Cellular senescence induced by cholesterol accumulation is mediated by lysosomal ABCA1 in APOE4 and AD

Background Cellular senescence is a hallmark of aging and has been implicated in Alzheimer’s disease (AD) pathogenesis. Cholesterol accumulation drives cellular senescence; however, the underlying mechanisms are unclear. ATP-binding cassette transporter A1 (ABCA1) plays an important role in cholesterol homeostasis. ABCA1 expression and its trafficking is afiltered in APOE4 and AD cellular and mouse models. However, whether ABCA1 trafficking is involved in cellular senescence in APOE4 and AD remains unknown. Methods We examined the association between cellular senescence and ABCA1 expression in human postmortem brain samples using transcriptomic, histological, and biochemical analyses. An unbiased proteomic screening was performed to identify targets that mediate cellular ABCA1 trafficking. APOE4-TR mice, immortalized, primary and induced pluripotent stem cell (iPSC) models were used to examine the cholesterol-ABCA1-senescence pathways. Results Bulk and single nuclei transcriptomic profiling of the human dorsolateral prefrontal cortex from the Religious Order Study/Memory Aging Project (ROSMAP) revealed upregulation of cellular senescence transcriptome signatures in AD, which was strongly correlated with ABCA1 expression. Immunofluorescence and immunoblotting analyses confirmed increased ABCA1 expression in AD brain tissues, which was associated with lipofuscin-stained lipids and mTOR phosphorylation. Using discovery proteomics, caveolin-1, a sensor of cellular cholesterol accumulation, was identified to promote ABCA1 endolysosomal trafficking. Greater caveolin-1 expression was found in both APOE4-TR mouse models and AD human brains. Cholesterol induced mTORC1 activation was regulated by ABCA1 expression or its lysosomal trapping. Reducing cholesterol by cyclodextrin in APOE4-TR mice reduced ABCA1 lysosome trapping and increased ABCA1 recycling to efflux cholesterol to HDL particles, reducing mTORC1 activation and senescence-associated neuroinflammation. In human iPSC-derived astrocytes, the reduction of cholesterol by cyclodextrin attenuated inflammatory responses. Conclusions Cholesterol accumulation in APOE4 and AD induced caveolin-1 expression, which traps ABCA1 in lysosomes to activate mTORC1 pathways and induce cellular senescence. This study provided novel insights into how cholesterol accumulation in APOE4 and AD accelerates senescence.


Background
Cellular senescence, de ned as permanent arrest of the proliferative state of the cell, is one of the hallmarks of aging and age-related diseases, including Alzheimer's disease (AD) [1,2].Senescent cells secrete extensive in ammatory mediators, such as chemokines, cytokines, growth factors, and proteases, collectively known as the senescence-associated secretory phenotype (SASP) [3,4], which contributes to AD pathology.Cellular senescence is triggered by many stimuli such as DNA damage, reactive oxygen species, telomere attrition, lysosomal dysfunction, tau and amyloid aggregation, and lipid accumulation [2,5,6].Senescent astrocytes, microglia, endothelial cells, and neurons have been detected in the brains of patients with AD and AD mice [4,7], particularly in relation to tau aggregation [8].The clearance of senescent cells by senolytic therapy alleviates cognitive de cits in AD mouse models [9,10].In addition, removing senescent cells in naturally aged mice lengthens their lifespan [11].
However, the key drivers of cellular senescence and their association with AD neuropathology markers remain largely unknown.
How cell-speci c senescence signatures relate to brain ABCA1 expression, APOE4, and AD neuropathology remains unknown.To address this gap, we examined cellular senescence signatures using bulk-RNA and single-nucleus RNA sequencing (snRNA-seq) in postmortem human brain tissues from the Religious Order Study/Memory Aging Project (ROSMAP).This was followed by the measurement of senescence markers and ABCA1 expression in postmortem human brain tissues from individuals with or without AD, which differed by APOE genotype.Mechanisms of ABCA1 tra cking were investigated using cell lines, primary cells, human induced pluripotent stem cells (iPSC), and humanized APOE-TR mouse models.

Study design
To determine whether cellular senescence is associated with ABCA1 expression in APOE4 and AD, we analyzed the databases of bulk-RNA sequencing (632 participants) and single-nucleus RNA sequencing (427 participants) that performed from the dorsolateral prefrontal cortex (DLPFC) of postmortem human brain samples in the Religious Orders Study or the Rush Memory and Aging Project ROSMAP.Frozen mid frontal lobe brain tissues (138 participants) and formalin-xed para n-embedded (FFPE) slides ( 12participants) of the postmortem human brains from ROS were used to validate the ndings of RNAsequencing analysis.To identify the proteins that regulate ABCA1 degradation, proteomics was performed after ABCA1 puri cation in ABCA1 overexpression HeLa cells treated with recombinant ApoE3 or ApoE4 proteins.The biochemistry and histochemistry analysis were used to validate the ndings from proteomics.To test the effect of cholesterol reduction on the APOE4-TR mice, 14-15 mice per group were determined by the publication.The mice were subjected to behavioral tests and treatment allocation was blinded to the testers.All animal experiments were approved by the Institutional Animal Care and Use Committee of the University of Southern California.Every effort was made to reduce animal stress and minimize animal use.For all experiments, the number of replicates, statistical tests used, and P values are reported in the gure legends.
Analysis of bulk-RNA sequencing data Database Filtered raw counts were downloaded from the Synapse AD Knowledge Portal (https://www.synapse.org/#!Synapse:syn9702085) with Synapse ID: syn8456637.
Source data were collected from a cohort of 632 subjects from the Religious Orders Study or the Rush Memory and Aging Project (ROSMAP) [26].Both studies were approved by the Institutional Review Board of Rush University Medical Center.All participants signed informed and repository consent forms and the Anatomical Gift Act.Based on the clinical diagnosis at the time of death, 83 study participants were diagnosed with no cognitive impairment (NCI), 162 with mild cognitive impairment (MCI), and 151 with AD dementia [27][28][29].CERAD Braak staging represents the amount and distribution of neuritic plaques and neuro brillary tangle pathology [30,31].The APOE genotype was determined as previously described [32].Detailed information regarding the samples is provided in Supplementary Table 1.

Gene Set Enrichment Analysis (GSEA)
The R package GSVA (method option "ssGSEA") was used to calculate the pathway enrichment score of each cell.The annotated gene sets were retrieved from MsigDB.REACTOME_CELLULAR_SENESCENCE (M27188) was selected to represent cellular senescence.The pathway score was tested to determine the differences between the variables of interest.
Analysis of single nucleus-RNA sequencing data

Database
Post-QC counts and ROSMAP metadata were downloaded from the Synapse AD Knowledge Portal (https://www.synapse.org/#!Synapse:syn52293417) with Synapse ID: syn2580853.Source data were collected from a sample of 427 subjects from ROSMAP [33].Detailed information about the samples is provided in Supplementary Table 2. Nuclei were isolated from frozen postmortem brain tissues and subjected to droplet-based single-nucleus RNA sequencing (snRNA-seq).

Gene Set Enrichment Analysis (GSEA)
The R package GSVA (method option "ssGSEA") was used to calculate the pathway enrichment score of each cell.The annotated gene sets were retrieved from MsigDB.REACTOME_CELLULAR_SENESCENCE (M27188) and REACTOME_SENESCENCE_ASSOCIATED_SECRETORY_PHENOTYPE_SASP (M27187) were selected to represent cellular senescence and SASP.The linear mixed effect was used to test if the pathway score was different between the variables of interest, using the formula: ~ size factor (the library size of each cell) + interested variables + (projid (sample id)).

SenTraGor (STG) staining
The formalin-xed para n-embedded (PPFE) human brain slides containing middle frontal lobe region (ROS) were depara nized following the protocol (xylene, 3 min each time for two times; xylene 1:1 with 100% ethanol, 3 min each time for two times; 100% ethanol, 3 min each time for two times; 95% ethanol, 3 min each time for two times; 70% ethanol, 3 min each time for two times; 50% ethanol, 3 min each time for two times; running cold tap water to rinse for 5 min).The slides were then subjected to antigen retrieval using a Sodium Citrate buffer (10mM Sodium Citrate, 0.05% Tween 20, pH 6.0) for 30 min.The slides were washed once with TBS for 5 min and incubated with STG (Bio-Techne, #7555, 2mM, dissolved in 100% ethanol, ltered with 0.45uM lter) for 10 min at room temperature (RT).After incubation, the slides were washed twice with 50% ethanol followed by one time wash with TBST (0.025% Triton X-100 in TBS) for 5 min each wash.Then, primary antibodies (goat anti-biotin, anti-ABCA1, and anti-GFAP) were added to the slides and incubated overnight at 4°C.After washing three times with TBST, the slides were incubated with uorescein-conjugated secondary antibodies at room temperature for 1 h.After washing three times with TBST, the slides were incubated with an auto uorescence quencher at room temperature for 5 min.After washing thrice with TBST, the slides were mounted with mounting medium (F4680; Sigma).Images were acquired using a slide scanner (Zeiss Axio Scan, Oberkochen, Germany).Z1, 20x objective) and analyzed using ImageJ software (NIH).

Human brain tissue homogenate preparation
Frozen mid-frontal lobes of postmortem human brains were obtained from the Rush Alzheimer's Disease Center (RADC) at the Rush University Medical Center.The Religious Orders Study (ROS) was approved by the Institutional Review Board (IRB) of Rush University Medical Center [34].Detailed information on these samples is presented in Supplementary Table 3.
To extract soluble and insoluble proteins, frozen middle frontal lobe brain tissues were weighed and homogenized with TBS (1:15, w/v) containing a protease inhibitor cocktail and phosphatase inhibitor cocktail.The homogenate was centrifuged at 15,000 g for 1 h at 4°C.The supernatant was collected as the TBS fraction, and the pellets were incubated with TBSX (1% Triton X-100) (same volume as TBS) overnight with agitation at 4°C.The next day, the homogenate was centrifuged at 15, 000 × g for 1 h at 4°C and the supernatant was collected as the TBSX fraction.The protein concentrations of all fractions were measured using a bicinchoninic acid kit and used for further experiments.For ABCA1 detection, TBSX fraction samples were added to 4x Laemmli sample buffer (without boiling) and subjected to 12% SDS-PAGE for western blotting.
The elution was quickly neutralized with 2µL of Tris buffer (1M, pH8.5).Then lamp-2 enriched samples were added to 4x Laemmli sample buffer (without boiling) and loaded onto a 7.5% SDS-PAGE gel for western blotting.
100uL of GFP-Trap Agarose (gta-20, Chromotek) was rst washed with IP buffer (PBS 7.5-7.9; 200mMNacl, 1mM EDTA, 0.5% Triton X-100, 1mM DTT and 0.2mM PMSF) three times and then added to 2mL supernatant of HeLa (ABCA1-GFP) cell lysates and incubated with rotation overnight at 4°C.The next day, agarose was washed with IP buffer three times with rotation at 4°C.After washing, a 2-fold volume of glycine (0.1M, pH 2.7) was used to elute the ABCA1 complex and neutralized with 1/10 volume of Tris-HCl (1M, pH 9.5) after elution.Parts of elution were used for SDS-PAGE analysis, and other parts were added with TCA (SA433-500, Fisher Scienti c) to a nal concentration of 20% and incubated for 15 min to precipitate proteins.After centrifugation (13,000rpm, 4°C, 30 min), the pellet was collected and washed once with 1mL of acetone (A949-1, Fisher Scienti c).After removing the acetone, the protein pellets were airdry for 3-4 minute at room temperature to ensure that no liquid remained.Finally, the protein pellet was sent to a mass spectrometry facility (Harvard Center for Mass Spectrometry).
For ABCA1 degradation experiments, Hela cells expressing ABCA1-GFP were treated with recombinant APOE3 or APOE4 (0.2µM) (Academy Biomedical Company) for different time points.After treatment, cells were lysed with radioimmunoprecipitation assay (RIPA) buffer (9806, Cell Signaling Technology, CST), and total ABCA1 protein levels were detected by western blotting.

Analysis of MS data
The normalized spectral abundance factors (NSAFs) [35] were used to measure the relative abundance and calculate the fold change for each protein.To obtain the NSAF value for a speci c protein, the spectral abundance factor (SAF) value was rst calculated by dividing the spectral count of the protein by its length (number of amino acids).The NSAF value was calculated by dividing the SAF value by the sum of the SAF values from all the identi ed proteins within each sample.

Cell culture
Primary astrocytes were obtained from APOE3-TR and APOE4-TR mouse pups and cultured as previously described [60].Brie y, cerebral cortices from 1 to 3 day-old neonatal mouse were dissected in ice-cold Hanks' balanced salt solution (HBSS) (Corning, 21-021-CV) and digested with 0.25% trypsin for 20 min at 37°C.Trypsinization was stopped by adding a 2-fold volume of DMEM (Corning, 10-013) with 10% fetal bovine serum (FBS) (Omega Scienti c, FB-12) and 1% antibiotic-antimycotic (anti-anti) (Thermo Fisher, 15240062).The cells were dispersed to a single-cell level by repeated pipetting and ltered through 100µm cell strainers (VWR, 10199-658).After ltering, the cells were centrifuged for 5 min at 1000 rpm and resuspended in culture medium (DMEM, Corning, 10-013) supplemented with 10% FBS and antibiotics.The cells were then seeded in a 75 cm 2 ask and cultured at 37°C in 5% CO2.The medium was changed the next day and replaced every three days.The mixed glial cultures reached con uence after 7-10 days.The cells were then shaken at 250 rpm for 16h at 37°C to remove the microglia and oligodendrocyte progenitor cells.The remaining cells were harvested by trypsin digestion.At this stage, the culture contained 95% astrocytes, which were used for further experiments.

Cell plasma membrane protein preparation
To study cell plasma membrane proteins, biotinylation of cell surface proteins was performed to isolated cell plasma membrane protein.Brie y, cells were washed twice with cold PBS followed by incubation with 0.5 mg/ml sulfo-NHS-SS-biotin (Thermo Fisher Scienti c, PG82077) in PBS for 30 min at 4°C with shaking.The reaction was quenched by rinsing cells with 50 mM glycine in PBS.The cells were then lysed with RIPA buffer (CST, 9806) containing a protease inhibitor cocktail, followed by centrifugation at 12,000 × g for 10 min at 4°C.The supernatant was collected, and protein concentrations were measured using a BCA kit (Thermo Fisher, 23225).The radioimmunoprecipitation assay (RIPA) fraction represented the total protein fraction.Then, 100µg of the RIPA fraction from each sample was incubated with 40ul of NeutrAvidin agarose (Thermo Fisher Scienti c, 29200) for 2 hours at 4°C.The agarose was then washed three times with PBS containing a protease inhibitor cocktail, followed by boiling with 100µL 2x sample buffer for 5 min at 95°C.After centrifugation (2,500 x g, 2 min, room temperature), the supernatants were collected as plasma membrane proteins.Membrane and total protein analyses were performed using western blotting.

Immunoprecipitation
HeLa (negative) or HeLa (ABCA1-GFP) cells were lysed in RIPA buffer containing protease and phosphatase inhibitors.10uL of GFP-Trap Agarose was added to 200uL cell lysates (1µg/µL) and incubated with rotation overnight at 4°C.The next day, agarose was washed with IP buffer three times with rotation at 4°C.After washing, a 2-fold volume of diluted (2x) sample buffer (Bio-Rad, 1610747) was added and boiled for 5 min at 95°C.After centrifugation (1,500 rpm, 5 min, room temperature), supernatants were collected for SDS-PAGE and western blotting.
Astrocytes were lysed in radioimmunoprecipitation assay (RIPA) buffer containing protease and phosphatase inhibitors.1.5ug antibody (anti-ABCA1) was added to 400uL cell lysates (1µg/µL) and incubated with rotation for 1 h at 4°C.Then, 25uL Dynabeads Protein G (Thermo Scienti c, 10004D) was added and incubated with rotation overnight at 4°C.After three washes with 1 mL of ice-cold 0.1% TBST, 25 µL of diluted (1.5 ×) sample buffer was added to the beads and boiled for 5 min at 95°C.The supernatants were collected and subjected to SDS-PAGE and western blotting.
The TBSX fraction of postmortem human cortical tissue was used to enrich the lysosomes.Anti-LAMP2 antibody was added to 300uL homogenate (1µg/µL) and incubated with rotation for 1 h at 4°C.Then, 20uL Dynabeads Protein A/G was added and incubated with rotation overnight at 4°C.After three washes with 1 mL of ice-cold 0.1% TBST, the beads were eluted with 20 µL elution buffer.The supernatants were collected and subjected to SDS-PAGE and western blotting.

Western blotting
The cell lysates, brain homogenates, or immunoprecipitation protein complexes were separated on 4-15% mini-precast protein gels (Bio-Rad, 4561086) under reducing conditions and then transferred onto nitrocellulose membranes (Bio-Rad, 1704270).After transfer, the membranes were blocked with 5% fatfree milk (Bio-Rad, 1706404) in TBST for 1 h at room temperature, followed by overnight incubation with the primary antibody in 5% BSA at 4°C.The membranes were then incubated with an HRP-conjugated secondary antibody for 1 h at room temperature.A chemiluminescent HRP substrate (Millipore, WBKLS0500) was used for detection.The Fuji lm LAS-4000 imager system was used to capture images, and densitometric quanti cation was performed using Gel Quant NET software.The antibodies used are listed in Supplementary Table 5.
The mouse Caveolin-1 (Myc-DDK-tagged) expression plasmid (MR201562) was purchased from Origene and puri ed using an endo-free plasmid DNA midi kit (Bioland Scienti c, PD03-21).The pCMV-GFP expression plasmid was used as the control.The cells were seeded in 24-well plates and cultured overnight.The plasmid (0.125µg) was transfected using the jetPRIME reagent.Protein expression was detected using western blotting 48 h after transfection.

Cholesterol e ux
Immortalized astrocytes were seeded in 24-well plate (0.12x10 5 cells/500µL) and cultured overnight.The cells were transfected with 10nM siRNA for 24 h, followed by labeling with 1µCi/mL (3H) cholesterol (Moravek, MT9112) using serum-free DMEM/F12 containing 2 mg/ml fatty acid-free BSA (Sigma-Aldrich, catalog #A9647), and 2ug/ml acyl-coenzymeA: cholesterol acyltransferase inhibitor SANDOZ (Sigma-Aldrich, catalog #S9318) for 24 h.After washing once with serum-free culture medium, the cells were treated with recombinant CS-6253 peptide (1µM in DMEM/F12 containing 2 mg/mL fatty acid-free BSA, 2 µg/mL SANDOZ, and 200 µL/well) for 4 h.After treatment, the cell culture medium was collected and transferred to scintillation vials lled with 3 mL of the scintillation mixture.The cells were solubilized in 0.5 N NaOH (200 µL), neutralized with PBS, and then transferred to scintillation vials lled with 3 mL of scintillation mixture.After vigorous mixing, vials were counted using a Beckman LS6500 liquid scintillation counter (Beckman Coulter).The e ux of cholesterol was assessed by the ratio of cholesterol in the medium to the total cholesterol (medium and cell lysate).

Cholesterol loading and depletion
Immortalized astrocytes or baby hamster kidney (BHK) cells were seeded in 24-well plates (0.5x10 5 cells/500µL) and cultured overnight.Cells were washed twice with serum-free medium, and human LDL (10µg/mL diluted in serum-free medium containing 2 mg/mL fatty acid-free BSA) (Sigma, LP2-2MG) was added for 24 h.The cholesterol levels in the cells were determined using a cholesterol assay kit ( lipin III staining, Abcam, ab133116).Cellular caveolin 1 protein levels were determined by western blotting.
Human iPSC astrocytes were seeded in 48-well plates (0.5x105 cell/300uL) and cultured for 48 h.Cells were rst loaded with LDL as described above and treated with 1mM methyl-β-cyclodextrin (MβCD) for 2 h.Cellular cholesterol levels were determined using lipin staining.For parallel experiments, after MβCD treatment, cells were stimulated with TNFα (100ng/mL) and IFNγ (100ng/mL) for 16 h.RNA was extracted from each condition using qPCR.
ABCA1 knock down cell line and cholesterol loading ABCA1 knocking down astrocyte line was created by transfection of ABCA1 CRISPR Gene Knockout Kit (ABCA1-sgRNAs and Cas9 recombinant protein, Synthego).Three days after transfection, the single cell was seeded and cultured for two weeks.ABCA1 knocking out or knocking down clones were validated by western blotting using ABCA1 antibody.

ABCA1 induction and cholesterol loading in BHK cells
BHK cells were seeded in 24-well plates (0.5x105 cell/500uL) and cultured overnight.ABCA1 expression was induced with Mifepristone (0.1nM) treatment for 16 hours.Then, cells with or without ABCA1 were loaded with LDL (100µg/mL) for 24 hours.The cells were lysed for further analysis.

Cell lysate and mouse brain homogenate preparation
Immortalized or primary astrocytes were lysed with 1x RIPA buffer (CST, 9806) containing protease inhibitor cocktail (Sigma, P8340) and phosphatase inhibitor cocktail (Sigma, P0044), followed by centrifugation at 14,000 × g for 10 min at 4°C.The supernatant was then collected for further analysis.
The mouse cerebral cortex was weighed, and RIPA buffer containing a protease inhibitor cocktail and phosphatase inhibitor cocktail was added at a ratio of 1:30 (w/v).The tissue was then homogenized using a 2 mL glass Dounce tissue grinder, followed by centrifugation at 15, 000 × g for 1 h at 4°C.The supernatant was collected, and the concentration was measured using a BCA kit.
To extract soluble and insoluble proteins, mouse and human brain samples were weighed and homogenized with TBS (1:15, w/v) containing protease and phosphatase inhibitor cocktails.The homogenate was centrifuged at 15,000 g for 1 hour at 4°C.The supernatant was collected as the TBS fraction, and the pellets were incubated with TBSX (1% Triton X-100) (same volume as TBS) overnight with agitation.After centrifugation (15,000 × g, 1 hour at 4°C), the supernatant was collected as the TBSX fraction.The pellets were dissolved in 150-200uL guanidinium chloride (GnHCl) (5M, pH7.5) and incubated overnight at room temperature with agitation.After centrifugation (15,000 × g, 1 hour at 4°C), the supernatant was collected and dialyzed with a dialysis device (MINI Dialysis Device, 3.5 K MWCO, 0.1 mL, Thermo Scienti c) in TBS overnight at 4°C.The solution was collected as the GnHCl fraction.The protein concentrations of all fractions were measured using a bicinchoninic acid kit and used for further experiments.

Immunohistochemistry or immuno uorescence on mouse brain slides
Fixed mouse brains were sectioned using a Leica cryostat at a thickness of 10 µm.Slides were equilibrated for 30 min at room temperature and rinsed twice with PBS.After blocking with blocking buffer (10% goat serum + 1% BSA + 0.2% Triton X-100 in PBS) at room temperature for 1 hour, the sections were incubated with diluted primary antibodies (GFAP, Iba-1) and incubated overnight at 4°C.After washing, the slides were incubated in a 0.3% H 2 O 2 solution in PBS at room temperature for 10 min to block endogenous peroxidase activity.Then, the SignalStain® Boost IHC Detection Reagent (#8114, Cell Signaling Technology) was added to the slides and incubated at room temperature for 1 hour.After washing, DAB substrate solution was added to the slides and developed for 2-5 min until the desired color intensity was reached.After washing, the slides were mounted with mounting medium (F4680; Sigma).Images were taken using an Olympus Microscope (10x objective).Images were analyzed using the ImageJ software (NIH).
For immuno uorescence staining, after overnight incubation with the primary antibody (synaptophysin), the uorescein labelled secondary antibody was added to the slides and incubated at room temperature for 1 hour (protected from light).After washing, the slides were mounted with mounting medium with DAPI (Sigma, F6057).Images were captured using a Zeiss Axiovert 200M Inverted Fluorescence Microscope (20x objective).Images were analyzed using the ImageJ software (NIH).
For ABCA1 and Lamp1 double staining, after overnight incubation with the primary antibody (ABCA1, 18180, Abcam; Lamp1, 99437, CST), the uorescein labelled secondary antibody was added to the slides and incubated at room temperature for 1 hour (protected from light).After washing, the slides were mounted with mounting medium with DAPI (Sigma, F6057).Images were taken using a slide scanner (Zeiss, Axio Scan.Z1, 20x objective) and analyzed using the ImageJ software (NIH).Representative images were taken using Leica SP8 confocal microscope (60x objective).

Lysosome isolation
Thirty to fty mg of brain tissues were used to isolate lysosomes with the lysosome isolation kit (Minute Lysosome Isolation Kit, Invent Biotechnologies, LY-034) following the manuals.Brie y, the tissues were homogenate with a plastic rod for 1 minute, followed by the centrifugation at 16,000 X g for 30 seconds.The lter was discarded, and the pellet was resuspend by vigorously vertexing for 10 seconds, followed by centrifugation at 2,000X g for 3 minutes.The supernatant was collected to a fresh 1.5 ml microfuge tube and centrifuged at 4°C for 15 minutes at 9,000 X g.After centrifugation, the supernatant was carefully transfer to a fresh 1.5 ml tube and spined at 16,000 X g at 4°C for 30 minutes.After centrifugation, the supernatant was removed completely.Then, the pellet was resuspended in 200 µl of cold buffer A by pipetting up and down 60-100 times and vortex vigorously for 20 seconds, followed by centrifugation at 2,000 X g for 4 minutes.The supernatant was transferred to a fresh 1.5ml tube and added 100 µl of buffer B, and vortexed brie y vortex to mix well (the supernatant to buffer B ratio is 2:1).The mixture was incubated on ice for 30 minutes and centrifuged at 11,000 X g for 10 minutes.After removing all the supernatant, the pellet (lysosomes) was resuspended in 50µl of RIPA buffer for further analysis.

Bodipy-cholesterol stain
Fixed mouse brains were sectioned using a Leica cryostat at a thickness of 10 µm.Brain sections were incubated overnight at 4°C in a solution of 10 µg/ml bodipy-cholesterol (Cayman Chemical, 24618) in PBS.The slides were then washed with PBS 4 times.Images were obtained using a Zeiss Axiovert 200M Inverted Fluorescence Microscope (20x objective) using the same imaging parameters for each image.

Isolation of cell types from mouse brain
Cell types isolation from mouse brain was performed following the published protocol [36].Brie y, mice were euthanized with CO2 and perfused with PBS.The brains were harvest and cerebra were collected.Then, the cerebra were dissociated in the enzyme digestion mix (collagenase A (Sigma, 10103586001) + DNase I (Sigma, 10104159001) by passing three separate ame-polished Pasteur pipets to a single cell suspension.After centrifugation, the debris/myelin was removed by percoll (Sigma, GE17-0891-02) gradient centrifugation.The myelin fraction and total cells mixture were collected separately.Microglia from total cells mixture were puri ed by incubating with CD11b microbeads (Miltenyi Biotec,130-093-636) and passing LS magnetic column (Miltenyi Biotec, 130-042-401).CD11b negative population were also collected.After collection, the different cell types were lysate with RIPA for western blotting detection.
Ion-mobility analysis HDL particles in mouse CSF and plasma were determined by ion mobility analysis as previously reported [37].Samples treated with dextran sulfate were introduced into a charge-reducing electrospray (TSI Inc., model 3482) every 13 min by automated loop injections using an integrated autosampler (Teledyne CETAC Technologies, model MVX-7100).The electrospray settings were as follows: voltage 2.0 kV, CO2 ow 0.15 slmp, and air ow 1.5 slmp.The differential mobility analyzer (TSI Inc., model 3085), coupled to a condensation particle counter (TSI Inc, model 3788), scanned particles 4.45 to 63.8 nm for 180 s.The generated data were analyzed using Fityk (version 1.3.1),as previously described, and graphed using OriginPro software (version 2021).Voigt probability distribution curves were generated from particle count (#/mL) vs. diameter range for lipoprotein subclasses and normalized by dividing subclasses by the sum of peak areas from all lipoproteins present within the spectrum.

Animals and treatment
All animal experiments were approved by the Institutional Animal Care and Use Committee of the University of Southern California.Animals were housed under a standard 12-h light/dark cycle with water and chow diet in a pathogen-free animal facility at the University of Southern California.Every effort was made to reduce animal stress and minimize animal use.ABCA1 / mice were obtained from Dr. John Parks of Wake Forest University School of Medicine.Nestincre mice (Strain #:003771) were purchased from Jax. ABCA1 conditional knockout mice were generated by crossing ABCA1 / mice with nestin-Cre mice.ABCA1 genotype was con rmed by PCR genotyping.
Ten-month-old of wild type and ABCA1 conditional knockout mice were euthanized with CO 2 and perfused with cold PBS.Brains were collected for further experiments.
The APOE3-TR and APOE4-TR mice were purchased from JAX. APOE3 and APOE4-TR mice of different ages were euthanized using CO 2 and perfused with cold PBS.Brains were collected for further experiments.
APOE4-TR male and female mice (8 months old) were randomized to receive subcutaneous injections of 2 g/kg body weight 2-hydroxypropyl-β-cyclodextrin (HPCD) in phosphate-buffered saline (PBS) or PBS twice a week for eight weeks.The mice were then subjected to behavioral tests, and the testers were blinded to the treatment allocation.The injection was continued during behavioral tests (two weeks).Finally, after collection of the cerebrospinal uid (CSF) and plasma, the mice were anesthetized and subjected to cardiac perfusion with ice-cold PBS.The brains were dissected, and one hemisphere was post-xed in 4% paraformaldehyde (PFA) for cryostat sectioning, while the other was used for biochemistry assays and RNA puri cation.

Novel Object Recognition (NOR)
This was performed as previously reported with some modi cations [38].On day 1, mice were acclimated to the test room for 1 h, followed by free exploration for 5 min in the test arena (without objects).On day 2, mice were placed into the same arena containing two identical objects and allowed to freely investigate both for 5 min (training trial).After an interval of two hours, the mice were placed back in the chamber with one novel object and one familiar object and allowed to explore for 5 min (testing trial).After each trial, the testing area and objects were thoroughly cleaned with 70% ethanol solution.
The training and testing trials were recorded with a high-resolution camera, and the number of touches to both objects of each mouse in the testing trial was analyzed using the EthoVision XT software (Noldus).Mice touching the object with the nose were identi ed as one instance of nose touch.The movement distance, total touch times for both objects, and ratio of nose touch times of the novel object to the old object were calculated.One mouse in the HPCD group was excluded because of a ghting wound.

Statistical analysis
GraphPad Prism software (version 10) or Program R (regression analysis, version 4.3.2) was used for all statistical analyses.One-way or two-way analysis of variance (ANOVA) was used to determine statistical signi cance, followed by Tukey's test for multiple comparisons.All quantitative data are presented as mean ± SD.Statistical signi cance was de ned as *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

ABCA1 expression associates with markers of cellular senescence in postmortem brain cells
To determine whether cellular senescence is associated with ABCA1 expression in APOE4 and AD, we rst analyzed bulk-RNA sequencing data from the dorsolateral prefrontal cortex (DLPFC) of postmortem human brain samples from 632 participants in ROSMAP.The sample included 261 APOE3 controls (APOE3/3) and 96 APOE4 carriers (APOE3/4 and APOE4/4).Based on the clinical diagnosis using clinical data collected over time and proximate to the time of death, 83 participants were classi ed as having no cognitive impairment (NCI), 162 as having mild cognitive impairment (MCI), and 151 as having AD dementia.A detailed neuropathological evaluation included Braak staging, which indicated the degree of AD pathology based on the severity of tau tangles.A total of 164 participants were classi ed as Braak Stages 0, I, II, and III.A total of 116 patients were classi ed as Braak stage IV disease.A total of 116 participants were classi ed into Braak stages V and VI.The CERAD score, which is a semiquantitative measure of neuritic plaques, was used to classify patients as neuritic plaque positive (n = 241; scores 1 and 2) and negative (n = 155; scores 3 and 4), as previously published [31].The additional clinical and pathological characteristics of the samples are presented in Supplementary Table 1.
The cellular senescence pathway enrichment score for each sample was calculated using single-sample Gene Set Enrichment Analysis (ssGSEA).Each pathway enrichment score represents the relative expression level of all genes in a speci c pathway within the sample.The cellular senescence pathway enrichment scores were analyzed to determine whether there were differences between the participant parameters of interest.Cellular senescence ssGSEA was higher in the AD dementia group than in the NCI or MCI group (p = 0.0053, AD vs. NCI; p = 0.0066, AD vs. MCI; Fig. 1A).In addition, cellular senescence ssGSEA increased with advanced Braak stages (Fig. 1B) and CERAD (Fig. 1C).The greater cellular senescence score was driven by APOE4 carriers with AD dementia compared with APOE4 carriers without AD dementia, whereas no difference was found among APOE3 carriers (Fig. 1D).Similar to the senescence ssGSEA scores, ABCA1 mRNA levels were higher in the AD dementia group than in the NCI or MCI groups (Fig. 1E).Cellular senescence ssGSEA was positively correlated with ABCA1 expression independent of the APOE genotype (p < 0.001, Fig. 1F).
To obtain cellular level information, we further analyzed single-nucleus (sn) RNA sequencing data from 2.3 million nuclei of 427 individuals from ROSMAP [33] in relation to AD neuropathology scores and ABCA1 expression.The cohort included patients without AD dementia (n = 189) and AD dementia (n = 238).Based on the APOE genotype, the participants were classi ed as APOE4 non-carriers (APOE2/2, APOE2/3, and APOE3/3) and APOE4 carriers (APOE2/4, APOE3/4, and APOE4/4).We tested whether cellular senescence is associated with AD characteristics (Global AD pathology: a quantitative summary of AD pathology derived from counts of three AD pathologies: neuritic plaques, diffuse plaques, and neuro brillary tangles; Braak stage: a semiquantitative measure of distribution and severity of neuro brillary tangle pathology; tangle density: immunohistochemistry for phosphorylated tau; overall amyloid level: immunohistochemistry for amyloid beta) and other AD risk factors, such as APOE4, sex, and ABCA1 mRNA expression.The association between the cellular senescence score and the variables of interest was examined using a linear mixed-effect model: size factor + variable + sample ID for each sub-cell type.The T-values are presented in a heat map (Fig. 1G) represent the effect size  divided by standard error of .The sample characteristics are listed in Supplementary Table 2.
AD pathological factors, such as neuro brillary tangle burden and neuritic plaque burden, were associated with increased cellular senescence ssGSEA scores in astrocytes, microglia, excitatory neurons, oligodendrocytes, and vascular cells (Fig. 1G).Sex was also associated with cellular senescence ssGSEA scores in astrocytes and inhibitory neurons (Fig. 1G).APOE4 was associated with cellular senescence ssGSEA scores in GRM3 (glutamate metabotropic receptor 3 positive) astrocytes, some excitatory neuron types, and oligodendrocytes (Fig. 1G).Importantly, ABCA1 expression was strongly associated with cellular senescence ssGSEA scores in most brain cell types including astrocytes, excitatory neurons, microglia, inhibitory neurons, oligodendrocytes, and oligodendrocyte progenitor cells (Fig. 1G).Similar pro les showed that ABCA1 expression was correlated with SASP ssGSEA scores in the same postmortem human brain samples (Supplementary Fig. 1A).Sn-RNA sequencing analysis also indicated that mTOR ssGSEA scores were positively associated with ABCA1 expression in astrocytes, excitatory and inhibitory neurons, microglia, oligodendrocytes, and oligodendrocyte progenitor cells (Supplementary Fig. 1B).AD characteristics such as Braak stage, NFT burden, and neuritic burden were also positively correlated with mTOR ssGSEA scores in excitatory neurons (Supplementary Fig. 1B).APOE4 was associated with increased mTOR expression in excitatory neurons, oligodendrocytes, and pericytes (Supplementary Fig. 1B).Taken together, these results indicate that in participants with AD dementia, particularly those with APOE4, cellular senescence scores were strongly correlated with ABCA1 expression and mTOR activation across different cell types.

Markers of cellular senescence and lysosomal ABCA1 proteins are increased in APOE4 AD human brain tissues
Lipofuscin accumulation is a marker of non-degradable lipid and protein aggregates in the cytoplasm of stressed or damaged cells and is one hallmark of cellular senescence [41].To con rm our RNA-seq ndings, we stained postmortem human brain slices (mid-frontal lobe region) from the Religious Order Study (ROS) with SenTraGor (STG, biotinylated Sudan black that detects lipofuscin) [42], anti-ABCA1 antibody, and anti-GFAP antibody (astrocyte marker) (Fig. 2A).Validation of STG and ABCA1 staining is shown in Supplementary Fig. 2A.The STG signal in GFAP-positive astrocytes was stronger in APOE3/3 and APOE3/4 carriers with AD dementia compared to APOE3/4 carriers with NCI (Fig. 2B).No difference in the STG signal was found between NCI APOE3/3 and AD APOE3/3 or between APOE3/3 and APOE3/4 in either the NCI or AD dementia groups (Fig. 2B).ABCA1 expression in GFAP astrocytes was higher in the APOE3/3 and APOE3/4 AD dementia groups compared to APOE3/3 and APOE3/4 NCI groups (Fig. 2C).No difference in ABCA1 expression was found between the APOE3/3 and APOE3/4 genotypes in either the NCI or AD group (Fig. 2C).Importantly, the STG signal in astrocytes strongly correlated with ABCA1 expression (Fig. 2D, R 2 = 0.81, p < 0.0001), con rming the association between ABCA1 expression and cellular senescence.
Next, ABCA1 protein levels were examined in proteins extracted from postmortem human middle frontal lobe tissues of individuals with NCI and AD dementia that differed by APOE4 status and who were enrolled in ROS.The tissue characteristics of the samples are listed in Supplementary Table 3.Because ABCA1 is predominantly localized in cell membranes [43], brain tissues were homogenized in TBS to remove cytosolic proteins, and proteins were extracted using 1% Triton X-100 in TBS (TBSX) (Fig. 2E).The TBSX fraction contained all membrane proteins, including plasma and organelle membranes.As expected, ABCA1 was found to be enriched in the TBSX fraction (Supplementary Fig. 2B).Total membrane ABCA1 levels were higher in APOE3/3 and APOE3/4 carriers with AD dementia than in APOE3/3 and APOE3/4 carriers in the NCI group (Fig. 2E).However, no differences in ABCA1 levels were detected between the APOE3/3 and APOE3/4 genotypes in either the NCI or AD groups (Fig. 2E, full blots are shown in Supplementary Fig. 2 C).Immunoprecipitation for LAMP-2 (a membrane marker of lysosomes) revealed that more ABCA1 was bound to LAMP-2 in the AD APOE3/4 group than in the NCI APOE3/4 group, indicating that more ABCA1 was trapped in the lysosomes in APOE4 AD (Fig. 2F).
To validate the association between ABCA1 and mTOR activation, we measured phosphorylated mTOR and ABCA1 protein levels in postmortem human brain ROS samples (Supplementary Fig. 2D).
Proteomics identify caveolin-1 as a mechanism for ABCA1 tra cking to lysosomes following ApoE4 treatment Our ndings indicated that APOE4 trapped ABCA1 in lysosomes in the human brain with APOE4 and AD dementia.We previously reported that APOE4 traps more ABCA1 in the lysosomes of astrocytes than does APOE3 [25].To explore the underlying mechanisms, we treated ABCA1-GFP overexpressing HeLa cells with recombinant ApoE3 (rE3) or ApoE4 (rE4).rE4 treatment resulted in lower total ABCA1 protein levels than rE3 treatment starting at 3 h, indicating greater ABCA1 degradation with rE4 (Supplementary Fig. 3A).The cells were then treated with either rE3 or rE4 for 4 h.Cell lysates were incubated with GFPagarose to enrich ABCA1-binding complexes for proteomic analysis using liquid chromatography-mass spectrometry (LC-MS/MS) (Fig. 3A).SDS-PAGE with silver staining con rmed the successful enrichment of ABCA1 complexes.A similar size distribution pro le of the ABCA1 complexes was observed between rE3 and rE4 treatments (Supplementary Fig. 3B).MS analysis revealed that 768 proteins among the ABCA1 co-precipitated complexes were shared by rE3 and rE4 treatments, 347 proteins were unique to the rE3 treatment, and 314 proteins were unique to the rE4 treatment (Supplementary Fig. 3C).We performed Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis of these 768 shared proteins using the Database for Annotation, Visualization, and Integrated Discovery (DAVID) platform [44] to identify speci c pathways mediating ABCA1 degradation.The top 14 pathways identi ed based on the number of enriched proteins and p-values are shown in Supplementary Fig. 3C.Subsequent experiments focused on the endocytic pathway, given its potential role in protein degradation [45,46].27 proteins were enriched in the endocytosis pathway, and nine of these proteins (Supplementary Table 4) showed greater expression in the rE4 treatment group than in the rE3 group (Supplementary Fig. 3D).
Co-IP in astrocyte lysates revealed that caveolin-1 and AP2B1 were pulled down by the ABCA1 antibody, but not by control IgG (Fig. 3C).ABCA1, caveolin-1, and AP2B1 colocalized in both the membrane and intracellular compartments of astrocytes (Fig. 3D, and 3E, respectively).To test the effects of AP2B1 and caveolin-1 on ABCA1 degradation, caveolin-1 and AP2B1 expression were reduced by siRNA in mouse primary astrocytes.Forty-eight hours after transfection with siRNA, the cells were labelled with biotin and the membrane proteins were puri ed with avidin agarose beads.Caveolin-1 and AP2B1 siRNAs reduced the expression of the target proteins (Supplementary Fig. 3E).Reducing caveolin-1, but not AP2B1, increased plasma membrane ABCA1 levels in mouse primary astrocytes (Fig. 3F), supporting that caveolae-mediated endocytosis regulated ABCA1 degradation.To test the effect of caveolin-1 on ABCA1-induced cholesterol e ux, we labeled astrocytes transfected with caveolin-1 siRNA and 3Hcholesterol and used recombinant ApoA1 or ApoE3 as cholesterol acceptors.Surprisingly, the results showed that reducing caveolin-1 expression signi cantly decreased ABCA1-mediated cholesterol e ux (Fig. 3G) despite the increase in total ABCA1 protein levels after reducing caveolin-1 expression (Supplementary Fig. 3F).Indeed, overexpression of caveolin-1 in astrocytes increased the cholesterol e ux (Supplementary Fig. 3G).These results indicate that caveolin-1 not only tra cs ABCA1 to the lysosome, but also plays a critical role in promoting ABCA1 mediated cholesterol e ux possibly by enriching the plasma membrane with cholesterol or participating in the bidirectional recycling of ABCA1 between the membrane and intracellular pools.

The association of APOE4 with increased caveolin-1 expression is a marker of cholesterol accumulation
Caveolae are vesicular invaginations of the plasma membrane that increase with cellular cholesterol accumulation [48].We tested whether cholesterol loading in two independent cell lines modulated caveolin-1 expression by adding LDL cholesterol, and veri ed cholesterol accumulation via lipin staining (Fig. 3H and 3I).As expected, caveolin-1 expression was increased after cholesterol loading in immortalized astrocytes (Fig. 3 J and 3 K) and increased in a dose-dependent manner in baby hamster kidney (BHK) broblast cells (Supplementary Fig. 3H).Because APOE4 is associated with greater cholesterol accumulation than APOE3, [49] we examined whether caveolin-1 protein levels were higher in the cortical tissues of APOE4-TR mice than in those of APOE3-TR mice at 8 and 18 months of age (Fig. 3L).In addition, cortices from 22-month-old APOE3-TR and APOE4-TR mice were fractionated using TBSX and guanidine HCL to isolate soluble membrane proteins and insoluble aggregated proteins (Supplementary Fig. 4A).Caveolin-1 protein levels in the membrane protein-enriched fraction were signi cantly higher in APOE4-TR mice than in APOE3-TR mice (Fig. 3L).No differences in caveolin-1 levels were observed in the insoluble aggregated protein fraction (Fig. 3L).In addition, higher caveolin-1 protein levels were found in the cortex of APP/PS1/APOE4 mice than in that of APP/PS1/APOE3 mice (Fig. 3M).In homogenates of human postmortem middle frontal lobe tissues from ROS, caveolin-1 protein levels were higher in APOE3/3 and APOE3/4 carriers with AD dementia compared to APOE3/3 and NCI APOE3/4 carriers in the NCI group (Fig. 3N).However, there was no difference between the APOE3/3 and APOE3/4 genotypes in either the NCI or AD group (Fig. 3N).These results support that APOE4 is associated with the accumulation of cellular cholesterol in the brain as re ected by increased caveolin-1 expression.
To determine whether astrocytes were the major cell type in which ABCA1 and caveolin-1 interacted, we isolated astrocytes from the brains of adult mice.First, the myelin fraction was separated by gradient centrifugation, followed by the removal of microglia using CD11b-coupled magnetic microbeads.
Astrocytes were found to be enriched in the residual fraction (Supplementary Fig. 4B).CD11b + cells were enriched in Iba-1 expression, consistent with the microglial population, whereas CD11b − cells were enriched in GFAP expression, consistent with the astrocyte population.Myelin fraction was enriched in beta-3 tubulin expression, consistent with enriching myelinated axons (Supplementary Fig. 4C).ABCA1 was mainly expressed in the astrocyte-enriched and axon-enriched fractions, whereas lower ABCA1 expression was detected in microglial cells (Supplementary Fig. 4D).Meanwhile, caveolin-1 was mainly expressed in astrocytes and microglia but not in axon-enriched fractions (Supplementary Fig. 4E).These results con rmed that the interaction between ABCA1 and caveolin-1 is largely a feature of astrocytes.

Lower ABCA1 function and lysosomal cholesterol accumulation induces mTORC1 activation
Cholesterol accumulation in the lysosomes of a broblast cell model via ABCA1 has been shown to activate mTORC1, which plays a role in cellular senescence and SASP.[50][51][52] To corroborate these ndings in brain cells, an ABCA1 knockdown (KD) astrocyte cell line was created using CRISPR-cas9 and cells were loaded with cholesterol.Upon cholesterol loading, the marker of mTORC1 activation p-S6K/S6K signi cantly increased in ABCA1 KD cells compared to WT (Fig. 4A).Consistent with the ABCA1 KD cell line, overexpression of ABCA1 in BHK cells using mifepristone reduced p-S6K/S6K (Fig. 4B).To validate these ndings in vivo, the cortex of 10-month-old conditional brain ABCA1 knockout mice (ABCA1 / mice crossed with Nestin Cre mice) revealed signi cantly higher p-S6K/S6K in ABCA1 brain KO mice than in WT mice (Fig. 4C).Unlike human AD brain tissues, the brain tissues of APOE4-TR mice had lower total membrane ABCA1 expression levels than APOE3.Similar to human AD brains, APOE4-TR mice brains showed greater p-S6K/S6K (Fig. 4D), and greater colocalization of LAMP1 with ABCA1 (Fig. 4E).Taken together, these ndings support that cholesterol accumulation in lysosomes from either lower ABCA1 function or ABCA1 tra cking to lysosomes induces an mTORC1 activation state.
Reducing cholesterol rescues lysosomal ABCA1 and promotes endosome-lysosome recycling Reducing cellular cholesterol by cyclodextrin reduces caveolin-1 expression [53].To test whether reducing cellular cholesterol rescues ABCA1 trapped in lysosomes, APOE4-TR mice were treated with PBS or 2-hydroxypropyl-β-cyclodextrin (HPCD) for 2 months starting at 8 months of age (n = 14 PBS-and n = 15 HPCD-treated mice, Fig. 5A).Mice body weight was not affected by HPCD treatment (Supplementary Fig. 5A).Bodipy-cholesterol staining showed that cholesterol levels in the hippocampus were lower in APOE4-TR HPCD-treated females, but not in males, compared to the PBS treatment group (Fig. 5B).Similar to bodipy-cholesterol staining, aggregated lipids stained by STG were lower in HPCDtreated females but not in males (Fig. 5C).Mouse brain tissues were fractionated using TBSX and guanidine HCl to isolate the soluble membranes and insoluble aggregated proteins, respectively.Total membrane caveolin-1 levels were signi cantly decreased in HPCD treated-APOE4-TR mice (Fig. 5D).This was associated with enhanced ABCA1 recycling, as evidenced by the higher membrane ABCA1 levels in the HPCD treatment group than in the control group (Fig. 5D).We then isolated lysosomes from mouse brains (Supplementary Fig. 5C) and con rmed that lysosomal ABCA1 levels decreased after HPCD treatment (Fig. 5E), and further con rmed lysosomal ABCA1 sequestration by double staining for ABCA1 and Lamp1 (Fig. 5F and 5G).
As increased ABCA1 activity induces small HDL formation to facilitate cholesterol e ux, [54] we measured the concentration of total HDL particles of different sizes in the cerebrospinal uid (CSF) and plasma of HPCD-treated mice using ion mobility (Supplementary Fig. 5D).The concentration of small HDL increased, while middle-sized HDL decreased in HPCD-treated mice compared to control mice in CSF (Supplementary Fig. 5E) and plasma (Supplementary Fig. 5F) although these changes did not reach statistical signi cance.
Since HPCD treatment reduced STG-stained aggregated proteins and lipids, we tested whether endolysosomal markers were increased in the membrane fraction compared to those in the aggregated cellular fractions.HPCD treatment resulted in more soluble and less aggregated levels of EEA1 (early endosomes), Rab9 (late endosomes), and LAMP1 (lysosomes) in the corresponding brain fractions (Fig. 5H).Together, these results indicated that the reduction of cholesterol accumulation by cyclodextrin reduced the proportion of ABCA1 trapped in aggregated lysosomes, restored endosome-lysosomal recycling, and allowed ABCA1 to recycle back to the plasma membrane and e ux cholesterol to HDL.

Reduction of cholesterol by cyclodextrin decreases mTORC1 activation and neuroin ammation in APOE4-TR mice
Lowering cellular cholesterol by HPCD treatment decreased the expression of mTORC1 activation markers (Fig. 6A).As the NF-kB family is located in the central axis of the SASP [5].NF-kB mRNA levels measured by qPCR were assessed after treatment.Three NF-kB members, Nfkb1, Rela, and Rel, were reduced in the brains of HPCD-treated mice compared to the control mice (Fig. 6B).In addition, the senescence and SASP markers, such as Cdkn2a (p16) and Il-6 mRNA levels, were also decreased in the brains of HPCD-treated mice compared to the control mice (Fig. 6B).In addition, activated astrocytes and microglia signi cantly decreased following HPCD treatment (Fig. 6C).To validate these ndings in human cells, we cultured human induced pluripotent stem cell (hiPSC)-derived astrocytes with the APOE4/4 genotype (Supplementary Fig. 6A) and treated them with methyl-β-cyclodextrin (MβCD) after LDL loading.As expected, cellular cholesterol accumulation was signi cantly decreased following MβCD treatment (Supplementary Fig. 6B).Importantly, IL-1β and CCL2 mRNA levels were signi cantly reduced in the MβCD treatment group compared to those in the LDL group in hiPSC astrocytes stimulated with TNF-α and IFN-γ (Supplementary Fig. 6C).Thus, reduction in cholesterol ameliorates some of the neuroin ammatory features of senescent cells.
Reduction of cholesterol by cyclodextrin increases functional neuronal markers and rescues recognition memory de cit in APOE4-TR mice Inducing ABCA1 recycling using the peptide CS-6253 in APOE4-TR mice enhanced synaptophysin and vGlut1 expression and animal [55].Similarly, we found that synaptophysin levels measured by immuno uorescence staining and western blotting were signi cantly increased in HPCD-treated APOE4-TR mice compared with controls (Supplementary Fig. 7A and 7B).PSD95 and vGluT1 (Vesicular glutamate transporter 1) were also signi cantly increased after treatment (Supplementary Fig. 7B).The novel object recognition (NOR) test performed eight weeks after treatment with HPCD (Fig. 6D) showed signi cant differences in nose touch times between the novel and old objects, and a higher discrimination index for novel objects compared to control mice (Fig. 6D).The total distance moved did not differ between the two groups of mice (Fig. 6D).

Discussion
This is the rst report to identify increased or "sequestrated" lysosomal ABCA1 expression as mechanism of cellular senescence in the AD brain.We show that ABCA1 expression strongly correlates with cellular senescence signatures of selective groups of astrocytes, excitatory neurons, microglia, and other immune cells in relation to AD neuropathology in a relatively large and well-characterized brain autopsy study.Using cellular and animal models, the accumulation of cellular cholesterol and sequestration of ABCA1 in lysosomes led to an increase in cellular senescence markers and SASP via mTORC1 activation.
Whether cellular senescence is a cause or consequence of neurodegeneration remains unclear [2], but growing evidence con rms the presence of cellular senescence in different cell types, including neurons, astrocytes, microglia, endothelial cells, and oligodendrocyte precursor cells in AD mouse models and, importantly, in AD human brains [4,10,[56][57][58][59].Our analysis is consistent with another study using single-cell transcriptomic data, which showed that excitatory neurons are the most senescent cells in the human brain [60].Aβ and Tau accumulation are major factors that induce cellular senescence in AD [8,59].Removing senescent cells genetically or pharmacologically ameliorates AD pathology and improves memory in AD mouse models [9,10].Whether cells that do not display senescent phenotypes, such as inhibitory neurons, are more vulnerable to cell death remains unknown.
Dysfunction in cholesterol homeostasis drives cellular senescence in AD.A high-fat diet increases senescent endothelial cells in LDLR (low-density lipoprotein receptor) knockout mice [61] and upregulates SASP in the kidneys of C57BL/6J mice [62].27-hydroxycholesterol accelerated cellular senescence in resident human lung cells in an in vitro assay [18].Cholesterol and lipid accumulation has been reported in cellular and mouse models of Aβ [63,64], tau [65, 66] and APOE4 [39,[67][68][69].A recent study identi ed more senescent neurons in the hippocampus of old ApoE4-TR mice than in ApoE3-TR mice owing to dysregulation of ATP and acetyl coenzyme A [14], which are substrates for cholesterol synthesis [70].Cholesterol localization shifts to lysosomes, leading to the formation of microdomains enriched for the mTORC1 scaffolding complex, priming mTORC1 for activation, and increasing proin ammatory SASP [5,71].ABCA1 plays a pivotal role in importing cholesterol into lysosomes, thereby causing lysosomal dysfunction [71].ABCA1 expression in lysosomal compartments is implicated in the accumulation of lipid droplets in astrocytes and microglia [72].ABCA1 degradation is mediated by caveolin-1, the principal protein marker for caveolae, which regulates high-density lipoprotein (HDL) metabolism and cholesterol e ux [73].Importantly, caveolin-1 can act as a cholesterol sensor and regulate cellular cholesterol homeostasis [73,74] and a potential biomarker for effective reducing sequestrated brain cholesterol pools.
Our results suggest that restoring cholesterol homeostasis is a potential therapeutic strategy for AD.A lower cholesterol burden induced by cyclodextrin increases axonal myelination and improves learning and memory in APOE4 mice [69].However, cyclodextrin is not a potent cholesterol chelator and toxic when used at higher doses.In AD mouse models, the reduction of cholesterol in astrocytes robustly reduces amyloid and tau burden [75].Clearing out of the cholesterol burden promotes the clearance of defective mitochondria via autophagy in cells and ameliorates brain in ammation and neurodegeneration in the APOE4/Tau mouse model [65,76].
Our ndings suggest that regulation of cholesterol metabolism, particular cholesterol e ux appears to be a druggable target for AD.Unfortunately, no clinical interventions focusing on modulating cholesterol metabolism have been successful in APOE4 carriers with AD.Examples of these interventions include rosiglitazone [77], pioglitazone (PPARγ agonists) [78], and bexarotene (RXR agonist) [79], which are ineffective for APOE4 carriers with AD.The challenge with such interventions is that drugs that induce both ABCA1 and APOE expression without accounting for endolysosomal dysfunction may inadvertently accelerate ABCA1 lysosomal sequestration in APOE4 or in conditions of impaired lysosomal recycling.
Our study has a few limitations.We do not identify the mechanisms by which APOE4 induced cholesterol accumulation affects lysosomal recycling.Prior studies implicate that targeting NHE6, the primary proton leak channel in the early endosome can reverse the ApoE4-induced recycling block of several receptors [80].It is plausible that cholesterol directly or indirectly affects endolysosomal pH and recycling.With the relatively small sample size of CSF obtained from mice studies, the changes in HDL particles after treatment did not reach statistical signi cance.Finally, caveolin-1 expression does not appear to be a good drug target to restore ABCA1 recycling, as reducing its expression was associated with lower cholesterol e ux.
Lower ABCA1 expression increases mTORC1 activation.(A) ABCA1 knock down astrocytes were treated with MβCD (5 mg/mL) for 2 h, followed by loading with medium (Med) or LDL (100μg/mL) for 2 h.ABCA1, total and phosphorylated mTOR, and S6K were detected by WB (n=3 biological replicates).(B) BHK cells were treated with mifepristone (0.1nM) for 16 hours to induce ABCA1 expression.The cells were then treated with LDL (100μg/mL) for 24 h.ABCA1, total and phosphorylated mTOR, and S6K were detected by WB (n=5 biological replicates).(C) The cortex of 10-month-old ABCA1 brain conditional knockout mice was homogenized in RIPA buffer.ABCA1, total and phosphorylated mTOR, and S6K were detected by WB. (n=6 for ABCA1 WT mice, n=8 for ABCA1 knockout mice of both sexes).(D) ABCA1, phosphorylated and total mTOR, and S6K in the cortex of 22-months old APOE3 and APOE4-TR mice were detected by WB (n=7 for each genotype, both sexes).(E) Representative images of LAMP1 and ABCA1 staining by immuno uorescence in the APOE3 and APOE4-TR mouse brain slides.Scar bar, 10μm.(n=9-15 LAMP1 + ROIs from 3 mice in each group).Data are represented as the mean ±SD and were analyzed using a two-tailed t-test or one-way ANOVA followed by Tukey's test.*p<0.05,***p<0.001,****p<0.0001.Cholesterol accumulation promotes senescence and neuroin ammation through lysosomal ABCA1 aggregation in APOE4 and AD cells.Compared to healthy individuals, a higher cholesterol burden in AD (including APOE4) induces more caveolin-1 expression, traps more ABCA1 in endosomes and lysosomes, decreases plasma membrane ABCA1 levels, slows down endosome-lysosome recycling, and promotes cholesterol and lipid accumulation in lysosomes, leading to senescence features and neuroin ammation through mTORC1 activation.Reduction of cholesterol by cyclodextrin reduces lysosomal ABCA1 levels and increases ABCA1 recycling to the plasma membrane to promote the e ux of cholesterol.Lowering the cholesterol burden in lysosomes reduces mTORC1 activation to decrease senescence and neuroin ammation.Illumination is created by BioRender.
AbbreviationsAD Alzheimer disease APOE Apolipoprotein E ROSMAP Religious order study/memory aging project SASP Senescence-associated secretory phenotype ABCA1 ATP-binding cassette A1 sn-RNA seq single-nucleus RNA sequencing iPSC induced pluripotent stem cells APOE-TR APOE-target replacement DLPFC Dorsolateral prefrontal cortex NCI No cognitive impairment MCI Mild cognitive impairment ssGSEA single-sample Gene Set Enrichment Analysis CERAD Consortium to Establish a Registry for Alzheimer's Disease mTOR

Figure
Figure 2

Figure 5 Reduction
Figure 5