YouTube Facebook LinkedIn Google+ Twitter Xinginstagram rss  

A Virtual Pharmacopeia 
By Michael Goldman

Supporting simulation: Physiome Sciences CEO Jeremy Levin believes in the power of simulation in pharmaceutical research.
Nov. 11, 2002 | In a high-ceilinged office at the edge of the University of Washington's medical school campus in Seattle, James Bassingthwaighte leads a global effort to codify a puzzle that is orders of magnitude more complex than the human genome itself. The modest goal of the Human Physiome Project is nothing less than the description of every physicochemical event that makes the human body run smoothly — or not.

The pharmaceutical industry is desperately in need of a shot in the arm. With stock prices plunging, pipelines bare, and reports that the majority of prescription drugs approved in the past decade were simply old medicines in new bottles, many observers are impatiently wondering what happened to the promise of the much-hyped genomics revolution.

"There is an industry movement away from drug discovery and a feeling that the Human Genome Project did not pay off," says Thomas Tibbitts, chief technical officer for Cernomics Inc., a Massachusetts-based maker of drug discovery software.

The cure the industry needs may come from an obscure intersection of biological and information science — in silico biology — the logical successor to traditional "wet lab" studies in test tubes (in vitro) and living organisms (in vivo). Through the identification of 30,000 or more genes and millions of variations in SNPs (single nucleotide polymorphisms), genomics will help reveal the basis, and ultimately the best therapy, for many inherited diseases and make it possible to "personalize" medicines.

"There is an industry movement away from drug discovery and a feeling that the Human Genome Project did not pay off."
Thomas Tibbitts, Cernomics Inc.
But the identification of new drug targets for common diseases, including cancer and heart disease, will require a rigorous description of changes in the concentrations and chemical states of 100,000 or more proteins and their interactions in the body. Taking inventory of specific genes or proteins and assigning their function will not explain the dynamic state of whole cells and organs — a prerequisite for the next phase in medicine. Rather, what is needed is a description of how genes, proteins, and other metabolites interact, and how these networks are perturbed in disease states and other conditions.

This transition from decades of reductionist biology, focusing on a single gene or protein, to a new systems biology approach will be ushered in by the next wave of "-omics" disciplines, including transcriptomics (the spectrum of RNA molecules in each cell type), metabolomics (all of the chemicals in the cell), and physiomics (the function of organs, tissues, and ultimately the whole organism). Critical to the success of this revolution are new computational strategies for modeling the properties of cells, organs, and disease states, producing virtual organs, even virtual patients.

In Silico Medicine 
A cardiologist by training, Bassingthwaighte no longer sees patients, but he firmly describes himself as "a physician first." His own research has involved detailed modeling of the function of the heart, which many say is an ideal candidate for in silico analysis because of the wealth of good quantitative data and its undoubted relevance to the pharmaceutical industry. Not only is heart disease the leading cause of mortality in the United States, but also the very drugs that are supposed to correct heart ailments often result in serious side effects.

Just a few patients with life-threatening cardiac arrhythmias prompted the FDA to pull potential blockbuster drugs, including Seldane, from the market. The ability to simulate the effect of a drug on ventricular depolarization-repolarization —thereby identifying a subset of patients who might be susceptible to serious side effects — could save lives by sharply reducing the more than 100,000 fatal adverse drug reactions every year.

Bassingthwaighte envisions a time when computer models could integrate and interpret medical data for an individual patient, guiding primary care physicians and specialists alike in determining an optimum treatment plan for the individual. In principle, this is possible today by considering a patient's medical history, laboratory tests, and current medications. In practice, Bassingthwaighte says, most large medical centers are so compartmentalized that a cardiologist may have no idea what an oncologist is doing for a particular patient.

But with the universal markup languages being designed to share diverse kinds of data —medical, genetic, proteomic, physiological —across platforms, a coordinated medical-care system could become a reality.

Enabling In Silico Pharmacology 
At the far end of the drug discovery and development scheme, Pharsight Corp., based in Mountain View, Calif., has carved out a niche in the design of clinical trials. Working with more than a dozen pharmaceutical and biotechnology companies, including Pfizer, Aventis, Eli Lilly, Millennium, Roche, Schering-Plough, Cephalon, and Vertex, Pharsight seeks to model clinical trials at both the physiological level and in the selection of doses and numbers of patients.

According to Pharsight president and CEO Michael S. Perry, "Modeling and simulation can add tremendous value, from hundreds of millions at the individual drug level, to billions of dollars at the industry level." The company's proprietary software and hardware are co-marketed with IBM's Life Sciences division.

Speaking a Common Language 
Exploiting human genome data requires changing the way we think and communicate.

Read More 
Pfizer has already publicized one of Pharsight's early success stories. Pfizer scientist David Hermann and colleagues simulated an Alzheimer's disease drug candidate, modeling factors such as disease progression, pharmacodynamics, placebo effects, absorption, clearance, and even the patient population's variability in response. Simulation results convinced the project team to conduct a trial using a 4x4 Latin Square design (which models trials based on several different designs in a single simulation) with a four-week crossover period.

Presenting in March at the 2002 American Society for Clinical Pharmacology and Therapeutics conference in Atlanta, Hermann said that this computer-assisted trial design "aided the decision to move forward with a more efficient go/no-go trial design, leading to savings of more than $1.4 [million] to $3.8 million in direct costs and a go/no-go decision eight to 12 months earlier" than would a traditional 12-week parallel group design. In addition, Pfizer was able to redeploy its team earlier, leading to considerable staff savings.

Perry cites another example in which, "based on our simulations, we recommended that a client abandon a compound prior to the completion of Phase II. Our software showed a high probability of failure, with a very tight confidence interval." The client chose to carry on, only to abandon the compound during Phase III trials after spending roughly $50 million more, Perry says.

Such in silico insights could also help companies avoid abandoning a product prematurely that could have succeeded. Traditional decision- making and prioritizing of the drug portfolio are far from infallible.

"In portfolio review," Perry says, "it is easy to move forward with the distinct winners and kill the loser candidates. But in the middle, which may comprise up to 80 percent of the portfolio, decisions can be arbitrary." Simulations can assign a meaningful probability of success on which to base such decisions.

While Pharsight has carved out a niche in the preclinical and clinical testing arena, companies such as Entelos Inc., Physiome Sciences Inc., and Molecular Mining Corp. are targeting the discovery phases, where success can be more difficult to quantify.

Molecular Mining, based in Kingston, Ontario, takes a data-driven approach, building its models from known experimental results rather than on the theoretical underpinnings. "We can tune our technology to handle proteomics, SNPs, clinical measurements, pre- and post-surgery data, cancer survival rates, etc.," CEO Evan Steeg says.

The Systematic Approach 
Delivering on the rich promise that the broad scope of systems biology offers

Read More 
In one example Steeg cites, a large East Coast drug discovery company had implicated some specific chromosomal regions in disease, but Molecular Mining's algorithms provided multi-SNP markers that predicted the disease with greater accuracy. Working with two to five genes at a time, rather than with entire metabolic pathways, Molecular Mining has derived predictive rules based on a National Cancer Institute database; the company has filed for patent protection on these findings.

Steven Horrigan, senior scientist with Avalon Pharmaceuticals in Gaithersburg, Md., is pleased with Molecular Mining's work. "We do drug discovery," he says. "We have lots of compounds and lots of data. [Molecular Mining] make sense of it." Avalon runs traditional analyses in parallel, and as Horrigan emphasizes, "We are not looking to substitute for assays." But Molecular Mining's analysis helps Avalon prioritize drug candidates correctly, a benefit Horrigan thinks will be significant.

Heart of the Matter 
Physiome Sciences in Princeton, N.J., one of the more established in silico biology leaders, concentrates on drug discovery at the early stages. Models are built on a biology-based knowledge of signal transduction and metabolic pathways. "We look at where drugs act in pathways," evaluating targets and lead drug candidates, says Mark Engelhart, Physiome's vice president of commercial development. "Where we come in is in taking the complexity of the biology and helping to simplify and narrow it down pretty quickly to a selected few compounds. You get to your winners faster."

But pharmaceutical companies must still do all of the wet lab work involved in drug discovery. Although Physiome models from the levels of gene, protein, and cell function based on kinetics, new data are continually used to refine models. Engelhart notes that one popular misconception is that exhaustive data are required to produce a robust model. "That's not the case," he says. "You can get a lot of value in a combination of wet bench data and modeling. Gaps in your understanding redirect the next round of experiments."

Physiome CEO Jeremy Levin offers a dramatic example of the power of simulation. A large European pharmaceutical company had selected a target in the tumor necrosis factor pathway and spent 18 months studying it by the classic experimental approach, without success. In one afternoon, Physiome's simulations demonstrated why the target failed. There were collateral pathways that resulted in little or no downstream effect of an inhibitor.

Physiome's approach also prioritized several alternative targets. Although living systems may be far too complicated to model in excruciating detail in silico, experiments at the laboratory bench are also painfully inadequate. "The message," Levin says, "is that even with really smart scientists, some of these systems are so complicated that decisions based on only intuition and experience are inefficient," or even incorrect.

Dennis Noble, Physiome Sciences' acting chief scientific officer (CSO) is, like Bassingthwaighte, a leader in modeling the human heart. Physiome has clients and partners in the pharmaceutical and biotechnology arena, including Aventis, Johnson & Johnson, and 3rd Millennium.

Entelos, based in Menlo Park, Calif., is "the only company creating models and representations at the disease level," according to James M. Karis, its president and CEO. Entelos, with clients including Johnson & Johnson, Pfizer, and the American Diabetes Association, produces software models of all biological systems involved in a particular disease state, including obesity, diabetes, asthma, and rheumatoid arthritis. Efficacy, rather than safety or toxicity, is the primary target of the modeling.

Although some of Entelos' work is at the single-gene level, the general approach is from the top down, with models dictated by the clinical manifestations of the systems involved. Entelos focuses on uncertainties about the underlying genetic and environmental factors in disease.

"No two diabetics look exactly the same," says Entelos CSO Thomas Paterson, and the software must take that into account. Paterson, a veritable cheerleader for the in silico biology approach, says 20 percent of the pharmaceutical industry's revenues are spent on R&D — and that's too much. "In the electronics and aerospace industry, that figure is only 7 percent," he says, and the difference is that in silico modeling is routine in those industries.

Computational modeling of disease pathways, organs — even patients — could transform drug discovery. Does salvation exist in silico?
Camille Wallwork, a medical informatics and drug discovery researcher with Johnson & Johnson, uses the Entelos PhysioLab platform in drug discovery and in the analysis of clinical trial data. "We expect the modeling to impact development timelines," she says. "With a more comprehensive understanding of diabetes, J&J is better equipped to be able to provide appropriate and feasible treatments, such as a device-drug combination."

Engelhart says there's little overlap between Physiome Sciences and Entelos. Entelos focuses on comprehensive models involving systems related to a particular disease, he says, while his company exploits metabolic networks in order to identify the more fundamental causes of disease. Pharmaceutical companies can benefit from using both Entelos and Physiome as complementary approaches, Engelhart says.

 From wet work to network: A human cancer cell pathway as visualized by Gene Network Sciences

Another approach is that of Gene Network Sciences Inc. (GNS), which is likely the most academically oriented of the in silico biology companies. The Institute for Systems Biology in Seattle (see "The Systematic Approach," page 60) recently began using GNS software to model specific diseases. Operating on a slim budget from Ithaca, N.Y., GNS has already released the largest computer simulation of a human cancer cell (see "GNS Bare World's Most Complex Cancer Cell Model," August Bio·IT World, page 14), with more than 500 genes and proteins representing most of the major signal transduction pathways and gene expression networks that control the human cell cycle. "This model," CEO Colin Hill says, "contains one-third to one-half of all cancer drug targets."

Down with disease: James M. Karis, Entelos president and CEO, says his is "the only company creating models and representations at the disease level."
As large as the model is, Hill emphasizes it "represents only a few percent of the circuitry of a human cell. What do you do about the other 95 percent?" Though the cancer cell simulation can run on a laptop, GNS hopes to look further with a dedicated 192-processor computer engine from IBM Life Sciences. "We're swinging for home runs," Hill says. "We are at the cusp of a major revolution in science."

Revolution or no, in this climate of close FDA scrutiny and a growing mistrust of the drug industry, what are the long-term prospects for in silico drug discovery and development?

In silico methods will be very important in the validation of new drug targets and in the design of clinical trials. Targets that are validated will be verified by traditional in vivo and in vitro methods, and these wet lab data may be the only information forwarded to the FDA. Clinical trial simulations will identify the best human drug trial designs. Actual data from the human trials are what the FDA will see. In silico methods may, therefore, be irrelevant to the FDA approval process.

But as Hiroaki Kitano, of the Systems Biology Institute in Tokyo, wrote in March in Science: "It is not inconceivable that the [FDA] may one day mandate simulation-based screening of therapeutic agents, just as plans for all high-rise buildings are required to undergo structural dynamics analysis to confirm earthquake resistance."

Physiome's Levin doesn't think the FDA will ever mandate a simulation, but he makes a good argument for why they should. "Drug-induced QT prolongation [associated with cardiac arrhythmia] is responsible for several FDA safety-based drug withdrawals since 1998," he says.

A retrospective simulation study of the drug D-sotalol, which was associated with a higher mortality in women than men in a Phase III trial, showed that a small difference in ion channels between men and women could explain the increased arrhythmia incidence in women. "There is no excuse to put this into a man or woman before putting it into a computer model," Levin says. "We've got to show [the FDA] that this type of approach can be dramatically helpful. Current methods for assessing cardiac safety are so diverse and troubled as to be unpredictable."

The use of in silico methods is too new to evaluate fully. But at the clinical trials end of the drug development process, concrete results are emerging. Geraldine Cruz, an industry analyst with Gartner, sees Pharsight as a traditional service provider for the pharmaceutical industry. "They aren't just providing information for R&D, but the bigger picture," she says. "They can help the decision-makers."

Prospects for an In Silico Pharmacopeia
Although Pfizer's estimated savings of several million dollars with Pharsight's analysis in a Phase II clinical trial are minuscule compared with the $400 million to $800 million typically involved in bringing a drug to market, it represents intervention in one small slice of the drug discovery and development process. PricewaterhouseCoopers estimates that in silico methods applied throughout the drug discovery and development timeline could save $200 million and two years for each drug reaching the market.

Further Reading 
Voit, E. O. "Metabolic modeling: a tool of drug discovery in the post-genomic era." Drug Discovery Today 7, 621-628: 2002.

Kitano, H. "Systems biology: A brief overview." Science 295, 1662-1664: 2002.

Davidson, E. H., et al. "A genomic regulatory network for development." Science 295, 1669-1678: 2002.

Lazarou, J., B. H. Pomerantz, & Corey, P. N. "Incidence of adverse drug reactions in hospitalized patients." Journal of the American Medical Association 279, 1200-1204: 1998.

Hermann, D., P. A. Lockwood, W. Ewy, N. H. G. Holford. "Use of computer assisted trial design (CATD) for a phase 2 go, no-go decision." Clinical Pharmacology and Therapeutics 71, 57: 2002.
Physiome's Engelhart doesn't make promises in dollars and years, but he says that simulation allows pharmaceutical companies to narrow the field of candidate drugs more rapidly and efficiently than traditional experimental methods.

Systems biology will become more important for drug discovery as the backlog of untapped data swell. Frost & Sullivan estimates the market will grow from $84 million in 2001 to about $6 billion by 2008. The competition between in silico companies will only intensify as the pharmaceutical industry recovers from its present slump.

The outlook for systems biology companies depends on whether pharmaceutical companies choose to outsource in silico biology or attempt to do it themselves. "Our biggest competition is in-house," Molecular Mining's Steeg says of the pharmaceutical companies. Eli Lilly, for instance, is establishing a systems biology center in Singapore.

But there is clearly a dearth of scientists trained in both information and biological sciences, and a handful of companies have locked up much of the talent. Bioinformatics companies have learned the hard way that large biotechnology and pharmaceutical companies will not pay top dollar for analyses they can do on their own.

But systems biology and in silico modeling are considerably more complex. Levin sees a parallel between their future and the existing partnerships between biotechnology and pharmaceutical industries. The current model of global academic centers such as the Human Physiome Project, private research institutes such as the Institute for Systems Biology, and a handful of pioneering companies is likely to remain pivotal for the next decade and beyond.

Michael A. Goldman is a professor in the department of biology at San Francisco State University. He can be reached at 


For reprints and/or copyright permission, please contact Angela Parsons, 781.972.5467.