|Graphic: Cathleen Heard|
This actual case1 illustrates the emerging role of DNA chips, also called microarrays, in understanding, diagnosing, and treating cancer. The value of chips arises from the nature of the disease. When defective genes cause cells to divide uncontrollably, they also disturb gene-expression patterns. These patterns may provide clues into what's going wrong in a cell; at the very least, they can identify a tumor. The technological leap represented by DNA chips lies in their ability to simultaneously detect the expression patterns of thousands of genes. Earlier cancer studies, in contrast, typically examined a handful of genes at most.
DNA chips are just starting to have an impact on the vast field of cancer research. Earlier this month, for example, a team of 31 academic and government researchers reported that chips could be used to characterize diffuse large B-cell lymphoma as two distinct diseases with large differences in survival rates.2 The potential utility of DNA chips, nevertheless, was grasped when they first became available in the mid-1990s. "We more or less committed right away to looking at cancer," recalls Patrick O. Brown, an associate professor of biochemistry at Stanford University School of Medicine and a pioneer in the field.3 He also cites a more practical reason to focus on cancer--the researcher's need for tissue samples. "People are generally reluctant to give up bits of their body," he observes. "The one exception is cancer."
DNA chips work as follows: One surface is typically specked with thousands of dots of single-stranded DNA embodying various genes or gene segments.4 When the chip is incubated with labeled cDNAs made from the RNA in a tissue, these cDNAs hybridize to complementary DNA sequences on the chip. By knowing the sites and identities of the DNA strands on the chip, and by seeing where labeled cDNAs hybridize, researchers can identify those cDNAs. This information indicates which RNAs were in the tissue and thus which genes were expressed.
A year ago, Jeffrey M. Trent, chief of the cancer genetics branch at the National Human Genome Research Institute (NHGRI), lamented that DNA microarray technology was restricted largely to a few labs and a growing number of companies.5 He now predicts that "in the next 12 to 18 months, you'll see a tremendous influx of investigators who have access to this technology."
Trent attributes his more optimistic outlook to several developments. First, more companies are selling equipment to fabricate the chips and "read" the pointillistic patterns created by hybridization.6 (This increase in suppliers has occurred despite a spate of patent litigation.) Second, well-characterized and sequenced DNA clones are increasingly available to put onto chips. Finally, better statistical tools have been developed to analyze data derived from chips.
The federal government has also pitched in. The National Cancer Institute (NCI) announced in September that it would spend $4.1 million to help 24 cancer research centers in the United States buy DNA microarray equipment. Three months later, NCI awarded 10 grants totaling $4.1 million to projects involving some use of DNA chips. These grants were the first awarded under a challenge that NCI director Richard D. Klausner has issued to researchers to develop a molecular classification of tumors.
Papers confidently applying DNA chips to cancer are just beginning to appear. Nature Genetics editor Barbara Cohen says she noted a "steep increase" last year in submissions of articles involving microarrays; many of these papers were about cancer. At a meeting on microarrays that Nature Genetics sponsored last September in Scottsdale, Ariz., one-third of the 178 abstracts dealt with cancer.7
Assuming that studies soon validate the methodology, direct medical applications of DNA chips should kick in within a year or two, according to Stephen H. Friend, president of Rosetta Inpharmatics Inc., a company in Kirkland, Wash., that develops microarray technology. Based on array patterns, "decisions will be made to give patients more or less of one therapy, or an alternative therapy," he says. (The case at the beginning of this article was an isolated incident.)
Still, using DNA chips to study solid tumors, in particular, is far from easy. To preserve nondegraded RNA, a tumor should be frozen soon after it is removed in an operation. But medical protocol dictates that a pathologist must first examine the tissue, which could occur hours later. Before freezing the tumor, a researcher must dissect malignant regions away from normal blood and lymph cells, fibroblasts, and connective tissue. So much tissue may be whittled away, however, that little RNA remains. Amplifying the genetic material may introduce errors and biases into the results.
|Courtesy of Patrick O Brown|
Two RNA samples labeled red and green, hybridized to a gene array. Spot color represents relative expression of the corresponding gene.
According to one observer, these difficulties--some requiring technical fixes; others, changes in hospital procedures--could take 10 years to resolve fully. The biggest challenge, meanwhile, may lie in fashioning an affordable set of chips that contains all human genes, now estimated to number from 100,000 to 200,000. Affymetrix Inc. of Santa Clara, Calif., sells a set of four stamp-size chips representing about 7,000 full-length genes, minus introns, and 33,000 expressed sequence tags (ESTs), which don't correspond precisely to 33,000 genes because some ESTs likely derive from the same gene. The price is $10,000 per set, though volume discounts are offered.
Of course, a tumor cell does not express every gene in the genome, and researchers might want to limit a study to a subset of human genes expressed in a particular cell type--if that subset is known. This is where gene databases and custom-built DNA chips come in handy.
In 1997, NCI began the Cancer Genome Anatomy Project (CGAP). Budgeted this year at $12 million, CGAP has created a tumor gene index by parceling out DNA sequencing to Washington University in St. Louis, tissue dissection to other groups at the National Institutes of Health, and cDNA-library construction to M. Bento Soares at the University of Iowa. As of December, the index had cataloged 90,000 genes (some of which may be alternatively processed forms of the same gene), produced over 160 cDNA libraries, and submitted more than 800,000 ESTs to various databases.
Investigators can gain access to this information at CGAP's Web site (www.ncbi.nlm.nih.gov/CGAP). Project director Robert L. Strausberg describes how the site may be used: Researchers studying prostate cancer, say, would click on the "Summary Tables of Libraries, Genes and Sequences" and select "prostate." They would then learn that in the prostate, CGAP has cataloged about 61,000 sequences and 13,300 genes, some unique to that organ.
Strausberg continues: "So they say, 'Gee, there are 1,500 genes that have been seen to be expressed only in the prostate. I'm going to order those clones, and I'm going to build DNA microarrays to study the [gene] expression across a broad set of tumors.'" Clones can be ordered through the IMAGE (Integrated Molecular Analysis of Genomes and their Expression) Consortium (www-bio.llnl.gov/bbrp/image). DNA chips also can be custom built to incorporate those clones.8 With these chips, Strausberg says, researchers can compare many tumor samples and ask whether information at the molecular level "might tell us that we can classify prostate tumors into multiple types."
Asking a similar question about human leukemia, Todd R. Golub, an assistant professor of pediatrics at Harvard Medical School, and 11 colleagues recently used DNA chips to differentiate two forms of the disease.1 These cancers were chosen as a test case because they were already distinguishable by several methods, which could thus serve to validate the microarray approach.
Applying a 6,817-gene Affymetrix chip to 38 bone marrow samples, the investigators identified 50 genes whose expression most distinguished acute myeloid from acute lymphoblastic leukemia. Thirty-four samples drawn from a wider range of sources were then analyzed. Expression patterns for the 50 genes, determined with DNA chips, accurately predicted the type of leukemia for 29 of the 34 samples.
Tissue Microarrays Exist, Too
After a DNA chip has identified a gene as possibly contributing to cancer, how does a researcher confirm that hypothesis? One option is to use a tissue microarray to study the gene or its protein product in as many as 1,000 tumor samples at a time.
Earlier arrays called multitissue blocks could accommodate only one-tenth as many samples. Tissue microarrays, unveiled in 19981 but still not widely used, were developed in the lab of Olli-P. Kallioniemi, a section head in the Cancer Genetics Branch of the National Human Genome Research Institute. Besides allowing the parallel processing of many samples, microarrays afford researchers greater access to tissue specimens, "a precious commodity in academic centers," he notes.
With conventional methods, Kallioniemi says, "you cut five-micron sections through a tumor, and you have destroyed your clinical material" after making 200 to 300 slices, each laid onto one slide. A microarray uses much less material from a single tumor per slide. A cylindrical biopsy with a 0.6 mm diameter is cut from the tissue. Specimens need not be specially prepared; archival tissues work also. Typically, 500 to 1,000 biopsies from various tumors are then arranged into an array in a single paraffin block. Five-micron-thick slices from that block are laid onto slides.
One molecular marker is analyzed per microarray. "Whatever is possible on a regular tissue section--whether it's in situ hybridization, immunostaining, or in situ PCR--will work on a tissue-microarray section," says Kallioniemi. He has used microarrays to study gene amplifications in breast cancer1 and to identify mechanisms leading to hormone therapy failure in prostate cancer.2
Beecher Instruments (www.beecherinstruments.com) of Silver Spring, Md., sells a machine to make tissue microarrays for $7,000, and an automated version is due out this spring, according to company president Stephen B. Leighton. An ordinary microscope might suffice to examine the tissues in a microarray, but several companies are developing automated imaging systems to speed up the process.
--- Douglas Steinberg1. J. Kononen, "Tissue microarrays for high-throughput molecular profiling of tumor specimens," Nature Medicine, 4:844-7, 1998.
2. L. Bubendorf et al., "Hormone therapy failure in human prostate cancer: analysis by complementary DNA and tissue microarrays," Journal of the National Cancer Institute, 91:1758-64, 1999.
Golub and his colleagues are now applying their method to lymphomas and to cancers of the brain, prostate, and lung. They're also using DNA chips to determine whether gene expression patterns predict which patients will respond to standard chemotherapy. If chips are predictive, Golub doubts that "the cost barrier or technology barrier will be significant at all" in preventing hospitals from obtaining microarray technology.
Stanford's Brown and David Botstein, NCI's Louis M. Staudt, and their colleagues used CGAP as their starting point in studying diffuse large B-cell lymphoma (DLBL), a disease diagnosed in more than 25,000 Americans each year that has long resisted subclassification.2 The group identified more than 15,000 genes in CGAP's tumor gene index that were uniquely expressed in B cells. These genes, along with 3,000 others involved in various immune system cancers, were then arrayed on a "Lymphochip" to which lymphocyte samples from 42 DLBL patients were applied.
Noting distinct differences in gene expression, the researchers focused on genes implicated in a stage of B-cell maturation during which the cells reside in the germinal center of the lymph node. Analysis of the DNA chips revealed two subtypes of DLBL. In one, germinal-center gene expression resembled that of normal B cells, while in the other, expression was low or undetectable. Significantly, patients with the first subtype of DLBL had a five-year survival rate of 76 percent, compared to a 16 percent rate in patients with the second subtype. Brown surmises that "this finding will warrant a change in the way this disease is managed." The study is now being scaled up.
Other cancer studies using DNA chips also have clinical implications but haven't been published yet:
* Brown is looking at other types of cancer, too. In 40 breast cancer samples obtained before and after chemotherapy, his group found gene expression to be remarkably stable and unique. This "recognizable signature that distinguishes one tumor from another" should aid in classifying and characterizing tumors, he says. In a recent chip study with NCI's John N. Weinstein, Brown examined 60 cell lines culled from all types of cancer and already tested for drug sensitivity.
* Trent's lab at NHGRI used DNA chips to distinguish hereditary breast cancers with a mutation in the gene BRCA1 from those with a mutation in the gene BRCA2 and from sporadic (presumably nonhereditary) tumors. Earlier studies could not make this distinction.9
* Melanoma has long resisted subclassification. But by applying DNA chips with some 10,000 genes to several dozen melanoma samples, a group including Trent identified samples that clustered together. "The genes responsible for forming that subset led us down a specific series of biologic studies that we think have given us really important insights into that disease," he says. An abstract from the 1999 Scottsdale meeting, with lead author Paul S. Meltzer of NHGRI, reveals that highly invasive melanoma cultures expressed genes consistent with a vasculogenic phenotype.7
* Patients with chronic myelogenous leukemia are treated with alpha-interferon, but only about one-third respond, according to Friend. The firm plans to use its proprietary data-analysis system and DNA microarrays to try to identify responders and nonresponders before treatment is administered.
With their scientific, engineering, and analytical expertise, companies such as Rosetta Inpharmatics exemplify how DNA chips are making biomedical research more collaborative than ever. Microarray studies "require people with training in clinical medicine, basic biology, computer science, mathematics, and technology development," notes Golub. "And there really is no single person who is likely to be an expert in all of those areas."
In cancer research, DNA chips are spreading their influence in other directions as well. Albert J. Fornace Jr. at NCI uses chips to study cellular responses to radiation, a cause and treatment of the disease.10 Various groups are applying microarrays to mouse models of cancer, and NCI has begun a Mouse Cancer Genome Anatomy Project. Chip technology could also foster gene discovery, though it apparently hasn't yet uncovered a gene as important to cancer as ras or p53.
Noting that use of that technology is still in its infancy, Trent adds: "All of us have data sets in which we are identifying candidate genes. I would just say, 'Hold on.' There will be a tremendous influx of information and important genes that flood from this."
Douglas Steinberg is a freelance writer in New York.
1. T.R. Golub et al., "Molecular classification of cancer: class discovery and class prediction by gene expression monitoring," Science, 286:531-7, 1999. Aspects of this case are described on page 536.
2. A.A. Alizadeh et al., "Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling," Nature, 403:503-11, 2000.
3. J. DeRisi et al., "Use of a cDNA microarray to analyse gene expression patterns in human cancer," Nature Genetics, 14:457-60, 1996.
4. Regarding the variety of DNA chips, see B. Sinclair, "Everything's great when it sits on a chip: a bright future for DNA arrays," The Scientist, 13:18, May 24, 1999.
5. J. Khan et al., "DNA microarray technology: the anticipated impact on the study of human disease," Biochimica et Biophysica Acta, 1423:M17-M28, 1999.
6. Links to some of these companies and major labs in the field: www.nhgri.nih.gov/DIR/LCG/15K/HTML/microlinks.html.
7. Abstracts and other information about the meeting are at genetics. nature.com/microarray/meeting.html.
8. See, for example, cmgm.stanford.edu/pbrown/mguide.
9. An abstract of this study is at www.faseb.org/genetics/ashg99/f42.htm.
10. S.A. Amundson et al., "Fluorescent cDNA microarray hybridization reveals complexity and heterogeneity of cellular genotoxic stress responses," Oncogene, 18:3666-72, 1999.
Guilt by Association
At the time of the study, LifeSeq contained data on 40,000 human genes in 522 cDNA libraries. In searching for genes, the investigators used a statistical approach they called "Guilt by Association" (GBA).
GBA finds genes with expression patterns similar to a gene already known to be associated with a disease--the original "guilty" party. Genes are deemed to be either on or off, in comparison to some other methods that try to quantify the degree of gene expression.
According to study leader Michael G. Walker, a consultant at Incyte and in Stanford University's department of medicine, the idea of GBA "is that if two genes are always on or always off at the same time, they're probably related." One gene of the pair presumably controls expression of the other, or both genes are controlled by a third gene.
"The really big advantage of GBA is that the same set of [cDNA] libraries can be used for every disease," says Walker. GBA also dispenses with an assumption underlying methods that quantify gene expression--that the expressions of related genes rise or fall proportionately. Given the unique set of influences on each gene, this assumption, in his view, "doesn't make much sense."
One of the eight prostate-cancer-associated genes discovered through GBA appears to encode a serine protease. (Prostate-specific antigen and kallikrein, current markers for prostate cancer, are also serine proteases.) The other genes "show no significant similarity to anything that's known," Walker notes, and thus would not have been detected through sequence homology searches. He says Incyte is now using GBA to search for genes associated with breast and colon cancers.
1. M.G. Walker et al., "Prediction of gene function by genome-scale expression analysis: prostate cancer-associated genes," Genome Research, 9:1198-1203, 1999.