Hippocampal transcriptomic profiling of Alzheimer’s disease patients and mouse models: Differences and similarities

Abstract

Rodent models of Alzheimer's disease (AD) have provide invaluable insights on AD pathophysiology in the past decades. Therefore, the evaluation of similarities between models and human pathology can improve AD research. Here, we compared hippocampal transcriptomic profiles of human late onset-AD (LOAD) and early-onset AD (EOAD) individuals with three mouse models (hAβ-KI, APP/PS1 and 5xFAD) to evaluate transcriptomic similarities between these animal models and human pathology. While APP/PS1 and 5xFAD are well established EOAD models, hAβ-KI was recently introduced as a potential LOAD model. We hypothesized that APP/PS1 and 5xFAD will have more similiatires to EOAD human AD and hAβ-KI with human LOAD. Five publicly available human AD/cognitively unimpaired transcriptomic profiles were collected from GEO (https://www.ncbi.nlm.nih.gov/geo/), merged and submitted to differential expression analysis (DEA) and functional enrichment analysis (FEA) using R. APP/PS1 mouse model RNAseq data from two datasets (GSE149661 and GSE145907) were also collected. Finally, RNAseq data of 5xFAD and hAβ-KI mouse models were obtained from AMP-AD Knowledge Portal (https://www.synapse.org/) using synapser and synapserutils packages. All animal models were submitted to DEA and FEA. DEA identified 1164, 3261 and 1782 differentially expressed genes (DEGs) for the comparisons between hAβ-KI, 5xFAD and APP/PS1 mutants versus wild type controls, respectively. The hAβ-KI mice model exhibited more DEGs in common with LOAD than with EOAD patients. Interestingly, APP/PS1 and 5xFAD mice also presented more DEGs in common with LOAD than with EOAD patients. However, 5xFAD showed significantly higher DEGs intersection with EOAD than with hAβ-KI and APP/PS1. FEA of Gene Ontology biological processes revealed 92% overlap of terms enriched in hAβ-KI with LOAD, but only 32% with EOAD. Unexpectedly, both 5xFAD and APP/PS1 presented more than 75% of their enriched terms in common with LOAD patients compared to ∼45% in EOAD. New animal models are being introduced constantly to better simulate AD pathology. Hence, the evaluation of their overlap with human pathology is paramount. We found that the new hAβ-KI shows a good fit towards LOAD, as was its original purpose. APP/PS1 and 5xFAD on the other hand, were less optimal than expected to model EOAD.