Towards high throughput and dealing with awkward cases:

A snapshot of recent crystallization articles in Acta Cryst. D

The past years have seen many advances in our ability to solve macromolecular structures and have considerably shortened the time to go from crystal to publication. Cryogenic techniques and more powerful beamlines have resulted in structures often being obtained from single crystals one or more orders of magnitude smaller than practical a few years ago. The success of the human genome sequencing project highlights the need for new methods for high throughput crystallization and structure determinations. One can imagine that the growth in this field will result in thousands of structures being published in a form similar to small-molecule papers rather than first announcing the crystallization conditions and subsequently publishing the structure and the coordinates. Increased demand has turned the attention of many researchers to crystallization, the main bottleneck to structure determination, i.e. finding conditions to get any crystals and optimization of conditions to turn microcrystals into usable crystals. Crystallization is a multidisciplinary science encompassing biology, chemistry, physics and computing.Embarking on the crystallization of a new protein, one is faced with the choice of where to begin the trial-and-error process. In the past this was guided by previous experience from the purification process, which indicated conditions where the protein is most stable and soluble, likely precipitants, etc., and included one's previous experience with other proteins. To some extent this historical culture has been supplanted by the introduction of rational screening kits. Here, the collected published wisdom has been distilled into a collection of the best recipes of the past. The kits are proving very successful, leading to a boom in crystals from all corners of the living world, from plant, insect, animal or human sources, but the results still rely on trial and error. Even when the screens do not produce crystals they can provide information on solubility useful to the design of conditions suitable for crystallization as in the case of insect neurotoxins from scorpions reported by Guan and co-workers [Acta Cryst. D56 (2000), 1012-1014]. Screening kits have been systematically expanded to include crystallization cocktails that are optimized for different types of macromolecules. Hennessy and co-workers [Acta Cryst. D56 (2000), 817-827] have taken the next logical step. They have mined the Biological Macromolecular Crystallization Database (BMCD), first dividing the macromolecules into hierarchical classes based upon their function and type, and then looking for similarity in the crystallization conditions for each class. A statistical analysis of their database showed significant correlation between the families and conditions employed. This is particularly fascinating, given the potential for variation in protein structure and the fact that even relatively modest modifications, proteolytic clipping, genetic mutations, etc., can determine whether or not one obtains crystals. This procedure has been incorporated into software, which can be used to track ongoing crystallization experiments and extend a laboratory's database. The input data for a macromolecule can include information on its biochemical properties, amino acid composition, solubility, α-helix and β-sheet content determined by circular dichroism and secondary and tertiary structure predicted by various algorithms.

One possible difficulty with the approach of Hennessy et al. is addressed in a paper on the crystallization of the matrix protein from Ebola virus [Dessen et al., Acta Cryst. D56 (2000), 758-760]. This protein lacks primary and secondary structural similarity with other known viral matrix proteins, i.e. it has functional but not structural similarity. According to the Hennessy approach, the functional requirements would lead to some similarity in structure, thereby giving a similarity in crystallization behavior. The convergence of the structural and functional similarity is presumably the reason why the hierarchical classification and crystallization approach works. However, other data may be able to correct a strictly functionally based approach. In the case of the Ebola protein, secondary-structure prediction analysis suggests a high β-sheet content, in contrast with the highly α-helical structures of matrix proteins from retroviruses [Conte & Matthews, Virology 5 (1998), 191-198] and influenza virus [Sha & Luo, Nature Struct. Biol. 4 (1997), 239-244]. When a primary sequence is known, structural inferences based upon it can influence the initial crystallization trials.

One can readily imagine that in the future, groups of laboratories would be able to combine their databases, or add their databases to a central collection, to create a crystallization superdatabase. One problem of the existing BMCD is that it reports only the conditions that gave the best crystals. Marginal crystallization conditions and failures are not reported. It is obvious that knowing where not to look can also be important for obtaining crystals.

Membrane proteins, as a class, are among the most reluctant to crystallize. The use of detergents was a significant breakthrough but has not produced a major increase in the number of membrane protein structure determinations. By binding amphiphilic detergents to the hydrophobic domains these proteins can be made more hydrophilic, and hopefully crystallizable by more conventional methods. Cubic phases are now taking center stage as an alternative approach in membrane protein crystallization [Chiu et al., Acta Cryst. D56 (2000), 781-784]. Rather than mask the hydrophobic part of the protein, these authors make the environment more hydrophobic with a lipid matrix. Use of this technique requires an understanding of the lipid phase diagram, particularly with respect to the detergents used to solubilize the protein. Once the lipid phase behavior was understood, the authors could use a single lipid, monoolein, as a crystallization matrix for several test membrane proteins (two photosynthetic reaction centers, a lightharvesting complex, halorhodopsin and bacteriorhodopsin). Cubic phases do not replace detergents, which are often needed to maintain the molecule in a soluble form prior to crystallization. Non-membrane proteins that have extensive hydrophobic regions may also benefit from detergent use, as in the case reported by Klaholz & Moras [Acta Cryst. D56 (2000), 933-935], where a single dodecyl-α-D-maltoside molecule was found to occupy a hydrophobic cleft between two molecules of the human retinoic acid receptor ligand-binding domain. The detergent molecule (Fig. 1a) is clearly visualized in the crystal structure (Fig. 1b) with the cavity apparently being specific for the α-isomer, and the maltose head group forming hydrogen bonds to protein side-chain residues. The detergent acts as a spline, helping to keep the two molecules in a fixed orientation and increasing the contact area (by 38%) to strengthen the interaction. A likely reason for the ability to crystallize macromolecules is their tendency to self-associate in defined structures. Additives or procedures that can initiate the formation of defined structures in solution may be key factors in obtaining macromolecular crystals.Macromolecular crystallographers are on a constant search for the Holy Grail of crystal growth. The use of screening kits and detergents is one approach. Manipulation of the physical environment as an alternative is seen as a more global approach, one that does not have to be tailored to a specific molecule. In recent years the use of microgravity and gels have been investigated. Neither is without problems, but steady progress is being made. Now another method is emerging, the use of high magnetic fields to suppress convective flow. Sato et al. [Acta Cryst. D56 (2000), 1079-1083] report measurable improvement in mosaicity and diffraction resolution in crystals of lysozyme grown in a 10 T magnetic field. The c axis of the crystals was found to be aligned with the magnetic field (Fig. 2) and the diffraction resolution was increased from 1.3 Å at 0 T to 1.13 Å at 10 T.

This short review highlights a variety of crystallization developments and case studies. Many more are described in greater detail in each issue of Acta Cryst. D. The vigorous crystal growth field is poised to achieve the high throughput required in today's structural genomics world.

Marc Pusey, NASA, Huntsville, AL, USA
Naomi Chayen, Imperial College School of Medicine, London, UK