Methodological considerations for gene expression profiling of human brain

Abstract

Gene expression profiles of postmortem brain tissue represent important resources for understanding neuropsychiatric illnesses. The impact(s) of quality covariables on the analysis and results of gene expression studies are important questions. This paper addressed critical variables which might affect gene expression in two brain regions. Four broad groups of quality indicators in gene expression profiling studies (clinical, tissue, RNA, and microarray quality) were identified. These quality control indicators were significantly correlated, however one quality variable did not account for the total variance in microarray gene expression. The data showed that agonal factors and low pH correlated with decreased integrity of extracted RNA in two brain regions. These three parameters also modulated the significance of alterations in mitochondrial-related genes. The average F-ratio summaries across all transcripts showed that RNA degradation from the AffyRNAdeg program accounted for higher variation than all other quality factors. Taken together, these findings confirmed prior studies, which indicated that quality parameters including RNA integrity, agonal factors, and pH are related to differences in gene expression profiles in postmortem brain. Individual candidate genes can be evaluated with these quality parameters in post hoc analysis to help strengthen the relevance to psychiatric disorders. We find that clinical, tissue, RNA, and microarray quality are all useful variables for collection and consideration in study design, analysis, and interpretation of gene expression results in human postmortem studies.

Figure 1

The significance of comparing groups separated by differences in high vs low RNA quality, clinical quality, and tissue quality is shown for multiple variables commonly used in controlling gene expression profiling experiments and in particular microarray results (%PC, SF, ACI, and Type 1 / 2). The data is from Table 2 for the anterior cingulate cortex and shows on the y-axis the significance of t-test (p value transformed by −log10) and the x-axis shows the individual variables. The abbreviations are same as shown in Table 1. The legend shows three different groups (AFS, pH, and RNA).

Figure 2A. Histogram of pH by AFS

Figure 2B. Histogram of RNA quality (28S/18S) by AFS

Figure 2C. Histogram of pH by Type ‘1’ and Type ‘2’ samples. The type ‘1’ and ‘2’ refer to groups formed in hierarchical clustering that were associated with pH differences between clusters

Figure 3

The diagnostic performance of test variables, or the ability of a variable to discriminate microarray outcome (Type ‘1’ vs Type ‘2’) is evaluated using Receiver Operating Characteristic (ROC) curve analysis. The ROC curves are shown for 6 quality variables that relate differentially to microarray outcome. The area under each curve shows a relationship to overall prediction of Type ‘1’ and ‘2’ microarray outcome. The comparison of ROC curves for tissue quality (pH, freezer time, PMI), clinical quality (AFS, age), or RNA quality (28S / 18S) shows that tissue quality pH measure is strongly related to Type 1 and Type 2 microarray outcome (Li et al., 2004). Table 6 shows the ROC values for each variable.

Figure 4

The sources of variation in an ANCOVA of multiple covariates shows that the RNA degradation accounts for the highest average effect (F-ratio) across the entire transcriptome in ACC measured on an Affymetrix U133A chip. The AffyRNAdeg program (Cope, 2005) provides a slope, which is a microarray chip based indicator of the decline in signal across a transcript, thus a putative index of RNA degradation. The abbreviations for each variable are described in Table 1.