Not Always Sticky: Specificity of Protein Stabilization by Sugars Is Conferred by Protein–Water Hydrogen Bonds

Solutes added to buffered solutions directly impact protein folding. Protein stabilization by cosolutes or crowders has been shown to be largely driven by protein–cosolute volume exclusion complemented by chemical and soft interactions. By contrast to previous studies that indicate the invariably destabilizing role of soft protein–sugar attractions, we show here that soft interactions with sugar cosolutes are protein-specific and can be stabilizing or destabilizing. We experimentally follow the folding of two model miniproteins that are only marginally stable but in the presence of sugars and polyols fold into representative and distinct secondary structures: β-hairpin or α-helix. Our mean-field model reveals that while protein–sugar excluded volume interactions have a similar stabilizing effect on both proteins, the soft interactions add a destabilizing contribution to one miniprotein but further stabilize the other. Using molecular dynamics simulations, we link the soft protein–cosolute interactions to the weakening of direct protein–water hydrogen bonding due to the presence of sugars. Although these weakened hydrogen bonds destabilize both the native and denatured states of the two proteins, the resulting contribution to the folding free energy can be positive or negative depending on the amino acid sequence. This study indicates that the significant variation between proteins in their soft interactions with sugar determines the specific response of different proteins, even to the same sugar.

as well as non-ideal mixing terms.S1,S2 The main advantage of using the FH theory is that the chemical potentials of all solution components can be expressed using a small set of parameters, specifically the excluded volume parameter, ν, the non-ideal mixing parameter, χ, and its non-ideal enthalpic and entropic contributions χ = χ H − χ T S .
The FH mixing free energy, Eq. 1 of the main text, is: where ϕ S and ϕ C are the solvent and cosolute volume fractions, M 0 = (N S + νN C ) is the total volume of solution (expressed in terms of number of lattice sites that solvent and cosolute inhabit), k is the Boltzmann constant, and T is temperature.∆G can be dissected into an enthalpic and entropic contributions: and the solvent and cosolute chemical potentials, µ S and µ C , can be derived as µ S/C = ∂∆G ∂N S/C P,T,N C/S (S4)   where P is pressure.We further extend the FH theory to ternary mixtures (cosolute, water, and protein) by dividing the mixture into two domains: the protein and bulk domains.The protein domain is defined as the volume in the vicinity of the protein surface that is buried upon folding, and the bulk spans the part of the mixture unperturbed by the protein.Specifically, the surface area of the protein domain corresponds to the change in solvent accessible surface areas upon folding, ∆SASA = SASA N −SASA D , and the extent of the protein domain into solution is defined by the length scale set by the cosolute, a = ν 1/3 .The values of ∆SASA for both proteins (MET16 and AQ16) were calculated using the ProtSa server S3,S4 and given in Table S1.
In our model, protein folding corresponds to the burial of a section of the protein surface, translating to the removal of some of the solvent accesible surface from contact with the mixture.This surface removal releases cosolute and solvent molecules that were previously confined to the protein domain into the bulk.Therefore, the folding free energy can be calculated from the difference between the mixing free energies with and without the removed protein surface in the mixture.The mixing free energy excluding the buried surface is given by the standard FH expression, Eq. 1, whereas the mixing free energy with the surface is comprised of the FH and an additional protein domain mixing free energy expressions.
For the protein domain, we modify the mixing free energy compared to the bulk in two ways.First, the mixing volume in the protein domain is reduced from M surf to M mix on account of the cosolute excluded volume that restricts the number of available lattice sites.Consequently, we rescale the volume fractions in the protein domain, where ϕ mix S/C and ϕ surf S/C are the rescaled and unscaled protein-domain volume fractions, respectively.Second, we add a soft interaction term between the cosolute and the protein surface.This term is expressed using the parameter ε, that describes the effective protein-cosolute interaction per exposed protein surface area in terms of free energy.We note that this constant is similar to the interaction term also introduced by Schellman.S5 These modifications to the free energy result in the following mixing free energy associated with the protein domain: The contribution from the term containing ε can be either stabilizing or destabilizing.Specifically, repulsive soft-interactions, ε > 0, result in a stabilizing contribution to the protein's native state, whereas attractive soft-interactions, ε < 0, result in a destabilizing contribution.Moreover, like χ, ε can in principle depend on temperature, with corresponding enthalpic and entropic terms, ε = ε H −ε T S .We have previously shown that this temperature dependence of the soft interaction is necessary to fully describe an enthalpically stabilizing and entropically destabilizing cosolute.S6 Eq.S6 is useful only as long as the volume fractions of solvent and cosolute in the protein domain are known.To determine ϕ surf S/C , we numerically solve the equilibrium condition: where µ bulk S/C are the bulk chemical potentials and µ surf S/C are the chemical potentials in the protein domain.The chemical potentials are calculated using Eq.S4 with Eq. 1 for the bulk or with Eq.S6 for the protein domain.Eq.S7 is solved with respect to ε for a specific set of ∆SASA, ν, and χ.

S1.3. Density Measurements and Determination of ν
Values for the excluded volume parameter, ν, were determined from density measurements of binary solutions.For each cosolute, the density of solutions for at least seven different cosolute concentrations were measured using a DMA 5000 M (Anton Paar) densitometer at 25°C.Each cosolute's concentration ranged from 0 (pure water) to near saturation, or up to ∼ 4.8 molal, the lower of the two.The measured densities as a function of cosolute concentration are in Fig. S1A and Table S6.The molar volume of the mixture, V , is defined as Where ρ is the density in units of gr/ml, X S and X C are the solvent and cosolute molar fractions, and M S and M C are the solvent and cosolute molar masses.Fig. 4B shows the molar volumes for different mixtures as a function of the cosolute molar fraction, X C .The linear relation that the data shows for V versus X C indicates that mixture molar volume is a linear combination of concentration-independent cosolute and solvent partial molar volumes, VC and VS , so that V = VS + X C VC − VS .The values of VS and VC can then be determined for each cosolute by extrapolating to infinite dilution or to the pure cosolute limits: VS = lim Then, ν is calculated as the ratio of the solvent and cosolute partial molar volumes, ν = VC VS

S1.4. Water Activity Measurements and Determination of χ
The non-ideal interaction parameter, χ, for each cosolute is determined from measurements of water activity, a S .Water activities of aqueous cosolute mixtures were measured at different temperatures using an AQUALAB 4TE activity meter as follows.The activity meter inner temperature controller was initially set to either 15, 20, 25, or 45°C.Once the sample chamber temperature had settled, the activity meter offset was calibrated using standard salt solutions.For each measurement, ∼3ml of aqueous cosolute mixture was loaded into a measurement plate that was then inserted into the sample chamber and the chamber was closed.After the sample temperature was allowed to equilibrate, the activity was measured.Each reported value represents and average over at least three repeats, Tables S7 -S12.
The measured water activities were converted to osmotic pressure using χ was then determined from fits to the FH expression for the osmotic pressure, Π: S7 Fig. S2A shows the FH fits to the osmotic pressure as function of volume fraction for each cosolute at 25°C.In addition, in order to verify the accuracy of our measurements, we also measured Π as a function of ϕ C , on an APRO 5520 Wescor osmometer at ambient temperature (with the room temperature set to 25°C), after calibration using standard salt solutions.The measured osmotic pressures are shown in the insets of Fig. S4 and in Table S13.Each data point represents the average of at least three repeats.
Finally, we further dissected χ into its enthalpic, χ H , and entropic, χ T S , contributions, as follows.First, we determined the values of χ for several temperatures as described above, see fits in Fig. S4.Then, the values of χ as function of T for each cosolute were fitted to either a linear function or to the following Padé approximant: The value of χ T S at 25°C was then calculated as χ T S = χ H − χ.Values of χ and χ T S are plotted in Fig. 4B, C and given in Table S3.

S1.5. Circular Dichroism Measurements and Determination of ε
Circular dichroism (CD) spectra of AQ16 and MET16 in aqueous solutions were measured for a range of cosolute concentrations and temperatures using a J-810 spectrophotometer (JASCO, Japan).Aqueous samples containing either ∼ 50µM AQ16 or ∼ 100µM MET16 were prepared gravimetrically from lyophilized powders and the protein concentrations were determined by following the absorbance of the single tyrosine residue at 274nm.The pH was maintained using 20mM phosphate buffer at pH 7. The samples were measured in a 1 or 2mm path length quartz cell (Starna).
To follow the effect of cosolute concentration on the protein folding equilibrium, the protein samples were titrated using concentrated cosolute stock solutions.The CD spectra of the titrated protein solutions were measured in quintuplicates from 190 or 200 to 260nm with 0.1nm steps at 25°C.The spectra were baseline-corrected by background subtraction of spectra of a solution with the same cosolute concentration but in the absence of protein.In addition, the ellipticities at 222nm for AQ16 and 215nm for MET16 were recorded and accumulated over 90 seconds.These accumulated ellipticities were used to determine the concentrations of the proteins' native, c N , and denatured, c D , states from the limiting ellipticity values of AQ16's and MET16's native and denatured states.
For MET16, the native ellipticity of −10, 342deg cm 2 dmol −1 was measured in 50%wt MeOH-water mixture, S8 while a signal of 0 was used for the fully denatured protein.S9-S11 For AQ16, the denatured ellipticity is 640deg cm 2 dmol −1 , while the native ellipticity depends on the protein length and is given by, θ N = θ 0 N (1 − 2.5/n), where θ N is the protein's native state ellipticity, θ 0 N = −42, 500deg cm 2 dmol −1 is the native state ellipticity of an infinitely long protein, and n = 16 is the number of AQ16's residues.S12-S14 The folding free energy was then determined from the native and denatured states concentrations by, where R is the gas constant.Additionally, temperature scans were conducted at several cosolute concentrations.For each cosolute concentration the reversibility of the folding process was ensured by taking a full spectra measurement (190 or 200 to 260nm) in the beginning and end of each scan at 25°C.Temperature scan experiments ranged from 5 or 10°C to 60°C in 5°C steps and the folding free energy versus temperature was determined from accumulated ellipticities as described for the titration experiments.
The value of ε is determined by fitting the change in folding free energy, ∆∆G 0 versus cosolute concentration, Fig. 3A, B of the main text.The fitting is preformed by solving the equilibrium condition, Eq.S7, with ε as the only variable, and using the previously determined values of ν and χ (sections S1.3 and S1.4).The resulting values of ε for AQ16 and MET16 with all cosolutes are in Fig. 5 and Table S4.
We dissect ∆G 0 into its enthalpic and entropic contributions by fitting the experimental ∆G 0 versus temperature to the integrated Van 't Hoff equation, Eq. 3, at different cosolute concentrations, Fig. 3C, D of the main text.The resulting change in folding enthalpy and entropy due to added sugars, ∆∆H 0 and ∆∆S 0 , were further fitted to our mean-field model to resolve the enthalpic and entropic contribution to the soft-interaction parameter, ε H and ε T S , respectively.Here, the value of χ T S as derived from experiments is added as an additional parameter, and the fit is performed with ε T S as variable.

S1.6. Determination of Uncertainties in ε
The uncertainties of ε and ε T S in Figs. 5, 8, and S8 and Table.S4 were determined by subsampling following the Monte Carlo cross-validation (MCCV) methodology.MCCV is usually used in validating the accuracy of fitted models, but it also allows estimation of variability of model parameters.S15,S16 In accordance with MCCV, values from the experimental set of ∆∆G 0 for a protein-cosolute pair were randomly sampled and assigned to a smaller training set that comprises two-thirds the full ∆∆G 0 set.By fitting our model to this training set we derived an estimate for the value of ε.This process was repeated N = 100 times, generating N estimates from which the uncertainties were determined.A similar procedure was applied for ε T S with N = 6 considering the smaller data set of ∆∆H 0 and T ∆∆S 0 .

S1.7. Concentration Dependence of ∆C P
To determine the enthalpic and entropic contributions to the folding free energies we have fitted the experimental ∆G 0 versus temperature using Eq. 3, the integrated van 't Hoff equation.In the fits shown in Fig. 3C, D we have assumed that the change in heat capacity upon folding, ∆C P , is independent of the identity and concentration of the cosolute, an assumption verified as follows.
Fig. S3A shows ∆C P derived from fits to the data for AQ16 with each cosolute versus cosolute concentration, treating ∆C P in Eq. 3 as a fitting parameter.Values of ∆C P for the different cosolutes and all concentrations fluctuate with rather small and evenly distributed deviations around the mean value, ⟨∆C P ⟩.The positive value for the folding heat capacity of AQ16, ⟨∆C P ⟩ = 0.476±0.006kJ/molK,and the corresponding negative value for MET16, −0.408±0.007kJ/molK,reflect the different thermodynamic folding mechanisms, as detailed in the main text.Fig. S3B shows the van 't Hoff fits to ∆G 0 of AQ16 in presence of trehalose (same experimental data as in Fig. 3C) derived in two ways: (i) fits that treat ∆C P as a fit parameter (as in panel A), shown as dark curves.(ii) fits to ∆G 0 where ∆C P is fixed to its mean values, ⟨∆C P ⟩ (full black line in Fig S3A ), shown in the lighter colored curves.The two ways of fitting resulted in ∆H 0 and T ∆S 0 that are almost identical, Fig. S3C.We conclude that treating the value of ∆C P as a constant for all cosolutes at all concentrations does not change our results within experimental error.

S3. Measured Solution Densities
Densities of aqueous cosolute binary solutions as a function of the cosolute molality at 25°C are given in Table S6.These values correspond to the densities shown in Fig. S1A.The cosolutes' molality was determined gravimetrically, section S1.3.

S5. AQ16's CD Spectra in Presence of Trehalose
Fig. S5 shows the CD spectra of AQ16 for different concentrations of the dissacaride trehalose.The isodichroic point at 201 ± 1nm that is observed in temperature variation measurements, Fig. 2A, is conserved in the presence of trehalose, indicating that the structures of AQ16's native and denatured states do not change significantly upon addition of cosolutes.ν , ∆∆G 0 χ , and ∆∆G 0 ε to the folding free energy versus cosolute size, at a cosolute concentration of 1 molal.The stabilizing contribution of ν and destabilizing contribution of χ invariably scale with the cosolute size in both proteins.By contrast, ∆∆G 0 ε does not simply scale with cosolute size.For AQ16, ∆∆G 0 ε becomes more destabilizing with size for the monosaccharides and sorbitol but levels off for the larger disaccharides.For MET16 ∆∆G 0 ε changes from destabilizing for glycerol to stabilizing for the monosaccharides and sorbitol.For the larger disaccharides, trehalose and sucrose, the stabilizing contribution of ε decreases to almost zero.

S7. Enthapic and Entropic Soft-Interaction Contributions are Strongly Compensating
Using our model, we can dissect the enthalpic and entropic contributions to ∆∆G 0 ν , ∆∆G 0 χ , and ∆∆G 0 ε .Fig. S7 shows a representative example of these enthalpic and entropic contributions for the dissacaride trehalose.Although the trends in protein stability seen in Fig. S7 are also reflected in the ∆∆G 0 of Fig. 7, the mapping of ∆∆G 0 onto the enthalpyentropy plane is useful, since it allows to visually distinguish the enthalpic and entropic terms and to asses the degree of their compensation.
The contributions of the excluded volume interactions are purely entropic, ∆∆G 0 ν = −T ∆∆S 0 ν , as they must always be, see vertical red lines.By contrast, the non-ideal mixing term is composed of a favorable enthalpic component, ∆∆H 0 χ < 0, and an unfavorable entropic component, T ∆∆S 0 χ < 0. The non-ideal mixing contributions, blue curves, are destabilizing since T ∆∆S 0 χ < ∆∆H 0 χ , and therefore reside below the diagonal that corresponds to ∆∆G 0 = 0.
The purple curves in Fig. S7 show that the stabilizing effect of the joint contribution of ν and χ, T ∆∆S 0 ν + T ∆∆S 0 χ versus ∆∆H 0 χ , is mostly entropic with a smaller stabilizing enthalpic contribution.Interestingly, the enthalpy-entropy signature of the sum contribution of ν and χ (but excluding the contribution of soft interactions) resembles the thermodynamic fingerprint that is often measured for polymeric crowders, S20,S21 suggesting that soft interactions may play a lesser role in large polymeric crowders.
The enthalpy-entropy curves for the soft interaction contribution, ∆∆H 0 ε − T ∆∆S 0 ε , vary from residing above the diagonal for MET16 to below the diagonal for AQ16.This difference between proteins seen in the entropy-enthalpy plot of the soft interaction contribution corresponds to the positive ε for MET16 and negative ε for AQ16.Nevertheless, we find that the soft interactions enthalpy-entropy curves are consistently close to the diagonal because the contributions of ε H and ε T S are strongly compensating.S8.Values of Protein-Sugar ε H Fig. S8 shows ε H for AQ16 and MET16 with the sugar cosolutes.We find that ε H > 0 for all cosolutes with both proteins, i.e., the enthalpic contribution to the soft interaction is repulsive.

S9. Simulation Details and Validation
The choice of a suitable simulation force field is crucial to properly describe the delicate interplay of forces between the solvent, cosolute, and protein in simulation, and to relate these molecular interactions to the experimental results.S22,S23 Here we use the modified CHARMM36 force field developed by Cloutier et al S24 because it reproduces the concentration dependence of ∆Γ S for proteins in presence of carbohydrates, S24 indicating that the forces between solvent, cosolute, and protein are well represented.
Molecular dynamic simulations were performed using the GROMACS package S25 and the TIP3P water model.S26 The Nosé-Hoover thermostat was utilized for temperature coupling, S27,S28 particle-mesh Ewald was employed for electrostatic calculations, S29,S30 and the LINCS method was used to fix the bond lengths to hydrogen atoms.S31 Van der Waals interactions were truncated smoothly with a switching distance of 10 Å and a cutoff distance of 13 Å.All simulations ran for 200ns and converged after 100ns, as demonstrated by the protein radius of gyration, Fig. S9.All subsequent analyses were carried out on the second half of each trajectory.We validated our force field selection by comparing ∆Γ S determined from simulations to ∆Γ S determined from model fits to experiments, Table S14.Γ S of the native and denatured states in simulation are determined from the relation, )   where N S , N C , n S and n C are the number of solvent (water) and cosolute (sugar) molecules in the bulk and protein domain, respectively.Eq.S12 converges asymptotically to the value of Γ S as the distance to boundary between protein and bulk domains,r, increases, Fig. S10A, B. S32-S36 The difference in preferential hydration ∆Γ S is calculated as the difference between the preferential hydration of the native and denatured states, Γ S,N -Γ S,D .Although in simulation, MET16's values of ∆Γ S are more negative than the experimentally determined values, Table S14, the values show the correct trend with respect to sugar identity and are in good qualitative agreement with the experimental ∆Γ S .Similarly, AQ16's values of ∆Γ S from simulations are more negative than the experimental values, yet the trend with sugar identity is different from experiments, with trehalose showing less negative ∆Γ S than sucrose in simulations but more negative ∆Γ S in experiments.Thus, we find that for MET16 the simulation is qualitatively in good agreement with the experiments and for AQ16 the sign of ∆Γ S determined from the simulations is in agreement with the experiments.with the backbone's carbonyl and amide nitrogen groups, Fig. S11A.By contrast, the same water-backbone interactions destabilize AQ16's native state, Fig. S11C.
For interactions with the side-chains, the increased stabilization of MET16 mainly stems from water Hbonds with the charged amine of the lysine (sc NpW) and the hydroxyl groups (sc HW) of serine, threonine, and tyrosine, mainly because of their high abundance in MET16.The interactions of water with lysine's charged amine groups is shown to even become favorable in MET16's native state in presence of sugars, further adding to the native state stability, Fig. S11B.For AQ16, the modest contribution from the side-chains mainly originates in water Hbonds with glutamine's amide groups, while the Hbonds of water with the tyrosine's hydroxyl group are practically the same for the native and denatured states, Fig. S11D.However, this modest stabilizing contribution from interactions of water with glutamine's amide groups is insufficient to overturn the destabilizing Hbonds that include AQ16's backbone.

Figure S1 :
Figure S1: Determination of cosolute partial molar volume.(A) Measured densities of aqueous cosolute mixtures at 25°C.(B) Molar volume of binary solutions, calculated using Eq.S8.Solid lines show linear fits to the data.

Figure S2 :
Figure S2: Determination of the FH non-ideal mixing parameter, χ. (A) Fits to the FH scaled osmotic pressure, Eq.S10.(B) Van 't Hoff analysis of the non-ideal mixing parameter, χ.

Figure S3 :
Figure S3: Concentration Dependence of ∆C P for AQ16.(A) Values of ∆C P versus cosolute concentration.The mean change in heat capacity, ⟨∆C P ⟩, is shown by the horizontal line.(B) AQ16's folding free energy versus temperature for different trehalose molal concentrations.Curves are fits to the integrated van 't Hoff equation, Eq. 3, with ∆C P as a free (relaxed) fitting parameter or with constant ∆C P = ⟨∆C P ⟩. (C) AQ16's folding enthalpies and entropies for different trehalose concentrations from the fits in panel B.

Figure S4 :
Figure S4: Scaled osmotic pressure versus volume fraction of (A) glycerol, (B) glucose, (C) galactose, (D) sorbitol, (E) trehalose, and (F) sucrose.Lines are fits to the FH osmotic pressure, Eq.S10.Insets compare the osmotic pressure that is derived from activity and osmometry measurements at 25°C.

Figure S5 :
Figure S5: CD spectra of AQ16 for different trehalose concentrations at 25°C.

Fig. S6
Fig.S6shows the contribution of ∆∆G 0 ν , ∆∆G 0 χ , and ∆∆G 0 ε to the folding free energy versus cosolute size, at a cosolute concentration of 1 molal.The stabilizing contribution of ν and destabilizing contribution of χ invariably scale with the cosolute size in both proteins.By contrast, ∆∆G 0 ε does not simply scale with cosolute size.For AQ16, ∆∆G 0 ε becomes more destabilizing with size for the monosaccharides and sorbitol but levels off for the larger disaccharides.For MET16 ∆∆G 0 ε changes from destabilizing for glycerol to stabilizing for the monosaccharides and sorbitol.For the larger disaccharides, trehalose and sucrose, the stabilizing contribution of ε decreases to almost zero.

Figure S6 :
Figure S6: The free energy contributions, ∆∆G 0 ν , ∆∆G 0 χ , and ∆∆G 0 ε versus cosolute size, ν.The subscript i indicates the type of contribution associated with ν, χ, or ε.Data points represent values derived from model fits to data at cosolute concentration of 1 molal for (A) AQ16 and (B) MET16.The lines are used as a guide to the eye.

Figure S8 :
Figure S8: Values of ε H for MET16 and AQ16 with sugar cosolutes.

Figure S10 :
Figure S10: Change in preferential hydration parameter, ∆Γ S , due to protein folding from simulations of proteins in presence of trehalose or sucrose.(A) ∆Γ S of MET16.(B) ∆Γ S of AQ16.

Figure S11 :
Figure S11: Dissected changes in Hbond free energy in presence of sugars.MET16's Hbond free energy for (A) backbone and (B) side-chains, and AQ16's Hbond free energy for (C) backbone and (D) side-chains.The schemes of MET16 and AQ16 in panels B and D highlight the side-chains that contribute most to the change in Hbond free energy.MET16's charged lysine amines are in orange and hydroxyls of serine, threonine, and tyrosine are in yellow.AQ16's amid group (carbonyl and nitrogen) of glutamine are in cyan and the hydroxyl of tyrosine are in purple.The background corresponds to the highlighted groups with the same colors.

Table S1
contains the values of ∆SASA of the proteins, TablesS2 to S4contain the determined mean-field model parameters, and TableS5contains the experimental ∆Γ S for AQ16 and MET16 with each cosolute.

Table S1 :
Change in solvent accessible surface area, ∆SASA, upon protein folding.

Table S2 :
Excluded volume parameter, ν. a,b a Figures in brackets indicate the fit error.b Ratio of cosolute to water (0.018 L/mol) partial molar volumes.

Table S3 :
Non-ideal mixing parameters, χ. a a Figures in brackets indicate the fit error.

Table S4 :
Soft interaction parameter, ε. a a Figures in brackets indicate the error, determined by Monte Carlo analysis, see section S1.6.

Table S5 :
Change in preferential hydration coefficient, ∆Γ S , determined from model fits to the experimental ∆∆G 0 at c = 1molal.

Table S6 :
Densities of aqueous mixtures at 25°C.

Table S7 :
Water activity of aqueous glycerol mixtures at different temperatures.a Figures in brackets indicates the fit error. a

Table S8 :
Water activity of aqueous glucose mixtures at different temperatures.a a Figures in brackets indicates the fit error.

Table S9 :
Water activity of aqueous galactose mixtures at different temperatures.a a Figures in brackets indicates the fit error.

Table S10 :
Water activity of aqueous sorbitol mixtures at different temperatures.a Figures in brackets indicates the fit error. a

Table S11 :
Water activity of aqueous trehalose mixtures at different temperatures.a a Figures in brackets indicates the fit error.

Table S13 :
Cosolute osmotic pressure from vapor pressure osmometer given in Osmolal.a a Figures in brackets indicates the fit error.

Table S14 :
Changes in preferential hydration parameter, ∆Γ S , in presence of trehalose and sucrose from simulations and model fits to experiments.Figures in brackets indicate the standard deviation in Γ S as determined in simulation.a Values for c = 1molal

Table S16 :
Abbreviations used for Hbond pairs.Protein groups that participate in Hbonds are grouped as either backbone (bb) or side chain (sd).Hbond pairs that are found only in MET16, AQ16, or in both are shaded red, blue, and green, respectively.