The three methods of isolating EVs compared are detailed in Fig. 1.

Steps of the different methods used to isolate and characterize EVs. Created with BioRender.com

To confirm efficient EV isolation, particle size distributions and yields for each isolation procedure were determined by NTA (Fig. 2).

Characterization of EVs using NTA to measure (a) particle distributions and (b) total numbers of particles isolated following ultracentrifugation (UC), size exclusion chromatography (SEC), or both (SEC + UC). ***p < 0.001. The presented data has been adjusted based on the dilution of samples utilized for NTA analysis

Mean particle size for the three methods was greater than 100 nm. The SEC + UC group showed the largest mean size (172.1 ± 15.9 nm; mean ± SD), followed by UC (163.8 ± 9.5 nm) and Sect. (160.2 ± 25.5 nm) (Fig. 2a). However, differences were non-significant.

Our NTA analysis revealed that the UC method gave rise to a significantly greater number of isolated particles (1.28 × 1010 ± 9.31 × 109 particles/mL; data ± SD) than Sect. (1.56 × 109 ± 4.08 × 108 particles/mL) or SEC + UC (1.47 × 109 ± 7.73 × 108 particles/mL) (p < 0.0001) (Fig. 2b), which failed to differ between each other.

Through TEM, we observed that all isolation procedures were successful in isolated EVs within the expected size range. Accordingly, all three methods yielded EV-like structures of characteristic cup-shaped appearance and heterogeneous sizes ranging from approximately 100–200 nm (Fig. 3). The microscopy image of EVs-UC (Fig. 3a) shows a blurry background with aggregates, indicating the co-isolation of other products. Also, the EVs-SEC image reveals the presence of rounded white vesicles resembling EVs, but these appear too small so they could be lipoprotein particles (Fig. 3b). For the combined method (Fig. 3c), a clean background may be observed with different sized EVs aggregated together as a consequence of ultracentrifugation.

Electron microscopy images of EVs yielded by the different isolation methods. (a) ultracentrifugation (UC), (b) size exclusion chromatography (SEC), and (c) SEC + UC. Scale bar = 200 nm

EV isolates were lysed and total protein contents (EVs plus soluble protein) of each sample were determined by BCA (Fig. 4a). Protein contents varied significantly according to the isolation method, with highest concentrations recorded for the UC group (1770.83 ± 286.4 µg/mL; mean ± SD) followed by SEC and SEC + UC (730.85 ± 4.04 and 236.77 ± 137.7 µg/mL, respectively), the combined approach yielding the lowest protein contents (p < 0.001).

Purity of EVs isolated using different methods. (a) Protein concentrations of intact isolated EVs were determined by the bicinchoninic acid (BCA) assay. (b) Coomasie staining gel of proteins according to the EV isolation method used. (c) Western blotting of EVs isolated using the different methods. Data represent means ± SD of three independent experiments. *p < 0.05; ***p < 0.001

To assess isolation efficiency, equal protein amounts in the UC-, SEC-, and SEC + UC-derived samples were placed on an SDS-PAGE gel followed by a Coomasie blue staining. As shown in Fig. 4b, there was great qualitative variation in protein patterns among the different preparations. Compared to the other EV fractions, EVs-UC were especially rich in high molecular weight proteins like immunoglobulins and albumin. Also, according to the different pattern of protein bands observed in SEC- and SEC + UC-derived EVs compared with UC-derived EVs, these two groups may have a similar protein composition (Fig. 4b).

Western blotting of the purified EV fractions confirmed the presence of the EV marker proteins CD63, CD81 (tetraspanins) and TSG101 (cytosolic protein) in 10 µg of the protein sample derived from each isolation method. As shown in Fig. 4c, all methods recovered clean EV populations. We also assessed the expression of albumin in all the EV preparations, as this marker serves to detect impurities as it is the most abundant protein in plasma. The expression of this protein was higher in EVs isolated by UC compared to SEC or SEC + UC. Additionally, we assessed lipoprotein contamination through the expression of apoA1, which was higher in both SEC and UC methods (Fig. 4c).

As plasma is a complex fluid, all MS-based proteomic procedures were conducted in duplicate samples to identify and validate as many proteins as possible. In total, 542 proteins were identified along with at least two unique peptides for all EVs derived from the different isolation methods.

The protein contents recorded for each isolation protocol revealed that the SEC + UC method yielded a more complex proteome including 364 identified proteins, compared to 212 proteins for UC and 276 proteins for SEC. These proteins were further analysed using the DAVID database, mapping them only for cellular component (CC) against the human genome as background to determine their associations with extracellular vesicles and plasma. The identification and classification of these proteins within the GO terms ‘extracellular exosome’ and ‘blood microparticle’ are provided in Additional Table 1. This analysis revealed that the isolation method SEC + UC gave rise to more EV proteins (273 proteins) and fewer blood-related proteins (70 proteins) compared to UC, which showed the largest number of plasma proteins (153 EV proteins vs. 116 plasma proteins), or SEC. (200 EV proteins vs. 92 plasma proteins).

We then conducted a more thorough assessment to determine whether commonly enriched terms were more associated with EVs or plasma across the isolation methods. The terms related to EVs examined were ‘extracellular exosome’ (GO: 0070062), ‘extracellular space’ (GO: 0005615), ‘extracellular region’ (GO: 0005576), ‘vesicle’ (GO: 0031982) and ‘extracellular vesicle’ (GO: 1,903,561). The plasma related terms considered were ‘plasma membrane’ (GO: 0005886) and ‘blood microparticle’ (GO: 0072562). Out of a total of 364 proteins identified for SEC + UC, 276 for SEC and 212 for UC, the DAVID database recognized 256, 195 and 157 gene entries, respectively.

Table 1 lists the GO terms associated with EV and plasma proteins. Terms related to EVs consistently showed comparable percentages among the different EV samples for the term ‘extracellular exosome’ with 76.2% for SEC + UC, 75.4% SEC and 75.2% UC or ‘extracellular vesicle’ with 3.9% in SEC + UC, 3.6% in SEC and 3.2% in UC. For the terms ‘extracellular space’, ‘extracellular region’, and ‘vesicle’ differences in the percentages were found in the different sample. However, for the plasma-related terms, especially ‘blood microparticle’ a higher percentage of gene entries was recorded in the UC group (21.1%, 33.3%, and 52.9% of identified proteins in SEC + UC, SEC and UC, respectively). No differences among the various methods were recorded for ‘plasma membrane’.

Finally, we conducted a detailed analysis to validate the presence of proteins related to EVs and potential contaminants in the samples examined previously according to MISEV2023 guidelines from ISEV [20]. The protein samples shown in Fig. 5 were classified into three distinct categories of markers. Categories 1 and 2 indicate the detection of EVs with traditional EV markers such as integrins (Uniprot: ITG), actins (Uniprot: ACTB), or glyceraldehyde-3-phosphate dehydrogenase (Uniprot: GAPDH) in all samples. Additionally, heat shock protein 71KDa (Uniprot: (HSPA8), tetraspanins (Uniprot: CD9), guanine nucleotide (Uniprot: GNA), disintegrins (Uniprot: ADAM10), tubulins (Uniprot: TUB) or caveolae-associated protein 2 (Uniprot: CAVIN2) were found in the SEC + UC and SEC sample groups. However, the UC sample showed the least identification of EV protein markers.

Characterization of EV protein contents based on MISEV2023 guidelines. Each row represents the identified protein in the samples within the different categories, and columns indicate the isolation method in which the protein was detected

Category 3 indicates the presence of common contaminants for purity assessment. As plasma samples were used, protein impurities from plasma such as albumin (Uniprot: ALB), apolipoproteins (Uniprot: APO), immunoglobulins, and others were found in all isolated EVs, but particularly in UC-EVs. A notable observation is that the number of apolipoproteins, common protein contaminants plasma-derived EVs, found in SEC + UC-EVs were clearly reduced compared to single-step techniques. Furthermore, proteins like 14-3-3 beta/alpha (Uniprot: YWHAH), heat shock protein 90 alpha (Uniprot: HSP90AA1) or lactate dehydrogenase (Uniprot: LDH) were exclusively identified in SEC and SEC + UC-EVs (Fig. 5).

After our initial assessment of EV and plasma proteins across all samples, we performed a comparative analysis to identify shared and unique proteins in the different isolation groups. Overlap in protein identification was visualized using a Venn Diagram generated with the FunRich tool (Fig. 6a).

Total protein contents of EV samples yielded by the three isolation methods. (a) Venn diagram showing the proteins identified according to the isolation method. (b) Percentages of shared cellular component terms of EVs and plasma: among the 80 proteins common to all three methods; (c) among the proteins common to pairs of isolation methods; and (d) among the proteins unique to each method

In all three isolation method groups, 80 proteins were found to be common. Our GO enrichment CC analysis revealed 5 enriched terms related to EVs or plasma among these shared proteins as detailed in Fig. 6b. GO classifications indicated 82.8% of these proteins were categorized as ‘extracellular exosomes’. Furthermore, 60.3% were linked to the term ‘extracellular region’, and 56.9% to ‘extracellular space’. These percentages were higher than the proportion of plasma-related proteins (‘blood microparticle’) detected (44.8%).

Upon closer examination of overlapping proteins across groups (Fig. 6a), it becomes evident that the method pairs SEC + UC and SEC gave rise to a higher number of shared proteins (80) than SEC and UC (47) or SEC + UC and UC [23].

In a more detailed analysis, we examined GO terms related to EVs and to plasma within proteins common to the different isolation methods compared as pairs.

As may be observed in Fig. 6c, at the general level, terms related to EVs appear in greater proportions than the plasma-related terms in all comparisons. In effect, proteins common to SEC and UC showed the highest proportions of the terms ‘extracellular exosome’, ‘extracellular region’ and ‘extracellular space’, at 81.2%, 87.5% and 90.6%, respectively, compared to the pairs SEC + UC and UC, and SEC + UC and SEC. However, it should be noted that all isolation methods yielding proteins common with UC (SEC + UC and UC, and SEC and UC) gave rise to greater proportions of the plasma-related term ‘blood microparticle’, at 75% for SEC and UC, and 38.9% for SEC + UC and UC.

Once we had established proteins shared among the different isolation methods, we focused our next analysis on identifying proteins exclusive to each isolation method. The Venn diagram in Fig. 6a reveals that 181 proteins were exclusive to EVs isolated from plasma using the SEC + UC method, 62 proteins were exclusive to the single-step UC method, and 69 proteins were exclusive to the SEC method.

We then annotated unique proteins arising from each isolation method by assessing enriched GO terms for CC and comparing the top 5 enriched GO terms shared by the different isolation method groups.

Remarkably, according to the data presented in Fig. 6d, all unique proteins showed a marked abundance of GO terms associated with EVs such as ‘extracellular space’, ‘extracellular region’, and ‘extracellular exosome’. This suggests that a great majority of these unique proteins can be attributed to EVs. Interestingly though, among proteins exclusive to UC, high proportions of enrichment in plasma-related terms were detected (‘blood microparticle’ and ‘plasma membrane’ at 51.2% and 46.3%, respectively). This pattern was not apparent for the other two methods.

Considering that most unique proteins could be associated with EVs, we then focused on these unique proteins as a measure of the sensitivity of the different isolation methods and also tried to determine whether any important information could be missing according to the isolation protocol used. To this end, we took as reference values theoretical concentrations of these unique proteins in plasma as defined in the Human Protein Atlas database (https://www.proteinatlas.org/ accessed on November 23, 2023). These data are provided in Table 2 and in Additional Table 2.

The dynamic ranges of the unique proteins identified for each isolation method were categorized by concentration rates (mg/L, µg/L, and ng/L). The percentages shown in the table indicate the proportions of proteins within each specified concentration range detected for each of the three isolation methods.

The SEC + UC method emerged more effective for the identification of different proteins across various concentration ranges. The highest detection rate, 81% of unique proteins, was observed within the µg/L range, followed by 9.01% and 9% in the ng/L and mg/L ranges respectively. Protein isolation using SEC followed a similar trend with higher detection rates of 58% and 17%, respectively for µg/L and ng/L concentrations compared to 23% for mg/L concentrations. Although SEC method was able to identify proteins with the lowest theoretical concentration in plasma (7.1 ng/L), SEC + UC detected a higher number of proteins within this concentration when compared to SEC (11 proteins vs. 8 proteins) (Additional Table 2).

Notably, in the isolates produced by UC, it was possible to identify distinct proteins present in plasma at mg/L concentrations (87%), whereas only six proteins detected were within the µg/L range.