The individual fates of the 1.3 million Americans diagnosed with cancer this year will be largely decided by one simple factor: at what stage was the disease spotted?
Ovarian cancer offers a fearsome example. Because of its vague symptoms, it is usually ignored or misdiagnosed, sometimes for years. Eighty percent of patients don’t find out they have it until it’s spread beyond the ovaries. At that point, it is usually incurable; only one patient in three survives five years after diagnosis. On the other hand, surgery can cure 90 percent of patients whose cancer is detected while still confined to the ovary. Even notoriously lethal cancers of the lung and pancreas are anything but a death sentence, if caught early enough. “Cancers can almost always be cured by simple, classical surgical techniques, if they’re detected early,” says Bert Vogelstein, a molecular geneticist at the Howard Hughes Medical Institute at Johns Hopkins.
The problem, of course, is that cancers, which begin with just a few deviant cells, are by their very nature hard to diagnose early. In the last few years, though, a new method has emerged that promises to deliver simple blood tests that identify the telltale molecular profiles of various cancers easily and accurately. It has long been known that cancer leaves traces in the blood, but these hints are confusing and ambiguous. “Blood is perfusing through every tissue in your body, 60 times a second, beating through it,” says National Institutes of Health pathologist Lance Liotta. “You’d imagine there’d be fragments of what’s going on in every cell, and every tissue, ending up in the circulation.”
Routine screening for most types of cancer does not exist today. With a few important exceptions, cancer screening has been a failure; even widely endorsed methods, such as breast and testicular self-exams and mammography, have come under fire. These tests miss too many cancers and pick up too many “false positives,” suspicious findings that turn out to be benign. The result: much anxiety, many unnecessary biopsies and exploratory operations-and relatively few cures.
The few existing blood tests for cancer aren’t any better. Take, for example, PSA, the test for prostate cancer, and CA-125, for ovarian cancer. Both are named after the proteins they look for, and both are terrible. PSA, which the American Urological Society recommends offering to every man over age 50, “misses about a third of patients with cancer,” says David Sidransky, a cancer researcher at Johns Hopkins, “and it falsely calls patients that are positive with PSA as having cancer about a third of the time.” For ovarian cancer, the picture looks even worse. Only about half of patients with early-stage ovarian cancer show elevated CA-125 levels, and the rate of false positives is high, because some benign conditions cause overproduction of the protein. As a result, CA-125 is only approved for monitoring the progression or recurrence of ovarian cancer, not for screening.
George Wright, a cell biologist at the Eastern Virginia Medical School in Norfolk, has spent his entire research career of more than 40 years trying to find better diagnostic markers for early cancer detection. He has been understandably frustrated. “Maybe a protein biomarker that [we] discovered would detect 20 to 30 percent, not 100 percent, of the cancers,” says Wright. Equally useless are markers that falsely diagnose healthy people as ill, regardless of how many cancers they reveal. To be useful for routine screening, a blood test for cancer has to be almost perfect.
In 1998, Wright read an article in a trade magazine about a biotech company in California that was looking at protein patterns; suddenly, that almost-perfect test seemed possible. The Fremont-based company, Ciphergen Biosystems, claimed that a few drops of blood could reveal hundreds of proteins simultaneously, when analyzed with a standard laboratory instrument called a mass spectrometer. The proteins, though, aren’t explicitly identified; instead, the machine prints out a pattern of sharp peaks and troughs, each peak representing the blood level of some unknown protein. Ciphergen thought that comparing the results from cancer patients to those from healthy subjects could aid the search for cancer biomarkers, because many proteins are overproduced in tumor cells. Properly identified and studied, those proteins could lead to better cancer tests. But Wright had an even bolder idea: the patterns themselves might provide a ready-made signature for cancer. The strategy of using patterns, if it worked, would shave years, even decades, off the time required to create a test, since it would eliminate the need to identify the individual proteins and perfect means to detect them.
Wright also suspected that tests based on cancer protein profiles could fingerprint cancer more accurately than any one protein. They “would be more effective than anything that was available,” he recalls telling Ciphergen’s CEO, who was skeptical. But Wright bought a Ciphergen machine in January 1999 and began looking for incriminating patterns himself.
Sketching a Solution
Wright was not alone. Around the time he was beginning his work, the same approach occurred to NIH’s Liotta and U.S. Food and Drug Administration researcher Emanuel Petricoin. Petricoin and Liotta knew that cancer, on the level of the cell, generates a cacophony of changes, both in the tumor tissue and in the normal tissue surrounding it. This complexity appears impenetrable. But the duo thought they could exploit that very complexity to generate a cancer fingerprint from traces of the disease circulating in the blood.
Like Wright, Petricoin and Liotta used a Ciphergen system to generate protein profiles from blood samples. Their early attempts to find cancer patterns failed, though, because they were simply trying to juggle too much information. Then, in June 1999, a solution appeared. Petricoin and his friend Peter Levine, a Maryland lawyer with a background in data analysis, were chatting about the problem over brunch; Levine suggested using pattern recognition algorithms to make sense of the massive amount of data. Levine, who had considered using such algorithms to analyze stock market trends and commodities trading, sketched out the cancer idea on a napkin. “In about five minutes, we both realized this would be a really fascinating approach,” Petricoin recalls.
So they tested it, together with Ben Hitt, a software engineer who borrowed the necessary algorithms from artificial-intelligence theory. In fact, cancer patterns did emerge, and in 2000 Levine and Hitt founded Correlogic Systems to develop blood tests for cancers. In early 2002, the researchers published results in the British medical journal Lancet, showing they could use a specific protein pattern to spot ovarian cancer. Their test correctly identified 50 out of 50 women with cancer and correctly scored negative for 63 out of 66 unaffected women. Later given the name OvaCheck, it promised to be the first blood test accurate enough to be used for general ovarian-cancer screening. By the end of 2002, Correlogic had licensed OvaCheck to two major commercial laboratories and planned a 2004 product launch.
Meanwhile, Wright’s group in Virginia was also pushing ahead. Using a different algorithm, Wright and Eastern Virginia molecular biologist John Semmes showed that a protein pattern could distinguish prostate cancer from a common noncancerous condition, benign prostatic hypertrophy, in 25 out of 30 cases. The PSA test, by contrast, is unable to distinguish the two conditions.
While Wright stresses that the results are preliminary, the technology continues to inch toward commercialization. A large initial trial across many medical centers should finish in about a year; a final validation trial will conclude, if all goes well, in 2006. And Eastern Virginia has already licensed its technology to an undisclosed company for eventual development into a full-blown diagnostic test.
Given the stakes, a cautious approach makes sense. In fact, Correlogic, which has advanced its technology much more aggressively than the Eastern Virginia group, suffered a critical setback last winter. In September 2003, Correlogic announced at the annual meeting of the Ovarian Cancer National Alliance, a patient advocacy group, that OvaCheck would be on the market early in 2004. But in February, marketing plans went on hold when the FDA notified Correlogic and its two partners that the test might need regulatory approval-something not usually required of diagnostic tests marketed by clinical laboratories. Now the company and the FDA are working out a plan for moving forward.
The biggest mistake was to announce that you have a blood test,” says Eastern Virginia’s Semmes, who notes that Correlogic hadn’t even finalized its diagnostic pattern, at least in published form, at the time it made the announcement. “That claim, I think, hurt the field tremendously.” Even Petricoin and Liotta have distanced themselves from the test they helped originate.
The field must also cope with a scientific backlash against the whole idea of protein pattern diagnostics. Eleftherios Diamandis, a cancer expert at the University of Toronto in Ontario, calls the original Lancet ovarian-cancer paper “complete junk.” Diamandis contends that the patterns don’t actually represent proteins produced by cancer cells. “The technology will fail, because the molecules they monitor are not the correct ones,” says Diamandis. “I don’t think mass spectrometry, the way they perform it, is sensitive enough.” Instead, he urges, identify the proteins behind the peaks first, to make sure they’re really cancer proteins, and then develop standard tests to detect them. “Then we can put them all together, and we can make a reasonably good clinical diagnosis,” Diamandis says.
Petricoin firmly believes that the instruments are following proteins from cancer but concedes that proof can come only from rigorous trials. “The only way to prove it’s real or not is by validation, like any biomarker,” he says. The effort is worth it, he adds, because generating patterns is relatively simple, while identifying proteins and translating that knowledge into a useful lab test could take years, even before clinical trials. “The debate about whether or not it’s critical to identify the particular proteins or other molecules that make up a pattern, that’s [arguing] how many angels dance on the head of a pin,” agrees Levine. “If what you can do in the near term is develop diagnostics that will save lives, to me that’s the beginning of the end of the discussion.”
Testing the Future
Petricoin and Liotta, also undeterred by critics, are moving steadily forward, albeit separately from Correlogic. They’re assessing their own ovarian-cancer test in a clinical trial for women in remission, to detect return of the disease. They intend to submit the method for FDA review and to license the related technology nonexclusively to any company interested in offering it.
The two scientists are also preparing similar tests for pancreatic, lung, and prostate cancers. They envision a future in which a small blood sample, periodically drawn in the doctor’s office, will reveal a complete image of the current disease status of the entire body. “Our goal is to show that in fact you can come up with a protein pattern that can discriminate disease and for the [National Cancer Institute] to take that all the way to FDA approval,” says Petricoin.
But the first diagnostic test based on protein profiling will probably be an ovarian-cancer test from Ciphergen, the company whose machine started it all. Ciphergen is not using a pattern per se but is instead sticking with its initial, more conservative approach: using protein patterns to find markers that are then individually identified and validated for their ability to distinguish cancer from noncancer.
The company is fashioning tests that use mass spectrometry to detect these specific markers, as opposed to the overall protein pattern. Ciphergen is also working on pancreatic- and prostate cancer tests, but its ovarian-cancer test-based on three protein markers-is the most advanced. “Our goal is to commercialize the test by the end of this year or early next year,” says Gail Page, president of Ciphergen’s diagnostics division.
Whether it will open the floodgates for similar tests depends on how well it performs. But Ciphergen is committed to using protein profiling as the basis for new and better cancer diagnostics. “We believe it will be the wave of the future,” says Eric Fung, Ciphergen’s director of clinical affairs and one of the originators of the ovarian-cancer test. “There is a transition from single markers to multiple markers, andsomeday it will evolve to patterns,” he says. “It’s my personal view that we may end up there, but we’re not there yet.”
Many enterprising scientists and companies, however, are betting that patterns will be ready for use within just a few years. And they expect patterns to diagnose cancer earlier, more accurately, and more reliably than a limited set of known markers like Ciphergen’s, however well chosen. Wright, for instance, though now retired and playing an advisory role, still continues his quest for an accurate cancer test based on pattern recognition. Four decades of failure have taught him to be cautious, but he can’t hide his excitement. “It will take several years for us to know whether this can be definitely proven to be useful,” he says. Still, he adds, “It’s very exciting, highly promising.”
Tests based on protein patterns, if they work, could help to save millions of lives. But as Wright and other cancer researchers well know, they aren’t the last word in cancer diagnosis. “There’s no magic elixir,” says Petricoin. “Nothing’s going to replace a smart doctor working with a patient.”
How AI is reinventing what computers are
Three key ways artificial intelligence is changing what it means to compute.
These weird virtual creatures evolve their bodies to solve problems
They show how intelligence and body plans are closely linked—and could unlock AI for robots.
We reviewed three at-home covid tests. The results were mixed.
Over-the-counter coronavirus tests are finally available in the US. Some are more accurate and easier to use than others.
A horrifying new AI app swaps women into porn videos with a click
Deepfake researchers have long feared the day this would arrive.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.