Several automation companies are currently mar- keting robots ... A host of other firms market systems capable of automating various ... systems includes intelligent error recovery; allowing ro- .... memory (RAM), a 3 1/2" floppy-disk drive, and from 100 ... When synthesis applications demand customized hard- ware, an ...
Perspective Molecular Diversity, 1 (1995) 270-274 ESCOM MOLDIV 027
Automating combinatorial chemistry: A primer on benchtop robotic systems Jill H. H a r d i n * a n d F r a n k R. S m i e t a n a Lilly Reseamh Laboratories, A Division of Eli Lilly & Co., Lilly Corporate Center, Indianapolis, IN 46285, US.A.
Received 6 April 1996 Accepted 21 May 1996 Keywords: Robotics; Automation; Combinatorial chemistry
Summary Benchtop robotic systems are inexpensive, flexible automation tools with potential applications in a wide array of disciplines such as combinatorial chemistry, high-throughput screening, and genomics. We explain the basic components of a benchtop system and explore factors to consider when purchasing or customizing a robot, such as automation benefits, vendor selection, and current system limitations. Issues involving system specification, software design, and hardware customization are then discussed. Additionally, system optimization, validation, and support are detailed. Given a properly designed and implemented system, the combinatorial laboratory can markedly increase compound synthesis and purification.
Accompanying the growing acceptance of combinatorial chemistry is the realization that manual processes need to be automated. Automation establishes consistency in the synthesis and screening of compounds, frees the scientist from the mundane aspects of the process, and channels scarce resources toward more productive arenas demanding human attention. This paper explores automation from the perspective of benchtop systems specifically designed for combinatorial chemistry. Benchtop systems are currently being used to automate a number of specific combinatorial chemisty applications. These include applying microliter-scale volumes of synthesized compounds to thin-layer-chromatography (TLC) plates; adding solvents, reagents, diluents and resin slurries to synthesis vessels; solid-phase extraction (SPE); transfer of products to vials or microtiter plates; and injection of final products into an autosampler attached to either a high-pressure liquid chromatograph (HPLC) or a mass spectrometer (MS). Some systems are capable of septum piercing, a useful capability for adding product *To whom correspondenceshould be addressed. 1381-1991/$ 6.00 + 1.00 9 1996 ESCOM SciencePublishers B.V.
to MS vials or removing product under an inert atmosphere. Several automation companies are currently marketing robots specifically built for solid-phase combinatorial chemistry. These firms include Advanced ChemTech [1], Bohdan Automation Inc. [2], Science Applications International Corp. (SAIC) [3], Tomtec [4], Tecan AG [5], Argonaut Technologies Inc. [6], and CombiChem Inc. [7]. A host of other firms market systems capable of automating various aspects of combinatorial chemistry, including Beckman Instruments Inc. [8], Hamilton Co. [9], RoSys Inc. [10], Gilson Inc. [11] and Packard Instrument Company, Inc. [12]. Custom-built solutions can also be purchased from Bohdan Automation Inc., Zymark Corp. [13], and Sagian Inc. [14]. Before attempting an automation project, the limitations of current benchtop systems must be clearly understood. Some applications are straightforward and can be automated using off-the-shelf systems requiring no further customization. Other applications, though capable of being automated, present a greater challenge since no
271 commercially available robotic system exists, necessitating custom hardware construction. Generally, automated systems do not perform tasks quicker than manual alternatives. However, this does not necessarily preclude the use of automated systems, as they may run well beyond the typical 40-h work week without an increase in hourly operating costs. There are reasons to automate a task even if it does not prove cost-effective, such as eliminating human exposure to toxic chemicals, human 'burnout' due to the performance of repetitive routine tasks, the necessity for very accurate and precise deliveries of liquids, or to standardize procedures across many laboratories within a given department or company. Therefore, automation may be able to compete with manual performance if the timing, expense, and accuracy issues provide adequate justification. In order to guage the applicability of automation, it is important to understand both current functionality as well as limitations. Functionality that is currently available for automated systems includes intelligent error recovery; allowing robots to assess an unexpected event, take corrective action and proceed rather than halting. Multi-tasking is a feature available with certain robotic systems software involving the simultaneous operation of multiple pieces of equipment within a given system. Multi-tasking may be invaluable in an automated system as it maximizes system throughput by minimizing the downtime of the various components comprising the robot. Commercially available databases may be used to capture relevent data generated in an automated system and provide convenient links to statistical data evaluation tools or allowing for the use of data mining tools, such as neural networks, to create 'intelligent' databases. There are some situations which commercially available robotic systems may not be able to address. The weight of containers that will be used in the automated process needs to be examined closely. Most robots are restricted to moving only a few kilograms at best. Heavier containers would demand manual intervention, eliminating the possibility of a totally automated system. If septurn puncturing is to be performed, robot-arm durability and ruggedness is paramount. Processes requiring subjective observations to determine a color change or product formation may be difficult or impossible to automate. Another factor to consider is the length of time a given automation system will be operational. This factor may prove to be substantial when comparing the cost of the automated system with the manual process. Budgetary constraints must be examined to determine how much of the process can reasonably be automated. A semi-automated solution involving minimal human intervention may meet the needs of a laboratory unable to finance a 'cradle-to-grave' solution. It is conceivable that several robots can be used in tandem to perform the desired automation. Complexity increases dramatically when
using multiple systems, and timing issues must be resolved to eliminate workflow bottlenecks. The length of time necessary to automate a given robotic system may prove to be substantial therefore, this factor must be weighed heavily when considering an automation project. The length of time required to automate may be increased with the increase in complexity of the desired task. Simple routine tasks, for example serial dilutions, may be programmed in a vendor-provided 'point and click' environment, while a more difficult task may need to be programmed in a more complex environment, using a language such as C++ or Microsoft Visual Basic. With the increase in complexity of the task one can expect the length of time required to program the robot to increase, as well as the complexity of the program. Other automation-system applications may require the use of customized hardware, requiring substantial investments of time and money, as it becomes necessary to work closely with a machine shop to design the hardware. After the automation specialist has delivered the desired system to the end users, requests for changes in the programming or hardware of the system typically begin to surface. Changes may be due to work load increases, evolvement of underlying processes or misunderstood expectations. The latter can be minimized by the automation specialist possessing a thorough understanding of the manual application to be automated and by providing users of the system with a system-requirements document. This is a document detailing everything a given system needs to do in order to be used, accepted, and valuable to the users. Once the users have 'signed off' on this document, the automation specialist may begin working on the system. Ideally, the automation specialist charged with implementing a benchtop system will have a clear understanding of the processes to automate. Failing that, it is imperative that the specialist has free access to the personnel engaged in chemical synthesis. Automation project failures are often the result of an incomplete understanding of the end user's expectations. If the automation specialist does not have a thorough understanding of the underlying process, she may choose to shadow laboratory personnel for several cycles of the manually executed synthesis to familiarize herself with the attendant details and potential difficulties. A number of commercially available off-the-shelf systems may currently be evaluated by a laboratory seeking to automate its applications. It is preferable to negotiate a risk-free evaluation period with each prospective vendor, typically 1 to 2 months in duration. This allows the specialist time to evaluate the chemistry, programming and durability of a system, and to perform an objective comparison between the various robots. If the applications demand a custom-built robot, the laboratory will obviously not have the luxury of performing a risk-free evaluation at the expense of the vendor.
272 Some benchtop systems are merely retooled versions of aqueous-based liquid-handling robots. Tubing, valves and any other components that are subjected to prolonged contact with organic solvents will therefore have to be replaced with components capable of resisting this exposure. The arm's construction is also critical to the longevity and success of the system. During the evaluation phase, periods of time should be blocked off to run diagnostics that check the reproducibility of the arm's positioning mechanisms. Some robots such as the Tecan RSP 5000 provide utility programs that automate this procedure. A typical benchtop robot consists of a work surface or deck, a robotic arm, a liquid-handling system and a controller, and a personal computer (PC). Roughly two feet deep by three feet long, the deck of an instrument may house numerous components, such as test-tube racks, microtiter plates, wash troughs, or autosamplers and injection ports. The robotic arm manipulates and transfers samples, reagents and diluents on the deck. The arm of the robot is capable of movement along the x-axis with range of motion bounded by the dimensions of the deck. The robot may have one, four or eight probes which are capable of movement in the yz plane and equipped with a variety of different tips used to manipulate liquids. Some vendors offer an optional second arm that, although identical in functionality to the primary arm, does not necessarily double throughput. Due to sample-scheduling demands and the synchronization of functions on the robot, the second arm may be idle for long periods of time. Instead of the two arms being configured identically, the second arm might operate as a plate handler capable of a wide range of motion. The RoSys Plato and the Tecan Roma systems exemplify this type of robot. The liquid-handling system interacts with the arm to aspirate and dispense liquids between designated locations on the deck. This system is flushed and filled with a system liquid, typically a solvent compatible with the reagents used during the synthesis. The system is composed of the aforementioned probes, a length of tubing connecting the probes to the syringes, a stepper-motor mechanism capable of driving the syringes, a switching valve that facilitates liquid transfer and a liquid-detection mechanism. The probes are usually individually plumbed, allowing each probe to handle different volumes, if necessary. Various tip configurations are commercially available. If the application specifies a sterile or pyrogen-free pipetting apparatus, disposable tips must be used which are available in polystyrene or polypropylene. These tips perform a single transfer and are then discarded by the robot. Fixed tips are another option and are composed of Teflon-coated stainless steel. These tips are washed between liquid transfers and generally have a lifetime of several months. Connecting each tip to its respective switching valve is a length of tubing with sufficient internal volume to con-
tain the maximum amount of liquid that could be drawn up by that syringe, thereby preventing samples and reagents from entering the syringe. Syringes vary in size from 0.10 ml to 25 ml. Syringes contain system liquid that is separated from samples and reagents by airgaps that are programmed into the liquid-handling operation. Syringes are driven by the stepper motors, which are capable of accurate movement at a very small step size, perhaps 8000 increments over the path of syringe travel. Switching valves toggle between inlet and outlet settings to aspirate and dispense liquids through the tip. When using organics, it is imperative to use syringes and valves that are Teflon-coated. Most robots use syringes and valves manufactured by either Cavro or Hamilton; some of both companies' components are resistant to a broad range of organics. Liquid-level sensing is a mechanism that ensures that the minimum amount of liquid required by an application is present prior to aspiration from a container. If sensors fail to detect this minimum, an error message is displayed or logged by the system informing the user of this condition. The majority of liquid-level-sensing systems are capacitance-based, relying on ionic activity to determine the presence of liquid. Problems may occur when pipetting organic solvents exhibiting weak ionic activity. A second approach, termed sonic liquid-level sensing, is used in the Beckman Biomek 2000. A sound-wave generator and a receiving device similar to a microphone embedded inside the Biomek's liquid-handling tools provide sensing capabilities that are not dependent on a liquid's ionic activity. The analyst may also choose to disable liquid-level sensing completely. If this approach is employed, probes are programmed to travel to the maximum coordinate on the z-axis of the respective container before aspirating liquids from the container. Disabling liquidlevel sensing forces the end user to maintain appropriate levels of liquid in all containers to ensure the proper execution of the application. The robot is controlled by an Intel 486 or Pentiumbased PC. Communications to the robot take place either via a custom interface card or the RS-232 port on the PC. Peripheral devices can also be controlled through additional communication ports on the PC. A generic PC configuration might include 8-16 megabytes of volatile memory (RAM), a 3 1/2" floppy-disk drive, and from 100 to 500 megabytes of permanent hard-drive storage. A current trend in the software industry is to package shrinkwrapped software on CD-ROMs, necessitating a CDROM drive. Additionally, the PC should be configured with at least one communications port. Two software environments are currently available for benchtop robotic systems. One approach caters to the scientist, providing a 'point-and-click environment' requiring no in-depth programming at the expense of some flexibility and extensibility. The Beckman Biomek 2000
273 provides a three-dimensional graphical depiction of the deck. The user builds an application program by selecting volumes, locations and containers from this depiction. The actual machine commands are then automatically created by the Biomek system software. Despite its extreme ease of use, these environments may not support file manipulation or peripheral device control. This may be problematic when querying external databases or devices or sending data to a laboratory-information-management system (LIMS). A second approach to creating robotics software provides a complete programming environment for experienced automation specialists. Designed for specialists accustomed to working with languages such as Microsoft Visual Basic, C++, or Borland Turbo Pascal, these environments provide total flexibility and extensibility. Robotics vendors generally define the software driving the mechanical operations of their robot in a Dynamic-Link Library (DLL). The programmer uses a language such as Microsoft Visual Basic to create data-entry screens, reports and the scheduling of the automated synthesis while using the DLL to perform robot-specific operations. Interfacing to an external database such as Oracle or Sybase is also handled through the use of a DLL. If the scientist wishes to send data directly to a spreadsheet or statistics package, Dynamic Data Exchange (DDE) may be used. Note that these capabilities are possible only when the application resides under a graphical operating system such as Microsoft Windows 3.1 or Microsoft Windows 95. Older applications written for the Disk Operating System (DOS) will not share this capability, although various vendors do provide user-friendly environments in their DOS applications. Regardless of which approach is used for software design, end-user acceptance is largely dependent on the application's ease of use and the software's ability to anticipate and correctly process improper or erroneous user input. Although it requires additional programming, the analyst may choose to create detailed audit trails that facilitate diagnosis of hardware problems, should they occur.
When contracting with an outside vendor for software development, an agreement regarding source code must be established. Source code is comprised of readable files that can be edited by a programmer when modifications to the application need to be made. These files are then compiled into machine-readable form or interpreted by the PC for use by the robot. If the vendor does not release the application's source code, the user will be dependent on the vendor for all subsequent updates and maintenance. In this case, purchasers should negotiate for the source code to be held in escrow by a third party in the event that the outside vendor becomes insolvent. The vendor may also choose to sell the source code or provide it as part of the robot's purchase price.
When synthesis applications demand customized hardware, an in-house machine shop may be indispensable for rapid hardware prototyping and testing. A second approach is to use an external machine shop, though it may be difficult to effectively communicate the exact needs of the desired prototype, and the subsequent turn-around time for prototype development may be unacceptably slow. A third approach is to have a vendor specializing in custom automation design and build a robot to meet the laboratory's needs. Bohdan Automation Inc. specializes in the construction of robots into which commercially available components are integrated whenever possible. Off-the-shelf systems are also available from this vendor. Bohdan writes Windows software capable of integration into external databases using text file transfers. Sagian Inc. also builds custom robotic systems interfaced with CRS [15] and Orca subsystems. The Orca robot is currently manufactured by Sagian Inc., as is the point-andclick SAMI software environment which allows rapid application development that intuitively models the workflow of the process being automated. Once the hardware and software have been implemented, the robot's liquid-handling parameters must be optimized. These parameters include the syringe speeds used to aspirate and dispense liquids, the size of the airgaps that surround the aspirated liquid and delay times that pause the probe following liquid dispensing. Depending on the fluids that will be used in the application, this optimization process ranges from straightforward to tedious. Parameters are very dependent on the volume and viscosity of the representative liquid; therefore, a universal parameter set, robust enough to handle the entire range of potential volumes, rarely exists. A statistical technique available in most PC-based statistics packages that can be used to expedite parameter optimization is E X P E R I M E N TAL DESIGN. Using a statistically generated matrix of potential parameter sets, the automation specialist generates gravimetric or volumetric measurements for those parameters, allowing the statistical routine to quickly determine a robust optimum for a given range of volumes and viscosities. Following optimization, the robot must be validated to ensure that volumes being dispensed are accurate and precise. A batch program can be written that aspirates and dispenses numerous aliquots across the range of volumes used in the application. The aliquots are then weighed and adjusted for density to determine whether the optimized parameter set delivers the correct volumes. If the liquid-detection unit is employed, it must be validated to determine whether detection takes place at volumes and ionic concentrations typical to the application. This is especially critical when using very small sample volumes. Though usually treated as an afterthought, system maintenance costs should be considered during the robot
274 selection process. With nonrobust systems, these costs may prove to be a significant percentage of the original purchase price. Systems that suffer a high failure rate tend to discourage further automation efforts and cast doubt on the credibility of results generated by the robot. Maintenance contracts can be negotiated with most vendors and may be necessary if the synthesis laboratory lacks trained instrumentation-service technicians. The system will require both users and instrumention technicians to support the hardware. It is necessary for users to be trained to perform routine maintenance on a given system. Though this may appear trivial, finding a person who is motivated to perform such tasks reliably may be difficult. Timetables for performing routine maintenance range from daily to yearly activities. Certain applications may be dependent on the actual time that the robot is in service. Time in service can be logged by the robot's software, eliminating the need for manual tracking. Examples of routine maintenance are the removal of dust and dirt from the instrument deck and arm, thorough lubrication of the arm, and inspection of syringes, valves, and tubing for wear and leaks. When done on a routine basis this maintenance will prove to be invaluable to the life of the automated system. Instrumentation technicians have a more in-depth understanding of system hardware than the typical end user. They are capable of thoroughly dismantling and troubleshooting an instrument down to the circuit board level. These technicians generally obtain their training from the system vendor, though some automation companies prefer to keep this level of expertise within their organizations, necessitating service contracts. In addition to diagnosing and repairing unexpected system malfunctions, instrumention technicians are also responsible for performing periodic preventative maintenance, such as removal and replacement of syringes, tubing, and valves and checking the system's pipetting accuracy and liquiddetection capabilities. A program that transfers replicate aliquots across a broad range of volumes to a container appropriate for gravimetric verification should also be run periodically to determine whether accuracy and preci-
sion is being maintained by the liquid-handling system. Systems that perform this calibration check using the optical density of a dispensed solution are commercially available and can be easily integrated into most benchtop systems. A properly designed benchtop robotics system can markedly increase the throughput of a synthesis laboratory. Carefully determining the scope of the automation project, clearly communicating the details of the synthesis and analytical procedures to the automation specialist, and becoming familiar with the host of commercially available benchtop systems should guarantee a successful implementation. Combinatorial chemistry applications lacking off-the-shelf solutions may also be automated by vendors specializing in customized systems.
References 1 Advanced ChemTech, 5609 Fern Valley Road, Louisville, KY 40228, U.S.A. 2 Bohdan Automation, Inc., 1500 McCormick Boulevard, Mundelein, IL 60060, U.S.A. 3 Science Applications International Corporation (SAIC), 10260 Campus Point Drive, MS C4, San Diego, CA 92121, U.S.A. 4 TomTec, 607 Harborview Road, Orange, CT 06477, U.S.A, 5 Tecan U.S. SLT Lab Instruments, RO. Box 13953, Research Triangle Park, NC 27709, U.S.A. 6 Argonaut Technologies, Inc., 887 Industrial Road, Suite G, San Carlos, CA 94070, U.S.A. 7 CombiChem, Inc., 9050 Camino, Santa Fe, San Diego, CA 92121, U.S.A. 8 Beckman Instruments, Inc., 846 Algonquin Road, Schaumburg, 1L 60173, U.S.A. 9 Hamilton Co., 4970 Energy Way, Reno, NV 89502, U.S.A. 10 Rosys, 1113 Forest Run Drive, Batavia, OH 45103, U.S.A. 11 Gilson Medical Electronics Inc., 3000 West Beltline Highway, Middleton, WI 53562, U.S.A. 12 Packard Instrument Company, 800 Research Parkway, Meriden, CT 06450, U.S.A. 13 Zymark Corporation, Zymark Center, Hopkinton, MA 01748, U.S.A. 14 Sagian Inc., 5601 West 74th Street, Indianapolis, IN 46278, U.S.A. 15 CRS Robotics Corp., 5344 John Lucas Drive, Burlington, ON, Canada L7L 6A6.
