Drug Discovery

Rarely does an argument over something like drug discovery technology make it beyond laboratory water coolers and drug industry technical conferences to the front page of the Wall Street Journal. The flap over combinatorial chemistry and high-throughput screening (HTS), however, did just that on Feb. 24.

Under the headline "Drug Industry's Big Push Into Technology Falls Short," the article aired the growing concern that these techniques, whereby researchers create and rapidly test a million or more chemical compounds for possible pharmaceutical activity, have failed to make good on their promise of bringing "a flood of medicines to patients and profits to investors."

What's more, the Journal contended, medicinal chemists are concerned that high-volume screening has taken them out of the loop at a crucial phase in discovery, replacing the ingenuity and intellectual capacity of the researcher with the programmed mechanics of robots. This new paradigm is seen as eliminating any chance for the kind of serendipity that led Sir Alexander Fleming, for example, to stumble upon penicillin in 1928.

The mainstream press coverage is not really surprising, given the steady drumbeat about the drug industry's new-product pipeline doldrums. The idea that drug companies have created a kind of numbers game in order to crack the code on cancer, AIDS, and other diseases makes for interesting copy.

Major pharmaceutical companies and other advocates of HTS are quick to counter--as they did in the Journal article--that expectations for new technologies are always unrealistic and that pipelines are beginning to fill up, thanks in large part to new screening techniques. HTS, they say, is a crucial tool that should never be mistaken for a silver bullet.

It is a tool, however, that many users, by their own account, are only beginning to use correctly. Screening-system suppliers and executives with smaller drug discovery firms, some offering alternatives to HTS, agree that the reappraisal detailed in the Journal article and in earlier C&EN articles (Oct. 13, 2003, page 77; Oct. 27, 2003, page 45) provides a fair assessment of how the technology has performed since its ramp-up in the mid-1990s.

Most sources agree that combinatorial chemistry is an important part of building a library of compounds from which to work and that HTS is needed at some point in the process of drug discovery. These techniques are seen as necessary steps in managing data and structuring experiments in what is inevitably becoming a more automated process.

Other technologies, such as microfluidics, are also making headway as fast, high-volume means of vetting compounds in drug discovery--sometimes in conjunction with HTS. Microfluidics, in fact, is catching on as a way of gaining control in the screening process by collecting absorption, distribution, metabolism, and excretion (ADME), as well as toxicology data, at an early stage in order to head off failures in clinical trials. Late-stage failure of compounds that originated from HTS is a major criticism of the technology.

In a recent survey by consulting firm HighTech Business Decisions, 51 HTS directors of drug companies reported a total of 74 clinical candidates that originated from leads found in HTS laboratories. Two drugs discovered through HTS methods are currently on the market. The report acknowledges disappointment over how few drugs have emerged in the 10 years since major HTS labs were established, but it argues that only in the past few years have scientists amassed the equipment, libraries, targets, and expertise necessary to advance compounds into the clinic with any efficiency.

WHILE GAINING a better understanding of compound libraries in order to conduct more targeted screening is typically the first step in boosting HTS efficiency, HighTech President Sandra J. Fox says the technique is also likely to be used more often in conjunction with other techniques such as microfluidics. "It's very interesting," she says. "Everyone is developing their own philosophy about how HTS should be done."

Given the basic philosophy that bigger--or more--is better, large drug companies are naturally the most involved in large-scale combinatorial chemistry and HTS. "For the last five or six years, high-throughput screening has been a rather central process in our drug discovery program," says Jeff Paslay, vice president for screening science at Wyeth. "We have a very good track record of developing small molecules as well as proteins. In the small-molecule world, high-throughput screening has been making a major impact."

The number of HTS campaigns at Wyeth--in which compounds are taken through three stages of screening--has quadrupled since 1995, doubling in the past two years, Paslay says.

The success rate at Wyeth has been "fairly steady," Paslay says. "We think we have a 70 to 75% success rate at coming out of the campaign with a compound that our chemistry colleagues are willing to work on." In an effort to improve the quality of its compound library, he says, Wyeth trimmed the collection two years ago from 1.5 million compounds to about 500,000.

Paslay says that, while he has worked on high-volume screening since the 1980s, the practice began in earnest across the pharmaceutical industry 10 years ago. Since then, HTS has sought its place in the drug discovery process. "High-throughput screening is clearly just one step in a larger process," Paslay says. But views vary from company to company, he notes.

"There are companies that say they are going to go to ultra-high-throughput screening--more or less an industrial approach. Others say they are going to continue to do high-throughput screening, but it will be on a more focused library. Libraries like ours with 500,000 compounds are still very large. I know of some companies that for certain targets are going down to 25,000 compounds."

At Aventis, HTS is a core technology, according to Tina K. Garyantes, the firm's head of lead discovery technologies in the U.S. "It is the bread-and-butter way for pharmaceutical companies to figure out what chemical space to explore during lead optimization," she says. "It is a foundation system."

Success comes down to experience with the technology, Garyantes says. "People are coming to realize that the downstream parts of the organization--medicinal chemistry and disease groups--can only handle a few chemical series. So the best thing to do in screening is give them a full appreciation of what are the best starting points."

Garyantes says some large drug companies are initiating pre-HTS tests for ADME and other variables that will determine success downstream. "It is important to know whether a molecule is bioavailable and soluble, and whether there are metabolism or toxicity problems. The question is, 'How early do I start looking at these issues?' " She says that, if the analysis is "cheap enough in time and materials," it is done during compound collection. Otherwise, testing is done on the active series of compounds coming out of HTS.

GlaxoSmithKline, which has invested more than $200 million in HTS, has seen a steady increase in the number of successful screens. Louise Dunn, GSK's director of science communications, says automation is giving a clear boost to early-stage pipeline productivity. She argues that since it takes 10 to 15 years to bring a new drug through R&D to market, it is too early to criticize HTS for failure to produce new drugs.

SOURCES AGREE that corporate size is not a paramount issue in effectively using HTS. Many drug discovery start-ups depend on the technology. PTC Therapeutics, a drug discovery company that recently moved its first compound into clinical trials, makes extensive use of HTS as part of its post-transcriptional-control discovery platform, which aims to develop drugs that treat genetic disorders and diseases by altering how ribonucleic acid produces proteins. Chief Executive Officer Stuart W. Peltz says the company runs some 16 HTS campaigns a year using its library of about 200,000 compounds.

"I wouldn't make the argument that HTS hasn't been effective," Peltz says. "I would say that you have to pick the appropriate targets, but it is somewhat of a numbers game to have enough programs going on so that you have a certain amount of success."

According to Peltz, it's unrealistic to judge HTS strictly on the number of new drugs it produces. "It is easy to criticize HTS," he says. "It takes a lot of work to move things forward. HTS is a tool, just as genomics is a tool. It is amazing that drug discovery is successful at all. It is always a combination of mission and science, and science is the great unknown."

On the other hand, the drug discovery services company SRI International has had a different experience with HTS. "We have a traditional drug discovery operation, with the exception of high-throughput screening," says James P. McNamara, director of business development at SRI's biosciences division. "That is something we never got into in a big way. Most of our success has just come from real medicinal chemistry, intuition, and insight using the more traditional tools of molecular modeling and quantitative structure-activity relationships." He adds, however, that the company has amassed a library of about 100,000 compounds over 40 years of drug discovery.

At one point, SRI dabbled in combinatorial chemistry, McNamara says. "That was a failure. When you start building these big libraries, there is no thought put into whether you are making druglike molecules. The molecules you get generally have very poor solubility, so there are formulation issues. They are not very bioavailable, so if you want to give them orally, the chance of them being absorbed readily is very low. There are a lot of things you would want to look at in drug discovery that high-throughput screening doesn't give you much information on." McNamara says he formed this opinion mostly in working with clients that have lead compounds developed through combinatorial chemistry and HTS.

McNamara admits that the cost of HTS has been a prohibitive factor for SRI, but he cites more fundamental objections. "I'm not sure we have the desire to do it from an intellectual, medicinal chemistry perspective," he says. "Screening seems like pretty much a thoughtless process. We will use combinatorial chemistry to a limited extent, but not to make hundreds of molecules. We might make 20 compounds around a scaffold." Most of the screening is done with in vitro models using technologies such as receptor binding assays, he says.

In some therapeutic categories, HTS has been a washout. "If there is one area where high-throughput screening has been a big disappointment, it is in antibacterial therapies," says Jeffrey L. Stein, chief scientific officer at Quorex Pharmaceuticals.

For most diseases, researchers need to discover and optimize an inhibitor for a single target, but antibacterial therapies require compounds that hit subtly different versions of the same target in different pathogens, Stein says. "Most big pharma companies go for broad-spectrum drugs by screening through millions of compounds. But there are different three-dimensional architectures in the active sites of antibacterial targets that make HTS difficult to use effectively. HTS has not produced a single compound that has gone into the clinic for an antibacterial therapy," he says.

Screening techniques have evolved in the antibacterial field along with cycles in demand, according to Stein. In the 1990s, when the surge of resistant bacterial strains caused a spike in interest in new treatments, drug firms were combining target-based HTS with genomics to develop novel targets. Generally, companies accessed enormous libraries that were not very high quality, Stein says.

Developing a robust assay was another challenge. For speed, most researchers used fluorescent assays rather than high-performance liquid chromatography. False positives and negatives forced them to step back, identify contaminants, and re-screen. "In the end," Stein says, "high-throughput screening became an oxymoron. It turned out to take longer, because you needed to take a lot of extra steps."

Quorex has developed an alternative in silico approach, called NanoLead, that quickly identifies compounds that serve as starting points for structure-based optimization, Stein says. The company scans an internal database of 2 million commercially available compounds. NanoLead also uses three-dimensional models.

Rather than collecting hits, NanoLead utilizes negative screening, first eliminating compounds with no chance of binding. Successive filters eliminate compounds based on ADME and other criteria. The system then assesses charge compatibility and the compatibility of the compound's shape with the shape of the target.

This process results in a pool of 2,000 to 5,000 compounds, Stein says. Because Quorex's filters are not biased to a structural class, the pool includes a diverse set of potential starting points that are organized into as many as 10 structural families. At this stage, the screening is taken out of the computer and biochemical assays begin, using HPLC or radioactive technology.

Protein, computational, and medicinal chemists go to work on cocrystallization, design, and synthesis, all of which are done in-house, Stein says. "The goal is to find something that can bind to a target," he says. "At the end of the day, you end up with a small set of compounds with a good chance of binding."

MANUFACTURERS of alternative screening equipment and informatics software, many of whom have established their own drug discovery operations, offer a range of views on the contribution that HTS makes to the discovery process. Most of these firms place more emphasis on the contribution of medicinal chemistry.

Dror Ofer, co-CEO of Keddem Bioscience, a newly formed drug discovery subsidiary of the Israeli bioinformatics firm Compugen, downplays the contribution of HTS. Nor is Keddem particularly oriented toward in silico discovery methodologies. "Our work has little to do with other projects at Compugen," Ofer says. "But it has everything to do with the philosophy at Compugen--the philosophy of trying to bring theory into drug discovery."

Compugen's basic assumption is that medicinal chemistry should be approached in the same way as other sciences and other branches of chemistry where the researcher is led by theory. "You need a clear, mathematically consistent model that has predictive power," Ofer says. "That is the way it is done in the more established sciences, and it is the way it should be done in biology and medicinal chemistry."

According to Ofer, medicinal chemistry has relied too heavily on an empirical approach, using "brute-force experiments" and simply hoping for the best. He sees an intuitive flaw in this. "People have millions of compounds in their chemical libraries. But when you think about it, chemical space is huge. A million is not a big number. It's the exact opposite. It's a ridiculously small number," he says. "When you ask researchers to explain the logic that leads them to believe they have a good chance of finding a lead in that million compounds, the best answer you get is that people have done it in the past and it seems to work."

That, Ofer says, is a valid answer, but indicative of the lack of understanding of the process. "If you look at medicinal chemistry over the past two decades, it's been all about volume--libraries have grown, throughput has increased, people are crystallizing more proteins. People keep doing the same thing, just more and more of it. There is no grasp of basic assumptions--where things can go wrong and what can be fixed."

Keddem wants to go back to basics, developing a system akin to structure-based drug design in which the researcher arrives at a full characterization of the active site on a target. "What you really are after is a map of the hot spots where molecules bind to the protein," Ofer says. Keddem uses a basic screening apparatus to do this. But rather than looking for leads, the firm is looking for units of information on the target.

"By approaching the problem in this manner, the relevant chemical space collapses to a finite set," Ofer says. "We feel we can arrive at a truly exhaustive set of compounds that, in the mathematical sense rather than the biochemical sense, tells you which options suit the target protein specifically. From this, you can deduce what the protein site looks like at a very high level of detail."

Ofer says the current computational algorithms employed in drug discovery are inadequate compared to the experimental screening approach Keddem envisions. He says, however, that the company will need to synthesize its own molecules. "We have gone through various huge collections and never found more than about 2% of what we need," he says, "because we don't care whether molecules are druglike. We want molecules designed around very rigorous mathematical and physical characteristics to maximize the amount of information we get from them." If things do not work out with a particular molecule, the company is still getting useful mapping information through experimental screening.

Other major bioinformatics companies have well-established drug discovery arms. Tripos, for example, acquired Receptor Research in 1997, establishing a base for the company's discovery work. Its closest competitor, Accelrys, recently spun off its drug discovery arm, Pharmacopoeia.

According to Tripos CEO John P. McAlister III, a lot of work has to be done to harness the information and know-how that has poured into the drug discovery research lab with the decoding of the human genome. For some time, however, there was too much attention being paid to one particular source of information, he says.

"In the mid-1990s, there was a hype around combinatorial chemistry and high-throughput screening. Everyone went crazy over it," McAlister says. "People had to come back to reality and figure out how to glean benefit from these technologies. People put their heads down and did good chemistry for a while, and now are getting the benefits of that good chemistry."

McAlister says information technology plays an important role in enabling chemists and biologists to work together in an environment where there is a heavy influx of new information. "Trial and error is always part of drug discovery," he says. "But now you have some opportunity to add rational approaches. You can take the information and knowledge stored in databases, combining pragmatic concerns with the more esoteric aspects of computational chemistry, to suggests things you should try."

Tripos is currently working with Pfizer on what the drug major calls its File Enrichment Program, an effort to i mprove the quality of its combinatorial compound library and the efficiency of its HTS campaigns. "I believe what Pfizer is engaged in is the embodiment of successful use of high-throughput chemistry to develop new ways of doing drug discovery," McAlister says.

SOURCES AGREE that as HTS finds its place in discovery, it is likely to be carried out in conjunction with emerging screening technologies that will assist in gleaning critical ADME, toxicity, and other information at early stages. Prominent among these is microfluidics.

Microfluidics, which has been called a variant on HTS, can be used to collect better ADME and toxicity information than standard HTS because it also separates product from substrate during testing, says E. Kevin Hrusovsky, CEO of the microfluidics firm Caliper Life Sciences. "Our focus is on helping researchers create better information that allows them not to progress a drug that shouldn't be progressed."

Hrusovsky says as many as 20% of HTS assays currently involve up to 1,536 simultaneous tests. This volume of testing means that smaller amounts of compounds and reagents are used per test point than on the standard 384-well HTS plate.

Miniaturization has been a primary focal point in discovery efficiency, Hrusovsky says, but it creates more bad data. "Reader technologies, assay development technologies, and reagent technologies haven't kept pace with miniaturization. New artifacts started to crop into the data package, increasing the rate of false positives, which are a kind of fool's gold."

Microfluidics can eliminate this problem, he claims, because the technology is more sensitive, precise, and repeatable than HTS. It even allows further miniaturization of samples. Microfluidics systems pull small amounts of liquid onto a chip where, through processes such as electrophoresis, product is separated from substrate. "Microfluidics gives you two orders of magnitude more information on whether or not the compound has activity and meets safety requirements," Hrusovsky says.

However, the greater efficiency of data generation through microfluidics does not eliminate the need to streamline compound libraries, according to Stephen D. O'Connor, CEO of Nanostream, a microfluidics developer. He says Nanostream markets microfluidics technology as part of a high-throughput liquid chromatography screening system that can be used simultaneously with HTS or in secondary screening to assess the solubility and purity of compounds.

Nanostream recently added fluorescence detection to its product, allowing simultaneous use of chromatography and fluorescent monitoring at biologically active concentrations. "We can do these in a medium- or high-throughput-screening apparatus," he says. Such medium-throughput screening, he predicts, may catch on faster than ultra-high-throughput screening as discovery researchers cut back their compound libraries to ones they understand better.

At Aventis, Garyantes says researchers are incorporating microfluidics using Caliper's system. "When you do HTS, you look at a huge number of compounds and get a large number of false positives just because of statistical noise," she says. "Microfluidics is allowing us to decrease the percentage of compounds that yield false positives for some classes of assays, and it allows us to do other assays that we could not do other ways. We can now do something close to a high-throughput assay that allows us to separate two molecules. That is very empowering."

Nanostream's O'Connor sees the discipline of drug discovery moving toward an approach that balances several scientific disciplines. "Pharma has gone through periods of thinking chemistry is all-important, or biology is all-important, or robots are all-important," he says. "I think they are coming to realize that you need a mixture of all of them. You need to have both the chemistry and the biology to know what's going on, as well as some tools to gather and make sense of all the information. This isn't a numbers game, and you really have to understand more about what you are doing."

Ofer at Keddem sees the tools as far less important than the researchers. Researchers, however, need to assert themselves in order to gain better control of discovery, he says.

"I don't view the machinery or robots as having any real significance," he says. "They have gotten orders of magnitude more attention than they're worth. The real issue in drug discovery is that we don't understand the key steps in developing a drug. We must say this openly and clearly. To understand, in science, means only one thing: the ability to predict the results. Medicinal chemists must study physical chemistry--how atoms really react to one another. You have to go back to the science when something doesn't work, rather than applying more brute force."