Designing Heterodimeric Two-stranded α-Helical Coiled-coils

The E/K coil, a heterodimeric coiled-coil, has been designed as a universal peptide capture and delivery system for use in applications such as biosensors and as an expression and affinity purification tag. In this design, heterodimer formation is specified through the placement of charged residues at the e and g positions of the heptad repeat such that the E coil contains all glutamic acid residues at these positions, and the K coil contains all lysine residues at these positions. The affinity and stability of the E/K coil have been modified to allow a greater range of conditions for association and dissociation. Increasing the hydrophobicity of the coiled-coil core, by substituting isoleucine for valine, gave increases in stability of 2.81 and 3.73 kcal/mol (0.47 kcal/mol/substitution). Increasing the α-helical propensity of residues outside the core, by substituting alanine for serine, yielded increases in stability of 2.68 and 3.28 kcal/mol (0.41 and 0.45 kcal/mol/substitution). These sequence changes yielded a series of heterodimeric coiled-coils whose stabilities varied from 6.8 to 11.2 kcal/mol, greatly expanding their scope for use in protein engineering and biomedical applications.

The coiled-coil is an oligomerization domain found in a wide variety of proteins, including transcription factors, motor proteins, chaperone proteins, and viral fusion proteins (1)(2)(3)(4). Recent surveys of genomic data bases suggest that up to 10% of eukaryotic proteins contain sequences predicted to be coiledcoils (5). This structural motif has been of considerable interest, both because of its diversity in structure and oligomerization state and because of its many advantages as a model system for protein design (6,7). Coiled-coils contain a single type of secondary structure, the ␣-helix, which is easy to monitor experimentally by circular dichroism (CD) spectroscopy. Their quaternary interactions yield a structure that is folded stably in aqueous solution at neutral pH, unlike most singlestranded ␣-helices.
The structural features of coiled-coils have been reviewed extensively (3,7,8). Their sequences are characterized by a heptad repeat, denoted abcdefg, in which positions a and d are occupied by hydrophobic residues. The side chains from the a and d residues pack against each other in a "knobs-intoholes" manner (9), forming a continuous hydrophobic core. Maintaining this packing along the length of the ␣-helices results in their wrapping around each other in a left-handed supercoil. The side chains of the residues in positions e and g lie alongside the hydrophobic core. These positions are typically occupied by charged residues that can participate in i to iЈϩ5 electrostatic interactions, which have been found to play an important role in specifying homo-and heteroassociation in native coiled-coils (1, 10 -13). The preference for electrostatic attractions over repulsions has been key to the de novo design of heterodimeric coiled-coils (14 -22).
Despite the apparent simplicity of coiled-coils, their structures display a surprising diversity. They can be composed of two to five ␣-helices, which may be identical or different and may be arranged in a parallel or antiparallel manner. In addition, coiled-coils have been observed to assemble into larger structures, such as the ␣-sheets and ␣-cylinders described by Walshaw and Woolfson (23). The sequence determinants that control these structural features are superimposed upon the heptad repeat and are only partially understood.
Coiled-coils have been used in numerous applications including affinity purification (24 -26), the directed assembly of extracellular receptor domains (27)(28)(29), the creation of miniaturized antibodies (30 -32), a library presentation scaffold (33,34), and the design of hydrogels with defined properties (35)(36)(37). We have designed the E/K coil, a heterodimeric coiled-coil for use in biotechnological applications including as an expression and purification tag and as a universal dimerization domain for biosensors (14,25,38). Heterodimerization is based on the placement of charged residues at the e and g positions. This system has several advantages, including its high stability and specificity. However, the affinity chromatography procedure required an elution buffer that was both acidic and contained a high percentage of acetonitrile. This procedure worked very well, but more benign elution conditions may be desired for some applications. Therefore, we have modified this design to obtain a set of heterodimerization domains with a range of stabilities and affinities. This will increase the flexibility of this design and allow users to tailor the E/K system for a particular application. Our first approach was to change the length of the coiled-coil peptides (39). We found a direct but nonlinear relationship between chain length and stability. The reduction from 5 to 4 heptads length yielded a useful E/K analog with a free energy of unfolding (⌬G H2O ) of 8.2 kcal/mol. The 3-heptad analog, however, was too labile and was only able to fold when stabilized by an interhelical disulfide bond.
In the present study, we examine sequence modifications * This work was supported in part by the Canadian Institutes of Health Research group in protein structure and function, the University of Colorado Health Sciences Center, and Sensium Technologies, Inc. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18  designed to stabilize the coiled-coils, making the shorter E/K analogs more useful. The first approach was to increase the hydrophobicity of residues in the coiled-coil hydrophobic core, which has been shown to be closely related to coiled-coil stability (40 -43). The second approach was to increase the ␣-helical propensity of residues outside the coiled-coil interface. Helical propensity differences have been shown to affect coiled-coil stability up to 0.77 kcal/mol/substitution (44) and up to 0.96 kcal/mol/substitution in a single-stranded amphipathic ␣-helix (45).

EXPERIMENTAL PROCEDURES
Peptide Synthesis and Purification-The peptides were synthesized by standard t-butyloxycarbonyl solid phase techniques developed by Erickson and Merrifield (46) on an Applied Biosystems peptide synthesizer model 430A (Foster City, CA), using 4-methylbenzhydrylamine resin (0.74 mmol of NH 2 /g of resin) (Bachem, Torrance, CA) on a 0.5-mmol scale. The synthesis methodology is similar to that described by Sereda et al. (47), except that activation and coupling were performed in situ and described as follows. A 4-fold molar excess of amino acid (2 mmol) was dissolved in N,N-dimethylformamide and activated with slightly less than equimolar amounts of HBTU (2-(1H-benzotriazole-1-yl)-1,1,3,3-tetramethyluronium hexafluorophosphate) and HOBt (N-hydroxybenzotriazole), and a 5-fold molar excess of DIEA (N,Ndiisopropylethylamine). The side chain protecting groups were 2-chlorobenzyloxycarbonyl for lysine, O-benzyl for glutamic acid, benzyl for serine, and 4-methylbenzyl for cysteine. A scale of 0.1 mmol was used for the synthesis of each peptide. The peptides were cleaved from the resin by reaction with hydrogen fluoride (20 ml/g of resin) containing 10% anisole and 2% 1,2-ethanedithiol for 1.5 h at Ϫ5°C. The resin was washed with diethyl ether to remove the organic scavengers. The peptides were extracted from the resin with glacial acetic acid, and the extract was lyophilized.
Crude peptides were purified by RP-HPLC, 1 using a semipreparative Zorbax 300SB-C8 column (250 ϫ 9.4 mm, inner diameter, 5-m particle size, 300-Å pore size) from Agilent Technologies (Englewood, CO). The following conditions were used: a linear AB gradient of 1% CH 3 CN/min from 0 to 10% CH 3 CN, followed by a gradient of 0.2% CH 3 CN/min, where eluent A was 0.05% aqueous trifluoroacetic acid (v/v) and eluent B was 0.05% trifluoroacetic acid in CH 3 CN (v/v). The flow rate was 2 ml/min. Peptides IAAL E4, IAAL E3, ISAL E4, VAAL E4, and VSAL E4 were purified at 70°C to prevent aggregation and improve resolution and peak shape. The remaining peptides were purified at room temperature. The homogeneity of the purified peptides was verified by analytical RP-HPLC, amino acid analysis, and mass spectrometry.
Peptide ISAL K3 required an additional mixed mode hydrophilic interaction/cation exchange chromatography step (48) because it contained serine acetylated impurities that were not removed by our RP-HPLC procedure. The hydrophilic interaction/cation exchange chromatography was performed on a PolySulfoethyl A strong cation exchange column (200 ϫ 4.6 mm, inner diameter, 5-m particle size, 300-Å pore size) from PolyLC (Columbia, MD). A linear AB gradient, where A ϭ 10 mM triethylammonium phosphate, 65% CH 3 CN, pH 6.5, and B ϭ 10 mM triethylammonium phosphate, 65% CH 3 CN, 350 mM NaClO 4 , pH 6.5, was performed, starting at 20% B (70 mM NaClO 4 ) with an increasing salt gradient of 0.7% B/min (2.5 mM NaClO 4 /min). The flow rate was 1 ml/min, and the temperature was ambient. The buffer was made by starting with a 10 mM solution of phosphoric acid (Anachemia, Toronto, ON) and raising the pH to 6.5 with triethylamine (redistilled before use) (Anachemia). A stock solution of 2 M NaClO 4 (HPLC grade, Fisher Scientific) was filtered through a 0.22-m filter (Millipore Corporation) before dilution to the desired concentration.
Electrospray mass spectrometry was carried out on a VG Quattro electrospray triple quadrupole mass spectrometer from VG BioTech (Altrincham, UK). Direct injections were performed by injecting 10 l of the sample, using 0.1% formic acid in 50% aqueous CH 3 CN as the solvent, and a flow rate of 50 l/min. The resulting spectra were scanned from 500 to 1,500 Da.
Concentrations were determined by amino acid analysis. Peptides were hydrolyzed in 6 N HCl containing 0.1% (v/v) phenol at 155°C for 1 h in sealed, evacuated tubes. Amino acid analyses were performed on a Beckman model 6300 amino acid analyzer.
CD Spectroscopy-CD spectra were recorded on Jasco J-500C and Jasco J-720 spectropolarimeters (Jasco, Easton, MD). The temperature was maintained at 20°C by a Lauda model RMS water bath (Brinkmann Instruments). The spectropolarimeters were routinely calibrated with an aqueous solution of recrystallized D-10-(ϩ)-camphorsulfonic acid at 290.5 nm. CD spectra were the average of four scans obtained by collecting data at 0.1-nm intervals from 250 to 190 nm, or as low as possible. The results are expressed as the mean residue molar ellipticity [] with units of degrees⅐cm 2 ⅐dmol Ϫ1 and calculated from Equation 1, where obs is the ellipticity measured 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 0.02 cm for the CD spectra scans and 0.05 cm for the data points in the denaturation studies. GdnHCl denaturation studies were carried out by monitoring the ellipticity at 222 nm (an average of five 1.0-s readings) as a function of GdnHCl concentration. Mixtures were prepared from stock peptide solutions in water (ϳ10 mg/ml), buffer (50 mM PO 4 , 100 mM KCl, pH 7.0), and a solution of 8 M GdnHCl in buffer.
GdnHCl Data Analysis-The GdnHCl denaturation curves were analyzed using a two-state unfolding model to determine the fraction folded, using Equation 2, where [] is the observed molar ellipticity and [] F and [] D are the ellipticities of the folded and denatured states, respectively (49). The free energy of unfolding was calculated by Equation 3 where R is the molar gas constant, T is the temperature in Kelvin, P t is the total peptide concentration, and F u is the fraction unfolded (Equation 4) . We then used the linear extrapolation method to calculate the free energy of unfolding in the absence of denaturant (⌬G H2O ), using Equation 5, . This assumes a linear relationship between ⌬G D and [GdnHCl]. The difference in the free energies of unfolding of two peptides (⌬⌬G) was calculated by subtracting the ⌬G values of a pair of peptides at the [GdnHCl] 1/2 of the peptide chosen as a reference, as described by Kohn et al. (16). Sedimentation Equilibrium-Sedimentation equilibrium experiments were performed on a Beckman model Optima XL-I ultracentrifuge, using a six-sector charcoal-filled Epon centerpiece and Rayleigh interference optics. Samples were dissolved in a 50 mM phosphate, 100 mM KCl, pH 7.0, buffer at ϳ1.5 mg/ml and then dialyzed overnight at 4°C against the same buffer. The samples were analyzed at three concentrations and three speeds (between 34,000 and 50,000 rpm) at 20°C. We determined that the samples had reached equilibrium by subtracting successive scans. The data were then analyzed by nonlinear least squares analysis, using the program Nonlin.

RESULTS
Peptide Design-Considerable experience has been gained in understanding the interactions and features important for the design and folding of coiled-coils (15,21,22,38). The major stabilizing features are the hydrophobic interactions in the core (41,42,(51)(52)(53), electrostatic attractions across the coiled-coil interface (7,11,13,22), and helical propensity effects (44, 45, 54 -56). This knowledge and experience were used in the design of the E/K heterodimeric coiled-coils, as illustrated in Fig. 1 and Table I (14, 38). The hydrophobic core, composed of positions a and d, was occupied by valine and leucine. Because of packing effects, ␤-branched hydrophobic residues are the most stabiliz-ing at position a (42,43), and leucine is the most stabilizing residue at position d (41,57). Serine was placed at position b because it is a small polar residue that will increase peptide solubility. Alanine was placed at position c to increase the overall helical propensity (44, 45, 54 -56). Heterodimerization was specified for by the placement of charged residues at the e and g positions (15,21,22,38). The e and g positions are occupied by glutamic acid in the E coils and by lysine in the K coils. Thus, potential homodimer formation will be destabilized, whereas the E/K heterodimer will be stabilized through electrostatic attractions (14,38). The charged residues at position f were opposite in charge to those at positions e and g and were incorporated to increase solubility and reduce the overall net charge. The NH 2 terminus was acetylated, and the COOH terminus was amidated to prevent repulsions between charged termini.
There are two principal targets for increasing the stability of this sequence: the hydrophobic core of the coiled-coil (positions a and d) and the ␣-helical propensity of surface exposed positions b, c, e, f, and g. We chose to stabilize the hydrophobic core by increasing its hydrophobicity. Isoleucine has been shown to be significantly more stable in the a position than valine because of its higher hydrophobicity (43). Accordingly, a series of peptides was made in which all of the a positions have either valine or isoleucine. The ␣-helical structures were stabilized by increasing the overall ␣-helical propensity. To do this, a series of peptides was made in which the low helical propensity residue serine was replaced by the high helical propensity residue alanine at all b positions. We made a series of E and K coils that contained one or both of these modifications and were 3 or 4 heptads long (21 or 28 residues). The nomenclature reflects the sequences (Table I) at positions a-d (i.e. VSAL), and the number reflects the length of the peptide in heptads. E/K denotes a heterodimer formed by a 1:1 association of the E and K coils.
CD Spectroscopy-The secondary structure of the peptides was evaluated by CD spectroscopy. VSAL E4/K4 exhibits a typical ␣-helical spectrum (Fig. 2B), with the characteristic minima at 208 and 222 nm (58). Because the molar ellipticity at 222 nm is directly proportional to the amount of helical structure (59,60), VSAL E4/K4 is folded (Table II). In contrast, the VSAL E4 and VSAL K4 peptides exhibited random coil spectra, with a broad minimum at 200 nm (Fig. 2B). This clearly shows that VSAL E4 and VSAL K4 specifically interact to form a heterodimeric ␣-helical coiled-coil. In contrast, the smaller VSAL E3/K3 analog is still a random coil ( Fig. 2A). This shows that this sequence is not stable enough at a length of 3 heptads to form a heterodimeric coiled-coil.
On the other hand, our more stable sequences were able to form coiled-coils at 3 heptads in length ( Fig. 2, C, E, and G, and Table II). The result with ISAL E3/K3 (Fig. 2E) demonstrates that increasing the hydrophobicity in the core, by substituting three isoleucines for three valines, provided sufficient stabilization for the coiled-coil to fold. VAAL E3/K3 also formed a coiled-coil, as seen in Fig. 2C, showing that increasing the ␣-helical propensity by substituting three alanine residues for three serines could also provide sufficient stabilization for the coiled-coil to fold. Additionally, both sequences have high specificity, demonstrated by the random coil character of the individual E and K peptides of these sequences. The peptide IAAL E3/K3 contains both substitutions and is, as expected, a fully folded ␣-helical coiled-coil.
All of the 4-heptad sequences formed heterodimeric coiledcoils (Fig. 2, B, D, F, and H, and Table II). Some, however, show a loss of specificity (Fig. 2, D, F, and H). These sequences were able to form homodimeric coiled-coils even in the presence of 8 Glu-Glu or 8 Lys-Lys electrostatic repulsions. This is most pronounced for the most stable sequence, IAAL, in which the E4 and K4 homodimers are nearly fully folded coiled-coils (Fig.  2H). The negative-negative charge repulsions are more destabilizing to coiled-coil formation than positive-positive repulsions (Fig. 2, D and F). In all cases, the 1:1 mixture of E and K coils had the greatest ellipticity at 222 nm, indicating that the maximum helical structure is still observed in the het-  Table I and under "Results."

TABLE I Peptide Sequences
The four-letter name (i.e. IAAL) denotes the peptide sequence in positions a, b, c, and d of the coiled-coil heptad repeat (abcdefg). E and K denote peptides in which all of the e and g positions are occupied by either glutamic acid or lysine, respectively. The number refers to the peptide length in number of heptads.
b The sequences are written in the one-letter amino acid code. Ac represents an N ␣ -acetyl group, and NH 2 represents a C ␣ -amide group. Positions a and d of the heptad repeat are underlined and form the hydrophobic core of the coiled-coil. erodimeric coiled-coil. This demonstrates that achieving a balance between stability and specificity is an important principle in protein design.
Trifluoroethanol is a helix-inducing solvent that is used to determine the ability of a sequence to adopt an ␣-helical struc-ture (61,62). All peptides in this study are ␣-helical in the presence of 50% trifluoroethanol (Table II). Trifluoroethanol is also known to disrupt quaternary interactions. This is reflected in the [] 222 /[] 208 ratio, typically Ͼ1.0 for interacting helices in a coiled-coil conformation and 0.85-0.95 for single ␣-helices   Table II. Some of the shorter coiled-coils have [] 222 /[] 208 ratios less than 1.0 in benign buffer. This can be attributed to end fraying, which is more significant in short helices.
Sedimentation Equilibrium-Weight-averaged molecular weights typical of a dimeric species were observed for IAAL E3/K3, ISAL E3/K3, VAAL E3/K3, IAAL E4/K4, and VAAL E4/K4 (Table III). ISAL E4/K4 had a molecular weight typical of a tetrameric species, and VSAL E4/K4 showed evidence of a dimer to tetramer association. Previous studies have shown that the VSAL E4 and VSAL K4 peptides interact in a 1:1 manner, ruling out a trimeric structure (39). VSAL E3/K3 showed signs of a weak monomer to dimer association in the high concentration gradient experienced in the analytical ultracentrifuge. However, it did not fold under the conditions used for our CD analysis, indicating that it is unsuitable for most applications.
The difference in the oligomerization states of ISAL E4/K4 and IAAL E4/K4 is interesting, given that they differ only in the presence of serine or alanine at position b. We cannot explain why this should happen, for this position is highly solvent-exposed and distant from the coiled-coil interface. To our knowledge, this is the first example of a b position substitution causing a change in coiled-coil oligomerization state, although interstrand electrostatic interactions between b and c position residues have been shown to affect the stability of tetrameric coiled-coils (64). This demonstrates that the sequence features that control coiled-coil oligomerization state are complex and still not fully understood. We are proceeding with crystallization studies of these peptides in the hope of answering this question with three-dimensional structure information. Although ISAL E4/K4 and VSAL E4/K4 form tetrameric species, this should not rule out their application as a tag system in which one strand is immobilized on a solid surface, because only the dimeric species can form under these conditions.
We have previously found many examples of coiled-coils that convert from trimers or tetramers to dimers at low levels of denaturant, in a transition that is silent to the CD signal (65). The major unfolding transition observed by CD spectroscopy was then a dimer to monomer transition. To determine whether ISAL E4/K4 behaved in a similar fashion, we performed sedimentation equilibrium experiments in a buffer containing 2 M GdnHCl, a concentration that does not induce unfolding in ISAL E4/K4 (Fig. 3). Molecular weights characteristic of the dimeric species were observed, confirming that these coiled-coils do undergo a silent tetramer to dimer transition before the major unfolding transition (data not shown). This allows us to compare the unfolding data (⌬G, ⌬⌬G) directly.
Conformational Stability-The conformational stabilities of the coiled-coils were determined by GdnHCl denaturations (Fig. 3 and Table III). The denaturation curves for the 3 heptad coiled-coils are shown in the upper panel. VSAL E3/K3 did not have sufficient stability to fold, and so it is not shown. The ISAL E3/K3 and VAAL E3/K3 peptides have similar denaturation midpoints (1.7 and 1.8 M GdnHCl) and free energies of unfolding (⌬G H2O values of 6.8 and 7.2 kcal/mol). Our two approaches to stabilizing this sequence, increasing the hydrophobicity of the core and increasing the ␣-helical propensity, yielded similar gains in stability. The double mutant, IAAL E3/K3, is remarkably stable for a coiled-coil that is only 21 residues long ([GdnHCl]1/2 ϭ 4.3 M and ⌬G H2O ϭ 9.6 kcal/mol), making it an excellent choice for a heterodimerization domain.
The denaturation curves of the analogous 4-heptad coiledcoils are shown in the lower panel of Fig. 3. The original VSAL sequence is the least stable, with a denaturation midpoint of 2.1 M and a ⌬G H2O of 8.1 kcal/mol (14,38). Increasing the hydrophobicity of the core residues (ISAL E4/K4) or the ␣-hel- ical propensity (VAAL E4/K4) resulted in similar increases in stability, with denaturation midpoints of 4.6 and 4.4 M and ⌬G H2O values of 11.0 and 11.2 kcal/mol, respectively. The double mutant, IAAL E4/K4, is an extremely stable molecule and was still ϳ70% folded at 7 M GdnHCl.
This series of coiled-coil sequences allows us to make a number of comparisons and to evaluate the contribution of different structural features to the stability of heterodimeric coiled-coils (Table IV). The ⌬⌬G values were acquired by subtracting the ⌬G values at the transition midpoints, as described under "Experimental Procedures." In addition, the ⌬⌬G value was divided by the number of substitutions (six or eight) to obtain the free energy change per substitution. Each valine to isoleu-cine substitution increased stability by 0.47 kcal/mol when comparing the VSAL E4/K4 and ISAL E4/K4 pair and the VAAL E3/K3 and IAAL E3/K3 pair. This is in excellent agreement with the values obtained by Zhu et al. (43) for a triple valine to isoleucine substitution at the a position, which yielded increases of 0.45 and 0.88 kcal/mol, in coiled-coils with and without disulfide bridges, respectively.
In addition, we could measure the effect of an increase in chain length from 3 to 4 heptads. There is a difference of 4.06 kcal/mol between ISAL E3/K3 and ISAL E4/K4 and a difference of 3.86 kcal/mol between VAAL E3/K3 and VAAL E4/K4. This agrees well with the previously observed difference of 4.34 kcal/mol between VSAL E4/K4 and VSAL E5/K5 (39). DISCUSSION We have designed a set of heterodimeric coiled-coils with a wide variety of stabilities by modifying three parameters: hydrophobicity, ␣-helical propensity, and chain length. Residues in the hydrophobic core of coiled-coils (positions a and d) have been shown to play key roles in determining such characteristics as stability, oligomerization state, and hetero-versus homoassociation (41, 42, 51, 66 -70). Single substitutions in the a and d positions of coiled-coils have yielded a series of coiledcoils whose stabilities covered a range of 7 kcal/mol (41,42,67). We compared the effects of valine and isoleucine in position a while position d was constantly occupied by leucine. The side chains of these residues differ in terms of side chain length (one methylene group) and therefore hydrophobicity. Numerous hydrophobicity scales of the amino acid side chains have been developed, based on partitioning between water and organic solvents, RP-HPLC retention behavior, and calculations of solvent-accessible surface area (for reviews, see Refs. 71 and 72). The relative rankings of the amino acids vary between scales, but isoleucine and valine are always classified as strong hydrophobes, and isoleucine is always the more hydrophobic of the two. The difference between these residues has been measured to be 0.40 -0.80 kcal/mol (72)(73)(74) and to cause an increase in the RP-HPLC retention time of 1.8 min (56). In our E/K coil heterodimers, each substitution of isoleucine for valine yielded an increase in stability of 0.47 kcal/mol (Table IV). This is well within the range given by the different hydrophobicity tables.
What effect have similar substitutions had in other coiled-  Fig. 2A); the guanidine hydrochloride midpoint for IAAL E4/K4 was not determined because the peptide was too stable and only partially unfolded at 7 M GdnHCl (Fig. 3).
f Previous studies have shown that VSAL E4 and VSAL K4 interact in a 1:1 manner thus ruling out a trimeric structure (39). They found that isoleucine was more stabilizing than valine by 0.45 kcal/mol when the coiled-coil was stabilized by a disulfide bridge, and by 0.88 kcal/mol in the reduced peptides. This is a larger effect per substitution than we observed, but still within a comparable range. When a single mutation is made in the hydrophobic core of a well packed protein, the environment surrounding the substitution site can have a significant effect on the degree of stabilization or destabilization observed. These effects are often referred to as context effects and are complicated by the ability of the side chains and the protein backbone to accommodate the substitution through conformational adjustments. Because of this, the contribution of a single methylene group to stability has been observed to vary from 0.1 to 1.5 kcal/mol (67,75). In contrast, we have examined the contribution of an extra side chain methylene group from multiple substitutions along the entire hydrophobic core. This eliminates most context effects, so the differences in stability which we have measured are primarily the result of the hydrophobicity differences between isoleucine and valine.
The second modification to our design was to substitute alanine for serine in position b, significantly increasing the ␣-helical propensity of the sequence. Serine and threonine are thought to have low ␣-helical propensities because their side chain hydroxyl groups compete with the peptide backbone for hydrogen bonds (76). The helical propensity of a series of basic side chains increased as their side chain length increased, suggesting that the presence of polar or charged groups near the peptide backbone is destabilizing (77). A survey of helical propensity scales showed a surprisingly poor correlation between different scales (78); however, alanine always ranks near the top of the scales and always has a greater helical propensity than serine does. We found that a serine to alanine substitution gave increases in stability of 0.41-0.45 kcal/mol (Table IV). This is in very good agreement with the results of O'Neil and DeGrado (44), who found a difference of 0.42 kcal/mol between serine and alanine in a coiled-coil model system. Studies in single-stranded ␣-helices have given helical propensity differences of 0.5-0.8 kcal/mol for a serine to alanine substitution (45,54,55,79). Helical propensity effects appear to be greater in single amphipathic ␣-helices than in coiled-coils presumably because of the additional stabilizing interactions available to coiled-coils (80).
The third aspect of our design was to reduce the chain length to 3 and 4 heptads (21 and 28 residues). Increases in chain length have been observed to cause increases in coiled-coil stability, but the relationship has been found to be nonlinear (39,81). The greatest stability gains were observed at chain lengths of 3 heptads; further increases in length were stabilizing but to lesser degrees. This is primarily because of end fraying, a partial and temporary unfolding of the helix termini (52,82,83). We found that an increase from 3 to 4 heptads caused an increase in stability of 4.06 kcal/mol for the ISAL sequence and an increase of 3.86 kcal/mol for the VAAL sequence. This is in good agreement with our previous results, in which an increase from 4 to 5 heptads in length for the VSAL sequence gave an increase in stability of 4.34 kcal/mol (39).
Electrostatic interactions have been widely shown to control homo-versus heterodimerization in natural coiled-coils (12,84,85). The placement of electrostatic attractions and repulsions has been used extensively in the design of heterodimeric (13-15, 22, 38), heterotrimeric (86), and heterotetrameric coiledcoils (87). Careful double mutant cycle analysis has shown that i to iЈϩ5 electrostatic attractions between glutamic acid and lysine are stabilizing by ϳ0.50 kcal/mol (10,88). The destabilizing effect of electrostatic repulsions is well established. A single i to iЈϩ5 electrostatic repulsion between glutamic acid residues has been shown to be destabilizing by 0.4 -0.8 kcal/ mol (17,88). Thus, electrostatic interactions have the potential to play a critical role in structural specificity.
These results illustrate the need to balance stability and specificity, an important principle in protein design. When we maximized the stability of our design, with IAAL E4/K4, its specificity was lost, as seen in the large amount of homodimer formed (Fig. 2H). In this case, the E4/E4 and K4/K4 homodimers were stable enough to tolerate the destabilizing effect of eight electrostatic repulsions. In nature, many interactions that specify particular coiled-coil oligomerization states are destabilizing. Examples include the presence of asparagine in the hydrophobic core of the GCN4 coiled-coil (51) and glutamic acid in the core of Max/c-Myc (12). Charged residues, though typically destabilizing in the hydrophobic core of coiledcoils, can specify a single oligomerization state (41,42).
In summary, we have created a series of heterodimeric coiled-coils whose conformational stabilities range from 6.8 to 11.2 kcal/mol (Table III). This expands the potential for tailoring heterodimerization domains to particular applications. For example, we can now develop an expression tag/affinity purification system that has more gentle elution conditions, extending the usefulness of this system to more sensitive proteins. The most promising of these heterodimerization domains is IAAL E3/K3. Despite its small size (3 heptads, or 21 residues), it is a fully folded coiled-coil with a high conformational stability ([GdnHCl] 1/2 ϭ 4.3 M, ⌬G H2O ϭ 9.6 kcal/mol) and small dissociation constant (70 nM). It is highly specific because there is no sign of homodimer formation in either the E or the K coil. We have significantly reduced the net charge (ϩ3 on K coil and Ϫ3 on E coil versus ϩ5 on K coil and Ϫ5 on E coil) and hydrophobicity (6 a and d position hydrophobic residues versus 10) compared with the original 5-heptad design. This reduced size is an advantage for the expression of recombinant proteins a The relative change of the free energy of unfolding in the coiled-coil due to the substitutions. The calculation of ⌬G is described under "Experimental Procedures." b Number of interactions in a two-stranded coiled-coil. c Relative change of the free energy of unfolding/substitution. This is obtained by dividing the coiled-coil ⌬G value by the number of substitutions.
for biophysical studies, in which nonnative residues can complicate results. The smaller the tag, the less likely it is to affect the conformation, function, and biophysical properties of the recombinant protein. Smaller tags are therefore advantageous for the many researchers who do not remove expression tags before characterization of recombinant proteins. We have also shown that the design of our heterodimeric coiled-coil has considerable flexibility, and its stability can be modulated. This is a considerable advantage compared with expression tags based on native protein sequences, which are much less amenable to redesign.