One of the last sessions on the opening day of the annual Association of Biomolecular Resource Facilities meeting in Portland, Ore., this week prompted Barry Merriman, a researcher in the Department of Human Genetics at the University of California, Los Angeles, to speculate, with a smile, that “maybe a fist-fight will break out.”
That was quite an unlikely sentiment to be voiced in a collegial scientific gathering of this type, but not so unusual when the future of a nearly $1 billion industry, or the financial pegs supporting the growth of the emerging microarray industry, were potentially at stake.
Merriman was serving as chairman of a roundtable discussion entitled “Choosing the Right Microarray,” attended by some 150 of the nearly 850 scientists at the annual meeting.
On the agenda was the presentation of results from an ambitious, broad study of microarray platforms conducted by a consortium sponsored by the National Institute of Neurological Disorders and Stroke and the National Institute of Mental Health. [To view the results, click here.] Members of the consortium, which was created to provide access to microarray technologies for sponsored researchers, include the microarray center of the Translational Genomics Research Institute, the Duke University microarray center, and UCLA’s microarray core laboratory. [http://arrayconsortium.cnmcresearch.org]
If any in the crowd came thinking that the session would provide a red-line delineation of exactly what microarray platform to purchase, they were surely disappointed.
There was no “Right Microarray” platform crowned here, but more questions were raised about how to quantify and objectively compare the accuracy, reliability, sensitivity, and reproducibility of the leading microarray technologies.
“Overall there is good repeatability and fair concordance between the major commercially fabricated microarray platforms and spotted cDNA/oligo arrays. However, substantial disagreements arise as well,” the study concluded.
The platforms give rather different results, Merriman told BioArray News. “This is easy to put in concrete terms, which says that if you made a greater than twofold change list for kidney vs. spleen in Platform A, and in Platform B, the overlap of the two lists would only vary by the fraction listed in the table — e.g., for Affymetrix vs. Amersham, it’s only 32 percent [e.g. if Affy found 500 and Amersham found 500, they would agree on 32 percent of 500 + 500 = 320] and similar[ly] for any other two platforms, so they are not very concordant on identifying differentially expressed genes.”
“Overall, he said, [while] “the cDNA arrays seemed to do worse than all other platforms, they are at least concordant with the others, in aggregate, suggesting that cDNAs are not the way to go due to cross-hybridizing.”
The study was based upon assaying aliquots from the same four total RNA samples derived from human liver, kidney, and spleen, as well as the Stratagene Human Reference pool. The subjects of the study were the Affymetrix, Agilent Technologies, and Amersham Biosciences platforms as well as spotted arrays produced from whole-genome oligo sets from Operon, ClonTech, and MWG Biotech, plus 33,000-clone spotted cDNA arrays. Validation of results was conducted with RT-PCR targeting the 100 genes most discordant between the platforms tested, as well as SAGE analysis.
Laboratories involved in the analysis outside of the consortium core labs included: NIH, which conducted experiments on the cDNA arrays; the University of Pennsylvania and TGEN, which conducted the Affymetrix experiments; Duke, which tested the Operon arrays; UCLA, which did Affymetrix, Agilent, and ClonTech-spotted arrays; the University of Southern California, which tested the MWG-based arrays, and Amersham, which conducted and analyzed experiments on its CodeLink line of microarrays. The QT-PCR validation was conducted by the laboratory of James Willey of the Medical College of Ohio, and Duke conducted the SAGE analysis of the universal RNA sample.
“Every lab that generated data was using a platform they were well- versed in the art for,” said Merriman.
The study, which generated 10 Gb of raw data and 30 Mb of condensed data, did not address the question of low-level expression.
Merriman said that the consortium is planning to make the data available publicly.
Stanley Nelson, a professor in the department of human genetics at UCLA, and director of the UCLA DNA Microarray core lab, as well as program chairman of the ABRF meeting, said the study actually showed improvement in the platforms from the last time a study of this type was conducted.
As for how the study might reflect on microarrays as potential diagnostics, he was confident.“For brain tumors, it’s more reliable than a pathologist,” he said.