Gut‐derived bacterial vesicles carrying lipopolysaccharide promote microglia‐mediated synaptic pruning

Abstract

INTRODUCTION

Growing evidence links gut microbiota (GM) to Alzheimer's disease (AD). Elevated lipopolysaccharide (LPS) levels, a Gram‐negative bacteria component, are found in AD brains, but how LPS breaches the blood–brain barrier (BBB) remains unclear. Hypotheses suggest that bacteria‐derived extracellular vesicles (bEVs) may transport LPS across the BBB.

METHODS

bEVs were extracted from human and mouse feces and blood, and LPS levels were measured. In vivo imaging and immunofluorescence confirmed the transport of blood LPS‐carrying bEVs across the BBB. The role of these bEVs in microglia was investigated both in vivo and in vitro.

RESULTS

Elevated LPS‐containing bEVs were detected in the plasma of AD patients compared to healthy individuals. These bEVs activated microglial Piezo1, consequently precipitating an excessive synaptic pruning process mediated by the C1q‐C3 complement pathway.

DISCUSSION

These findings illuminate the complex interplay between the gut microbiota, bEVs, neuroinflammation, and synaptic plasticity – a key early event in AD – offering insights for potential therapeutic interventions.

Highlights

• GM‐derived bEVs can traverse the BBB.

• LPS was necessary for bEVs’ penetration into the brain, and bEVs might be closely related to AD progression.

• bEVs mediated microglial activation and synaptic pruning via C1q‐C3 complement pathway.

• Microglia Piezo1 was involved in bEV‐induced excessive synaptic pruning.

1. BACKGROUND

The gut microbiota (GM), a complex community of microorganisms residing in the gastrointestinal tract (GIT), has emerged as a central player in neurodegenerative diseases, exerting a profound influence on their development and progression. ¹ , ² , ³ In‐depth genomic studies of GIT microbiota have brought light to a noteworthy pattern in patients with Alzheimer's disease (AD). These studies demonstrate a significant reduction in the diversity of the GM, accompanied by a discernible increase in the abundance of Gram‐negative (G−) bacteria. ⁴ This ubiquitous alteration of microbial composition in the GIT suggests a potential link between G− bacterial populations and the pathogenesis of neurodegenerative diseases, including AD.

Numerous researchers are currently delving into the potential correlation between a specific strain of gut bacteria and particular neurological ailments, ⁵ , ⁶ , ⁷ while others are examining shared signaling molecules present in various disorder categories. ⁸ , ⁹ One prevalent element among G− bacteria is lipopolysaccharide (LPS), a well‐known stimulant utilized in studies on neuroinflammation. ¹⁰ , ¹¹ , ¹² , ¹³ Significantly, there is an observed rise in LPS expression in the brains of individuals with neurological disorders, particularly in the cortical and hippocampal regions of individuals diagnosed with AD. ⁴ , ¹⁴ However, a puzzling aspect emerges when considering the integrity of the blood–brain barrier (BBB), which typically restricts the free passage of LPS into the brain. ¹⁵ This discrepancy often leads to speculation about potential contamination as an explanation for the presence of detected LPS in human brains.

Recent studies suggest that bacteria can produce extracellular vesicles (bacteria‐derived extracellular vesicles [bEVs]), which substantially facilitate the communication along the gut‐brain axis. ¹⁶ , ¹⁷ , ¹⁸ Similar to host EVs, bEVs are cup‐shaped nanoparticles with a phospholipid bilayer membrane, exhibit a size distribution of 50 to 150 nm, and harbor bacterial molecules like nucleic acids, proteins, and metabolites, ¹⁹ , ²⁰ , ²¹ with the difference being LPS and OmpA are found specifically on the membrane of G− bEVs. bEVs can be internalized readily by recipient cells, triggering various pathological processes related to the GM. ²² In the realm of AD, bEVs have been shown to successfully enter the brain via the vagus nerve, ²³ elevate the expression of TNF‐α and IL‐6, ²⁴ and cause neuronal dysfunction by activating the C3‐C3aR signaling pathway. ²⁵ Moreover, bEVs secreted by the GM of AD patients are shown to exacerbate neuroinflammation and tau phosphorylation. ²⁶ This raises a crucial question: Can bEVs act as carriers, aiding in the transportation of LPS into the brain? If so, what mechanisms are involved in mediating neurodegenerative changes?

In this investigation, we addressed the complex interconnections between the GIT microbiota, bEVs, and the AD pathogenesis by demonstrating that bEVs can cross the BBB and trigger microglial activation. Moreover, we observed that bEVs containing LPS activated microglial Piezo1, leading to an excessive synaptic pruning process mediated by the C1q‐C3 complement pathway. This investigation holds the promise of unveiling novel therapeutic targets and strategies for mitigating the impact of neurodegenerative conditions through the modulation of the gut‐brain axis.

2. METHODS

2.1. Human samples

Approval for this study was granted by the Institutional Review Board of First Affiliated Hospital of Zhejiang University School of Medicine, Hangzhou, Zhejiang, China ([2025B]IIT Ethics Approval No. 0591). Post mortem brain tissues were sourced from the China National Health and Disease Human Brain Tissue Resource Center (Hangzhou, China). All materials were collected from donors who had given written informed consent for brain autopsy and allowed the use of their clinical information for research. The demographics and clinical data of the participants used for immunostaining are summarized in Table S1. Plasma and feces samples from age‐ and sex‐matched neurologically healthy controls (HCs), AD patients, and mild cognitive impairment (MCI) patients were obtained from the First Affiliated Hospital, Zhejiang University School of Medicine. AD and MCI participants underwent extensive clinical evaluations and met the specified inclusion and exclusion criteria. The sample collection procedures are detailed in previous publications. ²⁷ , ²⁸ Demographics and clinical data of the participants used for plasma and feces collection are summarized in Table S2.

2.2. Animals

All mouse experiments received authorization from the Animal Care and Use Committee of the animal facility at the First Affiliated Hospital, Zhejiang University School of Medicine (2024559). Wild‐type mice (WT, Catalog No.: SM‐001, C57BL/6J) were sourced from Shanghai Model Organisms Center, Inc. APP/PS1 mice (Catalog No.: 005864, C57BL/6J background) were purchased from the Jackson Laboratory. Myeloid‐specific Piezo1 deficiency mice (Piezo1^∆LysM , C57BL/6J background) were generously provided by Professor Jing Li. These mice were generated using the Cre‐loxP system. ²⁹ To generate Piezo1^∆LysM mice and Piezo1^fl/fl littermate controls, Piezo1^fl/fl (Catalog No.: 029213, C57BL/6J background, Jackson Laboratories) were crossed with LysM‐Cre mice (Catalog No.: 004781, C57BL/6J background, Jackson Laboratories). All mice were randomly assigned to either the control or treatment group, and the experiment, outcome assessment, and statistical analysis were conducted blindly. All mice were kept on the same 12‐h light/dark cycle and had unrestricted access to food and water.

2.3. Collection of gut microbiota

GM from HC, AD patients, 2‐month‐old WT mice, 12‐month‐old WT mice, and 12‐month‐old APP/PS1 mice was collected and transported to laboratory on ice within 2 h. Fecal samples were homogenized with sterilized normal saline at a 1:5 ratio. To remove undigested food and smaller particles, the mixture was passed through stainless steel sieves with pore sizes of 2.0, 1.0, 0.5, and 0.25 mm and then centrifuged at 6000 × g for 15 min at 4°C. After the supernatant was discarded, the GM was resuspended in half of the original volume of sterilized normal saline and mixed with sterile glycerol to a final concentration of 10%. The entire procedure was conducted anaerobically and completed within 1 h. 90 mg/L vancomycin was used to obtain G− bacteria, while 100 mg/L polymyxin and 60 mg/L Linazolamine were used to separate G+ bacteria.

RESEARCH IN CONTEXT

1. Systematic review: Growing evidence links GM to AD, characterized by reduced diversity with augmented G− bacteria. Elevated LPS levels, an essential component of G− bacteria, are found in AD brains, but how LPS breaches the BBB or directly results in brain damage remains to be investigated.

2. Interpretation: This study unveiled a few key elements for the gut‐derived LPS‐carrying bacteria‐derived extracellular vesicles to mediate synaptic pruning, a pivotal early process related to neurodegeneration such as in AD. While the mechanism underlying the passage of bEVs across the BBB requires further investigation, the observed mediation of microglial activation via Piezo1 suggests a promising avenue for therapeutic intervention in neurodegenerative disorders.

3. Future directions: Future research will identify the complex interplay between the GM, bEVs, neuroinflammation, and synaptic plasticity in neurodegeneration.

2.4. Genetically engineered bacteria culture

Escherichia coli BL21 (DE3) and endotoxin‐free BL21 (ClearColi BL21(DE3), Research Corporation Technologies, USA) transfected with plasmids encoding HA and mPlum tags were grown at 37°C in Luria‐Bertani (LB) medium with shaking at 180 rpm. bEVs were extracted after 24 h incubation. The sequences of plasmid are listed in Table S3.

2.5. Isolation of bEVs from bacteria

Bacteria at a concentration of 1 × 10⁷ CFUs per 100 mL of medium were cultured in YCFA medium (LA4040, Solarbio, China) with 0.002 g/mL each of glucose, maltose, and cellobiose under anaerobic condition for two days. The conditioned media from these bacteria were then collected, centrifuged at 6000 × g for 30 min at 4°C, and filtered through a 0.22‐µm filter (Millipore, Billerica, USA) to eliminate any remaining bacteria. The supernatant was transferred to Amicon Ultra‐15 Centrifugal Filter Units (100‐kDa, Millipore, USA) and concentrated 400‐ to 500‐fold by centrifugation at 4000 × g at 4°C.

2.6. Isolation of bEVs from plasma

Blood samples were centrifuged at 2500 × g for 15 min at room temperature within 120 min of collection, followed by another centrifugation at 2500 × g for 15 min at 19°C to obtain platelet‐depleted plasma. Then, 2 mL plasma was then applied to a 10‐mL exclusion chromatography (SEC) column (Sepharose CL‐2B, S8731, Solarbio, China), and chromatography was performed according to the column specifications. As the plasma passed through the column, bEVs were effectively separated by size. Size exclusion chromatography (SEC) fractions (1 mL/each) were collected, with fractions 4 to 7 containing bEVs, which were then concentrated by ultrafiltration to 1 mL of crude bEVs. The crude bEVs were further purified using an Optiprep density gradient (ODG) centrifugation. Sequential layers of 3 mL of 40%, 20%, and 10% and 2.6 mL of 5% Optiprep, along with 0.75 mL of crude bEVs, were placed in a 12.5‐mL polyallomer Beckman Coulter tube. The gradient fractions were centrifuged at 100,000 × g in a Beckman Optima XPN ultracentrifuge (Type70 Ti rotor) for 18 h at 4°C. Fraction 12 and 13, containing bEVs, were collected, pooled, and applied to an exclusion chromatography column, where fractions 4 to 7 were collected. The SEC fractions were concentrated to 100 µL using a 10‐kDa centrifugal filter (Millipore, Billerica, USA).

2.7. In vivo imaging of bEVs

The bEVs were tagged using Vybrant DiR cell‐labeling solution (D‐12731, Thermo Fisher Scientific, USA). A total of 200 µL of DiR‐labeled bEVs, with a concentration of 1 × 10¹¹ particles/mL, were administered intravenously through the caudal vein to mice. After 3 h, mice were sacrificed and perfused with PBS prior to IVIS Spectrum imaging (PerkinElmer, Waltham, MA, USA).

2.8. Quantitative analysis of bEVs in plasma and brain

Mice were administered bEVs derived from E. coli BL21 (DE3) transfected with HA‐mPlum plasmid via intestine injection (1 × 10¹¹ particles/mL, 200 µL). Twenty‐four hours after the treatment, blood and brain tissues were collected, weighed, and analyzed for HA protein levels using Western blotting.

2.9. Immunofluorescence staining

The bEVs were labeled with Vybrant DiI cell‐labeling solution (D‐282, Thermo Fisher Scientific, USA) or the antibiotic probe polymyxin B‐cy3 (Ruixi biological Technology Co., China). 200 µL of DiI‐labeled bEVs, with a concentration of 1 × 10¹¹ particles/mL, was intravenously administered to mice via tail. The mice were transcardially perfused with PBS under deep anesthesia. The mice brain tissues were dissected, fixed in 4% paraformaldehyde (PFA), dehydrated in 30% sucrose, and sectioned using a cryostat (Leica Biosystems, Germany). Sections were then incubated sequentially with a blocking solution (1% BSA, 0.3% Triton X‐100, and 4% normal goat serum in PBS), primary antibodies, and corresponding secondary antibodies. Immunofluorescence staining images were randomly acquired using a STELLARIS 5 confocal microscope (Leica, Germany). The antibodies used in immunofluorescence staining included Iba1 (ab283319, Abcam, USA, 1:200), AQP4 (AF5146, Affinity Bioscience, USA, 1:200), NeuN (ab104224, Abcam, USA, 1:200), C1q (nbp1‐51139, novus, USA, 1:200), synaptophysin (14511‐1‐AP, Proteintech, China, 1:200), PSD95 (20665‐1‐AP, Proteintech, China, 1:200; 3490, Cell Signaling Technology, USA, 1:200), MAP2 (822501, Biolegend, USA, 1:500), glial fibrillary acidic protein (GFAP) (ab134436, Abcam, USA, 1:200), NeuN (ab104224, Abcam, USA, 1:200), and Piezo1 (15939‐1‐AP, Proteintech, China, 1:200).

2.10. Brain stereotaxic injection of bEVs

WT mice aged 2 to 3 months were anesthetized and positioned on a stereotaxic injection apparatus (RWD Life Science, USA). After making an incision to expose the skull, the mice were unilaterally injected with 0.1 µL bEVs pretreated with DiI (1 × 10⁸ particles/mL) on each side (from bregma, anterior posterior: −1.7 mm; mediolateral: ± 1.4 mm, dorsal ventral: −2.0 mm for the hippocampus) using a microliter syringe (Hamilton, USA) in 1 min. The needle was left in place for 10 min after the infusion before being removed. Mice were sacrificed 24 h after injection for staining.

2.11. LPS neutralization

bEVs with a concentration of 1 × 10¹¹ particles/mL were treated with 500 µg/ml polymyxin B sulfate for 30 min at 4°C, then the neutralization efficiency was verified by limulus amebocyte lysate (LAL) assay.

2.12. Nanoparticle tracking analysis (NTA)

To optimize particle count for bEVs derived from feces and blood of human and mice, the bEVs were diluted 1:100 with PBS filtered through a 0.22‐µm filter. Three 60‐s videos were recorded for each fraction and analysis using NTA 3.1 software (Nanosight, UK).

2.13. Western blot

Tissue samples were homogenized and the protein concentration was determined using the Protein Quantification Kit (23225, Thermo Fisher Scientific, USA). Proteins were separated on a 4% to 12% gel and transferred to a 0.22 µm PVDF membrane (Millipore, USA). The membranes were blocked with 5% skim milk for 1 h at room temperature, then incubated with specific primary antibodies at 4°C overnight. After three washes with TBST, the membranes were incubated with IRDye 800CW secondary antibodies (LI‐COR, USA) for 2 h at room temperature. Immunoreactive bands were visualized and quantified using the ChemiDoc MP imaging system (Bio‐RAD, USA) and ImageJ software, with GAPDH or β‐actin serving as internal controls. Antibodies used in Western blot included LPS (ab35654, Abcam, USA, 1:1000), OmpA (111120, Antibody Research, USA, 1:1000), CD9 (C‐4) (sc‐13118, Santacruz, USA, 1:1000), C1q (nbp1‐51139, Novus, USA; 11602‐1‐AP, Proteintech, China, 1:1000), C3 (21337‐1‐AP, Proteintech, China, 1:1000), CX3CR1 (nbp1‐76949, Novus, USA, 1:1000), SIRPα (14482‐1‐AP, Proteintech, China, 1:1000), CD47 (66301‐1‐AP, Proteintech, China, 1:1000), Piezo1 (15939‐1‐AP, Proteintech, China, 1:1000), anti‐GAPDH (G9545, Sigma Aldrich, USA, 1:5000), and anti‐β‐actin (66009‐1‐Ig, Proteintech, China, 1:10000).

2.14. Dot blot

Protein samples were diluted in PBS (1 µg/µL) and directly spotted onto the NC membrane. After air‐drying, the membrane was blocked with 5% BSA in TBST for 1 h to prevent non‐specific binding. Next, the membrane was incubated with the primary antibody diluted in blocking buffer overnight at 4°C, followed by three washes with TBST. A HRP‐conjugated secondary antibody was then applied for 1 h. Images were visualized using a ChemiDoc MP imaging system (Bio‐Rad, USA). Antibodies used in dot blot included HA‐Tag (3444, CST, USA, 1:1000) and HRP‐conjugated Affinipure Goat Anti‐Mouse IgG(H+L) (SA00001‐1, Proteintech, China, 1:1000).

2.15. LAL assay

The level of endotoxin in bEVs was determined by LAL Chromogenic Endpoint Assay Kit (MKC0505, Ximen Bioendo Technology, China) following the manufacturer's instructions. Briefly, a standard endotoxin stock solution was diluted to achieve concentrations within the linear range for assay calibration. The LAL reagent, standard endotoxin solutions, and bEVs were dispensed into 96 microplate wells. Spectrophotometry measured sample turbidity at 405 wavelengths every 30 s until 120 min, and a standard endotoxin calibration curve was established. The endotoxin concentration in the bEV samples was calculated based on the calibration curve and dilution factors. The entire procedure and materials adhered to pyrogen‐free standards.

2.16. Transmission electron microscopy (TEM)

The bEVs were suspended in PBS and applied to copper grids for 10 min and subsequently negatively stained using 2% phosphotungstic acid for 2 min. The morphology of the bEVs was examined with a JEM‐1400 PLUS microscope (JEOL, Japan). Six TEM images were captured for each sample.

2.17. Flow cytometry analysis

LPS and CD9 levels in bEVs were measured using flow cytometry analysis as previously described. ³⁰ Fluorophore‐conjugated antibodies were generated by labeling anti‐LPS antibody (ab35654, abcam, USA), with Zenon Alexa Fluor 488 mouse IgG2b Labeling Kit (z25208, Invitrogen), or anti‐CD9 antibody (sc‐13118, Santacruz, USA), with Zenon Alexa Fluor 488 mouse IgG1 Labeling Kit (z25002, Invitrogen, USA). A 10‐µL sample of bEVs was blocked with 2% BSA for 1 h at room temperature, then diluted with 10 µL PBS (pH 7.4, 0.22‐µm filter). The blocked bEVs were incubated overnight at 4°C with fluorophore‐conjugated antibodies (anti‐LPS or anti‐CD9 antibody [0.06 µg each], or corresponding IgG isotype control antibody [0.06 µg]). The samples were fixed with 20 µL 4% PFA (pH 7.4, 0.22‐µm filter) for 20 min at room temperature and analyzed using the CytoFLEX Lx (Beckman Coulter, Germany).

2.18. ELISA analysis of lipoteichoic acid (LTA) level

The level of LTA in bEVs was determined with an ELISA kit following the manufacturer's instructions (JM‐1303302, Jingmei Biological Technology, China).

2.19. Gram staining

A heat‐fixed bacterial smear was prepared on a glass slide and stained with crystal violet for 1 min, then rinsed with distilled water. The slide was then treated with iodine solution for 1 min, rinsed again, and decolorized with ethanol or acetone until the runoff was colorless, followed by another rinse with distilled water. The smear was subsequently counterstained with safranin for 1 min and rinsed with distilled water. After air‐drying, the slide was examined under a BX53 light microscope (Olympus, Japan). G+ bacteria exhibited a purple or blue coloration, while G− bacteria appeared pink or red.

2.20. Quantitative real‐time PCR (qRT‐PCR)

RNA was extracted and converted to cDNA using the SuperScript III First‐Strand Synthesis System (Invitrogen, USA) with oligo (dT) 20 primer. qRT‐PCR was conducted with ChamQ Universal SYBR qPCR Master Mix (Q711, Vazyme, China) on the QuantStudio 5 Real‐Time PCR System (Applied Biosystems, USA). The sequence of primers was listed in the Table S4.

2.21. Synaptic imaging and quantification

For synaptic puncta colocalization, brain sections (paraffin or frozen) were co‐immunostained with anti‐SYP and anti‐PSD95 antibodies and imaged with a ×63 oil objective and ×3 zoom on a STELLARIS 5 confocal microscope (Leica, Germany). Images were processed with ImageJ, and colocalized puncta was quantified with the Synapse Counter plugin. ³¹ To quantify PSD95 in Iba1‐positive microglia, brain paraffin or frozen sections were co‐immunostained with anti‐Iba1 and anti‐PSD95 antibodies, imaged with a ×63 oil objective in z‐stack mode on a confocal microscope, and analyzed with the surface function in Imaris software.

2.22. Cell culture and treatment

HMC3 cells, sourced from ATCC, were cultured in DMEM (Corning, USA) with 10% FBS (Quacell Biotechnology, China) and 1% penicillin/streptomycin (Corning, USA) at 37°C in 5% CO₂ atmosphere. The cells were seeded on six‐well plates (Thermo Fisher Scientific, USA) at 2 × 10⁵ cells per well for 24 h and then serum‐starved for 12 h. After replacing the medium with fresh serum‐free DMEM, the cells were treated with bEVs (1 × 10¹¹ particles/mL) from human feces for 24 h. 5 µM GsMTx4 was used to inhibit Piezo1, and 10 µM dynasore (HY‐15304, MCE, USA) was used to inhibit endocytosis of bEVs in HMC3 cells. Cells were subsequently collected for the Western blot or qPCR analysis.

2.23. Primary cell cultures and treatment

Primary microglia and astrocyte were isolated from the neonatal mouse cortex. Neonatal mouse cortices were collected, meninges removed, and tissues placed in cold DMEM (C11995500BT, Gibco, USA). The tissues were chopped, digested with 0.25% EDTA‐trypsin for 5 min at 37°C, and centrifuged. The pellet was resuspended in DMEM, filtered through a 70‐µm mesh to create a single‐cell suspension and seeded in tissue flasks at 1 × 10⁶ cells/mL. Once the cells reached 90% confluence, the flasks were shaken overnight at 220 rpm at 37°C. Primary microglia and astrocytes were then separately collected for in vitro experiments.

Primary neurons were obtained from the neonatal mouse cortex. Neonatal mouse cortices were dissected in ice‐cold Hank's balanced salt solution (HBSS, Invitrogen, USA), cut into small pieces, and digested with 0.25% EDTA‐trypsin for 2 min at 37°C. The tissue was mechanically dissociated in DMEM using a sterile Pasteur pipette, then centrifuged at 1000 × g for 5 min at room temperature. The neuronal suspension was plated at 3 × 10⁴ cells/cm² on poly‐D‐lysine‐coated dishes. After 2 days, 2 µM cytosine arabinoside was added to prevent glial cell proliferation. The medium was half‐replaced every 3 days, and experiments were performed on day 5.

Microglia or astrocytes were cocultured with primary neurons for 24 h at 37°C and then treated with a gradient concentration of bEVs (from 1×10⁷ to 1×10⁹ particles/mL) from mouse intestinal feces for 3, 6, 12, and 24 h in the incubator. The cells were pretreated with 1 ng/mL ANX005 (HY‐P990545, MCE, USA), a humanized recombinant antibody against C1q, for 1 h to inhibit the complement cascade. Immunocytochemical staining was performed to observe the synaptic pruning.

2.24. Ca²⁺ analysis

To measure the intracellular calcium level, cells were incubated with 10 nM Fluo 4‐AM probe (S1060, Beyotime, China) for 30 min at 37°C, followed by flow cytometry analysis (Beckman Coulter, Germany).

2.25. Statistical analysis

All statistical analyses were performed using Prism 9.0 (GraphPad Software). Results are presented as means ± standard error of the mean (SEM). To determine statistical significance, t‐tests were used for two comparisons, while one‐way ANOVA with Bonferroni's post hoc test was applied for multiple comparisons. Significance levels are indicated as follows: *p < .05; **p < .01; ***p < .001.

3. RESULTS

3.1. Characterization of gut‐derived bEVs in the human feces and blood

bEVs were extracted from feces and plasma of healthy individuals using established protocols with ultracentrifugation and SEC. ³² Nanoparticle tracking analysis (NTA) revealed a consistent particle size and distribution of 50 to 150 nm for bEVs isolated from both feces and plasma (Figure S1A). Transmission electron microscopy (TEM) further illustrated a double‐layered membrane structure aligning with the NTA results (Figure S1B). The purity of bEVs from feces and plasma was validated by the enrichment of LPS and OmpA, specific components of G− bacteria outer membranes (Figure S1C‐D). The presence of LPS in the extracted bEVs was confirmed using the LAL assay (Figure S1E), while LTA, a marker of G+ bacteria, was also assessed via ELISA (Figure S1F). Additionally, the absence of CD9, a canonical marker for host‐cell‐derived extracellular vesicles (EVs), was verified using Western blot and flow cytometry (Figure S1C,S1G). These findings confirm the detectability of bEVs in feces and plasma from healthy individuals. Parallel findings were observed in bEVs isolated from the fecal samples of healthy WT mice (Figure S2A–E).

To directly investigate whether bEVs could transit from the gut to the bloodstream, mice were intracolonically injected with bEVs derived from E. coli BL21 (DE3) transfected with plasmids encoding HA and mPlum tags. A subsequent dot blot assay clearly revealed HA and mPlum signals in both the intestine and blood, indicating that bEVs could enter the bloodstream after gut inoculation (Figure S1H–I).

3.2. Transportation of blood bEVs across BBB

To investigate the potential for blood bEVs to cross the BBB, we administered bEVs extracted from the gut microbiota of 2‐month‐old WT mice intravenously to WT mice of the same age. In vivo imaging conducted at 2 h revealed prominent DiR signals in the brain of mice treated with bEVs (Figure 1A). Given the complexity of cell types in the central nervous system (CNS), their differential responses to bEVs were explored. Immunofluorescence analysis demonstrated that, following intravenous injections of bEVs labeled with DiI at 0.5, 2, 6, 18, and 24 h, bEVs rapidly entered astrocytes marked by aquaporin 4 (AQP4) as early as 0.5 h. Their distribution peaked particularly in the cortex and hippocampus (Figure 1B,C). However, 6 h after the injection, microglia, visualized using the specific marker Iba1, surpassed astrocytes in bEV distribution and continued to accumulate in the cortex and hippocampus (Figure 1D,E). Notably, a small amount of bEV signals were also observed in neurons marked by NeuN (Figure 1F). To further assess the transport of bEVs from the gut to the brain, we measured the relative abundance of HA‐bEVs in the blood and brain. Mice were injected intestinally with bEVs derived from HA‐mPlum‐transfected E. coli, and HA levels were analyzed via Western blot. At 24 h after injection, HA levels in plasma and the brain were approximately 0.68 and 0.01 of those in the injected bEVs, respectively. While this does not precisely quantify the proportion of bEVs migrating from the gut to the brain, it indicates that a subset of bEVs eventually reaches cerebral tissue (Figure S1J,K). Taken together, these findings indicate that microbiota‐derived bEVs can traverse the BBB, accumulating predominantly in the cortex and hippocampus. They initially enter astrocytes, followed by a pronounced and sustained increase in microglia.

FIGURE 1.

Distribution of bEVs within mouse brain. (A) In vivo imaging shows the presence and distribution of bEVs labeled with DiR in the brain of 2‐month‐old mice treated with PBS or bEVs for 3 h. (B and C) Representative immunofluorescence images and quantitative analysis depict the colocalization of bEVs labeled with DiI and AQP4 in the cortex and hippocampus at various time points (0.5, 2, 6, 18, and 24 h) after injection (N = 3 mice in each group; five randomly taken images per mouse were measured and averaged into a single value). (D and E) Representative immunofluorescence images and quantitative analysis show the colocalization of bEVs labeled with DiI and Iba1 in the cortex and hippocampus at various time points (0.5, 2, 6, 18, and 24 h) after injection (N = 3 mice in each group; five randomly taken images per mouse were measured and averaged into a single value). (F) Representative immunofluorescence images depict the colocalization of bEVs labeled with DiI and NeuN in the cortex and hippocampus at various time points after injection. Scale bar: 25 µm for B, D, and F. Values are means ± SEM. bEV, bacterial extracellular vesicle; AQP4, aquaporin 4; Cx, cortex; Hp, hippocampus.

3.3. The role of LPS in transportation of blood bEVs to brain

We further explored whether LPS affected the transport of blood bEVs across the BBB. Our results showed that bEVs contained plenty of LPS (Figure S1C–E and S2C,D), but to delve deeper into the influence of LPS on the penetration of bEVs, bEVs were extracted from G− and G+ bacteria (Figure 2A–C), respectively. In vivo imaging (Figure 2D) and immunofluorescence analysis (Figure 2E,F) demonstrated signals in the brains of mice treated with G− bacteria‐derived bEVs, which were absent in those treated with G+ bacteria‐derived bEVs, suggesting that bEVs from G− bacteria possess an inherent advantage in traversing the BBB. To confirm that LPS was involved in this process, we used the antibiotic probe polymyxin B‐cy3 (Pxb‐cy3) to label LPS. Immunofluorescence results exhibited signals of Pxb‐cy3 in astrocytes and microglia in the cortex and hippocampus of mice (Figure 2G,H), suggesting that LPS may contribute to brain penetration of bEVs. To directly verify whether LPS helps bEVs to enter the brain, LPS in bEVs was neutralized by polymyxin B sulfate (Figure 2I), which is an antibiotic neutralizing endotoxin. In vivo imaging (Figure 2J) and immunofluorescence (Figure 2K) showed that bEV signals were absent in the brain of mice following intravenous injections of bEVs pretreated with polymyxin B sulfate. Finally, using endotoxin‐free BL21 as a control, we injected BL21‐derived bEVs, tagged with both mPlum and HA, into mice via the tail vein. Immunofluorescence analysis detected mPlum signals of BL21‐derived bEVs in astrocytes and microglia within the cortex and hippocampus, whereas no signals were observed from endotoxin‐free BL21‐derived bEVs (Figure S3A,B). These results further indicate that LPS is essential for bEVs to enter the brain.

FIGURE 2.

LPS is necessary for the entry of bEVs into the brain. (A) Visualization of G+ and G− bacteria through Gram staining. (B) Western blot analysis detects the levels of LPS in bEVs derived from G+ and G− bacteria. (C) Flow cytometry evaluates LPS level in bEVs from G+ and G− bacteria (N = 3 bEV samples, each derived from a different mouse in each group). (D) In vivo imaging illustrates the distribution of bEVs from G+ and G− bacteria labeled with DiR in the brain of 2‐month‐old mice treated with PBS or bEVs via tail vein injection for 3 h. (E and F) Representative immunofluorescence images show DiI‐labeled bEVs from G+ and G− bacteria colocalized with AQP4 and Iba1 in the cortex and hippocampus of 2‐month‐old mice treated with PBS or bEVs via tail vein injection for 3 h. (G and H) Representative immunofluorescence images show Pxb‐cy3‐labeled bEVs colocalized with AQP4 and Iba1 in the cortex and hippocampus of 2‐month‐old mice treated with PBS or bEVs via tail vein injection for 3 h. (I) LAL assay measures the endotoxin levels in bEVs treated with or without PxB (N = 3 bEV samples, each derived from a different mouse in each group). (J) In vivo imaging shows the distribution of DiR‐labeled bEVs pretreated with or without PxB in the brain of 2‐month‐old mice for 3 h. (K) Representative immunofluorescence images show DiI‐labeled bEVs colocalized with Iba1 in the cortex and hippocampus of 2‐month‐old mice treated with PBS or bEVs (pretreated with or without PxB) via tail vein injection for 3 h. Scale bar: 25 µm for E–H and K. Values are means ± SEM. *p < .05. bEV, bacterial extracellular vesicle; G+, Gram‐positive; G−, Gram‐negative; AQP4, aquaporin 4; Pxb‐cy3, polymyxin B‐cy3; PxB, polymyxin B.

As indicated earlier, previous studies showed that LPS levels were higher in the brains of AD patients compared to control brains, and LPS may promote AD progression. ⁴ Remarkably, analyses through Western blot and flow cytometry revealed increased LPS expression in bEVs from the feces and blood of AD patients compared to healthy individuals (Figure 3A–C). LPS levels were also higher in plasma EVs, including eukaryotic EVs (eEVs) and bEVs, from MCI and AD patients (Figure 3D), suggesting that increased LPS in the AD brain might come from blood bEVs derived from the gut. Similar results were obtained in WT mice and APP/PS1 transgenic (AD) mice (Figure 3E–G). We also explored whether the increased LPS levels in bEVs of AD mice could enhance their entry into the brain by intravenously injecting bEVs extracted from WT and AD mice into WT mice. In vivo imaging demonstrated more abundant entry of bEVs from AD mice into the brain compared to those from WT mice (Figure 3H,I). Notably, similar to WT mice but to a greater extent, more bEVs from AD mice were observed in astrocytes and microglia in the cortex and hippocampus compared to those from WT mice (Figure 3J–M). These results suggest that AD process is associated with both increased blood bEV levels and facilitated bEV penetration into the brain.

FIGURE 3.

Enhanced entry of bEVs into the brain induced by LPS. (A and B) Representative Western blot and corresponding densitometry analysis reveal LPS level in bEVs from feces of HC and AD patients (N = 4 bEV samples, each derived from a different individual in each group). (C) Flow cytometry analysis shows LPS level in bEVs from feces (N = 6 bEV samples, each derived from a different individual in each group) and plasma (N = 4 bEV samples, each derived from a different individual in each group) of HC and AD patients. (D) Flow cytometry analysis shows LPS level in total EVs extracted from plasma of HC, MCI, and AD patients (N = 10, 8, and 9 bEV samples, each derived from a different individual for HC, MCI, and AD group, respectively). (E–F) Representative western blot and corresponding densitometry analysis reveal LPS levels in bEVs from feces of 12‐month‐old WT and APP/PS1 (AD) mice (N = 3 bEV samples, each derived from a different mouse in each group). (G) Flow cytometry analysis shows LPS levels in bEVs extracted from feces of 12‐month‐old WT and APP/PS1 (AD) mice (N = 11 and 9 bEV samples, each derived from a different mouse for WT and AD group, respectively). (H–I) In vivo imaging and quantitative analysis show the distribution of bEVs labeled with DiR in the brains of 2‐month‐old mice treated with PBS or bEVs from 12‐month‐old WT or AD mice via tail vein injection for 3 h (N = 3 mice in each group). (J and K) Representative immunofluorescence images and statistical analysis of DiI‐labeled bEVs colocalized with AQP4 in the cortex and hippocampus of 2‐month‐old mice treated with bEVs from 12‐month‐old WT or AD mice via tail vein injection for 3 h (N = 3 mice in each group; five randomly taken images per mouse were measured and averaged into a single value). (L and M) Representative immunofluorescence images and statistical analysis of DiI‐labeled bEVs colocalized with Iba1 in the cortex and hippocampus of 2‐month‐old mice treated with bEVs from 12‐month‐old WT or AD mice via tail vein injection for 3 h (N = 3 mice in each group; five randomly taken images per mouse were measured and averaged into a single value). Scale bar: 25 µm for J and L. Values are means ± SEM. ns, not significant; *p < .05. bEV, bacterial extracellular vesicle; LPS, lipopolysaccharide; HC, healthy control; AD, Alzheimer's disease; MCI, mild cognitive impairment; WT, wild type.

3.4. bEV‐mediated microglial activation and synaptic pruning via C1q‐C3 complement pathway in vivo and in vitro

The mechanisms underlying LPS‐mediated bEV transmission through the BBB remain to be fully understood; however, when exposed to LPS directly, it has been established that microglia can be readily activated, leading to significant synaptic pruning in AD pathological process. ³³ , ³⁴ In our study, mice injected with bEVs via tail vein exhibited a noticeable microglial activation in the cortex and hippocampus at 6 h after injection, characterized by an increase in Iba1 fluorescence intensity, enlarged cell bodies, and retracted branching extensions (Figure 4A,B). In contrast, immunofluorescence of GFAP and AQP4 in the same brain regions revealed no significant activation of astrocytes in bEV‐treated mice (Figure S4A–C). In vitro experiments involving primary astrocytes and microglia treated with bEVs, at concentrations of 1 × 10^7, 1 × 10^8, and 1 × 10^9 particles/mL, at 3, 6, 12, and 24 h demonstrated that microglia were activated at low bEV concentrations at 12 h, while astrocytes showed only slightly activation at high bEV concentrations after 24 h (Figure S4D–I). Dual staining of synaptophysin and PSD95 in mice revealed a substantial decrease in synaptic density in the hippocampus 24 h after bEV treatment (Figure 4C–F). Costaining of PSD95 and Iba1 demonstrated increased microglial engulfment of synaptic elements in mice treated with bEVs compared to the control group (Figure 4G,H). To exclude the systematic effects caused by bEVs via intravenous injection, mice were additionally administered bEVs directly via stereotactic injection in the hippocampus. Immunofluorescence analysis revealed a significant increase in microglia activation and synapse loss following direct bEV injection (Figure 4I–N). To further validate the involvement of LPS in bEV‐induced microglial activation and synapse loss, a neutralization assay was performed. The results showed that neutralization of LPS by polymyxin B remarkably reversed the bEV‐induced microglial activation and excessive synaptic pruning (Figure 4I–N). In vitro experiments using a primary neuron and microglia/astrocyte cocultured system revealed a significant decrease in PSD95 intensity in neurons cocultured with microglia treated with bEVs (Figure 4O–P), while neurons cocultured with astrocytes treated with bEVs showed only a minimal reduction (Figure S4J–K). These results indicate that bEVs carrying LPS primarily modulate synaptic plasticity by activating microglia.

FIGURE 4.

bEVs elicit synaptic pruning mediated by microglia in vivo and in vitro. (A and B) Representative immunofluorescence images and quantitative analysis show Iba1 in the cortex and hippocampus of 2‐month‐old mice treated with bEVs for 0.5, 2, 6, 18, and 24 h via tail vein injection (N = 3 mice in each group; three randomly taken images per mouse were measured and averaged into a single value). (C and D) Representative immunofluorescence images and quantitative analysis show the colocalization of SYN and PSD95 in the hippocampus of 2‐month‐old mice treated with PBS or bEVs via tail vein injection for 24 h (N = 6 mice in each group; three randomly taken images per mouse were measured and averaged into a single value). (E and F) Representative high‐magnification immunofluorescence images and quantitative analysis show the colocalization of SYN and PSD95 in the hippocampus of 2‐month‐old mice treated with PBS or bEVs via tail vein injection for 24 h (N = 6 mice in each group; five randomly taken images per mouse were measured and averaged into a single value). (G and H) Three‐dimensional reconstruction and surface rendering demonstrate larger volumes of PSD95 puncta inside Iba1‐positive microglia in the cortex and hippocampus of 2‐month‐old mice treated with PBS or bEVs via tail vein injection for 24 h (N = 6 mice in each group; six randomly captured microglia per mouse were measured and averaged into a single value). (I and J) Representative immunofluorescence images and quantitative analysis show Iba1 in the hippocampus of 2‐month‐old mice treated with DiI labled PBS, Pxb, bEVs, or bEVs+PxB via brain stereotactic injection for 24 h (N = 5 mice in each group; five randomly taken images per mouse were measured and averaged into a single value). (K and L) Representative immunofluorescence images and quantitative analysis show the colocalization of SYN and PSD95 in the hippocampus of 2‐month‐old mice treated with PBS, Pxb, bEVs, or bEVs+PxB via brain stereotactic injection for 24 h (N = 5 mice in each group; three randomly taken images per mouse were measured and averaged into a single value). (M and N) Representative high‐magnification immunofluorescence images and quantitative analysis show the colocalization of SYN and PSD95 in the hippocampus of 2‐month‐old mice treated with PBS, PxB, bEVs, or bEVs+PxB via brain stereotactic injection for 24 h (N = 5 mice in each group; five randomly taken images per mouse were measured and averaged into a single value). (O and P) Representative immunofluorescence images and quantitative analysis show the presence of synaptic puncta (PSD95+) around microglia (Iba1+) in primary neuron and microglia cocultured system, and microglia were treated with or without bEVs (N = 3 independent rounds of experimentation, each using primary neurons and microglia isolated from five different neonatal mice. In each round, six randomly selected neurons per well were measured and averaged to generate a single data point). Scale bar: 25 µm for A, C, G, I, K, and O, 8 µm for E and M. Values are means ± SEM. *p < .05. bEV, bacterial extracellular vesicle; SYN, synaptophysin; PxB, polymyxin B.

Microglia‐mediated synaptic pruning involves multiple pathways, including the classical complement system, CX3CR1, and the SIRPα‐CD47 axis. ³⁵ , ³⁶ , ³⁷ In the complement pathway, microglial C1q, a key element, has been demonstrated to promote microglial engulfment of synaptic structures via activation of the C3 receptor (CR3). ³⁸ , ³⁹ Western blot and qPCR results revealed an increased trend of expression of C1q and C3 in a human microglial cell line derived from SV40 immortalized human fetal brain primary microglial cell cultures, HMC3 cells, treated with bEVs (Figure 5A–C). The expression of CX3CR1 and SIRPα were slightly increased or unchanged in HMC3 cells treated with bEVs (Figure S5A,B). The involvement of C1q in bEV‐induced synaptic pruning was further confirmed in vivo, with heightened mRNA and protein levels of C1q and its downstream protein C3 in the cortex and hippocampus of mice following bEV treatment (Figure 5D–G). Notably, no significant alterations were observed in the levels of CX3CR1, SIRPα, and CD47 in the cortex and hippocampus of mice subsequent to bEV treatment (Figure S5C–G). Furthermore, Western blot showed that neutralization of LPS significantly reduced the elevation of C1q levels in the hippocampus of mice induced by bEVs (Figure 5H,I). The role of C1q in bEV‐induced synaptic pruning was further confirmed in vitro with a primary neuron and microglia cocultured system, where inhibiting C1q by ANX005 reversed the bEV‐induced reduction in PSD95 density (Figure 5J–K).

FIGURE 5.

bEVs induced C1q‐mediated synaptic pruning through activating Piezo1. (A and B) Western blot and quantitative analysis demonstrate the protein levels of C1q in HMC3 cells treated with PBS or bEVs (N = 3 independent rounds of experimentation. In each round, three wells of HMC3 cells per group were measured and averaged to generate a single data point). (C) qPCR results and quantitative analysis demonstrate mRNA levels of C1q and C3 in HMC3 cells treated with PBS or bEVs (N = 5 independent rounds of experimentation. In each round, four wells of HMC3 cells per group were measured and averaged to generate a single data point). (D and E) Western blot and quantitative analysis demonstrate protein levels of C1q in the cortex and hippocampus of 2‐month‐old mice treated with PBS or bEVs via tail vein injection for 24 h (N = 6 and 8 mice for PBS and bEVs groups respectively). (F and G) qPCR results and quantitative analysis demonstrate mRNA levels of C1q and C3 in the cortex and hippocampus of 2‐month‐old mice treated with or without bEVs via tail vein injection for 24 h (N = 3 and 5 mice for PBS and bEVs group, respectively). (H and I) Western blot and quantitative analysis demonstrate protein level of C1q in the hippocampus of 2‐month‐old mice treated with PBS, Pxb, bEVs, or bEVs+PxB via brain stereotactic injection for 24 h (N = 3 mice in each group). (J and K) Representative immunofluorescence images and quantitative analysis show the presence of synaptic puncta (PSD95+) in a primary neuron and microglia cocultured system, and coculture cells were pretreated with ANX005 for 1 h, followed by treatment with PBS or bEVs (N = 3 independent rounds of experimentation, each using primary neurons and microglia isolated from five different neonatal mice. In each round, six randomly captured neurons per well were measured and averaged to generate a single data point). (L and M) Western blot and quantitative analysis demonstrate protein levels of C1q in HMC3 cells pretreated with or without 5 µM GsMTx4 for 1 h, followed by bEV treatment (N = 3 round of experimentation. In each round, four wells of HMC3 cells per group were measured and averaged to generate a single data point). Scale bar: 10 µm for J. Values are means ± SEM. ns, not significant; *p < .05. bEV, bacterial extracellular vesicle; PxB, polymyxin B.

3.5. Microglial Piezo1 in bEV‐mediated synaptic pruning in vitro and ex vivo

The foregoing results suggest that bEVs induce pronounced synaptic pruning through the C1q‐C3 complement pathway both in vivo and in vitro. However, the specific target modulated by bEVs in microglia has yet to be elucidated. The mechanosensitive ion channel Piezo1, which is readily activated by bacterial infections or LPS stimulation, ⁴⁰ exhibits high expression in microglia. ⁴⁰ , ⁴¹ In this study, we observed that (1) the treatment of HMC3 cells with bEVs led to increased mRNA and protein levels of Piezo1 (Figure S6A–C), accompanied by elevated intracellular calcium concentration ([Ca²⁺] i) (Figure S6D); this effect was partially attenuated by the Piezo1 non‐specific inhibitor GsMTx4 (Figure S6D); (2) pretreatment of HMC3 cells with the endocytosis inhibitor dynasore, followed by bEV treatment, demonstrated that inhibiting bEV endocytosis suppressed the elevation in [Ca²⁺]i induced by bEVs (Figure S6E); (3) immunofluorescence analysis showed that bEVs increased Piezo1 levels in primary cultured microglia, while no change in Piezo1 levels was observed in primary cultured astrocyte (Figure S4D–I). These findings suggest that bEV internalization in microglia is primarily responsible for Piezo1 activation.

Next, we explored whether bEV‐induced activation of Piezo1 was implicated in C1q‐mediated synaptic pruning. Western blot results demonstrated increased protein and mRNA levels of C1q in HMC3 cells treated with bEVs, which were partially suppressed by Piezo1 inhibitor GsMTx4 (Figure 5L,M). These findings suggest that bEVs might activate Piezo1 through endocytosis, subsequently activating the C1q‐C3 pathway, leading to excessive synaptic pruning and ultimately reducing the number of neuronal synapses.

To investigate whether Piezo1 was indeed involved in human diseases, brain cortex samples from patients with AD were examined. Immunofluorescence analysis revealed increased expression of Piezo1 and colocalization of Piezo1 with Iba1 in the cortex of AD patients compared to healthy individuals (Figure S7A). Additionally, C1q expression and PSD95 engulfment by microglia were also increased in the cortex of AD patients compared to healthy individuals (Figure S7B–D).

3.6. Microglial Piezo1: a potential therapeutic target for bEV‐induced synaptic pruning

To further dissect the role of microglial Piezo1 in bEV‐induced synaptic pruning, we conducted intravenous injection experiments in WT mice with a Piezo1 inhibitor. Immunofluorescence results demonstrated that bEV treatment significantly induced microglial activation (Figure S8A,B), reduced synaptic density (Figure S8C–F), and increased the engulfment of microglia toward PSD95 (Figure S8G,H) in the cortex and hippocampus of mice, with these effects being reversed by prior administration of the Piezo1 non‐specific inhibitor GsMTx4. Western blot and qPCR results demonstrated that the bEV‐induced elevation of C1q and C3 expression also decreased in mice prior to administration of GsMTx4 (Figure S8I–L).

To further confirm the role of Piezo1 in bEV‐induced synaptic pruning and loss of synapses, myeloid‐specific Piezo1‐deficient mice (Piezo1^∆LysM ) and Piezo1^fl/fl mice (Figure S9A,B) were treated with bEVs for 24 h. Immunofluorescence results demonstrated that bEV‐treated Piezo1^fl/fl mice exhibited activation of microglia (Figure 6A,B), decreased synaptic density (Figure 6C–F), and increased engulfment of microglia (Figure 6G,H) in the cortex and hippocampus compared to PBS‐treated control. Western blot and qPCR results also showed an increased expression of C1q and C3 in Piezo1^fl/fl mice treated with bEVs (Figure 6I–L). Remarkably, in Piezo1^∆LysM mice, bEV‐induced decreased synaptic density and increased expression of C1q and C3 were markedly suppressed (Figure 6I–L).

FIGURE 6.

Myeloid‐specific Piezo1 deficiency mice resisted the synaptic loss and activation of complement system induced by bEVs. (A and B) Representative immunofluorescence images and quantitative analysis show Iba1 in the cortex and hippocampus of Piezo1^fl/fl and Piezo1^∆LysM mice treated with PBS or bEVs via tail vein injection for 24 h (N = 6 mice in each group; five randomly taken images per mouse were measured and averaged into a single value). (C and D) Representative immunofluorescence images and quantitative analysis show the colocalization of SYN and PSD95 in the hippocampus of Piezo1^fl/fl and Piezo1^∆LysM mice treated with PBS or bEVs via tail vein injection for 24 h (N = 6 mice in each group; three randomly taken images per mouse were measured and averaged into a single value). (E and F) Representative high‐magnification immunofluorescence images and quantitative analysis show the colocalization of SYN and PSD95 in the hippocampus of Piezo1^fl/fl and Piezo1^∆LysM mice treated with PBS or bEVs via tail vein injection for 24 h (N = 6 mice in each group; five randomly taken images per mouse were measured and averaged into a single value). (G and H) Three‐dimensional reconstruction and surface rendering demonstrate larger volumes of PSD95 puncta inside Iba1‐positive microglia in the cortex and hippocampus of Piezo1^fl/fl and Piezo1^∆LysM mice treated with PBS or bEVs via tail vein injection for 24 h (N = 6 mice in each group; six randomly captured microglia per mouse were measured and averaged into a single value). (I and J) Western blot and quantitative analysis demonstrate protein levels of C1q in the cortex and hippocampus of Piezo1^fl/fl and Piezo1^∆LysM mice treated with PBS or bEVs via tail vein injection for 24 h (N = 3 Piezo1^fl/fl mice and N = 4 Piezo1^∆LysM mice in both PBS and bEV group, respectively). (K and L) qPCR and quantitative analysis demonstrate mRNA levels of C1q and C3 in the cortex and hippocampus of Piezo1^fl/fl and Piezo1^∆LysM mice treated with PBS or bEVs via tail vein injection for 24 h (N = 3 Piezo1^fl/fl mice and N = 4 Piezo1^∆LysM mice in both PBS and bEVs group, respectively). Scale bar: 25 µm for A, C, and G, 8 µm for E. Values are means ± SEM. *p < .05. bEV, bacterial extracellular vesicle; Cx, cortex; Hp, hippocampus; SYN, synaptophysin.

Finally, to further support the role of LPS and Piezo1 in synaptic pruning, we employed the cecal ligation and puncture (CLP) model to simulate sepsis, marked by a significant elevation in LPS level in the gut and blood. As shown in Figure S10, CLP for 24 h also induced activation of microglia and reduced synaptic density in the cortex and hippocampus of Piezo1^fl/fl mice, with these effects significantly suppressed in Piezo1^∆LysM mice. In summary, these findings demonstrate that the depletion of Piezo1 in microglia can ameliorate bEV‐induced synaptic pruning and the loss of synapses in the CNS.

4. DISCUSSION

The ongoing inquiry has yielded several novel observations, including the following: (1) a discernible elevation of bEVs containing LPS was observed in individuals with AD and MCI compared to healthy individuals; (2) LPS appears to play a regulatory role in facilitating the transit of GIT bEVs across the BBB; and (3) bEVs harboring LPS exhibited robust activation of microglial Piezo1, consequently precipitating an excessive synaptic pruning process mediated by the C1q‐C3 complement pathway.

Gut microbiota have emerged as a prevailing concept in neurodegenerative diseases, yet the underlying pathogenesis remains incompletely elucidated. ¹ , ² Compelling evidence suggests that bEVs, particularly those secreted by probiotics, can deliver bioactive molecules, modulate host signaling pathways, and influence distal organ functions. ¹⁸ , ⁴² Here, we designed a combination of bEV isolation techniques to extract bEVs from the feces and blood of both humans and mice. These bEVs exhibited typical EV characteristics, including a bilayer membrane structure and a size distribution of 50 to 150 nm, along with robust expression of OmpA and LPS, which are markers of host bacteria. We also discovered that EVs from the blood of AD and MCI patients showed higher LPS levels than those from healthy controls (Figure 5G), consistent with the results observed in AD mice, indicating that bEVs, together with LPS, likely serve as a key component in the AD pathogenesis.

Previous studies demonstrated that free forms of LPS can activate microglia, leading to significant synaptic pruning in the setting of AD pathogenesis. ⁴³ However, there is no conclusive evidence supporting the direct entry of free forms of LPS into the brain parenchyma. ¹⁵ , ⁴⁴ Therefore, this study focused on bEVs carrying LPS and found that these LPS‐containing bEVs could directly penetrate the BBB. However, we cannot entirely rule out the possibility that some LPSs may dissociate and be taken up by host EVs before bEVs enter the CNS, as host‐derived EVs have been reported to capture blood‐borne LPSs in vivo through the interaction of the lipid bilayer of EVs with the lipid A of LPS mediated by CD14. ⁴⁵ This mechanism warrants further investigation, as our study also found increased levels of host EVs associated with LPS in AD and MCI patients.

Growing evidence suggests that blood bEVs might penetrate the BBB via receptor‐mediated transcytosis, transcellular pathways, and endocytosis. ²² , ⁴⁶ However, a key observation from this study is that LPS plays a critical role in enabling bEVs to cross the BBB. Specifically, after intravenous injections of bEVs labeled with DiI to 2‐month‐old WT mice, the bEVs appeared predominantly in the cortex and hippocampus, following a chronological sequence involving astrocytes and microglia. In contrast, bEVs derived from G+ bacteria were barely detectable in the brain after injection, indicating an essential role of LPS in mediating bEVs’ penetration into the brain. Further evidence for the role of LPS in BBB penetration comes from experiments showing that bEVs treated with polymyxin B sulfate, which neutralizes the LPS on bEVs, could hardly cross the BBB. Moreover, only BL21‐derived bEVs exhibited both HA and mPlum signals in astrocytes and microglia in the cortex and hippocampus of mice, compared to endotoxin‐free BL21‐derived bEVs, suggesting LPS is necessary for bEVs’ penetration.

It remains unclear how bEVs carrying LPS traverse the BBB. Previous research indicated that OmpA enabled E. coli K1's bacterial adhesion to brain endothelial cells by triggering GRP94‐mediated vesicular transport. ⁴⁷ Additionally, Toll‐like receptors might mediate bEV penetration of the BBB, as demonstrated in a study involving neutrophils. ⁴⁸ , ⁴⁹ Furthermore, LPS‐laden EVs can facilitate cytosolic access for LPS, triggering non‐canonical inflammasome activation of gasdermin D and pyroptosis. ⁴⁵ Finally, AD is often associated with significant peripheral inflammation and dysfunctions of the BBB ⁵⁰ , ⁵¹ , ⁵² ; it is possible that LPS, transferred intracellularly by bEVs, exacerbates BBB damage and further contributes to the transportation of bEVs to the brain and AD progression.

Regardless of the mechanisms involved in bEV entry into the brain, in mice treated with Helicobacter pylori bEVs, both astrocytes and microglia can internalize bEVs and induce an inflammatory response, ⁴⁶ , ⁵³ together with Aβ accumulation and cognitive impairment. ²⁵ Consistent with this observation, our study demonstrated that, within 2 h of intravenous injection, as well as 24 h of stereotactic injection in hippocampus, bEVs induced prominent microglial activation in WT mice. As the primary phagocytic and immune‐effector cells in the CNS, besides removing unwanted debris, microglia are crucial for the maturation of synaptic connections and stable construction of neural circuits. ⁵⁴ , ⁵⁵ Indeed, within 24 h of bEV injection, a significant decrease in synaptic density in the cortex and hippocampus was observed, accompanied by an increase in PSD95‐positive synaptic density near bEV‐treated microglia compared to the control group. Recent studies underscored the crucial role of reactive microglia in neurodegenerative diseases and excessive synaptic pruning. Expanding on this concept, our research shows that bEVs carrying LPS influence synaptic plasticity by activating microglia, suggesting that microglia play a central role in the neurodegenerative process. Notably, since synaptic pruning is commonly regarded as an early event in AD progression, further studies using chronic models are needed to assess behavioral changes at various stages of AD, with and without bEV treatment.

Among the principal pathways regulating synaptic plasticity regulation mediated by microglia, ³⁷ , ⁵⁶ our findings revealed that bEVs resulted in significant alterations in microglial C1q and CR3 expression, indicating an active involvement of bEVs in the C1q‐C3‐mediated synaptic pruning process. A recent study reported that peripheral LPS could activate complement C3 in both peripheral and central systems, leading to aberrant synaptic pruning in microglia and contributing to depression. ³⁹ While peripheral LPS is unlikely to cross the BBB directly to activate C3, it may still contribute to CNS C3 activation through multiple mechanisms, particularly when peripheral LPS levels are high, as seen in the ligation model used in this study. More specifically, in addition to being transferred by host EVs via a capture mechanism, as discussed above, elevated peripheral LPS levels may also exacerbate BBB disruption, facilitating the entry of LPS contained in bEVs and triggering C3/CR3‐mediated synaptic pruning in microglia.

The specific targets modulated by bEVs carrying LPS in microglia remain to be characterized further. The mechanosensitive ion channel Piezo1, readily activated by bacterial infections or LPS stimulation, exhibits high expression in microglia and astrocytes. ⁵⁷ , ⁵⁸ In parallel with this observation, we identified Piezo1 as a significant regulator in the bEV‐induced synaptic pruning. It is important to note that, compared to astrocytes, the activation of Piezo1 was significantly greater in microglia, which may suggest a selective activation of Piezo1 in microglia to mediate synaptic pruning induced by bEVs. Interestingly, a recent investigation reported that Piezo1, robustly upregulated by Aβ plaques in the post mortem brain tissues of AD patients (Braak stages III and IV), ⁵⁹ may instead help rescue the brain from Aβ deposition through interplay with the engulfment role of disease‐associated microglia. ⁵⁹ , ⁶⁰ The varied outcome can be interpreted in several ways. First, previous studies reported that abnormal synaptic function led to the deposition of Aβ, ⁶¹ suggesting that synaptic construction might actually precede the accumulation of Aβ. Second, as a double‐edged sword, Piezo1 may have a greater effect on the loss of neuron function, compared with its influence on Aβ eliminating. To this end, it should be noted that several clinical trials have shown that simply depleting Aβ deposits, especially at late stages of the disease, does not improve clinical outcomes of patients. ⁶²

In summary, this study unveiled a few key elements for gut‐derived LPS‐carrying bEVs to mediate synaptic pruning, a pivotal early process related to neurodegeneration such as in AD. While the mechanism underlying the passage of bEVs carrying LPS across the BBB requires further investigation, the observed mediation of microglial activation via Piezo1 by these bEVs suggests a promising avenue for therapeutic intervention in neurodegenerative disorders.

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no conflicts of interest. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

The study related to human autopsy specimens was conducted in accordance with the Declaration of Helsinki, 2013, and approved by the Institutional Review Board of First Affiliated Hospital of Zhejiang University School of Medicine, Hangzhou, Zhejiang, China ([2025B]IIT Ethics Approval No. 0591).

Supporting information

Supporting information

Supporting information

Zhao X, Yu J, Xu B, et al. Gut‐derived bacterial vesicles carrying lipopolysaccharide promote microglia‐mediated synaptic pruning. Alzheimer's Dement. 2025;21:e70331. 10.1002/alz.70331