All animal experiments were conducted in accordance with the ethical guidelines of the Korea Brain Research Institute and with the approval of the Korea Brain Research Institute Ethical Committee for Animal Experimentation (approved protocol numbers IACUC-22-00014, M2-IACUC-21-00014). Mice were housed in standard ventilated cages with ad libitum access to food and water on a standard 12-h light/12-h dark cycle. 5xFAD (also known as Tg6799, B6SJL-Tg(APPSwFlLon, PSEN1*M146L*L286V)6799Vas/Mmjax), Ift88-floxed (B6.129P2-Ift88tm1Bky/J), and Cx3cr1-cre (B6J.B6N(Cg)-Cx3cr1tm1.1(cre)Jung/J) mice were purchased from the Jackson Laboratory and C57BL/6 mice via Orient Bio. (South Korea). Adult 5xFAD and their respective Ift88-flox/flox littermates were utilized in the experiments. To examine the effect of the loss of function of the Ift88 allele in a 5xFAD background, 5xFAD; Ift88-flox/flox and 5xFAD; Ift88-flox/wt mice were used.

A recombinant AAV vector was created by combining two different plasmids: pAAV-CD68-hM4D(Gi)-mCherry, a gift from Dr. Bryan Roth (Addgene plasmid #75033), and pAAV-GFP/Cre, a gift from Dr. Fred Gage (Addgene plasmid, #49056). The EGFP-Cre fragment was amplified using a forward primer (CCG​CGG​GTC​GAC​GCC​ACC​ATG​GTG​AGC​AAG​GGC​GA) and a reverse primer (CCG​CCC​GAA​TTC​CTA​ATC​GCC​ATC​TTC​CAG​CA) using pAAV-GFP/Cre as a template. The pAAV-CD68-Cre plasmid was generated by inserting a SalI-EcoRI fragment containing the EGFP-Cre into the pAAV-CD68-hM4D(Gi)-mCherry after removal of hM4D(Gi)-mCherry using SalI and EcoRI sites. The pAAV-pCD68-Cre was propagated in Stbl3 E. coli cells (Invitrogen™ One Shot™ Stbl3™ Chemically Competent E. coli, C737303, ThermoFisher Scientific) and sequenced (Bionics) to check for correct expression.

Generation and purification of AAV-PHPeB-pCD68-Cre and AAV1 (AAV-hSyn1-EGFP-P2A-EGFPf-WPRE-HGHpA, a plasmid generously provided by Dr. Guoping Feng from Addgene, plasmid #74513) were conducted following established protocols provided by Addgene. In brief, AAV vectors, rep/cap packaging plasmids, and adenoviral helper plasmids were added to AAV-293 cells after mixing with polyethyleneimine (PEI, Polysciences, Inc.). After 72 h of transfection, the supernatant was collected and precipitated overnight at 4°C by mixing it with 40% w/v PEG-8000 and 0.4M NaCl. The resulting mixture was centrifuged at 2,500 g for 30 min at 4°C, and the pellet was resuspended in PBS containing 0.001% Pluronic F68, and 200 mM NaCl (virus buffer). Simultaneously, the cell pellet was resuspended in virus buffer and sonicated with four 1-s pulses, followed by at least 15 min of incubation on ice. The cell debris was pelleted at 3,220 g for 15 min, and the clear lysate was transferred to the tube that contained the resuspended virus. The total crude virus was subjected to iodixanol gradient centrifugation. Recovered AAVs were concentrated, and the buffer was exchanged with virus buffer using Amicon 100K columns (EMD Millipore). To determine virus titers, RT-PCRs were performed.

Mice aged 6 months were fasted for 12 h prior to surgery. For anesthesia, 150 mg/kg of 2,2,2-tribromoethanol (Sigma-Aldrich) was administered, and 1.5 percent isoflurane in combination with oxygen was used to maintain deep anesthesia. The animals were placed in a stereotaxic frame (Kopf Instruments), and the surgical site on the head was cleaned with isopropyl alcohol. Stereotaxic coordinates were determined based on the location of the bregma. Following a minor incision to expose the skull, small holes with a diameter of less than 1 mm were drilled bilaterally for virus injection. Each animal received approximately 1.5 × 10^9 viral genomes of AAV1 (AAV-hSyn1-EGFP-P2A-EGFPf-WPRE-HGHpA) injected at a rate of 30 nL/min into specific regions: the mPFC (bregma +2.0, ±0.45, dura −1.7), vDG (vDG, lambda. 0.4, 2.9, dura-2.7), and the LS (bregma. 0.4, 0.3, dura-2.6). The injection was performed using a pulled glass pipette (20–30 µm inner diameter) controlled by a Nanoject III (Drummond Scientific) to regulate the injection speed. Following the procedure, the incision was sutured, and the mice were returned to their respective home cages once they had fully recovered from anesthesia. Post-operative care, including buprenorphine injections, was provided as required by the IACUC of Korea Brain Research Institute. One month after AAV injection, brain tissues were collected and perfused with 4% paraformaldehyde for subsequent immunostaining and three-dimensional imaging.

The X-Clarity tissue clearing system II (Logo Biosystems, Korea) was employed to achieve brain tissue transparency in mouse brains 14 days after AAV injection. The brains were polymerized for a period of 3 hours and subsequently cleared for 24 hours using the X-Clarity system. Following tissue clearing, brain samples were subjected to antibody staining utilizing the DeepLabel Antibody Staining Kit (Logos Biosystems, Korea). After thorough washing with PBST, the tissues were incubated with anti-GFP antibodies (1:200) in 0.1% tween 20 in PBS at 37°C for a duration of 1-2 days. This was followed by an additional day of washing and subsequent incubation with secondary antibodies conjugated with Alexa Fluor 488 (1:400, Invitrogen, United States). To facilitate imaging, the cleared brain tissues were immersed in an RI matching solution for 24–48 h, consisting of 50% sucrose and 25% urea. For three-dimensional circuit visualization using AAV tracers, the cleared samples were imaged using a light sheet fluorescence microscope (Ultramicroscope II, LaVision BioTec GmbH, Germany) equipped with a fiber laser source (SuperK EXTREME EXW-12, NKT photonics A/S, Denmark) to generate the light sheet. The emitted light was transmitted through a 2x objective lens (MVPLAPO, Olympus, Japan) and detected by a Neo sCMOS camera (ANDOR NEO, Oxford Instruments, United Kingdom). The GFP signals of the AAV tracers were observed using a bandpass filter set with an excitation range of 470/40 nm and an emission range of 520/50 nm. The numerical aperture of the light sheet was calculated to be 0.073 in the system software, and the thickness of the light sheet was approximately 5 μm at the manufacturer’s setting. Since the axial resolution to separate two different structures parallel to the laser beam was determined by the numerical aperture of the light sheet system, the step size in the Z-direction between each image was set to 5 µm. The resulting serial TIFF image files were subsequently converted to the Imaris file format, enabling image post-processing and three-dimensional rendering using Imaris software (Bitplane, Cologne).

BV2 cells (kindly provided by Dr. Hoe, Hyang-Sook) were maintained in DMEM high glucose (Thermo Fisher) supplemented with 10% fetal bovine serum (Thermo Fisher Scientific) and 1% penicillin/streptomycin (Thermo Fisher Scientific). Subculturing of the cells was performed every 3 days, and the cells were incubated at 37°C in a humidified atmosphere with 5% CO₂.

Aβ 1-42 peptide (Bachem AG) was dissolved in sterile distilled water to achieve a concentration of 500 µM and then incubated at 37°C for 5 days (Pike et al., 1993). For the phagocytosis assay, Aβ (1-42) was labeled with pHrodo using the pHrodo iFL red microscale protein labeling kit (Invitrogen) according to the manufacturer’s instructions.

SiRNA transfection was conducted using Lipofectamine RNAiMAX Transfection Reagent (Thermo Fisher Scientific). BV cells were transfected with 80 nM of mouse siIft88 or a negative control siRNA (Bioneer). The RNA sequence of the mouse siIft88 siRNA used was ACU​GGG​AGA​GUU​AUA​CGA​U. Following siRNA transfection, BV2 cells were treated with a final concentration of 1 μM Aβ (1-42), pHrodo-Aβ (1-42), or 50 nM LysoTracker (Invitrogen).BV2 cells were seeded at a confluence of approximately 80%. The following day, cells were transfected with siRNA and treated with Aβ 6 hours after siRNA transfection. BV2 cells transfected with siRNA were exposed to Aβ for 18 h. BV2 cells were harvested or fixed at 24 h after transfection. For the analysis of the intensity of pHrodo-Aβ and lysotracker, 3 z-stack images (1.5 µm thick) were acquired per experiment using a confocal microscope (100x oil objective, Plan Apo VC 100x Oil D IC N2, Nikon A1/Ni-E). Maximal intensity projection pictures from every z-stack were created using NIS-Elements AR analysis software (Nikon). The intensity of pHrodo-Aβ and lysotracker in each BV2 cell was analyzed using NIS-Elements AR analysis software (Nikon). We pooled data from the 3 technical replicates.

The mice were perfused with 4% paraformaldehyde and subsequently post-fixed for an additional 3 h. Brain tissues were cryoprotected in a 20% sucrose solution and embedded in OCT compounds to obtain thin sections of 12-20 µm thickness. The tissue sections were then rinsed in PBS containing 0.2% Tween20 (Merck Millipore) for 10 min, followed by overnight incubation at 4 °C with primary antibodies (refer to the antibody table) diluted in 5% BSA. After washing with PBST, the sections were incubated with appropriate fluorescent-conjugated secondary antibodies (refer to the antibody table, Thermo Fisher Scientific) for 4 h at room temperature. To stain EV with PSVue-550, the tissue sections were incubated 10 µM of PSVue-550 (Molecular Targeting Technologies) and Amylo-Glo for 1 h and then washed with washing buffer (pH7.4) including 5 mM TES and 145 mM NaCl. Immunofluorescence images were captured using a Nikon Ai Rsi/Ni-E confocal microscope, and the signals were analyzed using NIS-Elements AR analysis software (Nikon).

A confocal microscope (Nikon A1/Ni-E) with laser sources (Coherent; 403 nm (power 100 mW), 457/488/514 nm (power, 40 mW), 561 nm (power 20 mW), and 640 nm (power 40 mW) were used to capture the images. The bandpass filter configured with an excitation range of 520/50 nm and an emission range of 500/50 nm was used to observe Alexa 488 signals. The bandpass filter configured with an excitation range of 595/50 nm and an emission range of 570/50 nm was used to examine Alexa 594 signals. The bandpass filter configured with an excitation range of 700/75 nm and an emission range of 663/75 nm was used to observe Alexa 647 signals. The bandpass filter was configured with an excitation range of 450/50 nm and an emission range of 400/50 nm was used to detect Amylo-Glo signals.

For the analysis of the axon spheroid near the Aβ plaque in the LS, NAc, CP, and Hb (6-month-old, n = 4/each group, female), 5 z-stack images (1.5 µm thick) were acquired per experiment using a confocal microscope (20x dry objective, Plan Apo VC 20x DIC N2 with 3x digital zoom, Nikon A1/Ni-E). Microcopy acquisition settings were kept constant within the same experiment. Maximal intensity projection pictures from every z-stack were created using NIS-Elements AR analysis software (Nikon). Axon spheroids near each Aβ plaque were counted within the area (300 μm × 300 µm).

To analyze the percentage of AC3-positive microglia and measure the length of AC3 in the LS and cortex (5-month-old, n = 4/each group, female), 10 z-stack images (1 µm thick) were acquired per experiment using a confocal microscope (100x oil objective, Plan Apo VC 100x Oil DIC N2, Nikon A1/Ni-E). Maximal intensity projection pictures from every z-stack were created using NIS-Elements AR analysis software (Nikon). We counted the microglia and measured the length of the AC3 in the area (120 μm × 120 µm), where AC3 was present in the main process of the microglia.

To count the number of Aβ plaques (D54D2) and Amylo-Glo plaques and to measure the area of Aβ plaques (D54D2) and Lamp1 in the LS (10-month-old, n = 3/each group, one male and two females), 2D images were acquired per experiment using a confocal microscope (20x dry objective, Plan Apo VC 20x DIC N2 with 3x digital zoom, Nikon A1/Ni-E). Aβ plaques (D54D2) and Amylo-Glo plaques were counted, and the area of Aβ plaques (D54D2) and Lamp1 was measured within the area (500 μm × 500 µm) using NIS-Elements AR analysis software (Nikon). To get representative images of Aβ plaques (D54D2) and Lamp1, 3 z-stack images (2.5 µm thick) were acquired per experiment using a confocal microscope (100x oil objective, Plan Apo VC 100x Oil DIC N2, Nikon A1/Ni-E). Maximal intensity projection pictures from every z-stack were created using NIS-Elements AR analysis software (Nikon).

To count the number of microglia (10-month-old, n = 3/each group, one male and two females), located within <5 µm distance from the Amylo-Glo plaques in the LS, 3 z-stack images (2.5 µm thick) were acquired per experiment using a confocal microscope (100x oil objective, Plan Apo VC 100x Oil DIC N2, Nikon A1/Ni-E). Maximal intensity projection pictures from every z-stack were created using NIS-Elements AR analysis software (Nikon). Microglia, located at < 5 µm from the Amylo-Glo plaques, were counted in each image. Data from three animals in each group was combined.

To analyze the intensity of EV on each dense plaque stained with Amylo-Glo in the LS and cortex (10-month-old, n = 3/each group, one male and two females), 4 z-stack images (2 µm thick) were acquired per experiment using a confocal microscope (100x oil objective, Plan Apo VC 100x Oil DIC N2, Nikon A1/Ni-E). Maximal intensity projection pictures from every z-stack were created using NIS-Elements AR analysis software (Nikon). We measured the fluorescence intensity of CD63, CD81, or PSVue-550 on each dense plaque stained with Amylo-Glo using NIS-Elements AR analysis software (Nikon). Data from three animals in each group was combined. Information on the antibodies used in each experiment is listed in Table 1.

The conditioned medium was collected from BV2 cells 24 h after transfection with siIft88 or siCon and subsequent treatment with Aβ. EVs present in the conditioned medium were isolated using the ExoQuick^® ULTRA EV Isolation Kit (SBI System Biosciences) following the manufacturer’s instructions. The precipitated EVs were then resuspended in HBSS buffer for NTA. For Co-IP (co-immunoprecipitation), EV pellets were resuspended in lysis buffer containing 1% Nonidet P-40 (Sigma-Aldrich), 50 mM Tris-HCl, 150 mM sodium chloride (pH 7.4), and Halt™ Protease and Phosphatase Inhibitor Single-Use Cocktail (Thermo Fisher Scientific).

NTA was conducted using the NanoSight LM10 instrument (Malvern Instruments) following the manufacturer’s user manual (NanoSight LM10 User Manual). The capture and analysis of the particles were performed using the integrated NanoSight Software NTA3.2 (Malvern Instruments). For optimal particle visibility without signal saturation, the camera level was set to 10. Most observed particles were detected using a detection threshold of 15. The NTA measurements were performed at 25°C, and each sample was measured for 60 s, five times.

BV2 cells were seeded at a confluence of approximately 80%. The following day, cells were transfected with siRNA and treated with Aβ 6 hours after siRNA transfection. BV2 cells transfected with siRNA were exposed to Aβ for 18 h.BV2 cells and CM were harvested at 24 h after transfection. Cells and EVs were lysed using a lysis buffer composed of 1% Nonidet P-40 (Sigma-Aldrich), 50 mM Tris-HCl, 150 mM sodium chloride (pH 7.4), and Halt™ Protease and Phosphatase Inhibitor Single-Use Cocktail (Thermo Fisher Scientific). The cell lysates were then subjected to immunoprecipitation by incubating them overnight at 4 C with either anti-CD63 (Sicgen) or anti-CD81 (Novus) antibodies. Subsequently, the mixtures were incubated with Dynabeads Protein G (Thermo Fisher Scientific) for 4 h at room temperature. After several washes with PBST containing 0.01% Tween20 in PBS, the immunoprecipitated proteins bound to the beads were eluted using 2x sample buffer (BioRad) for subsequent western blot analysis, or the beads containing the binding proteins were directly analyzed using LCMS.

The BCA protein assay kit (Thermo Fisher Scientific) was used to access protein concentration before using 15 μg of proteins onto 4%–20% Mini-PROTEAN^® TGX™ Precast Protein Gels (Bio-Rad Laboratories). Subsequently, the proteins were transferred to Immobilon®-P PVDF membranes (Millipore). The membranes were blocked for 30 min in a PBS solution containing 5% skim milk (BD Biosciences) and 0.1% Tween 20 (Merck Millipore). Primary antibodies, including anti-CD63 (Sicgen, AB0047-200, 1:1,000), anti-CD81 (Novus Biologicals, NB100-65805, 1:1,000), anti-GAPDH (Origene, TA802519, 1:1,000), anti-Ift88 (Proteintech, 60227-1-Ig, 1:1,000), anti-β-Amyloid (D54D2) (Cell Signaling, 8243S, 1:1,000), anti-TMEM119 (Proteintech, 66948-1, 1:1,000) anti-P2RY12 (BioLegend, 848002, 1:1,000), anti-C1Q (Abcam, ab182451, 1:1,000), anti-SPARC (Cell Signaling, 8725, 1:1,000), anti-αTub (Abcam, ab21058, 1:1,000), anti-Trem2 (Novusbio, nbp1-44067, 1:1,000), anti-CD9 (Bio-Rad, MCA2749, 1:1,000), anti-SPP1(R&D System, AF808, 1:1,000), anti-Dap12 (LSbio, ls-ls-c749774/193791, 1:1,000), anti-Apoe (Abcam, ab 1906, 1:1,000), and anti-β-actin (Origene, TA811000S, 1:1,000), were then incubated with the membranes overnight at 4°C. Following washing steps, HRP-conjugated secondary antibodies were incubated with the membranes for 30 min at room temperature. The HRP signals were visualized using an Enhanced Chemiluminescence Reagent Kit (Thermo Fisher Scientific, United States). Information about the mice used in each experiment is listed in Table 2.

Protein samples and beads containing immunoprecipitated proteins were subjected to reduction by adding 2 μL of 500 mM DL-dithiothreitol and incubated for 30 min at 55°C. Alkylation was then accomplished by adding 4 μL of 500 mM iodoacetamide and incubating at room temperature in the dark for 20 min. Trypsin digestion was carried out overnight at 37°C using an enzyme/substrate ratio of 1:50. For label-free proteomics, the samples were cleaned using a PierceTM C18 spin column (Thermo Fisher Scientific), followed by vacuum drying and storage at −20°C until further use. Prior to analysis, the dried peptides were resuspended in 20 μL of 0.1% formic acid. The dried peptides were resuspended in 20 μL of 0.1% formic acid. Nano-liquid chromatography (UltiMate 3,000, Thermo Fisher Scientific) coupled to a Q-Exactive Plus Orbitrap mass spectrometer (Thermo Fisher Scientific) was utilized for peptide analysis. All analyses utilized a binary solvent system comprising 0.1% formic acid in water and acetonitrile, respectively, as previously described (Jang et al., 2023). Peptide fractions were separated using an Ultimate 3,000 RSLCnano System (Thermo Fisher Scientific) with a PepMap 100 C18 LC column (#164535, Thermo Fisher Scientific) as a loading column, followed by a PepMap RSLC C18 (#ES903, Thermo Fisher Scientific) analytical column using 10%–50% solvent B (0.1% formic acid in ACN) in 80 min, 50%–95% solvent B in 0.1 min, a 20 min hold of 90% solvent B, a return to 10% solvent B in 0.1 min, and finally a 20 min hold of 10% solvent B. All flow rates were 300 nL/min delivered using a Easy-nLC1000 liquid chromatography system (Thermo Scientific). Solvent A consisted of water and 0.1% formic acid. Peptides were ionized with an EASY Spray source (Thermo Scientific) held at 250°C and set to 1.8 kV. For MS1 (Full MS mode), precursor ion selection scanning between 350 and 2,000 m/z at a resolution of 70,000, a AGC target of 3e⁶, and a maximum IT of 100 ms in the Orbitrap mass analyzer were utilized. For MS2 (ddMS ² mode), precursor ion selection is scanned between 350 and 2,000 m/z at a resolution of 17,500. A AGC target of 1e⁵, and a maximum IT of 100 ms in the Orbitrap mass analyzer were utilized.

Proteome Discoverer™ (PD; version 3.0) software was used for protein identification and quantification. Raw MS data were searched against a protein database obtained from the UniProt Knowledgebase Release 2022_5 (14-Dec-2022). Searches are performed utilizing the preinstalled processing workflow (PWF_Hybrid_Precursor_Quan_and LFQ_SequestHT_Percolator) and consensus workflow (CWF_Comprehensive_Enhanced Annotation _LFQ_and Precursor_Quan) as provided by PD3.0. In the processing workflow steps (PWF_Hybrid_Precursor_Quan_and LFQ_SequestHT_Percolator), peptides are required to have a minimum length of ≥6 and a maximum allowable number of reported peptides equal to 10. In the consensus workflow steps (CWF_Comprehensive_Enhanced Annotation _LFQ_and Precursor_Quan), a minimum of one peptide sequence is used as a protein filter criterion. The Protein FDR Validator node in PD3.0 was used to analyze the protein data (confidence thresholds = 0.01). This node systematically assessed the list of target proteins, calculating the FDR as it progressed from the highest to the lowest-ranked proteins. Subsequently, a protein list consisting of high-confidence candidates (FDR <0.01) was selected for the differential analyses. A protein list with abundance values obtained at FDR <0.01 was used for the downstream differential expression analyses. The abundance values of identified proteins were log-transformed and normalized according to the default algorithms of PD3.0.

Lists of identified peptides with a fold-change value and a p-value were exported as a text file. The text file was transferred to the Jupyter environment with the R kernel (version 4.3.0) and analyzed following the analysis workflow in the Differential Enrichment Analysis of Proteomics Data (Zhang et al., 2018). An adjusted p-value less than 0.05 and an absolute value of fold change larger than 0.25 were used as cut-off criteria. The abundance values of identified proteins were log-transformed and normalized according to the algorithms of PD3.0. A comparison of differentially expressed proteins between two types of experimental groups is shown in Venn diagrams (Gao et al., 2021). The differential expression protein analysis was visualized in a volcano plot using the ggplot2 package (Wickham, 2016). For further functional pathway enrichment analysis of the differentially expressed proteins, the Enrichr R package with Reactome, KEGG, and GO datasets was employed. The enriched pathways were selected based on a p-value <0.01. Cytoscape (version 3.9.1) (Shannon et al., 2003) was used to generate a network of these proteins by mapping protein data onto a PPI network. The network was constructed in Cytoscape’s public database section using the STRING database (Doncheva et al., 2019).

Mice were sacrificed by carbon dioxide gas overdose, and the brains were rapidly dissected out. The hippocampus of 3-, 6-, 9-, and 12-month-old mice, the septum of 6-month-old mice, and the Hb of 3-, 9-, and 12-month-old mice were collected from 5xFAD mice and WT mice. The samples were cut into small pieces, placed in digestion media (Neurobasal media (Gibco) containing 1 mg/mL Collagenase type 1, 2, 4 (Worthington Biochemical Corp., United States) or papain (Worthington Biochemical Corp., United States), and 20 U/mL of DNaseⅠ (Sigma)), and incubated in a shaking incubator at 37 C for 45 min. The digested tissues were mechanically dissociated and filtered through a 40 μm cell strainer. The papain was additionally inactivated by adding 20 μg/mL ovoicoid protease inhibitor (Worthington Biochemical Corp.). The single-cell suspension was centrifuged at 500 g for 15 min at 4°C and the resulting pellet was suspended in ice cold neurobasal media containing B27. This neurobasal media wash was repeated twice with two different spins at 100 g and 200 g centrifugation to remove dead cells and cellular debris. The single-cell suspension was finally suspended in an ice-cold Hanks’ balanced salt solution with 0.04% bovine serum albumin (A3294, Sigma-Aldrich, United States) at approximately 2 × 10⁵ cells/mL cell population.

The single cells were profiled using in Drop™ barcoded beads in a home-made microfluidic device connected to the Drop-seq apparatus in the Korea Brain Research Institute. Libraries from isolated single cells were generated according to the in Drop™ Protocol with the following modifications (Zilionis et al., 2017). Amplified cDNAs were not fragmented, and cDNA cleanup and size selection were performed using SPRIselect beads (B23317, Beckman Coulter, United States). Amplified cDNAs and final libraries were assessed on an Agilent BioAnalyzer using the High Sensitivity DNA Kit (Agilent Technologies, United States). The libraries were pooled and sequenced on a NovaSeq 6,000 system or a NextSeq500 system at two sequencer companies (Macrogen, Korea/LAS Science, Korea).

The Python pipeline available at https://github.com/indrops/indrops was used to read raw sequencing data with modified parameters (Bowtie 1.2.3 m = 200, n = 1, l = 15, e = 1,000) followed by alignment to the Ensembl mouse reference genome (GRCm38 release 102) to obtain a cell-gene expression matrix. In the R environment (version 4.3.0), the Seurat package (version 4.3) (Butler et al., 2018; Stuart et al., 2019; Hao et al., 2021) was used to filter cells, reduce dimensionality, cluster with Louvain, and project cells with UMAP. Cells with low quality, defined as having less than 1,000 reads and 500 UMI-filtered mapped reads, were prefiltered. Doublets were identified and removed using the ScDblFinder package (Germain et al., 2021). Using the SCTtransform wrapper, the G2M phase score, the S-phase score, the UMI count, and the proportion of mitochondrial genes were regressed out. Our samples were integrated with parts of a public dataset (GSE129788) as a reference based on L2 normalization using Harmony (Korsunsky et al., 2019) for improved data integration and cell-type identification. Initial clustering was used to group cells using the Louvain graph clustering algorithm, and cells were annotated using the topmost significant upregulated cluster-specific genes. Microglia were isolated and preferentially expressed markers of microglia or macrophages. The subset of microglia was processed through dimensionality reduction, Louvain clustering, and projection into UMAP. To examine changes in gene expression, we used the MAST (Yajima, 2023) test based on a log fold change threshold of 0.25 and an adjust p-value of p < 0.1 (about p-value <1 × 10⁻⁵) in Seurat. The weighted kernel density estimation was calculated and visualized using the Nebulosa package (Alquicira-Hernandez and Powell, 2021). The results of the differential gene expression were plotted as volcano plots using the EnhancedVolcano package (Lewis, 2023). GSEA was performed with the clusterProfiler package (Yu et al., 2012; Wu et al., 2021). We identified pathways with p-value <0.01 as significant pathways. The pathways we showed in the dot plot were plotted using the ggplot2 package (Wickham, 2016). Pseudotime analysis was performed using the Monocle3 package (Cao et al., 2019). A beginning point for the trajectory was selected manually based on the expression of canonical markers of homeostatic microglia. To analyze gene expression dynamics along the pseudotime, we used the tradeseq package (Van den Berge et al., 2020). An association test was used to determine gene expression changes along the pseudotime lineage. Next, we applied the patternTest function to evaluate the smoothed gene expression patterns along the pseudotime between the two principal trajectories. The scores of the three gene-sets previously reported were calculated by the UCell package (Andreatta and Carmona, 2021).

We reanalyzed public datasets of single-cell transcriptome profiles from previously published research. The count data from the Gene Expression Omnibus (GEO) database with the accession numbers GSE121654, GSE127893, GSE129788, GSE165306, and GSE166548 datasets (Hammond et al., 2019; Sala Frigerio et al., 2019; Ximerakis et al., 2019; Grubman et al., 2021; Safaiyan et al., 2021) were downloaded. Subsequently, all datasets were integrated using Harmony, and the standard Seurat analysis workflows were implemented as previously described.

Information on the number of animals and the number of replicates for experiment

The statistical analyses involved in the western blot analysis and the immunostaining analysis were conducted with GraphPad Prism 9 (GraphPad Software Inc.). For the statistical analysis of the transcriptome, R environments were employed, while for the proteome analysis, the PD 3.0 software was utilized. All data were presented as the mean ± standard error of the mean (SEM).

Loss of synapses in AD patients is directly associated with cognitive decline (Selkoe, 2002; Barthet and Mulle, 2020). To investigate the regional selectivity of axonal dystrophies in an amyloidopathy mouse model, we used 5xFAD mice that carry mutations in the amyloid precursor protein (APP) and PSEN1 genes leading to Aβ accumulation to introduce AAV tracers expressing EGFP (AAV1-hSyn1-EGFP-P2A-EGFPf-WPRE-HGHpA) into the ventral dentate gyrus (vDG), the lateral septum (LS), and the medial prefrontal cortex (mPFC) (Figures 1A,B). These injections targeted the major axonal tracts of the vDG to the LS, the LS to the habenula (Hb), and the mPFC to the nucleus accumbens (NAc) or the caudate putamen (CP), respectively. To visualize three-dimensional axonal projection from each injection site, we took advantage of the brain clearing and three-dimensional rendering of EGFP tracers using the mouse brain expressing the AAV tracers. The dorsal and lateral brain-wide imaging showed major axonal projections targeting the LS, the Hb, and the subcortex (Figure 1B). The coronal sections of the AAV-expressing mouse brains were used to localize the EGFP signals in dystrophic axons adjacent to the Aβ plaques in 5xFAD mice compared to the normal axonal projections in WT mice. The strong EGFP signals of the axons originally injected in the vDG, the LS, and the mPFC were observed at the LS, the Hb, and NAc/CP, respectively, in both wild-type (WT) and 5xFAD mice (Figures 1C–E). While Aβ plaques were predominantly present in the vDG, the LS, and the mPFC of 5xFAD mice, the EGFP-expressing axons exhibited distinct signals of axonal spheroids near each plaque. The vDG-LS tracts showed the most drastic Aβ plaques (D54D2, red) and axonal spheroids (EGFP) signals (Figure 1F), whereas the LS-Hb tracts did not exhibit such axonal spheroid signals (Figure 1G). In the case of the mPFC-NAc/CP tracts, the NAc showed the signals of Aβ plaques with tiny axonal spheroids, and the CP did not exhibit any detectable signals (Figure 1H, CP not shown). We quantified the number of axon spheroids within each Aβ plaque in the terminal region of the axonal tracts and found that the LS of the vDG-LS tracts had a significantly higher number of axonal spheroids compared to the other tracts (Figure 1I). These results show that the vDG neurons projecting to the LS are susceptible to forming an axonal spheroid structure close to Aβ plaque.

Microglia are phagocytic cells that engulf a variety of debris, including Aβ plaques, myelin fragments, apoptotic cells, and extracellular harmful substances. This diversity of phagocytic targets is reflected in the heterogeneous microglial subtypes, each with its own unique set of receptors and signaling pathways (Ribeiro Xavier et al., 2015; Krasemann et al., 2017; Butler et al., 2021; Clayton et al., 2021; Huang et al., 2021). As demonstrated in stereotaxic experiments, the vDG-septum tract of 5xFAD mice implies the presence of microglia that are associated with these axonal abnormal structures in the septum. To profile the regional heterogeneity of microglia, including the septum, we performed single-cell transcriptome profiling isolated from the hippocampus, the septum, and the Hb of WT and 5xFAD mice using an inDrop™ platform (Klein et al., 2015). We captured a total of 144,185 cells with 38,075 feature genes, from 27 independent libraries (Figure 2A). The cell types of 50 clusters were annotated by examining marker genes and comparing them to those in a previous study (Ximerakis et al., 2019). A subset of microglia among these clusters was identified with the most significantly upregulated genes (Cst3, Ctss, C1qa, C1qb, B2m, Supplementary Figure S1). A total of 1,660 microglia were projected to new UMAP dimensions, and clustered into the twelve microglia clusters that were named MG-01 to MG-12 (Figure 2A bottom).

The microglia in WT mice were primarily characterized as homeostatic, expressing genes encoding transmembrane protein 119 (Tmem119) (Satoh et al., 2016), P2Y purinoreceptor 12 (P2ry12), and CX3 chemokine receptor 1 (Cx3cr1) (Butovsky et al., 2014). The MG-02 cell cluster had dense expression of the Tmem119 and P2ry12 genes, indicating that it is a homeostatic microglia-like cell cluster (Figure 2B, Supplementary Figure S2A,B). Microglia from WT and 5xFAD mice overlapped across the seven microglia clusters. Of the seven identified clusters, the MG-01 cell cluster demonstrated more elevated expression of Apoe (Supplementary Figure S2C,D), along with Trem2 and Cst7 genes previously known as DAM genes (Keren-Shaul et al., 2017; Krasemann et al., 2017). These findings suggested that the MG-01 cluster was comprised of DAM. The distribution of Axl and Dkk2 expression indicated that the MG-03 and the MG-10 cell clusters were identified as activated response microglia and/or a DAM cluster (Aghaizu et al., 2023). The MG-04, the MG-06, the MG-07, and the MG-09 cell clusters were distinguished by their expression of hypoxia-inducible factor-1α (Hif-1a). The MG-03 and the MG-10 cell clusters consisted primarily of microglia isolated from the hippocampal region, while the other four cell clusters, including the MG-01, consisted primarily of microglia isolated from the septum. This result suggests that there were regional differences in the microglial transcriptome and in the distribution of DAM. Interestingly, microglia from the septum, the hippocampus, and the Hb regions in the MG-01, the MG-04, the MG-07, and the MG-09 cell clusters overlapped regardless of age (Supplementary Figure S2A,B).

It is assumed that homeostatic microglia in the normal and healthy states progressively undergo a transition into DAM in a pathological condition via extracellular stimuli. The application of pseudotime analysis to the scRNAseq data has revealed the existence of five distinct trajectories with the microglia clusters. Two primary trajectories have been identified, originating from the MG-02 cell cluster and then bifurcating toward either the MG-01 (Route A) or the MG-03 cell cluster (Route B, Figure 2C). These trajectories were consistent with the spatial distribution of regional differences in DAM described earlier. To characterize the progress of hippocampus-enriched DAM (Route B) and septum-enriched DAM (Route A), we looked for the genes whose expression varied significantly over pseudotime within each trajectory. We found 1,912 genes in Route A, 76 genes in Route B, and 159 genes in both routes (Figure 2D). Given that both routes originated from a common MG-02 cell population, certain genes with different expression patterns may explain the characteristics of hippocampus-enriched DAM and septum-enriched DAM among the 159 genes found to be significantly associated with pseudotime. The Trem2 gene, as a canonical marker of DAM, exhibited comparable trends of slight fluctuations over time in both primary trajectories (Figure 2E). This aligned with Trem2 expression, which is initially increased in DAM. In contrast, it was observed that out of the 159 genes that were previously identified, 26 genes exhibited significantly different expression patterns in Route A and Route B (Figure 2F). The genes Kif5a, Cys1, and Armc2 were commonly associated with ciliary assembly (Tao et al., 2009; Novas et al., 2018; Hesketh et al., 2022). The finding indicates that the variation in the expression of genes associated with cilia may be one of the contributing factors to the regional heterogeneity observed in both routes.

Microglia isolated from WT and 5xFAD mice were similarly distributed across seven microglial clusters, as determined by UMAP dimensionality reduction analysis of the microglial subtypes. The findings suggested the presence of a microglial subtype in the brains of WT mice with gene expression features related to the DAM subtype identified in 5xFAD mice. To test whether the analysis of microglial single-cell transcriptome datasets supports these findings, the five publicly available datasets (Hammond et al., 2019; Sala Frigerio et al., 2019; Ximerakis et al., 2019; Grubman et al., 2021; Safaiyan et al., 2021) were acquired and integrated using the same method used in the previous analysis (Supplementary Figure S3A). This allowed us to identify a total of 43 clusters of microglia. To analyze marker gene expression in the microglial subclusters, we selected 9 clusters consisting of more than 1% microglia from 5xFAD mice and APP/PS1 mice (Supplementary Figure S3B). We further selected 10 clusters that had markers for Spp1, Trem2, Axl, Lpl, Lgals3, and Gpnmb, and 7 clusters that had markers for Cx3cr1, P2ry12, and Tmem119. This down-sampling process produced a total of seventeen clusters containing 36,013 cells (Supplementary Figure S3C). We examined the canonical marker gene expression in the microglial subclusters (Supplementary Figure S3D). This led to the identification of two microglial subtypes in the WT mouse brains that exhibited characteristics like DAM gene expression, supporting our previous findings using scRNAseq data from the septum.

Our previous scRNAseq analysis identified three microglial clusters in the mouse brain that exhibited comparable expression of DAM signature genes, such as MG-01 in Route A, MG-03, and MG-10 in Route B. These clusters all expressed DAM signature genes such as Apoe, Dkk2, and Axl (Figure 3A). To ascertain the similarity of DAM stages between the Route A and the Route B cell populations, we compared the gene expression profiles of these cell populations in WT and 5xFAD mice (Figure 3B). We found that both cell populations had two common DAM signature genes (Apoe and Ctdb) and three common axon tract microglia (ATM) signature genes (Apoe, Ctdb, and Ctds), which were significantly upregulated in 5xFAD mice. The Route B cell populations, which are predominantly hippocampus-derived microglia, had three DAM signature genes (Ftl1, Lyz2, and Fth1) that were significantly upregulated in 5xFAD mice. To determine if there were microglial subtypes in the WT mouse brain that exhibited DAM features in our data containing a regional distribution, we examined microglia derived from WT brains. Gene score values were calculated using known 14 DAM genes (Ank, Apoe, Axl, Ccl6, Cd63, Cd9, Csf1, Cst7, Ctsz, Igf1, Itgax, Lpl, Spp1, Tyrobp), 20 axon tract microglia (ATM) genes (Spp1, Gpnmb, Igf1, Lgals3, Cd9, Fabp5, Lpl, Syngr1, Pld3, Ctsl, Lgals1, Lilrb4a, Ccl9, Anxa5, Gm1673, Csf1, Cd63, Gm10116, Anxa2, Apoe), and 20 interferon-responsive microglia (IRM) signature genes (Ccrl2, Csprs, Dhx58, Ifit1, Ifit2, Ifit3, Ifit3b, Ifitm2, Ifitm3, Irf7, Isg15, Isg20, Ly6a, Oas1a, Oas1g, Oas3, Oasl1, Oasl2, Stat2, Usp18) (Figure 2C) (Hammond et al., 2019; Lall et al., 2021). The MG-02 cell cluster enriched in WT mice, which was considered the homeostatic microglia, had relatively low gene scores for DAM and ATM. The MG-01 cell cluster from WT mice, with a high proportion of septum-derived microglia, maintained high gene score values for DAM and ATM but had the lowest values for IRM. In contrast, the MG-03 cell cluster from WT mice, which was dominated by microglia from the hippocampus, had relatively high gene score values for DAM and ATM and the highest value for IRM. Similarly, the MG-10 cell cluster from WT mice showed high gene score values for DAM, ATM, and IRM. This may imply that WT microglia are regionally heterogeneous and transcriptionally primed to respond closely to the pathological needs of the cells comprising the tissues.

Subsequent differential expression gene analysis was conducted on the MG-01 and the MG-03 in comparison to the MG-02 in WT mice (Figure 3D, Supplementary File S1). The analysis yielded a list of genes associated with cilia that exhibited significant alterations in expression (with adjust p-value <0.1 and an average absolute Log₂ Fold change >0.25) based on data from CiliaCarta (van Dam et al., 2019). Only the Kif5a gene among those cilia-related genes exhibited downregulation in differential expression analysis between the MG-01 and MG-02 cell clusters in wild-type mice. However, it was discovered that 14 cilia-related genes overlapped with the differentially expressed genes in the MG-03 cell cluster in WT mice as compared to the MG-02 cell cluster, indicating the presence of ciliary dysfunction. To gain insights into the differential enrichment of signal pathways among regional microglial subtypes, we performed gene-set enrichment analysis (GSEA) using the differentially expressed genes. GSEA, based on the Gene Ontology (GO) and the Reactome databases, revealed a notable increase in genes related to cilium organization within the MG-03 cell cluster (Figure 3E, Supplementary File S1). On the other hand, the cluster of MG-01 cells displayed an upregulation of diverse vesicle-related genes such as secretory granules, lytic vacuole, and cytoplasmic vesicle. Additionally, we observed a decrease in genes related to immune system, lysosome, and endosome within the MG-01 cell cluster. These results suggest that the regionally heterogeneous microglial subtypes may be involved in distinct stages or priming status as determined by cilium-, vesicle-, and secretion-related genes.

The previous scRNAseq analysis of microglial subclusters in different brain regions revealed that cilium-related genes may regulate the status of microglia. To examine the role of the microglial primary cilia in a pathological condition such as amyloidopathy, we compared the expression of adenylate cyclase type III (AC3), a well-established cilia marker found in primary cilia throughout the brain (Bishop et al., 2007), in microglia from 6-month-old WT and 5xFAD mice in the LS and the cortex. We identified AC3-labeled primary cilia localized in the region between the cell body and the major process in microglia expressing Iba1. In the LS, both WT and 5xFAD mice showed microglia with primary cilia, regardless of Aβ plaques. However, in the cortex of 5xFAD mice, AC3 was not detected in microglia (Figures 4A,B). The percentage of AC3-positive microglia in the LS and the cortex was significantly lower in 5xFAD mice compared to WT mice. Furthermore, microglia in 5xFAD mice exhibited shorter primary cilia in the same brain region (Figure 4B). Since the primary cilia face outside cells, the protruding structure could be damaged by extracellular toxic substances such as Aβ. These results indicate microglia in the cortex are more likely to be affected in their primary cilia by extracellular amyloidopathy. To examine the involvement of primary cilia in microglial status, we used BV2 cells, a mouse microglial cell line, to downregulate the expression of Ift88 using siRNA transfection. Subsequently, we examined protein markers associated with both homeostatic and activated microglial states. The knockdown of Ift88 led to a decrease in the expression of homeostatic microglial cell markers, such as Tmem119, C1q, and Sparc, except for P2ry12, which exhibited a significant induction (Figure 4C). Additionally, the expression of protein markers associated with activated microglial cells, including Trem2, CD9, Spp1, and Apoe, was reduced upon siIft88 treatment (Figure 4D). These findings suggest that Ift88 or primary cilia may be involved in triggering or maintaining the activated status of microglia such as DAM.

Microglia in the DAM state exhibit upregulation of genes involved in EV secretion as well as phagocytosis of Aβ and dystrophic neurites. To assess whether the primary cilia of microglia regulate the biogenesis and secretion of EVs, we evaluated the protein expression levels of CD63 and CD81, a member of the tetraspanin family known to be specifically localized in EVs and crucial for EV biogenesis and secretion (Berditchevski and Odintsova, 2007). In BV2 cells transfected with siIft88, the levels of CD81 and Ift88 were significantly reduced in the cellular lysate. However, the expression of CD63 remained unchanged upon decreased Ift88 in BV2 cells (Figures 5A,B). Indeed, it is intriguing to note that upon the silencing of Ift88, a discernible elevation in CD63 expression was observed, accompanied by a statistically significant augmentation in CD81 levels (Figure 5C,D). These findings suggest that primary cilia may be involved in EV secretion and EV biogenesis in microglia. EV secretion can be influenced by lysosomal overload, and the removal of toxic substrates may alleviate cellular damage (Eitan et al., 2016; Hessvik and Llorente, 2018). To investigate microglial phagocytosis and lysosomal clearance, BV2 cells were treated with Aβ (1 µM) following transfection with either scrambled siRNA or siIft88. Remarkably, Aβ uptake was significantly augmented in BV2 cells transfected with siIft88 (Figure 5E). Furthermore, the release of Aβ within EVs was also enhanced (Figure 5F). To assess lysosomal activity in siRNA-transfected BV2 cells, we employed pH-sensitive dye-conjugated Aβ (pHrodo-Aβ) and LysoTracker. Silencing Ift88 in BV2 cells resulted in a greater proportion of cells positive for pHrodo-Aβ and LysoTracker compared to controls (Figure 5G). Moreover, the intensity of pHrodo-Aβ and LysoTracker signals was significantly increased in BV2 cells transfected with siIft88 (Figure 5H, I). These findings suggest that Ift88 deficiency may induce aberrantly heightened lysosomal activity while promoting the secretion of EVs containing Aβ in microglia.

To characterize the proteomic changes in microglia in response to Aβ treatment, we performed proteomic analysis of BV2 cells with the combination of Ift88 inhibition and treatment with Aβ. Figure 6A illustrates the experimental workflow utilized for proteomic analysis via LC-MS. Most proteins were identified in both lysates (88%) and EVs (85%) between the (siCon + Aβ) group and the (siIft88 + Aβ) group (Figure 6B). In the lysates of BV2 cells transfected with siIft88 and treated with Aβ, we observed 168 upregulated proteins (log2 fold-change ≥0.25, adjust p-value <0.05) and 126 downregulated proteins (log2 fold-change ≤ −0.25, adjust p-value <0.05) (Figure 6C). To gain insight into the functional implications of the differentially expressed proteins, we performed GSEA, which revealed significant alterations in various biological processes. The upregulated proteins were enriched in terms associated with Alzheimer disease, phagosome, ER-phagosome, ER-to Golgi anterograde, positive regulation of exocytosis, among others. Conversely, the downregulated proteins were enriched in terms such as interferon α/β signaling, autodegradation of Cdh1 by Cbh1:APC/C, APC/C:Cdc20 mediated degradation of Securin, and more (Figure 6D, Supplementary File S2). EV marker proteins such as Tsg101 were induced by Ift88, indicating the involvement of the primary cilia in the regulation of EV biogenesis and secretion. Overall, the enrichment analysis results suggest that the loss of Ift88 may induce an AD-associated state in microglia, stimulating secretion and decreasing phagocytic vesicle function. In the EVs of BV2 cells transfected with siIft88 and treated with Aβ, we observed that 76 proteins were upregulated (log2 fold-change ≥0.25, adjust p-value <0.05), while 196 proteins were downregulated (log2 fold-change ≤ −0.25, adjust p-value <0.05) (Figure 6E). The upregulated proteins were predominantly associated with cellular response to stress, regulation of cellular response to stress, cytokine signaling in immune system, and ribosome biogenesis. On the other hand, the downregulated proteins were enriched in terms related to autophagosome maturation, regulation of phagocytosis, lysosomal lumen, and vesicle cytoskeletal trafficking (Supplementary File S2). Representative proteins were depicted on the volcano plot, using distinct colors to denote their association with specific processes, such as dark red for cellular response to stress, bright red for cytokine signaling in immune system, green for vesicle cytoskeletal trafficking, and blue for autophagosome maturation, regulation of phagocytosis, and lysosomal lumen (Figure 6E, F). These findings suggest that the loss of Ift88 may induce proteomic alterations in EVs secreted by microglia, characterized by an increase in cellular response to stress and cytokine signaling in immune system and a decrease in autophagosome maturation and vesicle cytoskeletal trafficking. To validate the increase in vesicles, because of positive regulation of exocytosis, we performed nanoparticle tracking analysis (NTA) on EVs derived from BV2 cells transfected with siIft88 and treated with Aβ. Figure 6G illustrates the size distribution of EVs with and without Aβ treatment. The concentration of EVs was significantly increased in BV2 cells transfected with siIft88 and treated with Aβ, supporting the previous proteomic analysis (Figure 6H).

EVs exhibit heterogeneity, and their proteomic contents provide insights into the signaling cascade mediated by these vesicles and the alterations in extracellular protein homeostasis, also known as, extracellular proteostasis (Geraghty et al., 2021; Krause et al., 2022). As demonstrated in previous findings (Figures 5B,D), the loss of Ift88 resulted in a decrease in CD81 levels in the lysate but an increase in CD63 and CD81 levels. To investigate the specific proteomic alterations in EVs associated with CD63 and CD81, we conducted a proteomic characterization of CD63-or CD81-decorated EVs in the lysate and EVs secreted from BV2 cells transfected with siIft88 and treated with Aβ, utilizing co-immunoprecipitation (Co-IP) with anti-CD63 or anti-CD81 antibodies. Figure 7A illustrates the experimental workflow for analyzing CD63-and CD81-binding proteins through liquid-chromatography mass-spectrometry (LCMS). Representative results of the Co-IP experiment showed that CD63 was enriched in the elution of the CD63-Co-IP, both in the lysates and EVs, and Aβ was also detected in the CD63-binding proteins (Figure 7B), indicating a potential interaction between Aβ and CD63 in BV2 cells and their secreted EVs. Most CD63-binding proteins were identified in both the lysates (89%) and EVs (72%) when comparing the (siCon + Aβ) group with the (siIft88 + Aβ) group (Figures 7C,D). In the lysates of BV2 cells transfected with siIft88 and treated with Aβ, 96 proteins were upregulated (log2 fold-change ≥0.25, adjust p-value <0.05), while 127 proteins were downregulated (log2 fold-change ≤ −0.25, adjust p-value <0.05) among the CD63-binding proteins (Figure 7C). In EVs, 75 proteins were identified as upregulated proteins (log2 fold-change ≥0.25, adjust p-value <0.05), and 8 proteins were identified as downregulated proteins (log2 fold-change ≤ −0.25, adjust p-value <0.05) among the CD63-binding proteins (Figure 7D). In the lysates, the upregulated proteins associated with CD63 were involved in processes such as Rab regulation of trafficking, intracellular protein transport, cellular response to stress, and Fc gamma R-mediated phagocytosis. The downregulated proteins associated with CD63 were related to selective autophagy and secretory granule lumen (Figure 7E, Supplementary File S3). In EVs, the upregulated proteins associated with CD63 were linked to phagosome, cytoplasmic vesicle lumen, and extracellular vesicle. The downregulated proteins associated with CD63 were involved in micro GTPase and RHOBTB3, signaling by Rho GTPase and vacuolar lumen (Figure 7E, Supplementary File S3). In the lysates of BV2 cells transfected with siIft88 and treated with Aβ, we observed that 147 proteins were upregulated (log2 fold-change ≥0.25, p-value <0.05 adjust p-value <0.05), while 113 proteins were downregulated (log2 fold-change ≤ −0.25, adjust p-value <0.05) among the CD81 binding proteins. Furthermore, in EVs, 76 proteins were identified as upregulated (log2 fold-change ≥0.25, adjust p-value <0.05), and 13 proteins were identified as downregulated (log2 fold-change ≤ −0.25, adjust p-value <0.05). Regarding the CD81-binding proteins in the lysates, the upregulated proteins were associated with Rab regulation of trafficking, phagocytosis vesicle, and cytoplasmic vesicle lumen, vesicle. Conversely, the downregulated proteins among the CD81-binding proteins were involved in processing of capped intron -containing pre-mRNA, regulation of focal adhesion assembly, and peroxisomal membrane (Figure 7F, Supplementary File S4). On the volcano plots, representative proteins were marked with distinct colors, such as dark red for AD, bright red for the secretory granule lumen, pink for the interleukin-12-mediated signaling pathway, green for focal adhesion, and blue for mRNA processing (Supplementary Figure S4A). In the EVs, the upregulated proteins associated with CD81 were linked to phagosome, phagosome pathway, cargo trafficking to periciliary membrane, vesicle, extracellular vesicle, and cytoplasmic vesicle lumen. The downregulated proteins among the CD81-binding proteins were associated with immune system, post-translational protein modification, and metabolism of proteins (Figure 7F, Supplementary Materail S4). On the volcano plot, representative proteins were marked with distinct colors, such as dark red for phagosome, bright red for cargo trafficking to periciliary membrane, faint red for vesicle, green for the metabolism of proteins and post-translation protein modification, and blue for immune system (Supplementary Figure S4B). Figure 7G,H summarize the protein-protein interaction (PPI) network among the up- or downregulated proteins of CD63 or CD81 binding proteins in each cell lysate and EV. These findings suggest that the loss of Ift88 may lead to an upregulation of stress response-related proteome in the extracellular space via EVs altering the extracellular proteostasis. The changes in the EV proteome and extracellular proteomic composition may affect the amyloidopathy in AD.

To further investigate the role of microglial primary cilia in the development of AD, we utilized 5xFAD mice carrying the conditional Ift88-flox allele (5xFAD; Ift88-flox/flox). The chemokine receptor Cx3cr1 serves as a selective marker for microglia in the central nervous system. To examine the effects of endogenous Ift88 loss in microglia, we generated Cx3cr1-cre; Ift88-flox/flox; 5xFAD mice by crossing Cx3cr1-Cre mice (Yona et al., 2013) with 5xFAD; Ift88-flox/flox mice (Supplementary Figure S5A). We investigated the characteristics of Aβ plaques and dystrophic neurites in the LS of 3-month-old mice when the Aβ plaques started to occupy the region. The number of small-sized Amylo-Glo plaques (<50 µm) significantly increased, and there was a noticeable trend of increased Aβ plaques marked with D54D2 (Supplementary Figure S5B). Furthermore, the fluorescence intensity of GFAP and Iba1 was significantly elevated (Supplementary Figure S5C, D), indicating elevated gliosis possibly from the promoted EV secretion containing altered EV proteins from Ift88-inhibited microglia. To avoid the possible non-specific inhibition of Ift88 in the Cre-driver mouse line (Zhao et al., 2019), we also employed an intracranial injection of adeno-associated virus (AAV) carrying Cre recombinase under the control of the CD68 promoter (AAV-pCD68-Cre), which is active in activated microglial cells (Bodea et al., 2014). Figure 8A illustrates an experimental outline to investigate the function of microglial Ift88 in the axon terminal region of the vDG-LS tract and its impact on the pathogenesis of AD. We induced the loss of Ift88 in 5xFAD mice at 9 months of age through stereotaxic injection of AAV-pCD68-Cre. Immunohistochemical analysis of the LS in 5xFAD; Ift88-flox/flox mice injected with AAV-pCD68-Cre revealed pronounced Lamp1 signals, indicative of dystrophic neuronal membranes including axons, compared to 5xFAD mice injected with a sham control (Figures 8B,C). Notably, the LS of 5xFAD; Ift88-flox/flox mice injected with AAV-pCD68-Cre exhibited a significant increase in Aβ plaques (D54D2) and Amylo-Glo stained plaques, specifically in the small (<50 µm) and large (>200 µm) size ranges (Figures 8C,D). Considering that microglia lacking Ift88 tended to engulf and secrete more EVs containing Aβ, we hypothesized that the enlarged plaques may be contributed by microglial secretion, which also induced neuritic dystrophy adjacent to the plaques. Indeed, staining for Lamp1-positive dystrophic neurites adjacent to Aβ plaques revealed a significant increase in dystrophic neurites in 5xFAD mice with conditionally inhibited Ift88 expression in microglial cells (Figure 8D). These findings suggest that the loss of Ift88 in microglia may contribute to the increased occurrence of diffused and fibrillar Aβ plaques as well as dystrophic neurons affecting the axon projection. The toxic substances may also affect the dystrophy and recruitment of microglia to the Aβ plaques. Quantifying microglia near each plaque enabled us to examine the recognition and interaction of microglia with plaques in both 5xFAD and 5xFAD mice with Ift88-deficient microglia. We observed a reduced number of microglia located within 5 µm of the plaque in 5xFAD; Ift88-flox/flox mice injected with AAV-CD68-Cre compared to 5xFAD mice (Figures 8E,F). This finding suggests that the loss of microglial Ift88 affects the sensing and recruitment of microglia towards Aβ plaques. The facilitated EV secretion in Ift88-deficient microglia could aggravate the amyloidopathy via dampening of microglial primary cilia function and altering the extracellular proteostasis.

To investigate the dysregulated accumulation of EVs on the Aβ plaques in the cortex and the septum, we conducted immunohistochemistry analyses using 5xFAD (sham) and 5xFAD; Ift88-flox/flox mice injected with AAV-pCD68-Cre, employing anti-CD63 and anti-CD81, as well as the vesicle-specific dye PSVue-550. Interestingly, the intensities of CD63, CD81, and PSVue-550 were significantly elevated within the Aβ plaques in the LS of 5xFAD; Ift88-flox/flox mice injected with AAV-pCD68-Cre relatively to 5xFAD (sham) (Figures 8G,H). The markers of EVs such as CD63 and CD81 were also elevated in the cortical Aβ plaques after the inhibition of Ift88 with slight decreased signals of PSVue. Even though cortical microglia may behave differently than septal microglia, it is reasonable to assume that the lipid composition of the cortical plaques may confer resistance to PSVue staining, given that the EV markers were elevated in Aβ plaques in the LS, just as in the cortical plaques. These results provide compelling evidence supporting our initial idea that the absence of Ift88 in microglia may potentiate the secretion of EVs, potentially contributing to increased plaque formation.