Microarrays Are a Robust Platform Suitable for Diagnostics.

So far, gene expression profiling is used for the classification of malignant diseases and to identify novel prognostic markers and potential therapeutic targets in a research setting. We were interested whether the microarray platform also qualifies for a diagnostic setting which requires robustness of the technique itself and the markers used. Several types of microarray platforms exist, e.g. short DNA-oligonucleotide and cDNA microarrays. Platforms have evolved by increasing the information content on each array, mainly by shrinking the feature size of represented probes. These differences can cause difficulties in data comparisons and might impair the transferability of results from research studies to large-scale clinical trials. Therefore, we designed a validation study to address the robustness and accuracy of diagnostic marker patterns in a prospective cohort of patients. Gene expression signatures had been identified to discriminate different leukemia types using Affymetrix DNA-oligonucleotide microarrays HG-U133A and HG-U133B. These predefined diagnostic gene expression signatures were tested in a study using the HG-U133 Plus 2.0 microarray and the GeneChip 2.0 Dx diagnostics platform. The study included prospectively collected samples from leukemia patients at diagnosis. More than 90% of the samples had been sent consecutively to our laboratory between January and July 2004. The median shipment time was 24 hours. The median percentage of blasts in acute leukemias was 80%. A total number of 411 target preparations was performed and resulted in 400 hybridization cocktails (97.3% success rate). The sample target preparation protocol included the current assay recommendations for nucleic acid quantification and cleanup, IVT labeling and hybridization and allows to prepare samples in less than 24 hours. The microarrays were automatically washed and stained with the fluidics station. A scanner with removable 48-array autoloader carousel maximized array throughput. For statistical comparison both array designs were normalized using the recommended set of 100 housekeeping genes (scaling to common target intensity). The U133 set reference data matrix was applied to train the classifier (Support Vector Machines). The 400 independent blinded validation samples were predicted using the set of differentially expressed genes from the training data set. Overall 378/400 (94.5%) of the independent test samples were correctly classified. Some of the misclassifications could be explained by the underlying biology of the respective samples. In conclusion, new advances in gene expression profiling, particularly with regard to instrumentation and reliability of assays for sample target preparation have paved the way to now enroll patients in prospective multi-center trials. In these microarrays can be tested as an additional routine diagnostic method in parallel to gold standard procedures. Moreover, the design of a custom 11 microns feature-sized array will also optimize costs and needs for sample material.