BOSTON--Gene expression microarrays have made a profound impact on biomedical research. The current expectation is that this technology will extend its current role as an experimental tool for basic science research and be increasingly applied in clinical practice. Before this can occur, the issues of cross-platform comparison and integration of data must be addressed. The diversity of available platforms and analytical methods has made comparison of data from multiple platforms challenging. Different laboratories may use different platforms to profile the same genes, but how comparable are they?
Winston Patrick Kuo, DDS, MS, DMSc, post-doctoral researcher in the Department of Oral and Developmental Biology at the Harvard School of Dental Medicine, and colleagues tested nearly all the available commercial and "in-house" gene expression microarray platforms for cross-platform and cross-laboratory comparisons. They have produced a framework for comparisons across platforms and different laboratories, reported in the July 2 online edition of Nature Biotechnology.
This study comes on the heels of several large efforts to create standardized protocols for microarray experiments (from probe annotation to data analysis), such as the Minimum Information About a Microarray Experiment (http://mged.org/Workgroups/MIAME/miame.html), the External RNA Controls Consortium (http://www.cstl.nist.gov/biotech/Cell&TissueMeasurements/GeneExpression/ERCC.htm), and the Microarray Quality Control Project (http://www.fda.gov/nctr/science/centers/toxicoinformatics/maqc/). In addition, major portals for deposition and retrieval of microarray data, such as the Gene Expression Omnibus and ArrayExpress, will only be truly useful if experiments are sufficiently reliable and annotated so meaningful results can be extracted across different platforms. Given the existence of multiple platforms and the corresponding plethora of microarray data, an important issue is whether the platforms measure genes differently, and, if so, how data from different platforms can be compared or combined.
Over the past several years, several small-scale attempts have been made and the reported results have been conflicting. In Kuo's study, gene expression measurements were obtained from 10 different microarray platforms (Affymetrix, Applied Biosystems, Agilent Technologies, Compugen, GE Healthcare, Mergen, MWG BioTech, Operon, academic cDNA and an academic long oligonucleotide array) and two different QRT-PCR approaches (Applied Biosystems TaqMan Assays and Roche Diagnostics Universal Probe Library). It has been an emphasized goal during this effort to develop a sound and consistent framework for cross-platform comparisons.
Kuo aimed to provide unbiased results with clear metrics for evaluations of performance, using established analytical techniques, to conduct the experiments for different platforms as similarly as possible, and to be general enough to allow inclusion of novel academic and commercial platforms as they develop. In order to make the conditions for the different platforms as similar as possible, two distinct RNA samples with wide dynamic range of expression were selected.
"We demonstrated that, after stringent pre-processing of the data, commercial arrays were more consistent than 'in-house' arrays both on internal consistency and agreement with other platforms," Kuo said, "and by most measures, one-dye platforms were more consistent than two-dye platforms." For the most part, the microarray results were highly concordant with QRT-PCR results, especially for highly expressed genes, but variable for genes with lower expression values. Kuo also found that when gene expression measurements are generated from the same platformbut conducted at different laboratories, cross-laboratory variation were significantly smaller than cross-platform variations.
In 2002, Kuo published a cross-platform paper comparing the two most common platforms at the time, Affymetrix and cDNA arrays. The results were not too optimistic, and the analyses were constrained by many factors--one issue was the annotation and mapping of the genes across platforms. So Kuo, based on prior experience, decided to obtain probe sequences from every platform as negotiated prior to the induction of a platform to the study, and for a few of the platforms, this information remains proprietary.
Using probe sequences that were matched at the sequence level (RefSeq and RefSeq exon), improved consistency of measurements across different microarray platforms was observed as compared to using annotation-based (UniGene clusters and LocusLink identifiers) matches. Kuo also found evidence for the importance of careful assay design when using QRT-PCR to confirm microarray results. Higher concordance with microarray data was observed where QRT-PCR assays targeted the same exon as the microarray probes.
"The goal of this study was to illuminate a comparison framework that matched the transcripts at the sequence level," said Kuo. "This is a first report with a relatively large-scale initiative in which the sequences of all probes were known and utilized in such an analysis. The results indicate in the current state of microarrays, there are many available platforms that provide good quality data, or reliability, especially on highly expressed genes, and that between these platforms, there is generally good agreement, or consistency. The ability to detect low expressed genes is still a limitation of microarrays in general. However, there are substantial areas where the platforms disagree, despite considerable developments towards standardization of gene expression profiling, and therefore many issues remain open for investigation."
This is an ongoing study led by Kuo. Additional data has been generated for Agilix, Illumina, MPSS, and SAGE platforms and are in the process of analysis, which, along with other developing technologies (such as Phalanx Bio and Solexa platforms) will be also be included in future comparisons.
This work was supported in part by the National Institutes of Health and by the Bioinformatics Division of the Harvard Center for Neurodegeneration and Repair.
HARVARD SCHOOL OF DENTAL MEDICINE
http://www.hsdm.harvard.edu
Harvard School of Dental Medicine (HSDM) is the oldest university-affiliated dental school in the nation and is committed to educating leaders in dental health. HSDM's mission is to produce leaders who advance knowledge and discovery, serve the community through patient care and advocacy, and contribute to improved oral health and the quality of life. The school offers a unique problem-based learning pre-doctoral curriculum, as well as specialized advanced graduate education programs. HSDM also has a strong interdisciplinary research program.
Journal
Nature Biotechnology