Poster M5 - APOE4 Mouse to Human Alzheimer’s Disease Translational Modeling with Orexin Target Screening

Introduction: Alzheimer’s Disease (AD) is a brain disease that impedes cognition and behavior. Coupled with tau tangles and amyloid plaque, increased synapse and neuron death are prominent in AD patients. Age, genetics, including apolipoprotein E e4 (APOE4), and certain lifestyle factors, like sleep deprivation, increase susceptibility to AD development. In the brain, lipids are carried by APOE. There is a heightened risk of AD by expressing the APOE e4 allele due to increased levels of amyloid beta and tau aggregates compared to APOE e3 carriers and dysfunctional lipid metabolism. Also, AD hallmarks in APOE4 AD mice models are aggravated by extended periods of sleep disruption. Sleep is regulated by the orexin network, with its upregulation inducing wakefulness. Elevated orexin levels have been marked with higher amyloid beta than diminished orexin expression in mice. Orexin inhibitors could be a potential treatment to ameliorate AD burden. However, the interacting mechanisms with orexin and APOE4 exacerbating AD have not been fully understood. In addition, the insufficient effective human therapies from mice studies and drug screenings are a challenge due to models not fully encompassing AD pathology.

Materials and

Methods: We used GSE141509 from the Gene Expression Omnibus database to access bulk-RNA sequenced brain hemisphere samples of APOE4 KI (n=32) and control C57BL/6J (n=35) mice. We filtered for six and twelve-month-olds; we had (n=11) female APOE4 KI, (n=9) male APOE4 KI, (n=11) female control, and (n=12) male control samples. GSE44772 with bulk-RNA sequenced postmortem prefrontal cortex samples of AD and control subjects were retrieved for our human data. We selected (n=128, mean age = 80, female = 67, male = 61) AD and (n=59, mean age = 69, female = 14, male = 45) healthy subjects. Differentially expressed genes (DEGs) of Suvorexant, an orexin inhibitor, were accessed from the Library of Integrated Network-Based Cellular Signatures (LINCS). We chose mice genes orthologous to human genes and overlapped them with genes available from LINCS. Translatable Components Regression (TransComp-R) was conducted through the mouse data undergoing principal component analysis (PCA) and overlaid with human subject gene expression on the first mouse PCs (80% variance). Then, we used Boruta feature selection and Random Forest with human demographic factors to discover mouse components that predict human AD state. Gene set enrichment analysis (GSEA) was run for hallmark biological pathways of the genes from the selected mouse components. To determine potential therapeutic AD targets by Suvorexant, we did a z-test and a Benjamini-Hochberg correction to obtain significant DEGs (q-value ≤ 0.25). We performed a Spearman correlation analysis to discover an association between Suvorexant DEGs and mouse component loadings (Figure 1).

Results, Conclusions, and Discussions: We obtained 20 principal components from the mice data that explained ~80% variance through PCA. We input the integrated human gene expression data with mouse PCs to our Boruta model (500 runs), which selected 26 variables with sex interaction terms to be important in predicting human AD state (Figure 2). Then, we fit our Random Forest model (500 trees) with Boruta-selected features, resulting in an out-of-bag error of 12.37%. PC3 had the highest variance explained by mouse to human (69.22%), and the score plot demonstrated a clear separation between (N) control (negative) and (A) AD (positive). In addition, PC6 had a 2.83% variance explained from mice to humans and separated between control (positive) and AD (negative) (Figure 3A). GSEA of PC3 mouse genes demonstrated hallmark biological pathways on inflammation, including “complement,” “interferon-gamma response,” and “TNF-A signaling via NF-KB,” steroid hormone responses, including “estrogen response early” and “androgen response,” and metabolism, such as “fatty acid metabolism” and “xenobiotic metabolism” to be upregulated in AD subjects (Figure 3B). PC6 had enriched genes in AD subjects involved in intracellular signaling, such as “PI3K AKT mTOR,” and metabolism, such as “heme metabolism,” “fatty acid metabolism,” and inflammation (Figure 3C). After running the Spearman correlation, we found a significant positive association (R= 0.62, p-value = 0.035) between PC6 mouse loadings of the 12 genes also present in the significant Suvorexant DEGs (Figure 4). The genes upregulated in control (positive) in PC6 were upregulated by Suvorexant, while genes upregulated in AD (negative) were downregulated by Suvorexant. Some genes upregulated by Suvorexant were associated with synapse functions downregulated in AD. TYROBP, involved in immune response, was downregulated by the drug, serving as a potential target for delaying AD progression. Our model pipeline was able to predict between human AD and control groups from mouse PCs and inform for significant biological pathways involved, such as inflammatory and dysregulated metabolism with possible sex interactions with APOE4. Future work will further explore the therapeutic gene targets of Suvorexant to ameliorate AD symptoms.

Acknowledgements (Optional): Supported by Good Ventures Foundation and Open Philanthropy.