Genomic Test Predicted Breast Cancer's Spread to Lymph Nodes With 90 Percent Accuracy in Most Women, Duke Study Showed
DURHAM, N.C. -- Using a woman's genetic profile alone, a team of researchers has been able to predict, with 90 percent accuracy, whether a breast cancer tumor has extensively spread to her lymph nodes. Cancerous lymph nodes are critical in determining a woman's long-term survival because they imply a more aggressive tumor, said the researchers from Duke University and Duke University Medical Center.
The researchers also accurately predicted -- in a subset of women with none or just a few cancerous lymph nodes -- the chances of their cancers recurring within three years.
The researchers achieved such a high degree of accuracy by developing a novel gene-profiling method that analyzes large clusters of genes, then statistically interprets the genetic information to produce a personal risk profile for each woman.
Pending a larger clinical trial involving hundreds of women, the new genomic analysis could ultimately enable doctors to select the best treatment for each particular woman instead of prescribing chemotherapy across the board, said Mike West, Ph.D., professor of statistics and decision sciences at Duke University.
Results of the study, a collaboration between the Duke Institute for Genome Science and Policy (IGSP) and the Koo Foundation Sun Yat-Sen Cancer Center (KF-SYSCC) in Taipei, are published in the May 10, 2003 issue of The Lancet. The research was funded by Synpac–NC in Research Triangle Park, N.C.
"Our model is the clearest example to date of a step toward personalized medicine," said lead author Erich Huang, M.D., who will begin his surgical residency this summer at Duke University Medical Center. "You don't want to tell a woman only that she resides in risk category A or B, but rather, that we can predict her personal risk level based on her unique genomic profile and clinical parameters."
Currently, doctors use cancerous lymph nodes as the single most important factor in assessing a woman's long-term outcome and treatment needs. This measure and other clinical risk factors are imperfect diagnostic tools, as women with few or no cancerous lymph nodes can still relapse within a few years, and a percentage of women with very invasive tumors can be effectively cured on the first course of treatment, said Huang.
The Duke/KF-SYSCC analysis used gene-chip technology to measure the activity of thousands of genes that determine a tumor's behavior and its response to therapies. Scientists then applied a rigorous statistical analysis to the genetic information to develop genetic "signatures" that defined risk levels and likely prognoses.
Each woman's disease course and post-surgical follow-up were documented in detail, so the researchers were later able to verify the accuracy of their predictions against each woman's eventual outcome.
"Our approach is unique and innovative because it utilizes multiple collections of gene expression patterns that we call 'metagenes,'" said West. Each metagene is a cluster of perhaps 50 to 100 genes that have related characteristics. The scientists use statistical models to evaluate the collection of metagenes and identify those that define each tumor most effectively. Each metagene is then assigned a predictive value. Taken together, the multiple metagenes provide a broad picture of interlocking genomic patterns that characterize a woman's unique risk profile.
In fact, the Duke/KF-SYSCC team achieved a 90 percent accuracy rate in predicting outcomes of women whose genes clearly placed them in high- or low-risk categories. In contrast, the latest gene profiling tests developed elsewhere predict disease course in broad terms that reflect group differences, but they do not directly customize risk assessments to individual patients, said West.
Despite the team's success in accurately predicting outcomes for most women, they found that genetic profiles of a few women were contradictory -- meaning some genes suggested low-risk while other genes suggested high-risk. Or, in some cases, clinical risk factors contradicted genetic profiles. This ambiguity led to less accurate predictions for these women, the study showed. Such cases present a challenge to the researchers as they strive to refine and perfect the approach, said co-author Joseph Nevins, Ph.D., a Howard Hughes Medical Institute Investigator and professor of molecular genetics and microbiology at Duke University Medical Center.
"Our future studies will aim to refine the analyses to eliminate this ambiguity and thus improve our ability to accurately predict each patient's outcome," said Nevins. "Our goal is to develop an even sharper picture of a woman's status based on multiple aspects of genetic information and her clinical parameters."
With a clearer picture of the likely outcome, physicians will be able to treat each woman with the most appropriate therapy -- whether highly aggressive or more innocuous -- for her particular case, said Nevins.
While the Duke/KF-SYSCC approach is unique, it relies on an often-used technology for analyzing genes called a gene chip. This technique measures the relative abundance of the myriad genes in tumor tissue by measuring each gene's copying mechanism, called messenger RNA (mRNA.) Scientists process mRNA from a tumor and label it with fluorescent tags. The fluorescent mRNA then binds to its complementary DNA sequence on the gene chip. By measuring the resulting level of fluorescence when each gene is scanned with a special light, scientists can tell how much mRNA -- and hence how much of each gene transcript -- is present. Genes that are missing, or are overly active, are pinpointed as potential culprits in the disease process.
In particular, the current study revealed which metagenes are associated with lymph node metastasis and which metagenes are useful in predicting recurrence, said the researchers. The gene subsets that underlie these two events are different though related, suggesting that different but potentially interacting biological processes promote these two breast cancer traits.
"Each gene gives you one dimension of the tumor, so multiple genes give you multiple dimensions," said Andrew Huang, M.D., professor of medicine at Duke and president of the Koo Foundation-Sun Yat-Sen Cancer Center. "We're sorting through thousands of genes, so you get an incredible wealth of information that has the capacity to predict clinical outcomes."
Furthermore, the researchers can home in on defective genes or clusters of genes, dissect faulty metabolic pathways, and obstruct the progression of those particular pathways, said Huang.
Genomic data will not replace traditional clinical data but rather, will add significant detail to the clinical information, especially if it is ambiguous, the study found.
As the next step, researchers are continuing to develop and refine their statistical model so that the latest scientific technologies can be incorporated into their existing risk assessment model. The goal is to make their model "technology-independent" so it can utilize new genetic assays as they become available.