Screening the Screeners

A host of start-ups want to make high-volume screening the method of choice for finding drug leads and for discovering the uses for those new, mysterious targets pouring out of the genetic databases, But few of the new firms are basing their long-term plans on screening services; most are either using their systems to look for their own drugs or still evolving their business plans along with their technologies.

These start-ups want to make high-volume screening the method of choice both for finding drug leads and for discovering the uses for those new, mysterious targets pouring out of the genetic databases.

by Roger Longman

• A group of new screening companies wants to radically reduce the cost and increase the accuracy and speed of screening large libraries of compounds, enabling discovery researchers to find drug leads faster and cheaper.

• But combinatorial chemistry start-ups are creating smaller, smarter libraries; if screeners can use libraries of a few thousand, or even fewer, compounds, than high volume screening systems will be less valuable.

• Screeners are also hoping to use their systems as an alternative approach to discovering the uses of the many newly discovered targets of unknown function.

• Few of the screening companies are adopting a pure service strategy: most are either using their systems to look for their own drugs or still evolving their business plans along with their technologies.

Call it the triumph of the irrational. Following the discovery and development in the 1970s of Squibb Corp.'s captopril—the first drug more or less designed to fit into a particular receptor—the pharmaceutical industry poured millions into rational drug design technology. The investment grew with the launch of rational design start-ups like Agouron Pharmaceuticals Inc. and Vertex Pharmaceuticals Inc. Using never-before-available computer power, scientists figured they could tease out the structure of the biological target to be activated or inactivated, then build the molecule to fit. Traditional "irrational" drug screening—taking a library of chemical compounds and "irrationally" testing each one against a biological target hoping to find a hit—would become a thing of the past.

Things didn't work out that way. Although Agouron and Vertex have since gone on to some dramatic success, it is hardly due to rational design per se. Agouron's Viracept, its anti-HIV protease inhibitor, actually originated in Eli Lilly & Co. 's research labs—and only went to Agouron as part of the dénouement of the two companies' 1988 antiviral deal. Indeed, most drug firms still feel burned by the rational drug design mania: too many dollars, too few results.

Just as Agouron and Vertex were getting started, Oncogene Science Inc.(OSI) was going in a very different direction. By 1987, the Long Island-based, cancer-focused biotech had licensed out most of its partnerable assets: Pfizer Inc. had acquired its oncogene-based pharmaceutical targets; Becton Dickinson & Co. had made a deal for their diagnostic applications. Its new direction, inhibiting gene transcription, was a hard sell to potential partners, notes Gordon Foulkes, PhD, who had been recruited to the company in 1987 and tasked with the job of managing the Pfizer relationship and building a new, partner-able technology platform.

The problem with selling gene transcription as a pharmaceutical target area was that there was little evidence of its value: few drugs were then known to act at the transcription level. Foulkes believed that the quickest way to find new drugs that did was the old irrational way: to screen OSI's transcription targets against compound libraries. The more compounds one screened, he believed, the higher the likelihood of coming up with a hit. But screening was still largely a manual process, particularly for assays that were new and complex. Hiring the number of technicians required to screen large libraries would be prohibitively expensive for OSI. So the company decided to automate the procedure.

It took several years, notes Foulkes, to get its automated robotic system up and running reliably—and then plenty more convincing drug companies of the value of screening more than 20-30,000 compounds. Few screened more than 10,000 compounds against a target—and screening itself was hardly de rigeurin the late 1980s and very early ‘90s.

But OSI's strategy worked—eventually. Its twin arguments—the value of its targets and its ability, with its automated system for screening thousands of compounds very quickly (a technique now called high-throughput screening, or HTS), to maximize the chances of finding a compound that affects the target—have helped make OSI one of the most successful dealmakers in biotechnology—at least in terms of keeping its deals alive. Of the eight Big Pharma alliances it has signed since 1986, only one—its deal with American Home Products Corp. 's Wyeth-AyerstLaboratories unit—is no longer being funded, and Wyeth is apparently working to develop one of the products of that collaboration [See Deal]..

By the mid-1990s, OSI's demonstrable success in dealmaking had spawned a host of screening-oriented imitators. Foulkes himself became chief technical officer at Aurora Biosciences Corp. , which is focused on taking screening from OSI's HTS to the next level, ultrahigh throughput screening (UHTS)—not merely screening thousands of compounds a day, but hundreds of thousands. "My dream," says Foulkes, "is to build the fastest HTS and drug discovery system in the world." OSI, meanwhile, is focusing greater efforts on developing drugs on its own. Still, its recent announcement that it had received a broad patent covering the use of reporter genes which identify compounds that modulate gene transcription will likely net OSI further partnerships—and could, wrote Robertson Stephens analyst Jay Silverman, force a number of companies to license the technology for work they already have underway.

OSI's pioneering work in HTS reflected a need that would become even more acute following the industry's general disillusion with rational drug design. Automated screening seemed to be the best alternative for exploiting the new targets coming out of biotechnology—and, by the mid-90s, genomics. Moreover, by 1993, combinatorial chemistry was providing hundreds of thousands of new compounds to test against the new targets. But screening departments had largely atrophied with the industry's wholesale embrace of rational approaches, notes Martin Haslanger, PhD, Lilly's executive director, research technologies and head of its combinatorial chemistry and screening programs. Following the example of OSI, small companies have rushed in to fill the gap.

By screening more compounds faster, these companies aim to dramatically increase the chances of finding new drug leads. One major problem with screening today is the expense: the assay for a 100,000-compound screen, for example can cost $200,000-500,000. Many of the new companies propose miniaturization schemes in order to reduce the amount of reagent necessary. The lower the cost, the more screens that can be performed—and thus more leads generated.

Moreover, as the cost of screening falls, its uses expand. The number of targets whose functions are poorly understood or unknown has been multiplying ferociously thanks to the many new genes being constantly turned up from industry and academic databases and sequencing programs. Teasing out their uses has largely been the job of functional genomics researchers at companies like Millennium Pharmaceuticals Inc. and Exelixis Pharmaceuticals Inc.

But screeners—granted they can reduce the costs of the job—feel they've got a useful approach, too. They propose screening chemical libraries against these incompletely understood targets in order to find chemical compounds that make the targets stop or start doing something. When that something is known, scientists will not only have a good start on gaining an understanding of the target, they'll also have leads on compounds that modulate its function—and thus be further down the road than competitors to putting new compounds into the clinic.

One strategically important feature of these ambitious new companies is just how new they are: Aurora, one of the first start-ups to identify screening as a business opportunity in itself, apart from specific drug development programs, was incorporated just two years ago, in May 1995. A number of other companies have popped up in the months following it, including BioDx Inc., and Small Molecule Therapeutics Inc. Caliper Technologies Corp. , a Palo Alto-based start-up, has only in the last few months decided to focus its efforts almost entirely on developing a screening business, seeing both a crying need for its UHTS system and a willingness on the part of Big Pharma to pay significantly for it—far more than the in vitrodiagnostics and research markets would have paid for its technology.

All of these companies are racing to develop systems and entrench them into the drug industry's screening programs before incrementally better—or completely obsoleting—systems from competitors become available.

And before other fundamental scientific shifts can alter standard practices. A number of companies and scientists are leading the charge towards smaller, smart libraries; the smaller the library, the less necessary a massive HTS system.

But most screeners aren't openly dismayed by the possibility, particularly given the plethora of new targets. It makes little difference if the industry needs to scan 500,000 compounds against one target, they argue, or 5,000 compounds against 100 targets: either way, drug companies will require HTS systems.

Dividing the Screening World

For Jean-Francois Formela, MD, a partner at Atlas Venture, screeners generally trade off throughput for resolution, and vice versa. The more information required from a screen, the slower it is. So-called molecular or binding assays, the majority of those employed in screening today, are generally easier to set up and faster to run but give little information about how their targets function when agonized or antagonized by a compound; they simply note whether the compound binds to its target. Functional screens, on the other hand, can deliver lots of information about how hits affect targets but are generally slower and more complex to set up.

But as Atlas' own investments show, new technologies are changing the rules of such simple tradeoffs. New developments with fluorescent tags, Formela notes, have allowed companies to develop sophisticated assays which provide extraordinarily detailed quantitative data, often from living cells, and to do so on a high throughput basis. Indeed, one major development of the last decade has been the improvement in whole-cell screening—a technique which allows screeners to observe how binding affects a target within a more or less "real" biological environment, as opposed to the simpler and more common method of testing the purified target outside of its natural cellular home. According to experts, roughly 30% of all screening today is done in cellular assays.

Atlas has investments on both sides of the screening universe—in Scriptgen Pharmaceuticals Inc. for the high-throughput, low-information side and in the German start-up Evotec BioSystems GMBH for its highly informative functional screens. Evotec, founded in December 1993, isn't trading off much in speed: it aims to be screening 100,000 compounds a day relatively soon. But it is still limited in the number of assays that work with its system. Another high-information, high-speed company, Aurora, has as an internal goal a system that can perform 500,000 functional assays in a single day, says Foulkes. Aurora's key advantage: the wide variety of assays that can work on its system.

All of these new companies have sprung up to respond to a host of screening challenges the drug industry never had to address when compound libraries and the number of new targets were much smaller. One of the most fundamental problems is expense: standard 50 microliter assays can cost $2-6 apiece. A traditional 5-10,000-compound screen will be expensive enough—some $10-60,000. But making enough target molecules and formatting them into assays to screen against 500,000 samples can cost $1-3 million—not including the cost of other lab supplies, which aren't cheap either. The 96-well microtiter plates used in most screening programs, for example, cost from as little as $2 to as much as $50 a plate, with most averaging $5-10. Using each well for a different compound in a 500,000-compound screen would require 5200 plates at a cost of some $26-52,000, not including the labor, automation and time costs of storing and manipulating all of the plates.

For these reasons, among others, companies generally put multiple compounds in each well, preferring the problem of sorting through the higher false-hit rates—often because of cross reactivity between compounds—to the costs of the much cleaner, single-compound-per-well screening.

The compounds themselves aren't inexpensive either—or necessarily plentiful. Indeed, one of the problems with some of the combinatorial processes for creating libraries, such as Pharmacopeia Inc. 's, is that they don't make large amounts of each compound. Each new screening run can deplete the stores substantially.

Miniaturization

One answer to the above problems is miniaturization. If the amount of target in each assay can be reduced, costs should go down—both the cost of the assay itself and the cost of the compounds, associated reagents and lab supplies. Some screeners have already reduced assays to a single microliter—a 50-fold reduction in average size—and are achieving accurate, reproducible results, notes Mark Crawford, PhD, VP of Lasure and Crawford, a Seattle-based consulting firm specializing in screening issues. Some microtechnology (often called nanotechnology) companies want to reduce assay sizes much further, to 50 picoliters (a million-fold reduction in size), cutting reagent costs to negligible levels.

Reducing the amount of reagent should also allow screeners to increase their speed. Aurora, for example, is creating a microtiter plate the same size as a standard 96-well plate—but with 36 times the number of wells (though the plate won't cost, Foulkes hastens to add, 36 times the cost of a standard plate). Where 96 experiments can be done simultaneously with standard equipment, Aurora is aiming to do 3,456 shrunk-down assays. A research group at Du Pont Merck Pharmaceutical Co.claims to have developed a 9,600 well plate.

Finally, reducing the size of assays decreases the need for elaborate production infrastructure. Putting together standard assays requires dozens of people and expensive equipment in order to get enough of a particular target into a stable cell line, notes Larry Bock, a former partner at the now defunct Avalon Ventures, the founding venture firm for both Aurora and Caliper. "But if you're only using a very small amount [of cellular assay], you don't need a cell culture system; you can do it directly," that is, by producing enough usable cells simply with the tools available at the lab bench. "And if you can do it directly," Bock continues, "you can be more flexible in your experiments—mutating the target, for example, and seeing how that affects things." In short, smaller samples, he says, "gets you more functionality."

Signalling Technology

But miniaturization brings with it a host of technical challenges, not the least of which is a way to detect and report a reaction in very tiny samples. Fluorescence-based techniques are probably the most popular of the new signaling approaches. Screeners have developed techniques for using this basic ability of molecules to absorb and emit light to detect hits even at the level of single cells. Just as importantly, fluorescence and its detection doesn't have to interfere with the biological mechanism the researcher wants to monitor, making it an excellent real-time reporter system.

The problem with fluorescence-based reporter techniques is that different fluorescent systems are required for different classes of targets. One technique won't work for everything. For example, ion channels, which regulate the voltage in cellular membranes, require a reporter which can detect when a compound targeting one of them alters membrane voltage. Proteases and kinases perform quite different biological functions and require quite different ways of monitoring them.

Aurora recognized this issue early on and thus, at its founding, licensed a set of fluorescent technologies which are being adapted for many of the major classes of drug targets: proteases, protein kinases, ion channels, intra-cellular signaling proteins and a variety of receptor classes at cell surfaces and within cells. Most of the basic discoveries came from the lab of Aurora's principal scientific founder, Roger Y. Tsien, PhD, a biochemist and Howard Hughes Investigator at the University of California, San Diego ; still others came from the University of Oregon and from the California Institute of Technology lab of Melvin I. Simon, PhD. And Aurora continues to keep its eyes open for other assay technology, most recently licensing a testing technique from Lidak Pharmaceuticals for lipid metabolism [See Deal].

Evotec, with its high-throughput EVOscreensystem, approaches fluorescence-based screening from a very different angle, coupling it with mass spectroscopy in order to be able to increase the amount of information it gets from each assay. Evotec's fluorescence correlation spectroscopy (FCS) exploits the fact that the weight of molecules changes when they interact with others—when, for example, one molecule binds with another. EVOscreen reads weight changes in fluorescently tagged molecules of interest, theoretically allowing screeners to read interactions even at the single-molecule level. FCS (and a cousin, FCS+, which works in situations when there are no molecular weight changes) also takes advantage of another aspect of fluorescence—that different changes in fluorescence reflect different changes in the molecule itself. Alterations in brightness, in the color the fluorescing molecule emits, or in the length of time it stays bright, all reflect biological changes in the target being studied—changes prompted by compounds which bind to it.

Poorly Understood Targets

Still, most of the new functional fluorescence-based assay systems require considerable knowledge about the target before they can be used—often the identity of the molecule which naturally binds to the target and some clue as to its function. The problem, says Scriptgen's VP of R&D Michael G. Palfreyman, PhD, is that 40-50% of the new genes discovered are of totally unknown function. Only one-tenth of the remainder, he estimates, are well-enough understood to allow high-throughput screening by the newer functional assay systems or the more traditional binding assays.

Scriptgen's answer is to avoid functional assays in favor of a one-size-fits-all, high-throughput, binding-assay reporter system that it patented in January 1996. The technique is based on a well-known principle of biochemistry: a ligand—the molecule which attaches to the target—increases the stability of the protein it binds to. Scriptgen's Atlas(Any Target Ligand Affinity Screen) and Scan (Screen for Compounds with Affinity for Nucleic Acids) systems both measure the ability of compounds to increase the stability of a protein—and thus whether or not the compound is binding to the target. Since the company's systems work with virtually any drug target—Atlas with proteins, Scan with the RNA which is a crucial target for many antibiotics—they are, says Palfreyman, "universal systems."

The advantage of the system is considerable, even with targets of known function. The P-53 target implicated in various cancers, for example, has 11 or 12 known biological functions, says Palfreyman. Any screening program will need to find compounds that modulate some of P-53's functions but not others—which means that the screens will have to test against all 11 activities. A functional high-throughput P-53 screen with 100,000 compounds will require at least 1.1 million assays (100,000 compounds x 11 functional assays), notes Palfreyman. With Atlas, on the other hand, screening the same 100,000-compound library would find, on average, 100 compounds which bind in some way to P-53. Each of those 100 compounds would then be screened in 11 functional assays, bringing the total screening program, in this hypothetical example, to 101,100 assays—a far more manageable and less expensive number.

Scriptgen isn't the only company going after the difficult-to-screen targets: Novalon Pharmaceutical Corp. has a technology which in effect allows it to create probes which interact with the functional domains of various targets, well-known or newly discovered and uncharacterized; partners will screen libraries against the probe/target complex, looking for compounds that displace the probe. Novalon's first deal may be its last: Cubist Pharmaceuticals Inc. signed a research and equity agreement that also gives it a six month option to acquire the company [See Deal].

Instrumentation

But while chemists and biologists can propose plenty of novel reporter systems, based largely on their expertise, they have little understanding of the required instrumentation. The problem isn't simply in designing and building instruments which can do the screening—lab equipment manufacturers, such as Zymark Corp. and Tecan Group, already supply products and robots which can do many of the required tasks—but in making them robust enough. As Mark Crawford says, "Scientific equipment is made for research labs. It's finicky. Vendors are used to making things people turn on once a week and use for a few hours."

But HTS is an industrial process, with facilities in constant, heavy use. The result, says Mike Palfreyman, is that "you walk through a company's screening facility and get shown all this equipment but nine times out of ten it isn't working on anything. Technicians are fixing it." Agrees Lilly's Marty Haslanger: "We've been through an interesting learning process with lab instrumentation. These lab robots are toys. You need industrial instrumentation, which is much sturdier."

From all accounts, Aurora understood from the beginning that its UHTS system required industrial-level performance. In building its system, the company has ranged far outside biotech's usual haunts, in both sourcing equipment partners and in hiring scientists. Foulkes estimates that more than half his research staff are engineers and biophysicists culled from companies outside the drug and biotech industries.

Likewise, it has turned for development help for its microfluidic components, NanoPlates(its 3456-well microtiter plates), fluorescence detectors, liquid handling components, robotics and systems for the storage and retrieval of millions of compounds, to engineering firms like Packard Instruments Co. (which also invested $1 million in Aurora) [See Deal], and Universal Technology Inc.

But Aurora, and other miniaturizers like Evotec, have to contend with far more problems than simple reliability. Both companies are in effect developing shrunk-down versions of current equipment, albeit with very different reporter systems. But miniaturizing standard testing "is not a linear process," says Larry Bock. Simply adding more wells to a microtiter plate brings up plenty of problems that don't exist at larger scales. For example, tiny amounts of solution-based target evaporate quickly, requiring carefully controlled air temperature and humidity. Tiny samples can also be far more easily and damagingly contaminated by the micro-pipetting systems used to add chemicals to them. Adds Gordon Ringold, PhD, the president of Glaxo Wellcome PLC 's Affymax Research Institute : "Here's a problem Glaxo worries about a lot: once you're in a miniaturized assay, how do you know whether the compounds are staying in solution? Do you have a quality control assay to prove that?"

Because of these problems at the microliter level, advocates of chip-based screening technologies—in which all or most of the assay is done on tiny silicon or plastic wafers, not in microtiter plates—contend that HTS systems need to be altogether re-engineered.Bringing assay sizes down from Aurora's one-microliter level, these companies aim for picoliter-sized assays—a million-fold reduction in size.

Advocates of microtechnology point out that, with such small assay requirements, chip-based systems can theoretically bring screening down to the level of the individual biologist, decentralizing a research tool that has historically been built in single sites—such as Roche 's huge screening facility in Japan or Merck & Co. Inc. 's in Spain. For example, instead of requiring the kind of infrastructure necessary to produce robust cell lines large and stable enough for most HTS systems, biologists can, on their own, produce enough target to test against several thousand compounds—and then mutate the cell lines for further testing in response to new information gleaned from the screening.

Indeed, the idea of decentralized screening has some significant organizational advantages. Screening groups are themselves often isolated, from the chemists who make the libraries and optimize the leads, on the one hand, and from the biologists who need the screening information. One research manager at a large company pointed out the "histocompatibility problems"—chronic rejection problems—a competitors' biologists were having working with the company's newly created, centralized, state-of-the-art screening group. Other formerly centralized technologies are also beginning to be available to site workers: Argonaut Technologies Inc. and Hewlett-Packard Co. both offer instruments that allow biologists to make small-scale combinatorial libraries far from the company's centralized compound library.

Perhaps the most advanced of the various microtechnology screening systems is Caliper's. The Palo Alto-based start-up is building its system around variously sized chips, some as large as a school pencil eraser, others no bigger than a dime. Using photolithography and chemical etching, the company digs tiny channels, a few microns wide, into the chips and then puts tops on the channels, making them into conduits (these closed systems thus avoid the problem of evaporation). Picoliters of samples are moved through the conduits with electrokinetic forces generated by different "pin-outs"—tiny metal pins, about the size of pins at the end of a computer cable, inserted at particular places on the chip. New samples move through the chip every few seconds; in one day a single-channel chip can do nearly 6,000 assays, says co-founder Michael Knapp, PhD, the company's VP of science and technology. The company is also working on multi-channel chips so that one chip can perform multiple assays simultaneously. Results can be read by a variety of reader systems, including mass spectroscopy and fluoroscopy.

The requisite equipment is small, too: the desktop instrument within which the chip works is about the size of a large VCR, says Knapp. It's relatively cheap to make, and sturdy: lacking most of the robotics other miniaturization programs require, Caliper's technology requires few moving parts, which can break down, and is instead made up of inexpensive electronic components.

Caliper is by no means alone in going after chip-based systems. Soane BioSciences Inc., with its capillary electrophoresis techniques, and Orchid Biocomputer Inc., with its own microchip technology, are also developing chip-based screening systems. But Caliper, after some initial strategic dilly-dallying with various ideas for its technology, has now decided to focus its most intense efforts on HTS. In contrast, Orchid's first efforts are in combinatorial chemistry, in large part because of its 1985 alliance with SmithKline Beecham PLC, though it is also working with its majority owner, the David Sarnoff Research Center, on sophisticated fluorescence-based detection and screening systems. Soane is developing a variety of applications for its chips, collaborating with Hitachi Chemical Co. Ltd. on point-of-care immunodiagnostics and working on DNA analysis and sequencing tools.

Mine Won't Work with Yours

But for Gordon Ringold the argument between the miniaturization approaches of Aurora and Evotec on the one hand, and the microtechnologies of Caliper, Orchid and Soane, on the other, misses the point. "People are paying plenty of attention to the components of screening, but much less attention to the process. I want to know whether these companies are worried about how our compounds are stored? What if my one million compounds are stored this way and not that? Can I get my compounds into their reporter assays?" In short, he asks, "how well will they integrate their systems into the process we've already got set up?"

The problem of integrating new screening technology into the rest of discovery research is hardly limited to the screening start-ups. A number of drug firms have apparently built combinatorial chemistry libraries which won't interface easily with their existing screening systems. Lilly's Haslanger notes that one of the side advantages to his company's takeover of Sphinx Pharmaceutical Corp., acquired in 1994 for its combinatorial chemistry and screening programs [See Deal], was the fact that the two programs had grown up side-by-side and knew how to talk to and design things for each other.

For the start-ups who will depend on revenues from their screening assets, the integration issue is critical. Caliper's Knapp acknowledges the problem: "Yes, the first systems will be clunky because of the interface" with older ways of storing, moving and testing compounds and recording and using data from the screens. "Most customers' systems aren't chip friendly."

The interface problem is one big reason few screening companies envision, at least in the short term, a business in actually selling HTS systems broadly throughout the industry. Indeed, several companies—Caliper, for example—simply maintain that drug firms will probably have to adapt some of their screening programs to work with the start-ups' systems. Aurora probably has done the most work towards making a smooth interface: it has created an on-line storage system for a 1.5 million compound "working library" and since the system's informatics will be accessible over corporate intranets, a researcher will be able to see the results of his screening and order new screens from his desktop. Still, as sophisticated and smooth as Aurora is making the system, the company will outfit a maximum of a half-dozen companies with its screening system; the rest will likely buy screening services at Aurora's central facility.

Are We There Yet, Daddy

But even granted the interface problems are resolved, the timeline for delivering on the promises of these systems isn't clear. When will any of them work?

Karsten Henco, PhD, Evotec's CEO and CSO, believes his company's system is the closest to a true working miniaturization-based HTS system. The German firm will deliver a single-channel (one assay at a time) system to its partners in the first half of 1998, he says, enabling them to screen 100,000 compounds per day. In contrast, he says, "if you read Aurora's prospectus, they're talking about delivering a system in 1999—and that could mean...whenever."

Aurora's contractual goal is in fact to deliver a system for its consortium partners which will be up and running and screening 100,000 compounds a day by the end of 1999. But Foulkes thinks it can be on line by the end of 1998. Moreover, the system will be worth the wait, Aurora executives contend: when it comes on line it will be able to do a far greater variety of assays than any competitor's system.

Still longer term than either, say critics, will be the chip-based systems capable of testing in picoliter-level assays. "People can only get down to assays at the one microliter level with today's technology," says consultant Crawford, though he adds that Aurora is still mostly running assays between 10-50 microliters. "Their talk is ahead of the reality. But to do picolitertesting," he continues, "companies will need technology that hasn't been invented yet." As for Caliper's idea of decentralized testing: "If they get there in three years, I'll be breathless. If they get there in five years I'll be shocked," says one senior R&D executive.

Caliper's Mike Knapp shrugs off the criticism. "We'll be screening in ‘97," he asserts. Nonetheless, the system used will still be a prototype, screening what Knapp estimates will be about 10,000 compounds a day—not bad but still "orders of magnitude lower than what we're aiming to achieve by 1998." And, of course, Caliper hasn't yet solved the interface issues which remains a significant early hurdle for the company's commercialization efforts.

Lansing Taylor, PhD, CEO of BioDx, is skeptical of most HTS claims. "They're all selling futures," he maintains. But Taylor is hardly an unbiased observer. BioDx is trying to bridge the gap between what's achievable today and what the miniaturization companies are trying to achieve for tomorrow. BioDx is now selling, as its first generation entry into the screening competition, what Taylor calls a high content screening system for target validation and lead optimization.

Already, the primary, low-information screening programs run by many companies are finding plenty of hits—the problem is determining which hits to pursue. Most high-throughput assays give relatively simple yes/no answers—the compound binds to a target or doesn't. They don't distinguish between the quality of hits, which must be tested in largely manual secondary screens, ideally whole-cell screens, to determine a whole host of characteristics, from how tightly they bind to the targets and how quickly they come off of them, to their cytotoxicity, to their ability to be "competed off" their targets by other molecules, to whether they are actually causing any functional change—many hits end up binding to sites on the target which don't elicit any biological response at all.

Such secondary and lead-optimization screening can become a major bottleneck, according to Taylor, as companies focus their screeners' attention on increasing the throughput of their basic systems. But with higher throughput, and given the same basic hit rate of about .01-.02%, scientists will need to sort through the hits themselves on a relatively high-throughput basis.

BioDx is targeting this unmined area: making broadly available, today, an automated system for cell-based assays which can best provide the functional and cytotoxicity data that allow companies to distinguish among candidates with and without potential as drugs. BioDx's ArrayScanuses standard 96-well plates (and will soon introduce a 384-well plate) in a system that can measure, using a multicolor fluorescence reporter system, a variety of information about cellular responses to a compound. The activity is measured over time and cellular space "so you can study cells' responses to compounds in the context of cellular physiology," says Taylor.

BioDx's system, which it is selling now, has been put together as quickly as it has with the help of Carl Zeiss Jena GMBH, which is also the OEM supplier of Evotec's FCSreader. But BioDx is also interested in selling futures, eventually hoping to migrate its live-cell assay system to chips as well as selling what it calls its Cellomics database of cellular functions and responses to candidate drugs, which it's constructing partly from information gleaned from ArrayScan.

Tiny Necessaries

To a certain extent, the question of when the new systems will arrive begs a perhaps more important issue raised by Gordon Ringold: "Is miniaturization critical?"

Precisely, says Richard K. Brown, PhD, VP of marketing and sales for Irori Quantum Microchemistry Inc. Irori's main business is built around selling the synthesis equipment which enables companies to exploit Irori's unique approach to combinatorial chemistry. Most screening of large libraries is done with mixtures of up to several hundred compounds in a single well; should one of the wells register a hit, the problem is figuring out which compound of the mixture is the one that is binding to the target and signaling the hit. For this reason, most combinatorial chemistry companies have avoided making big libraries: the fewer the compounds, the more economically they can be screened in one-compound-per-well arrays—each compound's identity known by its address on the array.

Irori avoids the problem entirely by making each compound identify itself with radio-frequency waves. It makes its compounds on tiny silicon chips, each chip encoded with all the reaction steps used to make the compound—which is in effect the compound's identity because it tells chemists how to make it again. Also on each chip is a tiny copper antenna. Following the screening, a transmitter sends either radio- or microwaves to the chips' antenna and brings back the synthesis information encoded on the chip.

Now Irori is adapting its system to screening, proposing what it calls its "one pot" method. In effect, Irori puts a batch of the target receptor into a single pot, then dumps in its RF-tagged compounds. The compounds in the pot that bind to the target in the pot are identified by RF waves. There's no inter-compound contamination or interference because each of the compounds is separated by its own solid support.

Irori's method is in effect the very opposite of miniaturization, the point of which, notes Brown, is to divide up the batch of expensive target into more and more, tinier and tinier, portions, each of which is put into individual wells in microtiter plates. In contrast, Irori can theoretically use the same total amount of target Aurora would use for the same number of compounds—but completely avoid the problem and expense of having to parcel it out into individual micro-wells. If a company wants to screen 10,000 compounds, all it has to do is to put them into a single pot with enough target to go around—no robotics or micro-pipetting necessary.

Irori isn't by itself going to obsolete the information-rich systems Aurora, Evotec and others are trying to build. In the first place, Irori's system won't work with cell-based assays. Second, it won't screen existing libraries—unless they are somehow modified into Irori's microreactors. Nor will the Irori system deliver functional data: it basically registers hits, although, Brown notes, it is also "the ultimate in a competitive binding assay." Since all compounds in the library are competing to bind to receptors, he notes, "you'll end up with data on the relative binding abilities of a group of molecules."

The Smart Library Threat

Theoretically, Irori's technique, says Brown, will work for any size screening program—the one pot idea is readily adaptable to libraries of millions of compounds, so long as they are made with Irori's technology. But in fact, says Brown, Irori is finding few customers who are interested in making libraries of more than 10,000 compounds.

If what Brown is hearing is representative of a general industry trend, the prospect should be worrying for many of the miniaturization start-ups. Companies began developing their miniaturization technologies because they saw the need to screen large libraries quickly; if libraries are growing smaller and "smarter"—libraries which manage to represent a high degree of molecular diversity with a relative handful of representative compounds—the elaborate automation and miniaturization schemes now under development look unnecessary. Combinatorial chemistry companies like ArQule Inc. , the AlanexCorp. division of Agouron, and CombiChem Inc. are all hard at work trying to prove that bio- and chemoinformatics can make libraries of even just several hundred compounds adequate for coming up not only with hits but with optimized compounds. "If smart libraries work," says Mark Crawford, "this industry will have a huge overcapacity of high throughput screening infrastructure."

Indeed, one reason the HTS system developers like large libraries is that they backstop potential errors often found in miniaturized screening systems: large libraries contain redundant compounds so that if one assay misses a hit it should have registered—perhaps because a compound didn't really dissolve in its microvolume of target solution and therefore never had a chance to bind—its backup compound will work.

Evotec's Karsten Henco also foresees the trend to smaller libraries but doesn't believe that smart libraries mean the end of UHTS. As EVOscreenbecomes more versatile and its menu of assay types grows, its high throughput capacities will be more frequently required for the informationally rich secondary assays for which it is being built. "By the year 2000," Henco says, "combinatorial chemistry libraries will have become much smarter; the need won't be for basic hits—we could have 1000 a day" from screening what Henco believes will be 200-300 targets per year—"but for profiling those 1000 hits on 50-100 parameters," in effect requiring virtually the same total number of assays overall as with programs which screen large libraries against just a few targets.

Yet Evotec doesn't see itself as a screening company: but as a more traditional drug-focused screener. Evotec wants to develop anti-infective drugs based on technology from one of its scientific founders, the Nobel laureate Manfred Eigen, PhD, of the Max Planck Institute for Biophysical Chemistry. Eigen's drug development ideas stem from his understanding of the evolutionary defense mechanisms of viruses and his belief—proven in principle in bacteria—that it is possible to exploit these principles and create drugs that would make viruses lose their infectivity.

Henco continues: "Having an HTS capacity opens up possibilities for us [in terms of screening many viral targets] that pharmaceutical companies don't have." Evotec created a strategy, therefore, to get other companies to pay for EVOscreen's development—as Aurora recoups its systems costs through its consortium of funding partners—but largely so it could use it for its own drug programs, not as the central asset of an Aurora-like screening service business.

In fact, neither Caliper nor Aurora intends to limit itself to simple compound screening. One reason: Most of Aurora's potential customers are developing internal programs which in part or wholly overlap what Aurora wants to do for them. Lilly's Haslanger, for example, notes that rather than pay Aurora to create the assays it needs, Lilly is using Aurora's reporter technology to do its own assay development, worried that its proprietary biology could leak from Aurora to competitors pursuing the same or similar targets.

And Caliper is arguing for two distinct technology strategies. "You can build a better mousetrap—that is, do what other people are doing, but better," says Mike Knapp, "or you can change the process. We want to do both things."

Caliper's "better mousetrap" is its technology used for the more or less traditional purpose of compound screening. But rather than trying to guess how the argument over small and large libraries will come out, Knapp points to the target side: "No one has any clear idea of what to do with 150,000 human genes. There's lots of talk about functional genomics but that's hardly straightforward. We'd like to end-run the validation process."

Caliper's longer-term goal is to use its UHTS system to screen all genes-encoded targets—particularly those of unknown functions—against a reference library of compounds. The result would be a database of structure-activity relationship (SAR) and biological information that would give anyone working on a new target a tremendous headstart both in finding a compound that modulates it as well as a starting point from which to tease out function. Finding a compound that binds to the target should trigger the same function that the target's natural ligand would: a researcher could thus find the function of a new target without having the natural ligand—a process Knapp calls "chemical genetics," or the use of a synthetic chemical for the natural, genetically derived but still undiscovered ligand which turns a target on or off.

Such a system would also theoretically create a tidy way around the problem of mass-patenting of poorly understood but recently discovered targets, notes Larry Bock. Given a headstart with SAR information, "you could do research on the target until its patent issues, by which time you could already have a lead compound and have basically circumvented the original patent."

Caliper isn't the only company thinking in terms of using compound screening as an alternative approach to functional genomics. In its efforts to avoid having to pick a single target on which to base a chemistry and clinical program, Microcide Pharmaceuticals Inc. is screening internal and external libraries simultaneously against a large number of targets, characterized and uncharacterized. Information gleaned from the program—for example, targets which are hit by more than one compound and which therefore may be functionally related—will help the company choose better targets and better compounds for further development.

Gordon Foulkes notes that Aurora has not yet talked publicly about an internal program on target validation using a modification of the company's screening technology. The program "couldproduce large numbers of targets," he says.

But there's plenty of uncertainty about just whether such modified screening approaches can be used for real target validation. Morphosys GMBHis using its Hu-CAL system for creating libraries of antibodies and its SIP screening system to validate targets in a similar high-throughput manner to what Caliper is proposing. "There's lots of bluster," says Morphosys CEO, Simon Moroney, PhD, "about using combinatorial chemistry as a way to characterize targets, but there's no question that the best place to start is with libraries of antibodies, which are far more likely to bind, appropriately."

Aurora, too, is by no means depending on its target validation to bring it either a valuable database business or a large set of proprietary targets. Nonetheless, it does want to share in the proceeds of a drug's dramatic profitability, possible only if it has a proprietary position with targets and compounds. But Foulkes doesn't want to depend on Aurora's embryonic target validation technology. "We don't want to reinvent the wheel. We'll look at alliances [with biotechs] for targets," as it has in such announced deals as those with Allelix Biopharmaceuticals Inc. [See Deal] but, says Foulkes, "we'd like to have libraries of millions of compounds."

Rational Irrationality

The desire to provide more value than can be done by simply screening libraries is ubiquitous in the screening mini-industry. Scriptgen, for example, is trying to move into the target validation game using technology completely unrelated to its Atlasand Scan screening systems, working with technology which allows it to shut down a gene of interest for a short time in order to see the effect of doing so—and thereby tease out the gene's function.

This trend among screening companies—to look ahead to a time when screening alone won't satisfy investors' appetites for growth—is partly an artifact of a central principle in the biotech industry: as Mike Knapp says, the requirement that one invent a strategy at the same time one is developing a technology. HTS as a discipline is so new that, as with many areas of biotech, there are no companies which have succeeded in creating a long-term stable business. Caliper is a perfect example: the company had already signed a deal with Roche giving the Swiss firm an exclusive license to a screening system in a particular configuration and with a specific throughput [See Deal]. In the few months since that deal was signed in the fall of 1996, Caliper decided that screening would be its central activity—and will now have to re-work its relationship with Roche to give itself the flexibility to pursue the business.

Still, exactly what the screening companies will sell, long-term—if not drugs themselves—isn't completely clear. The success of Incyte Pharmaceuticals Inc. has caused a number of them to think in terms of database businesses, though few have concrete plans they were willing to share on the subject. Still there's a clear need in the area. Marty Haslanger calls the informatics issue "the big sleeper": given all the data HTS will generate, current information systems are simply inadequate. "We've just realized this in the last year," admits Haslanger. "The industry is used to having data on compounds you can manipulate with an Excel spreadsheet. The desktop system from my lab won't work anymore." Lilly's Sphinx screening unit recently hired a new information technology chief, Rickesh Kochhar, whose previous job had been putting together data systems for the space shuttle for Hughes Aircraft. "We needed someone who could manipulate terabytes of data—whatever a terabyte is," says Haslanger.

Most of the start-ups are confident they can help. For example, Axiom Biotechnologies Inc. , a San Diego-based start-up, is creating what it calls its Pharma Profile Database, with each module containing data on various natural cell lines and how compounds interact with them, intending to license it out non-exclusively to drug firms.

All of this data, it is hoped, will eventually make screening an even more valuable discovery technology. For the proponents of this new industry, indeed, screening puts together the most crucial drug-discovery intelligence: the activities of synthetic molecules interacting with natural targets. The more data these companies can ferret out from their programs, and the better they get at exploiting it in drug discovery operations, the closer to actual rational drug design will grow this "irrational" science. And as screening and its affiliated technologies metamorphose into design, its value will increase for companies and investors looking for those technologies which add a measure of certainty to the ultimately uncertain drug discovery business.