The effects of interhelical electrostatic repulsions between glutamic acid residues in controlling the dimerization and stability of two-stranded alpha-helical coiled-coils.

The effects of interhelical electrostatic repulsions in controlling the dimerization and stability of two-stranded α-helical coiled-coils have been studied using de novo designed synthetic coiled-coils. A native coiled-coil was synthesized, which consisted of two identical 35-residue polypeptide chains with a heptad repeat QgVaGbAcLdQeKf and a Cys residue at position 2 to allow formation of an interchain 2-2′ disulfide bridge. This peptide, designed to contain no intrachain or interchain electrostatic interactions, forms a stable coiled-coil structure at 20°C in benign medium (50 mM KCl, 25 mM PO4, pH 7) with a [urea]1/2 value of 6.1 M. Five mutant coiled-coils were designed in which Gln residues at the e and g positions of the heptad repeat were substituted with Glu systematically from the N terminus toward the C terminus, resulting in each polypeptide chain having 2, 4, 6, 8, or 10 Glu residues. These substituted Glu residues are able to form interchain i to i‘+5 electrostatic repulsions across the dimer interface. As the number of interchain repulsions increases, a steady loss of helical content is observed by circular dichroism spectroscopy. The effects of the interchain Glu-Glu repulsions on the coiled-coil structure are partly overcome by the presence of an interchain disulfide bridge; the peptide with six Glu substitutions is only 15% helical in the reduced form but 85% helical in the oxidized form. The stabilities of the coiled-coils were determined by urea and guanidine hydrochloride (GdnHCl) denaturation studies at 20°C. The stabilities of the coiled-coils determined by urea denaturation indicate a decrease in stability, which correlates with an increasing number of interchain repulsions ([urea]1/2 values ranging from 8.4 to 3.7 M in the presence of 3 M KCl). In contrast, all coiled-coils had similar stabilities when determined by GdnHCl denaturation (approximately 2.9 M). KCl could not effectively screen the effects of interchain repulsions on coiled-coil stability as compared to GdnHCl.

The two-stranded ␣-helical coiled-coil domain is an important structural motif in a wide variety of proteins. These include fibrous muscle proteins (McLachlan and Karn, 1982;Smillie, 1979) and DNA-binding proteins, which are transcriptional regulators (Alber, 1992), as well as a host of other protein types (Adamson et al., 1993;Cohen and Parry, 1990). The coiled-coil motif is characterized by a heptad repeat denoted abcdefg, where a and d are normally occupied by hydrophobic residues (Hodges et al., 1972;Hodges, 1992), which fall on the same side of the helix, resulting in a hydrophobic interface between the two helices which provides the major driving force for formation and stability of the coiled-coil (Hu et al., 1990;Zhou et al., 1992aZhou et al., , 1992b. The positions e and g of the heptad repeat flank the hydrophobic face of these amphipathic helices and can participate in interhelical interactions, as well as shielding the hydrophobic core from water by folding across the dimer interface and making direct interactions with the hydrophobes in the core through the methylene groups of their side chains O'Shea et al., 1991). These positions generally contain charged residues, which may lead to interhelical electrostatic attractions or repulsions, thereby stabilizing or destabilizing the coiled-coil (Cohen and Parry, 1990;Talbot and Hodges, 1982). While many coiled-coils, including tropomyosin (McLachlan and Stewart, 1975;Stone et al., 1975) and transcriptional factors (Hu and Sauer, 1992;O'Shea et al., 1991), contain interhelical Lys-Glu and Arg-Glu salt bridges, which have been shown to stabilize them (Krylov et al., 1994;Zhou et al., 1994a), many coiled-coils have also been shown to be even more stable at pH 3, where protonation of the Glu residues prevents the formation of salt bridges (Lowey, 1965;Noelken and Holtzer, 1964;O'Shea et al., 1992;Zhou et al., 1994b). When more hydrophobic residues appear in these positions through random mutagenesis, more stable mutants were actually obtained (Hu and Sauer, 1992;Pu and Struhl, 1993;Schmidt-Dorr et al., 1991). Similar combinatorial mutagenesis studies (Hu et al., 1993) showed that GCN4 coiled-coils containing alanine and threonine at these positions were functional, also suggesting that the ion pairs were not critical.
Subsequently, interhelical ionic interactions between the e and g positions of coiled-coils have been shown to be more relevant in the specificity of coiled-coil formation. Recent studies (Baxevanis and Vinson, 1993;Graddis et al., 1993;O'Shea et al., 1993;Schuermann et al., 1991;Zhou et al., 1994b) have suggested that the presence of electrostatic repulsions in the homodimer will favor heterodimer formation if the number of interchain electrostatic repulsions can be reduced through heterodimer formation. This is the case for the Fos/Jun heterodimer, which forms preferentially by a factor of 1000 over the respective homodimers because both homodimers are destabilized relative to the heterodimer by interchain electrostatic repulsions as well as differences in interhelical hydrophobic interactions (O'Shea et al., 1992(O'Shea et al., , 1989Schuermann et al., 1991). In addition, electrostatic interactions have been shown to affect the chain orientation (parallel versus antiparallel) in model coiled-coils (Monera et al., 1994a(Monera et al., , 1993. It is clear then that, in nature, fine tuning of the specificity of coiled-coil dimerization domains particularly in the DNA-binding proteins leads to a complex system of regulatory proteins each with a specific task (Jones, 1990;Ransone and Verma, 1990). In addition, other coiled-coils have been shown to form heterodimers, in particular tropomyosin (Lehrer et al., 1989;Lehrer and Stafford, 1991), and it is likely that interchain electrostatic interactions are important in dimerization specificity in these proteins as well.
In this paper we take a detailed look at the role of electrostatic charge repulsion in regulating the formation and stability of the ␣-helical coiled-coil dimerization motif, to obtain more insight into the importance of these interactions in regulating the dimerization specificity of various coiled-coil containing proteins in nature. Previous studies (Graddis et al., 1993;O'Shea et al., 1993;Zhou et al., 1994b) have shown that many glutamic acid residues at the positions e and g of model coiledcoil proteins prevented coiled-coil formation, presumably due to the effects of interhelical charge repulsion. Each Glu-Glu repulsion between a g and an e position has been estimated to destabilize the coiled-coil by 0.45 kcal/mol (Kohn et al., 1995) and 0.78 kcal/mol (Krylov et al., 1994). These repulsive destabilizing effects are over and above the intrinsic destabilization of the Gln to Glu substitutions resulting from differences in helical propensity and hydrophobic contributions to the dimer interface and are the dominating effect (Kohn et al., 1995). We would then predict that as the number of Glu residues in these positions is varied, the degree of coiled-coil formation should be altered significantly. This may be a function of the overall net charge present in the dimer interface as well as the effects of direct electrostatic repulsions. The present study uses a stable coiled-coil containing zero net charge and no interhelical electrostatic interactions at the dimer interface and systematically increases the net charge as well as the number of potential specific charge-charge repulsions by the substitution of glutamic acid residues for glutamine to observe how the formation and stability of the coiled-coil are affected. In particular, we are interested in determining the degree of negative charge repulsion allowed before coiled-coil formation is completely abolished. Finally, the ability of salt and changes in pH to affect the degree of formation and the stability of the model coiled-coil analogues is addressed.

MATERIALS AND METHODS
Peptide Synthesis and Purification-The peptides were synthesized by the solid phase method and purified by reversed-phase high performance liquid chromatography, and their identities were confirmed by amino acid analysis and mass spectrometry as described previously (Kohn et al., 1995).
Size-exclusion Chromatography-The effective molecular weights of the disulfide linked two-stranded peptides and the reduced peptides were determined by size-exclusion chromatography (SEC). 1 SEC was carried out on a Superdex 75 FPLC size-exclusion column (13 mm ϫ 290 mm (inner diameter), 10-m particle size, 125-Å pore size) (Pharmacia LKB Biotechnology, Uppsala, Sweden) with a Varian Series 5000 liquid chromatograph coupled to a Hewlett Packard (Avondale, PA) HP1040A detection system and an HP9000 series 300 computer. Unless otherwise specified, the samples were eluted with 50 mM PO 4 , 1 M KCl, pH 7, buffer at a flow rate of 0.5 ml/min. The high salt concentration was used to minimize nonspecific electrostatic interactions with the column matrix and to prevent nonspecific aggregation. Reduced peptides were dissolved in buffer containing DTT to prevent oxidation. Peptide retention times were compared with a standard curve constructed with peptide standards of 7, 14, and 28 residues based on the heptad repeat KgLaEbAcLdEeAf, where the 28-mer forms a dimer of 56 residues and the 7-mer and 14-mer remain as monomers.
Sedimentation Equilibrium Experiments-Sedimentation equilibrium experiments were carried out with a Beckman model E analytical ultracentrifuge containing a Rayleigh interference optical system as described previously (Zhu et al., 1993;Kay et al., 1991). Samples were run in 50 mM PO 4 , 10 mM DTT buffer containing 100 mM KCl for peptide E8r and 1 M KCl for peptide Nr. The experiments were run at 40,000 rpm for approximately 48 h prior to taking equilibrium photographs.
Laser Light Scattering Experiments-Molecular weight determinations using laser light scattering techniques were performed on a Dawn F multiangle laser light scattering photometer (Wyatt Technology Corp., Santa Barbara, CA) according to methodology described by Farrow et al. (1994). Relatively low concentration molecular weight determinations were made by injecting samples onto a gel filtration column attached to the instrument. Concentrations for the eluting peaks from the column were measured on the basis of the differential refractive index increment of the sample (dn/dc), for which a value of 0.185 was used.
Circular Dichroism Measurements-Circular dichroism spectra were recorded at 20°C on a Jasco J-500C spectropolarimeter (Jasco, Easton, MD). A water bath was used to control the cell temperature. The spectropolarimeter was routinely calibrated with an aqueous solution of recrystallized D-10-(ϩ)-camphorsulfonic acid at 290.5 nm. The results are expressed as mean residue molar ellipticity [] with units of degrees⅐cm 2 ⅐dmol Ϫ1 and calculated from the following equation, where [] obs is the observed ellipticity in millidegrees, MRW is the mean residue molecular weight (molecular weight of the peptide divided by the number of amino acid residues), c is the peptide concentration in mg/ml, and l is the optical path length of the cell in cm. Cell path lengths were 0.05 cm for the data points in the denaturation studies and salt and pH titrations and 0.02 cm for the CD spectra scans. CD spectra were the average of four scans obtained by collecting data at 0.1-nm intervals from 250 to 190 nm. For monitoring the ellipticity at 220 nm, eight 1.0-s readings were averaged and repeated five times at each data point and the resulting average was used (a total of 40 1.0-s readings at each data point). Stock peptide solutions were prepared in the appropriate buffer (0.1 M or 3 M KCl, 50 mM PO 4 , pH 7 or pH 3). For urea and GdnHCl denaturation studies, stock solutions of 10 M urea or 8 M GdnHCl were prepared in the same buffer. The ratios of buffer and denaturant solution added were varied to give the appropriate final denaturant concentrations, and 10 l of peptide stock solution was added to each to make a total volume of 120 l. Similarly, a stock 4 M KCl solution was used to make up the samples for the salt titration of E10x. The pH titrations were done by making up separate buffer solutions at the required pH values and adding 10 l of peptide stock solution to 110 l of the buffers at each pH. The pH of the final solutions for CD measurements were verified with a digital pH meter. Peptide concentrations of the stock solutions were determined by amino acid analysis as described previously (Kohn et al., 1995), and an internal norleucine standard was added to each sample.
Calculation of Differences in Free Energy of Unfolding, ⌬⌬G u -Assuming a two-state coiled-coil to random coil denaturation model, the molar fraction of folded peptide (f f ) was calculated from the equation where [] is the observed mean residue molar ellipticity at any particular denaturant concentration and [] n and [] u are the mean residue molar ellipticities of the native (folded) and denatured (unfolded) states, respectively. For reduced coiled-coils the free energy of unfolding (⌬G u ) is dependent on peptide concentration due to the monomer-dimer equilibrium, whereas in the oxidized state stability is concentration-independent (Zhou et al., 1992c). Therefore, for the oxidized coiled-coils, ⌬G u at each denaturant concentration was calculated from the equation ⌬G u ϭ ϪRTln[(1 Ϫ f f )/f f ] and plotted against [denaturant]. Estimates of the free energy of unfolding in the absence of denaturant (⌬G u H2O ) can then be obtained from linear extrapolation according to ⌬G u ϭ ⌬G u H2O Ϫ m[denaturant] (Pace, 1986). Since small errors in the slope term (m) lead to large errors in the extrapolated value of ⌬G u H2O , we calculated the difference in the free energy of unfolding between two peptides, denoted ⌬⌬G u , using the equation of Serrano et al. (1990), which gives the value of ⌬⌬G u at the denaturant concentration half-way between the [denaturant]1 ⁄2 values of the peptides.

RESULTS AND DISCUSSION
Peptide Design- Fig. 1 shows the amino acid sequences of the synthetic peptides that were used in this study. Each consisted of a 35-residue chain with an acetylated N terminus and an amide group at the C terminus. The 35-residue polypeptide chain of the native sequence consists of five repeats of the heptapeptide sequence QgVaGbAcLdQeKf, where the valine residues at position a of the heptad repeat and leucine residues at position d of the heptad repeat form a 3-4 hydrophobic repeat, which makes up a hydrophobic core in the dimerization interface (Hodges et al., 1990;Lau et al., 1984b;Zhu et al., 1993;Su et al., 1994). The sequence resembles the coiled-coils in bZIP transcription factors, which contain almost exclusively leucine at position d and a high percentage of ␤-branched amino acids, particularly valine, at position a of the hydrophobic repeat and are usually 30 -40 residues in length (Alber, 1992;Hu and Sauer, 1992).
The native sequence contains no interhelical or intrahelical electrostatic interactions. In a coiled-coil the e and g positions, which flank the hydrophobic interface made up of positions a and d, normally contain charged residues, which can participate in interhelical electrostatic interactions. In this sequence these positions contain neutral glutamine residues. Interhelical Lys-Glu interactions have been implicated as an important factor in stabilizing a parallel and in-register arrangement of the polypeptide chains (McLachlan and Stewart, 1975;Stone et al., 1975;Monera et al., 1993Monera et al., , 1994aZhou et al., 1994b) and were found to contribute about 0.4 kcal/mol per salt bridge to stability (Zhou et al., 1994a). However, the same study showed that replacement of the Lys and Glu residues with Gln led to a more stable coiled-coil despite the loss of the interhelical salt bridges. This is probably due to increased hydrophobic interactions at the dimerization interface between the side chains at positions e and g and the hydrophobic core a and d positions, which allow the hydrophobic core to be more sequestered from solvent.
In this study we introduced a cysteine residue at position 2 (position a in the heptad repeat) in order to form an interchain disulfide bridge, leading to a 70 residue two-stranded peptide. The disulfide bridge has the advantage of ensuring a parallel, in-register coiled-coil as well as removing peptide concentration as a determinant of the extent of coiled-coil formation and stability (Zhou et al., 1992c) since subunit association in solu-tion is not a factor in the folding process when the two chains are covalently linked. When positioned at the N terminus of a 35 residue coiled-coil, a disulfide bridge has been shown to dramatically increase coiled-coil stability (Hodges et al., 1990;Zhou et al., 1992cZhou et al., , 1993. The native sequence, designated N, was mutated by substitutions of glutamic acid for glutamine in a systematic fashion from the N terminus toward the C terminus as shown in Fig. 1. The resulting mutant peptides were designated E2, E4, E6, E8, and E10 according to the number of Glu residues at positions e and g in the 35 residue polypeptide chain (Fig. 1). The peptides with reduced cysteine at position 2 were designated with an "r" while those with an interchain disulfide bridge between cysteines at positions 2 and 2Ј were designated with an "x." Shown in Fig. 2 is a helical wheel diagram of the potential coiled-coil formed by E10. The diagram depicts with arrows the Van der Waals interactions between residues at the a and d positions on opposing chains as they pack together in the hydrophobic core and the potential electrostatic repulsions between glutamic acid residues at position e of one helix and at position g of the opposing helix. Although charged groups do not interact significantly in an aqueous environment, these residues are in a partially hydrophobic microenvironment since they are partially surrounded by hydrophobic residues. Therefore, despite being solvent-exposed, charged interactions between these residues can contribute to coiled-coil stability (Cantor and Schimmel, 1980;Krylov et al., 1994;Talbot and Hodges, 1982;Zhou et al., 1994a). The central boxes in Fig. 2 containing the a, d, e, and g positions on chain 1 and the aЈ, dЈ, eЈ, and gЈ positions on chain 2 signify the complete dimerization interface (Adamson et al., 1993;Hodges et al., 1994;O'Shea et al., 1991) where any potential interhelical contacts are made.
The net charge at the dimer interface in this series of peptides ranges from zero in the native coiled-coil to Ϫ20 in the E10 coiled-coil as shown in Fig. 1.
Structural Characterization of the Model Coiled-coils-The peptides were evaluated by several techniques to determine the extent of two-stranded ␣-helical coiled-coil formation. These techniques included circular dichroism spectroscopy, ultracentrifugation, laser light scattering, and high performance sizeexclusion liquid chromatography.
The CD spectra of the peptides are shown in Fig. 3. In panel A are the spectra for the disulfide-bridged analogs Nx, E4x, FIG. 1. Amino acid sequences of the synthetic peptides used in this study. Ac denotes N ␣ -acetyl, and amide denotes C ␣ -amide. Native refers to the peptide with all glutamine (Q) residues at positions e and g, shown in bold type, of the heptad repeat, which is designated by the letters abcdefg. In the peptide name, E stands for substitutions of glutamic acid for glutamine, shown in boxes, at positions e and g. The number following E is the total number of glutamic acid substitutions in the analogue, in which the substitutions proceed from the N terminus to the C terminus. All peptides contain a single cysteine residue at position 2, allowing formation of homodimers with an interchain disulfide bridge. The resulting two-stranded peptides with a 2-2Ј disulfide bridge are further designated with an "x" for oxidized and those without the disulfide bridge are denoted with an "r" for reduced.
E6x, E8x, and E10x in benign 25 mM PO 4 , 50 mM KCl buffer at pH 7. The double minima at 208 and 220 nm are characteristic of helical structure. The CD band at 220 nm is due to the n to * transition and is responsive to the amount of helical content. The predicted molar ellipticity for a 100% ␣-helical 35 residue polypeptide was calculated as [] 220,max ϭ Ϫ33,600 degrees⅐ cm 2 ⅐dmol Ϫ1 (Hodges et al., 1988) based on the theoretical equation of (Chen et al., 1974). Therefore, Nx, with a [] 220 of Ϫ31,900, is highly ␣-helical (Table I). The addition of the helixenhancing solvent TFE (Sönnichsen et al., 1992) did not increase the helical content of Nx, as indicated by the molar ellipticity at 220 nm (Table I).
As shown in both Fig. 3A and Table I, an increase in the number of glutamic acid residues per chain from 0 to 10 corresponding to an increase in the net charge at the dimer interface from 0 to Ϫ20 and an increase in the number of interhelical i to iЈϩ5 Glu-Glu repulsions from 0 to 10, results in a gradual decrease in the amount of helical content from approximately 100% to 4%, which is indicative of a random coil structure. As mentioned above, the native coiled-coil Nx contains a high degree of helical content in benign conditions of low salt. E4x contains nearly the same helical content as Nx, so a net charge of Ϫ8 in the dimer interface has not significantly affected the coiled-coil structure. As the number of glutamic acid residues per chain is further increased to 6, 8, and 10, a progressive loss in helical content is observed until with 10 Glu residues/chain the amount of helical content is only about 4%. Therefore, the helical content has been systematically reduced through a gradual increase in the net negative charge at the dimer inter-face. This result is likely due to a shift in the folding equilibrium between folded and unfolded protein rather than a decrease in helicity of the folded state.
The increase in net negative charge prevents dimerization primarily through interchain electrostatic repulsions. However, this increase in negative charge has a much smaller effect on the formation of monomeric ␣-helices in 50% TFE. TFE has been shown to disrupt tertiary and quaternary structure (Lau et al., 1984a). For example, the molar ellipticity of E10x at 220 nm is increased from Ϫ1,500°in benign buffer to Ϫ21,800°in 50% TFE (Table I).
The unfolding process for coiled-coils is not clearly understood, but the most commonly accepted model is the two-state transition model, as judged by the appearance of a monophasic denaturation curve. The two-state model requires that a fully folded coiled-coil dimer unfolds in one cooperative step to denatured monomers. Chain dissociation and loss of helix are presumed to be simultaneous events (Engel et al., 1991;Monera et al., 1994b;Skolnick and Holtzer, 1985;Thompson et al., 1993). This is consistent with the observation that isolated ␣-helices are generally unstable in solution (Cohen and Parry, 1990;Dyson et al., 1988;Thompson et al., 1993).
Not all coiled-coils have been found to follow a two-state transition. In some cases biphasic denaturation has been observed (Greenfield and Hitchcock-DeGregori, 1993;Lehrer et al., 1989;Lehrer and Stafford, 1991), indicating that folding intermediates can occur as previously suggested by the "continuum of states" theory (Skolnick and Holtzer, 1986) in which a broad spectrum of partially folded intermediates may occur in the unfolding process. Therefore, while the simple two-state theory appears to adequately explain the unfolding process of most coiled-coils, it is not necessarily a universally correct description for the process. The direction of propagation of the polypeptide chain is into the page from N to C terminus with the chains parallel and in-register. This diagram depicts interchain a-aЈ and d-dЈ Van der Waals interactions between the hydrophobic side chains, which pack in a knobs into holes fashion (the prime indicates the corresponding position on the opposing helix). Also indicated are g-eЈ (i to iЈϩ5) interchain repulsions between glutamic acid residues at position g of the heptad repeat on one helix and position eЈ of the following heptad on the other helix. The residues that comprise the entire dimerization interface at positions a, d, e, g, aЈ, dЈ, eЈ, and gЈ are within the central boxes. Also given are the net charges on the nonpolar and polar faces of each helix (Ϫ10 and ϩ5, respectively), as well as the net charge on each helix (Ϫ5) and on the dimer interface of the coiled-coil (Ϫ20). In Fig. 3B and Table I, the effect of reducing the interhelical disulfide bridge is illustrated. While the reduced native coiledcoil, Nr, is fully helical, the helical contents of E4r, E6r, and E8r are substantially decreased compared to their oxidized counterparts. A disulfide bridge between positions 2 and 2Ј has been shown to significantly stabilize model 35-residue coiledcoils (Hodges et al., 1990;Zhou et al., 1992Zhou et al., , 1993. The increased stability of the coiled-coil resulting from the disulfide bridge is able to overcome the destabilizing effects of the interhelical charge-charge repulsions to a significant extent. Effects of pH on Coiled-coil Formation-Because the loss of helical content upon the substitution of glutamic acid for glutamine appears to be due to interhelical electrostatic repulsions, protonation of the glutamic acid residues should promote a recovery in helical content by removing the repulsions. This is evident in Fig. 4A, where the CD spectrum of E10x is shown at three different pH values. At pH 7 the spectrum is essentially that of a random coil peptide, at pH 6.25 the peptide is approximately 50% helical, and at pH 5.4 the peptide is fully helical. The isodichroic point at 202 nm suggests a two-state transition (Engel et al., 1991;Thompson et al., 1993). The pH titration profile (Fig. 4B) shows that both E10x and E10r are random coils at pH 7. As the pH is decreased, there is a sharp transition to helical structure within a range of 1 pH unit as indicated by the large increase in negative ellipticity at 220 nm. Interestingly, the transition for E10x is shifted to higher pH relative to that for E10r, which suggests that E10x can form a coiled-coil with a lower degree of glutamic acid residue protonation (a higher net negative charge at the interface). These results and those of Fig. 3 show that the destabilizing effects of the negatively charged glutamate residues can be partially overcome by the presence of a disulfide bridge. It may be inferred that, in general, the more inherently stable the coiled-coil, due to factors such as the presence of an interhelical disulfide bridge or stronger packing of the hydrophobic residues in the dimer interface, the more interhelical charge repulsion the coiled-coil can accept.
It is important to realize that the midpoint of the folding transition curve is not the pK a of the glutamate side-chain carboxyl groups. In fact, the pK a values of the glutamate side chains, which appear to influence the folding, will not be constant but should be different in the folded and unfolded states. If this were not the case, they would have little or no effect on the pH dependence of stability (Yang and Honig, 1993). There-fore, if an unfolding transition takes place over a certain pH range, the pK a value(s) of the group(s) responsible for causing the transition does not correspond to the pH at the transition midpoint, but instead the pK a of these groups is shifted from one value that corresponds to the titration start point (pK a in FIG. 4. pH dependence of the ellipticity of E10x and E10r. All measurements were performed in 50 mM PO 4 , 100 mM KCl buffer at 20°C. A, CD spectra of E10x at pH 7, 6.25, and 5.4. B, mean residue molar ellipticity at 220 nm versus pH for E10x and E10r. Peptide concentrations were 43 and 156 M for E10x and E10r, respectively. (1 Ϫ k/n) where X H ϱ is Ϫ36,300 degrees⅐cm 2 ⅐dmol Ϫ1 for a helix of infinite length, n is the number of residues per helix, and k is a wavelength dependent constant (2.6 at 220 nm) (Chen et al., 1974). d Benign medium conditions combined 1:1 (v/v) with trifluoroethanol.
the folded protein) to another value that corresponds to the titration end point (pK a in the unfolded protein) (Yang and Honig, 1993). The correlation between the transition end points and the pK a values will not be exact, especially when more than one ionizable residue is responsible for the pH dependence. The pH-induced conversion from random coil monomer to coiled-coil dimer was also shown by the size-exclusion chromatograms of E8r (at a flow rate of 0.2 ml/min to increase resolution) at different pH values ranging from 7 to 5 (Fig. 5). At pH 7 peptide E8r elutes at about 56 min, corresponding to the monomer. As the pH is reduced, a peak at about 51 min due to the coiled-coil dimer begins to appear by pH 6. At pH 5.87, there are two approximately equal peaks in the chromatogram, indicating that this is the midpoint of the conversion. As the pH is further reduced, the dimerization becomes complete by pH 5.5. This pH-induced folding is very similar to that observed by circular dichroism in Fig. 4. In both cases, the folding transition is rapid, taking place within one pH unit, and the midpoints are similar. For E10r the apparent pH transition midpoint was 5.6 using CD, while for E8r the apparent pH transition midpoint was 5.87 using SEC. Thus, there is a correlation between the transition in helicity as measured by CD and subunit association as measured by SEC, which supports a two-state transition from helical dimer to random coil monomers.
The preference for the order of association of a coiled-coil has been shown to be dependent mainly on the residues in the a and d positions of the heptad repeat. Val at position a and Leu at position d, as in our sequences, have been shown to favor dimer and trimer formation over tetramer formation (Harbury et al., 1993;Lovejoy et al., 1993;Lupas et al., 1991). Recent studies with the GCN4 leucine zipper showed that a mutant with Val at all a positions and Leu at all d positions formed a mixture of dimer and trimer (Harbury, et al., 1993). Another study where the asparagine at position 16 (position a of the heptad repeat) of the 33-residue leucine zipper of GCN4 was substituted with valine yielded a triple-stranded ␣-helical coiled-coil with a temperature denaturation midpoint 40°C higher than the dimeric native sequence (Potekhin et al., 1994). Their conclusion was that having Val at a and Leu at position d favors trimer structure but that the presence of a polar group such as asparagine in the a and d positions directed twostranded coiled-coil formation instead.
Size-exclusion chromatography of the series of coiled-coil analogues in both the reduced and oxidized forms were carried out (data not shown). In the case of the oxidized peptides, all six eluted in a narrow range of retention times predicted from the standard curve to correspond to the two-stranded 70-residue monomer. In the case of the reduced analogues, Nr, E2r, and E4r were predicted by SEC to be present as dimers, while E6r, E8r, and E10r were predicted by SEC to be present as monomers, corresponding with the results of the CD spectral analysis.
Sedimentation equilibrium experiments in the analytical ultracentrifuge were carried out as described previously Zhu et al., 1993) to confirm the assignment of monomeric and dimeric species. The E8r peptide gave an apparent M r of 3600 (Fig. 6), which closely matches the calculated M r of 3696 for the monomer. This confirms our assignment of E6r, E8r and E10r, which coeluted in SEC, as monomers. Similarly, an apparent M r of 7790 was determined for Nr (Fig. 6), which closely agrees with the expected dimer M r of 7376.
Further evidence for the association state was obtained by use of a low angle laser light scattering detector (Arakawa et al., 1992) to estimate molecular weights of the peptides eluted from a Superdex 75 size-exclusion column. The results gave a molecular weight for E4r twice that for E6r, which indicates that E6r is present as the monomer and E4r is present as the dimer, in agreement with the assignment of association state by retention time.
We conclude from size-exclusion chromatography, ultracentrifugation, and laser light scattering that our sequence pref- FIG. 5. Effects of pH on the conformation of E8r as monitored by size exclusion chromatography. All runs were carried out in 50 mM PO 4 , 100 mM KCl buffer with a flow rate of 0.2 ml/min on a Pharmacia Superdex 75 column. The peptide was dissolved in running buffer containing 10 mM DTT to keep it reduced. An internal standard 10 residue unstructured peptide with the sequence Ac-RGAGGLGLGK-NH 2 was included in each run to confirm run to run reproducibility. M and D represent the monomeric and dimeric forms of peptide E8r, respectively.
FIG. 6. lnY versus r 2 plot from sedimentation equilibrium experiments on Nr and E8r. Samples were run in 50 mM PO 4 , pH 7 buffer with 10 mM DTT. The buffer contained 100 mM KCl for E8r and 1 M KCl for Nr. Both samples were brought into equilibrium at a rotor speed of 44,000 rpm in a Beckman model E analytical ultracentrifuge. Concentrations were 1.08 mg/ml (290 M) and 0.80 mg/ml (216 M) for E8r and Nr, respectively. erentially forms a two-stranded coiled-coil. Thus, either the cysteine at position a in the reduced coiled-coils or sequence differences at other positions could be destabilizing trimers, analogous to the effect of the buried Asn in GCN4. Recent studies by Krylov et al. (1994) and Alberti et al. (1993) have shown that residues in the positions e and g of the heptad repeat can greatly affect the order of association.
Denaturation of Coiled-coil Analogs in Urea and Guanidine Hydrochloride-The stabilities of the disulfide-bridged coiledcoils at 20°C were determined by urea denaturation at pH 7 and GdnHCl denaturation at pH 7 and pH 3 (Fig. 7). The stability of the coiled-coils is expressed as the urea or GdnHCl concentration at which the coiled-coil is 50% unfolded, designated [urea]1 ⁄2 or [GdnHCl]1 ⁄2 , as well as by the difference in free energy of unfolding for the mutant coiled-coils with respect to the native coiled-coil, designated ⌬⌬G u (Table II).
The [urea]1 ⁄2 for Nx is 6.1 M, indicating significant stability in the absence of interhelical electrostatic interactions and making this coiled-coil a good control with which to compare the mutant analogs (Fig. 7A, Table II). E4x has a [urea]1 ⁄2 of 5.1 M and a ⌬⌬G u value of Ϫ0.8 kcal/mol. With the addition of another 2 glutamic acid residues/chain in E6x the decrease in stability as compared with that of Nx is more pronounced, with a [urea]1 ⁄2 of 2.8 M and a ⌬⌬G u of Ϫ2.3 kcal/mol. E8x is only about 50% folded under benign conditions, so its [urea]1 ⁄2 by definition is 0 M and it has a ⌬⌬G u of Ϫ4.4 kcal/mol. Finally, the stability of E10x cannot be determined under these conditions (pH 7, 20°C) since it is too unstable to show appreciable helical content. As was observed for the amount of helical content, the increased charge repulsion at the coiled-coil interface has substantially decreased the [urea]1 ⁄2 in a progressive fashion.
The GdnHCl denaturation profiles at pH 7 (Fig. 7B) show a much different result from those obtained with urea. The [GdnHCl]1 ⁄2 values for Nx, E4x, E6x, and E8x are all very similar, in the range from 3.1 M to 2.8 M ( Table II). The results with GdnHCl do not reflect the large systematic decrease in the stability of the coiled-coils that was observed with urea denaturation as the net interface charge and degree of interhelical charge repulsion is increased. These observations support previous results from this laboratory (Monera et al., 1994a(Monera et al., , 1994b(Monera et al., , 1993, which showed that GdnHCl appears to mask electrostatic repulsions and attractions in two-stranded coiled-coils, giving the same measure of stability whether the residues at positions e and g were arranged to form interhelical attractions or repulsions as long as the hydrophobic packing at the dimer interface was the same. Similarly, Hagihara et al. (1994) have also recently illustrated this same phenomenon, where urea denaturation at pH 7 showed progressively lower stability as the degree of acetylation of horse heart cytochrome c was increased while GdnHCl denaturation gave the same stability regardless of the extent of acetylation. This indicates an ability of GdnHCl to mask the negatively charged residues and their interactions with other charged residues that contribute to stability. The ionic nature of GdnHCl is the probable reason for its ability to disrupt the effects of charged residues on protein stability. The positively charged guanidinium ion, which is responsible for binding to the protein surface and subsequent denaturation (Makhatadze and Privalov, 1992;Greene and Pace, 1974;Pace, 1986;Tanford, 1970), may initially bind specifically at the negatively charged glutamate carboxyl groups when present at low concentrations, thereby neutralizing the charge repulsions and removing their effects on stability.
At low pH, the stabilities of many coiled-coils have been observed to be higher than at pH 7, as shown for example in the muscle protein tropomyosin (Lowey, 1965;Noelken and Holtzer, 1964), the Fos-Jun heterodimeric leucine zipper (O'Shea et al., 1992), and synthetic model coiled-coils Zhou et al., 1992cZhou et al., , 1993. Zhou et al. (1994a) demonstrated that a protonated Glu residue (at pH 3) at the e position of the heptad repeat in a coiled-coil makes a 0.65 kcal/mol greater contribution to coiled-coil stability than an ionized Glu residue (at pH 7) and a 0.45 kcal/mol greater contribution to stability than a Gln residue in the absence of interhelical or intrahelical charge-charge interactions. Recently, Lumb and Kim (1995) have found that the pK a of a Glu residue at position 20 of the GCN4 leucine zipper (position e of the heptad repeat) is slightly higher in the folded state than in the unfolded state, also indicating that the protonated form is energetically favorable. Our study supports these results as shown in Fig. 7C, where the GdnHCl denaturation profiles at pH 3 show that as the number of protonated Glu residues is increased the stability of the coiled-coil is increased, with Nx being least stable and E10x the most stable at pH 3. Nx has the same stability as at pH 7, as one would expect, since there are no residues that change their ionization state over the pH range 3 to 7. E10x is so stable at pH 3 that it is only about 50% FIG. 7. Denaturation profiles of some of the disulfide-bridged coiled-coils at 20°C in 50 mM PO 4 , 100 mM KCl buffer using: urea at pH 7 (A), GdnHCl at pH 7 (B), and GdnHCl at pH 3 (C). The fraction of folded peptide was calculated from the observed mean residue ellipticity at 220 nm, as described under "Materials and Methods." For those analogues that are not fully helical, the fraction folded was calculated based on the ellipticity of Nx in benign conditions. The peptide concentrations in the final solutions for CD measurements ranged from 70 to 100 M. unfolded in 7.4 M GdnHCl (the maximum experimentally possible concentration). This peptide is fully folded in 9 M urea (data not shown), indicating a greater efficiency of GdnHCl as a denaturant versus urea as described previously (Greene and Pace, 1974). Protonated Glu residues have been shown to have higher helical propensity than Gln (Chakrabartty et al., 1994;Scholtz et al., 1993), and it has been shown that helical propensities of side chains can affect coiled-coil stability (Hodges et al., 1981;O'Neil and DeGrado, 1990). In addition, the higher hydrophobicity of protonated Glu compared to Gln (Guo et al., 1986;Sereda et al., 1994) can increase coiled-coil stability when located at positions e and g. The side chains at these positions can extend across the hydrophobic interface, thereby reducing solvent accessibility to the hydrophobic core. More hydrophobic residues in these positions have therefore been shown to lead to higher coiled-coil stability (Schmidt-Dorr et al., 1991;Zhou et al., 1994a;Hodges et al., 1994). Fig. 8A illustrates the dramatic contrast between the results of GdnHCl titrations of Nx and E10x. While Nx shows the typical two-state denaturation curve, E10x is initially unfolded in benign medium but is actually induced to a helical state by small concentrations of GdnHCl and achieves about 90% of the helical structure of Nx at 1 M GdnHCl. Upon further increase of the GdnHCl concentration, the coiled-coil is subsequently unfolded. E8x behaves similarly, as shown in Fig.  7B, going from 50% folded in benign buffer to 90% folded in 1 M GdnHCl and then unfolding with further addition of GdnHCl. The ability of GdnHCl to act as a stabilizer or inducer of protein structure when present at low concentration has been suggested previously (Hagihara et al., 1993(Hagihara et al., , 1994Mayr and Schmid, 1993;Morjana et al., 1993;Tsong, 1975) either through the masking of positive charge repulsions by the Cl Ϫ ions or negative charges by the guanidinium ϩ ions. The binding of guanidinium ions to the negatively charged glutamate groups at low GdnHCl concentration would neutralize the Glu-Glu charge repulsion, thereby inducing the coiled-coil structure. Alternatively, GdnHCl could promote dimerization through general electrostatic charge screening (Debye-Hückel screening). Goto et al. (1990) have proposed these two possible mechanisms for the salt-induced folding transition of melittin.

Effects of GdnHCl on the Electrostatic Interactions in the Coiled-coils-
We have compared the abilities of GdnHCl and KCl to induce helical structure in the E10x peptide as shown in Fig. 8B. In the case of GdnHCl, the maximum inducible helix (Ϫ29,000 degrees⅐cm 2 ⅐dmol Ϫ1 ) is achieved at about 0.75 M and is about 90% of the helical content of Nx. KCl is able to induce the same degree of helical structure but requires a concentration of over 2 M to do so. The potassium ion has the same plus one charge as the guanidinium ion but does not have the same apparent binding affinity for the glutamate groups. We propose that the guanidinium ion can bind to the glutamate groups via an electrostatic interaction and through hydrogen bonding while potassium ion can only bind to the glutamate carboxyl group through an electrostatic attraction. Therefore, the added effect of hydrogen bonding may increase the overall affinity of the guanidinium ion for the glutamate carboxyl groups and thereby increase its ability to mask the charge repulsions. The ability of salts to have general Debye-Hückel charge screening effects should depend on the ionic strength and therefore be the same for equal concentrations of GdnHCl and KCl. Thus, general charge screening appears not to be the major contribution to the induction of helix in E10x by GdnHCl. a The sequences of the peptides are shown in Fig. 1. b [GdnHCl] 1/2 or [urea] 1/2 is the concentration of urea or GdnHCl (M) at which the peptide is 50% folded, as determined by comparing its molar ellipticity at 220 nm to that of the native coiled-coil in the absence of denaturant (the fully folded state). c ⌬⌬G u is the apparent difference in the free energy of unfolding between the native coiled-coil and the specified mutant coiled-coil. This was calculated from the equation: ⌬⌬G u ϭ ([D] 1/2 mutant Ϫ [D] 1/2native )(m native ϩ m mutant )/2, where [D] 1/2 is the [urea] 1/2 or [GdnHCl] 1/2 and m is the slope of the equation: ⌬G u ϭ ⌬G water Ϫ m [denaturant] and ⌬G u ϭ ϪRTln(f u /f f ) (Serrano et al., 1990;Pace, 1986). f u is the fraction unfolded and f f is the fraction folded at the denaturant concentration at which ⌬G u is being calculated. Values of ⌬G u were calculated for each denaturant concentration, and plots of ⌬G u versus [D] were made to determine m. Only points near the transition midpoint were included in these plots (data not shown). A positive ⌬⌬G u indicates the mutant coiled-coil is more stable than the native coiled-coil. It has long been accepted that salts can affect hydrophobic interactions in proteins. KCl promotes protein stability mainly through its effect on the structure of water, which increases the stability of the folded state by decreasing the solubility of hydrophobic molecules and increasing the apparent hydrophobic effect (Creighton, 1993). This type of salt is predicted to be excluded from the protein surface, meaning that its concentration is lower around the protein molecules than in the bulk solvent, and the protein is described as preferentially hydrated. In contrast, a denaturing salt such as GdnHCl tends to increase protein solubility (including the hydrophobic core) via direct interactions of the guanidinium ion with the surface of the protein, as proposed in the denaturant binding model (Makhatadze and Privalov, 1992). The binding of the denaturant to the protein surface allows the exposure of the normally buried hydrophobic core and a much greater surface area of the protein exposed to solvent. Thus, the different apparent abilities of KCl and GdnHCl to promote the E10x coiled-coil structure may be due to different mechanisms of action by the two salts; GdnHCl potentially operates mainly by masking charge repulsions through direct ion binding, whereas KCl may act more through promoting a stronger hydrophobic effect that can override the effects of the charge repulsions on the coiled-coil folding (i.e. it is able to force the coiled-coil to form even in the presence of the charge-charge repulsions).
Possible evidence for this line of reasoning is given by the results of urea denaturation of the disulfide-bridged coiled-coils at pH 7 in the presence of 3 M KCl (Fig. 9A). The large KCl concentration has shifted the stabilities of the entire series of analogues significantly, to higher [urea]1 ⁄2 values. As a result, the stabilities are now such that even the lowest stability analogues E10x and E8x are almost fully helical in the absence of urea (Table I), and their entire denaturation with urea can be observed and compared with that of the other analogues. The shifting of these denaturation curves by the presence of 3 M KCl clearly illustrates the ability of KCl to increase the apparent hydrophobic effect. This effect on stability should be the same for all the analogues since they share the same hydrophobic core residues and is seen clearly from the increase in the [urea]1 ⁄2 of Nx (which is not affected by electrostatic interactions) from 6.1 M to 8.4 M (Table II) as the KCl concentration is increased from 0.1 M to 3 M. This corresponds to an increase in the ⌬G u (free energy of unfolding) of 1.8 kcal/mol. This extra stabilization is therefore able to counteract the destabilizing effects of the interchain repulsions enough to promote full coiled-coil formation. There may also be some degree of charge screening by the KCl contributing to the increased coiled-coil formation and stability of the analogues with repulsions, but this effect is not large since a wide range of stabilities (range in [urea]1 ⁄2 from 3.7 to 8.4 M; see Table II) is obtained in the high KCl concentration. If KCl were very effective at screening the charge-charge repulsions, all the analogues should have similar [urea]1 ⁄2 values. Similar results were obtained by Monera et al. (1993) where the inability of KCl to screen electrostatic interactions was also observed.
In contrast to KCl, GdnHCl does not increase protein stability through promoting stronger hydrophobic interactions. For example, the urea denaturation of Nx in the presence of 1 M GdnHCl gave a [urea]1 ⁄2 of 4.8 M (data not shown) versus 6.1 M without the GdnHCl present. This corresponds to the known effects of GdnHCl on decreasing the temperature stability (T m ) of proteins (Von Hippel and Wong, 1965).
Effects of Gln to Glu Substitutions- Table II compares the ⌬⌬G u values for the disulfide-bridged analogues with respect to the native coiled-coil Nx as obtained by GdnHCl denaturation at pH 7 and pH 3 in 100 mM KCl and by urea denaturation at pH 7 in 3 M KCl. These data were plotted as ⌬⌬G u versus the number of Glu residues per chain in Fig. 9B. The urea denaturation shows negligible destabilization for the Glu residues in the first two positions (2 Glu/chain). For the middle three heptads (increasing to 4, 6, and 8 Glu/chain), there is an approximately linear decrease in stability (a negative ⌬⌬G u ) with respect to Nx of about 1.2 kcal/mol per increase of 2 in the number of Glu residues per chain (4 in the dimer). Since we have previously determined a destabilization of about 0.15 kcal/mol per Gln to ionized Glu substitution (Kohn et al., 1995), 0.6 kcal/mol (4 ϫ 0.15) of this 1.2 kcal/mol destabilization is due to differences in the intrinsic properties of ionized Glu versus Gln, and the other 0.6 kcal/mol may be due to two Glu-Glu repulsions for every 4 residues. This would suggest two i to iЈϩ5 repulsions of 0.3 kcal/mol each, which is somewhat less than the 0.45 kcal/mol we previously determined in low salt (Kohn et al., 1995), suggesting partial screening of the electrostatics by KCl.
In contrast to the results of the urea denaturation, the denaturation in GdnHCl at pH 7 shows that ⌬⌬G u is essentially unaffected (Fig. 9B), demonstrating the ability of GdnHCl to screen the effects of the electrostatic repulsions on coiled-coil stability. At pH 3 the stability increases (a positive ⌬⌬G u , indicating the coiled-coil is more stable than the native coiledcoil) with the number of Glu residues per helix as indicated by Fig. 7C. In the middle part of the coiled-coil (going from 2 to 8 Glu residues/chain), there is a linear increase in stability of about 1.8 kcal/mol per increase of 2 Glu residues/chain (four in FIG. 9. Effects of KCl and pH on the apparent stability of the disulfide-bridged coiled-coils in the presence of urea and GdnHCl. A, urea denaturation profiles at 20°C in 50 mM PO 4 , 3 M KCl buffer at pH 7. The fraction folded was calculated as described under "Materials and Methods." The peptide concentrations in the final solutions for CD measurements ranged from 72 to 108 M. B, ⌬⌬G u versus the number of Glu residues per chain under different denaturation conditions. Results shown are for GdnHCl denaturation in 50 mM PO 4 , 100 mM KCl buffer at pH 7 and 3, and for urea denaturation in 50 mM PO 4 , 3 M KCl buffer at pH 7. All measurements were done at 20°C, and ⌬⌬G u was calculated as outlined in Table II. the dimer). This corresponds to 0.45 kcal/mol increase in stability per Gln to protonated Glu substitution, which is consistent with the study of Zhou et al. (1994a) in which protonated Glu was substituted for Gln only at the e positions. Therefore, the substituted Glu residues appear to have equal effects on stability at both the e and g positions of the heptad repeat.
Effects of N-terminal Gln to Glu Substitutions-The substitution of glutamic acid residues actually has an additional effect on coiled-coil stability, which is apparent in the behavior of the E2 analogue. It contains Glu residues at positions 1 and 6 of the peptide chain, and its stability as obtained by urea denaturation at pH 7 in low salt (100 mM KCl) is actually more than that of the native coiled-coil both in the oxidized and reduced forms. E2x and Nx have [urea]1 ⁄2 values of 6.8 M and 6.1 M, respectively (⌬⌬G u versus Nx of ϩ0.6 kcal/mol), and E2r and Nr have [urea]1 ⁄2 values of 3.2 M and 2.5 M, respectively (⌬⌬G u versus Nr of ϩ0.8 kcal/mol). Therefore, the stabilization of the coiled-coil by the Glu substitutions at the N terminus is independent of the presence of the disulfide bridge. It has been shown previously that a disulfide bridge at the N-terminal a position of a 35-residue coiled-coil does not distort the coiledcoil structure . This result indicates that the negative charges are interacting favorably with the helix macrodipole of the helices that make up the coiled-coil. This macrodipole is due to the lining up of the dipole moments of all the peptide bonds of the helices and is estimated to be equivalent to a charge of ϩ0.5 at the N-terminal end of the helix and Ϫ0.5 at the C-terminal end of the helix (Hol et al., 1981;Scholtz, et al., 1993). Previous studies have estimated the energetic effects of a favorable helix dipole-charge interaction in the range 0.5-2 kcal/mol (Sali et al., 1988;Sancho et al., 1992).
The addition of 3 M KCl to the solution brings the [urea]1 ⁄2 value for E2x back below that of Nx (Fig. 9A), indicating that the charge-dipole interaction is screened by high salt concentration. This result indicates that KCl is able to screen chargedipole interactions even though it could not effectively screen the charge-charge interactions between Glu side chains as shown in Fig. 9A, suggesting that the charge-charge interaction is stronger. Similarly the positioning of negatively charged residues at the C-terminal end of the coiled-coil is likely to have a destabilizing effect due to an unfavorable interaction with the negatively charged end of the helix macrodipole, and a high salt concentration should screen this interaction. This may explain why there is a large decrease in helicity between E8x and E10x under low salt conditions (Fig. 3A), but there is not a significant difference in the stability of E8X and E10x by urea denaturation in the presence of 3 M KCl (Fig. 9A) where charge-dipole effects are screened.
The model coiled-coils from our previous study (Kohn et al., 1995) serve as key controls for this paper since they unequivocally show that an interchain Glu-Glu repulsion in the i to iЈϩ5 orientation destabilizes a coiled-coil by 0.45 kcal/mol. We have subsequently shown in the present study that a series of coiled-coils containing a systematic increase in the number of such Glu-Glu repulsions result in a gradual loss of helical content and stability. The fact that the helical content and stability decrease gradually indicate that the repulsive effects introduced by Glu substitutions are additive. While there is not a critical point at which an all helical coiled-coil is transformed to a random coil by the substitution of two additional Glu residues, the loss of helical content is significant in going from E6x to E8x in the oxidized coiled-coils and from E4r to E6r in the reduced coiled-coils, indicating the importance of subtle changes in interchain electrostatic interactions in determining coiled-coil formation, specificity, and stability.
The question still remains as to the mechanism of destabili-zation of the coiled-coil by the interchain repulsions. As stated by Zhou et al. (1994b), interhelical electrostatic repulsions can disrupt the formation and stability of coiled-coils in two ways. First, specific i to iЈϩ5 (g-eЈ) repulsions between two likecharged side chains could destabilize the coiled-coil. The repulsion between two like-charged residues on opposing helices of the coiled-coil may not destabilize the coiled-coil through the electrostatic repulsion forcing the chains apart, since to do so, it would have to overcome the much more significant hydrophobic effect, which is the major force driving coiled-coil formation. Normally the side chains that occur in the e and g positions lie across the dimer interface and interact with each other forming salt bridges when they are oppositely charged; however, when they are similarly charged, the repulsions between the side chains may force them apart and away from the dimer interface. This would leave the hydrophobic core a and d positions more exposed and therefore lead to a decrease in stability by allowing greater solvent access and therefore reducing solvent entropy in the folded state. Second, the buildup of a large net charge on the dimer interface could also lead to general electrostatic effects in which net charges on the faces of the two helices repel.
Determining the relative roles of these two effects is somewhat difficult from the current understanding of interhelical electrostatic effects on coiled-coil formation and stability. However, we have found that a coiled-coil containing 5 interhelical Glu-Glu and 5 interhelical Lys-Lys repulsions but zero net charge at the dimer interface forms a coiled-coil in both the case of parallel and antiparallel chain orientation (Monera et al., 1993). 2 In contrast peptides containing 10 interhelical Lys-Lys or 10 interhelical Glu-Glu repulsions with high net charges at the prospective dimer interface will not form coiled-coils (Graddis et al., 1993;Zhou et al., 1994b;this study). In addition, a synthetic triple-stranded coiled-coil with a net charge of Ϫ3 in the trimer interface forms despite seven pairs of specific interhelical ionic repulsions between residues at the e and g positions of the helices, which lie antiparallel to each other (Lovejoy et al., 1993). Therefore, the high net charge may be preventing the chains from approaching each other through general electrostatic repulsions while specific charge repulsions between these residues may destabilize the coiled-coil by disrupting side chain packing around the dimer interface. However, the relative roles of these two forms of electrostatic effects will probably depend on the surrounding amino acids.
The effects of disulfide bridges on the stability of twostranded coiled-coils have been described previously (Hodges et al., 1990;Zhou et al., 1993). In the present study, we have illustrated the ability of disulfide bridges to overcome the effects of interhelical charge-charge repulsion on coiled-coil formation (Figs. 3 and 4). While the effect of the disulfide bridge on stability is difficult to interpret because it involves the comparison of a bimolecular concentration-dependent unfolding process for the reduced coiled-coil and a unimolecular concentration-independent unfolding process for the oxidized coiled-coil (Regan et al., 1994), we have estimated the interchain disulfide bridge to offer about 3-4 kcal/mol additional stability to the coiled-coil (Hodges et al., 1990;Kohn et al., 1995). This effect would be expected to be capable of counteracting a large number of interchain electrostatic repulsions, which as stated above have been estimated to destabilize by 0.45 kcal/mol per Glu-Glu repulsion. In the case of the E8x peptide with a net charge of Ϫ16 and containing 8 interhelical Glu-Glu repulsions, the disulfide bridge is able to promote 50% coiled-coil structure. Therefore, in this case the stabilizing fac-tors including the disulfide bridge are being almost offset by the charge repulsion so that the peptide is equally populating the folded and unfolded states.
Although the disulfide bridge does not generally apply to natural coiled-coils, there are some cases where redox control of transcription factor activity in eukaryotic cells has been suggested. Recently it has been found that the basic helix-loophelix transcription factor E2A binds DNA at physiological temperature as a homodimer only in the presence of an interhelical disulfide bond, while under reducing conditions E2A binds DNA only as a heterodimer with MyoD or Id (Benezra, 1994). Nuclear translocation of the transcription factor NF-B is activated by oxidative stress (Meyer et al., 1993;Schreck et al., 1991), while USF (a basic helix-loop-helix zipper protein) forms both intra-and inter-molecular disulfide bonds, which appear to decrease its affinity for DNA (Pognonec et al., 1992). Therefore, the fact that an intermolecular disulfide bond is required for high affinity DNA binding of certain transcription factors and that the presence or absence of this bond can profoundly direct dimerization specificity at physiological temperatures suggests that, as observed in this study, the ability of the disulfide bridge to overcome destabilizing effects on homodimerization such as intermolecular charge repulsion is the key to its role in transcription factor activity.
The dramatic effects of pH and salt on protein folding observed in this study and previously in model coiled-coils (Zhou et al., 1994b) and other model helical proteins (Ramalingam et al., 1992) have suggested major implications for the potential of de novo design of environmentally sensitive proteins. One of the best examples of salt effects on protein folding in native proteins is the case of extreme halophilic bacteria, which are adapted to living in high salt concentration environments and whose cytoplasm is close to saturated in KCl. The proteins of these bacteria are often rich in acidic groups and have lower hydrophobicity than their counterparts in non-halophilic bacteria (Lanyi, 1974). These proteins, when isolated, require high salt concentrations to stabilize their folded structures. This stabilization may be due to the combined effects of the salt stabilizing the hydrophobic core as well as interactions of hydrated salt ions with the surface of the folded protein (Zaccai and Eisenberg, 1990). It has been suggested that clustering of negatively charged residues on the surface of halophilic proteins may cause structural instability, possibly due to charge repulsion, that is removed by the effects of salts (Ramalingam, et al., 1992). pH sensitivity of coiled-coil formation has recently been illustrated for the influenza virus hemagglutinin protein, which is required for fusion of the viral and cellular membranes. The protein undergoes a conformational shift to become active under the mildly acidic (pH 5) conditions of the mature endosome (Carr and Kim, 1994). The protein forms a threestranded coiled-coil stem adjacent to a sequence of about 35 residues, which forms an extended loop at pH 7 but at pH 5 becomes helical and extends the triple-stranded coiled-coil, thereby inducing the activating changes in the structure. It has been found that this sequence probably forms an extended loop at pH 7 due to interchain electrostatic repulsion, which prevents coiled-coil propagation through this region but which is alleviated by protonation of acidic side chains at lower pH (Carr and Kim, 1993).
In conclusion, this study illustrates that a systematic increase in interhelical charge repulsion leads to a progressive loss of helical content and stability. These effects can be modulated by changes in pH and high salt concentrations.