Microarrays have enjoyed center stage in cancer research for a while now, spurring hope that they will become improved diagnostic and prognostic tools. In particular, oncologists would like to use them to predict whether or not a cancer is going to spread in the body, how likely it will respond to a certain type of treatment, and how long the patient will probably survive. But what will it take to translate results from early studies into clinical applications, and what role will microarrays play in this process?
After a first wave of studies that showed microarrays could differentiate between subtypes of tumors that could already be distinguished by other methods, a ”’second wave’ is being able to tell what a tumor is going to do in the future, and those data are just starting to emerge, and they are a lot messier,” said Dietrich Stephan, an assistant professor of pediatrics at Children’s National Medical Center in Washington, D.C.
It would be useful, for example, if gene expression profiles could distinguish between subtypes of tumors that standard methods, such as histological pathology from a biopsy, fail to discriminate, and that require different treatments. For example, subtypes of certain pediatric sarcomas ”that have the same look to them under standard pathological examination do have very different kinds of treatment, and it’s important to get them right,” said Michael Bittner, an associate investigator in the cancer genetics branch of NHGRI. At the molecular level, many tumors fall into different classes, he said, making researchers hopeful that microarrays will allow them to discriminate between these groups.
"Because these methods are able to look at a larger number of markers and because disease states that are currently given the same diagnosis are in fact quite heterogeneous in their clinical behavior in response to therapy, this gene expression profiling will result in better diagnostic and prognostic schemes than current methods, and will probably complement current methods," said Stephen Schmechel, a hematophathology research fellow at the University of Washington department of laboratory medicine.
Results From Different Platforms Don't Match Up
However, there are significant obstacles to overcome. One of the main problems in getting to the right profiles is variability between different microarray platforms. ”There is growing consensus that the data between an Affymetrix platform and spotted cDNA arrays may not be directly comparable,” said Schmechel, ”which simply says we don’t know what we are measuring at this point.” The problem, he believes, lies mainly in the different probes each system uses, leading to different degrees of cross-hybridization and sensitivity, even if the same gene is to be measured. As an example, he cited two studies, one from Stanford University using a cDNA platform, the other from the Whitehead Institute using an Affymetrix platform, that both analyzed diffuse large B-cell lymphoma and found almost entirely different lists of gene expression markers.
What is required in response is more validation of microarray data, according to Schmechel: ”There has been far too little corroboration of data in this field using secondary methods like quantitative PCR, Northern blotting, Western blotting, [and] other techniques. A lot of the literature is based on array data with all these concerns [of] what it is we are measuring.”
Another reason for concern is the heterogeneity of tumors themselves, which consist of a mixture of normal and malignant cells, with blood vessels in between. Separating cancer cells out before a microarray experiment, by techniques like laser capture microdissection, for example, does not necessarily solve the problem, according to Stephan. ”Even if you pulled out some cancer cells from that tumor, there is no guarantee that those are the cells that are going to metastasize, just because tumors are heterogeneous,” he said.
To establish complex correlations, he believes, large amounts of data must be collected: ”What’s really needed in the field is a very large meta-analysis of huge numbers of tumors, all done in the same way and all with really detailed information on them.” But getting people to participate in such a study might not be easy, he conceded. ”It’s very difficult to convince people to give up their samples.” Other researchers agree that only large datasets can provide enough assurance about the results. According to John Quackenbush, a bioinformaticist at The Institute for Genomic Research in Rockville, Md., scientists at a recent workshop at the NCI discussing bioinformatics issues complained that ”the tools themselves, and our evaluation of them, [are] currently limited by the availability of data. What we really need at this point are expression profiles from hundreds or thousands of tumors linked to relevant, and appropriate, clinical data.”
There is certainly consensus about the need for more clinical studies, in particular prospective studies that can establish a link between a treatment outcome and a gene expression profile. But opinions differ about what is the best way to assay gene expression markers in a clinical application. Most researchers do think it is possible to reduce the number of markers to one or two dozen without compromising results. Using appropriate statistical methods, ”a relatively small number of markers should fairly adequately diagnose most clinical conditions, such as different tumor types,” said Schmechel.
But microarrays, some believe, are just too complicated to handle since they require numerous steps to preserve the RNA, prepare the sample, and conduct the experiment. ”If you wanted to get [gene expression markers] into the clinic in a reasonable amount of time, you would want to use as your method of detection something that pathologists already use in the lab,” said Bittner, for example immunohistochemistry or ELISA-type assays. According to Schmechel, the method of choice might be quantitative RT-PCR, and he said that several clinical studies using this technique are currently being prepared. ”RT-PCR is easier, much less expensive, and already routinely used in molecular diagnostic labs,” he said, and appears to be more sensitive and provide a larger range of linearity than microarrays. ”Second generation” arrays that allow for the sensitivity and specificity expected from a clinical assay might be another suitable format, he added, but ”clinical implementations will first be in technologies that are already widely used in pathology and laboratory medicine.”
Finally, FDA approval will determine which platform will win in the end, and a large number of markers, such as 100, might not be the best starting point. ”There is – to my knowledge – no FDA-approved device with that many variables,” said Bittner. ”The FDA is certainly considering how they should be designing their assessment of such systems now.”
The timeline for developing gene expression-based diagnostics will largely depend on the duration of prospective studies, which can run for several years. ”I think within the next year, you will see many studies organized around expression profiling; within two years, you will start to see markers proven in prospective trials [that will then] move into more routine clinical use,” said Schmechel.