Single-cell RNA sequencing reveals disease associated changes in brain endothelial cells in the 5XFAD mouse

TL;DR

This study uses single-cell RNA sequencing to uncover changes in brain endothelial cells in a mouse model of Alzheimer's disease.

Contribution

The study provides a detailed molecular profile of endothelial cell changes in the 5XFAD mouse model using scRNAseq.

Findings

Differentially expressed genes in 5XFAD mice are linked to DNA damage, immune reactivity, and inflammation.

Transcriptomic changes are zonated along the arteriovenous axis and involve AD GWAS risk genes.

The study offers a resource to better understand vascular dysfunction in Alzheimer's disease.

Abstract

Vascular dysfunction is a key contributor to Alzheimer’s disease (AD) pathology, where changes to the endothelium and its crucial role in maintaining blood-brain barrier (BBB) integrity have been of particular emphasis. The transgenic 5XFAD (5X Familial Alzheimer’s Disease) mouse model, which exhibits AD-related amyloidosis through FAD associated mutations in amyloid precursor protein (APP) and presenilin-1 (PS1), has become a widely adopted preclinical model in AD-related research studies. The need for cross-study standardization, accessibility, and data reproducibility has led to the widespread implementation of the C57BL/6J genetic background for maintaining this model. However, its reliability for studying vascular dysfunction and BBB alterations has been questioned due to conflicting reports in the literature. This variation is often attributed to the previously documented protective nature of the C57BL/6J background and loss of genetic background diversity. Since prior studies have mostly relied on imaging or functional assays, we herein utilized single-cell RNA sequencing (scRNAseq) to investigate AD-related molecular changes to endothelial cell populations in the 5XFAD mouse model. To initially build this resource, we focused on 12-month-old male mice, which revealed differentially expressed genes between 5XFAD and wildtype animals that mapped to signaling pathways involved in DNA damage, immune reactivity, and inflammation, among others. Many of these transcriptomic changes were zonated along the arteriovenous axis and occurred in AD genome-wide association study (GWAS) risk-associated genes. Overall, we anticipate this resource will help clarify the use of the 5XFAD model for studying AD-associated vascular changes and provide the foundation for expanded molecular profiling of brain endothelial cells under AD-associated conditions.

The online version contains supplementary material available at 10.1186/s12987-026-00766-w.

Figures

Differentially expressed genes in endothelial cell clusters of arteriovenous zonation. Heatmap shows genes identified after negative binomial mixed modeling (NEBULA) of 5XFAD vs. WT. Upregulated genes (red) and downregulated genes (blue) were determined to be statistically significant with adjusted *p-*value < 0.05 and log_2_FC > 0.25. Cluster identity: aBECs (arterial brain endothelial cells), aCap BECs (arterial capillary brain endothelial cells), iBECs (immune reactive endothelial cells, mBECs (metabolic activated brain endothelial cells), vCap BECs (venous capillary brain endothelial cells

Brain endothelial differential gene expression of AD GWAS genes (Grubman et al., 2019), mapped to mouse orthologs. Upregulated genes (red) and downregulated genes (blue) were determined to be statistically significant with adjusted *p-*value < 0.05 and log_2_FC > 0.25 Cluster identity: aBECs (arterial brain endothelial cells), aCap BECs (arterial capillary brain endothelial cells), iBECs (immune reactive endothelial cells, mBECs (metabolic activated brain endothelial cells), vCap BECs (venous capillary brain endothelial cells), vBECs (venous brain endothelial cells). (*) = adjusted *p*-value <

Pathway analysis using KEGG on significantly differentially upregulated and downregulated genes in the 5XFAD condition

Full text

Introduction

The neurovascular unit (NVU) is a highly complex, multicellular physiological system that regulates brain homeostasis by controlling the exchange of ions, molecules, and cells between the blood and the parenchyma of the central nervous system (CNS) [1]. Specialized tight junctions between endothelial cells, in combination with cytoplasmic adapter proteins, form a high resistance electrical barrier defining of the blood-brain barrier (BBB) [1–5]. Coordinated signaling between specialized endothelial cells, astrocytes, pericytes, microglia/macrophages, and neurons regulate bidirectional exchange of material through active receptor mediated transport pathways, efflux transport, immune cell adhesion molecule expression, and pinocytosis in both health and disease [1, 6]. Vascular changes relating to BBB dysregulation are believed to occur relatively early in Alzheimer’s disease (AD), before the onset of dementia, rendering such changes as potential targets for therapeutic intervention [7–10]; as such, a key consideration for understanding these changes is their proper representation in experimental model systems. Although engineered neurovascular in vitro models have significantly progressed in recent years [11], these systems cannot fully recapitulate the complexity of the NVU, so animal models continue to be the gold standard for dynamic NVU research focused on understanding BBB changes in disease states.

The 5XFAD (5X Familial Alzheimer’s Disease) is a mouse model of AD-related amyloidosis. Developed in 2006 as a model of accelerated plaque development to contrast other transgenic AD models [12], the 5XFAD model was developed under the precept that deposited β-amyloid (Aβ) plaques contribute to AD etiology, and that murine replication of the human missense mutations in amyloid precursor protein (APP) and presenilin-1 and 2 (PS1/PS2) would trigger neurotoxic Aβ42 plaque formation based on altered post-translational proteolytic processing of APP by β- and γ-secretases. APP (APP695) is a transmembrane protein cleaved and processed by ɑ-β- and γ- secretases. In non-amyloidogenic conditions, ɑ-secretase cleaves APP at the ɑ-site to generate both a large soluble ectodomain (sAPPɑ) outside of the cell and a C83/a-CTF (C terminal fragment), which is the substrate for γ-secretase [13–15] (Fig. 1A). In contrast, the amyloidogenic pathway for APP processing involves β-secretase, which cleaves APP at the Met-Asp [16–18]. This cleavage then produces a large soluble ectoderm fragment (sAPPβ), and a C99/β-CTF fragment that becomes the substrate for γ-secretase (Fig. 1B) [13]. Processing of APP in this way leads to a higher level of Aβ42 peptide variants, which are more prone to aggregation and related neurotoxicity [12, 14]. This aggregation culminates in the plaque formation that is a hallmarking pathological marker linked to both familial Alzheimer’s Disease (FAD) and sporadic/late onset Alzheimer’s Disease. The 5XFAD model was motivated by mutations in APP that are associated with the predisposition of Aβ plaque formation in FAD [12]. Swedish mutations K670N/M671L at codon 670/671 increase C99 fragment presentation by enhancing β-secretase cleavage [19]. Florida I716V and London V717I mutations at codons 716 and 717, respectively, shift γ-secretase cleavage towards Aβ42 [20, 21] (Fig. 1C-D). Meanwhile, presenilin-1 (PS1), which is a subunit of the γ-secretase enzyme and responsible for cleavage of the C99/β-CTF, can also carry mutations that serve as one of the strongest risk factors for FAD [22–25]. PS1 mutations M146L (codon 146) and L286V (codon 286) both increase cleavage at the neurotoxic Aβ site, further exacerbating plaque accumulation [12, 26] (Fig. 1C). Combination of these five FAD related mutations downstream of the Thy1 promoter in the mouse leads to a high level of neurotoxic Aβ42 plaque formation in the brain [12].

Fig. 1**(A)** Non-amyloidogenic condition where α-secretase cleaves APP (amyloid precursor protein) at the α-site which results in a soluble ectodomain (sAPPα) and a C38/a-CTF (C-terminal fragment) that becomes the substrate for γ-secretase. (B) Amyloidogenic condition where β-secretase cleaves APP which results in a large soluble ectoderm fragment sAPPβ and a C99/b-CTF fragment that becomes the substrate for γ-secretase. The result is a higher level of Aβ42 variant which is more prone to aggregation. (C) 5XFAD mutations shift γ-secretase cleavage towards Aβ42. PS1 (presenilin-1) mutations M146L and L286V also increase cleavage at the neurotoxic Aβ42 site. (D) Swedish mutations K670N/M671L at codon 670/671, Florida I716V at codon 716, and London V717I mutation at 717 locations with respect to the Aβ APP domain and cleavage sites. Figure is adapted from Thinakaren and Koo 2008 [13], Kulikova, et al., 2014 [87], and O’Brien and Wong, 2011 [15], and is accessible on BioRender (https://biorender.com/8kccu6a)

The critical influence of genetic background on the resulting phenotype of transgenic models is a well-documented phenomenon and known confounder in transgenic animal modeling [27–29]. Although the 5XFAD model was originally developed on a mixed, C57BL/6J x SJL/J F1 hybrid background, there has been a preferential shift in the literature towards a pure C57BL/6J background [30]. This shift reflects a move towards greater standardization, which supports more meaningful cross-strain comparisons and strengthens experimental reproducibility [31]. The detailed and extensive genotypic characterization of the C57BL/6J strain makes it a preferred model, not only for comparative studies, but also for more controlled biological normalization when cross breeding with other strains. The 5XFAD on the C57BL/6J background now makes up approximately 50% of all 5XFAD related research [30]. A major drawback to the strain is its resistance to transgenic phenotypic traits due to the protective nature of the genetic background [32], which can delay comparable outcomes relative to other backgrounds or dampen disease characteristics making them difficult to measure [32, 33]. Reports of disease related neurovascular changes in the 5XFAD on the C57BL/6J background from the Jackson Laboratory are mixed in the field, which has led to varying opinions on its continued use as a model to study AD related vascular dysfunction. Some reports indicate that the BBB is largely preserved specifically in female 5XFAD mice [34], and there is an absence in cerebral perfusion defects, which might suggest a potential dissimilarity between the mouse model and the pathological perfusion changes observed in human AD [35]. Additionally, it has been reported that although spatial memory continues to decline with age in the animal, there are no obvious changes in barrier integrity as assessed by real-time two-photon imaging of tracer extravasation across the vasculature [36]. In contrast, multiple studies have documented a compromised BBB in 5XFAD mice, for example downregulation of tight junction proteins such as ZO-1 [37] as measured by immunoblotting and increased extravasation of tracer dyes as measured by quantification of dye signal in the brain at a fixed time point [38–40]. Given conflicting data, we sought to assess molecular aspects of BBB dysfunction through transcriptomic profiling of endothelial cells using single-cell RNA sequencing. Using this approach, we detail disease related transcriptomic profiles in endothelial cell populations from male, 12-month-old 5XFAD mice on the C57BL/6J background as an initial resource and starting point for future in-depth transcriptional profiling across multiple time points and both sexes.

Methods

Animal husbandry

Male hemizygous 5XFAD mice were obtained from The Jackson Laboratory (RRID: MMRRC_034848-JAX), and harem breeding pairs were established using wildtype C57BL/6J (RRID: IMSR_JAX:000664) female littermates to establish paternal inheritance of the transgene. To ensure genetic consistency across the cohort, all mice originated from the same paternal harem breeding lineage. All animals were maintained in the institutional vivarium under standard conditions on a 12-hour light/dark cycle under temperature control. Animals were provided food and water ad libitum in accordance with standard protocols. Progeny underwent tail biopsy and were genotyped using probes designed by Transnetyx (Cordova, TN) to determine presence of the transgene (huPSEN, APPsw, Chr3-6_WT).

Brain dissociation and endothelial enrichment

Male hemizygous 5XFAD mice aged 12 months (n = 3), and male wildtype controls aged 12 months (n = 3), were euthanized using isoflurane and adjunctive cervical dislocation. The cohort was constructed such that five of the six total mice were direct littermates (two WT controls paired with the three 5XFAD mice). The third WT control of the same age was supplemented from the same paternal harem breeding lineage (as described above). Brains were harvested and immediately dissociated into a single-cell suspension using a combined gentle mechanical and enzymatic dissociation approach adapted from the Miltenyi Adult Brain Dissociation Kit, Mouse and Rat (Miltenyi Biotec 130-107-677). After tissue homogenization, dissociation, myelin/debris removal, and red blood cell lysis, cells were incubated with fluorescently conjugated antibodies against CD31 (0.1 µg/mL, Invitrogen 170311-82) and CD45 (0.1 µg/mL, BD Pharmingen 553081) diluted in 0.5% BSA (w/v) (Research Products International A30075-100.0) in Dulbecco’s Phosphate-Buffered Saline (DPBS, Gibco 141190-144) for 30 min at 4 °C. After immunolabeling, cells were rinsed with DPBS containing 0.5% BSA and resuspended in DPBS containing 0.5% BSA with 4’6-Diamidino-2-phenylindole (DAPI, 0.2 µg/mL, Sigma Aldrich D9542) or without DAPI for single stained compensation controls and isotype controls (0.1 µg/mL, Miltenyi Biotec 130-123-273; 0.1 µg/mL, Invitrogen 17-4321-81). CD31+/CD45- cells were isolated using fluorescence activated cell sorting (FACS) using a 5-laser BD FACSAria III. Nonviable cells were distinguished from live cells using DAPI and removed from the cell population. Desired fractions were collected into DPBS containing 0.5% BSA and kept on ice until prepared for sequencing.

Single-Cell RNA sequencing

After dissociation and enrichment through FACS for live CD31+/CD45- fractions, cells were encapsulated and barcoded using the 10X Genomics Chromium platform to generate barcoded cDNA libraries. Libraries were then subjected to next-generation sequencing using a GEM-X Single Cell 5’ Gel Bead kit (v3) and a GEM-X 5’ Chip. Approximately 10,000 cells per sample were processed for 5’ single-cell RNA sequencing (Table 1). Libraries were sequenced on an Illumina NovaSeq XP, targeting ~ 50,000 reads per cell. Raw sequencing data were demultiplexed and processed with the Cell Ranger software package using mouse reference genome (GRCm39) − 2024-A from 10X Genomics. Downstream analysis was carried out using the Seurat package [41] (v.5.2.1) in R (v.4.4.2) (R Foundation for Statistical Computing, Vienna Austria). Normalization was performed using SCTransform [42], and Harmony was used for batch correction [43]. Data manipulation and visualization was conducted using dplyr [44], ggplot2 [45], SingleCellExperiment [46] glmGamPoi [47], Matrix [48], and patchwork [49]. Doublets were removed using scDblFinder [50]. Differentially expressed genes DEGs were calculated using DESeq2 [51], lme4 [52], and NEBULA [53]. Pathway analysis was performed using clusterProfiler [54], org.Mm.eg.db [55], and stringr [56]. GWAS data comparison was processed and visualized using readxl [57], ComplexHeatmap [58], tibble [59], biomaRt [60], and Matrix [48]. The complete source code used for analysis presented in this study is publicly available through GitHub (see Data and Code Availability).

Table 1. Summary of cell, gene, and unique molecular identifiers (UMI) metrics per sampleSampleCells CapturedMedian Genes/CellMedian UMI Counts/CellTotal Genes DetectedWT 17,7652,3614,75122,016WT 27,6562,4104,89221,753WT 39,0482,4985,20022,2765XFAD 17,2192,6095,47121,7125XFAD 26,9302,3634,71221,3315XFAD 36,1092,2114,36420,804

Results

Endothelial arteriovenous gradient zonal signatures in single-cell RNA sequencing

A major technical challenge in preparing brain endothelial cells for single-cell analysis lies in balancing a dissociation method that must be stringent enough to dislodge the cells from the basement membrane, while simultaneously gentle enough to preserve cell viability and transcriptional integrity. To circumvent this challenge, many labs turn to single-nucleus sequencing when profiling endothelial cells at the transcriptomic level, sacrificing overall assay sensitivity [61]. We demonstrate that we can obtain a high number of intact, viable brain endothelial cells from the mouse by compounding a gentle mechanical dissociation technique with optimizations to the enzymatic digestion method detailed in the protocol from the Miltenyi adult brain dissociation kit for rat and mouse. The resulting dissociated single cells were enriched for our desired population through fluorescence activated cell sorting (FACS) by gating for CD31+/CD45- cell populations. Single cells were sequenced as described in our methods, which resulted in 44,727 total cells sequenced after capture, with samples ranging from 6,109 to 9,048 total cells. Median genes per cell ranged from 2,211 to 2,609, and median unique molecular identifier (UMI) counts per cell ranged from 4,364 to 5,471. Quality control was conducted to remove low quality reads and potential doublets (Fig. S1). After quality control, the retained cells from our population were used for downstream analysis (Table 1). Data was processed using the Seurat package (v.5.2.1) [41, 62, 63] in R (v.4.4.2). Following quality control, each dataset was normalized using SCTransform [42] with the top 800 variable features, which was previously determined to be sufficient using the variance stabilization transformation method in Seurat in our initial preprocessing workflow. Datasets were then merged (n = 6) (Fig. S2) and integrated using Harmony to correct for batch effects [43]. After quality control and batch correction, we performed dimensionality reduction using principal component analysis, of which the top 13 principal components were retained (Fig. S2). To identify distinct cell populations, we performed graph based clustering [41], where a k-nearest neighbor graph was constructed using the first 13 principal components, and clusters were determined by applying the Louvain algorithm [64], with resolution set to 0.25, which yielded a total of 9 distinct cell clusters that were subsequently visualized using Uniform Manifold Approximation and Projection (UMAP) [65] plots (Fig. 2B). Differential gene expression between clusters was performed with a Wilcoxon Rank Sum test using the Seurat FindAllMarkers [41] function, where genes were considered cluster markers if they had an adjusted p-value < 0.05 and Log_2_-FC > 0.25. The ranked top 10 differentially expressed genes per cluster were used to assign subcluster identity (Fig. 2A). We characterized the endothelial subclusters and assigned identity using known markers of endothelial zonation [66–71] (Table 2). Specifically, known genes related to zonal identity (Fig. 2C) were mapped onto the UMAP to generate feature maps of zonal gene expression (Fig. 2D). We were able to capture robust populations of cells displaying a comprehensive set of zonal endothelial subtypes along the arteriovenous axis. Arterial brain endothelial cells (aBECs), arterial capillary brain endothelial cells (aCap BECs), venous brain endothelial cells (vBECs), venous capillary brain endothelial cells (vCap BECs), immune-reactive brain endothelial cells (iBECs), and metabolic activated brain endothelial cells (mBECs) were all identified in the endothelial supercluster. Immune-reactive brain endothelial cells were definitively named due to the enriched expression of Irf7,* Isg15*,* Ifit3*, and Usp18, which are effectors in the canonical Type I Interferon pathway [72–74]. This is distinct from the metabolic activated brain endothelial cells where enrichment in markers such as Lars2 and Camk1 indicate activated cells undergoing metabolic signaling [75, 76].

Fig. 2A) Differential gene expression between clusters using the Wilcoxon Rank Sum test plotted on a heatmap where high (yellow) - low (blue) expression is ramped by color. Significance was thresholded to adjusted p-value < 0.05 and Log_2_-FC > 0.25, with the top 10 genes ranked per Seurat cluster labeled on the y-axis. (B) Distinct cell clusters visualized using Uniform Manifold Approximation and Projection (UMAP) plots. (C) Selected markers of zonation for feature maps in panel D. (D) Genes related to zonal identity plotted for expression by feature onto UMAP projections

Table 2. Summary table of literature review highlighting key endothelial markers of arteriovenous zonationGeneZonal RegionReferenceGeneZonal RegionReference Vwf Arterial, Venous, Non-CapillaryVanlandewijck et al., 2018 [66] Hey1/2 ArterialFischer et al., 2004 [69] Vcam1 Arterial, Venous, Non-CapillaryVanlandewijck et al., 2018 [66] Mfsd2a CapillaryVanlandewijck et al., 2018 [66] Bmx ArterialVanlandewijck et al., 2018 [66] Car4 CapillaryGhandour et al.,1992 [70] Efnb2 ArterialVanlandewijck et al., 2018 [66] Tfrc Capillary, VenousVanlandewijck et al., 2018 [66] Vegfc ArterialVanlandewijck et al., 2018 [66] Slc16a1 Capillary, VenousVanlandewijck et al., 2018 [66] Sema3g ArterialVanlandewijck et al., 2018 [66] Slc38a5 VenousVanlandewijck et al., 2018 [66] Gkn3 ArterialVanlandewijck et al., 2018 [66] Nr2f2 VenousChen et al., 2020 [67] Mgp ArterialChen et al., 2020 [67] Il1r1 VenousChen et al., 2020 Clu ArterialChen et al., 2020 [67] Cfh VenousChen et al., 2020  [67] Stmn2 ArterialChen et al., 2020 [67] Ctsc VenousChen et al., 2020  [67] Cdh13 ArterialChen et al., 2020 [67] Tmsb10 VenousChen et al., 2020  [67] Efnb2 ArterialAdams et al., 1999 [68] Gm5127 VenousLee et al., 2021 [71]

Disease-associated alterations transcriptomic profiles in brain endothelial cells from 5XFAD mice

To assess transcriptomic differences in the endothelial clusters between the 5XFAD and wildtype (WT) control group, we performed pseudobulk differential gene expression analyses on the endothelial supercluster using DESeq2 [51]. Between groups, differentially expressed genes were identified as adjusted *p-*value < 0.05 and Log_2_FC > 0.25 (Table S1). Differentially expressed genes were visualized using ComplexHeatmap, where genes upregulated in the 5XFAD condition are highlighted in red, and genes downregulated in the 5XFAD condition are highlighted in blue (Fig. 3A). Functional enrichment analysis of genes upregulated and downregulated in the endothelial supercluster were used for pathway analysis using Kyoto Encyclopedia of Genes and Genomes (KEGG) [77, 78] (Fig. 6) and Reactome [79, 80] (Fig. 3C), and gene function was characterized using Gene Ontology (GO) [81, 82] (Fig. 3B). Pathways were considered statistically significant if adjusted p-value was ≤ 0.05, where p-values were adjusted using the Benjamini-Hochberg [83] method to correct for multiple testing. Pathway analysis across the endothelial supercluster revealed disease relevant transcriptomic alterations in the 5XFAD model relative to WT controls. Notably, pathways associated with DNA damage, immune reactivity, MAPK signaling, mTOR signaling, cell cycle regulation, endocytosis, and endosomal transport are dysregulated in the 5XFAD condition compared to the WT condition.

Fig. 3A) Significantly differentially expressed genes in endothelial cell supercluster as assessed by pseudobulk analysis. Upregulated genes (red) and downregulated genes (blue) were determined to be statistically significant with adjusted p-value < 0.05 and log_2_FC > 0.25. () = adjusted p-value < 0.05, () = adjusted p -value < 0.01, () = adjusted p-value < 0.001. B) Pathway analysis using GO on significantly differentially upregulated and downregulated genes in the 5XFAD condition. C) Reactome analysis on differentially upregulated and downregulated genes in the 5XFAD condition

We then analyzed individual endothelial clusters relating to zonal identity for DEG expression using negative binomial mixed modeling through NEBULA [53]. These analyses revealed 56 significant DEGs in the vCap BEC cluster, 19 significant DEGs in the aCap BEC cluster, 14 significant DEGs in the vBEC cluster, 5 significant DEGs in the mBEC cluster, 6 significant DEGs in the iBEC cluster, and 11 significant DEGs in the aBEC cluster. A DEG was determined to be statistically significant if a threshold of adjusted *p-*value < 0.05 and log_2_FC > 0.25 were reached (Fig. 4 and Table S2).

Fig. 4. Differentially expressed genes in endothelial cell clusters of arteriovenous zonation. Heatmap shows genes identified after negative binomial mixed modeling (NEBULA) of 5XFAD vs. WT. Upregulated genes (red) and downregulated genes (blue) were determined to be statistically significant with adjusted p-value < 0.05 and log_2_FC > 0.25. Cluster identity: aBECs (arterial brain endothelial cells), aCap BECs (arterial capillary brain endothelial cells), iBECs (immune reactive endothelial cells, mBECs (metabolic activated brain endothelial cells), vCap BECs (venous capillary brain endothelial cells), vBECs (venous brain endothelial cells). () = adjusted p-value < 0.05, () = adjusted p -value < 0.01, () = adjusted p-value < 0.001. Gray boxes indicate gene expression in a specific cluster could not be reliably measured

To evaluate whether human AD genetic risk factors correlate with differential gene expression in 5XFAD endothelial cells, we cross referenced our dataset with AD-associated genes identified via Genome Wide Association Study (GWAS). We utilized a prioritized list originally derived from Grubman, et al., 2019 [84], which was further refined down to 651 genes by Yang, et al., 2022 [85]. Human Ensembl IDs [86] were converted to mouse orthologs using biomaRT [60] in R. Corresponding GWAS Mouse ortholog IDs were then compared with our DEG dataset calculated using NEBULA by cluster and condition for GWAS associated genes. This analysis identified 11 GWAS-associated orthologs with significant expression changes in 5XFAD endothelial populations (Fig. 5). Notably, these AD-associated genes exhibited distinct zonal expression patterns across the arteriovenous axis. For example, Golim4 was differentially regulated across all endothelial populations. Overall, vCAP BECs exhibited the highest number of DEGs mapping to human AD GWAS risk loci.

Fig. 5. Brain endothelial differential gene expression of AD GWAS genes (Grubman et al., 2019), mapped to mouse orthologs. Upregulated genes (red) and downregulated genes (blue) were determined to be statistically significant with adjusted p-value < 0.05 and log_2_FC > 0.25 Cluster identity: aBECs (arterial brain endothelial cells), aCap BECs (arterial capillary brain endothelial cells), iBECs (immune reactive endothelial cells, mBECs (metabolic activated brain endothelial cells), vCap BECs (venous capillary brain endothelial cells), vBECs (venous brain endothelial cells). () = adjusted p-value < 0.05, () = adjusted p -value < 0.01, () = adjusted p-value < 0.001

Fig. 6. Pathway analysis using KEGG on significantly differentially upregulated and downregulated genes in the 5XFAD condition

To further assess the relevance of our findings on the context of human AD, we performed a cross-species transcriptomic comparison analysis between the 5XFAD endothelial DEGs and a human AD vascular single-nucleus RNA sequencing (snRNAseq) DEG dataset obtained from Yang et al., 2022 [85]. Because of the differences in zonal clustering resolution, we combined all endothelial DEGs from this dataset (arterial, venous, and capillary) into one list to compare to our 5XFAD pseudobulk DEGs dataset (Table S1). Human gene names were converted to mouse orthologs as described above, and DEGs were assessed for commonly upregulated and downregulated genes between both datasets. Assessment of directional regulation of DEGs between datasets revealed 21 orthologs with concordant upregulation, and 62 orthologs with concordant downregulation, indicating a shared transcriptomic trend between species (Table S3). Only Akap12 (human ortholog AKAP12), which was downregulated, met the criteria for significant concordant regulation across both datasets where the adjusted *p-*value < 0.05 and Log_2_FC > 0.25. Because of the technical variation in how the datasets between human and mouse were acquired, there is an obvious limitation to this cross-species analysis. Data collection on the 5XFAD mice utilized scRNAseq, whereas the human dataset utilized snRNAseq, limiting the ability to control technical differences in read depth and assay sensitivity in comparative analyses. Furthermore, differences in sample size, sequencing metrics, normalization, and bioinformatic pipelines between the two studies may constrain meaningful statistical overlap despite biological trends. Nonetheless, these comparisons provide a starting point for further analyses upon additional data collection and harmonization.

Discussion

Our study directly addresses the critical need for a high-resolution single-cell RNA sequencing dataset characterizing the endothelial cell transcriptome of the 5XFAD mouse model on the widely used C57BL/6J background. This resource is intended to accompany and enhance studies utilizing mouse models of AD on the C57BL/6J background to further elucidate AD related endothelial cell, BBB, and neurovascular dysfunction. We successfully isolated a high yield of viable endothelial cells using our optimized dissociation method, capturing key distinct endothelial subpopulations along the arteriovenous axis. Our initial analysis of the dataset demonstrates upregulation and downregulation of AD-associated genes and pathways, with findings correlating to human genetic risk factors identified in GWAS studies, thereby validating the model’s use to study markers associated with human disease at the transcriptomic level. We also observed zonal heterogeneity in AD-associated transcriptomic changes specific to zones on the arteriovenous axis. Our pathway analyses strongly implicate increases in chronic inflammatory signaling, immune reactivity, dysregulated endocytosis and endosomal transport, altered MAPK and mTOR signaling, and DNA damage as key features of AD related pathology represented in the endothelium in this model; which is also demonstrated in our analysis using KEGG [77, 78] (Fig. 6).

The scope of this initial analysis is confined to the use of 12-month-old male 5XFAD mice with paternal transgenic inheritance compared with wildtype controls. Future studies are warranted to fully delineate the endothelial transcriptomic landscape across key variables that are critical factors in the pathogenesis of AD murine models such as age, parental inheritance, and sex differences, which were not included in this study but should be recognized when comparing these findings to other studies. While acknowledging the known limitations of the 5XFAD model, its accessibility and standardization on the C57BL/6J background make it an invaluable tool for murine research related to amyloidosis and AD pathology. Despite its prominence, its utility for studying complex AD related vascular changes has been historically inconclusive in literature and a point of contention within the field of vascular biology. We directly address this gap by providing a robust, high-quality dataset for the field, which demonstrates evidence for significant AD-related transcriptomic changes in endothelial cell populations. Although we observe transcriptomic correlation with AD vascular pathology, future studies are needed to investigate the functional consequences of these transcriptomics changes in the model.

Conclusions

Ultimately, this work provides an initial molecular framework for researchers to investigate AD-related vascular pathology in the 5XFAD model. The observation of gene and pathway changes related to AD suggests the 5XFAD model on the C57BL/6J background, at the molecular level, is a useful tool for preclinical research on endothelial dysfunction in AD.

Supplementary Information

Below is the link to the electronic supplementary material.

Supplementary Material 1

Supplementary Material 2

Supplementary Material 3

Supplementary Material 4