The 2018 National Institute on Aging—Alzheimer’s Association (NIA-AA) research framework proposed a biomarker-based definition of Alzheimer’s disease (AD) reflecting the three core underlying neuropathological processes: amyloid-β (Aβ) deposition, pathologic tau, and neurodegeneration (ATN) (Jack et al., 2018). To date, positron emission tomography (PET), structural magnetic resonance imaging (MRI), and cerebrospinal fluid (CSF) are the most widely used modalities to derive ATN biomarkers (Hansson, 2021). However, it is recognized that the development and routine use of blood-based biomarkers will improve accessibility, minimize invasiveness and cost, and assist with diagnosis, prediction of disease progression, prognosis, and response to treatments (Angioni et al., 2022).

The soluble blood phase is increasingly being investigated as a potential source of biomarkers. Plasma Aβ₄₂/Aβ₄₀ may be able to detect abnormal amyloid status in cognitively impaired and cognitively unimpaired individuals even before amyloid PET reaches its positivity threshold (West et al., 2021; Hu et al., 2022). Additionally, several phospho-tau (P-Tau) isoforms can be reliably measured in plasma and demonstrate high precision in predicting Tau deposits in the AD brain (Palmqvist et al., 2020; Mielke et al., 2021). Finally, clinical trials are increasingly utilizing plasma biomarkers either as screening tools (Leuzy et al., 2022) or as surrogates to pharmacodynamic responses (Budd Haeberlein et al., 2022).

Extracellular vesicles (EVs) are membranous nanoparticles that are secreted by all cell-types. Their cargo, encapsulated by a lipid bilayer, comprises nucleic acids, lipids, and proteins that reflect the parental cell identity and homeostatic status (van Niel et al., 2018). EVs secreted by brain cells can be detected in blood, thereby providing a window into the brain, while remaining minimally invasive (Dickens et al., 2017). We and others have isolated neuron-derived EVs (NDEVs) by selective immunoaffinity capture targeting the neuronal cell-adhesion molecule L1CAM and have shown that they can provide predictive, diagnostic, prognostic, and treatment response biomarkers for various brain diseases (Bhargava et al., 2021; Vandendriessche et al., 2022; Blommer et al., 2023). In particular, AD pathogenic proteins (Aβ, Tau, and various P-Tau species) have repeatedly been measured in NDEV preparations and have shown promise for AD clinical and preclinical diagnosis (Fiandaca et al., 2015; Kapogiannis et al., 2019; Eren et al., 2020; Eren et al., 2022). However, studies that simultaneously study biomarkers in the soluble phase of plasma and in plasma NDEVs to assess their comparative biomarker potential are still lacking.

The aim of this study was, primarily, to examine whether core ATN biomarkers (Aβ₄₂, Aβ₄₀, total Tau, P181-Tau) measured cross-sectionally and longitudinally in both NDEVs and plasma reveal differences in individuals with amnestic MCI (aMCI) who progress to AD dementia compared to individuals with AD dementia or cognitively intact individuals at two time points. Additionally, we assessed the relationship between these NDEV and plasma ATN biomarkers across all individuals and separately within each group. The selection of the groups was based on the hypothesis that individuals with MCI who progress to dementia are at a biologically active disease stage, which may be expressed by dynamic biomarker changes compared to the other two more stable states. Differences between groups were also examined, secondarily, for exploratory biomarkers reflecting additional, and often, overlapping pathologies, synaptic integrity, which weakens early in the course of AD (Clare et al., 2010), and neurotrophic factors, which mediate neuronal survival, maintenance, and regeneration, and have a potentially pathogenetic role in AD with their expression levels being abnormal (Buchman et al., 2016; Lattanzio et al., 2016). These biomarkers included a-synuclein (a-syn), TAR DNA-binding protein 43 (TDP-43), neurogranin (NRGN), P231-Tau, heavy chain neurofilaments (Nf-H), brain-derived neurotrophic factor (BDNF), and precursor brain-derived neurotrophic factor (proBDNF). Specifically, a-syn may be involved in the earliest stages of AD development, reflected in altered CSF levels and associations with brain Aβ plaque deposition in both sporadic and familial AD cases (Twohig et al., 2018; Vergallo et al., 2018). Similarly, NRGN levels are elevated in CSF, depleted in the brain, and associated with poorer cognitive performance in AD patients (Thorsell et al., 2010). Finally, proBDNF and BDNF have been associated with cognitive impairment (Michalski and Fahnestock, 2003; Peng et al., 2005) with several studies showing decreased levels in patients with MCI and AD dementia (Yasutake et al., 2006; Laske et al., 2007) and others showing conflicting results with increased levels in AD patients (Angelucci et al., 2010; Faria et al., 2014).

In this study, we utilized a cohort of three distinct and well-characterized diagnostic groups and measured longitudinally the core ATN biomarkers, in both plasma and NDEVs, along with several other exploratory markers reflecting additional pathologies, synaptic integrity, and neurotrophic factors. Unlike past studies, NDEVs were isolated from plasma samples by immunoaffinity capture targeting three separate neuronal markers, and not solely L1CAM, to yield a wider range of neuronal origin EVs. NDEV-based measures revealed group differences at baseline even when plasma did not, while the two types of biofluids did not show any association for any assessed marker. The Aβ₄₂/Aβ₄₀ ratio measured either in plasma or NDEVs was lower in MCI-DEM than NRM-NRM group at baseline, whereas NDEV-associated proBDNF was lower in MCI-DEM and DEM-DEM groups compared to NRM-NRM, with its levels decreasing longitudinally as dementia was advancing. Interestingly, these two biomarkers achieved the highest AUCs for discriminating between diagnostic groups, while additionally, NDEV-associated proBDNF was positively correlated with MMSE score. Paradoxical findings emerged for NDEV-associated total Tau and P181-Tau, with the direction of differences at baseline being the opposite from what had previously been observed in other cohorts (Fiandaca et al., 2015; Kapogiannis et al., 2019; Eren et al., 2020). Besides proBDNF, no group differences were found for the rest of exploratory markers.

Additionally, we performed a comprehensive assessment of the EV composition of each sample by measuring particle composition by NTA, as well as measurement of CD9, CD63, and CD81 (tetraspanin transmembrane EV markers) as well as Syntenin-1, VLA-4, and Cytochrome C (intravesicular EV markers reflecting different pathways of biogenesis) by ProcartaPlex™ Simplex™ multiplex EV assay. If there were any systematic differences in EV composition between groups, one would expect this to be reflected in multiple measures. This was not the case, and the only group difference observed was for NDEV concentration by NTA, which was decreased in the DEM-DEM group at baseline compared to the other two groups. One should bear in mind that neither NTA nor any of the bead-based EV assays like the ones employed in this study, can detect total EVs originating from all EV biogenesis pathways. For example, NTA has a size limit resolution of approximately 50 nm, at best, which is influenced by the instrument’s light scattering detection sensitivity (practically, a large portion of EVs < 100 nm goes unobserved). Thus, NTA EV concentration reflects primarily larger EVs and excludes a large portion of smaller EVs mainly of endosomal origin (where many biomarkers used in this study originate). On the other hand, bead assays based on the capture and detection of pan-tetraspanins as canonical EV markers might exclude other subpopulations of small EVs, presumably originating from both the endosomal pathway as well as the plasma membrane, previously reported to be devoid of CD9/CD63/CD81 (Kowal et al., 2016). Therefore, the discrepancies in the NDEV concentration between diagnostic groups obtained from NTA and the multiplex EV assays render the observed difference for the DEM-DEM group at a single time point questionable.

The assessment of tetraspanin markers, along with Cytochrome C, Syntenin-1, and VLA-4 showed that NDEVs carried increased levels of CD9, CD81 and Syntenin-1, compared to CD63 and VLA-4 (Supplementary Figures S2C–H, S4B). These results are in line with proteomics studies (Pulliam et al., 2019; Anastasi et al., 2021) and the NanoView^® pan-tetraspanins fluorescent assay (Supplementary Figures S3C, D) showing low to undetectable CD63 levels, comparable to those of Cytochrome C. Additionally, diameter range of particles was 144–230 nm as determined by NTA, similar to those from JHADRC and other cohorts (Kapogiannis et al., 2019; Eren et al., 2022); however, others have shown deviating average values (e.g., 100–125 nm) (Nogueras-Ortiz et al., 2020). Aside from expected size differences due to subject-to-subject variability across cohorts, size determination via NTA is subject to considerable technical variability originating from fluctuating sample dilutions, which in turn are required for optimal image processing. As previously reported (Bachurski et al., 2019) and validated by our group, fold dilution of plasma derived EVs is negatively correlated with NTA particle size mode (Supplementary Figures S5D, S6H). Although this source of technical variability might result in inaccurate size determination across studies, our objective of comparing EV size differences across the experimental groups of the present study was met by measuring samples at an optimal particle density range achieving reduced size variability (Supplementary Figure S5D).

Regarding the Aβ₄₂/Aβ₄₀ ratio measured in plasma, our findings are in line with those of other studies that examined plasma amyloid levels in longitudinal cohorts. Perez-Grijalba et al. (2019) reported lower Aβ₄₂/Aβ₄₀ levels in MCI subjects compared to controls, which were also negatively correlated with cerebral amyloid deposition and predicted future development of dementia. Similarly, in another study, low plasma Aβ₄₂/Aβ₄₀ levels were associated with more pronounced cognitive decline over time (Giudici et al., 2020). Additionally, in our study, the numerator of the ratio, Aβ₄₂, measured in plasma was also lower in MCI-DEM compared to NRM-NRM group at baseline, decreased over time in DEM-DEM group, and was positively correlated with MMSE score. Similarly, in NDEVs, its levels were lower in MCI-DEM and DEM-DEM groups compared to NRM-NRM. Although the latter finding comes in contrast with some other studies of NDEVs (Fiandaca et al., 2015; Jia et al., 2019), in a study of pathologically confirmed AD cases from the same site (JHADRC), lower rather than higher NDEV-associated Aβ₄₂ levels were associated with worse cognitive performance (Eren et al., 2022). In the largest longitudinal study of blood NDEV biomarkers for AD prediction, no differences were seen in Aβ₄₂ between individuals who would go on to develop AD compared to those who would remain cognitively normal (Kapogiannis et al., 2019). These conflicting results may be attributable to the fact that Aβ is associated with the external surface of plasma EVs rather than true intra-vesicular cargo (Eitan et al., 2016). Operationally, they raise doubts as to whether NDEV-associated Aβ₄₂ is a reliable biomarker for disease diagnosis or monitoring.

NDEV-associated proBDNF produced significant results in multiple analyses, including lower levels in MCI and AD dementia compared to the cognitively normal state and further declines with dementia progression, a positive correlation with MMSE score, and the second best performance in group classification by ROC analysis. Unlike proBDNF, mature BDNF showed no differences between groups. These findings are in agreement with the study by Eitan et al. (2023), in which levels of NDEV-associated proBDNF were also found lower in AD compared to control individuals and were positively correlated with MMSE scores. Our results, also, indicate that levels of NDEV-associated proBDNF may reflect true relative concentrations in brain cells, since similar decreases in MCI and AD dementia compared to the cognitively normal state were found in an autopsy study using cerebral tissue samples (Peng et al., 2005). Moreover, NDEV-associated proBDNF has shown biomarker properties for additional age-related conditions. In a longitudinal study using samples from the Baltimore Longitudinal Study of Aging, increased levels of proBDNF in L1CAM (+) NDEVs were associated with walking speed decline (Suire et al., 2017). Overall, these results indicate that NDEV-associated proBDNF may be a useful biomarker for diagnosis and disease monitoring.

The inclusion of well-characterized subjects, the availability of longitudinal measurements spanning a mean of 3.1 years and the fact that both plasma and plasma NDEVs were assessed simultaneously are considerable strengths of this study. Moreover, this is the first study using a combination of antibodies against L1CAM, GAP43, and NLGN3 to derive NDEVs. Nevertheless, important limitations must be recognized. First, even though we utilized three well-characterized diagnostic groups, we did not study MCI individuals who remained stable over time as a potentially interesting contrast to those with MCI who progressed to dementia. Additionally, a paradoxical result emerged for NDEV-associated total Tau and P181-Tau biomarkers, with the differences observed at baseline being the opposite from what had previously been observed. On the other hand, no differences were found between groups for P231-Tau, contradicting previously reported strong correlations between the two phosphorylated forms (Bayoumy et al., 2021). Although, this may be attributable to the new type of NDEVs we isolated, it may also represent a peculiarity of this small cohort. Additionally, in this study, we measured P181-Tau levels using the SimoaⓇ technology. Nevertheless, in a study by Mielke et al. (2021) that comparatively assessed various analytical assays, P181-Tau levels were correlated with cognitive performance only when measured with the Meso Scale Discovery platform, and not with SimoaⓇ. Although, in that study, P181-Tau was assessed in the soluble phase of plasma, these findings regarding the analytical assays performance may also have implications and extend to the study of EVs. Moreover, NDEV-associated total Tau has shown a lack of association with cognitive performance (Eren et al., 2022), questioning its suitability as a disease monitoring biomarker. Collectively, these observations warrant further research on NDEVs isolated by immunoaffinity capture targeting the combination of L1CAM, GAP43, and NLGN3, using various analytical assays for the assessment of Tau biomarkers.

Importantly, our study generated a novel hypothesis to inform future research. The complete lack of association between NDEVs and plasma for any assessed marker may indicate that the two types of biofluids demonstrate distinct biochemical signatures, reflecting the intra- and extracellular environment, respectively. Additionally, although Tau is highly abundant in the central nervous system (CNS), it is also present in peripheral tissues (e.g., liver, kidney, heart) implying that Tau forms measured in plasma originate, to some extent, from non-CNS sources (Dugger et al., 2016). In line with this argument, Barthelemy et al. estimated that only a fifth of the signal measured by plasma total Tau is brain-derived while the remainder originates from peripheral sources (Barthelemy et al., 2020). Similarly, Aβ peptides are generated in appreciable amounts from sources outside CSN (e.g., skeletal muscles, platelets, vascular wall) allowing for an active and dynamic interchange between the brain and periphery (Li et al., 1998; Kuo et al., 2000; Van Nostrand and Melchor, 2001). Although, this lack of association between the two biofluids may be attributable to the different nature of their biomarkers, it also assuages concerns that NDEV markers of neurodegenerative diseases simply represent soluble contaminants or non-specific association with the EV protein corona. Multiple lines of evidence implicate EVs in AD pathogenesis: intracellularly generated Aβ may be released into the extracellular compartment associated with EVs, facilitating its uptake by recipient cells causing cytotoxicity (Toh and Gleeson, 2016; Sardar Sinha et al., 2018). Similarly, for Tau, convincing evidence exists that EVs act as vehicles for its trans-synaptic spreading (Wang et al., 2017). Our findings suggest that EV Aβ and Tau cargos may vary dramatically depending on the stage of disease, with levels varying over time, something that ought to be seriously considered when attempting to address their pathogenic significance. Moreover, our findings regarding NDEV-associated proBDNF encourage the conduct of further longitudinal cohort studies that could validate its potential as a diagnostic and disease monitoring biomarker, but also examine its prognostic and predictive value.