Robotics for the Small Scale

Five questions to see if you're ready to automate.

Jan 1, 2007
Jeffrey M. Perkel
<figcaption> Credit: COURTESY OF JIM METHERALL</figcaption>
Credit: COURTESY OF JIM METHERALL

Jim Metherall, associate professor of genetics at the University of Utah, got his first robot in the mid-1990s while buried by a project on cholesterol homeostasis. "We were trying to clone a gene by complementing a mutation in mammalian cells, and we couldn't do it with an entire cDNA library, because the signal-to-noise was too low," Metherall explains. So he broke the library into small pools, isolated DNA, and then tested each pool for activity.

A few thousand minipreps later, Metherall recalls, "A very qualified, capable, valuable technician approached me and said she was sick of it. It was either automate or find new personnel." Metherall applied for and won an onsite equipment competition grant, some $50,000 to $60,000, as he recalls, which went into a BioMek 2000 automated liquid handler (about $45,000 at the time) and accessory equipment such as a plate washer and pump.

Today, he still uses the handler, though it's part of a much larger system, including microplate hotels, PCR machines, a microarray spotter, a scintillation counter, and a spectrophotometer, all linked by two robotic arms. The entire assembly occupies 100 square meters of lab space. And with only four members (not counting collaborators), Metherall's group is smaller than what you might expect for a robotics-hub at the University of Utah.

Ten years in the making, Metherall's robot is well beyond the scope of what most researchers would need starting out. But if you've been wondering if automation is right for you, consider these five questions:

1. Is your workload big enough?

Metherall's technician ran thousands of preps by hand before the lab acquired a robot, which eventually would run another 5,000 preps or so before that particular project ended.

According to Douglas Gurevitch, incoming executive editor of the Journal of the Association for Lab Automation and a senior development engineer in the department of bioengineering at the University of California, San Diego, the robotics facility at UCSD runs perhaps six ELISA plates at a time. A high-throughput screening system in industry might run 1,000 a day, Gurevitch says, "So six is not a lot - but it is in academia." All that pipetting can get old, fast, not to mention the incubations: ELISAs typically take six hours to run.

Yet even a dozen plates will occupy the machine for only a little while. If the machine is going to sit idle 20 hours per day, you might want to send your jobs to a core facility, or alternatively, to an external service provider. Either option is probably more expensive on an hourly basis than running your own robot - the Recombinant DNA/Protein Resource Facility at Oklahoma State University, charges $5 per hour to use its Beckman Coulter Biomek 2000 workstation; UCLA's Sequencing and Genotyping Core Facility charges $60 - but of course, you save the capital investment. OSU's system cost roughly $40,000 five years ago. Both core facilities and service providers will provide better quality data than manual work, by reducing the potential for human error, says Gurevitch, who also teaches an annual course on the economic justification of lab automation (D. Gurevitch, J Assoc Lab Automation, 9:33-43, 2004).

2. Is reproducibility paramount?

Shawn Levy, assistant professor of biomedical informatics and molecular physiology and biophysics, and director of the Vanderbilt Microarray Shared Resource, recently purchased for his facility a PerkinElmer Janus liquid handling robot with three arms: a 96-well pipettor, an eight-channel pipettor, and a gripper arm to move things around the deck. His cost: just under $200,000.

For Levy, whose facility runs several hundred samples per day, the decision to automate was easy. In addition to the high number of samples run, Levy sometimes noted differences in the quality of real-time PCR assays depending on which technician set up the experiment. Levy saw SNP call rates in the mid-90% range. That's pretty good, but with 500,000 SNPs per array, each percentage point lost represents 5,000 SNPs. Now, says Levy, the call rates have jumped 2.5% to 3%, meaning they get 15,000 more called SNPs per array. "Whichever tech does the assay, as long as they make the master mix correctly, the reaction is always the same," he says.

3. Is the cost of mistakes prohibitive?

The difference in error rate between manual and robotic processes can be substantial, says Gurevitch. He estimates manual work has an error rate of about 30%, compared to 3% to 10% for automation. In industry, that kind of difference makes automation a no-brainer.

Imagine you have a process that costs $1,600 per day and has a 10% error rate when performed manually, he says. That's a loss of $160 per day, or $24,000 per year, if you run the experiment three days per week, 50 weeks per year. Now suppose you could lower your error rate to 3%; that would reduce your per-day loss to $48, saving $16,800 per year. "That's where the robots pay off," Gurevitch says. But, he cautions, workload is still a limiting factor.

On the other hand, says Metherall, some processes are so difficult and expensive, automation makes sense even without industrial-scale throughput. Consider the case of quantitative real-time PCR; the assays are sometimes run in 384-well plates. "Let's say you do 10 plates per week, and each cost $500 in reagents. If a couple of plates have problems due to manual manipulation, that's $1,000 per week that's lost. If your assay is set up that way, the cost arguments [for robotics] are pretty reasonable," he says.

4. Would scaling up offer additional benefits?

Rebecca Frederick, a graduate student at the University of Utah, was running synthetic genetic-array screens, looking for yeast mutants that exacerbate the effect of a particular mutation. It was a massive undertaking: 20,000 or so colonies per step, nine steps per screen, in duplicate - some 360,000 colonies in all.

"We thought about doing this by hand [in a 384-well plate format], and I had the equipment to do it by hand, but it became obvious we could do it robotically," Frederick says. That decision offered Frederick benefits beyond stemming the risk of carpal tunnel syndrome. It's difficult to manually pipette to 384-colony density, and impossible to do so at 1,536 colonies. Automating at 1,536-spot density enabled Frederick to place both controls and duplicates on a single plate, a luxury the 384-well format did not afford. And, going up in size lowered the lab's plastics expenditure considerably.

Frederick turned to Metherall. He helped her get the robot up and running as a colony picker. Because the robotics facility was already extant, her only expenditure was $100 for jeweler's cleaner, to clean the pins.

5. Is your protocol automation-friendly?

Gurevitch recalls a time when he worked at a pharmaceutical company that needed to extract genomic DNA from milliliters of blood at a time. Plenty of automated DNA extraction systems were available in the 100-300 microliter range, he says, but he found nothing in the milliliter range. "We figured we could automate it," he recalls. "It would take two years of work, and cost $300,000." Ultimately, the company decided against automation - it simply wasn't cost-effective.

Even if you have the workload to justify robotics, some protocols simply don't lend themselves to automation, says John Fink, product manager for automation technologies at Thermo Fisher Scientific. "Automation requires standardization," he says. Suppose you need to automate a process that involves a freezing step, says Fink. "If the lab is using a normal carbon dioxide freezer with a manual door, we can't automate that unless they bought an automated freezer."

More things to consider

Some guidelines from the experts: "Go with a highly flexible, modular design," says Metherall. "Your project will end at some point, and the equipment will be obsolete if it's highly specialized."

Metherall's robot, for instance, has been used for several applications, including genome-wide gene-expression analysis of Drosophila embryos; handling whole-genome RNAi libraries in Caenorhabditis elegans; Frederick's synthetic genetic arrays in yeast; and kinase screens. "You started with minipreps but you could end up doing kinase screens, expression screens - just a wide range of things if you concentrate on flexibility," he says. Post-market and refurbished robotics equipment are available from Web sites like www.labtrader.com, www.atlanticlabequip.com, www.biodirect-us.com says Gurevitch.

Other items to consider: Will you need a dedicated staff member to run the machine? Are you trying to automate for safety as opposed to throughput reasons? How will you handle the glut of data that robots inevitably produce? Some of these questions can be answered by the Laboratory Robotics Interest Group (www.lab-robotics.org), which has regional branches across the United States and Europe and may be able to help introduce you to robotics users near you.

jperkel@the-scientist.com