OBJECTIVE: Normalization of mRNA data, i.e., the calculation of mRNA expression values comparable in between different experiments, is a major issue in biomedical and orthopaedic/rheumatology research, both for single-gene technologies [Northern blotting, conventional and quantitative polymerase chain reaction (qPCR)] and large-scale gene expression experiments. In this study, we tested several established normalization methods for their effects on gene expression measurements. METHOD: Five standard normalization strategies were applied on a previously published data set comparing peripheral and central late stage osteoarthritic cartilage samples. RESULTS: The different normalization procedures had profound effects on the distribution as well as the significance values of the gene expression levels. All applied normalization procedures, except the median absolute deviation scaling, showed a bias towards up- or down-regulation of genes as visualized in volcano plots. Of interest, the P-values were much more depending on the normalization procedure than the fold changes. Ten commonly used housekeeping genes showed a significant variability in between the different specimens investigated. The gene expression analysis by cDNA arrays was confirmed for these genes by qPCR. CONCLUSION: This study documents how much normalization strategies influence the outcome of gene expression profiling analysis (i.e., the detection of regulated genes). Different normalization approaches can significantly change the P-values and fold changes of a large number of genes. Thus, it is of vital importance to check every individual step of gene expression data analysis for its appropriateness. The use of global robustness and quality measures for analyzing individual outcomes can help in estimating the reliability of final microarray study results.
OBJECTIVE: Normalization of mRNA data, i.e., the calculation of mRNA expression values comparable in between different experiments, is a major issue in biomedical and orthopaedic/rheumatology research, both for single-gene technologies [Northern blotting, conventional and quantitative polymerase chain reaction (qPCR)] and large-scale gene expression experiments. In this study, we tested several established normalization methods for their effects on gene expression measurements. METHOD: Five standard normalization strategies were applied on a previously published data set comparing peripheral and central late stage osteoarthritic cartilage samples. RESULTS: The different normalization procedures had profound effects on the distribution as well as the significance values of the gene expression levels. All applied normalization procedures, except the median absolute deviation scaling, showed a bias towards up- or down-regulation of genes as visualized in volcano plots. Of interest, the P-values were much more depending on the normalization procedure than the fold changes. Ten commonly used housekeeping genes showed a significant variability in between the different specimens investigated. The gene expression analysis by cDNA arrays was confirmed for these genes by qPCR. CONCLUSION: This study documents how much normalization strategies influence the outcome of gene expression profiling analysis (i.e., the detection of regulated genes). Different normalization approaches can significantly change the P-values and fold changes of a large number of genes. Thus, it is of vital importance to check every individual step of gene expression data analysis for its appropriateness. The use of global robustness and quality measures for analyzing individual outcomes can help in estimating the reliability of final microarray study results.