Gene Expression Tests Foretell Breast Cancer's Future

See allHide authors and affiliations

Science  19 Mar 2004:
Vol. 303, Issue 5665, pp. 1754-1755
DOI: 10.1126/science.303.5665.1754

The first patient profiling tests based on gene expression hit the market early this year. With others soon to follow, they will test genomic medicine's power

One of the hardest decisions a breast cancer patient faces is whether to undergo chemotherapy. Most patients whose cancer has not spread to the lymph nodes are cured by surgery and the estrogen-inhibiting drug tamoxifen, but a small minority will go on to develop distant metastases and die. Chemotherapy reduces that risk but has its own dangers, including heart failure, leukemia, and life-threatening infections.

Doctors can't guide patients very well because they too are on the fence. A pathology examination and molecular markers broadly hint at whether a tumor may metastasize, “but for each individual, we don't really know,” says Clifford Hudis, chief of the breast cancer medicine service at Memorial Sloan-Kettering Cancer Center in New York City.

Help may have arrived from genomics, a field long on publicity but, until now, short on products for the clinic. On 26 January, Genomic Health, a private company based in Redwood City, California, quietly launched Oncotype DX, a testing service for determining a breast cancer patient's chance of developing metastases. The test is based entirely on the pattern of genetic activity in her tumor. “Our hope is that the data just help shed some more light on the real risk for these women,” says Genomic Health co-founder and CEO Randy Scott.

This somewhat understated goal masks an intense competition for breast cancer patient business. Agendia, a Dutch start-up company, launched a genomic profiling test in early March. At least two other companies are developing similar tests. Oncotype DX costs $3400, and the potential market is huge. Over 215,000 women will be diagnosed with breast cancer this year in the United States alone, and the Genomic Health test “might be relevant to up to a third of breast cancer patients,” says Peter Ravdin, a medical oncologist at the University of Texas Health Science Center in San Antonio.

But the new tests are reaching the market without having been validated in true prospective clinical trials. They haven't been subjected to Food and Drug Administration (FDA) review, either, because they're being marketed as a lab service rather than as diagnostic kits. How well the tests perform is critical not only for breast cancer patients, but as one of the first tests of genomic medicine.

Predicting trouble

Genomic Health has a low profile, but it has pedigreed executives. Scott was a founder of Incyte, a Palo Alto, California, genomics company marketing a popular gene database, and Chief Medical Officer Steven Shak shepherded Herceptin, the first monoclonal antibody for treating solid tumors, to market while he was at Genentech in South San Francisco. For its first clinical project, Genomic Health decided to create a test that could anticipate the course of breast cancer, selecting 250 genes for their presumed influence on cancer progression. Company scientists measured the genes' expression in tumor samples. Sixteen genes, as a group, strongly predicted which tumors metastasized; they and five “control” genes were ultimately chosen for the screen. Genomic Health then tested this gene panel in 668 stored breast tumors from a different trial and unveiled the “validation” results in December 2003 at the San Antonio Breast Cancer Symposium.

In the validation study, the test was much more accurate than current methods oncologists use to estimate metastasis risk. The investigators analyzed and classified gene expression patterns in tumors without knowing whether the person's cancer had spread. The “patients”—all of whom had been treated years earlier—were separated into low-, intermediate-, and high-risk groups. Only 6.8% of the patients in the low-risk group had gone on to develop distant metastases after 10 years, their files revealed, compared to 30.5% in the high-risk group.

Reading the genes.

A new test developed by Steven Shak (above) and colleagues uses RNA extracted from breast tumors to predict cancer spread.


But the test has limitations. The biggest: It doesn't identify which patients are likely to benefit from chemotherapy. Researchers assume that patients most at risk of metastases would benefit most from chemotherapy, but responses to treatment will vary, and “other sets of genes [might] predict a benefit from chemo,” says Hudis, who sits on Genomic Health's scientific advisory board. Genomic Health, like many other companies, is busy profiling genes involved in chemotherapy response and hopes to add those tests later. Oncotype DX “is really just the first step,” says Scott.

Other limitations include the fact that Oncotype DX is based on tumors that express the estrogen receptor, taken from patients whose cancer had not spread to lymph nodes. Fewer than half of all breast cancer patients fall into this category, raising a question about how well the test would work for others. What's more, the patients took tamoxifen, so the test might have been picking up in part on their response to that drug and might not work as well in patients taking other anti-estrogen drugs.

Oncotype DX did not undergo FDA review. As a “clinical laboratory reference service,” or “home brew,” it is exempt from the standard review FDA requires for diagnostic kits. But it was approved by the California State Licensing Agency, which regulates labs in the state. FDA review might have aired several issues. The tumors themselves “come from a study that was 20 years old,” points out Daniel Hayes, a medical oncologist at the University of Michigan, Ann Arbor. They might have changed over that time, he suggests. Or perhaps the sample was selective, collected only from patients with large enough tumors to have material left over after pathology tests. The lack of FDA review “is too bad,” says Edison Liu, executive director of the Genome Institute of Singapore.

But Hayes and others consider the evidence strong, based on what they heard at the San Antonio meeting. “If [Oncotype DX] is as good as it seems to be, I'm certainly going to suggest for some patients—those patients with stage one breast cancer—that this test could be a very valuable test,” says Ravdin.

The ultimate test of Oncotype DX may be on the way. The National Cancer Institute and Breast Cancer Intergroup of North America, a consortium of clinical trial cooperative groups, are discussing a prospective clinical trial. In addition to tracking metastasis and survival, the trial may settle the lingering question of whether the test predicts patient response to chemotherapy.

Bright line.

Colored dots representing activated genes highlight tumors that are likely to metastasize.


The challengers

Neither Genomic Health nor its competition is waiting for future trials. Agendia, based in Amsterdam, began selling its Web-based breast cancer profiling test, called Mammaprint, last month, according to Chief Scientific Officer René Bernards. The test costs €1650.

Agendia's test is very different from Genomic Health's. Bernards and colleague Laura van't Veer were working at the Netherlands Cancer Institute (NKI), which had an extensive bank of frozen tumor samples dating back to 1983. Bernards contacted Stephen Friend, a pediatric oncologist who founded Seattle biotech company Rosetta Inpharmatics, a pioneer in DNA microarrays, chips that light up to reveal the activity of thousands of genes. “I went to see Stephen and said, ‘Listen, I have a beautiful tumor collection here in Amsterdam, you have beautiful technology,’” Bernards recalls. “‘Let's get together and do this project on gene expression profiling.’”

The group used an approach that scans the entire genome. They generated expression data for 25,000 different genes from a group of 117 young Dutch women with breast cancer that hadn't spread to the lymph nodes and whose clinical course had been followed for more than 5 years after surgery. Ultimately, they identified a 70-gene expression signature to indicate good or bad prognosis.

The work was proof of principle for the then-controversial idea that individual tumors are programmed early to metastasize or not (Science, 14 February 2003, p. 1002). “Before 4 years ago, biologists assumed that there were … neutral small tumors that would become good or bad tumors later,” says Friend. But “sometimes very early tumors can be predestined to be bad players.” Friend and Bernards were the first to show that global genomic expression patterns could identify such tumors.

The group validated their gene panel in 295 tumors from NKI—a study published in The New England Journal of Medicine in December 2002 (Science, 11 April 2003, p. 238). Almost 56% of lymph node-negative patients with a poor prognosis signature developed distant metastases after 10 years, compared to 13% of patients in the good prognosis group. Unlike Genomic Health, Rosetta and NKI looked only at tumors from patients who had not yet been treated with tamoxifen or other drugs, so there's no chance that treatment skewed the results. And the Rosetta group didn't preselect its genes as Genomic Health did. “If you start out in a completely unbiased fashion, you come up with, I think, a more powerful set of genes than if you take your own best guesses,” says Bernards.

But the Rosetta-NKI study also has flaws. It, too, used stored tumor samples and case records. It tested only women under 53 years old at a single institution, raising doubts that the results can be safely extrapolated. And it incorporated its test group—the group it derived the 70 genes from—into its validation group, thus potentially inflating the results. “To me that's just not kosher,” says Hayes. Although the authors made statistical adjustments, Friend concedes that the study should have been designed differently. Two prospective trials of the Agendia test are under way, one in Europe and one in the United States.

Although Genomic Health and Agendia are the furthest along, at least two other companies, Celera Diagnostics in Alameda, California, and Arcturus Applied Genomics in Carlsbad, California, are developing breast cancer prognostic tests. Each company has a different technology platform, but the goal is the same: to predict which women's cancers will recur.

Looking for validation.

Genomic test results from stored tumors were checked against actual patient outcomes.


“There will be a lot of demand for these tools, generated by the patients themselves,” predicts Liu. But how doctors and patients will apply test results is far less certain. Bernards says that such tests may lead many women with a low-risk signature to forgo chemotherapy. “We expect that in Europe we can achieve a 25% to 30% reduction of chemotherapy,” he says. Ravdin notes that “aggressive chemotherapy causes a risk of 1% of something really bad happening.”

But American doctors are generally more prepared than Europeans to order chemotherapy for marginal cases. So it's possible that testing won't decrease overall chemotherapy use in the United States. “Actual chemotherapy use will probably stay the same,” Scott says. “It's just that you'll now do a better [selection] job.”

Few doctors are likely to rely wholly on the test results. “Most of us aren't going to be brave enough to withhold chemotherapy from a patient who otherwise looks like they need it, on the basis of this test, which has not yet been validated for the chemotherapy decision,” says Hudis. But if traditionally low-risk patients test for a “bad” tumor, the test could save lives by tipping these patients toward chemotherapy, Hudis points out.

For all the weaknesses of these tests—the validation questions, the absence of prospective trials and FDA review, and the inability to predict chemotherapy response—they provide a new, more personalized dimension of information to women agonizing over a life-or-death decision. “Right now, even if it's a little bit wrong,” says Hayes, “it's better than what we've got.”

View Abstract

Stay Connected to Science

Navigate This Article