Background
Alzheimer’s disease (AD) brains are characterized by progressive neuron loss and gliosis which involves mostly microglia and astrocytes. Comparative transcriptomic analysis on AD vs. normal brain tissues helps to identify key genes/pathways involved in AD initiation and progression. However, many such studies using bulk brain tissue samples have not considered cell composition changes in AD brains, which may lead to expression changes that are not due to transcriptional regulation.
Methods
Using five large transcriptomic datasets including 1,681 brain tissue samples (882 AD, 799 normal) in total, we first mined frequent co-expression network modules across them, then combined differential expression and differential co-expression analysis on the mined modules in AD versus normal brains. Integrated with cell type deconvolution analysis, we addressed the question of whether the module expression changes are due to altered cellular composition or transcriptional regulation. We then used four additional large AD/normal transcriptomic datasets to validate our findings.
Results
The integrative analysis revealed highly elevated expression level of microglia modules in AD without co-expression change. Decreased expression and elevated co-expression are observed for neuron modules in AD, while significant over-expression and co-expression perturbation are observed in astrocyte modules, all of which has not been previously reported. The expression levels of astrocyte modules also show the strongest correlation with the clinicopathological biomarkers among all cell type specific modules.
Conclusion
Further analysis indicated that the overall increased expression of the core microglia modules can be well explained by the increased microglia cell population in AD brains instead of bona fide microglia genes’ upregulation. In contrast, the decreased expression and perturbed co-expression in AD neuron modules are due to both neuron cell loss and expression regulation of neuronal pathways including differentially expressed transcription factors such as BCL6 and STAT3, which previous study was not able to identify from the shadow of the cellular composition change. Similarly, the strong changes in expression and co-expression in the astrocyte modules may be also due to a combinatory effect from astrogliosis and astrocyte gene activation in AD brains. In this work, we demonstrated that the combinatorial analyses not only provide a powerful approach to delineate the origin of transcriptomic changes in bulk tissue data, but also lead to a deeper understanding of genes in AD.