INTRODUCTION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION CONCLUSIONS REFERENCES

Oppositely charged residue pairs often form ion pairs in proteins. Ion pairs play important roles in protein structure and function, such as in oligomerization, molecular recognition, domain motions, thermostability, and -helix capping (Perutz, 1970; Fersht, 1972; Barlow and Thornton, 1983; Musafia et al., 1995; Xu et al., 1997a,b; Kombo et al., 2000; Kumar and Nussinov, 2001a,b; Kumar et al., 2000a,b, 2001a,b). Experimental and theoretical estimates of the electrostatic free energy contribution of ion pairs range from highly stabilizing to highly destabilizing (e.g., Warshel and Russell, 1984; Hwang and Warshel, 1988; Sham et al., 1998; Horovitz and Fersht, 1992; Marqusee and Sauer, 1994; Kumar and Nussinov, 1999; Singh, 1988; Barril et al., 1998; Dao-pin et al., 1991; Waldburger et al., 1995; Grimsley et al., 1999; Sheinerman et al., 2000). Continuum electrostatics-based calculations using computer mutations of the salt bridging residue side chains to their hydrophobic isosteres allow estimation of the electrostatic free energy contribution of an ion pair toward protein stability. Using this method, Hendsch and Tidor (1994) have studied 21 ion pairs from nine proteins. They found that most of the ion pairs are destabilizing. However, similar continuum electrostatic calculations have indicated that salt bridges in cytochrome P450cam are highly stabilizing (Lounnas and Wade, 1997). Xu et al. (1997a,b) have shown that hydrogen bonds and salt bridges play a more significant role in protein binding than in folding. The inter-subunit salt bridges in their analysis had stabilizing electrostatic contributions. Using a greater value for effective dielectric constant for the protein interior and a different method, Schutz and Warshel (2001) have shown that the ion pair His 31-Asp70 in T4 lysozyme is not destabilizing.

Kawamura et al. (1997) have shown that the disruption of the Glu34-Lys38 salt bridge in a DNA-binding protein, HU from Bacillus stearothermophilus, reduced its thermal stability. Salt bridges are more frequent in many proteins from thermophiles as compared with their mesophilic homologs (Kumar et al., 2000a; Sterner and Liebl, 2001; Kumar and Nussinov, 2001b). Computational studies have suggested that salt bridges and their networks contribute substantially to protein stability in thermophiles (Elcock, 1998; Xiao and Honig, 1999; de Bakker et al., 1999; Kumar et al., 2000b). Experimentally, surface salt bridges have been shown to stabilize proteins (Spek et al., 1998; Strop and Mayo, 2000).

Recently, we have computed the electrostatic free energy contributions for 222 salt bridges from 36 nonhomologous monomeric proteins (Kumar and Nussinov, 1999) using the continuum electrostatics methodology. Most (~86%) of the salt bridges in these proteins have stabilizing electrostatic free energy contributions, regardless of whether they are buried or solvent exposed, isolated or networked, or contained hydrogen bonds or not. The total electrostatic free energy contribution of a salt bridge depends upon the location of the salt-bridging residues in the protein and the interactions of the salt-bridging residue side-chain charged groups with each other and with other polar as well as ionized groups in the protein. That earlier study indicated that the geometrical orientation of the side-chain charged groups with respect to each other may be a crucial factor in determining the total electrostatic free energy contribution of the salt bridge.

Crystal structures present a static picture of a protein structure. This yields a single side-chain orientation for the salt-bridging residues, with the variability in the position of the side-chain charged group atoms inferred from B-factors. Hence, electrostatic calculations using protein crystal structures classify an ion pair as either stabilizing or destabilizing. The outcome of these calculations also depends on the methodology and the value of dielectric constant used for the protein interior as was recently reviewed by Schutz and Warshel (2001).

Recently, we have performed continuum electrostatic calculations using individual conformers in the NMR conformer ensembles of proteins (Kumar and Nussinov, 2000, 2001a). The calculated electrostatic free energy contribution for a pair of oppositely charged residues often fluctuates and interconverts between being stabilizing and destabilizing. These fluctuations reflect variations in atomic positions of the oppositely charged residue pairs in the protein conformers, even though it is difficult to separate the true protein mobility in solution from artifacts due to NMR data collection and structure calculation protocols (Kumar and Nussinov, 2000, 2001a). Fluctuations in ion pairs and their electrostatic strengths are also seen when different crystal structures for the same protein are compared (Kumar and Nussinov, 2001a). These observations illustrate the sensitivity of the total electrostatic free energy contribution of oppositely charged residue pair to the geometrical positioning of the side-chain charged groups.

Here, we examine the relationship between the geometrical orientation of the side-chain charged groups and the electrostatic free energy contribution of ion pairs. We refer to the total electrostatic free energy contribution by an ion pair to protein stability as the electrostatic strength of the ion pair. The electrostatic strength of an ion pair depends on the ion pair geometry and on the protein structural context. The ion pair geometry denotes the orientation of the side-chain charged groups in the ion pair with respect to each other. It can be characterized by computing the distance between the centroids of the ion-pairing residue side-chain charged groups and the angular orientation of these side-chain charged groups with respect to each other (Kumar and Nussinov, 1999, 2000, 2001a; see also Materials and Methods). The protein structural context of an ion pair refers to the location of the ion-pairing charged residues in the protein along with the identity, number, and spatial distribution of other polar and ionized groups in the protein. In different NMR conformers of a protein, the protein structural contexts are expected to be similar. This facilitates the study of the relationship between geometry and electrostatic strength of the ion pair because the effect due to the variations in the protein structural context on its electrostatic strength is minimized, though not completely removed.

We have studied 22 ion pairs in 14 NMR conformer ensembles of 11 nonhomologous proteins. A total of 1174 computations of ion pair geometry and electrostatic strengths have been performed. Most of these calculations are homologous as they pertain to the ensemble behavior of the 22 unique ion pairs. We classify each ion pair in each conformer as a salt bridge, nitrogen-oxygen (N-O) bridge, or a longer-range ion pair. Salt bridges have the best geometrical orientation of the side-chain charged groups with respect to each other. The side-chain charged group centroids and at least one pair of side-chain nitrogen and oxygen atoms are within 4 Å distance. The ion pair geometry in N-O bridges is worse than that in salt bridges but better than that in the longer-range ion pairs. The N-O bridges have at least a pair of side-chain nitrogen and oxygen atoms within 4 Å distance, but the centroids of their side-chain charged groups are more than 4 Å apart. In the longer-range ion pairs, the side-chain charged group centroids as well as the side-chain nitrogen and oxygen atoms are more than 4 Å apart. In NMR conformer ensembles, the electrostatic interactions among the ion-pairing residues are strongest when they form salt bridges, weaker in N-O bridges, and the weakest in longer-range ion pairs. Most salt bridges stabilize the protein structures. The majority of the N-O bridges are also stabilizing but with a smaller proportion. Most longer-range ion pairs are destabilizing. Most of the ion pairs with side-chain charged group centroids within 5 Å distance are stabilizing toward protein structures.

	MATERIALS AND METHODS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION CONCLUSIONS REFERENCES

Definition of ion pair and ion pair geometry

A pair of oppositely charged residues, Asp or Glu with Arg, His, or Lys, is considered an ion pair. The geometrical orientation of the side-chain charged groups in the ion-pairing residues with respect to each other is characterized in terms of 1) the distance (r) between the side-chain charged group centroids and 2) the angular orientation () of the side-chain charged groups in the two ion-pairing residues. This is the angle between two unit vectors. Each unit vector joins a C atom and a side-chain charged group centroid in a charged residue. Fig. 1 presents a schematic diagram.

View larger version (5K): [in this window] [in a new window]   FIGURE 1   A schematic diagram characterizing geometrical orientation of the side-chain charged residues in an ion pair. Ion pair geometry can be measured in terms of r and . r is the distance in angstroms between the centroids, C1 and C2, of the side-chain charged groups in the ion-pairing residues 1 and 2. , in degrees, is the angle between two unit vectors C1C1 and C2C2. Note that we actually take the supplemental angle.

The coordinates for a side-chain charged group centroid are computed by taking the average of the coordinates of the heavy (nonhydrogen) atoms in the side-chain charged group. We have used only the heavy atoms because the protein crystal structures often lack coordinates for hydrogen atom positions. The present study uses protein structures solved both by x-ray crystallography and NMR spectroscopy. The following side-chain charged group atomic coordinates are used for computing the side-chain charged group centroid position: Asp, C, O¹, O²; Glu, C, O¹, O²; Arg, N, C, N¹, N²; His, C, N¹, C², C¹, N²; and Lys, N.

Geometrical criteria for various ion pair types

Ion pairs are divided into four geometrical categories, salt bridges, N-O bridges, C-C bridges, and longer-range ion pairs. An ion pair is classified as a salt bridge when 1) the side-chain charged group centroids are within a 4.0 Å distance and 2) at least one pair of Asp/Glu side-chain carbonyl oxygen and Arg/Lys/His side-chain nitrogen atoms are within a 4.0 Å distance (Kumar and Nussinov, 1999, 2000, 2001a). An ion pair is classified as a C-C bridge when it satisfies only the first criterion. The ion pair is a N-O bridge when it violates the first but satisfies the second criterion for salt bridge formation. In a longer-range ion pair both salt bridge criteria are violated.

Database composition

The ion pairs and NMR conformer ensembles used in the present study are shown in Table 1. Our database consists of 22 ion pairs in 14 NMR conformer ensembles for 11 nonhomologous monomeric proteins or single-protein domains that contain 50 amino acid residues. All NMR conformer ensembles but one contain 40 NMR conformers. The 22 ion pairs were selected because their oppositely charged residues form salt bridges in at least one crystal structure (if available) or in the NMR average energy-minimized structure or in the most representative conformer (Kelly et al., 1996) in the NMR conformer ensemble (if both the crystal structure and the NMR average energy-minimized structures are unavailable) of the proteins. The 11 nonhomologous proteins are -spectrin pleckstrin homology (PH) domain (Nilges et al., 1997); CheY (Bellsolell et al., 1994; Moy et al., 1994; Volz and Matsumura, 1991); c-Myb DNA-binding domain repeat 1 (R1), repeat 2 (R2), and repeat 3 (R3) (Ogata et al., 1995); cysteine-rich intestinal protein (CRIP) (Perez-Alvarado et al., 1996); CSE-I (Jablonsky et al., 1999); cyanovirin-N (Bewley et al., 1998; Yang et al., 1999); horse heart cytochrome c (reduced form) (Banci et al., 1999); high mobility group 1 (HMG1) box 2 (Read et al., 1993); ISL-1 (Ippel et al., 1999); B1 domain of protein G (Gallagher et al., 1994; Gronenborn et al., 1991; Kuszewski et al., 1999); and U1 SNRP A (Avis et al., 1996). The 14 NMR conformer ensembles involved in this study are taken from the Protein Data Bank (PDB) (Bernstein et al., 1977) entries 1B3C, 1BW5, 1CEY, 1FHT, 1GB1, 1HSN, 1IML, 1MBF, 1MBH, 1MBK, 1 MPH, 2EZN, 2GIW, and 3GB1. For each repeat of c-Myb DNA-binding domain, a separate NMR conformer ensemble containing 50 conformers is available. Hence, there are three NMR conformer ensembles (1MBF, 1MBH, and 1MBK) for c-Myb DNA-binding domain. Two conformer ensembles (1GB1 and 3GB1) are available for the B1 domain of protein G. The details of the selection of protein NMR conformer ensembles for the purpose of our study have been described as well as the advantages and limitations of using NMR conformers for computations of electrostatic strengths of ion pairs (Kumar and Nussinov, 2000, 2001a). The criteria and rationale for selecting the individual ion pairs are summarized in Table 1.

Computation of electrostatic free energy contributions by ion pairs

Free energy contributions due to electrostatic interactions in proteins are frequently computed using continuum electrostatics methodologies. The solute (protein) is represented in atomic detail, but the solvent (water) is represented only in terms of its bulk properties. These methods have found several applications (Honig and Nicholls, 1995). The methodologies to model electrostatic effects in proteins have been recently reviewed by Warshel and Papazyan (1998).

The electrostatic strength of an ion pair is calculated relative to the hydrophobic isosteres of the side chains of the charged residues. The hydrophobic isosteres are the charged residue side chains with their partial atomic charges set to zero (Hendsch and Tidor, 1994). This method has been used extensively (Hendsch and Tidor, 1994; Xu et al., 1997b; Lounnas and Wade, 1997; Kumar and Nussinov, 1999, 2000, 2001a; Kumar et al., 2000b, 2001), and its predictions are consistent with experimental observations (Waldburger et al., 1995, 1996; Spector et al., 2000). At the same time, disagreements between the results obtained by calculations using this method and by experiments have also been noted, and the origin of this inconsistency was discussed by Schutz and Warshel (2001). These include the use of a low value for the protein dielectric constant and the neglect of protein reorganization. The advantages and limitations of this method have been also been discussed by Hendsch and Tidor (1994).

The total electrostatic strength of an ion pair (G_tot) is
where G_dslv is the desolvation energy penalty incurred by the ion-pairing residues because of the desolvation of the charged side chains from the highly polar solvent water to the lower dielectric folded protein interior. This energy penalty depends on the location of the charged residues in the folded protein. G_brd is the favorable energy term that represents the electrostatic interaction between charged residue side-chain groups. This term is sensitive to the geometrical orientation of the side-chain charged groups with respect to each other. G_prt represents the electrostatic interaction of the ion-pairing residue side-chain charged groups with all other polar and ionized groups in the protein. For each ion pair, we have also computed an additional term, the association energy (G_assoc). G_assoc takes into account the desolvation energy penalty for the ion pair and the electrostatic interaction between its side-chain charged groups. However, it ignores the electrostatic interaction between the ion pair and the charges in the rest of the protein (Hendsch and Tidor, 1994). All the energy terms, including G_tot, are sensitive to the value of the dielectric constant used for the protein interior.

G_dslv depends upon the location (burial) of the charged residues (Hendsch and Tidor, 1994; Kumar and Nussinov, 1999). G_prt depends upon the distribution of partial atomic charges in the rest of the proteins with respect to the ion pair. Hence, both G_dslv and G_prt represent the free energy contributions caused by the environment of the ion pair and are sensitive to the variations in the protein structural context, such as the movement of the side chains with respect to other polar/ionized group(s) and to the protein surface. On the other hand, the bridge energy (G_brd) and association energy (G_assoc) terms are sensitive primarily to the ion pair geometry. The protein structural context also affects these terms. It determines the screening of the electrostatic interactions between the ion-pairing residue side-chain charged groups. All the free energy terms are sensitive to variations in side-chain conformations of the charged residues, explaining the relationship between the fluctuations in ion pairs and their geometries and electrostatic strengths in NMR conformer ensembles of proteins (Kumar and Nussinov, 2000, 2001a).

The detailed protocol for the computation of the electrostatic free energy contribution of an ion pair has been described previously (Hendsch and Tidor, 1994; Kumar and Nussinov, 1999, 2000, 2001a; Kumar et al., 2000b). We have used this protocol except with a finer grid spacing (0.5 Å per grid step). For the crystal structures, we have fixed the hydrogen atoms using the BIOPOLYMER module of INSIGHT II. We have energy minimized the crystal structure, keeping the nonhydrogen atom positions fixed. The minimization process consisted of 100 steps of steepest descent followed by 500 steps of conjugate gradient. Energy minimization is carried out using the CFF91 force field in the DISCOVER module of INSIGHT II. This procedure improves the accuracy of the continuum electrostatic calculations (Nielsen et al., 1999). The calculations are carried out for all ion pairs listed in Table 1, in the NMR conformer ensembles, energy-minimized average NMR structures, and crystal structures. All calculations have been carried out at pH 7.0 and at zero ionic strength. All energy values correspond to 25°C. All the calculations, except those shown in Table 5, have been carried out using a protein dielectric constant of 4.

	RESULTS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION CONCLUSIONS REFERENCES

Variations in ion pair geometries in NMR conformer ensembles

Table 1 shows the frequencies with which the ion pairs in our database form salt bridges, N-O bridges, and longer-range ion pairs. It also presents the average ion pair geometry in each class. Most ion pairs form salt bridges only in a minority of the conformers. The incidence of N-O bridge formation by the ion pairs is even more rare, and there are no C-C bridges in our database. Fig. 2 shows an example of the change in the interaction between the side-chain charged groups when an ion pair forms a salt bridge, a N-O bridge, and a longer-range ion pair. Overall, there are 1174 homologous observations for the 22 unique ion pairs in Table 1. These form 358 (30.5%) salt bridges, 96 (8.2%) N-O bridges, and 720 (61.3%) longer-range ion pairs. The ion pairs D48-R81 in c-Myb R1 (c-Myb DNA-binding domain repeat 1) and D100-R123 in c-Myb R2 occupy structurally equivalent positions. The same holds for E47-R73 in c-Myb R1 and E151-R176 in c-Myb R3 (Kumar and Nussinov, 2001a). However, their structural contexts are different because the repeats share only 31-46% sequence identity. These ion pairs are not grouped here. The observations on ion pairs in two NMR conformer ensembles of the B1 domain of protein G are also treated separately.

View larger version (32K): [in this window] [in a new window]   FIGURE 2   Different types of ion pairs in proteins defined in Materials and Methods. Salt bridge, N-O bridge, and longer-range ion pair types are shown for Glu 47-Arg 73 in the NMR conformer ensemble (1MBF) of c-Myb DNA-binding domain repeat 1 (R1). The side-chain charged groups are close and optimally oriented toward each other in a salt bridge. The separation of side-chain charged groups increases in N-O bridges and in longer-range ion pairs. Atoms are color coded. Oxygen atoms are in red, nitrogen in blue, and carbon in green. In general, ion pairs are strongest when they form salt bridges, considerably weaker when they form N-O bridges, and the weakest when they form longer-range ion pairs.

The distribution of ion pair geometries is shown in Fig. 3. The geometries of salt bridges, N-O bridges, and longer-range ion pairs are shown in blue, green, and red, respectively. The side-chain charged group centroid distances fall within 5 Å for almost all N-O bridges. In contrast, in the majority of the longer-range ion pairs, they are >5 Å apart. Overall, the spatial orientation of the side-chain charged groups with respect to each other is most favorable in salt bridges with the average geometric parameters being r_av = 3.5 ± 0.3 Å and _av = 104 ± 27°. N-O bridges also have good geometries (r_av = 4.4 ± 0.3 Å; _av = 105 ± 42°). In longer-range ion pairs, the geometry is most unfavorable with r_av = 7.6 ± 2.1 Å and _av = 118 ± 39° (Table 1). The geometries for longer-range ion pairs vary to a greater extent (Fig. 3 and Table 1). The average distance between the side-chain charged group centroids in the longer-range ion pair increases by 4.1 Å with respect to the salt bridges and by 3.2 Å with respect to the N-O bridges.

View larger version (10K): [in this window] [in a new window]   FIGURE 3   The spatial orientations of side-chain charged groups in 1174 ion pairs in 14 NMR conformer ensembles. The number of unique ion pairs is 22. The ion pair types are color coded. Salt bridges are in blue, N-O bridges in green, and longer range ion pairs in red. The geometry of an ion pair can be characterized in terms of distance between its side-chain charged group centroids (r (Å)) and angular orientation of its side-chain charged groups ( (°)). In this polar plot, radii of the concentric circles represent different r (Å) values. The dotted lines connecting the circles denote different  (°) values. Additional details are given in Materials and Methods. Most of the ion pairs with r  5 Å are stabilizing. In contrast, most of the ion pairs with r > 5 Å are destabilizing.

Electrostatic free energy contribution by various ion pair types

Table 2 presents the electrostatic strengths of each ion pair in the ensembles, classified by the ion pair types. In all NMR conformer ensembles the free energy terms that are sensitive to variations in ion pair geometry, G_brd and G_assoc, are most favorable for salt bridges followed by N-O bridges (Table 2). They are the weakest for the longer-range ion pairs. Because G_dslv and G_prt values do not depend upon ion pair geometry, these free energy terms do not show a consistent variation for ion pair types.

The favorable interaction (G_brd) between oppositely charged side-chain groups in an ion pair is mainly responsible for overcoming the unfavorable desolvation energy penalty, G_dslv (Kumar and Nussinov, 1999). Consistently, we find that salt bridges have better overall electrostatic strengths (G_tot) than N-O bridges and longer-range ion pairs (Table 2). The only notable exception is that of ion pair E151-R176 in the NMR conformer ensemble (1MBK) of c-Myb DNA-binding domain repeat 3 (R3). In this ensemble, the average G_tot (6.4 ± 3.9 kcal/mol) of the three longer-range ion pairs formed by Glu 151 and Arg 176 is slightly stronger than that (5.8 ± 2.8 kcal/mol) for the 36 salt bridges formed by these residues in different NMR conformers. The average G_brd weakens by ~7 kcal/mol from being 7.6 ± 2.0 kcal/mol for salt bridges to 0.9 ± 0.2 for longer-range ion pairs. Interestingly, G_prt becomes stronger by ~7 kcal/mol from 4.8 ± 3.1 kcal/mol for salt bridges to 11.8 ± 6.8 kcal/mol for longer-range ion pairs. Hence, the loss in stability due to deterioration in ion pair geometry appears to have been compensated by the change in the structural context for the longer-range ion pairs. We observe this compensation type between G_brd and G_prt for other ion pairs, too.

A closer examination suggests that the longer-range ion pair in conformer 12 may be responsible for this exception. In this conformer, the interaction between Glu 151 and Arg 176 incurs a desolvation penalty (G_dslv) of 9.9 kcal/mol. G_brd is only 1.1 kcal/mol. However, the interaction is particularly strong with the charges in the rest of the protein. G_prt is 19.6 kcal/mol, and G_tot is 10.8 kcal/mol. G_assoc, which indicates the stability of the ion pair in the absence of the charges in the rest of the protein, is only 0.4 kcal/mol. These free energy values differ from the average values for the free energy terms for E151-R176 in the whole NMR ensemble (G_dslv-av = 6.6 ± 2.1 kcal/mol, G_brd-av = 6.9 ± 2.5 kcal/mol, G_prt-av = 5.2 ± 3.8 kcal/mol, G_tot-av = 5.4 ± 2.9 kcal/mol, and G_assoc-av = 4.3 ± 1.4 kcal/mol). In conformer 12, Glu 151 forms a salt bridge with Lys 144, with a very favorable geometry (r = 3.0 Å;  = 34°). This alternative salt bridge (Kumar and Nussinov, 2001a), which is absent in the average energy-minimized structure of c-Myb DNA-binding domain repeat 3 (R3, PDB: 1MBJ) and other conformers in 1MBK, may be responsible for the large G_prt value in conformer 12. This type of compensation between G_brd and G_prt contributions toward G_tot has been reported earlier (Warshel and Russell, 1984; Cutler et al., 1989). In our own analysis, we have found that charged residue pairs constituting ion pairs in protein crystal structures and in NMR average energy-minimized structures often move farther apart and form alternative salt bridges with the other charged residues in the individual conformers in the ensembles (Kumar and Nussinov, 2000, 2001a). The average values for the ion pair geometric parameters (Table 1) and for the free energy terms (Table 2) are simple arithmetic averages computed over conformers in the ensembles. They do not reflect the experimentally measurable values of these parameters in solution.

Stabilizing and destabilizing electrostatic free energy contributions by ion pairs

Table 3 enumerates the times each ion pair has stabilizing and destabilizing electrostatic free energy contributions in the NMR conformer ensembles. In most cases, ion pairs are stabilizing when they form either salt or N-O bridges. Overall, ion pairs have stabilizing contributions in ~92.5% (331 of 358) of the salt bridges. This proportion is similar (86%) in a dataset of 222 salt bridges in 36 crystal structures of nonhomologous protein monomers (Kumar and Nussinov, 1999). Ion pairs have stabilizing contributions in ~68% (65 of 96) of the N-O bridges. Taken together, ion pairs have stabilizing electrostatic contributions in 396 (~87%) of 454 incidents where they form either salt bridges or N-O bridges. Only 237 (~33%) of 720 longer-range ion pairs are stabilizing. Overall, the ion pairs are stabilizing in 633 cases (salt bridges, 52.3%; N-O bridges, 10.3%; and longer-range ion pairs, 27.4%). In contrast, 483 of 541 (89.3%) destabilizing ion pairs are longer range. The changes in proportions of the stabilizing and destabilizing contributions of the ion pair types are significantly different at 95% level of confidence as indicated by a change in proportion test (Kumar and Bansal, 1998). These statistics relate to highly homologous observations.

Taking together, the results presented in this and previous sections, it appears that geometry plays a crucial role in determining the electrostatic strengths of ion pairs. From Tables 1-3 and Fig. 3, a distance of 5 Å between side-chain charged group centroids appears to be a natural cutoff between (largely stabilizing) close-range and (largely destabilizing) longer-range electrostatic interactions in proteins.

Role of protein structural context in ion pair stability

While ion pair geometry is crucial for electrostatic strength of an ion pair, our results also show the importance of the protein structural context. In 27 (of 358, 7.5%) cases, the ion pairs form salt bridges with destabilizing electrostatic contributions. These are distributed over 11 unique ion pairs (Table 3). In 13 of these, the electrostatic interaction between the salt bridges and the other polar and ionized groups in the protein is unfavorable. In the remaining 14 cases, the favorable bridge and protein energy terms are unable to overcome the unfavorable desolvation energy penalty. In 10 of the 27 cases, the salt bridges are only marginally destabilizing (0 kcal/mol < G_tot < 1 kcal/mol). All 27 destabilizing salt bridges have favorable association free energies (Table 2). Thirty-one N-O bridges with destabilizing contributions show similar features (Tables 2 and 3). Furthermore, we cannot rule out the possibility of uncertainties in the NMR conformer ensembles being a contributing factor in these observations. In our previous analysis (Kumar and Nussinov, 1999) of 222 unique salt bridges from 36 nonhomologous high-resolution (1.6 Å or better) crystal structures, 32 salt bridges also had destabilizing electrostatic contributions because of similar reasons as those cited above.

On the other hand, 237 of the longer-range ion pairs are stabilizing even though the electrostatic interactions among their side-chain charged groups are weak. Because of high homology among the protein structural contexts across the conformers in NMR ensembles, a majority of these longer-range ion pairs are for ion pairs D40-K42 and K91-E95 in -spectrin (1MPH), E49-K52 in c-Myb DNA-binding domain repeat 1 (R1) (1MBF), K92-E99 c-Myb DNA-binding domain repeat 2 (R2) (1MBH), E151-R176 in c-Myb DNA-binding domain repeat 3 (R3) (1MBK), E68-K84 in cyanovirin-N (2EZN), E61-K99 in horse heart cytochrome c (2GIW), and E108-R110 in U1 SNRP A (1FHT) (Table 2). These NMR conformer ensembles account for 188 (~79%) of 237 stabilizing longer-range ion pairs.

In 227 (~96%) of the 237 longer-range ion pairs the protein energy terms are favorable, compensating for the weak interaction between the charged residues. G_prt are stronger than the G_brd for the longer-range ion pairs in -spectrin, CheY (ion pairs: D12-K109 and D41-K45), c-Myb DNA-binding domain repeat 1 (R1) (E47-R73 and E49-K52), repeat 2 (R2) (K92-E99), repeat 3 (R3) (E150-R153 and E151-R176), horse heart cytochrome c, B1 domain of protein G (E27-K28 in 3GB1 and D47-K50 in 1GB1 and 3GB1), and U1 SNRP A (Table 2). G_prt terms are particularly strong for longer-range ion pairs in -spectrin, c-Myb DNA-binding domain repeats (R1, R2, and R3), and horse heart cytochrome c (Table 2). These proteins account for 147 (~62%) stabilizing longer-range ion pairs (Tables 2 and 3). The average desolvation energy penalties for longer-range ion pairs are also small for E68-K84 in cyanovirin-N, E61-K99 in horse heart cytochrome c, and E108-R110 in U1 SNRP A (1FHT) (Table 2). Of the 237, 103 (~43%) are only weakly stabilizing (1 kcal/mol < G_tot < 0 kcal/mol). The G_assoc values for the longer-range ion pairs are weak. Of the 237, 231 (~97%) have G_assoc values ranging between 1 and 0 kcal/mol. Hence, the association of the side-chain charged groups in these longer-range ion pairs would be only marginally favorable in the absence of the stabilizing effect of the neighboring charged residues. These conformer ensembles are also rich in alternative salt bridges that are not observed in the crystal structures, NMR average energy-minimized structures, or most representative conformers of the ensemble (Kumar and Nussinov, 2001a). Hence, the stronger G_prt contributions for many of these longer-range ion pairs could be caused by the presence of charged residue(s) closer to the charged residue(s) in the original ion pair. When such a situation occurs in a NMR conformer, the original ion pair breaks and alternative salt bridge(s) form. An original ion pair is the one formed in the reference structure. As suggested by Warshel and coworkers (Warshel and Russell, 1984; Cutler et al., 1989), this type of compensation between G_brd and G_prt may be an integral part of the ion pair energetics. The effects of other polar and ionized groups on the stability of a given ion pair has already been studied earlier (Hwang and Warshel, 1988).

Ion pair geometry and electrostatic strength relationship in protein crystal structures and NMR average structures

We have also analyzed ion pair geometry and stability in crystal structures and NMR average energy-minimized structures of the proteins in our database, where two or more sets of atomic coordinates are available. Crystal structures are available for CheY (PDB entries 1CHN and 3CHY), cyanovirin-N (PDB entry 3EZM), and B1 domain of protein G (PDB entries 1PGA and 1PGB). For cyanovirin-N and B1 domain of protein G, the NMR average energy-minimized structures are also available (PDB entries 2EZM and 2GB1, respectively). These proteins contain seven ion pairs with 17 sets of calculations (Table 4). The results are consistent with those on NMR conformer ensembles. In all seven cases, the ion pairs have stronger G_brd and G_assoc terms when they have better geometries. For six of the seven ion pairs, the better geometries result in better overall stability. The only exception is E68-K84 in cyanovirin-N. Interestingly, this protein forms domain-swapped dimers in crystals but is monomeric in solution (Bewely et al., 1998; Yang et al., 1999).

	DISCUSSION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION CONCLUSIONS REFERENCES

Here we study the relationship between geometrical orientation and electrostatic strengths of charged residue pairs in proteins. We categorize our observations into different geometrical types. We characterize the geometrical orientation of the charged residues in an ion pair through measurements of the distance and the angular orientation of the side-chain charged groups. We find that the electrostatic interaction between the two charged residues is mostly stabilizing when the side-chain charged groups (as in salt bridges) or at least a pair of side-chain nitrogen and oxygen atoms (as in N-O bridges) are close. Although both salt and N-O bridges may contain hydrogen bonds, they are not the focus of our study. Furthermore, here we use whole side-chain charged groups rather than just acceptor and donor atoms.

Electrostatic interaction is a cooperative phenomenon. When the distance between the charged residues is small, their interaction (proportional to 1/r) dominates. The interaction with other, more distant charges is proportional to 1/r³. However, when the residues in the ion pair have greater distances, this might not be the case. This situation occurs in some longer-range ion pairs. Here, to some extent we address this problem by recomputing alternative salt bridges as the original ones are broken. The NMR ensembles of -spectrin, c-Myb DNA-binding domain repeats (R1, R2, and R3), and cytochrome c contain several such alternative salt bridges (Kumar and Nussinov, 2001a).

Our analysis facilitates the formulation of empirical rules for detection of stabilizing electrostatic interactions in proteins with known three-dimensional structures. These may be useful in identifying appropriate sites for inclusion of electrostatic interactions in de novo designed proteins to potentially enhance their thermal stability. An interesting example is that of glutamate dehydrogenase. It has been shown that Pyrococcus furiosus glutamate dehydrogenase (PfGDH) derives its high thermal stability (Tm = 113°C) from an increased occurrence of ion pairs and their networks within and across its six subunits (Kumar et al., 2000a,b; Yip et al., 1995). Incorporation of an ion pair network, found in the hinge region of PfGDH, at structurally equivalent positions in homologous (55% sequence identity) and less stable (Tm = 89°C) Thermotoga maritima glutamate dehydrogenase (TmGDH) failed to improve the stability of TmGDH (Lebbink et al., 1998). However, a similar attempt to engineer a 16-residue ion pair network across subunit interfaces in TmGDH succeeded in marginally improving its stability (Lebbink et al., 1999). Our results show that incorporation of close-range electrostatic interactions in designed proteins is more likely to improve protein stability if the side-chain charged group centroids are within 5.0 Å. Because most salt bridges are between sequentially close charged residues (Kumar and Nussinov, 1999), this may help in keeping the charged residues close in space, although side-chain motions would nevertheless present a problem.

Protein conformer ensembles can be obtained via conformational sampling around the native state by molecular dynamics simulations and NMR spectroscopy. Both methods have advantages and disadvantages (Kumar and Nussinov, 2000, 2001a). The quality of the experimental data used here is high (Kumar and Nussinov, 2001a). The issue of whether NMR ensembles reflect the protein motion in solution has been controversial but does not bear directly on the analysis performed here. Although it is not possible to completely separate the effects of geometry and the protein structural context, we have attempted to minimize this difficulty through calculations on ion pairs in different conformers of the same protein. The protein structural contexts for an ion pair are expected to be homologous in different NMR conformers for same protein.

Computation of experimentally measurable values for the electrostatic strengths of the ion pairs in solution is not feasible using the NMR conformer ensemble data. NMR conformer ensembles available from PDB cannot be treated as true ensembles in the statistical-mechanical sense. The NMR conformers are obtained by optimizing an energy function consisting of force-field energy terms and experimental restraints. This function is not a Hamiltonian, and probabilities of the individual conformers do not follow a Boltzmann distribution. Additionally, the number of conformers obtained by typical NMR experiments is quite small (~10² to 10³). Of these, only a limited number (10-50) of the conformers that have energy function values below an arbitrary threshold are presented in the PDB files. Even for these conformers, the PDB files do not contain data on relative populations of different conformers. Hence, our analysis does not involve comparison with the experimental results. The values of various electrostatic energy terms computed for the ion pairs serve the qualitative purpose of comparison among various ion pair types. These should not be taken as the quantitative estimates of the ion pair stabilities in proteins in aqueous solution. The experimental support of this analysis is implicit and limited to the use of experimental protein structural (NMR and x-ray crystal) data.

The continuum electrostatics methodology has been widely used. Like any other technique, this method also has its drawbacks. Hendsch and Tidor (1994) have discussed the limitations and advantages of computing the electrostatic strengths of ion pairs with respect to their hydrophobic isosteres. More recently, Schutz and Warshel (2001) have highlighted the limitations of this methodology. Estimates of desolvation free energy penalty paid by the charged residues and screening of electrostatic interactions between the ion-pairing residues and between the ion pair and the rest of the charges in the protein depend upon the dielectric constant (_p) used for the protein. The interior of a protein is often considered largely apolar, and the use of a low dielectric constant for the protein has been quite common. In our calculations, we have used a value of 4 for _p (Kumar and Nussinov, 1999, 2000, 2001a; Kumar et al., 2000b, 2001a). Such a value for protein dielectric constant has also been used by others (e.g. Hendsch and Tidor, 1994; Lounnas and Wade, 1997; Xu et al., 1997b; Xiao and Honig, 1999). However, the effective dielectric constant experienced by an ion pair in a protein depends on the protein relaxation and reorganization of the other polar and ionized groups in the protein (Sham et al., 1998; Schutz and Warshel, 2001). Recently, Schutz and Warshel (2001) have recommended the use of higher values (e.g., 20) for _p when computing the electrostatic strengths of ion pairs. In our more recent continuum electrostatic calculations, a value of 20 for _p also yields reasonable estimates for electrostatic strengths of the salt bridges in citrate synthase (Kumar and Nussinov, unpublished results).

To examine the effect of a higher _p on our results in the present study, we have recalculated the electrostatic strengths of the ion pairs shown in Table 4 with a protein dielectric constant _p = 20. All other parameters were kept the same. The results of the new calculations are shown in Table 5. We have also computed the root mean square deviation (rmsd) values between various corresponding energy terms for the ion pairs in Tables 4 and 5. Due to the higher _p, the electrostatic energy terms G_dslv, G_brd, and G_prt have smaller magnitudes. The rmsd values for G_dslv, G_brd, and G_prt are 2.8 kcal/mol, 1.7 kcal/mol, and 0.9 kcal/mol, respectively. The differences in these three energy terms largely cancel according to Eq. 1 (Materials and Methods). Although the values of G_tot still differ in the two calculation sets, the differences are smaller. The rmsd value for G_tot is 0.2 kcal/mol. The rmsd value for G_aasoc is 0.4 kcal/mol. In three of the four incidents where the ion pairs in Table 4 had destabilizing electrostatic contributions, they now become marginally stabilizing (Table 5). In the remaining case, the ion pair is still destabilizing, but only marginally. Not withstanding these differences, the trend with respect to the ion pair geometries remains the same. When the ion pair geometries are better, the energy terms G_brd and G_assoc are stronger and, in all but one (E68-K84 in cyanovirin-N) case, the overall electrostatic strengths, G_tot, are stronger as well (Table 5).

View this table: [in this window] [in a new window]   TABLE 5   Ion pairs in protein crystal structures and NMR average energy-minimized structures for which at least two sets of calculations are available, with p = 20

The protein structure environment, i.e., the presence of the other polar and ionized groups, also critically affects the electrostatic strength of the ion pair. There appears to be a certain degree of compensation between the G_brd and G_prt terms for ion pairs across different conformers of the NMR ensemble. This compensation mechanism was first recognized by Warshel and coworkers (Warshel and Russell, 1984; Cutler et al., 1989; Hwang and Warshel, 1988; Sham et al., 1998; Schutz and Warshel, 2001). They have used this concept to explain why a simple reversal of charges on residues in enzyme-binding sites and substrates using genetic engineering will not succeed in altering the enzymes-binding specificities (Hwang and Warshel, 1988). We note the consistency between our results and those by Warshel and coworkers, even though the two groups use different methodologies to compute the electrostatic strengths of the ion pairs.

	CONCLUSIONS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION CONCLUSIONS REFERENCES

We have characterized the relationship between geometry and electrostatic strength of ion pairs in proteins. Most ion pairs with side-chain charged group centroids within 5 Å distance are likely to be stabilizing to the protein structure. These results may be useful in formulating guidelines for detecting stabilizing electrostatic interactions in proteins with known three-dimensional structures and for incorporating stabilizing electrostatic interactions in de novo protein design.