Researchers Profile Cancer Cells Through Gene Expression

For this article, Jim Kling interviewed Louis M. Staudt, senior investigator, Center for Cancer Research, National Cancer Institute; and Michael Bittner, a researcher with the National Human Genome Research Institute. Data from the Web of Science (ISI, Philadelphia) show that Hot Papers are cited 50 to 100 times more often than the average paper of the same type and age. A.A. Alizadeh et al., "Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling," Nature, 403:503-11, Feb 3, 2000. (Cited in 351 papers) M. Bittner et al., "Molecular classification of cutaneous malignant melanoma by gene expression profiling," Nature, 406:536-40, Aug. 3, 2000. (Cited in 107 papers) ©2001 MacMillan Magazines Ltd.Melanoma Exposed: Variation in biological properties of melanoma clusters. a-c, shows a representative member of the major melanoma cluster (UACC-1022). d-f, shows a sample falling outside of the major cluster (M93-047). (Reprinted with permission from Nature, 406:536-40, Aug. 3, 2000.) Cancer is not just one of the most difficult diseases to treat, it is also exceedingly difficult to classify because of its numerous developmental pathways. Morphological characteristics have been used to classify cancers for years, but advances in genetics have clinicians and researchers salivating for the chance to get detailed gene expression profiles of individual cancers that could help them diagnose and treat the disease. Courtesy of Louis M. StaudtLouis M. Staudt The trouble is, useful gene expression profiles of tumors have been slow in coming. These profiles provide reams of data about what cell types express which genes, and in what quantities, but because the techniques and protocols of cDNA expression are still being perfected, the results generated by different labs usually don't agree with one another, much less correlate to disease outcomes in patients. "Unless you have a diagnosis that changes the management of a patient, it is probably not going to gain widespread acceptance," says Louis M. Staudt, senior investigator, Center for Cancer Research, National Cancer Institute. In these papers, one coauthored by Staudt and the other by Michael Bittner of the National Human Genome Research Institute (NHGRI), researchers have broken through the barrier and established relationships between gene expression profiling and the biology and clinical outcome of diffuse large B-cell lymphoma and cutaneous malignant melanoma, respectively. Bittner's team accomplished the task by using strict statistical analysis of expression profiles, while Staudt's group relied more on an understanding of the underlying biology to make sense of otherwise mystifying patterns of global gene expression. (See "Roll-Your-Own Microarrays") Jeffrey Trent, director of intramural research at NHGRI and team leader of Bittner's group, led a melanoma study in which they used a cDNA microarray and organized the data using hierarchical clustering to group samples according to similarity in gene expression. The team selected a subset of genes that had the best measurement data, and then ran the samples through a series of calculations. The first calculation determines which of the 31 chosen pairs are most similar and links them together. The gene expression profile of those two samples was then averaged, and the average was substituted for the two samples, leaving 30 samples. The calculation was repeated, finding the next two most closely related samples. If one of the samples in the new pair was the average of a previous pair, the new sample was linked to the existing group. If two new samples were linked, they form a more distantly related group. Eventually, the technique produces a hierarchical tree of related samples. Avoiding Problems Courtesy of NHGRIMichael Bittner One problem that cropped up in the method was the frequent difficulty in judging the boundaries that separated groups. To get around that problem, "we embedded in the experiment a series of samples of melanoma and other cell lines picked for phenotypic similarity ... so we had some measures of similarity that we thought would give us useful classes," says Bittner. Of 31 melanoma samples used, 19 turned out to be as similar to each other as related controls were to one another, prompting Bittner and his colleagues to group those 19 into their own class. Other controls accounting for measurement noise confirmed the statistical significance of the relationship. "Most people at that time ... weren't applying any kind of tests to see how robust their clustering was," adds Bittner. The related controls turned out to be the critical step that allowed them to discern the class of 19, but another important moment came when Paul Meltzer, a senior investigator in NHGRI's cancer genetics branch, compared the expression patterns from this experiment with samples from associate director Mary Hendrix at the Holden Comprehensive Cancer Center, University of Iowa. Hendrix provided samples of an invasive uveal melanoma, which forms on the back of the eye. The team then tested the uveal melanoma samples for expression of a set of genes that were consistently different in the group of 19 as compared to the rest of the samples. In this subset of genes, the uveal samples had a gene expression profile that was almost an inverse of the profile for the 19 related melanomas. That finding gave Bittner's team an idea: If the expression phenotype was the inverse, might the invasive phenotype be inverted as well? They tested their theory in vivo and indeed found the 19 to be noninvasive. "I don't know of anyone who predicted a phenotype based on an expression pattern before," says Bittner. "It was a bit of a surprise to see that you could explain a bit of what you saw so directly." The work was also "a movement away from methods that only presented views of similarity (from gene expression profiles), to performing analytic methods that quantitated the strengths of similarity and dissimilarity," explains Bittner. Since the paper's publication, Bittner's team has identified, but not published, the gene that is responsible for the invasiveness. Determining the Cell of Origin Staudt's team, meanwhile, was working with diffuse large B-cell lymphoma (DLBCL), an aggressive malignancy of mature B lymphocytes that is the most common subtype of non-Hodgkin's lymphoma. There are two broad classes of DLBCL patients: 40 % have prolonged survival rates and the rest don't. The work "was based on my strong belief that the cell of origin of a tumor would have an influence on the clinical outcome," recalls Staudt. DLBCL cells come from a cell somewhere in the B-cell differentiation pathway. Previous studies had shown that DLBCL lymphomas, like most non-Hodgkin's lymphomas, bear mutations indicative of the hypermutation that generates antibody diversity. This follows exposure to an antigen that matches the B-cell's surface antibody. That process occurs in the germinal center, a microenvironment made up of B-cells and T-cells. The team conducted an expression analysis from 96 normal and malignant lymphoma samples, and then used a hierarchical clustering algorithm to view the results. Although the germinal center appeared important, "every time we produced clusters based on all the genes, there was no clear pattern," says Staudt. The key decision was made to deliberately subtract out genes involved in cell proliferation, and the host response to the tumor, and focus on the genes that were left. After that, the samples fell into the two categories. "Our problem had to do with competing biological influences in the same sample, and we had to separate out those influences so we could understand [them] one at a time," recalls Staudt. One class of lymphoma (since designated by Staudt as GCB) had all the hallmarks of the germinal center, suggesting that the cells of that type originated there. The second class (also dubbed by Staudt, now known as ABC) looked more like activated B-cells. The two categories confirmed Staudt's suspicion; the GCB tumors largely corresponded to the 40% of patients who had a better clinical outcome. The paper garnered so much attention because it was the first example of a clinically relevant classification of cancer using gene expression profiling, says Staudt. The team has gone on to repeat the study with many more samples, and is using statistical studies to glean finer details from the data. They have performed similar studies on chronic lymphocytic leukemia (Journal of Experimental Medicine, in press) and also show that the NFkB pathway is highly activated in the ABC class of cells, preventing apoptosis, and that deactivation of the pathway could kill those tumors (Journal of Experimental Medicine, in press). That result is the most significant, says Staudt. "We understood the biological basis [between the tumor classes], and that could lead perhaps to a new targeted therapy in the future. The most useful outcome of gene expression profiling will be [a better] understanding of tumor cell biology." Jim Kling (jkling@nasw.org) is a freelance writer in Washington, D.C.

Researchers Profile Cancer Cells Through Gene Expression

Interested in reading more?

Become a Member of