Immunomagnetic sequential ultrafiltration (iSUF) platform for enrichment and purification of extracellular vesicles from biofluids

Extracellular vesicles (EVs) derived from tumor cells have the potential to provide a much-needed source of non-invasive molecular biomarkers for liquid biopsies. However, current methods for EV isolation have limited specificity towards tumor-derived EVs that limit their clinical use. Here, we present an approach called immunomagnetic sequential ultrafiltration (iSUF) that consists of sequential stages of purification and enrichment of EVs in approximately 2 h. In iSUF, EVs present in different volumes of biofluids (0.5–100 mL) can be significantly enriched (up to 1000 times), with up to 99% removal of contaminating proteins (e.g., albumin). The EV recovery rate by iSUF for cell culture media (CCM), serum, and urine corresponded to 98.0% ± 3.6%, 96.0% ± 2.0% and 94.0% ± 1.9%, respectively (p > 0.05). The final step of iSUF enables the separation of tumor-specific EVs by incorporating immunomagnetic beads to target EV subpopulations. Serum from a cohort of clinical samples from metastatic breast cancer (BC) patients and healthy donors were processed by the iSUF platform and the isolated EVs from patients showed significantly higher expression levels of BC biomarkers (i.e., HER2, CD24, and miR21).

. The Immunomagnetic sequential ultrafiltration (iSUF) platform. (A) Schematic representation of iSUF stages 1, 2, 3, including tangential flow filtration (TFF) purification, centrifugal unit enrichment, and tEV immunoaffinity isolation, respectively. In stage 1, biofluids were processed using a 500 kDa TFF filter, EVs were retained in the retentate and enriched in 7 mL while free proteins and nucleic acids permeated through the TFF filter. Then the PBS valve was opened to start TFF diafiltration until removing up to 99% of free proteins. Finally, EVs were recovered in 2 mL of PBS after flushing the TFF system with air. In stage 2, EVs were centrifuged using a 10 kDa centrifugal unit at 3000×g and enriched in 100 μL. In stage 3, tEVs were captured using antibodies immobilized to streptavidin-coated magnetic beads (i.e., EGFR) and subsequently pulled out with a magnet. (B) iSUF stage 1 purification performance for different biofluids. Cell supernatant (CCM), urine, and serum took 80, 100, and 120 min to obtain up to 99% efficiency of free protein and nucleic acid removal.
Processing biofluids using commercially available EV isolation kits. Using the Total Exosome Isolation method (TEI, Invitrogen, Carlsbad, CA), EVs were isolated from 0.5 mL serum according to the manufacturer's instructions. Briefly, 0.5 mL of serum was mixed with a proprietary reagent provided in a kit and incubated for 30 min at 4 °C. After mixing, the sample was centrifuged at 10,000×g for 10 min at RT. EVs pellets were resuspended in 100 μL PBS. For size-exclusion chromatography, 0.5 mL serum was loaded into a qEV column (Izon Science, Medford, MA), and flushed with PBS, fractions 7-12 were collected and pooled in 3 mL according to the manufacturer's instructions.
EV quantification. The size and concentration of EVs were measured using a tunable resistive pulse sensing (TRPS) method (qNano, Izon Science). Samples were filtered before processed in the qNano instrument using a 1-μm syringe filter (Tisch Scientific, North Bend, OH). 45 μL of biofluid was pipetted into different nanopore stretchable membranes (NP150, NP300, NP600, NP800, and NP1000 from Izon Science) to cover the wide size range of EVs 36  Immunofluorescence staining. After  Flow cytometry and total internal reflection fluorescence (TIRF) microscopy. tEVs from U251 GBM cells were stained with a lipophilic fluorescent dye, SP-DiOC18(3) (Thermo Fisher Scientific) for 20 min, the excess dye was washed out using TFF. 100 μL fluorescent EVs (~ 10 11 particles/mL) were then spiked into 500 μL of serum (~ 10 12 particles/mL) and 100 mL of urine (~ 10 9 particles/mL) from healthy donors. Each of the samples was processed by iSUF and recovered in 100 μL of PBS (stage 1-3). Non-spiked fluorescent tEVs were also captured on functionalized beads as a positive control. After washing with PBS, the captured EVs were analyzed by imaging flow cytometry (Amnis, ImageStream X Mark II Imaging Flow Cytometer, Luminexcorp, Austin, TX); also images were taken using a TIRF microscope (Eclipse Ti Inverted Microscope System, Nikon, Melville, NY) with a 100 × oil immersion lens. For comparison, samples were also processed by UC and resuspended in 100 μL PBS. For total RNA quantification, captured EVs were lysed, and RNA was extracted and quantified using the same procedures mentioned above.

Definition of terms used in this study.
where V E0 and C E0 are the initial volume and concentration of the sample; V TFF (in 2 mL) and C TFF (measured by qNano pore size [NP150, NP300, NP600, NP800, and NP1000]) are the final volume and concentration of the TFF product.
where V 0 and C P0 are the initial volume of the sample and concentration of free proteins in the sample, respectively; V t and C Pt are the volume of the TFF retentate and concentration of free proteins in the retentate at a specified time t, respectively. Concentration of free proteins was measured using the bicinchoninic acid (BCA) where C E0 is the EV concentration in the initial sample; C Ef is the EV concentration in the final iSUF product. Concentration of samples were measured by qNano pore size [NP150, NP300, NP600, NP800, and NP1000]).
where V E0 is the initial volume of the product; V Ef is the volume of the final iSUF product.
where C Ef is the final concentration of the iSUF product (in 100 μL); C Pf is the remained free protein concentration in the product.
The coefficient of variation (CV) is a measure of relative variability. It is the ratio of the standard deviation (σ) to the mean (μ).

Statistical analysis.
Data are expressed as the mean ± STD. A significant test between different mean values was evaluated using one-way ANOVA in JMP Pro 16 software provided by The Ohio State University. Differences between samples were considered statistically significant for p < 0.05. Exponential simulation. Curve fitting and probability distribution calculation were conducted using exponential simulation in JMP Pro 16. The goodness of fit test was validated using Kolmogorov D. Hypothesis is the data comes from the exponential distribution (significance level α = 0.05).

Results
Optimization of the iSUF platform. To overcome current limitations of the enrichment and purification of EVs and on-demand EV subpopulation characterization, we developed the iSUF platform ( Fig. 1 A) which includes three stages: (1) tangential flow filtration (TFF) for the enrichment and purification of EVs, (2) centrifugation for further enrichment of EVs, and (3) immunomagnetic affinity selection for desired EV subpopulation isolation. We used the iSUF platform to process various volumes (0.5-100 mL) of different biofluids (CCM, serum, and urine).
To design stage 1, many parameters of TFF processing required optimization, including the selection of membrane pore size (MWCO), sample processing temperature, sample flow rate, pressure, and sample protein concentration. We tested membrane filters with two MWCO sizes (300 and 500 kDa) to determine the optimal MWCO that maximizes the removal of free proteins and nucleic acids while reducing processing time. Our experiments showed that 500 kDa filter membranes were able to remove up to 99% of free proteins with over 99% EV recovery rate (Eq. 1, Supplementary Fig. 1). When a 300 kDa membrane filter was used, only 80% of free proteins were removed. Moreover, a 500 kDa membrane filter was chosen since it processed samples 2-3 times faster than a 300 kDa membrane filter (Supplementary Table 1). The TFF stage 1 of iSUF ran at 4 °C to minimize EV degradation 37 . We further tried to optimize sample processing time, which was highly dependent on the flow rate. The flow rate was linearly associated with the shear rate generated by the filter based on the manufacturer's protocol (Repligen), which exerted a shear force on the EVs. Then, we used a flow rate of 35 mL/ min to maintain a shear rate below 5000/s 38 . High flow rates increased the system pressure, mainly when the protein concentration of the sample was high (> 15 mg/mL). We kept the pressure of the system below 68.9 kPa to avoid leakage and maximize the lifespan of the 500 kDa filter. We used dilutions of fetal bovine serum (FBS) to test the effect of protein concentration on system pressure at 35 mL/min. Our results showed that protein concentration must be equal or lower than 15 mg/mL to maintain the pressure of the system below 68.9 kPa to protect the filter (Supplementary Fig. 2).
At stage 2 of iSUF, EVs were loaded into centrifugal filter units and were spun down at 3000×g. We compared the recovery rates of EVs and processing time for different filter pore sizes. A 3 kDa filter unit obtained over 99% recovery rate and took 60 min to spin down, while a 10 kDa unit obtained a 95% recovery rate in 20 min, and a 30 kDa filter obtained only a 70% recovery rate and took 15 min to spin down ( Supplementary Fig. 3). We selected the 10 kDa filter unit to maintain a high recovery rate while reducing sample processing time. To enrich tEVs (stage 3 of iSUF), we tested incubation of 3 μm magnetic beads with different concentrations of biotinylated antibodies (10, 20, 100 μg/mL) for 1 h, 2 h, and 12 h at RT, and 4 °C. Overnight incubation with 20 μg/mL of antibody was able to yield an optimal bead-antibody coverage. Then, we examined the volume ratio of beads to EVs at 5 μL/100 μL, 20 μL/100 μL, and 80 μL/100 μL during the bead/tEVs incubation step. The 20/100 μL ratio achieved the highest bead-tEVs capture efficiency (> 90%).
Scientific Reports | (2021) 11:8034 | https://doi.org/10.1038/s41598-021-86910-y www.nature.com/scientificreports/ iSUF platform for processing biofluids. A flow rate of 35 mL/min for stage 1 of iSUF was applied since the protein concentrations of the different biofluids were below 15 mg/mL (Supplementary Table 2). We tested the ability of our platform to purify and enrich EVs from three different biofluids (i.e., CCM, serum, and urine; Fig. 1B). 50 mL of CCM, and 100 mL of urine were enriched to 7 mL. Subsequently, PBS diafiltration buffer was used to remove the remaining protein contaminants from CCM and urine in 80 min and 100 min, respectively. The percentage of remaining protein is defined as the mass of free proteins in the TFF retentate at a specified time divided by their initial mass in the sample (Eq. 2). For serum, the initial high concentration of proteins (> 80 mg/mL) required an initial dilution of 0.5 mL of the sample in 7 mL of PBS. Subsequently, PBS diafiltration buffer was used to remove the remaining protein contaminants from serum in 120 min. We first tested iSUF with a 10% BSA solution for which an SDS-PAGE gel showed extensive removal of albumin ( Supplementary Fig. 4). Moreover, analysis of the purified samples by an SDS-PAGE gel indicated that iSUF removed BSA from CCM, human serum albumin (HSA) and globulins from serum, and Tamm Concentration, size distribution and microscopy characterization of EVs. For 50 mL of CCM, 0.5 mL of serum, and 100 mL of urine, the enrichment factors were 489 ± 18, 4.8 ± 0.1, and 942 ± 19, respectively (n = 5; Fig. 2A) (Eq. 3). Accordingly, the EV recovery rate (the ratio of the total number of EVs post-processing by iSUF to the total number of EVs pre-processing) was 98% ± 3.6%, 96% ± 2.0% and 94% ± 1.9% for CCM, serum, and urine, respectively (Eq. 4). Considering that EVs are heterogeneous in size, we also tested the enrichment factor across a wide size range of EVs (70 nm-1 μm) in CCM with similar results (n = 5; p > 0.05; Fig. 2B). Moreover, the EV concentrations obtained by iSUF was greater than the concentrations obtained by UC. This difference was consistent across the 70 nm-1 μm size range (n = 5; p < 0.05; Fig. 2C), with iSUF enriching EVs at two to three orders of magnitude higher than UC (approximately 10 11 EVs concentrated by iSUF and 10 9 for UC). Similar results were obtained with comparisons for the enrichment of EVs present in serum and urine. With iSUF, EVs from serum and urine were enriched almost at the same level (10 12 EVs/mL; Fig. 2D). We confirmed the presence of EVs by using atomic force and electron microscopy on the different biofluids processed. The majority of isolated EVs exhibited a round morphology with heterogeneous size distribution. Cryo transmission electron microscopy images (TEM) images of isolated EVs showed the presence of a double-layered lipid membrane, a representative characteristic of EVs 39 (Fig. 2E, F).
To further test our iSUF platform, we performed comparative studies with different commercially available EV isolation methods: qEV, TEI, and UC. 0.5 mL of serum from healthy donors were processed with different EV isolation platforms. EVs have shown a comparable size range for all platforms, but iSUF obtained a higher EV concentration within the 70 nm-1 μm size range (Fig. 3A, Supplementary Fig. 8). Notice that all plots presented a right-skewed distribution. Moreover, we plotted the probability distribution as reported previously (Supplementary Fig. 9) 40 , which showed an exponential decay trend for all the isolation methods that were verified by the Kolmogorov D test (p > 0.05) 40 . However, qEV, TEI, and UC did not show this trend clearly (Fig. 3A) due to large losses of EVs during the isolation process. Following the MISEV2018 guidelines, subpopulations of EVs can be described independently of the EV biogenesis pathway 36 . Such that subpopulations of small EVs (sEVs) and medium/large EVs (m/l-EVs) can be defined based on size, with EVs smaller than 100 nm or 200 nm as sEVs and EVs larger than 200 nm as m/l-EVs. Considering that the stretchable nanopore membranes NP150 and NP300 cover the corresponding size range for sEVs and m/l-EVs, respectively 41 . we compared the mean size of these EV subpopulations by different EV isolation methods (Fig. 3B). For sEVs, EVs processed by iSUF were significantly smaller than UC (n = 5; p < 0.05). For m/l-EVs, EVs purified by UC were larger than all other techniques (n = 5; p < 0.05). Then, we compared the total concentration and purity of isolated EVs using different methods. iSUF enriched EVs significantly more efficiently than all the tested methods (n = 5; p < 0.05; Fig. 3C). The concentration of EVs enriched by iSUF was on average 51, 7, and 56 times higher than qEV, TEI, and UC. The purity of isolated EVs was normalized and evaluated in terms of the ratio between EV concentration and remaining contaminating protein concentration present in samples after purification using the different methods (Eq. 5). The purity of the isolated EVs by iSUF was 100-10,000 times higher than the different tested methods (Fig. 3D).

Molecular content quantification and characterization of EVs.
We quantified the total amount of protein and RNA present in EVs isolated by iSUF. For CCM, the quantity of RNA obtained was 11 ± 8.2 ng/mL, thus giving a ninefold higher concentration of RNA when compared to UC (Fig. 4A). For protein analysis, the protein concentration was 6 ± 1.8 μg/mL, almost eightfold more protein obtained than when the same biofluid was processed by UC (Fig. 4B). Total RNA and protein quantification were carried out for serum and urine from 5 healthy donors. 0.5 mL of serum and 100 mL of urine processed by iSUF produced 54 ± 40 ng/mL and 0.3 ± 0.1 ng/mL of RNA, respectively (Fig. 4C). The protein concentration was 1250 ± 480 μg/mL for serum and 3.1 ± 3.0 μg/mL for urine (Fig. 4D). The differences in protein concentration of EVs in original CCM, processed by UC, and processed by iSUF were also demonstrated using CD63 and CD9 western blot analysis. Also, the expression level of CD9 and CD63 in original urine and serum samples, and after iSUF processing demonstrated superiority enrichment and purification performance of the iSUF (Fig. 4E). Full-length western blot images are displayed in (Supplementary Fig. 11). Moreover, we compared the RNA content obtained from EVs   Table 3). The total RNA content obtained using iSUF was 54 ± 41 ng, which was significantly higher than other methods that only obtained 10 ± 9.8 ng (Fig. 4F). Next, we processed different volumes of serum and urine with iSUF to identify the equivalent volumes of biofluid that will produce comparable concentrations of EVs, total RNA, and total protein. We started with 0.5 mL of serum and different volumes of urine (i.e., 100, 75, 65, 10 mL). We found that the enriched concentrations of EVs were comparable (~ 10 12 particles/mL, Supplementary Fig. 12). The total RNA content obtained from EVs from serum and urine also exhibited comparable values of 60 ± 35 ng and 75 ± 40 ng, respectively. However, analysis of protein content of the isolated EVs from both biofluids showed that the protein content of EVs isolated from urine was tenfold lower than the protein content of EVs in serum, which was 0.4 ± 0.1 mg and 40 ± 10 mg, respectively.
Immunomagnetic affinity selection of the iSUF platform. tEVs are outnumbered by EVs from nontumor cells that require removal to perform accurate molecular analysis 42 . One way to isolate tEVs is to exploit the presence of specific surface markers 43 . Our iSUF platform enables the separation of subpopulations of EVs by capturing them on magnetic beads through immunoaffinity. To demonstrate our approach, we used a model system that consisted of spiking fluorescently labeled tEVs from a cancer cell line (i.e., U251 EVs) in 0.5 mL of serum and 100 mL of urine from healthy donors (n = 3). 3-μm magnetic beads were functionalized with EGFR as the capture antibody. All samples were processed through all stages of iSUF, and the captured subpopulation of EVs was characterized and quantified using TIRF and imaging flow cytometry (Fig. 5A). A comparable fluorescent signal was obtained between the positive control (tEVs from U251 cells) and spiked samples processed in serum and urine (88% ± 1% for serum, and 93% ± 2% for urine), which verified the high recovery efficiency and their purity of tEVs. The biofluids spiked with tEVs were also processed by UC, which yielded a significantly lower fluorescent signal (3.1 ± 0.1 times lower, Fig. 5A). Furthermore, to determine tEV RNA isolation efficiency by iSUF, lysed U251-EVs (positive control) and bead-isolated U251-EVs were extracted. iSUF showed over 90% RNA isolation efficiency for the EGFR + tEVs (Fig. 5B). Moreover, the RNA content of bead-extracted tEVs from serum using the iSUF platform was sevenfold higher than UC, which was mainly attributed to the high yield of Red dots were purities defined as the EV concentration divided by the remaining free protein concentration. iSUF isolated EVs from serum most efficiently with high purity (n = 5; p < 0.05) (mean ± STD), qEV yielded pure EVs, but at a lower concentration, TEI isolated more EVs but at relatively low purity, and UC recovered EVs with the lowest concentration and purity. www.nature.com/scientificreports/ total EVs recovered in iSUF (Fig. 5C). The high RNA content of tEVs from serum isolated using iSUF also confirmed the ability of iSUF to enrich and retrieve tumor targets from mixed populations of EVs.

Detection of proteins and miRNA in EVs from clinical samples. The emergence of targeted thera-
pies require a precise characterization of the molecular subtypes of a patient's tumor 44 . For BC patients (e.g., luminal A/B, triple-negative), the molecular subtype is typically determined by testing a tissue biopsy for the presence or absence of three essential proteins: estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2) 45 . However, a tissue-based tumor profile is subject to sampling bias, provides only a snapshot of tumor heterogeneity, and cannot be obtained repeatedly 46 . These limitations restrict early cancer detection and significantly contribute to the overdiagnosis and overtreatment of BC patients 47 . The analysis of EVs present in cancer patients biofluids provides a minimally invasive method to quantify different biomarkers that would enable precise diagnosis or response to treatment 48 . To test the potential use of the iSUF platform in screening biomarkers for BC, we processed 0.5 mL of serum from 10 metastatic BC patients and quantified the expression levels of HER2, CD24, and miR21 on patient isolated EVs [49][50][51] . After stage 3 of iSUF, magnetic beads were isolated using a magnet and incubated with detection antibodies and MBs. Then, the magnetic beads were processed with imaging flow cytometry and TIRF microscopy. Similar volumes of serum from healthy donors were processed and analyzed. The age of the healthy donors was between 21 and 68 years old. Although the age range was wider than from our BC patient cohort, a comparison of levels of expression for HER2, CD24, and miR21 between healthy donors samples did not show statistically significant differences (p > 0.05, Supplementary Fig. 13. We obtained representative images and histograms that quantify the number of beads and their corresponding fluorescent intensities (Fig. 6A, B) for BC patient and healthy donor samples. Both protein and miRNA show significantly higher expression in BC patient samples (p < 0.05; Fig. 6B), which suggested that iSUF can differentiate tumor biomarkers (Supplementary Table 4). Moreover, we have characterized the CD9 and CD63 levels of protein expression in serum samples from both cohort of samples. Although in this case, similar levels of expression were obtained (p > 0.05, Fig. 6A, B).

Discussion
Recently, EVs have been explored for diagnostic and therapeutic applications, including liquid biopsy assays for cancer diagnostics, and nanocarriers for drugs and nucleic acids 52,53 . The large number of tEVs compared to other rare biomarkers (i.e., circulating tumor cells, CTCs) makes them more statistical reliable 54 . Innovative methods developed for the enrichment and purification of EVs should remove all contaminants (e.g., free proteins), have a high yield, work amongst different biofluids, and maintain the integrity of the EVs. We have engineered a new platform that includes a TFF enrichment and purification stage, a centrifugal-unit enrichment stage, and a magnetic separation stage for specific isolation of subpopulations of EVs (e.g., tEVs). In stages 1 and 2, iSUF performs enrichment and purification of all EVs (e.g., tEVs and nEVs). In stage 3, tEVs are isolated based on their tumor-specific surface markers. 300 kDa and 500 kDa TFF filters were initially selected for this study because the molecular weights of the major free proteins in CCM (e.g., BSA, 65 kDa), serum (e.g., HSA, 65 kDa) and urine (e.g., THP, 98 kDa) were below 300 kDa [55][56][57] . After testing, a 500 kDa TFF filter was finally chosen for subsequent experiments because of its minimal EV loss and faster processing time (~ 2 h). Different biofluids required different TFF diafiltration buffer volumes and different scale of TFF hollow fiber filters to remove most contaminants. CCM and urine contain relatively low concentrations of protein contaminants 58,59 , so a small total diafiltration buffer volume was required to remove the majority of free proteins. Unlike CCM and urine, the serum has a significant amount of free proteins 60 . Therefore, a large total diafiltration buffer was necessary to remove the free protein in serum. iSUF was able to achieve up to 99% of free protein removal efficiency. Free proteins removed by iSUF were shown as decreased bands on the SDS-PAGE gel. Other bands were increasing on the SDS-PAGE gel because those correspond to different proteins present on EVs, which were enriched by the iSUF [61][62][63][64][65] . Also, there are very-low-density lipoproteins (VLDL)) that overlap in size with EVs that also might be enriched 2,66,67 .
Although TFF can concentrate and purify samples, the final product volume is mostly dependent on the dead volume of the specific TFF system. Therefore, further enrichment is necessary to concentrate the samples (e.g., spin down to 100 μL) for applications such as the diagnosis of rare biomarkers 68 . Centrifugal units with different MWCOs (3, 10, and 30 kDa) were evaluated in terms of processing time and yield in our study. A larger MWCO size required shorter centrifugation time, but more EVs were lost because prolonged high-speed centrifugation elongated EVs into an oval shape which made them squeeze through the filter membrane 69,70 . The centrifugal unit with the 10 kDa MWCO was optimal with the shortest processing time and highest yield.
We also compared EVs purified from serum and urine samples of healthy donors. EVs from urine and serum can serve as prognosis biomarkers for clinical analysis 71,72 . Serum is the most commonly used biofluid in the clinical setting with the highest EV concentration 73 . Compared to serum, urine collection is minimally invasive and can be obtained in larger volumes 74 , but it usually suffers from much lower EV concentration 75 . Interestingly, using our iSUF platform, the EV RNA concentration in the final products of urine samples (originating from ~ 100 mL of collected sample volume) were comparable to those of serum samples (originating from 0.5 mL of collected sample volume). This suggests that clinical diagnosis by urinary EVs is possible. However, there is still a concern towards urinary EV collection because of large variabilities in the urine volume and its EV concentration. More efforts are needed to come up with a gold standard protocol for urine collection and processing. For protein in urinary EVs, we found that the amount of protein was lower than the amount of protein obtained from serum EVs, which might be explained by the degradation of EV membrane proteins by urine proteases 76 .
In this study, we compared the performance of iSUF with three different EV isolation techniques (TEI, qEV, and UC) to determine EV concentration, purity, and quantity of RNA recovered. As other authors have www.nature.com/scientificreports/ discussed 77,78 , we found that UC, a traditional method for EV isolation, has raised concerns about the integrity, yield, and purity of EVs after purification. Interestingly, we found UC-isolated EVs were larger than other platforms, one possible explanation is the presence of extensive levels of protein aggregates that cause bias. For our UC method, washing steps using multi-rounds ultracentrifugation were not applied, since it was shown that purity of the samples did not increase significantly when adding washing steps, and a significant loss of enriched EVs could occur 55,[79][80][81][82][83] . Moreover, although qEV obtained a relatively pure product, its recovery rate was low; TEI obtained a relatively higher EV number, while retaining massive amounts of free protein (Fig. 3C, D). Therefore, low EV concentration and purity will impact the accuracy of molecular analysis of EVs 84 . We used the MISEV 2018 guidelines to define EVs: small EVs (sEVs) and medium and large EVs (m/l-EVs) 36 . Following these definitions, we found that the concentration and average size of EVs in patient samples and healthy controls varied (Supplementary Table 4). Different factors may explain these differences. Subtypes and stages of BC in the patient cohort, may contribute to the difference in the metrics. Moreover, we obtained our patient samples from a biobank, we had minimal control of the storage conditions, thus it was possible that some degradation of the samples could occur. It is known that factors such as material of the container, temperature of storage, and cycles of freezing and thawing may also influence the concentration and mean size of EVs [85][86][87][88] . Considering all these aspects, the size distributions and concentrations of EVs that we obtained for the different samples are within the reported values in the literature. Previously concentrations of EVs in serum from cancer patients are between 10 9 and 10 11 particles/mL [89][90][91][92] .
Differentiating EV particles and other non-EV particles (e.g., lipoproteins) when measuring size distribution and concentration is challenging for all current methods, including TRPS and nanoparticle tracking analysis (NTA) 66,[93][94][95][96][97] . This is because the size range of different lipoproteins [e.g., 7-13 nm (HDL); 21-27 nm (LDL); and 30-90 nm (VLDL)] 98 may overlap with the reported size ranges for sEVs and m/l-EVs 36 . Thus, current methods for EV size characterization and concentration have certain limitations 88,[99][100][101] . For iSUF, we used the TRPS method to measure size distribution and concentration of EVs. Different nanopore stretchable membranes (NP150, NP300, NP600, NP800, and NP1000) were selected, covering a broad size range of EVs such as 70-1000 nm 102 . The 70 nm cut-off were chosen since EVs in samples processed by iSUF, exhibited a range Figure 4. Characterization of protein and RNA content of EVs isolated from different biofluids. (A) Quantity of total RNA extracted from CCM-EVs using UC and iSUF (n = 6; p < 0.05). (B) Comparison of proteins extracted from CCM-EVs using UC and iSUF (n = 6; p < 0.05). (C) Quantification of total RNA extracted from serum and urine EVs isolated by iSUF. (D) Quantification of EV proteins isolated from serum and urine EVs isolated by iSUF. (E) Western blot analysis of CD63 and CD9 expression in original CCM, CCM-EVs processed using iSUF and UC, original serum and urine, serum and urine-EVs processed using iSUF. The expression level of CD63 and CD9 was calculated using an equal mass (40 μg) of protein lysates from samples. (F). Total RNA of EVs isolated using different EV isolation methods. Black boxplots were the absolute amount of RNA isolated from 0.5 mL of serum. iSUF isolated the most RNA contents from serum EVs (n = 5; p < 0.05) (mean ± STD). ▸ Figure 5. Characterization of stained tEVs isolated from U251-CCM and spiked in different biofluids. (A) Upper. Fluorescence images of tEVs on magnetic beads using TIRF microscopy. Lower. Comparison of fluorescence intensities quantified using imaging flow cytometry. Stained U251-EVs isolated by iSUF was a positive control. tEVs isolated from serum and urine samples using iSUF demonstrated no significant difference (n = 3; p > 0.05), while UC isolated tEVs differed with iSUF and the positive control (n = 3; p < 0.05) (mean ± STD). (B) Quantification of total RNA isolated from U251-EVs before and after bead extraction showed no differences (n = 3; p > 0.05). (C) Total RNA extracted from serum samples using iSUF was significantly higher than using UC (n = 3; p < 0.05) (mean ± STD).  Fig. 14). Thus, iSUF intrinsically filters out the majority of lipoproteins and EVs that overlap in size. This effect is beneficial since we are trying to characterize EVs with minimal contribution of lipoproteins. We consider the 70 nm cut-off used for the TRPS measurements as appropriate. Moreover, above this value the different characterization metrics are considered accurate. The particle count measurements can verify this for each one of the methods used in our study ( Supplementary Fig. 15). A linear relationship indicates an accurate measurement 30 . In Fig. 3, we found that the lower limit of detection varies for the different methods, indicating an intrinsic loss of EVs during processing samples with different methods. However, iSUF shows better recovery for sEVs and m/l-EVs. The wide upper range showed by UC could be indicative that we measure aggregates of EVs [103][104][105][106] . Our measurements are in agreement with MISEV 2018 since sEVs, and m/l-EVs are in the 100-150 nm and larger than 200 nm, respectively 36,63,107-111 . We are interested in the enrichment and isolation of tEVs to characterize tumor-related proteins and RNAs. Like other immunoaffinity methods [112][113][114] , iSUF captured and isolated EVs using specific tumor surface proteins. iSUF differentiated metastatic BC patients from healthy donors by detecting significantly higher expression levels of proteins and RNA biomarkers present in EVs (e.g., HER2, CD24, and miR21). Based on previous reports, HER2 and miR21 are cancer-associated protein and microRNA species, and are known to be overexpressed in metastatic BCs 115,116 . Compared to miR21 and HER2, CD24 is relatively less investigated in BC but was previously identified as being released from BC stem cells 117 . Furthermore, a recent study indicated that serum CD24 is elevated among BC patients 118 . Moreover, it is important to note that one of these biomarkers may not be a reliable predictor of BC alone. However, the combination of several biomarkers can serve as a tool for BC risk assessment.
In conclusion, iSUF was proposed for rapid, efficient, and specific isolation of EVs from different biofluids. The EV recovery rate mean value was above 94%, with 90% tumor RNA isolation efficiency and negligible concentrations of free proteins and nucleic acids. Although iSUF does not process a sample in the shortest time (80-120 min), its versatility working with different biofluids, sample volumes, and high purity after sample processing constitutes unmatched advantages over current methods used in the field. Overall, we found that the iSUF platform isolated and enriched EVs from a scaled-up sample volume with high purity and yield in a sterile and quick manner simultaneously, and isolated tEVs with high specificity, while other current methods could not guarantee all of those conditions at the same time. Furthermore, we recognize that the iSUF platform has potentially broad clinical applications beyond liquid biopsies for cancer diagnosis or monitoring. Characteristic fluorescence images of HER2, CD24, miR21, CD63, and CD9 expression on isolated EVs immobilized on magnetic beads from BC patients and healthy donors. Right. Fluorescence intensity histograms of HER2, CD24, miR21, CD63, and CD9 of isolated EVs for BC patients (n = 10) and healthy donors (n = 14). (B) Total fluorescence intensity quantification of HER2, CD24, miR21, CD63, and CD9 expression level on isolated EVs from BC patients and healthy donors. Patients demonstrated higher expression levels of HER2, CD24, and miR21 than healthy donors (p < 0.05). Patients and HDs demonstrated similar expression level of CD63 and CD9 (p > 0.05). P patient; H healthy donor. ◂