The crystallographically determined structures of atypical strained disulfides engineered into subtilisin.

The geometries of two disulfide bridges genetically engineered into subtilisin have been characterized by x-ray crystallography to determine the structural and energetic constraints involved in introducing disulfide bonds into proteins. Both disulfide bridges (Cys-24-Cys-87 and Cys-22-Cys-87) exhibit atypical sets of dihedral angles compared to those for other reported disulfide structures in proteins. The geometric trends for naturally occurring disulfides in protein crystal structures are examined. Comparison of the disulfide-containing mutant protein structures with the wild-type structure shows that, in both cases, disulfide incorporation is accommodated by relatively minor changes in local main-chain conformation. The Cys-22-Cys-87 disulfide has two high energy dihedral angles (X2 = 121 degrees, X2' = 143 degrees). Both disulfides produce short non-bonded contacts with the main-chain.

The geometries of two disulfide bridges genetically engineered into subtilisin have been characterized by x-ray crystallography to determine the structural and energetic constraints involved in introducing disulfide bonds into proteins. Both disulfide bridges (Cys-24-Cys-87 and Cys-22-Cys-87) exhibit atypical sets of dihedral angles compared to those for other reported disulfide structures in proteins. The geometric trends for naturally occurring disulfides in protein crystal structures are examined. Comparison of the disulfidecontaining mutant protein structures with the wildtype structure shows that, in both cases, disulfide incorporation is accommodated by relatively minor changes in local main-chain conformation. The Cys-22-Cys-87 disulfide has two high energy dihedral angles (X2 = 121", X2' = 143'). Both disulfides produce short non-bonded contacts with the main-chain.
A disulfide bond is one of several types of structural elements that impart conformational stability to proteins (1). A question being addressed by protein engineering studies is the degree to which engineered disulfides can produce the stabilizing effects of natural ones. Recently, a measurable increase in the stability of two proteins has been accomplished by introducing disulfides into their structures (2). ' Although it is thought that the location of insertion of a disulfide bridge into a protein is a principal determinant in its stabilizing effect, it was unknown how well a modeled disulfide group must conform to an optimal geometry or how closely it must adhere to the stereochemical trends observed for natural disulfides in proteins in order that the disulfide bond form. We felt that insight into these issues could be gained by comparing crystal structures of protein variants containing a disulfide whose modeled geometry was somewhat structurally incompatible with the native protein.
Here we describe the modeled and crystallographically determined geometries of two atypical disulfides incorporated into subtilisin. A primary goal of the study was to determine how the resulting perturbations to the structure would be partitioned between t h e engineered disulfide and adjacent main-and side-chain groups. The structural characterization of such disulfide-containing mutants provides data regarding the relative importance of the several physicochemical factors involved in introducing disulfide bonds into proteins.
* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "aduertisenent" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. 4 Supported by Grant GM29616 from the National Institutes of Health.
J. E. Villafranca, personal communication. (3) produced disulfide-containing mutants of subtilisin, a protease of about 27,500 M, containing no natural cyst(e)ine groups. We chose two of these mutants (expressed in Bacillus subtilis), each with a common residue (Cys-87), to study crystallographically. Wells and Powers (3) demonstrated by polyacrylamide gel electrophoresis that both mutants, Ser-24"*Cys/Ser-87+Cys and Thr-22+Cys/Ser-87+Cys,' contain a disulfide bond that forms quantitatively during fermentation of the bacteria (3). In both mutants the disulfide is on the surface of the molecule, about 24 A away from the active site, and does not affect enzymatic activity (3).

EXPERIMENTAL PROCEDURES
Potential sites for introducing disulfide bridges were identified by determining the locations in the structure where a modeled disulfide group of good stereochemistry could be inserted with a minimum of main-chain rearra~~gement.~ The X3 dihedral angle was constrained to be within 10" of *go", while X I , X2, X2', and X1' were varied to best approximate staggered conformations (see diagram in legend of Table I1 for definitions of the dihedrals). Theoretical dihedral energies of disulfide conformations were calculated with a subroutine of the program AMBER (4). Modeling indicated that for C24/C87,'1 a righthanded disulfide (X3 -90") could be built requiring very little mainchain movement. The disulfide in C22/C87 could not be modeled with any combination of staggered dihedral angles without also requiring significant main-chain rearrangements. The model requiring minimal main-chain movement for Cys-22-Cys-87 was a left-handed disulfide (X3 = -85") with two high energy dihedrals (X2 = 112", X2' = 130").
Purification and Crystallization-A culture of the C24 fC87-producing bacteria was obtained from J. Wells of Genentech and grown in The Cys-22-Cys-87 disulfide-containing subtilisin mutant studied here was actually a triple mutant Tyr-21-Ala/Thr-Z2+Cys/Ser-87-Cys. The substitution of Tyr-21 for Ala was made for a purpose independent of this disulfide study. The triple mutant will be called C22 fC87 throughout this paper.
The initial modeling was done by J. Wells and R. Bott of Genentech.
Data Collection and Structure Refinement-Data were collected on an ENRAF NONIUS CAD4 diffractometer by the step scan method (7) with monochromated copper K a radiation (40 kV, 27 mA). Data collection was done in 10 shells, starting from the high resolution limit. Typically, raw intensities were obtained by summing the highest five consecutive steps (0.02" (in w)/step). The data were corrected for absorption by the method of North et al. (8). Table I contains data collection parameters and statistics. C24/C87"For the C24/C87 mutant a complete data set to 2.15 A resolution was collected in 12 days with no detectable decay. Friedel pairs collected at low and high resolution (see Table I) were locally scaled' to correct for any systematic errors. Re, (defiqed in Footnote a of Table I)  Corrections are applied to both F+ and F with the constraint that the resulting local slope of Fversus F+ be unity and that the intercept be zero. The regions of reciprocal space used in locally scaling a Friedel pair are continually updated along with the local slope and intercept, so that each pair receives its own correction. The maximum number of neighboring reflections used to scale a particular Friedel pair can be varied with resolution. Thus, at low resolution, where neighbors are farther apart in reciprocal space, the localness of the scaling can be maintained by using fewer neighbors to scale the pair. The program is advantageous in correcting data where absorption due to glass and solvent is significant and insufficiently corrected with a cp curve alone. (The original version of the program was written by A. Goldman at Yale.) Because collecting Friedel pairs for all unique data to 2 A resolution is time consuming with a conventional diffractometer, we collected Friedel pairs at low resolution and at high resolution, deduced the 4, x, and 0 dependence of the corrections, and applied them to the data lacking Friedel mates. intensity); for 2266 pairs from 2.31 to 2.15 A, R, , was 9.2%. Local scaling of the intensities of C24/C87 to those of WT gave an RZim value (defined in Footnote b of Table I), reflecting the difference between the WT and the C24/C87 structure, of 7.8%.
C22/C87"For C22/C87 a complete data set to 2.05 A resolution was collected in 14 days. No decay was observed for the low-resolution data; the average decay at higher resolution (2.15-2.05 A) after 14 days was 7%, but because these data were collected first, their measured intensities were not affected by decay. Therefore, no decay correction was applied. Local scaling of F+ to F yielded an R,, value of 2.4% for the data to 3.3 A. Locally scaling the C22/C87 intensities to those of WT gave an RZ,, value of 7.8%.
Structure Refinement-Initial models of the disulfide structures were built from difference electron density maps calculated with terms

RESULTS
The C24/C87 difference (F(C24/C87) -F(WT), &T) map at 2.15 A resolution showed two resolved positive peaks corresponding to the scattering of the sulfur atoms and two negative peaks at the positions occupied by the Ser-24 Oy and Ser-87 07 atoms in the W T structure (Fig. 1). A model of a C24/C87 double cysteine mutant was made by replacing Ser-24 and Ser-87 with cysteines and rotating the X1 dihedral of Cys-24 by -85" and of Cys-87 by -173", with respect to the WT serine orientations, to yield a right-handed disulfide geometry ( X 3 = 85"). The fit was further optimized by a $ rotation of -9" about the CW87-C87 bond.
The program used to locally scale mutant data sets to the wild type data set determines, for each reflection, the local slope of Zhkl (mutant) uersus Zhkl (WT) and applies the slope to zhu (mutant). The calculations can be weighted by l/Zhy. The other algorithms of this program are similar to those in the program described in Footnote 5.
With the data collection strategies used here, such a local scaling procedure can greatly minimize the effects of x-ray decay on difference Fouriers. The signal-to-noise ratio in the F(mutant) -F(WT) difference electron density maps was shown to be dramatically improved compared to that in the corresponding maps obtained when the mutant data are scaled to the WT data with only one scale factor and one temperature factor.
A left-handed Cys-~2-Cys-87 disulfide group was fit to the density in a 2.05 A resolution (2F(C22/C87) -F(WT), dWT) map. The difference map is shown superimposed on the W T structure in Fig. 2. A good fit of the disulfide was accomplished by small $J,d rotations (less than 10") of residues 85-90, along with a 40" rotation of the Cys-87 X1 dihedral. No rotation of the Cys-22 dihedral was needed. The need for $J,d rotations to fit the density in the disulfide region was later determined to result from a translation in the crystal lattice of the entire C22/C87 structure with respect to WT.' 'The finding here that in highly isomorphous crystals there can be subtle but real changes in the packing of protein molecules in the lattice identifies the need, when comparing two structures, to consider the possibility that long range systematic differences could sometimes be artifacts of crystallization.
The resulting disulfide dihedral angles for the initial unrefined crystal structures of C22/C87 and of C24/C87 are given in Table 11. In most respects, the experimentally determined disulfide structures agree well with modeling predictions. The Cys-22-Cys-87 structure showed that the disulfide modeled with high energy X2 and X2' dihedral angles was the correct one. Formation of the Cys-24-Cys-87 disulfide requires much larger side-chain rotations than formation of Cys-22"Cys-87. Disulfide incorporation in both cases results in minor subtle local rearrangement in main-chain conformation.
No large changes in structures were observed between the initial and refined disulfide structures. As inferred from modeling, little main-chain alteration occurs in C24/C87 when compared to WT. In the crystal structury of C22/C87, the entire molecule is translated by about 0.3 A, primarily in the

I1 The most highly populated classes of leftand right-handed disulfide geometries, determined by inspection of disulfide dihedrals observed in protein crystal structures, along with the modeled and crystallographically observed geometries of Cys-24-Cys-87 and of Cys-22-Cys-87
The dihedral angles in a disulfide group are defined in the diagram below.
Note that the X3 dihedral energy function is a sum of a 2-fold term whose minima occur at k90" plus a smaller 3fold term whose minima occur at +60', 180"; the minima in the overall X3 dihedral energy function, therefore, occur at k83", close to the values observed in some small molecule crystal structures. The Ca-Ca and other distances were calculated assuming ideal bond lengths and bond angles. Angles given are in degrees, distances are in A, and energies are in kcal/mol. Standard deviations in dihedral angles are in parentheses.
Left-handed disulfide geometries in the protein crystal structures tabulated here. e Cys-58-Cys-42 in 7 homologous serine proteases. e Right-handed disulfide geometries in the protein crystal structures tabulated here as well as some "miscellaneous" right-handed geometries occurring in proteins (data not shown). direction of the.b axis, with respect to the W T molecule. After a rigid body rotation and translation matrix is applied to superimpose C22/C87 on WT (resulting root -mean square deviation between the two structures = 0.18 A), the mainchain in the disulfide region shows only minor differences. The structures of the two mutants compared to WT are shown in Fig. 3. Fig. 4 shows the region of the subtilisin molecule where the disulfides were introduced.

DISCUSSION
Geometric Trends of Naturally Occurring Disulfides in Proteins-Available protein structure coordinates indicate that two general disulfide families can be defined by the chirality of the dihedral angle about the sulfur-sulfur bond righthanded (X3 -90") and left-handed (X3 --90"). The families can be subdivided according to their other side-chain dihedral angles into several distinct classes (1,12). Because the data base of protein crystal structures has expanded since Richardson's compilation (12), we have re-examined the geometric trends in the disulfide families.
The dihedral angles for the most highly populated geometric types of right-and left-handed disulfides in the protein data bank (5) are given in Table 11. Right-handed and left-handed disulfide groups in proteins differ in several geometric characteristics. Whereas many (15 out of 39) left-handed disulfide groups in proteins adopt (approximately) a preferred geometry (called a left-handed spiral (X1 = -60°, X 2 = -60", X3 = -85", X2' = -60", X1' = -60")), the right-handed groups are more variable in their dihedral angle patterns. There are many more examples of right-handed disulfides with X1 (or XI') = 60" than there are left-handed ones. Finally, Richardson (12) pointed out that while the average Ca-C? distance of left- Ser-221 is in the foreground at the end of the long helix running through the middle of the molecule.
structures, tbat of right-handed ones is somewhat smaller (5.07 f 0.73 A in our survey). Table I1 are calculated values of the dihedral energies for each disulfide class. On average, the left-handed types have lower calculated dihedral energy (1.68 f 1.51 kcal/ mol uersus 3.19 f 1.43 kcal/mol). Crystallographic (13-16) and theoretical (17) studies on small molecules have established that the electronically and sterically preferred values of X3 are close to f83". In proteins, the average X3 dihedral angle for left-handed groups agrees well with these findings (X3 = -85" f 9"); however, for right-handed groups the average is 99" f 11".

Included in
Comparison of C24/C87 and C22/C87 Disulfide Geometries with Naturally Occurring Disulfide Geometries-From the current structures in the Protein Data Bank (5) we found that neither Cys-24-Cys-87 nor Cys-22-Cys-87 fit well into any geometric category with respect to all five dihedral angles.
The Cys-22-Cys-87 disulfide adopts a novel set of dihedral angles. Although X1 (53") is of low dihedral energy, the occurrence in proteins of disulfides with similar Xls (or Xl's) is rare, presumably because of unfavorable non-bonded interactions between ST and the main-$hain (12). C22/C87 has a short Sy22-N22 contact of 3.02 A and a short Sy87-N87 contact of 2.92 A. Short 1,4 sulfur-carbon (methylene or methyl) interactions have been observed in a significant number of small molecule organosulfur compounds (la), and semiempirical molecular orbital calculations suggest that favorable but weak 1,4 sulfur-carbon interactions can occur (18). The unique geometry of Cys-22-Cys-87 is also reflected in its X2 (121") and X2' (143") values. The former dihedral is of essentially maximal X2 dihedral energy (the maxima in the X2 dihedral energy function occur at OD, k120"). (Similar dihedral strain also occurs in some rare cases in natural proteins (vide infra).) The dihedral strain energy in disulfide groups can also be partitioned nearly equally among all five dihedrals. The dihedral energy barrier to rotation about a RCH,"SR bond (defining X2 or X2') is lower than the other dihedral energy barriers in a disulfide group, and thus X2 and X2' are expected to be the most variable?
Insight into the amount of strain in a disulfide due to its geometry and its local environment can be obtained by measuring the disulfide's redox potential. For example, we find a correlation between the redox potentials of the three disulfide groups in BPTI (19) and their calculated dihedral energies (data not shown). Since the disulfides in C24/C87 and in C22/C87 are in the same region of the molecule and share a common residue, these mutants comprise an ideal system to relate changes in disulfide structures to differences in redox potentials. The disulfide bond in C22/C87 has a higher calculated dihedral energy than that in C24/C87 (4.8 kcal/mol versus 2.5 kcal/mol). Wells and Powers (3) experimentally showed that the C22/C87 disulfide is the more strained by determining the redox potentials of C24/C87 and of C22/C87.
Attempts to quantify the thermal stabilities of C24/C87 and of C22/C87 have been complicated by aggregation and autolysis." Although Wells and Powers (3) have shown that neither mutant is more stable toward autolysis than WT, a definitive evaluation of the stabilities of these disulfides must await a full thermodynamic treatment.
With regard to developing strategies for engineering disulfides, it is important to recognize that a disulfide with a nonoptimal high energy geometry may still markedly increase thermodynamic stability; disulfides with high energy dihedrals are found in some very stable proteins. For example, reduction of the high energy Cys-14-Cys-38 disulfide in BPTI causes a 21" decrease in the reversible thermal denaturation midpoint temperature at pH 2.1 (20). Removal, by site-directed mutagenesis, of either the Cys-14-Cys-38 crosslink or of another high energy disulfide in BPTI, Cys-30-Cys-51, in each case effects a similar pronounced decrease in the thermal stability of the protein." The decrease in entropy in the denatured state of BPTI presumably outweighs the strain that these disulfides add to the native state. Ueda et al. (21) have described the effect of the strain of intramolecular cross-links of various lengths, introduced by chemical modification into lysozyme, on the stability of this protein.

CONCLUSIONS
The primary goals of this structural study were 1) to evaluate the nature and extent of the structural changes that accompany the introduction of disulfide bonds into proteins and 2) to identify the structural and energetic criteria and In subtilisin, we have observed that the dihedrals about the Cy-S6 bonds of 3 out of the 5 methionines are near +120". lo When phenylmethanesulfonyl subtilisin is heated, the Ser-221 Oy-sulfur bond is hydrolyzed regenerating active enzyme that autolyzes.
constraints most important in influencing conformations of disulfide cross-links. Summarizing our results, we come to the following conclusions.
1. Disulfides with atypical dihedral geometries can be inserted into proteins by site-directed mutagenesis. Note, however, that their description as atypical should be qualified by the limited size and accuracy of the data base available.
2. Both engineered and natural disulfides can manifest high energy dihedral angles as well as unfavorable non-bonded interactions. Dihedral strain can be reflected in X2 (or X2') dihedral angles near +120".
3. Modeling can give reasonable insight into locations in a protein structure where introduction of a disulfide cross-link is possible and into the resulting disulfide's geometry.
4. Insertion of the disulfide is accommodated by the protein's ability to relax around the structural constraints of the cysteine substitutions. The substitutions are accommodated by small subtle structural rearrangements and energy compromises localized about the mutagenesis site.
5. From the point of view of engineering disulfides into structures to increase their stabilities, one need not always be restricted to disulfides of optimal low energy geometries. Indeed, disulfides of relatively high dihedral energy can still be quite effective at imparting to the protein molecule increased stability toward unfolding.