Data-Driven Analysis of Age, Sex, and Tissue Effects on Gene Expression Variability in Alzheimer's Disease

Abstract

Alzheimer's disease (AD) has been categorized by the Centers for Disease Control and Prevention (CDC) as the 6 th leading cause of death in the United States. AD is a significant health-care burden because of its increased occurrence (specifically in the elderly population), and the lack of effective treatments and preventive methods. With an increase in life expectancy, the CDC expects AD cases to rise to 15 million by 2060. Aging has been previously associated with susceptibility to AD, and there are ongoing efforts to effectively differentiate between normal and AD age-related brain degeneration and memory loss. AD targets neuronal function and can cause neuronal loss due to the buildup of amyloid-beta plaques and intracellular neurofibrillary tangles. Our study aims to identify temporal changes within gene expression profiles of healthy controls and AD subjects. We conducted a meta-analysis using publicly available microarray expression data from AD and healthy cohorts. For our meta-analysis, we selected datasets that reported donor age and gender, and used Affymetrix and Illumina microarray platforms (8 datasets, 2,088 samples). Raw microarray expression data were re-analyzed, and normalized across arrays. We then performed an analysis of variance, using a linear model that incorporated age, tissue type, sex, and disease state as effects, as well as study to account for batch effects, and included binary interactions between factors. Our results identified 3,735 statistically significant (Bonferroni adjusted p < 0.05) gene expression differences between AD and healthy controls, which we filtered for biological effect (10% two-tailed quantiles of mean differences between groups) to obtain 352 genes. Interesting pathways identified as enriched comprised of neurodegenerative diseases pathways (including AD), and also mitochondrial translation and dysfunction, synaptic vesicle cycle and GABAergic synapse, and gene ontology terms enrichment in neuronal system, transmission across chemical synapses and mitochondrial translation. Overall our approach allowed us to effectively combine multiple available microarray datasets and identify gene expression differences between AD and healthy individuals including full age and tissue type considerations. Our findings provide potential gene and pathway associations that can be targeted to improve AD diagnostics and potentially treatment or prevention.

Microarray Gene Expression Dataset Re-Analysis Reveals Variability in Influenza Infection and Vaccination

“…As we have previously observed (16), meta-analyses using microarray expression data have multiple limitations: Our findings are limited to only genes that have been annotated and are existing probes on the arrays, and also have to be consistently utilized across array platforms. Hence, we are unable to probe global gene expression, and are limited to mRNA profiling.…”

“…The 18 datasets were from Affymetrix and Illumina microarray platforms (Table 1). We modified and implemented the data-analysis pipeline outlined by Brooks et al(16)). To achieve our goal, after curating the datasets, we used the R programming language (17) to pre-process the raw gene expression data and to fit linear mixed effects models to determine statistically significant differentially expressed genes by factor (Figure 1).…”

Microarray Gene Expression Dataset Re-analysis Reveals Variability in Influenza Infection and Vaccination

“…The 18 datasets were from Affymetrix and Illumina microarray platforms (Table 1). We modified and implemented the data-analysis pipeline outlined by Brooks et al (29). To achieve our goal, after curating the datasets, we used the R programming language (30) to pre-process the raw gene expression data and to fit linear mixed effects models to determine statistically significant differentially expressed genes by factor (Figure 1).…”

Meta-analysis of Gene Expression Microarray Datasets in Chronic Obstructive Pulmonary Disease

“…Our meta-analysis also highlighted disease genes that 48 interact with smoking status, and these genes can be used to further characterize the different microarray platforms: Affymetrix GeneChip Human Genome U133 Plus 2.0, 55 Affymetrix Human Gene 1.1 ST Array and Agilent Whole Human Genome Microarray 56 4x44K. Our current meta-analysis pipeline (similar to Brooks et al [18]), included 5 57 main steps (Fig 1): (1) data curation; (2) pre-processing of raw expression data; (3) 58 analysis of variance (ANOVA) on our linear model which compared gene expression 59 changes due to disease state, smoking status, sex and age group; (4) post-hoc analysis 60 using Tukey Honest Significance Difference test (TukeyHSD) for biological significance; 61 and (5) Gene ontology (GO) and pathway enrichment analysis of the differentially 62 expressed and biologically significant genes. 63 Microarray Data Curation from Gene Expression Omnibus and 64 Array Express 65 To gather the datasets for our meta-analysis, we searched the National Center for 66 Biotechnology Information (NCBI)'s data repository, Gene Expression Omnibus 67 (GEO) [19], and the European Bioinformatics Institute (EMBL-EBI)'s data repository, 68 Array Express (AE) [20] for microarray expression data.…”

“…The batch effects can introduce non-biological variation in the data, which affects 127 the interpretation of the results. In order to visualize variation in the expression data stats package (as previously described [18] BH-adjusted TukeyHSD p-values, and all GO terms and pathways with a BH-adjusted 169 p-value <0.05 were considered significant. To find genes that were significantly up-and 170 down-regulated, we further filtered the gene list by difference in means by using the 171 two-tailed 10 and 90% quantile.…”