Thriving tumors burn glucose and show up as bright spots on positron emission tomography screens. Those bright spots can disappear a week after a patient begins chemotherapy, signaling possible remission. Waiting for demonstrable tumor shrinkage on computed tomography scans takes another six months. Seeing whether a patient actually lives longer could take years.
Usually, changes in positron emission tomography (PET) results are not dramatic enough to be seen by the naked eye, says Gregory Sorensen, a radiologist at Massachusetts General Hospital (MGH). The signal must be interpreted through a series of calculations gauging the concentration of a radioactive tracer, like 18F-fluorodeoxy-glucose. Still, done right, Sorensen says, imaging can slash years off clinical trials and answer questions with studies of 100 patients that would otherwise require thousands.
Imaging and other biological markers, from genomics, proteomics, and even metabolomics, are being hailed for their potential to predict a patient's susceptibility to disease or response to drugs long before slow and expensive clinical studies confirm it.
"Without biomarkers, developing novel drugs would be almost impossible," says Paul Rolan, a pharmacologist at University of Adelaide, Australia. Technologies to find biomarkers are proliferating, but results are inconsistent. Even when assays are reproducible, demonstrating that a biomarker reliably correlates with a clinical outcome requires more data, time, and money than the average $800-million, 10-year trek getting a drug to market. If biomarkers are really going to produce more effective treatments with lower costs, the pharmaceutical industry must coordinate its efforts on an unprecedented scale.
PLAYING WELL WITH ACADEMIA
Drug companies are desperate for these tools. Last year, the US biopharmaceutical industry spent $49.3 billion on research and development, according to estimates by the industry group Pharmaceutical Research and Manufacturers of America (PhRMA). But for all the funds allocated, a majority of experimental drug candidates fail to become approved drugs. In 2004, the US Food and Drug Administration (FDA) estimated that a drug entering clinical trials had only an 8% chance of reaching the market, falling from the historic 14%.1 Reliable, early read-outs of whether and when an experimental treatment works would spur investment in innovative therapies.
© 2005 Society for Neuroscience
Diagnosing Alzheimer disease in living patients has presented significant challenges. In this multiphoton image at left, neurites (green) curve around senile plaques (blue) in the brain of a living mouse transgenic for amyloid precursor protein. The plaques were labeled with methoxy-X04, a systemically administered Congo-red derivative. A three dimensional reconstruction at right shows the close association between plaque and dendrite. (Adapted from T.L. Spires,
To find new biomarkers, companies are forming alliances with academic centers. Sorensen heads MGH's Center for Biomarkers in Imaging, which works with many pharmaceutical companies. His group is currently using magnetic resonance imaging (MRI) to test a statin-for-stroke treatment. They examine patients' blood vessels for plaques before and after receiving the drug. In another example, Pfizer and Yale University scientists will use PET and other scans to figure out where in the body medicines work. And Ciphergen Biosystems, a company that sells tools to analyze the proteins in blood, urine, and tissue samples, is funding work at Johns Hopkins to develop a blood test for ovarian cancer that could detect the disease in its earliest stages.
But identifying a putative biomarker is very different from showing that it can be trusted. The former might require only a matter of weeks and a few patient samples, yet the latter may require multiple multicenter trials. Skimping can have devastating results.
Past failures show how even seemingly straightforward, logical biomarkers can prove false. Some osteoporosis drugs increase bone density but boost the risk of fractures. A trial for chronic granulomatous disease, an immune disorder, was originally planned to last only long enough to evaluate whether patients' white blood cells overcame the disease's characteristic defect – an inability to generate a bacteria-killing oxygen burst. Fortunately, a longer trial plan was adopted: The drug had no detectable effect on oxygen production or bacterial death, but still cut the rate of recurrent serious infections by 70%.
Drug Rescue Just the Beginning
Biomarkers can rescue drugs, but the story doesn't necessarily end there. Herceptin, approved in 1998, emerged as a $480 million-per-year winner only a decade after clinical trials showed little or no efficacy. Only when the 20% to 30% of women whose tumors overexpress HER2 were singled out was the drug's efficacy indisputable. In the pivotal clinical trial of patients with metastatic breast cancer, tumor-response rates to Herceptin plus chemotherapy were 45%, compared to 29% for chemotherapy alone.
But response is not wholly predictable. Reported response rates for HER2-positive cancers vary from less than 20% to more than 75%. HER2-positive cells that don't respond to Herceptin may have more active forms of the kinase Akt. And HER2 belongs to a receptor family that can be activated by 11 different soluble proteins and combinations thereof. Researchers are already betting that working out the biology behind the biomarker will lead to better treatments. Another anti-cancer antibody based on this understanding is already in clinical trials.
Similarly, the lung-cancer drug Iressa (gefitinib) could be rescued by a diagnostic. Unfavorable clinical trial results dashed high hopes for big sales, but finding the patients most likely to benefit might change its prospects. Last year, separate studies found that in nine samples obtained from patients that responded to Iressa, eight had mutations in the gene for EGFR.12 Researchers recently identified a protein involved in developing resistance to the drug, the protein endothelial membrane protein-1 (EMP-1).3 In human samples, tumors sensitive to Iressa came up negative for EMP-1, but the biomarker is not a slam dunk. More than 70% of Iressa-resistant tumors also lacked the protein.
Most recently, and most controversially, the heart-failure drug BiDil was approved only for black people. Two trials studying the drug in a mixed population found no benefit. In a study of 1,050 self-identified black patients with severe heart failure, 6.2% of patients taking the drug along with standard therapy died, compared to 10.2% of patients taking placebo; the survival benefit was so strong that the trial was halted early. Still, using race as a biomarker is debated. One argument is that the latest trial showed efficacy not because the drug is more effective for black people, but because black people of the age studied are more prone to die from heart disease, providing more events from which to garner data. Another is that race correlates with some physiological (or even socioeconomic) factor that makes the drug work. Everyone agrees that using race to determine treatment will exclude some patients that can benefit and include some who can't.
For these and other drugs, getting to the real heart of the matter means plumbing the underlying biology.
The most notorious example, however, comes from former heart drugs encainide, flecainide, and morcizine. These drugs normalize dangerously irregular heartbeats and were once prescribed to hundreds of thousands of patients. Ethicists argued that multiyear trials would endanger the untreated placebo group. In fact, death rates for treated patients were 2.5 times higher than for controls, and the drugs were quickly pulled from the market.
Instead of bone density or heartbeat, biomarkers are now more likely to be differences in genes or protein expression. But techniques to find them produce diverging answers. For example, mass spectra can be analyzed through machine learning and various computer algorithms to identify sets of peaks that distinguish between, say, late-stage and early-stage cancer. Nevertheless, Paul Tempst and colleagues at the Memorial Sloan-Kettering Cancer Center recently found that in five published studies on prostate cancer, "only two of the many discriminatory masses appear in more than a single publication."2 Stephen Naylor, a computational biologist at the Massachusetts Institute of Technology and former chief technology officer of Beyond Genomics, says the same proteins tend to pop up for multiple unrelated diseases, making them useless for accurate diagnosis.
"You've got some new technology that tells you something has changed, but you have to know what that change means," says Stephen Williams, head of clinical technology platforms at Pfizer. To find that meaning – and to convince the regulatory authorities and scientific community of that meaning – you need strong data tying biomarker to clinical outcome. Often, that's more data than any individual company can afford to collect.
PLAYING WORSE TOGETHER
Companies are playing close to the chest both to maintain their competitive edge and protect information that could spin out diagnostic tests. Thus, scientists can't replicate each other's work or offer alternative hypotheses. The dream of making big money on a diagnostic biomarker just holds back the industry, says Don Stanski, scientific advisor to the director at the FDA's Center for Drug Evaluation and Research, who is working to foster collaboration among academia, industry, and government.3 "It's hard to think of a biomarker that one person owns." Resulting tests probably won't be predictive enough to generate huge profits, says Rolan. "If it's just done on a pure commercial model, you get one company, one test. But a combination of tests can give you a better result." Combinations, he says, would be much more likely to arise from consortia.
Though none so far have the goal of producing diagnostics, some successful cross-industry collaborations have emerged. In 2001, at Imperial College London, six pharmaceutical companies funded the Consortium for Metabonomic Toxicology (COMET) through a subscription fee that paid for up to five postdocs, equipment, and other costs. The companies provided blood and urine samples from drug safety studies on rats and mice. Imperial researchers tallied results of 147 treatments to find how chemicals known to be toxic affected concentrations and ratios of a few hundred metabolites. Subscribers got access to databases of the results along with models to predict toxicity. COMET-2 is already in the works.
The National Institutes of Health has launched two larger, better-funded efforts. The Alzheimer Disease Neuroimaging Initiative (ADNI) and Osteoarthritis Initiative have both received funding from multiple pharmaceutical companies to identify biomarkers that can be used to track disease progression and severity. ADNI will spend $60 million over five years to fund longitudinal prospective studies correlating cognitive assessments with serial MRI, PET, and biomarkers found in blood and cerebrospinal fluid.
Still, drug companies have been slow to embrace collaboration. Executives tend to be uncomfortable funding research that helps their competitors. Right now, says Sorensen, collaboration requires "a funny accident of particularly good questions with particularly big diseases where other approaches have failed." ADNI took years to get off the ground, he says, and attempts at collaboration in cancer have fizzled.
LEARNING TO SHARE
© 2003 Nature Publishing Group
Computed tomography sections of this lung nodule taken a month apart are hard to distinguish. A three-dimensional volume rendering of the lesion, automatically separated from surrounding vasculature, helps provide information for analysis. Volume has increased 24% and doubling time is estimated at 103 days. Biopsy confirmed the lesion to be large-cell carcinoma. (Adapted from R. Frank, R. Hargreaves,
Williams says industry has other reasons to be wary. Companies have diverse but specific questions, and academics' questions are often, well, academic. Collaborative studies address questions other than the ones a drug company values. Studies that aimed only at those questions could be smaller and faster, tempting companies to try studying solo. Still, Williams says, collaborations can be incredibly cost effective. "You get access to a big expensive experiment and you don't have to pay for all of it."
Sharing existing data would be even more cost effective than collective efforts that generate new data, since pharmaceutical companies already own heaps from clinical trials. That might already be happening. "Companies [are starting] to recognize the fact that they are spending their individual research dollars on something that another company is duplicating," says Caroline Loew, vice president for scientific and regulatory affairs at PhRMA. If, for example, companies have collected a substantial amount of data on a biomarker, the information could be blinded and pooled, and the FDA could determine whether the biomarker had sufficient evidence to back its use in regulatory decisions. An NIH-industry group is already evaluating a portfolio of biomarkers. Pfizer's Williams is hoping to organize such efforts. His colleagues are interested, but nervous. "Sharing data causes the biggest anxiety," he says.
The FDA's Stanski, also working to promote wide-ranging collaborations, is used to this attitude. "Some people think they have to protect everything," he says, "and some people are more visionary."
The technology to find the best biomarkers doesn't exist yet, says Lee Hartwell, Nobel laureate and president of the Fred Hutchinson Cancer Research Center (FHCRC) in Seattle. Humans make hundreds of thousands of proteins and peptides, some of which are a trillionfold more common than others, and the most informative biomarkers are probably among the least abundant.
The techniques necessary to enrich and fractionate low-abundance proteins are immature. Artifacts of sample preparation and processing are frequently indistinguishable from true peptide profiles. Current attempts to validate biomarkers are, Hartwell says, horribly inefficient.
Along with the University of Michigan, FHCRC is leading a multimillion-dollar project to figure out how to get reproducible results from multiple laboratories. (See more, page 18) The teams will use well-characterized cancers from transgenic mice, a common informatics program, and even common reagents. Ultimately, they plan to create common data sets and reference standards to produce a sort of proteomic Rosetta stone.