Courtesy Timothy Triche, National Cancer Institute

Metastasis, the leading cause of cancer deaths, remains a poorly understood phenomenon. In early-stage lung cancer, for example, 25% to 30% of patients will succumb to the disease even though the tumors are small and detected early. The use of gene-expression analyses has allowed the opportunity to ask whether these high-risk stage I lung cancers can be distinguished from benign tumors so that these patients might receive more aggressive therapy.

This issue's Hot Papers attempted to identify the hallmark features of metastatic cells through gene-expression profiles obtained by microarray analysis. In 2002 David Beer and Samir Hanash at the University of Michigan, and others, described an exhaustive gene-expression profile of lung adenocarcinoma involving 67 early-stage and 19 late-stage tumors.1 In testing the tumors, they devised a panel of 50 predictive genes, including genes previously not known to influence survival. This enabled them to...


Data derived from the Science Watch/Hot Papers database and the Web of Science (Thomson Scientific, Philadelphia) show that Hot Papers are cited 50 to 100 times more often than the average paper of the same type and age.

"Gene-expression profiles predict survival of patients with lung adenocarcinoma," Beer DG, Nat Med , 2002 Vol 8, 816-24 (Cited in 187 papers, HistCite Analysis)"A molecular signature of metastasis in primary solid tumors," Ramaswamy S, Nat Gen , 2003 Vol 33, 49-54 (Cited in 232 papers, HistCite Analysis)

Both papers were preceded by back-to-back reports in 2001 by a Harvard University group led by Matthew Meyerson,3 and a Stanford University group led by Iver Peterson.4 These publications were the first to report that lung adenocarcinomas could be classified into histologic subtypes on the basis of their gene-expression profiles. Both Beer's and Ramaswamy's groups used adenocarcinoma samples from Meyerson's collection in their studies.

The interest in whole-genome profiling hasn't always had a lot of support, says Hanash, now at the Fred Hutchinson Cancer Research Center in Seattle. "The field struggled with the idea that you could profile anything, without hard evidence that the data are applicable beyond the initial studies."

In their study, Beer, Hanash, and colleagues correlated gene-expression profiles from lung adenocarcinomas to patient clinical outcome. They confirmed these findings by Northern-blot and immunohistochemistry analyses. Then, using the Massachusetts-based lung adenocarcinoma set, consisting of 84 lung adenocarcinomas representing stage I, II, and III tumors, they further validated their results by showing that their risk index similarly correlated with patient outcome.

Even they didn't anticipate such a high degree of validation. "We were surprised we could use an independent set of data and find correlation," says Beer. "That was kind of remarkable," agrees Hanash.

"What this has heralded is that, within a subset of adenocarcinomas, there are different genetic changes," says Bruce Johnson, director of thoracic oncology at Dana-Farber and a coauthor on Meyerson's study.3 "This has huge implications on what we know about tumors," he says. Furthermore, since these differentially expressed patterns stem from underlying genetic differences, "those mutations have an impact on selecting treatment and outcome," Johnson adds.

Leukemias and lymphomas are considerably easier to diagnose given the cell types involved, the stages of cellular differentiation, and telltale surface markers. But solid epithelial tumors are nowhere near as revealing. "In solid tumors, we just didn't have that degree of information at all," says Meyerson. But that scenario is now changing, he says. "The genomic era makes it possible to accelerate the process of discovering the different [tumor] subtypes and achieve classification of solid epithelial cancers."

But the really exciting aspect of the Beer study, Meyerson says, "was that they built a model based on their data set, and they were then able to validate their model on our data set."

Currently, Beer and colleagues are in the process of completing a large study of 22,000 genes from 550 lung adenocarcinomas. "We're doing the most comprehensive analysis of lung adenocarcinoma to capture the heterogeneity in this tumor type," he says.



© 2002 Nature Publishing Group

Gene-expression patterns determined using agglomerative hierarchical clustering of 86 lung adenocarcinomas against 100 survival-related genes. The expanded area shows genes with extremely high expression levels for some tumors. (From D.G. Beer et al., Nat Med, 8:816–23, 2002.)

Ramaswamy and colleagues at Dana-Farber analyzed the expression profiles of 12 metastatic adenocarcinoma tumors of lung, breast, prostate, uterine, ovarian, and colorectal origin, comparing them to 64 primary adenocarcinomas representing the same tumor types from different individuals. They identified 128 genes associated with metastases, providing a distinct molecular signature that was also predictive of patient outcome. They further refined that signature to just 17 genes.

"The idea was that perhaps we'd find global differences between metastatic and primary tumors," says Ramaswamy, co-author on the study. "When we did analyze a relatively small set of metastatic and primary tumors, we were able to identify a gene-expression signature differentially expressed," he says. But the breadth of that signature surpassed even their expectations. "It was surprising to us because we didn't expect to find a signature that would be generically prognostic across different tumor types."

Furthermore, they found that some primary tumors were expressing genes characteristic of the metastatic profile. "One-third had the hallmark [metastatic] signature," he says. The remaining two-thirds demonstrated the non-metastatic profile.

Since gene-expression profiles are derived from whole tumors, it's hard to detect differences arising in a minority of cells within that tumor, he says. "We reasoned that primary tumors expressing the metastatic signature might have different clinical outcomes for patients." To address this, they then compared those primary tumors displaying the metastatic signature and correlated those profiles with patient outcome.

What they found was a "signature broadly applicable to different cancers, associated with a higher or lower frequency of metastases from the given primary tumor," says Meyerson. "They identified this signature both in our data set and in the [Stanford group's] data set."

"We found statistically significant differences between patients with primary tumors that lacked the metastatic signature and patients with tumors that expressed the signature," says Ramaswamy. These differences confirmed what they suspected: Patients whose primary tumors displayed the metastatic signature were more likely to die sooner than those who did not. "That was a very interesting and surprising result," says Meyerson.


They were able to identify the primary tumors most likely to metastasize, suggesting that some primary tumors are preprogrammed to metastasize. They found that the bulk of cells in these primary tumors already exhibited the metastatic signature from the outset, leading them to believe that metastasis is a relatively early event. The result is a cancerous time bomb that can be tripped at any time. Their theory stands in stark contrast to the prevailing view of metastasis as a late event that arises when rare cells in a primary tumor acquire metastatic potential through a series of mutations. Many groups are now trying to address this experimentally, says Ramaswamy.

David Sugarbaker, chief of the thoracic surgery division at Dana-Farber, says that physicians have determined clinically that primary tumors have a varying number of cells that have metastatic potential, and that this varies from tumor to tumor. Finding that a molecular signature distinguishing metastatic lesions and primary tumors fits intuitively with these clinical observations, he says.

The significance "is that for the first time, it provides a genomic basis for an observed clinical phenomenon," Sugar-baker says. "They have demonstrated that primary adenocarcinomas have a distinct genomic signature, suggesting different clinical behavior."

Ramaswamy says that he and his colleagues are focused on trying to understand the genetic basis of the metastatic signature and how these genes may be involved in driving the metastatic process mechanistically.

"Research on metastasis has picked up tremendously over the last two to three years," says Ramaswamy. Taken with other such studies that are increasingly moving toward clinical testing,5 genomic information about a tumor may reveal the likelihood that patients will develop metastasis or a recurrence of their cancer.

"Eventually, we're going to tease out that specific vulnerability of the metastatic tumor and exploit those differences for therapeutic gain," says Sugarbaker. Greater understanding of a broad number of tumors, their genetic signatures, and gene-expression differences, may lead to an understanding of the underlying pathways responsible for metastasis, which in turn may provide further targets for treatment, he says. "For research and therapy, it's a prime focus for intensive investigation."

Interested in reading more?

Magaizne Cover

Become a Member of

Receive full access to digital editions of The Scientist, as well as TS Digest, feature stories, more than 35 years of archives, and much more!
Already a member?