Using Genomics to Diagnose Breast Cancer
DURHAM, N.C. – Researchers at Duke University have developed a new statistical approach to classifying breast cancers based on profiles of their gene expression.
Their analysis of the activity of genes in breast tumor samples, as revealed by a laboratory tool known as a “DNA microarray,” demonstrated that researchers could discriminate the tumors’ status regarding two clinically important factors — estrogen receptor status and lymph node status.
Tumors whose cells show the presence of estrogen receptors are more likely to spread aggressively, as they respond to growth-inducing estrogen and progesterone. Also, the ability to predict whether breast tumors either have spread to lymph nodes, or have the potential to do so, constitutes the single most important risk factor in the metastasis of the cancer, the scientists noted.
The scientists’ report on the success of their statistical models in predicting tumor characteristics is reported in the Sept. 25, 2001, issue of the Proceedings of the National Academy of Sciences (PNAS). The study is posted on the PNAS Web site.
While the findings are preliminary and the approach new, Joseph Nevins, the study’s senior author, said he and his colleagues believe the technique eventually will lead to new diagnostic tests that can predict the status of tumors with considerable precision, enabling improved diagnosis and treatment. Nevins is a Howard Hughes Medical Institute investigator at Duke Medical Center and interim director of the Center for Genome Technology. The center is one of five research centers that are part of the Duke Institute for Genome Sciences and Policy (IGSP).
“This technique goes beyond the standard practice of visually examining a tumor and deciding on treatment and outcomes based on that examination,” Nevins said. “Since the visual information that guides a pathologist’s diagnosis is largely determined by the gene expression within the tumor, our ability to directly analyze gene expression on a very large scale, and ultimately to analyze every gene that is expressed in the tumor, affords the opportunity to diagnose tumors with much greater detail than now possible.”
The study was funded as a pilot IGSP project, and is also supported by the Duke Specialized Program of Research Excellence (SPORE) in Breast Cancer, the Early Detection Research Network, and the Duke Comprehensive Cancer Center.
Gene profiling using microarrays has been used before to analyze and classify such cancers as leukemias and lymphomas as well as breast cancers, the scientists said. However, the Duke study goes beyond this past work by developing predictive analyses of breast cancer clinical status, an important aspect of the future use of this technology in the clinic.
A DNA microarray — also known as a “gene chip” — allows researchers to examine thousands of genes from a single tumor sample.
In using the chips to measure gene activity, the scientists extracted messenger RNA (mRNA), the result of expression of a tumor’s genes, from each of the tumors. They then applied the mRNA from each tumor to a chip where the mRNA transcribed from a particular gene interacts with the gene sample on the microarray. The scientists use fluorescent markers on the mRNA to measure the level of mRNA, thus “reading out” which genes are actively expressed in which tumor.
The chips used by the Duke researchers were commercial microarrays, which consist of about 7,000 human genes. In their analysis, they concentrated on 100 genes whose activity maximally reflected the outcomes in the tumors. They used their statistical analytical approach to analyze the gene expression profiles of 49 tumor samples previously tested to be either positive or negative for estrogen receptors. These gene expression profiles revealed clear differences in the patterns of gene expression in breast tumors that could predict the estrogen receptor status of the tumors, Nevins said.
“The determination of estrogen receptor status is an important aspect of breast cancer diagnosis because of its role in promoting tumor growth. It also has implications for therapies. Although an analysis for estrogen receptor can be done now, the use of gene expression analysis provides much more detailed information about the nature of estrogen receptor status in these tumors,” Nevins said.
The PNAS study also reported the results of gene profiling of breast tumors to predict their lymph node involvement. In their profiling, the scientists compared tumors that had spread to lymph nodes at the time of diagnosis to those that had not. They found that statistical analysis of the profiles suggested the potential to classify tumors’ lymph node status, although the accuracy was less than that of the estrogen receptor analysis.
According to Nevins, DNA microarray analysis can be compared to examining the individual pixels of a digital photo rather than only being able to view the overall scene. Instead of examining a tumor under a microscope, one day physicians will likely have the technology to examine the individual genes of a tumor, he said.
“Large scale gene expression information holds the promise of improved clinical diagnosis and treatment strategies, but it depends on developing statistical tools needed to analyze the data,” Nevins said. Also, the scientists said, the ability to predict lymph node status using gene profiling could in many cases obviate the need for the extensive and risky surgery involved in removing lymph nodes.
Mike West, director of the Duke Institute of Statistics and Decision Sciences and lead author of the study, developed and conducted the statistical analysis in the study in collaboration with Nevins and other researchers on the project.
“The opportunities for these kind of data to aid clinical decision making has been recognized for some years, but the technology to develop the analytic methods needed to put it into practice hasn’t been available,” West said. “Our findings are extremely encouraging.
“In addition to isolating patterns of gene expression that accurately predict estrogen status of many tumors with high precision, the methods identify additional aspects of gene expression that characterize tumors that are not accurately predicted, so providing clues for further biological studies. We are now embarking on a much larger study and are currently refining and developing the analytic methods needed to extend our capacity to understand tumor genomics and improve the precision with which we can classify cancers,” he said.
Jeffrey R. Marks, a co-author of the study and director of the Duke Breast Cancer SPORE, said that analysis of gene activity levels in cancers constitutes only one set of parameters that can be used to characterize them.
“For a truer understanding, we will have to look at other data, such as protein levels, the details of an individual’s immune system and the natural genomic variation among people. Even though the gene activity data are incredibly rich and detailed, they are very one-dimensional and we haven’t yet completed the entire puzzle,” Marks said.
According to the researchers, the statistical methods for breast cancer gene activity analysis likely will not be perfected and clinically available for several years. However, they said, such tests to assess the potential of tumor spread and to guide treatment likely will become part of the routine care for cancer patients.
“Genomics is all about the promise and hope that one can precisely identify and classify a disease state and treat that disease with a much more precise and individualized therapy,” Nevins said. Also, the scientists said, such gene profiling offers an opportunity to identify the specific metabolic pathways underlying particular kinds of tumors, leading to a better basic understanding of breast cancer.
The IGSP, established in 2000 with $200 million in institutional funds, represents Duke University’s comprehensive response to the broad challenges of the genomic revolution. IGSP activities are organized through five research centers: the Center for Genome Technology, the Center for Human Genetics, the Center for Human Disease Models, the Center for Bioinformatics and Computational Biology and the Center for Genome Ethics, Law and Policy.