Scientists Report New Online Cancer Research Databases

National Cancer Institute
Friday, 10 March 2000

Like the old saying "art for art's sake," most medical researchers once believed that creating large computer databases of known genes or DNA sequences was a case of gathering information for the sake of gathering information. Though interesting, it had no immediate relevance to understanding or treating a disease.

But today, as technology has advanced to make data collection faster and far less expensive, online scientific Web sites have thrived. With a few clicks of a computer mouse, scientists now can scroll through vast databases of biological information, allowing them to form hypotheses at their desktops and quickly pursue new discoveries.

In two articles in the March issue of the journal Nature Genetics, scientists from Stanford University School of Medicine and the National Cancer Institute (NCI) collaborated to report a major harvest of new data in the study of cancer. The data will be posted on two Web sites, beginning on March 1, 2000.

In the first article, the scientists recorded the expression patterns of approximately 8,000 biologically interesting genes in 60 distinct tumor cell lines. All of the cell lines currently are used in NCI's national drug screening program, and they represent a range of common human tumors. The data from this study can be accessed online at: http://genome-www.stanford.edu/nci60 and http://discover.nci.nih.gov.

According to Patrick Brown, M.D., Ph.D., a senior author on the study and a scientist at the Howard Hughes Medical Institute at Stanford University School of Medicine in Stanford, Calif., the new database offers scientists detailed molecular profiles, which point to the inherent similarities and differences in the biology of the cell lines, helpful information in trying to understand tumor development.

"One of the major advantages in studying these 60 tumor cell lines is that, as a part of a national program, there is a complete record of all of the drugs to which these cells have been exposed during the screening process," said Brown. "It is hard to find such a well characterized resource, especially in the clinic where most tumor samples arrive with confusing or fragmentary treatment histories."

Brown said the study, which employed a customized cDNA microarray to record the patterns of gene expression, showed reproducibly that cells from a common tissue or organ tended to have similar gene expression patterns. With a few exceptions, melanoma and leukemia cells as well as those derived from the colon, kidney, central nervous system, and ovaries tended to cluster into their own distinct subgroups, based on similarities in their gene expression program.

"All of these lines have had to adapt to growth requirements in cell culture, all have dysregulated growth cycles, and all have been bombarded with chemotherapy drugs," said Douglas Ross, lead author on the paper and a scientist at Stanford School of Medicine. "What is striking is that through it all, these cell lines still maintain a genomic identity with the tissue or organ of their origin."

In the second paper, the same scientists correlated the gene-expression data in the 60 cell lines with the "activity profiles" of 1,400 compounds tested previously in the NCI drug screening program. Most of the compounds are either standard chemotherapy drugs or agents that have been tested extensively in clinical trials. The results of this analysis are posted at http://discover.nci.nih.gov

According to John N. Weinstein, M.D., Ph.D., a senior author on the study and an NCI scientist, when a compound is tested in NCI's standard 60 tumor cell lines to evaluate its growth-inhibiting ability, it generates a characteristic activity profile. For instance, the profile might show that the compound is active in estrogen receptor-positive breast cell lines, but it is less active in the rest of the screen. For another compound, the activity profile might be completely different.

To date, the NCI has generated activity profiles on over 70,000 compounds and stores them in an online public database (http://dtp.nci.nih.gov). When analyzed in the database, the profiles can suggest shared patterns of activity among a specific class of compounds, suggesting a common mechanism of growth inhibition and a likely point of attack against a tumor cell.

Weinstein said the second study builds on the finding that the cell lines, particularly those from the same tissue, could be reproducibly subgrouped based on their similar gene expression profiles. "If certain cell lines have similar expression patterns, indicating shared molecular characteristics, then an obvious question is whether or not they have similar activity profiles that would suggest a shared sensitivity to certain classes of drugs?" he said.

To approach this question, Weinstein and colleagues selected 1,376 genes from the microarray study that showed the greatest variation in expression among the cell lines. They then regrouped the cells based on similarities in the expression profiles of this smaller subset of genes. With some exceptions, as was the case in the other study, cells from the same type of cancer tended to share similar expression profiles.

The scientists next turned to the activity profiles of 1,400 compounds in the NCI drug screen database. Comparing the screening results in the 60 NCI tumor cell lines, they found that the activity profiles of these compounds correlated less strongly with cell type than did the gene expression profiles. The question was why?

"The answer seems to be that while an individual gene that is involved in drug resistance will not change the overall expression profile of a cell, it can adversely affect the activity profiles of many compounds," said Uwe Scherf, lead author on the study and formerly a scientist in Weinstein's laboratory, now at Gene Logic, Inc., in Gaithersburg, Md.

So, Weinstein and colleagues decided to look closer at the activity profiles. They focused their attention on 118 well-characterized compounds and grouped them into five broad categories, defined according to a compound's presumed general mechanism of action: DNA and DNA/RNA anti-metabolites, tubulin inhibitors, DNA damaging agents, topoisomerase I inhibitors, and topoisomerase II inhibitors.

The group next turned its attention to the gene-expression data. Applying a series of statistical calculations as their analytical tools, they compared the expression patterns of 1,376 genes with the activity profiles of these 118 compounds. The question was whether this next analysis would yield a grouping of these compounds different from that based on their mechanism of action?

After performing the analysis, they found that the subgrouping of many, but not all, of the compounds was different, as recorded on a colorful and now online "clustered" image map that correlates information on genes, drugs, and cells. In particular, the groupings of anti-metabolite and alkylating agents changed in ways not clearly associated with their structure or mechanism of action. Moreover, while the anti-tubulin cluster did not shift, the topoisomerase inhibitors showed mechanistic distinctions among the subclasses.

"Clearly these analyses have limitations, most notably that they were performed in cultured cells and involved fewer than 10 percent of all human genes," said Weinstein. "Nevertheless, these data provide a resource from which cancer researchers can be expected to generate new ideas about the biology of tumor cells and new leads for the development of improved cancer drugs."

** The article is entitled, "A Gene Expression Database For The Molecular Pharmacology Of Cancer," dated March, 2000. The authors are: Uwe Scherf, Douglas T. Ross, Mark Waltham, Lawrence H. Smith, Jae K. Lee, Lorraine Tanabe, Kurt W. Kohn, William C. Reinhold, Timothy G. Myers, Darren T. Andrews, Dominic A. Scudiero, Michael B. Eisen, Edward A. Sausville, Yves Pommier, David Botstein, Patrick O. Brown, and John N. Weinstein.

For more information, or to contact National Cancer Institute, see their website at: www.cancer.gov