Image: Courtesy of Biocat
In 1997, Juha Kononen, a postdoctoral fellow at the National Human Genome Research Institute, was pondering the significance of the recently developed DNA microarray. He was studying genetically altered genes in cancerous cells using fluorescence in situ hybridization and immunostaining of individual tissue sections. The process was, he says, "very laborious." Kononen wondered whether a technology based on DNA biochips could aid in his research. "I thought why could you not invert the concept? Instead of laying down hundreds or thousands of probes, how about laying down hundreds or thousands of tissue spots and probing them one antibody or gene probe at a time." Kononen ap-proached his advisor, Olli Kallioniemi, NHGRI section head for the cancer genetics branch, who gave the green light to proceed.
What Kallioniemi and Kononen developed was the tissue microarray (TMA), an ordered array of tissue cores--up to 1,000 of them--on a single glass slide.1 The approach allows researchers to analyze "many orders of magnitude more" tissue sections and stains than with normal approaches, says Matt van de Rijn, associate professor of pathology, Stanford University. In addition, these arrays use less tissue per analysis and offer greater internal consistency, all at a reduced cost. But they also pose significant challenges with data collection and management, which researchers are only now beginning to address.
MICROARRAY APPLICATIONS Researchers have quickly embraced the new technology. Stephen Hewitt, who runs the Tissue Array Research Program (TARP) at the National Cancer Institute (www.nci.nih.gov/tarp), estimates his facility has distributed "just shy of 4,000 slides" to approximately 350 investigators since March 2001, most of them outside of the National Institutes of Health.
No standard TMA configuration exists. The arrays can contain samples of every organ in a particular animal's body, or a wide variety of common cancers with normal controls. They can contain obscure cases, such as an array of salivary gland tumors, or more common ones, like breast and colon carcinomas. A researcher might array a single, specific tissue from a panel of knockout mice, or even cultured cells.
Scientists treat these slides like any other histological section, using in situ hybridization to detect gene expression or identify chromosomal abnormalities, or employing immunohistochemistry to localize protein expression. With serial sections of the master block, investigators can analyze numerous biomarkers over essentially identical samples. More broadly, researchers use TMAs to validate potential drug targets identified with DNA microarrays.
Arul Chinnaiyan, assistant professor of pathology and urology, University of Michigan, uses TMAs in his research. Chinnaiyan and colleagues queried a 9,984-element DNA microarray with cDNA samples from more than 50 individuals with normal, benign, or cancerous prostates. "We were trying to define a molecular signature for prostate cancer," he says. The group identified more than 200 potential targets--too many to reasonably pursue--that were overexpressed in cancerous prostate tissue. By applying a battery of statistical methods to the data, the team whittled the list of candidates down to a handful. They then used TMAs containing hundreds of clinically stratified prostate cancer specimens to relate expression of each particular protein with clinical outcome. In this way, Chinnaiyan's group identified several genes whose expression correlates with poor clinical prognosis.2-4 Ultimately, this type of research can help clinicians make better diagnoses and better decisions about patient care.
TISSUE ARRAYERS Unlike DNA microarrays, which are usually made by spotting microliter-scale sample volumes directly on a slide, scientists construct tissue microarrays in paraffin blocks. Each tissue core in the array is collected as a "punch"--generally 0.6 to 2.0 mm in diameter--from a donor block of paraffin-embedded tissue, using a needle. A second, slightly smaller needle is used to create a hole in the recipient block. The tissue cores are then arrayed in the recipient block to produce a master block, from which researchers can obtain around 200 individual 5-µm slices.
Steve Leighton, an engineer and former head of the NIH's advanced engineering program, designed the original manual tissue arrayer. The company he founded, Beecher Instruments of Sun Prairie, Wis., has now sold 430 arrayers based on the original design, says Beecher's president and CEO, Dan Rohwer-Nutter. But making arrays with this instrument is time-intensive: Kononen spent two to three weeks constructing his first 645- element microarray.
Beecher is currently alpha-testing a semiautomated instrument that uses a laser sight to help align the needle with the donor block, and an automated arrayer will be available in November. The base automated instrument holds 27 blocks, which can be any combination of donors and recipients, and can process up to 180 samples per hour. The company plans to offer an optional autoloader, which will increase the system's capacity to 500 donor blocks and 30 recipients.
This September, Chemicon International of Temecula, Calif., released its own manual arrayer, the Advanced Tissue Arrayer (ATA100). "The basic difference between the two machines," says Rohwer-Nutter, "is how the two needles are positioned." In Beecher's design, the two needles are on a pendulum-like mechanism; when the donor needle is in place, the recipient needle is not, and vice versa. Chemicon's machine has the two needles on separate posts.
The two designs also differ in the way a researcher aligns the donor needle with the donor block. With Beecher's manual arrayer, a pathologist can mark the regions to punch on the face of the donor block, or the scientist can "just sort of eyeball it," says Rohwer-Nutter. In comparison, Chemicon's ATA100 features a stereomicroscope that holds a reference slide prepared from the donor block. As the user moves the microscope's stage, the donor block moves under its needle as well. The researcher can therefore use the microscope to line up the needle with the donor block, improving the accuracy of sample collection.
Both companies are developing arrayers that can handle frozen tissue samples instead of paraffin-embedded ones. Martin Ferguson, senior vice president of bioinformatics at Ardais in Lexington, Mass., cites two reasons researchers would want to use frozen tissue samples instead of paraffin-embedded ones. First, some antibodies don't work on paraffin-embedded tissues, he says, because the formalin fixation step destroys the antigen. In addition, in situ hybridization experiments often fail with paraffin-embedded tissue samples, "either because you've destroyed the nucleic acid, or at least fixed it in such a way that it's not accessible anymore."
Rohwer-Nutter says Beecher's frozen-tissue arrayer will be semiautomated, because the engineering challenges in building a fully automated one "probably would drive the cost of the instrument well beyond what would be justified for the smaller frozen archives out there." The primary problem is temperature control: keeping the tissue frozen and preventing frost buildup in the instrument. On a more practical level, he says, unlike most paraffin sections, frozen tissue samples do not tend to exhibit uniform consistencies.
IT'S AN 'OMIC' Like standard DNA biochips, TMAs present researchers with significant challenges in data management and bioinformatics. TMA users must keep track of both clinical and experimental data, and they require an easy way to access the visual data associated with the experiments. Each new antibody that researchers use to probe a given array not only adds to that array's value, but also increases the data's complexity.
"Early on we realized what an informatics challenge this would be," says David Rimm, associate professor of pathology at Yale University and director of the Yale Cancer Center Tissue Microarray Facility, "but not only that, but what a goldmine it would be ... if we maintain all this data." Summing up the challenge, NCI's Hewitt says, "It's an 'omic.'" Kallioniemi characterizes it as "pathomics."
Mark Rubin, associate professor of pathology at Brigham & Women's Hospital, Harvard Medical School, helped develop a software system to deal with the image archiving problem while he was an associate professor at the University of Michigan. The software, called Profiler (portal.path.med.umich.edu), allows researchers to examine digital images of individual histological specimens, such as tissue cores from a TMA; evaluate and score them; and store all the data in a relational database.
But this scoring process, as performed by a pathologist, is inherently subjective and imprecise. Whereas DNA biochip information is analyzed by a computer, which provides a value over a large dynamic range, many pathologists score manually using a four-point scale: negative, weak positive, strong positive, or no data. "You're still limited by what I see as the nonquantitative nature of the pathologist," says Rimm.
To work around this problem, Rimm's team developed an automated microscopy device and software package called Aqua (R.L. Camp et al., "Automated subcellular localization and quantification of protein expression in tissue microarrays," Nature Medicine, advance online publication, DOI:10.1038/nm791, Oct. 21, 2002). Rimm says the system can analyze a slide and output a value that indicates the level of expression of a particular protein within a user-defined subcellular compartment. Rubin, one of Rimm's collaborators, says that though the system is not yet widely available, it can already in some respects exceed the capabilities of a pathologist. "You're actually asking the machine to first make a diagnosis, just as a pathologist would do, and then to quantify that evaluation, which a pathologist cannot do," he says.
Chih Long Liu, working with van de Rijn while an undergraduate student at Stanford, developed a solution to some TMA bookkeeping headaches.5 Now a graduate student at Harvard Medical School, Liu and his colleagues wrote two programs: TMA-Deconvoluter and Stainfinder (genome-www.Stanford.edu/TMA/). TMA-Deconvoluter is a series of Excel macros that helps researchers get TMA data into a format that can be read by conventional data analysis tools like Cluster and TreeView (rana.lbl.gov). Cluster runs a hierarchical cluster analysis on the TMA data, helping users to interpret the highly complex datasets obtained from TMAs stained with large numbers of antibodies, and TreeView allows researchers to browse the clustered data. Stainfinder is a Web interface that links the clustered TMA data to an online image database, allowing scientists to rapidly reevaluate the data and compare different stains on the same core.
To produce such an image database, researchers must capture digital images of the tissue cores. Most microscope manufacturers include software to process TMAs with their instruments, says Kallioniemi. Alternatively, scientists can use automated slide scanners like the BLISS imaging system, developed by Bacus Laboratories of Lombard, Ill., or the ScanScope slide scanner from Aperio Technologies of Vista, Calif.6
GETTING AND MAKING ARRAYS Several companies supply off-the-shelf TMAs; many of these companies offer custom services as well. Researchers supply the tissues to be arrayed, or sometimes even the animals from which the specimens should be harvested, and the company produces a master block to customer specifications.
Scientists can often obtain specimens for TMAs from tissue archives. Medical schools generally maintain sample repositories, which can grow to considerable size. Yale's pathology archive, for instance, contains three million tissue blocks representing one million cases, according to Rimm. Alternatively, researchers can query the NCI's Specimen Resource Locator (www.cancer.gov/specimens). But university resources tend to serve mostly academic researchers, shutting out private concerns, says Ardais' Ferguson. To fill this niche, the company is amassing a tissue repository of its own through its National Clinical Genomics Initiative, which caters to both academic and private entities.
Kallioniemi says the driving force behind the development of TMAs was to take the genomics approach and apply it to the study of clinical tissue samples. Four years later, he says, it now appears that early vision was prescient: "We really will need large sample sets of tumors to realize the potential and the clinical application of genomics in practice."
Jeffrey M. Perkel can be contacted at firstname.lastname@example.org.
1. J. Kononen et al., "Tissue microarrays for high- throughput molecular profiling of tumor specimens," Nature Medicine, 4:844-7, 1998.
2. S.M. Dhanasekaran et al., "Delineation of prognostic biomarkers in prostate cancer," Nature, 412:822-6, 2001.
3. M.A. Rubin et al., "a-methylacyl coenzyme A racemase as a tissue biomarker for prostate cancer," Journal of the American Medical Association, 287:1662-70, April 3, 2002.
4. S. Varambally et al., "The polycomb group gene EZH2 is involved in progression of prostate cancer," Nature, 419:624-9, Oct. 10, 2002.
5. C.L. Liu et al., "Software tools for high-throughput analysis and archiving of immunohistochemistry staining data obtained with tissue microarrays," American Journal of Pathology, in press.
6. J.M. Perkel, "Microscopy goes virtual--and global," The Scientist, 16:51, Sept. 30, 2002.