The process of automating compound screening is not easy. Here's how to avoid the pitfalls that commonly arise.

Patrick McGee
Senior Editor
Carol Ann Homon, PhD, saw her first laboratory robot in 1984 and was immediately converted. "I was an automation freak from day one," says Homon, associate director of biomolecular screening at Boehringer Ingelheim Pharmaceuticals Inc., Ridgefield, Conn. "In those days, people were really starting to feel the effects of hand pipetting. My original idea for automation was to eliminate stress on people from doing repetitive actions."

When designing their high-throughput screening lab, the most important thing for Amphora Discovery Corp. was flexibility. Most of the equipment is on wheeled platforms so researchers can change the lab's configuration as the need arises. (Source: Amphora Discovery Corp.)
It wasn't until the early 1990s that the pharmaceutical industry really began to realize the impact that automation could have on increasing process throughput and productivity, says Alastair Binnie, group director for discovery informatics and automation at Bristol-Myers Squibb (BMS), Wallingford, Conn. The mid- to late-1990s were "boom years" for automation, adds Binnie, an engineer who came to pharma from the aerospace industry. Companies invested heavily in automating more obvious platforms and instituted organizational changes to centralize intensive processes like high-throughput screening and compound management. "After that, there was a realization that although automation was very successful at tasks that we tried to automate, in fact there were still a huge number of discovery processes that were not yet automated. So there were a huge number of bottlenecks and until all the key parts of the discovery process had their productivity enhanced and their capacity increased to match processes like high-throughput screening, you weren't really getting full value for those initial early-on high-throughput technology investments."

As the new millennium dawned, companies tended to scale back on huge automation investments and began to face the fact that discovery is a combination of a large number of processes. Companies looked at how they could match the capacity or the impedance of processes, and for ways in which automation and informatics could improve them and bring the overall process to a greater level of productivity, Binnie says. "It seemed to be far more about application of automation informatics to more specialized scientific processes than the original big investments in screening compound management."

Things have changed greatly over the years, says William Janzen, PhD, vice president of operations for Amphora Discovery Corp., Research Triangle Park, N.C. "There's been a large influx of industrial engineering into screening, so there's a lot more robust automation. The robots are stronger now, and they're much more reliable." While some wonder if the automation investment pays off, Janzen is convinced it does. "The companies that have really focused on quality and a robust process are seeing value from their screening and are not asking if it is valuable. They're simply executing on it and have things flowing through the pipeline."

Like anything else, automating screening processes can result in major headaches for all involved if not planned and executed properly. Drug Discovery & Development spoke with researchers at pharmaceutical companies and academic research labs to get their perspective on what to do, and what not to do, when automating screening.

Do Think Before You Act
While it may seem obvious, many automation projects are undertaken before a clear plan is drawn up and goals are set. "What I've tried to inculcate in myself and in my group is thinking about some of the key success criteria that you should be able to identify up front before you do a big technology project," says Binnie. Following the automation boom of the mid to late 1990s, many managers became disillusioned after finding the labs they had been responsible for investing in and setting up were not delivering. In some cases, researchers turned back to manual processes because that was the only thing that was reliable. "You need to avoid those kind of white elephants, those projects that are big investments but don't actually prove value in the hands of the users. Any technology project is only really valuable if an end-user scientist can use it to make their job more productive or to generate better quality data."

The key to avoiding these white elephants is to identify project success criteria before any investment is made. What will the business impact of the automation be? Will the automation provide significant improvements in productivity, data quality, or decision making? Will it significantly decrease cycle time? After those questions are answered, the technology group needs to establish a clear goal line for the completion of the project. "It is always tempting to allow the scope to grow—we'll add a little bit more, we'll fiddle around with it a bit more—but unless the tool is in the hands of end users, it is not adding value," Binnie says.

If screening is going to be done continuously, setting up proper contracts with vendors for consumables and with hardware manufacturers for service is vital. Lastly, it is important to make sure that a database is ready to handle the information gathered. "A lot of people will ignore that and keep track of stuff on Excel spreadsheets or Notepad," says Peter Hodder, PhD, associate director and head of lead identification at Scripps Florida, Jupiter, Fla. A big part of his efforts at Scripps is to ensure that they have a high-throughput screening (HTS) ready database that keeps track of not just the assay data but other information, including how many compound plates are there and what their volume is. "These are numbers that are critical to help us run smoothly."

Do Talk With Users Before Acting
Extensive discussion with end users should be integral to the planning process. The technology group should be sure the project's sponsor, as well as the end users, are committed to using the new technology. "Whether it's a program or a piece of instrumentation, they've got to want it and they've got to be committed and proactive in helping you to develop this," Binnie says. If an in-house technology group is going to be handling the project, they need to make an honest assessment of whether they have the time and the skills to take the project on. "Any significant project takes up a large amount of bandwidth and you shouldn't over-commit. That's common sense, but it's very easy in the heat of the moment to get carried away and say, 'Yes, we can build that,' and then three years later it's still not built."

In addition, if an internal automation group is being used, they must not monopolize the project. Too often, these groups insist that the system has to be invented or at least identified internally by they alone. "Scientists are typically very technically aware and many of them are really enthusiastic technologists as well as scientists. The best way to work with them it to be partners and recognize the potential they have to lead technology innovation as well," Binnie says.

Getting core therapeutic area scientists on board can be difficult, however, especially at larger companies, says Janzen. While such scientists are clearly a valuable resource due to their knowledge of biology and chemistry in their core area, they usually have a set way of doing things that can be difficult to change. "Quite often, there are very good reasons for some of their methodology, but there's also usually a great deal of almost mythology around the experiments." Janzen says one of the most common myths centers on the need to incubate overnight. "When you trace it back, the reason they incubated overnight is it took four hours to set up the experiment and they started right after lunch."
A partial image of Amphora's 28,000,000-point database, this heat map displays the percent inhibition of compounds (rows) against targets (columns). The activity of each compound is displayed by varying the color from blue (inactive) to yellow (very active). Promising series of compounds are easily identifiable by examining the patterns of activity against the entire panel. (Source: Amphora Discovery Corp.)

Do Buy, Don't Build
Who would build a car from the ground up when they could go a dealer, sign on the dotted line, and drive away in the same car? Yet that is what many companies do when it comes to automating their screening processes. "That often creates a lot of problems when you're trying to get different pieces of equipment to work together," says David Mark, PhD, senior research director of discovery technologies at Hoffmann-La Roche Inc., Nutley, N.J. "If you can buy off-the-shelf systems, that would be the best way to go."

Binnie says that while it is natural for technology teams to get excited at the prospect of building their own system, they need to consider the reality of the situation. Building a system is one thing, maintaining it is another issue altogether. "You're going to have to keep supporting it, you're going to have to fix it when it breaks, you're going to have to upgrade the software when new external software becomes available, and so on. How are you going to do that, and are you just going to keep it in maintenance mode or are you actually planning to continually enhance it, which is the more expensive thing to do." Instead of building their own systems, most companies buy the best components for each application and integrate them.

For example, Hodder's department is in a central facility that will serve all of Scripps' ultra-HTS needs. In October, the facility will get two robots from Kalypsys Inc., San Diego. One will be for screening and the other will be a cherry-picking robot for library storage and reformatting. The screening robot can screen up to a million compounds in 24 hours using 1,536-well plates, and the robots are almost entirely customized for their use. Scripps is planning to ramp up its collection to about a million compounds, making a 1,536-well format screening robot very cost-efficient. In addition to the two main centerpiece technologies, there will be a great deal of ancillary equipment such as a large-scale compound storage unit, liquid handlers, detectors, and mass spectrometry units for determining sample purity.

It is also important to remember that HTS is not as simple as buying a robot and plugging it in, says Berta Strulovici, PhD, research vice president, automated biotechnology, Merck Research Laboratories, North Wales, Pa. "People think that if you buy a robot you do high-throughput screening….People focus on the robot, but that's not really what it's about; it's an integrated organization which will also require multiple skills," including assays, software, and instrumentation. Another problem is that many labs are set up based on the mistaken assumption that all robotics systems are turnkey. They are not and must be carefully integrated with other systems and then supported by engineers who can troubleshoot problems as they arise.

Do Test Before Buying
Homon strongly recommends testing automation equipment under real-life HTS conditions before buying. "There's too many things out there that look perfectly fine for 5, 10, 20 plates, but when you get up to a full HTS consideration of 120 to 150 plates, you just start seeing things happening that you would never see in a low number of plates. . . . Almost every vendor has a customer that you can go to and actually see a system running under real-life conditions, but don't just trust the vendor's word that this happens. Get out there and see things in action."

Hodder recommends bringing each vendor in and testing the equipment for a specific application. "You have to spend a lot of time evaluating these new technologies, giving them a test-run to see if they can do what [the manufacturer says] they can." He tests for qualities like precision, accuracy, and robustness. Homon adds that if money is an issue, companies should ask vendors about used equipment, some of which is refurbished and sold with a warranty.
Carl Zeiss assembled a screening system for Hoffmann-La Roche that consists of custom-designed and off-the-shelf components. (Source: Hoffmann-La Roche Inc.)

Do Build Flexible Systems
Amphora's Janzen has been involved in automating three screening labs and believes the most important thing he learned is to keep flexibility in mind. "Don't build any more walls than you have to and make everything on wheels. In our labs, literally every lab bench, every robot, is on a wheeled platform, everything that can be. Obviously, some of the automation needs to be more heavily fixed than that." Another key lesson is giving scientists in the lab the flexibility to modify the processes and the workflow. "They will know more about it after they've been in there working for a few months than you could possibly design in at the very outset." But that commonly doesn't happen. Janzen says the most common mistake he sees is people building a large inflexible automation core, and then forcing people to modify their workflows around it.

Homon says systems should be built in a modular manner so that modules can be added or deleted as technologies evolve and assay formats change. "The only way you can do that is by being able to incorporate different modules for the different assay formats. There is a bigger outlay of initial cash, but in the long run your system is going to go a lot longer." In addition, the systems should also be able to integrate with different readers, Homon believes, because in the future there will be more label-free technology. "These are entirely different types of readers or detectors."

Do Pay Attention To Details
Seemingly small things like washers can play a big role and cause problems if they are not attended to, Homon says. "Never let your washer go dry. People tend to use it and then not use it for months, and then while it's not being used they just let it sit off in the corner and it goes absolutely dry. Of course, all those fittings and things dry out and it doesn't work." She also recommends replacing the pumps that come with the washers, which she believes are not powerful enough. "They don't go into absolute, abject failure. They start failing and mis-pumping, and as it does you get uneven vacuum across the manifold. So just purchase a slightly bigger pump." People often don't pay attention to the pumps because they are out of sight, but they can have an effect when they malfunction. During one experiment, everything went smoothly when doing 10 or 20 test plates, but when the researchers ramped up to 120 plates, the washer pump was overheating, Hormon says. "Even if you cycle it, it's cycling so fast that it doesn't really help it out."

As with many things, the devil is in the details, down to the level of ensuring that temperature and humidity are controlled, says Hodder. The physical space that robots or other equipment will occupy need to be carefully considered as well. "Things related to facilities like changing out wash solvents, having tool kits nearby, having lab space where you can stage and store plates for HTS . . . . are things that you don't think about but which become very important later."

Don't Automate Everything
While it is tempting to see automation as a panacea for a compound pipeline that is in danger of drying up, avoid the temptation, says Homon. Her advice? "Automate where appropriate." Mark recommends automating assays that are more automation friendly. That may mean modifying procedures slightly to make them more automation friendly. Thinking about the necessary throughput is crucial as well. "Sometimes people think about automation for automation's sake, but if the assay doesn't require high volume, then automation may actually make it less efficient," Mark says. "You really need to be running it routinely and with a large number of samples in your assay. Otherwise, there's no point in doing your automation."

While it is important to automate where appropriate, companies should also avoid automating in a piecemeal manner. But that is often what happens, Janzen says, mainly due to budgetary constraints. "At the companies that have done a really good job, upper management allocated enough money to do a complete automation system in an entire division. But usually what happens is people get enough money to fix one part of the process, then they fix another part. It's actually pretty rare for someone in upper management to have the vision to say, 'Let's examine the whole process, find all the bottlenecks, and automate everything that we can to try to make this more efficient.'" Janzen and others say one company that has done a good top-down job of automating processes is GlaxoSmithKline.

Don't Automate To Mimic Manual Assays
"I think the biggest mistake I see people make as they go into automation is to perform the assay exactly as they do it by hand," says Homon. "That may not be the most efficient or even the best way to do it if you can do it in an automated manner."

The first thing she tells people to do is to sit down and describe the process, not necessarily what they do by hand, but what the process is that they want to automate. Mark says he has also witnessed people trying to perform automated assays in the same way that they do them by hand. "Sometimes, that becomes a little difficult, so you have to be a little more flexible in potentially modifying an assay if you want to automate it."

Don't Limit Your Range Of Products
Finally, it is important to remember that little things mean a lot. For example, people will often try to save money by limiting themselves to one type of pipetter, Homon says. "You can end up with sticky proteins, sticky peptides, sticky compounds. You really have to have a range of pipetting devices such as a plastic-tipped device, Teflon-coated tip, broadcast dispensers, ones with silicon tips, ones with glass syringes. You really have to have a toolbox full of different pipetting options. That's a must."

Another thing that people try to scrimp on are readers. "Not all readers are created equal or even the same, and so we recommend strongly that you have at least two different types of readers in your lab," Homon says. She adds that one reader may get no signal, leading researchers to believe that the assay is not working, but a second reader will often pick up the signal. "It's the reader that for one reason or another is not picking up that assay."