SAN FRANCISCO — Scientists at the Genomics Research Institute of the Novartis Foundation said they have designed a fully automated cellular-profiling system they claim can systematically investigate how panels of cellular assays react to molecular libraries in single dose-response experiments and in triplicate.
The researchers used the system to perform dose-response screens on approximately 5,000 compounds against approximately 47 tyrosine kinase cellular assays.
Jeremy Caldwell, executive director of molecular and cell biology at GNF, presented the results of the 10-day experiment at CHI’s High-Content Analysis 2008 meeting, held here last week.
In an interview with CBA News following the presentation, Caldwell said he originally suggested running a kinase of interest in a high-throughput screen to find an inhibitor. But, he said, “Instead of selecting only the compounds with the highest potency, why not look at the entire kinome and select the compounds with the most desirable specificity as well, and optimize the compound with the rest of the kinome data being generated at the same time?”
Such a format could enable researchers to look at the whole kinase family while they optimize a lead hit from an HTS screen for potency, which in turn could help them avoid “uninteresting or potentially untoward specificity issues with other kinases,” Caldwell said.
According to Caldwell, this study prompted the team to develop a robot that, instead of screening a million compounds against a single target, can screen thousands of compounds against an entire family of targets.
“That is the kind of data that a lead discovery team of chemists, biologists, and pharmacologists want to see before selecting what compounds to actually put into an animal model, and [which] management would like to see to determine which of those compounds to move into more expensive studies downstream,” Caldwell said.
The end result “plays on this whole idea of failing fast and failing early, to save money down the line,” he said.
According to Caldwell, the Genomics Research Institute of the Novartis Foundation employs about 40 engineers from the automotive industry. He said the institute hired the chief engineer from Saturn, “who was interested in getting into biotech, and he hired people from his previous life at General Motors.”
The first thing the GNF team did was build an ultra HTS (uHTS) screening system that aimed to solve a lot of the issues with cellular assays, including time requirements and cell maintenance requirements, as well as the cost of running HTS campaigns. He said that the HTS system, which uses 1,536-well plates, can screen 2.2 million compounds per day.
Caldwell said that by miniaturizing the screens in that format, the costs associated with the screen are reduced to 0.05 percent of those associated with conventional screens. “That allows us to run a lot of screens cheaply and really feed into the pipeline,” he said.
The other issue that the GNF investigators and engineers addressed is that in cellular assays run in 1,536-well plates, samples in the wells often partially or fully evaporate. “In 1,536-well plates, the wells can only hold 10 µL of liquid, which can evaporate in 24 hours if not handled properly,” Caldwell said.
The GNF team designed a stainless steel lid with holes on the top that are oriented in such a way as to optimize air exchange so that cells can breathe, Caldwell said. “A rubber gasket around the edge of the lid creates a seal with the plate so that air cannot escape around the perimeter,” he said.
These developments have allowed the researchers to run longer assays and use cells that are ordinarily difficult to culture, such as stem and primary cells. “Now, instead of being restricted to running assays that [take] 6 to 8 hours, we have gone out to three-week assays, which is a requirement for some of the stem cell-differentiation assays, such as adipocytes, because preadipocytes take 17 days to become adult adipocytes,” said Caldwell.
This uHTS system enabled Caldwell’s team to amass a database of about 350 screens across about 2 million compounds. “We wanted to mine this database for interesting attributes: ‘Could we identify all the compounds that were nonspecifically toxic? Could we identify only those that worked on immunology assays? Could we begin to predict the function of a compound or its target just on the basis of its activity in these disparate cellular and biochemical assays?’”
Caldwell said that one problem with HTS data is that they represent a single point at a single dose, so statistically “its not that significant to make these predictions about what the compounds are doing.”
Noting that it took his lab about six years to amass the 350 screens, Caldwell said the GNF team wanted to develop a system that would give them a richer data set in a much shorter period. Obtaining more robust data would require them to screen compounds in triplicate and in dose-response so they could look at the specificity of the compounds as well, he said.
“Why not look at the entire kinome … and optimize the compound with the rest of the kinome data being generated at the same time?”
According to Caldwell, dose-response data become important “because a compound hitting at 10 µM is a relatively high dose, and there is a relatively good chance that something hitting at that dose is still non-specific. But if it hits at 10 nM as well, [it] is really interesting and can be used.”
Caldwell said his team realized that it had to develop a system that could take hundreds of cellular assays at any given time and test them against selected sublibraries, or focused libraries, plated out in dose-response and in triplicate, then run a single screen and look at the data set. This, in turn, could enable them to make predictions about the compounds’ mechanism of action based on their specificity and activity across a battery of cellular assays.
“Once we built our cellular-profiling system and ran our first couple of experiments, we were able to get specificity and potency data and show how certain compounds had the same specificity profiles as one another,” said Caldwell.
He said that within a very short time, the researchers began to see side effects of some of their molecules on targets that they did not want to hit because they would lead to toxicity in man, or the compounds hit another target for a different disease.
“We realized in silico that we had a compound for an entirely different disease,” Caldwell said. “Maybe instead of developing the molecule for the first disease that we originally intended to develop it for, why don’t we go down this other path?
“What building the profiling system really involved, on a technical level, was taking all of our modules from HTS, from liquid handling to equipment such as fluorescence or luminescence readers, and then grafting on an automated, robotic tissue culture system,” Caldwell explained.
Then the researchers could call up whatever cells they wanted to have screened against whatever chemical library they were interested in, and have that data within a week or so, said Caldwell.
Caldwell said that the really new component of the profiling system is the tissue culture system. “It has incubators with carousels in them that carry flasks that we designed, called AutoFlasks,” Caldwell said.
The researchers built another station that can take a “mother” flask of cells and split it into however many “daughter” flasks they are interested in seeding. The daughter and mother flasks are returned to the incubator and grown into whatever state is needed, he said.
“The technology we used, which was already available, is the cell counter and densitometer,” Caldwell said. The data from those two pieces of equipment can be used to determine how much longer the flask needs to be in the incubator to grow up however many cells the investigators need to be in that flask.
“Once we built the incubators with the AutoFlasks and the tissue culture station, we integrated them using the robotic arm with the profiling equipment and the cell counter and densitometer to create our automated cell profiling system,” said Caldwell.
In order for the profiling system to handle so many cell lines at one time, the team had to develop an event scheduler, which is basically a script of everything that the robot is going to do, Caldwell explained. “An algorithm informs the robot of the most efficient path and use of time to move things around on the robotic system,” he said.
After the researchers built its first system and ran a couple of large-scale experiments, they realized that they could quadruple its capacity by using two robotic arms instead of one, Caldwell said.
“Instead of having one arm do both the tissue culture and the profiling, one arm takes care of all the cell lines and tissue culture, while the other arm performs of all the profiling and screening processes,” he said.
The researchers used this dual-arm system to run the HTS GPCR profiling experiments, one of which was 100,000 compounds against 41 GPCRs. Caldwell presented this data at the High-Content Analysis meeting.
“We have commercialized these systems. Merck, the NIH, and the Scripps Research Institute, among others, have adopted the technology,” said Caldwell. He said that the NIH has begun doing HTS experiments on the technology. “Our next step is to make these technologies available to those who are investigating neglected diseases or orphan diseases,” he said.