Column-based Technology for CD9-HPLC Immunoaffinity Isolation of Serum Extracellular Vesicles

Abstract

Serum-derived extracellular vesicles (EVs) are a promising source of biomarkers; however, major challenges in EV separation and proteomic profiling remain for isolating EVs from a small amount, that is, on the microliter scale, of human serum while minimizing the contamination of blood proteins and lipoprotein particles coeluting in EV preparations. Herein we have developed a column-based CD9-antibody-immobilized high-performance liquid chromatography immunoaffinity chromatography(CD9-HPLC-IAC) technology for EV isolation from a microliter scale of serum for downstream proteomic analysis. The CD9-HPLC-IAC method achieved EV isolation from 40 μL of serum in 30 min with a yield of 8.0 × 10⁹ EVs, where EVs were further processed with a postcolumn cleaning step using the 50 kDa molecular weight cut-off filter for the buffer exchange, concentration, and reduction of potentially coeluting serum proteins. In total, 482 proteins were identified in EVs by using liquid chromatography tandem mass spectrometry, including the common exosomal markers such as CD63, CD81, CD82, Alix, and TSG101. The statistical analysis of EV protein content showed that the top 10 serum proteins in EVs were significantly decreased by using the CD9-HPLC-IAC method compared with the use of ultracentrifugation (p = 0.001) and size exclusion chromatography (p = 0.009), and apolipoproteins were significantly reduced 4.8-fold compared with the SEC method (p < 0.001). The result demonstrates the potential of the CD9-HPLC-IAC method for the efficient isolation and proteomic characterization of EVs from a microscale volume of serum.

Graphical Abstract

INTRODUCTION

Serum-derived extracellular vesicles (EVs) are membrane-enclosed nanovesicles secreted by cells and released into the bloodstream.^1,2 EVs carry a complex cargo of functional proteins, microRNA (miRNA), and lipids as mediators of intercellular communication by transferring biological information between cells.^3,4 The common surface markers of EVs include CD9, CD63, and CD81, whereas individually, heterogeneous EV populations derived from different cell types can selectively load unique cargo proteins and genetic molecules from the cell of origin.^5–7 The lipid bilayer membrane of EVs protects the contents from degradation that may otherwise occur during exposure to various biofluids. They may thus serve as a promising source of biomarkers for the detection of disease progression based on minimally invasive assays.^8,9

For the isolation of EVs from serum/plasma, the conventional EV separation techniques include ultracentrifugation (UC),^10–12 polymeric precipitation,^13,14 peptide affinity,¹⁵ and size exclusion chromatography (SEC).^16,17 Recent efforts have also been devoted to the application of the conventional separation approaches to the analysis of low volumes of serum for liquid biopsy.^18–20 Brennan et al. reported a comparison study on the commonly used EV isolation techniques, including ExoQuick, SEC, UC, and density gradient centrifugation (DGC), for the isolation of EVs from 200 μL of human serum.²¹ In addition, novel approaches have been developed for EV isolation from microliter-scale human serum/plasma.^22,23 One such advance has included a polyester (PET) capillary-channeled polymer (C-CP) fiber column coupled to hydrophobic interaction chromatography (HIC), which allows the isolation of EVs from a small volume of plasma sample (20 μL) within 10 min.^24,25 This high-performance liquid chromatography (HPLC)-based method offers the advantages of a small scale, high efficiency, and real-time monitoring. Another study achieved exosome enrichment from 1 μL of serum using TiO₂ microspheres for downstream proteomic profiling with a data-independent acquisition (DIA) method.²⁶ However, these techniques isolate total/bulk EVs, whereas EV subpopulations could be important to identify signatures related to specific diseases. Major issues including the contaminating impurities coenriched with EVs, such as large EVs and blood proteins,^5,27 remain to be solved.

Alternatively, immunoaffinity extraction^28–30 using antibody-coated beads has been used for the isolation of subpopulations of EVs based on surface proteins (e.g., CD9, CD63, CD81, TSG101, Alix) recognized by antibodies. For example, Kowal et al. performed a proteomic comparison to characterize heterogeneous populations of EVs by immunoisolation EV subtypes from cell culture media using beads coated with antibodies targeting CD9, CD63, or CD81.⁶ Sharma et al. developed an immunoaffinity-based method to capture melanoma-cell-derived exosomes from the plasma of patients with melanoma using magnetic beads conjugated to anti-CSPG4 and anti-CD63 antibodies, respectively, where the EV subpopulations were evaluated by flow cytometry analysis.³⁰ Brahmer et al. performed a multiplex phenotyping analysis of immunobead-isolated EVs from human plasma using the CD9-, CD63-, or CD81-Exosome Isolation kit to examine the EV subclass dynamics.³¹ However, the bead-based immunoisolation of EVs is often coupled to flow sorting and Western blotting (WB) analyses but is incompatible with Nanoparticle Tracking Analysis (NTA) or proteomic analysis, as intact EVs cannot be efficiently eluted from the isolation beads.³¹ In the study by Kowal et al., the immunobead-isolated EVs were loaded onto sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) gel, where gel slices were excised for the subsequent proteomic analysis.⁶ An alternative immunoisolation platform that can facilitate the elution of intact EVs for downstream proteomic analysis is required.

Proteomic profiling is important to characterize the protein content in EVs for the discovery of protein signatures and biomarker candidates for disease diagnosis and prognosis. There are challenges in the proteomic profiling of serum/plasma-derived EVs, which include: (1) the low protein concentration in EVs and (2) the contaminating impurities of blood proteins and lipoprotein particles^22,32 that suppress the mass spectrometry (MS) signals of low-abundance EV proteins. Therefore, minimizing the impurities of serum-derived EVs for a higher degree of sample purity is important for the characterization of the proteomic profiling of EVs.

Herein we have developed a column-based CD9-antibody-immobilized HPLC immunoaffinity chromatography (CD9-HPLC-IAC) technology to isolate EVs from a small amount of serum for downstream proteomic analysis. This CD9-HPLCIAC approach enables the real-time isolation of EVs from microscale amounts of serum, that is, 40 μL, within 30 min. The EV fraction eluted from the CD9-IAC column retains its morphological integrity, where a postcolumn cleaning step using the 50 kDa molecular weight cut-off (MWCO) filter is employed for the efficient sample concentration, buffer exchange, and reduction of potentially coeluting serum proteins. The proteomic profiling of EVs shows that this CD9-HPLC-IAC approach greatly reduces the contamination of blood proteins and apolipoproteins compared with the UC and SEC methods. The result demonstrates that the CD9-HPLC-IAC approach coupled to the postcolumn cleaning procedure is highly efficient for the isolation and characterization of this subpopulation of EVs from a microscale volume of serum for proteomic profiling, which could have potential utilization in the clinic.

METHODS

Materials

Reagents were purchased from Sigma (St. Louis, MO) unless otherwise specified. The UltraLink hydrazide resin, Zeba spin desalting columns, Pierce centrifuge columns, radioimmuno-precipitation assay (RIPA) buffer, and chemiluminescence substrate kit were from Thermo Scientific (Rockford, IL). Sequencing-grade trypsin was from Promega (Madison, WI). The YM-30 kDa and YM-50 kDa MWCO centrifugal filters and C18 ZipTips were from Millipore (Billerica, MA). The 200 mesh Formvar/carbon-coated grid was from Electron Microscopy Sciences (Hatfield, PA). The pooled normal human serum sample was obtained from Innovative Research (Novi, MI). The monoclonal anti-CD9 antibody (no. ab92726) and horseradish peroxidase (HRP) conjugated secondary antibody were from Abcam (Cambridge, MA). The 4–20% SDS-PAGE gel was from Bio-Rad (Hercules, CA), and the ProteoSilver Plus Silver Stain Kit was from Sigma. The phosphate-buffered saline (PBS) buffer and 1 M NaCl were filtered with a 0.22 μm filter prior to use.

Experimental Design

We developed a CD9-HPLC-IAC method to enable the rapid isolation of EVs from a microscale volume of serum for downstream proteomic analysis. The workflow is summarized in Figure 1. In brief, an immobilized anti-CD9 column was developed and packed in house for EV isolation on an HPLC platform. A small amount of serum, that is, 40 μL, was loaded onto the column where the EV fraction was collected. A postpurification cleaning step using the 50 kDa MWCO filter was performed to desalt, concentrate, and reduce the coeluting serum proteins of the EV fraction. After the characterization with transmission electron microscopy (TEM), NTA, and WB, the EVs were processed for the proteomic analysis by nano liquid chromatography tandem mass spectrometry (nano-LC–MS/MS).

Figure 1.

Workflow of a CD9-HPLC-IAC approach for the isolation of EVs from a small volume of human serum (i.e., 40 μL) followed by proteomic analysis.

Anti-CD9 HPLC Column

The anti-CD9 HPLC column was developed using the UltraLink hydrazide resin conjugated with a CD9 antibody via hydrazide chemistry.³³ The selection of CD9 antibody was based on the product information from the supplier Web site, from which the application in immunoprecipitation should be included. The immobilization was performed according to the manufacturer’s instruction (Thermo). The ratio of the antibody to the hydrazide resin was ~20 μg of antibody per milliliter of resin. In brief, 100 μL of monoclonal anti-CD9 antibody was diluted with 900 μL of coupling buffer (0.1 M sodium acetate, 0.15 M sodium chloride, pH 5.5) and oxidized with sodium meta-periodate at room temperature for 30 min. The glycol groups of the sugar moieties on the Fc portion of the antibody were oxidized to generate aldehydes that can react with hydrazide groups on the resin to form stable hydrazone bonds. After buffer exchange into coupling buffer to remove excess sodium meta-periodate on a Zeba spin desalting column, the oxidized antibody was incubated with 1 mL of UltraLink hydrazide resin (average particle size 50–80 μm) on a Pierce centrifuge column at room temperature for 4 h with gentle end-over-end mixing. The antibody was coupled to the resin through the Fc portion only, leaving its antigen-binding sites unobstructed. The anti-CD9 immobilized hydrazide resin was washed sequentially with three bed volumes of coupling buffer, 1 M NaCl, and PBS with 0.05% sodium azide and then packed into a PEEK column (4.6 mm × 50 mm), where the column volume was ~0.8 mL. An underivatized column was used for the comparison where the raw UltraLink hydrazide resin was packed into a PEEK column without antibody.

CD9-HPLC-IAC Isolation of EVs from Serum

EVs were extracted from serum using the anti-CD9 column on a Beckman Coulter HPLC system with a UV absorbance detector. The HPLC isolation of EVs was achieved within 30 min. Microscale volumes of serum, that is, 10, 20, 40, 60, 80, and 100 μL, were tested. The serum sample was diluted with PBS in a 1:5 dilution ratio and then filtered through a 0.22 μm filter to remove any particulates and large EVs prior to injecting into the HPLC system. The mobile phases were PBS (buffer A) and 1 M NaCl (buffer B), with a flow rate of 0.5 mL/min. The UV absorbance was detected at 254 nm, a wavelength that has been reported for the detection of EVs.³⁴ After the filtered serum sample was injected, the column was flushed with buffer A (0–10 min), eluted with buffer B (10–19 min), and then re-equilibrated with buffer A (19–30 min). In the elution step, the binding interaction between the antigen and the antibody was broken with a high salt concentration. The eluted fraction between 16 to 19 min was collected and followed by a postcolumn cleaning procedure. The fraction delay time was caused by the volume of the tubing between the diverter valve and the UV detector.

Postcolumn Cleaning Procedure

A postcolumn cleaning procedure was performed according to the protocol established in our recent study.²² The eluted EV fraction was transferred into a 4 mL YM-50 centrifugal device and centrifuged at 3700g for 10 min followed by buffer exchange three times with PBS. The final sample volume was 100 μL. This postcolumn cleaning step functioned as buffer exchange, concentration, and further reduction of the contaminating serum proteins prior to LC–MS/MS. The EV fractions before versus after the postcolumn cleaning step, without lysis, were subjected to 4–20% SDS-PAGE followed by silver staining to evaluate the efficiency in reducing the coeluting serum proteins from the EV fraction. A total amount of 4.5 μg of protein ladder (Bio-Rad) was loaded on the gel as a reference. ImageJ was used to quantify protein bands on the gel.

Transmission Electron Microscopy

EVs were assessed by TEM using negative staining. First, 5 μL of EV sample was loaded on a 200 mesh Formvar/carbon-coated grid and incubated for 10 min followed by washing with 5 μL of water for 1 min. Then, the grid was negatively stained using 5 μL of 1% uranyl acetate for 3 min. TEM images were obtained on a JEOL 1400-plus transmission electron microscope.

NanoSight Analysis

The concentration and size distribution of EVs were measured using a NanoSight NS300 apparatus (Malvern, U.K.). Each EV sample was diluted with PBS to 1 mL and automatically infused into the NanoSight at a flow rate of 10 μL/min. The EV motion was video captured five times at 1 min each. After capture, the videos were analyzed by the built-in software, NTA, to generate the histogram of the particle size distribution and concentration.

EV Protein Extraction

EVs isolated from 40 μL of serum were dried down in a SpeedVac concentrator (Thermo) and suspended in RIPA buffer containing 25 mM Tris-HCl pH 7.6, 150 mM NaCl, 1% NP-40, 1% sodium deoxycholate, 0.1% SDS, and 1× protease inhibitor cocktail. RIPA buffer has been reported to outperform other buffers for the MS-based identification of EV proteins.³⁵ The sample was incubated at 4 °C for 30 min followed by an ice-cold sonication bath for 30 s and a gentle mix on ice for 15 min.

The amount of EV proteins was estimated by 4–20% SDS-PAGE followed by silver staining according to the manufacturer’s instruction. A total amount of 1.5 μg of protein ladder (Bio-Rad) was loaded on the gel as a reference. The gel image was analyzed using ImageJ to quantify the protein bands in EV samples.

Tryptic Digestion

The filter-aided sample preparation (FASP) method was used to perform the tryptic digestion. The EV proteins were diluted with 100 μL of 8 M urea, reduced with 10 mM 1,4-dithiothreitol (DTT) at 60 °C for 30 min, and alkylated with 25 mM iodoacetamide (IAA) for 30 min. The proteins were transferred to a YM-30 kDa centrifugal filter, which was subsequently washed with 8 M urea and 50 mM NH₄HCO₃, three times each. Then, 0.4 μg of trypsin was added and incubated at 37 °C overnight. The released peptides were collected by centrifugation and desalted with a ZipTip C18 tip and dried down in a SpeedVac concentrator for the MS analysis.

Nano-LC–MS/MS

EV peptides were dissolved in 0.1% formic acid (FA), and half of the sample was injected onto an Orbitrap Fusion Lumos Tribrid mass spectrometer (Thermo) coupled to a Dionex UltiMate 3000 ultra performance liquid chromatography (UPLC) system. Peptides were separated with 0.1% FA in water (A) and 80% CH₃CN containing 0.1% FA (B) on a 75 μm × 50 cm column (Thermo, P/N 164942) under a 90 min linear gradient from 2 to 40% B at a flow rate 300 nL/min. The MS instrument was operated in positive ion mode, and the data were acquired in a data-dependent mode. MS1 spectra (m/z 375–1700) were acquired in the Orbitrap (120 k resolution, 2e5 AGC target). Up to 20 of the most intense MS1 peaks were selected for the tandem MS2 analysis by collision-induced dissociation (CID) in the linear ion trap. The normalized collision energy was set at 34% for MS/MS. The isolation width was set at 1.6. Dynamic exclusion was enabled with an exclusion window of ±10 ppm and exclusion duration of 12 s.

All MS/MS spectra were searched against the human UniProt database using SEQUEST (Proteome Discoverer 1.4, Thermo Scientific). The search parameters were as follows: static modification, carbamidomethyl (C); dynamic modifications, oxidation (M); two missed cleavages allowed; MS1 mass tolerance 10 ppm; MS2 mass tolerance 0.6 Da; 1% false discovery rate (FDR). Proteins were quantified using a label-free quantitative method termed normalized spectral counting,³⁶ where the spectral count of a given protein was normalized against the sum of the spectral counts of all proteins identified in an MS run. All proteins identified in this study are listed in Supplemental Table S1. The MS proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE³⁷ partner repository with the data set identifier PXD027025.

Western Blotting

One-third of EV proteins were separated on a 4–15% SDS-PAGE gel and transferred to a polyvinylidene fluoride (PVDF) membrane (Bio-Rad). The membrane was incubated with 5% milk for 1 h at room temperature (RT) and then incubated with anti-CD9 antibody (1:1000) overnight at 4 °C followed by incubation with HRP conjugated secondary antibody (1:5000) for 1 h at RT. The membrane was washed with PBST (1× PBS, 0.1% Tween) for 10 min three times between each step and finally visualized using a SuperSignal West Pico Plus chemiluminescence substrate kit (Thermo). The membrane was incubated with the chemiluminescence substrate for 5 min prior to exposure on an imaging system (Bio-Rad).

Statistical Analysis

Experiments were independently performed in at least in triplicate. The statistical analysis of MS-based proteomic data of EVs was performed with Perseus.³⁸ The gene ontology (GO) analysis of all identified proteins was performed with FunRich version 3.1.4 using the Gene Ontology Database.³⁹ A P value of <0.05 indicated statistical significance.

RESULTS AND DISCUSSION

HPLC Fraction of EVs

Figure 2A shows a representative chromatogram of EV isolation from 40 μL of human serum using the CD9-HPLC-IAC method. The isolation of EVs was achieved within 30 min, where the peak of the EV fraction was observed between 16 and 19 min.

Figure 2.

(A) HPLC chromatogram of EV isolation from 40 μL of human serum with the EV fraction eluted between 16 and 19 min. (B) HPLC peak area under the curve of the EV fraction with different loading volumes of serum, that is, 20, 40, 60, 80, and 100 μL, respectively. Experiments were performed three times. Data are plotted as the mean ± SD.

We then evaluated the peak area under the curve of the EV fraction using different injection volumes of serum, that is, 10, 20, 40, 60, 80, and 100 μL, respectively. Experiments were performed in triplicate. Supplemental Figure S1A shows the chromatogram of 10 μL of serum, where only a minimal peak was observed for the EV fraction. As shown in Figure 2B, the HPLC peak area of EVs significantly increased by ~0.7-fold between 20 and 40 μL of serum (p < 0.001) and then slightly increased by ~0.1-fold between 40 and 60 μL of serum (p = 0.002). There was no significant increase in the peak area of EVs among 60, 80, and 100 μL of serum (p > 0.05), indicating that we had reached the loading capacity due to the column saturation. The size of the PEEK column used in this study was 4.6 mm × 50 mm, where the column volume was ~0.8 mL. One could scale-up the amount of antibody and the size of the PEEK column to accommodate larger sample volumes.

An underivatized column packed with the raw hydrazide resin without antibody was used to evaluate the nonspecific binding of the hydrazide resin. No EV peak was observed when 40 μL of serum was injected onto the underivatized column (Supplemental Figure S1B).

The CD9-IAC column allowed EV isolation from serum volumes as low as 20 μL, which is significantly lower than that used in the conventional methods. The column was run at the optimal flow rate, 0.5 mL/min, according to our previous study on HPLC-IAC purification.³³ Because the column was packed with antibody-conjugated resin (average particle size 50–80 μm), it is suitable for low-pressure procedures. As the flow rate is increased, the pressure will increase, which will result in lowering the binding capacity. Because 40 μL of serum gave a significant yield of EVs, we chose 40 μL of serum as the injection amount for the downstream characterization of EVs. A wash was applied between sample injections, which confirmed that there was no carryover between runs (Supplemental Figure S2). We also investigated the robustness of the CD9-IAC column by conducting the isolation experiment with 50 aliquots of the serum sample (40 μL each). The coefficient of variation (CV) of the peak areas of EVs among the 50 replicates was 9.7%, with a minor decrease of 8.4% in the yield for the 50th injection compared with the first injection (Supplemental Figure S3). The result indicated that the CD9-IAC column can be used for at least 50 times without a significant degradation of performance.

Postcolumn Cleaning

After the collection of the EV fraction, a postcolumn cleaning procedure was performed for buffer exchange and further reduction of potential coeluting serum proteins prior to EV characterization and proteomic analyses. Recently, we compared three different pore sizes of the Amicon Ultra-4 regenerated cellulose (RC) centrifugal filter (MWCO 10, 50, and 100 kDa, respectively) for the postcolumn cleanup of serum-derived EVs.²² A previous study showed that the Amicon 50 kDa filter dramatically reduced the serum albumin in EV samples compared with the 10 kDa filter, whereas it maintained a considerable recovery of EVs compared with the 100 kDa filter.²² In this study, the 50 kDa filter was used for postcolumn cleaning to reduce the contamination of serum proteins that coeluted with EVs. Because the EV fraction was collected from the HPLC between 16 and 19 min followed by the postcolumn cleaning step for 40 min, the whole procedure for EV isolation and cleaning required 60 min.

Supplemental Figure S4 shows the SDS-PAGE analysis of the EV fraction, without lysis, before versus after the postcolumn cleaning. With the reference of a total amount of 4.5 μg of a protein ladder on the gel (Figure S4), the amount of coeluting serum proteins in the EV fraction was estimated to be ~0.55 μg before the postcolumn cleaning and ~0.11 μg after the postcolumn cleaning. The coeluting serum proteins in the EV fraction were reduced by 80% after the postcolumn cleanup step as compared with before. With a typical range of 60–80 g/L of serum proteins, there was 2.4 to 3.2 mg of total serum proteins in 40 μL of serum. Therefore, >99.977% of serum proteins were depleted by the CD9 column before the postcolumn cleaning, and >99.999% of serum proteins were depleted with the postcolumn cleaning.

EV Characterization

The purified EVs were characterized by TEM, NTA, and WB. As shown in Figure 3A, the TEM image shows that the isolated EVs retained their morphological integrity as spherical or spheroidal membrane-encapsulated particles of ~35–150 nm in diameter. The result indicated that the CD9-HPLC-IAC procedure did not affect the membrane structure of the EVs. The size histogram of EVs retrieved from the TEM images is shown in Figure 3B, where the mean diameter of the EVs was ~70 nm.

Figure 3.

Characterization of the EVs isolated from human serum using the anti-CD9 HPLC column. (A) TEM image showing the morphology and size of the EVs; scale bar = 100 nm. (B) Size histogram of EVs retrieved from TEM images. (C) NTA of the particle size distribution and concentration of the EVs. (D) Western blotting image of the EV marker CD9.

The particle concentration was determined by NTA. Figure 3C shows the NTA result of five videos of the isolated EVs, where a major peak was observed at 96 nm and two less-concentrated peaks were observed at 71 and 140 nm, respectively. The NTA result showed that ~95% of the isolated particles were within the expected size range of exosomes (30–150 nm). Minor signals at ~200 nm accounted for <5% of the particles. Although the number of larger EVs was low, they could significantly contribute to the protein signal. It has been reported that the NTA produced substantially larger sizes and broader size distributions as compared with TEM imaging, which might result from the physical principle of size determination of the NTA, where the size is calculated based on the motion of particles in the solution and influenced by the surface protein composition of EVs.⁴⁰

The number of EVs isolated from 40 μL of serum was ~8.0 × 10⁹, which was equivalent to ~2.0 × 10¹¹ per mL of serum. Compared with the conventional UC and SEC methods that can recover (1.0 to 1.5) × 10⁹ EVs from 1 mL of serum,¹⁰ the current CD9-HPLC-IAC method achieved a 100 times higher yield. In addition, our method was comparable to the PET C-CP fiber HIC method developed by the Marcus group,⁴¹ which can isolate ~1 × 10¹⁰ EVs from 0.1 mL of serum.²²

To confirm that the vesicles were indeed EVs, a common EV marker, CD9, was analyzed using WB. As shown in Figure 3D, CD9 was observed in the WB result of EVs isolated from human serum by the CD9-HPLC-IAC method. Other EV markers such as CD63 and CD81 were detected in the subsequent MS analysis. An image of the entire membrane is shown in Supplemental Figure S5.

Proteome Profiling of EVs Isolated from 40 μL of Serum

First, we estimated the protein amount in EVs by using SDS-PAGE followed by silver staining. Figure 4A shows the gel image of EVs isolated from 40 μL of serum after postcolumn cleanup, with a total amount of 1.5 μg of a protein ladder as a reference. With ImageJ to quantify protein bands on the gel, the protein amount of EVs isolated from 40 μL of serum was estimated as ~0.6 μg.

Figure 4.

(A) SDS-PAGE analysis of the EVs isolated from 40 μL of human serum using the CD9-HPLC-IAC method followed by silver staining, with a total amount of 1.5 μg of a protein ladder as a reference. (B) Venn diagram showing the overlap of proteins identified in three replicates of EVs isolated from 40 μL of human serum. (C) Heat map of the Pearson correlation coefficient R² for the comparison of the MS results of the three replicates. (D) Gene ontology analysis of the cellular component, molecular function, and biological process of the identified EV proteins.

Half of the peptide samples were injected for LC–MS/MS analysis. The experiment was performed three different times with three aliquots of a human serum sample. In total, 482 proteins were identified among the three replicates (Supplemental Table S1). As shown in Figure 4B, the Venn diagram shows the overlap of proteins identified across the three replicates, where 412 (78.0%) proteins were found in at least two replicates and 212 (42.3%) proteins were commonly identified in three replicates. To obtain a desirable protein identification for EVs isolated from a microscale volume of serum (40 μL), two to three replicates are required. A previous study on EVs isolated from 1 mL of serum using UC, UC and SEC, and SEC, respectively, showed that 70–80% proteins were identified in at least two replicates and 45–55% proteins were commonly identified in all replicates.¹⁰ Herein the CD9-IAC method achieved a comparable result on the overlap of protein identities across the replicates. A heat map of the Pearson correlation coefficient R² values for the comparison of the MS results of the three replicates was used to reflect the reproducibility of the method (Figure 4C). The Pearson correlation coefficient R² across the three replicates was 0.90, 0.93, and 0.94, respectively.

The result of the GO analysis of the cellular component, molecular function, and biological process of the identified EV proteins is shown in Figure 4D, where the eight most significant items (p < 005) of the GO analysis are listed. For the cellular component, it revealed that 82% of proteins were associated with exosomes followed by cytoplasm (72%) and plasma membrane (38%), and a total of 23% of proteins were from the extracellular space. This suggests that these proteins originated from exosomes. The prevalent molecular functions of the identified proteins were assigned as transporter activity, cytoskeletal protein binding, cell adhesion molecule activity, and receptor activity. The GO analysis for biological processes discovered that the EV proteins were mainly involved in signal transduction (24%), cell communication (23%), protein metabolism (18%), cell growth and/or maintenance (17%), and transport (11%).

Identification of EV Markers

The EV markers identified in this study are summarized in Table 1. The identified EV markers included endosome-specific tetraspanins (CD63, CD81, CD82, and CD151), the components of the endosomal sorting complex required for the transport (ESCRT) complex (Alix, TSG101, HSP70, and VPS), and endosome-related membrane transport and fusion proteins (annexins, flotillin, and Rab). In addition to CD63/CD81/CD82, other tetraspanins including TSPAN4, TSPAN8, and TSPAN14 were also detected, whereas TSPAN8 has been reported to facilitate a selective recruitment of proteins into EVs and elicit the preferential binding of these EVs to endothelial cells.⁴² Tetraspanins also control the trafficking of integrin complexes; for example, the assembly of the complex between CD151 and integrin takes place early during the biosynthesis of the integrin heterodimer, which may promote cell migration.⁴³ Because of their endosomal origin, the accessory ESCRT pathway proteins, that is, Alix and TSG101, are often used as exosome markers, confirming the endocytic origin of vesicles.⁴⁴ In this study, VPS38B, an ESCRT-associated protein, and the exosome-excluded endoplasmic reticulum protein GP96 (HSP90B1) were also detected. Major vault protein (MVP), EH domain-containing protein 4 (EHD4), and syndecan binding protein 2 (SDCBP2) were involved in trafficking and membrane fusion.⁴⁵ MHC class I proteins (HLA-A/C), which are responsible for antigen presentation, and integrins α2/3/6/V and β1/3/4, which are responsible for cell adhesion, were detected in the EV samples. Among these proteins, some specifically enriched in exosomes are widely used as exosomal markers (e.g., CD63, CD81, CD82, Alix, and TSG101). We detected CD9 in the Western Blot analysis (Figure 3D), which confirmed the presence of CD9 in EV samples. CD9 was not detected by MS, probably due to the tryptic digestion step using the FASP method with a YM-30 kDa centrifugal filter, where CD9 has a molecular weight (MW) of <25 kDa, which might be filtered out during the FASP procedure.

Table 1.

Lists of the EV Markers Identified in This Study

category	description
tetraspanins	CD9, CD63, CD81, CD82, CD151, TSPAN4, TSPAN8, TSPAN14
ESCRT complex	TSG101, Alix, VPS37B, HSPA8, HSP90B1
membrane transport and fusion proteins	FLOT1, FLOT2, ANXA1, ANXA2, ANXA4, ANXA5, ANXA7, ANXA11, ANXA13, RAB7A, MVP, EHD4, SDCBP2
antigen presentation	HLA-A, HLA-C
cell adhesion	ITGA2, ITGA3, ITGA6, ITGAV, ITGB1, ITGB3, ITGB4, EPCAM

Relative Quantification of EV Protein Content

We then performed relative quantification of the proteomic data to assess the relative abundance of proteins in EV content and the efficiency of the CD9-HPLC-IAC method in reducing contaminating proteins in EV samples. The major challenge of the proteomic study of serum-derived EVs is the coelution of the high-abundance serum proteins and lipoprotein particles,^32,46,47 especially for EVs purified from a small amount of serum.¹⁹ The two major impurities can interfere with the analysis of EV protein content by MS.^16,46 The top 10 most abundant serum proteins include albumin, IgG, transferrin, fibrinogen, IgA, α−2-macroglobulin, α−1-antitrypsin, complement C3, IgM, and haptoglobin, accounting for 90% of the total serum protein mass,⁴⁸ whereas the main components of lipoprotein particles include apolipoprotein A/B/C/E and lipoprotein(a).³² The relative abundance of proteins in these two groups was summed, respectively, to represent the percentage content of the top 10 serum proteins and apolipoproteins identified in EVs.

The percentage content of the top 10 serum proteins in the three EV samples was 15.0, 11.9, and 9.6%, respectively, and that of apolipoproteins was 8.6, 8.8, and 6.3%, respectively (Table S1). The mean and SD values of the top 10 serum proteins and apolipoproteins were 12.2 ± 2.7% and 7.9 ± 1.4%, respectively. We compared the MS result with that reported in our recent study where EVs were isolated from 1 mL of human serum using an optimized UC or SEC (qEV) method.¹⁰ The content of the top 10 serum proteins was significantly decreased in the CD9-HPLC-IAC method as compared with the optimized UC (25.5 ± 0.8%, p = 0.001) and SEC (qEV) methods (19.7 ± 0.3%, p = 0.009). The content of apolipoproteins was also significantly reduced in the CD9-HPLC-IAC method compared with the SEC (qEV) method (38.2 ± 0.7%, p < 0.001), whereas it was comparable to the UC method (5.9 ± 0.7%, p = 0.095). We also compared the average percentage content of the top 10 serum proteins in the EVs between the CD9-IAC method and the PET C-CP fiber HIC method,²² which was 12.2 and 12.69%, respectively. This suggests that the IAC method is comparable to the HIC method in reducing the high-abundance serum proteins in EV isolation.

Comparison between the CD9-HPLC-IAC Method and the UC/SEC/Magnetic-Bead-based IAC/HIC

We compared the CD9-HPLC-IAC method versus the UC and SEC methods in the isolation of EVs from human serum, including serum sample volume, processing time, morphology, size, concentration, and reduction of the contamination of serum proteins and lipoprotein particles. The data of the UC and SEC (qEV) methods were retrieved from our recent study.¹⁰ The comparison is summarized in Table 2.

Table 2.

Comparison between the CD9-HPLC-IAC Method and the UC and SEC (qEV) Methods in Isolating EVs from Human Serum

	CD9-HPLC-IAC	five cycles of UCa	SEC (qEV)a
serum volume (μL)	40	1000	500
processing time (min)	60	400	60
EVs
morphology	intact	intact	intact
mean size	70 nm	67 nm	77 nm
concentration	8.0 × 109 per 40 μL serum (equivalent to 2 × 1011 per mL serum)	1.0 × 109 per mL serum	1.3 × 109 per mL serum

^aData of the UC and SEC methods were retrieved from our recent studies¹⁰

In the case of the morphology, the EVs isolated using the CD9-HPLC-IAC, UC, or SEC methods all had intact membrane structures, indicating that these approaches did not affect the morphology of EVs. Regarding the size of the EVs, the TEM result showed that the mean diameters of EVs from the CD9-HPLC-IAC, UC, and SEC methods were 70, 67,¹⁰ and 77 nm,¹⁰ respectively. When comparing the yields of the serum-derived EVs among the three methods, the CD9-HPLC-IAC method achieved 8.0 × 10⁹ EVs from 40 μL of serum, equivalent to 2.0 × 10¹¹ per mL of serum, whereas the UC and SEC methods could recover 1.0 × 10⁹ and 1.3 × 10⁹ EVs from 1 mL of serum.¹⁰ The current CD9-HPLC-IAC method achieved a 100 times higher yield than the UC and SEC methods. It has been reported that on average, from 1 mL of serum, EV concentrations can vary between (1 and 3) × 10¹² particles/mL,⁴⁹ and EV concentrations in the serum from patients with disease tend to be higher than those in healthy samples.⁵⁰ Taking into consideration the very small volume of the starting serum, that is, 40 μL, the CD9-HPLC-IAC method achieved a high yield of EVs.

According to the proteomic results of the serum-derived EVs, the percentage content of the top 10 most abundant serum proteins in the CD9-HPLC-IAC method was reduced 2.1- and 1.6-fold compared with that of the UC and SEC methods, respectively. The abundance of apolipoproteins in the CD9-HPLC-IAC method was reduced 4.8-fold compared with that of the SEC method while being similar to the UC method. The result demonstrated that the CD9-HPLC-IAC method distinctly reduced the contamination of the high-abundance serum proteins and apolipoproteins for the EV isolation from serum.

It should be noted that compared with the UC and SEC methods, the advantage of the CD9-HPLC-IAC method for the isolation of EVs from serum is that it scales down the starting volume of the serum to as little as 40 μL. It enables the proteome profiling of EVs derived from a microscale amount of serum, which is most important when patient samples are limited in volume size. In addition, the whole procedure of the CD9-HPLC-IAC method including the postcolumn cleaning takes 60 min, whereas the UC method is time consuming and usually takes 400 min. Although the SEC method also provides a rapid EV isolation like the CD9-HPLC-IAC method, an SEC column can be reused up to only five times, whereas the antibody-based HPLC column can be reused over 50 times. In addition, size- or density-based separation cannot distinguish subpopulations of EVs; however, antibody-based immunoisolation, such as CD9-HPLC-IAC, can capture EV subtypes based on the specific surface proteins of EVs.

Compared with the magnetic-bead-based IAC method, the column-based IAC method has the advantage of real-time monitoring of the EV fraction and minimizing the contamination of blood proteins by optimizing the column wash step. The use of magnetic beads is a useful strategy for the isolation of EVs from a cell culture supernatant or liquid biopsy for the downstream analysis of EV surface markers using flow cytometry or imaging.^30,31 However, lipoproteins in a liquid biopsy often bind to magnetic beads, as they are one of the major contaminants in the isolated EVs for downstream proteomic analysis.³² The CD9-HPLC-IAC method enables the elution of intact EVs and fast postcolumn cleaning, where the EV sample is obtained in PBS and ready for a full characterization including proteomic profiling. Another advantage is that the HPLC column is reusable multiple times, that is, >50 times.

The CD9-HPLC-IAC technology developed in this study is a different technique compared with the PET C-CP HIC method reported in our previous work.²² The CD9-HPLC-IAC method is based on immunoaffinity using a CD9 antibody, whereas the PET C-CP method is based on HIC. Compared with the PET C-CP column, which is suitable for isolating total/bulk EVs, the advantage of the CD9-HPLC-IAC method is that the column can be modified with different antibodies for the separation of EV subpopulations. The main disadvantages of the CD9-IAC method are the cost of the antibody and the fact that the IAC efficiency is highly dependent on the quality of the antibody, where a high-quality antibody with a high intrinsic affinity toward the EV surface protein is required, whereas the PET C-CP fiber is low-cost with the consistency of a hydrophobic interaction. Other differences between the two methods include the column size and the mobile phases. Because of the different types of matrix materials, the CD9-immobilized hydrazide resin was packed into a 4.6 mm × 50 mm PEEK column, whereas the PET C-CP fiber was packed into a 0.76 mm × 50 mm PEEK column.²⁴ Because of the different interaction patterns with EVs, the mobile phases in the CD9-IAC method were PBS (for loading and washing) and 1 M NaCl (for elution), whereas in the PET C-CP method,²² the mobile phases were 2 M (NH₄)₂SO4 in PBS (for loading), 25% glycerol in 1 M (NH₄)₂SO4 (for washing), and 50% glycerol in PBS (for elution). Although the EVs were eluted with different conditions, the same postcolumn cleaning step using a MWCO filter was performed in the two methods to exchange buffer into PBS, concentrate the EV sample, and further reduce potentially coeluting serum proteins.

An important advantage of both the CD9-HPLC-IAC and PET C-CP HIC methods is that they can provide the efficient isolation of EVs from small amounts of serum sample, down to 20 μL if needed. Both methods provide an equivalent yield of nearly 10¹¹ EVs per milliliter of serum. In addition, both methods can be rapidly performed on an HPLC platform with real-time monitoring, and the columns are reusable and reproducible. Most importantly, these methods provide improved performance in eliminating high-abundance blood proteins and apolipoproteins, which interfere with mass spec analysis, where extensive washing on the column can be achieved by these chromatographic methods. The two techniques complement each other in various studies and may be used together to purify EV populations.

CONCLUSIONS

We developed a CD9-HPLC-IAC method that enables the rapid and effective isolation of EVs from a microscale amount of serum, as little as 40 μL. This approach yielded 8.0 × 10⁹ EVs from 40 μL of serum in 30 min to achieve a pure sample of morphologically intact EVs. The proteomic analysis of the EVs after a postcolumn cleaning step using the 50 kDa MWCO filter demonstrated that this method greatly reduced the contamination of serum proteins and apolipoproteins, the major impurities in serum-derived EVs. This technique enables the real-time isolation of EVs from a microscale amount of serum for downstream proteomic analysis, which is most important when patient samples are limited in volume size. We used only 40 μL for this study to demonstrate the method, indicating that it is suitable for the development of a real-time EV assay, in particular, for trace amount sample analysis. The column could be scaled up to accommodate larger sample volumes. The high efficiency of the CD9-HPLC-IAC method in the isolation and characterization of EVs from a microscale volume of serum for proteomic profiling may provide a potential application for clinical analysis. In addition, this technology can be extended by using other EV markers such as CD63, CD81, and TSG101, which is especially useful where one might be interested in various subpopulations of EVs in correlation with diseases.