Large‐scale heparin‐based bind‐and‐elute chromatography identifies two biologically distinct populations of extracellular vesicles

ABSTRACT

Purifying extracellular vesicles (EVs) has been challenging because EVs are heterogeneous in cargo yet share similar sizes and densities. Most surface marker‐based affinity separation methods are limited to research or diagnostic scales. We report that heparin chromatography can separate purified EVs into two distinct subpopulations as ascertained by MS/MS: a non‐heparin‐binding (NHB) fraction that contains classical EV markers such as tetraspanins and a heparin‐binding (HB) fraction enriched in fibronectins and histones. Both fractions were similarly fusogenic but induced different transcriptional responses in endothelial cells. While EVs that were purified by conventional, non‐affinity methods alone induced ERK1/2 phosphorylation and Ki67, the NHB fraction did not. This result suggests heparin chromatography as an additional novel fractionation step that is inherently scalable, does not lead to loss of material, and separates inflammatory and pyrogenic EVs from unreactive EVs, which will improve clinical applications.

1. INTRODUCTION

Extracellular vesicles (EVs) are membrane‐bound particles secreted by all types of cells. Based on their origins of biogenesis, EVs are classified into two major categories: ectosomes (microvesicles) and exosomes. Ectosomes are released by direct outward budding from the plasma membrane (PM). Exosomes are formed by inward budding of the endosomal membrane during the maturation of multivesicular bodies (MVBs) and are released when MVBs fuse with the PM (Théry et al., 2018; van Niel et al., 2018; Witwer & Théry, 2019). Here, we use the term EVs to denote vesicles purified from extracellular fluids with a diameter of ≤200 nm, irrespective of their biogenesis.

EVs have received intense interest due to their ability to transport bioactive molecules, such as proteins and RNAs (mRNAs, miRNAs, lncRNAs), their function in human disease, and their ability to elicit functional paracrine responses (Chugh et al., 2013; Hinger et al., 2018; McNamara et al., 2018; Skog et al., 2008; Valadi et al., 2007). Ongoing clinical studies seek to understand the potential of EVs as drug‐delivery vehicles and whether EVs have specific tissue‐homing capabilities combined with low adverse effects as compared to other drug‐delivery modalities. Applied and research studies depend on having a reliable, reproducible, consistent, and defined source of EVs, ideally of low complexity.

EVs, however, are heterogeneous in their sizes, density, surface markers, and cargo composition. This prompted studies to isolate EV sub‐populations based on size range, density, and marker expression (Gould & Raposo, 2013; Théry et al., 2018). Different techniques are currently in use to purify EVs. These include differential ultracentrifugation (UC), density gradient centrifugation, filtration, polyethylene glycol (PEG) based precipitation, size‐exclusion chromatography (SEC), field flow fractionation, fluorescence‐activated cell sorting (FACS), high‐pressure liquid chromatography (HPLC) and bead‐based affinity separation techniques, to name but a few (reviewed in Royo et al. (2020)). For therapeutic applications, it is critical to avoid methods that can potentially damage EVs or that lack scalability in regulated manufacturing environments. Our impression is that no one method alone can obtain concentrated, pure EV preparations for therapeutic use and that better purification pipelines are needed to advance the field.

If not sufficiently purified, vesicles and non‐vesicular particles obscure each other's phenotypic readouts. Aside from ectosomes and exosomes, other types of particles, such as exomeres, supermeres, and apoptotic bodies, have been identified. Exomeres were first isolated by Zhang et al. (2018), whose study describes an abundant population of nanoparticles with sizes under 50 nm, which notably do not contain membranes. Supermeres refer to non‐membranous nanoparticles even smaller than exomeres (Jeppesen et al., 2019; Zhang et al., 2021). Apoptotic bodies, sizing between 800 and 5000 nm and released from dying cells, are the largest EVs (Crescitelli et al., 2013).

This work adds a novel biophysical purification step (heparin‐binding) to previously published EV purification methods, whether it be UC, SEC, tangential flow filtration (TFF), or PEG precipitation. Our prior method emphasizes scalability and can use several liters of biofluid as input (McNamara et al., 2018). It uses successive steps of slow‐speed centrifugation, TFF, PEG precipitation, and Capto Core 700 chromatography. This method is efficient and reproducible and yields EVs of comparable homogeneity to UC and gradient centrifugation. Here, we added a heparin column step, collected EVs eluting in the binding and non‐binding fractions, and characterized their molecular compositions and biological functions. One population flows through without binding to the column (non‐heparin‐binding, NHB), and the other binds to the column under physiological conditions but can be eluted by a higher salt concentration buffer (heparin‐binding, HB). NHB and HB contained extracellular particles detectable by nanoparticle tracking analysis (NTA) and transmission electron microscope (TEM). Western blot (WB) and proteomics analysis showed that NHB and HB contained different cargo. Importantly, NHB and HB showed different biological effects on endothelial cells, demonstrating that heparin‐based affinity binding chromatography can separate EVs into two subpopulations with distinct biological functions.

2. MATERIALS AND METHODS

2.1. Cell culture

U2OS cells (HTB‐96) were obtained from ATCC and maintained in DMEM medium (Gibco 11995‐065) with 10% EV‐free Fetal Bovine Serum (FBS VWR 97068‐085), 20 mM of L‐Glutamine (Gibco 25030‐081), and 100 U/mL PennStrep (Gibco 15140‐122). BCBL‐1 cells were from Don Ganem (Renne et al., 1996) and maintained in RPMI medium (Gibco 11875‐093) with 10% EV‐free FBS, 20 mM of L‐Glutamine (Gibco 25030‐081), and 100 U/mL PennStrep (Gibco 15140‐122). To make EV‐free FBS, PEG‐8000 was added to FBS to a final concentration of 40 mg/mL and incubated at 4°C overnight, followed by centrifugation at 1000 g for 1 h. The supernatant (EV‐free FBS) was used in cell culture. For routine maintenance, cells were trypsinized (Gibco 25300‐054) upon 80% confluency, and 5 × 10⁶ cells were seeded in a T175 flask.

2.2. Medium conditioning and EV isolation

To condition the medium, 1 × 10⁶ U2OS cells were seeded to 20 mL medium in a T175 flask (10% confluency) and grown for four days until about 80% confluency when we added another 80 mL fresh medium to a total of 100 mL per 10 million cells and condition for 2 days. The viability of cell culture was monitored daily by observing cell morphology, density, and medium pH. Healthy cell morphology was judged based on guidance from ATCC (U2OS, HTB‐96). The pH of the medium was between 7.0 and 7.5, as indicated by phenol red. Cells were checked by trypan blue staining to ensure more than 95% of them were viable at EV harvest. The conditioned medium was sampled daily, filtered through a 0.22 μm filter, and the particle concentration was measured by ZETAView (Particle Metrix, PMX‐120) to ensure the particle concentration was at >5 × 10⁹ particles/mL before harvesting.

EV isolation was performed as described before (McNamara et al., 2018). Briefly, the conditioned medium was centrifuged at 1000 g for 5 min, filtered through a 0.45 μm (GenClone 25‐230) and then a 0.22 μm (GenClone 25‐229) vacuum filter system to remove cell debris and vesicles larger than 220 nm diameter, then transferred to an ÄKTA Flux tangential flow filtration (TFF) system (Cytiva 29038437) paired with a 750 kDa cut‐off ultrafiltration hollow fiber cartridge (Cytiva UFP‐750‐C‐H24LA). The transmembrane pressure (TMP) was kept at 27 psi. The medium was condensed until the retentate reached approximately 1/10 of its starting volume before adding PEG‐8000 to a final concentration of 40 mg/mL and incubating with rotation at 4°C overnight. The sample was centrifuged the next day at 1000 g for 1 h. The pellet was resuspended in PBS (pH 7) (Gibco 14190144) and incubated with 50 μg/mL of DNase (Promega M6101) and Rnase (Thermo Scientific, EN0531) and 25 μg/mL of Cell Mask Deep Red (CM‐DeepRed) stain (Invitrogen C10046) at 4°C overnight. After incubation, the nucleases, remaining contaminants, and excess dye are removed by column chromatography on a 1 mL HiTrap Capto Core 700 column (Cytiva 29400461) at 0.5 mL/min on an ÄKTA Start chromatography system (Cytiva 29022094) using PBS (pH 7) as the running buffer. An additional incubation with PBS (pH 7) ‐primed Capto Core 700 slurry was added before Capto Core 700 column run to reduce viscosity and column pressure. Capto Core 700 combines the sizing capabilities of a size‐exclusion chromatography (SEC) resin with a charged inner bead core to retain soluble proteins, nucleic acids, and metabolites. EVs were in the flow‐through. At each step, EV size distribution was measured by NTA.

2.3. EV fractionation on heparin column chromatography

After Capto Core 700, the peak fractions were combined and loaded on a 1 mL HiTrap Heparin High‐Performance Column (Cytiva 17040601) at 0.5 mL/min using PBS (pH 7) as a binding buffer. The column was washed with three column volumes of PBS (pH 7), and the EVs were eluted with an increasing gradient of NaCl up to 1 M.

2.4. EV isolation by differential centrifugation after TFF

The conditioned medium (500 mL) was clarified by filtering through a 0.45 μm and a 0.22 μm vacuum filter system and then condensed by TFF to 1/10 of the starting volume as mentioned above. The condensed medium was either loaded directly onto the 1 mL HiTrap Heparin High‐Performance Column (Cytiva 17040601) (no‐spin); or further clarified by centrifuging at 10,000 g in a 5424R (Eppendorf) centrifuge with an FA‐45‐24‐11 rotor for 45 min (10K spin) before loading the supernatant on to the heparin column; or the EVs in the supernatant were centrifuged down at 100,000 g in an SW55 rotor (Beckman Coulter) for 90 min (100K spin), resuspended in 0.5 mL PBS before loading on to the heparin column. EVs were fractionated by the heparin column, as mentioned above.

2.5. Western Blot for EV marker

The cell pellet was lysed with RIPA buffer (20 mM Tris‐HCl, 150 mM NaCl, 1 mM EDTA, 1% NP‐40, 0.1% SDS, 0.5% sodium deoxycholate, pH 7.5) supplemented with protease inhibitor (Thermo Scientific 78429) on ice for 30 min and vortexed once every 10 min followed by centrifugation at 10,000 g for 10 min at 4°C to remove large debris. EVs were lysed with Strong Lysis Buffer (5% SDS, 10 mM EDTA, 8 M Urea, 120 mM Tris‐HCl, pH 6.8) at a 1:1 ratio with protease inhibitor (Thermo Scientific 78429). Protein concentrations were determined by BCA assay (Thermo Scientific 23225), and 10 μg of protein was loaded per well. Cell and EV lysates were combined with loading buffer (62 mM Tris‐HCl, 2% SDS, 0.05% bromophenol blue, 10% glycerol, pH 6.8) and heated at 95°C for 5 min. For reducing conditions, 3% β‐mercaptoethanol was added as the reducing agent. Samples were separated on a SurePage gel (GenScript M00654) or a NuPage gel (Invitrogen NP00336BOX) at 120 V for 90 min before transferring onto a nitrocellulose or PVDF membrane (20 V overnight or 100 V for 1 h 30 min). Transferred membranes were blocked with 5% non‐fat milk or 5% BSA in 1 x TBST (100 mM Tris‐Cl pH 7.5, 0.9% w/v NaCl, 0.1% Tween 20) and incubated with primary antibody overnight at 4°C and Li‐Cor secondary antibodies (Donkey anti‐Rabbit: Li‐Cor 926‐68073; Donkey anti‐Mouse: Li‐Cor 926‐32212) for 1 h in blocking buffer. The antibodies used in this study are listed in Table S1. The membranes were imaged using a Li‐Cor Odyssey Imager (Li‐Cor 9120).

2.6. Nanoparticle tracking analysis

ZetaView Nanoparticle Tracking Analyzer (Particle Metrix, PMX‐120) was used to determine particle concentration, size distribution, and zeta potential. The instrument was standardized using 100 nm Nanosphere beads (Thermo Scientific 3100A) per system requirement at the sensitivity of 65 and shutter of 100 (ZetaView 8.04.02). Samples were diluted in ddH₂O to between 50 and 200 particles per view window. For size and size distribution measurement, data was acquired with sensitivity at 88, shutter at 70, max area at 220, min area at 5, and min brightness at 20. For zeta potential measurement, data was acquired with sensitivity at 88, shutter at 70, max area at 220, min area at 5, min brightness at 20, and max brightness at 255. Each measurement was run for three cycles, and at least three measurements were performed with fresh dilutions for each sample.

2.7. TEM

Carbon‐coated grids (400‐mesh, copper) were first glow‐discharged. EV samples were loaded onto the grids, followed by a brief rinse with water, then stained with 2% (w/v) uranyl acetate and air‐dried. Images were obtained with an FEI Tecnai 12 transmission EM (Thermo Fisher Scientific) at 80 kV equipped with a Gatan Orius CCD camera with Gatan Digital Micrograph software as described (Al‐Turki & Griffith, 2023).

2.8. dSTORM

Protocols for dSTORM calibration and operation were published earlier (Chambers et al., 2021; McNamara et al., 2022). We included details most relevant to this work here. Briefly, to load EV onto the imaging chamber, a micro‐slide 8‐well glass‐bottom chamber (iBidi 80827) was prepared by adding 200 μL, 0.01% Poly‐L‐Lysine solution (MP Biomedicals, 0215017510) to each well and left at 4°C overnight. Poly‐L‐Lysine was removed and washed with PBS (pH 7), and 200 μL of heparin‐isolated EV samples were seeded in each well at a concentration of 1 × 10¹⁰ particles/mL in PBS and left at 4°C overnight to settle. EVs were fixed with 0.5% paraformaldehyde for 30 min at room temperature and then washed three times with PBS. For antibody probing, 5% BSA in PBS was used to block for 30 min at room temperature. The CD81 antibody (Invitrogen, MA5‐13548) and Fibronectin antibody (Abcam, ab2413) were labeled with the Alexa Fluor 488 (Thermo Fisher A20181) and Alexa Fluor 568 Antibody Labeling Kits (Thermo Fisher A20184), respectively. The labeled antibodies were diluted 1:100 in 5% BSA and added to each well for 3 h at room temperature. Excess antibody was removed, followed by the addition of 200 μL of pre‐mixed B‐cubed buffer (ONI, BCA0017) at room temperature for 30 min. EV samples were imaged using the Nanoimager (Oxford Nanoimaging) described in (Chambers et al., 2021). Data were exported and analyzed under ONI's CODI analysis platform: https://alto.codi.bio/user/login. Drift correction, filtering, clustering, and counting were performed on each image. For sizing analysis, the radius of gyration and length for EV surface marker analysis was performed using parameters listed in Table S2.

2.9. Fluorescence and immunofluorescence microscopy

For fusogenic analysis, hTERT‐HUVEC cells (gift from B. Damania) were seeded at 1 × 10⁶ cells/mL in 2 mL EBM‐2 medium supplemented with 2% EV‐depleted FBS on coverslips (Azer Scientific, 170 μm, ES0107052) and incubated at 37°C, 5% CO₂ for 24 h. EVs were added to each well at 1 × 10⁹ particles/mL (multiplicity of infection (MOI) equivalent ∼1000:1) and incubated for 24 h. Cells were then fixed with 4% paraformaldehyde for 10 min, blocked with 5% BSA/PBS for 30 min, and the nuclei were stained with 1 mL DAPI (1 μg/mL) in ddH₂O for 10 min. After each step, the cells were rinsed with 0.5 mL PBS three times. Coverslips were mounted to glass slides (Azer Scientific, UNIW75251+) with 20 μL of antifade mounting media (Cell Signaling 9071S), sealed with nail polish, and allowed to harden at room temperature for 10 min. Imaging was performed on a phase‐contrast microscope (Leica 24665) using a 40x objective lens, a blue filter (Ex480/40, Em527/30) for nuclei (DAPI), and a deep red filter (Ex620/60, Em700/75) for cells treated with CM‐DeepRed labeled EVs. Acquired images were exported and analyzed using Image‐J (v1.53k) software.

For Ki67 activation analysis, hTERT‐HUVEC cells were seeded at 2 × 10⁶ cells/mL in 3 mL EBM‐2 medium supplemented with 2% EV‐depleted FBS on coverslips and incubated at 37°C, 5% CO₂ for 24 h. EVs were added to each well at 3 × 10⁹ particles/mL (multiplicity of infection (MOI) equivalent ∼1500:1) and incubated for 24 h. Slides were prepared similarly as above except blocking with 5% Goat serum in PBST (0.1% Tween‐20), additional incubations with rabbit anti‐Ki67 antibodies (Epredia, RM‐9106‐S0, Clone SP6) for 3 h, anti‐rabbit AlexaFlour 488 antibody (Invitrogen, A32731) for 1 h, rhodamine‐phalloidin (Invitrogen, R415) for 20 min at room temperature using the dilutions specified in Table S1.

2.10. Flow cytometry

hTERT‐HUVEC cells were seeded in 1 mL of EGM‐2 medium in a 24‐well plate at 70,000 cells/mL. CM‐DeepRed labeled EV samples were added to each well at indicated concentrations and incubated at 37°C for 2 h. After incubation, the cells were trypsinized and resuspended in 1 mL EGM‐2 medium before flow. The mean fluorescence was measured on a BD flow cytometer (BD Accuri C6 Plus) calibrated with CS&T RUO beads (BD 661414) at an excitation wavelength of 640 nm and a 675/25 nm filter.

2.11. Mass spectrometry

Sample preparation: Input, NHB, and HB EV samples were subjected to mass spectrometry proteomics analysis. Three biological replicate batches of samples, isolated 1 week apart between each batch, were included. For each sample, 50 μg protein of each sample was dried down via vacuum centrifugation, resuspended in 8 M urea and reduced with 5 mM DTT at 56°C for 30 min, and alkylated with 15 mM iodoacetamide for 45 min at room temperature. Samples were then diluted to 1 M urea and subjected to digestion with LysC (Wako) for 2 h and trypsin (Promega) overnight at 37°C at a 1:50 enzyme:protein ratio. The resulting peptides were acidified to 0.5% TFA (trifluoroacetic acid) and desalted using Thermo Forma desalting spin columns. Eluates were dried via vacuum centrifugation, and peptide concentration was determined via Pierce Quantitative Fluorometric Assay. All samples were normalized to 0.25 μg/μL, and a pooled sample was created by combining 2 μL of each sample to assess technical reproducibility. All samples were spiked with internal retention time standards (iRT; Biognosys) and subjected to LC‐MS/MS analysis.

LC/MS/MS analysis: Samples were analyzed (n = 3) by LC‐MS/MS using an Ultimate 3000 coupled to an Exploris 480 mass spectrometer (Thermo Scientific). The pooled sample was analyzed twice, once before and once after the sample set. Samples were injected onto an IonOpticks Aurora series 2 C18 column (75 μm id × 15 cm, 1.6 μm particle size) (IonOpticks) and separated over a 200 min method. The gradient for separation consisted of 3%–41% mobile phase B at a 250 nL/min flow rate, where mobile phase A was 0.1% formic acid in water and mobile phase B consisted of 0.1% formic acid in 80% ACN. Exploris 480 was operated in product ion scan mode for Data Independent Acquisition (DIA). A full MS scan (m/z 350–1650) was collected; the resolution was set to 120,000 with a maximum injection time of 20 ms and an AGC target of 300%. Following the full MS scan, a product ion scan was collected (30,000 resolution) and consisted of stepped higher collision dissociation (HCD) set to 25.5, 27, 30; AGC target set to 3000%; maximum injection time set to 55 ms; variable precursor isolation windows from 350 to 1650 m/z.

2.12. Proteomics data analysis

Raw data were analyzed using DirectDIA within Spectronaut (v15.7.220308.50606, Biognosys). Data were searched against the reviewed human proteome database from Uniprot (UP000005640, downloaded January 2022 and containing 20,291 protein sequences); appended to a common contaminants database (MaxQuant; 245 proteins). Two separate searches were performed: a qualitative (Table S3) and a quantitative analysis (Table S4). For the qualitative analysis, samples of each group were searched separately (n = 3/search). For the quantitative analysis, all samples and the two pooled replicates were searched together. The following settings were used for both searches: enzyme specificity set to trypsin, up to two missed cleavages allowed, cysteine carbamidomethylation set as a fixed modification, methionine oxidation, and N‐terminal acetylation set as variable modifications. Precision iRT calibration was enabled. A false discovery rate (FDR) of 1% was used to filter all data, and single hits were excluded from the results. The datasets were queried against the Gene Ontology (GO) database within Spectronaut, and proteins with GO Cellular Compartment (GOCC) terms containing ‘extracellular’ were considered potential exosome proteins. The qualitative dataset was used to compare the number of identified proteins and their proportion of proteins containing the GOCC term ‘extracellular.’ The quantitative dataset was subjected to q‐value sparse filtering, global imputation, and cross‐run normalization. Pairwise comparisons were computed on this dataset using t‐tests (FDR‐corrected p‐value <0.05) within Spectronaut. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE (Perez‐Riverol et al., 2022) partner repository with the dataset identifier PXD039845. Data visualization was performed in R (Version 4.2.1) using the VennDiagram, ggplot2, and ggrepel packages to generate the Venn diagram, waterfall, and volcano plots. R code used in this paper is uploaded to Bitbucket at https://bitbucket.org/dittmerlab/heparin_exosome_figures/src/master/. The heparan sulfate binding protein database was obtained from a previous study (Gómez Toledo et al., 2021). For cellular compartmentalization analysis, proteins identified with at least three unique peptides and were enriched either in Peak 1 or Peak 2 fractions were analyzed on DAVID Bioinformatics Resources (https://david.ncifcrf.gov/home.jsp). Annotations with P‐values (Benjamini‐Hochberg correction) smaller than 0.5 were selected and ranked based on the percentage (%) of proteins, and the top 10 hits were shown.

2.13. Phospho‐ERK1/2 Western blot

hTERT‐HUVEC cells were kept in the complete EGM‐2 medium made by Endothelial Cell Basal Medium 2 (EBM‐2, PromoCell C‐22211), supplemented with Growth Medium 2 Supplement Pack (PromoCell C‐39211) and 10% EV‐depleted FBS (Avantor 97068‐085). Upon confluence, 12‐well plates were seeded with 150,000 hTERT‐HUVEC cells per well with 1 mL EBM‐2 medium supplemented with 2% EV‐depleted FBS and incubated at 37°C for 24 h. Cells were then treated with either 1 × 10⁹ particles of EVs (Input, NHB, HB) or control reagents (PBS, Complete EGM‐2 medium, 25 ng/mL of TPA) and incubated at 37°C. Cell lysates were harvested by directly lysing cells in the plate with 100 μL of cold RIPA buffer (20 mM Tris‐HCl, 150 mM NaCl, 1 mM EDTA, 1% NP‐40, 0.1% SDS, 0.5% sodium deoxycholate, pH 7.5) supplemented with protease inhibitor (Thermo Scientific 78429) and phosphatase inhibitor (Thermo Scientific 78426). Cell lysates were incubated on ice for 10 min and centrifuged at 10,000 g for 5 min before analyzing by western blot.

2.14. RNAseq

hTERT‐HUVEC cells were seeded in 6‐well plates at 600,000 cells per well with 3 mL EBM‐2 medium supplemented with 2% EV‐depleted FBS and incubated at 37°C for 24 h. PBS, Input, NHB, and HB EVs were added to each well at 3 × 10⁹ particles per well to treat the cells for 1 and 24 h before RNA extraction. Total RNA was extracted from cell pellets using Qiagen RNeasy Mini Kit (QIAGEN, 24104) per the manufacturer's instructions, including on‐column DNase digestion of DNA (QIAGEN, 79254). RNA sample concentration and quality were analyzed using the NanoDrop 1000 Spectrophotometer (ThermoFisher Scientific), Qubit RNA HS Assay Kit (Q32852) on the Qubit 3.0 Fluorometer (Invitrogen), and Infinite M200 PRO (TECAN). Sequencing and subsequent gene expression analysis were performed using AmpliSeq Transcriptome Human Gene Expression Panel, Chef‐Ready Kit (ThermoFisher Scientific, A31446). cDNA was made using SuperScript VILO cDNA Synthesis Kit (Invitrogen, 11754‐250) with 10 ng of RNA from each sample as the template. For targeted amplification, Ion Chef (ThermoFisher Scientific) was set to 15 cycles with a 16‐min anneal/extension time. Libraries generated by Ion Chef were templated onto two Ion 540 sequencing chips and sequenced by Ion S5 (ThermoFisher Scientific).

2.15. Transcriptomics analysis

Reads were mapped using the Ion Torrent Suite software, and the resulting data were processed in R (Team, 2022)(version >4.2.1). The DESeq2 package (Love et al., 2014) was used for Differential Expression Analysis; the stats package was used for PCA. Visualizations for PCA, MA plots, and volcano plots were made with the ggplot2 (Wickham, 2009) and ggpubr (Kassambara, 2020) packages. Visualizations for correlation heatmap were made with the pheatmap (Kolde, 2019) package. R code used in this paper is uploaded to Bitbucket at https://bitbucket.org/dittmerlab/heparin_exosome_figures/src/master/.

3. RESULTS

3.1. Heparin‐binding separates purified EVs into two subpopulations

Our four‐step purification pipeline generated size homogenous EVs of ∼100 nm median diameter at ≥10¹² particles/mL (McNamara et al., 2018). We define this as the input henceforth. The input was loaded onto a 1 mL heparin column at 0.5 mL/min using PBS (pH 7) as the running and washing buffer (three column volumes) and gradient eluted with a 0–1 M NaCl gradient in PBS (pH 7). A fraction of the EVs did not bind heparin and came out in the void volume. We term this the NHB population. The column‐bound fraction of EV was eluted at 100–300 mM NaCl (conductivity 20–60 mS/cm). We term this the HB population (Figures 1a and S1A). The NHB and HB combined recovery rates calculated based on particle and protein concentrations were 93% (40% for NHB and 53% for HB) and 96% (55% for NHB and 41% for HB), respectively (Table 1). A similar result was obtained with EV from a second cell line (Figure S1C). This experiment demonstrates that heparin chromatography distinguishes two EV types and that this biochemical characteristic can be exploited for large‐scale EV purification.

FIGURE 1.

Heparin chromatography separates EVs into two fractions. (a) Purified (post Capto‐Core‐700) EVs were loaded on a 1 mL Hitrap Heparin‐Sepharose column using PBS (pH 7) as the running and washing buffer. The NHB were pooled from the flow‐through fractions. HB were eluted with NaCl in PBS at a maximum concentration of 1 mol/mL in a gradient manner. A representative chromatography curve out of three biological repeats is shown. The horizontal axis shows buffer volume. The vertical axis on the left shows UV and the vertical axis on the right shows conductivity. (b) ZETA potential and (c) particle concentrations were measured by NTA. (d) Protein concentrations were determined by BCA assay. Five biological repeats were performed for ZETA potential and the p‐value was calculated by One‐way ANOVA. Three biological repeats were performed for particle concentrations and protein concentrations.

TABLE 1.

Recovery rate by heparin chromatography (n = 3)

	Volume (mL)	Particles (109 particles/mL)	Particle recovery (%)	Total protein concentration (μg/mL)	Total protein recovery (%)
Input	0.9	220 ± 10	N.A.	1620 ± 23	N.A.
NHB	3.0	26 ± 7	40 ± 9	266 ± 46	55 ± 9
HB	2.0	52 ± 7	53 ± 6	300 ± 25	41 ± 4
Total			93		96

3.2. Biophysical analysis of HB EVs

Size distribution analysis by NTA showed a size range between 50 and 150 nm with a peak of ∼100 nm for all three fractions: Input, NHB, and HB (Figure 2b bottom row, S1B, S1D). This suggests that the heparin‐binding step did not select a particular size of EVs. The ZETA potentials of Input and NHB were similar at an average of −58.2 ± 1.8 and −55.5 ± 2.7 mV, respectively, while HB was significantly more positively charged at an average of −40.9 ± 4.5 mV (Figure 1b), corresponding to the mobility results (Figure S1E). This difference in Zeta potential was consistent with differential binding to heparin, which has a net negative charge. The heparin column, similar to Capto Core 700, diluted the Input resulting in a concentration of ∼10¹¹ particles/mL (Figure 1c), a protein concentration of 700 μg/mL (Figure 1d), and a purity index of ∼10⁹ particles/μg protein (Figure S1F).

FIGURE 2.

NTA, TEM, and dSTORM report EV sizes with variations. (a) Representative TEM images of EV in the Input, NHB, and HB EV fractions. (b) Size distribution of Input, NHB, and HB EV samples determined by TEM, dSTORM, and NTA. ‘R. Gyr.’ stands for ‘radius of gyration,’ which measures the radial distance to where the inertia equaled the mass, hence roughly half the ‘true’ particle diameter. Here the R.Gyr. numbers were multiplied by two to correspond to EV diameter as reported by other methods. n: number of particles measured. d: diameter. Five biological repeats were performed for TEM and NTA. Three biological repeats were performed for dSTORM.

We used TEM to visualize the EVs directly (Figures 2a and S2A). The EVs in the input contained vesicles with characteristic cup‐shaped forms but also smaller particles and protein aggregates. The NHB fraction showed the characteristic cup‐shaped form. This is quantitated by the pronounced second peak in the TEM density distribution graph (Figure 2b, arrow). The HB fraction contained a purer fraction of particles. Note that TEM fixation and dehydration tend to result in the compaction of some vesicles but not necessarily all. This result established that adding a heparin column after prior purification removes any heparin‐binding contaminant proteins and a population of EVs that bind to heparin. This is the difference between heparin‐coupled Sepharose chromatography and conventional, uncharged Sepharose SEC. Both elute EVs in the flow through and retain smaller‐size protein aggregates. In addition, heparin‐resin retains EVs that specifically bind to heparin through a combination of charge and sugar interactions.

To confirm that both fractions contain lipid‐based EVs, we employed 2D direct stochastic optical reconstruction microscopy (dSTORM), a technique that can reach a 20 nm resolution (Chambers et al., 2021) (Figure S2B). dSTORM in this application depends on an intercalating lipid dye, Cell Mask Deep Red (CM‐DeepRed), to detect particles. It reports two parameters: the radius of gyration, which measured the radial distance to where the inertia equaled the mass, roughly half diameter, and the length, which is the furthest distance in an EV cluster of signals between two blinking events. Figure 2b middle shows the density distribution for both measurements. Unfortunately, the results were not identical. The length measurements for each fraction were ∼80 nm larger than the radial distance‐based measurements. Thus, length by dSTORM approximated the NTA result, while the radius of gyration by dSTORM approximated the TEM result. The reason for this discrepancy is under investigation by the manufacturer. Nevertheless, we conclude that input, NHB, and HB all contained lipid vesicles and that the majority population was 60–120 nm in size. Thus, it is unlikely that the HB identified here is comprised of previously described ‘supermeres’ or ‘exomeres,’ which are even smaller.

3.3. Proteomics analysis shows conventional EV protein markers in NHB

To determine the molecular composition of these EVs, Input, NHB, and HB samples were subjected to proteomics analysis by liquid chromatography‐coupled tandem mass spectrometry (LC‐MS/MS, Tables S3, S4, and S5). A total of 2606 proteins were identified. 89.1% (across all three fractions) were proteins previously identified in other human EVs (Figure S3A) and recorded in the Vesiclepedia database (Pathan et al., 2019). The experiment was conducted in biological triplicates, representing three different batches isolated three weeks apart. The biological replicates correlated well within each fraction group (Input, NHB, or HB) as determined by principal component analysis (PCA, Figure S3B) and Pearson's correlation analysis (Figure S3C), except replicate three of HB showed more variation from the other two replicates. A total of 114 (4.37%) proteins were present in the input fraction but neither in the NHB nor the HB fraction.

A total of 1497 proteins (57%) were common to Input, NHB, and HB samples. A total of 508 proteins (19%) were present within Input and NHB fractions, while 88 proteins (3%) were present in Input and HB (Figure 3a). A total of 297 (11%) proteins were only identified in NHB and 79 (3%) only in HB (Figure 3a). We assume that these were present in the Input below the cut‐off/sensitivity of MS/MS. This proteomic analysis demonstrates that an added heparin‐column purification step removes HB EVs from NHB EVs (or conversely enriches as a distinct heparin‐binding population of EV) and that the HB EVs have a different protein composition than the NHB EVs.

FIGURE 3.

Proteomics analysis revealed conventional EV markers in NHB. (a) Venn diagram comparison of proteins identified in the Input, NHB, and HB samples. (b) Volcano plot of the quantitative differences of proteins in NHB and HB. Red dots indicate proteins enriched either in NHB or HB (p‐value < 0.05, |Log2| > 0.58). Blue dots annotate the Top 100 EV proteins. (c) Waterfall plot of the relative abundance of proteins in NHB versus HB. Three biological replicate samples were used for proteomics analysis. Raw data is available for download from ProteomeXchange PXD039845. (d) Western blot confirmation of validated EV protein markers, specifically CD81, CD63, Alix, Flotillin‐2, Syntenin, Fibronectin, Histone H3. Representative images of three biological repeats were shown (complete blots are provided as supplemental material).

By differential analysis, 1,849 proteins were significantly (FDR adjusted p ≤ 0.01) enriched in NHB, and 807 proteins were enriched in HB (Figure 3b,c). Thirty‐three belong to the Top 100 previously identified EV proteins in the Vesiclepedia database (Pathan et al., 2019). They partitioned to either the NHB (21/33) or the HB (12/33) fraction, demonstrating that prior EV purification methods contained a mixture of both HB and NHB EVs.

CD9, CD63, CD81, TSG101, Alix (PDCD6IP), Syntenin (SDCBP), and Flotillin were enriched in the NHB fraction. By contrast, heparan sulfate‐binding proteins (HSBP) were enriched in HB fraction (Figure S3F, Table S5), confirming that heparin binding was the affinity interaction that separated the two EV populations. Proteins previously identified in non‐vesicle (NV) material (Jeppesen et al., 2019), such as fibronectin and histones, were enriched in HB (Figures 3b, S3D, and S3E, Table S5). The MS/MS results were validated by western blot (Figures 3d and S4). CD81, CD63, Alix, Flotillin‐2, Syntenin, as well as Fibronectin and Histone H3 were present in the input fraction, while actin, a marker for non‐specific cell debris, was not. The HB fraction was devoid of tetraspanins, Alix, Flotillin‐2, and Syntenin, but enriched in Fibronectin and Histone H3.

Cellular component analysis by DAVID Bioinformatics Resources showed ‘extracellular exosome’ as the most enriched pathway, as well as ‘extracellular region’ and ‘extracellular space’ (Figure 4). This classification affirms both fractions as containing EVs, but also highlights the limitation of broad classification schemas. NHB contained proteins mostly from lysosome and plasma membrane, corresponding to known or presumed EV biogenesis locations. HB, on the other hand, was enriched for proteins from the endoplasmic reticulum and nucleosome.

FIGURE 4.

Cellular component analysis of proteins identified in NHB and HB. Proteins enriched in either NHB or HB were analyzed by DAVID Bioinformatics Resources (https://david.ncifcrf.gov/home.jsp) and the top 10 hits are shown.

To confirm that the HB population was not an artifact and was not induced by PEG precipitation, we took medium concentrated by TFF only or purified by differential centrifugation as Input for the heparin chromatography (Figure S5). HB was present in the medium without centrifugation, in the supernatant after a 10K centrifugation, and in the pellet after a 100K UC (Figure S5B–G). The 100K EV preparation contained more protein aggregates (Figure S5G), and the size distribution was wider without 100K ultracentrifugation (Figure S5H). Thus, EVs purified by multiple methods that are standard in the field contain a population of EVs that binds to heparin.

In summary, purified EVs were separated by heparin chromatography into two populations. The EVs in the NHB displayed a marker profile, traditionally associated with EVs, while the HB consists of EVs enriched in fibronectin, and Histone H3, which were also considered among the top 100 EV‐associated proteins.

3.4. dSTORM analysis of surface markers on single particles

To overcome the unavoidable averaging and bias for the most abundant markers inherent to MS/MS proteomics, we applied 2D dSTORM analysis and investigated the marker composition on individual EVs. Input, NHB, and HB EVs were labeled with Cell Mask Deep Red (CM‐DeepRed) dye (coloured magenta) and probed for CD81 (coloured cyan), a canonical EV surface marker, and fibronectin (coloured yellow), a surface protein, which was enriched in the HB fraction. Representative images of each sample were presented in Figure 5a–c. Firstly, this analysis confirms that both the EVs in the NHB and HB contain lipids. Secondly, they confirm the enrichment of CD81‐positive, Fibronectin‐negative EVs in the NHB and conversely of CD81‐negative and Fibronectin‐positive EVs in the HB fraction. Thirdly, this experiment confirms the existence of EVs containing both CD81 and fibronectin.

FIGURE 5.

Quantitative EV surface marker profiling by dSTORM super‐resolution microscopy. Representative dSTORM images of (a) Input, (b) NHB, (c) HB EV samples labeled by CM‐DeepRed (magenta), CD81 (cyan), and fibronectin (yellow). Percentages of Fibronectin+ (Red), CD81+ (Green), and double positive (Blue) EVs in Input (d, n = 2025), NHB (e, n = 1545), HB (f, n = 1837) samples were quantified by CODI analysis (ONI Inc.).

At this level of resolution and analysis, the colocalization of two or three colours within a 200 nm diameter and distinct from any neighbouring signal is counted. Since CD81 and fibronectin were detected by antibodies, depending on the orientation of the EVs, sometimes the colour signals overlap exactly (yielding white) or are separated by 10 nm due to the size of the antibody. The sample size is too small to calculate exact enrichment scores, but this trend is consistent with the prior biochemical and proteomic analysis and suggests that CD81 and fibronectin may have utility as biomarkers to distinguish two populations of EVs.

The input sample of purified EVs contained almost equal fractions of CD81 single positive and fibronectin single positive EVs. This establishes that fibronectin‐positive EVs constitute a significant constituent of EVs purified from these cell lines. Adding a heparin column depleted the fraction of fibronectin‐positive EVs from (45% ± 6%) to (13% ± 1%). Conversely, heparin‐binding enriched fibronectin‐containing EVs from (45% ± 6%) to (58% ± 11%). Having thus established that NHB and HB EVs have distinct protein compositions by multiple orthogonal methods, we asked if they also had distinct biological properties.

3.5. Heparin‐depleted cancer EVs no longer activate ERK or Ki67

Before investigating biological phenotypes, we needed to establish that the EVs were fusogenic. Input, NHB, and HB EVs were loaded with CM‐DeepRed dye, quantitated, diluted, and added to human telomerase reverse transcriptase (hTERT) overexpressing human umbilical vein endothelial (hTERT‐HUVEC) cells. Endothelial cells line the blood and lymphatic system and are exposed to the highest physiological concentrations of EVs in the human body. The hTERT‐HUVEC are immortalized, but not cancerous. They do not form tumours in nude mice; they have low basal activation of the receptor tyrosine kinase and other cellular activation markers but are exquisitely sensitive to growth factors (McNamara et al., 2019). Images were taken upon 24‐hour incubations (Figure 6a). They showed equal and consistent EV uptake under our experimental conditions (multiplicity of infection (MOI) equivalent ∼1000:1). A dose‐dependent cellular intake of EV was verified by flow cytometry (Figure S6). This demonstrates that fusogenic activity was maintained throughout the purification and was equivalent across the three populations.

FIGURE 6.

Heparin fractionates EVs into two biologically active but functionally distinctive populations. (a) Representative immunofluorescent images of hTERT‐HUVEC cells treated with CM‐Red dye labeled Input, NHB, and HB EV for 24 h. Nuclei were labeled by DAPI (blue). Presence of CM‐Red dye (yellow) indicates EV taken in by cells. (b) WB of phosphorylated ERK (top) and total ERK (bottom) upon Input, NHB, and HB EV treatment for 1, 4, 8, and 24 h. (c) Quantification of phosphorylated ERK intensity on WB upon 24 h of EV treatment from three independent experiments. (d) Representative immunofluorescent images of hTERT‐HUVEC cells treated with PBS, Input, NHB, and HB EVs for 24 h. Ki67, nuclei (DAPI), and composite images were shown. (e) Percentage of Ki67 positive cells upon different treatments. ****p < 0.0001, ***p < 0.0005 by T‐Test.

U2OS cells can activate the mitogen‐activated protein kinase (MAPK)/extracellular signal‐regulated kinase (ERK1/2) pathway (Abarrategi et al., 2016; Ara & Declerck, 2010) to stimulate self (autocrine) and neighbouring (paracrine) cell proliferation (Roberts & Der, 2007). EVs isolated from BCBL‐1 cells have the same phenotype (McNamara et al., 2019). Hence, we sought to investigate if the EV populations differed by their ability to stimulate ERK1/2 phosphorylation and Ki67 expression in hTERT‐HUVEC cells. Under nutrient‐limiting conditions, ERK1/2 is not phosphorylated in hTERT‐HUVEC cells (Figure S7A), and the Ki67 expression level is low (Figures 6d,e, S8A–D). After adding a medium containing growth factor and tetradecanoylphorbol acetate (TPA), a small molecule drug that activates PKC and the MAPK/ERK1/2 pathway, ERK1/2 was rapidly phosphorylated (Figure S7A) and Ki67 was induced (Figure S8E–H), confirming that the MAPK/ERK1/2 pathway and cell proliferation marker Ki67 was functional in hTERT‐HUVEC cells.

Purified EVs (Input) activated ERK1/2 (Figure 6b–c) and induced Ki67 expression (Figure 6d–e), recapitulating our prior results and those of others previously reported (McNamara et al., 2019; Zhou et al., 2019). By contrast, the NHB EV population, which was highly enriched in tetraspanins (CD9, CD63, CD81), did not activate ERK1/2 nor induce Ki67 expression in this 24‐h time course experiment. All the ERK1/2 and Ki67 stimulating activity segregated with the fibronectin‐enriched HB fraction (Figure 6). Thus, adding a heparin‐column step to EV purification protocols depletes mitogenic EVs.

3.6. Transcriptomics reveal differences in endothelial cells treated by NHB and HB

To investigate how Input, NHB, and HB modulate gene expression on a broad scope, we performed RNAseq on EV‐treated hTERT‐HUVEC cells at post 1 and 24 h of treatment. We monitored to ensure RNA and sequencing read quality (Table S6). Correlation analysis (Figure S9) shows distinct differences between 1‐h and 24‐h samples, indicating a change in cell conditions over this time course. At each time point, cells without EV treatment (PBS) were different from the EV (Input, NHB, and HB) treated samples based on PCA, and NHB and HB treatment cells were also different (Figure 7a,b). Out of 32 samples, three (1hr_PBS_1, 1hr_PBS_2, 24hr_PBS_4) were the most variable (Figure S10) and thus were not included in the subsequent analysis.

FIGURE 7.

Principal component analysis (PCA) and expression fold change of transcriptomics in hTERT‐HUVEC cells upon EV treatment. hTERT‐HUVEC cells were treated with PBS, Input, NHB, and HB EVs and subjected to transcriptomics analysis by RNAseq. Each treatment included four replicates. After 1 h (a) and 24 h (b) of treatment, data were subjected to PCA. Expression fold change between NHB and HB treated cells were represented in (c). Red and blue dots indicate mRNA up‐regulated in HB and NHB, respectively (p‐value < 0.05, |Log2| > 0.5). For comparison, Interferon‐stimulated genes (ISG) were annotated by gene names. Fold changes between 1 and 24 h were represented in (d). Red and blue dots indicate mRNA up or down‐regulated after 24 h of treatment (FDR < 0.05, |Log2| > 0.58). Data and the R codes used for analysis are available on Bitbucket (see Section 2).

Between NHB and HB, at 1 h, 33 genes were up‐regulated by HB (Log2 >0.5), and 27 were up‐regulated by NHB (Log2 < ‐0.5). Three interferon‐stimulated (ISG) genes (Schoggins et al., 2011) PMM2, JUNB, and SIRPA, were included as indicated on the plot. At 24 h, the numbers were reduced to nine genes up‐regulated by HB and one gene up‐regulated by NHB (Figure 7c, Table S7). From 1 to 24 h post‐treatment, without EV (PBS), 1343 genes were up‐regulated, and 1548 were down‐regulated (Figure S11A), indicating a natural change of transcriptomics profile over treatment time in the negative control. To focus on EV‐related change in transcriptomics, we removed the significantly changed genes in PBS and still observed clusters of up and down‐regulated genes (Figure 7d, Table S8). Based on the Mean Expression level, significant genes can be further divided into low expression (Fold Change <5), medium expression (Fold Change 5–10), and high expression (Fold Change >10) categories (Figure S11B, Table S8). Each category included genes that were either up or down‐regulated by overlapping treatments (Figure S11B). Genes affected by all three EVs (Input, NHB, and HB) are responsive to generic EV components. Those only affected by Input are likely due to contaminants in the initial EV sample (Input). Those affected by both NHB and HB can be due to purification‐enriched components (Figure S11B, Table S8).

4. DISCUSSION

To summarize, adding a heparin column as an additional purification step to any established method increases the purity of EVs. EVs eluting in the void volume were positive for CD63, CD81, and CD9 tetraspanins. Compared to a non‐charged Sepharose SEC resin (via size), the heparin‐coupled Sepharose resin (via charge and sugar interactions) removed additional complexity, that is, a specific population of heparin‐binding EVs as well as additional non‐EV contaminants. Possible non‐EV contaminants, depending on the input source, are soluble growth factors, lipoproteins, and residual nucleic acids, which have been shown to bind tightly to heparin. These also include nuclear proteins, such as histones, which is expected since heparin fractionation is used routinely as a final step in nuclear transcription factor purification. Heparin‐binding EVs include EVs that are enriched in fibronectin. Unlike the EV input, the NHB fraction no longer had an acute mitogenic activity or induced ERK1/2 signaling.

EV manufacturing for therapeutic applications is challenged, first, by the heterogeneity of the sample and, second, by the difficulty of adapting traditional purification processes, such as UC or gradient flotation (Onódi et al., 2018) to an industrial scale, even though these techniques yield highly purified EVs at laboratory scale and are considered best practices by many. These two properties represent significant barriers to the therapeutic use of EVs as drug carriers, vaccines, or gene/protein delivery platforms. The problem with scaling EV production prompted us and others (McNamara et al., 2018; Visan et al., 2022) to explore non‐centrifugation‐based methods, such as TFF, PEG precipitation, and liquid chromatography. In addition to being scalable, these methods avoid potentially damaging g‐forces associated with UC (Dong et al., 2020; Foers et al., 2018; McNamara et al., 2018). Antibody‐based affinity purification on tetraspanin capture beads yields the most specific tetraspanin‐positive EVs for diagnostic purposes, but the scalability is still price prohibitive at the moment. ‘Semi‐affinity’ resins, such as heparin‐Sepharose, provide one possible alternative.

The physiological relevance of this study stems from the observation that HB and NHB EVs have different biological activities. Input and HB induced ERK1/2 phosphorylation and Ki67 expression, but the NHB did not. ERK1/2 phosphorylation is part of the MAPK pathway that controls cell proliferation and differentiation. Ki67 is a clinically validated cell proliferation marker. Furthermore, NHB and HB induced different mRNA profiles in target cells (Figure 7), although the fold change of significantly different genes was not large. Some prior studies reported that soluble heparin blocked EV uptake and the EV‐induced biological phenotype, while others have found no effect of heparin on EV‐induced phenotypes (Christianson et al., 2013; Usman et al., 2018). This is analogous to the experience with viruses. It has long been recognized that some enveloped viruses use heparin as an attachment enhancer, while others do not (Akula et al., 2001; Shukla et al., 1999). We posit likewise, that some EV populations bind heparin, others do not, and that those have different biological phenotypes. Alternatively, unknown impurities that co‐purify with EVs by standard methods may confound the physiological readouts previously attributed to EVs. These could not be small proteins as our purification process, analogous to differential centrifugation, removes proteins, lipids, or nucleic acids smaller than 750 kDa.

This discovery has important implications for the therapeutic applications of EVs. Adding a heparin column chromatography depletion step should yield EVs less prone to induce adverse and acute inflammatory reactions in clinical applications. Heparin‐purified EVs should show fewer complex phenotypes, which are more dependent on specific EV cargo than co‐purifying fibronectin‐containing EVs. Conversely, heparin column chromatography identified a highly homogenous population of fibronectin high EVs for further study and with exploitable phenotypes for human applications.

One protein enriched in the HB fraction stood out. Fibronectin is an important mediator of cellular interactions in the extracellular matrix. Usually existing as a dimer of two 250 kDa subunits, fibronectin is a large protein that binds to integrin, collagen, fibrin, and heparan sulfate (Pankov & Yamada, 2002). Fibronectin has a known association with histone proteins (Foers et al., 2018; Jeppesen et al., 2019), which is difficult to reconcile given the different cellular locations of these proteins. Fibronectin is regularly detected in crude EVs (Théry et al., 2018). It has been categorized by some as a co‐purified contaminant. Others consider it a bona fide component of EVs and, particularly in cancer‐EVs, ascribe biological phenotypes to it (Sung et al., 2015; Xavier et al., 2021). Now we can study fibronectin‐enriched CD81‐positive EVs separately from fibronectin‐depleted, CD81‐positive EVs, as well as particles that are fibronectin‐positive but tetraspanin negative.

There are limitations to this study. We were concerned that the particular EV purification pipeline artificially induced the HB fraction. This is unlikely. The HB EVs are contained in EVs purified by TFF alone without the PEG concentrated step, and we know that they are contained in EVs purified by differential UC (Figure S5) (Jeppesen et al., 2019; Zhang et al., 2018, 2021, 2023). Whereas we typically treat all EVs with DNAse and RNAse, we observed both NHB and HB EVs on samples not treated by DNase or RNase. At least formally, there is the possibility that DNAse and RNAse treatment unmasks a heparin‐binding activity as both DNA and RNA are negatively charged sugars, and EVs surrounded by a corona of nucleic acids (Buzas, 2022) may not bind to heparin as much. This is an understudied topic. Prior work using heparin for EV purification (Balaj et al., 2015; Barnes et al., 2022) was not designed to detect different phenotypes between EVs binding to heparin and those that do not. Thus, this work extends prior efforts. An analysis of nucleic acid corona requires the ability to investigate individual EVs by EM or dSTORM. Those tools have only recently become accessible.

We were dismayed to experience how little the different size estimation techniques concurred in the EV diameter estimates. This issue needs further study. We can say for sure, however, that none of the EVs studied here were larger than 220 nm. All sizing methods concurred up to their limits of sensitivity, and we used two filtration steps with 450 and 220 nm mean pore size before TFF. We can say for certain that none of the EV preparations studies here contained proteins of >800 kDa, which was the molecular weight cut‐off for the TFF (750 kDa, see Materials and Methods) as well as the CaptoCore multimodal resin (700 kDa, see Materials and Methods). This excludes most growth factors and secreted proteins. There is the potential that large protein aggregates (>800 kDa and <220 nm) are retained throughout this procedure. Depending on their surface composition, they could be present in both the NHB and HB populations.

The use of heparin‐based affinity purification of EV has been suggested previously (Balaj et al., 2015; Barnes et al., 2022; Reiter et al., 2019) but on a limited scale and without detailed investigations on EV functions. Balaj et al. (2015) employed bead‐based heparin affinity purification after an initial clarification step by centrifugation. We independently confirmed their work; however, centrifugation alone yields a very crude EV preparation. In our hands, additional filtration and multimodal chromatography steps are needed before any affinity‐based method. Centrifugation severely restricts the scale of EV isolation and increases the cost of GMP‐compliant manufacturing. A large effort in the manufacture of biologicals, which includes vaccines and gene therapy particles, is devoted to avoiding or replacing centrifugation steps. Responding to this challenge, Barnes et al. (2022) used TFF as the first step to isolate EVs before heparin chromatography. Reiter et al. (2019) used a Capto Core‐700 – Heparin chromatography method, but without prior TFF, thus lacking the concentrating step. When we and others validated TFF as a method of EV purification before (Heath et al., 2018; McNamara et al., 2018), it was apparent that both an additional concentration step and a column step were needed to obtain concentrated and pure EV. We used a multimodal resin (Capto Core), and now report that even greater EV purity can be achieved by adding heparin as a third step, which in the case of the HB population is a bind‐elute step, rather than a flow‐through step.

Barnes et al. (2022) found EVs containing all three tetraspanin proteins (CD63, CD81, CD9) binding to heparin. We also found CD81+ EVs in the heparin‐binding fraction (Figure 5), but they were not as enriched by comparison to fibronectin+ and tetraspanin‐ EVs. The discrepancy might be due to different EV inputs. Barnes et al. used a neural stem cell line while this report used cancer cell lines. Different cell lines yield EVs with different surface compositions thus affecting their overall heparin‐binding affinity. Further studies are needed to understand, EVs of which cell lineage binds to heparin with high affinity and which do not, which EVs are pyrogenic, and which are not.

Heparin is not a pure compound but a mixture of heterogeneous, complex sugars, so it is hard to define the specific interactions that are involved between EVs and heparin. Part of it is charge‐based since heparin is negatively charged. Affinity interactions mediated by specific residues are also possible, like those between heparin and antithrombin. HSBPs were enriched in the HB EVs and represent the most likely candidates to mediate this interaction. These interactions may contribute to heparin‐induced EV aggregation and blockage in transfer (Atai et al., 2013). Sugar and charge‐based interaction may be common in EV adhesion, fusion, and uptake. Exomeres were similarly more positively charged (Zhang et al., 2018). Others have used chitosan to isolate EV (Kumar et al., 2021). Chitosan is positively charged and likely to isolate a different population of EVs. The authors, however, did not ascribe different biological functions to the chitosan‐binding and non‐binding EVs.

A concept to consider for future studies is that of the EV corona (Buzas, 2022). The presumed EV corona is a layer of biomolecules that spontaneously form on the outside of the EV membrane. These may include proteoglycans, such as heparan‐sulfate proteoglycans (HSPG), that is, syndecan‐1 (SDC‐1) or glypican‐1 (GPC‐1). The EV corona may contain protein, nucleic acids, and lipids that are loosely associated with EV and sensitive to degradation by DNases, RNases, and proteases. Several proteins attributed to being part of the EV corona were detected in the HB, such as TGFβ, ApoB, ApoD, and ApoC3 (Figure S3F) (Buzas, 2022). We would not have detected nucleic acid compounds because of DNAse and RNAse treatment before chromatography. It may be that the heparin ligand on the column competes with EVs for EV corona proteins and selectively retains components of EV corona or EVs with extensive corona on the column. More research is needed to investigate this hypothesis, but heparin‐chromatography followed by SEC may have utility in enriching or depleting EVs with an extensive corona.

In summary, we demonstrate that heparin‐binding is a phenotype of some EVs, but not others and that heparin‐affinity chromatography is an easy, cost‐effective, and scalable step to reduce the complexity of EVs for subsequent research and therapeutic applications. Heparin‐chromatography separated the two populations of cancer cell‐derived EVs. These two populations had otherwise identical biophysical characteristics. NHB EVs, which are not mitogenic and do not induce ERK1/2 phosphorylation or Ki67 activation, and HB EVs, which induce ERK1/2 phosphorylation and Ki67 activation, are depleted for tetraspanin markers and enriched for fibronectin and histone.

AUTHOR CONTRIBUTIONS

Yijun Zhou: Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Project administration; Resources; Software; Supervision; Validation; Visualization; Writing—original draft; Writing—review & editing. Runjie Yuan: Conceptualization; Data curation; Formal analysis; Methodology; Software; Validation; Visualization; Writing—original draft; Writing—review & editing. Allaura S. Cone: Data curation; Formal analysis; Investigation; Methodology; Software; Supervision; Visualization. Kyle W. Shifflett: Data curation; Formal analysis; Methodology; Software; Visualization. Gabriel F. Arias: Data curation; Formal analysis; Methodology; Software; Visualization. Alice Peng: Formal analysis; Methodology; Software; Visualization. Meredith G. Chambers: Data curation; Formal analysis; Software; Visualization. Ryan P. McNamara: Methodology; Supervision. Smaranda Willcox: Data curation; Methodology; Visualization; Writing—original draft. Justin T. Landis: Formal analysis; Methodology; Software; Visualization. Yue Pan: Formal analysis. Jack D Griffith: Funding acquisition; Supervision. Dirk Dittmer: Conceptualization; Data curation; Formal analysis; Funding acquisition; Investigation; Methodology; Project administration; Resources; Software; Supervision; Validation; Visualization; Writing—original draft; Writing—review & editing.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Supplemental Figure 1. Heparin column reproducibly separates EV into NHB and HB subpopulations

Supplemental Figure 2. Representative examples of EV imaged by TEM and dSTORM for size measurements.

Supplemental Figure 3. Additional analysis of proteomics data

Supplemental Figure 4. Images of the representative western blot results shown in Figure 3D.

Supplemental Figure 5. EV isolation by differential centrifugation after TFF

Supplemental Figure 6. Dose‐dependent EV intake by HUVEC cells

Supplemental Figure 7. ERK phosphorylation image of complete blots

Supplemental Figure 8. Ki67 activation by heparin fractionated cancer EV

Supplemental Figure 9. Correlation map for principle component analysis of transcriptomics data

Supplemental Figure 10. Principle component analysis of transcriptomics data

Supplemental Figure 11. Up and down‐regulated genes in cells treated by EV for 1 hour versus 24 hours

Supplementary Table 1. List of Antibodies

Supplementary Table 2. Parameters used for 2D dSTORM sizing analysis by CODI

Supplementary Table 3. Qualitative search results from MS analysis

Supplementary Table 4. Quantitative search results from MS analysis

Supplementary Table 5. List of genes in volcano plots from proteomics analysis

Supplementary Table 6. RNAseq Quality Control

Supplementary Table 7. HB vs NHB Significantly Changed Genes at 1h and 24h.

Supplementary Table 8. Significantly changed genes 1h vs 24 h.

Zhou, Y. , Yuan, R. , Cone, A. S. , Shifflett, K. W. , Arias, G. F. , Peng, A. , Chambers, M. G. , McNamara, R. P. , Willcox, S. , Landis, J. T , Pan, Y. , Griffith, J. , & Dittmer, D. (2023). Large‐scale heparin‐based bind‐and‐elute chromatography identifies two biologically distinct populations of extracellular vesicles. Journal of Extracellular Vesicles, 12, e12327. 10.1002/jev2.12327