George Mack is a freelance writer based in Columbia, South Carolina.
Big pharma is dabbling in systems biology, a few brave entrepreneurs are trying to make a business out of it, but investor caution may keep those in the field from really making it big in drug development.
Today's high-throughput technologies have dumped reams of information on the disc drives of computers at every major drug developer. Some think systems biology players may hold the keys to putting all that information together to speed up drug discovery and development and to make it much more efficient. There is a crying need for a better approach to drug development, with the high cost—both in time and money—of producing a new blockbuster drug. Moreover, the US Food and Drug Administration (FDA; Rockville, MD) is currently seeing a 50% failure rate among drugs in late-stage phase 3 trials, according to Lester M. Crawford, acting commissioner of the FDA.
A systems approach is designed to integrate and analyze varying streams of biochemical information in ways that are not obvious to even the highest of human intelligence so that workable theories can be validated without countless years of trial-and-error research. Some large pharmas are bringing on the new technologies for discovery and development on their own (Box 1 and Table 1). Other companies are positioning themselves as 'tool' providers (Box 2 and Table 2). But a small cadre of systems biology companies has emerged (Table 3), and although some of these innovators have shown flashes of genius in helping the big pharmas discover and develop drugs, old habits die hard. Systems biology service companies are not finding it easy to sell a cryptic technology to major healthcare companies.
Table 1. Public companies with research programs on systems
Does systems biology mean business? The term 'systems biology' is not terribly informative; some have argued that it is just a fancy term for physiology. But for the purposes of this article, we consider system biology companies those that go beyond data integration, and use information from disparate data sets to create computational models that can describe and predict phenotype at the cell, tissue or organismal level. Some functional genomics companies are backing into a systems approach, as they try to find ways to use varied kinds of information (Box 3); others are approaching it head-on.
It isn't easy to get companies in this sector to talk about what they are doing and who their collaborators are, except in very general terms. Some even eschew the moniker of systems biology altogether. This may reflect the response that 'systems biology' companies are getting from the marketplace. Investors might see systems biology as another technology platform, something they have increasingly shied away from in favor of more product-oriented plays. This attitude from the moneymen is unlikely to change unless systems biology comes up with tangible results to show to investors.
Physiology modeling By extracting and reviewing all the information available on a disease, companies are creating computer models to simulate the course of that disease and its response to input—drugs, environment or merely the progression of time. To do this, though, requires the talents of many different kinds of scientists, from the biologist and chemist to the engineer and physicist, not to mention the information technology specialist. Whereas some companies choose to focus on specific questions in the study of specific diseases, others are aiming to predict changes of all the major components of an entire system at once, something you might need when studying a molecule with an unknown mechanism of action. With the present technology, modeling individual pathways is within reach. However, attempting to infer physiological changes purely from molecular events in a cell—what is sometimes called a bottom-up approach—may still be a stretch (Fig. 1). With a top-down approach, which characterizes much of today's systems biology, researchers start at the phenotypic or event level of a disease and drill down through functional pathways to only what is important in a specific disorder, because that disease phenotype is what they want to change.
Top-down approaches start with major systems (e.g., circulatory, immune, metabolic) and work from the top down to the relevant tissues, cells, proteins and genes. Bottom-up starts with thousands of genes and proteins and tries to fit them together in a representation of cells.
An early entry into the field, Entelos, founded in 1996, has roots going back to 1992 when its technical founders, information technology expert Alex Bangs, engineer Thomas Paterson and IT engineer Samuel Holtzman, began building computational models of human disease. Today the Foster City, California-based company has pharmaceutical and medical experts on its staff, and has developed disease-specific models for diabetes, obesity, asthma and is currently building one for rheumatoid arthritis. These models, which the company calls PhysioLabs, are large-scale mathematical representations of biological systems created from data extracted from literally thousands of peer-reviewed papers, according the Mikhail Gishizky, Entelos's CSO.
Because the models are quantitative, Entelos drug discovery customers may change a rate constant for a particular enzyme in a tissue, such as one affecting glucose utilization in the liver. A slight perturbation will ripple through the model and allow researchers to see how the perturbation affects the entire system and what the clinical output (glucose levels, in this example) would be. Gishizky says his models can predict whether a particular enzyme is a good target for drug development. But being a top-down company, they connect relatively few dots to make a mechanism-disease connection. Because of this, Entelos's PhysioLabs don't necessarily model down to the molecular level, though the company is currently working with Massachusetts Institute of Technology (Cambridge, MA, USA) to address this kind of issue in future versions of their models.
Richard Ho, head of medical informatics at Johnson & Johnson Pharmaceutical Research and Development (New Brunswick, NJ, USA), has been working with Entelos on a type II diabetes model that was created for his team a number of years ago, and that is still in use today. Ho points to a particular instance of great benefit when his endocrine discovery group wanted to pursue a specific clear target that had been identified for glucose regulation. The animal data had looked very good, but a PhysioLab simulation put all the information together and suggested that interference with that particular target would not be useful in humans. "Our drug discovery head in the endocrine group decided he would not pursue it," says Ho. For competitive reasons he will not divulge what specific products are being analyzed in his lab, or how much per year he's actually spending with Entelos. But the PhysioLab technology clearly has value to Johnson & Johnson.
Entelos, today with about 80 people, won't disclose much of anything about its business. President and CEO James Karis says annual revenue is in double-digits, meaning $10 million or more. But the company has not had to raise money since 2002, and they do make known high-value clientele (Table 3). "We have been involved with companies long enough," says Karis, "that drugs have advanced [to the clinic], but we cannot point to specifics publicly." Asked how many drugs he has helped clients get into the clinic, he says less than ten but more than five.
A company taking an approach similar to that of Entelos is Optimata (Ramat Gan, Israel). According to founder and CSO Zvia Agur, the company uses data gathered from the literature to create models of "virtual patients" that predict disease progression and study how drugs can affect the course of the disease. But with only $2 million in total investments and not a lot of revenue so far, the 20-employee company has a long way to go.
Optimata had a relationship with Genentech in studying a pipeline drug, thrombopoietin (TPO), which is used to stimulate production of platelets in oncology patients. In phase 1 clinical trials, Genentech reported some immunogenicity as an unwelcome side effect. Optimata created a model that predicted no toxic immunogenicity at all with the same optimal efficacy if the TPO was administered four times per day while using 10% to 45% of the standard daily dose. In rhesus monkeys treated with monkey recombinant TPO, the dose and schedule suggested by the model produced effective platelet stimulation with considerably reduced toxic immunogenicity1. Agur believes the simulator can be effective in predicting toxic and dose effects of other thrombopoiesis drugs. Although it appears Genentech's TPO never entered phase 3 trials, Agur considers the simulation to be a proof of concept.
From molecular profiling to stratified treatments Gene Network Sciences (Ithaca, NY, USA) has also developed a set of disease models, including cancer, inflammation and cardiovascular disease. But CEO and cofounder Colin Hill says their approach is different from others. "I think we've been more successful in making data-driven models and getting the respect of academic and industry collaborators." Hill doesn't want to refer to his enterprise as a systems biology company but rather as an "enabler" of drug development for his customers. The idea is to model the molecular profiles and predict the actual phenotypic behavior of the cells. For example, the cancer model describes how receptors interact with a given ligand, and how that signal is transmitted all the way through the cell circuitry. The researcher can effectively watch the simulated dynamics of genes going up and down and the proteins interacting. At the same time, a prediction can be made as to whether the cell will divide, proliferate uncontrollably or go through apoptosis (Fig. 2)
Figure 2. Cell cycle model showing G1-S transition and the highly connected p53 node.
The business model is a bit different from that of Entelos, in that Gene Network Sciences almost never lets its software out of the building. "We do the modeling work for our customers, and we provide answers," says Hill. And as for the secrecy about who his clientele might be, he says some of the early modeling companies did collaborations that didn't have a lot of impact on customers. "So we've taken longer to develop, and we've had a more academic evolution, being really focused on developing the science for the first few years of the company, and now we are making a transition to being a lot more commercially focused," he says.
Gene Network Sciences cofounder and vice president of research and development Iya Khalil says their models can be of best use to developers in phase 2 and phase 3 clinical trials. They are currently modifying the models to simulate what's happening in a particular patient subpopulation. The method will be to extract molecular data from human tissue biopsy and evaluate RNA and protein level profiles. The simulations can be trained with this data and modified to reflect genotype and phenotype of that subpopulation and essentially predict the effect of a drug. And after the drug is administered, the simulation should predict biomarkers that indicate either efficacy or lack of efficacy. The company is now collaborating on this very project with the National Cancer Institute (Bethesda, MD, USA) and the FDA.
Tissue profiling or systems biology? From its origin in 2000, Waltham, Massachusetts-based Beyond Genomics has been in the business of measuring transcripts, proteins and metabolites in human tissues and focusing on the differences between diseased and healthy tissues (Fig. 3)— an approach similar to that pioneered by Pharmagene in Royston, UK. Now, Beyond Genomics is using this approach to find biomarkers for the prediction or detection of disease. Acting CEO Muzammil Mansuri, who is also a venture capital investor in the company, says the FDA is encouraging companies to develop more biomarkers as decision-making tools in clinical trials, and they are beginning to acknowledge that transcripts, proteins and metabolites and their combinations are going to be important endpoints. "But one thing is clear," he says, "there's much more momentum and interest now around biomarkers for both diagnostics and monitored therapeutic response." The company says it has collaborative relationships with several companies (Table 3). It appears the business model will be internal drug discovery and partnership drug discovery and development. The company has no internally developed products in the clinic but does have revenue. So far, funding has added up to$26 million.
Figure 3. Beyond Genomics Correlation Network: Up- and downregulated genes, proteins and metabolites in a mammalian disease model versus controls ('normals').
A subset of the molecules depicted here could be used as an early biomarker of disease. The lines between and among the various molecules denote correlations, which are useful for understanding disease pathways and drug mechanisms of action from a biological system perspective. (Graphic courtesy of Beyond Genomics)
Stelios Papadapoulos, of investment bank S.G. Cowan (New York), who sits on Beyond Genomics's board, thinks that although the risks are higher with wet-lab approaches, the rewards are potentially greater. He says with pure computational systems-biology companies, you could make a nice return on investment if you have a product or service that can be sold repeatedly to pharmas and biopharmas. "But this is very different from the drug discovery business where you can fail and it will cost a lot of money," he says. "But if you win, you win big." The question is, however, whether most private biotech companies have sufficient resources and expertise to combine both the large-scale laboratory wet-work and the computational know-how in order to fully realize the systems approach.
Beyond drug discovery Outside of healthcare, there may also be commercial opportunities in other areas of systems biology. For example, METabolic Explorer (Saint Beauzire, France) is working with several large chemical companies to design new bacterial strains that should be capable of synthesizing fine chemicals more economically than conventional chemical manufacturing processes. The company's approach is to rationally design nonnatural metabolic pathways, use recombinant technology and directed evolution to engineer new strains, and then analyze strain performance using 13C substrate labeling and detection by nuclear magnetic resonance and gas chromatography/mass spectrometry. The results are incorporated into a flux map that shows the distribution of anabolic and catabolic fluxes over the metabolic network of interest. This is then used to judge the performance of the strain being tested and to identify possible targets for further genetic modification. After several iterative rounds of modeling and wet work, the company claims a strain can be generated with the desired properties.
According to Benjamin Gonzalez, president and CEO, METabolic Explorer is currently developing new strains capable of producing L-methionine and propylene glycol. Research is also underway on new bioconversion processes involving NADPH-dependent enzymes for the reduction of ketones and for hydroxylation, oxidation or epoxidation reactions. Partnerships currently under-way include an agreement with Degussa (Frankfurt) to optimize amino acid fermentation processes, a cooperation with Biogemma (a biotech subsidiary of the seed producer Limagrain; Paris) to characterize plant secondary metabolism and work with Bayer (Leverkusen, Germany) to optimize an undisclosed pharmaceutical product produced by fermentation.
The catch Whether in healthcare or fine chemicals, fewer and fewer companies are attempting to position themselves as systems biology companies, according to John Hamer of Burrill & Co., a venture capital boutique where he is entrepreneur-in-residence. (He formerly worked at Icoria.) He says the term has obvious appeal to academics and funding agencies, but the investment community is waiting to see what the value is. "Will it bring products to the clinic faster?" he asks. And he's not particularly sanguine about the prospects for the computational biology companies for one simple reason: there aren't that many potential customers. "There are only about a hundred companies or so that can actually buy your product," he says. "That makes it a very small market." One solution for software-based companies may be to use their systems biology tools to develop their own products through partnerships, he adds. "Of course, the most powerful way to demonstrate that you can do discovery quicker and faster is to do it yourself."
If the wet-lab companies can get drugs into the clinic or provide economic alternatives to fine chemicals, that will be the value-driver for them. But the big challenge to the computational systems biology companies is effective communication with healthcare or chemical companies, many of whom see the technology as a black box. Just understanding the concepts, which use not only biology and information technology but chemistry, engineering and physics, can be problematic. This makes it a hard sell to some biologists and many clinicians, and it may ultimately require a leadership change in the R&D departments of some large companies.
For the pharmaceutical industry, which has spent a king's ransom on proteomic and genomic technologies without a lot of reward for their efforts, there are some deeper issues. "It's a bit of a Catch-22," says Andrew F. Branca, senior vice president and group director at Cambridge Healthtech Advisors, a research firm to the pharmaceutical industry. "These systems biology companies need the data and resources of pharma to build effective computational models, but pharma research leaders are largely disinterested in providing either until they've seen a relevant model that actually works." Despite this, he's not altogether without optimism. "There's no doubt that in the long run a systems biology approach is the way to go," says Branca. "The challenge for these companies lies in surviving long enough for the promise of their models to be realized."
Skomorovski, K. et al. New TPO treatment schedules of increased safety and efficacy: pre-clinical validation of a thrombopoiesis simulation model. British J. Haematol.123, 683−691 (2003). | Article | ISI |
