Microglia-derived tiny vesicles coordinate two cell-cleaning processes for Alzheimer's treatment

To further investigate whether toxic proteins contribute to autophagy dysfunction, we established an Aβ1‒42 oligomer-induced HT22 cell model in vitro. Aβ1‒42 treatment reduced the LC3BII/LC3BI ratio but had no effect on p62 or Lamp2A expression levels (Fig. 2c, d). These results suggest that CMA and macroautophagy impairment may not be solely caused by toxic Aβ proteins. Instead, as previous studies have suggested,¹⁴ autophagy dysfunction may precede Aβ deposition, whereas Aβ1‒42 oligomers may accelerate macroautophagy impairment.

Given the observed dual impairment of macroautophagy and CMA in AD, therapeutic strategies that restore both pathways may help counteract disease progression. However, a major challenge in autophagy-targeting therapies is efficient drug delivery across the BBB, which severely limits the bioavailability of many promising modulators. To overcome this, we sought to develop a biomimetic nanodrug delivery system capable of synchronizing CMA and macroautophagy activation while efficiently crossing the BBB. To achieve this goal, we developed Microglia-Liposome Fusion Extrusion (MiLi-FE), a bioengineering strategy for generating microglia-derived nanovesicles that encapsulate and codeliver therapeutic agents.

In the MiLi-FE process, AR7 and rapamycin, both highly hydrophobic, were separately encapsulated into liposomal formulations (AR7@LP and Rapa@LP) to increase their solubility and delivery efficiency. Microglia were incubated with these drug-loaded liposomes, facilitating internalization, followed by extrusion through nanoporous membranes to generate AR@ENV nanovesicles that retain microglial membrane properties while incorporating encapsulated drugs (Fig. 2e). miLi-FE improves drug loading efficiency while preserving the innate targeting capabilities of microglia, enabling the precise delivery of autophagy modulators across the BBB.

Characterization of AR7@LP and Rapa@LP revealed particle sizes of 160.3 ± 3.4 nm and 127.3 ± 2.6 nm, respectively (Supplementary Fig.), with polydispersity indices (PDIs) of ~0.2, indicating a uniform size distribution. The encapsulation efficiencies were 92.9 ± 1.3% (AR7@LP) and 84.5 ± 1.5% (Rapa@LP), while the drug loading efficiencies were 1.30 ± 0.012% and 3.91 ± 0.58%, respectively (Supplementary Table). Zeta potential analysis revealed negative surface charges of −37.9 ± 4.6 mV (AR7@LP) and −30.8 ± 1.5 mV (Rapa@LP) (Supplementary Fig.). Transmission electron microscopy (TEM) imaging confirmed a well-defined phospholipid bilayer structure with a regular spherical morphology (Supplementary Fig.). 1a, b 1 1c, d 1a, b

The optimal drug loading time of 8 h was determined through comprehensive evaluation via a CCK8 assay, liposome‒lysosome colocalization analysis, and HPLC quantification of drug loading efficiency (Supplementary Figs. 2, 3 and Supplementary Table 2). To determine the distribution of encapsulated drugs within AR@ENV, we performed confocal microscopy using FITC- and DiD-labeled liposomes. The lack of significant colocalization of fluorescent markers in microglia indicated that AR7 and rapamycin were present in free form within AR@ENV rather than being retained within liposomes (Fig. 2f).

To compare AE@ENV with naturally secreted extracellular vesicles (EVs), we isolated EVs via ultracentrifugation. Nanoparticle tracking analysis (NTA) revealed that the diameters of the EVs, drug-free extruded nanovesicles (ENVs), and AR@ENVs were 175.4 ± 6.5 nm, 142.6 ± 5.2 nm, and 134.0 ± 4.7 nm, respectively (Fig. 2g). Zeta potential measurements revealed values of −35.33 ± 0.68 mV (EV), −21.42 ± 0.95 mV (ENV), and −19.40 ± 1.04 mV (AR@ENV) (Fig. 2h). TEM confirmed that ENVs and AR@ENVs maintained a disc-like morphology similar to that of EVs (Fig. 2i). Western blot analysis confirmed the presence of the EV-associated marker proteins TSG101 and CD9 (Fig. 2j), confirming the vesicular nature of AR@ENV.

Notably, the MiLi-FE process significantly increased the vesicle yield, with AR@ENV production being 33.8-fold greater than that of EVs (Fig. 2k). Furthermore, the drug-loading capacities of AR7 and rapamycin in AR@ENV were 19.9 ± 0.22 μg/mg and 7.02 ± 0.34 μg/mg, respectively (Supplementary Table 3), demonstrating high drug-loading efficiency via the MiLi-FE method.

The CX3CL1‒CX3CR1 axis represents a crucial signaling pathway between neurons and microglia and is essential for the neuronal targeting of AR@ENV. CX3CR1 expression in AR@ENV was examined via Western blotting. The results demonstrated that ENVs retain the CX3CR1 protein derived from microglia, with no effect on CX3CR1 expression observed during the drug loading process (Fig. 2l).

Furthermore, the complete proteome of AR@ENV was analyzed via proteomics. Proteomic analysis revealed a total of 7771 proteins (Supplementary Table). Specifically, 7206 proteins were detected in AR@ENV, 7058 in BV2 cells, and 6963 in microglial plasma membrane fractions (Supplementary Fig.). The number of proteins detected in AR@ENV cells was even greater than that detected in BV2 cells. This may be attributed to the retention of low-abundance proteins during the extrusion preparation of ENV, leading to an increase in their relative concentration and subsequent detection, whereas these proteins were undetectable in whole BV2 cells due to their low abundance. 4 4

Principal component analysis (PCA) further revealed noticeable shifts in the main components (PC1 and PC2) among samples derived from microglia, microglial plasma membranes, and AR@ENV (Supplementary Fig.). 5

Subsequent comparative analysis revealed protein changes in AR@ENV relative to whole BV2 cells and microglial plasma membrane fractions. Compared with those in whole BV2 cells (6931 proteins analyzed), 1792 proteins were upregulated (including 1396 with a ≥2-fold increase), 2074 were downregulated (including 1625 with a ≥2-fold decrease), and 164 proteins expressed in BV2 cells were not detected in AR@ENVs (Fig. 2m and Supporting Information Supplementary Table 5). Compared with those in microglial plasma membrane fractions (6836 proteins analyzed), 1469 proteins were upregulated (including 1018 with a ≥2-fold increase), 1820 were downregulated (including 1187 with a ≥2-fold decrease), and 78 proteins expressed on the microglial plasma membrane were not detected in ENVs (Fig. 2m).

To determine whether AR@ENV upregulates autophagy, we performed western blot analysis to assess autophagy-related protein expression in HT22 neurons treated with different nanovesicles (Fig. 3b). Compared with the ENV and R@ENV groups, the A@ENV and AR@ENV groups presented a significant increase in Lamp2A expression (Fig. 3c), indicating CMA activation. Moreover, p62 levels decreased in the R@ENV and AR@ENV groups (Fig. 3d), whereas the LC3II/LC3I ratio, a hallmark of macroautophagy activation,³² was significantly elevated (Fig. 3e). These findings confirm that AR@ENV effectively activates macroautophagy in neuronal cells.

As demonstrated in our previous experiments, macroautophagy impairment is not solely caused by Aβ toxicity. TEM analysis revealed a significant reduction in the number of autophagosomes in Aβ1-42-treated neurons, indicating that autophagy was suppressed (Fig. 3f, g). ENV treatment did not increase autophagosome formation, suggesting a lack of autophagy-activating properties. In contrast, AR@ENV treatment significantly increased the number of autophagosomes, confirming its ability to increase macroautophagy in neuronal cells.

Given the established impact of rapamycin-induced macroautophagy on mTOR signaling, q‒PCR analysis was performed to evaluate the effects of AR@ENV on core mTOR pathway gene expression. Following 48 h of Aβ1‒42 induction, the results indicated a significant increase in the expression of mTOR and AKT1, alongside a decrease in Pten gene expression, which is indicative of mTOR pathway activation. However, the addition of AR@ENV suppressed this activation, potentially due to mTOR pathway inhibition resulting from autophagy activation. Moreover, ENV alone inhibited this activation, which was likely attributable to the high expression of autophagy-related proteins within ENVs (Fig. 3h).

In addition to its role in Aβ clearance, autophagy dysfunction in AD also contributes to mitochondrial damage, leading to reactive oxygen species (ROS) accumulation, neuronal apoptosis, and neuroinflammation.¹⁹ To assess whether AR@ENV affects mitophagy, its impact on Aβ1–42-treated HT22 neuronal cells was analyzed via combined Western blotting and mitochondrial–lysosomal colocalization assays. AR@ENV significantly upregulated the expression of the mitophagy-related proteins PINK1 and Parkin (Fig. 3i and Supplementary Fig. 6). Furthermore, confocal microscopy revealed increased mitochondrial–lysosomal colocalization signals upon AR@ENV treatment, collectively indicating enhanced mitophagy (Fig. 3j). Moreover, AR@ENV effectively protected against Aβ1‒42-induced mitochondrial impairment, leading to recovery of the mitochondrial membrane potential (Fig. 3k).

In summary, in vitro pharmacological assays demonstrated that AR@ENV activates both macroautophagy and CMA in neuronal cells. Additionally, AR@ENV protects neurons from Aβ1‒42-induced mitochondrial dysfunction by restoring mitochondrial autophagy, highlighting its therapeutic potential in AD.

While AR@ENV has neuroprotective effects in vitro, its therapeutic success relies on efficient brain delivery, which is hindered by the restrictive nature of the BBB.^33,34 We therefore investigated whether microglia-derived ENV, the precursor of AR@ENV, could effectively cross the BBB and reach neuronal cells.

After crossing the BBB, we evaluated the uptake of ENV-DiD by HT22 neurons or microglia (BV2) in the lower chamber. Confocal microscopy further confirmed that ENV-DiD was internalized by the neuronal cells in the lower chamber (Fig. 4d; Supplementary Fig. 7b). Flow cytometry analysis revealed greater fluorescence intensity in the Aβ1–42-induced group than in the noninduced control and free DiD groups (Fig. 4e).

To determine whether drug-loaded AR@ENV exerts neuroprotective effects, we performed a CCK-8 assay. Aβ1-42 exposure reduced cell viability to less than 70%, but 24-hour AR@ENV treatment significantly restored viability in neurons (Fig. 4f; Supplementary Fig. 7c).

Furthermore, we investigated the mechanism of ENV translocation and examined the expression of ZO-1, a key tight junction protein, in bEnd.3 cells following ENV exposure. ENV treatment significantly reduced ZO-1 expression in the BBB model (Fig. 4g; Supplementary Fig. 8a, b). To determine whether this effect was mediated by surface proteins or internal contents, ENVs were fractionated into membrane components and luminal contents, which were then independently cocultured with bEnd.3 cells. Both the ENV membrane and luminal contents downregulated ZO-1 expression (Fig. 4g; Supplementary Fig. 8a, b), indicating that microglia-derived ENVs facilitate BBB transmigration via tight junction modulation.

Moreover, to assess the impact of AR@ENV on the integrity of the BBB in vivo, we utilized the Evans blue assay. The Evans blue assay is a widely recognized method for evaluating BBB integrity. The BBB, a highly selective semipermeable barrier, safeguards the brain from noxious substances in the bloodstream. When the BBB is disrupted, Evans blue, a large molecular dye, can infiltrate brain tissue. The quantification of Evans blue in brain tissue enables the assessment of BBB integrity. The experimental methods are illustrated in Fig. 4h. The results from the quantitative analysis revealed a significant increase in the Evans blue concentration within the brain tissues of the AR@ENV-injected mice at the 8 h time point relative to that of the control mice, which was indicative of potential BBB disruption. However, at 24 h, the concentration of Evans blue in the brain tissues did not significantly differ from that in the control group, suggesting that the disruption of the BBB elicited by AR@ENV is ephemeral and reversible (Fig. 4i).

An additional complex model was further established, which yielded results consistent with the aforementioned findings (Supplementary Fig.). These findings confirm that AR@ENV not only successfully crosses the BBB but also alleviates Aβ1-42-induced neurotoxicity, underscoring its potential as a therapeutic strategy for neurodegenerative disorders. 9

To further validate these findings, major organs were collected and prepared into cryosections. Confocal microscopy confirmed that the fluorescence distributions in the tissues were consistent with the in vivo imaging results (Supplementary Fig. 11; Supplementary Fig. 12). The ENV is distributed mainly in the hippocampal region (Supplementary Fig. 13). Immunofluorescence staining revealed preferential ENV accumulation in neurons (Fig. 5f, g; Supplementary Fig. 14) rather than in microglia (Fig. 5h, I; Supplementary Fig. 14), suggesting a greater affinity for neuronal targeting. Furthermore, a minor fraction of ENV was also observed to be taken up by astrocytes (Supplementary Fig. 15). This may be attributed to the preservation of key receptor proteins, such as CX3CL1 and CD200, on the ENV surface, which interact with their respective neuronal ligands, CX3CR1 and CD200R, facilitating selective neuronal uptake.^39,40

Given the susceptibility of nanodelivery systems to rapid clearance in vivo, the uptake of AR@ENV by immune cells and platelets was further investigated. Eight-month-old male APP/PS1 mice were administered ENV-DiD (DiD-labeled ENV substituting for the drug payload). At 8 h postinjection, the peak AR@ENV brain accumulation platelets and lymphocytes were isolated for flow cytometry analysis. The results demonstrated negligible uptake of ENV-DiD by platelets (Supplementary Fig. 16). Lymphocytes exhibited modest uptake of ENV-DiD, which was significantly lower than that of long-circulating liposomes. Furthermore, ENV-DiD uptake was predominantly localized to CD3E-positive lymphocytes (Fig. 5j, k).

To mimic the in vivo environment, the stability of rapamycin and AR7 within AR@ENV was assessed in 10% serum. Drug degradation was compared at 0 h and 8 h post-incubation. Approximately 12.8% degradation of rapamycin and 3.8% degradation of AR7 were detected after 8 h of serum exposure (Supplementary Fig.and Supplementary Table). 17 6

These findings confirm the successful development of ENVs as BBB-penetrating nanovesicles, demonstrating efficient brain targeting in AD models.

Compared with WT mice, APP/PS1 mice exhibited significantly prolonged escape latency in the Morris water maze (MWM) test, indicating that spatial learning and memory impairments are characteristic of AD (Fig. 6b, c). In contrast, AR@ENV-treated mice presented a marked reduction in escape latency, indicating an increased ability to locate the hidden platform. Moreover, AR@ENV treatment increased both the frequency of platform crossings and the time spent in the target quadrant, further confirming improvements in spatial memory and navigation skills (Fig. 6d, e). Similar cognitive improvements were observed in Aβ1-42-injected mice, reinforcing the broad applicability of AR@ENV across different AD models (Supplementary Fig. 18).

To further assess working memory and anxiety-related behaviors, we conducted Y-maze and open field tests. APP/PS1 mice exhibited a significant reduction in spontaneous alternations, indicating working memory deficits, whereas AR@ENV-treated mice restored alternation performance to WT-like levels, reflecting improved cognitive flexibility and spatial recognition (Fig. 6f, g). Additionally, in the open field test, APP/PS1 mice spent significantly less time in the central area, a behavior associated with heightened anxiety and cognitive decline (Fig. 6h, i). AR@ENV-treated mice presented a noticeable increase in center time, indicating reduced anxiety-like behavior and enhanced cognitive function.

To further validate cognitive restoration, we performed novel object recognition (NOR) and nest-building assays. In the NOR test, APP/PS1 mice exhibited reduced object recognition abilities, as indicated by their lower preference and discrimination indices. However, compared with WT mice, AR@ENV-treated mice presented a novel object recognition index and preference index, suggesting enhanced memory retention and cognitive restoration (Fig. 6j–l).

Similarly, in the nest-building assay, APP/PS1 mice and Aβ-42-injected mice exhibited impaired nest-building behavior, producing shallow, poorly structured nests (Fig. 6m; Supplementary Fig. 19a). In contrast, the AR@ENV-treated mice constructed well-formed nests with three-dimensional architecture and intact walls, similar to the WT mice (Fig. 6m). The higher nest-building scores in the AR@ENV group further support its role in restoring cognitive and motor functions (Fig. 6n and Supplementary Fig. 19b).

In addition to its cognitive benefits, AR@ENV also demonstrated superior safety compared with free drug administration. Hematological analysis revealed that free drug administration significantly reduced white blood cell (WBC) counts compared with those in WT mice (Supplementary Fig.). Additionally, the saline and AR7+Rapa groups presented increased platelet (PLT) counts (Supplementary Fig.). Importantly, biochemical indices revealed no significant abnormalities, suggesting that AR@ENV encapsulation effectively mitigated the systemic toxicity associated with free drug administration (Supplementary Fig.). 20a 20d 21

Taken together, these results demonstrate that AR@ENV significantly improves cognitive deficits in AD model mice by enhancing memory, reducing anxiety-like behaviors, and restoring cognitive function. Additionally, its encapsulated delivery system minimizes systemic toxicity, highlighting AR@ENV as a promising and safe therapeutic strategy for neurodegenerative disorders.

Since neuroinflammation is a major driver of AD progression, we next examined the effects of AR@ENV on the microglial response and brain inflammation. In AD, microglia accumulate around Aβ plaques and shift between proinflammatory (M1, neurotoxic) and anti-inflammatory (M2, neuroprotective) states.⁴¹ Confocal microscopy revealed a significant increase in M1 microglia in saline-treated AD mice, contributing to widespread neuroinflammation (Fig. 7e; Supplementary Fig. 24). However, AR@ENV treatment significantly reduced M1 microglial activation, restoring the levels to those observed in WT mice, indicating that AR@ENV modulates microglial activation to suppress neuroinflammation.

Reactive astrogliosis promotes Aβ deposition and generates neurotoxic byproducts (e.g., putrescine, ammonia, H₂O₂) via the urea cycle, contributing to neuronal damage and memory impairment. Consequently, the effect of AR@ENV treatment on astrogliosis was examined. Compared with wild-type (WT) mice, APP/PS1 mice exhibited significant astrogliosis in brain tissue. AR@ENV intervention effectively ameliorated this pathological proliferation (Supplementary Fig.). 25

To further confirm its anti-inflammatory effects, we quantified the proinflammatory cytokines TNF-α, IL-6, and IL-1β in AD mouse brains via ELISA (Fig. 7f; Supplementary Fig. 26). AR@ENV treatment significantly reduced all three inflammatory markers, confirming that it effectively suppresses neuroinflammation and improves the AD brain microenvironment.

Since neuronal loss is correlated with memory deficits in AD, we next examined neuronal survival in the hippocampal CA1 and CA3 regions, areas crucial for learning and memory.^42,43 Immunofluorescence staining revealed that AR@ENV-treated mice had significantly more NeuN⁺ neurons than saline-treated AD mice did (Fig. 7g; Supplementary Fig. 27), indicating substantial neuroprotection and reduced neuronal loss.

These findings were further supported by Nissl staining, where AR@ENV-treated mice presented well-preserved Nissl bodies and intact neuronal structures, with no significant difference from those of WT mice (Fig. 7h; Supplementary Figs. 28, 30). Similarly, H&E staining of saline-treated AD mice revealed severe neurodegeneration characterized by pyramidal cell necrosis, nuclear pyknosis, and widespread neuronal damage (Fig. 7i; Supplementary Figs. 29, 30). AR@ENV treatment significantly reversed these degenerative changes, restoring normal hippocampal morphology.

Since Aβ accumulation is a hallmark of AD, we next examined whether AR@ENV reduces the Aβ burden in the APP/PS1 mouse brain. AR@ENV-treated mice presented a marked reduction in Aβ plaque deposition, particularly in the hippocampus (Fig. 7j). These findings indicate that AR@ENV not only protects neurons but also enhances the clearance of neurotoxic Aβ aggregates.

Taken together, these results demonstrate that AR@ENV enhances autophagy, reduces neuroinflammation, preserves neuronal integrity, and promotes Aβ clearance in AD model mice. By restoring autophagic flux and modulating the inflammatory microenvironment, AR@ENV confers neuroprotection and represents a promising therapeutic strategy for AD.

Given the therapeutic efficacy of AR@ENV in ameliorating AD pathology, we next evaluated its biocompatibility and systemic safety to assess its suitability for in vivo applications. Hemolysis assays demonstrated that AR@ENV exhibited minimal hemolytic activity, with hemolysis rates remaining below 5% across various concentrations within 8 h. Erythrocytes retained a normal morphology without signs of crenation or rupture, indicating a negligible risk of red blood cell damage (Supplementary Fig.). 31a–c

We further assessed cytotoxicity in vitro in HT22 cells. Four types of nanovesicles were tested: ENV, A@ENV, R@ENV, and AR@ENV. After 24 h of treatment, the cell viability in all the groups exceeded 80%, confirming favorable biocompatibility (Supplementary Fig.). 32a

To evaluate systemic safety in vivo, healthy Kunming mice were treated daily with seven doses of saline, free AR7+RAPA, ENV, or AR@ENV. H&E staining of major organs revealed no tissue lesions in any group (Supplementary Fig.), supporting the absence of organ toxicity. Hematological parameters remained within normal ranges across all treatments (Supplementary Fig.), and biochemical analysis revealed that AR@ENV significantly reduced the hepatotoxicity observed with free drug administration (Supplementary Fig.). Additionally, AR@ENV treatment led to a significant reduction in circulating low-density lipoprotein (LDL) levels, suggesting potential metabolic benefits. The body weights are shown in the supplementary table. across the study (Supplementary Fig.), further supporting systemic safety. 32b 32c 32d 32e

To evaluate potential immune responses, we quantified the inflammatory cytokines TNF-α, IL-6, and IL-1β in both serum and brain tissue after treatment. No significant elevation in cytokine levels was observed in any group, indicating that AR@ENV does not elicit measurable immunogenicity or proinflammatory responses (Supplementary Fig.). 32f

To systematically evaluate the in vivo immune compatibility of AR@ENV, Kunming mice (6–8 weeks old) received ten intravenous tail vein administrations every 3 days. Serum and spleens were collected 7 days after the final drug administration. ELISAs revealed no significant differences in total IgG, IgG1, or IgG2a levels versus controls, indicating the absence of humoral immune activation (Supplementary Fig.). Flow cytometric analysis further demonstrated that while AR@ENV did not alter CD4⁺ T-cell populations (Supplementary Fig.), it significantly increased the proportion of CD8⁺ T cells (Supplementary Fig.). Given the documented evidence that CD8⁺ T cells mitigate AD progression by suppressing hyperactivated microglia, this immunomodulatory effect may provide complementary therapeutic synergy to AR@ENV's primary efficacy. 33 34a, c 3b, c

DSPE-PEG2000 was purchased from Hunan Huateng Biological Technology (China). The Cell Counting Kit-8 (CCK-8) and BCA protein assay kits were purchased from Beyotime Biotechnology Co., Ltd. (Shanghai, China). Mouse IL-6, IL-1β, and TNF-α ELISA kits were purchased from Thermo Fisher Scientific (USA). AR7 and rapamycin were purchased from MedChemExpress (USA). Mouse Lamp2A, NeuN, Iba-1 and CX3CR1 antibodies were purchased from Abcam (USA). LC3B, p62, and GAPDH antibodies were purchased from Abways (Shanghai, China). CD9, TSG101, PINK1, and Parkin antibodies were purchased from HUABIO (Hangzhou, China). IgG, IgG1 and IgG2a were purchased from ABclonal (Wuhan, China). Aβ1-42 was purchased from Sangon Biotech (Shanghai, China).

All chemicals used were of analytical or reagent grade. Mouse brain microvascular endothelial cells (bEnd.3), mouse hippocampal neuronal cells (HT22) and a murine microglial cell line (BV2) were obtained from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). Male Kunming mice (6–8 weeks old) were obtained from Chengdu Dashuo (Chengdu, China). Male APP/PS1 mice (8 months) were obtained from Changzhou Cavens (Changzhou, China). All animal experiments were approved by the ethics committee of Sichuan University.

The preparation of AR7 and rapamycin liposomes (LPs) followed previously established methods. Briefly, AR7 or rapamycin was dissolved with soybean phospholipids, cholesterol, and DSPE-PEG-2000 (in a 3:1:1 mass ratio), and each component was separately dissolved in dichloromethane. The drug-to-lipid ratio was either 40:1 or 20:1. Liposomes are prepared via the thin-film hydration method, followed by probe sonication after hydration to obtain the desired liposomes.

The specific methodology employed for coencapsulating lipophilic DiD and hydrophilic FITC into liposomes was the thin-film hydration-extrusion method, which was chosen for its reliability and controllability. This involved first dissolving DiD, soybean phospholipid, cholesterol, and DSPE-PEG-2000 (at a mass ratio of 3:1:1) in dichloromethane, with a DiD-to-lipid mass ratio of 1:100. The mixture was then rotary evaporated at 40 °C to form a homogeneous thin film. The lipid film was subsequently hydrated via a buffer solution containing 0.1 mg/mL FITC, with hydration assisted by probe sonication at 80 W for 2 min. Finally, the suspension was extruded 5 times through a 200 nm polycarbonate membrane to yield liposomes coloaded with both DiD and FITC.

(1) BV2 cells were seeded in 96-well plates and cultured to 50–60% confluence. The medium was then replaced with liposomes loaded with AR7 and rapamycin. Cell viability was assessed via the CCK-8 assay at 2 h, 4 h, 6 h, 8 h, 12 h, and 24 h posttreatment. (2) BV2 cells were seeded in 24-well plates and cultured to 50–60% confluence. After treatment with DiD-loaded liposomes for 2 h, 4 h, 6 h, or 8 h, the lysosomes were stained with a LysoTracker probe. The cells were fixed with 4% paraformaldehyde, the nuclei were counterstained with DAPI, and mounted slides were imaged via confocal microscopy to evaluate liposome‒lysosome colocalization. (3) BV2 cells were seeded in 6-well plates and cultured to 50–60% confluence. Following treatment with AR7/rapamycin-loaded liposomes for 2 h, 4 h, 6 h, or 8 h, the cells were harvested and lysed with acetonitrile to extract AR7 and rapamycin. After centrifugation to remove the precipitates, the drug concentrations were quantified via HPLC.

The preparation of AR@ENV was modified on the basis of the literature.⁵² First, BV2 cells were cultured to 80% confluence and then treated with AR7@LP (containing 30 μg/mL AR7) or Rapa@LP (containing 10 μg/mL rapamycin) for 8 h. The cells were collected and extruded through polycarbonate membranes with pore sizes of 5 µm, 0.8 µm, 0.4 µm, and 0.1 µm (each size was passed through three times). The precipitate was subsequently collected by ultracentrifugation at 1,000,000 × g, and the resulting AR@ENV was stored at −80 °C.

To simulate the degradation of AR@ENV in vitro, the samples were incubated in PBS containing 10% FBS at 37 °C with shaking (60 rpm/min) for 8 h. Acetonitrile was added to lyse the vesicles, and rapamycin and AR7 were extracted. The drug content was quantified by HPLC, with the 0-h incubation group serving as the control. The degradation profiles of rapamycin and AR7 were calculated on the basis of residual drug content.

The distinctive proteins TSG101 and CD9, which are characteristic of vesicles, and the microglia-specific receptor CX3CR1 were identified through Western blotting (WB). Moreover, comprehensive proteomic profiling was conducted to analyze the entire proteome of BV2 cells, BV2 plasma membranes, and extracellular vesicles (ENVs). The morphology of the native vesicles, ENVs, and AR@ENVs was observed via transmission electron microscopy (TEM), and nanoparticle tracking analysis (NTA) was used to measure the particle size, Span and zeta potential of the vesicles. To verify whether the liposomes disperse after being taken up, DID and FITC were coencapsulated within the liposomes, and the colocalization results were used.

AR@ENV viability was evaluated through the CCK8 assay. HT22 and BV22 cells were seeded in 96-well plates and cultured to 50–60% confluence prior to treatment with ENV, A@ENV, R@ENV, or AR@ENV at defined concentrations: total vesicle protein concentrations of 10, 20, 50, and 100 μg/mL, corresponding to rapamycin loads of 0.07, 0.14, 0.36, and 0.72 μg/mL and AR7 loads of 0.20, 0.40, 1.00, and 2.00 μg/mL. After a 24-h incubation, the medium was replaced, and 10 µL of CCK8 reagent was added to each well. The cells were incubated until a color change occurred, and the absorbance was measured at 450 nm.

HT22 cells were inoculated onto 96-well plates and cultured for 24 h. The original medium was then aspirated and replaced with medium containing Aβ1-42 at a concentration of 15 μM. The cells were subsequently treated with ENV, A@ENV, R@ENV, or AR@ENV at the following concentrations: total vesicle protein at 1, 2, 5, and 10 μg/mL; corresponding rapamycin concentrations of 7, 14, 35, and 70 ng/mL; and AR7 concentrations of 20, 40, 100, and 200 ng/ml. After a 24-h incubation period, the medium was changed, and 10% CCK8 reagent was added to each well. The plates were then incubated until a color change was observed, after which the absorbance was measured at a wavelength of 450 nm.

bEnd.3 cells were seeded at a density of 1 × 10⁵ cells per well in the upper chambers (12-well, pore size: 3 μm; Corning, USA). The culture medium was replenished daily, and the transendothelial electrical resistance (TEER) values were measured until they reached a value of 200 Ω/cm². Subsequently, HT22 cells were seeded in the lower chambers and induced with fresh culture medium containing Aβ1-42 (15 μM) for 24 h. Next, the upper chamber was treated with DiD-labeled ENV at a specified concentration. After incubation for 4 h, the fluorescence intensity in the culture medium of each lower chamber was quantified via a microplate reader. The average fluorescence intensity in HT22 cells was assessed through flow cytometry, and the penetration of DiD-ENV through the bEnd.3 monolayer was examined via 3D confocal microscopy.

Western blotting was performed following methods described in the literature.⁵³ Briefly, cells or tissues were collected posttreatment. The cells or tissues were lysed via lysis buffer, with additional homogenization of the tissues. The obtained protein lysates were quantified via a BCA assay kit. The proteins were subsequently denatured by heating at 100 °C for 10 minutes. The proteins were separated via 10% SDS‒polyacrylamide gel electrophoresis (SDS‒PAGE) and then electrophoretically transferred onto polyvinylidene fluoride (PVDF) membranes. The membranes were blocked with 5% skim milk at room temperature for 2 hours, followed by an overnight incubation with specific primary antibodies at 4 °C. Afterward, the membranes were incubated with secondary antibodies at room temperature for 1 h. Finally, the protein bands were visualized via a gel imaging system.

To assess the impact of AR@ENV on the mTOR pathway via qRT‒PCR, HT22 cells were seeded in 12-well plates and pretreated with 15 μM Aβ1‒42 for 12 h. Cells were then exposed to ENV or AR@ENV at specified concentrations: 5 μg/mL total vesicle protein, 35 ng/mL rapamycin, and 100 ng/mL AR7. Untreated cells (without Aβ1‒42) and Aβ1‒42-only treatment were used as control groups. After 12 h of incubation, total RNA was extracted, and qRT‒PCR was performed to quantify the expression of core mTOR pathway genes (Mtor, Akt1, and Pten).

Amyloid-β1-42 (Aβ1-42) was dissolved in artificial cerebrospinal fluid (CSF) at a concentration of 1 mg/mL and preincubated at 37 °C for 7 days. Male Kunming mice were subsequently anesthetized and mounted in a stereotactic frame. Aβ1-42 (5 μL) was injected into the right hippocampus of Kunming mice, and the injection needle was slowly withdrawn 8 minutes post-injection.

APP/PS1 mice and Aβ-injected model mice were randomly assigned to different groups, with littermate control mice and healthy Kunming mice serving as the control groups. Subsequently, free DiD or DiD-labeled ENV (DiD dosage of 10 mg/kg) was administered, and imaging was performed at specific time points of 1, 2, 4, 8, 12, and 24 h postadministration via the Lumina III Imaging System (PerkinElmer, USA). At the designated time points, the mice were euthanized for tissue collection, which was then used for ex vivo imaging. The tissues were fixed and dehydrated, followed by sectioning via a freezing microtome (Leica CM1950, Germany). The sections were subjected to immunofluorescence staining, and after counterstaining with DAPI, the slides were mounted. The slides were then observed under a confocal microscope.

Eight-month-old male APP/PS1 mice were administered DiD-labeled ENVs (ENV-DiD) as a drug surrogate. At 8 h postinjection, blood was collected via retro-orbital puncture into EDTA-anticoagulated tubes and centrifuged at 800 rpm/min for 15 min. The upper plasma layer was transferred to a new tube and centrifuged at 3500 rpm/min for 10 min to obtain a platelet-rich pellet. Separately, whole blood was treated with lymphocyte separation medium for lymphocyte isolation according to the manufacturer's protocol. Uptake of ENV-DiD by lymphocytes and platelets was quantified via flow cytometry.

APP/PS1 mice and Aβ-injected mice were randomly assigned to six groups, each receiving intravenous injections of either saline (APP/PS1 group), free drug (AR7+Rapa), ENV, A@ENV, R@ENV, or AR@ENV. The dosing concentrations are ~0.2 mg/kg for rapamycin, 0.6 mg/kg for the agent AR, and 30.0 mg/kg for vesicular carrier proteins.⁵⁴ Injections were administered every 3 days for a total of 10 times. C57BL/6 mice or healthy Kunming mice served as the wild-type (WT) control groups and were injected with saline. Body weight was recorded every 3 days. After completion of the treatment, behavioral experiments were conducted, and the mice were euthanized for further research.

The Morris water maze test is a widely used behavioral assay in neuroscience for evaluating spatial learning and memory in mice. The experiment involved a circular pool divided into four invisible quadrants, with a platform hidden 1 cm below the water surface in one of the quadrants. Various colored cues—namely, a triangle, a circle, a square, and a pentagon made from construction paper—were placed around the pool to provide spatial reference points. The water was made opaque with a small amount of titanium dioxide, and its temperature was maintained at 22 ± 1 °C. An overhead video tracking system recorded the swimming trajectories of the mice. The testing period spanned 6 days and included a hidden platform test (days 1–5) and a probe test (day 6). During the hidden platform test, the mice were placed randomly in each quadrant of the pool and were given 60 s to locate the hidden platform. Testing occurred daily in three different quadrants, with a 4-h interval between trials, and escape latencies were recorded. On the final day, a 24-h probe test was conducted with the platform removed, allowing the mice to swim freely for 60 s. The video tracking system captured their trajectories, and the data were analyzed to assess spatial learning and memory by examining escape latencies and swimming speeds.

The Y-Maze test is used to assess spatial working memory and cognitive flexibility in mice. The maze consists of three arms of equal length, typically colored black or blue, and intersects at a central point to form a Y shape. During the experiment, the apparatus was placed in a quiet testing room with constant lighting, and a camera tracking system was used to monitor the movements of the mice. At the start of the test, the mouse was placed at the end of one arm, facing the end of the arm, and was allowed to freely explore the maze for 8 minutes. The trajectory of each mouse was recorded throughout the experiment.

Prior to the experiment, the testing chamber was cleaned, and the environment was quiet with uniform lighting. During the experiment, the animal was placed in the center of the chamber, the experimenter quickly left, and behavior analysis software was used to record data on the animal's movement trajectory, velocity, dwell time, and other parameters for 10 min.

Before the experiment, the mice were allowed to acclimate to the arena for 10 min. During the familiarization phase, they explored two identical objects (Object A and Object A) for 10 min. In the testing phase, one object was replaced with a novel object (Object B), and the mice explored the arena for an additional 10 min. The time spent by the mice exploring the novel object (Object B) and the familiar object (Object A) was recorded.

Before the experiment begins, the mice should be acclimated to the environment to minimize the impact of environmental factors on the results. The mice were subsequently placed in prepared cages, and an adequate amount of nesting material was provided. We observed and recorded how the mice used these materials to construct nests over the following 72 h. When evaluating nesting behavior, the focus is on the structure, comfort, and integrity of nests to gain a comprehensive understanding of their quality and complexity.

Eight-month-old male APP/PS1 mice received intravenous tail vein injections of AR@ENV at therapeutic concentrations. Age-matched APP/PS1 mice without AR@ENV injection served as blank controls. At 8 h and 24 h postinjection (time points corresponding to peak cerebral accumulation of AR@ENV as determined by distribution studies), 2% Evans blue solution (2 ml/kg) was administered via the tail vein. After 2 h, the mice were euthanized and subjected to transcardiac perfusion: the thoracic cavity was surgically exposed, and heparinized 0.9% saline was slowly infused through the cardiac apex until the effluent from the incised right atrium became clear. The brain tissues were then harvested, weighed, and homogenized. The tissue samples were incubated in formamide (100 mg of tissue per 1 ml of formamide) at 37 °C with 80 rpm orbital shaking for 48 h. The supernatants were subsequently analyzed via spectrophotometry at 620 nm.

Kunming mice were randomized into four treatment groups: saline control, AR7 + Rapa, ENV, and AR@ENV. Intravenous administration was performed every other day for seven doses at approximate concentrations of 0.2 mg/kg rapamycin, 0.6 mg/kg AR7, and 30.0 mg/kg vesicular carrier protein across all drug-treated cohorts. Body weight fluctuations were recorded throughout the intervention period. The final procedures included the collection of major organs for hematoxylin and eosin (H&E) histopathological examination, alongside the procurement of whole blood for complete hemogram analysis and serum for biochemical profiling and cytokine quantification.

Healthy Kunming mice were randomized into two groups receiving intravenous injections of either saline (control) or AR@ENV nanocomplexes at approximate concentrations of 0.2 mg/kg rapamycin, 0.6 mg/kg AR7, or 30.0 mg/kg vesicular carrier protein. The drugs were administered every 72 h for a total of 10 doses. Posttreatment, serum and spleen samples were collected. Serum samples were subjected to ELISA to quantify antigen-specific antibody titers (IgG, IgG1, and IgG2a), while splenocytes were isolated via mechanical dissociation for flow cytometric quantification of CD4⁺ and CD8⁺ T-cell populations.

All the experiments were performed via GraphPad Prism 9. All experiments were repeated at least three times. The data are presented as the means ± standard deviations (SDs). Statistical comparisons were performed via two-tailed t tests and one-way ANOVA. A p value < 0.05 was considered significant.