Validation of oligonucleotide microarray data using microfluidic low-density arrays: a new statistical method to normalize real-time RT-PCR data

Abstract

Profiling studies using microarrays to measure messenger RNA (mRNA) expression frequently identify long lists of differentially expressed genes. Differential expression is often validated using real-time reverse transcription PCR (RT-PCR) assays. In conventional real-time RT-PCR assays, expression is normalized to a control, or housekeeping gene. However, no single housekeeping gene can be used for all studies. We used TaqMan Low-Density Arrays, a medium-throughput method for real-time RT-PCR using microfluidics to simultaneously assay the expression of 96 genes in nine samples of chronic lymphocytic leukemia (CLL). We developed a novel statistical method, based on linear mixed-effects models, to analyze the data. This method automatically identifies the genes whose expression does not vary significantly over the samples, allowing them to be used to normalize the remaining genes. We compared the normalized real-time RT-PCR values with results obtained from Affymetrix Hu133A GeneChip oligonucleotide microarrays. We found that real-time RT-PCR using TaqMan Low-Density Arrays yielded reproducible measurements over seven orders of magnitude. Our model identified numerous genes that were expressed at nearly constant levels, including the housekeeping genes PGK1, GAPD, GUSB, TFRC, and 18S rRNA. After normalizing to the geometric mean of the unvarying genes, the correlation between real-time RT-PCR and microarrays was high for genes that were moderately expressed and varied across samples.