Changed relative to what? Housekeeping genes and normalization strategies in human brain gene expression studies

Many studies in biological psychiatry compare the abundance of individual messenger RNAs between cases and control subjects or, more recently, between genotype groups. Most utilize some form of normalization procedure, usually expressing the transcript(s) of interest relative to that of a housekeeping gene or genes (also called reference genes), to overcome various sources of experimental error. Indeed, normalization is such a standard procedure that its purpose, principles, and limitations are sometimes overlooked, and some papers lack sufficient information as to its implementation. Here, we review the rationales for normalization and argue that in well-conducted psychiatric gene expression studies using human brain tissue, it is reducing intersubject variability rather than experimental error that is the major benefit of normalization. We also review the conceptual and empirical basis for the category of housekeeping genes-i.e., genes with a ubiquitous and invariant expression. We conclude that the evidence is against any such simple categorization and that a more pragmatic, less dogmatic, approach to the selection and implementation of reference genes is required, which takes into account the particular issues that pertain to human brain tissue studies. This pragmatism extends to the issue of whether normalization should be to one or multiple reference genes. We end by making several recommendations toward a more flexible, transparent, and comprehensive approach to data presentation and analysis. We illustrate the review with examples from studies of schizophrenia and mood disorder.