When the patient—a retired physician—walked into oncologist Janessa Laskin’s office at the British Columbia Cancer Agency in October of 2008, both of them knew he was dying. He had a rare type of cancer, adenocarcinoma of the tongue, which had metastasized to his lungs. Laskin had two choices for treatment. Traditional chemotherapy could prove dangerous to the 77-year-old man and carried little chance of success. Targeted therapies had been developed to attack genetic dysfunctions found in some other types of cancer, but they had never been tested in his variety.

Meanwhile, across the street at the agency’s Genome Sciences Centre, bioinformatics director Steven Jones and his collaborators happened to be finishing a herculean task. They had used the center’s bank of sequencing machines to read the entire DNA sequence of a series of ovarian tumor cells, one base, or letter, at a time. The researchers hoped that comparing the order of these chemical units in the DNA of the cancer cells and the patients’ healthy cells would reveal the genetic anomalies that gave the cancer its destructive powers of growth and invasion.

The patient asked Laskin’s colleagues at the Genome Sciences Centre if there was anything they could do to help him, and Jones and his team suggested a plan, which the patient and Laskin decided to follow. The researchers would take a sample of his tumor, isolate the DNA, and feed it through the sequencing machines, searching for clues that might advise them on how to treat it. They spent the next three weeks working around the clock. After sequencing the DNA, they used the genetic mistakes they had identified, along with their knowledge of the molecular pathways that have been implicated in cancer, to create a model of what might be going wrong in this patient’s body. They painstakingly narrowed the search until they hit on a genetic defect that increases the activity of a molecular pathway that has been linked to the growth of cancer cells.

Laskin began treating her patient with a drug that inhibits the activity of the defective gene, and “within a month we saw 20 percent shrinkage of his cancer, whereas the previous six weeks it had undergone 20 percent growth,” she says. The cancer stabilized for six months, but then it started to grow again. The physicians tried a second set of drugs, again chosen with the guidance of the genome model. For another four months, the cancer stayed under control. Then once again it began to spread, with new tumors sprouting in the patient’s lungs and neck. Laskin performed a biopsy on these tumors and had their DNA sequenced “to see if we could explain what had changed and predict the next round of drugs,” she says. “But unfortunately, the cancer was growing really fast.” Unable to wait another three weeks for the results, she treated him with standard chemotherapy. The patient died in November of 2009.

The case, the first in which scientists used the entire DNA sequence of a tumor to help choose drugs for a cancer patient, provides a peek at both the great promise and the difficulties of using our increasingly detailed knowledge of cancer genetics to inform treatment. In many ways, cancer cells are rapidly evolving genetic mistakes. They originate from normal cells in the body, starting with the same genetic makeup as any other cell. But every time a cell divides and copies its DNA, it has the potential to make errors. Accumulate the wrong assortment of these mistakes—in genes that influence cell growth and survival, for example—and you get a cell that can outgrow and outlive its healthy neighbors. Cancer cells continue to mutate through the course of the disease, some ultimately acquiring the ability to break free of the tissue in which they originated and enter the bloodstream. Until recently, physicians have lacked the information to understand this process, let alone to make calculated attacks against its most dangerous aspect: metastasis, the spread of cancer from one part of the body to another. “A patient’s tumor is a living thing, changing all the time,” says Matthew Ellis, an oncologist at Washington University School of Medicine. “We have never been able to track that completely.”

By using DNA sequencing to decipher the precise nature of these changes, scientists hope to understand the ever-evolving cells. The results could lead to better diagnostic tests and more accurate prognoses, distinguishing patients whose cancer will probably not grow further from those whose disease is likely to spread and who would benefit from more aggressive treatment. And as scientists accumulate a list of mutations linked to cancer, they can map them against the complex molecular networks in the cell and begin to delineate the specific pathways that play a central role in the disease. With this map in hand, they will be able to tell when superficially similar types of cancer, such as two lung cancers, have different molecular causes. Then they’ll be able to design safer and more effective medicines and target them specifically to the patients who are most likely to respond. “In the next few years, we’ll create an essentially complete catalogue of mutations linked to each different type of cancer,” says Michael Stratton, who heads the Cancer Genome Project at the Wellcome Trust Sanger Institute, in the United Kingdom. “We will understand all the genes that can generate cancer and have a list of mutated genes that may be potential drug targets.”

This vision epitomizes the promise of personalized medicine—treatment based on a patient’s individual genetics and other molecular factors. But it also suggests its challenges. It’s now clear that thousands, perhaps millions, of genetic mutations are capable of triggering cancer. Capturing that information is feasible with existing technology. But understanding it is a different matter. “We really don’t have the tools to take advantage of this information today,” says Tyler Jacks, director of the David H. Koch Institute for Integrative Cancer Research at MIT. And until those tools are available, no one will be able to answer the crucial question: whether a detailed map of a patient’s cancer will actually help that person live longer.

MAP MAKING
About seven years ago, Matthew Meyerson, a pathologist at the Dana-Farber Cancer Institute in Boston, began getting desperate calls and letters. Meyerson had recently discovered that lung cancer patients with a mutation in a gene for the epidermal growth factor receptor (EGFR) were likely to respond to specific drugs, and patients wanted to find out where they could get tested for the mutant gene. But no such test existed yet (one was introduced in 2005). Over the next few years, Meyerson made similar findings about more and more mutations. “Patients and physicians were both writing to us for help in getting these diagnostic tests,” he says. “But with few exceptions, they were not available.”

Despite the huge volume of knowledge generated by research labs over the last few years, only a handful of genetic tests are currently available to most cancer patients, and those typically test only one or two genes. So about five years ago, Meyerson and Levi Garraway, an oncologist and scientist at Dana-Farber, set out to create a more comprehensive test for cancer-linked mutations. The goal was to identify genetic signatures, encompassing hundreds of mutations, that would help physicians diagnose cancers on the basis of molecular traits rather than the way cells look under a microscope. Some genetic classification was already under way—breast cancer patients are routinely tested for markers such as the HER2 protein, which predicts who will respond to the drug Herceptin. But as scientists began discovering a much broader array of genetic mutations linked to cancer, oncologists needed a way to screen individual patients for numerous mutations.

This type of screen is necessarily expensive. Because scientists want to pin down how cancer cells differ genetically from normal ones, they must sequence the DNA from both healthy tissue and tumor tissue, doubling the cost of sequencing. To make matters even more complicated, tumor tissue itself is often a mix of normal cells and cancer cells. Scientists typically run healthy DNA through the machine 10 to 40 times to make sure they can piece together an accurate sequence, but Gordon Mills, chair of the department of molecular therapeutics at M. D. Anderson Cancer Center, in Houston, estimates that DNA from cancer tissue may need to be sequenced a thousand times in order to detect rare mutations that could make tumors resistant to drugs.

When Meyerson and Garraway began developing their test, in 2005, it cost more than a million dollars to sequence a human genome. So the researchers chose to limit their test to certain “hot spots”—regions in the genome that were known to harbor a high concentration of cancer-causing mutations. They also chose a relatively cheap technology, called mass spectrometry, to analyze DNA. It could detect mutations that had been previously identified, but it couldn’t uncover new ones. By 2008, however, a revolution in DNA sequencing technology had made it possible to read entire cancer genomes affordably, and cancer researchers quickly began such experiments. As the cost of the technology continues to plummet (sequencing a human genome now costs about $10,000 to $20,000), hundreds of cancer genomes are being sequenced in labs around the world, adding to the database of cancer-linked mutations.

The picture emerging from sequencing studies suggests that the genomics of cancer are even more complicated than scientists had supposed. A few years ago, researchers thought that about five to seven mutations were needed to trigger the uncontrolled cellular proliferation that defines the disease. But recent estimates are as high as 20 in some cancers, and on average scientists are finding five to 15 mutations involved. And the vast majority of newly discovered mutations are rare, occurring in fewer than 5 percent of specific cancers. Indeed, each newly sequenced cancer genome reveals mutations that have never before been seen.

BUILDING A FOUNDATION
Early in 2010, Garraway, Meyerson, and two colleagues from the Broad Institute—its director, genomics pioneer Eric Lander, and Todd Golub, an expert on cancer genomics—started Foundation Medicine with funding from a Boston-based venture capital group called Third Rock Ventures. The startup’s goal is to create a real-world, clinical-grade test that will help physicians determine not only whether someone has one of the growing number of mutations implicated in specific cancers but how severe that patient’s cancer is and which drugs it is likely to respond to. Foundation’s test will use sequencing technology to identify all the mutations in hundreds of genes that have been linked to cancer in previous studies.

Because the test encompasses so many genes, Foundation’s product will be entirely different from the single-gene tests in use today. These are fairly simple to interpret: if a patient has a particular mutation, he or she will probably respond to a certain drug. But analyzing the meaning of hundreds of mutations simultaneously is orders of magnitude more complex. To go through the genome of Laskin’s cancer patient, Steven Jones needed a large team with diverse expertise in medicine, oncology, genomics, and information technology, and the process took three weeks.

It’s not yet possible to analyze and interpret cancer genomes on a scale large enough to make this approach a routine part of treatment. But Foundation Medicine aims to take a step in that direction: the company will focus on a subset of genes, a somewhat easier task. Part of the company’s product will be a database that compiles the rapidly evolving scientific literature on a number of specific mutations and the way they affect a patient’s response to different drugs. The software will need to predict which mutations in a tumor cell’s DNA actually drive the cancer, and which are genetic mistakes of little consequence. And it will need to cope with mutations so rare or novel that at most a handful of studies offer any basis for predictions. (In such cases, researchers typically try to predict the impact of a mutation by looking at how it affects the structure and function of the protein the gene produces, and then examining the role this protein plays in the cell’s various signaling networks.) Along with the database, Foundation Medicine is building a user interface to help oncologists interpret information.

Alexis Borisy, a partner at Third Rock and Foundation’s acting chief executive officer, is careful to say that the software won’t tell physicians what to do. Rather, it will provide layers of information to help them make choices about which drugs and other treatments to try. Only a handful of genetically targeted cancer drugs are widely available to patients today, but Borisy contends that Foundation’s test will be useful even when it doesn’t point to any of those medications. Hundreds of targeted drugs are now in human testing, so patients who test positive for certain mutations could be directed toward appropriate clinical trials. In addition, some more traditional cancer drugs, which are currently prescribed without genetic testing, are known to act on specific molecular pathways. Physicians who find that their patients’ cancers involve these pathways could suggest trying these drugs.

Still, not everyone will benefit. According to Borisy, Foundation researchers who have begun evaluating an early version of the company’s test say their results suggest that about half the patient tissue samples analyzed would yield plausibly “usable” information, meaning that the analysis might suggest a particular class of drugs or better define the type of cancer. (The proportion will grow as pharmaceutical companies develop new targeted drugs.) The company plans to begin testing patients early next year, in collaboration with academic medical centers and pharmaceutical companies. As evidence builds that the test improves patient care, “we expect it to spread to the broader oncology community,” Borisy says.

Foundation Medicine’s test will initially read the DNA sequences of a limited set of cancer-linked genes, but the company expects to sequence the entire genome of patients’ tumor cells as soon as the technology becomes cheap enough. The cost of sequencing is quickly coming into line with that of other medical tests and cancer treatments; an MRI scan, for example, costs about $6,000, and cancer patients typically have several. Still, if experts like Gordon Mills are correct, it will take hundreds of sequencing runs to analyze a patient’s cancer DNA accurately enough for medical decision making. In that case, genome-wide sequencing of patients is at least several years off.

THE BURDEN OF PROOF
In 2001, entrepreneur Mark Levin, an early proponent of personalized medicine, made a grand prediction to Technology Review: “Over the next five to 10 years, we’re going to see an explosion of not only [diagnostic] tests, but the integration of tests and therapy for personalized medicine” (see “Medicine’s New Millennium,” December 2001).

Levin had bet big on that idea, founding Millennium Pharmaceuticals in the early 1990s; the drug development company wanted to create personalized therapies based on knowledge of the genome. The predicted explosion has so far been more of a trickle, but Levin, who went on to cofound Third Rock Ventures, is optimistic that the time is finally right. “When we founded Millennium, the technology was still primitive; there was not enough data, and not enough people in the industry believed in it at the time,” he says. “It’s been 20 years since this really all started, but it’s just now coming together where we can do this cost-effectively and make a big difference for patients.” (Millennium was acquired by Takeda Pharmaceuticals, Japan’s largest pharmaceutical company, for $8.8 billion in 2008.)

Levin believes that Foundation Medicine will finally realize his vision of personalized medicine in cancer treatment. Indeed, oncology is already more personalized than most areas of medicine. “Cancer treatment, more than other diseases, has been based on a molecular understanding of the disease,” says Michael Stratton. “And that is increasing all the time.”

But many of the questions that have plagued personalized medicine for the last 20 years remain unanswered. No one yet knows whether developing complex genetic profiles will help patients live longer, healthier lives. The answer depends on a number of factors: whether drugs exist to target an individual’s genetic quirks and whether such drugs work better than ones that aim at more widespread cancer-causing mechanisms. It may turn out to be practical to develop only drugs that tackle the more common mutations—say, those present in 5 percent or more of certain cancers. In that case, patients with rare mutations might be left behind. Or perhaps the most successful treatments will target molecular mechanisms common to many cancers, making comprehensive genetic testing meaningless.

By its very nature, personalized medicine is difficult to assess. In the case of Janessa Laskin’s patient, it is impossible to know whether the genomic information that guided her selection of drugs helped him live any longer than he would have otherwise. Scientists typically test new drugs and diagnostics by comparing the outcomes for patients who get these interventions and those who do not, but Laskin may never come across another patient with the same set of rare mutations.

Researchers and regulators will need to invent an entirely new way to evaluate treatments targeted at patients whose genetic malfunctions may be unique. “We’re not measuring an individual gene or one specific algorithm,” says Borisy. “We’re asking the broader question of how this set of information supports physician decision making. That requires a different approach to clinical trials.”

As an example, Borisy cites the small number of ovarian cancer patients who have mutations in the gene for EGFR, the same genetic mistake Meyerson linked to drug response in lung cancer several years ago. While several clinical trials have shown that specific drugs, known as EGFR inhibitors, work best in lung cancer patients with these mutations, similar trials would be impossible to conduct with ovarian cancer patients. “There are maybe only a thousand women in the U.S. in this situation. It would be decades before you could collect enough patients to have a typical clinical trial,” he says. “But if you are the patient, you want access to this drug.”

Emily Singer is the biomedicine editor of Technology Review.