Optimized AF4 combined with density cushion ultracentrifugation enables profiling of high‐purity human blood extracellular vesicles

Abstract

Extracellular vesicles (EVs) have emerged as a promising tool for clinical liquid biopsy. However, the identification of EVs derived from blood samples is hindered by the presence of abundant plasma proteins, which impairs the downstream biochemical analysis of EV‐associated proteins and nucleic acids. Here, we employed optimized asymmetric flow field‐flow fractionation (AF4) combined with density cushion ultracentrifugation (UC) to obtain high‐purity and intact EVs with very low lipoprotein contamination from human plasma and serum. Further proteomic analysis revealed more than 1000 EV‐associated proteins, a large proportion of which has not been previously reported. Specifically, we found that cell‐line‐derived EV markers are incompatible with the identification of plasma‐EVs and proposed that the proteins MYCT1, TSPAN14, MPIG6B and MYADM, as well as the traditional EV markers CD63 and CD147, are plasma‐EV markers. Benefiting from the high‐purity of EVs, we conducted comprehensive miRNA profiling of plasma EVs and nanosized particles (NPs), as well as compared plasma‐ and serum‐derived EVs, which provides a valuable resource for the EV research community. Overall, our findings provide a comprehensive assessment of human blood EVs as a basis for clinical biopsy applications.

1. INTRODUCTION

EVs have gained increased attention because of their complex cargoes, including proteins, lipids, metabolites and nucleic acids such as DNA, messenger (m) RNA and non‐coding (nc) RNAs (Tkach & Théry, 2016), and their ability to mediate cell–cell communication under physiological or pathological conditions (Becker et al., 2016; van Niel et al., 2018; Wei et al., 2021). Notably, nearly all body fluids contain EVs, which are suitable for potential diagnostic tools or predictors of disease progression (De Wever & Hendrix, 2019; Huang et al., 2022; Lucotti et al., 2022; Melo et al., 2015). Plasma and serum are attractive sources of EV‐based markers because blood specimen acquisition is a minimally invasive procedure (Zhou et al., 2020). Therefore, it is necessary to recover the maximum number of vesicles from human blood, while preserving their structure and purity. Although there are several isolation techniques, including UC, density gradient centrifugation, size‐exclusion chromatography (SEC), affinity purification and polymeric precipitation (Bojmar et al., 2021; Kalra et al., 2013; Li et al., 2017; Yang et al., 2020), these procedures typically have drawbacks, particularly for human plasma or serum, where non‐EV contaminants such as albumin, immunoglobulins, lipoproteins, fibrinogen and other protein aggregates are more abundant than EVs, limiting the ability to identify low‐abundance signalling proteins (Dong et al., 2020; Simonsen, 2017; Welsh et al., 2024; Yuana et al., 2014). Therefore, it is necessary to integrate different strategies to improve the purity of EVs from plasma and serum (Karimi et al., 2018).

AF4, an open‐channel separation technique, has attracted increased attention because it is highly compatible with the efficient separation and characterization of nanoparticles, including viruses, microprotein complexes and polymeric particles (Roda et al., 2009). AF4 separates macromolecules and particles according to their diffusion coefficients using two perpendicular liquid flows, namely channel flow and cross‐flow, where the analytes pass through the channel with a parabolic flow and against with a perpendicular cross‐flow (Contado, 2017). Since separation occurs in empty and unobstructed channels, AF4 has superior resolution advantages for isolating macromolecules or particles over a broad size range; thus, making it suitable for EV separation. Accordingly, the AF4 technique coupled with different detectors was used to fractionate EVs harvested from cell cultures for further analysis (Kang et al., 2008; Sitar et al., 2015; Zhang & Lyden, 2019), and two EV sub‐populations, including large and small EVs, and an abundant population of non‐EV particles (exomeres) were identified by the AF4 approach (Zhang et al., 2018). Recently, AF4 was introduced to isolate EVs from human blood (Kim et al., 2020; Wu et al., 2020). However, the pre‐treatment of the sample and setting of the separation parameters results in different separation efficiencies, and separating high‐purity EVs from blood remains a challenge.

Here, we provided an optimized AF4 set‐up coupled with density cushion ultracentrifugation (UC) for the isolation of EVs from human blood with low lipoprotein particle contamination compared to SEC coupled with the same density cushion UC. Due to the high‐purity of EVs, we identified traditional EV markers and a group of EV‐associated membrane proteins, such as Tetraspanin family (Tetraspanin‐14, CD9, CD36, CD47, CD63 and CD151), transport proteins (SLC44A1, SLC29A1 and SLC43A3), Annexin family (Annexin A5, A6 and A7) and other membrane proteins, a large portion of which have not been identified before. Specifically, our results support the use of MYCT1, TSPAN14, MPIG6B, MYADM and the traditional EV proteins CD63 and CD147 as plasma‐EV markers. Further fractionation of EVs emphasized the importance of the actin‐filament process in the biogenesis and trafficking of EVs. Moreover, we provided comprehensive miRNA expression profiles of high‐purity plasma‐EVs and NPs. In addition, plasma‐EVs contain more membrane proteins than serum‐EVs, indicating that plasma‐EVs are more suitable for clinical application.

2. MATERIALS AND METHODS

2.1. Blood sample preparation

Blood samples were collected with the ethics permission from the First Affiliated Hospital of Guangzhou Medical University (No. 2020‐KY‐076), and informed consent was given by all donors. Nineteen healthy donors (age 23–43 years, six males and 13 females) were included (details for Dataset S1), and fasting blood collection was conducted in First Affiliated Hospital of Guangzhou Medical University. Peripheral blood was collected with a 21G butterfly needle (10 mL/person) into a BD Vacutainer blood collection tube (#367863 for plasma [EDTA anticoagulant], #368774 for serum, BD Company, USA). Blood samples were kept at room temperature (RT) and processed within 2 h after collection. The first tube drawn was discarded. The samples were pre‐centrifuged at 500 × g for 10 min at RT, further centrifuged at 1500 × g for 15 min at RT to obtain clear plasma or serum and stored as 0.5‐mL aliquots at −80°C.

2.2. Isolation of blood EVs based on AF4

The asymmetric‐flow field‐flow fractionation combined with Multi‐Angle Light Scattering (AF4‐MALS) system (Wyatt Technology, USA) was applied downstream of the Agilent quaternary pump and autosampler (Agilent, USA). The 152 mm separated channel was assembled with a 10‐kDa regenerated cellulose (RC10) membrane. In general, 250‐µL plasma or serum samples were diluted to 1.25 mL with PBS and 250 µL 20% glycerol cushion (in PBS) was added to the bottom of the centrifuge tube with a long 1‐mL syringe before being loaded into the TLS‐55 rotor. Then, the samples were centrifuged at 150,000 × g for 1.5 h at 4°C to obtain EV‐containing pellets (Optima MAX‐XP, Beckman Coulter, USA). The pellets were then fully re‐suspended in 100‐µL PBS. The insoluble components were removed by 3000 × g centrifugation for 10 min before injection into the AF4 system, where the degassing PBS was the equilibrium and elution buffer. The detailed flow rates and parameters used in this study could be found in the supplemental material (Table S1). Four major peaks (P1, P2, P3 and P4) were observed, and EVs were found in the P4 fraction. The fractions were transferred to a 50‐kDa ultrafiltration tube (Merck Millipore) and centrifuged at 3000 × g at 4°C to 100 µL for subsequent analysis. The data were recorded and processed by ASTRA software with UV + RI + MALS pipeline, and the ‘sphere’ mode was chosen for particle analysis (ver. 6.1, Wyatt Technology, USA). We submitted all relevant data of our experiments to the EV‐TRACK knowledgebase (EV‐TRACK ID: EV240025) (EV‐TRACK Consortium et al., 2017).

2.3. Isolation of plasma EVs based on SEC

SEC was carried out in accordance with the 35‐nm qEV column manufacturer's instructions (#ICO‐35, IZON Science). The crude EVs were obtained by the same density cushion UC as the AF4 isolation protocol. The EV‐containing pellet was re‐suspended in 500‐µL PBS and applied to the top of the qEV column. Before loading the sample, the column was equilibrated with two times the bed volume of degassing PBS. The separation was performed at RT. Seven millilitres PBS were added to start the elution, and collection began immediately. A total of 14 fractions (500 µL each tube) were collected.

2.4. Mass spectrometry

For sample preparation, all P3 or P4 fractions were concentrated to 0.1 mg/mL, subjected to ultrasonic lysis (2 s‐2 s, 5 cycles), mixed with LDS (lithium dodecyl sulfate) loading buffer and separated by sodium dodecyl sulfate‐polyacrylamide gel electrophoresis (SDS‐PAGE) to remove the insoluble fraction. The resolving slices were collected and in‐gel digested.

2.4.1. In‐gel digestion

After de‐staining, reduction (10‐mM DTT in 25‐mM NH₄HCO₃ for 45 min at 56°C) and alkylation (40‐mM iodoacetamide in 25‐mM NH₄HCO₃ for 45 min at RT in the dark), the gels were washed twice with 50% acetonitrile, dried using a SpeedVac and digested with trypsin in 25‐mM NH₄HCO₃ overnight at 37°C to complete digestion. The reaction was terminated by adding formic acid to a final concentration of 1%.

2.4.2. Data‐independent acquisition mass spectrometry (DIA‐MS)

All nanoLC‐MS/MS experiments were performed on an Orbitrap Exploris 480 (Thermo Scientific) equipped with an Easy n‐LC 1200 HPLC system (Thermo Scientific). The peptides were loaded onto a 100 µm id × 2 cm fused silica trap column packed in‐house with reversed phase silica (Reprosil‐Pur C18 AQ, 5 µm, Dr Maisch GmbH) and then separated on a 75‐µm id × 25 cm C18 column packed with reversed phase silica (Reprosil‐Pur C18 AQ, 1.9 µm, Dr Maisch GmbH). The peptides bound on the column were eluted with a 103‐min linear gradient. Solvent A consisted of 0.1% formic acid in water, and solvent B consisted of 80% acetonitrile and 0.1% formic acid. The segmented gradient was 4%–11% B, 4 min; 11%–21% B, 28 min; 21%–30% B, 29 min; 30%–42% B, 27 min; 42%–99% B, 5 min; 99% B, 10 min at a flow rate of 300 nL/min.

The MS data were acquired at a high resolution 120,000 (m/z 200) across the mass range of 400–1210 m/z. The target value was 4.00E + 05 with a maximum injection time of 50 ms. One full scan was followed by 40 windows with an isolation width of 16 m/z for fragmentation in the Ion Routing Multipole with HCD normalized collision energy of 30%. MS/MS spectra were acquired at a resolution 30,000 at m/z 200 across the mass range of 200–2000 m/z. The target value was 4.00E + 05 with a maximum injection time of 50 ms. For nano electrospray ion source setting, the spray voltage was 2.0 kV with no sheath gas flow, and the heated capillary temperature was 320°C.

2.4.3. For DIA data analysis

The raw DIA data from Orbitrap Eclipse were analysed using Spectronaut (ver.14 Biognosys) with the “Direct‐DIA” mode for protein identification and quantification. The UniProt human proteome database was used for searching. The most important search parameters were set to the default settings: trypsin was selected as the enzyme, and two missed cleavages were allowed; the mass tolerance of MS1 and MS2 were set as correction factor 1; cysteine carbamidomethylation was specified as fixed modifications; The methionine oxidation and acetylation of the protein N‐term were chosen as variable modifications. FDR < 1% was used for PSM, peptides and proteins identification. The data were filtered by the Qvalue, and “Global Normalization” was set as “Median” with enabled cross run normalization.

2.5. Nanoparticle tracking analysis (NTA)

NTA was performed using NanoSight NS300 with a green laser (Malvern Panalytical, UK) to examine the size distribution and particle density of the EVs. Briefly, the samples were diluted with PBS and loaded into the sample carrier cell. All the samples were buffer exchanged to PBS and measured at 23.5°C. The camera type used was sCMOS, the camera level was 13 and a total of 749 frames were captured. All other settings were set to automatic. The image data were analyzed by NanoSight NTA software (ver. 3.4) with three replicates.

2.6. Transmission electron microscopy (TEM)

The morphology of the indicated fractions was observed by TEM using negative staining. Briefly, peak fractions (P1–P4) isolated from AF4 were pooled and concentrated to 500 µL, respectively. Three microlitres liquid sample were applied to a pre‐clean carbon film‐coated round‐hole copper grid (#BZ110223b, Beijing Zhongjingkeyi Technology) and incubated for 1 min at RT. Then, the liquid was removed with filter paper (#1004110, Whatman), and the sample was stained with 2% uranyl acetate (Cat# 22451, Electron Microscopy Sciences) for 1 min. The uranyl acetate was removed with filter paper, and the grid should be dry completely before imaging. Imaging was performed using a Tecnai Spirit electron microscope (FEI, Hillsboro, OR, USA) operated at 120 kV with 49,000× magnification.

2.7. Immunoblotting and ImmunoGold labelling

All fractions with equivalent volumes were prepared with 4x LDS loading buffer (#M00676, GeneScript). After being treated at 70°ll fract min, the samples were separated via homemade 12% SDS‐PAGE and then transferred for 1.5 h onto a 0.2‐µm polyvinylidene difluoride membrane (ISEQ00010, Merck Millipore). The membranes were blocked with 5% milk in Tris‐buffered saline with 0.1% Tween (TBST) and then incubated with primary antibodies in 3% milk in TBST at 4°C overnight. The following antibodies were used: apolipoprotein A1 (66206‐1‐Ig, Proteintech, 1:5000), syntenin‐1 (ab133267; Abcam, 1:1000), TSPAN14 (15314‐1‐AP; Proteintech, 1:2000), MYCT1 (22004‐1‐AP; Proteintech, 1:2000), CD147 (11989‐1‐AP; Proteintech, 1:1000), Alix (12422‐1‐AP, Proteintech, 1:2000), CD9 (ab263019, Abcam, 1:1000), apolipoprotein B (sc‐393636, Santa Cruz, 1:200) and Albumin (66051‐1‐IG, Proteintech, 1:5000). Following incubation, the membranes were washed three times for 15 min with TBST and incubated with anti‐rabbit HRP‐conjugated or anti‐mouse HRP‐conjugated recombinant secondary Antibodies (RGAR001 and RGAM001, Proteintech; 1:5000) for 1 h at RT in 3% milk in TBST. After the membranes were washed three times again for 15 min with TBST, the immunolabels were visualized using enhanced chemiluminescence (iBright FL1500 imaging system, Thermo scientific).

For immunogold labelling, 1.4‐nm gold particles [Fab’ goat‐anti‐mouse IgG (H+L), 2002‐0.5ML, Nanoprobes] were blocked in PBS containing 5% BSA for 1 h at 4°C. Then, 1 µg (Fab’) immunogold particles were incubated with primary antibodies [anti‐APOB (sc‐393636, Santa Cruz, 5 µg)/anti‐APOE (ab183596, Abcam, 2 µg)] and anti‐CD9 (ab263019, Abcam, 5 µg)] for 1 h at RT. The antibody‐coupled gold particles were further incubated with the isolated fractions in a final 20‐µL volume for 4 h at 4°°C. Then, the labelled samples were imaged by 120‐kV TEM.

2.8. RNA isolation, small RNA sequencing and profiling

EVs from plasma were isolated with PBS and re‐suspended in TRIzol (#15596018CN, Invitrogen, USA). RNA pellets were obtained by an isopropanol‐precipitated protocol, re‐suspended in nuclease‐free water and quantified, and the 260/280 ratio was assessed using a NanoDrop2000 (Thermo Scientific). For small RNA sequencing, the library was constructed using the TruSeq Small RNA Sample Prep Kit (Illumina, San Diego, USA) according to the manufacturer's protocol. Single‐end sequencing (50 bp) was performed on an Illumina HiSeq 2500 platform. The raw sequencing fastq files were first adapter‐cut by Trimmomomatic (Bolger et al., 2014) and subjected to quality control by FastQC. The clean reads were filtered by snRNA, snoRNA, rRNA and tRNA summarized though Rfam (Griffiths‐Jones et al., 2003) database (Ver. 14.9). All the reads were kept in length > 17 nt for downstream mapping and calculating. Human mature miRNAs were obtained from miRbase (Griffiths‐Jones et al., 2006) database and prepared by bowtie (Langmead et al., 2009) as the mapping index. Then reads were collapsed and aligned by miRDeep2 (Friedländer et al., 2012) to the human miRNA library. Reads per million (RPM) values were calculated for miRNA expression levels.

2.9. Quantification and statistical analysis

Bioinformatic analysis was performed using modified scripts based on OpenSource R tools (ver. 4.1.0). The main packages included: ‘limma’(Ritchie et al., 2015) for QC, differential analysis of proteomic data; ‘clusterProfiler’(Yu et al., 2012) for Gene Ontology (GO) functional enrichment; ‘Mfuzz’(Kumar & E Futschik, 2007) for soft clustering of expression data; ‘DESeq2’(Love et al., 2014) for QC and differential analysis of miRNA expression matrix; ‘multiMiR’(Ru et al., 2014) for miRNA target predication (type = validated); ‘WGCNA’(Langfelder & Horvath, 2008) for weighted correlation network analysis and ‘ggplot2’(Wickham, 2016) for graphics of specific data. The protein interaction network was generated by the STRING web platform (www.string‐db.org/) and further processed in Cytoscape (Shannon et al., 2003) (ver. 3.9.1). in which the ClueGO (Bindea et al., 2009) plugin generated the biological process terms.

Data availability

The MS raw data have been deposited to the ProteomeXchange Consortium via the iProx partner repository (Chen et al., 2021; Ma et al., 2019) with the dataset identifier PXD041146. The raw small RNA sequence data have been deposited in the Genome Sequence Archive (https://ngdc.cncb.ac.cn/gsa) with accession number HRA004256. The processed data were included as Datasets S2‐S5.

3. RESULTS

3.1. Optimized AF4 coupled with density cushion UC strategy yields high‐purity EVs isolated from human plasma

Accumulating protocol was described for EVs separation from the cultured cells and body fluids (Gardiner et al., 2016; Visan et al., 2022), and AF4 has provided convincing separation of EVs based on particle size (Gao et al., 2022; Kim et al., 2020; Wu et al., 2020; Zhang & Lyden, 2019); however, the combination of local instruments, detailed fractionation program and sample preparation should be optimized to support stable and efficient vesicle separation from blood. In the pilot design, 250‐µL plasma was injected into the AF4 system to evaluate the separation of particles, in which the elution procedure was optimized in four stages to improve the separation efficiency (Figure S1A and Table S1). As shown in Figure S1B, the plot of the laser 90° displayed three major peaks, P‐i, P‐iii and a continuous P‐ii. Analysis of the 42 output collections using SDS‐PAGE revealed that most plasma proteins were primarily separated into P‐i and P‐ii, which was further confirmed by negative staining (Figure S1C,D). However, P‐iii exhibited a mixture of various nanosized particles (NPs), in which immunogold labelling indicated that the dehydrated EVs were cup‐shaped, whereas the lipoprotein particles were presented in fully spherical shapes (Figure S1E,F).

To avoid the interference of irrelevant components in AF4 isolation, the plasma was pre‐treated with UC to obtain crude EVs before AF4 isolation. Moreover, 20% glycerol cushion coupled UC efficiently blocked the enrichment of soluble proteins and lipoprotein particles (Figure S1G). Subsequently, the UC pellets were fully re‐suspended, injected into the AF4 system and isolated using an optimized separation program (Figure 1a). Remarkably, the laser 90° and UV280 signals displayed four sharp peaks (Figure 1a). Immunoblotting analyses revealed that the EV markers syntenin‐1, Alix and CD9 were enriched in the P4 fraction, whereas a large proportion of albumin and lipoproteins was well excluded (Figure 1b), suggesting that the plasma pre‐treatment with density cushion UC improved AF4 isolation to obtain high‐purity plasma‐EVs with low lipoprotein contamination. Furthermore, the similar separation behaviour of independent replicates confirmed the consistency and stability of this method (Figure S1H).

FIGURE 1.

Optimized AF4 coupled with density cushion UC strategy yields high‐purity EVs from human plasma. (a) Schematic diagram of the separation strategy based on AF4 (left panel), and the representative experimental record from AF4, UV and laser 90° signals were shown here (right panel). The plasma was diluted and pre‐centrifuged with density cushion to obtain crude EVs. The pellet was then injected into the AF4 system and isolated with optimized procedures. (b) Immunoblotting analysis of the AF4 isolated peaks. Syntenin‐1, Alix and CD9 indicate EVs. UC for crude pellet before AF4 injection. (c, d) Representative negative‐stain EM images and zoom‐view of fractions (P1–P3) from AF4 isolation. Scale bar = 200 nm. (e) Schematic diagram of plasma EVs isolation by SEV. The pretreatment of plasma was the same as that for AF4, and a commercial qEV gravity SEC column (IZON) was applied. (f) Immunoblotting analysis of SEC elution with EV markers, lipoinproteins and albumin; EVs were eluted into fractions 6–11. (g) NTA analysis of EVs isolated by AF4 and SEC; three replicates were included. (h) Negative staining shows that the EVs isolated by SEC from (D) were co‐eluted with various lipoproteins, the square box indicates lipoproteins with non‐dehydrated sphere forms. Scale bar = 200 nm. (i) AF4 significantly reduced the contamination of lipoproteins in isolated EV fractions. More than 10 TEM records from (g) were used to identify EVs and lipoproteins. Mean ± sd was shown, and p‐values were calculated by unpaired two‐tailed Welch's t‐test.

Next, all fractions were morphologically examined by negative staining. Both P1 and P2 contained plasma proteins and aggregates (Figure 1c). MS analysis revealed that albumin, α‐2‐macroglobulin, haptoglobin and immunoglobulin were mainly eluted into P1 and P2 fractions (Figure S2A,B). In addition, protein complexes that function in the complement cascade, hemostasis and immune system were also enriched in P1 and P2 (Figure S2C,D). As expected, we observed plasma‐EV enrichment in the P4 fraction, which had a low protein contamination background (Figure 1d), consistent with the immunoblotting results.

To better reflect the superior purity of the AF4‐based plasma‐EVs separation, SEC column isolation was introduced as a control. The same crude EVs obtained by density cushion UC were loaded into an SEC column (CL‐2B beads) to produce 14 fractions (Figure 1e). Immunoblotting of the EV markers showed that the SEC‐EVs were eluted into fractions 6–11, while non‐EV components, such as lipoproteins and albumin, were also observed in these fractions (Figure 1f). Then, the EV samples obtained by the two different separation methods were characterized by nanoparticle tracking analysis (NTA), which showed that the SEC‐EVs had a wider range of particle distributions (20–300 nm) (Figure 1g), implying a complicated composition of the SEC‐EVs. Negative staining also indicated that various lipoprotein particles co‐eluted with the EV fraction, consistent with the immunoblotting results (Figure 1h). Further TEM quantitative analysis and immunoblotting analysis suggested that AF4 enables low‐lipoprotein co‐elution for plasma‐EV separation (Figures 1i and S2E). The yield analysis by NTA showed that the number of particles isolated by AF4 was lower than that isolated by SEC (Table S2). However, NTA was unable to distinguish lipoproteins and non‐vesicular particles from vesicles. Considering the contamination of lipoproteins and non‐vesicle particles, the yield of AF4 was comparable to that of SEC methods, suggesting that EVs were enriched without bias from plasma by our improved AF4 isolation.

3.2. Proteomic characterization of plasma‐derived EVs and NPs

The high‐purity EVs obtained using optimized AF4 coupled with density cushion UC are compatible with high sensitivity protein and miRNA downstream analyses. Thus, two groups of NPs (P3) and EVs (P4) isolated from 19 healthy middle‐aged adults (Dataset S1) using the protocol described above were subjected to analysis by DIA‐MS (Figure 2a). Principal component analysis (PCA) revealed that the components of the EV fraction were highly distinct from those of the NP fraction, as also reflected by their sizes (Figures 2b and S3A). A total of 1040 proteins and 1002 proteins were identified in P3 and P4 (FDR < 1%), respectively, of which 955 proteins overlapped (Figure S3B and Dataset S2).

FIGURE 2.

Proteomic characterization of plasma‐EVs. (a) Schematic representation of the process of obtaining NP and EV fractions from 19 human plasma samples by optimized UC‐AF4 isolation. (b) The average diameter of the NP and EV fractions was measured by AF4 software (ASTRA), and the p‐value was calculated by unpaired two‐tailed t‐test. (c) Heatmap showing the identification frequency of traditional EV markers in NPs and EVs. The frequency (%) of the specified protein is shown in each box. (d) Heatmap showing the expression of the specified proteins between NPs and EVs. The expression levels are presented as the log2‐transformed LFQ intensity. (e) Volcano plot showing DE proteins of EVs versus NPs; the upregulated DE proteins are shown in blue, and downregulated DE proteins are shown in grey. P‐values were adjusted (padj) by the Benjamini–Hochberg method. (f) Heatmap showing DE protein expression in (e) between NPs and EVs. Annexin, tetraspanin and transporter proteins were enriched in plasma‐EVs. The expression levels are presented as the log2‐transformed LFQ intensity.

We first investigated the frequency of traditional EV markers identified in the P3 and P4 fractions. Alix, syntenin‐1, CD63, FLOT1 and FLOT2 were identified in most P4 fractions rather than in the P3 fractions, further confirming the efficiency of EVs isolation using AF4 (Figure 2c). However, a previous study showed that these traditional EV markers are rarely identified in plasma extracellular vesicles and particles (EVPs) isolated by sequential UC (Hoshino et al., 2020), indicating that the low plasma lipoproteins contamination in this study aided in the identification of traditional EV markers. Furthermore, 11 of the 13 previously reported human plasma EVP markers (FN1, LGALS3BP, A2M, JCHAIN, HBB, GSN, ACTB, B2M, STOM, MSN, PRDX2, RAP1B and FLNA) were identified in >80% of both the P3 and P4 samples (Hoshino et al., 2020) (Figure S3C), indicating the specificity of the AF4 method. For CD9, CD63 and CD81, the identification frequencies of CD9, CD63 and CD81 were 100, 96 and 21%, respectively. Consistently, the highest abundance score was for CD9, followed by CD63 and CD81 in plasma‐EVs (Figure S3E), consistent with previous reports of a higher abundance of CD9⁺ EVs than of CD63⁺ EVs in plasma (Rydland et al., 2023; Saftics et al., 2023). CD81 was identified in only 38% of the P4 fractions, consistent with a previous report that CD81⁺ vesicle is a rare sub‐population in plasma (Karimi et al., 2022). These results suggest that EVs are enriched by the optimized AF4 combined with the density cushion UC.

In addition, 22 proteins were identified as universal cell‐derived EV markers, including CLTC, LGALS3BP, GAPDH, RPS27A, RAB10, RAP1B, ITGB1, SDCBP, EEF2, GNAI2, CD147, GNB2, CD47, PDCD6IP SLC3A2, ATP1A1, RAN, SLC1A5, GNAI3, GNB1, TSG101 and RRAS, irrespective of the cellular source or isolation method used (Kugeratski et al., 2021). We analyzed our dataset for these EV proteins and found that CLTC, LGALS3BP, GAPDH, RPS27A, RAB10 and RAP1B were identified in almost all NP and EV fractions (Figure S3D). However, ITGB1, SDCBP, EEF2, GNAI2, CD147, GNB2, CD47 and PDCD6IP were identified in most EV samples instead of NP samples (Figure S3D), suggesting that these proteins are also plasma‐EV‐associated proteins. Intriguingly, TSG101 and RRAS were completely undetectable in both NP and EV samples (Figure S3D), which is consistent with a previous study, as it was also completely unidentified in 120 plasma EVP samples (Hoshino et al., 2020), suggesting that TSG101 and RRAS are cell line‐derived EV markers, not plasma‐EV markers. Taken together, these results suggest that cell line‐derived EV markers are incompatible with plasma‐EV identification.

3.3. Identification of human plasma‐EV markers

Next, the overall protein identification frequencies were assessed to identify the plasma‐EV markers. We filtered our dataset with the following criteria: totally unidentified in NP samples but identified in more than 80% of EV samples or totally identified in EV samples but identified in <20% of NP samples. Six proteins met the criteria, including the traditional EV markers CD63 and CD147 and four proteins, MYCT1, TSPAN14, MPIG6B and MYADM (Figure 2d), which were ranked in the top 50% of all identified EV proteins according to intensity‐based absolute protein quantification (iBAQ) values (Figure S3E). The expressions of CD147, TSPAN14 and MYCT1 in the P4 fraction were further confirmed by immunoblotting (Figure S3F). Taken together, these results suggest that the MYCT1, TSPAN14, MPIG6B, MYADM, and the traditional EV markers CD63 and CD147, are plasma‐EV markers.

Differential expression (DE) analysis was performed based on EV samples versus NP samples, where NPs were a suitable negative control for the characterization of EVs. In total, 196 DE proteins were identified, of which 128 proteins were up‐regulated and 68 proteins were down‐regulated (|log2 − fold change| > 1 and adjusted p‐value < 0.05) (Figure 2e). The top up‐regulated proteins were CD63, MYCT1, TSPAN14, MPIG6B, CD147 and MYADM (Figure 2e), which was consistent with the above findings, further suggesting that these proteins are plasma‐EV markers. In addition, a group of membrane proteins, including tetraspanin family proteins (tetraspanin‐14, CD9, CD36, CD47, CD63, and CD151), transport proteins (SLC44A1, SLC29A1 and SLC43A3) and the Annexin A5, A6 and A7, were identified in the up‐regulated subset (Figure 2f). GO analysis revealed that the upregulated proteins were involved in the biological processes of phagocytosis and vesicle‐mediated transport (Figure S3G). In contrast, the down‐regulated proteins were involved in the humoural immune response and complement activation. Consistently, the iBAQ values of the NPs showed that fibrinogens, immunoglobulins and α‐2‐macroglobulin were ranked in the top‐10 proteins (Figure S3H). Further, several secreted proteins, including MAP4, FBN1 and SAA2, were classified as NP‐associated proteins compared to plasma‐EVs (Figure S3I).

3.4. Fractionation of P4 promotes the valid identification of EV‐associated proteins

The flexible cross‐flow setting of the AF4 separation produced a 6‐mL P4 fraction per injection, enabling further division of the P4 collection into six fractions (F1–F6) (Figure 3a). Negative staining demonstrated that the vesicles were mainly enriched in F5, and some large vesicles were observed in F6 (Figure 3b). Both F1 and F2 contained particles with diameters <20 nm, whereas F3 and F4 contained a mixture of vesicles and protein particles (Figure 3b). In general, the average diameters of the particles and vesicles increased with flow time; however, the highest number density was observed in F5 (Figure 3c).

FIGURE 3.

Fractionations of P4 reveal actin filament enrichment in plasma‐EVs. (a) Schematic representation of the fractionation of P4 into six aliquots, namely, F1–F6. (b) Representative negative‐staining TEM images of fractions F1–F6 from the P4 fraction in (a). Scale bar = 200 nm. (c) Diameter and number density distributions of vesicles from F3 to F6 measured by AF4 software (ASTRA). (d) Soft clustering of the normalized protein abundance of F1–F6 with Mfuzz R package (fuzzy c‐means algorithm) to the identification of four clusters and GO biological process enrichment of proteins in each cluster by ClueGO. (e) Heatmap showing EV markers and EV‐associated protein expression across the F1–F6 fractions. Expression levels are presented as the log2‐transformed LFQ intensity. (f) Protein interaction network was generated using the STRING database, and the nodes are coloured according to the indicated biological processes. The interaction score was set as high confidence (0.7).

Next, the F1–F6 were subjected to MS analysis (Dataset. S3). We first performed time‐course analysis by c‐means‐based soft clustering to generate four protein expression patterns, including sequential‐up proteins (cluster #3), sequential‐down proteins (cluster #2), peak expression proteins in F4 (cluster #1) and peak expression proteins in F5 (cluster #4) (Figure 3d). Subsequent GO analysis indicated that the proteins in cluster #4 were associated with the exosomal secretion. As expected, traditional EV markers, such as CD9 and SDCBP, and the plasma‐EV markers MYCT1, TSPAN14, MPIG6B, CD147 and MYADM were classified into cluster #4 (Figure 3e). In addition, tetraspanins, annexins, transporter proteins and several membrane‐trafficking elements, including syntaxin‐7, were also enriched in cluster #4 (Figure 3e). Strikingly, the analysis of cluster #4 using the STRING database showed that the interaction complexes were significantly enriched in regulated exocytosis, vesicle‐mediated transport and actin filament‐based processes (Figure 3f), where ACTA1 and ITGB1 established hubs for core interactions, emphasizing the role of the cytoskeleton in plasma‐EVs.

3.5. Systemic analysis of plasma EV‐ and NP‐associated miRNA profiles

As carriers of intercellular communication, EVs contain various types of nucleic acids, of which miRNAs are important components (Tkach & Théry, 2016; Xu et al., 2022). Subsequently, small RNA libraries from plasma‐EVs and NPs for next‐generation sequencing were generated. In total, 384 and 272 known mature human miRNAs were identified in the EV and NP fractions, respectively, of which 197 miRNAs overlapped (Figure S4A and Dataset. S4). PCA revealed that EV‐associated miRNAs were highly distinct from those of NPs, and this difference was also reflected in their protein compositions (Figure S4B). The three most abundant miRNAs in EVs were hsa‐miR‐451a, hsa‐miR‐21‐5p and hsa‐let‐7a‐5p (Figure 4a), which are consistent with the EV‐associated miRNA profiles isolated by other methods (Sun et al., 2023). The most abundant miRNAs in NP samples were highly distinct from those in EV samples (Figure S4C). Furthermore, plasma‐EV miRNAs were generalized enrichment of 5p, and the AAG[U/C]AA motif was identified (Figure 4a); however, no significant motif was observed in NP‐associated miRNAs, indicating that the specific motif of miRNAs may be involved in their sorting into EVs (Garcia‐Martin et al., 2022). GO analysis revealed that EV‐associated miRNAs function in RNA processing and apoptotic signalling (Figure 4b). Using the filtering criteria (|log2 − fold change| > 1 and adjusted p‐value < 0.05), 56 unique DE miRNAs were identified in the comparison of EVs to NPs (Figure 4c), of which 28 miRNAs were up‐regulated, including hsa‐miR‐191‐5p, hsa‐miR‐217‐5p, hsa‐let‐7 g‐5p and hsa‐miR‐144‐5p (Figure 4d). GO enrichment of the target genes of the DE miRNAs revealed that the up‐regulated DE miRNAs were associated with the cell cycle and DNA checkpoint signalling, while the down‐regulated DE targets were associated with kinase activity and the apoptotic pathway (Figure 4e). Overall, we provided comprehensive miRNA expression profiles of plasma‐EVs.

FIGURE 4.

miRNA expression profiles in plasma‐EVs. (a) Rank abundance (reads per million/RPM value, log10 transformed) of miRNAs in plasma‐EVs; zoom‐view of the top 20 items. Motif was identified by the Multiple Em for Motif Elicitation (MEME) suite. (b) The target gene GO biological process enrichment of miRNAs in (a); the gene counts are labelled in each column. p‐values were adjusted (padj) by Benjamini–Hochberg method. (c) Volcano plot showing the distribution of DE proteins of the P4 versus P3 fractions; the up‐DE miRNAs are shown in yellow, and the down‐DE miRNAs are shown in blue. p‐values were adjusted (padj) by Benjamini–Hochberg method. (d) Heatmap showing the expression of the top DE miRNAs in (c). (e) The enrichment of the GO biological process of the target genes of the DE miRNAs in (c); the enrichment of the upregulated DE miRNAs is shown on the left, and the enrichment of the downregulated DE miRNAs is shown on the right. The gene counts are labelled in each column. p‐values were adjusted (padj) by Benjamini–Hochberg method.

3.6. Proteomics comparison of plasma‐ and serum‐derived EVs

To explore the potential differences between EVs derived from human plasma and serum, the same volumes of plasma and serum from one individual were used to isolate EVs using AF4 with established parameters. The results showed that the separation behaviour, morphology, number and radius of the EVs did not significantly change (Figure 5a–d).

FIGURE 5.

Differences between plasma‐ and serum‐derived EVs. (a) The isolation behaviors of plasma and serum using AF4 and the geometric radius analysis of their P4 fractions. (b) Representative negative‐stain TEM images for each peak in the plasma and serum AF4 isolation from (a) and the zoom view. Scale bar = 200 nm. (c, d) Average diameter and number density of P4 isolated from plasma and serum; mean ± sd was shown, and p‐values were calculated by unpaired two‐tailed t‐test. (e) Venn diagram showing proteins identified in plasma‐ and serum‐derived EVs. The proteins identified in all plasma or serum samples were included for analysis. (f–h) Heatmap showing the expression of the specified proteins in plasma‐ and serum‐EVs. The traditional EV markers and membrane‐associated proteins enriched in plasma‐EVs are shown in (f); the proteins specifically expressed in plasma‐ or serum‐EVs are shown in (g); the selected proteins from (g) showing their enrichment in plasma‐EVs are shown in (h). The expression levels are presented as the log2‐transformed LFQ intensity.

Further proteomic analysis revealed that the proteins identified in 100% of all samples were highly overlapped (Figure 5e and Dataset. S5). In addition, most EV markers were identified in both plasma‐EVs and serum‐EVs. However, CD81, a protein with the low identification frequency in plasma‐derived EVs, was highly expressed in serum‐derived EVs (Figure 2c). Conversely, CD63, FLOT1, MYCT1 and CD147 were highly expressed in plasma‐EVs but not in serum‐EVs (Figure 5f). Compared to the serum‐EVs, the up‐regulated proteins in plasma‐EVs were associated with blood coagulation and actin filament organization (Figure S5A,B). The top DE proteins are summarized based on the frequencies identified between the two groups (Figure 5g). For instance, APOM, KRT23 and CSTB are serum‐EV‐specific proteins, while TSPAN33, CAVIN2 and DNM1L are plasma‐EV‐specific proteins that were also highly identified in the above 19 plasma‐EVs (Figure 5h). Together, these results showed that plasma‐EVs are similar to those of serum‐EVs but exhibit more detectable membrane proteins, which emphasizes that plasma‐EVs may be more suitable for clinical applications.

4. DISCUSSION

The isolation of EVs from blood is of great importance for understanding the biological role of circulating EVs and developing EVs as disease biomarkers. However, the isolation of EVs from blood is challenging due to its complexity. The number of EVs in the blood is estimated to be at least 10³‐fold less than that of lipoproteins (Sódar et al., 2016; Yuana et al., 2014). In addition, due to the overlapping characteristics of EVs compared to lipoprotein particles and protein complexes, employing traditional strategies for separating EVs from blood, including UC, ultrafiltration and SEC, resulted in more lipoproteins being co‐isolated with EVs. Several strategies have been developed to yield high‐purity EVs from blood, such as immunocapture and sequential UC combined with other traditional separation methods (Hoshino et al., 2020; Logozzi et al., 2020). Although immunocapture using monoclonal antibodies may yield highly purified EVs, only sub‐populations of EVs can be isolated using this approach. However, subsequent protein analysis using MS revealed only a limited number of EV‐associated proteins were identified, suggesting that high‐abundance lipoproteins were still present in the samples and impeding the identification of low‐abundance EV‐associated proteins. In this study, we introduce an optimized AF4 strategy coupled with density cushion UC to obtain high‐purity EVs isolated from human blood, resulting in the identification of the maximal amount of EV‐associated proteins with minimal plasma protein contamination.

A systemically optimized parameters of the AF4 technology study were reported to separate EVs from high‐density lipoproteins (HDL) and low‐density lipoproteins (LDL) in plasma and serum with rapidity, simplicity and high reproducibility (Bria et al., 2019). However, as we observed during the initial AF4 separation (Figure 1b), after the untreated plasma was directly injected, the majority of the particles in fraction P‐iii were lipoproteins according to TEM analysis. In addition, AF4 coupled with pre‐treatment methods, including UC, ultrafiltration and commercial EV isolation kits, was reported to effectively separate EVs from human blood (Kim et al., 2020). However, the plots of the UV and laser 90° signals showed continuous peaks, indicating that the EVs were not completely separated from the lipoproteins, which was also reflected in the TEM results and the average radius of the EV‐enriched fractions. In this study, pre‐density cushion UC, such as sucrose, iodixanol or glycerol, could decrease lipoprotein contamination. The optimized parameters of sufficient pre‐focus time before separation and a four‐stage cross‐flow gradient resulted in four completely separated peaks, in which the P1 and P2 fractions were composed of plasma protein aggregates, and the P3 fraction contained some small NPs. High‐purity EVs with rare lipoprotein contamination were obtained from the P4 fraction.

High plasma protein contamination presents great challenges for the analysis and characterization of obtained EVs; therefore, their purities and isolation efficiencies must be emphasized. For example, a previous study revealed that a total of 1038 proteins were identified in seven replicates of purified EVs from pooled human plasma samples by AF4 (Wu et al., 2020). However, only a limited number of common EV proteins, such as RAB27b and RAB10, have been identified. Traditional markers, such as CD63, CD81, SDCBP, ALIX, FLOT1 and FLOT2, were not identified, suggesting that high‐purity EVs with low contamination of plasma proteins and lipoproteins facilitate the characterization of low‐abundance EV‐associated proteins. In addition, blood EVP markers from 426 human blood defined multiple human cancers (Hoshino et al., 2020), the EVP proteomic analysis identified 13 common EVP markers in humans, 11 of which were demonstrated to be highly expressed in EVs and NPs isolated by our AF4 method (Figure S3C). However, the proteomic profile of EVPs revealed that traditional EV markers, such as CD63, CD81, FLOT1, FLOT2 and Alix, were not identified in human plasma or serum EVs, further suggesting that purity is critical for the identification of EV‐associated proteins. Moreover, the data indicated that EV concepts from cultured cells are insufficient to describe blood EVs, rendering traditional EV markers inadequate for identifying blood EVs. For example, CD81 was identified in only 38% of plasma‐EV samples, and TSG101 was absent in plasma‐EVs. Based on the comprehensive proteomics dataset, we propose the use of the previously uncharacterized proteins MYCT1, TSPAN14, MPIG6B and MYADM and the traditional EV markers CD63 and CD147 as plasma‐EV markers.

The term ‘EVs’ encompasses various vesicles in the blood, including microvesicles and apoptotic body (Lötvall et al., 2014). In this study, we obtained vesicles of different sizes through fractionation of P4. Our results demonstrated that F5 fraction was enriched with vesicles ∼200 nm in diameter, and traditional EV markers and membrane‐associated proteins were also presented in the F5 fraction, suggesting that this fraction was composed of classical EVs. It is controversial whether the cytoskeleton is the cargo of classical EVs (Jeppesen et al., 2019; Zhang et al., 2018), our data show that actin and actin‐related proteins are enriched in blood classical EVs and that the role of actin filaments in EVs is traceable (Holliday et al., 2020). For instance, actin and myosin activation are required for the release of microvesicles from the ciliary tip (Nager et al., 2017). ARF6 signalling, which functions in cytoskeleton remodelling, has been shown to regulate the shedding of microvesicles (Muralidharan‐Chari et al., 2009). The large vesicles (>500‐nm diameter) in F6 are associated with blood coagulation, which is similar to the functions of microvesicles (Nomura & Shimizu, 2015). Thus, plasma‐EVs are heterogeneous.

EV‐associated miRNAs have attracted attention owing to their diverse biological functions and potential as cancer biomarkers. It is important to note that in addition to EVs, NPs, including protein aggregates and plasma lipoprotein particles, also contain miRNAs. It has been demonstrated that the miRNA expression patterns of cell line‐derived EVs are distinct from those of NPs (Zhang et al., 2021). However, characterization of blood EV‐ and NP‐associated miRNAs remains limited because of the lack of an efficient separation approach. This study revealed that EV‐associated miRNAs are highly distinct from those in NPs, where EV‐associated miRNAs play a role in RNA processing and apoptotic signalling. This study provides the comprehensive miRNA expression profiles of plasma‐EVs and NPs.

The initial purpose of isolating EVs from the plasma and serum was to evaluate the reliability of the optimized AF4 methods from different sources. Both plasma and serum‐EVs exhibited similar purification behaviours and morphological characteristics. However, serum‐derived EVs contain fewer membrane‐associated proteins, particularly those that are highly abundant in plasma‐EVs. For example, MYCT1 and CD147, which are highly related to plasma‐EVs, were lost in serum‐EVs. The heterogeneity of EV proteins between plasma‐ and serum‐EVs might be explained by the fact that the exclusion of fibrinogen during serum collection affects some biochemical features of EVs or the lack of clotting factors causes a fundamental difference between plasma and serum EVs. Our data suggest that plasma is a more attractive source for EV application instead of serum. In addition, there is no consensus regarding the platelet‐derived EVs between the plasma‐ and serum‐EVs. Previous report showed that serum‐EVs contain more platelet‐derived EVs than acid citrate dextrose (ACD) or EDTA plasma‐EVs, which were enriched by SEC (Dhondt et al., 2023; Palviainen et al., 2020). However, our study showed that complement and coagulation cascades were enriched in plasma‐EVs, consistent with the results of plasma‐EVs enriched by affinity‐capture isolation (Muraoka et al., 2022). The controversial results may be resulted from different isolation methods and we should be cautious about interpreting the changes in platelet‐derived EVs enriched by different isolation methods.

AUTHOR CONTRIBUTIONS

Liqiao Hu: Conceptualization (equal). Tao Xu and Zonghong Li: Conceptualization (equal). Liqiao Hu: Data curation (equal). Xinyue Zheng and Zonghong Li: Data curation (equal). Liqiao Hu: Formal analysis (equal). Xinyue Zheng and Zonghong Li: Formal analysis (equal). Liqiao Hu: Resources (equal). Xinyue Zheng and Ming Dong: Resources (equal). Liqiao Hu: Software (equal). Liqiao Hu: Supervision (equal). Liqiao Hu: Writing—original draft (equal). Liqiao Hu: Writing—review and editing (equal). Tao Xu and Zonghong Li: Writing—review and editing (equal). Xinyue Zheng: Investigation (equal). Maoge Zhou: Data curation (supporting). Maoge Zhou and Lingjun Tong: Resources (supporting). Jifeng Wang: Methodology (supporting). Zonghong Li: Funding acquisition (equal).

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Hu, L. , Zheng, X. , Zhou, M. , Wang, J. , Tong, L. , Dong, M. , Xu, T. , & Li, Z. (2024). Optimized AF4 combined with density cushion ultracentrifugation enables profiling of high‐purity human blood extracellular vesicles. Journal of Extracellular Vesicles, 13, e12470. 10.1002/jev2.12470