Neuron-derived extracellular vesicles in blood reveal effects of exercise in Alzheimer’s disease

We studied all remaining 162 plasma samples from the ADEX cohort, sourced from 47 participants in the control group (n = 38 at baseline, n = 43 at 16 weeks), and 48 participants in the exercise group (n = 44 at baseline, n = 37 at 16 weeks).

Plasma was processed in accordance with guidelines for pre-analytical factors for EV isolation and biomarker analysis [22, 23]. Blood was collected in EDTA polypropylene tubes and within 1 h centrifuged at 3000 rpm for 10 min at 4 °C. Supernatant plasma was divided into 250 µL aliquots and stored in a central biobank at − 80 °C. Plasma aliquots were received and processed blindly by investigators at the National Institute on Aging. NDEVs were isolated following an immunoaffinity capture methodology extensively characterized by us [17‐20] and others [24, 25] targeting L1 cell adhesion molecule (L1CAM), a transmembrane neuronal protein sorted to EVs. Briefly, 250 µL of plasma was defibrinated via incubation with 100 µL of Pacific Hemostasis Thromboplastin-DS (cat. no. 100354; Thermo Fisher Scientific) for 45 min at room temperature (RT), followed by dilution with 150 µL of Dulbecco’s phosphate-buffered saline (DPBS) -1X supplemented with 1X protease (cOmpleteTM Protease Inhibitor Cocktail; cat no. 04693116001; Roche) and phosphatase inhibitors (Halt™ Phosphatase Inhibitor Cocktail; cat no.78427; Thermo Fisher Scientific), and sedimentation at 3000 × g for 15 min at RT. The supernatant was transferred to a sterile 1.5 mL microtube and total EVs were sedimented via incubation with 126 µL of Exoquick™ (cat no. EXOQ100A-1; System Biosciences) for 60 min at RT, followed by centrifugation at 1500 × g for 30 min at RT. The crude EV pellet was resuspended in 350 µL of ultra-pure distilled water supplemented with protease/phosphatase inhibitors overnight with gentle rotation mixing at 4 °C. Crude EVs were incubated for 30 min at RT with 4 µg of biotinylated anti-human L1CAM antibody (clone 5G3) (cat. no. 13–1719-82; Thermo Fisher Scientific) or a three-antibody cocktail against canonical EV tetraspanins CD81 (cat. no. 302–030; Ancell), CD9 (cat. no. 558749; BD Pharmingen) and CD63 (cat. no. MAB15361; Abnova), to derive NDEVs or pan-tetraspanin expressing EVs (panTET-EVs), respectively. The EV-antibody complexes were then incubated with 25 µL of Pierce™ Streptavidin Plus UltraLink™ Resin (cat. no. 53117; Thermo Fisher Scientific) for 30 min at RT. After centrifugation at 600 × g for 10 min at 4 °C and removal of supernatant, NDEVs or panTET-EVs were eluted with 100 µL of 0.1 M glycine (stock solution at 1 M, pH = 2.7; cat. no. 24074–500; Polysciences, Inc.). Beads were sedimented by centrifugation at 4000 × g for 10 min at 4 °C, and supernatant containing immunoprecipitated EVs was transferred to a clean tube, where pH was immediately neutralized with 10 µL of 1 M tris hydrochloride (Tris–HCL, pH = 8; cat. no. CAS1185-53–1; Fisher Scientific). 10 µL of intact EVs were stored at − 80 °C for nanoparticle tracking analysis (NTA) and the remaining volume was subjected to EV lysis via two freeze–thaw cycles in 25 µL of 10% bovine serum albumin and 365 µL of 1X RIPA lysis buffer (stock solution at 10X; cat. no. 20–188; EMD Millipore Corp.) supplemented with 1X protease/phosphatase inhibitors. Lysed EVs were stored at − 80 °C.

We used enzyme-linked immunosorbent assays (ELISAs) to quantify the concentration of proBDNF (cat. no. BEK-2237; Biosensis) and humanin, also known as human putative humanin peptide or MT-RNR2 (cat. no. CSB-EL015084HU; Cusabio), in NDEV lysates; plates were read using the Synergy™ H1 microplate reader set to 450 nm and the Gen5™ microplate data collection software (BioTek Instruments). NDEV levels of BDNF, irisin, apelin, fractalkine, erythropoeitin (EPO), osteonectin (SPARC), interleukin-15 (IL-15), myostatin (MSTN)/GDF8, FABP3, follistatin-like 1 protein (FSTL-1), oncostatin-M (OSM) and osteocrin/musclin were quantified utilizing a Milliplex® Human Myokine Magnetic Bead panel (cat. no. HMYOMAG-56 K; EMD Millipore Corporation). Milliplex plates were read using the Luminex® 100/200™ instrument with the xPOTENT® acquisition software (Luminex Corporation). NDEV and panTET-EV samples were run undiluted based on experiments determining the optimal input volume for each assay. Standard curve equations of ELISAs were determined using four-parameter logistic (4-PL) regression, whereas for Milliplex assays, protein concentrations were extrapolated from a five-parameter logistic (5-PL) curve.

Samples were assessed in duplicate in all assays. The limit of detection (LOD) was defined as mean signal of the blank plus 2.5 times its standard deviation (SD). The lowest limit of quantification (LLOQ) was set by the following rules: (1) signal above LOD, (2) coefficient of variation (CV) > 20%, and (3) 80–120% recovery. A zero value was imputed to samples with signals below LOD; the LLOQ was imputed for signals between LOD and LLOQ; duplicate measurements with %CV ≥ 30 were excluded from the analysis. Excluded samples per analyte: 2 for proBDNF, 15 for humanin in NDEVs, 0 for humanin in panTET-EVs, 27 for apelin, 10 for fractalkine, 2 for BDNF, 2 for EPO, 8 for SPARC, 1 for IL-15, 3 for MSTN/GDF8, 13 for FABP3, 30 for irisin, 28 for FSTL-1, 21 for OSM and 21 for osteocrin/musclin.

An internal control (IC) was included in all plates to assess inter-plate variability. NDEVs from a healthy participant and a quality control provided were used as ICs for ELISAs and the Milliplex assay, respectively. CVs for the ICs across all plates were below 30% and hence, signals were not normalized (CVs for proBDNF, 9.9%; humanin, 23.1%; apelin, 5.3%; fractalkine, 11.8%; BDNF, 21.7%; EPO, 3.8%; SPARC, 8.4%; IL-15, 8.7%; MSTN/GDF8, 9.8%; FABP3, 12.0%; irisin, 3.6%; FSTL-1, 9.3%; OSM, 7.4%; and osteocrin/musclin, 17.3%).

Previous findings have shown that plasma contains a high concentration of soluble L1CAM that could significantly interfere with the enrichment of NDEVs via L1CAM immunoprecipitation [26]. Hence, we first sought to validate the EV composition of our NDEV preparations by NTA and quantify the enrichment of L1CAM⁺ EVs via flow cytometry analysis (FCA), following established guidelines [27]. NTA showed that NDEVs had sizes between 50 and 450 nm, which is typical for a mixed population of exosomes and smaller microvesicles (Fig. S1C). NTA results were consistent with FCA of total EVs and NDEVs, with a violet side scatter (vSSC) signal range mainly within that of nanobeads under 500 nm (Fig. S1D). The achieved enrichment was determined by quantifying L1CAM⁺ nanoparticles before and after L1CAM immunoprecipitation using FCA (Fig. S1D–H). Among EVs gated based on the detection of blue succinimidyl ester (BSE) staining all EVs (Fig. S1D), ~ 0.5% were positive for L1CAM (Fig. S1E); this portion drastically increased to over 40% after L1CAM immunoprecipitation (Fig. S1F). A vast percentage of L1CAM⁺ EVs in NDEV preparations were double-positive for the canonical EV markers CD9, CD63, and CD81 (Fig. S1G), confirming their EV identity. The specificity of the anti-L1CAM antibody was confirmed by FCA of NDEVs labeled with an isotype control antibody showing the absence of EVs within the L1CAM gate (Fig. S1H). The abolition of detected events after treatment with NP40 detergent further demonstrated that FCA signals originated from membrane-enclosed nanoparticles (Fig. S1D).

Given variable reports regarding the acute effects of exercise on the concentration and size of total circulating EVs [28, 29], we examined the effect of 16-week exercise on NDEVs. There were no group differences or changes over time in the concentration (Fig. S1A) and particle size (Fig. S1B and S1C) of NDEVs based on NTA. To account for any effects of differential NDEV yield on NDEV protein biomarkers, the concentration of NDEVs was used as a covariate in all models, as previously done [30].

NDEV proBDNF, BDNF, and humanin increased in the exercise group after 16 weeks compared to baseline, whereas no differences over time were seen in the control group (Fig. 1A–C, Table 2). The greatest change occurred for proBDNF, which showed a 1.8-fold increase in the exercise group [from 139.7 (56.3–223 95% confidence interval; CI) to 274.7 (183.8–365.6 95% CI) pg/mL; p = 0.007] resulting higher in the exercise compared to the control group at 16 weeks [274.7 (183.8–365.6 95% CI) vs. 150.5 (65.5–235.5, 95% CI) pg/mL, p = 0.047]. NDEV humanin was lower at baseline in the exercise compared with the control group [91.88 (10.0–173.7 95% CI) vs. 232.8 (147.9–317.8 95% CI) pg/mL], but this difference disappeared after 16 weeks [184.0 (93.1–274.9 95% CI) vs. 194.1 (107.6–280.5 95% CI) pg/mL]. A similar trend at baseline was observed for BDNF. These observations suggest that for NDEV humanin and BDNF, randomization was not as efficient in eliminating baseline differences as it was for proBDNF, with very similar baseline levels between the control and exercise groups. Since humanin, unlike proBDNF/BDNF, is ubiquitously produced, we sought to also assess baseline differences and exercise-induced changes in humanin across all cells using panTET-EVs. PanTET-EV levels of humanin were no different at baseline and its levels remained unchanged within and between groups (Fig. 1D).

In an exploratory analysis, we examined exercise effects in stratified sub-groups of apolipoprotein-E (APOE) ε4 carriers and non-carriers (Table S1), as has recently been suggested by multiple experts in the AD field [31]. Interestingly, in the exercise group, NDEV levels of proBDNF and humanin increased only in ε4 carriers; proBDNF: ε4 non-carriers [from 131.9 (− 17.1–281.0 95% CI) to 185.2 (77.0–293.5 95% CI) pg/mL; p = 0.577], ε4 carriers [from 165.9 (63.7–268.1 95% CI) to 355.5 (194.9–516.0 95% CI) pg/mL; p = 0.016]; humanin: ε4 non-carriers [from 127.0 (76.4–177.5 95% CI) to 115.0 (70.5–159.6 95% CI) pg/mL; p = 0.858], ε4 carriers [from 120.8 (79.0–162.5 95% CI) to 254.3 (134.1–374.4 95% CI) pg/mL; p = 0.005 (Fig. 2A and C). No significant changes were observed for BDNF in this stratified analysis (Fig. 2B).

As physical activity can profoundly alter the neuronal proteome [32], we explored whether exercise modulates NDEV cargo for exerkines expressed by neurons (as confirmed by https://​www.​proteinatlas.​org/​) (Table S2). We were particularly interested in irisin, an exercise-induced hormone shown to regulate cognitive function by promoting BDNF expression [33]. Although no changes were observed in NDEV irisin with exercise, as was also the case for all exerkines (Table S2), we observed a positive correlation with BDNF at baseline that was preserved at 16 weeks for both groups (Fig. S2C; Tables S3 and S4). Similar positive correlations were observed between levels of NDEV irisin and humanin and between humanin and proBDNF (Fig. S2A and B; Tables S3 and S4). We also examined the relationships between exercise-induced changes in NDEV biomarkers and the only significant association observed was between the change in NDEV irisin and that of BDNF in the exercise group (Fig. S2D; Table S5).