Date Published:
09/2004
Publication Type:
Journal Article
Authors:
Secondary:
Frontiers in Bioscience: A Journal and Virtual Library
Volume:
9
Pagination:
2935-46
URL:
https://pubmed.ncbi.nlm.nih.gov/
Keywords:
Cluster Analysis;External;Models;Principal Component Analysis;Statistical
Abstract:
<p>We previously showed that specific strains of human immunodeficiency virus (HIV)-1 infect the brain and contribute to Neuropathology, Cognitive Distress, and Neuropsychiatric Disease. To study further brain disease that results from HIV-1 infection, we commenced analysis of changes in gene expression in brain. We analyzed RNA purified from Frontal Cortex of 5 HIV-1 infected and 4 HIV-1 negative control subjects RNA was amplified and Affymetrix technology was used to analyze gene expression using the 12,585 gene Affymetrix Human Genome U95A chip. The expressed genes showed highly significant Pearsons correlations with each other within the two groups. Expression intensities were transferred to Microsoft Excel and Spotfire was used to analyze the results. Twenty-group K-means cluster analysis was done for HIV+ and HIV- subjects. Genes that were expressed in the same cluster numbers in the two groups were removed from further analysis. Analysis of Gene expression in the top 13 HIV+ clusters showed expression in the 40 gene categories designated in our prior studies. Genes from several categories occurred in more than one K-means cluster. Genes identified in these lists included several genes that have been previously studied: MBP, Myelin-PLP, NMDA receptor, MAG, astrocytic protein, Notch 3, APP, Senescence, proteasome, Ferritin, signaling, cell cycle, iNOS, Chemokine, splicing, synapse, protein tags, and ribosomal proteins. The first (primary significant) axis of both Principal Component Analyses ordered the genes in the same patient groups as the K-means cluster analysis for the respective patient groups. PCA was thus not more informative than K-Means cluster analysis. Ratios of HIV+ to HIV- intensities were calculated for all the averaged gene expression intensities. The ratio range was 0.14 to 9.26. The genes at the extremes (ad extrema) did not correspond to the gene order by K-means clustering (or PCA). The genes in the top 13 K-means clusters showed low-level changes by expression ratio. Genes ad extrema by ratio were in clusters with very large memberships. Mann-Whitney analysis confirmed expression ratio results. Several inferences result from our preliminary study. First, study design will be different in future studies involving additional replicates. Second, ratios inform us of the extent of changes in gene expression quantitatively. Third, Cluster methodology provides us with more subtle information, how bunches (clusters) of genes behave in terms of their centroids (attractors). Fourth, genes that change extensively by ratio tend to be in the larger k-Means clusters. We conclude that ranking gene expression with the use of expression ratio or by K-means clustering, yield different representations of the data.</p>