Comparative Assessment of Methods for Isolation and Imaging of Extracellular Vesicles Authors: Edward Carson, Yue Xu, Muthulekha Swamydas, Grace Duginski, Viviane Luangphakdy, Hannah Simmons, Vijay Krishna, Ronald Midura, George Muschler Published: 2026 ABSTRACT ABSTRACTExtracellular vesicles (EVs) from defined cell sources are emerging as potential vehicles with therapeutic efficacy. The EV and non‐EV secretome of individual cell populations may also be used as a measure of cell identity or state. EV attributes and composition may therefore be useful in defining the “fit‐for‐purpose” quality of cell populations for drug testing, biological models, and cell therapy products. As such, methods for cost‐effective isolation and rigorous characterization of EVs are needed. This work aims to evaluate current methods and establish effective and potentially improved techniques to address these needs.First, ultracentrifugation, ultrafiltration, and precipitation methods were evaluated as means for EV isolation from culture‐expanded human mesenchymal stromal cell (hMSC) populations. Second, traditional formvar and updated lacey carbon transmission electron microscopy (TEM) grid preparations were compared. Third, electron energy loss spectroscopy (EELS) was evaluated for EV elemental content characterization. Fourth, the efficacy of dynamic light scattering (DLS) was evaluated as a tool for characterizing variation in EV size distribution between hMSC populations. Finally, DLS measurements were compared to direct TEM size measurements using EV populations isolated from 28 culture‐expanded hMSC populations derived from 5 human donors.Ultracentrifugation was most effective in providing higher EV purity and yield per visual TEM analysis. Lacey carbon TEM grids provided the most effective imaging for visual characterization of EV morphology and enabled durable archiving of samples for discontinuous assessment. EELS identified phosphorus in the lipid bilayer but was ineffective at identifying sulfur (protein) and nitrogen (protein, RNA, DNA) contents of EVs. DLS was generally consistent in characterizing EV population size distributions, but discrepancies and limitations were identified based on direct comparative TEM observation.This study demonstrated that ultracentrifugation, DLS size characterization, and confirmation with lacey carbon TEM grid preparation provided high quality samples and effective assessment of EV size distribution and morphology. Introduction As the potential of extracellular vesicles (EVs) in cell‐based manufacturing, cell‐based models, and cellular therapeutics emerges, so does the need for practical, rigorous, and reproducible methods for EV isolation and characterization (Giebel and Lim 2025). Established isolation methods have included density‐based isolation, size‐based isolation, precipitation, and targeted affinity (Ayala‐Mar et al. 2019; Doyle and Wang 2019; Hou et al. 2019; Konoshenko et al. 2018; Kurian et al. 2021; Sidhom et al. 2020; Théry et al. 2018; Welsh et al. 2024; Yang et al. 2020). Each of these methods differ in relative yield and purity of the end product. It should be noted that comparisons of yield and purity of EV isolation methods in literature have largely relied on relative comparisons without absolute numerical standards. Density‐based isolation includes methods such as ultracentrifugation and density gradients. Differential ultracentrifugation consists of sequentially increasing centrifugation forces to ≥ 100,000 × the force of gravity. Cell components and debris are removed first, followed by isolation of EVs at 100,000 × g. Differential ultracentrifugation has often been considered the gold standard for isolating EVs due to its ability to produce higher yields with minimal debris (Ayala‐Mar et al. 2019; Doyle and Wang 2019; Hou et al. 2019; Konoshenko et al. 2018; Kurian et al. 2021; Sidhom et al. 2020; Théry et al. 2018; Welsh et al. 2024; Yang et al. 2020). Density gradient methods utilize differences in density to isolate EVs from other media components. Density gradients tend to produce lower yields with varying degrees of purity due to the inability to fully remove basal liquid (Ayala‐Mar et al. 2019; Doyle and Wang 2019; Hou et al. 2019; Konoshenko et al. 2018; Kurian et al. 2021; Sidhom et al. 2020; Théry et al. 2018; Welsh et al. 2024; Yang et al. 2020). Size‐based isolation includes methods such as ultrafiltration and size‐exclusion chromatography. Ultrafiltration consists of specialized filter membranes that, when centrifuged at low speeds, can separate EVs from the remaining media supernatant components. This method is reported to provide similar purity to density gradients, but with lower yields than ultracentrifugation (Ayala‐Mar et al. 2019; Doyle and Wang 2019; Hou et al. 2019; Konoshenko et al. 2018; Kurian et al. 2021; Sidhom et al. 2020; Théry et al. 2018; Welsh et al. 2024; Yang et al. 2020). Similarly, size exclusion chromatography utilizes a porous substrate to separate EVs from other media components and does not require centrifugation. This method is reported to provide higher purity, but the yield is reported to be less than ultracentrifugation (Ayala‐Mar et al. 2019; Doyle and Wang 2019; Hou et al. 2019; Konoshenko et al. 2018; Kurian et al. 2021; Sidhom et al. 2020; Théry et al. 2018; Welsh et al. 2024; Yang et al. 2020). Precipitation methods often use water‐soluble polymers such as polyethylene glycol (PEG) to increase the osmotic pressure of the media supernatant solvent, causing water molecules to move towards the PEG and allow the EVs to precipitate out of solution. This method can produce higher yields; however, the purity is lower than other methods (Ayala‐Mar et al. 2019; Doyle and Wang 2019; Hou et al. 2019; Konoshenko et al. 2018; Kurian et al. 2021; Sidhom et al. 2020; Théry et al. 2018; Welsh et al. 2024; Yang et al. 2020). Targeted affinity includes methods such as immunoaffinity chromatography and microfluidics. Immunoaffinity chromatography is a type of liquid chromatography that utilizes antibody‐coated magnetic beads or other agents to bind to surface markers antigens to target and separate EVs from other media components. Although this method is reported to produce higher purity isolations, the yield is lower in comparison to other methods (Ayala‐Mar et al. 2019; Doyle and Wang 2019; Hou et al. 2019; Konoshenko et al. 2018; Kurian et al. 2021; Sidhom et al. 2020; Théry et al. 2018; Welsh et al. 2024; Yang et al. 2020). Microfluidics often utilizes targeted affinity for EV separation, but may also utilize size and density for separation of EVs (Ayala‐Mar et al. 2019; Doyle and Wang 2019; Hou et al. 2019; Konoshenko et al. 2018; Kurian et al. 2021; Sidhom et al. 2020; Théry et al. 2018; Welsh et al. 2024; Yang et al. 2020). This method similarly can produce high purity, but low yield (Ayala‐Mar et al. 2019; Doyle and Wang 2019; Hou et al. 2019; Konoshenko et al. 2018; Kurian et al. 2021; Sidhom et al. 2020; Théry et al. 2018; Welsh et al. 2024; Yang et al. 2020). In considering options for isolation and assessment in this study, we focused on ultracentrifugation, ultrafiltration, and precipitation methods with higher reported purity, as well as higher reported yield. These attributes were most important to enable direct comparison of variation between EVs and the non‐EV secretome derived from different hMSC populations. Next, we focused on identification of the most effective methods for assessing the size, morphology, and cargo composition of the EVs from different hMSC populations. Traditional methods for visualization of EVs include TEM and DLS. Both of these methods can be limited by confounding factors from processing procedures that compromise the quality of visualization and measurement (Khan et al. 2022; Pascucci and Scattini 2021; Szatanek et al. 2017; Théry et al. 2006; Zhang et al. 2020). DLS requires a suspension of particles of interest in fluid medium and estimation of hydrodynamic size based on light scattering related to Brownian motion. DLS does not provide quantitative information about the number of particles assessed and may measure both particles of interest as well as debris (Khan et al. 2022; Szatanek et al. 2017; Zhang et al. 2020). TEM provides direct visualization of individual EVs on a grid surface, allowing a user to exclude non‐EV materials or aggregates. However, measurement requires capturing and processing many fields of view to acquire a sufficient data set for population assessment (Khan et al. 2022; Szatanek et al. 2017; Zhang et al. 2020). Traditional TEM grid preparation (i.e., fixation and negative staining on formvar) can affect EV number and morphology due to the harsh chemical processing that may disturb EV structure and may result in contamination artifacts (Pascucci and Scattini 2021; Théry et al. 2006). We explored an alternative TEM preparation method using lacey carbon grids without fixation and negative staining. These grids have long been utilized in applications for non‐biologic nanomaterials (Hofstraat et al. 2025; Radnóczi et al. 2016). However, lacey carbon grids have not been widely utilized for the presentation of biologic preparations of EVs. Additionally, this study challenged the perception that fixation is necessary to preserve the integrity of biologic samples. Finally, we investigated the prospect of utilizing EELS for characterizing EV elemental contents. EELS has the potential to identify localized content by measuring electron energy loss across the spatial area of an EV (Aronova and Leapman 2012; Dumaresq et al. 2024). However, this method comes with a potential disadvantage of damaging the EVs themselves due to the EVs being exposed to a more concentrated beam of energy (Aronova and Leapman 2012). Methods EV Isolation EV isolation methods were evaluated by parallel processing of conditioned media from an initial human bone marrow derived sample of culture‐expanded mesenchymal stromal cells (hMSCs). A total of 150 mL of α‐MEM‐based media with 10% fetal bovine serum was collected from 14.8 million hMSCs at the time of media change and separated into 3 × 50 mL aliquots designated for ultracentrifugation, ultrafiltration, and precipitation EV isolation methods respectively (detailed methods below). The EVs isolated using each method were assessed via TEM analysis. After establishing ultracentrifugation as the preferred method, a total of 31 EV populations were isolated from 31 passage 3 and passage 5 hMSC populations. To avoid contamination from exogenous serum‐derived proteins and EVs, cells were cultured for 48 hours in STEMCELL Technologies (Vancouver, BC) Mesencult ACF serum‐free media. The 31 hMSC populations were derived from 5 separate human patients. Sixteen of the 31 EV populations were polyclonal, derived from competitive expansion of the progeny resulting from mixed colony founding connective tissue progenitors (CTPs) in the starting cell population (Mantripragada et al. 2021; Muschler et al. 2003). Fifteen of the 31 EV populations were clonal and represented the isolated expansion of cells derived from a single colony founding CTP. To explore the possible effects from patient, expansion method (polyclonal vs. clonal), and batch‐to‐batch variation, EV morphology and composition were evaluated via DLS, TEM, and EELS. Ultracentrifugation The ultracentrifugation method was adapted from Current Protocols in Cell Biology (2006) (Théry et al. 2006). The published protocol defines a sequence of 5 centrifugation steps of increasing g‐force to eliminate debris and effectively isolate the EVs. To start, 50 mL of the conditioned media was placed in 4 × 15 mL conical Falcon tubes and centrifuged via Beckman Coulter (Indianapolis, IN) GH‐3.8 swinging bucket rotor at 300 × g (1,250 rpm on Beckman Coulter Allegra 6) for 10 min at room temperature (25°C) to remove any suspended cells from the media supernatant. The media supernatant was separated from the pellet, placed in new 4 × 15 mL conical Falcon tubes and centrifuged via Beckman Coulter GH‐3.8 swinging bucket rotor at 2,000 × g (3,150 rpm on Beckman Coulter Allegra 6) for 10 min at 25°C to remove any remaining dead cells from the media supernatant (Note that this second centrifuge step was unproductive and removed in subsequent protocol iterations for the set of 31 EV populations). The media supernatant was again separated from the pellet and 22 mL total was placed in 2 × 12.5 mL Beckman Coulter polypropylene tubes, and then centrifuged via Beckman Coulter SW‐41 swinging bucket rotor at 10,000 × g (7,600 rpm on Beckman Coulter XE‐90) for 30 min at 4°C to remove any remaining cell debris. The remaining media supernatant was separated from the pellet, placed in new 2 × 12.5 mL Beckman Coulter polypropylene tubes, and ultracentrifuged via Beckman Coulter SW‐41 swinging bucket rotor at 100,000 × g (24,100 rpm on Beckman Coulter XE‐90) for 70 min at 4°C to isolate the EVs from the media supernatant in pellet form. The EV pellet was subsequently washed via the addition of 10 mL of PBS per tube and the same tubes were ultracentrifuged a second time using the Beckman Coulter SW‐41 swinging bucket rotor for a final 100,000 × g (24,100 rpm on Beckman Coulter XE‐90) for 70 min at 4°C to form a purified EV pellet. Ultrafiltration The ultrafiltration method was adapted from Sigma Aldrich (St. Louis, MO) for the Amicon Ultra‐15 Centrifugal Filters Ultracel‐100K (100 kDa membrane pore cut‐off) (Extracellular Vesicle Protocol 2021). The adapted protocol also uses a sequential centrifugation process to isolate the EVs through the filter membrane but does not reach ultracentrifugation speeds. Note that 2 mL of PBS must be added to the Ultra‐15 filter tube and centrifuged at 4,000 × g (5,000 rpm on Beckman Coulter J2‐MC) for 10 min at 4°C to prime the filter with liquid to optimize filtration flow before use with sample. To start, 50 mL of the conditioned media was initially placed in 4 × 15 mL conical Falcon tubes and centrifuged via Beckman Coulter GH‐3.8 swinging bucket rotor at 300 × g (1,250 rpm on Beckman Coulter Allegra 6) for 10 min at 25°C to remove any suspended cells from the media supernatant. The media supernatant was separated from the pellet, placed in new 4 × 15 mL conical Falcon tubes and centrifuged via Beckman Coulter GH‐3.8 swinging bucket rotor at 2,000 × g (3,150 rpm on Beckman Coulter Allegra 6) for 10 min at 25°C to remove any remaining dead cells from the media supernatant (similar to the ultracentrifugation method, this step was removed in subsequent protocol iterations due to established redundancy). Therefore, these initial steps were identical to the ultracentrifugation method above. The media supernatant was again separated from the pellet and 22 mL total was placed in 2 × 12.5 mL Beckman Coulter polypropylene tubes, and then centrifuged via Beckman Coulter SW‐41 swinging bucket rotor at 10,000 × g (7,600 rpm on Beckman Coulter XE‐90) for 30 min at 4°C to remove any remaining cell debris. After priming, 15 mL of the remaining supernatant was transferred to the 15 mL Ultracel‐15 filter tube and centrifuged via Beckman Coulter JA‐12 fixed angle rotor at 4,000 × g (5,000 rpm on Beckman Coulter J2‐MC) for 30 min at 4°C to separate the EVs from the media supernatant. The filtered media was then removed from the bottom of the Ultracel‐15 and the EV retentate in the filter was washed with 14 mL of PBS and centrifuged via Beckman Coulter JA‐12 fixed angle rotor at 4,000 × g (5,000 rpm on Beckman Coulter J2‐MC) for 30 min at 4°C. The filtered PBS wash was then removed from the bottom of the Ultracel‐15 and the EV retentate in the filter was washed 2 more times with another 14 mL of PBS and centrifuged via Beckman Coulter JA‐12 fixed angle rotor at 4,000 × g (5,000 rpm on Beckman Coulter J2‐MC) for 15 min at 4°C to ensure the maximum amount of media supernatant was removed from the EV retentate left in the filter. Precipitation The precipitation method was adapted from Forever Labs (Ann Arbor, MI) protocol for use with SuperShot Polyethylene Glycol/Dextran (PEG/DEX) precipitation solution (Hozeska‐Solgot 2020). Similarly, the adapted protocol also uses a sequential centrifugation process to isolate the EVs via precipitation but does not reach ultracentrifugation speeds. Note that the 10X SuperShot solution must be vortex mixed prior to addition to the media supernatant. To start, 50 mL of the conditioned media was initially placed in 4 × 15 mL conical Falcon tubes and centrifuged via Beckman Coulter GH‐3.8 swinging bucket rotor at 300 × g (1,250 rpm on Beckman Coulter Allegra 6) for 10 min at 25°C to remove any suspended cells from the media supernatant. The media supernatant was separated from the pellet, placed in new 4 × 15 mL conical Falcon tubes and centrifuged via Beckman Coulter GH‐3.8 swinging bucket rotor at 2,000 × g (3,150 rpm on Beckman Coulter Allegra 6) for 10 min at 25°C to remove any remaining dead cells from the media supernatant (similarly to the above methods, this step was removed in subsequent protocol iterations due to established redundancy). Again, these initial steps were identical to the prior two isolation methods. The media supernatant was again separated from the pellet and 22 mL total was placed in 2 × 12.5 mL Beckman Coulter polypropylene tubes, and then centrifuged via Beckman Coulter SW‐41 swinging bucket rotor at 10,000 × g (7,600 rpm on Beckman Coulter XE‐90) for 30 min at 4°C to remove any remaining cell debris. Next, 2 mL of the 10X SuperShot solution was added to 18 mL of the media supernatant in a 50 mL Falcon conical tube (1:10 dilution of the SuperShot) and centrifuged via Beckman Coulter JA‐12 fixed angle rotor at 2,700 × g (4,200 rpm on Beckman Coulter J2‐MC) for 4 min at 4°C to precipitate out the EVs. The media supernatant was then removed, leaving the EV pellet. TEM Assessment Using Traditional Formvar vs. Simplified Lacey Carbon TEM Grid Preparation for EVs Traditionally, standard TEM preparation of biologic EV samples has consisted of fixation on formvar carbon grids with negative staining (Pascucci and Scattini 2021; Théry et al. 2006). This study compared the standard formvar carbon preparation with a simplified lacey carbon grid preparation (Pascucci and Scattini 2021; Théry et al. 2006). To compare grid preparations with a consistent isolation method, ultracentrifugation was used for EV isolation from an initial human bone marrow derived sample of culture‐expanded hMSCs. An initial 50 mL of 10% serum media was collected from hMSCs in culture for this experiment. The isolated EV pellet was resuspended in 100 µL of PBS and then was divided into 2 × 50 µL aliquots to be processed for either formvar or lacey carbon grid preparations. The formvar preparation procedure used in this study was also adapted from Current Protocols in Cell Biology (2006) (Théry et al. 2006). For traditional formvar preparation, Ted Pella (Redding, CA) Carbon Type‐B, 300 mesh, copper grids were used. The first 50 µL EV aliquot was fixed via the addition of 50 µL of 4% paraformaldehyde (PFA) for a final 2% PFA concentrated suspension. From this suspension, a 5 µL drop was placed on the formvar grid and allowed to dry for 20 min. The grid was then transferred to a 50 µL drop of 1% glutaraldehyde for 5 min and then washed with 100 µL of DI H2O 8 × for 2 min each. The grid was subsequently transferred to a 50 µL drop of uranyl‐oxalate (pH 7) for 5 min, followed by embedding in 50 µL of a 4% uranyl acetate (pH 4) and 2% methyl cellulose mixture (100 µL/900 µL respectively) to enhance contrast through negative staining. The excess solution was blotted, and samples were allowed to dry for 5 min before storage under vacuum in advance of viewing on TEM. In total, the process took around 60 min. The lacey carbon procedure for biologic samples was developed in‐house. For the lacey carbon preparation, Ted Pella Lacey Carbon, 300 mesh, copper grids were used. No fixation or negative staining/embedding was used. This procedure used 5 µL directly from the second 50 µL EV aliquot with the droplet placed on a clean petri dish. The lacey carbon grid was then placed directly on top of the droplet with the carbon side facing down for 35 seconds. The grid was then removed and the excess suspension was blotted and allowed to dry for 1 min before storage under vacuum prior to viewing on TEM. In total, the process took less than 5 min. TEM/EELS TEM was used to identify EVs and confirm relative sizes and quantities present among samples. Initially, TEM (Tecnai G2 Spirit BioTWIN TEM (Thermo Fisher Scientific, Waltham, MA) at 80 kV) was used to evaluate differences between formvar and lacey carbon grid preparations for the first human MSC sample derived EVs along with subsequent evaluations of ultracentrifugation, ultrafiltration, and precipitation EV isolation methods via lacey carbon grids. However, 28 of the remaining 31 EV samples derived from 5 human patients were evaluated via Titan3 G2 60–300 S/TEM (Thermo Fisher Scientific, Waltham, MA) at 60 kV on lacey carbon grids due to its higher resolution capabilities. Initial grid preparation and isolation methods were compared based on number of EVs per field of view (FOV) (standardized at 200 nm scale). The latter 28 EV samples were evaluated based upon blind sampling of sizes for the first 100 EVs identified from multiple fields of view per sample to compare with DLS acquired size distributions. Additionally, EELS via Titan3 G2 60–300 S/TEM at 300 kV was used as an analytical technique to further evaluate composition of EVs. Lacey carbon grids (consisting of a carbon lattice with holes) were also used in this case since the continuous carbon membrane present on formvar grids would interfere with EELS data acquisition. The collected EELS spectra were aligned by zero loss peak in the low‐loss spectrum, and the plural scattering in the high‐loss spectrum was removed by Fourier‐log deconvolution using DigitalMicrograph (Gatan Inc., Warrendale, PA). The goal was to identify the presence of phosphorus (lipid bilayer), sulfur (proteins), and nitrogen (protein, RNA, DNA). DLS DLS via Wyatt Technologies (Santa Barbara, CA) Mobius was used to evaluate the hydrodynamic size (diameter) distributions of the 31 EV samples derived from 5 patient samples (16 polyclonal and 15 clonal‐derived hMSC sample populations). Per sample, 15 acquisitions were captured via DLS with a read time of 30 seconds per acquisition. Comparison of EV Isolation Methods Ultracentrifugation, ultrafiltration, and precipitation EV isolation methods were compared for yield and purity through visual analysis via TEM. Yield measurement consisted of EV counts per FOV at similar magnification. Purity measurement consisted of visual evaluation of the presence of non‐EV components and structures present per FOV. Comparison of EV Characterization Methods DLS and TEM measurements were evaluated based upon comparison of the respective size distribution results for 28 of the 31 EV populations (3 EV samples could not be evaluated via TEM due to grid damage). Results from DLS were expressed as mean diameter of 15 individual measurements for each EV population (excluding measurements that failed to identify any EV diameters). TEM data sets comprised a random sampling of 100 EVs in TEM FOVs per each EV population. Histograms were compared from each method. Overall means were statistically evaluated via t‐test (Welch's two‐way with unequal variances) to determine the presence of significant differences between mean EV size outputs of DLS and TEM methods (p ≤ 0.05). Results EV Isolation Methods TEM imaging was collected via Tecnai G2 Spirit BioTWIN TEM at 80 kV and can be seen in Figure 1. Ultracentrifugation produced cleaner high yield samples and better preservation of EV morphology per FOV when compared to ultrafiltration and precipitation methods. EV morphology, including the lipid bilayer, was clearly evident after ultracentrifugation. Comparison of ultracentrifugation, ultrafiltration, and precipitation EV isolation techniques via TEM on lacey carbon grids. Ultracentrifugation (A and B) produced superior preservation of EV morphology with clear lipid bilayer (yellow arrows). Ultrafiltration (C and D) produced images with visible EVs (yellow arrows), however a sticky film of residual media components obscured EV morphology (green arrows). Precipitation (E and F) produced images with visible EVs (yellow arrows). However in addition to a sticky film (green arrows), there were many instances of additional debris components obscuring the EVs (blue arrow). Ultracentrifugation (A and B) also indicated possible evidence of a small number of exomeres (10–50 nm), but this population was a minority in comparison and is not believed to have largely influenced results (red arrows). Traditional Formvar vs. Simplified Lacey Carbon TEM Grid Preparation for EVs TEM imaging was collected via Tecnai G2 Spirit BioTWIN TEM at 80 kV and can be seen in Figure 2. Lacey carbon grids without fixation or staining provided clearer images of EVs with a distinct lipid bilayer. They also demonstrated higher numbers of EVs for assessment with a larger diversity of sizes when comparing the same samples on formvar grids. EVs assessed using the formvar preparation with fixation and staining were smaller in size with a less distinct lipid bilayer and a larger proportion of surrounding debris. Furthermore, lacey carbon samples were reimaged at 1 month and 1 year and maintained a non‐degraded EV appearance in all instances. This observation is important in that, in addition to higher yield and preservation of diversity, lacey carbon preparations may enable durable archiving of samples for discontinuous assessment. Formvar vs. lacey carbon grid TEM images. Traditional formvar grid TEM image (A): One EV is circled in yellow. Other dark particulates indicate debris and other non‐vesicular components, residual from the grid preparation process. Image represents one of the few fields of view with a discernable EV. In most fields of view, no EVs were visible. TEM image using lacey carbon mesh (B): Three EVs are circled in yellow showing varying size with clear lipid bilayers. The darker gray background regions represent the lacey carbon mesh used to capture EVs for imaging. Image is representative of multiple consistent fields of view showing distinct EVs. EELS EV Evaluation EELS was performed via Titan3 G2 60–300 S/TEM. Two samples from Pt4 clonal (PBS2) EV isolations were assessed with EELS for elemental content identification. A representative spectrum from one of these samples is shown in Figure 3. EELS was able to identify phosphorus (lipid bilayer) peaks around 132 eV. However, EELS was unsuccessful in demonstrating sulfur (protein) peaks at 165 eV or nitrogen (protein, RNA, DNA) peaks at 402 eV, likely due to the overlap with one of several phosphorus peaks (160 eV) and broad chlorine (200 eV), as well as carbon (starting at 284 eV) peaks. This was likely the result of the limited spectral resolution of the Titan3 G2 60–300 S/TEM detector. EELS spectrum of EV sample. An average sized EV (50–100 nm) is shown in the initial green box spectrum image and the zoomed in red box #5 represents the spatial region of the EV that generated this EELS spectrum—Note the capture of the EV bilayer and contents while minimizing overlap with the lacey carbon grid structure on the right side. Also note the grid pattern indicates EELS measurements across the EV and that EELS processing results in some permanent damage to the EV morphology. The red line is fitted to the blue spectrum curve to produce the green zero‐loss spectrum line that normalizes the peaks for better visualization. The EELS spectrum demonstrates phosphorus (lipid bilayer) peaks starting around 132 eV through 140 and 160 eV (P region). A clear sulfur (protein) peak around 165 eV was not visible, possibly due to overlap with the 160 EV phosphorus peak and a broad chlorine peak at 200 eV from residual PBS (S and Cl marked regions). Nitrogen (protein, RNA, DNA) peaks around 402 eV could not be seen either due to poor detection or due to the dominance of the broad carbon peak starting at 284 eV from the lacey carbon grid (N and C marked regions). TEM vs. DLS for Measurement of EV Size and Variation Hydrodynamic size was assessed via Wyatt Technologies Mobius on all 31 EV samples derived from 5 patient samples (16 polyclonal and 15 clonal‐derived hMSC sample populations) (Table 1). Of these 31 samples, 28 were then imaged with Titan3 G2 60–300 S/TEM to assess the fidelity between TEM and DLS EV size estimates (Table 1). Three remaining EV samples could not be imaged due to grid damage. The TEM images from each of these 28 samples were evaluated based upon unbiased sampling of 100 EVs (Figure 4) and blinded with respect to patient and clonality of each sample. Measurement of Feret's diameter was collected for each EV via ImageJ and used to calculate the mean EV diameter and size distribution. The results of DLS and TEM were compared via Welch's two‐way t‐test with unequal variances (α = 0.05). TABLE 1DLS and TEM size (diameter) data for the 31 EV samples derived from 5 patient samples comprised of 16 polyclonal (CEM) and 15 clonal (PBS) derived hMSC sample populations.Patient #Culture type DLS TEM T‐test (Welch's two‐way with unequal variances) Mean (nm)Std Dev (nm)Mean (nm)Std Dev (nm)Std error (nm) TEM and DLST‐testDegrees of freedomCritical T‐value P‐Value (α = 0.05) Pt1 CEM1 189.870.4222.279.120.41.58617.92.1010.065 CEM2 138.739.9196.0107.715.43.71041.02.0200.0003 CEM3 153.179.6150.273.523.3−0.12514.82.1310.451 CEM4 201.5122.7174.374.633.6−0.80914.42.1450.216 CEM5 119.746.6189.9100.216.04.39433.42.0350.0001 CEM6 129.680.0163.591.224.01.41016.32.1200.089 PBS1 139.864.1155.473.919.90.78414.72.1310.223 PBS2 125.061.3183.578.718.73.12617.62.1010.003 PBS3 194.4147.8106.971.340.2−2.17913.92.1450.023 PBS4 149.455.1215.578.816.34.06923.62.0640.0002 PBS5 187.699.5191.873.728.60.14513.82.1450.443 Pt2 CEM1 265.0236.9 PBS1 103.049.9248.8111.417.08.55639.62.0217.50E‐11 PBS2 152.649.8266.2125.219.15.95932.02.0376.35E‐07 Pt3 CEM1 * 245.6178.7165.490.354.6−1.46710.62.2010.085 CEM2 123.0103.5 CEM3 171.894.9125.967.927.2−1.68913.62.1450.057 Pt4 CEM1 139.556.3123.844.616.2−0.96114.02.1450.176 CEM2 234.7165.6177.579.145.0−1.27313.82.1450.112 CEM3 252.7189.3280.7121.952.00.54014.52.1310.299 PBS1 124.544.0137.260.013.60.94018.42.1010.180 PBS2 257.4231.9109.949.762.2−2.37213.22.1600.017 Pt5 CEM1 179.9112.2149.078.831.0−0.99714.92.1310.167 CEM2 201.0222.1 CEM3 172.8124.0138.752.147.2−0.7246.12.4470.248 PBS1 82.462.4241.7108.819.48.20228.82.0602.13E‐08 PBS2 118.188.3163.078.827.81.61811.82.1790.066 PBS3 148.055.8136.755.917.7−0.64112.32.1790.267 PBS4 141.253.0168.175.917.11.57417.02.1100.067 PBS5 138.488.7132.756.927.3−0.21110.92.2010.419 PBS6 118.883.8126.967.323.40.34715.42.1310.367Of the 31 EV samples measured by DLS, 3 samples could not be imaged by TEM due to grid damage (blank cells in table), resulting in only 28 samples being compared. The mean EV diameter across all 28 samples measured by DLS and TEM was 161.1 nm and 172.9 nm respectively (not significantly different, p = 0.341). Of the 28 individual EV samples compared between DLS and TEM, 19 were not significantly different in mean size measurement (p > 0.05) while 9 were significantly different (p ≤ 0.05).Difference Between DLS and TEM Data is Statistically Not Significant if P‐Value > 0.05.Difference Between DLS and TEM Data is Statistically Significant if P‐Value ≤ 0.05.*Statistical Calculations Based on 98 Total EVs Measurable by TEM. DLS and TEM size (diameter) data for the 31 EV samples derived from 5 patient samples comprised of 16 polyclonal (CEM) and 15 clonal (PBS) derived hMSC sample populations. Of the 31 EV samples measured by DLS, 3 samples could not be imaged by TEM due to grid damage (blank cells in table), resulting in only 28 samples being compared. The mean EV diameter across all 28 samples measured by DLS and TEM was 161.1 nm and 172.9 nm respectively (not significantly different, p = 0.341). Of the 28 individual EV samples compared between DLS and TEM, 19 were not significantly different in mean size measurement (p > 0.05) while 9 were significantly different (p ≤ 0.05). Difference Between DLS and TEM Data is Statistically Not Significant if P‐Value > 0.05. Difference Between DLS and TEM Data is Statistically Significant if P‐Value ≤ 0.05. Statistical Calculations Based on 98 Total EVs Measurable by TEM. TEM image of EVs and counted via ImageJ. This TEM image is one representative field of view. Unbiased sampling of 100 EVs was used to evaluate EV size differences for each of the 28 samples, blinded with respect to patient and clonality of each sample (A). Feret's diameter was collected for each EV via ImageJ and used to calculate the mean EV size per sample (B). The overall mean EV diameter across all 28 samples measured by DLS and TEM was 161.1 nm and 172.9 nm respectively, which was not significantly different (p = 0.341). Of the 28 individual EV samples compared between DLS and TEM, 19 were found to be not significantly different in mean size measurement (p > 0.05) while 9 were found to be significantly different (p ≤ 0.05). There was no evidence of systematic under‐ or over‐estimation of size between DLS and TEM measurements. Of the 9 significantly different EV samples, 7 samples had larger mean sizes in TEM than DLS and 2 were smaller in TEM than DLS. Further, across the 28 samples compared, 16 had larger individual mean EV sizes measured by TEM than DLS, potentially due to flattening of the EV on the lacey carbon grid. Example histograms demonstrating the measured EV size distributions for the DLS and TEM comparisons indicate that major EV size ranges identified in 50 nm bins are often similar in the 40–200 nm range (Figure 5). However, there are instances where DLS and TEM identified some differences. In some cases, DLS identified subpopulations of larger sized EVs (> 200 nm), which could point to possible particle agglomerates or debris being captured with sizes approaching 1000 nm. Such events can be excluded through analysis of comparative TEM images. Example histograms of EV size distributions measured by DLS and TEM. These histograms demonstrate three examples of direct comparison of EV size estimates from DLS and TEM measurements for Pt3 CEM1 (A. DLS and B. TEM), Pt3 CEM3 (C. DLS and D. TEM), and Pt4 PBS2 (E. DLS and F. TEM). Overall, means were often comparable. However, distributions and ranges of subpopulation particle sizes sometimes diverged. For example, in Pt4 PBS2 (E. DLS and F. TEM), DLS identified a wider range of EV sizes than those visualized in the TEM measurement. This can likely be attributed to the fact that DLS is an indirect light scattering measurement of Brownian motion which is used to calculate relative size distribution for the whole population. It is therefore incapable of separating debris or agglomerated EVs from individual EVs. In contrast, TEM provides direct visualization and confirmation of EVs for analysis. DLS data suggested that EVs from clonal population were significantly smaller than EVs from polyclonal hMSCs (p = 0.049) (Welch's two‐way t‐test with unequal variances comparing the 15 clonal and 13 polyclonal samples) (Figure 6). However, TEM data did not support this conclusion (p = 0.941) (Figure 6). Boxplot comparison of EVs derived from clonal (PBS) and polyclonal (CEM) hMSC populations. The boxplots demonstrate different results for comparing clonal (PBS) and polyclonal (CEM) hMSC EV populations. Discussion EV Isolation Methods Differential ultracentrifugation provided highest yield and the cleanest sample preparation for EV isolation based on TEM visualization of morphology. This finding with respect to purity is consistent with the available literature comparing these methods such as Sidhom et al. (2020) and Ayala‐Mar et al. (2019) who noted high purity for ultracentrifugation over alternatives (Ayala‐Mar et al. 2019; Sidhom et al. 2020). This finding with respect to yield is also consistent with MISEV2023 and 2018 recommendations and work by Kurian et al. (2021) indicating ultracentrifugation is particularly effective for high yield (Kurian et al. 2021; Théry et al. 2018; Welsh et al. 2024). However, this is in contrast to Sidhom et al. (2020) and Hou et al. (2019) who reported lower yields for ultracentrifugation compared to both ultrafiltration and precipitation (Hou et al. 2019; Sidhom et al. 2020). Traditional Formvar vs. Lacey Carbon TEM Grid Preparation for EVs To the best of our knowledge, this paper reports the first use of the lacey carbon TEM grid preparation method for presentation of EVs from any tissue source, including hMSCs without fixation and staining for analysis. Lacey carbon methods have been reported in several other applications for imaging nanoparticles, including size quantification of gold nanoparticle formulations for RNA delivery to myeloid cells and their bone marrow progenitors and imaging the microstructure of silicon nanorods for solar cells (Hofstraat et al. 2025; Radnóczi et al. 2016). Additionally, lacey carbon grids have been previously used in conjunction with cryo‐electron microscopy (cryo‐EM) for applications including visualization of EVs from cerebrospinal fluid (Emelyanov et al. 2020). The lacey carbon grid preparation method outperformed the traditional formvar carbon method. Lacey carbon grids provided clear resolution of the intact lipid bilayer in the EVs without fixation or staining. This represents a potentially valuable advancement, not only for the reduction of preparation time and debris, but also because the formvar method with fixation changes the morphological attributes, decreasing the size of the observed EVs as well as the number of EVs available for assessment. Moreover, the lacey carbon samples without fixation showed no evidence of degradation up to 1 year when stored under vacuum at room temperature (25°C), opening opportunity for sample archiving and asynchronous analysis. EELS EV Evaluation EELS is a sensitive method for localizing the presence of specific elements with high (nanoscale) spatial resolution (Aronova and Leapman 2012; Dumaresq et al. 2024). EELS uses the energy loss of electrons after interaction with a material to measure elemental composition and has been used to characterize a range of biologic samples including organ tissues, organelles, neurons, and cell components (Aronova and Leapman 2012). Literature is limited on the usage of EELS to examine and characterize EVs, but Stapane et al. (2020) demonstrated use with avian‐derived EVs to identify the presence of calcium bonds and mineralization in eggshell formation (Stapane et al. 2020). In this study, EELS was able to identify phosphorus (lipid bilayer) peaks around 132 eV; however, it was unsuccessful in demonstrating sulfur (protein) peaks at 165 eV or nitrogen (protein, RNA, DNA) peaks at 402 eV, likely due to overlap with phosphorus, chlorine, and carbon peaks. PBS was necessary to maintain structural integrity of the lipid bilayer of EVs for TEM. Carbon peaks starting at 284 eV resulted from the lacey carbon mesh. TEM vs. DLS EV Characterization DLS and TEM are both valuable tools to consider for analyzing EV size and size distributions. These data demonstrated some discrepancies between the outputs of the two methods using the same sample. DLS measures all particles and cannot ignore debris or agglomerates, often skewing the distribution of particle size above a range measured in TEM images. Regardless, the overall mean diameter estimated by DLS and TEM analysis were not statistically different in 68% of the tested samples, and there was no evidence of a systematic difference in size estimates (over or under) between these two methods. We tested the hypothesis that EVs from clonal (PBS) and polyclonal (CEM) hMSCs would be different with respect to mean size. DLS data suggested that EVs from clonal populations were significantly smaller than EVs from polyclonal hMSCs (p = 0.049) (Welch's two‐way t‐test with unequal variances for the 15 clonal and 13 polyclonal samples). However, TEM data did not support this conclusion (p = 0.941). There was also no suggestion of a difference in the size of EV subpopulations when comparing histogram patterns between clonal and polyclonal hMSC populations, based on EV size distribution. Traditionally, larger EVs (often termed microvesicles) typically ranging from 100 to 1,000 nm are thought to form by direct outward budding from the cell plasma membrane, whereas smaller EVs (often termed exosomes) typically ranging from 30 to 150 nm are thought to form by an inward budding of the limited membrane of early endosomes (Doyle and Wang 2019; Théry et al. 2018; Welsh et al. 2024). However, it should be noted that these biogenesis size range conventions are not clearly defined nor currently standardized (Welsh et al. 2024). Overall, DLS is a rapid and expedient method for particle size estimation, which can be useful in EV analysis. However, while more resource intensive, TEM imaging provides direct observation, sample quality assessment, and direct validation that is preferred when precision is needed. Limitations EV Isolation Methods A variety of methods are available for EV isolation. These methods vary with respect to equipment demands, expertise, time, and associated costs. We chose ultracentrifugation as a superior method for EV isolation. However, the ultracentrifuge and associated rotor needed are expensive. This method may not be readily available to every lab. Ultracentrifugation is labor‐intensive (approximately 3 hours of centrifugation time per sample). In contrast, ultrafiltration and precipitation methods require roughly 2.5 hours and 1 hour respectively and can be accomplished in high‐speed centrifuges. It has been suggested that ultrafiltration process time could be decreased to close to 1 hour by reducing wash steps, but this could further reduce purity (Extracellular Vesicle Protocol 2021). We prioritize yield and purity to enable the most effective comparison between hMSC populations. However, we recognize that if and when EVs derived from hMSCs or other cell sources are used as a drug therapy, alternative automated methods with greater scaling opportunities will be needed, even if ultracentrifugation remains the gold standard for research purposes. It is worth noting that we chose not to assess alternative EV isolation methods, such as targeted affinity (i.e., microfluidics and immunoaffinity chromatography). In the future these methods may provide opportunity for high purity isolations with increased yield at a lower cost than ultracentrifugation (Baranyai et al. 2015; Liga et al. 2015). Additionally, ultracentrifugation has the potential to co‐isolate small lipoproteins, exomeres, or viral particles (Yu et al. 2025). The TEM images in this study do not show evidence of lipoproteins or viral particles, however a small number of round structures between 10 and 50 nm can be seen in Figure 1 A and B that are in the size range and shape of exomeres. TEM and DLS results in Figure 5 suggest that there could be a small number of exomeres, but that they are only a small minority in the population. We do not believe these co‐isolates had any significant influence on the results. Traditional Formvar vs. Lacey Carbon TEM Grid Preparation for EVs We propose, based on this experience, that lacey carbon TEM grid preparations should be considered as a less time‐consuming method with greater visualization capabilities for EV analysis. Traditional formvar preparation for biologic samples, including EVs, has the benefit of a long track record of use (Pascucci and Scattini 2021; Théry et al. 2006). However, direct comparison to lacey carbon grid preparation without fixation or staining has demonstrated the significant disadvantages of formvar methods. Formvar preparations demonstrated fewer and smaller observable EVs as well as poor visualization of the EV lipid bilayer when compared to lacey carbon grid preparations. In addition, traditional formvar samples demonstrated an increase in debris, likely related to the more extensive pre‐processing required. The simplicity and value of the lacey carbon grid preparation cannot be overstated. This adaptation of an established tool for nanomaterials characterization has additional advantages for EV analysis. We have shown that EVs on lacey carbon grid without fixation can be stored under vacuum at room temperature for over a year without evident degradation. Note however, that unfixed EVs on lacey carbon grids are more susceptible to degradation from the electron beam and must be viewed at lower beam strength (60–80 kV max instead of traditional 80–120 kV). In such cases where additional resistance to beam degradation is desired, although requiring additional preparation steps, cryo‐EM has been employed to further address this need (Emelyanov et al. 2020). EELS EV Evaluation EELS was not considered to be a primary or essential outcome parameter in this work, but was included in our assessment as a means of validation of intact EVs with lipid bilayer using phosphorus as a marker and protein/RNA cargo using sulfur and nitrogen as a marker. Phosphorus was clearly seen in three predicted peaks. However, the EELS analysis was limited in detection of sulfur and nitrogen by the need for PBS to stabilize the EV structure, which added a broad obscuring peak from chlorine. The lacey carbon grid overlapping with or adjacent to all EVs also added a prominent broad carbon peak. Higher resolution EELS devices may be capable of clearer identification of content differences between the EV interior and the outer lipid bilayer as well as separate EV components from the adjacent carbon mesh. Note that EELS analysis results in structural degradation to the unfixed EV samples and can only be run once per EV. TEM vs. DLS EV Characterization DLS is theoretically most effective for evaluating size distributions among large numbers of particles. However, it is limited in its application to EV analysis by the inability to provide a quantitative measure of the number of particles evaluated, and the inability to separate particles of interest from debris or agglomerates. This was evident in some of the disparities we identified in paired analysis of samples using both DLS and TEM. TEM is superior to DLS in providing direct validation of the identity and number of individual sample data sources through visualization of particles present. However, TEM is laborious and requires greater resources (equipment, time, and money) and technical skills. With efficient image processing and standardized procedures, it may be possible to integrate TEM for larger batch sample processing. However, it is important to recognize that the visual per FOV measurement inherent in TEM assessment is semi‐quantitative in nature and that methods for more automated counting are still valuable to consider when assessing purity and yield. Currently, DLS is useful for rapid large‐scale analysis and screening. TEM is more appropriate for screening in settings where individual batch identity and morphology is needed as well as for precision comparison between EV sources and batches. DLS utilizes the hydrodynamic radius of the suspended sample for the indirect diameter measurement whereas TEM uses a direct measurement of diameter of EVs dried on a flattened grid surface. However, we found no evidence of a systematic over‐ or under‐estimation of mean EV size using these two methods. Although not within the scope of this study, alternative EV characterization methods such as nanoparticle tracking analysis (NTA) would also be appropriate to evaluate in comparison to DLS and TEM (Kowkabany and Bao 2024). Lastly, it is important to note that this work only focused on assessing EV morphology (TEM) and size distribution (DLS) metrics for comparison between EV populations. However, for complete assessment of EV purity and yield for research and clinical purposes in accordance with MISEV2023 standards, it is critical to confirm EV identity via a multiparametric combination of methods to evaluate morphology (i.e., TEM, cryo‐EM), protein biomarkers (i.e., flow cytometry), particle counting and size distribution (i.e., NTA, DLS), and/or other orthogonal approaches (Welsh et al. 2024). Conclusions EVs may have value as a readout of biological state or functional attributes of individual cell populations (Giebel and Lim 2025). EVs may also evolve into a therapeutic vehicle in their own rights (Giebel and Lim 2025). This paper provides insight into effective methods for isolation and characterization of hMSC‐derived EV population, but these findings are likely applicable to EV populations from many other cell sources (Welsh et al. 2024). It is important to define the most effective, accurate, and reliable methods for isolation of EVs and assessment (Welsh et al. 2024). This study has demonstrated that ultracentrifugation is one of the most effective methods for EV isolation with higher yield and purity that should further be evaluated by EV biomarker analysis to confirm. We provide the first report of TEM imaging and assessment of EVs using a lacey carbon TEM grid preparation without fixation, staining, or use of cryo‐EM. This provided a highly effective method for morphological assessment of EVs, including durable archiving of samples for discontinuous assessment, and should be considered a preferred method for TEM preparations going forward. TEM images provide the most direct and robust means of assessment of EV size and morphology. Automated methods for image segmentation and analysis to replace manual target identification and tracing are desirable. DLS is a less expensive and less resource‐intensive method that is common in nanoparticle materials science. DLS can provide estimates of EV size. However, as an indirect measurement, DLS can be sensitive to debris and agglomerate artifacts. It should be validated with TEM. EELS methods for EV characterization were not critical to this analysis. EELS for EV analysis is limited by region resolution and peaks from contaminating chlorine and carbon in these samples. Sample preparation and validation may be improved to increase resolution. Conflicts of Interest The authors declare no conflicts of interest. Acknowledgements We thank the Cleveland Clinic Research Imaging Core and The Ohio State University Center for Electron Microscopy and Analysis (CEMAS) for their support in capturing TEM and EELS imaging. This work was supported by NIH grant no. R01 DE029634, the Mary Lou Hammond Fund, and Lerner Research Institute Seed support. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request. References Associated Data Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request.