關(guān)鍵詞： 蛋白質(zhì)芯片
The next generation of protein microarrays from the likes of Protometrix and Molecular Staging may threaten the early leads of Biacore and Ciphergen — and work so well that drug companies won't want them.
BY MARK D. UEHLING
Once upon a time, on a planet similar to ours, a promising anti-inflammatory drug was in trouble. Efficacy was fine. But the company developing the drug still worried: Could it have unwanted effects on the immune system? Animal studies could provide only indirect insight. Clinical trials could take forever.
Molecular Staging got a call. In a matter of months, for less than $1 million, the company deposited blood samples from dosed patients on one of its protein chips — a microarray loaded with immunological proteins. Then the company watched if the drug affected the levels of the proteins. It didn't.
Hardest-working robot in pharma: Stephen Kingsmore in a temperature-controlled room where the company's Proteodyne Biocube Robot prepares chips.
"We surveyed the immunological universe. Even at the highest dose, we didn't see any off-target effect on the immune system," says Stephen Kingsmore, chief operating officer of Molecular Staging, who acknowledges that the company's contract work for pharma and biotech companies does not always yield such unambiguous answers. And so development of that particular drug proceeded.
Kingsmore says his company has worked with half a dozen major companies, but only one has authorized a news release about a project. Molecular Staging helped Eli Lilly find unpublished biomarkers for sepsis — proteins that would definitively indicate a deadly hospital infection that physicians have difficulty diagnosing. "We have candidate markers in hand," Kingsmore says. "Now it's a matter of looking at many more patients and saying, 'Are these generalizable? Are they going to be useful?'"
Molecular Staging's relationship with Lilly is evolving and could include taking the existing work to another level: "How do we turn this information into something that would affect medical practice?" Depending on the answer, Kingsmore says, the companies might develop a diagnostic test or new therapeutics.
With encouraging results like that, protein chips are poised to make the transition from mythic to everyday, from isolated salvage projects for drugs in trouble to wider rollouts throughout drug development. "It's a very robust technology," says George Grills, director of the new Protein Microarray Facility at the Harvard Partners Center for Genetics and Genomics. "I don't normally need to sell the technology. People want it."
New Experiments
On one level, a protein chip is just dozens, hundreds, or thousands of ELISAs (enzyme-linked immunosorbent assays) run simultaneously. But the multivariate data — and the mere hours required to read the chips — mean that a protein chip is not just an ELISA on steroids. Says Grills: "It is really a new paradigm that allows you to think of new types of experiments that you could never have done before."
As part of a three-year deal, Grills is handling contract work using the Zyomyx chip for scientists inside the university and in industry. "You can really do high-throughput profiling and marker discovery using this system," he says. "It's really enabling you to look at a whole panel of these important biomarkers all at once using a limited amount of sample."
"Demand for these systems is very high," agrees Steven Bodovitz, principal consultant at Select Biosciences, who estimates 36-percent annual growth in the protein chip market through 2007. But even Bodovitz concedes that the protein chip field has been dogged by vapor chips — promised arrays that have never quite materialized. "The big industry bugaboo is the lack of product launches," he says. (Bodovitz's 177-page report about the protein array industry is available for purchase at www.selectbiosciences.com.)
Published in April, Bodovitz's exhaustive report is already a mite obsolete, with a few companies either out of business or radically reducing their protein chip effort. Companies like Prolinx are gone; Large Scale Biology has suspended chip work. The field remains in flux, with basic specifications and metrics unclear. What's more, basic terms like "capture" chips and "interaction" chips can be confoundingly overlapping. In protein interaction chips, proteins or peptides help scientists localize other proteins, illuminating their binding with other proteins, lipids, or small molecules. Protein capture chips, on the other hand, generally rely on antibodies or antibody mimics to detect the presence or quantity of a protein. Getting data from either type of array in a reliable, reproducible manner requires deep biochemical and lab-automation expertise.
Early enthusiasm, in short, has hardened into a deeper appreciation of the immense challenge of working with proteins in quantity. "It's more daunting than people realized," Bodovitz says. "What these companies are proposing is not what you would think if you read Biochemistry 101."
Reinterpreting the Protein Gospel
The most important piece of biochemical dogma that may be rewritten is the idea that proteins will be denatured if attached to a sliver of glass or plastic. The leading proponent of a revised view is Yale University's Michael Snyder, a biochemist with a storeroom filled with boxes of reprints from prestigious science journals.
Protein king: Yale's Michael Snyder has founded a company, Protometrix, with the mission of detecting thousands of proteins at once.
Snyder and his postdoc wunderkinds say they've dispatched several key challenges. They can store proteins in their native form. They don't need to generate antibodies that theoretically bind to one and only one protein — but which in fact are far more promiscuous. Most important, Snyder's lab can attach proteins to a substrate without altering their exquisite 3-D shapes — and can validate what they have detected.
Heng Zhu, a postdoc in the Snyder lab, describes a recent challenge from a nearby Ivy League lab. In basic terms, the question was: With which proteins do these two drugs fit?
Zhu recalls: "They saw our [July 2001 Science] paper, and said, 'Wow, why don't we send you the drug, toss it on your chip, and find the target?' I did the experiment in two days and generated a short list for them." Zhu had sliced thousands of possible proteins down to 14 for one drug and 20 for the other. The original suppliers have since winnowed those down to a single protein for each drug.
That project used a chip with thousands of yeast proteins (virtually the entire proteome of that organism) developed by Zhu. Two-thirds of the yeast proteins have rough counterparts in humans, and 50 percent of those proteins (or one-third of the total) have very similar copies in man. The chip is now being commercialized by Snyder's new company, Protometrix, which Snyder launched in part to satisfy demand — and in the hope of becoming the Affymetrix of protein arrays. "You can take these small molecules and figure out what they bind to," Snyder says. "If you have any lead compounds, it strikes me as obvious that you'll want to put them on one of the protein arrays and look at its reactivity profile." Or see which proteins aside from its designated biological target the drug may bind with.
The chips are up: Standard genomic lab equipment can read this slide.
If it works — still a Jupiter-sized if, in the absence of peer-reviewed data — such a chip could offer early, reliable insights into the most tantalizing unknown in pharmaceutical research. Aside from its intended destination, what other cells, tissues, or organs will a drug hit? That will become even more efficient when Protometrix releases a series of chips with desirable classes of clinically relevant human proteins such as kinases.
The company hints this could happen later this year or early next. "I see this as revolutionizing drug discovery," says Hollis Kleinert, president and CEO of Protometrix. "This will become a common practice in all types of research labs in pharmaceutical discovery and to some extent in development as well."
Kleinert concedes that animal studies and clinical trials will not be supplanted but nevertheless believes protein chips will "be used to study efficacy and safety because you can look at specificity and selectivity of small molecules. You can do the dose-response [curves] right on the chip. It's going to make drug discovery more efficient; it's going to make it smarter. It's going to take some of the guessing games away."
The speed is impressive, usually requiring just "one morning or one afternoon to get an answer," Kleinert says. And she points to even more tantalizing possibilities ahead: "You'll be able to compare animal species that are used in drug efficacy and safety studies with human data" — something impossible today.
Given the impressive nature of Snyder's published results, it's fair to ask whether unusual meteorological circumstances in New Haven might not be easily duplicated elsewhere. Kleinert concedes the early beta testers of the Protometrix chips are skeptical but then come around. "They can't believe how simple and valuable [the chips] are. They're trying them and saying, 'Gosh, this is something we can do and can incorporate.' We expect the adoption is going to be quite favorable. This is the kind of thing that can really impact that 99-percent failure rate in drug companies."
Older Chips Still Tasty
The current euphoria over protein chips strikes the scientists at Swedish firm Biacore as a bit odd. After all, the company has been making protein chips for more than a decade and published more than 2,500 scientific articles. One of the signature advantages of the company's platform, says Clive Seymour, vice president and head of life science research, is its label-free approach. "We're not putting any radioactive labels, fluorescent tags, etc.," he says. "That does not happen in a Biacore assay."
Liquid gold: Protein-oriented microfluidics from Biacore.
As Seymour discusses a schematically oversimplified sensorgram — a graph of a protein's activity — it's clear that the chips are providing glimpses into the secret lives of proteins, not merely detecting their presence. The sensorgram curve depicts a simplified view of what happens between proteins in their natural, cellular milieu: binding, saturation, and separation. "You can look at what the characters of that binding are: kinetics, specificity, affinity, concentration," Seymour says. "There are a lot of data."
As scientists grow accustomed to the curves, they learn what works well in the body. "If that up curve is very, very steep," Seymour says, "that means the interaction is very fast. That means the two molecules clearly like each other a lot, and they get together in a big way."
Some drug companies may shun molecules that produce protein curves that don't subside in a timely way, worrying that the protein would persist in the body and cause unwanted effects. But the opposite may also be true: If it doesn't have a sufficiently strong affinity to another molecule, it may not last long enough to have any effect.
Traditional protein-detection technologies are more static, Biacore suggests. "You're looking at these molecules in a completely different manner," Seymour says. "You're confirming their having an effect, but you're getting down to just what that effect is in terms of its kinetics and its affinity. This is unique. There are very few systems out there on the market that will get close to this. You are essentially building up a profile of each individual molecule."
Seymour ridicules traditional mass spectrometry, describing the process in ways that sound like small children smashing things. "This is working with a live protein, or a live peptide, or a live molecule," he says. "We would be able to look at the protein in a much more detailed manner, much earlier on, and start making real decisions about whether this thing is what we think it is."
Some Biacore machines come with a data analysis suite. "It allows you to look at the sensorgrams, to fit them, to get the kinetic constants, the affinity constants," Seymour says. One machine even has software built in. "The control and the evaluation is pretty much all together in a single package. That allows the user to work in a more high-throughput mode."
Services Model
Of course, the field of protein chips is filled with companies, some making only one piece of the puzzle. In Colorado, SomaLogic Protein Chips: Time to Jump In?
Gavin MacBeath is a professor in Harvard's Department of Chemistry and Chemical Biology.
Read More
has endured a few delays along the way. "We've struggled, like everybody else," chairman Larry Gold says. "A lot of people who said they were coming out with a chip at time T didn't. Some of the delays are going to be forever." (For a rundown on a few companies, see "Protein Chips: Time to Jump In?", right.)
SomaLogic will use a service model, and hopes to be accepting blood samples from customers by October. Gold thinks pharmaceutical customers will want data before they invest heavily in capital equipment for protein chips. "The best way to get them data is to do it for them," he says. "This is more about seeding the market. Before they build a group like they did to service 2-D gels, or a service to do mass spec, before they invest in all the hardware, they're going to want data."
The key selling point of SomaLogic's array will be both the limits of detection and the quantity of proteins it can find. "Mass spec, for all of its great qualities, is not very different than 2-D gels were in 1975," Gold says, launching into what is almost a comedy routine about the limitations of the current technology. "You see a thousand things; you see them at around one nanomolar [concentration]; it's expensive; it's hard to quantify; you don't know for sure what you're looking at until you do a lot more work. There is no hope for 2-D gels getting very much better than they are today."
Like a few protein chip efforts, SomaLogic has given up the use of antibodies to help anchor proteins to be measured. True, antibodies can be designed to fit snugly together with proteins, and successful companies like Phylos are developing libraries of them.
But the problems with antibodies are vast. Instead, Gold is using aptamers — short stretches of nucleotides that can lock onto targeted molecules just like antibodies and proteins. But Gold believes aptamers will generate fewer nonspecific binding interactions. "The real difference is the stuff will scale to high feature density," Gold says.
Ciphergen's Definition of a Chip
If a company calls itself "e;The ProteinChip Company"e; and trademarks the phrase "e;ProteinChip,"e; one would predict it is in the business of selling protein chips.
Read More
Those are fighting words. As Gold and others have discovered, some protein chips today can detect only 30 to 40 proteins, chiefly for reasons of interactions between the chemicals being detected and doing the detecting. For that reason, free-floating bead systems to detect proteins are an entirely different proteomic strategy, and one being developed by leading vendors including Ciphergen (see "Ciphergen's Definition of a Chip," right). Some companies, such as Protometrix, predict they will be able to detect thousands of proteins per chip.
In some cases, these chips can be read by the same equipment used to read DNA microarrays. Likewise, some of the robots used to daub nucleic acids onto chips can be easily adapted to protein applications, as can the software.
Coming to an ER Near You
Indeed, the protein chip — despite the caveats and doubts of some industry leaders — has already made the leap into clinical practice. Biosite, in California, makes simple, inexpensive protein tests that are widely used in hospital emergency rooms.
Ken Buechler, senior vice president of R&D, says the company has tried to interest the pharmaceutical world in partnerships, lining up deals at Amgen and Lilly. But Buechler has more latitude to discuss projects with Johns Hopkins University, Brigham and Women's Hospital, and Duke University. "With those," he says of disease-related blood samples, "we're mining them for proteins we think will be important."
Good lines: Curves like this, from Biacore, helped AstraZeneca prioritize an anticlotting drug, Melagatran. Under a variety of conditions, Biacore's graphs show how quickly drugs bind with particular proteins — and then split apart from them. That allows scientists to rank prospective new drugs by predicting how they will interact with their molecular targets over time.
Biosite is best known for a test for drugs of abuse, but the company can detect other proteins, allowing doctors to differentiate between different causes of chest pain, stroke, sepsis, and shortness of breath. The company funds much of its own research into biomarkers — definitive indicators of the presence or absence of disease.
The Biosite chips operate passively, sorting molecules through capillary channels. "The key here is to get a result immediately or within 15 minutes, so we can define acute disease," Buechler says. In the case of ischemic stroke, a drug such as TPA (tissue plasminogen activator) can be administered after a test.
Curiously, Buechler suggests that drug companies are really not interested in what Biosite can do. "The [partnering] activity has decreased a little bit" in recent years, he says. "Are [Biosite chips] actually being used now, to any great extent? The answer is no. We have talked to drug companies and had lukewarm responses from them. At a lot of these companies, you find people who are really interested — then somehow it ends. Quickly. Why that is would be a very interesting thing to find out."
Buechler declines further comment, but other observers say that a drug tightly linked with a diagnostic test is something most drug companies energetically avoid. Aside from increasing cost and complicating the prescription process for a new drug, a diagnostic test that effectively excludes a certain percentage of patients could decimate a drug's market share.
So is protein chip-assisted personalized medicine impossible? No. In the years ahead, as more studies demonstrate what these chips can do, the chips may ultimately get their biggest boost from federal regulators or insurance companies. Both may look to save money or increase the quality of care by asking the chips to sort out whether a particular drug will work in a particular patient. And that, in time, could finally deliver on some of the hopes and promises of pharmacogenomics.
