Decreased Amyloid-β Pathologies by Intracerebral Loading of Glycosphingolipid-enriched Exosomes in Alzheimer Model Mice*

Background: Exosome, a type of extracellular vesicles, can associate with Aβ in vitro. Results: Intracerebrally injected exosomes trapped Aβ on surface glycosphingolipids and transported it into microglia in AD mouse brains, resulting in reductions in Aβ pathology. Conclusion: Exogenous exosomes act as potent scavengers for Aβ in mouse brains. Significance: The findings provide a novel therapeutic approach for AD. Elevated levels of amyloid-β peptide (Aβ) in the human brain are linked to the pathogenesis of Alzheimer disease. Recent in vitro studies have demonstrated that extracellular Aβ can bind to exosomes, which are cell-secreted nanovesicles with lipid membranes that are known to transport their cargos intercellularly. Such findings suggest that the exosomes are involved in Aβ metabolism in brain. Here, we found that neuroblastoma-derived exosomes exogenously injected into mouse brains trapped Aβ and with the associated Aβ were internalized into brain-resident phagocyte microglia. Accordingly, continuous intracerebral administration of the exosomes into amyloid-β precursor protein transgenic mice resulted in marked reductions in Aβ levels, amyloid depositions, and Aβ-mediated synaptotoxicity in the hippocampus. In addition, we determined that glycosphingolipids (GSLs), a group of membrane glycolipids, are highly abundant in the exosomes, and the enriched glycans of the GSLs are essential for Aβ binding and assembly on the exosomes both in vitro and in vivo. Our data demonstrate that intracerebrally administered exosomes can act as potent scavengers for Aβ by carrying it on the exosome surface GSLs and suggest a role of exosomes in Aβ clearance in the central nervous system. Improving Aβ clearance by exosome administration would provide a novel therapeutic intervention for Alzheimer disease.

Elevated levels of amyloid-␤ peptide (A␤) in the human brain are linked to the pathogenesis of Alzheimer disease. Recent in vitro studies have demonstrated that extracellular A␤ can bind to exosomes, which are cell-secreted nanovesicles with lipid membranes that are known to transport their cargos intercellularly. Such findings suggest that the exosomes are involved in A␤ metabolism in brain. Here, we found that neuroblastomaderived exosomes exogenously injected into mouse brains trapped A␤ and with the associated A␤ were internalized into brain-resident phagocyte microglia. Accordingly, continuous intracerebral administration of the exosomes into amyloid-␤ precursor protein transgenic mice resulted in marked reductions in A␤ levels, amyloid depositions, and A␤-mediated synaptotoxicity in the hippocampus. In addition, we determined that glycosphingolipids (GSLs), a group of membrane glycolipids, are highly abundant in the exosomes, and the enriched glycans of the GSLs are essential for A␤ binding and assembly on the exosomes both in vitro and in vivo. Our data demonstrate that intracerebrally administered exosomes can act as potent scavengers for A␤ by carrying it on the exosome surface GSLs and suggest a role of exosomes in A␤ clearance in the central nervous system. Improving A␤ clearance by exosome administration would provide a novel therapeutic intervention for Alzheimer disease.
Alzheimer disease (AD), 2 a common dementia, is pathologically characterized by the presence of amyloid-␤ peptide (A␤)containing senile plaques within the brain. In familial AD, genetic mutations cause increased production of A␤ (1), whereas in far more common sporadic cases, A␤ generation is normal, but its clearance is impaired (2). Elevated levels of A␤, caused by an imbalance in its metabolism, are linked to synaptic and nerve loss, which likely manifest as progressive cognitive deficits in AD (3).
Exosomes represent a subtype of secreted membrane vesicles (40 -100 nm in diameter) of endosomal origin that are released from various types of cells including neurons (4). Exosomes serve to remove and discard unwanted proteins into a drainage system; they are also known to intercellularly shuttle their cargo: a specific set of proteins, RNAs, and lipids (5). Recently, exosomes were reported to associate with a portion of extracellular amyloid-␤ precursor protein (APP) and its metabolites, including C-terminal fragments (CTFs), amyloid intracellular domain (AICD), and A␤, in cultures of human wild-type or mutant human APP-expressing neuroblastoma cells (6,7). In addition, exosomal proteins such as Alix and flottilin-1 were identified around neuritic plaques in AD brains (6). Similarly, our previous study demonstrated that exosomes released from neuroblastoma or primary cortical neurons can bind to synthetic or endogenous A␤ and promote A␤ fibril formation on their surface in vitro (8). Furthermore, exosome-bound A␤ is incorporated into microglia for degradation, suggesting that exosomes may act as a mediator for A␤ elimination in brains (8). Here, we demonstrated that long term intracerebral admin-* This work was supported by Creation of Innovation Centers for Advanced istration of exosomes to the brain of APP transgenic mice resulted in a marked reduction in A␤ levels, amyloid depositions, and A␤-mediated synaptotoxicity. We also clarified that glycosphingolipids (GSLs) abundant in the exosomes were essential for A␤ binding on the exosome surface.

EXPERIMENTAL PROCEDURES
Cell Cultures-Murine neuroblastoma Neuro2a (N2a) cells were maintained in Dulbecco's modified Eagle's medium (Invitrogen) supplemented with 10% fetal bovine serum. The murine microglial cell line BV-2 was purchased from National Cancer Institute (Istituto Nazionale per la Ricerca sul Cancro, Genova, Italy) and was cultured in RPMI1640 (Invitrogen) supplemented with 10% fetal bovine serum and 2 mM L-glutamine.
Animals-All animal experiments were conducted under a protocol approved by the animal care committees of Hokkaido University. Wild-type C57BL/6 mice were purchased from Japan SLC Inc. (Hamamatsu, Japan). Heterozygotic transgenic mice that express the human APP bearing the Swedish and Indiana (KM670/671NL, V717F) mutations (APP sweInd or J20 strain) were from The Jackson Laboratory (Bar Harbor, ME) and maintained in barrier facilities.
Exosome Isolation-Exosomes were prepared from culture supernatants of N2a cells as described previously (9). Briefly, 1 day before exosome isolation, culture medium was replaced with serum-free medium. The culture supernatants were collected and sequentially centrifuged at 3000 ϫ g for 10 min, 4,000 ϫ g for 10 min, and 10,000 ϫ g for 30 min to remove cells, dead cells, and debris then spun again at 100,000 ϫ g for 1 h to obtain exosomes as pellets.
For sucrose gradient analysis, each exosome pellet (100 g of protein) was loaded onto 10 ml of a sucrose gradient (0.25-2.3 M sucrose in 20 mM HEPES) and centrifuged at 100,000 ϫ g for 18 h. After centrifugation, 1-ml fractions were collected, diluted with 20 mM HEPES, and precipitated by centrifugation for 1 h at 100,000 ϫ g. The resulting pellets were resuspended in PBS and subjected to Western blot analysis.
Electron Microscopy-Exosomes (100 g of protein/ml) were resuspended in 50 mM Tris, 150 mM NaCl buffer (pH 7.6) (TBS), and applied to a grid covered with collodion. For A␤ binding experiments exosomes (100 g of protein/ml) were incubated with A␤ 1-42 (15 M) in TBS at 37°C for 5 h after pretreating with or without EGCase. Exosome mixtures were then applied to the grid. Exosomes were negatively stained with 2% phosphotungstic acid. Transmission images were acquired using an HD-2000 (Hitachi, Tokyo, Japan) or JEM-1400Plus (JEOL Ltd. Tokyo, Japan) transmission electron microscope.
Dynamic Light Scattering-Exosomes (untreated or treated with EGCase) were suspended in TBS at 100 g of protein/ml. The particle size of the exosomes was measured by dynamic light scattering using a DelsaNano HC (Beckman Coulter).
Injection and Isolation of Biotinylated Exosomes-Exosomes were biotinylated with EZ-link sulfo-NHS Biotin (Pierce) according to the manufacturer's protocol with minor modifications. Briefly, the exosomes were suspended in PBS (150 g of protein/ml) and incubated with biotin reagent (1 mg/ml) at room temperature for 30 min. The biotinylated exosomes were isolated by ultracentrifugation at 100,000 ϫ g for 1 h at 4°C and resuspended in PBS. For detecting co-precipitated A␤, 2 l of the 5 g/l biotin-exosome solution was injected into the right hippocampus of APP mouse (4 months old) using stereotaxic coordinates and kept for 3 h. The biotinylated exosomes in the homogenates of the hippocampus were precipitated using streptavidin microbeads according to the protocol in a MACS streptavidin kit (Miltenyi Biotech, Bergisch Gladbach, German). Co-precipitated A␤ was analyzed with Western blotting or enzyme-linked immunosorbent assay (ELISA) following the solubilization of the exosomes in SDS sample buffer or guanidine buffer, respectively.
SDS-PAGE and Western Blotting-SDS-PAGE and Western blot analysis were performed according to the standard methods of Laemmli. To detect target proteins, we employed as a primary antibody monoclonal antibodies against Alix (BD Bioscience), APP CTFs (Sigma), actin (Sigma), A␤ (6E10, Signet, Dedham, MA), or neprilysin (Santa Cruz Biotechnology) or rabbit polyclonal antibodies against flottilin-1, endothelin converting enzyme-1 (ECE-1), (Santa Cruz Biotechnology), or insulin degrading enzyme (IDE, Abcam) and as a secondary antibody an anti-mouse IgG-HRP antibody(GE Healthcare), or anti-rabbit IgG-HRP antibody (GE Healthcare). To detect the ganglioside GM1, we used horseradish peroxidase-conjugated cholera toxin B subunit from Sigma. Bands were visualized using a combination of an ECL Plus kit (GE Healthcare) and an LAS4000 imaging system (Fuji Film, Tokyo).
Fluorescence Labeling for the Exosomes-Exosomes were stained with the red fluorescence dye PKH26 (Sigma) as described previously (8). Briefly, the exosomes were resuspended in diluent C (Sigma) and incubated with PKH26 at room temperature for 5 min. The reaction was stopped by the addition of 1% bovine serum albumin. The PKH26-labeled exosomes were precipitated again by ultracentrifugation at 100,000 ϫ g at 4°C for 1 h.
Exosome Isolation from Murine Cerebrospinal Fluid (CSF)-CSF was collected from the cisterna magna of 2-month-old C57BL/6 mice as previously described (10). Exosomes were isolated from the CSF using a method similar to that described above for isolation from culture medium.
Analysis of Exosomal Particle Number-A qNano System (Izon Science, Ltd) was employed to analyze the particle densities of N2a-and mouse CSF-derived exosomes resuspended in PBS.
Exosome Administration in Mouse Brains-Mice were continuously treated with exosome solution (2 mg protein/ml) or vehicle (PBS) by Alzet minipump (model 1002) at 0.25 l/h for 14 days. Mice were placed in a stereotactic instrument (NARISHIGE, Tokyo, Japan), and stainless steel cannulas of Alzet Brain Infusion Kit3 were implanted into the right lateral ventricle (mediolateral, Ϫ0.8 mm; dorsoventral Ϫ3.0 mm) or hippocampus (anteroposterior, Ϫ2.0 mm; mediolateral, Ϫ1.3 mm; dorsoventral Ϫ2.2 mm). After a 14-day infusion, mice were then transcardially perfused with cold heparin/PBS. The right hemibrain was fixed with 4% paraformaldehyde, PBS at 4°C for 48 h for use in immunohistochemistry, and the left hemibrain was rapidly frozen with liquid nitrogen and stored at Ϫ80°C for later analysis.
For single injection studies, PKH-labeled exosomes or a conjugate of PKH-exosomes with fluorescent A␤ (4 g of exosome protein in 2 l PBS) were injected into the right hippocampus or the lateral ventricle of non-transgenic mice using stereotaxic coordinates as described above. To obtain the conjugates of the exosomes with A␤, PKH-exosomes (100 g/ml) were incubated with A␤  in TBS at 37°C for 24 h then centrifuged at 100,000 ϫ g for 1 h to remove free A␤. At 3 or 24 h post injection, the mouse brains were prepared as described above for immunohistochemistry.
Immunohistochemistry-The tissue sections were cut with a cryostat (Leica CM3050S) and post-fixed with 4% paraformaldehyde, PBS. After blocking with 5% bovine serum albumin (BSA), 16-m-thick sections were immunostained with monoclonal antibodies against Iba1 (Wako), ␤III tublin (Promega), or glial fibrillary acidic protein (SHIMA laboratory) followed by visualization with AlexaFluor488-conjugated anti-IgG. Serial 30-m thickness of brain sections were immunostained with monoclonal antibody against A␤ (4G8, Covance) after a brief formic acid treatment, and the signals were visualized using ABC elite kit (Vector Laboratories). Confocal images were obtained using an Olympus Fluoview FV10i microscope. The A␤ plaques were estimated as the percentage of the immunopositive area (positive pixel) to the examined area (total pixel) using ImageJ software.
A␤ ELISA-A␤ levels were determined using a sandwich ELISA. The kits for A␤ 1-40 and A␤ 1-42 were obtained from Wako (Osaka, Japan), and that for A␤ 1-38 was from IBL (Gunma, Japan). Mouse hippocampus or exosomes were homogenized in 4 M guanidine-HCl buffer (pH 8.0) with an ultrasonic homogenizer (TAITEC, Saitama, Japan). After incubation at room temperature for 3 h, the homogenates were further diluted with 0.1% BSA, PBS and centrifuged at 16,000 ϫ g for 20 min. The resulting supernatants were then applied to the ELISA. All samples were measured in duplicate.
Evaluation of Synaptic Densities-Synaptophysin-immunoreactive synaptic densities were quantified according to the methods of Mucke et al. (11) with minor modifications. The right hemisphere of each APP mouse brain was sagittally cut into 16-m-thick sections using a freezing microtome. The serial sections were incubated with a monoclonal antibody against synaptophysin (D35E4, Cell Signaling) followed by incubation with AlexaFluor488-bound anti-IgG. Immunofluorescent signals were visualized using an Olympus Fluoview FV10i microscope. The linear range of the synaptophysin-positive fluorescence intensities in nontransgenic control sections was determined, and the same setting was used to analyze all of the following images. For each mouse 9 confocal images were captured in three sections per the right hemisphere of the brain, and each image covered an area 5500 m 2 in the molecular layer of the dentate gyrus. The synaptophysin-immunoreactive synaptic densities were estimated as a percentage of the immunostained area (positive pixel) to the selected image area (total pixel) using ImageJ software.
Thioflavin-S Staining-Brain sections (30 m thick) were oxidized with 0.25% potassium permanganate for 20 min followed by 3 min of bleaching in 2% potassium metabisulfite and 1% oxalic acid. Sections were stained with 0.015% thioflavin-S in 50% ethanol in the dark for 10 min. After developing in two changes of 50% ethanol for 4 min each, images were captured with Olympus Fluoview FV10i microscope, and thioflavin-Spositive plaques were counted in three sections per mouse hippocampus.
Measurement of GSLs-The extraction of GSLs from the culture cells and the exosomes, the enzymatic digestion of GSLglycans with EGCase I and II (Takara Bio, Shiga, Japan), and the purification of glycans using glycoblotting were performed as described previously (12). Purified GSL-glycans were analyzed by MALDI-TOF MS using an Ultraflex II TOF/TOF mass spectrometer equipped with a reflector, which was controlled by the FlexControl 3.0 software package (Bruker Daltonics, Bremen, Germany). All spectra were obtained as positive ions and were annotated using the FlexAnalysis 3.0 software package (Bruker Daltonics). The glycan structures were then identified by online database SphinGOMAP.
Proinflammatory Cytokine ELISA-The levels of proinflammatory cytokines, including tumor necrosis factor-␣ (TNF-␣), interleukin-6 (IL-6), IL-1␤, and interferon (IFN-␥), were determined by ELISA (Multi-Analyte ELISArray, Qiagen) according to the manufacturer's instructions. Briefly, each APP mouse hippocampus was homogenized in 4 M guanidine-HCl buffer (pH 8.0) using an ultrasonic homogenizer (TAITEC). After an incubation at room temperature for 3 h, the homogenates were further diluted in 0.1% BSA, PBS and centrifuged at 16,000 ϫ g at 4°C for 20 min. The resulting supernatants were applied to the ELISA. All samples were handled in duplicate.
EGCase and Sialidase Treatment-Exosomes (1 mg of protein/ml) were incubated with 0.5 units/ml EGCase II (Takara Bio Inc., Shiga, Japan) at 37°C for 15 h in PBS containing 20 mM HEPES (pH 7.4) or 1 unit/ml of sialidase from Clostridium perfringens (Sigma) at 37°C for 16 h in 50 mM acetate buffer (pH 5.5). Each mixture was centrifuged at 100,000 ϫ g for 1 h. The resultant precipitates were resuspended in TBS or Hepes-buffered saline and used for further examination.
Seed-free A␤ Preparation-Seed-free A␤ solutions were prepared essentially according to a published report (13).
A␤ Binding Assay-PKH26-labeled exosomes (untreated or treated with EGCase) were plated on chamber slideglass (Thermo Fisher Scientific, Waltham, MA) by staying in PBS for 1 h at room temperature. Fluorescence-labeled A␤  (1 M) was then added into the chamber of cultured N2a cells or the labeled exosomes and co-incubated in serum-free medium at 37°C for 5 h. After a wash with PBS to remove free A␤, fluores-cent images were captured using Olympus Fluoview FV10i microscope.
Thioflavin-T Assay-Seed-free A␤ 1-42 solutions (15 M) were incubated at 37°C for various times with the exosomes (100 g of protein of exosomes in 100 l of TBS) that had been untreated or treated with EGCase. Fluorescence intensities of Thioflavin-T (Sigma) were determined as described before (8) using an Appliskan spectrofluorophotometer (Thermo Fisher Scientific).
Exosome Uptake Assay-Uptake of PKH26-labeled exosomes into microglial BV-2 cells were measured as described previously (8). Briefly, fluorescent exosomes were administered to BV-2 cells and incubated for various times in serum-free conditions. After a wash, the cells were then fixed, and confocal images were acquired using an Olympus Fluoview FV10i microscope. The fluorescence intensity of each sample was analyzed with ImageJ software.

Exogenously Injected Exosomes Trap A␤ and Are Incorporated into Microglia in Mouse
Brains-To explore the hypothesis that providing exogenous exosomes in vivo would enhance A␤ clearance by facilitating engulfment of exosome-bound A␤ by microglia, we loaded isolated exosomes into mouse brains and evaluated the effect on the balance of A␤ metabolism. Exosomes were collected from culture supernatants of mouse neuroblastoma N2a cells using sequential ultracentrifugation; the exosomes typically consisted of membrane vesicles of 70 -120 nm in diameter (Fig. 1, A and B) as previously described (14). The presence of exosomes was confirmed by detecting the exosomal markers Alix and flotillin-1 as well as the membrane glycolipid GM1 ganglioside in sucrose density gradient fractions corresponding to a density of 1.12 and 1.16 g/ml (Fig. 1C). A␤ was not identified in N2a cells and the exosomes used in this study (Fig. 1C). Exosomes isolated from N2a cells were biotinylated and injected into the hippocampus of APP SweInd transgenic (APP) mice. Hippocampal A␤ was detectable in streptavidin-precipitated exosomes together with the marker Alix 3 h after the injection (Fig. 1D). Accompanied by A␤, the intrahippocampal-injected exosomes co-localized with the microglial marker Iba1 (Fig. 1E), agreeing with our previous in vitro study (8). These results demonstrate that in mouse brains exogenous exosomes bind A␤ and are then incorporated into microglia together with the bound A␤ for degradation.
Continuous Exosome Administration Ameliorates A␤ Pathology and Synaptic Dysfunction in APP Mouse Brains-we next continuously administered exosomes into the lateral ventricles of 4-month-old APP mice for 14 days using osmotic minipumps. The exosome solution (2 mg of protein/ml) contained 2.67 ϫ 10 12 particles/ml, which is ϳ35 times higher than the concentration of the exosomes in mouse CSF (7.51 ϫ 10 10 particles/ml). The intraventricularly injected exosomes have been reported to penetrate into brain parenchyma (15). After 2 weeks, there was an approximate 50% reduction in total A␤  and A␤ 1-42 levels and 35% reduction in A␤ 1-38 levels in the hippocampus of the exosome-treated APP mice compared with those infused with vehicle ( Fig. 2A). A similar decline in A␤ levels was confirmed by direct administration of the exosomes into mouse hippocampus, with ϳ50% less observed in the ipsi- Exosomes were isolated from culture supernatant of N2a cells by sequential centrifugation, eventually to 100,000 ϫ g pellets. A, an electron microscopic image of phosphotungstic acid-stained exosomes. B, exosomes were measured by dynamic light scattering. C, exosomes were density-fractionated by sucrose gradient, and the fractions were analyzed by Western blotting to detect the exosome markers Alix, flotillin-1 (Flot-1), and ganglioside GM1 (GM1) as well as actin and A␤. CL, cell lysates. D, biotinylated exosomes stereotaxically injected into the hippocampus of 4-month-old non-transgenic or APP mice were isolated using biotin binding affinity beads, then analyzed by Western blotting to detect co-precipitated full-length (FL)-APP, A␤, and Alix. E, conjugates of fluorescence-labeled A␤ (green) and exosomes (red) were administered into the hippocampus of nontransgenic mice. The hippocampal images were captured 3 h after the injection, following antibody staining for the microglial marker Iba1 (blue). Scale bar, 50 m and 10 m (inset). lateral injection side than the contralateral (Fig. 2B). APP mice given vehicle treatment exhibited ϳ50% decreased densities in synaptophysin immunoreactivities in the hippocampus as compared with wild-type mice (Fig. 2, C and D), agreeing with previous reports (11). After a 14-day infusion of the exosomes, the synaptophysin immunoreactivities were markedly increased in the APP mouse hippocampus (85 and 50% of those observed in WT mice after exosome and PBS treatment, respectively; Fig. 2D). Thus, the exosomes mediated significant recovery from synaptic impairment in the APP mice. Taken together, these findings validated that in vivo, exogenously added exosomes induce reductions in A␤ levels and A␤-associated synaptotoxicity.
To assess the effect of exogenous exosomes on brain amyloid deposition, we continuously administered the exosomes into the hippocampus of 13-month-old APP mice for 14 days. We found that the exosomes markedly decreased the A␤ immunoreactive burden (65% reduction; Fig. 3, A and B) and number of thioflavin-S-positive plaques (38%; Fig. 3C) in the treated hippocampus compared with the untreated side. Supporting these histological results, tissue levels of A␤ 1-40 and A␤ 1-42 were also significantly decreased after exosome infusion, as determined by ELISA (Fig. 3D). We also performed exosome infusion into the lateral ventricles of 13-month-old APP mice for 14 days and found significant reductions in A␤ 1-40 and A␤ 1-42 levels in the mouse hippocampus (Fig. 3E). These findings demonstrate the evident efficacy of long term treatment with exogenous exosomes in A␤ deposition, even deposition of the fibrillar species of A␤ aggregates, in APP mice.
To exclude the possibility that the changes observed in vivo are due to effects of exosomes on A␤ generation and enzymatic degradation, we examined the expression levels of APP and its cleaved fragments, including CTF-␣ and CTF-␤, in the hippocampus. No obvious differences in the expression levels were apparent between the exosome-and PBS-treated APP mice (Fig. 4A). In addition, the expression levels of well known proteases for A␤ peptides, including insulin-degrading enzyme, neprilysin, and angiotensin-converting enzyme, were investigated, but there were no variations even after exosome infusion (Fig. 4A). Although the exogenous exosomes were taken up by the microglia, there was no obvious activation in releasing the proinflammatory cytokines like TNF␣, IL-6, IL-1␤, or IFN-␥ in the exosome-treated mouse hippocampus (Fig. 4B). The above findings indicate that the exosomes reduce A␤ levels by promoting an alternative pathway for A␤ clearance, i.e. exosomeassociated A␤ uptake by microglia, without any stimulation in APP processing, A␤ degradation, or microglial inflammatory reactions.
Exosomal GSL-glycans Are Critical for Their Association with A␤ in Vitro and in Vivo-The above experiments clearly indicate that the exogenous exosomes associated with the A␤ in the mouse brains, although how they interact with each other remained to be determined. Increasing evidence with synthetic liposomes or membranes has demonstrated that gangliosides (sialic acid-containing GSLs) clustering on the membrane surface bind to A␤ and that this A␤-GSL complex then acts as a template for catalyzing the reaction of A␤ fibril formation (16,17). In vivo, the monosialoganglioside GM1 was found to associate with A␤ in human brains that exhibit AD pathology (18,19). GSLs are reported to exist in exosomes (6), but details regarding which species of GSLs have not yet been described. We summarized the profiles of GSL-derived glycans from the exosomes and the cells from which the exosomes derived by quantitative GSL-glycomics ( Table 1). The total amount of GSLs was much higher in exosomes than in the parent cells (ϳ2300%, Fig. 5A). In either, the vast majority of GSLs were GM2 (Ͼ84%), with distinct minor compositions (Table 1). Except for GM2, the levels of sialylated GSLs were much higher in the exosomes (Fig. 5B). To investigate whether the GSLs abundant in the exosomes might affect A␤ binding and fibril formation, we deglycosylated exosomal GSLs using EGCase, which specifically cleaves the linkage between the oligosaccharide and glucosylceramide in GSLs (20). The particle size of the exosomes was stable up to 5 h after EGCase digestion, although larger aggregates commonly formed during the 24-h incubation (Fig. 5C). Within the 5-h period, the EGCase-treated exosomes associated very little with the A␤, as compared with intact exosomes, which overtly colocalized with the A␤ (Fig.  5D). We performed surface plasmon resonance studies to evaluate the specificity of the interaction between N2a-derived exosomes and individual A␤s. As shown in Fig. 5E, when the exosomes were injected onto the immobilized A␤ 1-42 and A␤ 42-1 , a peptide with reverse sequence of A␤ 1-42 , only the former gave a significant increase in resonance signal, demonstrating the specific nature of the interaction. Specific interactions of the exosomes were observed not only with immobilized A␤ 1-42 but also with A␤ 1-40 and A␤ 1-38 , and we found that the interactions were almost completely diminished when the exosomes were pretreated with EGCase (Fig. 5F). These results suggest that A␤s directly binds to the exosomes through the GSL glycans, particularly those sialic acid moieties on their surface. Pretreatment with EGCase also inhibited exosome-dependent amyloid fibril formation in incubation mixtures of exosomes and A␤ 1-42 , as assessed by thioflavin-T assay and electron microscopic observation (Fig. 6, A and B). Cleavage of sialic  acids with sialidase resulted in similar reductions in fibril formation (Fig. 6C). Steric blocking of GM1 or GM2 ganglioside by cholera toxin subunit B or anti-GM2 antibody, respectively, also partially but significantly suppressed amyloid formation (Fig. 6C). In contrast to intact exosomes, the EGCase-treated exosomes nearly failed to coprecipitate with A␤ when injected into the hippocampus of APP mice (Fig. 6D). Our data verify that cleavage of GSL glycans from exosomes can sufficiently prevent the association of exosomes with A␤. This suggests that there may be multiple species of GSLs, especially sialylated forms, on exosomal membranes that organize into unique sites of high potency able to induce A␤ binding and assembly. In addition to gangliosides, cholesterol and SM are also known to promote A␤ assembly via the lateral packing of gangliosides on membranes (21,22). We found that both cholesterol and SM were highly abundant in exosomes compared with their parent cells (Fig. 7), suggesting that high densities of these two lipids would promote GSL binding to A␤. Another lipid, Cer, which is the hydrophilic backbone of GSLs, is known to be involved in exosome generation (23). We found higher levels of ceramide in exosomes than in cells (Fig. 7), consistent with a previous report (23).
Exosomes Are Incorporated into Microglia in Vitro and in Vivo, in a GSL-glycan-independent Manner-Our previous in vitro experiments demonstrated that engulfment of exosomes by mouse primary microglia occurred in a partially phosphatidylserine-dependent manner (8). However, deglycosylated proteins, e.g. immunoglobulin FC receptor, reportedly exhibit low affinity toward microglia (24). To determine whether cleavage of GSL-glycans would affect microglial uptake of exosomes, we exposed fluorescent-labeled exosomes pretreated with EGCase or PBS to microglial BV-2 cells. We found no decrease in microglial uptake of EGCase-treated exosomes (Fig. 8A). Both of the labeled exosomes were co-localized with the microglial marker Iba1 when intracerebrally injected into mouse brains (Fig. 8B), and there were no obvious differences observed between untreated or EGCase-treated exosomes. Few fluorescent exosomes were apparent in merged images of cells stained for either a neuronal or astroglial marker (Fig. 8C). The above data indicate that exosomes can be incorporated into microglia undisturbed by the absence of GSL-glycans on the membrane.

DISCUSSION
Our study presented here clearly demonstrated that intracerebral exosome infusion leads to a decrease in A␤ levels and ameliorates A␤-related pathologies in APP mice. In the mouse brains, A␤ was trapped at the exosome surface by glycan moieties of GSLs and transported into microglia for degradation. Mass spectrometry-based analysis has revealed that GSLs are abundant in exosomes compared with parental cells. Just how GSLs are packed so much more into the exosomes than in parental cells remains an unanswered question. Exosomes are produced by intraluminal budding of the limited membrane of endosomes. Accumulation of Cer, which is generated by the hydrolysis of SM, has been reported to initiate the budding (23). Cer can induce a coalescence of small microdomains into larger microdomains to drive domain-induced budding of biological membranes (25), which results in highly loading Cer and its vicinal lipid molecules into the generated vesicle. Indeed, Cer and SM are concentrated in exosome membranes (Fig. 7) (23). In addition, SM forms a distinct membrane domain, namely a lipid raft, in the plasma membrane together with GSLs and cholesterol (26). Various raft-resident proteins have also been reported to be abundant in exosomes (27). In the present study, providing GSLs-enriched exosomes to the APP mouse brains resulted in recovering synaptic impairment and decreasing A␤ plaques. However, the effect of GSLs on AD pathogenesis is a controversial issue. GSL storage disorders, which are subtypes of lysosomal storage diseases caused by genetic dysfunction in GSL catabolism, share pathological features with AD, such as A␤ burden (28,29). Accumulated gangliosides are observed in human brains exhibiting AD, and they are proposed to contribute to AD development through promoting A␤ fibril formation (16). These discrepant effects of GSLs are likely to stem from their life span in brain tissues. Pathologically accumulated GSLs are pooled within cells to form complexes with A␤ and its polymer (19,28), which might be retained to exert neuronal damages. On the other hand, exosomal GSLs capture A␤ in extracellular fluid and are rapidly taken up by phagocytes without persistent harm to the brain.
Our present study demonstrated that exosomes derived from N2a cells can promote A␤ fibril formation on their surface (Fig.  6, A and B). The exosome-bound A␤ was then incorporated into microglia for degradation (Figs. 1E and 8, A and B) (8). Therefore, continuous infusion of exosomes induced a reduction in amyloid depositions in aged APP mouse brains (Fig. 3C). These results provide a notion that the exosomes in the brains are rapidly cleared by microglia before the exosome-bound A␤s form amyloid fibrils for depositions. Amyloid plaques were reported to change their sizes over days in the brains of AD model mice (30). The exogenously added exosomes might prevent further A␤ depositions by blocking the supply of the soluble A␤. Alternatively, exosomes might support the clearance of amyloid deposit, which already formed. The complex of GM1-A␤ has been reported to localize at the ends of extended A␤ fibrils in the incubation mixture of GM1 and A␤ (19). Alix, a marker for exosomes was enriched around the small A␤ plaques in brain sections from AD patients (6). Similarly, once the exogenous exosome-associated A␤s are attached to the A␤ fibrils in the amyloid plaques, they might provoke microglia gathering toward the plaques and accelerate their clearance.
Here, we used seed-free A␤ to perform the A␤ binding assay (Fig. 5D) and the surface plasmon resonance analysis (Fig. 5E and F) and demonstrated that A␤s directly bind to the exosomes through the GSL glycans on their surface. Seed-free A␤ was reported to contain soluble species of A␤ but not insoluble amyloid forms (13). The GM1-A␤ complex, which acts as a seed for A␤ amyloidogenesis, is known to consist of a clustered GM1 and a monomeric A␤ molecule (17,31). Accordingly, our previous report has demonstrated that the exosomes derived from N2a cells almost prevented the oligomeric A␤ formation from seed-free A␤, but not those preformed A␤ oligomers, which are recognized by A11, a specific antibody against oligomer (8). Thus, the above findings suggested that the exosomes released . Cleavage of exosomal GSL-glycans does not affect their uptake by microglia. A, fluorescence-labeled exosomes (untreated as Ctrl or treated with EGCase) were exposed to microglial BV-2 cells for 3 h, and the fluorescence intensities of exosomes taken up into the cells were determined by confocal microscopy. B, representative hippocampal sections of non-transgenic mice (4 months) injected with fluorescence-labeled exosomes (untreated as Ctrl or treated with EGCase, red) and stained with anti-Iba1 antibody. Arrows indicate significant exosomal fluorescence in Iba1-positive microglia. Scale bars, 50 and 10 m (insets). C, hippocampal sections from non-transgenic mice (4 months old) injected with fluorescence-labeled exosomes (untreated as Ctrl or treated with EGCase) stained with antibodies against the neuronal marker ␤III tubulin or the astroglial marker glial fibrillary acidic protein (GFAP). Bar, 50 m. from N2a cells would be mostly associated with monomeric A␤ through their surface GSLs. However, a recent study demonstrated that soluble A␤ oligomer strongly binds to GM1-containing membranes in vitro and in vivo, and GM1-bound A␤ is detected in human cerebrospinal fluid (32). An additional investigation may be required in the future to clarify which form of A␤s can be associated with the exosomes.
It is worth noting that other aggregate-prone proteins, including ␣-synuclein and prion protein, which cause Parkinson and Creutzfeldt-Jakob diseases, respectively, are also associated with exosomes (33,34). In addition, ␣-synuclein and prion protein have been reported to associate with GSLs on the surface of synthetic liposomes (35,36). A challenging subject of future studies will be determining whether exosomes are involved in the clearance of these proteins.
The normal phagocytotic function of microglia is conceivably important for exosome-bound A␤ clearance in this study. Increasing evidence has indicated that a large portion of secreted exosomes is convincingly taken up by microglia (8,37). In contrast, a small amount of exosomes can be incorporated into neurons (38). If the clearance function of microglia is decreased or absent, then the exosome-bearing aggregateprone proteins would trigger pathological events (i.e. formation of senile plaque) or even perform minor interneuronal transfer to propagate their toxic assemblies. Indeed, exosome-associated prion proteins, in which their folded species are infectious, can spread between neuronal cells in a monoculture system (33). The transmissibility of amyloids, a characteristic feature of many neurodegenerative diseases including Alzheimer disease and spongiform encephalitis, might emerge under a lack of glial activity for removing exosomes. Improvement of A␤ clearance by exosome administration or enhancement of exosome generation provides a novel therapeutic approach for AD therapy. It is noteworthy that the A␤-degrading enzymes, insulin-degrading enzyme and neprilysin, have been reported to be found in exosomes secreted from microglia and adipose tissue-derived mesenchymal stem cells, respectively (39,40). Exosomes have been used as a delivery platform, encapsulating reagents or siRNAs (41,42). Peripheral injection of the exosomes holding siRNA (against an APP-processing enzyme, BACE1) succeeded in brain targeting and specific gene knockdown in mice (41). In the future, development of engineered nanovesicles that regulate multiple processes in AD pathogenesis might be a valuable tool for the therapy.