A meta-analysis of kidney microarray datasets: investigation of cytokine gene detection and correlation with rt-PCR and detection thresholds
Department of Surgery, Mayo Clinic College of Medicine, Rochester, MN, USA
BMC Genomics 2007, 8:88 doi:10.1186/1471-2164-8-88Published: 30 March 2007
Microarrays provide a means to simultaneously examine the gene expression of the entire transcriptome in a single sample. Many studies have highlighted the need for novel software and statistical approaches to assess the measured gene expression. Less attention has been directed toward whether genes considered undetectable by microarray can be detected by other strategies or whether these genes can provide accurate gene expression determinations. In the kidney this is a concern for genes such as cytokines which dramatically influence the immune response but are often considered low abundance genes produced by a small number of cells.
Using both publicly available and our own microarray datasets we analyzed the detection p-value and detection call values for 81 human kidney samples run on the U133A or U133Plus2.0 Affymetrix microarrays (Affymetrix, Santa Clara, CA). For the cytokine genes, the frequency of detection in each sample group (normal, transplant and renal cell carcinoma) was examined and revealed that a majority of cytokine related genes are not detectable in human kidney by microarray. Using a subset of 29 Mayo transplant samples, a group of seven transplant-related cytokines and eight non-cytokine genes were evaluated by real-time PCR (rt-PCR). For these 15 genes we compared the impact of decreasing microarray detection frequency with the changes in gene expression observed by both microarray and rt-PCR. We found that as microarray detection frequency decreased the correlation between microarray and rt-PCR data also decreased.
We conclude that, when analyzing microarray data from human kidney samples, genes generally expressed at low abundance (i.e. cytokines) should be evaluated with more sensitive approaches such as rt-PCR. In addition, our data suggest that the use of detection frequency cutoffs for inclusion or exclusion of microarray data may be appropriate when comparing microarray and rt-PCR gene expression data and p-value calculations.