Co-crystallizing Success

By Mark F. Russo, PhD, Stephen Michalczyk, Matthew Cahn and Herbert Klei, PhD

The design, construction, testing and deployment of a large, custom-integrated automated laboratory system is a significant undertaking. The decision to initiate such a project in-house rather than adopt a commercially available solution is not to be taken lightly. A shortage of the requisite skills and proficiency, limited budgets and shifting priorities all represent significant obstacles — any one of which, if left unaddressed, could severely limit success. Here we describe a project for automating protein crystallization, including our philosophy, partnerships, setbacks, accomplishments and lessons learned.

A Philosophy for Internal Automation Groups

The philosophy followed by the automation group at the Princeton, N.J.-based Pharmaceutical Research Institute of Bristol-Myers Squibb Co. (New York, NY) is one of balance. In-house automation groups have certain advantages over external technology providers, including relationships with key decision-makers, a thorough awareness of the organization’s goals and priorities, and a general context that can only be obtained as employees of a company. We focus on our strengths and partner with external technology providers for other needs.

In appropriate situations, we may take on project development and do the hands-on work ourselves. These situations occur most often when there exists no commercially available solution, or when delivery time must be minimized and the project can be completed more quickly in-house. As any seasoned technologist knows, it can be very hard to resist the temptation to “do it yourself,” but that same technologist will also tell you that if a good solution exists commercially, it is almost always the more cost-effective one.

A Philosophy for Internal Automation Groups

The philosophy followed by the automation group at the Princeton, N.J.-based Pharmaceutical Research Institute of Bristol-Myers Squibb Co. (New York, NY) is one of balance. In-house automation groups have certain advantages over external technology providers, including relationships with key decision-makers, a thorough awareness of the organization’s goals and priorities, and a general context that can only be obtained as employees of a company. We focus on our strengths and partner with external technology providers for other needs.

In appropriate situations, we may take on project development and do the hands-on work ourselves. These situations occur most often when there exists no commercially available solution, or when delivery time must be minimized and the project can be completed more quickly in-house. As any seasoned technologist knows, it can be very hard to resist the temptation to “do it yourself,” but that same technologist will also tell you that if a good solution exists commercially, it is almost always the more cost-effective one.

The Problem

The 3-D structure of proteins, or protein-ligand complexes, can provide substantial insight into drug discovery. They are prized possessions worth investing substantial resources to obtain. The Midwest Center for Structural Genomics, based at the U.S. Department of Energy’s Argonne National Laboratory (Argonne, IL), reports its success rate at various stages of the process that begins with the DNA sequence of a protein target and ends with a 3-D structure.¹ The data show that the two stages related to protein crystals — identifying the initial crystallization conditions and refining these conditions to obtain diffraction-quality crystals — involve the most risk. These two stages have a combined success rate that is very low, approximately 14 per cent. Coaxing a protein to form a diffraction-quality crystal can be time-consuming, labour-intensive, and expensive in terms of reagent consumption.

Most experiments performed to grow protein crystals operate by vapour diffusion. In a vapour-diffusion experiment, a small volume of two distinct solutions separated by an air gap are closely positioned in a sealed container. One solution comprises a mixture of crystallization reagents buffered at some pH; the other is formed by mixing an aliquot of the purified protein solution with an aliquot of the first solution. During the course of the experiment the volatile components in the second solution diffuse across the air gap to the first in an effort to reach equilibrium. The net result is a gradual concentrating of the protein. If the conditions are just right, the protein will crystallize.

There are several common configurations used to perform vapour-diffusion experiments. One is the hanging drop experiment (Fig. 1).² The extent to which a protein crystal forms, if at all, is determined by the crystallization reagents present in the mixture and their concentrations, as well as temperature, pH and other factors. The more conditions tested, the better the chances a useful crystal will be obtained. Automation is key to maximizing the number of hanging-drop experiments that can be performed.

We use several of the successful commercially available automated workstations that prepare screen mixtures and set up hanging-drop experiment trays. Once the trays are ready, the burden shifts to the crystallographer who must use a microscope to inspect each drop of every tray on a frequent basis to identify crystal formation. A hanging-drop experiment can last from one day to several weeks, and the arduous task of repeatedly inspecting the many thousands of drops is tedious at best. This step can quickly become a bottleneck in the process. For this reason, we sought to automate the imaging, inspection and scoring of hanging-drop experiments.

To Buy or Build

Initially, a small internal group was formed that included members from the Crystallography, Informatics and Automation departments. The goal was to seek out and evaluate commercial systems to solve the crystal imaging, inspection and scoring problem.

We outlined several criteria with which to evaluate commercial systems. One item of particular importance was vibration, which can lead to overnucleation and crystals too small for X-ray analysis. Any automation equipment selected, which typically has multiple sources of vibration such as compressors and actuators, should expose experiment trays to vibrations equal to or less than what would normally be experienced in our “manual” crystallization rooms.

Another critical feature related to the quality of images captured, which must be of a very high resolution with minimal shadows. Shadows interfere with analysis, and occur because a hanging droplet can act as a lens, diffracting light and generating large dark circles around the drop's perimeter.

A third critical feature was the ability to modify and expand the system. We anticipated the introduction of new instruments and tray types, and we required the ability to adapt the system to accommodate new capabilities.

After a thorough search, no commercially available solution was identified that met all our needs. We decided to take on responsibility for integrating the system ourselves by combining external partnerships with in-house development of certain key components.

Running the Project
The system was designed in a modular fashion with well-defined interfaces. Responsibility for the modules was divided among the team members, and a timeline for completing development was established.

The system was to be built in a customized environmental room that included a thick concrete floor and specialized vibration isolating fixtures for all equipment. The size of the room permitted storage of a large number of trays with space for future expansion. Within the room a standard articulated arm robot on a two-metre rail would be used to move trays. Tray storage positions on custom fabricated shelving would hold both standard SBS footprint microplates3 or a larger 24-well Linbro® plate (Hampton Research Corp., Aliso Viejo, CA) commonly used in crystallization experiments.

The component most critical to the system’s success was the imager. Our experiences led us to the decision to build an instrument. We sought features such as multiple tray nests, automatic drop finding and zoom, actuated polarizing filters, carefully constructed lighting, an ability to capture multiple regularly spaced image “slices” through the depth of a single drop, and the remote control of the imager’s operation. The heart of the instrument design is a Thales Optem Inc. (Fairport, NY) automated microscope with motorized zoom and focus.

A second critical component was the image classification software, which would analyse captured images and calculate scores that measured the likelihood that an image contained a crystal. Not satisfied with commercial offerings, we decided to take on this challenge as well.

Expanding the Team

Project development tasks far exceeded the bandwidth of internal project team resources. To complement our own areas of expertise, we sought experts in optics and image processing. We teamed up with experts from Edmunds Industrial Optics (Barrington, NJ) and National Instruments Corp. (Austin, TX). To help us with the automated storage and retrieval system, we teamed up with Thermo Electron Corp.’s LAI Division (Burlington, ON).

Working out the kinks of a new instrument requires several design iterations and testing by many more people than we had available. Along with National Instruments and Edmunds Industrial Optics, we decided to team up with Coleman Technologies Inc. (Glen Mills, PA) to manage details of the imager design, as well as to develop control software and perform the instrument build. Coleman Technologies also developed the image classification software. Both products are currently available from Coleman Technologies.

According to plan, all hardware was built, delivered and installed, and all software was written, debugged and made ready for testing. Figure 2 depicts a photograph of the completed integrated system.

Testing — A Problem Emerges

Once completed, a classic approach to testing was followed with the isolated testing of all system components, followed by the testing of subsystems, and then the entire integrated system. The process proceeded with little out of the ordinary. Eventually, the complete system was being tested with images captured, classified and stored.

The final test and proof that the system was performing to specifications would be a comparison of the images generated by the imager over several thousand runs. Unfortunately, this turned out to be a problem. Classification scores from repeated imaging of the identical drop were inconsistent. A careful analysis revealed unexpected variations in focus. Further analysis revealed that the problem was with the imager, which was not capturing image slices through the depth of the drop in a consistent manner. It was only after extensive debugging that we determined the open-loop control of the lens motor offered by the vendor was not adequate for our application. By working with our external engineering partners, the lens vendor and the lens motor manufacturer, we were able to retrofit the focus and zoom lenses with motors that would operate in a closed-loop fashion, providing a much tighter control over lens operation. A new test of the integrated system completed with flying colours (Fig. 3).

Conclusions and Lessons Learned

Only after a careful definition of critical features and a thorough search for commercial solutions did we decide to undertake the in-house design and integration of a custom automated crystallization robotic system that stores, images and classifies hanging-drop crystallization trays. Given the unproven nature of the underlying approach, a modular design was very useful because it permitted, and continues to permit, frequent adjustments to the system to optimize operation. Creative arrangements with external partners can be used to minimize expense, share risk, and achieve the best result in the long run. Finally, a careful balance between internal strengths and those of external experts ensures the best possible outcome.

References
(1) “Retention Statistics.” Midwest Center for Structural Genomics. Last accessed May 1, 2005. http://olenka.med.virginia.edu/mcsg/phtml/statistics_viewer.phtml?plot=retention”>http://olenka.med.virginia.edu/mcsg/phtml/statistics_viewer.phtml?plot=retention.

(2) “Crystal Growth Techniques.” Hampton Research Corp. Last accessed May 16, 2005.
www.hamptonresearch.com/support/pdf101/CG101CGT.pdf”>www.hamptonresearch.com/support/pdf101/CG101CGT.pdf.

(3) “ANSI/SBS 1-2004: Microplates - Footprint Dimensions.” Society for Biomolecular Screening. Last accessed April 29, 2005. www.sbsonline.org/msdc/pdf/ANSISBS1-2004.pdf”> www.sbsonline.org/msdc/pdf/ANSISBS1-2004.pdf.

Mark F. Russo, PhD is a director of the Association for Laboratory Automation.