Systems Biology's Clinical Future
Imagine going to a doctor's office. your complete genome sequence, which provides a probabilistic prediction of your future health, is part of your medical file. To see how that genetic component is playing out and to obtain a snapshot of your current health status, your doctor orders a standard test of more than 2,000 proteins and metabolites. According to the results of those tests, your doctor recommends ways for you to maintain or improve your health through either medication or behavior modification.
Sounds futuristic? Perhaps, but it's not as far off as it seems, and systems biology will help make it a reality. Such a personalized approach to medicine is only one of the clinical applications of systems biology on the horizon.
"We've struggled for eons to figure out how to handle biological complexity," says H. Steven Wiley, director of the biomolecular systems initiative at Pacific Northwest National Laboratory (PNNL). Biological research has traditionally taken a one-at-a-time approach to studying genes and proteins, the so-called reductionist approach. Now, tools such as DNA and protein microarrays and mass spectrometry have made it possible to study many components and clarify how they work together to regulate and carry out biological processes. The goal of systems biology is to combine molecular information of various types in models that describe and predict function at the cellular, tissue, organ, and even whole-organism levels.
Leroy E. Hood, cofounder and president of the Institute for Systems Biology, Seattle, cautions against defining systems biology simply in opposition to reductionist approaches. "We want to understand how the individual biological networks work and how they work together with other networks within the cell or organism. A lot of the fundamental experiments that we do are reductionistic," he says. For instance, a typical systems biology experiment might involve making a single genetic perturbation and seeing how it affects everything else.
Some diseases will be effectively treated only with a systems approach. "Obesity, diabetes, many heart diseases, and hypertension are system problems, and the solution will have to be a system solution," says Joseph H. Nadeau, chairman of the genetics department at Case Western Reserve University School of Medicine. "Look at all the things that are problematic, compromised, or dysfunctional in an obese individual. To think that we can go into a laboratory and synthesize some drug that will fix all the problems and not cause any adverse side effects, the premise is goofy."
Even some of the simpler diseases caused by a single gene defect will probably require a systems approach to truly understand them, according to Nadeau. "Cystic fibrosis was the prototypic single-gene disorder, yet other genes in the genome influence the age of onset, the severity, and the clinical associations. The biology of individuals with cystic fibrosis is telling us that it may be a simple genetic cause, but there are many systems properties that confound it." The same is true for sickle cell anemia. "Even in the simplest cases, it's not as simple as we've told ourselves it is," he adds.
So far, systems biology approaches have served as research tools, but biologists, physicians, and pharmaceutical companies have high hopes for clinical applications.
"We're still in the early days of systems biology for diagnosing and treating disease," Nadeau says. He anticipates tremendous advances in the next five to 10 years in treating disorders for which we already know some of the essential circuitry and have rudimentary models, such as diabetes and other metabolic diseases.
When it comes to clinical applications, systems biology is still "fairly immature," says Jeffrey R. Balser, a professor of anesthesiology and pharmacology and associate vice chancellor of research at Vanderbilt University Medical Center. "We don't have the infrastructure in place to deliver systems information to the point of care," he says. "There is great promise, but we haven't even begun to realize the potential of systems biology in health care."
Systems biology will bring unprecedented amounts of molecular information and large-scale data sets to medicine in the form of DNA sequences and levels of messenger RNA, proteins, and metabolites. An important part of systems biology is taking all of those measurements and using them to construct models of what's going on in the cell, tissue, organ, or organism. In the pharmaceutical industry, some companies are already using such models to help discover, develop, and test drugs.
Cambridge, Mass.-based Genstruct combines data from a variety of sources to create cause-and-effect models of disease. "We deal with the complexity of biology at the level of complexity," rather than trying to simplify the system, says Keith O. Elliston, president and chief executive officer.
Genstruct's models include all the genes, proteins, and metabolites present in human cells. They then apply artificial intelligence tools to work through all the predicted and observed relationships among these components and put them in context within the complex system. The equations within the models are not mathematical; instead, they lay out cause-and-effect relationships in much the same way as chemical equations do for reactions.
"The outcome is an understanding of how things change in disease or with drug treatment at the molecular level," Elliston says. That understanding can then be used to identify how disease and drugs work and to design better clinical trials.
Genstruct has compared animal models of human disease with the human diseases themselves. For instance, they have studied mouse and rat models that are used to approximate type 2 diabetes in humans. "In many cases, we found that, other than having aberrations in insulin-signaling and glucose levels, there was no similarity between the animal model and the human disease condition," Elliston says. "If you're developing a drug in that animal model, it's clearly not going to work in humans because they have a different disease."
Entelos, in Foster City, Calif., uses mathematical modeling with differential equations to construct mechanistic disease models known as "virtual patients," according to Alex L. Bangs, the firm's chief technology officer. The goal is to identify the patients who are best suited to particular drugs and then find markers that could help differentiate those patients. There can be many virtual patients for a given disease, each one reflecting a different underlying biology of the disease. For example, patients can have mild or severe diabetes and can be lean or obese. The company then weights these different virtual patients to build a "virtual population" that mirrors the clinical trial population.
Entelos is focusing on pharmaceutical applications, but Bangs says it should be feasible to adapt virtual patients for use by physicians to guide treatment of individual patients.
So-called personalized medicine will be the ultimate clinical application of systems biology, and the first step toward that goal will be to use systems biology to stratify patient populations according to their responses to drugs. For example, some drugs work only in a fraction of the population, but for that fraction they are completely effective. Other subpopulations may have no response whatsoever. "Right now, we don't know who's who," Balser says. "The dream is to know that ahead of time and target therapies only to those who will be helped."
The same is true for most adverse drug effects. They occur in only a small percentage of patients, but as the Vioxx case has emphasized, a small percentage of patients can turn into a large number of people if a large enough patient population takes a drug. "If we knew prospectively who would be likely to have some kind of complication from Vioxx, all the millions of patients who are actually helped by that drug could still have it," Balser says.
Entelos is using its models to stratify patient populations. "We're trying to understand what patients might respond best to a particular therapy and then identifying biomarkers that could help differentiate these patients," Bangs says.
Elliston believes that understanding the underlying cause and mechanism of each patient's disease will allow clinical trials to become smaller and more focused. The trials can be based on cause and effect rather than correlations. "You simply can't do clinical trials big enough when you have personalized therapies to get statistically significant results unless you understand the underlying cause and can select patients based on that cause," he says.
Merrimack Pharmaceuticals, Cambridge, Mass., plans to take such an approach to the clinical testing of its protein drugs. In cancer trials, the company plans to screen each patient to determine the molecular signature of their tumor. Merrimack's stratification of types of cancer, based on the signaling networks that drive tumor growth, turns out to be mostly independent of the location of the tumor. Those patients whose cancer is best suited to one of Merrimack's signaling-targeted protein drugs would then be included in the clinical trials.
"From a clinical perspective, you'd be working with much smaller trials, very highly enriched in the patients who are preselected for the right molecular background for the drug," says Robert J. Mulroy, the company's president and CEO.
Such an approach might seem to rule out too many patients to be a financially viable strategy, but Mulroy doesn't think so. "Our stratification of cancer has not reduced the target population for each of our drugs but has actually increased it," he says. "Our analysis reduces the complexity of cancer because it replaces the standard definitions based on the stage of disease or tissue of origin with much smaller classifications based on molecular signatures that span all types of solid tumors. We're developing a panel of drugs that are tailored to a specific molecular signature. It is our hope that everybody with cancer will be ruled in and simply get the right drug from the outset."
Avalon Pharmaceuticals, Germantown, Md., is likewise using its RNA-based gene expression technology to identify patients who will or won't respond to a drug, according to David K. Bol, vice president for pharmaceutical development. Larger numbers of responsive patients can be included in clinical trials, making the company better able to show that drugs are active.
One thing that may slow down the clinical application of systems biology is that the data are not yet being collected from enough patients to be able to determine the extent to which changes in genes, proteins, and other pathway components will actually be predictive of health and disease. "Unfortunately, it's going to be a good while before we can really see the full benefit of even our current knowledge and understanding of systems," says Fred Sanfilippo, CEO of Ohio State University Medical Center. "We're only able to use a very small amount of that knowledge, because we don't measure those systems in most people."
Some of that data collection will happen at Vanderbilt, where the medical center is launching a new DNA bank that will link DNA samples from discarded blood with de-identified clinical information that cannot be linked back to the patient, Balser says. By the end of the decade, the medical center hopes to have half a million DNA samples.
"The DNA databank is being developed as a research hypothesis-testing tool," Balser says. "You don't want to be making decisions about changing health care without doing prospective, randomized, controlled trials. These large data sets should be used to develop the new ideas and exciting trends, but we will never escape the need to go back and test those ideas in the right population to confirm they're correct."
Another major roadblock to advancing systems biology into the clinic is the lack of analytical technologies to make such a broad range of measurements with high sensitivity at a low enough cost.
"Analytical technologies are so crude right now," PNNL's Wiley says. Technologies for gene expression are better able than those for proteins to gather large data sets. "One of the issues when you're starting to look at unbiased patterns in protein expression or phosphoproteins is that it is extraordinarily expensive and extremely slow," Wiley says. "What you really want to do is take a drop of blood from a patient and look at every protein, every metabolite, everything that's in that blood."
Wiley-previously a professor of pathology for 18 years-joined a national lab precisely because he saw the need for technology development. "One of the things national laboratories are very good at is developing analytical technologies," he says. He and his colleagues are developing high-throughput mass spectrometric approaches to measure biomarkers in blood and the bioinformatics infrastructure and software necessary to handle large amounts of data.
"The whole field of proteomics is stuck right now because we can't quantify proteins very well," says Allen W. Cowley Jr., professor and chair of physiology at the Medical College of Wisconsin, Milwaukee. "We can measure whether they're there with mass spectrometry, but it's very laborious to separate the proteins and measure them, and it's very difficult at this point in any clinical setting to quantify how much of that protein is present."
Cowley believes that a great deal of technology development is needed to make measurements more sensitive and noninvasive. "We have to get beyond measuring markers in blood and urine. That's going to be very limiting." He thinks the answer lies with imaging technologies that will allow measurements within tissues and organs deep within the body.
Such analytical measurements will need to be sensitive and "practically free" for systems biology to make clinical impact, says James R. Heath, a chemistry professor at California Institute of Technology. He collaborates with Hood at the Institute for Systems Biology and Michael E. Phelps at the University of California, Los Angeles, in a program called the Nanosystems Biology Alliance. The goal of the alliance is the development of analytical tools that can be used in the clinic.
It's not enough to do only one type of measurement, such as gene expression or proteins, Heath says. "You have to actually do true multiparameter measurements for remarkably low cost." The alliance is developing nanotechnology platforms to analyze mRNA, proteins, and cells in the same microenvironment. "We have technologies and methods that are beginning to let us do on the order of 20 or 30 measurements on a skinny-needle biopsy, something like 1,000 cells," Heath says.
One problem for protein analysis is the use of antibodies, Heath says. Many of the target proteins are present at such low concentrations that antibodies are used to enrich them. "Developing a high-affinity antibody can take a lot of money and a lot of time. Antibodies are notoriously unstable, and you're left with no guarantee of success," he says. "Finding molecules that have the same specificity for proteins that antibodies do, that can distinguish between slightly modified proteins versus unmodified ones, that are stable, that are relatively inexpensive to make, and that can be prepared in high throughput is a really tough chemical problem."
To truly affect patient care, Heath believes, nanotechnology must eventually lead to devices that patients can use in their own homes, much as diabetics can use glucose monitors today. "It's hard to imagine doing 2,000 measurements on blood if you can't do them electronically," he says.
Such devices will most likely be based on label-free methods with sensors made of nanowires. "The challenge is that they're still really expensive to make," Heath says. A lot of nanotechnology devices look remarkably like a graduate student's attempt to make one or two measurements. "That's hardly a clinical tool," Heath says. "If you're going to go into a doctor's office or take something home and make clinical decisions based on measurements you make in the blood, you want to be darn sure that that measurement is reproducible."
Emanuel Petricoin III, codirector of the Center for Applied Proteomics & Molecular Medicine at George Mason University, Fairfax, Va., agrees that measurements need to become much more reproducible. Petricoin came to George Mason because he was attracted by the opportunity to take proteomics much closer to the patient's bedside. His center works directly with pharmaceutical and biotech companies, as well as Inova Fairfax Hospital, a community hospital in Northern Virginia.
He and his coworkers are developing a "reverse-phase" protein array that can provide reliable, reproducible proteomic measurements in the clinic. Instead of immobilizing antibodies on the surface, as is usually done, they immobilize tissue or even individual cells. Using a technology called laser capture microdissection, they can remove cells from biopsy specimens, lyse them, and immobilize the cell lysates. "It's probably the most sensitive proteomic multiplexing technology that exists, as far as being able to look at many things at once from very small amounts of tissue." Petricoin is now using the protein array in clinical trials to analyze cellular circuitry before, during, and after therapy.
Despite all the emphasis on large data sets in systems biology, Petricoin cautions against collecting large data sets for their own sake. "At the bedside, you only need to measure what you need to measure," he says. "If you only need to measure seven proteins that have all the predictive value, that's all you need to be measuring."
Once the technical hurdles are overcome, Hood has a "grand vision" for how systems biology will transform medicine. Understanding biological networks and how disease perturbs them "gives us the ability to really revolutionize diagnostics, therapeutics, and ultimately disease prevention," Hood says. He sees medicine advancing in stages to become predictive, preventive, personalized, and finally participatory-what he calls "P4 medicine."
Hood predicts that within 10 years, DNA-sequencing technologies will be rapid and inexpensive enough that everybody's genome can be sequenced. Having this information will make it possible to make correlations and predictions about disease onset. In addition, researchers are finding unique molecular fingerprints of proteins that are secreted into the blood by each tissue type and organ in the body. Measuring these proteins will give a snapshot of the patient's current health status and presumably enable researchers to make predictions about appropriate treatments.
As we learn more about biological networks and are better able to make predictions, medicine will become less reactive, Hood predicts. Rather, the focus will shift from disease treatment to health maintenance and disease prevention. "How do we keep people healthy, rather than waiting until they get sick and trying to make them healthy?" he asks. Finally, medicine will become both personalized, with treatment tailored for each individual, and participatory, with patients helping to make decisions about their own health care.
Just as the ready availability of investment information has given people the power to make their own financial decisions, so will information about their current and projected health status allow people to direct their own health care. "Information liberates people," Hood says.
Hood is reluctant to put a firm timeline on his predictions about systems biology fulfilling its clinical promise, but he is sure that it will be sooner than anybody expects. "If you want to think about the future, you have to take into account the exponential change of technology," he says. "I've made a lot of predictions throughout my life. Inevitably, my predictions have been too conservative."