Lipidomic Analysis of Plasma Extracellular Vesicles Derived from Alzheimer’s Disease Patients

Analysis of blood-based indicators of brain health could provide an understanding of early disease mechanisms and pinpoint possible intervention strategies. By examining lipid profiles in extracellular vesicles (EVs), secreted particles from all cells, including astrocytes and neurons, and circulating in clinical samples, important insights regarding the brain’s composition can be gained. Herein, a targeted lipidomic analysis was carried out in EVs derived from plasma samples after removal of lipoproteins from individuals with Alzheimer’s disease (AD) and healthy controls. Differences were observed for selected lipid species of glycerolipids (GLs), glycerophospholipids (GPLs), lysophospholipids (LPLs) and sphingolipids (SLs) across three distinct EV subpopulations (all-cell origin, derived by immunocapture of CD9, CD81 and CD63; neuronal origin, derived by immunocapture of L1CAM; and astrocytic origin, derived by immunocapture of GLAST). The findings provide new insights into the lipid composition of EVs isolated from plasma samples regarding specific lipid families (MG, DG, Cer, PA, PC, PE, PI, LPI, LPE, LPC), as well as differences between AD and control individuals. This study emphasizes the crucial role of plasma EV lipidomics analysis as a comprehensive approach for identifying biomarkers and biological targets in AD and related disorders, facilitating early diagnosis and potentially informing novel interventions.


Introduction
Numerous studies indicate a link between Alzheimer's (AD) and perturbed lipid metabolism, highlighting the pivotal role of lipids in the development of AD [1][2][3].Plasma lipidomic analysis has been performed in mild cognitive impairment (MCI) and AD patients, showing that variations observed in lipid metabolites could signify transformations occurring in the neuronal membrane [4].In a previous study, quantitative analysis of controls and patients with MCI and AD dementia indicated alterations in individual plasma lipid metabolites, such as 40 acylcantinites, 15 sphingolipids and 90 glycerophospholipids and altered levels of phosphocholine (PC) and lypoPC [5].Another study identified differences in the plasma lipidome between AD patients and healthy age-matched controls for sphingomyelins, phosphatidylcholines, phosphatidylethanolamines, phosphatidylinositols and triglycerides, with a higher abundance of PC lipids in AD compared to control individuals, while more than 30% of lipids in ChE, PE and TG classes were linked with AD-related SNPs [6].A recent study identified blood lipid modules enriched in sphingolipids, fatty acid pathways and diacylglycerols/phosphatidylethanolamines that were associated with brain volume measures, with some connections also found with amyloid-β (Aβ) and genetic risk for AD [7,8].A targeted lipidomic exploration was performed in plasma samples from preclinical AD, MCI due to AD and healthy individuals and statistically significant alterations were detected for the diglycerol, lysophosphatidylethanolamine, lysophosphatidylcholine and sphingomyelin families, while 18:1 LPE showed statistically significant changes between preclinical/MCI-AD and healthy groups [9].When it comes to cerebrospinal fluid, a study analyzed samples from cognitively healthy, MCI and late-onset AD participants showing a progressively diminution of 1-radyl-2-acyl-sn -glycerophosphocholine (PC), 1-radyl-2-acyl-sn -glycerophosphoethanolamine, and 1,2-diacyl-sn -glycerophosphoserine levels from cognitively healthy to MCI, and then to AD [8].On the other hand, analysis of postmortem human brains of persons with mild AD, severe AD and cognitively normal controls showed alterations in different subgroups of lipids, such as glycerolipids, sphingolipids and glycerophosholipids, among them DAG 14:0/14:0, TAG 48:4/FA18:3, PE (p-18:0/18:1) and PS (18:1/18:2) [10].The above highlights the apparent impossibility of establishing a consistent lipidomic expression pattern for AD, partly due to the heterogeneous origin of lipids in the plasma.This realization motivates the search for a more targeted starting material, specifically, extracellular vesicles (EVs) of brain cell origin, to enhance our understanding of lipidomic alterations in AD.
EVs are small (30-150 nm) membranous nanoparticles continually produced and secreted by all cells into their extracellular environment, playing vital roles in cellular homeostasis and inter-cellular communication.EVs are encircled by a phospholipid bilayer membrane that is enriched in cholesterol and ceramides.EVs are composed (mainly) of exosomes, vesicles that are generated within the late endosome by inward budding of the endosomal membrane, and microvesicles, which are produced by outward budding of the plasma membrane, thereby reflecting the composition of the respective membranes [11,12].Cell-type-specific extracellular vehicles (EVs) have differential effects on the occurrence and progression of AD, such as participation in Aβ production and oligomerization as well as promotion of the clearance of toxic aggregates [13,14].Recently, exploration of the lipid composition of brain-derived EVs from human postmortem frontal cortex tissues of AD patient and control subjects was performed, identifying significant changes in glycerophospholipids and sphingolipids levels, mainly in PE lipids and polyunsaturated fatty acid-containing lipids [15].However, even this exploration was limited by the fact that brain-derived EVs have multiple cellular sources (neurons, glial cells) and by the singular origin of the frontal cortex, which may not reflect changes in other brain areas affected by AD.
EVs circulate in all biofluids, including blood, which contains EVs of diverse cellular origins, including of brain cellular origin.Different EV subpopulations carry cargos potentially revealing pathological processes in their cells of origin and may be used as biomarkers.Our group was one of the first that recognized the biomarker potential of neuronal-origin EVs in plasma for AD [16] and over the years, evidence has accumulated that EVs may provide diagnostic, prognostic and therapeutic response biomarkers for the disease [17].Although previous studies focused on the EV proteome, characterization of district groups of lipids in different subfractions of EVs fractionated from human blood of AD patients may also be used as diagnostic or prognostic biomarkers due to the differential enrichment in phospholipids, ceramides and diacylglycerol levels [18].However, apart from lipids constituting EVs, plasma lipoproteins co-precipitate along EVs.Plasma as such is a very challenging media for EV extraction.In another study, the ratio between lipid particles and EVs in plasma has been reported to be 10 12 vs. 10 7 -10 9 particles/mL, respectively [19].There is scrutiny regarding EV isolation techniques, as most of them may co-precipitate lipoproteins, which have either comparable size or density.Karimi et al. reported a 100-fold decrease in lipid particle contamination by combining two isolating techniques [20].Therefore, to perform lipidomic measurements in plasma that actually and accurately reflect EV composition, it is crucial to maximally deplete lipoproteins samples.To that end, first, we used immunoprecipitation as a means of negative selection to deplete ApoA1 and ApoE-containing lipoproteins, and, second, we used immunoprecipitation as a means of positive selection to capture three distinct EV subpopulations from plasma.Specifically, from the same plasma samples, we derived EVs of all-cell origin, derived by immunocapture of CD9, CD81 and CD63, transmembrane Tetraspanin proteins that are present (one or more of them) on all EVs; EVs of neuronal origin, derived by immunocapture of L1CAM; and EVs of astrocytic origin, derived by immunocapture of GLAST [21].We sought to identify both AD-associated differences and differences by EV subtype to gain further insights in the identity of these EV subpopulations and the role of EV-associated lipids in AD pathogenesis.

Participants and Blood Draws
We analyzed plasma samples from 5 individuals with high-probability early Alzheimer's disease (AD) (2 women, 3 men; age = (75.8± 4.5)) and 5 cognitively normal healthy controls (2 women, 3 men; age = (75.2± 4.9)).Diagnosis of high-probability early AD at the stage of mild cognitive impairment was based on the NIA-AA criteria [22] with typical clinical features and abnormal CSF levels of amyloid β-peptide (Aβ) 1-42 (Aβ42) < 192 pg/mL and of p181-tau > 23 pg/mL [23].All participants were evaluated at the Clinical Research Unit of the U.S. National Institute on Aging (NIA; Baltimore, MD, USA) under NIH IRPapproved protocols.Blood was collected in 10 mL purple top (EDTA) tubes and was processed within 2 h by centrifugation at 2500× g for 15 min at room temperature (RT) to derive plasma.Plasma was collected, leaving 1 cm above the buffy coat, and spun again at 2500× g.Pre-analytical parameters affecting EV isolation and blood lipids (i.e., blood draws acquired in the morning after 12 h fasting, processing into EDTA plasma within 2 h and immediate aliquoting in cryo-vials, freezing in −80 • C, avoidance of additional thaws, slow thawing on ice on the day of processing into EVs) [24] were identical for samples of AD and control individuals.

Extracellular Vesicle Isolation
All samples were processed on the same day.Immediately after thawing, we added 5 thrombin to 0.5 mL plasma, mixed by inversion and incubated for 30 min.Afterwards, we centrifuged at 4000× g for 15 min, which generated defibrinated plasma that was filtered through a 0.22 µm filter and transferred to a fresh tube.We isolated total EVs using the Exoquick-LP Exosome Isolation kit (System Biosciences (SBI), Inc., Palo Alto, CA, USA), which includes a step of lipoprotein depletion by immunocapture.Following the manufacturer's instructions, we conjugated magnetic beads separately with anti-ApoA1 (enriched in HDL) and anti-ApoE (enriched in LDL) and incubated with defibrinated plasma for 3 h at 4 • C. Following centrifugation and magnetic separation of the lipoproteinbead complexes, lipoprotein-depleted plasma was transferred into clean tubes.Next, we proceeded with concentration of the remaining particles by adding Exoquick solution, mixing by inverting and subjecting to overnight incubation at 4 • C. The next day, the suspensions were centrifuged at 14,000 rpm, supernatants were removed and pellets containing crude EVs were resuspended in distilled water containing 1x protease and phosphatase inhibitors.

Targeted Lipidomics
Analysis was conducted at the Biomarkers Core Laboratory (BCL) of Columbia University.Briefly, lipidomics profiling was performed using ultra performance liquid chromatography-tandem mass spectrometry (UPLC-MSMS) [26,27].Lipid extracts were prepared from samples spiked with appropriate internal standards using a modified Bligh and Dyer [28] method and analyzed on a platform comprising an Agilent 1260 Infinity HPLC integrated to an Agilent 6490A QQQ mass spectrometer controlled by Masshunter v 7.0 (Agilent Technologies, Santa Clara, CA, USA).Glycerophospholipids and sphingolipids were separated with normal-phase HPLC as described before [29], with a few modifications.An Agilent Zorbax Rx-Sil column (2.1 × 100 mm, 1.8 µm) maintained at 25 • C was used under the following conditions: mobile phase A (chloroform: methanol: ammonium hydroxide, 89.9:10:0.1,v/v) and mobile phase B (chloroform: methanol: water: ammonium hydroxide, 55:39:5.9:0.1, v/v); 95% A for 2 min, decreased linearly to 30% A over 18 min and further decreased to 25% A over 3 min, before returning to 95% over 2 min and held for 6 min.Separation of sterols and glycerolipids was carried out on a reverse-phase Agilent Zorbax Eclipse XDB-C18 column (4.6 × 100 mm, 3.5 µm) using an isocratic mobile phase, chloroform, methanol, 0.1 M ammonium acetate (25:25:1) at a flow rate of 300 µL/min.

Statistical Methods
The data were expressed as the mean value ± standard deviation (SD) of lipid classes from the measurement of five samples per group.Statistical significance of the results was calculated by unpaired two-tailed Student's t-test using GraphPad Prism™ software version 6.01 for Windows (GraphPad Software Inc., La Jolla, CA, USA).In addition to the main analysis presented above, we performed a multivariate analysis to replicate and increase our confidence in univariate results.For these analyses, percent total lipid composition values were further normalized by log-transformed EV particle concentrations by NTA.Multivariate principal component analysis (PCA) and follow-up univariate analyses were performed on the processed data.Natural logarithmic-transformed fold changes in lipid concentrations between AD and control groups vs. negative natural logarithmic-transformed p-values are depicted in the volcano plots (Figures 8,10 and 12).Comparisons between AD and control samples were made within each EV subtype.Additionally, the three EV subtypes were independently compared to each other within each patient group (healthy controls or AD).A standard scaler was applied to the input data before PCA was performed.PCA is an unsupervised dimension reduction methodology that highlights the axes, or principal components (PC), that account for the greatest variance between samples.PCs that best depicted separation between clusters were chosen to be visualized in the PCA plots (Figures 5a-c-7a-c, 9a-c and 11a-c).Loading vectors of the individual molecules from each lipid species were averaged and compared to the vector describing the relative direction of the AD cluster compared to the control cluster (separation vector) (Figures 5a-c-7a-c).The separation vector for the EV subtype characterization was calculated as the vector describing the relative position of the all-cell-origin EV cluster compared to the two brain-related cell-type-derived EV clusters (Figures 9a-c and 11a-c).The average loading vector of each species was then projected on the separation vector.The magnitude and direction of influence for species with relatively high projection vector magnitudes (top 10) are displayed in the corresponding PC plots (Figures 5a-c-7a-c, 9a-c and 11a-c).Individual lipids with large absolute loadings (top 25) are also presented (Figures 5d-f-7d-f, 9d-f and 11d-f).

Results
There were no differences in EV average concentration of NEVs, AEVs, or multi-origin EVs between AD and healthy control (HC) individuals.HC samples contained a lower concentration of AEVs compared to NEVs, a difference not seen for AD samples.Moreover, NEVs of AD individuals had a higher-diameter mode compared to that of NEVS of HC (Figure S2).Targeted lipidomics was performed on the three EV fractions, allowing the quantification of 593 lipid species, including lipid classes from the sterol lipid, glycerolipid, sphingolipid, glycerophospholipid and lysophospholipid categories (Table S1).The relative composition of EVs in the identified lipid categories expressed as normalized relative abundance differed across groups (AD and control individuals) and EV classes (Table 1).We observed a significant enhancement in glycerophospholipid abundance in neuronal EVs (NEVs, marked as L1CAM) compared to multi-origin EVs (marked as Tetraspanin-EVs) in control samples (p = 0.023), as well as in the lysophospholipids pool between astrocytic EVs (AEVs, marked as GLAST) and multi-origin EVs in both AD and control samples (p = 0.021 and p = 0.005, respectively).  1 The values are given as mean ± SD (n = 5). 2 Sterol lipids include FC + CE. 3 Glycerolipids are MG + DG + RG. 4 Sphingolipids are CEr + dhCer + SM + dhSM + MhCer + sulf + LacCer + GM3 + GB3. 5 Glycerophospholipids are PA + PC + PCe + PE + Pep + PS + PI + PG + BMP. 6Lysophospholipids are LPC + LPE + LPI + LPS.Values for each lipid class are presented in Table S1.Statistically significant samples: * (p < 0.05), ** (p < 0.01); for GLPs (L1CAM-control vs. Tetraspanin-control), for LPLs (GLAST-AD vs. Tetraspanin-AD; GLAST-control vs. Tetraspanin-control, Table S2).
Comparing the specific lipid composition of each class of EVs, profound differences were detected, as Figure 1 depicts (see Tables S1 and S2 for detailed values).S1.

Neuronal EVs
Focusing on sphingolipids, dihydroceramide (dhCer) levels were significantly decreased in NEVs compared to AEVs (p = 0.008) in AD samples, while enhanced levels of monohexosylceramide (MhCer) were observed between samples of controls (p = 0.007).The analysis showed limited alterations in glycerolipid levels among NEVs and multiorigin EVs families, as Figure 1 depicts (see Tables S1 and S2 for detailed values).Increased levels of diacylglycerol were seen in NEVs from AD samples compared to multiorigin EVs (p = 0.015, Figure 1B).No differences were found in triacylglycerol levels between control and AD NEVs (Figure S2).Raised levels of sulfatides were also indicated between multi-origin EVs and NEVs derived from AD patients (p = 0.031), as well as in controls (p = 0.006).Lastly, significantly decreased levels of GM3 were depicted in NEVs of controls compared to multi-origin EV samples (p = 0.013).No differences were found in ceramide, sphingomyelin and lactoceramide levels between control and AD samples (Tables S1 and S2).Among glycerophospholipids (Figure 1, total GLP classes in Figure S3 and Table S1), higher levels of phosphatidylcholine (PC) were found for control NEVs compared to AEVs (p = 0.041) and multi-origin EVs (p = 0.017).Significantly raised levels  S1.

Neuronal EVs
Focusing on sphingolipids, dihydroceramide (dhCer) levels were significantly decreased in NEVs compared to AEVs (p = 0.008) in AD samples, while enhanced levels of monohexosylceramide (MhCer) were observed between samples of controls (p = 0.007).The analysis showed limited alterations in glycerolipid levels among NEVs and multi-origin EVs families, as Figure 1 depicts (see Tables S1 and S2 for detailed values).Increased levels of diacylglycerol were seen in NEVs from AD samples compared to multi-origin EVs (p = 0.015, Figure 1B).No differences were found in triacylglycerol levels between control and AD NEVs (Figure S2).Raised levels of sulfatides were also indicated between multiorigin EVs and NEVs derived from AD patients (p = 0.031), as well as in controls (p = 0.006).Lastly, significantly decreased levels of GM3 were depicted in NEVs of controls compared to multi-origin EV samples (p = 0.013).No differences were found in ceramide, sphingomyelin and lactoceramide levels between control and AD samples (Tables S1 and S2).Among glycerophospholipids (Figure 1, total GLP classes in Figure S3 and Table S1), higher levels of phosphatidylcholine (PC) were found for control NEVs compared to AEVs (p = 0.041) and multi-origin EVs (p = 0.017).Significantly raised levels of phosphatidylethanolamine (PE) and plasmalogen phosphatidylethanolamine (PEp) were also observed in NEVs derived from AD patients compared to multi-origin EV samples (p = 0.0006 for PE; p = 0.002 for Pep).On the other hand, phosphatidylglycerol (PG) was significantly lower in NEVs compared to multi-origin EV samples (p = 0.005 Figure 1F).Lysophospholipids (LPLs), deacylated forms of phospholipids with a single fatty acid chain, were also measured, with an enhancement of lysophosphatidycholine (LPC) and lysophosphatidylethanolamine (LPE) in multi-origin EVs compared to NEVs of AD patients (p = 0.038, Figure 1G and p = 0.032, Figure 1H).Furthermore, significantly lower levels of lysophosphatidylserine (LPS) were observed in NEVs compared to AEVs in AD (p = 0.032.Figure 1I).

Astrocytic EVs
Phosphatidylserine (PS) was increased in AEVs derived from AD compared to control samples (p = 0.017, Figure 1, Table S1).The analysis also showed significantly lower levels of phosphatidylglycerol (PG) in AEVs compared to multi-origin EVs in AD samples (p = 0.024; Figure 1, Table S1).Lysophosphatidylcholine (LPC), a glycerophospholipid derived from the hydrolysis of one fatty acid ester of PC by phospholipase A 2 , was significantly reduced in AEVs compared to multi-origin EVs in AD and control samples (p = 0.038 and p = 0.003, respectively, Figure 1 and Table S1), as well as ether lysophosphatidylcholine (LPCs) (p = 0.032 and p = 0.0009, respectively, Figure S4 and Table S1).Moreover, a diminution of lysophosphatidylethanolamine (LPE) was found in AEVs derived from AD samples compared to multi-origin EVs (p = 0.032, Figure 1 and Table S1), while lower levels of LPS were indicated in AEVs compared to NEVs (p = 0.032 in AD, p = 0.026 in controls).Significantly higher levels of LPI were observed in multi-origin EVs of AD samples compared to NEVs (p = 0.002) and astrocytes EVs (p = 0.002), as well as multi-origin EVs of control samples (p = 0.006 for neuronal EVs and p = 0.004 for astrocytic EVs), as Figure S4 shows.Among glycerolipids, monoacylglycerol levels were significantly higher in AEVs derived from AD patients compared to multi-origin EVs (p = 0.026), whereas a decrease in diacylglycerol levels was also seen between these EV types (p = 0.009), as Figure 1 shows.No differences were found in triacylglycerol levels between control and AD samples (Figure S2).Furthermore, specific classes of sphingolipids such as dihydrosphingomyelin, monohexosyl ceramidewere, sulfatide, GM3 and GB3 were diminished in control AEVs compared to multi-origin EVs (p = 0.045 for dhSM; 0.023 for MhCer; 0.032 for Sulf; 0.008 for GM3; 0.021 for GB3), without significant alterations between EV types in AD samples (Figure S2 and Tables S1 and S2).Lastly, important insights were gained for the comparison between AEVs and multi-origin EVs controls, such as a diminution of bis(monoacylglycero)phosphate (BMP) (p = 0.048), an enhancement of free cholesterol (p = 0.023), as well as a decrease in cholesterol ester levels (p = 0.027) (Tables S1 and S2).

Multi-Origin EVs
Two important glycerophospholipids, phosphatidic acid (PA) and phosphatidylserine (PS), were increased in multi-origin EVs of AD patients compared to control samples (p = 0.003 and p = 0.021, respectively, Figure 1, Table S1).All the other significant changes between multi-origin EVs and NEVs or AEVs have been extensively described above.

Cholesterol Ester and Glycerolipid Levels in EV Cargo
The most highly abundant lipid class detected was cholesterol esters (CE), with some of their groups presenting significant differences between EV types (Table S3).As shown in Figure 2, an increase in CE 22:3 was observed in AEVs derived from AD patients compared to multi-origin EVs (p = 0.027), in CE 22:4 between AD AEVs and NEVs (p = 0.042), as well as in CE 22:6 abundance between AEVs and multi-origin EVs in AD samples (p = 0.041).A diminution of CE 20:3 was found in control multi-origin EVs compared to control AEVs (p = 0.001), as well as in control NEVs compared to multi-origin EVs for CE 22:4 species (p = 0.028).CE 18:2 and CE 20:4 were the most abundant among CE lipids (Table S3).For specific values, see Table S3.

Conclusions
Lipidomics holds substantial promise in investigating the etiology of AD and pinpointing biomarkers for early diagnosis, as perturbed lipid metabolism is linked to disease pathophysiology.This study reports an optimized workflow for EV isolation followed by targeted lipidomic analysis of distinct EV subpopulations (neuronal, astrocytic, multiorigin) in healthy control and AD patients.Significant differences were observed in the levels of lipids in GL, GPL, SL and LPL classes among neuronal, astrocytic and multiorigin EV subtypes, validating their differential derivation from the respective cells.Importantly, more differences were seen when comparing AEVs or NEVs to multi-origin EVs rather than when comparing them to each other.The relative similarity in the AEV and NEV lipidome is consistent with their derivation by developmentally related cell types that reside in the same organ, the brain.These results provided new insights into the role of MS-based lipidomic methodology used to explore plasma EVs.Further investigations encompassing an expanded cohort of AD patients will ascertain whether the lipid constitution derived from subfractions of EVs can be employed as viable tools for diagnostic and prognostic assessment of disease activity.
The present exploratory study suggests that, in NEVs derived from AD samples, PG levels were significantly decreased compared to controls.In NEVs, dhCer levels were significantly diminished compared to AEVs, as well as DG, GM3 and LPI levels compared to multi-origin EVs.On the other hand, PE and PEp were increased in NEVs compared to multi-origin EV samples.Moreover, in AEVs, raised levels of PS were observed in AD compared to control samples, while a decrease in PG, LPC and LPE and an enhancement of MG were found in AEVs derived from AD samples compared to multi-origin EVs.Finally, in multi-origin EVs from AD compared to control samples, PA and PS were significantly increased.

Conclusions
Lipidomics holds substantial promise in investigating the etiology of AD and pinpointing biomarkers for early diagnosis, as perturbed lipid metabolism is linked to disease pathophysiology.This study reports an optimized workflow for EV isolation followed by targeted lipidomic analysis of distinct EV subpopulations (neuronal, astrocytic, multiorigin) in healthy control and AD patients.Significant differences were observed in the levels of lipids in GL, GPL, SL and LPL classes among neuronal, astrocytic and multi-origin EV subtypes, validating their differential derivation from the respective cells.Importantly, more differences were seen when comparing AEVs or NEVs to multi-origin EVs rather than when comparing them to each other.The relative similarity in the AEV and NEV lipidome is consistent with their derivation by developmentally related cell types that reside in the same organ, the brain.These results provided new insights into the role of MS-based lipidomic methodology used to explore plasma EVs.Further investigations encompassing an expanded cohort of AD patients will ascertain whether the lipid constitution derived from subfractions of EVs can be employed as viable tools for diagnostic and prognostic assessment of disease activity.
The present exploratory study suggests that, in NEVs derived from AD samples, PG levels were significantly decreased compared to controls.In NEVs, dhCer levels were significantly diminished compared to AEVs, as well as DG, GM3 and LPI levels compared to multi-origin EVs.On the other hand, PE and PEp were increased in NEVs compared to multi-origin EV samples.Moreover, in AEVs, raised levels of PS were observed in AD compared to control samples, while a decrease in PG, LPC and LPE and an enhancement of MG were found in AEVs derived from AD samples compared to multi-origin EVs.Finally, in multi-origin EVs from AD compared to control samples, PA and PS were significantly increased.

Figure 5 .
Figure 5. Multivariate analysis was conducted on individual lipid molecules that were detected via MS in EV samples isolated with a pan-Tetraspanin IP.Crosses indicate the center of the subjectgroup cluster, with the vertical and horizontal components representing the standard deviation of a group along the corresponding PC (a-c).Asterisks indicate a significant difference (p < 0.05) between AD and control groups along a PC.Black lines represent the magnitude and direction of the pooled lipid species' relative contribution to any separation between AD and control clusters.The top 10 most influential lipid species are depicted.The 25 largest individual lipid contributions to variability along principal components 1, 2 and 3 are depicted through loadings plots (d-f).

Figure 5 .
Figure 5. Multivariate analysis was conducted on individual lipid molecules that were detected via MS in EV samples isolated with a pan-Tetraspanin IP.Crosses indicate the center of the subject-group with the vertical and horizontal components representing the standard deviation of a group along the corresponding PC (a-c).Asterisks indicate a significant difference (p < 0.05) between AD and control groups along a PC.Black lines represent the magnitude and direction of the pooled lipid species' relative contribution to any separation between AD and control clusters.The top 10 most influential lipid species are depicted.The 25 largest individual lipid contributions to variability along principal components 1, 2 and 3 are depicted through loadings plots (d-f).

Figure 6 .
Figure 6.Multivariate analysis was conducted on individual lipid molecules that were detected via MS in EV samples isolated with a GLAST IP.Crosses indicate the center of the subject-group cluster, with the vertical and horizontal components representing the standard deviation of a group along the corresponding PC (a-c).Asterisks indicate a significant difference (p < 0.05) between AD and control groups along a PC.Black lines represent the magnitude and direction of the pooled lipid species' relative contribution to any separation between AD and control clusters.The top 10 most influential lipid species are depicted.The 25 largest individual lipid contributions to variability along principal components 1, 2 and 5 are depicted through loadings plots (d-f).

Figure 6 .
Figure 6.Multivariate analysis was conducted on individual lipid molecules that were detected via MS in EV samples isolated with a GLAST IP.Crosses indicate the center of the subject-group cluster, with the vertical and horizontal components representing the standard deviation of a group along the corresponding PC (a-c).Asterisks indicate a significant difference (p < 0.05) between AD and control groups along a PC.Black lines represent the magnitude and direction of the pooled lipid species' relative contribution to any separation between AD and control clusters.The top 10 most influential lipid species are depicted.The 25 largest individual lipid contributions to variability along principal components 1, 2 and 5 are depicted through loadings plots (d-f).

Figure 7 .
Figure 7. Multivariate analysis was conducted on individual lipid molecules that were detected via MS in EV samples isolated with an L1CAM IP.Crosses indicate the center of the subject-group cluster, with the vertical and horizontal components representing the standard deviation of a group along the corresponding PC (a-c).Black lines represent the magnitude and direction of the pooled lipid species' relative contribution to any separation between AD and control clusters.The top 10 most influential lipid species are depicted.The 25 largest individual lipid contributions to variability along principal components 1, 2 and 3 are depicted through loadings plots (d-f).

Figure 7 .
Figure 7. Multivariate analysis was conducted on individual lipid molecules that were detected via MS in EV samples isolated with an L1CAM IP.Crosses indicate the center of the subject-group cluster, with the vertical and horizontal components representing the standard deviation of a group along the corresponding PC (a-c).Black lines represent the magnitude and direction of the pooled lipid species' relative contribution to any separation between AD and control clusters.The top 10 most influential lipid species are depicted.The 25 largest individual lipid contributions to variability along principal components 1, 2 and 3 are depicted through loadings plots (d-f).Cells 2024, 13, x FOR PEER REVIEW 16 of 23

Figure 8 .
Figure 8. Volcano plots describe the magnitude and significance of differences in individual lipid concentrations between AD and control EV samples.EVs were isolated with pan-Tetraspanin (a), GLAST (b) or L1CAM (c) IP. Green points above the horizontal red line depict lipids with concentrations significantly different (p < 0.05) between AD and control samples.Black diamonds above the horizontal red line depict lipids with concentrations significantly different (p < 0.05) after multiple-testing (Bonferroni) correction between AD and control samples.Positive values on the horizontal axis indicate enrichment in AD samples and vice versa.

Figure 8 .
Figure 8. Volcano plots describe the magnitude and significance of differences in individual lipid concentrations between AD and control EV samples.EVs were isolated with pan-Tetraspanin (a), GLAST (b) or L1CAM (c) IP. Green points above the horizontal red line depict lipids with concentrations significantly different (p < 0.05) between AD and control samples.Black diamonds above the horizontal red line depict lipids with concentrations significantly different (p < 0.05) after multiple-testing (Bonferroni) correction between AD and control samples.Positive values on the horizontal axis indicate enrichment in AD samples and vice versa.

Figure 9 .
Figure 9. Multivariate analysis was conducted on individual lipid molecules that were detected via MS in EV samples from healthy human subjects.Crosses indicate the center of an EV subset cluster, with the vertical and horizontal components representing the standard deviation of a group along the corresponding PC (a-c).* indicates a significant difference (p < 0.05) between Tetraspanin IP EVs and L1CAM IP EVs along a PC.† indicates a significant difference (p < 0.05) between Tetraspanin IP EVs and GLAST IP EVs along a PC.Black lines represent the magnitude and direction of the pooled lipid species' relative contribution to any separation between pan-EV (Tetraspanin IP EVs) and brain-associated EVs (L1CAM and GLAST IP EVs).The top 10 most influential lipid species are depicted.The 25 largest individual lipid contributions to variability along principal components 1, 2 and 4 are depicted through loadings plots (d-f).

Figure 9 .
Figure 9. Multivariate analysis was conducted on individual lipid molecules that were detected via MS in EV samples from healthy human subjects.Crosses indicate the center of an EV subset cluster, with the vertical and horizontal components representing the standard deviation of a group along the corresponding PC (a-c).* indicates a significant difference (p < 0.05) between Tetraspanin IP EVs and L1CAM IP EVs along a PC.† indicates a significant difference (p < 0.05) between Tetraspanin IP EVs and GLAST IP EVs along a PC.Black lines represent the magnitude and direction of the pooled lipid species' relative contribution to any separation between pan-EV (Tetraspanin IP EVs) and brain-associated EVs (L1CAM and GLAST IP EVs).The top 10 most influential lipid species are depicted.The 25 largest individual lipid contributions to variability along principal components 1, 2 and 4 are depicted through loadings plots (d-f).

Cells 2024 ,
13,  x FOR PEER REVIEW 18 of 23 between EV subtypes.Overall, both univariate differences and holistic multivariate analysis show similarities between AEVs and NEVs that are not present in multi-origin EVs.

Figure 10 .
Figure 10.Volcano plots describe the magnitude and significance of differences in individual lipid concentrations between EV subsets in healthy control subjects (L1CAM vs Tetraspanin (a), GLAST vs Tetraspanin (b), L1CAM vs GLAST (c)).EVs were isolated with L1CAM, GLAST or pan-Tetraspanin IP. Green points above the horizontal red line depict lipids with concentrations significantly different (p < 0.05) between AD and control samples.Black diamonds above the horizontal red line depict lipids with concentrations significantly different (p < 0.05) after multiple-testing (Bonferroni) correction between AD and control samples.

Figure 10 .
Figure 10.Volcano plots describe the magnitude and significance of differences in individual lipid concentrations between EV subsets in healthy control subjects (L1CAM vs. Tetraspanin (a), GLAST vs. Tetraspanin (b), L1CAM vs. GLAST (c)).EVs were isolated with L1CAM, GLAST or pan-Tetraspanin IP. Green points above the horizontal red line depict lipids with concentrations significantly different (p < 0.05) between AD and control samples.Black diamonds above the horizontal red line depict lipids with concentrations significantly different (p < 0.05) after multiple-testing (Bonferroni) correction between AD and control samples.

Figure 11 .
Figure 11.Multivariate analysis was conducted on individual lipid molecules that were detected via MS in EV samples from AD patients.Crosses indicate the center of an EV subset cluster, with the vertical and horizontal components representing the standard deviation of a group along the corresponding PC (a-c).* indicates a significant difference (p < 0.05) between Tetraspanin IP EVs and L1CAM IP EVs along a PC.† indicates a significant difference (p < 0.05) between Tetraspanin IP EVs and GLAST IP EVs along a PC.Black lines represent the magnitude and direction of the pooled lipid species' relative contribution to any separation between pan-EV (Tetraspanin IP EVs) and brain-associated EVs (L1CAM and GLAST IP EVs).The top 10 most influential lipid species are depicted.The 25 largest individual lipid contributions to variability along principal components 1, 3 and 4 are depicted through loadings plots (d-f).

Figure 11 .
Figure 11.Multivariate analysis was conducted on individual lipid molecules that were detected via MS in EV samples from AD patients.Crosses indicate the center of an EV subset cluster, with the vertical and horizontal components representing the standard deviation of a group along the corresponding PC (a-c).* indicates a significant difference (p < 0.05) between Tetraspanin IP EVs and L1CAM IP EVs along a PC.† indicates a significant difference (p < 0.05) between Tetraspanin IP EVs and GLAST IP EVs along a PC.Black lines represent the magnitude and direction of the pooled lipid species' relative contribution to any separation between pan-EV (Tetraspanin IP EVs) and brain-associated EVs (L1CAM and GLAST IP EVs).The top 10 most influential lipid species are depicted.The 25 largest individual lipid contributions to variability along principal components 1, 3 and 4 are depicted through loadings plots (d-f).

Figure 12 .
Figure 12.Volcano plots describe the magnitude and significance of differences in individual lipid concentrations between EV subsets in AD patients (L1CAM vs Tetraspanin (a), GLAST vs Tetraspanin (b), L1CAM vs GLAST (c)).EVs were isolated with L1CAM, GLAST or pan-Tetraspanin IP. Green points above the horizontal red line depict lipids with concentrations significantly different (p < 0.05) between AD and control samples.Black diamonds above the horizontal red line depict lipids with concentrations significantly different (p < 0.05) after multiple-testing (Bonferroni) correction between AD and control samples.

Figure 12 .
Figure 12.Volcano plots describe the magnitude and significance of differences in individual lipid concentrations between EV subsets in AD patients (L1CAM vs. Tetraspanin (a), GLAST vs. Tetraspanin (b), L1CAM vs. GLAST (c)).EVs were isolated with L1CAM, GLAST or pan-Tetraspanin IP. Green points above the horizontal red line depict lipids with concentrations significantly different (p < 0.05) between AD and control samples.Black diamonds above the horizontal red line depict lipids with concentrations significantly different (p < 0.05) after multiple-testing (Bonferroni) correction between AD and control samples.

Table 1 .
Relative quantitative percentage (%T-lip) of the average of main lipid classes (sterol lipids,