Stabilizing Salt-Bridge Enhances Protein Thermostability by Reducing the Heat Capacity Change of Unfolding

Most thermophilic proteins tend to have more salt bridges, and achieve higher thermostability by up-shifting and broadening their protein stability curves. While the stabilizing effect of salt-bridge has been extensively studied, experimental data on how salt-bridge influences protein stability curves are scarce. Here, we used double mutant cycles to determine the temperature-dependency of the pair-wise interaction energy and the contribution of salt-bridges to ΔCp in a thermophilic ribosomal protein L30e. Our results showed that the pair-wise interaction energies for the salt-bridges E6/R92 and E62/K46 were stabilizing and insensitive to temperature changes from 298 to 348 K. On the other hand, the pair-wise interaction energies between the control long-range ion-pair of E90/R92 were negligible. The ΔCp of all single and double mutants were determined by Gibbs-Helmholtz and Kirchhoff analyses. We showed that the two stabilizing salt-bridges contributed to a reduction of ΔCp by 0.8–1.0 kJ mol−1 K−1. Taken together, our results suggest that the extra salt-bridges found in thermophilic proteins enhance the thermostability of proteins by reducing ΔCp, leading to the up-shifting and broadening of the protein stability curves.


Introduction
To survive in the hot habitats, proteins from thermophilic organisms are more thermal stable than their mesophilic homologs.The conformational stability of proteins is defined as the free energy difference between the native and the unfolded states, or the free energy of unfolding (DG u ).DG u varies with temperature as a curve function (i.e. the protein stability curve), which is described by the Gibbs-Helmholtz equation: where T m is the melting temperature, DH m is the enthalpy change of protein unfolding at T m , and DC p is the heat capacity change of unfolding.
Nojima and co-workers pointed out that protein thermostability, or increase in T m , can in theory be enhanced by: (i) up-shifting, (ii) broadening, and (iii) right-shifting of the protein stability curves [1].Nussinov and co-workers studied the correlation between different thermodynamic parameters of 5 protein families and showed that thermophilic proteins prefer to increase T m by upshifting and broadening of their protein stability curves [2].In a later study, Razvi and Scholtz systematically compared the protein stability curves of 26 thermo-and mesophilic homologous pairs of proteins.Regardless to the physical origins, they showed that over 70% of thermophilic proteins in their study achieve higher T m by up-shifting and/or broadening of their protein stability curves as compared with their mesophilic homologous [3].It is clear that most thermophilic proteins achieve higher thermostability by upshifting and broadening of their protein stability curves.
The Gibbs-Helmholtz equation predicts that a smaller DC p can up-shift and broaden a protein stability curve.For example, the curvature of the protein stability can be defined as the second derivative of the Gibbs-Helmholtz equation: for T s is the temperature where DG u is maximum [2].A reduction in DC p will make the curvature less negative and, therefore, the protein stability curve is broadened.Similarly, it can be shown that a smaller DC p can increase the maximum DG u : DG u (T s ) = DH m 2DC p (T m 2T s ), or in other words, the protein stability curve is up-shifted [2], provided that DH m is increased or remains constant.Consistent with the observation that most thermophilic proteins achieve higher thermostability by up-shifting and broadening of their protein stability curves, thermophilic proteins tend to have a much smaller value of DC p than their mesophilic homologs [4][5][6][7][8][9][10][11].For example, we have shown that the thermophilic ribosomal protein L30e from Thermococcus celer has a DC p value of ,5 kJ mol 21 K 21 , which is much smaller than the value of ,10 kJ mol 21 K 21 obtained for the mesophilic L30e from yeast [12].
In thermophilic proteins, one common strategy to enhance thermostability is to have more favorable surface charge-charge interactions.When compared with their mesophilic homologues, thermophilic proteins have more surface charged residues [13] and have an increased number of salt bridge [14][15][16].The stabilizing role of the electrostatic interaction was first suggested by Perutz and Raidt based on their modeling studies [17,18], and was experimentally verified by various strategies including optimization of surface charges [19,20], removal of surface charges [21,22], addition of new ion pairs [23,24], and double mutant cycles [23,[25][26][27][28][29][30][31][32][33][34].To study the contribution of charge-charge interactions to the thermostability and the reduced DC p of thermophilic proteins, our group had systematically removed 26 surface charges on T. celer L30e by single charge-to-alanine substitutions.Most of the mutants results in decreases in T m [21], indicated that the surface charges are mostly stabilizing in thermophilic protein.In another study, we showed that removal of favorable charge-charge interaction by single charge-to-neutral substitutions increases the DC p value [12].
Here, we demonstrated that stabilizing salt-bridges enhance the thermostability of proteins by reducing the DC p .We used the double-mutant cycle to investigate the effect of pair-wise interaction of two salt bridges (E6/R92 and E62/K46) on protein thermostability and DC p .We showed that the two salt-bridges stabilized the T. celer L30e protein by ,2-5 kJ mol 21 , and the stabilizing effect was insensitive to temperature changes from 298-348 K.The contribution of the two salt-bridges to DC p was determined independently by Gibbs-Helmholtz and Kirchhoff analyses.Our results showed that each salt-bridge contributed to a reduction of DC p by 0.8-1.0kJ mol 21 K 21 .That salt-bridge reduces DC p provides a structural basis for the large differences in DCp observed between thermophilic and mesophilic proteins.

Design of variants
In this study, we used the double-mutant cycle to investigate how salt-bridges contribute to the thermostability of proteins.We have selected two salt-bridges (E6/R92 and K46/E62), which are located on opposite sides of T. celer L30e (Figure 1).Charged residues were substituted with alanine.For Arg and Lys residues that have long side chain, substitutions to Met were made to mimic their long hydrophobic side chains.As a result, two double-mutant cycles were applied for each salt-bridge.As a negative control, we have also used double-mutant cycles to study the pair-wise interaction between E90 and R92, which have a long separation distance of 10.8 A ˚.A total of seven single mutants and six double mutants were generated (Table S1).

Pair-wise Interaction energy between charge residues was determined by double-mutant cycles
Single charge-to-neutral substitutions suffer from the limitation that the residue being substituted may also form other interactions with the rest of the proteins.By canceling out these interactions using the double-mutant-cycle approach, one can estimate the contribution of the pair-wise interaction between the two oppositely charged residues in a salt-bridge [23].The scheme presented in Figure S1 explains how the pair-wise interaction energy is determined by the double-mutant-cycle approach.For the theoretical background on the use of double-mutant cycle to determine the pair-wise interaction energy of salt-bridges, please refer to the work of Fersht and co-workers [23].In brief, if pairwise interaction exists between two oppositely charged residues, the DDG u for removing a negative charge from the wild-type protein (process A) should be smaller than that from M 2ve in which the positive charged residues has been substituted in prior (process B) (Figure S1).It is because in addition to the interaction made by the negative charge residue to the rest of the protein, the process A also removes the pair-wise interaction.Similar arguments could be applied to the DDG for process C and D. We obtained the pair-wise interaction energy between the two charge residues (DDG int ) by: We have determined the free energy of unfolding (DG u ) of the wild-type T. celer L30e and its variants by urea-induced denaturation at 298 K (Table 1), and calculated the values DDG int for the cycles E6A/R92A(M), E62A/K46A(M), and E90A/R92A(M) (Figure S1).The values of DDG int were in the range of 1.9-3.6 kJ mol 21 for the pairs of charged residues (E6/ R92 and E62/K46) involved in salt-bridges.In contrast, the values of DDG int were close to zero for the control pairs (E90/ R92).Taken together, our results suggest that the two saltbridges of E6/R92 and E62/K46 contributed favorably to the stability of L30e.Salt-bridges are stabilizing and their interaction energies are insensitive to temperature changes Next we investigated the temperature dependency of the pairwise interaction energy.The measurement of DG u was extended to 308, 318, 328, 338 and 348 K (Table 1).The values of DDG int were determined accordingly and summarized in Table S2 and Figure S2.Within each double-mutant cycle, there is no significant difference among the DDG int values obtained at different temperatures, and between those derived from R/KRA cycles and from R/KRM cycles.On the other hand, the DDG int values for different pairs of charge residues were significantly different from each other.The average values of DDG int for the salt-bridges E6/R92 and E62/K46 were 2.360.3 and 3.960.3,respectively, while the value for the control pair was 0.660.3kJ mol 21 .Our results suggest that the salt-bridges E6/R92 and E62/K46 are stabilizing, and the pair-wise interaction energy appears to be independent of temperatures.

DC p is reduced by pair-wise interaction of salt-bridges
Values of DG u at temperatures 298-348 K and their T m values were fitted to the Gibbs-Helmholtz equation to obtain the values of DC p (Figure 2 and Table 2).Compared to the wild-type value of 5.3 kJ mol 21 K 21 , DC p were increased to 5.7-6.8kJ mol 21 K 21 for substitutions (E6A, E62A, K46A/M, R92A/M) that break a salt-bridge interaction.On the other hand, for the E90A substitution that did not break any salt-bridge, there was no significant change in the value of DC p (Table 2).These results suggest that single substitutions that break a salt-bridge would increase the values of DC p .
To address the question if the pair-wise interaction of saltbridges affects the values of DC p , we determined the DDC p(int) by double-mutant cycle in an analogy to the determination of the DDG int (Figure 3A).Take the double-mutant cycle of E6A/R92A as an example (Figure 3B).Removal of a negative charge by E6A substitution from the wild-type L30e resulted in an increase of DC p for 0.8 kJ mol 21 K 21 .On the other hand, the difference in DC p between R92A and E6A/R92A was only 20.1 kJ mol 21 K 21 .These data suggest that the two substitutions are not independent, and the pair-wise interaction between E6A and R92A affects the DC p .
Similar to the argument for the determination of DDG int , we have DDC p(int) = [DC p (DM)2DC p (M 2ve )]2[DC p (M +ve )2DC p (WT)] (Figure 3A).The values of DDC p(int) for the six double-mutant cycles were determined by the double-mutant cycle (Table 3 and Figure 3B).The values of DDC p(int) for the control cycle, E90A/ R92A(M), were close to zero (20.1 to 20.3 kJ mol 21 K 21 ).In contrast, for the cycles, E6A/R92A(M) and E62A/K46A(M), that involves breakage of a salt-bridge, values of DDC p(int) were from 20.8 to 21.0 kJ mol 21 K 21 .The negative values of DDC p(int) strongly suggest that the pair-wise interaction of salt-bridges reduces the DC p .
To further confirm the hypothesis that the pair-wise interaction of salt-bridge contributes to the reduction of DC p , we have determined the values of DC p independently by the Kirchhoff analysis [35][36][37].Values of T m and DH m at pH 2.5-6.0 for L30e and its variants were obtained by thermal denaturation.DC p values for wild-type and variant L30e were derived from the slope of the DH m vs. T m plot (Figure 4), and summarized in Table 2.The DC p value for wild-type L30e was 3.960.2kJ mol 21 K 21 .For substitutions (E6A, K46A, E62A, R92A) that break a saltbridge, the DC p values were increased to 4.6-4.9kJ mol 21 K 21 (Table 2).On the other hand, for E90A substitution that did not break any salt-bridge, the DC p value was 4.060.1 kJ mol 21 K 21 , which was similar to that of wild-type L30e.
We noticed that DC p values obtained using the Kirchhoff analysis based on thermal denaturation data were smaller than those using Gibbs-Helmholtz analysis based on chemical-induced denaturation experiments.This observation is consistent with our previous report, in which we pointed out that the systematic differences in DC p values were probably due to the thermal denatured state having more residual structures than the chemicalinduced denatured state [12].
Regardless of the systematic differences in DC p values, the values of DDC p(int) determined by the Kirchhoff analysis were in striking agreement with those obtained by the Gibbs-Helmholtz analysis (Table 3).For the double-mutant cycles involving the breakage of a salt-bridge, the values of DDC p(int) were 20.860.4 and 20.960.5 kJ mol 21 K 21 for E6A/R92A and E62A/K46A, respectively.In contrast, the DDC p(int) was close to zero for the control cycle E90A/R92A (20.260.4 kJ mol 21 K 21 ).Taken together, our results suggest that the pair-wise interaction of saltbridge reduces the DC p by ca.0.8-1.0kJ mol 21 K 21 .

No major structural changes were observed in the double charge-to-Ala variants
The crystal structures of the E6A/R92A, K46A/E62A, and E90A/R92A variants were determined at resolution ranging from 1.8 to 2.0 A ˚(Table S3).The structures of all these variants can be superimposable with the wild-type structures (Figure S3).The root-mean-square deviations (r.m.s.d.) between C a atoms of the wild-type L30e and its variants were ,0.5 A ˚(Table S3), suggesting there were no major structural change in these variants.

Discussion
Whether salt-bridge contributes to protein stability is controversial, and is probably context dependent [23,30,[38][39][40][41][42].Elcock proposed that salt-bridge should be more stabilizing at high temperatures because the unfavorable desolvation penalty [43][44][45] and the entropic cost of fixing two charged side-chains [33,46,47] would decrease with temperatures [48].Here, we used the doublemutant-cycle approach to study how salt-bridge contributes to the thermostability of proteins.The two salt-bridges, E6/R92 and E62/K46, stabilizes the protein by ,2-5 kJ mol 21 (Figure S2 and Table S2).That values of DDG int for R/KRA and R/KRM cycles were similar suggests that the stabilization is mainly due to the charge-charge interaction, rather than hydrophobic interaction, between the salt-bridging residues.We showed that the pairwise interaction energy, DDG int , is insensitive to temperature changes (Figure S2).This observation is consistent with a previous study by Ge and co-workers [26], which showed that the pair-wise interaction energies of salt-bridges in a hyperthermophilic protein Ssh10b at 298 and 353 K were similar.Since the free energy of unfolding is decreasing with temperatures, the more-or-less constant stabilizing effect of salt-bridges should contribute more in proportion to the overall protein stability at high temperatures.
We further demonstrated unambiguously that the stabilizing salt-bridges reduce the heat capacity change of unfolding (DC p ).We showed that single-substitutions that break a salt-bridge increased the DC p value.This observation is consistent with our previous report in that removal of favorable electrostatic interactions by single charge-to-neutral substitutions increases the DC p [12].Using the double-mutant-cycle approach, we determined the values of DDC p(int) , which estimates how much the pair-wise interaction between the salt-bridging residues contributes to the heat capacity change of unfolding.For the double-mutant cycles that break a salt-bridge (i.e.E6/R92 and E62/K46), negative values of DDC p(int) suggest that the pair-wise interaction of the salt-bridges reduces the DC p by 0.8 to 1.0 kJ mol 21 K 21 .
Using guanidine-induced denaturation and Gibbs-Helmholtz analysis, we have previously showed that the DC p for the mesophilic L30e from yeast (10.5 kJ mol 21 K 21 ) was much larger than that for the thermophilic T. celer L30e (5.3 kJ mol 21 K 21 ) [12].It is in fact a common observation that thermophilic proteins tend to have smaller values of DC p than their mesophilic homologues [4][5][6][7][8]10,11].Here, we demonstrated by double-mutant cycle that the pair-wise interaction between the salt-bridging residues reduces the DC p , which provide a structural basis of why thermophilic proteins have smaller values of DC p .This conclusion is consistent with our previous observation that removal of favorable charge-charge interactions by single substitutions resulted in increases in DC p [12].Using a simple spherical model, Zhou predicted that favorable interaction between two oppositely charge residues should decrease DC p [49].Our experimental results provide unambiguous evidence supporting the conclusion that stabilizing salt-bridge reduces the DC p .The structural basis of why thermophilic proteins have smaller values of DC p is controversial.It has been well documented that DC p correlates well with the changes in solvent accessible surface area (DASA) upon unfolding [50][51][52].As we have pointed out previously, due to the similarity in their native conformation, homologous proteins tend to bury similar amount of ASA upon folding assuming the denatured states are random coil [12].To explain the differences in DC p between thermophilic and mesophilic proteins, it has been proposed that thermophilic proteins may have more residual structures in their denatured states so that the DASA would be smaller than that calculated for a random-coil [53,54].However, it is uncertain if the differences in residual structures, if any, can explain the large differences in DC p observed.Moreover, Zhou pointed out that the presence of more residual structures may increase the free energy of the denatured state and destabilize the protein, which is counter-intuitive to the fact that thermophilic proteins are more stable than their mesophilic homologs [49].Apparently, the correlation of DC p to DASA fails to account for the large differences in DC p commonly observed for thermophilic and mesophilic pairs of homologous proteins [8,9], suggesting that factors other than the hydration effect may also contribute to DC p .
Our results showed that the DH m for the wild-type protein was slightly higher than that for the variants (Figure 4).Under this condition, having a smaller DC p always enhances protein thermostability by up-shifting and broadening the protein stability curve.Figure S4 simulates the shape of the protein stability curve of two hypothetical proteins with DC p values of 5.3 and 7.3 kJ mol 21 K 21 .The simulation shows that a decrease of DC p by 2 kJ mol 21 K 21 shifts the protein stability upward, and increases its maximum stability by ,10 kJ mol 21 .It also broadens the protein stability curve so that the protein remains stable at a wilder range of temperatures.Our previous study also showed T. celer L30e has an up-shifted and broadened protein stability curve when compare with that of the mesophilic yeast homologues [12].
In a survey of 26 protein families where thermodynamics data were available for both mesophilic and thermophilic homologs, Razvi and co-workers found that most protein enhances their thermostability by up-shifting and broadening of the protein stability curves [3].Since thermophilic proteins tend to have more salt-bridges than their mesophilic homologs [14][15][16], our observation that salt-bridge reduces DC p may provide a general   mechanism for enhancing thermostability -thermophilic proteins have more stabilizing salt-bridges that reduce the DC p , leading to the up-shifting and broadening of the protein stability curve.

Site-directed mutagenesis
All site-directed mutagenesis were performed by a two-stage PCR procedure modified from the QuikChange site-directed mutagenesis protocol using the mutagenic primers listed in Table S1 [55].Wild-type T. celer L30e cloned in expression vector pET3d (Novagen) was used as the template in all polymerase reactions.Mutations introduced were confirmed by DNA sequencing.

Protein expression and purification
All protein samples were expressed and purified as described [21,56].
Thermal-induced denaturation 20 mM protein samples were dialyzed in 10 mM sodium acetate buffer at pH 5.4 for 16 hours before circular dichroism (CD) measurement.After degassing thoroughly, all protein samples were heated in a securely stoppered 1 mm path-length cuvette from 298 K to 383 K at a heating rate of 1 K min 21 .The thermal denaturation was then monitored by molar ellipticity at 222 nm using a JASCO J810 spectropolarimeter equipped with a peltiertype temperature control unit.
The melting temperature (T m ) and enthalpy of unfolding DH m were obtained by fitting the thermal denaturation curve to a twostate model (Figure S5): where y obs is the observed molar ellipticity at 222 nm; y n and m n are the y-intercept and slope of the pre-transition baseline; y u and m u are the y-intercept and slope of the post-transition baseline; R is the gas constant; T is the temperature in Kelvin.

Urea-induced denaturation
20 mM protein samples were equilibrated with 0 M-10.2M urea in 10 mM sodium acetate buffer at pH 5.4 for 30 minutes before CD measurement.Concentration of urea was determined from refractive index measurements [57] using Leica AR200 refractometer.The urea-induced denaturation was monitored by molar ellipticity at 222 nm using a JASCO J810 spectropolarimeter equipped with a peltier-type temperature control unit.The urea-induced denaturation was analyzed by a two-state model [58] (Figure S6):   variants of L30e (E6A, K46M, E6AR92A, and E6AR92M) that exhibited irreversible thermal denaturation, values of apparent T m were used.The program PRISM (GraphPad Software, La Jolla, USA) was used to estimate the errors in DC p due to the uncertainty in DG u .

Estimation of DC p by Kirchhoff analysis
Thermal-induced denaturation curves were measured for protein samples of T. celer L30e in 10 mM sodium citrate/ phosphate buffer at pH 2.5 to 6.0.T m and DH m were obtained from by fitting the data to the two-state model described above.DC p values were then obtained from the slope of the DH m vs. T m plot.Only the data obtained from reversible thermal denaturation were included in the Kirchhoff analysis.

Crystal structure determination
Crystals of L30e variants were grown by sitting-drop-vapordiffusion method at 289 K. 2 ml of 10 mg ml 21 protein sample was mixed with 2 ml of precipitant solution (Table S3).Datasets were acquired and collected at 100 K using an in-house rotating anode X-ray source.The diffraction data were processed, merged, scaled, and reduced with programs (MOSFLM, SCALA, TRUN-CATE) from the CCP4 suite [60].The structures were solved by molecular replacement using PHENIX with the wild-type T. celer L30e crystal structure (PDB code: 1H7M) as the search model.The structures were refined using PHENIX [61], and were validated using WHATCHECK [62] and MOLPROBITY [63,64].

Figure 2 .
Figure 2. The protein stability curves of T. celer L30e and its variants.Values of DG u at 298-348 K were obtained by urea-induced denaturation experiments for the variants of L30e in the double-mutant cycles (A) E6A/R92A, (B) E6A/R92M, (C) E62A/K46A, (D) E62A/K46M, (E) E90A/ R92A, and (F) E90A/R92M.Values of DG u for the wild-type L30e are shown in circles, ERA variants in squares, R/KRA/M variants in diamonds, and the doubly-substituted variants in triangles.Values of DG u together with T m were fitted to the Gibbs-Helmholtz equation to obtain values of DC p .doi:10.1371/journal.pone.0021624.g002

Figure 3 .
Figure 3. Determination of DDC p(int) by double-mutant cycle analysis.The scheme shown in panel (A) is in analogy to that used to calculate DDG int in Figure S1.(B) DDC p(int) for all six double-mutant cycles analyzed.The substitutions are indicated inside the boxes.The values of DDC p for processes A-D were shown along the arrows, and the values of DDC p(int) were shown in the middle of the cycles.All values are in kJ mol 21 K 21 .doi:10.1371/journal.pone.0021624.g003 y obs ~(y n zm n ½D)z(y u zm u ½D)e {DG (D) =RT 1ze {DG (D) =RT where y obs is the observed molar ellipticity at 222 nm; y n and m n are the y-intercept and slope of the pre-transition baseline; y u and m u are the y-intercept and slope of the post-transition baseline; R is the gas constant; T is the temperature in Kelvin; [D] is the concentration of urea; DG (D) is the free energy change of unfolding at [D].The free energy change of unfolding without denaturant, DG u , was obtained by linear extrapolation model [58]: DG (D) = DG u 2m[D], using the average m-value approach[59].DG u at 298 K, 308 K, 318 K, 328 K, 338 K, and 348 K were measured for T. celer L30e and its mutants.
Estimation of DC p by Gibbs-Helmholtz analysisDG u at temperatures from 298 K to 348 K and T m were fitted to the Gibbs-Helmholtz equation to obtain the values of DC p .For

Figure 4 .
Figure 4. Determination of DC p by the Kirchhoff analysis.DH m and T m were obtained at pH 2.6-6.0 by thermal denaturation for the variants of L30e in the double-mutant cycles (A) E6A/R92A, (B) E62A/ K46A, and (C) E90A/R92A.Values of DH m were plotted as a function of T m for T. celer L30e and its variants.DC p was obtained by the slope of the plot.doi:10.1371/journal.pone.0021624.g004

Figure S1
Figure S1 Coupling energies (DDG int ) were determined by double-mutant cycles.(A) The scheme explaining how DDG int are calculated from values of DG u for wild-type (WT), single-mutants (M +ve and M 2ve ), and double-mutant (DM) by the double-mutant cycle analysis.(B) DDG int for all six double-mutant cycles analyzed.The substitutions are indicated inside the boxes.The values of DDG u for processes A-D were shown along the arrows, and the values of DDG int were shown in the middle of the cycles.All values are in kJ mol 21 .(PDF) Figure S2 Temperature dependency of the coupling energy.Values of DDG int derived from double-mutant cycles (A) E6A/ R92A(M) (circles), (B) E62A/K46A(M) (squares), and (C) E90A/ R92A(M) (diamonds) at temperatures 298 K to 348 K are shown.

Table 1 .
Free energy of unfolding (kJ mol 21 ) of T. celer L30e and its variants at 298-348 K.

Table S1
Oligonucleotide primers used in the mutagenesis.(DOC)