Systems-Based Analysis of Developmental Programming of Mammary Gland using Bayesian-Variable-Selection

For microarray analysis, there is need for better computational methods in identifying biologically-relevant changes in global gene expression. Bayesian-based variable selection algorithms (BVS) are good candidates for this task because they can identify discriminative variables in high-dimensional data. Objectives: It is hypothesized that BVS can extract the informative gene expression changes in microarray data in a study of diet history effects on offspring gene expression. Methods: BVS was used for analysis of a recent dataset where maternal dietary exposures of regular low-fat (5%) diets, or high-fat (24%) diets based on corn (omega-6) and flaxseed oils (omega-3), were fed to female mice from 2-weeks prior to breeding through gestation, lactation, and early-juvenile stages. At 6-weeks, all mice were shifted to the low-fat diet until termination at 10-weeks. Mammary glands were used for gene expression profiling via microarrays. Postnatal growth was recorded to 10-weeks. Partek software was used for data normalization and ANOVA. BVS identified a discriminative gene subset. Biological processes and molecular functions of this subset were attained by gene ontology enrichment (GOid). Results: Weight increased for high-fat groups with flaxseed > corn. BVS identified a 17 gene subset that optimally discriminates between control and flax groups (100% pup-classification). GOid enrichment correlated a decrease in expression of energy pathways with a reciprocal increase in inflammatory pathways. These inflammation-related pathways include chemotaxis, cytokine and integrin pathways, and T-cell and neutrophil activity. Previous (non-Bayesian) analysis was limited to genes involving KEGG pathways; BVS required no such focusing yet returned similar metabolic results and detection of inflammatory changes. Conclusion: These findings suggest BVS methodologies can extract the informative gene expression changes from microarray data. Dietary fat exposure during pregnancy is shown to significantly influence postnatal growth, metabolic programming, and signaling-networks in the mammary gland of female offspring. Supported by R21 ES013821 and R25 CA044789.