Principal component analysis, using all probe sets that passed through the filters described in Methods, indicated that regional differences in gene expression were greater than strain differences, as illustrated by a biplot of the first and second principal components (). The clusters representing arrays from the ACB, CPU, and FC are tightly grouped, with the exception of one outlier in each region. Those from the HIPP give a fairly good grouping, but not as tight as the other three regions; the AMYG gives the least grouping of data, with one outlier that groups with the HIPP cluster. As can be seen from the figure, there is no clear pattern of separation of the iP and iNP lines within brain region. Visualization of the third and fourth principal components (not shown) also did not resolve line differences within regions.
Fig. 1 Principal components plot of all 59 arrays. Only processed, filtered data were used (see Methods for details) in the computation. Principal component 1 is plotted on the horizontal axis and principal component 2 is plotted on the vertical axis. Clusters (more ...)
We used LIMMA (Smyth, 2004
) on RMA preprocessed (Irizarry et al., 2003
transformed, filtered data (see Methods for details), to determine differences between the inbred strains. The FDR, as calculated by the method of Benjamini and Hochberg (1995)
, was set at ≤0.10. In an analysis of individual brain regions (n
= 6 iNP, 6 iP in all regions except the FC, in which data were missing from one iP animal), significant differential gene expression was found in the five brain regions (Tables and ). The number of differences between the two inbred lines in each of the five individual regions ranged from 8 to 63, with the order of number of differences being AMYG > CPU > HIPP > ACB > FC (). Across regions, the total number of genes that demonstrated differential expression ranged from 8 to 54; the number of genes that were located within established alcohol QTLs (Bice et al., 1998
; Carr et al., 1998
; Foroud et al., 2002
; Radcliffe et al., 2004
; Terenina-Rigaldie et al., 2003
) ranged from 1 to 16. lists the significant genes that were different within each region along with individual fold changes and gene symbols. contains detailed expression levels and standard deviations for all the regionally significant probe sets.
Summary of biostatistical and bioinformatic analyses for individual regions and for the combined regions examining the main effect of straina
Regionally significant genes (FDR < 0.10)
Detailed expression levels and S.D.s for all regionally significant probe sets
Over-represented GO biologic process categories (P
<.05) were statistically sought using the Bioconductor package GOstats (Gentleman et al., 2004
). Briefly, GO categories of significant genes were tested for over-representation using the hypergeometric distribution. An analysis of each of the five individual brain regions did not reveal any over-represented GO biologic process categories, perhaps because there were too few differentially expressed genes in each region.
An overall analysis of all brain regions, which used a linear model with strain and brain region as factors, demonstrated significant differential expression of 296 genes (351 probe sets) (FDR ≤0.10). Of these, 173 genes (198 probe sets) demonstrated higher and 123 genes (153 probe sets) lower expression levels in the iP rats than in the iNP rats (). Although almost all of the genes identified as being differentially expressed within the individual regions were also identified in the overall analysis, the latter method detected many that fell below the threshold of statistical detection for individual regions, most likely because of the increased power gained by including a much larger number of arrays in the linear model (). There were 53 genes (71 probe sets) found significant in at least one individual region and in the overall analysis; 241 genes (280 probe sets) were found significant only in the overall analysis. Additionally, there were 15 genes (17 probe sets) found significant in the AMYG that were not significant in the overall analysis, one of which was also significant in the ACB.
Complete list of probe sets that were significantly different between iP and iNP rats in the overall analysis
Significant fold differences in combined region analysis and in individual region analyses
Analysis for over-represented GO categories in the set of significant genes found with the overall analysis indicated that 17 GO categories were over-represented (P <.05). Six of the most neurobiologically interesting categories are axon guidance, gliogenesis, negative regulation of programmed cell death, regulation of programmed cell death, regulation of synapse structure function, and transmission of nerve impulse (); the full list can be viewed in . More genes demonstrate increased expression in the iP versus iNP rat comparison in the categories of regulation of cell death, negative regulation of cell death, regulation of synapse structure and function, and gliogenesis; whereas, more genes demonstrate decreased expression in the iP versus iNP in the category of axon guidance. Equal numbers of genes with increased and decreased expression in the iP versus iNP are present in the category of transmission of nerve impulse.
Biologically interesting, nonredundant, significantly over-represented GO biologic process categoriesa
All significant Gene Ontology (GO) biologic process categories
For the purposes of further discussion, the authors found it desirable to use a gene classification schema that would include genes found significant in individual regions and that would include functional neuroinformatic categories such as neuroplasticity. Such customized, neuroinformatic classifications of genes found significant in any of the individual regions, or in the combined analysis, were obtained by first assigning all genes to their appropriate GO biologic process categories, regardless of statistical significance of the category. Of the categories obtained, further selection was made based on the number of genes within category and degree of neuroinformatic interest. The resultant GO categories were synaptic transmission, regulation of action potential, positive regulation of transport, potassium ion transport, chloride transport, metal ion homeostasis (collapsed to form “neurotransmission,” ); apoptosis, axon guidance, axonogenesis, cell migration, central nervous system development, cytoskeleton organization and biogenesis, gliogenesis, myelination, negative regulation of apoptosis, neuron differentiation, positive regulation of apoptosis, regulation of apoptosis, regulation of axonogenesis, regulation of neurogenesis, regulation of programmed cell death, and regulation of synapse structure and function (collapsed to form “neuroplasticity,” ); and small GTPase mediated signal transduction, negative regulation of protein kinase activity, negative regulation of enzyme activity, transmembrane receptor protein tyrosine kinase signaling pathway (collapsed to form “intracellular messaging,” ). Finally, the lists of genes thus classified were supplemented with manual curation based on a search of the literature. One last category was obtained by classifying genes as transcription factors based on results obtained with Genomatix Suite (Genomatix, Munich, Germany; ).
Neurotransmission-related significant genesa
Neuroplasticity-related significant genesa
Intracellular messaging-related significant genesa
Transcription-related significant genesa
3.1. qRT-PCR confirmation
qRT-PCR measurements were conducted on genes to verify differences observed with microarray hybridization. Genes were selected on the basis of significant differential expression in at least one brain region and reasonable fold changes (). GSTa4 and Scn1a are not differentially expressed using the RMA analysis presented here, but were with an earlier analysis using MAS5 (unpublished observation). The direction of the expression differences using qRT-PCR was the same as in the microarray data for all genes and regions tested. For the Akap11 gene, the magnitude of the change of the qRT-PCR data for all five regions was far greater than that observed in the microarray data (fold change in CPU was 785, and in HIPP 1,900), which probably reflects technical differences between the two assays. Expression levels of Akap11 in iNP were near back-ground levels in the array experiment; qRT-PCR assays allow one to measure much lower levels of RNA with more accuracy in that range (background is much lower), which allows one to better estimate the fold change.
Quantitative RT-PCR confirmation of microarray fold change (iP vs. iNP) data