Automated Protocols for Macromolecular Crystallization at the MRC Laboratory of Molecular Biology

When high quality crystals are obtained that diffract X-rays, the crystal structure may be solved at near atomic resolution. The conditions to crystallize proteins, DNAs, RNAs, and their complexes can however not be predicted. Employing a broad variety of conditions is a way to increase the yield of quality diffraction crystals. Two fully automated systems have been developed at the MRC Laboratory of Molecular Biology (Cambridge, England, MRC-LMB) that facilitate crystallization screening against 1,920 initial conditions by vapor diffusion in nanoliter droplets. Semi-automated protocols have also been developed to optimize conditions by changing the concentrations of reagents, the pH, or by introducing additives that potentially enhance properties of the resulting crystals. All the corresponding protocols will be described in detail and briefly discussed. Taken together, they enable convenient and highly efficient macromolecular crystallization in a multi-user facility, while giving the users control over key parameters of their experiments.


Introduction
X-ray crystallography is extensively applied to further advance our understanding of biological and disease mechanisms at the atomic level and to subsequently assist rational approaches to drug discovery 1 . For this, purified and concentrated (2-50 mg/mL) macromolecular samples of protein, DNA, RNA, other ligands, and their complexes are trialed for their propensity to form ordered three-dimensional lattices through crystallization 2,3,4 . When high quality crystals are obtained that diffract X-rays, the crystal structure may be solved at near atomic resolution 5,6 . Crucially, the conditions to crystallize a novel sample cannot be predicted and the yield of high quality crystals is usually very low. An underlying reason is that many samples of interest have challenging biochemical properties, which make them unstable on the corresponding timescale for crystallization (typically a few days). Finally, the process is compounded by the time required to produce samples and sample variants, and to optimize their purification and crystallization 7,8 .
A crystallization condition is a solution with a precipitant that reduces sample solubility, and conditions often also contain buffers and additives. Hundreds of such reagents are well suited to alter the parameters of the crystallization experiments as they have low propensity to interfere with sample integrity (such as protein or nucleic acid unfolding). While testing millions of combinations of crystallization reagents is not feasible, testing several to many screening kits -formulated with various strategies 9,10 -is possible with miniaturized trials and automated protocols. In this perspective, the most amenable technique is probably vapor diffusion with 100-200 nL droplets sitting on a small well above a reservoir containing the crystallization condition (25-250 µL), implemented in specialized crystallization plates 11,12 . The protein sample and condition are often combined in a 1:1 ratio for a total volume of 200 nL when setting up the droplets in the upper-wells. Robotic nanoliter protein crystallization can be implemented with alternative techniques and plates such as the under-oil batch 13 and the Lipidic Cubic Phase 14 (the latest one being applied specifically to trans-membrane proteins that are very poorly soluble in water).
The crystallization facility at the MRC-LMB was started in the early 2000s and an early summary of our automated protocols was presented in 2005 15 . A historical introduction to protein crystallization was presented and also an outline of the advantages of robotic nanoliter approach (then a novel approach to routine experimentation). Since macromolecular crystallization is essentially a stochastic process with very little or no useful prior information, employing a broad variety of (suitable) initial conditions increase the yield of quality diffraction crystals Figure 1A shows the system 1 based on a liquid handler operating with a liquid-system (deionised water). The liquid-system comprises a container, a pump, tubing, 8 syringes equipped with valves and 8 fixed tips. The liquid class settings were optimized to aspirate/dispense a broad variety of solutions and multi-dispense into 4 plates (multi-dispense requires a relatively large excess of aspirated volume). A 22 L water container and a smaller container (5 L) with 20 % v/v ethanol are stored underneath the system to feed the liquid-system. Each container is equipped with two coupling inserts. One insert (colored blue) feeds the system with liquid, the other one (red) is a flow-back to reduce excess pressure. After use, flushing with water and then 20% v/v ethanol solution prevents microbial growth. A cooling unit (also located underneath the system) is connected to a custom-built tube-cooling carrier. System 1 is 2050 mm wide, including the carousel, 760 mm deep and 88 mm high. An additional 550 mm is required in the front for the worktable that holds the control unit of the inkjet printer and the adhesive plate sealer. Additional space is also needed next to the system for the controlling PC. The program to produce 72 pre-filled plates (i.e. 18 rounds of 4 plates) takes 3 h and 50 min. Figure 1B is a close-up of the main deck that is equipped with 8 fixed tips ( Figure 1C) mounted on an automated pipetting arm, a second arm with a gripper, a tip wash station, 2 x 4-position carriers for SBS plates and the tube-cooling carrier. The main program processes 4 plates at a time that are taken out automatically from the carousel and placed on the deck where they are filled with crystallization conditions (80 µL in reservoirs, Figure 1D). Liquid levels are automatically detected as the tips are conductive. While the liquid handler dispenses conditions into a set of plates, another set of empty plates is taken out from the carousel, labeled and placed onto the main deck. A small custom-built holder and window in the rear panel of the liquid handler were required to position the printer head and its sensor at the back of the main deck. The 8 tips are flushed and washed with water from the main container after each dispensing step that consists of 4 aliquots in the corresponding columns of reservoirs. After 4 plates have been filled, they are automatically sealed and placed back in their original position in the carousel. The plate sealer is triggered by a specific driver (called by the controlling software). The sealer uses a roll of 3-inch wide adhesive tape which is applied to a plate with rollers under mechanical pressure. Upon completion of the program, the pre-filled plates are removed manually from the carousel stacks and stored in a 10 °C incubator located within the facility. The LMB plates are stocked in 10 °C incubators located within the facility where they are available to users anytime.  20 was integrated on the right-hand side of the system. Aspirate/dispense on the nanoliter dispenser also operates with positive displacement using disposable microsyringes (supplied in large spools). As a stand-alone robot, a removable strip-holder block is used to load protein sample(s). For the fully automated process, a 384-well PCR plate replaces the strip-holder. A similar adhesive plate sealer to system 1 was integrated on the left-hand side. The sealer and nanoliter dispenser stand on custom-built, raised worktables in order for the main gripper to reach the plate carriers of these two integrated robots (a window had to be cut in both side panels of the liquid handler for the gripper to gain access outside the main deck). The main program calls specific drivers to trigger nanoliter dispensing programs and the sealer in due time. The system 2 is 2,850 mm wide, 800 mm deep and 800 mm high. Additional space is needed next to the robot for the display of the controlling PC. Two waste bins are located underneath the system for the tips and microsyringes discarded during the process (the controlling PC is also stored underneath). An important characteristic of the system 2 is the overall layout, which retains easy access to the three robots so that they can be used individually for the optimization protocols described earlier, or other protocols described elsewhere such as crystallization of membrane proteins in lipidic mesophases 21 and random microseed matrix screening 22 . Figure 2B is a close-up of the deck that is equipped with a single robotic arm. The arm integrates 12 independent pipetting probes and the main gripper. The positions of the probes can be controlled individually in one direction in order to access different locations along the axis between the probes or even single tubes. The probes can pick up either disposable tips (1-12) or a pair of plate-moving adaptors. Two sets of 4 stacks containing 50 µL disposable tips are initially loaded. Liquid levels are automatically detected as the tips are conductive. The plate-moving adaptors are designed with 2 sharp pins on the inside that form an optional gripper when needed. On the main deck is located a 24-position microtube cooling carrier for the sample, although only 1-2 positions are used here. In addition, there is a carrier for the PCR plate and also 2 storage carriers for stackable, custom-built SBS lids ( Figure 2C). Finally, there are 4 sliding carriers, each with 5 locations for crystallization plates (4 x 5 = 20 plates).

System 2 and requirements for setting up droplets
Because the precipitant in the droplet is less concentrated than in the condition, the concentrations of all the components in the drop rise through water loss during the process of equilibration by vapor diffusion (very schematically represented by an arrow). (E) Light micrographs of 100 nL and (F) 200 nL droplets produced by the nanoliter dispenser with a test solution (20% v/v Polyethylene glycol 400, 0.001% w/v Safranin as red dye). The size and shape of the drops may vary according to chemicophysical properties of the condition. Please click here to view a larger version of this figure.
The main program starts with the optional gripper removing the SBS lid from the first crystallization plate. After the lid was transferred to the corresponding storage carrier, the crystallization plate is transported to the deck of the nanoliter dispenser. Then the pipetting arm transfers the amount of sample required for setting up droplets from the normally chilled microtube to the first column of the PCR plate. Subsequently, the main gripper moves the PCR plate onto the deck of the nanoliter dispenser that will prepare the droplets following the options selected by the user at the start (e.g. droplet size). For this, the protein sample is first dispensed into the upper-wells (using a single set of 8 microsyringes and multi-dispensing), then the crystallization conditions are dispensed onto the protein droplets (each row now requires 8 new microsyringes to avoid cross-contamination). Upon completion of the program to set up droplets, the main gripper transports the corresponding plate to the plate sealer, then places the PCR plate back in its original position (more protein will be dispensed in the next column of the PCR plate for the following plate). Finally, the sealed crystallization plate is moved back to its original position on the deck: Vapor diffusion experiments have already started in this plate ( Figure 2D). This cycle is repeated according to the number of plates. When needed, the optional gripper removes an empty tip rack allowing the pipetting arm to access more tips. The program takes 2 hr and 20 min and 440 µL of sample to prepare single droplets in 20 plates using 100 nL sample + 100 nL condition (Figure 2E and 2F).
When using the nanoliter dispenser as stand alone robot for two additional plates (see Protocol, step 1.

Formulation, preparation and handling of the 4-corner solutions
Both formulations of the 4 corner solutions ('A, B, C and D') and the corresponding optimization screen (Figure 3A) are automatically generated by an Excel spreadsheet. There are different spreadsheets for different numbers of conditions -essentially 24, 48, and 96 conditions -and also spreadsheets to prepare two different optimization screens in a single plate. One typically prepares a set of 4 x 10 mL corner solutions in test tubes from which 2-3 optimization screens can be prepared, depending on the number and volume of conditions required. The solutions are poured into troughs from which they are aspirated by a syringe-based liquid handler and later dispensed directly into plates ( Figure 3B). The two linear gradients of concentrations result from mixing A, B, C, and D at systematically varying ratios ( Figure 3C).   The 96-well plate can be used to prepare 96-or 48-condition optimization screen (when two 48-condition screens are prepared simultaneously, the robot must be equipped with 8 syringes and 8 troughs). Other programs enable the use of 48-well plates. The listed volumes in troughs include the required dead volume (0.5 mL). Figure 4 shows the steps to perform an additive screening starting either with 96 additive solutions already in the reservoirs of the crystallization plate (protocol 1) or in the wells of a low-profile cell culture plate used as a re-usable additive screen (protocol 2). Table 4 lists the programs available on the nanoliter dispenser according to the two types of protocols.    Because of the dilution of the condition with the additive during protocol 1, the condition needs to be prepared at a proportionally higher concentration than initially. The increase in final concentrations is most easily achieved by reducing the final addition of water to the calculated final volume. After this, one simply proceeds with the normal set up of droplets that mix the condition and sample (e.g., 100 nL protein + 100 nL condition already mixed with additives). Protocol 2 (e.g., 200 nL protein + 200 nL condition + 100 nL additive) facilitates the screening at different concentrations of additives by simply varying the volume of additive screen added. Protocol 2 implies more or less dilution of the droplets (which may alter crystallization).

Additive screening
The liquid handler from system 2 can be used to aspirate enough condition in 12 tips and dispense 8 aliquots into the reservoirs of a 96-well plate (see Protocol, step 2.2.2), although this step can of course be done manually with a multichannel pipette (Figure 4). Several plates can be filled at a time with the same condition when using the liquid handler (to test different additive screens later). When preparing 2 plates, fill the reagent container with at least 23 mL. When preparing 3 plates, fill the reagent container with at least 31 mL.

1-Preparation and use of initial screens stored in plates
Screening kits should be mixed before being dispensed into plates because light precipitation or phase separation occurs in some tubes during storage. When a screen is composed of two kits (2 x 48 tubes), the first tube of the second kit is placed in in location E1 of the cooling carrier. When a screen is composed of 4 kits (4 x 24 tubes), the first tube of the second kit is placed in in location C1, the first tube of the third kit is placed in location E1 and the first tube of the fourth kit is placed in location G1. While inserting the tubes in their cooling carrier, lids are placed on a tray following the standard 96-well landscape layout. Since well numbers are indicated on top of the lids by the manufacturers, this enables cross-checking if all tubes have been placed in the right order. This also helps to replace the correct lids on the tubes when filling a reduced number of plates.
We store the pre-filled plates at 10 °C, a compromise to avoid freezing and storage at 4 °C that may cause deterioration of conditions and issues with sealing. Plates are stored for up to several months with normally no noticeable condensation on the inner face of the seal. This is less true for LMB05, LMB06, LMB09 and LMB10 plates as these contain conditions with relatively high concentrations of volatile reagents ( Table 1). Small amount of condensation on the inner side of the seal reduces sealing efficiency and can cause cross-contamination between wells while unsealing the plates. To help with preventing condensation during initial cooling, plates can be first transferred from the carousel into an insulated picnic cooler which is stored in a 4 °C cold room overnight. The very slow cooling minimizes the development of temperature gradients within the sealed wells and hence reduces condensation overall 15 . In addition, once the plates are stored in the 10 °C incubator, an in-house custom SBS polystyrene lid is placed on the plate at the top of each stack (not shown).
The entire set of our pre-filled plates can be used as a large initial screen against a novel, water-soluble, protein sample. Alternatively, fewer plates may be selected to match specific requirements. For example, LMB15 and LMB19 are screens formulated specifically for membrane protein samples 26,27 , or LMB20 is a screen formulated with heavy-atoms to facilitate the experimental phasing of diffraction data 28 (see also: Formulation of the MORPHEUS protein crystallization screens).

Setting up crystallization droplets
When using the system 2, screening kits with significant amounts of volatile reagents should be processed first. This avoids condensation forming on the rubber of the SBS lids, which could affect lid handling and plate sealing. An SBS lid has a bit of clearance when on the top of a plate, which is why they need to be aligned initially (see Protocol, step 1.2.6). The protein dead volumes in the wells of the PCR plate are relatively generous (0.8 µL, see legend of Table 2). Note that equally generous dead volumes are employed when using the nanoliter dispenser individually with protein in 8-well strips ( Table 4). Smaller dead volumes may work, however some samples adhere to the tips, calibration of a robot may become slightly inaccurate, the room may be warmer than usual, etc. All lead to sample losses covered by the generous dead volumes in order to consolidate the approach.
Recent developments enabled further miniaturization of experiments and hence the volume of sample required for screening crystallization conditions can be significantly reduced by integrating the corresponding technology 29,30 . However, some aspects of further miniaturization need careful consideration, such as the evaporation of droplets 31 and the manipulation of microcrystals 32 .
Finally, centrifugation of the plate (2,000 rpm, 1 min) could be integrated as a routine final step when setting up crystallization droplets (in spherical upper-wells). A more consistent size and shape of droplets resulting from centrifugation may reduce reproducibility issues 33,34 . Surely, centered drops will ease the later assessment of experiments using a microscope as the required focal length will be similar across the entire plate.

Advantages of the 4-corner method
The most significant advantage of the 4-corner method is its simplicity, which minimizes errors and facilitates straightforward automated protocols. For example, the 4 corner solutions will always be placed on the deck of a liquid handler following the same layout. Also, all programs are based on fixed ratios between the solutions (Figure 3C). Manual preparation of the 4 corner solutions is preferred to automated handling of solutions at high concentrations which can be highly viscous. Relatively fast and accurate aspiration/dispensing is then possible on most types of liquid handlers with minimum requirements for optimization of liquid classes. Nevertheless, some corner solutions may still be too viscous for a robot operating with a liquid-system to operate efficiently. This is why we opted for a liquid handler operating with positive displacement (Figure 3B).
In addition to the 2 linear gradients of concentrations, a third component (i.e., a set of buffers/additives) can be tested at a constant concentration in a convenient way. For this, a relatively large volume of a core set of corner solutions at a suitably higher concentration, excluding the component to be varied, is prepared first. Then, stock solutions including this component is added to adjust the final concentrations. For example, 50 mL of a set of 4 corner solutions are prepared at 10 % higher concentrations than initially. This core set is then split into 5 smaller subsets of 4. Finally, 10% in volume of different buffer-pH solutions is added to each subset.

Formats and types of additive screens
The screens are normally stored at -20 °C (Figure 4) since they are not used regularly and contain volatile/unstable compounds. The use of a frozen additive screen stored in a deep well block (1 mL in wells) must be planned early because it will take 12-24 hr for all the additive solutions to thaw completely at room temperature. Also, a multitude of users share the same additive screen, potentially causing problems with cross-contamination. Finally, the height of deep well blocks makes them unsuitable for most nanoliter dispensers. As a convenient solution to circumvent these issues, the screen should be transferred from the deep well block to low-profile plates (Figure 4). Historically, additive screens which include a broad variety of single reagents (with single concentrations) have been very popular 35,36 . However, other types of additive screens have been developed that integrate mixes of additives 37 or a reduced number of single additives found at different concentrations 38 . Finally, a complementary approach is to investigate the effect of additives on the samples prior to crystallization 39,40 .

More considerations
Good practice: Most screens contain harmful or even toxic substances and hence adequate personal protection must be employed during the protocols. Equally, moving parts of the robots may lead to injuries, especially when trying to manually interfere while a program is running (although most of the robots have emergency stop button/system). Because of the technical complexities involved, regular checking of robots, screens and programs with previously characterized test samples are important for sustained high levels of reproducibility.
Throughput: As an indication, between 4,000 to 8,000 LMB plates are produced yearly with the system 1 (and subsequently employed by users for initial screening). It is not adapted to stock a large amount of pre-filled plates at 10 °C when the expected turnover is much lower, as after 4-5 months, some conditions will start to deteriorate and evaporate. Different approaches to automation protocols have been implemented for smallto medium-size laboratories 41 .
Storing and assessing experiments: After preparing the droplets, plates are stored on low-vibration shelves in a room at 4 or 18 °C with tightly controlled temperature (+/-0.5 °C maximum deviation). Experiments are assessed using cold light source microscopes. Various automated imaging systems are commercially available, however one should carefully consider all aspects: Will the speed required to scan a plate be sufficient for high throughput? Will objects other than crystals interfere with autofocus? Will the resulting quality of images be sufficient to spot very small crystals (especially around the edge of the droplets)? 42,43,44 Comparison of crystallization conditions: After careful investigations about the nature of the initially-obtained crystals, one can analyze trends and similarities across conditions using the LMB screen database or the C6 Web Tool 45 .

Disclosures
We hereby state a conflicting commercial interest since LifeArc commercializes the following items: the 96-and 48-well MRC plates, the MRC hanging drop seal, the MORPHEUS, Pi, ANGSTROM and LMB crystallization screens.