The MORPHEUS II protein crystallization screen

MORPHEUS II is a 96-condition initial crystallization screen formulated de novo. The screen incorporates reagents selected from the Protein Data Bank to yield crystals that are not observed in traditional conditions. In addition, the formulation facilitates the optimization and cryoprotection of crystals.


Introduction
The technique of single-crystal X-ray diffraction enables the routine and precise structure determination of biological macromolecules at high resolution (Blow, 2002;Rupp, 2010). It has been applied extensively to proteins, DNA and RNA, with the PDB recently celebrating the amazing milestone of 100 000 deposited structures. This is essentially the result of a multitude of innovations and technological developments spanning the last few decades (Abola et al., 2000;Fersht, 2008). Nonetheless, behind this success hides the struggle to produce purified samples, to obtain crystals with suitable diffraction quality and to subsequently reproduce/optimize them (Bergfors, 2009;Khurshid et al., 2014).
The poor yield of suitable crystals can be explained by the concept called the 'curse of dimensionality': there are so many dimensions associated with the parameter space to be explored that it is problematic or impossible to perform an analysis that has any statistical significance. The underlying reason is that the combinations of reagents employed in crystallization trials alter the combinations of variables associated with the main parameters of crystallization (McPherson et al., 1995); for example, the variables that are related to the nature of the protein and the experiment, the type of proteinprotein interactions and so on. Hence, initial crystallization screening is a stochastic process (Carugo & Argos, 1997;Lomakin et al., 1999) that usually requires various approaches and conditions. A crystallization condition traditionally contains a precipitant, a buffer and an additive. There are now hundreds of well known crystallization reagents and the possible combinations used to formulate conditions in a systematic manner has grown to a very large number that cannot be captured in any practical way because the amount of sample and the screening ISSN 2053-230X Table 1 Formulation of MORPHEUS II.
technology are limiting (Carter & Carter, 1979;Gorrec, 2014). As a consequence, many laboratories have chosen an approach with a minimum number of conditions. A widespread minimal approach is to employ a set of conditions selected empirically to form a 'sparse-matrix' screen (Jancarik & Kim, 1991). In the last two decades, advances in automation and the increase in the number of crystal structures solved and deposited have stimulated the optimization of sparse-matrix screens, mostly in the form of sets of 96 conditions, as this is an automation-friendly format (Kimber et al., 2003;Rupp & Wang, 2004;Newman et al., 2005;Stock et al., 2005;Newstead et al., 2008;Fazio et al., 2014). Nevertheless, we have argued that the minimal approach may mean undersampling of conditions and therefore essential hits could be missed (Gorrec, 2013). Subsequently, novel formulations should still be investigated when possible. Besides, screens should evolve in parallel with the increasing complexity of the samples and the technical difficulties encountered during the process of structure determination. Notably, the demands of cryo-crystallography (Petsko, 1975; x3.1.1) as well as current and future solutions to the phase problem (Taylor, 2003;x3.1.2) should be taken into account.
Previously, we presented an innovative approach in a screen formulation called MORPHEUS (Gorrec, 2009). In order to reduce bias towards a subset of samples, the conditions were formulated de novo by integrating a larger number of reagents than traditionally employed. For this, novel mixes of reagents were investigated. Reagents that aided protein stabilization, crystallization and crystal cooling were included in the final formulation. The multiplexing of reagents has been performed previously for cryoprotectants (Garman & Mitchell, 1996), precipitants (Majeed et al., 2003), buffer systems (Newman, 2004) and additives (McPherson & Cudney, 2006) (x3.1.3). The original MORPHEUS combined all of these innovations. It is worth highlighting two other particular design principles of MORPHEUS. Firstly, the inclusion of PDB-derived small molecules (as potential ligands) that were gathered into mixes of additives, sorted according to their chemical nature to avoid incompatibilities. Secondly, mixes of precipitants, additives and buffers were combined within a 96-condition threedimensional grid screen using fixed ratios to facilitate easier screen preparation and follow-up optimizations.
Here, we present MORPHEUS II, a 96-condition protein crystallization screen formulated in continuity with previous work. MORPHEUS II follows the design principles of MORPHEUS; however, we introduced less common additives, for example metals that are amendable to the collection of anomalous data sets directly from screening conditions. We also included nondetergent sulfobetaines (NDSBs), polyamines, amino acids and monosaccharides, which are known to enhance the solubility and stability of many proteins. To complete the formulation of MORPHEUS II, four unusual 'glycerol-like' polyols have been added as cryoprotectants to aid flash-cooling. Finally, innovative buffer systems were included as part of the formulations. The suitability of the resulting conditions is shown by the crystallization of eight different protein samples and their efficiency is compared with commercially available conditions (x3.2).   2. Materials and methods

Screen formulation
The mixes of ligands, precipitants and buffers were combined using a fixed ratio of volumes for the stock solutions as employed in the original MORPHEUS screen: 0.5 stock precipitants + 0.1 stock additives + 0.1 stock buffer system + 0.3 water. Methods used to select the PDB-derived ligands, to design the screen and to prepare the stock solutions were also as described previously (Gorrec, 2009). Further details can be found in the Supporting Information concerning the four precipitant mixes (Supplementary Table S1) and the three buffer systems (Supplementary Table S2).
Although the additive-to-protein ratio preferably needs to be maximized (Danley, 2006), the concentrations of the mixes Divalent cations II, Akalis, Oxometalates and Lanthanides had to be lowered compared with other, more soluble and less reactive additives, such as monosaccharides and carboxylic acids. Relatively low concentrations (around 1 mM) are suitable for these particular additives according to others (Petsko, 1985;Trakhanov & Quiocho, 1995).
By empirical experimentation, stable and suitable combinations of reagents were found. Unfortunately, some tradi-tional heavy atoms had to be excluded, such as those of the platinum group (chloride salts of platinum, osmium, iridium, ruthenium, rhodium and palladium) since they could not be solubilized and/or were unstable in solution. Of course, they can still be tested later on crystals already formed when necessary. The buffer system was removed from conditions B5-B8 to avoid precipitation (probably owing to formation of a chelate between a divalent cation and one of the corresponding buffers).
Also following an empirical approach, it was found that four small polyols that are not currently found in any commercially available screens vitrified samples during flash-cooling as efficiently as glycerol (i.e. typically 20-25% required to cryoprotect conditions): 1,2,4-butanetriol, 1,2,6-hexanetriol, 1,5-pentanediol and 1,1,1-tris(hydroxymethyl)propane. They were thus integrated into the precipitant mixes (Table 1 and  Supplementary Table S1). X-ray diffraction tests with the mixes of polyols and polyethylene glycols (PEGs) flash-cooled in cryoloops were then used to adjust the concentration of cryoprotectants and ensured that the resulting diffraction patterns were free of background from ice. NDSBs are another group of reagents often used in sample preparation and crystallization additive screens. They have a relatively low  Table 2 Selection of the 35 PDB-derived ligands selected to formulate MORPHEUS II.
Number of structures (i.e. occurrences in the PDB) showing a main chemical as an ordered ligand (as of December 2014). Note that a few reagents that were not found in the PDB were integrated to complete the formulations (the four polyols used as cryoprotectants and five of the six buffers used to formulate the buffer systems). frequency of occurrence in crystal structures, which suggests that their role may be less specific and therefore they have been integrated to the precipitant mixes.
The original MORPHEUS screen was purchased from Molecular Dimensions Ltd ('MORPHEUS I', Lot No. 021-1-46) and the sparse-matrix screen 'The JCSG+ Suite' was purchased from Qiagen (Lot No. 54806713). Triplicate droplets of 300 nl final volume with a 2:1 protein-to-condition ratio were formed in MRC plates at 20 C using a Mosquito robot (TTP Labtech). Single droplets of 200 nl final volume (1:1 ratio) were prepared similarly for the 'Bar' sample only. The plates were swiftly sealed and centrifuged (1000 rev min À1 , 1 min) and then kept at 18 C. Droplets were visualized with a stereo microscope after one week. Only obvious hits were taken into account (i.e. drops with crystals larger than 5 mm and with sharp edges). Details of the corresponding crystallization results can be found in the Supporting Information (Supplementary Table S3).

Formulation of MORPHEUS II
The formulations of the 96 MORPHEUS II crystallization conditions are listed in Table 1. The 35 PDB-derived ligands selected to formulate MORPHEUS II can be found in Table 2.
The recipes for preparing the eight additive mixes are listed in Table 3.
3.1.1. Integration of cryoprotecting agents. To obtain vitrification after flash-cooling, crystals are usually soaked briefly in a solution containing a cryoprotectant (most commonly glycerol). Many crystals are lost using this approach owing to the extensive handling and/or resulting variations in the composition of the mother liquor. A logical remedy for these issues is the use of crystallization conditions that are already cryoprotected. From this perspective, the amount of glycerol needed to successfully vitrify the conditions of Jancarik & Kim (1991) was determined by Garman & Mitchell (1996). Later, another study expanded these data with PEG 400, ethylene glycol and 1,2-propanediol (McFerrin & Snell, 2002). Adding the cryoprotectant directly to the formulations seems straightforward; however, the impact of introducing an additional reagent at high concentration on the yield of quality crystals was not investigated.
Cryoprotection should not bias formulations towards only a few cryoprotectants as main reagents since this would contribute to a further undersampling of initial screen conditions. We thought that more cryoprotectants should be tested as part of the development of novel conditions. Therefore, we integrated other polyols into the MORPHEUS II screen. Using polyols as cryoprotectants is beneficial since they are typically easy to handle and they display other interesting properties. For example, polyols are somewhat hygroscopic (Cohen et al., 1993) and hence they also act as precipitants, altering both the hydration of proteins and the kinetics of vapour-diffusion experiments (Forsythe et al., 2002;Collins, 2004). Finally, it should be pointed out that different cryoprotecting solutions cause different degrees of contraction upon flash-cooling and also affect cooling rates (Berejnov et al., 2006;Alcorn & Juers, 2010). These parameters will affect the differential contraction between the macromolecular crystal and the mother liquor surrounding it, and hence the quality of the diffraction data obtained.
3.1.2. Importance of additives. Additives may alter the parameters of crystallization experiments in a myriad of ways. If used correctly, they can increase the chances of obtaining useful crystals. One approach is to test how reagents alter the stability and solubility of the sample prior to crystallization  Table 3 Formulation of the eight additive mixes of MORPHEUS II.
Note that the mix called Alkalis includes two of the alkali-earth metals, strontium and barium, and two of the alkali metals, cesium and rubidium. Also, the mix called Lanthanides includes one of the rare-earth elements chemically very similar to the lanthanides (yttrium).

Row
Mix  (Ericsson et al., 2006;Izaac et al., 2006). Also, binding to the protein may be investigated (Boggon & Shapiro, 2000). Nevertheless, other essential parameters are specific to crystallization. For example, it has been demonstrated that crystal growth can be altered by additives such as divalent metal cations (Trakhanov & Quiocho, 1995). Therefore, it is probably advisable to integrate as many additives as possible into our initial screen. The positive impact of additives on the yield of crystals can be explained through the formation of new crystal contacts (Carugo & Djinović-Carugo, 2014). More complex molecules, for example polycarboxylic acids, sugars and polyamines, can bind to pockets in macromolecules and stabilize them or help them to adopt a particular conformation (Arakawa & Timasheff, 1982;Sauter et al., 1999;Maclean et al., 2002).
Polyamines were originally used with polynucleotides as they form favourable electrostatic interactions with DNA and RNA, leading to stable complexes (Bolton & Kearns, 1978;Drew & Dickerson, 1981). However, polyamines are now also regularly observed in complexes with proteins (e.g. spermine, PDB residue ID SPM, 105 occurrences in the PDB, Table 2). Zwitterionic organic chemicals, such as NDSBs (Table 2 and  Supplementary Table S1) and also HEPES-like buffers (e.g. MES) can be used as additives/buffers for solubilization and may prevent aggregation or polymerization (Vuillard et al., 1996). This may be explained through the abilities of molecules to shield specific apolar surface patches (Pusey et al., 2007).
3.1.3. Heavy atoms. In order to solve a structure by SAD, some early work suggested that derivatization should be performed by soaking the crystals since crystal growth may be altered by heavy-atom binding and hence prevent lattice contacts or eventually produce a different crystal form (Petsko, 1985). However, the increasingly challenging nature of proteins studied with crystallography means that different crystal forms occur less frequently. In addition, it is well known that when heavy atoms are tested for derivatization by soaking crystals (Garman & Murray, 2003) the diffraction usually worsens or may even be lost (as for cryoprotectants; x3.1.1).
Integrating heavy atoms into the initial screen is therefore very desirable. This was demonstrated when crystals of the nitrogen regulation-related protein NreA grew in two similar MORPHEUS conditions but with two different mixes of additives. Subsequently, closely related crystal forms were obtained that contained either iodide (NreA-I; PDB entry 4iuh) or nitrate (NreA-NO 3 ; PDB entry 4iuk) and enabled structure determination ab initio with experimental phasing (Niemann et al., 2014). Although we cannot yet share similar results from MORPHEUS II, it is worth mentioning that we did not observe issues with diffuse scattering caused by conditions containing heavy atoms (Luft et al., 2014), probably because of the very small proportion of the corresponding heavy atoms in the samples.
Finally, before thinking about derivatization, our goal was to increase our yield of initial (and novel) crystals. In this context, it is worth mentioning other work suggesting that multivalent metal ions such as yttrium can modulate proteinprotein interactions and even mediate crystal contacts that help to form the crystal lattice: they specifically bind to acidic surface patches as well as bridging acidic side chains from neighbouring subunits (Zhang et al., 2011).
3.1.4. Integration of mixes of additives. Use of mixes of additives during initial crystallization screening has been tested before with success (McPherson & Cudney, 2006). This strategy carries the risk that one component of a mix might have a deleterious effect and thereby mask the positive contribution of another. Nevertheless, if one of the additives from the mix participates in specific effects or interactions, the less specific addtive should be less pronounced. Also, by selecting components that have been regularly observed as ordered parts of crystal structures, the chances of incorporating molecules that play a positive role should be increased (Gorrec, 2009).
Finally, more than one type of additive may be required for crystal growth, as many structures contain multiple additives. This was demonstrated when the original MORPHEUS screen was used to crystallize the human endoplasmic reticulum aminopeptidase HERAP2 (PDB entry 3se6), where four additives and a buffer component are part of the crystal structure (Birtley et al., 2012).

Crystallization experiments
The overall yield of crystals was 11.3% with JCSG+, 17.2% with MORPHEUS and 16.5% with MORPHEUS II (Supplementary Table S3). Some may argue that our panel of test proteins may be more likely to crystallize under conditions containing high-molecular-weight PEG as the main precipitant (MORPHEUS screens; Page & Stevens, 2004). Different proteins may have specifically required relatively high salt concentrations and hence preferably crystallized in a sparse matrix (JCSG+). Others could also argue the most efficient pH range could have been anticipated (Kantardjieff & Rupp, 2004). Ultimately, the results will strongly depend on the subset of proteins tested.
The underlying problem is the numerous biases participating in the curse of dimensionality, notably those associated with the preparation of the screens/buffers (Wooh et al., 2003), the type/number of protein samples selected (McPherson & Cudney, 2006) etc. In the end, no matter how sophisticated the statistical analysis and data mining of crystallization space, any of the approaches will only provide a basis for increasing the probability of crystallization success, but will never guarantee success for any particular protein (Rupp, 2003).
As a consequence, no strict conclusions should be drawn when comparing efficiency between screens. Surely, we have demonstrated the suitability of nontraditional and de novo formulated MORPHEUS II conditions for protein crystallization, at least for samples with a propensity to crystallize in conditions with high-molecular-weight PEG precipitant, while two other screens widely used by the protein crystallography community were used as controls. Nevertheless, we hope that MORPHEUS II will efficiently extend the range of available iccbm15 conditions and hence enable the crystallization of recalcitrant samples. A cryoprotected screen certainly reduces reproducibility issues observed during the cryoprotection of crystals.

Conclusions
We have demonstrated that unusual and under-represented reagents can be combined to formulate suitable and useful conditions for protein crystal growth. The resulting screen is based on principles that have proven to be successful previously. The strategy of formulation reduces bias towards a subset of conditions or samples and integrates new mixes of reagents. MORPHEUS II has already increased our overall effectiveness with de novo structure determination enabled by including more heavy atoms in our initial screen.