An approach to crystallizing proteins by synthetic symmetrization

Abstract

Previous studies of symmetry preferences in protein crystals suggest that symmetric proteins, such as homodimers, might crystallize more readily on average than asymmetric, monomeric proteins. Proteins that are naturally monomeric can be made homodimeric artificially by forming disulfide bonds between individual cysteine residues introduced by mutagenesis. Furthermore, by creating a variety of single-cysteine mutants, a series of distinct synthetic dimers can be generated for a given protein of interest, with each expected to gain advantage from its added symmetry and to exhibit a crystallization behavior distinct from the other constructs. This strategy was tested on phage T4 lysozyme, a protein whose crystallization as a monomer has been studied exhaustively. Experiments on three single-cysteine mutants, each prepared in dimeric form, yielded numerous novel crystal forms that cannot be realized by monomeric lysozyme. Six new crystal forms have been characterized. The results suggest that synthetic symmetrization may be a useful approach for enlarging the search space for crystallizing proteins.

Although some of the steps involved in determining crystal structures of macromolecules have become routine, important obstacles remain. In fact, the successful automation of much of the process has brought increasing pressure on the remaining problematic steps. Among these, obtaining ordered three-dimensional crystals remains one of the most challenging and least predictable bottlenecks in the structure-determination pathway (1–4). Because structural analysis plays a vital role in biomedical research and is also the subject of numerous national and international science initiatives, further developments in protein crystallization could have a significant impact.

When a particular target protein fails to crystallize readily, it can be more fruitful to expand crystallization trials to include variations on that protein, rather than continuing with exhaustive crystallization trials (5). The use of homologous proteins is one way to increase the chances of obtaining good crystals, because even highly similar proteins tend to exhibit distinct crystallization behavior (6, 7). The structure determination of the transmembrane mechanosensitive channel provides an example of this strategy, where crystallization trials of homologous proteins from nine species were investigated to obtain well ordered crystals for a single protein (8).

Other established approaches for expanding the opportunities for protein crystallization include mutagenesis and chemical modification. Mutagenesis can be applied strategically, for example by changing flexible side chains such as lysine and glutamate residues to less flexible ones, such as alanine (9–12). Chemical modification offers less versatility, but reductive lysine methylation has been used effectively to alter the crystallization properties of proteins (13–17).

Here we suggest a novel approach to enhance crystallization, by synthetically symmetrizing proteins (Fig. 1). Two advantageous elements are combined. First, variation is provided by the ability to dimerize a protein about a single cysteine residue introduced at various positions. Second, we have argued from earlier studies that symmetric (i.e., homooligomeric) proteins might crystallize more readily than asymmetric (i.e., monomeric) proteins (18, 19). Why might this be the case? An argument can be made on intuitive grounds, noting that homooligomeric proteins are inherently rotationally symmetric, a property that is found in the majority of macromolecular crystals. Oligomeric proteins can therefore be viewed as existing in a state that is partway along a path to the crystalline state. More quantitatively, Wukovitz and Yeates (19) showed that an oligomeric macromolecule needs to participate in fewer distinct fortuitous crystal contacts to form a connected network in three-dimensional space. Whether such theoretical considerations actually translate to a greater success in crystallizing symmetric proteins is difficult to assess. However, supporting statistical arguments can be made based on a comparison of the crystallization behavior of monomeric vs. oligomeric proteins. An estimate is obtained that dimeric proteins crystallize more readily than monomeric proteins by a factor of ≈1.5. Combining symmetrization with site directed mutagenesis therefore makes it possible to generate, from a given protein, several variants having potentially favorable crystallization properties. Here, we demonstrate the utility of the approach in preliminary experiments, using T4 lysozyme as a model system, by producing numerous new crystal forms of this widely studied enzyme.

Fig. 1.

Diagram illustrating the idea of symmetrization combined with cysteine mutagenesis. The protein molecule is represented by the figure “5.” Cysteine residues are yellow. Each construct would be expected to present independent crystallization opportunities. Dimerization is achieved by direct disulfide bond formation. Trimerization would require a trivalent thiol-specific cross-linking reagent. Trimerization experiments are not discussed here.

Results

Synthetic Homodimerization.

We selected a naturally monomeric protein, bacteriophage T4 lysozyme, as a well studied model crystallization system in which to investigate whether artificial symmetrization, applied to various mutants, might lead to additional opportunities for successful crystallization. The method chosen for synthetic symmetrization was homodimerization via disulfide bond formation between monomers containing a single-cysteine surface mutation. Three single-cysteine variants were constructed in a pseudo-WT background from which the two native cysteine residues (Cys-54 and Cys-97) had been removed. This was done to avoid any unwanted reactivity of the native cysteine residues, both of which have limited solvent accessibility but might interfere nonetheless. The experiments reported here are similar in some respects to those of Heinz and Matthews (20), who introduced cysteines at sites packed at a known crystal lattice interface. They found that disulfide bond formation before crystallization enhanced the rate of formation of the anticipated WT crystal form (P3₂21; a, b, c = 61 Å, 61 Å, 97 Å; γ = 120°). Here, our purpose is different: to see whether creation of essentially arbitrary dimer constructs will open up entirely new crystallization possibilities not available to the protein in its monomeric form.

Sites for engineering cysteine residues were chosen from three different regions on the surface of the lysozyme molecule (Fig. 2). The three residues chosen for mutation to cysteine (S44, V131, and D72) fall in regions referred to as the “tail,” the “head,” and the “backbone,” respectively. Based on the shape of the lysozyme molecule, the S44C and V131C variants were chosen to minimize steric clash between the two fused subunits. In contrast, the D72C construct fuses T4 lysozyme along its backbone, with an expectation of greater steric hindrance in the dimerized state. All three of the constructs dimerized readily when oxidized, resulting in dimer yields exceeding 95% in all cases (see Methods). All three of the synthetic dimers formed crystals in <24 h during crystal screening experiments, in a variety of conditions. For all three dimeric constructs, crystals that diffracted well (1.9–2.5 Å) were obtained from preliminary screens without optimization of crystallization conditions.

Fig. 2.

The sites for generation of three single-cysteine lysozyme mutants used in synthetic dimerization and crystallization experiments. The native cysteine residues (C54 and C97) that had been removed by earlier mutagenesis (30) are shown in red. The individual sites for new cysteine residues (S44, D72, and V131) are shown in yellow.

Synthetic Dimers Produce Six New Crystal Forms.

In total, six new crystal forms of lysozyme were characterized by x-ray diffraction and their structures were determined by molecular replacement (Fig. 3 and Table 1). In all six cases, the dimeric lysozyme constructs appeared as symmetric dimers (i.e., obeying twofold rotational symmetry). All six crystal forms are distinct from the many forms already reported for lysozyme (see Table 2, which is published as supporting information on the PNAS web site). In some cases, a subset of the crystal contacts were similar to contacts seen previously in structures of monomeric lysozyme. For example, in one of the D72C crystal forms (Table 1, iv) the molecules make crystal contacts similar to the head-to-tail mode of association noted earlier (21). From the six distinct crystal forms, one representing each of the three dimeric constructs was subjected to crystallization optimization, structure determination, and atomic refinement (see Table 3, which is published as supporting information on the PNAS web site). Three other crystal forms for D72C were used, without further efforts to improve the crystals beyond the screening stage, to determine the packing of molecules in the crystal by molecular replacement methods. The six crystal forms are described in Table 1.

Fig. 3.

Six new crystal forms of synthetic T4 lysozyme dimers. (A) Crystal packing diagrams. In each case, red dots indicate the location of the single engineered disulfide bond. The annotation below each image indicates the site of the cysteine residue, the crystallization conditions (see Methods), and the crystal space group (see Table 1). (B) Electron density maps confirming the presence of disulfide bonds in the synthetic dimers: crystal form i (Left) V131C, crystal form ii (Center) S44C, and crystal form v (Right) D72C. 2F_obs − F_calc maps are shown, contoured at 1.0 standard deviations, showing that the disulfide bonds (yellow) are present in all cases.

Table 1.

Six new crystal forms of dimerized T4 lysozyme mutants

Fig. 3 assignment	Mutant*	Crystallization condition	Space group	Unit cell parameters, Å, °	Resolution, ņ
i	V131C	20% PEG 8000	C2	184.3, 29.4, 31.2	2.0
		0.1 M Na cacodylate, pH 6.5		90, 97.4, 90
		0.2 M Ammonium acetate
ii	S44C	2.0 M ammonium sulfate	C2	92.6, 28.9, 88.8	1.8
		0.1 M cacodylate, pH 6.7		90, 119.4, 90
		0.2 M NaCl
iii	D72C (CSC9)	4.0 M sodium formate	P21	36.5, 86.9, 55.4	2.1
				90, 102.9, 90
iv	D72C (IndH7)	0.15 M DL-malic acid, pH 7.0	P21	49.9, 56.0, 80.7	2.2
		20% PEG 3350		90, 93.0, 90
v	D72C (PegD9)	0.2 M tri-lithium citrate	P21	55.3, 55.9, 80.3	2.35
		20% PEG 3350		90, 108.1, 90
vi	D72C (WizC4)	20% PEG 3000	P3221	59.5, 59.5, 149.6	2.2
		0.1 M Hepes, pH 7.5		90, 90, 120
		0.2 M sodium chloride

*For theD72C mutant, the annotation includes the initial crystallization condition (see Methods).

^†For crystal forms i, ii, and v, resolution is the minimum Bragg spacing of the diffraction data used for refinement. For the other crystal forms, the resolution is the minimum Bragg spacing of the diffraction data obtained during data collection and processing. Refinement data for crystal forms i, ii, and v are described in Tables 2 and 3.

In each of the three structures that were refined, the presence of the engineered disulfide bond was confirmed by examining electron density maps. The disulfide bonds were determined to be in configurations typical of natural protein disulfide bonds, with calculated dihedral strain energies that were slightly lower than the average values for natural disulfide bonds in the S44C and V131C mutants and somewhat higher than average in the D72C mutant (22). This higher strain along the D72C disulfide bond might be explained by the greater chance for steric collision caused by cross-linking lysozyme along its backbone.

It is difficult to quantify how many additional crystal forms could have been obtained from the synthetic dimers, because only a subset of the preliminary crystals observed were carried through to structure determination. However, the results of a control experiment are notable. The D72C construct was subjected to preliminary crystallization trials in both its dimeric and monomeric forms (without cross-linking, and in the presence of the reductant 50 mM 2-mercaptoethanol). Whereas the dimeric form yielded preliminary crystals in a total of 98 of 384 conditions upon screening (14 days at room temperature), the monomeric form yielded preliminary crystals under only 14 of 384 conditions (14 days at room temperature). These values almost certainly do not account for all of the crystal forms that could be obtained from the two samples, but the results do support the notion that the success rate for crystallization is increased measurably in the dimeric form.

Crystallization Tendencies of Oligomeric Proteins.

The experimental approach demonstrated here is motivated in part by the idea that internal symmetry is an advantage for crystallization. To evaluate what effect oligomeric state has on the crystallization behavior of proteins, we surveyed the Protein Data Bank (PDB) (23), comparing the space group preferences that are exhibited by dimeric and monomeric proteins. Oligomeric proteins sometimes crystallize so that their internal symmetry is embodied within the crystal symmetry, such as when a dimer sits with its axis of symmetry on an axis of twofold symmetry in the crystal space group. In other cases, oligomeric proteins crystallize in such a way that the crystal symmetry takes no advantage of the internal symmetry of the oligomer. In such cases, the symmetric oligomer is crystallizing essentially as if it were an asymmetric monomer. We assert that the presence of internal symmetry in an oligomer does not detract from its opportunities to crystallize in an asymmetric fashion. To the degree that this is true, those situations where oligomers take advantage of their internal symmetry can be fairly counted as extra opportunities for crystallization that were made available by virtue of the built-in symmetry. To compare overall likelihoods of crystallization, the monomeric and dimeric space group distributions were normalized to put them on the same scale. This was done according to the behavior of the two sets of proteins across only those space groups for which the crystal symmetry would not be expected to have a significant effect on crystallization, i.e., space groups lacking pure twofold axes of symmetry. Then, a number was calculated corresponding to the hypothetical number of dimer crystals (in all space groups) that one would predict if the dimeric state conferred no advantage (i.e., basing the prediction on the space group preferences exhibited by monomeric proteins). Comparing the total actual number of dimer crystals with this hypothetical number leads to a ratio for the enhancement of crystallization in the dimeric state.

For monomeric proteins, one finds a 57% (1,905 of 3,345) occurrence of crystals in space groups that do not contain a pure twofold axis of symmetry, and a 43% (1,440 of 3,345) occurrence of crystals in space groups that do contain such a symmetry operator. The crystallization behavior of dimeric proteins is notably different (Fig. 4A). Of the 2,565 distinct dimer crystals, only 999 (40%) are in space groups that do not support an internal twofold axis of symmetry. According to the percentages obtained for monomeric proteins, in the absence of special space group preferences, one might expect to find 753 (999 × 0.43/0.57) dimeric proteins in space groups containing a twofold axis of symmetry. This would contribute to a total of 1,752 (999 + 753) expected dimer crystal structures in the data set. In contrast, we observe 1,566 dimeric proteins in space groups containing a twofold axis of symmetry, for a total of (1,566 + 999) 2,565 observed crystals of dimeric proteins. According to this reasoning, the overall enhancement in the likelihood of crystallization for a dimeric protein is a factor of ≈1.46 [obtained from (999 + 1,566)/(999 × (1 + 0.43/0.57)) = 2,565/1,762 = 1.46]. This finding is illustrated in Fig. 4B. A parallel calculation for trimeric proteins leads to a crystallization enhancement ratio of 2.0 [obtained from values corresponding to those in the dimeric calculation, (294 + 336)/(294 × (1 + 209/3,136))].

Fig. 4.

An analysis of crystal space group preferences in the PDB. (A) Comparison of the results for monomeric and dimeric proteins. (B) An illustration of the calculation of increased crystallization likelihood for dimeric proteins. The height indicated by “expected” is the number of dimer crystals that would be expected if dimeric proteins gained no crystallization advantage from their internal symmetry. The ratio of observed to expected heights is 1.46.

Some complicating issues can be raised regarding the arguments above. For example, if certain dimers tended to crystallize in several space groups (including some with twofold axes of symmetry and some without) while other dimers tended to not crystallize at all, then the appearance of extra cases in space groups with twofold symmetry axes might not constitute an increased success rate. Such considerations might lower the enhancement factors obtained but would not entirely offset the advantage gained by internal symmetry.

As noted above, our statistical arguments are based on the assumption that the presence of internal symmetry in an oligomer does not diminish its capacity to crystallize in a space group that lacks a corresponding symmetry element. To make an argument free of such an assumption would require available data on the success and failure rates for crystallization for proteins of various types. Structural genomics efforts may make such analyses possible in the future.

Discussion

Lysozyme is well known for the ease with which it crystallizes. A BLAST search (24) in the PDB using the pseudo-WT T4 lysozyme sequence [PDB entry 3LZM (25)] produced 446 structures (E-value cutoff = 10) of which 428 are of T4 lysozyme in an uncomplexed form (E-value cutoff = 1 × 10⁻⁶⁶). Most of these structures were determined by the Matthews group (21). An examination of the reported structures (which include both native and mutant forms of lysozyme) finds that monomeric lysozyme is able to crystallize in some 37 distinct crystal forms. Interestingly, the last new crystal forms were reported by He et al. (26) in 2004 using T4 lysozyme mutants generated by alanine scanning mutagenesis. In three of the four constructs, each bearing multiple mutations, new crystal forms were obtained owing partly to structural changes at the surface of the molecule. In fact, the great majority of the new crystal forms reported since the survey by Zhang et al. (21) in 1995 are for molecules bearing numerous amino acid changes compared with WT T4 lysozyme.

This suggests that the possible crystal forms that can be obtained by WT or near-WT T4 lysozyme have been effectively exhausted. It is therefore remarkable how easily numerous additional crystal forms can be obtained through dimerization. Inspection of the new crystal forms suggests that the disulfide bonds between proteins contribute critically to the connectivity of the protein molecules in the various crystal forms. It seems unlikely that any of them could have been obtained from native monomeric lysozyme. Furthermore, the crystal forms achieved by one mutant could not have been formed by the other mutants. Each of the dimerized species benefits from unique crystallization opportunities.

In some of the new crystal forms, the dimers take advantage of crystallographic symmetry as discussed above. This is the case for the individual structures that were characterized for mutants S44C and V131C. In both cases, the engineered dimer sits with its internal symmetry axis on a crystallographic twofold axis of symmetry in space group C2. These cases, where the internal symmetry of the dimer is used to build up the full symmetry of the crystal, illustrate the incremental advantage in crystallization that is conferred by symmetrization. In the four structures determined for mutant D72C, the dimers do not exploit crystalline symmetry but instead are arranged to have a complete dimer in the asymmetric unit of the crystal. This mixture of situations, where the local symmetry may or may not be used by the crystal, is consistent with the space group preferences exhibited by crystalline dimeric proteins.

There are a number of potential limitations to the strategy proposed here for crystallization of a protein with unknown structure. One anticipated limitation was that the engineered disulfide bond might adopt several discrete rotameric states, potentially offsetting any advantage that could be gained by engineering internal symmetry. At least in these first experiments, this does not appear to be the case. In fact, despite the potential flexibility, all six structures revealed dimers that were symmetric, whether or not the internal symmetry of the dimer was exploited as part of the crystal symmetry. This is a favorable observation with regard to the anticipated tradeoff between undesirable flexibility and advantageous internal symmetry. It is also interesting in the sense that it demonstrates the perhaps unexpected tendency of even artificially connected monomers to associate in strictly symmetric fashion, despite the potential for rotation about the bonds that link the two monomers. Bond rotations could have given rise to asymmetric arrangements, especially in space groups lacking twofold axes of symmetry.

Other limitations might also be noted. When the structure of the protein is not known (and cannot be modeled) in advance, the surface accessible residues for cysteine placement will generally not be known. This problem might be mitigated by the strategy of testing multiple sites for mutation, as well as by the generally good accuracy with which surface residues can be predicted from sequence data. Polar or charged residues that are not conserved between homologues are generally found on the surface of the protein, and algorithms for making such predictions are available (27–29). Potential surface sites could also be verified experimentally by reacting the cysteine mutant with a commercially available methane-thiosulfonate spin label and then examining the X-band EPR spectrum of the spin labeled protein. The EPR spectrum of a helix surface site is readily identifiable (30). Such a strategy could be useful in limiting the cost of scaling up purification of unreactive cysteine mutants.

Overall, the results support the potential utility of synthetic symmetrization as a crystallization strategy, especially when combined with mutagenic variation. The experimental techniques involved are straightforward and may be suited to both the individual research laboratory and high throughput crystallography settings. The favorable preliminary results motivate further experiments on a wide range of proteins, including those that are difficult to crystallize, to evaluate the generality of the approach described here.

Methods

Mutagenesis, Dimerization, and Purification.

Single-cysteine mutants were generated previously from T4 lysozyme pseudo-WT as described by McHaourab et al. (30). Mutant plasmids were chemically transformed into Escherichia coli strain BL21(DE3) (Invitrogen). Cultures, grown in LB (100 μg/ml ampicillin), were induced with 1 mM IPTG (final) at an OD₆₀₀ of 1.0 and then allowed to shake at 225 rpm for an additional 1.5 h at 37°C. Cell pellets were harvested and then stored at −20°C. Frozen cell pellets were lysed by a combination of osmotic shock and sonication in lysis buffer A (25 mM Tris/25 mM Mops/1 mM EDTA, pH 7.5). Soluble lysate was filtered (22-μm pore size, Millex; Millipore) before further purification on an ÁKTA FPLC (GE Healthcare). Soluble protein was bound to a cation-exchange chromatography column (HiTrap SP Sepharose High Performance; Amersham Biosciences) in lysis buffer A with 5 mM DTT to keep cysteines reduced. Initial purification occurred over an elution of 0 mM to 500 mM NaCl over 150 ml. Relatively pure lysozyme eluted ≈200–300 mM NaCl. Additional protein purification with simultaneous removal of DTT was achieved by gel-filtration chromatography (Superdex 75; Amersham Biosciences) in lysis buffer A with 150 mM NaCl.

Single-cysteine mutants were dimerized by spontaneous air oxidation over several days in the absence of reducing reagents, by addition of CuSO₄ (10 mM final concentration), or by addition of the reagent 4,4′-dithiodipyridine (one-half the molar protein concentration), which produces a direct disulfide bond between protein subunits and allows the cross-linking reaction to be monitored spectrophotometrically at 323 nm (31). Dimers were separated by gel-filtration chromatography and then dialyzed overnight into 50 mM Tris·HCl/25 mM NaCl (pH 7.5). Pure dimers were then concentrated to 40 mg/ml in preparation for crystallization.

Structure Determination.

A total of 384 crystallization conditions were screened by the hanging-drop, vapor-diffusion method. Commercial screens (Crystal Screen HT, Index HT, PEG/Ion Screen, Natrix, Hampton Research), and Wizard 1 and 2 (Emerald Biosciences) were purchased and prepared in 96-well format (100-μl reservoir volume). One-microliter hanging drops (500-nl protein plus 500-nl reservoir) were prepared by using the Mosquito pipetting robot (TTP LabTech). Crystal trays were incubated at room temperature over the course of screening. Space groups and crystallization conditions for the six crystal forms analyzed are given in Table 1.

Crystals were mounted in loops by using either the mother liquor or oil as cryo-protectant and checked for diffraction on the RAXIS IV. Diffraction data were collected at a wavelength of 1.54 Å on the RAXIS IV++ or at 1.0 Å using beamline 8.2.2 at the Lawrence Berkeley National Laboratory synchrotron facility. Diffraction data were processed with the Denzo/Scalepack package (32). Structures were solved by molecular replacement with PHASER (University of Cambridge, Cambridge, U.K.) and starting models PDB entries 168L (21) or 1C6T (33). Models were iteratively refined and rebuilt with REFMAC (34) and COOT (35), respectively. Refined structures for crystal forms i, ii, and v have been deposited in the PDB (ID codes 2HUK, 2HUL, and 2HUM, respectively).

Tabulation of Oligomeric Space Group Preferences.

Data sets for the statistical analysis of space group preferences for oligomers were obtained by searching the Protein Quaternary Structure Database for either monomers or homodimers (36). To remove redundancy in the data sets, both lists of monomers and dimers generated from the PQS search were compared with a list of PDB files clustered by 90% sequence homology with the program CD-HIT (37). Within each cluster, only those PDB structures having unique space-group and unit-cell parameters were included. To reduce the number of false-positive homodimers from the data set (monomers misclassified as homodimers by PQS), structures with a change in accessible surface area (ΔASA) upon dimerization of <830 Ų per subunit were removed from the data set (38). In addition, because monomeric T4 lysozyme is represented hundreds of times in the PDB, a subset of the lysozyme structures was erroneously classified because of misclassification of a crystal contact as a dimer interface. These were also removed from the data set. Finally, structures classified as homodimers by PQS were checked for twofold rotational symmetry by using the program LSQMAN (39). This was necessary to limit the number of false-positive homodimer identifications arising from crystal contacts. Deviations of 1° from 180° rotation and 1 Å from pure translation were allowed. This filtering reduced the number of dimers to be analyzed from 10,970 to 2,565 and the number of monomers from 13,973 to 3,345.

Abbreviation

PDB

Protein Data Bank.