω) − iΔε″(ω) + σ/iε0ω (i2 = −1).ε″(ω) shows a low-frequency divergence which depends on the static (dc) conductivity, σ. The latter contribution can be determined at low frequencies, where only the σ/iε0ω term survives. ε0 is the permittivity of the vacuum. The relaxational contributions, Δε′ and Δε″(ω), depend on the same set of microscopic parameters.
In the first set of experiments, we used the network analyzer 8712 ES from Agilent Technology (former Hewlett Packard) in combination with a coaxial line terminating in a home-made sample cell described by Kaatze and coworkers[39] to measure ε*(ω) in the range from ν = 1 MHz to ν = 1.3 GHz. In the second series, the network analyzer HP 8720 (Hewlett Packard) was used at 200 MHz ≤ ν ≤ 20 GHz in combination with the commercially available probe HP 85070B.
Results and data evaluationFig. 1 shows the background-corrected DSC traces of the buffered RNase A solution in H2O and D2O at a protein concentration of 0.5 wt.%. In H2O, the protein begins to unfold at about 55 °C, and the temperature of the unfolding midpoint is Tm = (62.0 ± 0.2) °C. Table 1 summarizes the resulting thermodynamic data. The enthalpy change, obtained by integration over the DSC peak, is ΔH = (435 ± 4) kJ mol−1. The increase in the apparent molar heat capacity from the native to the unfolded state, ΔCp = (5.2 ± 0.2) kJ mol−1 K−1, is typical for proteins in H2O. Substitution by D2O shifts Tm to (66.2 ± 0.2) °C. The enthalpies of unfolding in H2O by D2O are almost the same, but ΔCp in D2O is only about half of that in H2O.
Fig. 1 shows that the maxima in Cp at the unfolding transition correspond to minima in the temperature dependence of the apparent thermal expansion coefficient, α(T). Tm values extracted from the PPC curves agree within experimental error with those determined from the DSC traces. Volumetric data derived from the PPC curves are summarized in Table 1. In the native regime the PPC curves in both H2O and D2O show two major features: First, the apparent thermal expansion coefficient, α, of the protein is high, for example α10 = 0.76 × 10−3 K−1 at 10 °C in H2O. Second, there is a steep slope in the temperature dependence of α(T). In H2O between 10 and 40 °C α decreases almost linearly with a slope dα/dT = −4.3 × 10−6 K−2. The curve of α(T) in D2O lies above that of H2O, and α is more steeply decreasing with increasing temperature. Quantitatively, α in D2O is by about 17% higher than in H2O, and the negative slope, dα/dT, is enhanced by 46% (Table 1). In both H2O and D2O, α increases upon unfolding. The increase, Δα = 1.6 × 10−4 K−1, is typical for proteins. The relative volume change upon unfolding can be calculated by integration of the α(T) transition curve after baseline subtraction. In H2O, we obtain ΔV/V = −2.7 × 10−3. Based on the partial specific volume of RNase A (0.704 mL g−1), the absolute volume change is ΔV = −26 mL mol−1, which amounts to 0.27% of the total protein volume. In D2O, the volume change is 40% below the value found for H2O (Table 1).
Dielectric relaxation dataWe have determined dielectric relaxation spectra of solutions of RNase A (1.5 wt.%) and ubiquitin (2.5 wt.%) in H2O and D2O at 25 °C. The RNase A concentration was chosen as a compromise between the desire to reproduce the conditions of the PPC experiments and the need for sufficient amplitudes of the protein peaks in the spectra. Fig. 2 shows as an example the real and conductivity-corrected imaginary parts of the permittivity of RNase A in H2O. The spectra exhibit a multi-modal structure with diffusive, so-called “Debye behavior” of each dispersion step, characterized by amplitudes Sj and relaxation times τj:[38] For an accurate representation of the spectrum of RNase A, four terms were needed. Numbering the dispersions from low to high frequencies, processes 1 and 4 are dominant and cause the apparent bimodal shape of the spectrum shown in Fig. 2. The low-frequency dispersion, often termed “β-relaxation”, results from protein tumbling, the high-frequency dispersion 4 reflects bulk water reorientation.[32–37] The interpretation of two further processes (2 and 3) at intermediate frequencies is subject to some debate, but probably they are related to protein–water interactions. At the low concentrations applied here, their amplitudes were too weak to be able to extract meaningful parameters. Processes 2 and 3 will not be considered further. In the case of ubiquitin, it made only sense to fit a three-term expression, as the difference in frequency of dispersions 2 and 3 was small. The fit parameters are summarized in Table 2.
A scenario for the volumetric behaviourBefore considering changes in the volumetric behavior induced by H2O–D2O substitution, we discuss a simple scenario, in which the partial specific volume of a protein may be decomposed into three contributions:[40–42]V ≈ Vintr + δVhydr + Vtherm.The intrinsic volume, Vintr, of the protein results from the van der Waals volume of the atoms plus the volume of water-inaccessible voids in its interior. The hydrational or interaction term, δVhydr, describes the solvent volume changes associated with the hydration of the solvent-accessible hydrophobic, polar or charged protein atomic groups. The thermal volume, Vtherm, the volume of the void space surrounding the solute molecule, arises from mutual thermally induced vibrations and reorientations of the solute and solvent. Scaled particle theory, by employing statistical mechanical and geometrical arguments to describe the dissolution of a solute, allows one to evaluate the intrinsic and thermal contributions ([ref. 42] and references therein). The thermal volume contribution has been found to be slightly larger in D2O than in H2O, respectively.[42] The sum of the intrinsic (geometrical) volume and the thermal volume thus represents the partial molar volume of the cavity enclosing the solute. In eqn. (5), a minor term taking into account the coefficient of isothermal compressibility of the solvent has been neglected.[40] Certainly, as there is no rigorous way of disentangling the partial molar volume of a protein into its components here, other dissections of V may be conceivable. Owing to the qualitative discussion of the various contributions, this does not affect the conclusions drawn, however.
From the measured partial specific volume and simple models for Vintr and Vtherm, a rough estimate of δVhydr can be given for the native state. Such an analysis, conducted for RNase A in H2O in[ref. 30], yields as a highly negative value for δVhydr, about δVhydr = −0.2 cm3 g−1. A negative value implies a smaller partial molar volume, i.e. a higher density, of water at the protein surface compared to bulk water. Such a higher density is evidenced by combined neutron and X-ray scattering experiments.[43] Its origin in terms of the properties and topology of the protein–water surface has recently been addressed by MD simulations.[8]
Eqn. (5) implies that a similar dissection may hold for the temperature derivatives of the partial specific volume, i.e. the apparent thermal expansion coefficient, α, and its temperature coefficient, dα/dT. Only δVhydr and Vtherm contribute significantly, to α and dα/dT, because the intrinsic volume of the native protein does not depend markedly on temperature. In fact, the thermal expansivity of the protein interior has been measured over a limited temperature range, and the changes observed are rather small.[44–48] The thermal volume is expected to increase with temperature, giving a positive contribution to α. As pointed out by Lin et al.,[29]α and, in particular, dα/dT are primarily controlled by the hydrational contributions, suggesting that dα/dT is a direct measure of the effect of solvation upon volumetric properties of proteins. In fact, the hydrational contribution to the thermal expansibility is known to depend drastically on the nature of the protein–water interface. Hydrophilic groups in contact with water show the characteristic pattern of structure-breakers with large positive values of α, which decrease drastically with increasing temperature. The rationale is that the hydrophilic groups bind more adjacent water at low temperatures, which at higher temperatures are released by thermal agitation, and do no more contribute to α. In contrast, solvent-exposed hydrophobic groups act as structure makers, resulting in a decrease in the water density around hydrophobic groups. In this case α is negative, and the temperature coefficient, dα/dT, is positive. This picture adopted here for proteins, also evolves from studies of volumetric properties of inorganic and organic electrolytes and small model compounds such as hydrophilic and hydrophobic amino acids (see [refs. 41,42]).
Our data for RNase A in H2O and D2O indicate high apparent thermal expansion coefficients of the protein and steeply decreasing slopes in their temperature dependence. Thus, they classify RNase A as a protein with a significant number of hydrophilic side groups at the protein surface.[30] Even much steeper slopes of dα/dT are found for proteins with more charged side groups, such as SNase[31].
Volumetric behavior and protein stability in D2OWater deuteration quite generally displaces the transition temperature of unfolding, Tm, to higher values (Table 1).[17–21] Intuitively, this temperature shift is attributed to a higher stability of the native protein in D2O. However, the transition temperature only signals the equality of the Gibbs free energy of the native and unfolded state which results from enthalpy-entropy compensation. By a more detailed analysis of entropy and enthalpy behavior, some authors have concluded that at low temperatures the native protein is destabilized by D2O.[20,21] Guzzi et al[21]. have argued that at the microscopic level this destabilization in D2O originates from a less efficient solvation of solvent-exposed apolar side groups.
For RNase A in D2O, both the absolute value of the thermal expansion coefficient at low temperatures, as well as its temperature coefficient, are markedly higher than in H2O (Table 1). In particular, the large isotope effect on dα/dT is notable. These volumetric effects of H2O–D2O substitution are in the same direction as obtained for proteins with a significant number of charged groups, for instance SNase,[31] and also observed by addition of kosmotropic co-solvents such as glycerol or sorbitol.[30] This evidences a prime role of the enhanced solvation of the hydrophilic amino acid groups by D2O, and clearly points towards a stabilization of the native form by D2O. This would also be in agreement with the value of the Gibbs free energy of unfolding at room temperature, ΔG°(25 °C), deduced from the calorimetric data by assuming that the heat capacity change upon unfolding is independent of temperature. In that case, ΔG°(25 °C) values of 37 and 46 kJ mol−1 K−1 are obtained in H2O and D2O, respectively.
Finally, the volume change, ΔV, that accompanies the unfolding in D2O, is 40% lower than in H2O (Table 1). A considerable part of this reduction may arise from a strong temperature dependence of ΔV, because as Tm increases, ΔV decreases significantly.[48,49] It might, at least partially, also be due to a more compact state of RNase A in the unfolded state, however.
Protein stabilization as inferred from dielectric relaxation dataAn intriguing question is, whether the increased hydration and stabilization in D2O seen in the thermal and volumetric experiments leads to a change in the protein fold. Our strategy is to compare the isotope effect upon the protein relaxation time, τ1, with the isotope effect upon the bulk viscosity, and also to compare the amplitudes S1 of the protein peaks in H2O and D2O, respectively. Both types of information can be converted into structural data: τ1 contains information on the hydrodynamic radius of the protein, and S1 permits the calculation of the protein's dipole moment. We have conducted these experiments for RNase A and, for comparison, also for ubiquitin.
As noted, the relaxation time τ1 reflects protein tumbling. This process is considered to be a prototypical example for a process controlled by the hydrodynamic friction of the solvent,[32,33] and can be described by Debye's equation for the rotation of a dipolar sphere in a surrounding medium of viscosity η[38]Vhydr is the hydrodynamic volume of the protein, a the hydrodynamic radius, and kB is Boltzmann's constant. Because large deviations from spherical shape are needed to invalidate eqn. (6),[38] this relation is also applicable to RNase A, which can be pictured as an ellipsoid with an axial ratio of about 1.5. The viscosity of the buffer solutions were found to differ less than 1% from those of pure H2O, ηH = 0.894 × 10−3 kg m−1 s−1 at 25 °C, and of pure D2O, ηD = 1.101 × 10−3 kg m−1 s−1, respectively. Thus their ratio is ηD/ηH = 1.232,[24] where the subscripts H and D stand for H2O and D2O, respectively. Table 3 shows that, within experimental error, this ratio is indeed observed for the bulk water relaxation times τ4. For the protein relaxation times,τ1, this ratio is significantly smaller.
Before discussing this isotope effect in detail, we compare the hydrodynamic radius deduced for RNase A with estimates from other sources. The most direct comparison is possible with 15N magnetic relaxation data. Magnetic relaxation monitors protein tumbling, as described by correlation functions associated with second rank (l = 2) spherical harmonics. Dielectric relaxation refers to first rank (l = 1) spherical harmonics. In the diffusive limit of single-particle reorientation, we have τ1 = 3τNMR. From 15N relaxation data for RNase A by Cole and Loria,[50] we then estimate τ1 = 20 ns and a ≅ 1.90 nm, in comparison with τ1 = 27 ns and a ≅ 2.10 nm actually observed. The ifference is beyond experimental uncertainty, but one may note that the applied conversion, τ1 = 3τNMR, holds only for processes reflecting single-particle motions. For many simple systems, including water, collective reorientational motions cause the dielectric relaxation time τ1 to differ substantially from 3τNMR.[51] We have shown elsewhere[35] that protein-protein dipolar correlations lead to a concentration dependence of τ1 which can account for the discrepancy of dielectric and NMR data.
In all cases, the hydrodynamic radii are larger than the radii estimated from structural data. Molecular mass–volume correlations for globular proteins, for example given in [refs. 40 and 54], predict an intrinsic radius of bare RNase A of 1.54 nm only (a thermal contribution of about 1 nm may be added). Similar discrepancies between hydrodynamic and structural radii are found for other proteins as well,[32,33] and have been commonly attributed to a shell of hundreds of water molecules contributing to the hydrodynamic radius of the protein. A common assumption is that these water molecules are bound on the time scale of protein reorientation, i.e. in the microseconds time regime. This result is highly inconsistent with results of NMR exchange studies[13,52,53] and MD simulations,[54,55] which clearly show that residence times of water molecules at protein surfaces are only on the sub-nanosecond time scale. The apparent differences between hydrodynamic and structural radii can be removed by computing the friction tensors from the protein shape on an atomic scale,[56,57] and by assuming a non-uniform solvent viscosity near the protein surface.[52]
Turning to the effect of H2O–D2O substitution, we find the viscosity ratio of ηD/ηH = 1.232 to be reflected by the ratio of the relaxation times τ4 of bulk water, but not by the ratio of the protein relaxation times τ1 of RNase A and ubiquitin. There are two options for rationalizing this observation. First, as already noted, one expects a non-uniform solvent viscosity near the protein surface.[52] Thus, one possible explanation may presume that, owing to differences in hydration, the local isotope effect ηD/ηH differs from that in the bulk solvent. Alternatively, one may rationalize the observed behavior by assuming that, upon H2O–D2O substitution, the hydrodynamic radii of RNase A and ubiquitin shrink by about 5%.
Two observations indicate that the isotope effects may indeed reflect a change in the hydrodynamic radius, and that this effect results from changes in the protein fold. First, it has been found in an elaborate small-angle neutron scattering (SANS) study[43] that the radius of gyration of lysozyme decreases from Rg = 1.38 nm in H2O to 1.24 nm in D2O. The order of magnitude of this effect (11%) is even larger than that observed here. In SANS, the contrast at the protein–water interface is higher than obtained by small-angle X-ray scattering (SAXS), and the radius of gyration essentially refers to the bare protein.[43] Thus, the change in the radius of gyration may well result from a change in the protein fold. In a more indirect way, such an effect has also been deduced from phosphorescence lifetime studies of several proteins.[25]
Second, an interpretation in terms of a more compact fold in D2O is consistent with the behavior of the amplitudes S1 of the protein tumbling mode, which provide values for the effective dipole moments, μeff, of the proteins in solution (which differ from the dipole moments of the isolated proteins). The evaluation of S1slightly depends on details of the dielectric model. For example, the Onsager–Oncley model[32,33] yields values of 276 D and 175 D (1 D = 3.30 × 10−30 C·m) for RNase A and ubiquitin, respectively. Such high dipole moments are typical for proteins.[32,33] Any model yields, however, the same expression for the amplitude ratio where ρ is the number density of the protein molecules. Table 3 indicates that μeff,D is by about 5–10% smaller than μeff,H, indeed indicating a more compact state of the protein in D2O, in which the charge distribution is spread over a smaller space than in H2O.
H2O–D2O substitution has been found to be an effective tool for singling out hydration effects on the thermodynamic behavior of proteins and their stability. In particular, the volumetric behavior reflected by the apparent thermal expansion coefficient, α, and its temperature coefficient, dα/dT, changes drastically upon isotopic substitution. These changes are intrinsically related to hydration properties and the interplay between hydrophilic and hydrophobic groups at the protein–water interface. The volumetric data clearly point toward a stabilization of the native form of RNase A in D2O.
An assessment of the underlying molecular processes can certainly not been done by considering the thermal and volumetric properties alone. As noted by Finney,[4] in studying the subtle effects associated with protein hydration, as many techniques as possible should be applied. Here, we have complemented the DSC and PPC experiments by dielectric relaxation experiments for RNase A and ubiquitin. These experiments yield evidence that the stabilization of the native form in D2O reflects a more compact fold. A more compact fold in D2O follows also from SANS data for another protein, lysozyme.[43] Presumably, the stronger solvation of hydrophilic groups and less effective solvation of hydrophobic groups at the protein surface cause the protein to reduce the solvent-exposure of the hydrophobic groups. These changes, in particular for enzymes with highly flexible surface groups, may very well alter the protein function in D2O.