Participants and sample collection

Participating centers obtained ethics approval before enrollment in the study, and all participants provided written informed consent before blood collection. Samples were collected from a total of 523 participants in the multicenter study. The discovery cohort consisted of 132 patients with AD, 93 patients with NAD, and 85 HC, all meeting the clinical inclusion criteria and enrolled in the First Affiliated Hospital, Zhejiang University School of Medicine, and Peking Union Medical College Hospital. All participants underwent comprehensive clinical evaluation with the criteria described previously [40, 41]. The selected AD patients had CSF molecular signature and/or PET patterns consistent with AD based on previous cutoffs [17, 20, 50]; subjects with NAD exhibited clinical symptoms of dementia, but did not show molecular/PET evidence of AD; HC had normal clinical evaluations and majority did not have CSF/PET data (Table 1). Sample collection and plasma separation were also performed as previously described [43]. Regarding the classification of different dementias with CSF, it is important to note that the binding capacity of tau antibodies may vary depending on the nature of tau species, as well as the presence or absence of mutations, which occur in certain NAD cases. However, it is widely acknowledged that the absence of characteristic changes in CSF biomarkers effectively excludes the diagnosis of AD in each individual. In fact, a few years ago, the National Institute on Aging and Alzheimer’s Association (NIA-AA) introduced an “ATN classification system” that serves as the “gold standard” for evaluating dementia patients’ biomarker-based and biological definitions [22].

Table 1 Characteristics of the discovery and validation cohorts

The validation cohort consisted of 99 patients with AD, 44 patients with NAD and 70 HC, and was obtained from the Second Affiliated Hospital, Zhejiang University School of Medicine and Daping Hospital, Third Military Medical University, using the same inclusion and exclusion criteria. For assay development, reference plasma samples (n = 3, each sample was pooled from 10 healthy controls) were obtained from the First Affiliated Hospital, Zhejiang University School of Medicine as previously described [41,42,43].

Human brain tissue samples (3 subjects, each comprising 4 brain regions) were obtained from the National Human Brain Bank for Health and Diseases, Zhejiang University. These tissue samples were from males aged over 60 years with no apparent pathological alterations.

Mouse cortical or hippocampal neurons culture and EVs enrichment

Primary neuron cultures were prepared from mouse (postnatal day 1) cortex or hippocampus. Briefly, after the mice were decapitated, the hippocampus and cortex were isolated, and were quickly and separately put in ice-cold dissection media (neurobasal medium containing 2% B27, 0.5 mM L-glutamine). After careful removal of all meninges, the tissues were roughly dissociated, and digested with 0.25% trypsin for 10–20 min at 37 °C. Fetal bovine serum was added to terminate the digestion reaction. After trituration, the cells were harvested by centrifugation 1000×g for 5 min at 4 °C and resuspended in culture media (DMEM containing 10% fetal bovine serum, 1% penicillin/streptomycin). Then, cortical and hippocampal neurons were plated onto poly-D-lysine-coated T25 flasks respectively. After 1 day, culture media was changed to neurobasal media (neurobasal medium containing 2% B27, 2 mM L-glutamine, 1% penicillin/streptomycin, and 5 µM cytosine β-D-arabinofuranoside; Cytosine β-D-arabinofuranoside exhibits the capacity to impede glial cell proliferation while concurrently augmenting neuronal population to a remarkable extent surpassing 95% [58]). The change of half media was performed on day 3, 6, 9 and 14. The B27 and fetal bovine serum used above were depleted of EVs by ultracentrifuging at 100,000×g for 18 h at 4 °C (Beckman Coulter Optima XPN-100 centrifuge, SW41 Ti rotor) [51]. The media (~ 50 mL), collected on day 21, was centrifuged at 2000×g for 20 min at 4 °C. The supernatant was ultracentrifuged at 100,000×g for 2 h at 4 °C (Beckman Coulter Optima XPN-100 centrifuge, SW41 Ti rotor), and the resulting pellet was resuspended with phosphate buffered saline (PBS) (pH 7.4, 0.22 μm-filtered), and ultracentrifuged again at 100,000×g for 2 h at 4 °C (Beckman Coulter Optima XPN-100 centrifuge, SW41 Ti rotor). The obtained pellets (fractions enriched with EVs) were resuspended with 100 μL PBS and stored at − 80 °C before use.

Mass spectrometry

Mass spectrometry-based detection was employed to explore and identify biomarkers associated with region-specific EVs. Cortical/hippocampal derived EVs were lysed in buffer containing 8 M urea (pH 8.5) and protease inhibitor. Samples were sonicated before centrifugation at 12,000×g for 10 min at 4 °C. Protein concentrations of the supernatant were determined using a BCA kit. The protein (100 μg) was subsequently reduced with 5 mM DTT for 30 min at 65 °C and alkylated for 15 min with 11 mM iodoacetamide. Then the solution was replaced with 0.1 M NH4HCO3 using a 10 kD ultrafilter. Trypsin was added at a 1:50 mass ratio of trypsin to total protein and incubated overnight at 37 °C (> 12 h). Following digestion, peptides were acidified with TFA and dried with a vacuum concentrator evaporator. Then, peptides were dissolved with 0.1% TFA, desalted over a C18-stage tip, and dried again. Samples were then resuspended in 15 μL of bRP (basic reverse phase) buffer A (10 mM NH4HCO3, pH 10, 5% ACN); and 1ug peptides were analyzed on timsTOF Pro 2 (Bruker, Germany).

The MS data were processed using Proteome Discoverer software v2.5 (Thermo Fisher), and were searched against SwissProt Mouse database (17,201 sequences) using the SEQUEST algorithm. Trypsin(full) was specified as cleavage enzyme allowing up to 2 missing cleavages. The minimum peptide length was 6 amino acids with a maximum of 5 modifications per peptide. The mass tolerance for precursor ions was set as 10 ppm, and the mass tolerance for fragment ions was set as 0.02 Da. Carbamidomethyl on Cys was specified as fixed modification. The oxidation of Met (M), the acetyl, met-loss, and met-loss + acetyl of protein N-terminal were set as dynamic modifications. Proteins and peptide-spectrum matches (PSMs) were filtered with a maximum false discovery rate (FDR) of 1%.

EV enrichment from plasma sample

EVs were enriched from plasma utilizing an ultracentrifugation methodology. Frozen plasma samples were thawed quickly at 37 °C and centrifuged at 2000×g for 15 min at 4 °C to obtain platelet-free plasma, followed by 12,000×g for another 30 min at 4 °C to remove large cell debris. The supernatant (100 μL) was further diluted with PBS (pH 7.4, 0.22 μm-filtered) at a ratio of 1:10 and then ultracentrifuged at 100,000×g for 1 h at 4 °C (Beckman Coulter Optima MAX-XP centrifuge, TLA-55 rotor). The pellet was resuspended with PBS, and ultracentrifuged again at 100,000×g for 1 h at 4 °C. The obtained pellets (EV-enriched fractions) were resuspended with 100 μL PBS and stored at − 80 °C before use. EVs-depleted plasma supernatant was used to test the specificity of the NanoFCM assay.

Immunohistochemistry

IHC examination was performed on 4 μm paraffin-embedded sections (cortex, hippocampus, caudate, and cerebellum) and stained with a Ventana BenchMark staining device (Ventana Medical Systems Inc). Slides were dried at 60 °C for 1 h and then deparaffinized with EZ Prep (Ventana, 950-102) for 15 min. Endogenous peroxidases were blocked for 10 min with a 3.0% hydrogen peroxide solution from the OptiView DAB IHC Detection Kit (Ventana, 760-700). Then, slides were heated to 100 °C for 36 min in ULTRA Cell Conditioning Solution 1 (Ventana, 950-224). Primary anti-GABRD rabbit polyclonal antibody (Invitrogen, PA5-26307, 1:100 dilution); anti-GPR162 rabbit polyclonal antibody (Proteintech, 15254-1-AP, 1:200 dilution), anti-NeuN mouse monoclonal antibody (Sigma-Aldrich, MAB377, 1:200 dilution) and anti-Histone H3 rabbit polyclonal antibody (Abcam, ab5103, 1:300 dilution) were diluted in Tris buffered antibody diluent (pH 7.2, 15 mM NaN3 and stabilizing protein, Dako, Santa Clara, CA, USA) followed by overnight incubation at 4 °C. Visualization was performed using the OptiView DAB IHC Detection Kit (Ventana, 760-700) followed by nuclear counterstaining by hematoxylin II (Ventana, 790-2208), bluing staining (Ventana, 760-2037), dehydration, transparency, and mounting.

Electron cryo-microscopy

EV-enriched fractions from reference plasma (EVs fractions; 5 μL) were deposited on electron microscopy (EM) grids coated with perforated carbon film for 5 min; and plunge-frozen in liquid ethane using a Vitrobot (Thermo Fisher). Images were acquired on a Talos F200C (Thermo Fisher) operating at 200 kV.

Nanoparticle Tracking Analysis

We utilized the conventional Nanoparticle Tracking Analysis (NTA) method to characterize the distribution and quantify the abundance of EVs in plasma. NTA was performed using a NanoSight NS300 with a 405 nm violet laser (Malvern, UK) according to the manufacturer’s instructions. Briefly, ultracentrifuged reference plasma EVs were diluted with PBS (pH 7.4, 0.22 μm-filtered) to 1 × 108–1 × 109 EVs mL−1 with a final volume of 1 mL for direct scattering measurement. For each measurement, videos of 3 random views were captured with the following settings: temperature controller, on; temperature, 25 °C; camera level, 14; Syringe speed, 40 μL s−1; capture duration, 60 s. The videos were analyzed using NanoSight NTA 3.4 at automatic mode with the detection threshold of 5 to assess mean and modal particle diameters, D50 values and particle number concentration.

Western blot

EVs were lysed in RIPA buffer. After centrifugation at 14,000×g for 10 min, the supernatants were collected, mixed with SDS sample buffer, and boiled for 5 min. The protein samples were subjected to SDS–Polyacrylamide Gel Electrophoresis (SDS-PAGE) and transferred to polyvinylidene difluoride membranes. The criterion employed for loading the gels with equivalent quantities relied upon the overall protein concentration. After blocking with 5% nonfat milk, the membranes were incubated with primary antibodies overnight at 4 °C, followed by incubation with IRDye 800CW secondary antibody (LI-COR) [diluted in Tris Buffered Saline with Tween 20 (TBST)]. The blots were visualized by Odyssey CLx Imaging System (LI-COR). The primary antibodies used in the present study included anti-GABRD antibody (Proteintech, 15623-1-AP), anti-GPR162 antibody (Proteintech, 15254-1-AP), anti-pTau217 antibody (Invitrogen, 44-744), anti-TSG101 antibody (Proteintech, 67381-1-Ig), anti-CD9 antibody (Proteintech, 60232-1-Ig) and anti-albumin antibody (Proteintech, 16475-1-AP). Primary antibodies were diluted in universal antibody diluent (NCM biotech, WB100D).

Stochastic optical reconstruction microscopy

Zenon immunoglobulin G (IgG) labeling kits (Invitrogen) were used to prepare fluorophore-conjugated antibodies according to the manufacturer’s instructions. For stochastic optical reconstruction microscopy (STORM) experiments, anti-GABRD antibody (Proteintech, 15623-1-AP) or anti-GPR162 antibody (Proteintech, 15254-1-AP) was labeled with Zenon™ Alexa Fluor™ 488 rabbit IgG Labeling Kit (Invitrogen, Z25302), anti-pTau217 antibody (Invitrogen, 44-744) was labeled with a Zenon™ Alexa Fluor™ 647 rabbit IgG Labeling Kit (Invitrogen, Z25308). Ultracentrifuged plasma (EVs fractions; 10 μL) was thawed and blocked with an equal volume of 2% BSA for 1 h at room temperature before diluted with 10 μL PBS (pH 7.4, 0.22 μm-filtered). Labeled anti-GABRD antibody or anti-GPR162 antibody (0.06 μg), anti-pTau217 antibody (0.1 μg), together with PE anti-human CD9 antibody (0.06 μg) (BioLegend, 312106) were added to the blocked EVs sample and incubated overnight at 4 °C. The labeled sample was then fixed with 20 μL 4% PFA (0.22 μm-filtered) for 20 min at room temperature. Labeled EVs were washed three times with PBS (pH 7.4, 0.22 μm-filtered) and 200 μL of specialized STORM imaging buffer [7 μL of oxygen-scavenging GLOX buffer (14 mg of glucose oxidase, 50 μL of 17 mg mL−1 catalase in 200 μL of 10 mM Tris, 50 mM NaCl, pH 8.0), 70 μL of MEA buffer (1 M), plus 620 μL of Buffer B (50 mM Tris–HCl (pH 8.0), 10 mM NaCl, 10% Glucose)] was added before image acquisition with a STORM. All images were acquired on a Nikon N-STORM super-resolution system (Nikon Instruments Inc.) with a Nikon Eclipse Ti inverted microscope with a 100 × TIRF lens (numerical aperture 1.49). 2000 frames with a 60 ms exposure time were recorded to image one field by an electron multiplying CCD camera (Andorixon DU-897). During the fluorescence acquisition, Nikon microscopic imaging device provided a Perfect Focus System (PFS) to achieve real-time correction of focus drift in Z-axis direction.

EV analysis with NanoFCM Flow NanoAnalyzer

A NanoFCM Flow NanoAnalyzer (NanoFCM Inc., XiaMen, China), which readily detects 30–1000 nm nanoparticles, was used to analyze particle concentration, size distribution and protein marker phenotyping according to the manufacturer's instructions and reported protocols [52]. Two single photon counting avalanche photodiodes (APDs) were used for the simultaneous detection of side scatter (SSC) (FF01-488/6 bandpass filter for a 488 nm laser or a FF01-524/24 bandpass filter for a 532 nm laser) and fluorescence (FF01-525/45 bandpass filter for green fluorescence, FF01-579/34 bandpass filter for orange fluorescence, or FF01-630/69 bandpass filter for red fluorescence) of individual particles. The 250 nm PE and AF488 fluorophore-conjugated polystyrene beads of known concentration were used to calibrate the sample particle concentration. The Silica Nanosphere Cocktail (NanoFCM Inc., S16M-Exo) that contained a mixture of 68 nm, 91 nm, 113 nm and 155 nm beads were used as the particle size standards to test the size distribution of EVs. Particles passed by the detector during a 1 min interval were recorded in each test. Samples were diluted to attain a particle count within the optimal range of 4000–8000 min−1. Using the calibration curve, the flow rate and side scattering intensity were converted into corresponding vesicle concentration and size on the NanoFCM software (NanoFCM Profession V1.0).

Fluorophore-conjugated antibodies were generated as mentioned above. In short, anti-GABRD antibody (Proteintech, 15623-1-AP), or anti-GPR162 antibody (Proteintech, 15254-1-AP), or anti-NLGN3 antibody (Abcam, ab192880) was labeled with Zenon™ Alexa Fluor™ 488 rabbit IgG Labeling Kit (Invitrogen, Z25302); anti-pTau217 antibody (Invitrogen, 44-744) was labeled with a Zenon™ Alexa Fluor™ 647 rabbit IgG Labeling Kit (Z25308, Invitrogen).

For optimization of Nanoscale flowcytometry assays experiment, immunoglobulin isotype controls of corresponding species were also labeled at the same final concentrations as all the antibodies. Another negative control (no antibody “Blank”, i.e., dye only) was done with the same volume of PBS instead of specific antibodies during the labeling reaction. EVs, or EVs-depleted plasma, or PBS (10 μL) were blocked with an equal volume of 2% BSA for 1 h at room temperature before diluted with 10 μL PBS (pH 7.4, 0.22 μm-filtered). Then, the blocked EVs, or EVs-depleted plasma, or PBS were incubated with fluorophore-conjugated antibodies [anti-GABRD antibody (0.06 μg), or anti-GPR162 antibody (0.06 μg), or anti-NLGN3 antibody (0.06 μg), or anti-pTau217 antibody (0.1 μg)], or corresponding IgG isotype control, or dye, overnight at 4 °C. The labeled sample was then fixed with 20 μL 4% PFA (0.22 μm filtered) for 20 min at room temperature, followed by analyzing with NanoFCM. The samples were diluted linearly (final volumes of 50 μL, 100 μL and 150 μL, respectively) to evaluate the accuracy of the assays. And the day-to-day stability was evaluated with a single reference EV sample that was analyzed in duplicate and repeated over five days.

For cohort study, the blocked EVs were incubated with fluorophore-conjugated antibodies [anti-GABRD antibody (0.06 μg), or anti-GPR162 antibody (0.06 μg), together with anti-pTau217 antibody (0.1 μg)] overnight at 4 °C. The labeled sample was then fixed with 20 μL 4% PFA (0.22 μm-filtered) for 20 min at room temperature, followed by analyzing with NanoFCM. All samples were kept at 4 °C and tested within 8 h after labeling, and labeling was stable under these conditions. A reference plasma sample was added into each day’s measurements to help to assess day-to-day variations (≤ 10%).

Ethics oversight

The study was performed in accordance with the ethical standards as laid down in the 1964 Declaration of Helsinki and its later amendments for experiments involving humans. Meanwhile, all human sample studies, including those for the Cryo-EM, NTA, WB, STORM and NanoFCM study protocols, were approved by the clinical research ethics committees of the First Affiliated Hospital, College of Medicine, Zhejiang University (approval number: 2021-400). The IHC protocol for human brain investigations was approved by the clinical research ethics committees of the First Affiliated Hospital, College of Medicine, Zhejiang University (approval number: 2022-043). The experiment involving mice was approved by the animal experimental ethical inspection committees of the First Affiliated Hospital, College of Medicine, Zhejiang University (approval number: 2021-667) and performed in accordance with Chinese Laboratory Animal Guideline for ethical review of animal welfare (2018/09) for the care of laboratory animals.

Statistical analysis

All analyses were performed with SPSS 25.0 (IBM, Chicago, IL, USA) or Prism 8.0 (GraphPad Software, La Jolla, CA, USA). The Mann–Whitney U test (for two groups) or the one-tailed nonparametric ANOVA, Kruskal–Wallis test (for three groups) were used to compare the mean total number of particles detected by scatter, or ratio of a given positive marker to total events. Receiver Operating Characteristic (ROC) curves were generated to evaluate their sensitivity and specificity in distinguishing AD from HC or NAD. Logistic regression was used to create an integrative model that included multiple plasma biomarkers. The bootstrap method was used to estimate a 95% confidence interval through 1000 sampling iterations. Delong’s test was used to confirm whether the integrated model has a significantly different AUC from single-factor diagnostic model. p < 0.05 was regarded as significant.