Control Genes and Variability: Absence of Ubiquitous Reference Transcripts in Diverse Mammalian Expression Studies

Control genes, commonly defined as genes that are ubiquitously expressed at stable levels in different biological contexts, have been used to standardize quantitative expression studies for more than 25 yr. We analyzed a group of large mammalian microarray datasets including the NCI60 cancer cell line panel, a leukemia tumor panel, and a phorbol ester induction time course as well as human and mouse tissue panels. Twelve housekeeping genes commonly used as controls in classical expression studies (including GAPD, ACTB, B2M, TUBA, G6PD, LDHA, and HPRT) show considerable variability of expression both within and across microarray datasets. Although we can identify genes with lower variability within individual datasets by heuristic filtering, such genes invariably show different expression levels when compared across other microarray datasets. We confirm these results with an analysis of variance in a controlled mouse dataset, showing the extent of variability in gene expression across tissues. The results show the problems inherent in the classical use of control genes in estimating gene expression levels in different mammalian cell contexts, and highlight the importance of controlled study design in the construction of microarray experiments.