More hydrophobic proteins have more compact denatured states

Auton et al.³⁶ showed that urea m values could be predicted using the ΔG_tr values from water to 1M urea measured for the backbone group and the side chains of proteins. In their original analysis, 36 proteins were included. We have analyzed the three proteins in Table ​TableII in the same way, and the combined results for these 39 proteins are shown in Figure ​Figure6.6. There is a good correlation between the measured and predicted m values: m (exp) = (0.96 ± 0.03) (m (calc)) with a 0.92 correlation coefficient. However, it is clear that some of the measured m values deviate substantially from the predicted values. To predict the m values, it is necessary to assume a model to calculate the ASA for the denatured state. The model assumed was the average of the upper and lower bound models suggested by Creamer et al.⁴¹ (These models will be discussed further below.) It seems likely that the predicted m values that deviate to the greatest extent do so because the proteins unfold to a greater or lesser extent than the model. Using the measured m values and those predicted for the upper and lower bounds,³⁶ we can estimate the extent of unfolding for each of the 39 proteins. This approach shows that apomyoglobin (12% unfolded) is the least and staph nuclease (90% unfolded) the most unfolded proteins. This difference in extent of unfolding is striking and will be discussed further.

Open in a separate window

Figure 6

Experimental m values as a function of calculated m values for 39 proteins. The m values were calculated as described by Auton et al.³⁶ Three of the m values are from Table ​TableII and the rest are from Ref. 36.

In Table ​TableIVIV at the top, we list five proteins that are on average 80% unfolded. We chose not to use a fragment (Nank 4–7, 71% unfolded) of the full-length notch ankyrin domain because it is partly unfolded under native state conditions. At the bottom, we list five proteins that are on average 34% unfolded. We chose not to use apomyoglobin (12% unfolded) because the native protein binds a heme group; cytochrome c (21% unfolded) because of the presence of the heme groups, and von Willebrand factor A3 domain (26% unfolded) and ribonuclease A (36% unfolded) because they both contain disulfide bonds. The ten proteins in Table ​TableIVIV are all monomeric proteins without disulfide bonds that closely approach a two-state folding mechanism. Their urea m values were measured at pH 7 ± 1.

Uversky et al.¹³ published one of the first papers that attempted to predict whether a protein would be “natively unfolded” under native state conditions. (For the remainder of the paper, we will refer to these as IDPs.) They showed that the key characteristics were a low overall hydrophobicity and a large net charge. For their sample of 91 IDPs, the mean hydrophobicity was 0.39 ± 0.05 and the mean charge per residue was 0.12 ± 0.09. For their sample of 250 folded proteins, the mean hydrophobicity was 0.48 ± 0.03 and the mean charge per residue was 0.04 ± 0.04. In the extensions of these prediction procedures,⁶³ these have remained important parameters. They are also important in predictions of amyloid formation.⁶⁴ We reasoned that these same two characteristics might be important in determining the extent of unfolding of proteins in urea.

In Table ​TableIV,IV, we compare characteristics of the proteins that unfold to the smallest extent with those that unfold to the largest extent. The proteins range in size from 56 to 159 residues, and size does not appear to be an important factor. Similarly, Gly and Pro contents do not appear to be an important characteristic. None of the proteins that are the most unfolded have a net charge per residue as large as the average for the IDPs, and the averages for the two groups of proteins are very similar. Thus, the net charge per residue does not seem to be as important here as it does in the case of IDPs.¹³ (It is important at pH 3, as discussed below.) The two groups of proteins differ significantly in their content of nonpolar residues, and there is a corresponding change in their content of polar and charged residues. We will focus on the nonpolar residues. The average nonpolar residue content is 37% for the most unfolded and 43% for the least unfolded proteins. Two of the proteins stand out: Calbindin contains only 33% nonpolar residues and CheY contains 49% nonpolar residues (Table ​(TableIV).IV). These results encouraged us to examine the differences in hydrophobicity in more detail.

The results of Auton et al.³⁶ show that when a protein unfolds, 75% of the change in accessible surface area, ΔASA, is due to the side chains and 25% is due to the backbone groups. However, the urea m values are dominated by the contribution of the backbone groups (See Fig. ​Fig.44 in Auton et al.³⁶). Cannon et al.³⁵ reached a similar conclusion. So, to explain the results in Table ​TableIV,IV, we need to know why peptide groups are much more accessible to urea in the most unfolded proteins than in the least unfolded proteins. The column giving the percent nonpolar residues from Table ​TableIVIV is shown as the second column in Table ​TableV.V. This is not the best measure of a protein's hydrophobicity because it does not take into account the differences in hydrophobicity among the amino acid side chains. Uversky et al.¹³ used the Kyte and Doolittle scale (K-D)⁷³ to calculate the mean hydrophobicity of their sample of proteins. The K-D scale assigns each amino acid a hydrophobicity on a scale that ranges from −4.5 to +4.5. These values were normalized to a scale ranging from 0 to 1 and used to calculate the total and mean hydrophobicity of each protein. This calculation can be done with the program PONDR that is available at http://www.pondr.com.⁶⁵ We used this program to calculate the mean hydrophobicity of each of the proteins in Table ​TableIVIV and the results are shown in the third column in Table ​TableV.V. The least unfolded proteins are 6% more hydrophobic that the most unfolded proteins, a difference that is considerable smaller than the difference based on just the nonpolar amino acid content (16% higher). If the nonpolar side chains are important in determining the interaction of backbone groups from urea, the K-D scale is probably not a good one to use. For example, it classes the Tyr and Trp side chains as hydrophilic, in contrast to most other scales. For this reason, we also calculated the mean side chain hydrophobicities using six hydrophobicity scales determined in different ways. The results are shown in Table ​TableV.V. For each of the six scales, the least unfolded proteins are substantially more hydrophobic than the most unfolded proteins. As the units on the various scales differ, we have calculated the % difference in mean hydrophobicity based on each scale and these are shown at the bottom of Table ​TableV.V. The % differences range from 10% to 17% with an average of 13%. Staph nuclease is the least hydrophobic protein on all of the scales and either CheY or barstar are the most hydrophobic.

It is hard to envision how this 13% greater hydrophobicity can reduce the exposure of the backbone groups of a protein from 90% to 30%. However, it is now clear that “ ··· hydrophobic interactions remain strong even in 8M urea solution.”¹² If the increased hydrophobicity of the side chains causes pockets of structure to form, they would have to remove 60% of the backbone groups from access to solvent to explain these results. This seems unlikely. This subject will be discussed further below.

The m values double at low pH because charge–charge repulsion leads to DSE expansion

We showed previously that the m values for RNase A, RNase T1, Barnase, and RNase Sa increase markedly when the pH is lowered from 7 to 3.^42,43,46 This has also been observed with staph nuclease⁴⁷ and other proteins.^44,45,48,49 The original explanation was that electrostatic repulsion among the positive charges at low pH causes the molecules in the DSE to expand so that they are able to interact more extensively with urea. More recently, Whitten et al.⁵¹ suggested that the m value is low at pH 7 due to the presence of intermediate states and increases at lower pHs because the concentration of the intermediates decreases. To investigate this further, we prepared a series of variants of RNase Sa that differ in charged group content (Table ​(TableII)II) and measured their m values at pH 3 and 7 (Table III). The pI, net charge, and estimated content of positive and negative charges on the proteins at pH 3 and 7 are shown in Table ​TableII.II. For all of these variants, the number of residues and the hydrophobic residue content is identical.

At pH 3, there is a steep increase in the m value as the net positive charge on the molecule increases (Fig. ​(Fig.7).7). The amino acid substitutions are all on the surface of RNase Sa, so it is not surprising that the folded CD spectra of the variants are similar and they all have RNase activity (data not shown). In addition, the structure of 5K has been determined and the results are presented in the Supporting Information. The structure (PDB ID code 3A5E) is very similar to RNase Sa (1RGG). Consequently, the folded conformations of the variants do not differ significantly from that of RNase Sa. The difference between the m values at pH 7 and 3 increases as the net charge goes up, and for 3K, 6K and 7K, the m value more than doubles. The equation describing the line shown on the figure is m = 0.082 (net charge) + 1.10 with a correlation coefficient of 0.96. It seems unlikely that this pH dependence would be observed if it resulted from a decrease in the concentration of a folding intermediate. It is more consistent with an expansion of the molecules in the denatured state ensemble as the positive charge on the molecules increases. This expansion of the DSE at low pH is supported by electrostatic effect calculations on the unfolded state of staph nuclease by Fitzkee and Garcia-Moreno.⁷⁴

Open in a separate window

Figure 7

Experimental m values as a function of net charge for the charge variants of RNase Sa (Table ​(TableII)II) at pH 7 (•) and pH 3 (□).

At pH 7, the dependence of the m value on net charge is much smaller (Fig. ​(Fig.7).7). The equation describing the results at pH 7 is m = 0.011 (net charge) + 1.02 with a correlation coefficient of 0.87. To aid in understanding these results, Table ​TableIIII shows the net charge and the contribution from positive and negative charges for each of the variants. For wild type RNase Sa (0K), for example, at pH 3 the net charge is 6.8 due to 1.2 (−) and 8.0 (+) charges, but at pH 7 it is −6.6 due to 13 (−) and 6.4 (+) charges. Despite the similarity in the net charges, the m value is much higher at pH 3 (1.6) than it is at pH 7 (0.95). This probably results, in part, because the number of charges is a more balanced at pH 7 and there are both favorable and repulsive electrostatic interactions among the charges at pH 7 and this makes the molecules in the DSE more compact. There is now considerable evidence showing that electrostatic interactions in the DSE can affect both the stability and kinetics of folding of proteins.^75,76

Another observation is interesting. It is clear that an excess of positive charges on the molecules causes a much larger increase in the m value than an excess of negative charges. For example, 7K has a net charge of + 14.4 at pH 3 and the m value is 2.4, but 5REA has a net charge of −13.6 at pH 7 and the m value is 0.9 (Tables II and III). Consequently, in the DSE, there may be greater repulsion among positive charges than among an equivalent number of negative charges. We have also observed a difference in the effect of positive and negative charges on the solubility of RNase Sa. Adding negative charges to the protein surface always increases the solubility, but adding positive charges may either increase or decrease the solubility.⁷⁷ This may have to do with the extent of hydration of charged groups on proteins. As Collins⁷⁸ has pointed out: “ ··· the negative charges on proteins (carboxylates) are strongly hydrated, whereas the positive charges on proteins (derivatives of ammonium) are weakly hydrated.”

For three of the proteins, the m values were measured in the presence of 1M NaCl at pH 3 and 7, and the results are shown in Table ​TableVI.VI. At pH 3, the m values are always lower at the higher salt concentration and the decrease depends on the net charge on the molecule. These results are consistent with the higher salt concentration reducing the electrostatic interactions among the positive charges at pH 3 so that the molecules in the DSE are more compact. At pH 7, the m values do not differ significantly with salt concentration, suggesting that both repulsive and attractive electrostatic interactions are present so the effect of higher salt concentration is smaller.

The results for urea and GdnHCl m values are quite different. For the same three proteins shown in Table ​TableVI,VI, the GdnHCl m values decrease from an average value of 2.6 ± 0.1 kcal/mol/M at pH 7 to 2.3 ± 0.1 at pH 3 (Richard Thurlkill, unpublished results). The GdnHCl m values for RNase A were identical within error at pH 7 and 3, but for RNase T1, the GdnHCl m value is slightly higher at pH 3 than it is at pH 7.⁴² Thus, the pH dependence of the GdnHCl m values is quite different from that of the urea m values. Lim et al.⁷⁹ have suggested that urea but not GdnHCl destabilizes proteins by hydrogen bonding to the peptide group. They suggest that “··· stacking rather than H-bonding is the most likely mode for guanidinium interaction with the peptide group.” This, plus the fact that GdnHCl is a salt, screening charge–charge interactions, may contribute to the differences observed.

All experimental m values can be accounted for using the approach of Auton et al.³⁶

Auton et al.³⁶ measured ΔG_tr values from water to 1M urea for all of the amino acid side chains and the backbone units in a polypeptide chain. These data and an approach developed by Tanford²³ were used to predict the m values shown in Figure ​Figure3.3. We will use the equations shown below and the data used previously³⁶ to calculate m values.

(2)

where is the free energy difference between native (N) and denatured (D) protein when transferred from dilute buffer to 1M urea, GTFE*_i is the corrected transfer free energy values for the side-chains, GTFE^bb is the transfer free energy for the backbone groups, ni is the number of residues of type i in the protein, and and are the side chain and backbone group average fractional change in solvent accessibility of residue i for the N→D reaction (SI Appendix of Ref. 36).

(3)

where ASA_i,j,N and ASA_i,j,D are the native and denatured solvent accessible surface areas of residues in the native and denatured states calculated as described in Auton et al.³⁶ In the denominator, we use the ASA for the side chains and backbone groups in the fully extended chain—the upper bound of Creamer et al.⁴¹—rather than the ASA based on Gly-X-Gly tripeptides used by Auton et al.³⁶ The Gly-X-Gly model overestimates the exposure of the side chains in denatured proteins.⁴¹ In addition, we used ΔG_tr = −43 cal/mol for the backbone group rather than the value of 39 cal/mol used by Auton et al.³⁶ The more negative value is based on model peptides which should be a better model for an unfolded protein than the cyclic diglycine which was used to obtain the other value. We predicted m values for seven proteins using the lower and upper bounds described by Creamer et al.⁴¹ The results are shown in Table VII.

For their upper bound model, Creamer et al.⁴¹ calculated the ASA using an extended conformation of the protein with the backbone dihedral angles set to φ = −120° and Ψ = +120° and side chain torsion angles set to 180°. This is the same approach used by Spolar et al.,⁸³ but Creamer et al.⁴¹ use the average of results from 43 proteins. When applied to individual proteins, the results are generally in reasonable agreement, but, for example, for staph nuclease, the ΔASA from Creamer et al.⁴¹ is 15% higher than the value from Hong et al.⁶² Unfolded proteins are not expected to be as unfolded as this upper bound model.

For their lower bound model, the average ASA values of 17 residue fragments from 43 proteins with the same conformation they have in folded proteins were used.⁴¹ The fragments retain the secondary structure they have in the folded proteins, but there will be no long-range interactions. Unfolded proteins are not expected to be as compact as this lower bound model.

Creamer et al.⁴¹ analyzed the accessible surface area (ASA) of five proteins with from 110 to 275 residues (See their Table ​TableII).II). For these proteins, the ASA is ∼30% greater for the upper bound model than for the lower bound model. (For comparison, the ASA of the Gly-X-Gly tripepetide model is ∼22% higher than the upper bound model.) The % difference in ASA between the upper and lower bounds is greater for the backbone groups than for the side chains. For staph nuclease, for example, the ASA for the upper bound model is 73% higher for the backbone groups, but only a 19% increase for the side chains. This is expected based on the model used for the lower bound where the native α-helices and β-turns will be retained, but longer-range tertiary interaction will not.

The m values for the two proteins that unfold to the smallest extent, barstar and CheY, are close to the m values calculated using the lower bound model; the experimental m values are only 5 and 9% higher (Table VII). This suggests that the proteins with the most compact denatured states approach the lower bound model of the DSE. For the two proteins than unfold to the greatest extent, barnase and staph nuclease, the measured m values are intermediate between the lower and upper bound models. This suggests that at pH 7 they contain substantial structure, but considerably less than barstar and CheY.

For the two proteins that unfold to the greatest extent, barnase and staph nuclease, and for 0K, 5K, and 7K, experimental m values have been measured at pH 3 where the m values are much higher than at pH 7 (Table VII). It is of interest to compare these m values with those predicted using the upper bound model. For barnase and 5K, the experimental m values are in reasonable agreement with the m values calculated using the upper bound model. For RNase Sa (0K), the calculated m value is greater than the experimental value. We think this result because the net charge per residue is much less for 0K than for barnase and 5K (Table VII). This suggests that barnase and 5K approach the upper bound model which has no structure, but that 0K, because of the lower net charge per residue, retains some structure at pH 3.

For staph nuclease, the measured urea m value increases from 2380 at pH 7, to 3100 at pH 4, to ∼4200 at pH 3.7.⁴⁷ The m value cannot be measured at lower pHs because the native state is not stable and the value at pH 3.7 is less certain because only part of the urea denaturation curve can be measured. The urea m value at pH 3 for staph nuclease will clearly be much greater than the predicted m value of 3510 (Table VII).

There is no evidence in Figure ​Figure77 that the m values at pH 3 for the charge variants of RNase Sa are leveling off as the net charge on the molecules increases. The largest measured m value at pH 3 is 2370 for 7K (Table III), but the predicted value is only about 1910 (Table VII). (The structure of 7K has not been determined, but this variant should have a structure similar to RNase Sa and 5K and, consequently, a similar predicted m value. The structure of 5K is reported here. The details are given in the Supporting Information, and the pdb ID is 3A5E.) The fact that the m values predicted using the upper bound model are less than the measured m values at pH 3 for 7K and staph nuclease is puzzling. The upper bound model is a fully extended chain and should give the maximum possible m value for a given protein. There are only a few possibilities to account for this and they are discussed next.

The presence of intermediates would lower the m value so this need not be considered. Another possibility is the effect of an increase in the PPII content, as observed in the CD spectra. Urea promotes the formation of PPII,⁵⁶ and we observe this with the RNase Sa charge variants. For 0K, 5K, and 7K, the content of PPII increases substantially when the urea concentration is raised from 3.5 to 8.5M [Fig. ​[Fig.4(B)].4(B)]. The increase is greatest for 5K and 7K, which have the highest m values. This suggests that when the net charge per residue is high, the DSE contains more PPII than when it is lower. Since the net charge per residue is higher for staph nuclease than any of the other proteins, it may well have an even higher content of PPII. This is a possible explanation for why the m value calculated using the upper bound is less than the measured m value for staph nuclease. Related to this, the ΔG_tr value for a peptide group might be higher if the peptides used for the measurements were in a PPII conformation. The ΔG_tr values were determined in 1M urea and this might not be high enough to induce a PPII conformation in the peptides. A decrease from −43 kcal/mol to −50 kcal/mol would account for the measured m values of 7K and staph nuclease. Auton et al.³⁶ point out that if the transfer free energies of the glycine peptides are corrected for the dipolar ion contribution, the value obtained is −56 cal/mol per backbone unit, and the Pettit group (personal communication) has computed a value of −54 cal/mol so a decrease to 50 cal/mol is a reasonable possibility.

In summary, for the two proteins that unfold to the smallest extent, barstar and CheY, the m values predicted by the lower bound model are in reasonable agreement with the measured m values. For the proteins that unfold to the greatest extent at pH 3, the m values predicted by the upper bound model can account for m values for 0K, 5K, and barnase, and only a moderate increase in the ΔG_tr for the backbone group is needed to account for the m values for 7K and staph nuclease. This is encouraging. It suggests that the wide range of measured m values can be accounted for using the approach of Auton et al.³⁶ and that some of the more complicated explanations that have been suggested may not be necessary.⁵¹

The DSE of proteins at pH 7 must contain substantial amounts of secondary structure

Studies by the Bolen and Record labs suggest that urea m values are determined mainly by the backbone groups of a protein that are exposed to solvent when the protein unfolds.^35,36 Thus, the large difference between the least and most unfolded proteins in Table ​TableIVIV must result from differences in the extent of burial of backbone groups in the DSE. This suggests that the proteins that unfold to the smallest extent contain more structure that reduces their interactions with urea. This would most likely be α-helices, β-sheets, and β-turns. It has been shown that proteins with more compact DSEs contain more secondary structure.^12,84 The results from the previous section suggest that for the least unfolded proteins the content of secondary structure in the DSE is not much less than observed in folded proteins. It seems unlikely that pockets of structure of the type observed in several proteins could account for these results, although in some cases they may play an important role.^85–87 There is little evidence from the CD spectra for the presence of α-helices in the unfolded proteins (Fig. ​(Fig.4),4), perhaps because the helices form transiently in the DSE and will generally be shorter than the helices in folded proteins. However, there is abundant evidence from NMR studies that the denatured states of proteins contain native and nonnative α-helices and other structures that could inhibit the interactions with urea.^11,88,89 Our results indicate that these structures occur to a greater extent in more nonpolar proteins.

For the proteins that unfold to the greatest extent, considerable structure must be present in the DSE at pH 7, but it is clearly less than the amount present in the proteins that unfold to a smaller extent. The results in the previous section suggest that for the proteins with the largest net charge per residue the structure is disrupted by repulsive charge–charge interactions at pH 3. So, the importance of net charge per residue seen in IDPs can also be seen in proteins denatured by urea.

We thank Susan Marqusee and Kath Ratcliff for supplying the most recent urea m values for the RNases H. We also thank Doug Laurents, George Rose, Wayne Bolen, Matt Auton, Luis Holthauzen, Tom Record, and Ruth Saecker for help provided. However, none of them are likely to agree with all of the conclusions in this paper.