Synaptic proteins in neuron-derived extracellular vesicles as biomarkers for Alzheimer's disease: novel methodology and clinical proof of concept

Standard protocol approvals, registrations, and consents

This study relied on de-identified samples from commercial or Government [National Institute of Aging (NIA, Baltimore, MD)] biobanks. Commercial samples were purchased from two companies, BioIVT (Westbury, NY) and precision for medicine (PMED) (Elk Grove, CA), which adhere to the most current regulations for sample collection and use. The regulatory requirements met include Institutional Review Board (IRB) approval, privacy officer authorization, appropriate government licenses, and industry accreditations, as applicable. They also adhere to the General Data Protection Regulations. Samples from the NIA were collected through NIH IRB-approved protocols, and their use in this research is allowed based on the subjects’ original consent. Regarding patients with AD, legally authorized representatives or patients consented to participate in the original studies, depending on the assessment of consent capacity at the time of blood draws. All samples were de-identified prior to their transfer to NeuroDex for processing under the terms of a Collaborative Research and Development Agreement.

Human plasma specimens: Human EDTA-K2 plasma was used in all experiments. Pooled human EDTA-K2 plasma (BioIVT, HMN69947-X, ten donors per pool) was used for quality control experiments and as internal references. Samples procured from BioIVT and PMED comprised a single cohort for most analyses. The NIA cohort comprised 20 individuals with high probability (early-stage) AD as established by the NIA-AA criteria^[20] and 19 healthy, cognitively normal age- and sex-matched controls, participating in NIH IRB-approved studies (NIA Clinical Unit; Baltimore, MD, USA). AD diagnosis was based on clinical criteria and abnormal CSF levels of amyloid-beta (Aβ) peptide 1-42 (Aβ42 < 192 pg/mL) and p181-Tau > 23 pg/mL^[21]. Demographic and clinical data for all cohorts are summarized in Table 1. All NeuroDex employees involved in sample processing and analysis were blinded to the identity of the samples, and only internal sample IDs were used.

NDEV isolation: NDEV enrichment was performed using the NeuroDex ExoSORT kit (Cat. No NDX_ESNeuro, NeuroDex, Natick, MA). Briefly, plasma samples were precipitated with 1/2 plasma volume of a NeuroDex total EV isolation reagent (Cat. No. NDX_TPC) in 1/2 plasma volume, and pellets were resuspended in a binding buffer. Magnetic beads were conjugated with NeuroDex proprietary antibodies against GAP43 and NLGN3 and blocked with the NDX Blocking Reagent. The beads were incubated overnight with plasma samples at 4 °C with slow rotation. The following day, beads-NDEVs complexes were collected using a magnetic separator, washed three times using ExoSORT wash buffer, and transferred into ExoSORT elution buffer. The elution of EVs was performed for 5 min at 50 °C, followed by the removal of the beads. Eluates were collected and transferred into clean tubes with ExoSORT lysis buffer.

RNA isolation: NDEVs captured on ExoSORT beads, as described above, were washed twice with NDX ExoSORT wash buffer (provided in the kit) and transferred to QIAzol reagent (Qiagen, Cat. No. 79306). RNA was isolated using the miRNAeasy serum and plasma kit (Qiagen, Cat. No. 740004) according to the manufacturer’s instructions.

PCR analysis: cDNA was generated using SuperScript™ IV VILO™ reagent mix (Thermo-Fisher, Cat. No. 11766050). qPCR was performed using TaqMan master mix (Thermo Fisher Scientific, Cat. No. 4305719) and predesigned TaqMan gene expression assays (see Table 2) in a Step One Thermal Cycler (Applied Biosystems, Cat. No. 4369074). The data were expressed as ΔCt, without using any gene for normalization, as there is no consensus in the field on a normalization gene for plasma EV RNA.

EV isolation from cell culture media: Conditioned media (CM) from a culture of cortical neurons differentiated from induced pluripotent stem cells (iPSCs) and maintained for 4-8 weeks in differentiation media) was obtained from BrainXell (Cat. No. BX-0200). The cells were maintained in a proprietary serum-free, EV-free medium. CM was centrifugated at 1,000 x g for 10 min, and supernatants were collected and centrifugated at 3,000 x g for an additional 10 min. Multiple collections were combined for a total of 200 mL, and EVs were isolated by ion exchange chromatography (IEC) and concentrated by ultrafiltration as previously described^[22]. Briefly, CM was loaded on Q Sepharose Fast Flow columns (Cytiva, Cat. No. 17-0510-10). The unbound material was washed off with equilibration buffer, followed by wash and elution with buffers containing sequentially increasing concentrations of NaCl. The EVs were concentrated by dialysis in 100 kDa MWCO spin filtration units (Pall Corporation, Cat. No. MAP100C38) against NeuroDex storage buffer supplemented with protease and phosphatase inhibitors (Thermo Fisher Scientific, Cat. No. 78429, 78426).

Immunoassays: Tau, p181-Tau, Aβ40, and Aβ42 were measured according to the manufacturer’s instructions in undiluted NDEVs lysates using commercial Luminex kits (EMD Millipore, Cat. No. MXHABTM0N02010 and HNABTMAG-68K-04). Mature BDNF and proBDNF were measured in NDEVs lysates diluted 1:10 using a commercial Enzyme-Linked Immunosorbent Assay (ELISA) kit (Biosensis, Cat. No. BEK-2241). Apolipoprotein A (ApoA) (R&D system DY3664-05) and Albumin (R&D system DY1455) were also measured using commercial ELISA kits according to manufacturers’ instructions. Rab3a, GLUR2, NLGN3, and L1CAM, as well as the EV markers Flotillin (FLOT)-1 and CD9, were measured by intact EV ELISA as described elsewhere^[23]. FLOT-1 and CD9 values were used to normalize ELISA results since the high content of non-EV particles in plasma challenges the accuracy of particle counts by NTA or light scatter.

Intact EV Luminex analysis was performed with the NeuroDex Lumin-EV kit (NDX_LUMTET). Briefly, MagPlex microspheres (MC100XX-01) were conjugated with antibodies against CD9, CD63, CD81, or with antibodies against synaptic proteins using an ABC coupling kit (Luminex Corp., Cat. No. 4050016). The resultant capture beads were used in a multiplex format to capture EVs directly from plasma as described previously^[22]. The captured EVs were detected with a pan-tetraspanin antibody cocktail (for specific antibodies, see Table 3).

An intact EV ELISA assay was performed with the NeuroDex ELISA kit (NDX_ELISA). High-binding plates (Corning, Cat. No. 9018) were coated overnight with antibodies against the synaptic proteins of interest (antibody catalog numbers in Table 3). Then, the plates were blocked for 2 h, washed, and incubated for 2 h with detection antibodies at room temperature on a plate shaker. Next, the plates were washed three times, and biotinylated GAP43 antibody was added for 2 h, followed by a wash step. Next, streptavidin-HRP was added for 30 min, and the plates were washed and developed by TMB.

Internal reference/standards: For all assays, two internal reference standards were included, comprising pre-evaluated pooled human plasma or NDEV isolation, as appropriate. In cases of inconclusive or missing reference data, the test was repeated.

Transmission electron microscopy: Isolated NDEVs resuspended in 3% PFA and incubated with negative stain (Uranyl Acetate) were visualized at the Brandeis University Cell Imaging Facility (Dr. Berith Isaaks) using a Morgagni transmission electron microscope (FEI, Hillsboro, OR), operating at 80 kV and equipped with a Nanosprint5 CMOS camera (AMT, Woburn, MA).

Western blotting: NDEVs protein lysates and reference samples were processed using the WES apparatus and 25-sample cartridges containing loading buffer and secondary antibodies (ProteinSimple), according to the manufacturer’s instructions (for specific antibodies, see Table 3).

Proteomic analysis: Isolated NDEVs samples were sent to Tymora Analytical LLC for proteomic analysis using a proprietary procedure optimized for EV analysis.

Lipidomic analysis: Lipidomic profiling of NDEVs was performed by ultra performance liquid chromatography-tandem mass spectrometry (UPLC-MSMS) at the Columbia University Lipidomic Facility, as described previously^[24].

Nanoparticle tracking analysis: NDEV preparations were diluted 10 times in pre-filtered PBS (20 mm filters), and NTA analysis was performed using NanoSight 500 (Malvern Panalytical) as described previously^[25].

Statistical analysis

Statistical analysis was conducted with SPSS v27.0 (IBM). To assess group differences for individual biomarkers, we used linear mixed models including each biomarker as an independent variable, sex and group (early-stage AD vs. control) as factors, and age as a covariate. Pairwise comparisons were assessed using least-squares means. Although this exploratory study was not powered to correct for multiple comparisons, when results met Bonferroni correction (i.e., significance level P < 0.003125 for 8 independently measured biomarkers in 2 independent cohorts), this is noted in Results and Figures. To assess whether results may be replicable across diverse populations, the NIA and BioIVT/PMED Cohorts were analyzed separately. We also assessed performance in diagnostic classification for AD/MCI vs. control status. To determine the simplest and most accurate classifier model based on multiple biomarkers, we performed discriminant classifier analysis stepwise with the Wilks’ Lambda method, allowing biomarkers to “compete” against each other in each step with a minimum partial F of 3.84 to enter and 2.71 to remove. To determine the ability of individual biomarkers in group classification, receiver operator characteristic (ROC) analysis was conducted under the non-parametric distribution assumption. To assess the relationships among biomarkers and between biomarkers and clinical and cognitive scores, we computed zero-order and partial Pearson correlations (controlling for age and sex). Discriminant, ROC, and correlation analyses were conducted after combining all cohorts into one.

Data availability: De-identified data are available upon request from a qualified investigator.

ExoSORT captures extracellular vesicles

ExoSORT is a novel immunoaffinity-based method for EV isolation and was therefore evaluated for compliance with MISEV guidelines^[19] [Figure 1]. Spherical morphology was shown by electron microscopy [Figure 1A], and average particle diameter in the 50-250 nm range was confirmed by nano-tracking analysis (NTA; Figure 1B), even though it seems that NDEVs display a non-normal size distribution. The presence of typical EV markers, CD9 and flotillin (FLOT1), as well as reduced levels of the non-EV protein albumin, were shown by Western blot/WES [Figure 1C]. The depletion of contaminating plasma proteins was assessed using ELISAs for Apolipoprotein A1 (ApoA1) and albumin, with depletion of 95.95 ± 0.94% and 99.8 ± 0.2%, respectively (N = 6, P < 0.001; Figure 1D and E). The relatively higher level of ApoA1 contamination compared to albumin is consistent with previous reports demonstrating EV interactions with HDL/LDL^[26]. Lipidomic analysis showed high concentrations of characteristic EV lipids, such as cholesterol, phosphatidylserine, and sphingomyelin [Supplementary Figure 1A and Supplementary Table 1A], and unbiased proteomic analysis revealed multiple EV-specific proteins in the NDEV preparations [Supplementary Figure 1A and Supplementary Table 1B].

Figure 1. NDEV isolation and characterization. The isolated NDEVs demonstrate key EV characteristics. (A) NDEVs were eluted from ExoSORT capture beads, fixed in 3% PFA, and visualized by transmission electron microscopy. Vesicular shape and size under 200 nm were observed. (B) The size distribution of eluted NDEVs was determined by nanoparticle tracking analysis. The NDEVs average diameter ± SEM was 127 ± 72 nm. (C) Western blot analysis (WES instrument) showed that NDEVs (N) contained enriched levels of typical EVs markers CD9 and FLOT1, and reduced levels of albumin, compared to unprocessed plasma (P). Mouse brain extract (B) was used as a control. Total protein loaded (ug) is shown for each lane. (D, E) Levels of albumin and ApoA1 were much lower in isolated NDEVs than in unprocessed plasma, as quantified by ELISA (R&D Systems, Cat. No. DY1455 and DY366405). The fractions of albumin and ApoA1 in NDEVs were 0.78 ± 0.2% and 4.05 ± 0.94% of the levels observed in unprocessed plasma, respectively (P < 0.0001 for both proteins). (F, G) The enrichment for neuronal material was ascertained by comparing neuron-specific proteins and RNA in plasma and NDEVs. (F) WES analysis showed that NDEVs (lanes labeled N) are enriched for neuronal markers NeuN and GluR2 compared to unprocessed plasma (lanes labeled P). Protein amounts loaded onto each lane are shown above. (G) mRNA encoding markers of EV origin from erythrocytes (hemoglobin, HBB), platelets (platelet factor 4, PF4), liver (albumin, ALB), and neurons (neurogranin, NRGN) were measured by QPCR in plasma and NDEVs. The resultant values were adjusted to the input volume. Note that there were lower levels of HBB, PF4, and ALB mRNA in NDEVs compared to plasma (1,4, 34, and 68-fold, respectively; P < 0.0001 for all comparisons). In contrast, NRGN mRNA levels were similar between NDEVs and plasma (P = 0.66).

NDEVs captured by ExoSORT are enriched for neuronal markers

NDEVs captured by ExoSORT showed higher levels of neuron-specific proteins and 24-hydroxycholesterol, a lipid enriched in the brain^[27], compared to the procedural control, material immunocaptured using IgG isotype antibody [Table 4]. The enrichment of neuronal proteins was also demonstrated by WES with antibodies against neuron-specific proteins, GluR2 and NeuN [Figure 1F]. Unbiased proteomic analysis also revealed significant enrichment for neuronal proteins, as was shown by gene ontology analysis [Supplementary Table 2]. Since EVs are the major carriers of circulating mRNAs^[28] and RNA measurements are less sensitive to matrix effects, neuron-specific mRNA content was compared between NDEVs and unprocessed plasma [Table 4 and Figure 1G]. The NDEV levels of mRNA encoding for hemoglobin beta chain (HBB, erythrocyte marker), as well as platelet factor 4 (PF4, platelet marker), and albumin (ALB, liver marker), were less than 6% of those in unprocessed plasma, suggesting successful depletion of non-specific material; in contrast, the levels of mRNA encoding for the neuronal marker NRGN were similar in NDEVs and plasma, pointing to efficient recovery of neuron-specific EVs [Figure 1G]. The levels of neuronal markers in NDEVs were 35 times higher than in the material isolated with control IgG (procedural control, P < 0.0001; Table 4). The diminished levels of non-neuronal mRNAs and the preservation of neuronal mRNAs and proteins corroborate the specificity of the NDEV isolation.

ExoSORT isolates endogenous and spiked NDEVs with high efficiency and precision

To estimate the NDEV isolation efficiency, we compared NRGN mRNA levels in NDEVs isolated by ExoSORT and in unprocessed plasma and found that NRGN levels in NDEVs comprised 86 ± 7% of those in the parent plasma [Figure 1G]. Further estimates of NDEV isolation efficiency were based on conducting sequential rounds of ExoSORT, wherein the supernatant remaining after capture bead removal in each round was retained and subjected to a new round of ExoSORT [Supplementary Figure 2]. With NRGN mRNA level used as a measure of NDEVs extraction, the difference between two sequential yields was 87 ± 9% providing a measure of isolation efficiency [Figure 2A and B]. Next, we used a classical spiking approach using EVs isolated from human iPSC-derived cortical neurons. Culture-derived EVs, which were characterized according to MISEV guidelines^[19] [Supplementary Figure 3A-C], contain exceptionally high Tau levels [Supplementary Figure 3D], at least ten times higher than those expressed in endogenous plasma NDEVs. This difference enabled the use of Tau measurements to estimate spiked iPSC EV recovery from plasma since their contribution far exceeds that of endogenous NDEVs. iPSC EV in amounts containing 250-2000 pg Tau (30-200 ul) was spiked into 300 ml plasma aliquots, and Tau recovery was measured using a Luminex-based Milliplex assay (EMD Millipore). The recovery of the spiked reference material was estimated at 51 ± 10%, and the efficiency was not dependent on the input amount (N = 4,Figure 2C). The same spiking approach was used to evaluate the consistency between NDEVs recovery from plasma samples of disease-free controls (N = 11) and AD patients (N = 6) [Figure 2D]. These results further demonstrate the compatibility of the NDEV isolation methodology with the plasma matrices of both control and AD individuals. Of note, each of the capture antigens used in ExoSORT offered an improvement in NDEV capture compared to L1CAM [Supplementary Figure 4].

Figure 2. Efficiency and precision of NDEV isolation. (A) NDEV isolation was performed on unprocessed plasma samples (Round 1, N = 4, aliquoted from the same plasma pool). After the first round of ExoSORT, supernatants were retained and subjected to an additional round of capture with fresh ExoSORT beads (Rounds 2 and 3, see Supplementary Figure 3). NRGN mRNA was measured by TaqMan QPCR. Ct values are shown. (B) Tau-rich EVs generated by iPSC-derived neurons were spiked into plasma aliquots. The recovery of spiked NDEVs in three consecutive rounds of ExoSORT was measured as above, and residual NDEV amounts after each round were estimated. Fold change differences between initial plasma pools (Round 1) and the two consecutive rounds are shown (Rounds 2 and 3). The signals in Rounds 2 and 3 were 8.7 ± 5.5% and 0.02 ± 0.01% in Round 1, respectively. (C) EVs isolated from a culture of human iPSC-differentiated cortical neurons were spiked into 300 ml plasma aliquots in 4 different concentrations based on known amounts of Tau. EV recovery was calculated by the following formula: $$ \text { Recovery }=\frac{\text { Tau concentration in spiked sample-Tau concentration in non-spiked sample }}{\text { Tau input }} $$ Tau recovery after ExoSORT remained uniform for a range of spiked-in EV-associated Tau amounts and was much higher than with IgG control (P = 0.026, two-way ANOVA). (D) Identical EV amounts from a culture of human iPSC-differentiated cortical neurons were spiked into plasma samples from 11 healthy individuals and six dementia patients. The variability of recovery was < 13% for all samples, with no difference between groups (P = 0.6). (E-G) Five different ExoSORT procedures were performed on different days, using aliquots derived from the same two plasma samples (QC1 and QC2, respectively). The levels of p181-Tau, total-tau, and Aβ42 were measured in the same assay (Luminex-based Milliplex), yielding variability estimates of 20.3%, 22.7%, and 14.9% for p181-Tau, Tau, and Aβ42, respectively. (H) ExoSORT was performed in triplicate technical replicates by two operators on plasma from 4 distinct donors, and the results were compared using Tau levels as an output. We note similar variations between donors and similar levels determined by different operators for each donor (P = 0.83).

ExoSORT precision was also evaluated in the absence of spiked reference standards, yielding a coefficient of variance (CV) between 8.0 and 22.7% for two random plasma samples processed repeatedly in five independent experiments [Figure 2E-G]. The variability between two proficient operators was lower than the variability between donors (N = 4, CV < 21%; Figure 2H). Moreover, the isolation of NDEVs was successfully performed by a proficient operator in a different lab (the Kapogiannis lab at the NIA) [Supplementary Figure 5]. Importantly, the estimated variability combines the variance from two sources: NDEV isolation (ExoSORT) and biomarker measurements (ELISA or Luminex).

NDEVs-associated p181-Tau and Aβ42 confirm ExoSORT diagnostic potential and robust performance

The diagnostic potential of the NDEVs isolated using L1CAM-based immunocapture has been demonstrated by multiple groups, including the authors of this study^[29,30]. To examine the ability of ExoSORT-derived NDEVs to also detect previously reported biomarker differences, we first tested their performance in a cohort of 10 early-stage AD patients and 10 age-matched control participants (BioIVT). Similar to reports using L1CAM isolation, p181-tau and Aβ42 levels were higher in AD samples compared to control specimens [Figure 3A and B]. The robust assay performance was ascertained by repeated analyses of independently stored aliquots of the same samples [Figure 3C and D], yielding a strong correlation for p181-Tau (R² = 0.93 P < 0.001) and a medium-strength correlation for Aβ42 (R² = 0.55, P = 0.002). Furthermore, similar results were obtained in two additional well-characterized cohorts (20 high-probability early AD samples and 19 control samples from the NIA and 30 early AD samples, and nine control samples from PMED) [Figure 3E-H]. Importantly, the differences were statistically significant, and the direction of the differences between control and AD samples was the same across different cohorts; however, the absolute Aβ42 levels varied across cohorts, suggesting assay sensitivity to preanalytical conditions unique to each cohort.

Figure 3. p181-Tau and Aβ42 in NDEVs in three cohorts. The levels of p181-Tau, total Tau, and Aβ42 were measured in three cohorts of early-stage AD patients and control donors. (A, B) NDEVs-associated p181-Tau and Aβ42 were measured by a Milliplex assay in the BioIVT cohort (N = 10 per group, P = 0.007 and P < 0.0001, respectively). (C, D) Different aliquots of the same 20 plasma samples from the BioIVT cohort were analyzed again at a later date to assess the correlation between the two measurements (R² = 0.93, P < 0.01 and R² = 0.55, P < 0.002, respectively). (E, F) p181-Tau measurements were conducted in NDEVs derived from NIA samples (20 early AD and 19 controls, P < 0.0001) and PMED samples (30 early AD and 9 controls, P = 0.02). (G, H) NDEVs-associated Aβ42 was measured by Luminex in the NIA and PMED cohorts, and significant differences were detected between the disease and control groups (P = 0.01 and P = 0.001, respectively).

NDEVs-associated proBDNF shows potential as a biomarker for AD

Using ExoSORT, we reproduced published findings, wherein proBDNF, rather than mature BDNF, was predominantly found in association with NDEVs^[31] [Figure 4A and B]. This also provides further support for the enrichment of NDEVs for neuronal cargo, as their levels of proBDNF were much higher than in plasma. Interestingly, only NDEVs-associated proBDNF was significantly reduced in early AD compared to control samples [Figure 4A], while mature BDNF showed no differences [Figure 4B], suggesting yet another advantage of NDEV isolation for biomarker discovery. Assay reproducibility was verified by repeated analysis of multiple aliquots from the same sample (Figure 4C; R² = 0.7, P < 0.001), and the significance and direction of the differences were reproducible between the NIA and PMED cohorts [Figure 4D and E]. However, similar to Aβ42, levels of proBDNF varied widely between cohorts, again suggesting sensitivity to preanalytical conditions.

Figure 4. NDEVs and their source plasma samples were used to measure BDNF and proBDNF in early AD and control donors. (A) Measurements in the BioIVT cohort showed high enrichment of proBDNF levels in NDEVs (P < 0.001) compared to unprocessed plasma. Moreover, we observed lower levels of NDEVs-associated proBDNF in early AD compared to control individuals (P = 0.002), while no difference was observed in unprocessed plasma. (B) BDNF was 2.6-fold lower in NDEVs than in plasma (P < 0.01) and did not vary between AD and controls. (C) ProBDNF measurements in two independently stored aliquots from 20 samples showed a moderately strong correlation (R² =0.7, P < 0.0001). (D, E) NDEVs-associated proBDNF was measured in two additional cohorts from NIA and PMED; the decrease in NDEVs-associated proBDNF in early AD compared to control individuals was reproduced in the NIA cohort (P = 0.001), while the PMED cohort generated a similar trend that did not reach statistical significance (P = 0.09), potentially due to an insufficient number of control samples.

A comparison with CSF levels of classical AD markers revealed a significant negative correlation between NDEVs and CSF levels of Aβ [Supplementary Figure 6], consistent with a similar finding by Jia et al.^[32].

NDEV-associated synaptic proteins present biomarker potential for AD

Synaptic loss is a major pathological feature of AD. Previous studies have shown high levels of synaptic proteins in NDEVs^[33-35]. Many of these proteins were notably decreased in NDEVs of AD patients compared to control donors, and in some cases, these changes correlated with cognitive performance^[33]. Here, we measured five NDEVs-associated synaptic proteins: Syntaxin 1 (STXN1), GAP43, GluR2, PSD95, and NRGN. In the NIA cohort, the levels of NDEVs-associated NRGN were elevated in AD samples compared to controls, whereas GAP43, PSD95, STXN1, and GluR2 were decreased (Figure 5A-E; NRGN, P = 0.025; GAP43, P < 0.001; PSD95, P = 0.002; STXN1, P = 0.011; GluR2, P < 0.001). The AD-associated decreases in NDEV-associated STXN1 and GluR2 were also present in the combined BioIVT-PMED cohort (P = 0.031 and P < 0.001, respectively). However, for NRGN, PSD95, and GAP43, directionally consistent trends did not reach significance, with P = 0.1, P = 0.07, and P = 0.3, respectively [Figure 5F-J]. Importantly, for GluR2 and proBDNF, the differences between early AD and controls were reproducible across multiple cohorts and, moreover, survived Bonferroni correction for multiple comparisons (i.e., P < 0.003125). The biomarkers examined in this study showed neither age- nor sex-associated differences.

Figure 5. The levels of five synaptic proteins, NRGN, GAP43, PSD95, Syntaxin-1, and GluR2, were measured in three independent cohorts. (A-I) NDEVs-associated NRGN, GAP43, PSD95, Syntaxin-1, and GluR2 were measured in the NIA cohort (20 AD and 19 controls; NRGN: P = 0.025, GAP43: P < 0.001, PSD95: P = 002, Syntaxin-1: P = 0.011, and GluR2: P < 0.001). (B-J) The same proteins were measured in the combined BioIVT/PMED cohort (40 AD and 19 controls total), NRGN: P = 0.087, GAP43: P = 0.063, PSD95: P = 0.2, Syntaxin-1: P = 0.03 and GluR2: P < 0.001). (K) ROC analysis was conducted for all three cohorts combined. Each line corresponds to a different NDEV biomarker; higher values indicate AD/MCI, and lower values indicate CN Control status.

Performance of multiple NDEV biomarkers in AD diagnosis

ROC analysis for the classification of individual samples as belonging to early AD vs. control diagnostic categories [Figure 5K] showed that p181-Tau had the highest AUC (AUC = 0.832, P < 0.001), which was closely followed by GluR2 (AUC = 0.755, P < 0.001), proBDNF (AUC = 0.736, P < 0.001), GAP43 (AUC = 0.729,P < 0.001), NRGN (AUC = 0.716, P = 0.001), STXN1 (AUC = 0.709, P = 0.001), and PSD95 (AUC = 0.675, P = 0.005). Stepwise discriminant analysis for early AD vs. control classification incorporated stepwise GluR2, proBDNF, NRGN, and GAP43. The final model (cross-validated by leaving 1 out) yielded AD classification with 81.3% accuracy (P < 0.001), 94.7% sensitivity, and 61.5% specificity. Finally, we computed partial correlations (controlling for age and gender) in early-stage AD participants to examine the relationship between NDEV biomarkers and clinical scores. p181-Tau, proBDNF, and GluR2 were significantly associated with the MMSE and the Clinical Dementia Rating sum of boxes (CDR-SOB) [Table 5].