Endothelial Cell APOE3 Regulates Neurovascular, Neuronal, and Behavioral Function

Abstract

BACKGROUND:

Specialized brain endothelial cells and human APOE3 are independently important for neurovascular function, yet whether APOE3 expression by endothelial cells contributes to brain function is currently unknown. In the present study, we determined whether the loss of endothelial cell APOE3 impacts brain vascular and neural function.

METHODS:

We developed APOE3^fl/fl/Cdh5(PAC)-CreERT2^+/− (APOE3^Cre+/−) and APOE3^fl/fl/Cdh5(PAC)-CreERT2^−/− (APOE3^Cre−/−, control) mice and induced endothelial cell APOE3 knockdown with tamoxifen at ≈4 to 5 weeks of age. Neurovascular and neuronal function were evaluated by biochemistry, immunohistochemistry, behavioral testing, and electrophysiology at 9 months of age.

RESULTS:

We found that the loss of endothelial APOE3 expression was sufficient to cause neurovascular dysfunction including higher permeability and lower vessel coverage in tandem with deficits in spatial memory and fear memory extinction and a disruption of cortical excitatory/inhibitory balance.

CONCLUSIONS:

Our data collectively support the novel concept that endothelial APOE3 plays a critical role in the regulation of the neurovasculature, neural circuit function, and behavior.

Highlights.

• Knockdown of endothelial cell APOE3 resulted in higher neurovascular permeability and lower vessel coverage in vivo.

• Reducing endothelial cell APOE3 expression also resulted in deficits in behavior and neural circuit function in the prefrontal cortex.

• We propose that endothelial APOE3 plays an important role in the regulation of neural function in both physiological aging and neurodegeneration.

Human APOE plays a key role in brain function, which is highlighted in neurodegenerative disorders. Compared with APOE3, APOE4 is associated with greater cognitive decline during aging, poorer outcomes following stroke and traumatic brain injury, and higher risk of developing Alzheimer disease.^1,2 Although APOE modulates multiple functions throughout the body, increasing evidence suggests that APOE impacts neurovascular function. Compared with APOE3, markers of neurovascular dysfunction are greater with APOE4 including higher permeability, lower vessel coverage, and altered cerebral blood flow with aging and neurodegeneration in humans and mouse models.^3–12 Therefore, identifying fundamental mechanisms by which APOE regulates neurovascular function is important for ultimately understanding neurodegenerative disorders.

Endothelial cells comprising the neurovasculature are highly specialized compared with other vascular beds. The unique properties of brain endothelial cells selectively control bidirectional movement of molecules, regulate inflammation, and control cerebral blood flow. Ongoing research into the effects of APOE genotype on the neurovasculature has primarily focused on how APOE expression by astrocytes, pericytes, and macrophages affects brain endothelial cells.^13–16 Although these studies have identified novel mechanisms by which APOE impacts neurovascular function, the contribution of APOE expressed by brain endothelial cells has been potentially overlooked.

The physiological role played by endothelial APOE in neurovascular function is unclear and controversial. Human and in vivo research is limited to expression data, and although studies have shown endothelial cells express APOE, levels are varied, and functionality is debated.^17–23 In vitro models have also produced conflicting results on whether there are functional differences comparing APOE3 and APOE4 brain endothelial cells.^16,24,25 Therefore, at present, it is still unknown whether brain endothelial cell APOE modulates neurovascular function. Addressing this issue is crucial to advance our fundamental knowledge of APOE biology and its impact on the neurovasculature.

The goal of this study was to determine whether endothelial APOE3 regulates neural function in vivo. To achieve this goal, we addressed whether the loss of endothelial cell APOE3 impacts brain function. APOE3 is used as the control genotype in APOE research. We reasoned that if endothelial cell APOE was functional, then we would be most likely to find changes after APOE3 knockdown, without the confounds of APOE4 expression by other cells. Therefore, we created a mouse model to enable the conditional knockdown of APOE3 in endothelial cells and measured neurovascular, behavioral, and neuronal function.

METHODS

The authors declare that all supporting data are available within the article (and its Supplemental Material).

Mouse Model and General Design

All experiments were approved by the Institutional Animal Care and Use Committee at the University of Illinois at Chicago. Mice were group housed on a 12-hour light/dark cycle with food (Teklad 7912 from Inotiv) and water ad libitum. The overall design is presented in Figure S1. APOE3^fl/fl/Cdh5(PAC)-CreERT2 mice were developed via MTA (materials transfer agreement) with Cure Alzheimer’s Fund (to obtain APOE3^fl/fl mice) and contract with Taconic. APOE3^fl/fl/Cdh5(PAC)-CreERT2 mice were generated by crossing APOE3^fl/fl mice, which express floxed human APOE3 under the endogenous mouse APOE promoter,²⁶ with Cdh5(PAC)-CreERT2 mice that express CreERT2 under the control of the cadherin-5/VE-cadherin endothelial cell–specific promoter.²⁷ We refer to APOE3^fl/fl/Cdh5(PAC)-CreERT2 mice as APOE3^Cre+/− mice and control mice as APOE3^Cre−/− mice. For APOE knockdown, 4- to 5-week-old APOE3^Cre+/− and APOE3^Cre−/− mice were injected with 100 µL of tamoxifen (ip 20 mg/mL in corn oil) every 24 hours for 5 days. Experiments were conducted at 9 months of age.

Mice were weaned at P21, before determining the Cre genotype; this provided a natural randomization of genotypes in each cage. For cohort assignment, cages containing mice of similar age were selected such that the number of APOE3^Cre+/− and APOE3^Cre−/− mice per group were near equal in number. Cages contained 4 to 5 mice at the start of housing, which continued unless there was premature loss of mice due to natural causes or fighting before the start of experiments. For quantification of all experiments, investigators were blinded to the Cre genotype. Behavior was conducted in the mouse active/dark cycle (except trace fear conditioning) following our previously described protocols^11,28,29 with slight modifications and analyzed using the ANY-maze video tracking software (Stoelting Co) unless stated otherwise. Mice were placed in the room for 30 minutes before behavior tests for acclimatization, except for trace fear conditioning.

Open Field

Mice were placed in an opaque, white arena (46×46×40 cm; Ugo Basile) for 10 minutes, and the total distance traveled was measured.

Novel Object Recognition

Novel object recognition was conducted 24 hours after open field, using an opaque, white arena (46×46×40 cm; Ugo Basile) and 3-dimensional printed novel objects that were magnetically attached to the arena floor. Mice were placed in the arena for 7 minutes with 2 familiar objects (red shot glass shaped plastic) and then back in the arena 1 hour later for 7 minutes with a familiar (red shot glass shaped plastic) and novel (grey flag shaped plastic) object. Total investigation time and preference index (time spend with novel object/total investigation time) were calculated by 4 investigators.

Morris Water Maze

Morris water maze (MWM) was performed 2 days after novel object recognition in a white-walled pool, 120 cm in diameter, containing a circular escape platform (10 cm in diameter) submerged 7 mm below the surface of the water. Water was maintained at a temperature of 20±1 °C and made opaque using nontoxic white paint. High-contrast visual cues were affixed to the top of the pool walls for spatial orientation (3-dimensional printed black plastic circle, square, cross, and triangle). MWM consisted of visual acuity, acquisition, and probe trial phases. Visual acuity was performed in a nonspatial context for habituation and to ensure no visual impairment. A red plastic flag was placed on the platform in the center of the MWM pool without spatial cues. Two trials of 60 s were performed with entry positions different from any of those used in the acquisition or probe trial phases. All mice found the platform within the 2 trials (observational data). Two days later, the spatial learning acquisition phase was conducted, where mice were trained for 5 days to locate the platform, with each day consisting of 4 trials separated by a 25- to 30-minute inter-trial interval. Trials ended either when the mice found the platform or 60 s had elapsed; mice that did not find the platform were guided by following the investigator’s hand. Mice were left on the platform for 15 s or until they had performed at least 1 full investigative rotation on the platform. The platform location remained constant, and starting location changed with each trial; mice were never placed in the quadrant containing the platform or starting point of the probe trial. The time to locate the submerged platform from each of the 4 trials was averaged and plotted for each day. The platform was placed in the SW quadrant and the mouse entry order was N, E, SE, NW (day 1); SE, N, NW, E (day 2); NW, SE, E, N (day 3); E, NW, N, SE (day 4); and N, SE, E, NW (day 5). To test memory retention, the platform was removed 24 hours after the final acquisition trial, and mice were placed in the NE quadrant and allowed to swim in the arena for 60 s. The number of entries and time spent in the previous platform area were measured.

Trace Fear Conditioning

Fear conditioning was conducted over 3 days.^30,31 The conditioning phase consisted of 2-minute habituation, followed by 5 paired trails of a neutral tone (10 s, 1500 Hz, 85 dB) with a foot shock (1 s, 0.4 mA) at a delay of 20 s from the end of the tone, with an inter-trial interval of 220 to 280 s. Cue-induced fear extinction was evaluated 24 hours after conditioning (day 2) by measuring the freezing response to the conditioned tone over 15 trials at 60-s interval and then again 24 h later (day 3). On days 2 and 3, the average freezing response over 3 trials was calculated and used for statistical analysis.

In Vitro Brain Endothelial Cell Isolation

Brain endothelial cells were isolated from 9-month-old mice as described by Marottoli et al²⁴ with slight modification. Briefly, cortical tissue was diced, centrifuged (1000g, 5 minutes, 4 °C), resuspended in papain (20 U/mL)/DNase (2000 U/mL), and lightly triturated through a 19G needle. Brain homogenates were then incubated for exactly 15 minutes at 37 °C, triturated with a 21G needle, mixed thoroughly with 2 mL of 25% BSA per cortex, vortexed, and centrifuged (4000g, 5 minutes, 4 °C) to separate out the myelin. The supernatant was collected, vortexed, and centrifuged a second time (4000g, 5 minutes, 4 °C). Pellets from both centrifugations were combined in complete growth media (EBM-2 containing EGM-2 MV Microvascular Endothelial SingleQuots and 5.5 U/mL heparin), passed through a 100-µm cell strainer and pelleted (1000g, 5 minutes, 4 °C). Resuspended cells were plated at ≈1 cortex (ie, 2 hemispheres) to 1.32 cm² (ie, 4 wells of a 96-well plate) on a growth matrix of fibronectin, collagen, and laminin and placed at 37 °C with 5% CO₂. The following morning, cells were incubated in 8 μg/mL puromycin for 48 hours to negatively select for brain endothelial cells. Puromycin was removed and brain endothelial cells were grown to confluence (1 week) in complete growth media. All experiments were conducted 48 hours after the final media change at passage 0.

In Vivo Brain Endothelial Cell Isolation

For in vivo applications, the above procedure was modified to enrich for a single-cell suspension of brain endothelial cells without the need for culture. Cortices were homogenized and myelin was removed as above. Following the myelin separation, the pellets were combined and incubated in ACK buffer for 5 minutes followed by a 1-hour digestion in collagenase/dispase (Roche) and passed through a 70-µm strainer to produce a single-cell suspension. Samples were enriched for brain endothelial cells by depletion of remaining myelin debris with myelin separation beads and 3 rounds of positive selection with CD31 beads on a MulitMACS Cell24 Separator Plus. A fraction of the brain endothelial cell–enriched sample (15%) was assessed by flow cytometry. The remaining sample was pelleted for extraction of protein or RNA (see qRT-PCR). For protein extraction, pellets were lysed in 70 μL of RIPA buffer+1× HALT protease inhibitor cocktail, sonicated (20% amplification, 3 cycles), centrifuged (25 000g for 20 minutes at 4 °C), and supernatant was collected. Protein was not quantified as the BSA in the elution buffers masked the true protein levels of the samples. Experiments that typically require normalization to protein were instead normalized to actin values measured by Western blot.

Flow Cytometry

Brain endothelial cells were incubated in Zombie Green viability dye (1:200) for 15 minutes at room temperature and washed with 1% BSA/PBS. Fc receptors were blocked for 10 minutes with TruStain FcX PLUS (1:200). Cells were then incubated with CD31-PE (1:100) for 1 hour at room temperature, washed with 1% BSA/PBS, and fixed overnight at 4 °C in True-Nuclear 1X Fix. The following morning, cells were washed and resuspended in 1% BSA/PBS and measured on a CytoFLEX S. Analysis was performed using Kaluza software.

qRT-PCR

qRT-PCR and bioinformatics analysis was performed by the Genomics Research Core and Research Informatics Core at the University of Illinois at Chicago. RNA was extracted from brain endothelial cell–enriched samples with a miRNeasy Micro Kit (Qiagen) according to the manufacturer’s protocol. RNA samples were quantified with a Qubit RNA HS Assay Kit (Invitrogen) and evaluated for quality using an RNA ScreenTape Assay and Agilent 4200 TapeStation (Agilent Technologies). All steps of RNA processing were performed according to the manufacturer’s protocol “Gene Expression Analysis with the GE 96.96 IFC Using TaqMan Assays.” TaqMan gene expression assays (Actb, APOE, Cldn5, Ocln, Cdh5, Ctnnb1, Ctnnd1, Ccl3, Cd74, Cxcl1, B2m, Vcam1, and Vwf) were ordered from Life Technologies (Thermo Fisher Scientific [Major Resources Table]). In brief, RNA in the amount of 0.5 to 5 ng per sample was converted to cDNA and followed by target-specific amplification preamplification performed with a master mix of all intended assays according to the protocol “Gene Expression Preamplification with Fluidigm Preamp Master Mix and TaqMan Assay” (P/N 100-5876 C2). All samples were preamplified with 18 polymerase chain reaction cycles. Final cDNA products were diluted 1:1 and 1:5, and both dilutions were used for setting up qRT-PCR reactions with each individual assay. Each sample was tested in 3 technical replicates. The setup included no template controls. Reaction setup was performed according to the protocol “Gene Expression with the 96.96 IFC Using Fast TaqMan Assays” (P/N 100-2638 E1). After the completion of the run, IFC was checked for the quality of loading, consistency of the technical replications, and absence of the amplification signal in no temple control wells. Expression profiling was performed using the 96.96 Gene Expression Dynamic Array IFC, and data were collected using a BioMark HD real-time PCR system (Standard Biotools). The Fluidigm Real-Time PCR Analysis software was used to collect the data.

Raw data (CT values) were processed in R. Briefly, raw CT values for each assay were filtered to remove any assays in which the quality score was <0.65. The mean CT value for each gene for each sample was computed from the retained values. Normalized, that is, ΔCT, values were computed using Actb as the reference assay for each sample. Differential analysis of the normalized (ΔCT) values was performed using limma (v3.50.3) in R.³² Before differential analysis, genes were removed in which >12 samples did not have a detectable signal. Furthermore, normalized values were multiplied by −1 to create log scaled values required by limma. The transformed, normalized values were modeled as a function of the experimental covariates, that is, sex and genotype, with batch as a controlling factor as samples were collected in 3 batches of 8 that contained 2 of each group. Specific contrasts of the experimental covariates were then tested using empirical Bayes moderated t statistic (limma R package v3.50.3), that is, empirical Bayes function. For graphical representation, ΔΔCT was calculated for each sample using ΔCT (sample)−ΔCT (control average) using either male APOE3^Cre−/−, APOE3^Cre−/− or male mice as the control group average as supported by statistical analysis. Fold gene expression was then calculated as 2^−ΔΔCT.

Cortical Harvest

Mice were anesthetized via intraperitoneal injection with ketamine (100 mg/kg) and xylazine (10 mg/kg). For plasma isolation, blood was drawn by cardiac puncture of the right ventricle followed by centrifugation at 1500g for 20 minutes at 4 °C. Plasma was flash frozen in liquid nitrogen and stored at −80 °C. Following blood draw, transcardial perfusion was performed with ice-cold TBS at a rate of 4 mL/min for 4 minutes. For immunohistochemical analysis, left hemi brains were frozen in OCT and stored at −80 °C until processing. For biochemical analysis, right hemicortices were isolated and homogenized using a bead mill (Fisherbrand) at 6 m/s for 1 cycle of 30 s in SDS lysis buffer (1% SDS+10 mmol/L NaF+1× HALT protease inhibitor cocktail in 20 mmol/L HEPES; pH 7.4). Samples were then centrifuged at 500g for 5 minutes at 4 °C, sonicated (20% amplification, 3 cycles), and then centrifuged again (25 000g for 20 minutes at 4 °C). Aliquots of the resulting supernatant were then flash frozen in liquid nitrogen and stored at −80 °C. Total protein was quantified using the Pierce BCA Protein Assay Kit. Brains from a separate cohort of mice were drop-fixed overnight in 4% paraformaldehyde (PFA), cryopreserved in 30% sucrose, sectioned, and then stored at 4 °C in cryoprotectant until processing.

Western Blot

Gel electrophoresis was used to separate either 10.5 μL of brain endothelial cell lysate (total sample volume of 15 μL) or 20 μg of protein from cortical lysates on 26-well 4% to 12% Bis-Tris Midi gels (Invitrogen). Proteins were transferred onto low-fluorescence PVDF membranes, which were cut, blocked with 5% milk in TBS, and probed with primary antibody (list of antibody and dilutions provided in the Major Resources Table) in 1% BSA in TBS with 0.02% sodium azide overnight at 4 °C. After washing (3×5 minutes, 0.1% Tween 20 in TBS), membranes were incubated for 45 minutes in the appropriate secondary fluorescent secondary antibody (LI-COR) in 1% milk in 0.1% Tween 20 in TBS and 0.01% SDS. Proteins were imaged and quantified using the Odyssey Fc Imaging System. When required, blots were reprobed for a second antigen. To ensure signal specificity in reprobing, all antibodies were paired so there was no overlap between signals, and, where possible, primary antibodies from another host species were paired with secondary antibodies conjugated to a different fluorophore. Optical densities were quantified with Image Studio Lite v5.2 and normalized to β-actin.

The nature of our study required us to evaluate the contribution APOE3^Cre genotype, sex, and their interaction to each readout. For our study, this is difficult as all of the samples from every group will not fit on a single gel (4 groups: 2×APOE^Cre; APOE3^Cre−/− and APOE3^Cre+/− ×2 sex; male and female, n≈12). Furthermore, running a portion of each group across multiple gels can result in variability across blots from sample loading, transferring, probing, etc. Therefore, dividing our groups evenly across multiple gels could potentially result in artificial alterations in data such as bimodal populations, false positives, and false negatives. To compensate for this variability, and to enable a reliable evaluation across all groups, cortical samples were run across 3 different Western blots to calculate a correction factor/ratio to compare all samples at once. This correction factor allows us to express all data relative to male APOE3^Cre−/− mice and, therefore, analyze the contributions of both Cre genotype and sex to changes in protein levels analyzed by the Western blot. In the first blot, male and female APOE3^Cre−/− mice were analyzed. All data were normalized to the average of male APOE3^Cre−/− mice and the normalized ratio of female APOE3^Cre−/− mice:male APOE3^Cre−/− mice was used as the female correction factor for subsequent analysis. In the second blot, female APOE3^Cre−/− mice were run with female APOE3^Cre+/− mice. All values were normalized to the average of female APOE3^Cre−/− mice and then multiplied by the female correction factor. In the third blot, all male APOE3^Cre−/− mice were run with male APOE3^Cre+/− mice and values were normalized to the average of male APOE3^Cre−/− mice. Normalized data from blots 2 and 3 were then used for plotting and statistical analysis. The following is a summary of our normalization approach on actin-normalized data: (1) on blot 1, divide each value by the average of the APOE3^Cre−/− male values to normalize all values to APOE3^Cre−/− male mice; (2) average the normalized APOE3^Cre−/− female values to determine the female correction factor; (3) on blot 2, divide each value by the average of the APOE3^Cre−/− values to normalize all values to APOE3^Cre−/− mice; (4) multiply the normalized values from blot 2 by the female correction factor; (5) repeat step 3 on blot 3. Both blots 2 (female) and 3 (male) are now normalized to their respective APOE3^Cre−/− group; (6) all data are now normalized to the APOE3^Cre−/− male group for statistical analysis.

Immunohistochemistry

CD31, Laminin, IgG, and Fibrinogen

Fluorescent immunohistochemistry for CD31, laminin, IgG, and fibrinogen was performed on fresh frozen samples as described previously^11,28 with slight modifications. To summarize, 9×12-µm nonadjacent sagittal sections per animal were fixed with 3% PFA in TBS for 10 minutes. Sections were then permeabilized with 0.25% Triton X-100 in TBS (dilution media), blocked in 5% BSA in dilution media, and incubated in primary antibodies diluted in 2% BSA and 0.1% Triton X-100 in TBS (list of antibodies and dilutions provided in the Major Resources Table) overnight at 4 °C in a humidified chamber. Following the primary incubation, sections were washed 3×5 minutes in dilution media and incubated in the appropriate Alexa Fluor–conjugated secondary antibodies in 2% BSA and 0.1% Triton X-100 in TBS, followed by 3×5-minute washes in dilution media, and 1×5-minute wash in TBS. Following immunohistochemical staining, sections were coverslipped and imaged at ×10 magnification on an ImageXpress Micro (Molecular Devices) under identical exposure settings and equally thresholded. The area fraction of CD31 and laminin was quantified in the full cortex in ImageJ (Fiji) software.

GLUT1 and PDGFRβ

To expand upon our vascular analysis, drop-fixed brains were sectioned at 40 µm on a sliding microtome, and immunohistochemistry was performed according to the above procedure on free-floating sections. Following immunohistochemical staining, sections were mounted, coverslipped, and 10 images per animal were acquired at ×25 magnification on a Zeiss LSM710 confocal microscope. An established script³³ was used in ImageJ (Fiji) software to measure area fraction of vessels and pericytes, vessel segment length, vascular branching, and distance between blood vessels. Briefly, images were processed to remove noise and binarized. After measuring the area fraction, binarized images were skeletonized for quantification of total length, number of branches, and nearest neighbor distance. To quantify pericyte coverage, the pericyte area fraction was divided by the vessel area fraction.

ApoE

Immunohistochemical detection of apoE was performed as described previously³⁴ with slight modification. Sections were fixed with 3% PFA in PBS overnight at 4 °C, washed (2×5 minutes in PBS), permeabilized for 30 minutes (0.1% Triton X-100 in PBS), and background PFA quenched (0.05 M ammonium chloride, 10 minutes). Sections were then rinsed (1×PBS), blocked (4% BSA in PBS, 30 minutes), rinsed (1×PBS), incubated in primary antibodies overnight (4 °C, humidified chamber, 1% BSA in PBS), washed (3×5 minutes in PBS), incubated with Alexa Fluor–conjugated secondary antibodies (2 hours, room temperature), washed (3×5 minutes in PBS), and coverslipped. Slides were viewed on a Zeiss LSM710 confocal microscope at ×200.

In Situ Hybridization

APOE mRNA was measured in isolated brain endothelial cells and brain tissue sections using BaseScope Detection Reagent Kit v2-Red (ACD) following the manufacturers protocol and as described by Marottoli et al²⁴ with minor modifications. Briefly, samples were fixed in 3% PFA (30 minutes, 4 °C), washed (PBS), incubated with hydrogen peroxide (10 minutes, room temperature), washed (3×PBS), treated with RNAscope protease III (diluted 1:15, cultured brain endothelial cells, 10 minutes) or protease IV (tissue sections, 30 minutes) solution, and washed (2×PBS). Probe hybridization and amplification were performed according to manufacturer instructions except PBS was used in place of distilled water for the final washes. Isolated brain endothelial cells were then stained for ZO1 (zonula occludens-1) and tissue sections for FITC-lectin and then counterstained with DAPI.

ELISA

ApoE was measured by ELISA in brain endothelial cell–enriched lysates (1:8 dilution), cortical lysates (1:160 dilution), and plasma (1:20 000 dilution) as described in previous studies^11,28 using anti-apoE (1:2000; Millipore) and biotinylated anti-apoE (1:5000; Meridian) for capture and detection antibodies, respectively. Cortical lysates were analyzed for fibrinogen (1:3 dilution), IgG (1:3 dilution), and albumin (1:100 dilution) with commercial ELISA kits (Immunology Consultants) with 2 minor modifications to the manufacturer’s protocol; sample incubation time was increased to 2 hours at 37 °C, and secondary antibody was increased to 1 hour at 37 °C, and standards were buffer matched. Cortical samples were normalized to protein concentration, and isolated brain endothelial cell samples were normalized to actin quantified by the Western blot. Plasma levels (1:10 dilution) of cardiovascular dysfunction were measured using MILLIPLEX MAP CVD magnetic bead panel according to the manufacturer’s protocol.

Dextran Extravasation

Dextran extravasation was performed as described previously^35,36 with modification. Mice were injected intraperitoneally with 100 μL of fluorescently conjugated 10-kDa (Alexa Fluor-555; Invitrogen) and 70-kDa (Texas Red; Invitrogen) dextran tracers at a concentration of 3.75 mg/mL and returned to their home cages for 30 minutes to allow the dextrans to circulate. Animals were then transcardially perfused with TBS. Cortices were dissected, weighed, and incubated in formamide for 48 hours at 56 °C. Dye fluorescence was then measured using a SpectraMax i3× Multi-Mode Plate Reader (Molecular Devices) at the appropriate emission and excitation wavelength and normalized to tissue weight.

Cerebral Blood Flow

To quantify cerebral blood flow, endogenous arterial spin labeling magnetic resonance imaging based on flow-sensitive alternating inversion recovery was performed on a Varian Horizon 9.4T small-animal magnetic resonance imaging (Agilent Technologies, Santa Clara, CA) to acquire 2 interleaved inversion recovery data sets, a global select and a slab select, with 85 inversion times varying from 1 ms to 12 s. The slab thickness was 8× of the slice thickness (1 mm). Images were acquired with magnetization prepared steady-state free precession (FID-SSFP) pulse sequence using center-out k-space order with variable flip angles to start acquisition in a pseudo-steady state, with an FOV of 18×18 mm², 1 mm slice thickness, imaging TR/TE of 1.9/1 ms, a per inversion time TR of 10 s, and an acquisition time of 38 minutes. To reduce respiratory effects on the images, both the inversion pulse and the image acquisition were respiratory triggered. The actual inversion times were recorded using an Arduino Nano connected to the system TTL output of the scanner and used in the curve fitting.

The complex inversion recovery data (Sss, Sgs) were phase corrected to bipolar real values and fitted pixel by pixel for inversion recovery via SIRTI=M01-2αe-TIT1app. M0 is the signal of tissue at the fully relaxed equilibrium, α is the inversion efficiency, and T1_app is the apparent spin-lattice relaxation time T₁ of tissue (including blood).

The blood flow contrast data, Sss−Sgs, were fitted using SBFTI=SssTI-SgsTI=M0BW1-e-TI-t0/τe-TIT1BW. M0_BW is the signal of blood water at the fully relaxed equilibrium, t₀ is the initial arrival time of the blood, τ is the voxel filling time rate, and T1_BW is the T1 of the blood water. The blood flow cerebral blood flow=λ·M0_bw/M0 (mL blood per g of tissue, where λ is the tissue-blood partition coefficient, 0.9 [g of water per g of tissue]/[g of water per mL of blood]). The cortex was manually drawn in reference to T₂-weighted magnetic resonance images. The mean signal within each ROI was used during regression.

Electrophysiological Recordings

Ex vivo whole-cell patch-clamp recordings from layer V pyramidal neurons of the medial prefrontal cortex (PFC; prelimbic region) were conducted as described in previous studies.^30,31 Briefly, 350-µm PFC slices were obtained using a vibrating blade microtome (PELCO, Ted Pella, CA) in ice-cold artificial cerebrospinal fluid (95% O_2%–5% CO₂). PFC slices were then incubated in warm artificial cerebrospinal fluid (33–35 °C) with constant oxygenation (95% O₂–5% CO₂) for at least 60 minutes before transferring them to the recording chamber. All recordings were conducted at 33 to 35 °C using a low chloride–based internal solution and an external solution free of glutamate and GABA (gamma-aminobutryic acid) blockers to enable concurrent acquisition of excitatory and inhibitory postsynaptic currents (PSCs) at the single-cell level.³⁷ The internal solution also contained 0.1% neurobiotin (Vector Laboratories, CA) and (in mM): 10 CsCl, 130 Cs-gluconate, 10 HEPES, 2 MgCl2, 5 NaATP, 0.6 NaGTP, and 3 QX-314 (pH 7.23–7.28, 280–282 mOsm). Both spontaneous excitatory and GABA-inhibitory synaptic events can be readily assessed by recording the frequency of PSC at the −60-mV (PSC_−60mV) and +15-mV (PSC_+15mV) holding potentials, respectively. Only neurons with at least 10 minutes of stable baseline activity were included for analyses. Changes in the frequency of PSC_−60mV and PSC_+15mV events across the different experimental groups were determined from at least 2 noncontiguous epochs of 60 s each.

Statistical Analysis

All data are expressed as either the mean±SEM or as a box plot depicting the minimum score, the lower quartile (25%), the median (50%, horizontal line), the upper quartile (75%), maximum values, and mean (+). Due to the nature and feasibility of in vivo research, we considered the n value too low to give an accurate representation of normality or the variance between our groups. As our primary question was related to the mean differences, data were analyzed with general linear models in SPSS (v28.0.1.1; 15) with Bonferroni post hoc as required. For all data except behavior, the general linear models report between-subject significance for Cre, sex, and Cre×sex. For the acquisition phase of MWM and for trace fear conditioning, we included trial as a repeated measure in the general linear model, which, therefore, reports significance for within-subject (trial, trial×sex, trial×Cre, trial×Cre×sex using Pilai trace, Wilks lamdba, Hotelling trace, or Roy largest root) and between-subject effects (Cre, sex, and Cre×sex). Most data in this article were significant due to a Cre effect or a sex effect with no interaction. To visually represent the results of that statistical analysis, we, therefore, plotted data as either Cre genotype (our main comparison of interest, Figures 1 through 4) or sex (Figures S1 through S10) due to the absence of a Cre×sex interaction. Cre×sex interactions were found for data presented in Figures S3 and S6. We also conducted Pearson correlation coefficient and linear regression analysis in Figure 4C. All n sizes, data points, general linear model details, and statistical analyses are available in Source Data File 1 in the Supplemental Material. Statistical analysis for qRT-PCR is described above.

Figure 1.

Brain endothelial cell APOE expression and apoE levels are lower in APOE3^Cre+/− mice. A, In situ hybridization of APOE (red) with immunostaining for ZO1 (zonula occludens-1; green) in cultured brain endothelial cells at passage 0. No APOE3 was found in brain endothelial cells isolated from APOE3^Cre+/− mice. Scale bar, 10 µm. B, ApoE measured in the media from cultured brain endothelial cells (ELISA). ApoE levels were at the limit of detection for APOE3^Cre+/− mice, n=6. C, Confocal images of APOE (red, in situ hybridization) with a FITC-lectin stain for endothelial cells (green) and DAPI nuclear stain (blue). In APOE3^Cre−/− mice, there is APOE expression (white arrowheads) in and around brain endothelial nuclei that is absent in APOE3^Cre+/− mice. Scale bar, 5 µm. Images representative from 4 mice per Cre genotype. D, qRT-PCR for APOE in brain endothelial cells. APOE3 levels are substantially lower in brain endothelial cells from APOE3^Cre+/− mice compared with APOE3^Cre−/− mice (Cre genotype P=0.008 using empirical Bayes method), n=11/12 per Cre genotype. Data are presented as 2^−ΔΔCT using the average of APOE3^Cre−/− mice for calculations. E, Immunostaining of apoE (red), CD31 (green), and DAPI in the cortex of APOE3^Cre+/− and APOE3^Cre−/− mice. There are reduced levels of apoE (white arrowheads) in brain endothelial cells of APOE3^Cre+/− mice. Scale bar, 5 µm. Images representative from 4 mice per Cre genotype. F and G, ApoE levels in isolated brain endothelial cells. Brain endothelial cells demonstrated 40.66% lower ApoE levels by (F) ELISA (F_(1,15)=7.16; P=0.017), n≈5 and 86.19% lower levels by (G) Western blot (F_(1,15)=50.76; P<0.001), n≈5. Differences are due to antibody specificity (Figure S2). H and I, Cortical and plasma levels of apoE measured by ELISA. There were no effects of endothelial cell APOE3 knockdown on (H) cortical (F_(1,36)=2.08; P=0.16), n=10, or (I) plasma (F_(1,36)=1.33; P=0.26), n=10, apoE levels. All data are expressed as a box plot depicting the minimum score, the lower quartile (25%), the median (50%, horizontal line), the upper quartile (75%), maximum values, and mean (+). Data were analyzed using general linear models. D was analyzed using the Empirical Bayes method. *P<0.05 for APOE3^Cre+/− vs APOE3^Cre−/− mice.

Figure 4.

Endothelial cell APOE3 knockdown disrupts and preferentially reduces GABAergic synaptic events in APOE3^Cre+/− mice. A, Ex vivo recordings of spontaneous postsynaptic current (PSC) from excitatory (−60 mV) and inhibitory (+15 mV) synapses onto layer V pyramidal neurons from the prefrontal cortex (n=10–14 neurons per Cre group). The frequency of inhibitory events (PSC_+15mV) was lower in APOE3^Cre+/− mice compared with APOE3^Cre−/− mice (F_(1,21)=62.41; P<0.001). Such a difference was not apparent when neurons were held at −60 mV (PSC_−60mV; F_(1,21)=0.051; P=0.82), indicating that the level of excitatory synaptic events was not affected by the endothelial cell APOE3 knockdown. B, Summary of the excitatory/inhibitory (E/I) ratio (PSC_−60mV/PSC_+15mV) analyses. Relative to APOE3^Cre−/− mice, APOE3^Cre+/− mice exhibited an imbalanced (>1) E/I ratio (F_(1,21)=49.31; P<0.001) due to the lower inhibitory activity. C, The frequency of PSC_+15mV (inhibitory component) and the E/I ratio were correlated when evaluated by Pearson correlation coefficient (P<0.0001; r²=0.7649). D, Traces of spontaneous postsynaptic events recorded at both −60 and +15 mV holding potentials (scale bar, 40 pA/1 s). E, Levels of synaptic proteins in cortical lysates from APOE3^Cre+/− mice (protein levels normalized to actin and then to APOE3^Cre−/− mice). There were no effects of Cre genotype on general (left), glutamatergic (middle), or GABAergic (right) neuronal markers (Source Data File 1 in the Supplemental Material). There was an effect of sex on some synaptic markers (Figure S9). n=24 per Cre genotype. All data are expressed as a box plot depicting the minimum score, the lower quartile (25%), the median (50%, horizontal line), the upper quartile (75%), maximum values, and mean (+). Data were analyzed using general linear models (A and E through G). *P<0.05 for APOE3^Cre+/− vs APOE3^Cre−/− mice.

RESULTS

Whether APOE expression by brain endothelial cells plays a role in regulating brain function remains controversial. To gain insight on this issue, we evaluated the extent that endothelial APOE3 knockdown impacts neurovascular and neural function. To achieve our goal, humanized APOE3^fl/fl mice obtained from the Cure Alzheimer’s Fund were used to develop APOE3^fl/fl/Cdh5(PAC)-CreERT2^+/− (APOE3^Cre+/−) and APOE3^fl/fl/Cdh5(PAC)-CreERT2^−/− (APOE3^Cre−/−, control) mice, which enable tamoxifen inducible knockdown of endothelial cell APOE3. We treated mice with tamoxifen at ≈4 to 5 weeks of age and evaluated brain function at 9 months (see Figure S1 for study design). Although the goal was not the comparison of APOE3 and APOE4, 9 months was selected based on our previous studies on the role of human APOE (hAPOE) in neurovascular and neural function during aging in vivo. We found deficits in neurovascular function and cognition at 9 months of age with APOE4 compared with APOE3.^9,11,29 Therefore, we predicted that if endothelial APOE3 is functional, then APOE3 knockdown in endothelial cells may produce a phenotype by 9 months of age.

Brain Endothelial Cell ApoE3 Levels Are Lower in APOE3^Cre+/− Mice

The extent that APOE (gene) is expressed by and contributes to brain endothelial cell apoE (protein) levels is highly contested. We first validated APOE3 expression in endothelial cells cultured from APOE3^Cre−/− mice and its knockdown in APOE3^Cre+/− mice (Figure 1A and 1B). Due to the controversy, we next focused on evaluating levels of APOE (Figure 1C and 1D) and apoE (Figure 1E through 1G) in vivo. When assessed by in situ hybridization in APOE3^Cre−/− mice, we found perinuclear expression of APOE3 in brain endothelial cells, which was greatly reduced in APOE3^Cre+/− mice (Figure 1C). To quantify APOE levels, we positively selected brain endothelial cells from cortical homogenates (≈97% enriched for brain endothelial cells; Figure S2). Consistent with our in situ results, APOE transcripts were low in brain endothelial cell–enriched samples from APOE3^Cre+/− mice compared with APOE3^Cre−/− mice when measured by qRT-PCR (Figure 1D). We next confirmed that APOE knockdown resulted in lower apoE levels in brain endothelial cells. Immunohistochemical analysis showed apoE located within the brain endothelial cells, which was reduced in APOE3^Cre+/− mice (Figure 1E). We also found ≈41% and ≈86% lower apoE levels by ELISA (Figure 1F) and Western blot (Figure 1G), respectively, in brain endothelial cell–enriched samples. The difference in the relative lowering of apoE levels in brain endothelial cells between the 2 techniques is most likely due to the nonspecific protein binding of the apoE capture antibody (Figure S2). Lower APOE3 expression in brain endothelial cells did not affect overall cortical (Figure 1H; Figure S2) or plasma (Figure 1I) apoE levels. Overall, our data provide evidence that brain endothelial cells do express APOE3, which was knocked down in APOE3^Cre+/− mice and resulted in lower levels of apoE.

Endothelial Knockdown of APOE3 Is Sufficient to Cause Neurovascular Dysfunction

As APOE is implicated in regulation of the neurovasculature,¹⁰ we evaluated the functional impact of endothelial APOE3 knockdown on common markers of neurovascular dysfunction: higher permeability, lower vessel coverage, and altered cerebral blood flow.

Higher permeability to plasma proteins is a prominent marker of neurovascular dysfunction. We began by measuring extravasated plasma proteins in the cortex by ELISA. We found that cortical levels of both IgG and fibrinogen were ≈29% higher in APOE3^Cre+/− mice (Figure 2A and 2B; Figure S3) while levels of albumin were unchanged (Figure S3). Albumin transport at endothelial cells is complex and can involve both transcellular and paracellular routes,^38–42 potentially in a bidirectional manner.⁴⁰ Therefore, the lack of an effect on albumin extravasation in APOE3^Cre+/− mice may be related to changes in these transport processes in a way to compensate for higher paracellular permeability after endothelial cell APOE3 knockdown. Evaluation of neurovascular permeability by ELISA reflects the physiological accumulation of plasma proteins over 9 months. To determine neurovascular permeability at 9 months of age, we measured cortical extravasation of exogenously administered fluorescent dextrans. For both 10- and 70-kDa dextrans, there was a sex effect with females showing higher dextran extravasation than males (Figure S3). In addition, there was an interaction between Cre genotype and sex, driven by higher permeability with APOE3 knockdown in males (10 kDa dextran and trending for 70 kDa dextran). In female mice, with endothelial cell knockdown of APOE, the higher accumulation of endogenous plasma proteins but no effect on exogenously added dextran may be due to a combination of the technique and age. Plasma protein levels reflect accumulation over time, whereas dextran permeability represents a single timepoint. Thus, it is possible age-dependent increases in permeability in female mice resulted in comparable dextran extravasation between APOE3^Cre−/− and APOE3^Cre+/− at 9 months. Tight junctions between brain endothelial cells are crucial for maintaining the integrity and selective permeability of the neurovasculature.^43–45 Therefore, we next evaluated whether differences in levels of junctional proteins were associated with the increased cortical extravasation of IgG and fibrinogen in APOE3^Cre+/− mice. At the mRNA level, we did not find any differences between APOE3^Cre−/− and APOE3^Cre+/− mice for tight junction (Ocln and Cldn5) and adherens junction (Cdh5, Ctnnb1, and Ctnnd1) transcripts (Figure 2C; Figure S4). However, at the protein level and consistent with higher neurovascular permeability, APOE3 knockdown resulted in lower levels of the tight junction protein claudin-5 when assessed by Western blot (Figure 2C; Figure S4), which may suggest greater degradation. Therefore, endothelial cell APOE3 knockdown resulted in higher markers of neurovascular permeability and lower levels of claudin-5.

Figure 2.

Endothelial cell APOE3 knockdown disrupts neurovascular function in APOE3^Cre+/− mice. A and B, Levels of IgG and fibrinogen in cortical homogenates (ELISA). There was an effect of Cre genotype on extravascular (A) IgG (F_(1,51)=4.9; P=0.031) and (B) fibrinogen (F_(1,53)=4.5; P=0.038) levels, n≈29 per Cre genotype. Representative images highlight the higher levels of IgG and fibrinogen in APOE3^Cre+/− mice. There was also an effect of sex for IgG (Figure S3). Scale bar, 10 µm. C, qRT-PCR for Cldn5 in brain endothelial cells. There were no effects of Cre on Cldn5 levels (Cre genotype P=0.7 using empirical Bayes method), n=11/12 per Cre genotype. Data presented as 2^−ΔΔCT using the average of APOE3^Cre−/− mice for calculations. D, Cortical levels of claudin-5 were lower in APOE3^Cre+/− mice when measured by Western blot (F_(1,42)=5.32; P=0.026). n=24 per Cre genotype. E and F, Vessel area fraction quantified from laminin and CD31 immunostaining in the cortex. Cre genotype impacted (E) laminin (F_(1,27)=14.79; P<0.001) and (F) CD31 (F_(1,27)=23.89; P<0.001) levels. Scale bar, 10 µm. n≈16 per Cre genotype. G, Cortical vessel length quantified from GLUT1 (glucose transporter 1) immunostaining was lower in APOE3^Cre+/− mice (F_(1,16)=4.8; P=0.044), n = 10 per Cre genotype. H, Pericyte coverage quantified from GLUT1 and PDGFRβ (platelet-derived growth factor receptor beta) immunostaining showed no difference between APOE3^Cre+/− and APOE3^Cre−/− mice (F_(1,16)=0.78; P=0.39). All data are expressed as a box plot depicting the minimum score, the lower quartile (25%), the median (50%, horizontal line), the upper quartile (75%), maximum values, and mean (+). Data were analyzed using general linear models. C was analyzed using the Empirical Bayes method. *P<0.05 for APOE3^Cre+/− vs APOE3^Cre−/− mice.

We next used immunohistochemical staining to quantify alterations in vessel coverage with knockdown of endothelial APOE. APOE3^Cre+/− mice had both a lower vessel area fraction (Figure 2E and 2F) and a lower total vessel length (Figure 2G) compared with APOE3^Cre−/− mice, but there were no differences in the number of branches or the distance to the nearest vascular neighbor (Figure S5). Consistent with these data, multiplex ELISA of a vascular damage panel revealed that plasma levels of ICAM-1 (intercellular adhesion molecule 1) were also higher in APOE3^Cre+/− mice (Figure S5). We also measured levels of inflammatory and adhesion molecules (Ccl3, Cd74, Cxcl1, B2m, Vcam1, and Vwf) in brain endothelial cells via qRT-PCR, as changes in their levels may also indicate cellular dysfunction. Of these markers, we found that levels of Ccl3 and Cd74 were higher in APOE3^Cre+/− mice compared with APOE3^Cre−/− mice (Figure S5). These data support the idea that lowering endothelial cell APOE expression results in brain endothelial cell dysfunction, including lower vessel coverage and higher levels of inflammatory markers.

Interestingly, cortical cerebral blood flow, another common marker of neurovascular dysfunction, was reduced only in female APOE3^Cre+/− mice (Figure S6). Pericytes have been reported to be involved in regulation of cerebral blood flow.⁴⁶ We looked at pericyte coverage by immunohistochemistry to determine whether there were changes in pericyte coverage that corresponded to the reduced cerebral blood flow and vascular deficits. However, we found that pericyte coverage was unaffected by APOE3 knockdown in both male and female mice (Figure 2H).

In summary, our data support the idea that endothelial cell APOE3 plays a role in the regulation of neurovascular function, as endothelial cell APOE3 knockdown results in higher levels of plasma proteins in the brain and lower vessel coverage/length.

Endothelial Knockdown of APOE3 Disrupts Behavior

Our finding that the loss of endothelial cell APOE3 increased markers of neurovascular dysfunction led us to evaluate whether there was an impact on behavior. APOE3^Cre mice were tested with a range of behavioral tasks including open-field, novel object recognition (Figure S7), MWM, and trace fear conditioning (Figure 3). Out of these behavioral tasks, we found that the loss of endothelial cell APOE caused dysfunction in MWM and trace fear conditioning. In MWM, both APOE3^Cre+/− and APOE3^Cre−/− mice learned the location of the platform (Figure 3A); however, memory was lower in APOE3^Cre+/− mice (Figure 3B through 3D). Trace fear conditioning tests the natural response of a rodent to acquire, remember, and then extinguish a fear-conditioned response.^30,31 We found that APOE3^Cre+/− and APOE3^Cre−/− mice both acquired the fear-conditioned response, although APOE3^Cre+/− mice had slightly lower overall freezing behavior (Figure 3E). In the memory extinction phase, the level of conditioned freezing to the tone diminished over repeated trials in APOE3^Cre−/− mice that is consistent with normal memory extinction. However, APOE3^Cre+/− mice exhibited a level of fear response to the tone that remained high throughout the end of the session (Figure 3F and 3G), which is indicative of an impairment of fear memory extinction. Therefore, the loss of endothelial cell APOE3 results in deficits to spatial memory and fear memory extinction.

Figure 3.

Endothelial cell APOE3 knockdown disrupts behavior. A through D, Spatial learning and memory assessed in the Morris water maze test. In the acquisition phase (A), there was an effect of trial ([F_(4,37)=21.7; P<0.001] P<0.05 for day 1 vs 2, 3, 4 and 5 when assessed by Bonferroni post hoc comparisons; Source Data File 1 in the Supplemental Material) but not Cre genotype. In the probe trial phase that tests memory (B and C), the number of entries (F_(1,40)=6.16; P=0.017) and time spent (F_(1,40)=4.31; P=0.044) in the previous platform area were lower in APOE3^Cre+/− mice compared with APOE3^Cre−/− mice. n=20 per Cre group. Data are expressed as a box plot depicting the minimum score, the lower quartile (25%), the median (50%, horizontal line), the upper quartile (75%), maximum values, and mean (+) and analyzed using general linear models. *P<0.05 for APOE3^Cre+/− vs APOE3^Cre−/− mice. D, Representative track plot of the probe trial phase. E through G, Acquisition and extinction of fear conditioned freezing measured in trace fear conditioning. E, In the acquisition phase, there was an effect of Cre genotype (F_(1,34)=7.15; P=0.011) with overall responses lower in APOE3^Cre+/− mice. In the extinction phases, there was no effect of Cre genotype on day 2 (F_(1,34)=0.19; P=0.66; F); however, on day 3 (G), there was a Cre genotype×trial interaction (F_(4,31)=3.76; P=0.013). Post hoc analysis revealed that these differences were driven by a lack of extinction in APOE3^Cre+/− mice and, therefore, higher freezing responses compared with APOE3^Cre−/− mice in later trials. There was also an effect of sex on the fear conditioned response (Figure S8) but no Cre×sex interaction. n≈20 per Cre genotype. A, E through G, Data expressed as mean±SEM analyzed using general linear models with repeated measures. *P<0.05 for APOE3^Cre+/− vs APOE3^Cre−/− mice when assessed by Bonferroni post hoc comparisons.

Endothelial Knockdown of APOE3 Preferentially Disrupts GABAergic Activity

As the loss of endothelial cell APOE3 caused behavior dysfunction, our final goal was to evaluate whether there were also changes in neuron function. The PFC is important for cognitive function including memory retrieval and extinction. In fact, changes in neuron activity in the PFC have been directly linked to deficits in extinction of fear responses,^30,31 which, therefore, may have been impacted by endothelial APOE knockdown. To address this, pyramidal neurons from the PFC of APOE3^Cre+/− and APOE3^Cre−/− mice were recorded using a protocol that enables concurrent acquisition of excitatory and GABA-inhibitory synaptic activity.^30,31 We found that APOE3^Cre+/− mice exhibited a lower frequency of inhibitory PSC (PSC_+15mV), whereas the level of excitatory (PSC_−60mV) events remained unaltered (Figure 4A). As a result, pyramidal neurons recorded from APOE3^Cre+/− mice exhibited an imbalanced (>1) excitatory/inhibitory ratio (Figure 4B) in a manner that correlated linearly with the frequency of inhibitory PSC (Figure 4C). These results indicate that endothelial cell APOE3 knockdown preferentially reduces GABA function. Interestingly, the levels of multiple neuronal markers in the cortex were not affected by endothelial cell APOE knockdown (Figure 4E), supporting the view that the loss of endothelial APOE3 results in neuronal functional deficits that are not apparent at the synaptic protein level measured via Western blot. Thus, consistent with our behavior data, the loss of endothelial cell APOE3 preferentially reduced GABAergic activity in the PFC, resulting in a higher excitatory/inhibitory ratio.

DISCUSSION

The link between human APOE genotype and age-related neurodegenerative disorders demonstrates the involvement of apoE in regulating brain function. Previous studies have identified APOE modulated functions in models of aging and disease. Compared with APOE3, APOE4 is associated with detrimental changes in behavior, neuron function, metabolism, inflammation, neurovascular function, lipoprotein biology, and disease-specific pathology.^47–49 Research is now further defining these functions for therapeutic development. As a result of these studies, the need for a complementary strategy of identifying cell type–specific functions of APOE has emerged. The major sources of apoE in the body are hepatocytes and astrocytes.^{13–15,50,51} However, it is now recognized that APOE is also expressed by multiple cell types⁵² including neurons,^53,54 macrophages,^34,55–57 pericytes,¹⁶ adipocytes,^58–60 epithelial cells,⁶¹ and microglia.²⁶ Although they might not contribute to global apoE levels, local cell type APOE expression has been found to have significant functional effects.^52,57 Therefore, further investigation of the cell type–specific role of APOE in regulating peripheral and central biology will ultimately allow a deeper understanding of the fundamental functions of APOE, which can then be applied to disease and therapeutic development.

APOE4 is associated with higher permeability and lower vessel coverage during aging, in AD and in respective in vivo models.^{8–12,28,29,62–65} Specialized brain endothelial cells are central to the neurovascular function, primarily controlling the bidirectional movement of essential and unwanted molecules to and from the brain. Mechanistically, research focuses on how APOE genotype-dependent modulation of peripheral and central functions impacts brain endothelial cells (eg, signaling molecules, inflammation). In addition, the contribution of APOE expressed by astrocytes, pericytes, and macrophages to endothelial cell/neurovascular function is under investigation.^13–16 These studies are logical given what is known about APOE expression and function and have identified novel pathways. However, the importance of endothelial cell APOE may have been overlooked. We previously found that mouse brain endothelial cells express APOE in vitro, prompting us to reevaluate our own conceptions about APOE. Based on this finding, we proposed the concept that in a cell as specialized as brain endothelial cells, there may be a distinct advantage of using apoE for maintaining proper cellular function. However, the idea that endothelial cells express APOE to result in functional effects is contested.

The extent that brain endothelial cells express APOE is unclear. APOE expression by human and mouse brain endothelial cells has been demonstrated in both single-cell transcriptomic studies^17–23 and in vitro.^24,66 However, there is variability in human data, leading some to argue endothelial cell APOE expression is negligible. Likely contributing to the variability is difficulty in isolating healthy brain endothelial cells in high yield and the extent that mRNA is degraded postmortem in different tissues⁶⁷ and cell types. Although cultured brain endothelial cells have been shown to produce apoE at the protein level, in vivo data are lacking. Thus, there are questions on whether brain endothelial apoE exists and, if it does, its origin (ie, production versus uptake from interstitial fluid and plasma). Our data demonstrate that APOE is expressed by brain endothelial cells and that knockdown results in much lower levels of the protein. These data support the concept that a significant proportion of intracellular endothelial cell apoE is derived autonomously. This idea is somewhat surprising given that apoE endocytosis occurs in other cell types via multiple receptors (eg, LDLR [low-density lipoprotein receptor], VLDLR [very low-density lipoprotein receptor], and LRP1 [LDL receptor-related protein 1]), which are also expressed by brain endothelial cells.^68,69 However, relatively low rates of apoE clearance from the brain parenchyma by brain endothelial cells have been reported in vivo,⁷⁰ and apoE does not readily cross into the brain from the plasma.⁷¹ Future studies could focus on understanding the relative contribution and subcellular localization of apoE from endogenous and exogenous sources.

In terms of whether endothelial cell APOE is functional, all data thus far are in vitro.^16,25 It has been previously shown in mouse brain endothelial cells that there is higher permeability in response to stress with APOE4 compared with APOE3.²⁴ In addition, APOE4 expression by endothelial/epithelial cells differentiated from human iPSC (inhibitory postsynaptic current) is associated with cellular dysfunction compared with APOE3.²⁵ In contrast, a separate study found no effect of APOE genotype on basal endothelial cell permeability in isolated mouse brain endothelial cells.¹⁶ Due to the expression controversies and these in vitro functional data, there is debate on whether endothelial cell APOE is important for brain function. On one side is the opinion that endothelial cell APOE expression is too low to be functional, especially as other sources may compensate and functional data are an in vitro artifact. On the other side is evidence that APOE is expressed by brain endothelial cells in humans and mice, which has been demonstrated as functional in vitro. Our findings agree with in vitro data and support a novel cell type–specific role of endothelial cell APOE3 in regulating brain function. This concept raises important future questions including whether APOE3 expression is altered by age and disease and whether APOE4 is a loss of positive function or a toxic gain of function.

Our findings raise the question of which protective pathways are regulated by APOE3 in endothelial cells. In general, most research on the cellular function of APOE compares APOE3 with APOE4, and differences have been found in signaling, metabolism, inflammation, reactive oxygen species, and organelle damage. Overall, APOE appears to be integral to maintaining cellular function in response to acute and chronic stress (eg, age, disease, injury, and inflammation), which may be particularly important for brain endothelial cells. Brain endothelial cells are critical for maintaining homeostasis of the central nervous system, while being continually exposed to acute and chronic fluctuations in circulating molecules from the brain and plasma in physiological and pathological states. Thus, APOE may help maintain brain endothelial function in normal physiology, particularly in response to convergent stressors. More specifically, APOE has been found to modulate metabolism and stress or inflammatory pathways in brain endothelial cells in vitro.^24,25 In fact, we previously proposed that compared with APOE3, APOE4 brain endothelial cells have altered metabolism resulting in generation of mitochondrial reactive oxygen species, oxidative stress, and inflammation, which lowers the threshold of endothelial cells to dysfunction in the presence of stressors. In this study, we also found higher levels of Ccl3 and Cd74 at the mRNA level, and plasma ICAM levels, which may indicate that the loss of endothelial cell APOE3 has resulted in an activated or stressed phenotype. Indeed, Ccl3⁷² and Cd74^73,74 are upregulated in brain endothelial cells in response to stress, and we previously found higher CCL3 production by APOE4 brain endothelial cells after lipopolysaccharide treatment compared with APOE3.²⁴ Therefore, the loss of endothelial cell APOE3 may have triggered metabolic and stress-related pathways during aging. Future studies could reveal whether brain endothelial cell APOE3 regulates metabolism and stress responses in vivo and the difference compared with APOE4.

Integrating endothelial cell APOE with other identified pathways may allow a more complete understanding of how APOE regulates neurovascular function. In general, most data on this topic are comparisons between APOE genotypes, which is different to the physiological role of APOE in neurovascular function. Nonetheless, broadly, APOE-dependent modulation of inflammatory and metabolic signaling molecules in the periphery and within the brain can impact brain endothelial cells. In addition, APOE expressed by specific cell types has been shown to modulate pathways that alter brain endothelial cells. Astrocyte-produced apoE3 is thought to bind LRP1 on pericytes initiating a cascade that suppresses MMP9 production to limit basement membrane and tight junction degradation¹³ and lower endothelial cell permeability in vitro.¹⁵ ApoE3 production by pericytes has also been reported to promote extracellular matrix production and barrier function of brain endothelial cells in vitro.¹⁶ Peripherally, hepatocyte apoE3 alters brain endothelial cell function in a way that is considered protective.¹⁴ Thus, APOE3 expressed by astrocytes, pericytes, endothelial cells, and hepatocytes is involved in protecting neurovascular integrity. Such complex regulation by APOE3 may serve the advantage of allowing changes in peripheral and central environment (astrocytes, pericytes, and hepatocytes) to signal to brain endothelial cells for protection. In this context, brain endothelial apoE3 may be important for maintaining local homeostatic cellular function.

We also found that the loss of endothelial cell APOE3 caused deficits in the cortical excitatory/inhibitory balance due to a preferential lowering of GABAergic activity and associated dysfunction in fear extinction behavior. The alterations in GABAergic activity could be mediated via changes in permeability, vessel coverage, or other brain endothelial cell functions such as signaling cross talk between neurons and brain endothelial cells. Independent of precise functional connection, these data raise the novel idea that GABAergic synapses are particularly vulnerable to deficits in conditions with altered endothelial cell function, such as aging or neurodegeneration.

We would like to acknowledge limitations in this study. One limitation is the unresolved question of why cerebral blood flow was lower in female mice, but not male mice, with endothelial cell knockdown. A potential explanation is that with endothelial cell APOE knockdown, alterations in neurovascular function are age and sex dependent, such that cerebral blood flow alterations would occur in older male mice. Alternatively, the phenotype may be unique to female mice, due to sex-related changes in endothelial cells, smooth muscle cells, or pericytes that increases susceptibility to cerebral blood dysregulation after endothelial cell APOE3 knockdown. For example, sex hormones or stress and inflammatory processes are all known to be modulated by sex and can alter cerebral blood flow. In addition, due to technical limitations, we could not evaluate whether vasoreactivity/neurovascular coupling was altered by endothelial cell APOE knockdown in male and female mice. By design, our studies were focused on resolving whether endothelial cell APOE3 is functional, and, therefore, there is the unresolved question of how endothelial cell APOE3 maintains brain endothelial cell function. For example, as described above, metabolism, stress-related signals, or inflammation may be proximal to brain endothelial cell degeneration. In addition, our data suggest that claudin-5 degradation may contribute to the higher permeability after endothelial cell APOE3 knockdown, which could be related to recycling, signaling, and/or proteasomal degradation. To aid in pathway identification, it will be important to track age-dependent changes in neuronal function after endothelial cell APOE knockdown. Also, an important question is whether endothelial cell APOE4 is a toxic gain of function or loss of positive function. Our planned future studies are to identify whether the loss of endothelial cell APOE4 impacts neural function in a positive or negative manner and then to dissect the underlying pathways for APOE3 and APOE4.

In summary, our data demonstrate that the loss of endothelial cell APOE3 is sufficient to disrupt neurovascular, neuronal, and behavioral function. These data support the novel cellular mechanism that endothelial APOE directly impacts neural function in normal physiology and could, therefore, play a role in neurodegenerative disorders.

ARTICLE INFORMATION

Acknowledgments

The authors would like to acknowledge the Cure Alzheimer’s Fund for supplying the humanized APOE3^fl/fl mice.

Sources of Funding

This work was supported by National Institutes of Health grants R01AG061114 (L.M. Tai), R61NS114353 (L.M. Tai), R01MH086507 (K.Y. Tseng), and University of Illinois at Chicago Institutional funds (L.M. Tai and K.Y. Tseng). qRT-PCR was performed by the University of Illinois at Chicago (UIC) Genomics Research Core and Bioinformatics analysis by the UIC Research Informatics Core, supported, in part, by NCATS (National Center for Advancing Translational Sciences) through grant UL1TR002003.

Disclosures

None.

Supplemental Material

Figures S1–S10

Major Resources Table

Source Data File 1