View all hidden authors and organizations

Edited by Robert H. Austin, Princeton University, Princeton, New Jersey, approved on July 29, 2021 (review received on April 8, 2021)

Extracellular vesicles (EV) are ubiquitous in humans and contribute to cell-to-cell communication. Their content, size, number and surface markers depend on their cytoplasmic source and protein complexes involved in membrane transport. Therefore, vesicles can be used as valuable biomarkers and can be obtained by liquid biopsy. However, as of today, phenotype-specific EV subgroups have not been clearly determined, mainly because EVs can be absorbed and modified by other cells, which hinders the production of heterogeneous EVs in a single cell. The proposed method can robustly classify EVs secreted by single cells and evaluate their heterogeneity in surface markers. In the future, the tool can be used to evaluate the release and absorption dynamics of electric vehicles in a controlled environment.

Extracellular vesicles (EV) are continuously secreted from eukaryotic and prokaryotic cells. EVs, including those called exosomes, may have an impact on cell signaling and the incidence of diseased cells. In this manuscript, a platform for capturing, quantifying and phenotypic classification of EVs secreted by single cells is introduced. A microfluidic chamber of approximately 300 pL is used to capture and separate individual cells. EVs secreted in these chambers are then captured by monoclonal antibodies (mAb) immobilized on the surface, regardless of their intracellular origin. The immunostaining of plasma membrane and cytoplasmic proteins is combined with a highly sensitive multicolor total internal reflection fluorescence microscope to characterize immobilized vesicles. Data analysis of high-resolution images allows each detected EV to be assigned to one of 15 unique populations and proves that there is a highly heterogeneous phenotype even at the single-cell level. The analysis also showed that each mAb separated EVs with different phenotypes, and when targeting CD63 instead of CD81, more vesicles were effectively fixed. Finally, we showed how to obtain heterogeneous inhibition in secreted vesicles when the enzyme neutral sphingomyelinase is inhibited.

Mammalian cells secrete compartments separated by lipid bilayers, commonly called extracellular vesicles (EV) (1). The heterogeneous whole of these vesicles includes differences in membrane composition, encapsulated content, size, and cell origin (2, 3). Vesicles are usually divided into small vesicles (40 to 200 nm) of intracellular origin, large microvesicles (200 to 1,000 nm) derived from the plasma membrane of the cell, and larger apoptotic bodies (>1 µm). Smaller EVs, also called exosomes, have received increasing attention in the past 30 years. Since then, EVs have been shown to affect cells at different levels, from cell movement to stem cell immune regulation (4, 5). Due to the biologically active nature of its contents (for example, membrane proteins [Figure 1A], nucleic acids, carbohydrates, and lipids), the regulation of exosomes facilitates the identification of their biological purposes and functions under physiological and pathological conditions. Such as diabetes, cancer and neurodegenerative diseases (6⇓ ⇓ –9). In addition, these vesicles have attracted people's interest as hypothetical biomedical diagnostic tools or therapeutic drug carriers (10⇓ ⇓ –13).

The schematic diagram shows the motivation and experimental method. (A) EVs carrying nucleic acids, proteins and lipids have attracted much attention as potential biomarkers, but they cannot be distinguished correctly. In the context of its biological origin, electric vehicles are also regarded as possible carriers of pharmaceutical active compounds. (B) PDMS-based two-layer microfluidic device describes the fluid layer (blue), with two inlets (top), one outlet (bottom) and a second pneumatic control layer on top (red). (Scale bar 5 mm.) (C) Phase contrast image of cell culture wells. The external valve closes the area where the electric vehicle is fixed. The internal valve allows the isolation of hydrodynamically trapped cells during subsequent coatings. (Scale bar 100 µm.) (D) A functionalized surface with biotinylated BSA, NeutrAvidin, and biotinylated monoclonal antibody, used to immobilize the secreted EV, then labeled with a fluorescent-conjugated antibody and imaged in the four-color TIRFM. (E, i) The pneumatic valve on the microfluidic device can realize the surface functionalization of space control. (ii) A single cell is trapped in a hydrodynamic trap. (iii) In the cell culture process, the EV fixing surface coating and BSA between the ring valves are introduced to prevent non-specific adsorption outside the outer ring valve. (iv) During cell culture, the cells secrete directly fixed EVs. (v) Secreted and fixed EVs are immunostained. (vi) Then use TIRFM to image the EV.

Although EVs are usually enriched from cell culture supernatants using ultracentrifugation, they are also concentrated from body fluids, such as blood (14), urine (15), and saliva (16). The immunochemical characteristics of EVs can also be used to isolate them, for example, by targeting a four-span membrane protein integrated into the plasma membrane (2). Recent methods use microfluidic technology through passive transmission methods [e.g. filtration (17), hydrodynamic operations (18) and deterministic lateral displacement (19)] or through the application of external fields [e.g. acoustophoresis (14) and electromotive force (20) )]. The data collected using these techniques highlights the diversity of EVs/exosomes and the difficulty of accurately identifying or classifying them. In fact, the usage and definition of terms such as "exosomes" and "microvesicles" are unclear (2, 8, 21, 22). The production and transport of EVs are severely affected by cell-to-cell communication, and their release and absorption mechanisms contribute to all levels of (patho)physiology (6, 7, 9).

However, it is not yet clear how EVs secreted by cells are taken up, modified, and re-released into large numbers of samples by other cells. In order to clarify the biogenesis and secretion of EVs, it is important to evaluate the statistics of single-cell EV secretion without potentially interacting with other cells. In addition to direct contact with other cells and extracellular matrix, as well as the most important signal factors (usually the focus of single-cell research), this will further understand the most important clues to the interaction between cells.

Here, we introduced a microfluidic strategy to capture and culture single cells, fix their secreted EVs, and classify them according to their phenotype, regardless of their cytoplasmic origin. Therefore, a microfluidic device based on polydimethylsiloxane (PDMS) with two concentric pneumatic valve arrays is proposed. The sequential actuation of these valves enables independent functionalization of the areas where cells and EVs will be incubated and fixed respectively. Although other platforms have been reported for the analysis of EVs derived from single cells (23), the proposed design allows the analysis of EV populations without any cross-contamination with other cells. In our method, the antibody-based coating (for EV capture) is performed after individual cells are separated in the central chamber. Only EVs secreted from the captured cells are effectively captured in a small area between the two pneumatic valves. Therefore, the immobilization of EVs (suspended in large cell cultures) during sample introduction is prevented, which is a key challenge for studying EVs secreted by single cells (23, 24). After immunostaining several proteins, the fixed EV was imaged using a four-color total internal reflection fluorescence microscope (TIRFM), where TIRFM provides the necessary optical sensitivity and lateral resolution to classify a single EV into 15 unique Phenotypic group. We demonstrate that EV subpopulations occur even at the single-cell level and can be identified by the proposed method.

Our focus is to collect EVs secreted by single cells and perform their phenotypic-specific classification (Figure 1A). In addition to isolating individual cells, this task requires selective functionalization to promote the immobilization of cells and EVs in different areas. Therefore, we designed a double-layer microfluidic device with two concentric valves, as shown in Figure 1B. The first layer includes the fluid layer, where cells are introduced and captured by 72 PDMS columns that act as hydrodynamic traps (Figure 1C). The second (pressure) layer contains two circular donut-shaped valves on each column, which can be lowered around the trapped cells to create an isolation chamber (Figure 1C). Although it is common to use microfluidic valves to create cell culture chambers (25, 26), these two concentric valve designs allow once the cells are isolated and protected by the central chamber, the area where the EV will be fixed can be functionalized, avoiding initial crossover -Pollution.

The operation of the device can be described as follows. The area inside the inner flap serves as a cell culture area, initially coated with fibronectin to promote cell adhesion (Figure 1D, i). Then the cell suspension is washed, and individual cells are hydrodynamically captured by the PDMS column (Figure 1D, ii). Once the cells are separated and protected, the annular area between the valves is coated with antibodies against the epitopes in the EV membrane, and the area outside the outer valve is coated with bovine serum albumin (BSA) (Figure 1D, iii). Final cell culture The chamber is realized by closing and opening the external and internal valves respectively (Figure 1D, iv). During the culture process, the EV secreted by the cells is fixed in the antibody-containing area between the valves (Figure 1D, iv). After 24 hours, the valve was reopened to provide fluorescently labeled antibodies to the EV (Figure 1D, v), and the system was imaged by four-color TIRFM (Figure 1D, vi). The area outside the outer valve is coated with BSA to prevent non-specific adsorption (Figure 1E).

The device was initially evaluated for its ability to promote adhesion and fixation of cells and EVs in adjacent non-overlapping areas, respectively. To verify the differential functionalization, the surface between the two valves is coated with BSA-biotin and streptavidin-phycoerythrin (PE), while the area within the inner valve (cell capture area) is coated with BSA-biotin , NeutrAvidin and biotin-fluorescein isothiocyanate (FITC). The fluorescence generated in the non-overlapping area is shown in Figure 2A. Next, check the ability of the fibronectin coating area to promote cell adhesion by washing the cells, capturing them, and verifying their adhesion (Figure 2B-D). The membrane-stained, biotinylated, large unilamellar vesicles (LUV) were tested using antibodies to prepare annular regions for vesicle capture. After functionalizing these areas with BSA-Biotin and NeutrAvidin, rinse the LUV into the chip, incubate for a period of time (Figure 2E), and then wash. Then check and enumerate the fixed LUV (Figure 2F and 3A and B) (Figure 3C). The area blocked by BSA showed ≤6 LUV per image, while the fixed vesicles in the unblocked area increased with the increase of incubation time. After 0.5 hours, about 10 LUVs per image were detected (double-sided Kolmogorov-Smirnov [KS] test, DKS = 0.98, P = 2.58 × 10−12). After 2 hours of incubation, the number increased by 10 times (two KS Test, DKS = 1.0, P = 5.27 × 10−11). These results demonstrate the blocking of non-specific vesicle binding in the BSA inactivation area and the continuous immobilization of LUV in the functionalized area. Please note that in TIRFM, the incident excitation beam is reflected by the glass surface of the chip, creating an evanescent field within approximately 200 nm on top of the fixed surface of the vesicle. Therefore, non-immobilized vesicles above the surface will not be recorded.

Space surface functionalization for cell culture and vesicle fixation. (A) After incubating the respective areas, observe the spatial surface functionalization in an epi-fluorescence microscope with false-color PE (red) and FITC (green). These areas are separated by actuating valves. The image shows a central green area containing hydrodynamic cell traps (two unstained central black ellipsoids because they are plasma bound to the microscope cover glass). The cell culture area is surrounded by a black unstained ring, reflecting the PDMS-PDMS valve slide interface, and is surrounded by a fixed area stained red. (BD) Brightfield images showing single (B and C) and multiple (D) incubating cells directly captured (B) and <1 hour later (C and D). (Scale bar 50 µm.) (E) After showing the main surface coating, we used BSA-Biotin and NeutrAvidin to cultivate and immobilize biotinylated LUVs containing biotinylated lipids in a circular area enclosed by a double valve, and Prevent them from being fixed outside by passivating the area with 4% by mass heat-denatured BSA. (F) Fluorescence microscope image of immobilized LUV. (The scale bar is 200 µm.)

TIRFM displays functionalized surfaces for LUV and EV fixation. (A) The TIRFM image of the BSA passivation control area. The EV is not fixed. Next to it is the (B) BSA-Biotin-NeutrAvidin functionalized LUV capture area. The detected LUV has increased significantly. (Scale bar is 5 µm.) (C) Quantifying the immobilized LUV of each image shows that the number of immobilized vesicles increased after 0.5 and 2 hours compared with the control. The experiment was repeated twice, with at least 93 measurements each time. (DH) TIRFM images of MCF-7 cells excited in 405 (CD63-BV421, blue false color), 488 (HSP70-FITC, green), and 561 (TSG101-PE, yellow) 24 hours after the immobilized EV, And 640 nm (ANXA5-Alexa Fluor 647, red) and merge. The signal is highlighted with a white arrow. (The scale bar is 5 µm.)

Next, evaluate the platform's ability to quantify EVs secreted by single cells. The central chamber is coated with fibronectin, and the secreted EV is captured by CD81. After 24 hours of culture, the cells and EVs were fixed with paraformaldehyde (PFA), and the entire device was treated with BSA to prevent non-specific adsorption. Then the target epitopes (ie CD63, HSP70 [70 kilodalton heat shock protein] and TSG101 [tumor susceptibility gene 101]) of the EV immunostaining and the phosphatidylserine (PS) in the outer membrane lobules (different EVs ) Use Annexin V (ANXA5). The four-color TIRFM reveals the location of the immobilized EVs and their fluorescence signals, which are used to determine whether they are combined with any of the four mAbs provided (Figure 3D-H). Then enumerate the EVs detected in each well: if they bind to at least one antibody conjugate, the EVs are counted as positive, as shown in Figure 3H. To determine whether wells containing single cells show more EVs than empty wells, the signal distributions identified as vesicles in the two cases were compared. Figure 4A shows this vesicle distribution after 24 hours of incubation, where >1,000 TIRFM images were analyzed. The frequency distribution of EVs in single-cell wells is different from that of empty wells (two-sided KS test, 1 vs. 0 cells, DKS = 0.51, P <2.2 × 10−16). The distribution core of the empty wells is located in the low EV of each image area, in which 45% of the zero signals are identified as vesicles, while the proportion of the wells occupied alone is 8%. The signal detected in the unoccupied wells can be attributed to non-specific background adsorption. However, such signals are rare, as only 26% of unoccupied wells show more than 1 EV per image, and 76% of single-cell wells show more than 1 EV. Using the median as an indication of central tendency, wells occupied by single cells contained three times higher signal than unoccupied control wells (Figure 4B).

EV is detected at the single-cell level. (A and B) The relative frequency distribution of images taken from the EV capture area after MCF-7 cells were cultured for 24 hours. The chip is functionalized with CD81 for fixation. Compared with unoccupied control wells, the area in the wells occupied by single cells has a higher signal density. (B) Compared to the control, the relative frequency distribution shows an increased number of signals (larger median). (C) Represents the relative frequency distribution of the signal density of the image in the wells occupied by 1 cell, 2 cells, and 3 or more cells. (D) Corresponding to the relative increase in the EV population, we detected an absolute increase in the number of EVs, and the quadruple positive EV population increased by 10 times. (E) Quantifying the classification ratio of EV signals in each image shows that most EV populations in the wells occupied by single cells have increased. (F) To test the generation of biased image analysis and artifacts, we performed a two-tailed Pearson correlation coefficient analysis on the wells occupied by single cells and the control wells, showing that there is no correlation bias based on the analysis.

When comparing single-cell wells with wells with multiple cells, as the number of cells increased, it was observed that the EV distribution of each image gradually "flattened out" (Figure 4C). The EV distribution of wells with one cell shows a prominent peak, each image has about 3 EVs (82% of the wells contain 1 to 5 EVs per image) and light tail (only 7% of the wells contain more than 6 EVs). In contrast, wells containing a pair of cells showed less localized peaks (83% of wells contained 1 to 9 EVs per image). Data analysis revealed a flatter distribution of wells containing three or more cells. The prominent peaks observed at approximately three EVs per image of single-cell wells are now distributed over a larger range, with 83% of wells containing 1 to 16 EVs. In addition, the tail of the distribution becomes heavier because a larger proportion of wells contain >10 EVs (4%, 6%, and 27% for wells with 1, 2, and 3 or more cells, respectively). Therefore, the proposed technique is sensitive enough to distinguish the number of EVs secreted by one, two, and three or more cells (two-sided KS test: 1 pair 2 cells, DKS = 0.25, P = 2.7 × 10−7 ; 1 compared with 3+ cells, DKS = 0.31, P = 6.4 × 10−13; 2 and 3+ cells, DKS = 0.21, P = 7.7 × 10−4). The non-linear nature of EV secreted as a function of the number of cells per well indicates the complex dynamic behavior behind EV uptake and release (27, 28). However, further experiments on paired cells are needed to clarify the effects of mixed EV (secreted by different cells) and uptake-release kinetics. The results presented in this manuscript achieve this goal by evaluating the statistical behavior of individual cells.

Next, our platform is used to analyze the phenotype of EVs secreted by single cells. In addition to ANXA5 (against PS), antibodies against HSP70, TSG101 and CD63 were also provided to the EV. After washing, TIRFM analysis showed that although many EVs tested positive for the four markers, not all EVs showed signals above the detection threshold in the four channels. Population occurrence is described in Figures 4D and E. The quadruple-positive vesicle subgroup (ANXA5+HSP70+TSG101+CD63+) is the largest subgroup, followed by the double-positive CD81+CD63+ subgroup (Figure 4D). The observed distribution of EV populations emphasizes the heterogeneity of their secretion and strengthens the hypothesis that these subpopulations are secreted at different frequencies (29). In addition, these results emphasize the need for several markers to adequately identify EVs or exosomes. In order to test whether there is any correlation between the detected fluorescent signals, in this case, artificial similarity between EV populations will be introduced, and the Pearson correlation coefficients of all fluorescent channel pairs will be calculated (Figure 4F). The correlation matrix shows that there is no strong positive or negative correlation between the detected signals, either in the control or in the wells occupied by single cells.

It is currently under discussion whether there are membrane integral proteins targeting EVs. (2, 27) The reported "exosomal-specific" markers include four transmembrane proteins (eg, CD9, CD63, and CD81), HSP70, and flotillin-1 (2, 27, 30). The main difficulty in selecting a single confirmed marker is the lack of understanding of the complete biogenesis and pathways of EVs (including exosomes) in cells. Therefore, the proposed platform was used to evaluate and compare the fixation efficiency of two mAbs with the common four-span membrane protein classes CD63 and CD81 (Figure 5). The analysis of about 500 TIRFM images per channel showed that after a single MCF-7 cell was cultured for 24 hours, the signal distribution detected in the CD63-mAb functionalized well was significantly greater (two-sided KS test, DKS = 0.94, P <1 × 10-15) than those detected in the CD81-mAb wells (Figure 5A). The distribution core of CD81 immobilized EVs is concentrated at about 3 EVs per image, and 29% of the images show less than or equal to one vesicle. The other 29% contained 4 to 8 vesicles, and only 4% had a maximum of 22 EVs per image. In contrast, the peak of the CD63 positive distribution is located at approximately 40 EVs per image. More than half (57%) of the images displayed 20 to 40 EVs, only 10% contained 10 to 20 EVs, and 4% displayed <10 EVs per image. The proposed combination of the microfluidic device and the four-color TIRFM can characterize the 95th percentile increase of each functionalization strategy by a factor of 10 (Figure 5B). These results indicate that the choice of marker introduces a bias in the measurement (this is the case in all affinity positive selection methods), which should be taken into account in the analysis process.

Fixation technology affects fixation efficiency and detected population composition. (A) Histogram and (B) Scatter plots show the number of fixed vesicles for fixing EVs from a single MCF-7 cell using CD81 or CD63 in 24 hours. When the EV is fixed with CD63 instead of CD81, the EV of each image will increase. (C) The bar graph shows the standardized change in the number of vesicles after fixation with CD63 or CD81. (D) The bar graph shows the fold change of EVs aggregated into the ANXA5+ and CD63+ EV populations. After fixation with CD63, the population of ANXA5+ increased, and the number of CD63+ vesicles decreased (tested against CD81-immobilized EV). (E) The color-coded pie chart summarizes the relative composition of the EV population when fixed with CD81 (left) or CD63 (right), corresponding to the EV population clustered by ANXA5+- and CD63+-.

In order to determine the effect of the fixation method on the EV phenotype composition, the EV populations that occurred were clustered considering the different phenotypes. The incidence of CD63+ and ANXA5+ EVs showed that compared with CD63-immobilized EVs, the population of CD81-immobilized EVs in ANXA5+ vesicles was lower (χ2 test was used to detect differences after treatment; P values ​​were both <3.574 × 10-4 cases; The exact value of each comparison can be found in the SI appendix). We found that the relative abundance of EVs that were immobilized by anti-CD63 and carried ANXA5 increased by 20 and 50-fold, depending on the respective phenotypic population (Figure 5D). However, the same EV does not contain a secondary detectable CD63 antigen. The loss of co-detected secondary CD63 (when using CD63 instead of CD81 for fixation) occurred evenly across the entire CD63-positive population. The overall change in the composition of the EV population is shown in Figure 5E, where the phenotypic EV population is positively aggregated by CD63 or ANXA5, which is not related to the immobilized mAb. When using CD81, the CD63+ and ANXA5+ populations accounted for 65% and 35% of all vesicles (Figure 5E). In contrast, when the anti-CD63 mAb was used to immobilize EVs, the CD63-positive EV population was reduced to 2% of all CD63+ populations (Figure 5D and E). However, capturing EVs via CD63 resulted in an increase in PS-rich EV levels to approximately 98% (Figure 5D and E).

Using our platform, we studied the effect of GW4869, an enzyme inhibitor that down-regulates the enzyme neutral sphingomyelinase. It has previously been reported that the number of exosomes will decrease after blocking with GW4869 (31), because the inhibitor is involved in the invagination process of multivesicular bodies (32). Therefore, we provided a 5 µM GW4869 solution for the isolated cells and analyzed the changes in the composition of the secreted EV. If the entire population is compared, it is observed that the number of secretory vesicles of the cells treated with the inhibitor is reduced (two-sided KS test, DKS = 0.87, P <1 × 10-15) (Figure 6A and B). However, more in-depth Studies have shown that the observed population reduction is caused by the downregulation of a small number of subpopulations (Figure 6C). The effect of GW4869 is evaluated by comparing the number of vesicles produced by each subgroup when treated and untreated. We focus on those vesicles that show significant differences (all the compared P values ​​and D statistics are in the SI appendix Given). After fixing the EV with CD63, it was found that the CD81+ANXA5+ and CD81+STAM1+ANXA5+ subgroups showed the strongest downregulation (three times reduction; KS test on both sides, DKS = 0.83, P = 0.026). These categories were followed by CD81+TSG101+ANXA5+ and ANXA5+, which showed a two-fold decline (two-sided KS test, DKS = 0.83, P = 0.026). In contrast, ESCRT-dependent STAM1+TSG101+ and TSG101+ classes showed a relative increase in vesicle production. These results seem to indicate a strong effect on the secretion of vesicles rich in PS, CD81, STAM1, or TSG101, but a small effect on vesicles secreted through the ESCRT-dependent pathway (31, 33).

The sphingomyelinase inhibitor GW4869 reduces the secretion of certain EV populations. (A) A histogram of the relative frequency distribution of EVs secreted by a single MCF-7 cell incubated in 5 µM GW4869 for 24 hours without (control) shows that vesicle secretion is reduced when GW4896 is present. (B) The relative detection frequency of EVs decreased after incubation with GW4869. (C) Population analysis of EV secretion showed that under GW4869 incubation, the relative reduction of EVs was up to three times. Compared with the control, CD81+ANXA5+ (and CD63+) and the relative up-regulation of TSG101+ and TSG101+STAM1+ EVs. The P values ​​for all pairs (compared to the control) are shown in the figure.

Mammalian cells secrete EVs, and their morphology and biochemical composition are different. Given the large heterogeneity and small size of these vesicles, it is extremely challenging to identify them, distinguish their phenotypes, and trace their biological origins. Therefore, the enrichment and classification of EVs usually depend on their physical properties, such as density and diameter. In this manuscript, we propose a microfluidic strategy to fix EVs secreted by single cells and classify them by phenotype. Although the creation of a separate incubation chamber is a common method in microfluidic devices (26, 27), the proposed two-valve design enables selective functionalization of the annular area next to the cultured cells. In this way, EVs are fixed and analyzed in situ to prevent them from breaking or other changes, which usually occurs when an external enrichment step is required (Figure 1D). More importantly, the design allows for the functionalization of this circular area with antibodies after the introduction and protection of cells. This prevents any contamination of fixed EVs when introducing large amounts of cell culture supernatant, which is a common problem in single-cell research dealing with EVs (24).

After the EV is fixed, immunostaining of several proteins and imaging using four-color TIRFM facilitates the enumeration of each of their surface areas, where TIRFM provides the necessary optical sensitivity and lateral resolution to approximate the phenotypic specificity of a single EV Classification. Based on their use in the characterization and identification of EVs in previous studies, several membrane integral proteins and cytoplasmic protein markers were selected for the fixation and immunostaining of EVs (2, 23). However, the cytoplasmic origin of EVs was not considered in the analysis.

In principle, lipophilic dyes such as DiO can be used as a control for staining and counting EVs (34⇓ ⇓ –37). However, morphological artifacts of DiO-stained EVs were observed during TIRFM imaging (SI Appendix, Figures SI5 and SI6). Three lipophilic dyes are used (octadecyl rhodamine B chloride [R18], 9-(diethylamino)-5H-benzo[a]phenoxazine-5-one [Nile Red] and PKH26 [ SI appendix, Figure SI7] An additional control experiment was performed]). They all show significantly greater non-specific background adsorption, which prevents reliable EV quantification in wells occupied by one or more cells. Therefore, we gave up the use of non-specific tags.

The largest EV population was positive for all selected markers, confirming the immobilization of EVs (including exosomes) on the microfluidic device. In addition, these results demonstrate that antibody staining is not sterically hindered. In contrast, populations that only carry one, two, or three epitopes demonstrate the heterogeneity of EVs in their origin and/or size. These results show deviations in the measurement (relative to the selected antibody) and need to be considered before the results are generalized.

As expected, and according to most reports on EVs (30), four transmembrane proteins were detected. In addition to the four transmembrane proteins representing integrated membrane epitopes, STAM1 and TSG101 were also detected. The former is part of the ESCRT-0 and -I subunits and is involved in EV trafficking (38). In addition, EVs carrying HSP70 have been detected [a member of the heat-inducible and constitutively expressed HSP family that have been reported in EVs (23)]. PS-rich vesicles (via ANXA5) have also been identified to confirm the intracellular origin of EVs (2, 39). In order to confirm the membrane perforation effect of PFA, we were able to stain the cytoplasmic protein (ie, the cavity localization protein), and the mAb could thus pass through the lipid bilayer, and the PS was localized through the cytoplasm of ANXA5 (SI appendix, Figure 2). SI2A) and CD63 (SI appendix, Figure SI2B) have been verified.

Due to the high incidence of EVs carrying CD63 epitopes, the efficiency of EV capture using anti-CD81 and anti-CD63 was compared. When anti-CD63 was used, the number of immobilized EVs increased significantly, and the results were consistent with previous comparisons with the four-transmembrane protein EV capturing EVs (40, 41), indicating that the number of vesicles with CD81 epitopes is small. The larger capture efficiency may also be related to the larger binding constant of the anti-CD63 antibody (42).

Further experiments on immobilized electric vehicles have shown that the proportions of subgroups are different. For example, EVs immobilized by anti-CD63 no longer show CD63 epitopes on the free (surface avoidance) side. We speculate that either the free diffusion of the antibody bound to the surface (in the lipid membrane of CD63) exhausts the presence of CD63 in other areas of the EV membrane, or the random distribution of microdomains (TEM) rich in four-span membrane proteins leads to EVs (Via CD63) and potential CD63-negative (but CD81-positive) TEM presentation, as described by Nydegger et al. (43), Deneca et al. (44) and Charrin et al. (45, 46) When the immobilized antibody was changed from CD81 to CD63, a significant change in the population composition of EVs was observed, leading to an increase in the proportion of ANXA5+ vesicles. Jeppesen et al. earlier reported similar results regarding ANXA-positive EVs. (47). ANXA5 is a commonly used marker to detect early cell apoptosis and changes in cell membrane asymmetry caused by apoptosis (48). However, ANXA5 was used here to detect phosphatidyl-rich EVs. Therefore, EVs of cytoplasmic origin, such as apoptotic cells, are excluded from the analysis (using brightfield microscopy). Therefore, we considered the possibility that EVs secreted by MCF-7 cells and fixed by CD63 are rich in phosphatidylserine. This result may indicate that the CD63+PS+ subpopulation has a cytosolic source (39, 49, 50).

GW4869 inhibits neutral sphingomyelinase as a negative control and is often used to identify specific epitopes of EVs and the physiological activities of EVs and so-called subpopulations (eg, on other cell types) (31, 51). GW4869 acts on the secretion pathway of exosomes independent of ESCRT (52, 53), down-regulating the production of exosomes. Therefore, the effect of this enzyme inhibitor at the single-cell level was quantified. Although data analysis shows that EV secretion is generally down-regulated, only some people are strongly affected. The vesicle production in the CD81+ANXA5+ and CD81+ANXA5+STAM1+ cell populations was largely suppressed, while the vesicle production in the TSG101+ and STAM1+TSG101+ cell populations was up-regulated. This observation is similar to other cell types. It is also reported below (54).

In conclusion, due to the introduction of antibodies after cell fixation, the proposed microfluidic device can robustly isolate individual cells and prevent any cross-contamination. The combination of immunostaining and four-color TIRFM can reliably count and classify EVs secreted by single cells. Our findings indicate that the fixation strategy severely affects the phenotypic composition of the detected EVs, and the negative regulation of GW4896 has a non-uniform effect on EVs. The ability to track changes in phenotypic EV units, in terms of applied fixation methods, will allow in the future to link metabolic and genetic conditions with the formation and secretion pathways of EVs, thereby linking cells to single-cell communication.

Michigan Cancer Foundation (MCF)-7 cells are maintained in standard continuous cell culture conditions in Dulbecco's Modified Eagle Medium (DMEM) and supplemented with 1 g ⋅ L−1 glucose, pyruvate and 10 vol-% fetal bovine serum (both From Thermo Fisher Scientific) at 37 °C, 5 vol-% CO2 and 95% relative humidity. Cells are trypsinized (0.05% 2,2',2'',2'''-(Ethane-1,2-diyldinitrilo)tetraacetic acid [EDTA], Thermo Fisher Scientific), at a ratio of 1:5 per week Passed twice. The on-chip culture conditions include 1× penicillin-streptomycin (Thermo Fisher Scientific) and 2 vol% fetal bovine serum with exosomes removed (System Biosciences). GW4869 was purchased from Sigma-Aldrich and diluted according to the dealer's recommendation.

The PDMS device is manufactured using Sylgard 184 silicone elastomer matrix and curing agent (Dowsil, formerly Dow Corning Midland) with a mass ratio of 10:1. The mixture was degassed until there were no bubbles visually. Next, pour 40 grams of the mixed PDMS onto the master mold containing the pressure layer and place the mold at 80°C for >2 or 24 hours. Subsequently, 5 g of PDMS was spin-coated (20 seconds at 500 rpm, 40 seconds at 2,800 rpm) on the fluid layer mold and cured at 80°C for 1 hour. This process produces a flexible film approximately 25 µm thick. After curing, the pressure layer is peeled from the master mold. Then use a 1 mm biopsy punch (Integra) to punch the pressure inlet hole. To bond the pressure layer to the fluid layer, pour 2 to 3 milliliters of curing agent onto a blank 4-inch silicon wafer and spin-coat at 6,000 rpm for 1 minute. The stamping pressure layer equipment is dropped onto a blank silicon wafer coated with curing agent, peeled off, and manually aligned with the fluid layer. The edges of the combined layer were sealed with a degassed PDMS mixture. The assembled chip is cured for 2 hours at 80°C.

After curing, the assembled PDMS layer was peeled from the fluid mold, and the inlet/outlet was punched using a 1.5 mm biopsy punch (Integra Miltex York). Use tape to clean the assembled equipment. At the same time, 1 g of the PDMS mixture was spin-coated on the No. 1 microscope slide (6,000 rpm for 60 seconds), covering them with a 10 µm thick PDMS layer. The PDMS layer was then allowed to reflow for more than 30 minutes at room temperature (RT), and then cured overnight at 80 °C. The double-layer PDMS device and the PDMS-coated microscope slide were plasma activated (PDC-32G, Harrick Plasma) at approximately 0.77 mbar for 45 seconds (18 W) and bonded together. Place the glass bonded device on a hot plate at 100°C for 10 minutes and store at room temperature before use.

All devices are filled with miliQ water before use, by rotating at 800 g for 10 minutes, and incubating at 37 °C, 95% relative humidity and 5 vol% CO2 for >30 minutes. These devices were then connected to a 10 ml syringe (Becton Dickinson) via polytetrafluoroethylene tubing (PKM SA), silicon tubing (Gobatec) and polyetheretherketone micro-fittings. The syringe is loaded on the syringe pump (NE-1002X-ES, World Precision Instruments, or NanoJet, Chemyx, Inc.). The pressure channel is connected to the silicone tube through a bent metal pin, and the silicone tube is connected to the pressurized air through a manually customized control unit (Cole-Parmer). Rinse the device with sterile phosphate buffered saline (PBS) without (w/o) Ca2+ and Mg2+ (Sigma-Aldrich). In order to promote cell adhesion, each cell capture area on the entire device was rinsed with 100 ng ⋅ mL-1 fibronectin (Sigma-Aldrich) and incubated for 30 minutes (internal valve closed), and then rinsed with Ca2+-free PBS . After incubation, clean the chip with fetal bovine serum supplemented with exosomes removed (2 vol%, 1× penicillin-streptomycin, 1 g ⋅ L-1 glucose DMEM [exoPSFDMEM]). Next, the trypsinized and filtered (35 µm) cells are washed, captured, and incubated at 37 °C, 5 vol-% CO2. After the cells are captured, the inner loop valve is activated. In order to minimize the adverse effects on the isolated cells, such as the accumulation of metabolic compounds or the appearance of nutritional stress, which may affect the biogenesis and secretion of EVs, all incubation steps are limited to 24 hours. The medium for each cell is approximately 5.5 × 10-2 µL, which corresponds to the volume of fresh medium for each cell when the confluence is 60% under standard cell culture conditions. In contrast, after 72 hours of incubation, the culture medium per cell is approximately 3.3 × 10-2 µL, which is equivalent to 100% confluence. The EV capture area was incubated with 2 mg ⋅ mL-1 biotinylated BSA (Thermo Fisher Scientific) for 30 minutes, followed by 100 g ⋅ mL-1 Neutravidin and 5 ng ⋅ mL-1 biotinylated conjugates. Immobilize antibody, anti-CD63 or anti-CD81 (BioLegend), for 30 minutes. After incubation in the EV capture area, rinse the device with exoPSFDMEM without exosomes.

For the preparation of biotinylated LUV, a 20 mM lipid solution composed of 95 mol-% 1-palmitoyl-2-oleoyl-glycero-3-phosphocholine (16:0 to 18:0) and 5 mol-% 1, 2dioleoyl-composition prepared sn-glycerol-3-phosphoethanolamine-N-(biotinyl) (DOPE) in chloroform (Sigma-Aldrich). DOPE is a biotinylated lipid that is introduced into the LUV membrane to make the LUV model as similar to EV as possible. All lipids were purchased from Avanti Polar Lipids, Inc. Use a rotary evaporator to remove chloroform (200 mbar and room temperature for 10 minutes, then at 5 mbar for 15 minutes) from a 10 mL lipid solution to generate a dry lipid film in a glass flask and room temperature, and at 20 mbar and room temperature Dry for 1 hour). The lipid film was then rehydrated in PBS. The suspension has undergone 10 freeze-thaw cycles in liquid nitrogen and a water bath (T = 40 °C). The resulting vesicles were then extruded in a micro extruder (Avanti Polar Lipids, Inc.) using a polycarbonate membrane (Sigma-Aldrich) with a pore size of 100 nm. The size distribution of the vesicle suspension was determined in Zetasizer Nano ZSP (Malvern Panalytical GmbH). Coating microfluidic chips with BSA-Biotin and NeutrAvidin has led to a similar approach to EV targeting via BSA-Biotin, Neutravidin, and biotinylated mAbs.

The cultured cells were washed in PBS containing Ca2+ and Mg2+, fixed in 4 vol-% paraformaldehyde (pH 7.2) in PBS containing Ca2+ and Mg2+, and then blocked in 4 mass% heat shock denaturing BSA, and then 0.1 Quality to stain the respective epitopes -% Heat shock denaturation BSA contains Ca2+ and Mg2+ in PBS. Antibodies were purchased from Biolegend UK Ltd (biotinylated anti-CD63 and anti-CD81 [both 5 μg ⋅ mL-1], anti-CD63-PerCPCy5.5, anti-CD81-PE, ANXA5-Alexa Fluor 647) and from Santa Cruz Biotechnology (HSP70 [HSC70/HSPA8]–FITC, STAM1-FITC, TSG101PE).

Several experiments have been performed to select from empty wells (0 cells, n ≥ 448) and containing single cells (1 cell, n ≥ 448), two cells (2 cells, n ≥ 696), and three or more cells (+ 3 cells, n ≥ 696). After image analysis (described in detail in the SI appendix), the frequency distribution of the detection signal for each image is calculated; each image represents an area of ​​4,356 µm2. If the frequency distribution of the detected signal is significantly different among all the pair combinations in the set {0-cell, 1-cell, 2-cell, +3-cell}, the platform is considered to be sensitive. Use the two-sided KS test to test the difference in distribution. However, the KS test hypothesis is not related and is known to produce conservative P values ​​when applied to discrete data. To solve this problem, we adopted the bootstrap method with nb = 10,000 samples to calculate the zero distribution of the D statistic and provide a corrected P value (55). In all cases, the value of P is lower than P ≤ 7.7 × 10−4, indicating that there is enough information to reject the equality in the signal distribution. In addition, the permutation test of the equation is performed by approximating the density distribution (56), and all combinations show that the P value is approximately zero. Although noisy signals (ie non-specific adsorption) were detected in the empty wells, almost half (45%) of the processed images showed no signal at all. In contrast, only 7% of the images obtained from wells containing single cells showed no signal. The same ratio (7%) was found in wells containing two cells and wells containing three or more cells, indicating that the suggested workflow is reproducible. The proposed method fixes a part of the vesicles secreted by the cells on the surface of the isolation chamber, and then uses TIRFM to detect those attached to the bottom substrate (see description, operation and verification of the microfluidic device). We want this kind of bottom substrate attachment to be random. Therefore, the independence distance covariance test is used to confirm the independence between images sampled from the same well (57). Follow the guide method of nb = 10,000 repetitions per case. In all cases, the P value is greater than 0.001 (distance covariance DCov> 1,932), indicating that there is not enough information to reject the non-correlation between variables, except in some isolated cases (less than 0.1 and 1.0% of the analysis is significant The sex level is 0.001 and 0.01 respectively).

Derived data sets generated and/or analyzed during the current study are available at Dryad, https://doi.org/10.5061/dryad.dz08kprz5 (58). The image processing code can be obtained as a supplementary file with Zenodo online, https://doi.org/10.5281/zenodo.5211393 (59).

We thank Carola Alampi and Evi Bieler (BioEM Laboratory and the Swiss Institute of Nanosciences of the University of Basel) for TEM and scanning electron microscope imaging, the Department of Biological Systems Science and Engineering (Eidgenössische Technische Hochschule [ETH] Zurich) and Chao-Chen Lin ( ETH Zurich) provided assistance with microscopes and data analysis, Darius Rackus (ETH Zurich) performed proofreading, and Dirk Loeffler, Marina Nikova (ETH Zurich) and Jasmin John (F. Hoffmann-La Roche) performed science discuss. We are very grateful for the funding from the European Research Council (Consolidator Grant No. 681587, HybCell, to PSD).

Author contributions: JMN and PSD design research; JMN conducted research; JMN and MAS-E. Analyze the data; MAS-E. And the master model made by AK; JMN and PSD wrote this paper.

The author declares no competing interests.

This article is directly contributed by PNAS.

This article contains online support information https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.2106630118/-/DCSupplemental.

This open access article is distributed under Creative Commons Attribution 4.0 (CC BY).

Thank you for your interest in advertising on PNAS.

Note: We only ask you to provide your email address so that the people you recommend the page to know that you want them to see it and that it is not spam. We do not capture any email addresses.

Feedback privacy/legal

Copyright © 2021 National Academy of Sciences. Online ISSN 1091-6490. PNAS is a partner of CHORUS, COPE, CrossRef, ORCID and Research4Life.