For experiments that characterized CSF EVs by immunoblot, tunable resistive pulse sensing (TRPS), transmission electron microscopy (TEM), and single vesicle flow cytometry (vFC) (Figure 1), we pooled human CSF from neurologically normal male and female participants obtained from the Oregon Health & Science University (OHSU) Oregon Alzheimer’s Disease Research Center (OADC), core program of the OHSU Layton Aging & Alzheimer’s Disease Center or purchased from BioChemed (Winchester, VA). For CSF EV miRNA quantification by qPCR, we used individual human CSF samples from 28 CTL and 28 AD participants (Figure 1; Table 1) obtained from the OADC or from the University of California San Diego Shiley-Marcos Alzheimer’s Disease Research Center (UCSD ADRC). All procedures were approved by the Institutional Review Boards of OHSU (IRB 6845) and UCSD (IRB 80012). All participants at both the institutions provided written informed consent and underwent detailed evaluations consisting of medical history, physical and neurological examinations, laboratory tests, and neuropsychological assessments, including cognitive tests and interview with a collateral historian. CTL participants were volunteers in good health with no symptoms of cognitive impairment or neurological disease and had normal performance on a detailed battery of neuropsychological tests. Sex, age, MMSE, and APOE genotype (e3,3 or e3,4) status for all 56 participants in this study are shown in Table 1.

The OADC and USCD ADRC both used a standardized CSF collection protocol corresponding to that of other AD centers engaged in biomarker research (Shi et al., 2011). Lumbar punctures were done in the morning under fasting conditions in the lateral decubitus position with a 24-gauge Sprotte spinal needle. The OADC sends the first 3–5 ml of CSF collected to the clinical lab for cell counts and determination of glucose and total protein levels, and the UCSD ARDC sends the first 2 ml of CSF collected to the clinical lab for testing. At both sites, serial syringes with 5 ml of CSF were collected, mixed, and transferred to polypropylene tubes in 0.5-ml aliquots, and the tubes were numbered to account for any gradient effect in subsequent experiments. The tubes were labeled with a subject number, but no other identifying information was mentioned. The CSF aliquots were flash frozen on dry ice and stored at −80°C. All CSF samples were only thawed once on ice prior to use.

APOE-e4 genotypes at OHSU were determined by sequencing the amplicon of APOE exon 4 (e4 allele). Genomic DNA was isolated from blood and amplified by Touchdown PCR with 250 µM dNTPs, 1 unit Taq DNA polymerase, buffer, 1X Q-solution (Qiagen, Valencia, CA), and 0.5 µM forward (5′-GGC​GCT​GAT​GGA​CGA​GAC​C-3′) and reverse (5′-CCT​GGG​CCC​GCT​CCT​GTA​G-3′) primers to amplify the APOE exon 4. A product size of 443 nucleotides identified on a 1% agarose 1X TBE gel was excised, cleaned with ExoSAP-IT reagent (Affymetrix), and sequenced on a model 377 automated fluorescence sequencer (Applied Biosystems). Chromatogram traces were examined, and nucleotide sequences were determined using FinchTV (Geospiza, Inc.). APOE-e4 genotypes at UCSD were determined by PCR amplification and restriction fragment length polymorphism (RFLP) (Hixson and Vernier, 1990), as previously described (Wierenga et al., 2012). Venous blood was drawn from participants and genomic DNA was extracted using the QIAamp DNA Blood Mini Kit (Qiagen) followed by PCR amplification. The APOE-e4 gene sequences were amplified using forward (5′-ACG​CGG​GCA​CGG​CTG​TCC​AAG​GA-3′) and reverse (5′-GCG​GGC​CCC​GGC​CTG​GTA​CAC-3′) primers. The amplification products were digested with HhaI (restriction enzyme site GCG^C) and then subjected to electrophoresis on polyacrylamide gels; the gels were stained with ethidium bromide, and the digested fragments were visualized by ultraviolet illumination. A unique combination of HhaI fragment sizes enabled typing of all homozygotic and heterozygotic combinations: HhaI cleaves at GCGC encoding 112arg (e4) and 158arg (e3, e4) but does not cut at GTGC-encoding 112cys (e2, e3) and 158cys (e2).

For single vFC studies, we examined total CSF that was not concentrated as well as 1.0 ml of CSF that was concentrated by UF using Microcon®-100 kDa Centrifugal Filters (MPE100025, Millipore Sigma, Burlington, MA) to a final volume of 40 μL (25x). We analyzed the CSF by single vFC using fluorescence to estimate vesicle size, concentration, and surface cargo with a commercial flow cytometer (Beckman Coulter CytoFlex S, Brea, CA) and an assay kit (vFC™ EV Analysis kit, Cellarcus Biosciences, San Diego, CA). In brief, 5 µL each of the unconcentrated and 25x concentrated CSF were stained according to the manufactures recommendations with a membrane stain (vFRed) and pool of fluorescence-labeled antibodies against tetraspanins (TS PE mix: CD9, CD63, and CD81) for 1 h at room temperature. The stained samples were diluted 1:1,000 prior to reading on the flow cytometer and detected using fluorescence triggering (excitation: 488 nm; emission: 690/50 nm). To demonstrate the vesicular nature of the vFRed and/or TS-positive (TS+) events, an aliquot of the stained, concentrated 25x CSF was detergent treated (0.05% Triton X-100) for 1 min prior to being assayed. Data analysis was performed using FCS Express (Version 7, De Novo Software, Pasadena, CA). Events were gated with respect to time (to eliminate spurious background that occurs at the start of each sample), vFRed pulse shape (to eliminate short pulse width background events), and violet side scatter (VSSC) vs. vFRed fluorescence (to include events with characteristic membrane fluorescence and light scatter) (Supplementary Figure S4). NanoRainbow beads (Cellarcus Biosciences) were used to characterize critical performance metrics that enable the evaluation of laser alignment and fluorescence resolution (Supplementary Figure S1). Membrane fluorescence was calibrated in terms of vesicle size (surface area) using a synthetic vesicle size standard (Lipo100 beads, Cellarcus Biosciences), and diameter was calculated assuming a spherical shape, resulting in an acceptable size distribution, in accordance to Cellarcus Biosciences (Supplementary Figure S3A). Standardized preparations of platelet EVs were used as reference samples for surface marker immunofluorescence (Supplementary Figure S3B). Particle concentrations were determined by subtracting the background values of buffer control containing vFRed and TS PE Mix (without CSF) from the experimental sample values.

EVs were isolated from CSF by a combined approach of UF and SEC for downstream use in immunoblots, TRPS, TEM, and miRNA qPCR arrays according to Figure 1. Depending on the downstream application, aliquots of 0.5, 2.0, or 5.0 ml of CSF were thawed on ice, pooled, and then concentrated by UF using a 0.5-ml Microcon®-30 kDa Centrifugal Filters (MRCF0R030, Millipore Sigma) and centrifugation at 14,000 x g for 15 min at 4°C. Concentrated samples were recovered by inverting the centrifugal filter into a clean collection tube and then centrifuged at 2,000 x g for 3 min at 4°C. The concentrated CSF was then brought to a final volume of 150 µL with 0.22-µm-filtered PBS. CSF EVs were isolated from the concentrated CSF by SEC using qEV single 35 nm or 70 nm columns (IZON Science, Christchurch, New Zealand), according to the manufacturer’s recommendations. Prior to use the qEV columns were brought to room temperature and equilibrated with 0.22-µm-filtered PBS. The void volume of the column consisting of fractions (Fxs) 1–5 (1 ml) and pools of subsequent Fxs 6–9, 10–13, and 14–17 (800 μL/pool) were collected. For immunoblot, TRPS, and TEM, each pool was then UF using a 0.5-ml Microcon®-30 kDa Centrifugal Filter to a final volume of 105 μL for immunoblot, 50 μL for TRPS, and 30 μL for TEM, as described before. For qPCR, the 800 μL pools were directly processed for RNA isolation without a second UF. For immunoblot, TRPS, and qPCR, the pools were frozen at −80°C until use. For TEM, the pools were stored on wet ice or 4°C until processing within 6 h. All centrifugation steps in this subsection of the methods were conducted using the Microfuge 22R centrifuge equipped with an F241.5P fixed angle rotor (Beckman Coulter).

Protein concentrations were measured using the Qubit protein assay kit and Qubit 4 fluorometer (Thermo Fisher Scientific, Waltham, MA) for the concentrated UF-SEC pools using 5 μL of the void and Fxs 6–9, 1 μL of Fxs 10–13, and 1 μL of 1:10 diluted Fxs 14–17. The remaining volume of each pool was prepared for SDS-PAGE by diluting in 4x NuPAGE™ LDS sample buffer (Thermo Fisher Scientific). To assess the separation of EV associated proteins from vesicle-free proteins and lipoproteins, we performed a total protein stain and immunoblots for APOA1, APOE, and albumin by loading equal volumes (37 μL) of the void, Fxs 6–9, 10–13, and 14–17 on a NuPAGE™ 4–12% Bis-Tris 1.5 mm x 10 well gels (Thermo Fisher Scientific). To identify the SEC Fxs that contained an enrichment of EV markers, equal concentrations (0.1 μg) of void, Fxs 6–9, 10–13, 14–17, and neurologically normal human frontal cortex lysate (positive control) were run on 4–12% gels, as described above, and immunoblotted for CD9, CD63, CD81, flotillin, TSG101, annexin V (AnnV), synaptophysin (SYP), NCAM-1, GLAST, CD11b, and TMEM119 proteins. In addition, gels loaded with 1.0 µg of void, Fxs 6–9, 10–13, 14–17, and human frontal cortex were immunoblotted for SYP and CD11b. Following electrophoresis, proteins were transferred to PVDF membranes and kept overnight at 4°C (30 V constant current). Membranes were incubated for 1 h with a blocking buffer (5% non-fat milk in TBS with 0.05% Tween 20 (TBST)) and then incubated with primary antibody diluted with TBST overnight at 4°C, and for 1 h with secondary antibody (1:10,000 TBST) at room temperature. Membranes were developed using the West Dura, Pico, or Femto chemiluminescence kits (Thermo Fisher Scientific) and imaged using the ChemiDoc Touch imaging system (Bio-Rad, Hercules, CA). The following antibodies were used for immunoblotting: albumin 1:1,000 (#4929, Cell Signaling Technology, Danvers, MA), APOA1 (12C8) 1:200 (sc-080551, Santa Cruz Biotechnology, Dallas, TX), APOE 1:2000 (50A-G1A, Academy Bio-medical Company, Inc., Houston, TX), AnnV 1:5,000 (GTX103250, GeneTex, Irvine, CA), CD9 (C-4) 1:200 (sc-13118, Santa Cruz Biotechnology), CD11b 1:1,000 (ab133357, Abcam, Cambridge, United Kingdom), CD63 1:1,000 (ab134045, Abcam), CD81 (B-11) 1:100 (sc-166029, Santa Cruz Biotechnology), flotillin 1:10,000 (ab133497, Abcam), GLAST 1:500 (NB100-1869, Novus Biologicals, Littleton, CO), NCAM-1 1:125 (NBP2-38452, Novus Biologicals), SYP 1:1,000 (#36406, Cell Signaling Technology), TMEM119 (#66948-1-Ig, Proteintech, Rosemont, IL), and TSG101 1:1,000 (ab125011, Abcam). Secondary antibodies were purchased from Jackson ImmunoResearch (West Grove, PA) and included horseradish peroxidase (HRP)–conjugated donkey anti-mouse (#715-035-150), donkey anti-rabbit (#711-035-152), and donkey anti-goat (#705-035-003).

Pools of the void volume (Fxs 1–5) and Fxs 6–9 were analyzed for particle size and concentration by TRPS using a qNANO gold instrument fitted with a NP80 nanopore (IZON Science). Prior to TRPS, 800 µL SEC pools were subjected to UF using Microcon®-30 kDa centrifugal filters and then brought to a final volume of 50 µL in 2x PBS (Figure 1). TRPS measurements were recorded with the NP80 stretched to 47.50 mm and using 0.34 V at a pressure reading of 5 (mbar). Each sample was analyzed until at least 500 particles were recorded or until the maximum recording time of 10 min elapsed. Data were acquired and analyzed using the IZON Control Suite version 3.4 software (IZON Science). The size (nm) and concentration (particles/mL) of particles measured in the SEC pools were calibrated against CPC100 calibration beads (IZON Science).

Pools of CSF SEC fractions including the void, Fxs 6–9, 10–13, and 14–17 were concentrated by UF to a final volume of 30 μL and then stored on wet ice until processing for TEM within 6 h (Figure 1) by the OHSU Multiscale Microscopy Core. A volume of 5 µL of the SEC preparations were deposited on glow-discharged (120 s 15 mAmp, negative mode) carbon formvar 400 mesh copper grids (01822-F, Ted Pella, Inc., Redding, CA) for 3 min, rinsed 15 s in water, wicked on Whatman filter paper 1, stained for 60 s in filtered 1.33% (w/v) uranyl acetate in water, wicked again, and air dried. Samples were imaged by using the Multiscale Microscopy Core at 120 kV on a FEI Tecnai™ Spirit TEM system (Thermo Fisher Scientific Electron Microscopy, Hillsboro, OR). Images were acquired as 2048 × 2048 pixel, 16-bit gray scale files using the FEI’s TEM Imaging & Analysis interface on an Eagle™ 2K CCD multiscan camera.

EVs were isolated from 500 μL of CSF that was first concentrated by UF to 150 μL then fractionated by SEC (Figure 1). Fxs 6–9, 10–13, and 14–17 were pooled to final volume of 800 μL, and total RNA was isolated using the mirVana™ PARIS™ RNA and a Native Protein Purification Kit (AM1556, Thermo Fisher Scientific), with modification (Burgos et al., 2014), as we have previously reported for CSF and plasma EVs (Lusardi et al., 2017; Wiedrick et al., 2019; Sandau et al., 2020a; Sandau et al., 2020b; Lusardi et al., 2021). Isolated RNA was then stored at −80°C until use. For qPCR, the RNA samples were concentrated (R1013, RNA Clean and Concentrator™-5 Kit, Zymo Research, Irvine, CA) and then eluted into 6 μL of RNase/DNase-free water. Total RNA was assayed on Custom TaqMan™ Advanced Array Cards (192a Format) that include probes for 191 human CSF miRNAs, plus five endogenous non-changing miRNA controls used for normalization: miR-191-5p, -204-3p, -204-5p, -342-3p, and -574-3p (Supplementary Table S1). Using a pool of Custom RT Primers for the CSF array (Thermo Fisher Scientific) and MultiScribe™ Reverse Transcriptase (4, 311, 235, Thermo Fisher Scientific), 3.2 μL of concentrated RNA was reverse transcribed in a total reaction volume of 7.5 μL. A volume of 5 μL of cDNA was pre-amplified (PreAmp) for 14 cycles using Custom PreAmp Primers for the CSF array (Thermo Fisher Scientific) and TaqMan™ PreAmp Master Mix (4, 391, 128, Thermo Fisher Scientific) in a final reaction volume of 25 μL. Reverse transcription and PreAmp reactions were run with a Bio-Rad T100 thermal cycler (Thermo Fisher Scientific), following the manufacturer’s instructions, for detection of miRNAs with PreAmp. All cDNA was stored at −20°C. Prior to qPCR the PreAmp cDNA was diluted to 1:2 in RNase/DNase-free water and 18 μL was mixed with TaqMan™ Universal Master Mix II, no UNG (444,047, Thermo Fisher Scientific). The samples were loaded on the custom CSF arrays, and the qPCR amplifications and data acquisition were carried out on a QuantStudio™ 12K Flex real-time PCR system (Thermo Fisher Scientific).

MiRNAs were analyzed using relative quantification (∆∆Cq) based on Applied Biosystems recommendations (Thermo Fisher Scientific, Part Number 4371095 Rev B). Cq values were calculated using automatic baseline, and threshold values were determined by ExpressionSuite Software v.1.3 (Thermo Fisher Scientific). The Cq value for each well was reported along with the amplification score (AmpScore) and a Cq confidence (CqConf), which are metrics for the quality of each amplification, as we have previously reported (Lusardi et al., 2017; Wiedrick et al., 2019; Sandau et al., 2020a; Sandau et al., 2020b; Lusardi et al., 2021). Prior to data analysis, amplifications were filtered according to each well’s Cq value, AmpScore, and CqConf: 1) PCR products with a Cq > 34, or reported as “Undetected”, were considered below the detection threshold and assigned a Cq value of “34”; 2) amplifications with a Cq ≤ 34 and an AmpScore <1.0 or a CqConf <0.8 were excluded from analysis. Based on these criteria, miRNAs with a Cq ≤ 34, AmpScore >1.0, and CqConf >0.8 were deemed considered for further analysis. Next, we excluded miRNAs that were not expressed in at least 20% of the samples (11 out of 56), which would enable analysis of miRNAs differentially detected in one out of the eight subgroups (e.g., APOE-e3,4 females with AD). We also performed qPCR reactions with water only (no RNA—no template control) as a control for spurious PCR amplifications (Supplementary Table S2). MiRNAs that had good quality amplifications (Cq ≤ 34, AmpScore >1.0, and CqConf >0.8) in the water only control that were within 1 Cq of the average Cq for the 56 experimental samples were excluded from downstream analysis. For miRNAs that did meet all of the inclusion criteria, we then used the following formula for calculating the ∆∆Cq for each miRNA: ∆∆Cq = mean ∆Cq of the test samples (e.g., AD) − mean ∆Cq of reference samples (e.g., CTL). Within each sample, the ∆Cq for a miRNA was calculated by the following formula: ∆Cq = miRNA Cq − mean Cq of endogenous control miRNAs. MiRNAs selected as endogenous control normalizers showed 1) stable good quality expression values in all samples regardless of experimental group and 2) best endogenous control scores identified in ExpressionSuite. For each miRNA, the fold change (RQ value) was calculated by the 2^−ΔΔCq method, with RQ > 1 indicating increased miRNA expression in either AD vs. CTL or females vs. males. Conversely, RQ < 1 indicates decreased expression in AD or females, compared to their respective reference group. The miRNA microarray data is MIAME compliant and submitted to the Gene Expression Omnibus site: ncbi.nlm.nih.gov/geo/.

Data were analyzed with GraphPad Prism software v9.3.1 (GraphPad Software, Inc., San Diego, CA). Data are shown as either a Volcano plot (−Log10 (q-value) vs. the Log2(Fold change)) or the ΔCq mean ± standard error mean (SEM). Where possible the biological replicates are displayed as individual symbols. To assess the effects of AD on miRNA levels, participants were categorized as either AD (n = 28) or CTL (n = 28) and analyzed by Welch’s unpaired t-test with a Benjamini–Hochberg false discovery rate (B-H FDR) of 0.20, a cutoff that is in line with a hypothesis-generating study, and the preference for some false positive findings in exchange for a larger list of potentially interesting true findings (Efron, 2010). To assess the effects of sex on miRNA levels, data generated from the participants were re-analyzed for females (n = 28) vs. males (n = 28) and analyzed by Welch’s unpaired t-test with a B-H FDR of 0.20. To assess the effect of disease state and APOE status, data were analyzed within each sex using a two-way ANOVA (disease x APOE genotype) followed by a Tukey’s multiple comparisons post hoc tests (n = 7/group). Given the small number of participants in this study, a covariate analysis of disease by sex by APOE genotype was not conducted. Thus, conclusions drawn from the analysis assume no interaction between all three variables.

We used TargetScan 7.2 (Agarwal et al., 2015) and miRDB (Liu and Wang, 2019; Chen and Wang, 2020) to predict targets of the four CSF EV miRNAs significantly associated with AD, as these programs are widely used and frequently updated. As pathway analysis is most effective for predictions generated from a limited gene set, predicted targets were excluded if they had a Cumulative Weighted Context Score (CWCS) > −0.3 in TargetScan or a target score of <60 in miRDB. To determine whether to use a union, an intersection, or an individual target list in subsequent pathway analysis, we calculated sensitivity, specificity, and precision values using validated miRNA and target pairs from miRTarBase, an experimentally validated miRNA–target interaction database (Chou et al., 2018). The values in miRTarBase range from 0–1, with high-quality results closer to 1 (Fan and Kurgan, 2015; Oliveira et al., 2017). The union of TargetScan and miRDB showed the highest values of sensitivity, specificity, and precision with values of 0.35, 0.83, and 0.98, respectively. The use of a union between the two target prediction algorithm outputs increases the sensitivity of the targets predicted and the chance of predicting a novel target, as opposed to an intersection where only targets predicted by both algorithms are included in analysis (Fan and Kurgan, 2015; Oliveira et al., 2017). Pathway analysis was then performed on the union set of 2,319 unique targets using the ingenuity pathway analysis (IPA; QIAGEN Inc., qiagenbioinformatics.com/products/ingenuity-pathway-analysis). For this analysis, we excluded cancer-related tissues and cell lines to avoid knowledge bias toward cancer in IPA and applied a B-H FDR of 0.01.

The isolation and characterization of CSF EVs followed recommendations from the Minimal Information of Studies of Extracellular Vesicles 2018 (MISEV 2018), a position study from the International Society of Extracellular Vesicles (ISEV) (Thery et al., 2018). The vFC experiments followed recommendations from MIFlowCyt-EV, a framework for standardized reporting of EV flow cytometry experiments from ISEV (Welsh et al., 2020). Relevant experimental parameters can be accessed in the EV-TRACK (evtrack.org) knowledgebase (EV-Track et al., 2017).

CTL CSF was pooled and used for vFC, immunoblot, TRPS, and TEM experiments, while individual CSF samples from CTL and AD participants were used for the miRNA qPCR assays. Characteristics for the 28 AD and 28 CTL participants in this study (56 total participants) are shown in Table 1. The ratio of participants was 1:1 female to male overall, with equal numbers of females and males in the AD and CTL groups. Furthermore, the APOE-e3,3 to APOE-e3,4 ratio was 1:1 within each sex to obtain a total of eight groups (disease by sex by genotype) with n = 7/group. Participants were well matched for age at the time of lumbar puncture across all the groups with no significant differences. CTL participants were in good health with a mean MMSE (Folstein et al., 1983) score of 29.3 ± 1.3. AD patients were diagnosed with probable AD according to ADRDA-NINDS criteria (McKhann et al., 1984; McKhann et al., 2011), with a mean MMSE score of 21.5 ± 3.1. There was a significant effect of disease state on the MMSE score F_(1,48) = 139.8, p < 0.0001, but no significant interactions with disease state and sex and/or genotype.

To measure the size and concentration of EVs in total CSF, we performed single vFC, which uses sensitive measurements of light scatter from a fluorescent lipid probe (vFRed) and surface immunofluorescence from staining with a pool of phycoerythrin (PE)-conjugated TS antibodies (CD9, CD63, and CD81; TS PE mix). In our prior studies on plasma, we diluted the sample 150-fold prior to staining to enable detection of TS + membranous particles in the diameter range of ∼75–600 nm at mean concentration of 1.3e10/mL (Sandau et al., 2020a). However, when we performed vFC with undiluted total CSF that was stained with vFRed and the TS PE Mix, we detected TS + membranous particles just above background levels with mean size of 136 nm (range: ∼75–400 nm) and at a concentration of 1.7e6/mL (Figures 2A–C). To increase counts for the CSF assay, we subsequently used UF to concentrate 1 ml of CSF to 40 µL prior to staining with vFRed and TS PE Mix (Figure 1). The 25x concentrated CSF yielded more robust results with 1.5e7/mL TS + particles and mean particle size of 115 nm (range: ∼75–275 nm) (Figures 2D,E). Next, we detergent treated an aliquot of the stained, 25x concentrated CSF, which resulted in loss of >72% of gated events (Figures 2F,G), indicating that the majority of detected events were detergent-labile, as expected for EVs. Together, these data demonstrate that CSF contains a low concentration of TS + EVs compared to plasma.

To isolate EVs from human CSF, we first performed a series of SEC protocol refinement experiments. Using CSF pooled from multiple CTL samples, we tested 35 and 70 nm SEC columns for EV enrichment as indexed by the presence of established EV protein markers (Figure 3A) and depletion of vesicle-free lipoproteins and proteins (Figure 3B). To overcome the limitations of the low concentration of EVs in total CSF (Figure 2), we started with 5 ml of CSF that was concentrated by UF to 150 µL prior to fractionating the samples into SEC pools. We collected the void volume of the column (Fxs 1–5), Fxs 6–9, 10–13, and 14–17 (Figure 1). EVs are expected to elute in Fxs 6–9 based on the manufacturer’s recommendations and our prior studies using human plasma (Sandau et al., 2020a). The SEC pools were subsequently concentrated by UF to 105 µL then assayed by immunoblot. Gels loaded by equal concentration of protein (0.1 µg) showed a strong enrichment of flotillin and CD81 in Fxs 6–9 when using the 35 nm columns. In comparison, trace amounts of flotillin were detected in Fxs 6–9 of the 70 nm columns and CD81 was undetectable (Figure 3A). Equal volume loading of concentrated SEC pools demonstrated that both the 35 and 70 nm columns effectively depleted albumin and APOA1 (vesicle free-protein and lipoprotein, respectively) from Fxs 6–9 (Figure 3B). We next used CSF fractionated by SEC on 35 and 70 nm columns to obtain size and concentration measurements by TRPS (Figure 1). The void volume for the 35 and 70 nm columns both had <10 particles of the expected size that were detected during the entirety of the maximum recoding time of 10 min (data not shown). Fxs 6–9 from the 35 nm column had 1.45e10 particles/mL with a mean size of 138 nm (range: 76–620 nm), while the 70 nm column had 1.54e9 particles/mL with a mean size of 151 nm (range: 86–360 nm) (Figure 3C). The increased abundance of smaller-sized particles obtained in Fxs 6–9 of the 35 nm column is in agreement with the immunoblot results. Together, these results demonstrate that the 35 nm SEC column is optimal for CSF EV enrichment.

We further characterized CSF fractionated using the 35 nm SEC columns by TEM and immunoblots (Figure 1). Figure 4 shows representative wide and narrow field of view TEM images of the SEC pools. The images revealed that the void contained a small amount of diffuse material that ranges in size from 30–100 nm (Figures 4A,B). In line with our TRPS results, EVs with a lipid membrane of the expected size range for exosomes (∼40–150 nm) or small MVs (∼150–1,000) were detectable in Fxs 6–9. Furthermore, there was a greater proportion of smaller nanoparticles (<40 nm), which are the expected size range for exomeres (∼30–50 nm), supermeres (∼25–30 nm), or vesicle-free lipoproteins (∼5–35 nm) (Zhang et al., 2018; Zipkin, 2019; Zhang et al., 2021). Fxs 6–9 showed negligible levels of background protein, as expected for SEC isolates (Figures 4C,D). TEM images from Fxs 10–13 and 14–17 showed minimal numbers of EVs discernible by the presence of lipid membrane. However, there was a high degree of background, consistent with increasing amounts of vesicle-free lipoproteins and proteins that elute later (Figures 4E–H). Next, we used total protein stain and immunoblots to show a depletion of bulk proteins, albumin, and APOA1 when SEC pools were loaded by equal volume (Figure 5A). We also found that while a majority of APOE was present in Fxs 10–13 and 14–17, APOE was also detectable in the Fxs 6–9 (Figure 5A). Equal concentration loading of the SEC pools showed an enrichment of multiple EV protein markers that are associated with the endolysosomal pathway including CD9, CD63, CD81, flotillin, and TSG101 (Figure 5B) (Jeppesen et al., 2019). Furthermore, we identified an enrichment of AnnV (Figure 5B), which is associated with MVs and/or apoptotic bodies (Cocucci et al., 2009; Gouwens et al., 2018; Jeppesen et al., 2019). We next sought to determine if putative markers for brain-derived EVs were enriched in the CSF SEC pools. After loading 0.1 µg of CSF SEC Fx protein lysate, we detected an enrichment of GLAST, which is a marker for astrocyte-derived EVs (Figure 5B) (You et al., 2020). However, we were unable to detect NCAM-1 and SYP, which are markers for neuronal-derived EVs; as well as TMEM119, which is a microglia-derived EV marker (Figure 5B) (Togashi et al., 2009; Satoh et al., 2016; Sathe et al., 2019; Vukojevic et al., 2020; Cohn et al., 2021; Crews et al., 2021; Ruan et al., 2021; You et al., 2022). We also performed an immunoblot assay using 0.1 µg protein for CD11b, which is another marker for microglia-derived EVs (Gu et al., 2021), and detected faint bands in Fxs 6–9, 10–13, and 14–17 (Figure 5B). To further assess whether low concentrations of neuron- and/or microglial-derived EVs are present in CSF, we performed additional immunoblots with 1.0 µg of protein and probed for SYP and CD11b (Figure 5C). Despite a 10x increase in the amount of protein loaded, we were still unable to detect SYP in any of the CSF Fxs (Figure 5C). However, we did detect CD11b in Fxs 10–13 and 14–17, but just a faint CD11b band in Fxs 6–9 (Figure 5C). Together, these data support that using a combination of UF and SEC enriches for CSF EVs that contain markers for glial-derived EVs. However, we cannot definitively state that Fxs 6–9 are void of vesicle free lipoproteins and/or proteins based on the presence of APOE in these Fxs (Figure 5A).

Considering that extracellular miRNAs are associated with exosomes, MVs, apoptotic bodies, RNA binding proteins, and high-density lipoproteins (Lasser, 2019; Mori et al., 2019), we sought to assess the miRNA profile of CSF fractionated by SEC. We concentrated 500 µL of CSF (n = 4) to 150 µL by UF; performed SEC to collect pools of Fxs 6–9, 10–13, and 14–17; and assayed miRNAs previously shown to be expressed in CSF and miRNAs validated as candidate biomarkers for AD (Wang et al., 2017; Wiedrick et al., 2019). The experiment was performed with CTL CSF balanced for sex and APOE genotype (n = 4): female APOE-e3,3, female APOE-e3,4, male APOE-e3,3, and male APOE-e3,4 (Table 1). MiRNAs were excluded from the analysis if they were not expressed in at least two of the four participants. In total, Fxs 6–9, which is enriched for CSF EVs, expressed the greatest number of miRNAs (48 total), compared to Fxs 10–13 (27 total) and 14–17 (30 total) (Figure 6). Furthermore, 26 of the 48 miRNAs that were expressed in the CSF EVs (Fxs 6–9) were not detectable in the other SEC pools, while only a few miRNAs were exclusively expressed in Fxs 10–13 and 14–17. Of the 48 miRNAs expressed in Fxs 6–9, 12 were our candidate AD biomarkers (miR-125-5p, -140-5p, -142-3p, -145-5p, -146a-5p, -146b-5p, -19b-3p, -223-3p, -24-3p, -29a-3p, -328-3p, and -331-3p) (Wiedrick et al., 2019). Together, these data demonstrate that a majority of the miRNAs expressed in CSF, including AD biomarkers, are detectable in SEC fractions enriched for EVs.

Next, we assessed the impact of AD on the miRNA cargo of CSF EVs. MiRNAs were assayed in CSF Fxs 6–9 from eight biological groups that included AD vs. CTL, female vs. male, and APOE-e3,3 vs. APOE-e3,4 participants (Table 1, n = 56 total). We identified 71 miRNAs expressed in the AD (n = 28) and/or CTL (n = 28) participant samples (Figure 7A). Of these four miRNAs were (miR-16-5p, -331-3p, -409-3p, and -454-3p) significantly increased by at least a 1.5-fold in AD relative to CTL (Figures 7A–E and Supplementary Table S3), two of which we previously identified and validated as candidate AD biomarkers (miR-16-5p and -331-3p) (Wiedrick et al., 2019). There were also four miRNAs (let-7d-5p, miR-100-5p, -374a-3p, and -378e) that increased by at least a 1.5-fold change in expression with p < 0.05 but were not considered significant based on the FDR (Figure 7A, pink). Next, we used the four significant miRNAs (miR-16-5p, -331-3p, -409-3p, and -454-3p) in our established target prediction pipeline using TargetScan 7.2 (Agarwal et al., 2015) and miRDB (Liu and Wang, 2019; Chen and Wang, 2020) as depicted in Figure 7F. Target prediction returned a total of 272 mRNAs by TargetScan (Supplementary Table S4) and 2,245 mRNAs by miRDB (Supplementary Table S5). Of these, 198 targets were predicted by both TargetScan and miRDB. Thus, the total number of unique predicted mRNA targets is 2,319, which were used in IPA (Qiagen) to identify pathways that have a significant overrepresentation of the predicted targets and potentially regulated by the miRNAs. IPA identified 189 significant canonical pathways (Supplementary Table S6). Of these, 10 of the top 17 pathways are known to be relevant to AD, including the top three overall pathways: senescence, TGF-β signaling, and epithelial adherens junction signaling (Figure 7G) (Caraci et al., 2018; Pearson et al., 2020; Saez-Atienzar and Masliah, 2020).

Considering that female APOE-e4 carriers are more likely to progress from MCI to AD, develop AD more frequently than age-matched males, and have more brain atrophy and memory loss (Fleisher et al., 2008; Altmann et al., 2014; Sampedro et al., 2015), we next sought to determine if these biological factors impact CSF EV miRNA cargo. First, we assessed the effect of sex on CSF EV miRNAs, independent of AD and APOE genotype. We identified 71 miRNAs expressed in female (n = 28) and/or male (n = 28) participant samples (Figure 8A). Of these, three miRNAs (miR-146b-5p, -150-5p, and -342-3p) were significantly increased by at least 1.5-fold in females relative to males (Figures 8A–D and Supplementary Table S3), and we had previously identified miR-146b-5p as a candidate AD biomarker (Wiedrick et al., 2019). There were also seven miRNAs (miR-19b-3p, -188-5p, -223-3p, -320c, -320d, -483-5p, and -92a-3p) that were decreased by at least a 1.5-fold in females relative to males and p < 0.05 but were not considered statistically significant based on the FDR (Figure 8A, light blue).

We next sought to determine if there was an interaction between AD and APOE-e4 status within a sex. First, in females, we analyzed the 71 CSF EV miRNAs for an effect of disease state and/or genotype status and identified 16 significant miRNAs (Supplementary Table S7). Of these, multiple comparisons testing identified five miRNAs with a significant group effect (Figure 9). Let-7d-5p was increased in AD APOE-e3,3 females vs. CTL APOE-e3/e4 females (Figure 9A). miR-16-5p was increased in AD APOE-e3,4 females vs. CTL APOE-e3,3 and -e3,4 females (Figure 9B). miR-125b-5p and -320a were both increased in CTL APOE-e3,4. CTL APOE-e3,3 (Figures 9C,D). miR-331-3p was increased in AD APOE-e3,4 vs. CTL APOE-e3,4 (Figure 9E). Of these five miRNAs, we previously identified three (miR-16-5p, -125b-5p, and -331-3p) as candidate AD biomarkers (Wiedrick et al., 2019). Next, our analysis of the 71 CSF EV miRNA levels in males identified five miRNAs that had a significant effect of disease state and/or genotype status (Supplementary Table S7). Of these, multiple comparisons testing identified four miRNAs with a significant change in expression (Figure 10). miR-140-5p was increased in AD APOE-e3,4 males vs. AD and CTL APOE-e3,3 males (Figure 10A). miR-20a-5p was decreased in AD APOE-e3,4 males vs. CTL APOE-e3,3 males (Figure 10B). miR-30b-5p was increased in AD APOE-e3,4 males vs. AD APOE-e3,3 males (Figure 10C). miR-454-3p was increased in AD APOE-e3,4 males vs. all other groups (Figure 10D). Of these four miRNAs, we previously identified miR-140-5p as a candidate AD biomarker (Wiedrick et al., 2019). Together, these data demonstrate that biological factors such as sex and genotype can impact the levels of miRNA cargo of CSF EVs. Note that due to the low sample size we were unable to perform a statistical analysis to assess the effect of AD by sex by APOE genotype.