Upping the ante in biochip research

March 2000
Cover Story

William Check, PhD

Microminiaturization techniques developed in the computer industry have been adapted in the last 10 to 15 years to making silicon or glass chips, called biochips, on which biological reactions, typically nucleic acid hybridization, take place. Ulysses Balis, MD, instructor in pathology at Massachusetts General Hospital, uses the analogy of computers to describe progress in biochips.

"Computers started out very expensive," he notes. "People weren’t sure what they would do. But they became a mass market commodity, with technology doubling every 18 months or so." Biochips also started as a powerful tool for gathering thousands of pieces of information concurrently on random bits of gene segments. "Yet initially there was a dearth of tools for processing that data," Dr. Balis says.

Since the invention of the biochip around 1988 by Stephen Fodor, PhD, the ability to use information from these devices has grown exponentially. "Taking that into account, we can expect maybe in five to 10 years a transformation in diagnostics," predicts Dr. Balis, "not just tumor diagnostics but inborn errors of metabolism and genetic predispositions as well."

For example, screening for the few main mutations and the hundreds of sporadic point mutations responsible for cystic fibrosis by current methods is laborious and impractical. But a relatively simple biochip could theoretically perform this task quickly and accurately.

Patrick Brown, MD, PhD, associate professor of biochemistry at Stanford University and associate investigator at the Howard Hughes Medical Institute, quantifies the capacity of biochips. "We make microarrays where we try to represent all genes that can be expressed from the human genome," he says. "So far we have about 20,000. These are not all the human genes, but we are heading in that direction. Assuming that about 100,000 human genes are expressed, there is no fundamental reason we can’t get them all on a single array."

From a diagnostic perspective, biochips represent "an incredibly powerful new set of tools for metabolic disorders, neoplastic predisposition, and neoplasia," Dr. Balis says. These tools will classify neoplasias with a prognostic accuracy not previously possible. Just as acute lymphoblastic leukemia and acute myeloblastic leukemia looked similar before flow cytometry and other tools distinguished them, gene technologies are revealing differences in tumor types that look the same morphologically.

"But I don’t think gene chips are going to replace morphology," Dr. Balis says. "It will just be a new set of tools. Pathologists, especially surgical pathologists, don’t need to look on this technology with a jaundiced eye. They should embrace it as a better way to serve clinicians."

Advocates of bio-chips believe eventually physicians will be able to look at the spectrum of expression of 10,000 genes in a tissue sample, compare that profile to a database containing prognostic data associated with clinical histories and the chemotherapy used to achieve those outcomes, and select the best antineoplastic agents for that patient’s tissue type. "We will shift from empiric use of neoplastic therapy to targeting disease based on the molecular bases of oncogenesis," Dr. Balis says.

He cites a preliminary report in which 31 of 31 chronic myelogenous leukemia patients went into remission following treatment with a chemotherapeutic agent targeted to a key molecule in the process that deranges the cell cycle.

Dr. Brown presents a similar view. "A pathologist can look at tissue sections and recognize which are similar and different," he says. "If it is done systematically, one can recognize even subtle features that connect with clinical behavior, features that carry information about underlying biology." He continues, "A microarray too has double value, since we can trace prognostic differences in expression to particular genes that then become potential targets for therapy."

So far almost all biochips are intended for research and commercial uses rather than for clinical application. Says Tom Gingeras, PhD, vice president of biological sciences at the biochip pioneer Affymetrix, "All arrays that we make are for research purposes only." Bob Burrows, director of corporate communications at Gene Logic, which makes expression databases, sounds a similar note: "We want to be content providers to the pharmaceutical, biotechnology, and life science research communities." But, he adds, "the ultimate downstream use of this information is to put it into the hands of doctors for patient management and on the Internet for direct access by patients."

Underscoring the distance that bio-chips have to go to reach the clinic is the experience of LabCorp’s Center for Molecular Biology and Pathology. Says Tim Alcorn, PhD, director of infectious diseases at the center, "Gene chips do not have much clinical utility right now." LabCorp was using Affy-met-rix’s chip for detecting drug resistance in HIV. But, Dr. Alcorn says, "There were limitations to what that HIV chip could do that forced us to move away from it." One problem is that the chip can detect only what it has been instructed to look for. "HIV resistance testing evolved so rapidly that by the time a chip design was set, it was obsolete," Dr. Alcorn says. "Ultimately," he believes, "a chip type format may replace gel-based sequencing and be standard for HIV resistance testing, but not now."

Most people working in this field agree with the twin ideas that bio-chips will revolutionize diagnostics but will not make pathologists obsolete. Kieran Gallahue, chief financial officer at Nanogen, predicts, "The real excitement will be surrounding expansion of testing—pharmacogenomics and susceptibility and resistance testing. The objective is not to replace 30-cent tests, but to provide a whole new level of information that is much more specific and direct." He adds,"I don’t think anybody believes that laboratorians are going to go away, but they may have a higher level of information from which to interpret. Like many other aspects of modern life, people don’t go away; tools empower them to make better and more timely decisions."

As Fred Waldman, MD, PhD, professor of laboratory medicine at the University of California, San Francisco, puts it, "Our goal is to identify genetic signatures for subtypes of cancers. Gene chips could distinguish between renal and bladder cancer, but a pathologist can do that already with 99.9 percent accuracy. So the real goal is to define prognosis of tumors better than histology can do alone." Ideally, chips would not displace pathologists, but would provide new information for them to use.

Lawrence True, MD, associate professor of pathology at the University of Washington, envisions a scenario in which the initial workup of a patient with suspected prostate cancer, for example, will involve diagnosis by conventional his-to-pa-thol-ogy—"which is probably the most specific and cost-effective and efficient way to diagnose prostate cancer"—while expression array data will supplement some of the traditional measures used to distinguish that cancer’s malignant potential, such as histologic assessment of degree of differentiation. Expression profiles could replace assays of proliferation, such as DNA flow cytometry, and assays for over- or underexpression of genes such as HER-2/neu in pros-tate cancer.

 Dr. Gingeras sums up these ideas by saying, "We are looking to begin moving toward more tailored and individualized treatment."

 He thinks about the impact of biochips on medicine in terms of patient management. "One can think about the issue as a three-dimensional box," he says. These dimensions are the cost-effectiveness of treating an individual patient; the patient’s status, asymptomatic to symptomatic; and the likelihood of treatment success in that patient.

"Where we most often are today," he says, "is in the corner with the lowest likelihood of a drug being successful as well as being fairly expensive, and we use it most often on symptomatic patients. Using genetics and profiling, you would like to move the population to the corner where you are treating before they have serious disease, where there is a higher drug effectiveness, and where you are getting more from your money by virtue of treating the right people in the right way." Genetic profiling could help match treatments to patients. "Then you know the patient will respond," Dr. Gingeras says. "You will not be doing an experiment on each individual."

Two assumptions underlie this approach. The first is that variations in response to drugs are based on differences in metabolism that reflect detectable differences in genes and gene expression. The second is that there will be tissues available, such as blood cells, that will serve as surrogates when access to the affected tissue is not possible. This latter situation might arise in neurological conditions, for instance. And if you move from treating symptomatic to asymptomatic persons, it won’t be possible to take tissue from an apparently healthy heart or other organ.

W hether these researchers’ visionary predictions become reality is still an open question. But one fact is certain—an overview of the several novel ways in which biochips are being exploited bears striking testimony to human ingenuity. These devices are being used to measure the relative number of genes in tissues, to quantitate gene expression, as discovery platforms for finding potential therapeutic targets in cancer, for screening libraries of chemicals, to generate reference databases, as concentrating devices—each hopeful researcher and company has a unique wrinkle.

 We can start with an exception, Cepheid. "We are different from other companies," says Curt Petersen, PhD, Cepheid’s president. "Our intention is to attack the clinical laboratory market first through pathogen detection with DNA probes." Cepheid uses chips as an adjunct device—for concentrating samples.

Cepheid’s goal is automated sample preparation, including DNA extraction from biological samples. "We are trying to make a hands-off cartridge-based approach," Dr. Petersen says. "You take a specimen such as urine or blood, put it into our machine and it does everything." A prototype using a cartridge for chlamydia/gonococcus detection has been demonstrated. "It accepts up to 5 mL of urine and gives you a DNA probe result in 30 minutes," Dr. Petersen says. Potentially this will allow a small clinic or operating room to take a sample and get a rapid result for many infectious diseases or genetic disorders.

Cepheid uses chips to concentrate DNA. Its micro-fluidic chip contains an array of single-crystal silicon etched pillars that capture DNA as a biological fluid flows through. With their high surface area, the pillars adsorb DNA by nonspecific interactions.

Initial applications will be in areas where an answer is needed fast, although Dr. Petersen won’t specify which applications the company is working on. He does note that Ce-pheid’s prototype chlamyd-ia/gono-coccus cartridge "gave results in less than one-half hour that were comparable to commercially available, FDA-approved methods taking many hours." He adds that Cepheid has done a lot of work on biowarfare detection, including anthrax, for the Department of Defense.

Dr. Petersen predicts that Ce-pheid’s products will have a strong impact on culture-based bacterial assays. "In the next three years," he asserts, "we will dramatically change that industry, making identification a lot easier, quicker, and more accurate."

Dr. Waldman of UCSF uses bio-chips in a more conventional and research-oriented manner. In his work on genetic alterations in solid tumors, he uses them to evaluate changes in the copy number of genes. This application evolved from previous experiments on chromosomal copy number alterations using a technique called comparative genomic hybridization, in which tumor DNA and normal human DNA are labeled with fluorochromes of different colors and hybridized to human metaphase chromosomes. Changes in copy number show up as altered fluorescence ratios, allowing one to identify and quantify gains or amplifications in genetic material located to a particular chromosome region at low resolution—10 megabases.

To increase resolution, Dr. Waldman performs the same type of experiment on biochips using much smaller 100-kb clones of DNA as probes rather than metaphase chromosomes. A "profusion" of such clones have been isolated and provided by laboratories funded by the Human Genome Project, Dr. Waldman notes. Each clone or DNA segment is placed into one well of an 864-well plate. A robot spots a sample of each clone onto a slide via pins, creating the array. Thousands of different DNA clones can be spotted onto one array. Then labeled tumor and normal DNA are hybridized to the arrays, revealing relative amounts of the two DNAs binding to each spot, which defines genetic gains or losses at high resolution.

Dr. Waldman estimates that a useful map of tumor-related alterations in copy number can be achieved with about 3,000 cloned fragments distributed at one-megabase intervals along the genome. For regions of particular interest, such as a breakpoint in a translocation, analysis can be done at higher density.

His goal is to identify candidate markers of clinical outcome either with therapy or in the absence of therapy. Already more than 30 oncogenes are known to be amplified in one or more forms of cancer. And many chromosomal regions are amplified or deleted but the identity of the presumed genes in those regions is not known. "Our hope is that identifying these genes that are altered in copy number will lead to better diagnosis and prognosis of cancers," Dr. Waldman says.

One project involves renal cancer, for which other methods have defined specific genetic alterations associated with differing histologic subtypes. Dr. Waldman has worked with Guyla Kovacs, MD, of Germany, who provided 40 primary human renal cancers of differing histologic subtype—for example, clear cell, chromophobe, capillary. Dr. Waldman and his colleagues have produced arrays containing cloned fragments representing regions known to be associated with the different renal cancer subtypes. Hybridizing labeled DNAs from the human cancer samples onto arrays allowed them to classify these cancer DNAs with high accuracy. Although they have not yet totally unblinded the study, Dr. Waldman says,"We can see clear differences in genetic alterations of different renal tumors. So this becomes potentially a renal cancer diagnostic chip.

"Right now there are not good therapies or adjuvant treatment for renal cancer," he continues. "But identification of new genetic markers opens up the possibility of gene-based therapies in the future."

Dr. Waldman’s work concerns direct genomic analysis. More common today is expression analysis of genes, using a similar approach to ask whether some genes are over- or underexpressed in tumors. In this approach, messenger RNA (mRNA) is extracted from tumors, reverse transcribed into cDNA, cloned, and applied to an array.

Dr. True, of the University of Washington, uses expression arrays in his research on prostate cancer. "Our goal is to use arrays of ultimately all genes whose expression is associated with growth and differentiation of human prostate tissue," ranging from development to benign proliferative diseases like nodular hypertrophy to prostate cancer, he says. "These arrays will allow us to identify patterns of gene expression that characterize prostate cancers that will progress to metastatic tumors and to see which features of those patterns are most strongly associated with progression." Once the specificity of such putative patterns of gene expression is validated, a standardized expression array of at least 10,000 unique genetic sequences can be used as a laboratory assay in the initial diagnostic workup to help tailor management to the individual patient.

"We are increasingly able to detect prostate cancers based primarily on more sensitive serum PSA assays," Dr. True points out. "But only a minority of men have prostate cancer that will progress. The problem is that our present tools are not sufficiently specific to distinguish those patients whose cancers will progress from those whose cancers will not. The promise of expression arrays is to make that distinction with more certainty than we can presently do."

Optimism that such distinctions are possible is based on precedents. Dr. True cites differences in molecular abnormalities of neuroblastomas in children, where the degree of amplification of the n-myc gene is a predictor of the probability of metastasis. In breast cancer, overexpression of the HER-2/neu protein characterizes tumors that progress more rapidly, although the link is weaker.

"We have identified about 20 novel partial sequences of genes that remain to be identified, and we have reported on a couple of known genes that seem to be differentially expressed in the prostate," Dr. True reports. "We have been collecting fresh tumor tissue with information about stage and grade for the last six years. Probably within the year we will take some of those samples and see what expression differences we can detect. Using an array with a couple of thousand sequences, we will see if we can correlate expression patterns with clinical outcomes." (Dr. True’s collaborators at the University of Washington are Lee Hood, MD, PhD, and Peter Nelson, MD, of the Department of Molecular Biotechnology, and Paul Lange, MD, Alvin Liu, PhD, and Robert Vessella, PhD, of the Department of Urology.)

Technically, handling tissue for expression arrays differs from handling tissue for anatomic pathology, Dr. True notes. Hybridization to an array now requires greater care and immediacy in handling of the tissue to minimize RNA degradation. Fresh, not formalin-fixed, tissue is required, and tissue must be put into solutions that inhibit enzymes that degrade RNA. Second, solid tumor biopsies contain mostly nontumor cells, so techniques to enrich for tumor cells are needed. Dr. True is optimistic about flow cytometric sorting using cell surface antigens, a method now used for lymphomas. A third, more general, issue is how representative a biopsy sample is of the whole tumor. "We don’t yet know the answer to that question," he says.

Dr. Brown of Stanford is also an expert in expression arrays. "Recently," he says, "it has become possible to come close to looking at the activity of every gene in the genome in clinical samples." With this level of detail, he notes, "It is crucially important that our approach is systematic, making a series of detailed molecular characterizations of clinical samples using a rather standardized quantitative method. That allows us to look at large sets of samples and to use a rigorous statistical method to try to recognize correlations between variations seen at the molecular level and differences in the clinical behavior and biology of tumors."

Dr. Brown’s method is to "print" about 140 arrays at a time with 20,000 to 30,000 genes each using a mechanical printing process. Then, using a selective hybridization method, he quantitatively measures the abundance of mRNA from each of the genes in a clinical sample relative to a standard scale.

In an article published in early February in Nature, Dr. Brown and collaborators at the National Cancer Institute and other centers reported results of a study of variation in gene expression in diagnostic biopsies of diffuse large-cell B cell lymphoma. Based on gene expression patterns, they identified two subtypes of this malignancy not previously recognized. "Not only do these subtypes have a distinct difference in their gene expression pattern that suggests underlying biological differences," Dr. Brown says, "but they have a highly statistically significant difference in clinical outcome, a difference that has never before been possible to see." In this case the differences in gene expression suggest possible molecular differences that could account for the differences in outcome. These findings are now being expanded with a larger group of collaborators. "There will be many more such findings in the next year or two," Dr. Brown predicts.

In a variation on gene expression analysis, David Per-sing, MD, PhD, vice president of diagnostics development at Corixa Corp. and medical director of the Infectious Diseases Research Institute, both in Seattle, is using expression arrays as what he calls "gene discovery" platforms. "We use expression microarrays as screening devices to identify potential targets for T-cell vaccines for cancer," Dr. Persing says. Such vaccines are based on the concept that tumor-specific proteins or nucleic acids can prime a patient’s own T cells to become cytotoxic to a tumor or pathogen.

"For several tumor types, we aim to identify from the roughly 100,000 genes in the human genome between six and 20 candidate genes for which expression is essentially restricted to the tumor," Dr. Persing explains. Candidate genes identified on expression microarrays are subjected to real-time PCR (RT-PCR), which is more sensitive than expression analysis, to verify specificity, and to additional verification steps. "These candidate genes can be used for diagnostics, because their expression is consistent with the presence of tumor, and for vaccine targets as well as therapeutic antibody targets," Dr. Persing says.

Immunohistochemical testing of a prostate-specific candidate protein discovered by Corixa showed that 65/65 prostate cancers or prostate tissues were reactive while all 4,635 nonprostate tissues were nonreactive. (For prostate cancer, markers only need to be tissue-specific, not cancer-specific, since vaccines will be used in men who have had prostatectomy.)

Another candidate gene is mammoglobin, which Dr. Persing calls "the most breast tissue-specific gene that has ever been described in terms of its expression profile." Mammoglobin and two other markers were used to detect breast cancer cells in the blood of women who had had cytoreductive surgery. (Tumor cells can be detected in lymph nodes and blood of such patients even without obvious lesions.) Markers in addition to mammoglobin were needed since one-fourth of breast cancers either do not express it or lose the chromosomal segment that carries the mammoglobin gene.

Dr. Persing says that, using RT-PCR to measure the expression of these three genes, expression of one or more targets was present in nearly 100 percent of breast cancers. Dr. Persing is "very optimistic" about using these techniques for detection of cancers of various types and for monitoring therapy. "We plan to use RT-PCR profiles on whole blood to determine if we can detect circulating cancer cells at a very early stage," he says. Cells will be captured on magnetic particles using antiepithelial antibodies and assayed with RT-PCR to detect particular types of cancer. "I think this will be a huge area of opportunity for diagnostics," Dr. Persing says. Corixa is involved in clinical trials to evaluate this hypothesis.

Affymetrix, the company that trademarked the term "GeneChip," employs a unique approach to making microarrays. "Arrays are usually made with premade oligonucleotides or cloned DNAs that are bound to a substrate," Dr. Gingeras explains. In Affymetrix’s proprietary method, oligonucleotide probes, usually 20 to 25 nucleotides long, are synthesized on the substrate in a combinatorial fashion using photo-catalyzed reactions. By masking sites in a preset pattern as oligonucleotides are lengthened, each position on the surface identifies a unique sequence.

"This approach allows us to achieve much higher densities of information than competitors," Dr. Gingeras says. Affymetrix’s chips routinely carry 240,000 to 280,000 unique oligonucleotides.

 With these gene chips, three kinds of questions are answerable. In resequencing applications, a consensus sequence is defined for a portion of the genome. One can then do base-by-base interrogation of a sample testing for identity with each of the bases in the consensus sequence. Discovery of SNPs (single-nucleotide polymorphisms) is a direct application of this approach, Dr. Gingeras says.

Affymetrix has relationships with bioMérieux and Roche Molecular Systems to develop chips for this purpose. Clinical applications include bacterial identification, viral genotyping for drug resistance mutations, and genotyping of p53 genes and some human cytochrome P450 genes. Both companies are building proprietary platforms for diagnosis based on GeneChip probe arrays. Affymetrix will manufacture and provide arrays to the companies, who will handle development and certification.

A subset of resequencing applications arises when one wants to focus not on identifying each base in a long string of bases, but on looking at snapshots across the whole genome. "A large amount of work is being done identifying SNPs on the human genome," Dr. Gingeras says. "We are beginning to develop assays and arrays to interrogate many of those polymorphisms." Affymetrix has just released a product that interrogates 1,500 SNP sites mapped on the human genome.

Quantitative mRNA profiling—gene expression—is a third application. "We currently make a variety of arrays that have varying numbers of genes for evaluating message levels in a cell," Dr. Gingeras says. "For each gene in the expression array, the sample is interrogated with a set of 16 to 20 oligonucleotide pairs." Each pair consists of an oligonucleotide that is the perfect complement of the message, which interrogates a single base positioned along the length of the mRNA molecule, along with its companion, which has a deliberate mismatch incorporated at its center, and which is designed to serve as a negative control.

"This approach allows one to have multiple interrogations for each sequence," Dr. Gingeras says, "so you are not dependent on a single interrogation by a single probe as with a single PCR product or a single spotted clone." With this redundancy, results are less likely to be compromised if a sequence in the public database has errors.

Nanogen also supplies chips for researchers, but with a difference: Its chips can be custom designed in the investigator’s laboratory. As CFO Gallahue puts it, "We supply open platforms for targeted genetic analysis. Our products are unique, since their integrated electronics allows researchers to create their own genetic array." With a Nanogen chip, it is possible to do an analysis and get results in a single day, an advantage in research and eventual clinical applications. One of today’s biggest challenges, Gallahue notes, is to translate clinical research into clinical practice. "Nanogen’s technology," he asserts, "allows users to bridge that transition by designing their chips to meet requirements for either clinical research or clinical application."

More specifically, Gallahue notes, most biochips are passive systems, whether lithographically synthesized or spotted. In either case a genetic sequence is put down on a glass slide and heat and chemistry are manipulated to facilitate hybridization and stringency. "You have to apply the same characteristics to all of the chip at the same time," he says. Nanogen’s product integrates electronics into the analysis process. The company’s first product will be a 100-site chip with 100 individual controllable electrodes. A user can create an environment around each electrode without affecting the others.

"In effect you have 100 independent test sites," Gallahue says. "You can use the electronics to turn each site positive or negative or neutral. So a user can manipulate the electronics to attach genetic content to a blank chip at sites of their choice, then use reporter probes to interrogate the whole chip or only portions of it at one time."

Later this year Nanogen will sell its first product—NanoChip molecular biology workstation—to the research market—pharmaceutical companies, genomics firms, biotechnology firms, and clinical investigators. Last year, Dennis O’Kane, PhD, of Mayo Clinic, one of the beta-testing sites for the product, reported that he and colleagues had successfully used a customized Nanogen chip for a pharmacogenomic application, looking at SNPs in the gene for thiopurine methyl transferase (TPMT), which metabolizes immunosuppressant thiopurine drugs used to treat leukemia, Crohn’s disease, and other conditions. Variants in the TPMT gene result in high and low activity forms of the enzyme, which can affect drug toxicity in individual patients. A Nanogen chip customized by the Mayo researchers was 100 percent accurate in identifying SNPs in the TPMT gene in 115 patient samples, while two widely used comparison methods had six percent and 11 percent error rates.

 A similarly positive result was reported by the Bode Technology Group, a leading private forensics laboratory, using a custom-designed Nanogen chip to identify short tandem repeats, genetic identifiers used by the government for populating criminal identification databases.

Caliper Technologies makes no microarray products as such, but it uses biochips in a unique way—as total reaction platforms. Michael Knapp, PhD, vice president of science and technology at Caliper, notes that a conventional array has one partner of a single biochemical reaction, often an oligonucleotide probe, immobilized on its surface. As a result, such chips are unifunctional: They perform one step of a complex series of reactions, such as hybridization, with high throughput. But other steps in the process, such as amplification, may be done in a more labor-intensive format.

Caliper uses photolithography and photoetching techniques to create a custom network of channels on the order of 10 µm deep and 50 µm wide in a chip. "These channels," Dr. Knapp says, "amount to an instruction set for a series of fluid manipulations."

Such chips incorporate three innovations: They miniaturize functional steps to volumes of 100 picoliters to a nanoliter; integrate multiple steps on the same device; and automate the reaction sequence with a custom-designed software program that can exert forces on fluids inside the chip using pressure or vacuum or electrokinetic forces.

Caliper’s main product, targeted to the pharmaceutical industry, is a high-throughput device for screening large libraries of compounds for desired activities, such as antagonism of a cell surface receptor, inhibition of a kinase or calcium flux, or DNA hybridization—all multistep assays. Several such high-throughput chips are in production. A reservoir feeding into one processing unit of a chip stores enough reagent for a 24-hour run, perhaps 10 µL for each reaction, adequate for about 10,000 assays. Chips can have up to 12 processing units, thereby having a capacity of more than 100,000 assays per day.

Finally, at Gene Logic, the product is applied expression arrays targeted to the research market. Gene Logic uses Affymetrix GeneChip expression arrays, supplemented by a proprietary method to quantitate low-abundance transcripts, to generate expression databases. It then sells these databases in two formats: tailored databases sold exclusively to pharmaceutical customers, or their large reference database called Gene-Express by subscription. The company just signed up its first two subscribers to GeneExpress, one biotechnology company and one large pharmaceutical company, says director of corporate communications Burrows.

He notes that the goal for GeneExpress is to have quantitative expression profiles for all 100,000 genes across 30,000 human samples across all tissue types and all disease types plus all underlying clinical data. This will enable researchers to query on hundreds of factors for their influence on clinical outcome. Says Burrows, "It will be possible to say,
’Show me all genes that are expressed in livers of men aged 15 to 45 who drink a lot of alcohol and take Prozac and compare that profile to similar individuals who don’t take Prozac.’’" So far GeneExpress contains about 1,200 profiles, and profiles are being added at 400 samples per month; the company plans to accelerate to 800 per month later this year. Samples are obtained from volunteers recruited through an international network of centers.

If the scope of Gene Logic’s database seems overwhelming, consider a pharmacogenetic application conceived at Caliper. "With massive genotyping in the developmental phase of a drug," Dr. Knapp says, "one can identify different responses and different toxicities that are genetically determined." This requires 100,000 different genotypes on perhaps 1,000 patients. To enable this project, Caliper has developed an integrated chip that does amplification and genotyping as a high-throughput screening operation. Says Dr. Knapp, "Generating such a database requires a scale of experimentation that is unprecedented in any field."

William Check is a freelance medical writer in Wilmette, Ill.