November 2002
Cover Story
Karen Titus
As the director of molecular pathology at Baylor College of Medicine, Daniel H. Farkas, PhD, HCLD, has his hands full setting up a molecular pathology service at Methodist Hospital—which is part of the Texas Medical Center and which, until now, lacked such a program.
He’s also setting his sights beyond the molecular. In choosing the capital equipment for his new laboratory, Dr. Farkas wants to look at more than just DNA and RNA.
Dr. Farkas plans to look at proteins.
Why not? Proteomics has become the ambition of many a company and research lab. There’s big money to be made, and plenty of diseases in need of targeted therapies, better detection methods, and prognostic markers.
Nonetheless: Is Dr. Farkas crazy?
(Of course he isn’t. But hold that thought for just a minute.)
Some of the biggest names in proteomics entreat caution when it comes to their field. Take Emanuel Petricoin, PhD, co-director of the FDA/NCI Clinical Proteomics Program. Dr. Petricoin, along with fellow co-director and NCI chief of pathology Lance Liotta, MD, PhD, heads what is arguably one of the most successful clinical proteomics endeavors, having identified, in ongoing work, proteomic patterns characteristic of ovarian cancer with 100 percent sensitivity. (See "Catching ovarian cancer with time to spare," July 2002.)
"When you talk about clinical applications of proteomics right now, where the rubber hits the road, there just isn’t that much there," says Dr. Petricoin, who is also senior investigator at FDA’s Center for Biologics Evaluation and Research.
And while there’s no shortage of new proteomics technologies popping up in the marketplace—"There’s a whole cottage industry around protein array technology," Dr. Petricoin observes—most lack the solid, peer-reviewed data that’s needed to assess their worth. "We don’t know how these types of technology truly work, beyond what is claimed in company press releases," Dr. Petricoin says. "It just hasn’t made it past that propagandizing point yet."
Proteomics technology is expensive and not as sensitive as it needs to be, says Michael Shi, MD, senior director of genomics at Sequenom, a genetics discovery company based in San Diego. Work such as that done by Drs. Liotta and Petricoin is mighty impressive, says Dr. Shi. "But it’s not easy."
"Don’t overestimate what proteomics can do at this point," agrees Tim Veenstra, PhD, director of the Biomedical Proteomics Program at NCI-Frederick (Md.).
But—and here’s what pathologists and laboratory directors need to keep in mind—no one in proteomics plans to cry "uncle" anytime soon, and probably not ever. No one, least of all laboratorians, should ignore proteomics.
Now is the time for pathologists and laboratory directors to start thinking, hard, about proteomics, suggests Ruedi Aebersold, PhD, professor and co-founder of the Institute for Systems Biology, Seattle. "It is important to make a mental adjustment away from the idea that we’re going to measure just a particular molecule to the idea that there will be diagnostic patterns—constellations of molecules—that we’ll use for detailed diagnoses," he says.
Proteomics will mean more than "just buying a more sensitive machine that measures the same thing we’ve measured in the past," he continues. It won’t be a matter of moving from an ELISA to mass spectrometry to measure PSA. "The issue is that a completely different type of analysis or result will emerge, which is probably much harder to grasp," he says. "It will be harder to differentiate between ’yes’ and ’no,’ because the answers won’t be black and white." Get ready, in other words, to interpret—and explain to clinicians—many, many, many shades of gray.
Rest assured, you’ve got plenty of time to prepare for this eventuality. Frankly, even getting people to agree on a standard definition of "proteomics" is a challenge, and the relative newness of the field means people are bandying about words like "proteomically" and "bioinformatician," then asking, "Is that even a word?"
"’Proteomics’ is a broad statement," says Tim Fleming, president and general manager of Vivascience AG, a subsidiary of Sartorius AG and a maker of, among other things, protein purification kits. "It’s a buzzword now, and a lot of companies use it for almost anything, just to get people’s attention."
Likewise, any number of researchers are jumping on the proteomics bandwagon without rhyme or reason. "Too many times, too many investigators have totally altered their research plan because of the ’wonders of proteomics,’" says Dr. Veenstra, whose lab is involved in two major initiatives: the ovarian cancer studies spearheaded by Drs. Liotta and Petricoin, and assessing protein expression differences between various cell types and between similar cells treated with different drugs.
There are wonders, to be sure. There are also wonders of a different sort, chiefly, how to move proteomics into clinical practice. Most agree it will require making a molehill out of a mountain. How do you go about finding, characterizing, measuring, and monitoring the most important targets out of the god-knows-how-many proteins that fill a human being?
It is perhaps the ultimate wheat-from-the-chaff exercise. "I find, as somebody who runs a mass spectrometry facility, that people come to me and expect, essentially, miracles," says Dr. Veenstra. "They want me to take their cell lines, tell them every protein that’s in it, and then compare it to another cell line and tell me which proteins are not in that cell line." It’s impossible to know, he says. "I don’t know if it’s a case of it’s not there, or I just didn’t detect it. You’re looking at, potentially, a hundred thousand species and trying to identify 2,000 of them."
So why bother?
"Proteins are the business end of the cell," says Dr. Petricoin.
"It’s the protein that does the dirty work," agrees Dr. Farkas, who is also president-elect of the Association for Molecular Pathology. "It’s becoming clear that protein actions and levels and how they interact are highly diagnostic of what’s going on in a patient’s clinical and subclinical disease."
The trick will be to identify the right suite of protein markers, he says. "The genetic level is a clue, and sometimes a darn good clue, but it is, after all, just a genotype with potentially limited penetrance. The protein, in fact, causes the phenotype." That distinction, he says, is the future of clinical proteomics.
That’s why Dr. Farkas is contemplating proteomics even as he sets up his molecular pathology service at Methodist Hospital. "It’s time for pathologists to start thinking about proteomics," he says.
Lucky for him, many others are not only thinking about proteomics, but are doing the heavy lifting needed to bring proteomics into everyday use. What they’re encountering are heartening successes that march lockstep with daunting challenges.
To get a sense of the difficulty of proteomics, compare it to genomics.
"Genetics is straightforward," says Dr. Shi. "You have a genome, you have the gene, it’s embedded in every cell, so whatever you want to do, you know where you are. You can trust that you have a human genome map. It’s easier to navigate."
Adds Dr. Veenstra: "I don’t want to say it was easy to do the human genome, but you’re really dealing with a single, large-molecule DNA, which is not that heterogeneous. You basically have four bases to deal with, and you’re interested in doing one thing—you want to sequence it." Proteomics, on the other hand, "is a whole new nut to crack. You need to get the entire amino acid sequence of the protein as well as all the associated modifications." In gene expression transfertomics, there may be only two to three log orders of difference—that is, a hundred- to a thousandfold—between the most-expressed gene product and the least-expressed gene product. In proteomics, the difference could be as much as a billionfold.
That huge dynamic range makes it tough for researchers, who end up seeing only a small percentage of the proteome, says Dr. Petricoin. High-abundance proteins may be apparent on a 2D electrophoresis gel using various stains, but low-abundance proteins are elusive. "You’re only getting the low-hanging fruit. The low-abundance proteins you never even see. But the way nature sets this up, it’s usually the low-abundance proteins that are the more interesting drug targets."
Proteomics lacks a direct amplifying technology; there are no PCR-like methods for proteins. "You can’t amplify the low-abundance proteins even if you wanted to," says Dr. Petricoin. "So sensitivity is a huge issue in proteomics."
There are sheer numbers to consider, too. While the genome may contain only 30,000 to 50,000 genes, there might be millions of different proteins or different isoforms for modified proteins. One gene product could have a protein that is clipped eight different ways. Each of those eight clippings are localized to different sites on the cell and perhaps modified five different ways by glycosylation. Those glycosylations, in turn, may be phosphorylated two to three different ways. "So one gene form could have hundreds of protein isoforms, each of which could be important biologically," Dr. Petricoin says. "It’s mind-boggling, the complexity of proteomics."
Posttranslational modification—especially phosphorylation, on which cell signaling events are based—adds to the muddle. In many cases, says Dr. Veenstra, two different cell types may exhibit similar protein expression levels, but the protein in one will be phosphorylated. "That changes activity 180 degrees from the same protein in the other cell type," he explains. "And we have to be able to try to monitor that."
It can be done with mass spectrometry. "If we noticed that a protein’s mass was 80 daltons higher than we predicted it to be, that would tell us it’s phosphorylated, because phosphorylation has a specific mass associated with it," Dr. Veenstra says. "But I’m making it sound a lot easier than it actually is. Monitoring subtle variations in phosphorylation is still very difficult to do."
The real need, says Dr. Aebersold, is to measure overall changes in a protein’s phosphorylation profile, because rarely does phosphorylation consist of only one event. "And right now there is no technology that can measure that," he says.
Another limitation, says Dr. Aebersold, is that many samples used for proteomics, particularly blood serum and cerebrospinal fluid, are extremely difficult to work with. While they may be more accessible than tissue, these samples’ protein mixtures are dominated by the presence of a few, highly expressed proteins. The serum proteome, for example, is marked by an overwhelming presence of albumin protein, which effectively masks the presence of other, possibly important, proteins. Any analytical technique for looking at large numbers of proteins "has to deal with this rather difficult distribution," he says. The trick is to selectively remove these proteins. (This is harder than that last sentence would have you believe. "It is difficult to completely remove a protein," says Dr. Aebersold, a man who knows from difficult. Nor can researchers tell, with certainty, what else is being removed along with the targeted proteins.)
Jim Wittliff, PhD, MD, professor of biochemistry and molecular biology, James Graham Brown Cancer Center, University of Louisville, Ky., notes some other not-so-simple problems that widen the already Grand Canyon-sized gulf between proteomics research and clinical applications.
Proteomics "requires skilled folks to perform technically complex assays. This is not the kind of thing where you write a set of instructions and prepare some reagents. We’ve got to simplify this," says Dr. Wittliff, who developed CAP’s proficiency Survey for estrogen and progestin receptors.
Then there’s cost. "These analyses aren’t cheap right now," Dr. Wittliff acknowledges.
Just as important is teaching physicians how to interpret proteomics data. Dr. Wittliff would like to see gene expression profiling and proteomic expression profiling made part of studies by cooperative clinical trial groups. "It’s not going as quickly, in my opinion, as it should."
Dr. Petricoin cites one final challenge, one that may be the biggest hurdle of all. "Proteomics, unlike genomics, is completely context-dependent," he says. "We can’t do human proteome project, because there is no such thing as the human proteome. Every cell type in your body and in my body has a different proteome, which changes due to environmental impacts, time of day, hormonal changes, aging. So when you talk about clinical applications, what you really need to have are technologies that can take a few hundred cells from a patient biopsy and analyze quantitatively the many, many proteins in that cell simultaneously. And that technology doesn’t exist."
Plenty of other proteomics technologies do exist, but it’s anyone’s guess whether they’ll migrate from research to everyday use. Some platforms are already available for clinical labs, but "it’s akin to buying a computer with no software," says Dr. Farkas. And until the boundaries of proteomics are better defined—until diagnostically relevant proteins are more clearly identified—no one will know the best way to perform proteomic analysis on a day-to-day basis.
Each of the current proteomics technologies have different strike zones, as Dr. Petricoin puts it. "So right now it’s not a matter of committing to any one technology platform, but rather using many at the same time."
Many such platforms should be adaptable to the clinical laboratory once it’s clear which proteins are of interest. In fact, many will need to be adapted, suggests David Hicks, director of market and business planning for the discovery proteomics and small molecule business of Applied Biosystems, based in Foster City, Calif. "There’s not going to be any single, one-stop shop for all things proteomics," he says.
Naturally, many hope to borrow a page or two from genomics technology. There’s even talk of putting proteins on biochips. "And rightfully so," says Dr. Veenstra. Such chips could be used to look at specific protein pathways repeatedly, in rapid fashion.
Among the current methods, "The protein array, I think, will gradually become part of clinical laboratory practice. But we are talking about at least three to five years down the road," says Haifeng M. Wu, MD, assistant professor of pathology and director of the clinical coagulation laboratory at Ohio State University. "We need to identify those proteins first and then make protein or antibody arrays."
At OSU, Dr. Wu and his colleagues are using 2D electrophoresis to identify proteins and protein patterns related to the pathogenesis, clinical development, and clinical progress of acute myocardial infarction. Within two years, he says, they hope not only to have pinpointed key proteins/protein patterns, but to have transformed current methods into more user-friendly technologies. "And then, with these tests, we’d like to extend our study from acute MI to other diseases, such as autoimmune diseases."
"We’re trying to be pioneers," he adds, cheerfully acknowledging the magnitude of these tasks. "I’ve got up to 3,000 proteins on each 2D gel, and it’s difficult to analyze the data since there’s no established method at this time."
The Institute for Systems Biology is in the thick of developing quantitative methods for measuring proteins. Its base technology combines chemistry, mass spectrometry, and computer informatics; the general strategy, says Dr. Aebersold, involves using selected chemical reactions to introduce stable isotope-coded affinity tags at specific sites in proteins or protein subclasses. These tags represent signatures that can be read quantitatively by a mass spectrometer, then analyzed and interpreted via computer.
As the technology is refined, says Dr. Aebersold, "we are working toward getting better ’depth perception,’ to see deeper down into the low-abundance species of the proteome." Other adaptations of the platform permit researchers to determine the composition and changes in composition of proteins, protein-protein interactions, how proteins direct themselves as well as DNA, and changes in phosphorylation. Though these technologies are not quite ready for clinical use, they’re close, Dr. Aebersold contends.
When Dr. Veenstra and his colleagues at NCI generate protein patterns to be used as diagnostics, they now use the entire pattern. "But it’s really a few key areas within the pattern that are diagnostic. Ultimately, we’d like to be able to develop simpler assays that focus just on those differences within the pattern, instead of having to judge the whole thing."
Currently they use surface-enhanced laser desorption ionization/time of flight, or SELDI-TOF, a mass spectrometry technique that uses protein chips containing chemical surfaces that bind to particular protein classes. The beauty of this approach is that it allows researchers to work with "very, very crude samples," Dr. Veenstra says. "We take raw serum and apply 5 microliters onto this chip surface, wash it a couple times, put it in the mass spec, and the SELDI-TOF basically reads the pattern."
The pattern is not pretty. "It’s not a nice pattern, like a circle or series of figure eights," Dr. Veenstra says. "You can’t look at it with your eye and say, ’Yes, this person has ovarian cancer.’ We have to use it in combination with that computer software, which reads the patterns and measures them and compares them to previous patterns we’ve measured. The patterns are that complex."
To identify individual proteins, he uses a higher-end mass spectrometer, with higher mass accuracy and higher resolution. It also has tandem mass spectrometry capability; tandem MS permits researchers to fragment peptides, thus determining their sequence—and the identity of a protein.
Often lost in the shuffle is proeomics’ predecessor—genomics. "There’s a lot of jealousy and anxiety in the genomics field about being left behind," notes Dr. Petricoin.
Don’t bet on it. Proteomics couldn’t have gotten its start without genomics, and it’s unlikely to supplant it. Right now, no one actually knows what happens between the gene level and the protein level. "A lot of people use complex bioinformatic acrobatics to look at gene arrays to see if they can predict what’s happening at the protein level," says Dr. Petricoin. What they’re finding is no or very little correlation between gene expression analysis and protein expression analysis, and even less of a connection between gene expression analysis and cell signaling events, which occur before the genome can sense them. Therefore, proteomics will not, by itself, revolutionize molecular medicine, he suggests. It will add tremendously to the study, analysis, and treatment of disease—but only in conjunction with genomics. Using either one alone means physicians will be missing, and misinterpreting, crucial information.
Separating genomics and proteomics, biologically or technologically, is an artificial boundary at best, says Hicks. "Certainly many of our customers would like to combine gene expression and protein expression work. The bioinformatics folks we’re talking to are working with both sets of data, and with other data sets as well."
Indeed, says Dr. Wittliff, physicians will need to incorporate yet another layer of information—the metabolome, or metabolic state, as determined by functionally active proteins and their impact on energy-containing molecules, substrates, signal molecules, and so on. "This is the greater and perhaps more clinically revelant aspect of the picture," he says.
Dr. Wittliff, in fact, can barely contain his excitement when he talks about proteomics. Already the language and preliminary technologies of proteomics and genomics are creeping into everyday practice, he notes. "We’re teaching the medical students the very principles of gene expression profiles. They’re learning how to read 2D gel profiles. They’re learning what mass spectrometry is, what MALDI-TOF is." At Louisville, he and his colleagues are working with transplantation immunologists to figure out how proteomics information might help determine better tissue matches.
"This is moving quickly," he says.
It is and it isn’t. While the word "proteomics" has gathered steam in recent years, the phrase "clinical proteomics" is chugging behind a bit more slowly. "It’s almost oxymoronic," says Dr. Farkas. "I’m certainly not practicing, and anyone who is is in a translational research mode."
And yet that hasn’t kept him from seeking out platforms flexible enough to allow him to do proteomics work down the road. "This is the next big thing," he says. "Of course, the next big thing could be two, five, or 10 or more years away."
Karen Titus is CAP TODAY contributing editor and co-managing editor.