Integrated imaging and molecular profiling reveals APOE4-associated neurovascular and glial disruptions in young adult mice

Abstract

Background

The Apolipoprotein-E ε4 (APOE4) allele is the strongest genetic risk factor for late-onset Alzheimer’s disease (LOAD) and may contribute to neurodegeneration through a multi-hit hypothesis, in which vascular dysfunction, glial activation, and impaired lipid metabolism play central roles. Alterations in neurovascular unit (NVU) have emerged as an early APOE4-related phenotype, independent of amyloid and tau pathology. Astrocytes, as the primary source of APOE in the brain and key regulators of NVU homeostasis, may play a central role in these processes. This study investigates APOE4-associated NVU water exchange dynamics and astrocyte-vascular interactions using integrated in vivo MRI, ex vivo histology, and transcriptomic profiling.

Methods

Non-contrast multimodal MRI, including multi-echo time arterial spin labeling (multi-TE ASL), T1-weighted imaging, and diffusion-weighted MRI, were applied in 6-9-month-old APOE3-KI and APOE4-KI mice. Multi-TE ASL was used to estimate regional NVU water exchange dynamics, while diffusion MRI assessed tissue microstructural alterations. Immunohistochemistry evaluated perivascular matrix metalloproteinase-9 (MMP9) activity, vascular-associated markers, astrocytic AQP4 expression, and glial reactivity. Single-nucleus RNA sequencing (snRNAseq) characterized cell-type-specific transcriptional profiles, and inferred cell-cell communication analysis between astrocytes, pericytes, and other NVU components. Integrated analyses compared MRI-derived measures with molecular and cellular findings.

Results

APOE4-KI mice showed regionally specific alterations in NVU water exchange dynamics, particularly in the hippocampus, accompanied by trends toward altered microstructural complexity. Immunohistochemistry demonstrated increased perivascular MMP9 expression and evidence of extracellular matrix remodeling without prominent structural disruption of blood-brain barrier (BBB) markers in APOE4 mice. Astrocytes showed increased AQP4 expression, heightened proinflammatory gene signatures, and morphological reactivity. Molecular findings aligned with MRI, supporting the sensitivity of non-contrast MRI to early NVU alterations. Exploratory snRNAseq suggested an APOE4-enriched astrocyte subpopulation associated with immune activation and matrix-related pathways and suggested potential glial-vascular interactions that require validation in larger samples.

Conclusions

This integrated imaging and molecular analysis suggests that non-contrast multimodal MRI detects early APOE4-related changes in NVU exchange dynamics and glial-vascular interactions. By providing converging multiscale neuroimaging and cellular observations, this work provides a foundation for developing non-invasive biomarkers to monitor neurovascular vulnerability and guide early intervention strategies in individuals at risk for LOAD.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12974-026-03888-y.

Background

Sporadic or late-onset Alzheimer’s disease (LOAD) is a neurodegenerative condition characterized by the progressive β-amyloid (Aβ) and tau accumulation, with cognitive symptoms typically emerging after the age of 65 [1]. Recent genome-wide association studies identified the apolipoprotein E (APOE) ε4 allele as the strongest genetic risk factor for LOAD, increasing the risk up to 4-fold in heterozygous carriers and 15-fold in homozygous carriers [2, 3]. The APOE gene encodes a 34-kDa glycoprotein primarily expressed by astrocytes in the brain, where it supports neuronal and glial maintenance through lipid redistribution and receptor-mediated signaling [2, 4]. Compared to APOE2 and 3, recent studies suggest APOE4 may increase the risk of LOAD through multiple pathways, including neurovascular dysfunction, glial activation, and lipid dysregulation that occur during the early stages of brain aging [2, 5, 6]. These observations highlight a central role of the neurovascular unit (NVU), where APOE4 may disrupt coordinated cellular interactions that regulate molecular and fluid exchange between blood and brain.

The blood-brain barrier (BBB), as a key interface within the NVU, regulates the exchange of water, solutes, and signaling molecules between the peripheral circulation and the central nervous system [7–9]. The NVU is comprised of multiple components, including neurons, astrocytes, microglia, endothelial cells (including tight junctions), mural cells (i.e., pericytes), and the basement membrane [10]. Recent in vivo neuroimaging studies showed that middle-aged and older people and transgenic mice carrying the APOE4 allele have higher hippocampal BBB permeability than non-carriers, independent of their Aβ and tau status [11, 12]. The hippocampal BBB is particularly vulnerable to age-related degradation. This breakdown may compromise other components of the NVU by allowing peripheral immune and metabolic factors to influence neural function, and may initiate a cascade of glial activation, oxidative stress, and demyelination [7]. At an older age, people who carry the APOE4 allele also showed significant hippocampal atrophy and loss of tissue microstructural integrity of WM tracts in the medial temporal lobe (MTL) compared to non-carriers [13–15]. Together, these findings support the concept that altered NVU exchange function represents an early and sensitive indicator of APOE4-associated brain vulnerability.

Mechanistically, APOE4-related alterations in NVU function have been linked to activation of the cyclophilin A (CypA)-matrix metalloproteinase 9 (MMP9) pathway, particularly in the hippocampus, a region highly susceptible to both vascular injury and age-related atrophy [12, 16]. Specifically, compared to other APOE2 and APOE3, APOE4 has a lower affinity for the low-density lipoprotein receptor-related protein 1 (LRP1) receptor on the pericytes [16]. This signaling pathway is involved in suppressing the activation of the CypA that can be released from vascular smooth muscle cells and pericytes in response to inflammatory stimuli [17]. As a consequence, CypA induces nuclear factor-κB (NF-κB) activity that upregulates the MMP9 expression in the pericytes, which in turn contributes to the degradation of BBB tight junctions and basement membrane proteins [16]. MMP9 is a zinc-dependent endopeptidase of the gelatinase subfamily, which degrades extracellular matrix and tight junction proteins [18]. Consistent with this hypothesis, a previous study in older adults (aged 66–85 years) showed an elevation of cerebrospinal fluid (CSF) MMP9 in APOE4 carriers [19]. Similarly, 18-24-month-old APOE4-KI mice showed increased MMP9 colocalization with pericytes in the hippocampus compared to APOE3-KI mice [12]. These findings suggest that early molecular remodeling within the NVU may precede prominent barrier breakdown and instead manifest as altered exchange dynamics.

Astrocytes are central regulators of NVU function and play a critical role in modulating perivascular water and solute exchange through endfeet-localized proteins such as aquaporin-4 (AQP4) [20]. Evidence suggests that astrocytes are not only reactive responders to injury but also early modulators of vascular dysfunction in individuals carrying the APOE4 allele. In APOE4 carriers, astrocytes exhibit altered lipid metabolism and immune signaling, leading to reactive phenotypes that may disrupt NVU homeostasis [12, 21–23]. While these pathways implicate vascular and astrocytic dysfunction as early drivers of APOE4-related brain vulnerability, the temporal onset, regional specificity, and transcriptomic underpinnings of early APOE4-mediated NVU and astrocyte alterations remain poorly defined. A critical unresolved question is how early NVU disruption and astrocyte signaling changes interact to initiate downstream neurodegenerative processes.

In vivo characterization of NVU exchange function is therefore critical for understanding early neurovascular alterations. Traditional methods such as dynamic contrast-enhanced MRI (DCE-MRI) and cerebrospinal fluid (CSF)/serum albumin quotient collection are limited by contrast burden or invasiveness [24, 25]. Moreover, both methods may have limited sensitivity in detecting subtle regional BBB damage at early timepoints. Other conventional methods used in animal models to measure BBB permeability also involve tracers, including Evans Blue and Fluorescein, which cannot be applied to study humans. As an alternative, the multiple echo time arterial spin labeling (multi-TE ASL) technique was developed to provide insights on quantifying the kinetics of labeled blood water as it transitions from the vascular to tissue compartments [26]. It revealed faster water exchange rate in older mice (27 months old) compared to their younger counterparts (7 months old), which correlated with the upregulation of AQP4 expression transcript level in the brain [26]. Their results suggest that multi-TE ASL may be sensitive to detect changes in the NVU properties such as blood-brain barrier (BBB) permeability. This technique has also been applied to healthy participants, where several studies have demonstrated age-related decline in water exchange times, suggesting higher BBB permeability [27–30]. This technique has not yet been applied to study the effect of APOE4 on the mouse brain. Compared to gadolinium-based DCE-MRI, multi-TE ASL enables repeated measurements in the same subject with no systemic contrast burden, allowing for the dynamic assessment of BBB changes during aging or treatment. This is particularly advantageous for APOE4 models, where longitudinal monitoring of subtle permeability differences is crucial for accurate diagnosis and treatment. Diffusion MRI (dMRI) provides sensitivity to microstructural changes such as white matter integrity, tissue complexity, and extracellular space expansion [31, 32]. These measures have revealed age- and disease-related disruptions in fiber tracts, often preceding significant neuronal loss and brain atrophy [33–35]. Previous studies reported changes in dMRI metrics in APOE4 carriers, suggesting potential white matter degeneration [36, 37]. Moreover, recent studies demonstrate that dMRI can capture distinct signatures of microglia and astrocyte activation in vivo by modeling glia-specific morphological features, and detected increased glial activation and disrupted microstructure in the hippocampus and cingulate cortex in a toxicant-exposed animal model [32, 38]. However, neuroimaging alone cannot resolve cell-type-specific mechanisms. Ex vivo approaches such as immunohistochemistry and single-nucleus RNA sequencing enable high-resolution characterization of interactions between different components of NVU [39, 40]. Integrating MRI with molecular and cellular assays provides a comprehensive framework for linking mesoscale neuroimaging phenotypes to microscale mechanisms.

The goal of this study is to investigate early APOE4-associated alterations in neurovascular unit (NVU) exchange function, astrocyte reactivity, and intercellular signaling using a multimodal approach combining in vivo non-contrast multimodal MRI (multi-TE ASL, T1-weighted imaging, diffusion MRI) with ex vivo immunohistochemistry and single-nucleus RNA sequencing. APOE3 or APOE4 knock-in (APOE-KI) mouse models were examined at 6–9 months old, translating to approximately 30–40 years of age in humans [41]. The main hypothesis is that APOE4 induces region-specific alterations in transvascular water exchange dynamics, reflecting early NVU functional dysregulation, alongside astrocyte-mediated signaling changes during early adulthood. By characterizing vascular, astrocytic, and transcriptional alterations and integrating them with neuroimaging biomarkers, this work aims to clarify early mechanisms through which APOE4 accelerates brain aging and AD risk, ultimately informing the development of non-invasive biomarkers and therapeutic targets for neurovascular vulnerability.

Materials and methods

Mice

Experiments were conducted in APOE knock-in mice in which the endogenous murine APOE gene is replaced with human APOE3 or APOE4 (C57BL/6NTac-Apoe<tm4206.1(APOE*C130,R176)Tac> and C57BL/6NTac-Apoe< tm4207.1(APOER130,*R176)Tac> ) [42]. This mouse model is homozygous for human APOE isoforms (ε3 or ε4) and was generated on a C57BL/6NTac background by the Cure Alzheimer’s Fund. These mice express human APOE isoforms without spontaneous amyloid or tau deposition and therefore model APOE-isoform–specific effects independent of classical AD pathology. Male and female mice aged 6–9 months were studied to capture early-to-midlife vascular and glial alterations relevant to preclinical stages of AD [43]. Animals were housed in an AAALAC-accredited facility under a 12-hour light–dark cycle with ad libitum access to food and water. All procedures were approved by the Boston University IACUC.

In Vivo MRI acquisition

General procedures

In vivo multimodal MRI was performed at ~ 6 months of age using a 9.4T Bruker scanner at the Boston University Medical Campus (BUMC). Mice were anesthetized with isoflurane (2–3% induction; 0.5–2% maintenance), positioned in a prone orientation, and scanned with a surface receive coil and 86-mm transmit volume coil. Respiration and body temperature were continuously monitored and maintained within physiological ranges. Total scan time was approximately two hours per animal (Fig. 1A). Experiments were conducted under the guidelines of the NIH Committee on Laboratory Animal Resources and with the approval of Boston University Medical Campus’s IACUC. Details of MRI sequences are listed in Table 1. Methodological details are described in Supplementary Methods. A list of all MRI outcome measures obtained after image processing are listed in Supplementary Table 1.

Fig. 1.

In vivo MRI detected neurovascular and tissue microstructural disruptions in APOE4 mice. A APOE-KI mice underwent MRI at 6–9 months of age. This figure was generated in part using BioRender. B Representative images from each image modality and ROI definition. C Regions of interest rendered on a representative T2-weighted MRI. D Comparisons of NVU water exchange time (T_ex) between groups across all mice. E Group comparison of DKI AK and RK in the hippocampus. F Pearson correlation between T_ex and DKI RK. Correlation coefficients (R) were presented for analyses across all animals (black), within E3 animals (blue), and within E4 animals (red). Asterisk indicated significant correlations. A total of 22 animals were included in these analysis (n = 8 E3, n = 14 E4). Group comparisons were performed using independent t-tests. *FDR-p < 0.05. Data presented as median, interquartile range and full range. E3 = APOE3-KI; E4 = APOE4-KI; T_ex = exchange time; CBF = cerebral blood flow; ATT = arterial transit time; DTI = diffusion tensor imaging; DKI = diffusion kurtosis imaging. Refer to Supplementary Table 1 for the full list of imaging measures included in the study

Table 1.

MRI scanner information and acquisition parameters

Image Sequence	Imaging Parameters	Duration
2D T1-weighted rapid imaging with refocused echoes (RARE)	TR=800ms, TE=6ms, excitation FA = 90°, refocusing FA = 180°, FOV = 20 × 20mm2, MX = 256 × 256, ST=1 mm, Number of slices = 16.	3 min
Multi-TE ASL	TR=5000ms, TE = 8,10,12,15,18,23,30,40,50,65ms, FA = 90°, FOV = 20 × 20mm2, MX = 64 × 64, ST=2 mm, TI=800ms and 1500ms, Number of slices = 1. A separate short-inflow-time ASL acquisition was performed to estimate arterial transit time (ATT) with TE=10ms and TI = 200,300,400,500ms.	50 min
2D T2-weighted rapid imaging with refocused echoes (RARE)	TR=3845ms, TE=11ms, excitation FA = 90°, refocusing FA = 180°, FOV = 20 × 20mm2, MX = 256 × 256, ST = 0.05 mm, Number of slices = 32.	5 min
Multi-B dMRI: DTI and DKI	TR=2000ms, TE=18ms, FA = 90°, FOV = 20 × 12mm2, MX = 128 × 80, b-values = 1000, 2000, Number of slices = 32.	40 min

TR Repetition time, TE Echo time, FA Flip angle, FOV Field of view, MX Matrix, ST Slice thickness, TI Inflow time, PLD Post-labeling delay

Multi-TE ASL processing (NVU Water exchange dynamics)

Multi-echo time arterial spin labeling (multi-TE ASL) is a non-contrast MRI technique that allows estimation of transvascular water exchange rate by capturing signal changes at different echo times (TEs) after magnetically labeling arterial blood water [44]. The ASL image acquisition was based on a flow-alternating inversion recovery (FAIR) sequence with single-shot spin-echo EPI readout. Acquisition parameters and detailed protocol were described in Table 1 and Supplementary Methods. Multi-TE ASL data were processed using a custom MATLAB pipeline based on Ohene et al. to extract outcome measures including water exchange time (T_ex), arterial transit time (ATT), cerebral blood flow (CBF) and intravascular fraction (IV) [45]. Region-of-interest (ROI) analyses were performed in five predefined regions (hippocampus, global cortex, retrosplenial cortex, thalamus, peripheral cortex) using a semi-automated template-guided pipeline (Fig. 1B&C). ROIs defined on the Mortimer Space Atlas and DSURQE templates were registered to each subject’s native T2 space using FSL FLIRT [46–48]. Quality control involved visual inspection of all ROI overlays.

Diffusion MRI processing (surrogate markers of tissue microstructure)

Diffusion-weighted MRI (dMRI) was performed using a single-shot spin-echo echo-planar imaging (EPI) sequence optimized for the mouse brain. Diffusion data were denoised, motion-corrected, and eddy-current–corrected using MATLAB and FSL [49]. The imaging protocol was designed to capture both conventional diffusion tensor imaging (DTI) and diffusion kurtosis imaging (DKI) metrics. DTI metrics, including Fractional Anisotropy (FA), Mean Diffusivity (MD), Axial Diffusivity (AD), and Radial Diffusivity (RD), were computed from b = 0/1000 s/mm² images using DIPY [50]. To assess non-Gaussian diffusion properties, DKI metrics were computed using the full b-value set, including Kurtosis Fractional Anisotropy (KFA), Mean Kurtosis Tensor (MKT), Mean Kurtosis (MK), Axial Kurtosis (AK), and Radial Kurtosis (RK). DKI measures are particularly sensitive to microstructural complexity and heterogeneity and were used to capture subtle changes in glial reactivity and tissue architecture not detectable by DTI alone. ROIs for gray matter and white matter were defined by aligning T2-weighted atlas labels to diffusion space and by tractography-based segmentation of the corpus callosum and peripheral WM using DSI Studio (GQI reconstruction: https://dsi-studio.labsolver.org) [51, 52]. ROI values were extracted using FSL and custom scripts.

Tissue processing and immunohistochemistry (IHC)

Sample preparation

Following MRI, mice were euthanized by CO₂ overdose. Brains from four mice (one per sex per genotype) were hemisected: left hemispheres were flash-frozen for snRNAseq; right hemispheres were post-fixed. All other brains were fixed in 4% paraformaldehyde for 48 h. A total of 23 brains (19 whole brains and 4 half hemispheres) were cut into 100-µm coronal sections using a Leica VT1200 vibratome.

CLARITY clearing and IHC

Sections underwent passive CLARITY (Clear Lipid-exchanged Acrylamide-hybridized Rigid Imaging/Immunostaining/In situ-hybridization-compatible Tissue-hYdrogel) to clear tissue as previously described [53]. IHC followed previously established protocols as described in more details in Supplementary Methods [54]. Briefly, sections underwent PBS washes (3 × 10 min), glycine quenching, blocking (BSA, NDS, Triton X-100), and incubation with primary antibodies for 24–48 h at 4 °C. Secondary antibodies were applied overnight at 4 °C. Tissue was mounted with ProLong Gold and cured for 48–72 h. For mouse-host primary antibodies, the M.O.M. (Mouse on Mouse) Immunodetection Kit (Vector Laboratories) was used according to manufacturer instructions, replacing the standard blocking and diluent steps. Microwave antigen retrieval using the low-wattage PELCO Biowave (TED PELLA Inc, CA) was applied before antibody incubation to enhance signal penetration.

The following markers were used to examine different components of the neurovascular unit. Pericyte marker: goat anti-CD13 (1:100; R&D Systems AF2335). Matrix metalloproteinase-9 (MMP9): rabbit anti-MMP9 (1:100; Abcam ab38898). Cyclophilin A (CypA): rabbit anti-CypA (1:200; Proteintech 10270-1-AP). Brain vasculature: Dylight 594-conjugated Lycopersicon esculentum lectin (1:200; Vector Labs DL-1177). Tight junction proteins: rabbit anti-ZO-1 (1:200; Invitrogen 402200), anti-Occludin (1:100; Invitrogen 331500) and Claudin-5 (1:200; Invitrogen 352588). Perivascular IgG with goat anti-IgG (1:1000; Abcam AB197767). Reactive astrocyte markers: chicken anti-GFAP (1:500; Abcam AB4674), and rabbit anti-C3d (1:500; Bioss BS4877R). Oligodendrocyte maker: goat anti-Olig2 (1:200; R&D AF2418). Additionally, amyloid or tau pathology was examined using mouse anti-6E10 (1:100, Biolegend, 803004), mouse anti-phospho-tau (AT8, 1:500, Invitrogen MN1020), and mouse anti-phospho-tau (AT180, 1:500, Invitrogen MN1040).

Confocal imaging and quantification

Confocal images were acquired using an Andor BC43 microscope with a 40x objective. Confocal z-stacks (10 μm; 1 μm steps) were acquired from hippocampal and cortical regions with identical imaging parameters across animals. ImageJ was used for area-fraction analysis, vessel colocalization (CD13, MMP9, CypA, junctional proteins), astrocyte morphology [55], GFAP/C3d colocalization, and AQP4 expression and polarization. All animals that underwent MRI were included in this analysis. Values from 4 random fields were averaged to generate one data point per animal per region. Colocalization analyses were performed with the ImageJ colocalization plugin on each z-step unless specified. Group-wise statistical comparisons were performed in MATLAB (R2023b) using two-sample t-tests, with FDR correction for marker families (vascular markers and glial markers). Results are reported as t-values, p-values, and effect sizes (Cohen’s d). Statistical significance was defined as p < 0.05.

Single-nucleus RNA sequencing

Nuclei isolation and sorting

To examine cell-type-specific transcriptional changes associated with APOE genotype, an exploratory single-nucleus RNA sequencing (snRNAseq) experiment was performed on the hippocampus and cortex of four APOE-KI mice (n = 1 per sex per genotype). Frozen tissue was homogenized in nuclei isolation buffer, filtered, and DAPI-stained. ~50,000 nuclei per sample were FACS-sorted into RNase-free buffer while maintaining cold conditions.

Library preparation and sequencing

Sorted nuclei were delivered to the Boston University Single-cell Sequencing Core for quality assessment and library preparation. Nuclei were processed using the Chromium Single Cell 3’ v3.1 kit (10x Genomics) and sequenced on an Illumina NovaSeq (target: ~2000 nuclei/sample; ~50,000 reads/nucleus).

Data processing

Reads were aligned using Cell Ranger (--include-introns). Quality control was performed with singleCellTK; doublets, low-UMI nuclei, and high-contamination nuclei were removed. Data were normalized with SCTransform, corrected for batch/sex using Harmony, and clustered using PCA and the Louvain algorithm. UMAPs (Uniform Manifold Approximation and Projection) were computed for visualization [56–58]. Cell-type annotation was performed by comparing cluster-enriched genes to canonical markers for major brain cell types. Cell-type annotation was performed using canonical markers (e.g., SOX10, CLDN5, GJA1, SNAP25), cluster markers identified by FindAllMarkers, and comparison with reference datasets [59, 60]. Detailed description of the snRNAseq analysis can be found in Supplementary Methods. A list of outcome measures can be found in Supplementary Table 2.

Astrocyte subclustering and differential expression

Astrocytes were subset by region and re-analyzed with SCTransform. Subclusters were annotated as homeostatic or reactive based on gene markers. Differential expression between E3 and E4 conditions within each cell type and astrocyte subcluster was assessed using the Wilcoxon rank-sum test implemented in Seurat, with FDR correction applied. Given that only one biological sample per genotype, sex, and region was available, analyses were performed at the cell level and are interpreted as exploratory. Accordingly, emphasis was placed on effect sizes, expression patterns, and consistency with known biology rather than strict statistical significance. Additional methodological details are provided in the Supplementary Methods.

Cell-cell communication analysis

After annotation of major cell types and astrocyte subclusters, cell-cell communication analysis was conducted using the CellChat v1.6.1 R package, which infers intercellular signaling based on known ligand-receptor interactions [61]. Separate models were generated for APOE3 vs. APOE4, hippocampus vs. cortex. Communication probability, interaction count, centrality, and top ligand-receptor pathways were compared across groups. Due to limited representation of glial populations, a minimum cell threshold of 3 cells per group was used to retain rare cell types. Results are interpreted as exploratory and hypothesis-generating.

Statistical analysis

MRI outcome measures (ASL, DTI and DKI metrics) and IHC measures (% colocalization area, reactive astrocyte count, proximity measures, etc.) were compared between APOE3-KI and APOE4-KI mice using two-sample t-tests corrected with false discovery rate (FDR) method [62]. Gene expressions from snRNAseq data were compared between groups/astrocyte clusters using Wilcoxon rank-sum tests (adjusted via FDR) and interpreted as exploratory due to limited biological replications. Results are reported as t-values, p-values, FDR-corrected p-values. Correlations between measures were conducted using Pearson correlation and reported as the correlation coefficient (r) and p-values. Effect sizes (Cohen’s d) were reported alongside p-values for MRI and IHC outcomes. Primary findings were considered at FDR-p < 0.05, but uncorrected p < 0.05 findings were also reported for this exploratory study in young mice. All analyses were performed using MATLAB (MathWorks, R2023b), and figures were generated using GraphPad Prism (version 10.5.0) or R package (v4.2.2).

Results

A total of 23 mice completed MRI (mean age at time of MRI = 7.7 ± 1.3 months old), which included 9 APOE3-KI mice (n = 4 females, n = 5 males) and 14 APOE4-KI mice (n = 7 females, n = 7 males). One of the animals (E3 male 94455RN) was excluded due to an enlarged ventricle, resulting in failed QC during image processing. The remaining 22 mice were used for further statistical analyses.

Neurovascular dysfunction in APOE4-KI mice

Comparing the APOE4-KI (hereafter E4) to the APOE3-KI (hereafter E3) mice, we observed significantly shorter hippocampal water exchange time (T_ex) in the E4 mice, which may reflect changes in NVU water exchange dynamics (Fig. 1D). The hippocampal (HP) T_ex was significantly reduced (t = -4.9, p = 0.0001, FDR-p = 0.0005, d = -2.1) compared to E3 mice. In the retrosplenial and somatosensory cortex (CTX), E4 mice also exhibited significantly lower T_ex (t = -3.2, p = 0.0044, FDR-p = 0.0073, d = -1.4), in the retrosplenial cortex (RSC) alone (t = -3.6, p = 0.0016, FDR-p = 0.0040, d = -1.6), and in the thalamus (t = -2.6, p = 0.019, FDR-p = 0.024, d = -1.2). A similar trend was found in the peripheral cortex (t = -1.6, p = 0.13, FDR-p = 0.13, d = -0.72). When performing sex-stratified comparisons, the group differences in hippocampal T_ex remained significant in both female mice (t = -3.7, p = 0.0050, FDR-p = 0.025, d = -2.1) and male mice (t = -3.0, p = 0.016, FDR-p = 0.040, d = -1.7). Interestingly, the female groups had a bigger effect size (absolute value of Cohen’s d) compared to the male group, suggesting a sex-dependent APOE4 effect on neurovascular exchange time in the hippocampus (Supplementary Fig. 1). In addition to the T_ex measure, E4 mice showed lower cerebral blood flow (CBF) in the hippocampus compared to E3 mice (t = -3.1, p = 0.0061, FDR-p = 0.012, d = -1.3) (Supplementary Fig. 2A). Across all animals, hippocampal CBF was positively correlated with hippocampal T_ex (r = 0.70, p = 0.0004), and this relationship remained significant within the E4 group (r = 0.62, p = 0.024) but not within the E3 group, suggesting a genotype-dependent association (Supplementary Fig. 2B), while cortical T_ex and CBF were not correlated with each other in whole sample or subgroup analyses. After adjusting for hippocampal CBF in a generalized linear model, differences in hippocampal T_ex between E3 and E4 mice remained significant (Supplementary Fig. 2C). Intravascular fraction (IV) was lower in APOE4 mice than APOE3 mice at both TI=800ms and 1500ms for the hippocampus (TI 800: t = -4.2, p = 0.0005, FDR-p = 0.0047, d = -1.8; TI 1500: t = -2.9, p = 0.010, FDR-p = 0.024, d = -1.2) and for the cortex (TI 800: t = -3.6, p = 0.0016, FDR-p = 0.0080, d = -1.6; TI 1500: t = -2.6, p = 0.017, FDR-p = 0.029, d = -1.1) (Supplementary Fig. 2D). For arterial transit time (ATT), group differences were not significant after FDR correction, where APOE4 mice showed a trend of increased ATT in the peripheral cortical region (t = -2.4, p = 0.031, FDR-p = 0.15, d = -1.1) than E3 mice (Supplementary Fig. 2C).

APOE4 is associated with altered grey matter microstructural environment

Comparisons of dMRI measures in grey matter regions showed trends that did not survive FDR corrections, suggesting that microstructural changes are subtle at this early stage. Specifically, there was a trend toward lower AK in the hippocampus of APOE4 mice (t = -2.4, p = 0.028, FDR-p = 0.19, d = -1.2), suggesting early signs of reduced directional complexity and potential axonal or glial disruption in this region (Fig. 1E). Additionally, there appears to be a trend for a reduced RK in the hippocampal region, albeit not statistically significant (t = -2.1, p = 0.054, FDR-p = 0.15, d = -1.1), indicating loss of microstructural complexity in the perpendicular direction (Fig. 1E). These diffusion-derived metrics reflect changes in overall microstructural complexity and heterogeneity, which may arise from multiple factors, including cellular, extracellular, and vascular alterations, rather than being specific to any single cell type. These findings align with the observed differences in NVU water exchange dynamics estimated using multi-TE ASL, further supporting early APOE4-associated microstructural and neurovascular disruptions in hippocampal and related cortical circuits. To examine the relationship between NVU water exchange time (T_ex) and diffusion measures, two GM regions, HP and CTX, which showed the most significant change in T_ex, were included for correlation analysis. Across all animals, hippocampal T_ex was positively associated with hippocampal RK (r = 0.49, p = 0.029). When examined within E3 and E4 groups separately, the associations were not statistically significant, but both groups exhibited trends in the same direction as the pooled analysis (Fig. 1F). This pattern suggests that the overall relationship may not be solely driven by between-genotype differences, but that the lack of significance within subgroups may reflect limited statistical power due to smaller subsample sizes. Accordingly, this finding is interpreted cautiously as a consistent but underpowered within-genotype trend rather than a robust individual-level association.

APOE4 is associated with reduced white matter microstructural integrity

For WM regions, significant group differences were observed in the corpus callosum (CC). Compared to E3 mice, FA was significantly lower in APOE4 mice (t = -3.0, p = 0.0074, FDR-p = 0.019, d = -1.5), potentially reflecting reduced microstructural integrity and axonal loss in the CC (Supplementary Fig. 2F). E4 mice also had higher RD (t = 3.1, p = 0.0068, FDR-p = 0.019, d = 1.5) and MD (t = 2.7, p = 0.017, FDR-p = 0.023, d = 1.3) in the CC compared to APOE3-KI mice, which may reflect compromised myelin integrity or increased extracellular space and greater overall water diffusivity within the CC. Similar patterns were also observed in the peripheral WM regions, including higher MD (t = 3.0, p = 0.0091, FDR-p = 0.019, d = 1.5), RD (t = 2.7, p = 0.017, d = 1.3), and AD (t = 3.0, p = 0.0092, FDR-p = 0.019, d = 1.5) in APOE4 mice compared to APOE3 mice. These changes collectively reflect early APOE4-associated white matter microstructural degeneration, particularly in major commissural pathways.

APOE4 is associated with NVU remodeling and increased perivascular MMP9 expression

To investigate the molecular mechanisms underlying APOE4-associated NVU alterations, we performed ex vivo tissue analysis (immunohistochemistry and single-nucleus RNA sequencing) in the hippocampus and cortex (Fig. 2A). In the hippocampus, E4 mice had greater MMP9 expression in vasculature-associated compartments. MMP9 colocalized with CD13 + pericytes was significantly higher in E4 compared to E3 mice (t = 3.1, p = 0.0054, FDR-p = 0.016, d = 1.3) (Fig. 2B&C). When examining males and females separately, the APOE4-related MMP9 upregulation in the pericytes was only significant in the female group comparisons (Supplementary Fig. 3), consistent with the sex-dependent effect in the exchange time result (Supplementary Fig. 1). Across all animals, the percentage area of MMP9 and CD13 colocalization was negatively correlated with water exchange time (T_ex) measured by multi-TE ASL; however, this relationship was not significant within genotype, suggesting that the pooled association may be influenced by genotype-related differences (Fig. 2D). A similar elevation was found for MMP9 colocalization with lectin-labeled endothelial cells (t = 2.9, p = 0.0081, FDR-p = 0.016, d = 1.2) and negatively correlated with water exchange time (Supplementary Fig. 4A&B). When comparing the %area of CypA that colocalizes with CD13 + pericytes, E4 mice showed higher colocalization of the two markers than E3 mice, but the group difference diminished after FDR correction (t = 2.2, p = 0.047, FDR-p = 0.063, d = 0.99) (Supplementary Fig. 4C). However, other vascular integrity markers were not different between E3 and E4 animals, including total pericyte coverage and tight junction coverage on endothelium, and total %area of perivascular IgG deposits (Supplementary Fig. 4E). Perivascular IgG deposits, which are markers of macromolecular barrier leakage, did not differ significantly between genotypes (t = 1.5, p = 0.16, d = 0.61) (Supplementary Fig. 4). Tight junction-associated markers (Claudin-5, Occludin, ZO-1) were also not different between genotypes in the hippocampus (all p > 0.2, d < 0.52), suggesting structural tight junctions were preserved at this age (Fig. 2E&F). In the cortex, vascular MMP9-CD13 colocalization was also higher in APOE4 mice (t = 2.1, p = 0.047, FDR-p = 0.063, d = 0.93), which had a smaller effect size compared to the hippocampus (Fig. 2C Right Panel). In the cortex, MMP9-lectin colocalization signal trended toward significance (t = 1.9, p = 0.070, FDR-p = 0.070, d = 0.84). Lastly, amyloid (6E10) or tau (AT8 or AT180) depositions were not detected in the brains of either group (Supplementary Fig. 4D). These findings suggest functional changes of neurovascular unit components, such as early matrix remodeling and pericyte-linked protease activation in E4 mice, where the hippocampus has greater vulnerability than the cortex, even in the absence of severe vascular structural damage or well-known AD pathology (i.e., amyloid or tau depositions).

Fig. 2.

Ex vivo tissue analysis showed increased MMP9 expression in pericytes of APOE4 mice. A Schematic of mouse brain used for immunohistochemistry and single-nucleus RNA sequencing. This figure was generated using BioRender. B Representative images of Lectin (red), CD13 (green) and MMP9 (magenta) immunostaining in APOE3 (top) and APOE4 (bottom) mice in the hippocampus. C Group comparison of MMP9 and CD13 colocalization across all animals in hippocampus and cortex. D Pearson correlation between MMP9 and CD13 colocalization area with T_ex derived from multi-TE ASL (n = 22 including 8 E3 mice and 14 E4 mice). Correlation coefficients (R) were presented for analysis across all animals (black), within E3 animals (blue), and within E4 animals (red). Asterisk indicated significant correlations. E Representative images of tight junction markers Claudin-5 (green) and ZO1 (magenta) with Lectin (red), and (F) group comparison on ZO1 and Lectin colocalization in the hippocampus. Group comparisons were performed using independent t-tests. *p < 0.05. **p < 0.005. Data presented as mean ± SEM. HP = hippocampus; CTX = cortex

APOE4 is associated with astrocyte reactivity without affecting perivascular AQP4 polarization

To examine glial responses to vascular damage, we analyzed astrocytic markers in the hippocampus (Fig. 3) and cortex. Astrocytes are also a critical component of the NVU, not only to maintain NVU integrity but also to comprise the perivascular glymphatic system, which regulates waste clearance from the interstitial fluid-CSF exchange that is crucial for the cellular environment. When examining astrocyte morphology and functions, E4 mice demonstrated elevated astrocytic reactivity in the hippocampus. The percentage of reactive astrocytes, identified based on morphology (hypertrophy), suggested in previous literature [55], was higher in APOE4 hippocampus than in APOE3 hippocampus (t = 3.4, p = 0.0033, FDR-p = 0.017, d = 1.5) (Fig. 3A&B). Across all animals, hippocampal reactive astrocyte count was positively correlated with MMP9 + CD13 colocalization (r = 0.57, p = 0.012). Within genotype-stratified analyses, this relationship was not statistically significant; however, a directional trend was apparent in the APOE4 group, whereas the APOE3 group showed relatively flat patterns (Fig. 3C). This suggests that the pooled association may be partly driven by APOE4-related effects, while the lack of significance within subgroups may reflect limited power and potentially non-linear relationships at the individual level. Complement component 3d (C3d) was previously shown to be upregulated in the neurotoxic subtype of astrocytes [63]. APOE4 mice had a significant increase in GFAP+C3d colocalization (t = 2.7, p = 0.015, FDR-p = 0.037, d = 1.2), indicating a shift toward a reactive neuroinflammatory phenotype (Supplementary Fig. 4F). AQP4 was previously suggested to be elevated in reactive astrocytes and may regulate inflammatory cytokine secretion by astrocytes [64, 65]. The expression of AQP4 on GFAP+ astrocytes was higher in E4 mice (t = 2.2, p = 0.041, FDR-p = 0.069, d = 0.93), but AQP4 perivascular polarization was not significantly altered (t = 0.26, p = 0.80, d = 0.11) (Fig. 3D&E&F). These measures were not significantly different in the cortex, regardless of FDR correction. In the cortex, the GFAP+AQP4 signal was elevated in E4 but did not reach statistical significance (t = 1.1, p = 0.29, d = 0.45). AQP4 polarization index in astrocytes was also similar between E3 and E4 mice in the cortex (t = -0.40, p = 0.69, d = -0.16).

Fig. 3.

Ex vivo tissue analysis showed astrocyte morphological and functional changes in APOE4 mice. A Representative images of GFAP (green) immunostaining in APOE3 (left) and APOE4 (right) mice in the hippocampus. B Group comparison of percentage of reactive astrocyte count based on morphology. C Correlation between reactive astrocyte count and IHC MMP9 + CD13 colocalization in hippocampus. Correlation coefficients (R) were presented for analysis across all animals (black), within E3 animals (blue), and within E4 animals (red). Asterisk indicated significant correlations. D Representative images of AQP4 (blue), GFAP (green), Lectin (red) in the hippocampus. E Group comparisons of percentage area of colocalization between AQP4 and GFAP normalized to total GFAP, and (F) AQP4 polarization index calculated by the amount of AQP4 + GFAP+Lectin triple colocalization normalized to total area of Lectin and GFAP. A total of 23 animals were included in these analysis (n = 9 E3 mice, n = 14 E4 mice). Group comparisons were performed using independent t-tests, and correlation using Pearson Correlation. *p < 0.05. **p < 0.005. ***p < 0.0005. E3 = APOE3-KI; E4 = APOE4-KI; HP = hippocampus; CTX = cortex

Exploratory single-nucleus RNA sequencing suggests cell-type-specific transcriptional responses to APOE4

To identify cell-type-specific transcriptional changes associated with APOE genotype, exploratory snRNAseq was performed on hippocampal and cortical tissue from four APOE-KI mice (n = 1 per sex per genotype) and therefore does not permit formal genotype-level inference. After quality control filtering, nuclei were clustered using graph-based clustering in Seurat and visualized using UMAP projections (Fig. 4A and colored by genotype in Supplementary Fig. 5). Primary cell types were annotated based on canonical marker expression suggested by previous literatures [59, 60], including n = 711 astrocytes (Aldh1l1, Aqp4), n = 6175 oligodendrocytes (Plp1, Mbp), n = 549 oligodendrocyte precursor cells, n = 1862 microglia (C1qa, Trem2), n = 1014 endothelial cells (Cldn5, Pecam1), n = 137 pericytes (Pdgfrb, Rgs5), n = 9226 excitatory neurons (Slc17a7), and n = 4332 inhibitory neurons (Gad1). Subgroup-specific cell count can be found in Supplementary Table 3. Cell-level expression patterns were qualitatively consistent with imaging and IHC findings (Supplementary Fig. 6A&B). For the current study, primary focus was put on astrocytes and pericytes, which play critical roles in the relationship between APOE and NVU disruption, as discussed previously. In pericytes and endothelial cells, higher Ppia (encoding cyclophilin A) expression was observed in pericytes from the APOE4 sample, aligning with the IHC findings (Fig. 4B). Similarly, in astrocytes, higher Aqp4 expression was observed in astrocytes from the E4 sample, suggesting altered water homeostasis and potential astrocytic endfeet dysfunction (Fig. 4C). Also, it showed a trend of correlations with markers from IHC that target the same pathways (Supplementary Fig. 6B). These changes were more prominent in hippocampal samples (i.e., astrocyte Aqp4 and pericyte Ppia were not differentially expressed in cortex), supporting the regional specificity of APOE4-associated vulnerability. Other well-known astrocyte-related markers, such as Snta1 (syntrophin alpha 1, encoding the anchoring protein for AQP4) and Gfap, were not differentially expressed when comparing E4 to E3 mice (p > 0.05 for both measures). This APOE4-associated pattern is an exploratory observation that aligns with other findings but does not independently establish genotype effects due to limited biological replication.

Fig. 4.

snRNAseq identified unique astrocyte subclusters with distinct genetic signatures in APOE4 mice. A UMAP of major cell type cluster from snRNAseq. B Cell-level differential expression (using Wilcoxon test) analysis comparing APOE4 vs. APOE3 log(expression) of Ppia in pericytes, and (C) of Aqp4 in astrocytes. D Left: UMAP visualization of astrocyte subclusters in hippocampus (top) and cortex (bottom). Right: Proportion of astrocyte subclusters in APOE3 and APOE4 mice in each region. E Comparison of astrocyte subclusters in log (expression) Aqp4 in hippocampus (top) and cortex (bottom). A total of 4 animals were included in these analysis (n = 1 per genotype per sex). Because each genotype is represented by a single biological replicate, all snRNAseq analyses are descriptive and hypothesis-generating. *p < 0.05. **p < 0.005. ***p < 0.0005. ****p < 0.00005

Regionally distinct astrocyte subclusters enriched in APOE4 hippocampus

To investigate heterogeneity within astrocytes, subclusters were identified for each region. Three astrocyte subclusters were identified in the hippocampus and four in the cortex, each displaying distinct transcriptional signatures (Fig. 4D). Among hippocampal clusters, Cluster 0 was found relatively uniform in E3 and E4 mice. Cluster 1 was E3-enriched, expressing genes related to homeostatic functions at a high level (Table 2). In contrast, Cluster 2 contained a higher proportion of APOE4-derived astrocytes in this dataset (26 E4 astrocytes and 1 E3 astrocyte) and showed higher expression of reactive astrocyte (reactivity, glial scar formation, and stress response) markers compared to other astrocytes (Table 2). Based on these top-expressed genes in each cluster, Cluster 2 was identified as “reactive astrocytes” and Cluster 1 as “homeostatic astrocytes”. Compared to clusters 0 and 1, cluster 2 also showed significantly higher expression of Aqp4 (Fig. 4E). In the cortex, four astrocyte subclusters were identified, with cluster 2 also showing enrichment of E4 cells (101 E4 cells and 16 E3 cells). However, the expression of Aqp4 was not different between the four cortical astrocyte subclusters (Fig. 4E).

Table 2.

Top genes expressed by hippocampal astrocyte subclusters

	Gene Name	Function
Cluster 1 (E3-enriched)	Slc1a2	Major astrocytic glutamate transporter, clears ~ 90% of synaptic glutamate → protects neurons from excitotoxicity. Functions in homeostasis and is downregulated in neurodegeneration.
	Slc4a4	Na⁺-bicarbonate cotransporter in astrocytes, critical for ionic & pH homeostasis; typical of mature astrocyte phenotype.
	Erbb4	Astrocyte-neuron signaling receptor; enhances calcium signaling and potential synaptic modulation.
	Nrxn1 / Cadm2 / Lsamp / Gpc5	Adhesion-associated genes involved in astrocyte-neuron interactions and synaptic support.
Cluster 2 (E4-enriched)	Csgalnact1	Initiates chondroitin sulfate chain synthesis → key proteins involved in reactive glial scarring.
	Lama2	Laminin subunit contributing to scar matrix; upregulation is typical in reactive astrocytes.
	Il1rapl1	Receptor linked to inflammatory signaling; IL-1 pathway is prominent in reactive and scar-forming astrocytes.
	Gria2 / Gabrb1	Glutamate and GABA receptor subunits hint at astrocyte involvement in altered neuronal signaling during reactivity or stress.
	Adk / Glud1	Metabolic enzymes in reactive astrocytes; alterations impact glutamate metabolism and may exacerbate neurotoxicity.
	Nfia / Sox6 / Hdac8 / Phlpp1	Transcriptional regulators involved in astrocyte reactivity, glial scar formation, and cellular stress responses.

Exploratory cell-cell communication analysis suggests potential astrocyte-pericyte signaling changes

After defining the astrocyte subcluster in the hippocampus that showed distinct signatures (homeostatic vs. reactive), CellChat was used to explore putative astrocyte-pericyte communication patterns in APOE3 and APOE4 samples. Exploratory CellChat analysis suggested potential differences in astrocyte-pericyte signaling patters in E4 mice, specifically from reactive astrocytes (Cluster 2) to pericytes (Supplementary Fig. 6C). Given that each genotype is represented by a single biological sample, these networks are illustrative and do not represent population-level estimates of cell–cell communication. Top ligand-receptor pairs contributing to increased communication (weight and count) from reactive astrocytes to pericytes include Vegfa-Vegfr1, Ncam1-Fgfr1, Glutamate (via Slc1a3 + Gls)-Grm7/Gria, and Fgf1-Fgfr1, etc. Statistical significance was assessed using a permutation-based method to generate a distribution of communication weights. APOE4 mice also showed a decreased outgoing signaling weight and an increased count from homeostatic astrocytes (Cluster 1) to pericytes. These observations suggest potential differences in astrocyte-pericyte signaling patterns in the APOE4 sample. However, given the limited biological replication and sparse representation of pericytes, these findings should be interpreted cautiously and are represented as hypothesis-generating.

Discussion

The present study demonstrates that non-invasive MRI detects early neurovascular and microstructural abnormalities in 6-9-month-old APOE4 knock-in mice, supporting the hypothesis that vascular dysfunction represents a primary and early component of APOE4-associated brain aging. Specifically, APOE4-KI mice showed significantly shortened NVU water-exchange times and regionally specific microstructural alterations in the hippocampus, indicating subtle but measurable vascular functional disruptions and tissue perturbation without pathological amyloid deposition or neurodegeneration. In vivo imaging findings were also supported by APOE4-related changes in brain biomarker expression and transcriptomic profiles consistent with vascular dysfunction and astrocyte inflammation. Importantly, reactive astrocytes and vascular changes co-occur, suggesting potential interactions between glial and vascular components that may further compromise vascular integrity and promote downstream damage. These results extend previous literature by demonstrating that neurovascular dysfunction, such as altered water exchange across the neurovascular unit, can be detected in young adult mice, provide converging evidence that APOE4 impacts may begin with subtle neurovascular dysfunction, and highlight the importance of early glial-vascular signaling changes as a therapeutic target in preclinical AD.

APOE4, NVU exchange function, and MMP9 activation

Altered NVU exchange function in APOE4 mice is consistent with prior studies showing that APOE4 activates the pericyte-specific CypA-NF-κB-MMP9 pathway, leading to degradation of basement membrane and tight-junction proteins [12, 16, 66]. In the current study, APOE4-KI mice showed increased colocalization of MMP9 with CD13 + pericytes and lectin-labeled endothelial cells, and smaller water exchange times (T_ex). Although MMP9-related measures and T_ex were inversely correlated when combining all animals, the relationship was not significant within smaller genotype subgroups, suggesting that further investigations in larger samples are necessary to provide direct evidence of a linear mechanistic relationship between MMP9 activity and T_ex at the individual level. Previous mouse and human studies showed that APOE4 carriers exhibited elevated MMP9 expression and decreased NVU integrity, particularly in the hippocampus [11, 12, 16, 19, 67]. APOE4’s reduced interaction with the LRP1 receptor in pericytes leads to disinhibition of CypA expression, which activates NF-κB and upregulates MMP9, promoting degradation of basement membranes and tight junctions [16, 66]. These changes emerged in the absence of amyloid or tau pathology, supporting a vascular-first hypothesis of APOE4-mediated brain aging. However, it is important to note that APOE and LRP1 signaling may also be involved in amyloid clearance [68]. The interaction between the APOE-LRP1 amyloid clearance pathway and the Cyp-MMP9 pathway is worth exploring using an alternative model, such as the APOE4*APP/PS1 model [69].

Importantly, these molecular changes occurred in the absence of detectable loss of tight junction coverage or increased perivascular IgG deposition, suggesting that structural disruption of the vascular barrier is not yet prominent at this stage. Instead, our results support a model in which APOE4 primarily alters the dynamics of transvascular water exchange and NVU signaling (early subclinical NVU dysfunction), rather than inducing significant barrier breakdown. This interpretation is further supported by the observed reductions in water exchange time (T_ex derived from multi-TE ASL), which likely reflect shifts in compartmental exchange kinetics and micro-environmental properties rather than gross structural failure.

Region-specific vulnerability was observed, with the hippocampus showing the most pronounced changes in T_ex, intravascular fraction (IV), cerebral blood flow (CBF), and accompanied by loss of microstructural integrity. This is captured by the in vivo neuroimaging measures showing reduced hippocampal T_ex, IV and CBF measures in the E4 mice compared to E3 mice. It is important to note that prior studies have observed a correlation between CBF and the estimated intravascular fraction of the ASL signal [70]. The current study also found association between CBF and T_ex in the hippocampus across all animals and within E4 mice, suggesting that the T_ex measure may be influenced by perfusion (primarily driven by the E4 group). However, the reduction of CBF in APOE4 mice shown by the current study was relatively subtle compared to the range of CBF values investigated in Silva et al. [70], meaning that our observations of markedly different water exchange times are likely to be primarily reflective of differences in the permeability of BBB to water. Results from the generalized linear model in the current study supported this notion by showing that after adjusting for CBF, the main effect of APOE4 carrier status on T_ex remained significant. A recent study using multi-TE ASL to examine the effect of caffeine in human brains showed CBF changes but not T_ex, suggesting partial independence between the two measures [71].

The current study highlighted the hippocampus as the key region affected by APOE4 expression in young adult mice. A previous study in cognitively normal adults (age range 20–91 years) showed that the hippocampus (CA1 and dentate gyrus regions) had an early and progressive increase in BBB permeability with age, while other cortical regions were not significantly impacted [8]. Interestingly, hippocampus and thalamus had the lowest permeability (DCE-MRI K_trans about 8 × 10^− 4/min) at young age range (20–40 years old) compared to other regions (DCE-MRI K_trans >1 × 10^− 3/min), which may suggest that they rely heavily on tightly regulated interactions between endothelial cells, pericytes, and astrocytic endfeet that are particularly sensitive to APOE4 risks. Pericytes are densely distributed in hippocampal capillaries and are particularly vulnerable to age-related oxidative stress and APOE4-mediated degeneration [72]. Pericyte loss leads to tight junction disruption, increased transcytosis, and diminished astrocyte polarization, all of which compromise NVU integrity [73]. Additionally, hippocampal glial populations exhibit regionally distinct transcriptional responses to BBB stress, including early upregulation of inflammatory and lipid metabolism pathways in astrocytes and microglia [74]. These glial responses can amplify neurovascular dysfunction through cytokine release and impaired neurovascular couplings. Collectively, these findings suggest that the hippocampus is a vulnerable region where tightly controlled cellular interactions are more susceptible to the impact of aging and the APOE4 genotype, making it a critical area for early neurovascular dysfunction.

Astrocyte reactivity and glial-vascular crosstalk

Astrocyte reactivity and vascular remodeling were prominent features observed in APOE4 mice, as evidenced by increased GFAP hypertrophy, AQP4 expression, and perivascular MMP9 activity. These changes occurred in parallel across imaging and histological measures, suggesting coordinated glial and vascular alterations in early APOE4-associated neurovascular dysfunction. Hippocampal reactive astrocyte burden was also associated with MMP9 + CD13 colocalization in pericytes across all animals, consistent with coordinated glial and vascular changes. However, this association was not statistically significant within genotype-stratified analyses. A directional trend was observed in subgroup, and the lack of significance within subgroups may reflect limited statistical power and potentially non-linear relationships and therefore warrants further validation in larger samples.

Astrocytes maintain NVU function via their endfeet and modulate extracellular fluid balance and neuroinflammation [75, 76]. Increased signaling from reactive astrocytes may impair pericyte stability and basement membrane maintenance, contributing to neurovascular dysfunction [77]. Previous evidence showed that pericytes influence endothelial barrier properties both directly and indirectly, including via gap junction communication, integrins, and secretion of proteases that remodel the extracellular matrix [73]. Astrocytes, through their endfeet, secrete factors (e.g., laminin) that maintain pericyte attachment, survival, and function [78]. Conversely, pericytes regulate the assembly of astrocytic endfeet and the expression of astrocytic AQP4, which are key for maintaining NVU integrity [79]. Dysfunctional crosstalk can lead to pericyte detachment or differentiation into a contractile phenotype, compromising NVU integrity, including the BBB [80]. This represents a novel insight into the early glial-vascular interactions in APOE4 carriers: reactive astrocytes may not only respond to damage but may also participate in processes associated with vascular remodeling, suggesting potential astrocyte-pericyte interactions, supported by previous studies [75, 77, 81]. Interestingly, AQP4 polarity or SNTA1 expression in astrocytes were similar in young APOE4 and APOE3 mice. The perivascular localization of AQP4 appeared to be preserved, suggesting that glymphatic clearance remains largely intact [82, 83]. Another study detected loss of astrocytic endfeet coverage on blood vessels in 10-month-old APOE4 mice using electron microscopy, suggesting glymphatic damage may occur during aging [67]. These findings support a model in which BBB breakdown is an early and vulnerable point of the NVU, emerging before prominent dysfunction in glymphatic or neuronal systems. The astrocyte-pericyte interaction is a critical regulatory pathway through which early APOE4-related glial changes may initiate or exacerbate vascular damage.

Exploratory snRNAseq analyses provided additional cell-type-specific transcriptional context, identifying an APOE4-enriched astrocyte subpopulation with gene expression patterns consistent with reactive and matrix-related processes. CellChat analysis further illustrated potential astrocyte-pericyte signaling pathways in these samples, including Vegfa-Vegfr1 and Ncam1-Fgfr1 pathways, which have been implicated in vascular remodeling, neuroinflammation, and cognitive disorders [84–86]. However, these transcriptomic findings should be interpreted cautiously. The dataset includes one biological sample per genotype, sex, and region, limiting the ability to distinguish biological variability from sample-specific effects. In addition, sparse representation of key cell types, including pericytes, constrains downstream analyses. As such, differential expression and CellChat-based communication analyses are considered exploratory and hypothesis-generating. While these findings suggest potential molecular pathways linking astrocyte reactivity and vascular remodeling, they do not provide definitive evidence of altered astrocyte-pericyte signaling and require validation in larger datasets.

Sex-specific effects of APOE4 on NVU vulnerability

This study also suggests sex-specific effects in APOE4-associated neurovascular dysfunction. Female APOE4-KI mice showed greater NVU water exchange alterations, stronger MMP9 activation, and more prominent astrocyte reactivity than males. Sex may modify the impact of APOE4 on brain aging, with females showing increased vulnerability to neurovascular and cognitive decline, aligning with prior research linking estrogen decline to impaired cerebrovascular regulation and heightened AD risk [87]. These observations are consistent with prior studies showing that APOE4 increases AD risk more significantly in women than men [88, 89]. Estrogen is known to regulate multiple aspects of NVU function, including astrocyte-mediated NVU support, anti-inflammatory signaling, and synaptic plasticity [90–92]. mouse models demonstrate that estrogen receptor β (ERβ) in astrocytes preserves hippocampal structure during aging, and its loss leads to hippocampal shrinkage and memory decline [93]. Moreover, human imaging studies suggest that timely hormonal replacement therapy (HRT) may mitigate APOE4-related brain aging in women if initiated near the onset of menopause, although delayed HRT could increase tau accumulation [94]. On the contrary, studies also suggest that older females undergoing hormonal replacement therapy have smaller hippocampal volume compared to non-users or past-users, an effect that is independent of APOE4 [95]. Future studies investigating hormonal effects need to carefully account for the type of regimens, initiation, and duration of the treatment. These factors may partially account for the inconsistent results from previous studies examining the interactive effect of APOE4 and HRT on brain aging [87].

Limitations and future directions

Although this study provides mechanistic insights, several limitations warrant consideration. First, cross-sectional design precludes definitive conclusions about temporal causality and should be further validated with functional manipulations and longitudinal assessments. While increased MMP9 expression and perivascular remodeling were observed, evidence for activation of the full CypA-MMP9 pathway remains partial. The smaller effect size of CypA relative to MMP9 in CD13 + pericytes suggests that additional mechanisms, such as hypoperfusion or oxidative stress, may contribute to MMP9 activation [96, 97]. Targeted experimental manipulation will be required to establish causality. For example, astrocyte-specific conditional knockout of APOE4, or pharmacological inhibition of CypA or MMP9 in APOE4 mice, could directly test whether modulating these components mitigates BBB disruption and glial reactivity [67, 98]. Preclinical evidence suggests that MMP9 inhibitors, such as minocycline and small-molecule zinc chelators, can attenuate neurovascular damage and improve cognitive outcomes in other neurodegenerative and ischemic models [99, 100]. In addition, the enzymatic activity of MMP9 can be measured using gelatin zymography or other gelatin assays, which may be relevant compared to measuring total MMP9 protein level [101, 102]. Similarly, employing gain- or loss-of-function approaches in in vitro BBB models, such as iPSC-derived neurovascular units or microfluidic BBB-on-a-chip platforms to recapitulate tight junctions and cell-cell communication, could enable high-resolution analysis of pathway dynamics under APOE4 versus APOE3 conditions and validate the functional relevance of these pathways in a controlled setting [103–105]. Second, the snRNAseq dataset is limited by the lack of biological replication and sparse representation of certain key cell types (e.g., pericytes), introducing potential pseudoreplication. Differential expression analyses were therefore performed using cell-level Wilcoxon tests and interpreted as exploratory. Cell-cell communication analyses using CellChat were conducted with a permissive threshold to retain rare cell populations, which may reduce robustness. These findings are therefore considered hypothesis-generating and require validation with complementary approaches and in a larger dataset. It is important to note that the transcriptomic findings should not be interpreted as mechanistic evidence but rather as supportive context for imaging and histological observations. Third, imaging-based biomarkers, including multi-TE ASL and diffusion metrics, remain indirect proxies of BBB and glial health. The water exchange time (T_ex) derived from multi-TE ASL represents an indirect measure of NVU exchange dynamics and may be influenced by cerebral blood flow and astrocyte-mediated water transport. Although APOE4-related differences remained significant after CBF adjustment, T_ex should be interpreted as reflecting composite neurovascular exchange processes rather than direct structural barrier disruption. Future studies using complementary approaches (e.g., DCE-MRI or tracer-based assays) will help further delineate underlying mechanisms. In addition, incorporating spatial transcriptomics, proteomics, and advanced imaging techniques (e.g., 3D light sheet microscopy [106] or electron microscopy [67]) may help refine mechanistic understanding and identify early biomarkers at higher resolutions. Finally, increasing evidence suggests that the multi-TE ASL techniques are sensitive to NVU water exchange alterations in aging and neurodegeneration in the human brain [27–29, 107, 108], but potential change due to APOE4 carrier status has not yet been explored. Our group has recently started testing multi-TE ASL in human participants for studying APOE4 and aging. Future experiments should also explore other mouse models, such as the APOE2 or APOE Christchurch knock-in mice, which may provide novel insights into the potential protective effects of APOE variants [109, 110]. The association between APOE and other members of metalloproteinases, such as MMP2 and ADAMTS, also warrants further investigations [18, 111].

Conclusions

This study demonstrates that APOE4 may be associated with alterations in NVU function, reflected by changes in transvascular water exchange, glial and vascular alterations. By integrating in vivo imaging and ex vivo molecular profiling, this study establishes that altered NVU exchange dynamics and reactive glial changes are among the earliest detectable features of APOE4-related brain aging, especially in the hippocampus and in females. These findings highlight the potential of noninvasive imaging biomarkers and glial-vascular targets for early detection and intervention in individuals at risk for Alzheimer’s disease, laying the groundwork for future mechanistic and therapeutic studies.

Abbreviations

Aβ

Amyloid-beta

LOAD

Late-onset Alzheimer’s disease

APOE

Apolipoprotein E

APOE3

Apolipoprotein E ε3 allele

APOE4

Apolipoprotein E ε4 allele

ASL

Arterial spin labeling

AQP4

Aquaporin-4

BBB

Blood-brain barrier

CD13

Aminopeptidase N

CLARITY

Clear Lipid-exchanged Anatomically Rigid Imaging/Immunostaining-compatible Tissue hYdrogel

CNS

Central nervous system

CSF

Cerebrospinal fluid

CypA

Cyclophilin A

DEG

Differentially expressed gene

dMRI

Diffusion magnetic resonance imaging

GFAP

Glial fibrillary acidic protein

IHC

Immunohistochemistry

IgG

Immunoglobulin G

KI

Knock-in

LRP1

Low-density lipoprotein receptor-related protein 1

MMP9

Matrix metalloproteinase-9

MRI

Magnetic resonance imaging

NF-κB

Nuclear factor kappa B

NVU

Neurovascular unit

PPIA

Peptidylprolyl isomerase A

ROI

Region of interest

SnRNAseq

Single-nucleus RNA sequencing

TE

Echo time

WM

White matter

ZO-1

Zonula occludens-1

Authors’ contributions

YG conducted the data collection and analysis and drafted the manuscript. CC supported data collection and analysis. SK supported analysis of transcriptomic data. FD supported histology and confocal imaging data collection. YO supported data analysis and interpretation. JW provided multi-TE ASL data analysis pipelines and was involved in study design and analysis. BK designed the study, developed the analysis framework, lead the data analysis and manuscript writing.

Funding

This project was supported by the Department of Defense / Congressionally Directed Medical Research Programs (CDMRP) TERP HT94252410934, and Boston University Marion & Henri Gendron Fund 2023 (YG, CC, SK, FD, KB). JAW is supported by the Wellcome Trust (225345/Z/22/Z). YO is supported by the Wellcome Trust (316339/Z/24/Z).

Data availability

The datasets and codes generated during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

All animal procedures were approved by the Boston University IACUC. No human data or tissue were included in the current study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.