The Rise of Free, Global Gene Expression Data Sets

See related Techlink, "The Data Analysis Grand Prix". For this article, Jim Kling interviewed Patrick O. Brown, Howard Hughes Medical Institute investigator and professor in the department of biochemistry, Stanford University Medical School in Stanford, Calif., and John N. Weinstein, senior investigator at the National Cancer Institute and head of the genomics and bioinformatics group, in Bethesda, Md. Data from the Web of Science show that Hot Papers are cited 50 to 100 times more often than the average paper of the same type and age. D.T. Ross et al., "Systematic variation in gene expression patterns in human cancer cell lines," Nature Genetics, 24:227-35, March 2000. (Cited in 173 papers) U. Scherf et al., "A gene expression database for the molecular pharmacology of cancer," Nature Genetics, 24:236-44, March 2000. (Cited in 137 papers) In 1990, the National Cancer Institute (NCI) began a screening project that to date, has involved more than 70,000 compounds tested against 60 human cancer cell lines of varying origins, including colon, breast, prostate, kidney, brain, ovary, leukemias, and melanomas. Besides identifying numerous drug development leads, experimental results were maintained in a vast, publicly available data set. As that resource increased in the 1990s, microarray technology was simultaneously developed and refined. It was perhaps inevitable that the data and the powerful technique would eventually come together. A few years ago, they did. John N. Weinstein, a senior investigator at NCI and head of the genomics and bioinformatics group, spoke at Stanford University, where he discussed his ongoing work characterizing protein and RNA expression of the cell lines in response to test compounds. Patrick O. Brown, who was pioneering the new microarray technology, using it to take "snapshots" of the gene expression profile in yeast cell lines, was in attendance at the talk. At the time, he had just begun to work with a few human cell lines. Courtesy of Patrick O. BrownPatrick O. Brown For Brown, Weinstein's presentation struck a cord. The cells represented a well-characterized group of samples that had been chosen for their diverse background and biological properties. "This seemed like a wonderful opportunity," says Brown, Howard Hughes Medical Institute investigator and professor in the department of biochemistry at Stanford University Medical School. The two men decided to collaborate: Weinstein's team focused on the cell lines' gene expression profiles in response to compounds while Brown and his coworkers focused on the basic cell biology. Weinstein's group carried out most of the cell biology work while Brown and company performed hybridizations and scanned the arrays. Both teams did data analysis. In March 2000, this collaboration reached an important milestone with the publication of two papers in Nature Genetics—one covering the basic biology work that was the focus of Brown's lab; the other centering on the pharmacology-driven work done by Weinstein and his team. The papers were "one of the first examples of using this systematic approach to look at gene expression in a large, self-consistent set of examples as a way of getting insights into the functions of genes on a large scale," says Brown. "Also, the fact that it was one of the first large, freely available global gene expression data sets means it was used as raw material for a lot of computational biologists developing algorithms for analysis of global gene expression." ©2000 Nature GeneticsOur Cells on Drugs: A Clustered-Image Map, relating the activity patterns of 118 tested compounds, to the expression patterns of 1,37 genes in the 60 cell lines. (Nature Genetics, 24:236-44, 2000.) Getting the System Down After searching through gene databases for known human genes and eliminating various "housekeeping" genes that were expressed at similar levels throughout the cell lines, the teams settled on 9,703 cDNA fragments representing about 8,000 genes. The cDNA fragments were spotted onto glass microscope slides to form the microarray. ©2000 Nature GeneticsExpress Yourselves: A side-by-side look at gene expression patterns in clinical breast cancer specimens and cultured breast cancer and leukemic cell lines. (Nature Genetics, 24:227-35, 2000.) Some of the 8,000 were known human genes, while others were homologues of genes found in other organisms; the rest were expressed sequence tags (ESTs). Brown's group then hybridized fluorescently labeled cDNA derived from the samples. Comparing the signal from each sample to a reference allowed them to identify the variation in gene expression between samples. "One of the characteristics of this kind of data that distinguishes it from other data looking at gene expression is that it is quantitative and systematic, and that allows the data to be analyzed in a lot of systematic ways," says Brown. Those systematic ways make it useful to bioinformatics specialists, who are still scratching their heads over the best way to analyze this sort of data, according to Kevin Coombes, an associate professor and head of the bioinformatics section at the MD Anderson Cancer Center in Houston, Texas, who won the 2001 Critical Assessment of Microarray Data Analysis (CAMDA) competition. "[CAMDA] has a very nice structure to it. You take one or two data sets and get a whole bunch of people to apply their methods to the same sets," says Coombes. The good news, he says, is that "most of these methods tend to find the same things." It's the subtle differences that often vary with the analyses. Doing It Cheaply Brown wanted to do microarray experiments that "were cheap [so] we could do tons of these experiments and it would be no big deal. The most critical thing about this paper is it was part of a strategy of building a map of the gene expression program that would be an interpretive framework, so that each [future] experiment of this kind would be a lot more meaningful than if it were done in isolation." To accomplish that goal, Brown printed the purified cDNA onto the glass microarray, using a robot with four printing tips. The work paid off right away from Weinstein's pharmacological perspective, revealing a potentially new clinical application of the drug L-asparaginase. It was recognized in the 1970s that acute lymphocyte leukemia cells often lack asparagine synthetase, which forces them to accumulate asparagine by scavenging it from the bloodstream. L-asparaginase operates by digesting free asparagine, depriving the cell of this critical nutrient. Some previous work had suggested that other cancer types might be vulnerable to the enzyme L-asparaginase, but the data were not conclusive enough to merit clinical trials. Weinstein's group looked at the expression of asparagine synthetase across the cancer cell lines. When they looked at the leukemias, "it was gangbusters," recalls Weinstein. Some ovarian cancer cells also showed susceptibility. Weinstein contacted pharmacologist and oncologist Daniel Van Hoff, director of the Arizona Cancer Center, who performed additional studies. "Those are ongoing, but the results have been interesting enough to suggest the possibility of Phase II clinical trials [for ovarian and pancreatic cancers] if things can be worked out with the FDA," says Weinstein. Not long after these papers came out, a similar analysis of the cell lines appeared in work that used the Santa Clara, Calif.-based Affymetrix, Inc. gene chip approach.1 Weinstein is also an author on this paper and sees it as a second resource to validate findings based on his and Brown's data. But such analyses should be only the tip of the iceberg, says Weinstein. "We generated these [data] and threw them open to the public because there's no way that we can do all of what's there. A subject matter expert looking up his or her own favorite gene or drug will see relationships that we wouldn't know to pay attention to. That's a fundamental characteristic of the work. Initially it worried some of the reviewers, who said we should spend our time figuring out which relationships are significant. Luckily, the editors didn't see it that way." Jim Kling (jkling@nasw.org) is a freelance writer in Washington, DC. 1. J.E. Staunton et al., "Chemosensitivity prediction by transcriptional profiling," Proceedings of the National Academy of Sciences, 98:10787-92, Sept. 11, 2001.

The Rise of Free, Global Gene Expression Data Sets

Interested in reading more?

Become a Member of